1 //===-- ARMInstrNEON.td - NEON support for ARM -------------*- tablegen -*-===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file describes the ARM NEON instruction set.
12 //===----------------------------------------------------------------------===//
15 //===----------------------------------------------------------------------===//
16 // NEON-specific Operands.
17 //===----------------------------------------------------------------------===//
18 def nModImm : Operand<i32> {
19 let PrintMethod = "printNEONModImmOperand";
22 def nImmSplatI8AsmOperand : AsmOperandClass { let Name = "NEONi8splat"; }
23 def nImmSplatI8 : Operand<i32> {
24 let PrintMethod = "printNEONModImmOperand";
25 let ParserMatchClass = nImmSplatI8AsmOperand;
27 def nImmSplatI16AsmOperand : AsmOperandClass { let Name = "NEONi16splat"; }
28 def nImmSplatI16 : Operand<i32> {
29 let PrintMethod = "printNEONModImmOperand";
30 let ParserMatchClass = nImmSplatI16AsmOperand;
32 def nImmSplatI32AsmOperand : AsmOperandClass { let Name = "NEONi32splat"; }
33 def nImmSplatI32 : Operand<i32> {
34 let PrintMethod = "printNEONModImmOperand";
35 let ParserMatchClass = nImmSplatI32AsmOperand;
37 def nImmVMOVI32AsmOperand : AsmOperandClass { let Name = "NEONi32vmov"; }
38 def nImmVMOVI32 : Operand<i32> {
39 let PrintMethod = "printNEONModImmOperand";
40 let ParserMatchClass = nImmVMOVI32AsmOperand;
42 def nImmVMOVI32NegAsmOperand : AsmOperandClass { let Name = "NEONi32vmovNeg"; }
43 def nImmVMOVI32Neg : Operand<i32> {
44 let PrintMethod = "printNEONModImmOperand";
45 let ParserMatchClass = nImmVMOVI32NegAsmOperand;
47 def nImmVMOVF32 : Operand<i32> {
48 let PrintMethod = "printFPImmOperand";
49 let ParserMatchClass = FPImmOperand;
51 def nImmSplatI64AsmOperand : AsmOperandClass { let Name = "NEONi64splat"; }
52 def nImmSplatI64 : Operand<i32> {
53 let PrintMethod = "printNEONModImmOperand";
54 let ParserMatchClass = nImmSplatI64AsmOperand;
57 def VectorIndex8Operand : AsmOperandClass { let Name = "VectorIndex8"; }
58 def VectorIndex16Operand : AsmOperandClass { let Name = "VectorIndex16"; }
59 def VectorIndex32Operand : AsmOperandClass { let Name = "VectorIndex32"; }
60 def VectorIndex8 : Operand<i32>, ImmLeaf<i32, [{
61 return ((uint64_t)Imm) < 8;
63 let ParserMatchClass = VectorIndex8Operand;
64 let PrintMethod = "printVectorIndex";
65 let MIOperandInfo = (ops i32imm);
67 def VectorIndex16 : Operand<i32>, ImmLeaf<i32, [{
68 return ((uint64_t)Imm) < 4;
70 let ParserMatchClass = VectorIndex16Operand;
71 let PrintMethod = "printVectorIndex";
72 let MIOperandInfo = (ops i32imm);
74 def VectorIndex32 : Operand<i32>, ImmLeaf<i32, [{
75 return ((uint64_t)Imm) < 2;
77 let ParserMatchClass = VectorIndex32Operand;
78 let PrintMethod = "printVectorIndex";
79 let MIOperandInfo = (ops i32imm);
82 // Register list of one D register.
83 def VecListOneDAsmOperand : AsmOperandClass {
84 let Name = "VecListOneD";
85 let ParserMethod = "parseVectorList";
86 let RenderMethod = "addVecListOperands";
88 def VecListOneD : RegisterOperand<DPR, "printVectorListOne"> {
89 let ParserMatchClass = VecListOneDAsmOperand;
91 // Register list of two sequential D registers.
92 def VecListDPairAsmOperand : AsmOperandClass {
93 let Name = "VecListDPair";
94 let ParserMethod = "parseVectorList";
95 let RenderMethod = "addVecListOperands";
97 def VecListDPair : RegisterOperand<DPair, "printVectorListTwo"> {
98 let ParserMatchClass = VecListDPairAsmOperand;
100 // Register list of three sequential D registers.
101 def VecListThreeDAsmOperand : AsmOperandClass {
102 let Name = "VecListThreeD";
103 let ParserMethod = "parseVectorList";
104 let RenderMethod = "addVecListOperands";
106 def VecListThreeD : RegisterOperand<DPR, "printVectorListThree"> {
107 let ParserMatchClass = VecListThreeDAsmOperand;
109 // Register list of four sequential D registers.
110 def VecListFourDAsmOperand : AsmOperandClass {
111 let Name = "VecListFourD";
112 let ParserMethod = "parseVectorList";
113 let RenderMethod = "addVecListOperands";
115 def VecListFourD : RegisterOperand<DPR, "printVectorListFour"> {
116 let ParserMatchClass = VecListFourDAsmOperand;
118 // Register list of two D registers spaced by 2 (two sequential Q registers).
119 def VecListDPairSpacedAsmOperand : AsmOperandClass {
120 let Name = "VecListDPairSpaced";
121 let ParserMethod = "parseVectorList";
122 let RenderMethod = "addVecListOperands";
124 def VecListDPairSpaced : RegisterOperand<DPair, "printVectorListTwoSpaced"> {
125 let ParserMatchClass = VecListDPairSpacedAsmOperand;
127 // Register list of three D registers spaced by 2 (three Q registers).
128 def VecListThreeQAsmOperand : AsmOperandClass {
129 let Name = "VecListThreeQ";
130 let ParserMethod = "parseVectorList";
131 let RenderMethod = "addVecListOperands";
133 def VecListThreeQ : RegisterOperand<DPR, "printVectorListThreeSpaced"> {
134 let ParserMatchClass = VecListThreeQAsmOperand;
136 // Register list of three D registers spaced by 2 (three Q registers).
137 def VecListFourQAsmOperand : AsmOperandClass {
138 let Name = "VecListFourQ";
139 let ParserMethod = "parseVectorList";
140 let RenderMethod = "addVecListOperands";
142 def VecListFourQ : RegisterOperand<DPR, "printVectorListFourSpaced"> {
143 let ParserMatchClass = VecListFourQAsmOperand;
146 // Register list of one D register, with "all lanes" subscripting.
147 def VecListOneDAllLanesAsmOperand : AsmOperandClass {
148 let Name = "VecListOneDAllLanes";
149 let ParserMethod = "parseVectorList";
150 let RenderMethod = "addVecListOperands";
152 def VecListOneDAllLanes : RegisterOperand<DPR, "printVectorListOneAllLanes"> {
153 let ParserMatchClass = VecListOneDAllLanesAsmOperand;
155 // Register list of two D registers, with "all lanes" subscripting.
156 def VecListDPairAllLanesAsmOperand : AsmOperandClass {
157 let Name = "VecListDPairAllLanes";
158 let ParserMethod = "parseVectorList";
159 let RenderMethod = "addVecListOperands";
161 def VecListDPairAllLanes : RegisterOperand<DPair,
162 "printVectorListTwoAllLanes"> {
163 let ParserMatchClass = VecListDPairAllLanesAsmOperand;
165 // Register list of two D registers spaced by 2 (two sequential Q registers).
166 def VecListDPairSpacedAllLanesAsmOperand : AsmOperandClass {
167 let Name = "VecListDPairSpacedAllLanes";
168 let ParserMethod = "parseVectorList";
169 let RenderMethod = "addVecListOperands";
171 def VecListDPairSpacedAllLanes : RegisterOperand<DPair,
172 "printVectorListTwoSpacedAllLanes"> {
173 let ParserMatchClass = VecListDPairSpacedAllLanesAsmOperand;
175 // Register list of three D registers, with "all lanes" subscripting.
176 def VecListThreeDAllLanesAsmOperand : AsmOperandClass {
177 let Name = "VecListThreeDAllLanes";
178 let ParserMethod = "parseVectorList";
179 let RenderMethod = "addVecListOperands";
181 def VecListThreeDAllLanes : RegisterOperand<DPR,
182 "printVectorListThreeAllLanes"> {
183 let ParserMatchClass = VecListThreeDAllLanesAsmOperand;
185 // Register list of three D registers spaced by 2 (three sequential Q regs).
186 def VecListThreeQAllLanesAsmOperand : AsmOperandClass {
187 let Name = "VecListThreeQAllLanes";
188 let ParserMethod = "parseVectorList";
189 let RenderMethod = "addVecListOperands";
191 def VecListThreeQAllLanes : RegisterOperand<DPR,
192 "printVectorListThreeSpacedAllLanes"> {
193 let ParserMatchClass = VecListThreeQAllLanesAsmOperand;
195 // Register list of four D registers, with "all lanes" subscripting.
196 def VecListFourDAllLanesAsmOperand : AsmOperandClass {
197 let Name = "VecListFourDAllLanes";
198 let ParserMethod = "parseVectorList";
199 let RenderMethod = "addVecListOperands";
201 def VecListFourDAllLanes : RegisterOperand<DPR, "printVectorListFourAllLanes"> {
202 let ParserMatchClass = VecListFourDAllLanesAsmOperand;
204 // Register list of four D registers spaced by 2 (four sequential Q regs).
205 def VecListFourQAllLanesAsmOperand : AsmOperandClass {
206 let Name = "VecListFourQAllLanes";
207 let ParserMethod = "parseVectorList";
208 let RenderMethod = "addVecListOperands";
210 def VecListFourQAllLanes : RegisterOperand<DPR,
211 "printVectorListFourSpacedAllLanes"> {
212 let ParserMatchClass = VecListFourQAllLanesAsmOperand;
216 // Register list of one D register, with byte lane subscripting.
217 def VecListOneDByteIndexAsmOperand : AsmOperandClass {
218 let Name = "VecListOneDByteIndexed";
219 let ParserMethod = "parseVectorList";
220 let RenderMethod = "addVecListIndexedOperands";
222 def VecListOneDByteIndexed : Operand<i32> {
223 let ParserMatchClass = VecListOneDByteIndexAsmOperand;
224 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
226 // ...with half-word lane subscripting.
227 def VecListOneDHWordIndexAsmOperand : AsmOperandClass {
228 let Name = "VecListOneDHWordIndexed";
229 let ParserMethod = "parseVectorList";
230 let RenderMethod = "addVecListIndexedOperands";
232 def VecListOneDHWordIndexed : Operand<i32> {
233 let ParserMatchClass = VecListOneDHWordIndexAsmOperand;
234 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
236 // ...with word lane subscripting.
237 def VecListOneDWordIndexAsmOperand : AsmOperandClass {
238 let Name = "VecListOneDWordIndexed";
239 let ParserMethod = "parseVectorList";
240 let RenderMethod = "addVecListIndexedOperands";
242 def VecListOneDWordIndexed : Operand<i32> {
243 let ParserMatchClass = VecListOneDWordIndexAsmOperand;
244 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
247 // Register list of two D registers with byte lane subscripting.
248 def VecListTwoDByteIndexAsmOperand : AsmOperandClass {
249 let Name = "VecListTwoDByteIndexed";
250 let ParserMethod = "parseVectorList";
251 let RenderMethod = "addVecListIndexedOperands";
253 def VecListTwoDByteIndexed : Operand<i32> {
254 let ParserMatchClass = VecListTwoDByteIndexAsmOperand;
255 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
257 // ...with half-word lane subscripting.
258 def VecListTwoDHWordIndexAsmOperand : AsmOperandClass {
259 let Name = "VecListTwoDHWordIndexed";
260 let ParserMethod = "parseVectorList";
261 let RenderMethod = "addVecListIndexedOperands";
263 def VecListTwoDHWordIndexed : Operand<i32> {
264 let ParserMatchClass = VecListTwoDHWordIndexAsmOperand;
265 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
267 // ...with word lane subscripting.
268 def VecListTwoDWordIndexAsmOperand : AsmOperandClass {
269 let Name = "VecListTwoDWordIndexed";
270 let ParserMethod = "parseVectorList";
271 let RenderMethod = "addVecListIndexedOperands";
273 def VecListTwoDWordIndexed : Operand<i32> {
274 let ParserMatchClass = VecListTwoDWordIndexAsmOperand;
275 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
277 // Register list of two Q registers with half-word lane subscripting.
278 def VecListTwoQHWordIndexAsmOperand : AsmOperandClass {
279 let Name = "VecListTwoQHWordIndexed";
280 let ParserMethod = "parseVectorList";
281 let RenderMethod = "addVecListIndexedOperands";
283 def VecListTwoQHWordIndexed : Operand<i32> {
284 let ParserMatchClass = VecListTwoQHWordIndexAsmOperand;
285 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
287 // ...with word lane subscripting.
288 def VecListTwoQWordIndexAsmOperand : AsmOperandClass {
289 let Name = "VecListTwoQWordIndexed";
290 let ParserMethod = "parseVectorList";
291 let RenderMethod = "addVecListIndexedOperands";
293 def VecListTwoQWordIndexed : Operand<i32> {
294 let ParserMatchClass = VecListTwoQWordIndexAsmOperand;
295 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
299 // Register list of three D registers with byte lane subscripting.
300 def VecListThreeDByteIndexAsmOperand : AsmOperandClass {
301 let Name = "VecListThreeDByteIndexed";
302 let ParserMethod = "parseVectorList";
303 let RenderMethod = "addVecListIndexedOperands";
305 def VecListThreeDByteIndexed : Operand<i32> {
306 let ParserMatchClass = VecListThreeDByteIndexAsmOperand;
307 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
309 // ...with half-word lane subscripting.
310 def VecListThreeDHWordIndexAsmOperand : AsmOperandClass {
311 let Name = "VecListThreeDHWordIndexed";
312 let ParserMethod = "parseVectorList";
313 let RenderMethod = "addVecListIndexedOperands";
315 def VecListThreeDHWordIndexed : Operand<i32> {
316 let ParserMatchClass = VecListThreeDHWordIndexAsmOperand;
317 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
319 // ...with word lane subscripting.
320 def VecListThreeDWordIndexAsmOperand : AsmOperandClass {
321 let Name = "VecListThreeDWordIndexed";
322 let ParserMethod = "parseVectorList";
323 let RenderMethod = "addVecListIndexedOperands";
325 def VecListThreeDWordIndexed : Operand<i32> {
326 let ParserMatchClass = VecListThreeDWordIndexAsmOperand;
327 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
329 // Register list of three Q registers with half-word lane subscripting.
330 def VecListThreeQHWordIndexAsmOperand : AsmOperandClass {
331 let Name = "VecListThreeQHWordIndexed";
332 let ParserMethod = "parseVectorList";
333 let RenderMethod = "addVecListIndexedOperands";
335 def VecListThreeQHWordIndexed : Operand<i32> {
336 let ParserMatchClass = VecListThreeQHWordIndexAsmOperand;
337 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
339 // ...with word lane subscripting.
340 def VecListThreeQWordIndexAsmOperand : AsmOperandClass {
341 let Name = "VecListThreeQWordIndexed";
342 let ParserMethod = "parseVectorList";
343 let RenderMethod = "addVecListIndexedOperands";
345 def VecListThreeQWordIndexed : Operand<i32> {
346 let ParserMatchClass = VecListThreeQWordIndexAsmOperand;
347 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
350 // Register list of four D registers with byte lane subscripting.
351 def VecListFourDByteIndexAsmOperand : AsmOperandClass {
352 let Name = "VecListFourDByteIndexed";
353 let ParserMethod = "parseVectorList";
354 let RenderMethod = "addVecListIndexedOperands";
356 def VecListFourDByteIndexed : Operand<i32> {
357 let ParserMatchClass = VecListFourDByteIndexAsmOperand;
358 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
360 // ...with half-word lane subscripting.
361 def VecListFourDHWordIndexAsmOperand : AsmOperandClass {
362 let Name = "VecListFourDHWordIndexed";
363 let ParserMethod = "parseVectorList";
364 let RenderMethod = "addVecListIndexedOperands";
366 def VecListFourDHWordIndexed : Operand<i32> {
367 let ParserMatchClass = VecListFourDHWordIndexAsmOperand;
368 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
370 // ...with word lane subscripting.
371 def VecListFourDWordIndexAsmOperand : AsmOperandClass {
372 let Name = "VecListFourDWordIndexed";
373 let ParserMethod = "parseVectorList";
374 let RenderMethod = "addVecListIndexedOperands";
376 def VecListFourDWordIndexed : Operand<i32> {
377 let ParserMatchClass = VecListFourDWordIndexAsmOperand;
378 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
380 // Register list of four Q registers with half-word lane subscripting.
381 def VecListFourQHWordIndexAsmOperand : AsmOperandClass {
382 let Name = "VecListFourQHWordIndexed";
383 let ParserMethod = "parseVectorList";
384 let RenderMethod = "addVecListIndexedOperands";
386 def VecListFourQHWordIndexed : Operand<i32> {
387 let ParserMatchClass = VecListFourQHWordIndexAsmOperand;
388 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
390 // ...with word lane subscripting.
391 def VecListFourQWordIndexAsmOperand : AsmOperandClass {
392 let Name = "VecListFourQWordIndexed";
393 let ParserMethod = "parseVectorList";
394 let RenderMethod = "addVecListIndexedOperands";
396 def VecListFourQWordIndexed : Operand<i32> {
397 let ParserMatchClass = VecListFourQWordIndexAsmOperand;
398 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
401 def dword_alignedload : PatFrag<(ops node:$ptr), (load node:$ptr), [{
402 return cast<LoadSDNode>(N)->getAlignment() >= 8;
404 def dword_alignedstore : PatFrag<(ops node:$val, node:$ptr),
405 (store node:$val, node:$ptr), [{
406 return cast<StoreSDNode>(N)->getAlignment() >= 8;
408 def word_alignedload : PatFrag<(ops node:$ptr), (load node:$ptr), [{
409 return cast<LoadSDNode>(N)->getAlignment() == 4;
411 def word_alignedstore : PatFrag<(ops node:$val, node:$ptr),
412 (store node:$val, node:$ptr), [{
413 return cast<StoreSDNode>(N)->getAlignment() == 4;
415 def hword_alignedload : PatFrag<(ops node:$ptr), (load node:$ptr), [{
416 return cast<LoadSDNode>(N)->getAlignment() == 2;
418 def hword_alignedstore : PatFrag<(ops node:$val, node:$ptr),
419 (store node:$val, node:$ptr), [{
420 return cast<StoreSDNode>(N)->getAlignment() == 2;
422 def byte_alignedload : PatFrag<(ops node:$ptr), (load node:$ptr), [{
423 return cast<LoadSDNode>(N)->getAlignment() == 1;
425 def byte_alignedstore : PatFrag<(ops node:$val, node:$ptr),
426 (store node:$val, node:$ptr), [{
427 return cast<StoreSDNode>(N)->getAlignment() == 1;
429 def non_word_alignedload : PatFrag<(ops node:$ptr), (load node:$ptr), [{
430 return cast<LoadSDNode>(N)->getAlignment() < 4;
432 def non_word_alignedstore : PatFrag<(ops node:$val, node:$ptr),
433 (store node:$val, node:$ptr), [{
434 return cast<StoreSDNode>(N)->getAlignment() < 4;
437 //===----------------------------------------------------------------------===//
438 // NEON-specific DAG Nodes.
439 //===----------------------------------------------------------------------===//
441 def SDTARMVCMP : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisSameAs<1, 2>]>;
442 def SDTARMVCMPZ : SDTypeProfile<1, 1, []>;
444 def NEONvceq : SDNode<"ARMISD::VCEQ", SDTARMVCMP>;
445 def NEONvceqz : SDNode<"ARMISD::VCEQZ", SDTARMVCMPZ>;
446 def NEONvcge : SDNode<"ARMISD::VCGE", SDTARMVCMP>;
447 def NEONvcgez : SDNode<"ARMISD::VCGEZ", SDTARMVCMPZ>;
448 def NEONvclez : SDNode<"ARMISD::VCLEZ", SDTARMVCMPZ>;
449 def NEONvcgeu : SDNode<"ARMISD::VCGEU", SDTARMVCMP>;
450 def NEONvcgt : SDNode<"ARMISD::VCGT", SDTARMVCMP>;
451 def NEONvcgtz : SDNode<"ARMISD::VCGTZ", SDTARMVCMPZ>;
452 def NEONvcltz : SDNode<"ARMISD::VCLTZ", SDTARMVCMPZ>;
453 def NEONvcgtu : SDNode<"ARMISD::VCGTU", SDTARMVCMP>;
454 def NEONvtst : SDNode<"ARMISD::VTST", SDTARMVCMP>;
456 // Types for vector shift by immediates. The "SHX" version is for long and
457 // narrow operations where the source and destination vectors have different
458 // types. The "SHINS" version is for shift and insert operations.
459 def SDTARMVSH : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
461 def SDTARMVSHX : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
463 def SDTARMVSHINS : SDTypeProfile<1, 3, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
464 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
466 def NEONvshl : SDNode<"ARMISD::VSHL", SDTARMVSH>;
467 def NEONvshrs : SDNode<"ARMISD::VSHRs", SDTARMVSH>;
468 def NEONvshru : SDNode<"ARMISD::VSHRu", SDTARMVSH>;
469 def NEONvshlls : SDNode<"ARMISD::VSHLLs", SDTARMVSHX>;
470 def NEONvshllu : SDNode<"ARMISD::VSHLLu", SDTARMVSHX>;
471 def NEONvshlli : SDNode<"ARMISD::VSHLLi", SDTARMVSHX>;
472 def NEONvshrn : SDNode<"ARMISD::VSHRN", SDTARMVSHX>;
474 def NEONvrshrs : SDNode<"ARMISD::VRSHRs", SDTARMVSH>;
475 def NEONvrshru : SDNode<"ARMISD::VRSHRu", SDTARMVSH>;
476 def NEONvrshrn : SDNode<"ARMISD::VRSHRN", SDTARMVSHX>;
478 def NEONvqshls : SDNode<"ARMISD::VQSHLs", SDTARMVSH>;
479 def NEONvqshlu : SDNode<"ARMISD::VQSHLu", SDTARMVSH>;
480 def NEONvqshlsu : SDNode<"ARMISD::VQSHLsu", SDTARMVSH>;
481 def NEONvqshrns : SDNode<"ARMISD::VQSHRNs", SDTARMVSHX>;
482 def NEONvqshrnu : SDNode<"ARMISD::VQSHRNu", SDTARMVSHX>;
483 def NEONvqshrnsu : SDNode<"ARMISD::VQSHRNsu", SDTARMVSHX>;
485 def NEONvqrshrns : SDNode<"ARMISD::VQRSHRNs", SDTARMVSHX>;
486 def NEONvqrshrnu : SDNode<"ARMISD::VQRSHRNu", SDTARMVSHX>;
487 def NEONvqrshrnsu : SDNode<"ARMISD::VQRSHRNsu", SDTARMVSHX>;
489 def NEONvsli : SDNode<"ARMISD::VSLI", SDTARMVSHINS>;
490 def NEONvsri : SDNode<"ARMISD::VSRI", SDTARMVSHINS>;
492 def SDTARMVGETLN : SDTypeProfile<1, 2, [SDTCisVT<0, i32>, SDTCisInt<1>,
494 def NEONvgetlaneu : SDNode<"ARMISD::VGETLANEu", SDTARMVGETLN>;
495 def NEONvgetlanes : SDNode<"ARMISD::VGETLANEs", SDTARMVGETLN>;
497 def SDTARMVMOVIMM : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisVT<1, i32>]>;
498 def NEONvmovImm : SDNode<"ARMISD::VMOVIMM", SDTARMVMOVIMM>;
499 def NEONvmvnImm : SDNode<"ARMISD::VMVNIMM", SDTARMVMOVIMM>;
500 def NEONvmovFPImm : SDNode<"ARMISD::VMOVFPIMM", SDTARMVMOVIMM>;
502 def SDTARMVORRIMM : SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
504 def NEONvorrImm : SDNode<"ARMISD::VORRIMM", SDTARMVORRIMM>;
505 def NEONvbicImm : SDNode<"ARMISD::VBICIMM", SDTARMVORRIMM>;
507 def NEONvbsl : SDNode<"ARMISD::VBSL",
508 SDTypeProfile<1, 3, [SDTCisVec<0>,
511 SDTCisSameAs<0, 3>]>>;
513 def NEONvdup : SDNode<"ARMISD::VDUP", SDTypeProfile<1, 1, [SDTCisVec<0>]>>;
515 // VDUPLANE can produce a quad-register result from a double-register source,
516 // so the result is not constrained to match the source.
517 def NEONvduplane : SDNode<"ARMISD::VDUPLANE",
518 SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisVec<1>,
521 def SDTARMVEXT : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
522 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
523 def NEONvext : SDNode<"ARMISD::VEXT", SDTARMVEXT>;
525 def SDTARMVSHUF : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisSameAs<0, 1>]>;
526 def NEONvrev64 : SDNode<"ARMISD::VREV64", SDTARMVSHUF>;
527 def NEONvrev32 : SDNode<"ARMISD::VREV32", SDTARMVSHUF>;
528 def NEONvrev16 : SDNode<"ARMISD::VREV16", SDTARMVSHUF>;
530 def SDTARMVSHUF2 : SDTypeProfile<2, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
532 SDTCisSameAs<0, 3>]>;
533 def NEONzip : SDNode<"ARMISD::VZIP", SDTARMVSHUF2>;
534 def NEONuzp : SDNode<"ARMISD::VUZP", SDTARMVSHUF2>;
535 def NEONtrn : SDNode<"ARMISD::VTRN", SDTARMVSHUF2>;
537 def SDTARMVMULL : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
538 SDTCisSameAs<1, 2>]>;
539 def NEONvmulls : SDNode<"ARMISD::VMULLs", SDTARMVMULL>;
540 def NEONvmullu : SDNode<"ARMISD::VMULLu", SDTARMVMULL>;
542 def SDTARMFMAX : SDTypeProfile<1, 2, [SDTCisVT<0, f32>, SDTCisSameAs<0, 1>,
543 SDTCisSameAs<0, 2>]>;
544 def NEONfmax : SDNode<"ARMISD::FMAX", SDTARMFMAX>;
545 def NEONfmin : SDNode<"ARMISD::FMIN", SDTARMFMAX>;
547 def NEONimmAllZerosV: PatLeaf<(NEONvmovImm (i32 timm)), [{
548 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
549 unsigned EltBits = 0;
550 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
551 return (EltBits == 32 && EltVal == 0);
554 def NEONimmAllOnesV: PatLeaf<(NEONvmovImm (i32 timm)), [{
555 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
556 unsigned EltBits = 0;
557 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
558 return (EltBits == 8 && EltVal == 0xff);
561 //===----------------------------------------------------------------------===//
562 // NEON load / store instructions
563 //===----------------------------------------------------------------------===//
565 // Use VLDM to load a Q register as a D register pair.
566 // This is a pseudo instruction that is expanded to VLDMD after reg alloc.
568 : PseudoVFPLdStM<(outs DPair:$dst), (ins GPR:$Rn),
570 [(set DPair:$dst, (v2f64 (load GPR:$Rn)))]>;
572 // Use VSTM to store a Q register as a D register pair.
573 // This is a pseudo instruction that is expanded to VSTMD after reg alloc.
575 : PseudoVFPLdStM<(outs), (ins DPair:$src, GPR:$Rn),
577 [(store (v2f64 DPair:$src), GPR:$Rn)]>;
579 // Classes for VLD* pseudo-instructions with multi-register operands.
580 // These are expanded to real instructions after register allocation.
581 class VLDQPseudo<InstrItinClass itin>
582 : PseudoNLdSt<(outs QPR:$dst), (ins addrmode6:$addr), itin, "">;
583 class VLDQWBPseudo<InstrItinClass itin>
584 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
585 (ins addrmode6:$addr, am6offset:$offset), itin,
587 class VLDQWBfixedPseudo<InstrItinClass itin>
588 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
589 (ins addrmode6:$addr), itin,
591 class VLDQWBregisterPseudo<InstrItinClass itin>
592 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
593 (ins addrmode6:$addr, rGPR:$offset), itin,
596 class VLDQQPseudo<InstrItinClass itin>
597 : PseudoNLdSt<(outs QQPR:$dst), (ins addrmode6:$addr), itin, "">;
598 class VLDQQWBPseudo<InstrItinClass itin>
599 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
600 (ins addrmode6:$addr, am6offset:$offset), itin,
602 class VLDQQWBfixedPseudo<InstrItinClass itin>
603 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
604 (ins addrmode6:$addr), itin,
606 class VLDQQWBregisterPseudo<InstrItinClass itin>
607 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
608 (ins addrmode6:$addr, rGPR:$offset), itin,
612 class VLDQQQQPseudo<InstrItinClass itin>
613 : PseudoNLdSt<(outs QQQQPR:$dst), (ins addrmode6:$addr, QQQQPR:$src),itin,
615 class VLDQQQQWBPseudo<InstrItinClass itin>
616 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
617 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
618 "$addr.addr = $wb, $src = $dst">;
620 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
622 // VLD1 : Vector Load (multiple single elements)
623 class VLD1D<bits<4> op7_4, string Dt>
624 : NLdSt<0,0b10,0b0111,op7_4, (outs VecListOneD:$Vd),
625 (ins addrmode6:$Rn), IIC_VLD1,
626 "vld1", Dt, "$Vd, $Rn", "", []> {
629 let DecoderMethod = "DecodeVLDST1Instruction";
631 class VLD1Q<bits<4> op7_4, string Dt>
632 : NLdSt<0,0b10,0b1010,op7_4, (outs VecListDPair:$Vd),
633 (ins addrmode6:$Rn), IIC_VLD1x2,
634 "vld1", Dt, "$Vd, $Rn", "", []> {
636 let Inst{5-4} = Rn{5-4};
637 let DecoderMethod = "DecodeVLDST1Instruction";
640 def VLD1d8 : VLD1D<{0,0,0,?}, "8">;
641 def VLD1d16 : VLD1D<{0,1,0,?}, "16">;
642 def VLD1d32 : VLD1D<{1,0,0,?}, "32">;
643 def VLD1d64 : VLD1D<{1,1,0,?}, "64">;
645 def VLD1q8 : VLD1Q<{0,0,?,?}, "8">;
646 def VLD1q16 : VLD1Q<{0,1,?,?}, "16">;
647 def VLD1q32 : VLD1Q<{1,0,?,?}, "32">;
648 def VLD1q64 : VLD1Q<{1,1,?,?}, "64">;
650 // ...with address register writeback:
651 multiclass VLD1DWB<bits<4> op7_4, string Dt> {
652 def _fixed : NLdSt<0,0b10, 0b0111,op7_4, (outs VecListOneD:$Vd, GPR:$wb),
653 (ins addrmode6:$Rn), IIC_VLD1u,
654 "vld1", Dt, "$Vd, $Rn!",
655 "$Rn.addr = $wb", []> {
656 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
658 let DecoderMethod = "DecodeVLDST1Instruction";
659 let AsmMatchConverter = "cvtVLDwbFixed";
661 def _register : NLdSt<0,0b10,0b0111,op7_4, (outs VecListOneD:$Vd, GPR:$wb),
662 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1u,
663 "vld1", Dt, "$Vd, $Rn, $Rm",
664 "$Rn.addr = $wb", []> {
666 let DecoderMethod = "DecodeVLDST1Instruction";
667 let AsmMatchConverter = "cvtVLDwbRegister";
670 multiclass VLD1QWB<bits<4> op7_4, string Dt> {
671 def _fixed : NLdSt<0,0b10,0b1010,op7_4, (outs VecListDPair:$Vd, GPR:$wb),
672 (ins addrmode6:$Rn), IIC_VLD1x2u,
673 "vld1", Dt, "$Vd, $Rn!",
674 "$Rn.addr = $wb", []> {
675 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
676 let Inst{5-4} = Rn{5-4};
677 let DecoderMethod = "DecodeVLDST1Instruction";
678 let AsmMatchConverter = "cvtVLDwbFixed";
680 def _register : NLdSt<0,0b10,0b1010,op7_4, (outs VecListDPair:$Vd, GPR:$wb),
681 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1x2u,
682 "vld1", Dt, "$Vd, $Rn, $Rm",
683 "$Rn.addr = $wb", []> {
684 let Inst{5-4} = Rn{5-4};
685 let DecoderMethod = "DecodeVLDST1Instruction";
686 let AsmMatchConverter = "cvtVLDwbRegister";
690 defm VLD1d8wb : VLD1DWB<{0,0,0,?}, "8">;
691 defm VLD1d16wb : VLD1DWB<{0,1,0,?}, "16">;
692 defm VLD1d32wb : VLD1DWB<{1,0,0,?}, "32">;
693 defm VLD1d64wb : VLD1DWB<{1,1,0,?}, "64">;
694 defm VLD1q8wb : VLD1QWB<{0,0,?,?}, "8">;
695 defm VLD1q16wb : VLD1QWB<{0,1,?,?}, "16">;
696 defm VLD1q32wb : VLD1QWB<{1,0,?,?}, "32">;
697 defm VLD1q64wb : VLD1QWB<{1,1,?,?}, "64">;
699 // ...with 3 registers
700 class VLD1D3<bits<4> op7_4, string Dt>
701 : NLdSt<0,0b10,0b0110,op7_4, (outs VecListThreeD:$Vd),
702 (ins addrmode6:$Rn), IIC_VLD1x3, "vld1", Dt,
703 "$Vd, $Rn", "", []> {
706 let DecoderMethod = "DecodeVLDST1Instruction";
708 multiclass VLD1D3WB<bits<4> op7_4, string Dt> {
709 def _fixed : NLdSt<0,0b10,0b0110, op7_4, (outs VecListThreeD:$Vd, GPR:$wb),
710 (ins addrmode6:$Rn), IIC_VLD1x2u,
711 "vld1", Dt, "$Vd, $Rn!",
712 "$Rn.addr = $wb", []> {
713 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
715 let DecoderMethod = "DecodeVLDST1Instruction";
716 let AsmMatchConverter = "cvtVLDwbFixed";
718 def _register : NLdSt<0,0b10,0b0110,op7_4, (outs VecListThreeD:$Vd, GPR:$wb),
719 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1x2u,
720 "vld1", Dt, "$Vd, $Rn, $Rm",
721 "$Rn.addr = $wb", []> {
723 let DecoderMethod = "DecodeVLDST1Instruction";
724 let AsmMatchConverter = "cvtVLDwbRegister";
728 def VLD1d8T : VLD1D3<{0,0,0,?}, "8">;
729 def VLD1d16T : VLD1D3<{0,1,0,?}, "16">;
730 def VLD1d32T : VLD1D3<{1,0,0,?}, "32">;
731 def VLD1d64T : VLD1D3<{1,1,0,?}, "64">;
733 defm VLD1d8Twb : VLD1D3WB<{0,0,0,?}, "8">;
734 defm VLD1d16Twb : VLD1D3WB<{0,1,0,?}, "16">;
735 defm VLD1d32Twb : VLD1D3WB<{1,0,0,?}, "32">;
736 defm VLD1d64Twb : VLD1D3WB<{1,1,0,?}, "64">;
738 def VLD1d64TPseudo : VLDQQPseudo<IIC_VLD1x3>;
740 // ...with 4 registers
741 class VLD1D4<bits<4> op7_4, string Dt>
742 : NLdSt<0, 0b10, 0b0010, op7_4, (outs VecListFourD:$Vd),
743 (ins addrmode6:$Rn), IIC_VLD1x4, "vld1", Dt,
744 "$Vd, $Rn", "", []> {
746 let Inst{5-4} = Rn{5-4};
747 let DecoderMethod = "DecodeVLDST1Instruction";
749 multiclass VLD1D4WB<bits<4> op7_4, string Dt> {
750 def _fixed : NLdSt<0,0b10,0b0010, op7_4, (outs VecListFourD:$Vd, GPR:$wb),
751 (ins addrmode6:$Rn), IIC_VLD1x2u,
752 "vld1", Dt, "$Vd, $Rn!",
753 "$Rn.addr = $wb", []> {
754 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
755 let Inst{5-4} = Rn{5-4};
756 let DecoderMethod = "DecodeVLDST1Instruction";
757 let AsmMatchConverter = "cvtVLDwbFixed";
759 def _register : NLdSt<0,0b10,0b0010,op7_4, (outs VecListFourD:$Vd, GPR:$wb),
760 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1x2u,
761 "vld1", Dt, "$Vd, $Rn, $Rm",
762 "$Rn.addr = $wb", []> {
763 let Inst{5-4} = Rn{5-4};
764 let DecoderMethod = "DecodeVLDST1Instruction";
765 let AsmMatchConverter = "cvtVLDwbRegister";
769 def VLD1d8Q : VLD1D4<{0,0,?,?}, "8">;
770 def VLD1d16Q : VLD1D4<{0,1,?,?}, "16">;
771 def VLD1d32Q : VLD1D4<{1,0,?,?}, "32">;
772 def VLD1d64Q : VLD1D4<{1,1,?,?}, "64">;
774 defm VLD1d8Qwb : VLD1D4WB<{0,0,?,?}, "8">;
775 defm VLD1d16Qwb : VLD1D4WB<{0,1,?,?}, "16">;
776 defm VLD1d32Qwb : VLD1D4WB<{1,0,?,?}, "32">;
777 defm VLD1d64Qwb : VLD1D4WB<{1,1,?,?}, "64">;
779 def VLD1d64QPseudo : VLDQQPseudo<IIC_VLD1x4>;
781 // VLD2 : Vector Load (multiple 2-element structures)
782 class VLD2<bits<4> op11_8, bits<4> op7_4, string Dt, RegisterOperand VdTy,
784 : NLdSt<0, 0b10, op11_8, op7_4, (outs VdTy:$Vd),
785 (ins addrmode6:$Rn), itin,
786 "vld2", Dt, "$Vd, $Rn", "", []> {
788 let Inst{5-4} = Rn{5-4};
789 let DecoderMethod = "DecodeVLDST2Instruction";
792 def VLD2d8 : VLD2<0b1000, {0,0,?,?}, "8", VecListDPair, IIC_VLD2>;
793 def VLD2d16 : VLD2<0b1000, {0,1,?,?}, "16", VecListDPair, IIC_VLD2>;
794 def VLD2d32 : VLD2<0b1000, {1,0,?,?}, "32", VecListDPair, IIC_VLD2>;
796 def VLD2q8 : VLD2<0b0011, {0,0,?,?}, "8", VecListFourD, IIC_VLD2x2>;
797 def VLD2q16 : VLD2<0b0011, {0,1,?,?}, "16", VecListFourD, IIC_VLD2x2>;
798 def VLD2q32 : VLD2<0b0011, {1,0,?,?}, "32", VecListFourD, IIC_VLD2x2>;
800 def VLD2q8Pseudo : VLDQQPseudo<IIC_VLD2x2>;
801 def VLD2q16Pseudo : VLDQQPseudo<IIC_VLD2x2>;
802 def VLD2q32Pseudo : VLDQQPseudo<IIC_VLD2x2>;
804 // ...with address register writeback:
805 multiclass VLD2WB<bits<4> op11_8, bits<4> op7_4, string Dt,
806 RegisterOperand VdTy, InstrItinClass itin> {
807 def _fixed : NLdSt<0, 0b10, op11_8, op7_4, (outs VdTy:$Vd, GPR:$wb),
808 (ins addrmode6:$Rn), itin,
809 "vld2", Dt, "$Vd, $Rn!",
810 "$Rn.addr = $wb", []> {
811 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
812 let Inst{5-4} = Rn{5-4};
813 let DecoderMethod = "DecodeVLDST2Instruction";
814 let AsmMatchConverter = "cvtVLDwbFixed";
816 def _register : NLdSt<0, 0b10, op11_8, op7_4, (outs VdTy:$Vd, GPR:$wb),
817 (ins addrmode6:$Rn, rGPR:$Rm), itin,
818 "vld2", Dt, "$Vd, $Rn, $Rm",
819 "$Rn.addr = $wb", []> {
820 let Inst{5-4} = Rn{5-4};
821 let DecoderMethod = "DecodeVLDST2Instruction";
822 let AsmMatchConverter = "cvtVLDwbRegister";
826 defm VLD2d8wb : VLD2WB<0b1000, {0,0,?,?}, "8", VecListDPair, IIC_VLD2u>;
827 defm VLD2d16wb : VLD2WB<0b1000, {0,1,?,?}, "16", VecListDPair, IIC_VLD2u>;
828 defm VLD2d32wb : VLD2WB<0b1000, {1,0,?,?}, "32", VecListDPair, IIC_VLD2u>;
830 defm VLD2q8wb : VLD2WB<0b0011, {0,0,?,?}, "8", VecListFourD, IIC_VLD2x2u>;
831 defm VLD2q16wb : VLD2WB<0b0011, {0,1,?,?}, "16", VecListFourD, IIC_VLD2x2u>;
832 defm VLD2q32wb : VLD2WB<0b0011, {1,0,?,?}, "32", VecListFourD, IIC_VLD2x2u>;
834 def VLD2q8PseudoWB_fixed : VLDQQWBfixedPseudo<IIC_VLD2x2u>;
835 def VLD2q16PseudoWB_fixed : VLDQQWBfixedPseudo<IIC_VLD2x2u>;
836 def VLD2q32PseudoWB_fixed : VLDQQWBfixedPseudo<IIC_VLD2x2u>;
837 def VLD2q8PseudoWB_register : VLDQQWBregisterPseudo<IIC_VLD2x2u>;
838 def VLD2q16PseudoWB_register : VLDQQWBregisterPseudo<IIC_VLD2x2u>;
839 def VLD2q32PseudoWB_register : VLDQQWBregisterPseudo<IIC_VLD2x2u>;
841 // ...with double-spaced registers
842 def VLD2b8 : VLD2<0b1001, {0,0,?,?}, "8", VecListDPairSpaced, IIC_VLD2>;
843 def VLD2b16 : VLD2<0b1001, {0,1,?,?}, "16", VecListDPairSpaced, IIC_VLD2>;
844 def VLD2b32 : VLD2<0b1001, {1,0,?,?}, "32", VecListDPairSpaced, IIC_VLD2>;
845 defm VLD2b8wb : VLD2WB<0b1001, {0,0,?,?}, "8", VecListDPairSpaced, IIC_VLD2u>;
846 defm VLD2b16wb : VLD2WB<0b1001, {0,1,?,?}, "16", VecListDPairSpaced, IIC_VLD2u>;
847 defm VLD2b32wb : VLD2WB<0b1001, {1,0,?,?}, "32", VecListDPairSpaced, IIC_VLD2u>;
849 // VLD3 : Vector Load (multiple 3-element structures)
850 class VLD3D<bits<4> op11_8, bits<4> op7_4, string Dt>
851 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
852 (ins addrmode6:$Rn), IIC_VLD3,
853 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn", "", []> {
856 let DecoderMethod = "DecodeVLDST3Instruction";
859 def VLD3d8 : VLD3D<0b0100, {0,0,0,?}, "8">;
860 def VLD3d16 : VLD3D<0b0100, {0,1,0,?}, "16">;
861 def VLD3d32 : VLD3D<0b0100, {1,0,0,?}, "32">;
863 def VLD3d8Pseudo : VLDQQPseudo<IIC_VLD3>;
864 def VLD3d16Pseudo : VLDQQPseudo<IIC_VLD3>;
865 def VLD3d32Pseudo : VLDQQPseudo<IIC_VLD3>;
867 // ...with address register writeback:
868 class VLD3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
869 : NLdSt<0, 0b10, op11_8, op7_4,
870 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
871 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD3u,
872 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn$Rm",
873 "$Rn.addr = $wb", []> {
875 let DecoderMethod = "DecodeVLDST3Instruction";
878 def VLD3d8_UPD : VLD3DWB<0b0100, {0,0,0,?}, "8">;
879 def VLD3d16_UPD : VLD3DWB<0b0100, {0,1,0,?}, "16">;
880 def VLD3d32_UPD : VLD3DWB<0b0100, {1,0,0,?}, "32">;
882 def VLD3d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
883 def VLD3d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
884 def VLD3d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
886 // ...with double-spaced registers:
887 def VLD3q8 : VLD3D<0b0101, {0,0,0,?}, "8">;
888 def VLD3q16 : VLD3D<0b0101, {0,1,0,?}, "16">;
889 def VLD3q32 : VLD3D<0b0101, {1,0,0,?}, "32">;
890 def VLD3q8_UPD : VLD3DWB<0b0101, {0,0,0,?}, "8">;
891 def VLD3q16_UPD : VLD3DWB<0b0101, {0,1,0,?}, "16">;
892 def VLD3q32_UPD : VLD3DWB<0b0101, {1,0,0,?}, "32">;
894 def VLD3q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
895 def VLD3q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
896 def VLD3q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
898 // ...alternate versions to be allocated odd register numbers:
899 def VLD3q8oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
900 def VLD3q16oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
901 def VLD3q32oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
903 def VLD3q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
904 def VLD3q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
905 def VLD3q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
907 // VLD4 : Vector Load (multiple 4-element structures)
908 class VLD4D<bits<4> op11_8, bits<4> op7_4, string Dt>
909 : NLdSt<0, 0b10, op11_8, op7_4,
910 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
911 (ins addrmode6:$Rn), IIC_VLD4,
912 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn", "", []> {
914 let Inst{5-4} = Rn{5-4};
915 let DecoderMethod = "DecodeVLDST4Instruction";
918 def VLD4d8 : VLD4D<0b0000, {0,0,?,?}, "8">;
919 def VLD4d16 : VLD4D<0b0000, {0,1,?,?}, "16">;
920 def VLD4d32 : VLD4D<0b0000, {1,0,?,?}, "32">;
922 def VLD4d8Pseudo : VLDQQPseudo<IIC_VLD4>;
923 def VLD4d16Pseudo : VLDQQPseudo<IIC_VLD4>;
924 def VLD4d32Pseudo : VLDQQPseudo<IIC_VLD4>;
926 // ...with address register writeback:
927 class VLD4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
928 : NLdSt<0, 0b10, op11_8, op7_4,
929 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
930 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD4u,
931 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn$Rm",
932 "$Rn.addr = $wb", []> {
933 let Inst{5-4} = Rn{5-4};
934 let DecoderMethod = "DecodeVLDST4Instruction";
937 def VLD4d8_UPD : VLD4DWB<0b0000, {0,0,?,?}, "8">;
938 def VLD4d16_UPD : VLD4DWB<0b0000, {0,1,?,?}, "16">;
939 def VLD4d32_UPD : VLD4DWB<0b0000, {1,0,?,?}, "32">;
941 def VLD4d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
942 def VLD4d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
943 def VLD4d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
945 // ...with double-spaced registers:
946 def VLD4q8 : VLD4D<0b0001, {0,0,?,?}, "8">;
947 def VLD4q16 : VLD4D<0b0001, {0,1,?,?}, "16">;
948 def VLD4q32 : VLD4D<0b0001, {1,0,?,?}, "32">;
949 def VLD4q8_UPD : VLD4DWB<0b0001, {0,0,?,?}, "8">;
950 def VLD4q16_UPD : VLD4DWB<0b0001, {0,1,?,?}, "16">;
951 def VLD4q32_UPD : VLD4DWB<0b0001, {1,0,?,?}, "32">;
953 def VLD4q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
954 def VLD4q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
955 def VLD4q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
957 // ...alternate versions to be allocated odd register numbers:
958 def VLD4q8oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
959 def VLD4q16oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
960 def VLD4q32oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
962 def VLD4q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
963 def VLD4q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
964 def VLD4q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
966 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
968 // Classes for VLD*LN pseudo-instructions with multi-register operands.
969 // These are expanded to real instructions after register allocation.
970 class VLDQLNPseudo<InstrItinClass itin>
971 : PseudoNLdSt<(outs QPR:$dst),
972 (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
973 itin, "$src = $dst">;
974 class VLDQLNWBPseudo<InstrItinClass itin>
975 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
976 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
977 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
978 class VLDQQLNPseudo<InstrItinClass itin>
979 : PseudoNLdSt<(outs QQPR:$dst),
980 (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
981 itin, "$src = $dst">;
982 class VLDQQLNWBPseudo<InstrItinClass itin>
983 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
984 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
985 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
986 class VLDQQQQLNPseudo<InstrItinClass itin>
987 : PseudoNLdSt<(outs QQQQPR:$dst),
988 (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
989 itin, "$src = $dst">;
990 class VLDQQQQLNWBPseudo<InstrItinClass itin>
991 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
992 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
993 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
995 // VLD1LN : Vector Load (single element to one lane)
996 class VLD1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
998 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd),
999 (ins addrmode6:$Rn, DPR:$src, nohash_imm:$lane),
1000 IIC_VLD1ln, "vld1", Dt, "\\{$Vd[$lane]\\}, $Rn",
1002 [(set DPR:$Vd, (vector_insert (Ty DPR:$src),
1003 (i32 (LoadOp addrmode6:$Rn)),
1006 let DecoderMethod = "DecodeVLD1LN";
1008 class VLD1LN32<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1010 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd),
1011 (ins addrmode6oneL32:$Rn, DPR:$src, nohash_imm:$lane),
1012 IIC_VLD1ln, "vld1", Dt, "\\{$Vd[$lane]\\}, $Rn",
1014 [(set DPR:$Vd, (vector_insert (Ty DPR:$src),
1015 (i32 (LoadOp addrmode6oneL32:$Rn)),
1018 let DecoderMethod = "DecodeVLD1LN";
1020 class VLD1QLNPseudo<ValueType Ty, PatFrag LoadOp> : VLDQLNPseudo<IIC_VLD1ln> {
1021 let Pattern = [(set QPR:$dst, (vector_insert (Ty QPR:$src),
1022 (i32 (LoadOp addrmode6:$addr)),
1026 def VLD1LNd8 : VLD1LN<0b0000, {?,?,?,0}, "8", v8i8, extloadi8> {
1027 let Inst{7-5} = lane{2-0};
1029 def VLD1LNd16 : VLD1LN<0b0100, {?,?,0,?}, "16", v4i16, extloadi16> {
1030 let Inst{7-6} = lane{1-0};
1031 let Inst{5-4} = Rn{5-4};
1033 def VLD1LNd32 : VLD1LN32<0b1000, {?,0,?,?}, "32", v2i32, load> {
1034 let Inst{7} = lane{0};
1035 let Inst{5-4} = Rn{5-4};
1038 def VLD1LNq8Pseudo : VLD1QLNPseudo<v16i8, extloadi8>;
1039 def VLD1LNq16Pseudo : VLD1QLNPseudo<v8i16, extloadi16>;
1040 def VLD1LNq32Pseudo : VLD1QLNPseudo<v4i32, load>;
1042 def : Pat<(vector_insert (v2f32 DPR:$src),
1043 (f32 (load addrmode6:$addr)), imm:$lane),
1044 (VLD1LNd32 addrmode6:$addr, DPR:$src, imm:$lane)>;
1045 def : Pat<(vector_insert (v4f32 QPR:$src),
1046 (f32 (load addrmode6:$addr)), imm:$lane),
1047 (VLD1LNq32Pseudo addrmode6:$addr, QPR:$src, imm:$lane)>;
1049 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
1051 // ...with address register writeback:
1052 class VLD1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1053 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, GPR:$wb),
1054 (ins addrmode6:$Rn, am6offset:$Rm,
1055 DPR:$src, nohash_imm:$lane), IIC_VLD1lnu, "vld1", Dt,
1056 "\\{$Vd[$lane]\\}, $Rn$Rm",
1057 "$src = $Vd, $Rn.addr = $wb", []> {
1058 let DecoderMethod = "DecodeVLD1LN";
1061 def VLD1LNd8_UPD : VLD1LNWB<0b0000, {?,?,?,0}, "8"> {
1062 let Inst{7-5} = lane{2-0};
1064 def VLD1LNd16_UPD : VLD1LNWB<0b0100, {?,?,0,?}, "16"> {
1065 let Inst{7-6} = lane{1-0};
1066 let Inst{4} = Rn{4};
1068 def VLD1LNd32_UPD : VLD1LNWB<0b1000, {?,0,?,?}, "32"> {
1069 let Inst{7} = lane{0};
1070 let Inst{5} = Rn{4};
1071 let Inst{4} = Rn{4};
1074 def VLD1LNq8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
1075 def VLD1LNq16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
1076 def VLD1LNq32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
1078 // VLD2LN : Vector Load (single 2-element structure to one lane)
1079 class VLD2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1080 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2),
1081 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, nohash_imm:$lane),
1082 IIC_VLD2ln, "vld2", Dt, "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn",
1083 "$src1 = $Vd, $src2 = $dst2", []> {
1085 let Inst{4} = Rn{4};
1086 let DecoderMethod = "DecodeVLD2LN";
1089 def VLD2LNd8 : VLD2LN<0b0001, {?,?,?,?}, "8"> {
1090 let Inst{7-5} = lane{2-0};
1092 def VLD2LNd16 : VLD2LN<0b0101, {?,?,0,?}, "16"> {
1093 let Inst{7-6} = lane{1-0};
1095 def VLD2LNd32 : VLD2LN<0b1001, {?,0,0,?}, "32"> {
1096 let Inst{7} = lane{0};
1099 def VLD2LNd8Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
1100 def VLD2LNd16Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
1101 def VLD2LNd32Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
1103 // ...with double-spaced registers:
1104 def VLD2LNq16 : VLD2LN<0b0101, {?,?,1,?}, "16"> {
1105 let Inst{7-6} = lane{1-0};
1107 def VLD2LNq32 : VLD2LN<0b1001, {?,1,0,?}, "32"> {
1108 let Inst{7} = lane{0};
1111 def VLD2LNq16Pseudo : VLDQQLNPseudo<IIC_VLD2ln>;
1112 def VLD2LNq32Pseudo : VLDQQLNPseudo<IIC_VLD2ln>;
1114 // ...with address register writeback:
1115 class VLD2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1116 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
1117 (ins addrmode6:$Rn, am6offset:$Rm,
1118 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VLD2lnu, "vld2", Dt,
1119 "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn$Rm",
1120 "$src1 = $Vd, $src2 = $dst2, $Rn.addr = $wb", []> {
1121 let Inst{4} = Rn{4};
1122 let DecoderMethod = "DecodeVLD2LN";
1125 def VLD2LNd8_UPD : VLD2LNWB<0b0001, {?,?,?,?}, "8"> {
1126 let Inst{7-5} = lane{2-0};
1128 def VLD2LNd16_UPD : VLD2LNWB<0b0101, {?,?,0,?}, "16"> {
1129 let Inst{7-6} = lane{1-0};
1131 def VLD2LNd32_UPD : VLD2LNWB<0b1001, {?,0,0,?}, "32"> {
1132 let Inst{7} = lane{0};
1135 def VLD2LNd8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
1136 def VLD2LNd16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
1137 def VLD2LNd32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
1139 def VLD2LNq16_UPD : VLD2LNWB<0b0101, {?,?,1,?}, "16"> {
1140 let Inst{7-6} = lane{1-0};
1142 def VLD2LNq32_UPD : VLD2LNWB<0b1001, {?,1,0,?}, "32"> {
1143 let Inst{7} = lane{0};
1146 def VLD2LNq16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
1147 def VLD2LNq32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
1149 // VLD3LN : Vector Load (single 3-element structure to one lane)
1150 class VLD3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1151 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
1152 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3,
1153 nohash_imm:$lane), IIC_VLD3ln, "vld3", Dt,
1154 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn",
1155 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3", []> {
1157 let DecoderMethod = "DecodeVLD3LN";
1160 def VLD3LNd8 : VLD3LN<0b0010, {?,?,?,0}, "8"> {
1161 let Inst{7-5} = lane{2-0};
1163 def VLD3LNd16 : VLD3LN<0b0110, {?,?,0,0}, "16"> {
1164 let Inst{7-6} = lane{1-0};
1166 def VLD3LNd32 : VLD3LN<0b1010, {?,0,0,0}, "32"> {
1167 let Inst{7} = lane{0};
1170 def VLD3LNd8Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
1171 def VLD3LNd16Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
1172 def VLD3LNd32Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
1174 // ...with double-spaced registers:
1175 def VLD3LNq16 : VLD3LN<0b0110, {?,?,1,0}, "16"> {
1176 let Inst{7-6} = lane{1-0};
1178 def VLD3LNq32 : VLD3LN<0b1010, {?,1,0,0}, "32"> {
1179 let Inst{7} = lane{0};
1182 def VLD3LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
1183 def VLD3LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
1185 // ...with address register writeback:
1186 class VLD3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1187 : NLdStLn<1, 0b10, op11_8, op7_4,
1188 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
1189 (ins addrmode6:$Rn, am6offset:$Rm,
1190 DPR:$src1, DPR:$src2, DPR:$src3, nohash_imm:$lane),
1191 IIC_VLD3lnu, "vld3", Dt,
1192 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn$Rm",
1193 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $Rn.addr = $wb",
1195 let DecoderMethod = "DecodeVLD3LN";
1198 def VLD3LNd8_UPD : VLD3LNWB<0b0010, {?,?,?,0}, "8"> {
1199 let Inst{7-5} = lane{2-0};
1201 def VLD3LNd16_UPD : VLD3LNWB<0b0110, {?,?,0,0}, "16"> {
1202 let Inst{7-6} = lane{1-0};
1204 def VLD3LNd32_UPD : VLD3LNWB<0b1010, {?,0,0,0}, "32"> {
1205 let Inst{7} = lane{0};
1208 def VLD3LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
1209 def VLD3LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
1210 def VLD3LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
1212 def VLD3LNq16_UPD : VLD3LNWB<0b0110, {?,?,1,0}, "16"> {
1213 let Inst{7-6} = lane{1-0};
1215 def VLD3LNq32_UPD : VLD3LNWB<0b1010, {?,1,0,0}, "32"> {
1216 let Inst{7} = lane{0};
1219 def VLD3LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
1220 def VLD3LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
1222 // VLD4LN : Vector Load (single 4-element structure to one lane)
1223 class VLD4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1224 : NLdStLn<1, 0b10, op11_8, op7_4,
1225 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
1226 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
1227 nohash_imm:$lane), IIC_VLD4ln, "vld4", Dt,
1228 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn",
1229 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4", []> {
1231 let Inst{4} = Rn{4};
1232 let DecoderMethod = "DecodeVLD4LN";
1235 def VLD4LNd8 : VLD4LN<0b0011, {?,?,?,?}, "8"> {
1236 let Inst{7-5} = lane{2-0};
1238 def VLD4LNd16 : VLD4LN<0b0111, {?,?,0,?}, "16"> {
1239 let Inst{7-6} = lane{1-0};
1241 def VLD4LNd32 : VLD4LN<0b1011, {?,0,?,?}, "32"> {
1242 let Inst{7} = lane{0};
1243 let Inst{5} = Rn{5};
1246 def VLD4LNd8Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
1247 def VLD4LNd16Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
1248 def VLD4LNd32Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
1250 // ...with double-spaced registers:
1251 def VLD4LNq16 : VLD4LN<0b0111, {?,?,1,?}, "16"> {
1252 let Inst{7-6} = lane{1-0};
1254 def VLD4LNq32 : VLD4LN<0b1011, {?,1,?,?}, "32"> {
1255 let Inst{7} = lane{0};
1256 let Inst{5} = Rn{5};
1259 def VLD4LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>;
1260 def VLD4LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>;
1262 // ...with address register writeback:
1263 class VLD4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1264 : NLdStLn<1, 0b10, op11_8, op7_4,
1265 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
1266 (ins addrmode6:$Rn, am6offset:$Rm,
1267 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
1268 IIC_VLD4lnu, "vld4", Dt,
1269 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn$Rm",
1270 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4, $Rn.addr = $wb",
1272 let Inst{4} = Rn{4};
1273 let DecoderMethod = "DecodeVLD4LN" ;
1276 def VLD4LNd8_UPD : VLD4LNWB<0b0011, {?,?,?,?}, "8"> {
1277 let Inst{7-5} = lane{2-0};
1279 def VLD4LNd16_UPD : VLD4LNWB<0b0111, {?,?,0,?}, "16"> {
1280 let Inst{7-6} = lane{1-0};
1282 def VLD4LNd32_UPD : VLD4LNWB<0b1011, {?,0,?,?}, "32"> {
1283 let Inst{7} = lane{0};
1284 let Inst{5} = Rn{5};
1287 def VLD4LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
1288 def VLD4LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
1289 def VLD4LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
1291 def VLD4LNq16_UPD : VLD4LNWB<0b0111, {?,?,1,?}, "16"> {
1292 let Inst{7-6} = lane{1-0};
1294 def VLD4LNq32_UPD : VLD4LNWB<0b1011, {?,1,?,?}, "32"> {
1295 let Inst{7} = lane{0};
1296 let Inst{5} = Rn{5};
1299 def VLD4LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
1300 def VLD4LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
1302 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
1304 // VLD1DUP : Vector Load (single element to all lanes)
1305 class VLD1DUP<bits<4> op7_4, string Dt, ValueType Ty, PatFrag LoadOp>
1306 : NLdSt<1, 0b10, 0b1100, op7_4, (outs VecListOneDAllLanes:$Vd),
1307 (ins addrmode6dup:$Rn),
1308 IIC_VLD1dup, "vld1", Dt, "$Vd, $Rn", "",
1309 [(set VecListOneDAllLanes:$Vd,
1310 (Ty (NEONvdup (i32 (LoadOp addrmode6dup:$Rn)))))]> {
1312 let Inst{4} = Rn{4};
1313 let DecoderMethod = "DecodeVLD1DupInstruction";
1315 def VLD1DUPd8 : VLD1DUP<{0,0,0,?}, "8", v8i8, extloadi8>;
1316 def VLD1DUPd16 : VLD1DUP<{0,1,0,?}, "16", v4i16, extloadi16>;
1317 def VLD1DUPd32 : VLD1DUP<{1,0,0,?}, "32", v2i32, load>;
1319 def : Pat<(v2f32 (NEONvdup (f32 (load addrmode6dup:$addr)))),
1320 (VLD1DUPd32 addrmode6:$addr)>;
1322 class VLD1QDUP<bits<4> op7_4, string Dt, ValueType Ty, PatFrag LoadOp>
1323 : NLdSt<1, 0b10, 0b1100, op7_4, (outs VecListDPairAllLanes:$Vd),
1324 (ins addrmode6dup:$Rn), IIC_VLD1dup,
1325 "vld1", Dt, "$Vd, $Rn", "",
1326 [(set VecListDPairAllLanes:$Vd,
1327 (Ty (NEONvdup (i32 (LoadOp addrmode6dup:$Rn)))))]> {
1329 let Inst{4} = Rn{4};
1330 let DecoderMethod = "DecodeVLD1DupInstruction";
1333 def VLD1DUPq8 : VLD1QDUP<{0,0,1,0}, "8", v16i8, extloadi8>;
1334 def VLD1DUPq16 : VLD1QDUP<{0,1,1,?}, "16", v8i16, extloadi16>;
1335 def VLD1DUPq32 : VLD1QDUP<{1,0,1,?}, "32", v4i32, load>;
1337 def : Pat<(v4f32 (NEONvdup (f32 (load addrmode6dup:$addr)))),
1338 (VLD1DUPq32 addrmode6:$addr)>;
1340 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
1341 // ...with address register writeback:
1342 multiclass VLD1DUPWB<bits<4> op7_4, string Dt> {
1343 def _fixed : NLdSt<1, 0b10, 0b1100, op7_4,
1344 (outs VecListOneDAllLanes:$Vd, GPR:$wb),
1345 (ins addrmode6dup:$Rn), IIC_VLD1dupu,
1346 "vld1", Dt, "$Vd, $Rn!",
1347 "$Rn.addr = $wb", []> {
1348 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1349 let Inst{4} = Rn{4};
1350 let DecoderMethod = "DecodeVLD1DupInstruction";
1351 let AsmMatchConverter = "cvtVLDwbFixed";
1353 def _register : NLdSt<1, 0b10, 0b1100, op7_4,
1354 (outs VecListOneDAllLanes:$Vd, GPR:$wb),
1355 (ins addrmode6dup:$Rn, rGPR:$Rm), IIC_VLD1dupu,
1356 "vld1", Dt, "$Vd, $Rn, $Rm",
1357 "$Rn.addr = $wb", []> {
1358 let Inst{4} = Rn{4};
1359 let DecoderMethod = "DecodeVLD1DupInstruction";
1360 let AsmMatchConverter = "cvtVLDwbRegister";
1363 multiclass VLD1QDUPWB<bits<4> op7_4, string Dt> {
1364 def _fixed : NLdSt<1, 0b10, 0b1100, op7_4,
1365 (outs VecListDPairAllLanes:$Vd, GPR:$wb),
1366 (ins addrmode6dup:$Rn), IIC_VLD1dupu,
1367 "vld1", Dt, "$Vd, $Rn!",
1368 "$Rn.addr = $wb", []> {
1369 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1370 let Inst{4} = Rn{4};
1371 let DecoderMethod = "DecodeVLD1DupInstruction";
1372 let AsmMatchConverter = "cvtVLDwbFixed";
1374 def _register : NLdSt<1, 0b10, 0b1100, op7_4,
1375 (outs VecListDPairAllLanes:$Vd, GPR:$wb),
1376 (ins addrmode6dup:$Rn, rGPR:$Rm), IIC_VLD1dupu,
1377 "vld1", Dt, "$Vd, $Rn, $Rm",
1378 "$Rn.addr = $wb", []> {
1379 let Inst{4} = Rn{4};
1380 let DecoderMethod = "DecodeVLD1DupInstruction";
1381 let AsmMatchConverter = "cvtVLDwbRegister";
1385 defm VLD1DUPd8wb : VLD1DUPWB<{0,0,0,0}, "8">;
1386 defm VLD1DUPd16wb : VLD1DUPWB<{0,1,0,?}, "16">;
1387 defm VLD1DUPd32wb : VLD1DUPWB<{1,0,0,?}, "32">;
1389 defm VLD1DUPq8wb : VLD1QDUPWB<{0,0,1,0}, "8">;
1390 defm VLD1DUPq16wb : VLD1QDUPWB<{0,1,1,?}, "16">;
1391 defm VLD1DUPq32wb : VLD1QDUPWB<{1,0,1,?}, "32">;
1393 // VLD2DUP : Vector Load (single 2-element structure to all lanes)
1394 class VLD2DUP<bits<4> op7_4, string Dt, RegisterOperand VdTy>
1395 : NLdSt<1, 0b10, 0b1101, op7_4, (outs VdTy:$Vd),
1396 (ins addrmode6dup:$Rn), IIC_VLD2dup,
1397 "vld2", Dt, "$Vd, $Rn", "", []> {
1399 let Inst{4} = Rn{4};
1400 let DecoderMethod = "DecodeVLD2DupInstruction";
1403 def VLD2DUPd8 : VLD2DUP<{0,0,0,?}, "8", VecListDPairAllLanes>;
1404 def VLD2DUPd16 : VLD2DUP<{0,1,0,?}, "16", VecListDPairAllLanes>;
1405 def VLD2DUPd32 : VLD2DUP<{1,0,0,?}, "32", VecListDPairAllLanes>;
1407 // ...with double-spaced registers
1408 def VLD2DUPd8x2 : VLD2DUP<{0,0,1,?}, "8", VecListDPairSpacedAllLanes>;
1409 def VLD2DUPd16x2 : VLD2DUP<{0,1,1,?}, "16", VecListDPairSpacedAllLanes>;
1410 def VLD2DUPd32x2 : VLD2DUP<{1,0,1,?}, "32", VecListDPairSpacedAllLanes>;
1412 // ...with address register writeback:
1413 multiclass VLD2DUPWB<bits<4> op7_4, string Dt, RegisterOperand VdTy> {
1414 def _fixed : NLdSt<1, 0b10, 0b1101, op7_4,
1415 (outs VdTy:$Vd, GPR:$wb),
1416 (ins addrmode6dup:$Rn), IIC_VLD2dupu,
1417 "vld2", Dt, "$Vd, $Rn!",
1418 "$Rn.addr = $wb", []> {
1419 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1420 let Inst{4} = Rn{4};
1421 let DecoderMethod = "DecodeVLD2DupInstruction";
1422 let AsmMatchConverter = "cvtVLDwbFixed";
1424 def _register : NLdSt<1, 0b10, 0b1101, op7_4,
1425 (outs VdTy:$Vd, GPR:$wb),
1426 (ins addrmode6dup:$Rn, rGPR:$Rm), IIC_VLD2dupu,
1427 "vld2", Dt, "$Vd, $Rn, $Rm",
1428 "$Rn.addr = $wb", []> {
1429 let Inst{4} = Rn{4};
1430 let DecoderMethod = "DecodeVLD2DupInstruction";
1431 let AsmMatchConverter = "cvtVLDwbRegister";
1435 defm VLD2DUPd8wb : VLD2DUPWB<{0,0,0,0}, "8", VecListDPairAllLanes>;
1436 defm VLD2DUPd16wb : VLD2DUPWB<{0,1,0,?}, "16", VecListDPairAllLanes>;
1437 defm VLD2DUPd32wb : VLD2DUPWB<{1,0,0,?}, "32", VecListDPairAllLanes>;
1439 defm VLD2DUPd8x2wb : VLD2DUPWB<{0,0,1,0}, "8", VecListDPairSpacedAllLanes>;
1440 defm VLD2DUPd16x2wb : VLD2DUPWB<{0,1,1,?}, "16", VecListDPairSpacedAllLanes>;
1441 defm VLD2DUPd32x2wb : VLD2DUPWB<{1,0,1,?}, "32", VecListDPairSpacedAllLanes>;
1443 // VLD3DUP : Vector Load (single 3-element structure to all lanes)
1444 class VLD3DUP<bits<4> op7_4, string Dt>
1445 : NLdSt<1, 0b10, 0b1110, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
1446 (ins addrmode6dup:$Rn), IIC_VLD3dup,
1447 "vld3", Dt, "\\{$Vd[], $dst2[], $dst3[]\\}, $Rn", "", []> {
1450 let DecoderMethod = "DecodeVLD3DupInstruction";
1453 def VLD3DUPd8 : VLD3DUP<{0,0,0,?}, "8">;
1454 def VLD3DUPd16 : VLD3DUP<{0,1,0,?}, "16">;
1455 def VLD3DUPd32 : VLD3DUP<{1,0,0,?}, "32">;
1457 def VLD3DUPd8Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1458 def VLD3DUPd16Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1459 def VLD3DUPd32Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1461 // ...with double-spaced registers (not used for codegen):
1462 def VLD3DUPq8 : VLD3DUP<{0,0,1,?}, "8">;
1463 def VLD3DUPq16 : VLD3DUP<{0,1,1,?}, "16">;
1464 def VLD3DUPq32 : VLD3DUP<{1,0,1,?}, "32">;
1466 // ...with address register writeback:
1467 class VLD3DUPWB<bits<4> op7_4, string Dt>
1468 : NLdSt<1, 0b10, 0b1110, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
1469 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD3dupu,
1470 "vld3", Dt, "\\{$Vd[], $dst2[], $dst3[]\\}, $Rn$Rm",
1471 "$Rn.addr = $wb", []> {
1473 let DecoderMethod = "DecodeVLD3DupInstruction";
1476 def VLD3DUPd8_UPD : VLD3DUPWB<{0,0,0,0}, "8">;
1477 def VLD3DUPd16_UPD : VLD3DUPWB<{0,1,0,?}, "16">;
1478 def VLD3DUPd32_UPD : VLD3DUPWB<{1,0,0,?}, "32">;
1480 def VLD3DUPq8_UPD : VLD3DUPWB<{0,0,1,0}, "8">;
1481 def VLD3DUPq16_UPD : VLD3DUPWB<{0,1,1,?}, "16">;
1482 def VLD3DUPq32_UPD : VLD3DUPWB<{1,0,1,?}, "32">;
1484 def VLD3DUPd8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1485 def VLD3DUPd16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1486 def VLD3DUPd32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1488 // VLD4DUP : Vector Load (single 4-element structure to all lanes)
1489 class VLD4DUP<bits<4> op7_4, string Dt>
1490 : NLdSt<1, 0b10, 0b1111, op7_4,
1491 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
1492 (ins addrmode6dup:$Rn), IIC_VLD4dup,
1493 "vld4", Dt, "\\{$Vd[], $dst2[], $dst3[], $dst4[]\\}, $Rn", "", []> {
1495 let Inst{4} = Rn{4};
1496 let DecoderMethod = "DecodeVLD4DupInstruction";
1499 def VLD4DUPd8 : VLD4DUP<{0,0,0,?}, "8">;
1500 def VLD4DUPd16 : VLD4DUP<{0,1,0,?}, "16">;
1501 def VLD4DUPd32 : VLD4DUP<{1,?,0,?}, "32"> { let Inst{6} = Rn{5}; }
1503 def VLD4DUPd8Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1504 def VLD4DUPd16Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1505 def VLD4DUPd32Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1507 // ...with double-spaced registers (not used for codegen):
1508 def VLD4DUPq8 : VLD4DUP<{0,0,1,?}, "8">;
1509 def VLD4DUPq16 : VLD4DUP<{0,1,1,?}, "16">;
1510 def VLD4DUPq32 : VLD4DUP<{1,?,1,?}, "32"> { let Inst{6} = Rn{5}; }
1512 // ...with address register writeback:
1513 class VLD4DUPWB<bits<4> op7_4, string Dt>
1514 : NLdSt<1, 0b10, 0b1111, op7_4,
1515 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
1516 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD4dupu,
1517 "vld4", Dt, "\\{$Vd[], $dst2[], $dst3[], $dst4[]\\}, $Rn$Rm",
1518 "$Rn.addr = $wb", []> {
1519 let Inst{4} = Rn{4};
1520 let DecoderMethod = "DecodeVLD4DupInstruction";
1523 def VLD4DUPd8_UPD : VLD4DUPWB<{0,0,0,0}, "8">;
1524 def VLD4DUPd16_UPD : VLD4DUPWB<{0,1,0,?}, "16">;
1525 def VLD4DUPd32_UPD : VLD4DUPWB<{1,?,0,?}, "32"> { let Inst{6} = Rn{5}; }
1527 def VLD4DUPq8_UPD : VLD4DUPWB<{0,0,1,0}, "8">;
1528 def VLD4DUPq16_UPD : VLD4DUPWB<{0,1,1,?}, "16">;
1529 def VLD4DUPq32_UPD : VLD4DUPWB<{1,?,1,?}, "32"> { let Inst{6} = Rn{5}; }
1531 def VLD4DUPd8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1532 def VLD4DUPd16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1533 def VLD4DUPd32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1535 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
1537 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
1539 // Classes for VST* pseudo-instructions with multi-register operands.
1540 // These are expanded to real instructions after register allocation.
1541 class VSTQPseudo<InstrItinClass itin>
1542 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src), itin, "">;
1543 class VSTQWBPseudo<InstrItinClass itin>
1544 : PseudoNLdSt<(outs GPR:$wb),
1545 (ins addrmode6:$addr, am6offset:$offset, QPR:$src), itin,
1546 "$addr.addr = $wb">;
1547 class VSTQWBfixedPseudo<InstrItinClass itin>
1548 : PseudoNLdSt<(outs GPR:$wb),
1549 (ins addrmode6:$addr, QPR:$src), itin,
1550 "$addr.addr = $wb">;
1551 class VSTQWBregisterPseudo<InstrItinClass itin>
1552 : PseudoNLdSt<(outs GPR:$wb),
1553 (ins addrmode6:$addr, rGPR:$offset, QPR:$src), itin,
1554 "$addr.addr = $wb">;
1555 class VSTQQPseudo<InstrItinClass itin>
1556 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src), itin, "">;
1557 class VSTQQWBPseudo<InstrItinClass itin>
1558 : PseudoNLdSt<(outs GPR:$wb),
1559 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src), itin,
1560 "$addr.addr = $wb">;
1561 class VSTQQWBfixedPseudo<InstrItinClass itin>
1562 : PseudoNLdSt<(outs GPR:$wb),
1563 (ins addrmode6:$addr, QQPR:$src), itin,
1564 "$addr.addr = $wb">;
1565 class VSTQQWBregisterPseudo<InstrItinClass itin>
1566 : PseudoNLdSt<(outs GPR:$wb),
1567 (ins addrmode6:$addr, rGPR:$offset, QQPR:$src), itin,
1568 "$addr.addr = $wb">;
1570 class VSTQQQQPseudo<InstrItinClass itin>
1571 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src), itin, "">;
1572 class VSTQQQQWBPseudo<InstrItinClass itin>
1573 : PseudoNLdSt<(outs GPR:$wb),
1574 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
1575 "$addr.addr = $wb">;
1577 // VST1 : Vector Store (multiple single elements)
1578 class VST1D<bits<4> op7_4, string Dt>
1579 : NLdSt<0,0b00,0b0111,op7_4, (outs), (ins addrmode6:$Rn, VecListOneD:$Vd),
1580 IIC_VST1, "vst1", Dt, "$Vd, $Rn", "", []> {
1582 let Inst{4} = Rn{4};
1583 let DecoderMethod = "DecodeVLDST1Instruction";
1585 class VST1Q<bits<4> op7_4, string Dt>
1586 : NLdSt<0,0b00,0b1010,op7_4, (outs), (ins addrmode6:$Rn, VecListDPair:$Vd),
1587 IIC_VST1x2, "vst1", Dt, "$Vd, $Rn", "", []> {
1589 let Inst{5-4} = Rn{5-4};
1590 let DecoderMethod = "DecodeVLDST1Instruction";
1593 def VST1d8 : VST1D<{0,0,0,?}, "8">;
1594 def VST1d16 : VST1D<{0,1,0,?}, "16">;
1595 def VST1d32 : VST1D<{1,0,0,?}, "32">;
1596 def VST1d64 : VST1D<{1,1,0,?}, "64">;
1598 def VST1q8 : VST1Q<{0,0,?,?}, "8">;
1599 def VST1q16 : VST1Q<{0,1,?,?}, "16">;
1600 def VST1q32 : VST1Q<{1,0,?,?}, "32">;
1601 def VST1q64 : VST1Q<{1,1,?,?}, "64">;
1603 // ...with address register writeback:
1604 multiclass VST1DWB<bits<4> op7_4, string Dt> {
1605 def _fixed : NLdSt<0,0b00, 0b0111,op7_4, (outs GPR:$wb),
1606 (ins addrmode6:$Rn, VecListOneD:$Vd), IIC_VLD1u,
1607 "vst1", Dt, "$Vd, $Rn!",
1608 "$Rn.addr = $wb", []> {
1609 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1610 let Inst{4} = Rn{4};
1611 let DecoderMethod = "DecodeVLDST1Instruction";
1612 let AsmMatchConverter = "cvtVSTwbFixed";
1614 def _register : NLdSt<0,0b00,0b0111,op7_4, (outs GPR:$wb),
1615 (ins addrmode6:$Rn, rGPR:$Rm, VecListOneD:$Vd),
1617 "vst1", Dt, "$Vd, $Rn, $Rm",
1618 "$Rn.addr = $wb", []> {
1619 let Inst{4} = Rn{4};
1620 let DecoderMethod = "DecodeVLDST1Instruction";
1621 let AsmMatchConverter = "cvtVSTwbRegister";
1624 multiclass VST1QWB<bits<4> op7_4, string Dt> {
1625 def _fixed : NLdSt<0,0b00,0b1010,op7_4, (outs GPR:$wb),
1626 (ins addrmode6:$Rn, VecListDPair:$Vd), IIC_VLD1x2u,
1627 "vst1", Dt, "$Vd, $Rn!",
1628 "$Rn.addr = $wb", []> {
1629 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1630 let Inst{5-4} = Rn{5-4};
1631 let DecoderMethod = "DecodeVLDST1Instruction";
1632 let AsmMatchConverter = "cvtVSTwbFixed";
1634 def _register : NLdSt<0,0b00,0b1010,op7_4, (outs GPR:$wb),
1635 (ins addrmode6:$Rn, rGPR:$Rm, VecListDPair:$Vd),
1637 "vst1", Dt, "$Vd, $Rn, $Rm",
1638 "$Rn.addr = $wb", []> {
1639 let Inst{5-4} = Rn{5-4};
1640 let DecoderMethod = "DecodeVLDST1Instruction";
1641 let AsmMatchConverter = "cvtVSTwbRegister";
1645 defm VST1d8wb : VST1DWB<{0,0,0,?}, "8">;
1646 defm VST1d16wb : VST1DWB<{0,1,0,?}, "16">;
1647 defm VST1d32wb : VST1DWB<{1,0,0,?}, "32">;
1648 defm VST1d64wb : VST1DWB<{1,1,0,?}, "64">;
1650 defm VST1q8wb : VST1QWB<{0,0,?,?}, "8">;
1651 defm VST1q16wb : VST1QWB<{0,1,?,?}, "16">;
1652 defm VST1q32wb : VST1QWB<{1,0,?,?}, "32">;
1653 defm VST1q64wb : VST1QWB<{1,1,?,?}, "64">;
1655 // ...with 3 registers
1656 class VST1D3<bits<4> op7_4, string Dt>
1657 : NLdSt<0, 0b00, 0b0110, op7_4, (outs),
1658 (ins addrmode6:$Rn, VecListThreeD:$Vd),
1659 IIC_VST1x3, "vst1", Dt, "$Vd, $Rn", "", []> {
1661 let Inst{4} = Rn{4};
1662 let DecoderMethod = "DecodeVLDST1Instruction";
1664 multiclass VST1D3WB<bits<4> op7_4, string Dt> {
1665 def _fixed : NLdSt<0,0b00,0b0110,op7_4, (outs GPR:$wb),
1666 (ins addrmode6:$Rn, VecListThreeD:$Vd), IIC_VLD1x3u,
1667 "vst1", Dt, "$Vd, $Rn!",
1668 "$Rn.addr = $wb", []> {
1669 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1670 let Inst{5-4} = Rn{5-4};
1671 let DecoderMethod = "DecodeVLDST1Instruction";
1672 let AsmMatchConverter = "cvtVSTwbFixed";
1674 def _register : NLdSt<0,0b00,0b0110,op7_4, (outs GPR:$wb),
1675 (ins addrmode6:$Rn, rGPR:$Rm, VecListThreeD:$Vd),
1677 "vst1", Dt, "$Vd, $Rn, $Rm",
1678 "$Rn.addr = $wb", []> {
1679 let Inst{5-4} = Rn{5-4};
1680 let DecoderMethod = "DecodeVLDST1Instruction";
1681 let AsmMatchConverter = "cvtVSTwbRegister";
1685 def VST1d8T : VST1D3<{0,0,0,?}, "8">;
1686 def VST1d16T : VST1D3<{0,1,0,?}, "16">;
1687 def VST1d32T : VST1D3<{1,0,0,?}, "32">;
1688 def VST1d64T : VST1D3<{1,1,0,?}, "64">;
1690 defm VST1d8Twb : VST1D3WB<{0,0,0,?}, "8">;
1691 defm VST1d16Twb : VST1D3WB<{0,1,0,?}, "16">;
1692 defm VST1d32Twb : VST1D3WB<{1,0,0,?}, "32">;
1693 defm VST1d64Twb : VST1D3WB<{1,1,0,?}, "64">;
1695 def VST1d64TPseudo : VSTQQPseudo<IIC_VST1x3>;
1696 def VST1d64TPseudoWB_fixed : VSTQQWBPseudo<IIC_VST1x3u>;
1697 def VST1d64TPseudoWB_register : VSTQQWBPseudo<IIC_VST1x3u>;
1699 // ...with 4 registers
1700 class VST1D4<bits<4> op7_4, string Dt>
1701 : NLdSt<0, 0b00, 0b0010, op7_4, (outs),
1702 (ins addrmode6:$Rn, VecListFourD:$Vd),
1703 IIC_VST1x4, "vst1", Dt, "$Vd, $Rn", "",
1706 let Inst{5-4} = Rn{5-4};
1707 let DecoderMethod = "DecodeVLDST1Instruction";
1709 multiclass VST1D4WB<bits<4> op7_4, string Dt> {
1710 def _fixed : NLdSt<0,0b00,0b0010,op7_4, (outs GPR:$wb),
1711 (ins addrmode6:$Rn, VecListFourD:$Vd), IIC_VLD1x4u,
1712 "vst1", Dt, "$Vd, $Rn!",
1713 "$Rn.addr = $wb", []> {
1714 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1715 let Inst{5-4} = Rn{5-4};
1716 let DecoderMethod = "DecodeVLDST1Instruction";
1717 let AsmMatchConverter = "cvtVSTwbFixed";
1719 def _register : NLdSt<0,0b00,0b0010,op7_4, (outs GPR:$wb),
1720 (ins addrmode6:$Rn, rGPR:$Rm, VecListFourD:$Vd),
1722 "vst1", Dt, "$Vd, $Rn, $Rm",
1723 "$Rn.addr = $wb", []> {
1724 let Inst{5-4} = Rn{5-4};
1725 let DecoderMethod = "DecodeVLDST1Instruction";
1726 let AsmMatchConverter = "cvtVSTwbRegister";
1730 def VST1d8Q : VST1D4<{0,0,?,?}, "8">;
1731 def VST1d16Q : VST1D4<{0,1,?,?}, "16">;
1732 def VST1d32Q : VST1D4<{1,0,?,?}, "32">;
1733 def VST1d64Q : VST1D4<{1,1,?,?}, "64">;
1735 defm VST1d8Qwb : VST1D4WB<{0,0,?,?}, "8">;
1736 defm VST1d16Qwb : VST1D4WB<{0,1,?,?}, "16">;
1737 defm VST1d32Qwb : VST1D4WB<{1,0,?,?}, "32">;
1738 defm VST1d64Qwb : VST1D4WB<{1,1,?,?}, "64">;
1740 def VST1d64QPseudo : VSTQQPseudo<IIC_VST1x4>;
1741 def VST1d64QPseudoWB_fixed : VSTQQWBPseudo<IIC_VST1x4u>;
1742 def VST1d64QPseudoWB_register : VSTQQWBPseudo<IIC_VST1x4u>;
1744 // VST2 : Vector Store (multiple 2-element structures)
1745 class VST2<bits<4> op11_8, bits<4> op7_4, string Dt, RegisterOperand VdTy,
1746 InstrItinClass itin>
1747 : NLdSt<0, 0b00, op11_8, op7_4, (outs), (ins addrmode6:$Rn, VdTy:$Vd),
1748 itin, "vst2", Dt, "$Vd, $Rn", "", []> {
1750 let Inst{5-4} = Rn{5-4};
1751 let DecoderMethod = "DecodeVLDST2Instruction";
1754 def VST2d8 : VST2<0b1000, {0,0,?,?}, "8", VecListDPair, IIC_VST2>;
1755 def VST2d16 : VST2<0b1000, {0,1,?,?}, "16", VecListDPair, IIC_VST2>;
1756 def VST2d32 : VST2<0b1000, {1,0,?,?}, "32", VecListDPair, IIC_VST2>;
1758 def VST2q8 : VST2<0b0011, {0,0,?,?}, "8", VecListFourD, IIC_VST2x2>;
1759 def VST2q16 : VST2<0b0011, {0,1,?,?}, "16", VecListFourD, IIC_VST2x2>;
1760 def VST2q32 : VST2<0b0011, {1,0,?,?}, "32", VecListFourD, IIC_VST2x2>;
1762 def VST2q8Pseudo : VSTQQPseudo<IIC_VST2x2>;
1763 def VST2q16Pseudo : VSTQQPseudo<IIC_VST2x2>;
1764 def VST2q32Pseudo : VSTQQPseudo<IIC_VST2x2>;
1766 // ...with address register writeback:
1767 multiclass VST2DWB<bits<4> op11_8, bits<4> op7_4, string Dt,
1768 RegisterOperand VdTy> {
1769 def _fixed : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1770 (ins addrmode6:$Rn, VdTy:$Vd), IIC_VLD1u,
1771 "vst2", Dt, "$Vd, $Rn!",
1772 "$Rn.addr = $wb", []> {
1773 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1774 let Inst{5-4} = Rn{5-4};
1775 let DecoderMethod = "DecodeVLDST2Instruction";
1776 let AsmMatchConverter = "cvtVSTwbFixed";
1778 def _register : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1779 (ins addrmode6:$Rn, rGPR:$Rm, VdTy:$Vd), IIC_VLD1u,
1780 "vst2", Dt, "$Vd, $Rn, $Rm",
1781 "$Rn.addr = $wb", []> {
1782 let Inst{5-4} = Rn{5-4};
1783 let DecoderMethod = "DecodeVLDST2Instruction";
1784 let AsmMatchConverter = "cvtVSTwbRegister";
1787 multiclass VST2QWB<bits<4> op7_4, string Dt> {
1788 def _fixed : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
1789 (ins addrmode6:$Rn, VecListFourD:$Vd), IIC_VLD1u,
1790 "vst2", Dt, "$Vd, $Rn!",
1791 "$Rn.addr = $wb", []> {
1792 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1793 let Inst{5-4} = Rn{5-4};
1794 let DecoderMethod = "DecodeVLDST2Instruction";
1795 let AsmMatchConverter = "cvtVSTwbFixed";
1797 def _register : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
1798 (ins addrmode6:$Rn, rGPR:$Rm, VecListFourD:$Vd),
1800 "vst2", Dt, "$Vd, $Rn, $Rm",
1801 "$Rn.addr = $wb", []> {
1802 let Inst{5-4} = Rn{5-4};
1803 let DecoderMethod = "DecodeVLDST2Instruction";
1804 let AsmMatchConverter = "cvtVSTwbRegister";
1808 defm VST2d8wb : VST2DWB<0b1000, {0,0,?,?}, "8", VecListDPair>;
1809 defm VST2d16wb : VST2DWB<0b1000, {0,1,?,?}, "16", VecListDPair>;
1810 defm VST2d32wb : VST2DWB<0b1000, {1,0,?,?}, "32", VecListDPair>;
1812 defm VST2q8wb : VST2QWB<{0,0,?,?}, "8">;
1813 defm VST2q16wb : VST2QWB<{0,1,?,?}, "16">;
1814 defm VST2q32wb : VST2QWB<{1,0,?,?}, "32">;
1816 def VST2q8PseudoWB_fixed : VSTQQWBfixedPseudo<IIC_VST2x2u>;
1817 def VST2q16PseudoWB_fixed : VSTQQWBfixedPseudo<IIC_VST2x2u>;
1818 def VST2q32PseudoWB_fixed : VSTQQWBfixedPseudo<IIC_VST2x2u>;
1819 def VST2q8PseudoWB_register : VSTQQWBregisterPseudo<IIC_VST2x2u>;
1820 def VST2q16PseudoWB_register : VSTQQWBregisterPseudo<IIC_VST2x2u>;
1821 def VST2q32PseudoWB_register : VSTQQWBregisterPseudo<IIC_VST2x2u>;
1823 // ...with double-spaced registers
1824 def VST2b8 : VST2<0b1001, {0,0,?,?}, "8", VecListDPairSpaced, IIC_VST2>;
1825 def VST2b16 : VST2<0b1001, {0,1,?,?}, "16", VecListDPairSpaced, IIC_VST2>;
1826 def VST2b32 : VST2<0b1001, {1,0,?,?}, "32", VecListDPairSpaced, IIC_VST2>;
1827 defm VST2b8wb : VST2DWB<0b1001, {0,0,?,?}, "8", VecListDPairSpaced>;
1828 defm VST2b16wb : VST2DWB<0b1001, {0,1,?,?}, "16", VecListDPairSpaced>;
1829 defm VST2b32wb : VST2DWB<0b1001, {1,0,?,?}, "32", VecListDPairSpaced>;
1831 // VST3 : Vector Store (multiple 3-element structures)
1832 class VST3D<bits<4> op11_8, bits<4> op7_4, string Dt>
1833 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1834 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3,
1835 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn", "", []> {
1837 let Inst{4} = Rn{4};
1838 let DecoderMethod = "DecodeVLDST3Instruction";
1841 def VST3d8 : VST3D<0b0100, {0,0,0,?}, "8">;
1842 def VST3d16 : VST3D<0b0100, {0,1,0,?}, "16">;
1843 def VST3d32 : VST3D<0b0100, {1,0,0,?}, "32">;
1845 def VST3d8Pseudo : VSTQQPseudo<IIC_VST3>;
1846 def VST3d16Pseudo : VSTQQPseudo<IIC_VST3>;
1847 def VST3d32Pseudo : VSTQQPseudo<IIC_VST3>;
1849 // ...with address register writeback:
1850 class VST3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1851 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1852 (ins addrmode6:$Rn, am6offset:$Rm,
1853 DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3u,
1854 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn$Rm",
1855 "$Rn.addr = $wb", []> {
1856 let Inst{4} = Rn{4};
1857 let DecoderMethod = "DecodeVLDST3Instruction";
1860 def VST3d8_UPD : VST3DWB<0b0100, {0,0,0,?}, "8">;
1861 def VST3d16_UPD : VST3DWB<0b0100, {0,1,0,?}, "16">;
1862 def VST3d32_UPD : VST3DWB<0b0100, {1,0,0,?}, "32">;
1864 def VST3d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1865 def VST3d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1866 def VST3d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1868 // ...with double-spaced registers:
1869 def VST3q8 : VST3D<0b0101, {0,0,0,?}, "8">;
1870 def VST3q16 : VST3D<0b0101, {0,1,0,?}, "16">;
1871 def VST3q32 : VST3D<0b0101, {1,0,0,?}, "32">;
1872 def VST3q8_UPD : VST3DWB<0b0101, {0,0,0,?}, "8">;
1873 def VST3q16_UPD : VST3DWB<0b0101, {0,1,0,?}, "16">;
1874 def VST3q32_UPD : VST3DWB<0b0101, {1,0,0,?}, "32">;
1876 def VST3q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1877 def VST3q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1878 def VST3q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1880 // ...alternate versions to be allocated odd register numbers:
1881 def VST3q8oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1882 def VST3q16oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1883 def VST3q32oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1885 def VST3q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1886 def VST3q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1887 def VST3q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1889 // VST4 : Vector Store (multiple 4-element structures)
1890 class VST4D<bits<4> op11_8, bits<4> op7_4, string Dt>
1891 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1892 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
1893 IIC_VST4, "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn",
1896 let Inst{5-4} = Rn{5-4};
1897 let DecoderMethod = "DecodeVLDST4Instruction";
1900 def VST4d8 : VST4D<0b0000, {0,0,?,?}, "8">;
1901 def VST4d16 : VST4D<0b0000, {0,1,?,?}, "16">;
1902 def VST4d32 : VST4D<0b0000, {1,0,?,?}, "32">;
1904 def VST4d8Pseudo : VSTQQPseudo<IIC_VST4>;
1905 def VST4d16Pseudo : VSTQQPseudo<IIC_VST4>;
1906 def VST4d32Pseudo : VSTQQPseudo<IIC_VST4>;
1908 // ...with address register writeback:
1909 class VST4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1910 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1911 (ins addrmode6:$Rn, am6offset:$Rm,
1912 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST4u,
1913 "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
1914 "$Rn.addr = $wb", []> {
1915 let Inst{5-4} = Rn{5-4};
1916 let DecoderMethod = "DecodeVLDST4Instruction";
1919 def VST4d8_UPD : VST4DWB<0b0000, {0,0,?,?}, "8">;
1920 def VST4d16_UPD : VST4DWB<0b0000, {0,1,?,?}, "16">;
1921 def VST4d32_UPD : VST4DWB<0b0000, {1,0,?,?}, "32">;
1923 def VST4d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1924 def VST4d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1925 def VST4d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1927 // ...with double-spaced registers:
1928 def VST4q8 : VST4D<0b0001, {0,0,?,?}, "8">;
1929 def VST4q16 : VST4D<0b0001, {0,1,?,?}, "16">;
1930 def VST4q32 : VST4D<0b0001, {1,0,?,?}, "32">;
1931 def VST4q8_UPD : VST4DWB<0b0001, {0,0,?,?}, "8">;
1932 def VST4q16_UPD : VST4DWB<0b0001, {0,1,?,?}, "16">;
1933 def VST4q32_UPD : VST4DWB<0b0001, {1,0,?,?}, "32">;
1935 def VST4q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1936 def VST4q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1937 def VST4q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1939 // ...alternate versions to be allocated odd register numbers:
1940 def VST4q8oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1941 def VST4q16oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1942 def VST4q32oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1944 def VST4q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1945 def VST4q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1946 def VST4q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1948 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
1950 // Classes for VST*LN pseudo-instructions with multi-register operands.
1951 // These are expanded to real instructions after register allocation.
1952 class VSTQLNPseudo<InstrItinClass itin>
1953 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
1955 class VSTQLNWBPseudo<InstrItinClass itin>
1956 : PseudoNLdSt<(outs GPR:$wb),
1957 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
1958 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1959 class VSTQQLNPseudo<InstrItinClass itin>
1960 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
1962 class VSTQQLNWBPseudo<InstrItinClass itin>
1963 : PseudoNLdSt<(outs GPR:$wb),
1964 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
1965 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1966 class VSTQQQQLNPseudo<InstrItinClass itin>
1967 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
1969 class VSTQQQQLNWBPseudo<InstrItinClass itin>
1970 : PseudoNLdSt<(outs GPR:$wb),
1971 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
1972 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1974 // VST1LN : Vector Store (single element from one lane)
1975 class VST1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1976 PatFrag StoreOp, SDNode ExtractOp, Operand AddrMode>
1977 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1978 (ins AddrMode:$Rn, DPR:$Vd, nohash_imm:$lane),
1979 IIC_VST1ln, "vst1", Dt, "\\{$Vd[$lane]\\}, $Rn", "",
1980 [(StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane), AddrMode:$Rn)]> {
1982 let DecoderMethod = "DecodeVST1LN";
1984 class VST1QLNPseudo<ValueType Ty, PatFrag StoreOp, SDNode ExtractOp>
1985 : VSTQLNPseudo<IIC_VST1ln> {
1986 let Pattern = [(StoreOp (ExtractOp (Ty QPR:$src), imm:$lane),
1990 def VST1LNd8 : VST1LN<0b0000, {?,?,?,0}, "8", v8i8, truncstorei8,
1991 NEONvgetlaneu, addrmode6> {
1992 let Inst{7-5} = lane{2-0};
1994 def VST1LNd16 : VST1LN<0b0100, {?,?,0,?}, "16", v4i16, truncstorei16,
1995 NEONvgetlaneu, addrmode6> {
1996 let Inst{7-6} = lane{1-0};
1997 let Inst{4} = Rn{4};
2000 def VST1LNd32 : VST1LN<0b1000, {?,0,?,?}, "32", v2i32, store, extractelt,
2002 let Inst{7} = lane{0};
2003 let Inst{5-4} = Rn{5-4};
2006 def VST1LNq8Pseudo : VST1QLNPseudo<v16i8, truncstorei8, NEONvgetlaneu>;
2007 def VST1LNq16Pseudo : VST1QLNPseudo<v8i16, truncstorei16, NEONvgetlaneu>;
2008 def VST1LNq32Pseudo : VST1QLNPseudo<v4i32, store, extractelt>;
2010 def : Pat<(store (extractelt (v2f32 DPR:$src), imm:$lane), addrmode6:$addr),
2011 (VST1LNd32 addrmode6:$addr, DPR:$src, imm:$lane)>;
2012 def : Pat<(store (extractelt (v4f32 QPR:$src), imm:$lane), addrmode6:$addr),
2013 (VST1LNq32Pseudo addrmode6:$addr, QPR:$src, imm:$lane)>;
2015 // ...with address register writeback:
2016 class VST1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
2017 PatFrag StoreOp, SDNode ExtractOp, Operand AdrMode>
2018 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
2019 (ins AdrMode:$Rn, am6offset:$Rm,
2020 DPR:$Vd, nohash_imm:$lane), IIC_VST1lnu, "vst1", Dt,
2021 "\\{$Vd[$lane]\\}, $Rn$Rm",
2023 [(set GPR:$wb, (StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane),
2024 AdrMode:$Rn, am6offset:$Rm))]> {
2025 let DecoderMethod = "DecodeVST1LN";
2027 class VST1QLNWBPseudo<ValueType Ty, PatFrag StoreOp, SDNode ExtractOp>
2028 : VSTQLNWBPseudo<IIC_VST1lnu> {
2029 let Pattern = [(set GPR:$wb, (StoreOp (ExtractOp (Ty QPR:$src), imm:$lane),
2030 addrmode6:$addr, am6offset:$offset))];
2033 def VST1LNd8_UPD : VST1LNWB<0b0000, {?,?,?,0}, "8", v8i8, post_truncsti8,
2034 NEONvgetlaneu, addrmode6> {
2035 let Inst{7-5} = lane{2-0};
2037 def VST1LNd16_UPD : VST1LNWB<0b0100, {?,?,0,?}, "16", v4i16, post_truncsti16,
2038 NEONvgetlaneu, addrmode6> {
2039 let Inst{7-6} = lane{1-0};
2040 let Inst{4} = Rn{4};
2042 def VST1LNd32_UPD : VST1LNWB<0b1000, {?,0,?,?}, "32", v2i32, post_store,
2043 extractelt, addrmode6oneL32> {
2044 let Inst{7} = lane{0};
2045 let Inst{5-4} = Rn{5-4};
2048 def VST1LNq8Pseudo_UPD : VST1QLNWBPseudo<v16i8, post_truncsti8, NEONvgetlaneu>;
2049 def VST1LNq16Pseudo_UPD : VST1QLNWBPseudo<v8i16, post_truncsti16,NEONvgetlaneu>;
2050 def VST1LNq32Pseudo_UPD : VST1QLNWBPseudo<v4i32, post_store, extractelt>;
2052 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
2054 // VST2LN : Vector Store (single 2-element structure from one lane)
2055 class VST2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
2056 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
2057 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, nohash_imm:$lane),
2058 IIC_VST2ln, "vst2", Dt, "\\{$Vd[$lane], $src2[$lane]\\}, $Rn",
2061 let Inst{4} = Rn{4};
2062 let DecoderMethod = "DecodeVST2LN";
2065 def VST2LNd8 : VST2LN<0b0001, {?,?,?,?}, "8"> {
2066 let Inst{7-5} = lane{2-0};
2068 def VST2LNd16 : VST2LN<0b0101, {?,?,0,?}, "16"> {
2069 let Inst{7-6} = lane{1-0};
2071 def VST2LNd32 : VST2LN<0b1001, {?,0,0,?}, "32"> {
2072 let Inst{7} = lane{0};
2075 def VST2LNd8Pseudo : VSTQLNPseudo<IIC_VST2ln>;
2076 def VST2LNd16Pseudo : VSTQLNPseudo<IIC_VST2ln>;
2077 def VST2LNd32Pseudo : VSTQLNPseudo<IIC_VST2ln>;
2079 // ...with double-spaced registers:
2080 def VST2LNq16 : VST2LN<0b0101, {?,?,1,?}, "16"> {
2081 let Inst{7-6} = lane{1-0};
2082 let Inst{4} = Rn{4};
2084 def VST2LNq32 : VST2LN<0b1001, {?,1,0,?}, "32"> {
2085 let Inst{7} = lane{0};
2086 let Inst{4} = Rn{4};
2089 def VST2LNq16Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
2090 def VST2LNq32Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
2092 // ...with address register writeback:
2093 class VST2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
2094 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
2095 (ins addrmode6:$Rn, am6offset:$Rm,
2096 DPR:$Vd, DPR:$src2, nohash_imm:$lane), IIC_VST2lnu, "vst2", Dt,
2097 "\\{$Vd[$lane], $src2[$lane]\\}, $Rn$Rm",
2098 "$Rn.addr = $wb", []> {
2099 let Inst{4} = Rn{4};
2100 let DecoderMethod = "DecodeVST2LN";
2103 def VST2LNd8_UPD : VST2LNWB<0b0001, {?,?,?,?}, "8"> {
2104 let Inst{7-5} = lane{2-0};
2106 def VST2LNd16_UPD : VST2LNWB<0b0101, {?,?,0,?}, "16"> {
2107 let Inst{7-6} = lane{1-0};
2109 def VST2LNd32_UPD : VST2LNWB<0b1001, {?,0,0,?}, "32"> {
2110 let Inst{7} = lane{0};
2113 def VST2LNd8Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
2114 def VST2LNd16Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
2115 def VST2LNd32Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
2117 def VST2LNq16_UPD : VST2LNWB<0b0101, {?,?,1,?}, "16"> {
2118 let Inst{7-6} = lane{1-0};
2120 def VST2LNq32_UPD : VST2LNWB<0b1001, {?,1,0,?}, "32"> {
2121 let Inst{7} = lane{0};
2124 def VST2LNq16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
2125 def VST2LNq32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
2127 // VST3LN : Vector Store (single 3-element structure from one lane)
2128 class VST3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
2129 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
2130 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3,
2131 nohash_imm:$lane), IIC_VST3ln, "vst3", Dt,
2132 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn", "", []> {
2134 let DecoderMethod = "DecodeVST3LN";
2137 def VST3LNd8 : VST3LN<0b0010, {?,?,?,0}, "8"> {
2138 let Inst{7-5} = lane{2-0};
2140 def VST3LNd16 : VST3LN<0b0110, {?,?,0,0}, "16"> {
2141 let Inst{7-6} = lane{1-0};
2143 def VST3LNd32 : VST3LN<0b1010, {?,0,0,0}, "32"> {
2144 let Inst{7} = lane{0};
2147 def VST3LNd8Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
2148 def VST3LNd16Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
2149 def VST3LNd32Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
2151 // ...with double-spaced registers:
2152 def VST3LNq16 : VST3LN<0b0110, {?,?,1,0}, "16"> {
2153 let Inst{7-6} = lane{1-0};
2155 def VST3LNq32 : VST3LN<0b1010, {?,1,0,0}, "32"> {
2156 let Inst{7} = lane{0};
2159 def VST3LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
2160 def VST3LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
2162 // ...with address register writeback:
2163 class VST3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
2164 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
2165 (ins addrmode6:$Rn, am6offset:$Rm,
2166 DPR:$Vd, DPR:$src2, DPR:$src3, nohash_imm:$lane),
2167 IIC_VST3lnu, "vst3", Dt,
2168 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn$Rm",
2169 "$Rn.addr = $wb", []> {
2170 let DecoderMethod = "DecodeVST3LN";
2173 def VST3LNd8_UPD : VST3LNWB<0b0010, {?,?,?,0}, "8"> {
2174 let Inst{7-5} = lane{2-0};
2176 def VST3LNd16_UPD : VST3LNWB<0b0110, {?,?,0,0}, "16"> {
2177 let Inst{7-6} = lane{1-0};
2179 def VST3LNd32_UPD : VST3LNWB<0b1010, {?,0,0,0}, "32"> {
2180 let Inst{7} = lane{0};
2183 def VST3LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
2184 def VST3LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
2185 def VST3LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
2187 def VST3LNq16_UPD : VST3LNWB<0b0110, {?,?,1,0}, "16"> {
2188 let Inst{7-6} = lane{1-0};
2190 def VST3LNq32_UPD : VST3LNWB<0b1010, {?,1,0,0}, "32"> {
2191 let Inst{7} = lane{0};
2194 def VST3LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
2195 def VST3LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
2197 // VST4LN : Vector Store (single 4-element structure from one lane)
2198 class VST4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
2199 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
2200 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4,
2201 nohash_imm:$lane), IIC_VST4ln, "vst4", Dt,
2202 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn",
2205 let Inst{4} = Rn{4};
2206 let DecoderMethod = "DecodeVST4LN";
2209 def VST4LNd8 : VST4LN<0b0011, {?,?,?,?}, "8"> {
2210 let Inst{7-5} = lane{2-0};
2212 def VST4LNd16 : VST4LN<0b0111, {?,?,0,?}, "16"> {
2213 let Inst{7-6} = lane{1-0};
2215 def VST4LNd32 : VST4LN<0b1011, {?,0,?,?}, "32"> {
2216 let Inst{7} = lane{0};
2217 let Inst{5} = Rn{5};
2220 def VST4LNd8Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
2221 def VST4LNd16Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
2222 def VST4LNd32Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
2224 // ...with double-spaced registers:
2225 def VST4LNq16 : VST4LN<0b0111, {?,?,1,?}, "16"> {
2226 let Inst{7-6} = lane{1-0};
2228 def VST4LNq32 : VST4LN<0b1011, {?,1,?,?}, "32"> {
2229 let Inst{7} = lane{0};
2230 let Inst{5} = Rn{5};
2233 def VST4LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
2234 def VST4LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
2236 // ...with address register writeback:
2237 class VST4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
2238 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
2239 (ins addrmode6:$Rn, am6offset:$Rm,
2240 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
2241 IIC_VST4lnu, "vst4", Dt,
2242 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn$Rm",
2243 "$Rn.addr = $wb", []> {
2244 let Inst{4} = Rn{4};
2245 let DecoderMethod = "DecodeVST4LN";
2248 def VST4LNd8_UPD : VST4LNWB<0b0011, {?,?,?,?}, "8"> {
2249 let Inst{7-5} = lane{2-0};
2251 def VST4LNd16_UPD : VST4LNWB<0b0111, {?,?,0,?}, "16"> {
2252 let Inst{7-6} = lane{1-0};
2254 def VST4LNd32_UPD : VST4LNWB<0b1011, {?,0,?,?}, "32"> {
2255 let Inst{7} = lane{0};
2256 let Inst{5} = Rn{5};
2259 def VST4LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
2260 def VST4LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
2261 def VST4LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
2263 def VST4LNq16_UPD : VST4LNWB<0b0111, {?,?,1,?}, "16"> {
2264 let Inst{7-6} = lane{1-0};
2266 def VST4LNq32_UPD : VST4LNWB<0b1011, {?,1,?,?}, "32"> {
2267 let Inst{7} = lane{0};
2268 let Inst{5} = Rn{5};
2271 def VST4LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
2272 def VST4LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
2274 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
2276 // Use vld1/vst1 for unaligned f64 load / store
2277 def : Pat<(f64 (hword_alignedload addrmode6:$addr)),
2278 (VLD1d16 addrmode6:$addr)>, Requires<[IsLE]>;
2279 def : Pat<(hword_alignedstore (f64 DPR:$value), addrmode6:$addr),
2280 (VST1d16 addrmode6:$addr, DPR:$value)>, Requires<[IsLE]>;
2281 def : Pat<(f64 (byte_alignedload addrmode6:$addr)),
2282 (VLD1d8 addrmode6:$addr)>, Requires<[IsLE]>;
2283 def : Pat<(byte_alignedstore (f64 DPR:$value), addrmode6:$addr),
2284 (VST1d8 addrmode6:$addr, DPR:$value)>, Requires<[IsLE]>;
2285 def : Pat<(f64 (non_word_alignedload addrmode6:$addr)),
2286 (VLD1d64 addrmode6:$addr)>, Requires<[IsBE]>;
2287 def : Pat<(non_word_alignedstore (f64 DPR:$value), addrmode6:$addr),
2288 (VST1d64 addrmode6:$addr, DPR:$value)>, Requires<[IsBE]>;
2290 // Use vld1/vst1 for Q and QQ. Also use them for unaligned v2f64
2291 // load / store if it's legal.
2292 def : Pat<(v2f64 (dword_alignedload addrmode6:$addr)),
2293 (VLD1q64 addrmode6:$addr)>;
2294 def : Pat<(dword_alignedstore (v2f64 QPR:$value), addrmode6:$addr),
2295 (VST1q64 addrmode6:$addr, QPR:$value)>;
2296 def : Pat<(v2f64 (word_alignedload addrmode6:$addr)),
2297 (VLD1q32 addrmode6:$addr)>;
2298 def : Pat<(word_alignedstore (v2f64 QPR:$value), addrmode6:$addr),
2299 (VST1q32 addrmode6:$addr, QPR:$value)>;
2300 def : Pat<(v2f64 (hword_alignedload addrmode6:$addr)),
2301 (VLD1q16 addrmode6:$addr)>, Requires<[IsLE]>;
2302 def : Pat<(hword_alignedstore (v2f64 QPR:$value), addrmode6:$addr),
2303 (VST1q16 addrmode6:$addr, QPR:$value)>, Requires<[IsLE]>;
2304 def : Pat<(v2f64 (byte_alignedload addrmode6:$addr)),
2305 (VLD1q8 addrmode6:$addr)>, Requires<[IsLE]>;
2306 def : Pat<(byte_alignedstore (v2f64 QPR:$value), addrmode6:$addr),
2307 (VST1q8 addrmode6:$addr, QPR:$value)>, Requires<[IsLE]>;
2309 //===----------------------------------------------------------------------===//
2310 // NEON pattern fragments
2311 //===----------------------------------------------------------------------===//
2313 // Extract D sub-registers of Q registers.
2314 def DSubReg_i8_reg : SDNodeXForm<imm, [{
2315 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
2316 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/8, MVT::i32);
2318 def DSubReg_i16_reg : SDNodeXForm<imm, [{
2319 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
2320 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/4, MVT::i32);
2322 def DSubReg_i32_reg : SDNodeXForm<imm, [{
2323 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
2324 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/2, MVT::i32);
2326 def DSubReg_f64_reg : SDNodeXForm<imm, [{
2327 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
2328 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue(), MVT::i32);
2331 // Extract S sub-registers of Q/D registers.
2332 def SSubReg_f32_reg : SDNodeXForm<imm, [{
2333 assert(ARM::ssub_3 == ARM::ssub_0+3 && "Unexpected subreg numbering");
2334 return CurDAG->getTargetConstant(ARM::ssub_0 + N->getZExtValue(), MVT::i32);
2337 // Translate lane numbers from Q registers to D subregs.
2338 def SubReg_i8_lane : SDNodeXForm<imm, [{
2339 return CurDAG->getTargetConstant(N->getZExtValue() & 7, MVT::i32);
2341 def SubReg_i16_lane : SDNodeXForm<imm, [{
2342 return CurDAG->getTargetConstant(N->getZExtValue() & 3, MVT::i32);
2344 def SubReg_i32_lane : SDNodeXForm<imm, [{
2345 return CurDAG->getTargetConstant(N->getZExtValue() & 1, MVT::i32);
2348 //===----------------------------------------------------------------------===//
2349 // Instruction Classes
2350 //===----------------------------------------------------------------------===//
2352 // Basic 2-register operations: double- and quad-register.
2353 class N2VD<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2354 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
2355 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
2356 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
2357 (ins DPR:$Vm), IIC_VUNAD, OpcodeStr, Dt,"$Vd, $Vm", "",
2358 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vm))))]>;
2359 class N2VQ<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2360 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
2361 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
2362 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
2363 (ins QPR:$Vm), IIC_VUNAQ, OpcodeStr, Dt,"$Vd, $Vm", "",
2364 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vm))))]>;
2366 // Basic 2-register intrinsics, both double- and quad-register.
2367 class N2VDInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2368 bits<2> op17_16, bits<5> op11_7, bit op4,
2369 InstrItinClass itin, string OpcodeStr, string Dt,
2370 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2371 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
2372 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2373 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm))))]>;
2374 class N2VQInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2375 bits<2> op17_16, bits<5> op11_7, bit op4,
2376 InstrItinClass itin, string OpcodeStr, string Dt,
2377 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2378 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
2379 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2380 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm))))]>;
2382 // Same as above, but not predicated.
2383 class N2VDIntnp<bits<2> op17_16, bits<3> op10_8, bit op7,
2384 InstrItinClass itin, string OpcodeStr, string Dt,
2385 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2386 : N2Vnp<op17_16, op10_8, op7, 0, (outs DPR:$Vd), (ins DPR:$Vm),
2387 itin, OpcodeStr, Dt, ResTy, OpTy,
2388 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm))))]>;
2390 class N2VQIntnp<bits<2> op17_16, bits<3> op10_8, bit op7,
2391 InstrItinClass itin, string OpcodeStr, string Dt,
2392 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2393 : N2Vnp<op17_16, op10_8, op7, 1, (outs QPR:$Vd), (ins QPR:$Vm),
2394 itin, OpcodeStr, Dt, ResTy, OpTy,
2395 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm))))]>;
2397 // Narrow 2-register operations.
2398 class N2VN<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2399 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2400 InstrItinClass itin, string OpcodeStr, string Dt,
2401 ValueType TyD, ValueType TyQ, SDNode OpNode>
2402 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$Vd),
2403 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2404 [(set DPR:$Vd, (TyD (OpNode (TyQ QPR:$Vm))))]>;
2406 // Narrow 2-register intrinsics.
2407 class N2VNInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2408 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2409 InstrItinClass itin, string OpcodeStr, string Dt,
2410 ValueType TyD, ValueType TyQ, SDPatternOperator IntOp>
2411 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$Vd),
2412 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2413 [(set DPR:$Vd, (TyD (IntOp (TyQ QPR:$Vm))))]>;
2415 // Long 2-register operations (currently only used for VMOVL).
2416 class N2VL<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2417 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2418 InstrItinClass itin, string OpcodeStr, string Dt,
2419 ValueType TyQ, ValueType TyD, SDNode OpNode>
2420 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$Vd),
2421 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2422 [(set QPR:$Vd, (TyQ (OpNode (TyD DPR:$Vm))))]>;
2424 // Long 2-register intrinsics.
2425 class N2VLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2426 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2427 InstrItinClass itin, string OpcodeStr, string Dt,
2428 ValueType TyQ, ValueType TyD, SDPatternOperator IntOp>
2429 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$Vd),
2430 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2431 [(set QPR:$Vd, (TyQ (IntOp (TyD DPR:$Vm))))]>;
2433 // 2-register shuffles (VTRN/VZIP/VUZP), both double- and quad-register.
2434 class N2VDShuffle<bits<2> op19_18, bits<5> op11_7, string OpcodeStr, string Dt>
2435 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 0, 0, (outs DPR:$Vd, DPR:$Vm),
2436 (ins DPR:$src1, DPR:$src2), IIC_VPERMD,
2437 OpcodeStr, Dt, "$Vd, $Vm",
2438 "$src1 = $Vd, $src2 = $Vm", []>;
2439 class N2VQShuffle<bits<2> op19_18, bits<5> op11_7,
2440 InstrItinClass itin, string OpcodeStr, string Dt>
2441 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 1, 0, (outs QPR:$Vd, QPR:$Vm),
2442 (ins QPR:$src1, QPR:$src2), itin, OpcodeStr, Dt, "$Vd, $Vm",
2443 "$src1 = $Vd, $src2 = $Vm", []>;
2445 // Basic 3-register operations: double- and quad-register.
2446 class N3VD<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2447 InstrItinClass itin, string OpcodeStr, string Dt,
2448 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
2449 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2450 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2451 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2452 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
2453 // All of these have a two-operand InstAlias.
2454 let TwoOperandAliasConstraint = "$Vn = $Vd";
2455 let isCommutable = Commutable;
2457 // Same as N3VD but no data type.
2458 class N3VDX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2459 InstrItinClass itin, string OpcodeStr,
2460 ValueType ResTy, ValueType OpTy,
2461 SDNode OpNode, bit Commutable>
2462 : N3VX<op24, op23, op21_20, op11_8, 0, op4,
2463 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2464 OpcodeStr, "$Vd, $Vn, $Vm", "",
2465 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>{
2466 // All of these have a two-operand InstAlias.
2467 let TwoOperandAliasConstraint = "$Vn = $Vd";
2468 let isCommutable = Commutable;
2471 class N3VDSL<bits<2> op21_20, bits<4> op11_8,
2472 InstrItinClass itin, string OpcodeStr, string Dt,
2473 ValueType Ty, SDNode ShOp>
2474 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2475 (outs DPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2476 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2478 (Ty (ShOp (Ty DPR:$Vn),
2479 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),imm:$lane)))))]> {
2480 // All of these have a two-operand InstAlias.
2481 let TwoOperandAliasConstraint = "$Vn = $Vd";
2482 let isCommutable = 0;
2484 class N3VDSL16<bits<2> op21_20, bits<4> op11_8,
2485 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
2486 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2487 (outs DPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2488 NVMulSLFrm, IIC_VMULi16D, OpcodeStr, Dt,"$Vd, $Vn, $Vm$lane","",
2490 (Ty (ShOp (Ty DPR:$Vn),
2491 (Ty (NEONvduplane (Ty DPR_8:$Vm), imm:$lane)))))]> {
2492 // All of these have a two-operand InstAlias.
2493 let TwoOperandAliasConstraint = "$Vn = $Vd";
2494 let isCommutable = 0;
2497 class N3VQ<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2498 InstrItinClass itin, string OpcodeStr, string Dt,
2499 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
2500 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2501 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2502 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2503 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
2504 // All of these have a two-operand InstAlias.
2505 let TwoOperandAliasConstraint = "$Vn = $Vd";
2506 let isCommutable = Commutable;
2508 class N3VQX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2509 InstrItinClass itin, string OpcodeStr,
2510 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
2511 : N3VX<op24, op23, op21_20, op11_8, 1, op4,
2512 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2513 OpcodeStr, "$Vd, $Vn, $Vm", "",
2514 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]>{
2515 // All of these have a two-operand InstAlias.
2516 let TwoOperandAliasConstraint = "$Vn = $Vd";
2517 let isCommutable = Commutable;
2519 class N3VQSL<bits<2> op21_20, bits<4> op11_8,
2520 InstrItinClass itin, string OpcodeStr, string Dt,
2521 ValueType ResTy, ValueType OpTy, SDNode ShOp>
2522 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2523 (outs QPR:$Vd), (ins QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2524 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2525 [(set (ResTy QPR:$Vd),
2526 (ResTy (ShOp (ResTy QPR:$Vn),
2527 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2529 // All of these have a two-operand InstAlias.
2530 let TwoOperandAliasConstraint = "$Vn = $Vd";
2531 let isCommutable = 0;
2533 class N3VQSL16<bits<2> op21_20, bits<4> op11_8, string OpcodeStr, string Dt,
2534 ValueType ResTy, ValueType OpTy, SDNode ShOp>
2535 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2536 (outs QPR:$Vd), (ins QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2537 NVMulSLFrm, IIC_VMULi16Q, OpcodeStr, Dt,"$Vd, $Vn, $Vm$lane", "",
2538 [(set (ResTy QPR:$Vd),
2539 (ResTy (ShOp (ResTy QPR:$Vn),
2540 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2542 // All of these have a two-operand InstAlias.
2543 let TwoOperandAliasConstraint = "$Vn = $Vd";
2544 let isCommutable = 0;
2547 // Basic 3-register intrinsics, both double- and quad-register.
2548 class N3VDInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2549 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2550 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp, bit Commutable>
2551 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2552 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), f, itin,
2553 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2554 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
2555 // All of these have a two-operand InstAlias.
2556 let TwoOperandAliasConstraint = "$Vn = $Vd";
2557 let isCommutable = Commutable;
2560 class N3VDIntnp<bits<5> op27_23, bits<2> op21_20, bits<4> op11_8, bit op6,
2561 bit op4, Format f, InstrItinClass itin, string OpcodeStr,
2562 string Dt, ValueType ResTy, ValueType OpTy,
2563 SDPatternOperator IntOp, bit Commutable>
2564 : N3Vnp<op27_23, op21_20, op11_8, op6, op4,
2565 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), f, itin, OpcodeStr, Dt,
2566 ResTy, OpTy, IntOp, Commutable,
2567 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>;
2569 class N3VDIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2570 string OpcodeStr, string Dt, ValueType Ty, SDPatternOperator IntOp>
2571 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2572 (outs DPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2573 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2575 (Ty (IntOp (Ty DPR:$Vn),
2576 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),
2578 let isCommutable = 0;
2581 class N3VDIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2582 string OpcodeStr, string Dt, ValueType Ty, SDPatternOperator IntOp>
2583 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2584 (outs DPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2585 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2587 (Ty (IntOp (Ty DPR:$Vn),
2588 (Ty (NEONvduplane (Ty DPR_8:$Vm), imm:$lane)))))]> {
2589 let isCommutable = 0;
2591 class N3VDIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2592 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2593 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2594 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2595 (outs DPR:$Vd), (ins DPR:$Vm, DPR:$Vn), f, itin,
2596 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
2597 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (OpTy DPR:$Vn))))]> {
2598 let TwoOperandAliasConstraint = "$Vm = $Vd";
2599 let isCommutable = 0;
2602 class N3VQInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2603 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2604 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp, bit Commutable>
2605 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2606 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), f, itin,
2607 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2608 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
2609 // All of these have a two-operand InstAlias.
2610 let TwoOperandAliasConstraint = "$Vn = $Vd";
2611 let isCommutable = Commutable;
2614 class N3VQIntnp<bits<5> op27_23, bits<2> op21_20, bits<4> op11_8, bit op6,
2615 bit op4, Format f, InstrItinClass itin, string OpcodeStr,
2616 string Dt, ValueType ResTy, ValueType OpTy,
2617 SDPatternOperator IntOp, bit Commutable>
2618 : N3Vnp<op27_23, op21_20, op11_8, op6, op4,
2619 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), f, itin, OpcodeStr, Dt,
2620 ResTy, OpTy, IntOp, Commutable,
2621 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]>;
2623 class N3VQIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2624 string OpcodeStr, string Dt,
2625 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2626 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2627 (outs QPR:$Vd), (ins QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2628 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2629 [(set (ResTy QPR:$Vd),
2630 (ResTy (IntOp (ResTy QPR:$Vn),
2631 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2633 let isCommutable = 0;
2635 class N3VQIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2636 string OpcodeStr, string Dt,
2637 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2638 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2639 (outs QPR:$Vd), (ins QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2640 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2641 [(set (ResTy QPR:$Vd),
2642 (ResTy (IntOp (ResTy QPR:$Vn),
2643 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2645 let isCommutable = 0;
2647 class N3VQIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2648 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2649 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2650 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2651 (outs QPR:$Vd), (ins QPR:$Vm, QPR:$Vn), f, itin,
2652 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
2653 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (OpTy QPR:$Vn))))]> {
2654 let TwoOperandAliasConstraint = "$Vm = $Vd";
2655 let isCommutable = 0;
2658 // Multiply-Add/Sub operations: double- and quad-register.
2659 class N3VDMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2660 InstrItinClass itin, string OpcodeStr, string Dt,
2661 ValueType Ty, SDPatternOperator MulOp, SDPatternOperator OpNode>
2662 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2663 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2664 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2665 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
2666 (Ty (MulOp DPR:$Vn, DPR:$Vm)))))]>;
2668 class N3VDMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2669 string OpcodeStr, string Dt,
2670 ValueType Ty, SDPatternOperator MulOp, SDPatternOperator ShOp>
2671 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2673 (ins DPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2675 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2677 (Ty (ShOp (Ty DPR:$src1),
2679 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),
2681 class N3VDMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2682 string OpcodeStr, string Dt,
2683 ValueType Ty, SDNode MulOp, SDNode ShOp>
2684 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2686 (ins DPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2688 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2690 (Ty (ShOp (Ty DPR:$src1),
2692 (Ty (NEONvduplane (Ty DPR_8:$Vm),
2695 class N3VQMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2696 InstrItinClass itin, string OpcodeStr, string Dt, ValueType Ty,
2697 SDPatternOperator MulOp, SDPatternOperator OpNode>
2698 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2699 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2700 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2701 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
2702 (Ty (MulOp QPR:$Vn, QPR:$Vm)))))]>;
2703 class N3VQMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2704 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2705 SDPatternOperator MulOp, SDPatternOperator ShOp>
2706 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2708 (ins QPR:$src1, QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2710 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2711 [(set (ResTy QPR:$Vd),
2712 (ResTy (ShOp (ResTy QPR:$src1),
2713 (ResTy (MulOp QPR:$Vn,
2714 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2716 class N3VQMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2717 string OpcodeStr, string Dt,
2718 ValueType ResTy, ValueType OpTy,
2719 SDNode MulOp, SDNode ShOp>
2720 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2722 (ins QPR:$src1, QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2724 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2725 [(set (ResTy QPR:$Vd),
2726 (ResTy (ShOp (ResTy QPR:$src1),
2727 (ResTy (MulOp QPR:$Vn,
2728 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2731 // Neon Intrinsic-Op instructions (VABA): double- and quad-register.
2732 class N3VDIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2733 InstrItinClass itin, string OpcodeStr, string Dt,
2734 ValueType Ty, SDPatternOperator IntOp, SDNode OpNode>
2735 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2736 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2737 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2738 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
2739 (Ty (IntOp (Ty DPR:$Vn), (Ty DPR:$Vm))))))]>;
2740 class N3VQIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2741 InstrItinClass itin, string OpcodeStr, string Dt,
2742 ValueType Ty, SDPatternOperator IntOp, SDNode OpNode>
2743 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2744 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2745 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2746 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
2747 (Ty (IntOp (Ty QPR:$Vn), (Ty QPR:$Vm))))))]>;
2749 // Neon 3-argument intrinsics, both double- and quad-register.
2750 // The destination register is also used as the first source operand register.
2751 class N3VDInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2752 InstrItinClass itin, string OpcodeStr, string Dt,
2753 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2754 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2755 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2756 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2757 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$src1),
2758 (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>;
2759 class N3VQInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2760 InstrItinClass itin, string OpcodeStr, string Dt,
2761 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2762 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2763 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2764 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2765 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$src1),
2766 (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]>;
2768 // Long Multiply-Add/Sub operations.
2769 class N3VLMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2770 InstrItinClass itin, string OpcodeStr, string Dt,
2771 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2772 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2773 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2774 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2775 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
2776 (TyQ (MulOp (TyD DPR:$Vn),
2777 (TyD DPR:$Vm)))))]>;
2778 class N3VLMulOpSL<bit op24, bits<2> op21_20, bits<4> op11_8,
2779 InstrItinClass itin, string OpcodeStr, string Dt,
2780 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2781 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$Vd),
2782 (ins QPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2784 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2786 (OpNode (TyQ QPR:$src1),
2787 (TyQ (MulOp (TyD DPR:$Vn),
2788 (TyD (NEONvduplane (TyD DPR_VFP2:$Vm),
2790 class N3VLMulOpSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2791 InstrItinClass itin, string OpcodeStr, string Dt,
2792 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2793 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$Vd),
2794 (ins QPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2796 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2798 (OpNode (TyQ QPR:$src1),
2799 (TyQ (MulOp (TyD DPR:$Vn),
2800 (TyD (NEONvduplane (TyD DPR_8:$Vm),
2803 // Long Intrinsic-Op vector operations with explicit extend (VABAL).
2804 class N3VLIntExtOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2805 InstrItinClass itin, string OpcodeStr, string Dt,
2806 ValueType TyQ, ValueType TyD, SDPatternOperator IntOp, SDNode ExtOp,
2808 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2809 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2810 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2811 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
2812 (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
2813 (TyD DPR:$Vm)))))))]>;
2815 // Neon Long 3-argument intrinsic. The destination register is
2816 // a quad-register and is also used as the first source operand register.
2817 class N3VLInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2818 InstrItinClass itin, string OpcodeStr, string Dt,
2819 ValueType TyQ, ValueType TyD, SDPatternOperator IntOp>
2820 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2821 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2822 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2824 (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$Vn), (TyD DPR:$Vm))))]>;
2825 class N3VLInt3SL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2826 string OpcodeStr, string Dt,
2827 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2828 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2830 (ins QPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2832 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2833 [(set (ResTy QPR:$Vd),
2834 (ResTy (IntOp (ResTy QPR:$src1),
2836 (OpTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2838 class N3VLInt3SL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2839 InstrItinClass itin, string OpcodeStr, string Dt,
2840 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2841 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2843 (ins QPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2845 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2846 [(set (ResTy QPR:$Vd),
2847 (ResTy (IntOp (ResTy QPR:$src1),
2849 (OpTy (NEONvduplane (OpTy DPR_8:$Vm),
2852 // Narrowing 3-register intrinsics.
2853 class N3VNInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2854 string OpcodeStr, string Dt, ValueType TyD, ValueType TyQ,
2855 SDPatternOperator IntOp, bit Commutable>
2856 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2857 (outs DPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINi4D,
2858 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2859 [(set DPR:$Vd, (TyD (IntOp (TyQ QPR:$Vn), (TyQ QPR:$Vm))))]> {
2860 let isCommutable = Commutable;
2863 // Long 3-register operations.
2864 class N3VL<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2865 InstrItinClass itin, string OpcodeStr, string Dt,
2866 ValueType TyQ, ValueType TyD, SDNode OpNode, bit Commutable>
2867 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2868 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2869 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2870 [(set QPR:$Vd, (TyQ (OpNode (TyD DPR:$Vn), (TyD DPR:$Vm))))]> {
2871 let isCommutable = Commutable;
2873 class N3VLSL<bit op24, bits<2> op21_20, bits<4> op11_8,
2874 InstrItinClass itin, string OpcodeStr, string Dt,
2875 ValueType TyQ, ValueType TyD, SDNode OpNode>
2876 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2877 (outs QPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2878 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2880 (TyQ (OpNode (TyD DPR:$Vn),
2881 (TyD (NEONvduplane (TyD DPR_VFP2:$Vm),imm:$lane)))))]>;
2882 class N3VLSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2883 InstrItinClass itin, string OpcodeStr, string Dt,
2884 ValueType TyQ, ValueType TyD, SDNode OpNode>
2885 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2886 (outs QPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2887 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2889 (TyQ (OpNode (TyD DPR:$Vn),
2890 (TyD (NEONvduplane (TyD DPR_8:$Vm), imm:$lane)))))]>;
2892 // Long 3-register operations with explicitly extended operands.
2893 class N3VLExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2894 InstrItinClass itin, string OpcodeStr, string Dt,
2895 ValueType TyQ, ValueType TyD, SDNode OpNode, SDNode ExtOp,
2897 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2898 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2899 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2900 [(set QPR:$Vd, (OpNode (TyQ (ExtOp (TyD DPR:$Vn))),
2901 (TyQ (ExtOp (TyD DPR:$Vm)))))]> {
2902 let isCommutable = Commutable;
2905 // Long 3-register intrinsics with explicit extend (VABDL).
2906 class N3VLIntExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2907 InstrItinClass itin, string OpcodeStr, string Dt,
2908 ValueType TyQ, ValueType TyD, SDPatternOperator IntOp, SDNode ExtOp,
2910 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2911 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2912 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2913 [(set QPR:$Vd, (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
2914 (TyD DPR:$Vm))))))]> {
2915 let isCommutable = Commutable;
2918 // Long 3-register intrinsics.
2919 class N3VLInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2920 InstrItinClass itin, string OpcodeStr, string Dt,
2921 ValueType TyQ, ValueType TyD, SDPatternOperator IntOp, bit Commutable>
2922 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2923 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2924 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2925 [(set QPR:$Vd, (TyQ (IntOp (TyD DPR:$Vn), (TyD DPR:$Vm))))]> {
2926 let isCommutable = Commutable;
2928 class N3VLIntSL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2929 string OpcodeStr, string Dt,
2930 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2931 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2932 (outs QPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2933 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2934 [(set (ResTy QPR:$Vd),
2935 (ResTy (IntOp (OpTy DPR:$Vn),
2936 (OpTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2938 class N3VLIntSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2939 InstrItinClass itin, string OpcodeStr, string Dt,
2940 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2941 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2942 (outs QPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2943 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2944 [(set (ResTy QPR:$Vd),
2945 (ResTy (IntOp (OpTy DPR:$Vn),
2946 (OpTy (NEONvduplane (OpTy DPR_8:$Vm),
2949 // Wide 3-register operations.
2950 class N3VW<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2951 string OpcodeStr, string Dt, ValueType TyQ, ValueType TyD,
2952 SDNode OpNode, SDNode ExtOp, bit Commutable>
2953 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2954 (outs QPR:$Vd), (ins QPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VSUBiD,
2955 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2956 [(set QPR:$Vd, (OpNode (TyQ QPR:$Vn),
2957 (TyQ (ExtOp (TyD DPR:$Vm)))))]> {
2958 // All of these have a two-operand InstAlias.
2959 let TwoOperandAliasConstraint = "$Vn = $Vd";
2960 let isCommutable = Commutable;
2963 // Pairwise long 2-register intrinsics, both double- and quad-register.
2964 class N2VDPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2965 bits<2> op17_16, bits<5> op11_7, bit op4,
2966 string OpcodeStr, string Dt,
2967 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2968 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
2969 (ins DPR:$Vm), IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
2970 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm))))]>;
2971 class N2VQPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2972 bits<2> op17_16, bits<5> op11_7, bit op4,
2973 string OpcodeStr, string Dt,
2974 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2975 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
2976 (ins QPR:$Vm), IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
2977 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm))))]>;
2979 // Pairwise long 2-register accumulate intrinsics,
2980 // both double- and quad-register.
2981 // The destination register is also used as the first source operand register.
2982 class N2VDPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2983 bits<2> op17_16, bits<5> op11_7, bit op4,
2984 string OpcodeStr, string Dt,
2985 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2986 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
2987 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vm), IIC_VPALiD,
2988 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
2989 [(set DPR:$Vd, (ResTy (IntOp (ResTy DPR:$src1), (OpTy DPR:$Vm))))]>;
2990 class N2VQPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2991 bits<2> op17_16, bits<5> op11_7, bit op4,
2992 string OpcodeStr, string Dt,
2993 ValueType ResTy, ValueType OpTy, SDPatternOperator IntOp>
2994 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4,
2995 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vm), IIC_VPALiQ,
2996 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
2997 [(set QPR:$Vd, (ResTy (IntOp (ResTy QPR:$src1), (OpTy QPR:$Vm))))]>;
2999 // Shift by immediate,
3000 // both double- and quad-register.
3001 let TwoOperandAliasConstraint = "$Vm = $Vd" in {
3002 class N2VDSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
3003 Format f, InstrItinClass itin, Operand ImmTy,
3004 string OpcodeStr, string Dt, ValueType Ty, SDNode OpNode>
3005 : N2VImm<op24, op23, op11_8, op7, 0, op4,
3006 (outs DPR:$Vd), (ins DPR:$Vm, ImmTy:$SIMM), f, itin,
3007 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
3008 [(set DPR:$Vd, (Ty (OpNode (Ty DPR:$Vm), (i32 imm:$SIMM))))]>;
3009 class N2VQSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
3010 Format f, InstrItinClass itin, Operand ImmTy,
3011 string OpcodeStr, string Dt, ValueType Ty, SDNode OpNode>
3012 : N2VImm<op24, op23, op11_8, op7, 1, op4,
3013 (outs QPR:$Vd), (ins QPR:$Vm, ImmTy:$SIMM), f, itin,
3014 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
3015 [(set QPR:$Vd, (Ty (OpNode (Ty QPR:$Vm), (i32 imm:$SIMM))))]>;
3018 // Long shift by immediate.
3019 class N2VLSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
3020 string OpcodeStr, string Dt,
3021 ValueType ResTy, ValueType OpTy, Operand ImmTy, SDNode OpNode>
3022 : N2VImm<op24, op23, op11_8, op7, op6, op4,
3023 (outs QPR:$Vd), (ins DPR:$Vm, ImmTy:$SIMM), N2RegVShLFrm,
3024 IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
3025 [(set QPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vm),
3026 (i32 imm:$SIMM))))]>;
3028 // Narrow shift by immediate.
3029 class N2VNSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
3030 InstrItinClass itin, string OpcodeStr, string Dt,
3031 ValueType ResTy, ValueType OpTy, Operand ImmTy, SDNode OpNode>
3032 : N2VImm<op24, op23, op11_8, op7, op6, op4,
3033 (outs DPR:$Vd), (ins QPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, itin,
3034 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
3035 [(set DPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vm),
3036 (i32 imm:$SIMM))))]>;
3038 // Shift right by immediate and accumulate,
3039 // both double- and quad-register.
3040 let TwoOperandAliasConstraint = "$Vm = $Vd" in {
3041 class N2VDShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
3042 Operand ImmTy, string OpcodeStr, string Dt,
3043 ValueType Ty, SDNode ShOp>
3044 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
3045 (ins DPR:$src1, DPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, IIC_VPALiD,
3046 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
3047 [(set DPR:$Vd, (Ty (add DPR:$src1,
3048 (Ty (ShOp DPR:$Vm, (i32 imm:$SIMM))))))]>;
3049 class N2VQShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
3050 Operand ImmTy, string OpcodeStr, string Dt,
3051 ValueType Ty, SDNode ShOp>
3052 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
3053 (ins QPR:$src1, QPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, IIC_VPALiD,
3054 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
3055 [(set QPR:$Vd, (Ty (add QPR:$src1,
3056 (Ty (ShOp QPR:$Vm, (i32 imm:$SIMM))))))]>;
3059 // Shift by immediate and insert,
3060 // both double- and quad-register.
3061 let TwoOperandAliasConstraint = "$Vm = $Vd" in {
3062 class N2VDShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
3063 Operand ImmTy, Format f, string OpcodeStr, string Dt,
3064 ValueType Ty,SDNode ShOp>
3065 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
3066 (ins DPR:$src1, DPR:$Vm, ImmTy:$SIMM), f, IIC_VSHLiD,
3067 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
3068 [(set DPR:$Vd, (Ty (ShOp DPR:$src1, DPR:$Vm, (i32 imm:$SIMM))))]>;
3069 class N2VQShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
3070 Operand ImmTy, Format f, string OpcodeStr, string Dt,
3071 ValueType Ty,SDNode ShOp>
3072 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
3073 (ins QPR:$src1, QPR:$Vm, ImmTy:$SIMM), f, IIC_VSHLiQ,
3074 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
3075 [(set QPR:$Vd, (Ty (ShOp QPR:$src1, QPR:$Vm, (i32 imm:$SIMM))))]>;
3078 // Convert, with fractional bits immediate,
3079 // both double- and quad-register.
3080 class N2VCvtD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
3081 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
3082 SDPatternOperator IntOp>
3083 : N2VImm<op24, op23, op11_8, op7, 0, op4,
3084 (outs DPR:$Vd), (ins DPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
3085 IIC_VUNAD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
3086 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (i32 imm:$SIMM))))]>;
3087 class N2VCvtQ<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
3088 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
3089 SDPatternOperator IntOp>
3090 : N2VImm<op24, op23, op11_8, op7, 1, op4,
3091 (outs QPR:$Vd), (ins QPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
3092 IIC_VUNAQ, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
3093 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (i32 imm:$SIMM))))]>;
3095 //===----------------------------------------------------------------------===//
3097 //===----------------------------------------------------------------------===//
3099 // Abbreviations used in multiclass suffixes:
3100 // Q = quarter int (8 bit) elements
3101 // H = half int (16 bit) elements
3102 // S = single int (32 bit) elements
3103 // D = double int (64 bit) elements
3105 // Neon 2-register vector operations and intrinsics.
3107 // Neon 2-register comparisons.
3108 // source operand element sizes of 8, 16 and 32 bits:
3109 multiclass N2V_QHS_cmp<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3110 bits<5> op11_7, bit op4, string opc, string Dt,
3111 string asm, SDNode OpNode> {
3112 // 64-bit vector types.
3113 def v8i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 0, op4,
3114 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
3115 opc, !strconcat(Dt, "8"), asm, "",
3116 [(set DPR:$Vd, (v8i8 (OpNode (v8i8 DPR:$Vm))))]>;
3117 def v4i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 0, op4,
3118 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
3119 opc, !strconcat(Dt, "16"), asm, "",
3120 [(set DPR:$Vd, (v4i16 (OpNode (v4i16 DPR:$Vm))))]>;
3121 def v2i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
3122 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
3123 opc, !strconcat(Dt, "32"), asm, "",
3124 [(set DPR:$Vd, (v2i32 (OpNode (v2i32 DPR:$Vm))))]>;
3125 def v2f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
3126 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
3127 opc, "f32", asm, "",
3128 [(set DPR:$Vd, (v2i32 (OpNode (v2f32 DPR:$Vm))))]> {
3129 let Inst{10} = 1; // overwrite F = 1
3132 // 128-bit vector types.
3133 def v16i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 1, op4,
3134 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
3135 opc, !strconcat(Dt, "8"), asm, "",
3136 [(set QPR:$Vd, (v16i8 (OpNode (v16i8 QPR:$Vm))))]>;
3137 def v8i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 1, op4,
3138 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
3139 opc, !strconcat(Dt, "16"), asm, "",
3140 [(set QPR:$Vd, (v8i16 (OpNode (v8i16 QPR:$Vm))))]>;
3141 def v4i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
3142 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
3143 opc, !strconcat(Dt, "32"), asm, "",
3144 [(set QPR:$Vd, (v4i32 (OpNode (v4i32 QPR:$Vm))))]>;
3145 def v4f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
3146 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
3147 opc, "f32", asm, "",
3148 [(set QPR:$Vd, (v4i32 (OpNode (v4f32 QPR:$Vm))))]> {
3149 let Inst{10} = 1; // overwrite F = 1
3154 // Neon 2-register vector intrinsics,
3155 // element sizes of 8, 16 and 32 bits:
3156 multiclass N2VInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3157 bits<5> op11_7, bit op4,
3158 InstrItinClass itinD, InstrItinClass itinQ,
3159 string OpcodeStr, string Dt, SDPatternOperator IntOp> {
3160 // 64-bit vector types.
3161 def v8i8 : N2VDInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3162 itinD, OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
3163 def v4i16 : N2VDInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3164 itinD, OpcodeStr, !strconcat(Dt, "16"),v4i16,v4i16,IntOp>;
3165 def v2i32 : N2VDInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3166 itinD, OpcodeStr, !strconcat(Dt, "32"),v2i32,v2i32,IntOp>;
3168 // 128-bit vector types.
3169 def v16i8 : N2VQInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3170 itinQ, OpcodeStr, !strconcat(Dt, "8"), v16i8,v16i8,IntOp>;
3171 def v8i16 : N2VQInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3172 itinQ, OpcodeStr, !strconcat(Dt, "16"),v8i16,v8i16,IntOp>;
3173 def v4i32 : N2VQInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3174 itinQ, OpcodeStr, !strconcat(Dt, "32"),v4i32,v4i32,IntOp>;
3178 // Neon Narrowing 2-register vector operations,
3179 // source operand element sizes of 16, 32 and 64 bits:
3180 multiclass N2VN_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3181 bits<5> op11_7, bit op6, bit op4,
3182 InstrItinClass itin, string OpcodeStr, string Dt,
3184 def v8i8 : N2VN<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
3185 itin, OpcodeStr, !strconcat(Dt, "16"),
3186 v8i8, v8i16, OpNode>;
3187 def v4i16 : N2VN<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
3188 itin, OpcodeStr, !strconcat(Dt, "32"),
3189 v4i16, v4i32, OpNode>;
3190 def v2i32 : N2VN<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
3191 itin, OpcodeStr, !strconcat(Dt, "64"),
3192 v2i32, v2i64, OpNode>;
3195 // Neon Narrowing 2-register vector intrinsics,
3196 // source operand element sizes of 16, 32 and 64 bits:
3197 multiclass N2VNInt_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3198 bits<5> op11_7, bit op6, bit op4,
3199 InstrItinClass itin, string OpcodeStr, string Dt,
3200 SDPatternOperator IntOp> {
3201 def v8i8 : N2VNInt<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
3202 itin, OpcodeStr, !strconcat(Dt, "16"),
3203 v8i8, v8i16, IntOp>;
3204 def v4i16 : N2VNInt<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
3205 itin, OpcodeStr, !strconcat(Dt, "32"),
3206 v4i16, v4i32, IntOp>;
3207 def v2i32 : N2VNInt<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
3208 itin, OpcodeStr, !strconcat(Dt, "64"),
3209 v2i32, v2i64, IntOp>;
3213 // Neon Lengthening 2-register vector intrinsic (currently specific to VMOVL).
3214 // source operand element sizes of 16, 32 and 64 bits:
3215 multiclass N2VL_QHS<bits<2> op24_23, bits<5> op11_7, bit op6, bit op4,
3216 string OpcodeStr, string Dt, SDNode OpNode> {
3217 def v8i16 : N2VL<op24_23, 0b00, 0b10, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
3218 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode>;
3219 def v4i32 : N2VL<op24_23, 0b01, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
3220 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
3221 def v2i64 : N2VL<op24_23, 0b10, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
3222 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
3226 // Neon 3-register vector operations.
3228 // First with only element sizes of 8, 16 and 32 bits:
3229 multiclass N3V_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3230 InstrItinClass itinD16, InstrItinClass itinD32,
3231 InstrItinClass itinQ16, InstrItinClass itinQ32,
3232 string OpcodeStr, string Dt,
3233 SDNode OpNode, bit Commutable = 0> {
3234 // 64-bit vector types.
3235 def v8i8 : N3VD<op24, op23, 0b00, op11_8, op4, itinD16,
3236 OpcodeStr, !strconcat(Dt, "8"),
3237 v8i8, v8i8, OpNode, Commutable>;
3238 def v4i16 : N3VD<op24, op23, 0b01, op11_8, op4, itinD16,
3239 OpcodeStr, !strconcat(Dt, "16"),
3240 v4i16, v4i16, OpNode, Commutable>;
3241 def v2i32 : N3VD<op24, op23, 0b10, op11_8, op4, itinD32,
3242 OpcodeStr, !strconcat(Dt, "32"),
3243 v2i32, v2i32, OpNode, Commutable>;
3245 // 128-bit vector types.
3246 def v16i8 : N3VQ<op24, op23, 0b00, op11_8, op4, itinQ16,
3247 OpcodeStr, !strconcat(Dt, "8"),
3248 v16i8, v16i8, OpNode, Commutable>;
3249 def v8i16 : N3VQ<op24, op23, 0b01, op11_8, op4, itinQ16,
3250 OpcodeStr, !strconcat(Dt, "16"),
3251 v8i16, v8i16, OpNode, Commutable>;
3252 def v4i32 : N3VQ<op24, op23, 0b10, op11_8, op4, itinQ32,
3253 OpcodeStr, !strconcat(Dt, "32"),
3254 v4i32, v4i32, OpNode, Commutable>;
3257 multiclass N3VSL_HS<bits<4> op11_8, string OpcodeStr, SDNode ShOp> {
3258 def v4i16 : N3VDSL16<0b01, op11_8, OpcodeStr, "i16", v4i16, ShOp>;
3259 def v2i32 : N3VDSL<0b10, op11_8, IIC_VMULi32D, OpcodeStr, "i32", v2i32, ShOp>;
3260 def v8i16 : N3VQSL16<0b01, op11_8, OpcodeStr, "i16", v8i16, v4i16, ShOp>;
3261 def v4i32 : N3VQSL<0b10, op11_8, IIC_VMULi32Q, OpcodeStr, "i32",
3262 v4i32, v2i32, ShOp>;
3265 // ....then also with element size 64 bits:
3266 multiclass N3V_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3267 InstrItinClass itinD, InstrItinClass itinQ,
3268 string OpcodeStr, string Dt,
3269 SDNode OpNode, bit Commutable = 0>
3270 : N3V_QHS<op24, op23, op11_8, op4, itinD, itinD, itinQ, itinQ,
3271 OpcodeStr, Dt, OpNode, Commutable> {
3272 def v1i64 : N3VD<op24, op23, 0b11, op11_8, op4, itinD,
3273 OpcodeStr, !strconcat(Dt, "64"),
3274 v1i64, v1i64, OpNode, Commutable>;
3275 def v2i64 : N3VQ<op24, op23, 0b11, op11_8, op4, itinQ,
3276 OpcodeStr, !strconcat(Dt, "64"),
3277 v2i64, v2i64, OpNode, Commutable>;
3281 // Neon 3-register vector intrinsics.
3283 // First with only element sizes of 16 and 32 bits:
3284 multiclass N3VInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3285 InstrItinClass itinD16, InstrItinClass itinD32,
3286 InstrItinClass itinQ16, InstrItinClass itinQ32,
3287 string OpcodeStr, string Dt,
3288 SDPatternOperator IntOp, bit Commutable = 0> {
3289 // 64-bit vector types.
3290 def v4i16 : N3VDInt<op24, op23, 0b01, op11_8, op4, f, itinD16,
3291 OpcodeStr, !strconcat(Dt, "16"),
3292 v4i16, v4i16, IntOp, Commutable>;
3293 def v2i32 : N3VDInt<op24, op23, 0b10, op11_8, op4, f, itinD32,
3294 OpcodeStr, !strconcat(Dt, "32"),
3295 v2i32, v2i32, IntOp, Commutable>;
3297 // 128-bit vector types.
3298 def v8i16 : N3VQInt<op24, op23, 0b01, op11_8, op4, f, itinQ16,
3299 OpcodeStr, !strconcat(Dt, "16"),
3300 v8i16, v8i16, IntOp, Commutable>;
3301 def v4i32 : N3VQInt<op24, op23, 0b10, op11_8, op4, f, itinQ32,
3302 OpcodeStr, !strconcat(Dt, "32"),
3303 v4i32, v4i32, IntOp, Commutable>;
3305 multiclass N3VInt_HSSh<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> {
3310 // 64-bit vector types.
3311 def v4i16 : N3VDIntSh<op24, op23, 0b01, op11_8, op4, f, itinD16,
3312 OpcodeStr, !strconcat(Dt, "16"),
3313 v4i16, v4i16, IntOp>;
3314 def v2i32 : N3VDIntSh<op24, op23, 0b10, op11_8, op4, f, itinD32,
3315 OpcodeStr, !strconcat(Dt, "32"),
3316 v2i32, v2i32, IntOp>;
3318 // 128-bit vector types.
3319 def v8i16 : N3VQIntSh<op24, op23, 0b01, op11_8, op4, f, itinQ16,
3320 OpcodeStr, !strconcat(Dt, "16"),
3321 v8i16, v8i16, IntOp>;
3322 def v4i32 : N3VQIntSh<op24, op23, 0b10, op11_8, op4, f, itinQ32,
3323 OpcodeStr, !strconcat(Dt, "32"),
3324 v4i32, v4i32, IntOp>;
3327 multiclass N3VIntSL_HS<bits<4> op11_8,
3328 InstrItinClass itinD16, InstrItinClass itinD32,
3329 InstrItinClass itinQ16, InstrItinClass itinQ32,
3330 string OpcodeStr, string Dt, SDPatternOperator IntOp> {
3331 def v4i16 : N3VDIntSL16<0b01, op11_8, itinD16,
3332 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp>;
3333 def v2i32 : N3VDIntSL<0b10, op11_8, itinD32,
3334 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp>;
3335 def v8i16 : N3VQIntSL16<0b01, op11_8, itinQ16,
3336 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16, IntOp>;
3337 def v4i32 : N3VQIntSL<0b10, op11_8, itinQ32,
3338 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32, IntOp>;
3341 // ....then also with element size of 8 bits:
3342 multiclass N3VInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3343 InstrItinClass itinD16, InstrItinClass itinD32,
3344 InstrItinClass itinQ16, InstrItinClass itinQ32,
3345 string OpcodeStr, string Dt,
3346 SDPatternOperator IntOp, bit Commutable = 0>
3347 : N3VInt_HS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
3348 OpcodeStr, Dt, IntOp, Commutable> {
3349 def v8i8 : N3VDInt<op24, op23, 0b00, op11_8, op4, f, itinD16,
3350 OpcodeStr, !strconcat(Dt, "8"),
3351 v8i8, v8i8, IntOp, Commutable>;
3352 def v16i8 : N3VQInt<op24, op23, 0b00, op11_8, op4, f, itinQ16,
3353 OpcodeStr, !strconcat(Dt, "8"),
3354 v16i8, v16i8, IntOp, Commutable>;
3356 multiclass N3VInt_QHSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3357 InstrItinClass itinD16, InstrItinClass itinD32,
3358 InstrItinClass itinQ16, InstrItinClass itinQ32,
3359 string OpcodeStr, string Dt,
3360 SDPatternOperator IntOp>
3361 : N3VInt_HSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
3362 OpcodeStr, Dt, IntOp> {
3363 def v8i8 : N3VDIntSh<op24, op23, 0b00, op11_8, op4, f, itinD16,
3364 OpcodeStr, !strconcat(Dt, "8"),
3366 def v16i8 : N3VQIntSh<op24, op23, 0b00, op11_8, op4, f, itinQ16,
3367 OpcodeStr, !strconcat(Dt, "8"),
3368 v16i8, v16i8, IntOp>;
3372 // ....then also with element size of 64 bits:
3373 multiclass N3VInt_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3374 InstrItinClass itinD16, InstrItinClass itinD32,
3375 InstrItinClass itinQ16, InstrItinClass itinQ32,
3376 string OpcodeStr, string Dt,
3377 SDPatternOperator IntOp, bit Commutable = 0>
3378 : N3VInt_QHS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
3379 OpcodeStr, Dt, IntOp, Commutable> {
3380 def v1i64 : N3VDInt<op24, op23, 0b11, op11_8, op4, f, itinD32,
3381 OpcodeStr, !strconcat(Dt, "64"),
3382 v1i64, v1i64, IntOp, Commutable>;
3383 def v2i64 : N3VQInt<op24, op23, 0b11, op11_8, op4, f, itinQ32,
3384 OpcodeStr, !strconcat(Dt, "64"),
3385 v2i64, v2i64, IntOp, Commutable>;
3387 multiclass N3VInt_QHSDSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3388 InstrItinClass itinD16, InstrItinClass itinD32,
3389 InstrItinClass itinQ16, InstrItinClass itinQ32,
3390 string OpcodeStr, string Dt,
3391 SDPatternOperator IntOp>
3392 : N3VInt_QHSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
3393 OpcodeStr, Dt, IntOp> {
3394 def v1i64 : N3VDIntSh<op24, op23, 0b11, op11_8, op4, f, itinD32,
3395 OpcodeStr, !strconcat(Dt, "64"),
3396 v1i64, v1i64, IntOp>;
3397 def v2i64 : N3VQIntSh<op24, op23, 0b11, op11_8, op4, f, itinQ32,
3398 OpcodeStr, !strconcat(Dt, "64"),
3399 v2i64, v2i64, IntOp>;
3402 // Neon Narrowing 3-register vector intrinsics,
3403 // source operand element sizes of 16, 32 and 64 bits:
3404 multiclass N3VNInt_HSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3405 string OpcodeStr, string Dt,
3406 SDPatternOperator IntOp, bit Commutable = 0> {
3407 def v8i8 : N3VNInt<op24, op23, 0b00, op11_8, op4,
3408 OpcodeStr, !strconcat(Dt, "16"),
3409 v8i8, v8i16, IntOp, Commutable>;
3410 def v4i16 : N3VNInt<op24, op23, 0b01, op11_8, op4,
3411 OpcodeStr, !strconcat(Dt, "32"),
3412 v4i16, v4i32, IntOp, Commutable>;
3413 def v2i32 : N3VNInt<op24, op23, 0b10, op11_8, op4,
3414 OpcodeStr, !strconcat(Dt, "64"),
3415 v2i32, v2i64, IntOp, Commutable>;
3419 // Neon Long 3-register vector operations.
3421 multiclass N3VL_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3422 InstrItinClass itin16, InstrItinClass itin32,
3423 string OpcodeStr, string Dt,
3424 SDNode OpNode, bit Commutable = 0> {
3425 def v8i16 : N3VL<op24, op23, 0b00, op11_8, op4, itin16,
3426 OpcodeStr, !strconcat(Dt, "8"),
3427 v8i16, v8i8, OpNode, Commutable>;
3428 def v4i32 : N3VL<op24, op23, 0b01, op11_8, op4, itin16,
3429 OpcodeStr, !strconcat(Dt, "16"),
3430 v4i32, v4i16, OpNode, Commutable>;
3431 def v2i64 : N3VL<op24, op23, 0b10, op11_8, op4, itin32,
3432 OpcodeStr, !strconcat(Dt, "32"),
3433 v2i64, v2i32, OpNode, Commutable>;
3436 multiclass N3VLSL_HS<bit op24, bits<4> op11_8,
3437 InstrItinClass itin, string OpcodeStr, string Dt,
3439 def v4i16 : N3VLSL16<op24, 0b01, op11_8, itin, OpcodeStr,
3440 !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
3441 def v2i32 : N3VLSL<op24, 0b10, op11_8, itin, OpcodeStr,
3442 !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
3445 multiclass N3VLExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3446 InstrItinClass itin16, InstrItinClass itin32,
3447 string OpcodeStr, string Dt,
3448 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
3449 def v8i16 : N3VLExt<op24, op23, 0b00, op11_8, op4, itin16,
3450 OpcodeStr, !strconcat(Dt, "8"),
3451 v8i16, v8i8, OpNode, ExtOp, Commutable>;
3452 def v4i32 : N3VLExt<op24, op23, 0b01, op11_8, op4, itin16,
3453 OpcodeStr, !strconcat(Dt, "16"),
3454 v4i32, v4i16, OpNode, ExtOp, Commutable>;
3455 def v2i64 : N3VLExt<op24, op23, 0b10, op11_8, op4, itin32,
3456 OpcodeStr, !strconcat(Dt, "32"),
3457 v2i64, v2i32, OpNode, ExtOp, Commutable>;
3460 // Neon Long 3-register vector intrinsics.
3462 // First with only element sizes of 16 and 32 bits:
3463 multiclass N3VLInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
3464 InstrItinClass itin16, InstrItinClass itin32,
3465 string OpcodeStr, string Dt,
3466 SDPatternOperator IntOp, bit Commutable = 0> {
3467 def v4i32 : N3VLInt<op24, op23, 0b01, op11_8, op4, itin16,
3468 OpcodeStr, !strconcat(Dt, "16"),
3469 v4i32, v4i16, IntOp, Commutable>;
3470 def v2i64 : N3VLInt<op24, op23, 0b10, op11_8, op4, itin32,
3471 OpcodeStr, !strconcat(Dt, "32"),
3472 v2i64, v2i32, IntOp, Commutable>;
3475 multiclass N3VLIntSL_HS<bit op24, bits<4> op11_8,
3476 InstrItinClass itin, string OpcodeStr, string Dt,
3477 SDPatternOperator IntOp> {
3478 def v4i16 : N3VLIntSL16<op24, 0b01, op11_8, itin,
3479 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
3480 def v2i32 : N3VLIntSL<op24, 0b10, op11_8, itin,
3481 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3484 // ....then also with element size of 8 bits:
3485 multiclass N3VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3486 InstrItinClass itin16, InstrItinClass itin32,
3487 string OpcodeStr, string Dt,
3488 SDPatternOperator IntOp, bit Commutable = 0>
3489 : N3VLInt_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt,
3490 IntOp, Commutable> {
3491 def v8i16 : N3VLInt<op24, op23, 0b00, op11_8, op4, itin16,
3492 OpcodeStr, !strconcat(Dt, "8"),
3493 v8i16, v8i8, IntOp, Commutable>;
3496 // ....with explicit extend (VABDL).
3497 multiclass N3VLIntExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3498 InstrItinClass itin, string OpcodeStr, string Dt,
3499 SDPatternOperator IntOp, SDNode ExtOp, bit Commutable = 0> {
3500 def v8i16 : N3VLIntExt<op24, op23, 0b00, op11_8, op4, itin,
3501 OpcodeStr, !strconcat(Dt, "8"),
3502 v8i16, v8i8, IntOp, ExtOp, Commutable>;
3503 def v4i32 : N3VLIntExt<op24, op23, 0b01, op11_8, op4, itin,
3504 OpcodeStr, !strconcat(Dt, "16"),
3505 v4i32, v4i16, IntOp, ExtOp, Commutable>;
3506 def v2i64 : N3VLIntExt<op24, op23, 0b10, op11_8, op4, itin,
3507 OpcodeStr, !strconcat(Dt, "32"),
3508 v2i64, v2i32, IntOp, ExtOp, Commutable>;
3512 // Neon Wide 3-register vector intrinsics,
3513 // source operand element sizes of 8, 16 and 32 bits:
3514 multiclass N3VW_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3515 string OpcodeStr, string Dt,
3516 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
3517 def v8i16 : N3VW<op24, op23, 0b00, op11_8, op4,
3518 OpcodeStr, !strconcat(Dt, "8"),
3519 v8i16, v8i8, OpNode, ExtOp, Commutable>;
3520 def v4i32 : N3VW<op24, op23, 0b01, op11_8, op4,
3521 OpcodeStr, !strconcat(Dt, "16"),
3522 v4i32, v4i16, OpNode, ExtOp, Commutable>;
3523 def v2i64 : N3VW<op24, op23, 0b10, op11_8, op4,
3524 OpcodeStr, !strconcat(Dt, "32"),
3525 v2i64, v2i32, OpNode, ExtOp, Commutable>;
3529 // Neon Multiply-Op vector operations,
3530 // element sizes of 8, 16 and 32 bits:
3531 multiclass N3VMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3532 InstrItinClass itinD16, InstrItinClass itinD32,
3533 InstrItinClass itinQ16, InstrItinClass itinQ32,
3534 string OpcodeStr, string Dt, SDNode OpNode> {
3535 // 64-bit vector types.
3536 def v8i8 : N3VDMulOp<op24, op23, 0b00, op11_8, op4, itinD16,
3537 OpcodeStr, !strconcat(Dt, "8"), v8i8, mul, OpNode>;
3538 def v4i16 : N3VDMulOp<op24, op23, 0b01, op11_8, op4, itinD16,
3539 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, OpNode>;
3540 def v2i32 : N3VDMulOp<op24, op23, 0b10, op11_8, op4, itinD32,
3541 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, OpNode>;
3543 // 128-bit vector types.
3544 def v16i8 : N3VQMulOp<op24, op23, 0b00, op11_8, op4, itinQ16,
3545 OpcodeStr, !strconcat(Dt, "8"), v16i8, mul, OpNode>;
3546 def v8i16 : N3VQMulOp<op24, op23, 0b01, op11_8, op4, itinQ16,
3547 OpcodeStr, !strconcat(Dt, "16"), v8i16, mul, OpNode>;
3548 def v4i32 : N3VQMulOp<op24, op23, 0b10, op11_8, op4, itinQ32,
3549 OpcodeStr, !strconcat(Dt, "32"), v4i32, mul, OpNode>;
3552 multiclass N3VMulOpSL_HS<bits<4> op11_8,
3553 InstrItinClass itinD16, InstrItinClass itinD32,
3554 InstrItinClass itinQ16, InstrItinClass itinQ32,
3555 string OpcodeStr, string Dt, SDNode ShOp> {
3556 def v4i16 : N3VDMulOpSL16<0b01, op11_8, itinD16,
3557 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, ShOp>;
3558 def v2i32 : N3VDMulOpSL<0b10, op11_8, itinD32,
3559 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, ShOp>;
3560 def v8i16 : N3VQMulOpSL16<0b01, op11_8, itinQ16,
3561 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16,
3563 def v4i32 : N3VQMulOpSL<0b10, op11_8, itinQ32,
3564 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32,
3568 // Neon Intrinsic-Op vector operations,
3569 // element sizes of 8, 16 and 32 bits:
3570 multiclass N3VIntOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3571 InstrItinClass itinD, InstrItinClass itinQ,
3572 string OpcodeStr, string Dt, SDPatternOperator IntOp,
3574 // 64-bit vector types.
3575 def v8i8 : N3VDIntOp<op24, op23, 0b00, op11_8, op4, itinD,
3576 OpcodeStr, !strconcat(Dt, "8"), v8i8, IntOp, OpNode>;
3577 def v4i16 : N3VDIntOp<op24, op23, 0b01, op11_8, op4, itinD,
3578 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp, OpNode>;
3579 def v2i32 : N3VDIntOp<op24, op23, 0b10, op11_8, op4, itinD,
3580 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp, OpNode>;
3582 // 128-bit vector types.
3583 def v16i8 : N3VQIntOp<op24, op23, 0b00, op11_8, op4, itinQ,
3584 OpcodeStr, !strconcat(Dt, "8"), v16i8, IntOp, OpNode>;
3585 def v8i16 : N3VQIntOp<op24, op23, 0b01, op11_8, op4, itinQ,
3586 OpcodeStr, !strconcat(Dt, "16"), v8i16, IntOp, OpNode>;
3587 def v4i32 : N3VQIntOp<op24, op23, 0b10, op11_8, op4, itinQ,
3588 OpcodeStr, !strconcat(Dt, "32"), v4i32, IntOp, OpNode>;
3591 // Neon 3-argument intrinsics,
3592 // element sizes of 8, 16 and 32 bits:
3593 multiclass N3VInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3594 InstrItinClass itinD, InstrItinClass itinQ,
3595 string OpcodeStr, string Dt, SDPatternOperator IntOp> {
3596 // 64-bit vector types.
3597 def v8i8 : N3VDInt3<op24, op23, 0b00, op11_8, op4, itinD,
3598 OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
3599 def v4i16 : N3VDInt3<op24, op23, 0b01, op11_8, op4, itinD,
3600 OpcodeStr, !strconcat(Dt, "16"), v4i16, v4i16, IntOp>;
3601 def v2i32 : N3VDInt3<op24, op23, 0b10, op11_8, op4, itinD,
3602 OpcodeStr, !strconcat(Dt, "32"), v2i32, v2i32, IntOp>;
3604 // 128-bit vector types.
3605 def v16i8 : N3VQInt3<op24, op23, 0b00, op11_8, op4, itinQ,
3606 OpcodeStr, !strconcat(Dt, "8"), v16i8, v16i8, IntOp>;
3607 def v8i16 : N3VQInt3<op24, op23, 0b01, op11_8, op4, itinQ,
3608 OpcodeStr, !strconcat(Dt, "16"), v8i16, v8i16, IntOp>;
3609 def v4i32 : N3VQInt3<op24, op23, 0b10, op11_8, op4, itinQ,
3610 OpcodeStr, !strconcat(Dt, "32"), v4i32, v4i32, IntOp>;
3614 // Neon Long Multiply-Op vector operations,
3615 // element sizes of 8, 16 and 32 bits:
3616 multiclass N3VLMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3617 InstrItinClass itin16, InstrItinClass itin32,
3618 string OpcodeStr, string Dt, SDNode MulOp,
3620 def v8i16 : N3VLMulOp<op24, op23, 0b00, op11_8, op4, itin16, OpcodeStr,
3621 !strconcat(Dt, "8"), v8i16, v8i8, MulOp, OpNode>;
3622 def v4i32 : N3VLMulOp<op24, op23, 0b01, op11_8, op4, itin16, OpcodeStr,
3623 !strconcat(Dt, "16"), v4i32, v4i16, MulOp, OpNode>;
3624 def v2i64 : N3VLMulOp<op24, op23, 0b10, op11_8, op4, itin32, OpcodeStr,
3625 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
3628 multiclass N3VLMulOpSL_HS<bit op24, bits<4> op11_8, string OpcodeStr,
3629 string Dt, SDNode MulOp, SDNode OpNode> {
3630 def v4i16 : N3VLMulOpSL16<op24, 0b01, op11_8, IIC_VMACi16D, OpcodeStr,
3631 !strconcat(Dt,"16"), v4i32, v4i16, MulOp, OpNode>;
3632 def v2i32 : N3VLMulOpSL<op24, 0b10, op11_8, IIC_VMACi32D, OpcodeStr,
3633 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
3637 // Neon Long 3-argument intrinsics.
3639 // First with only element sizes of 16 and 32 bits:
3640 multiclass N3VLInt3_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
3641 InstrItinClass itin16, InstrItinClass itin32,
3642 string OpcodeStr, string Dt, SDPatternOperator IntOp> {
3643 def v4i32 : N3VLInt3<op24, op23, 0b01, op11_8, op4, itin16,
3644 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
3645 def v2i64 : N3VLInt3<op24, op23, 0b10, op11_8, op4, itin32,
3646 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3649 multiclass N3VLInt3SL_HS<bit op24, bits<4> op11_8,
3650 string OpcodeStr, string Dt, SDPatternOperator IntOp> {
3651 def v4i16 : N3VLInt3SL16<op24, 0b01, op11_8, IIC_VMACi16D,
3652 OpcodeStr, !strconcat(Dt,"16"), v4i32, v4i16, IntOp>;
3653 def v2i32 : N3VLInt3SL<op24, 0b10, op11_8, IIC_VMACi32D,
3654 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3657 // ....then also with element size of 8 bits:
3658 multiclass N3VLInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3659 InstrItinClass itin16, InstrItinClass itin32,
3660 string OpcodeStr, string Dt, SDPatternOperator IntOp>
3661 : N3VLInt3_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt, IntOp> {
3662 def v8i16 : N3VLInt3<op24, op23, 0b00, op11_8, op4, itin16,
3663 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
3666 // ....with explicit extend (VABAL).
3667 multiclass N3VLIntExtOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3668 InstrItinClass itin, string OpcodeStr, string Dt,
3669 SDPatternOperator IntOp, SDNode ExtOp, SDNode OpNode> {
3670 def v8i16 : N3VLIntExtOp<op24, op23, 0b00, op11_8, op4, itin,
3671 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8,
3672 IntOp, ExtOp, OpNode>;
3673 def v4i32 : N3VLIntExtOp<op24, op23, 0b01, op11_8, op4, itin,
3674 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16,
3675 IntOp, ExtOp, OpNode>;
3676 def v2i64 : N3VLIntExtOp<op24, op23, 0b10, op11_8, op4, itin,
3677 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32,
3678 IntOp, ExtOp, OpNode>;
3682 // Neon Pairwise long 2-register intrinsics,
3683 // element sizes of 8, 16 and 32 bits:
3684 multiclass N2VPLInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3685 bits<5> op11_7, bit op4,
3686 string OpcodeStr, string Dt, SDPatternOperator IntOp> {
3687 // 64-bit vector types.
3688 def v8i8 : N2VDPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3689 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
3690 def v4i16 : N2VDPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3691 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
3692 def v2i32 : N2VDPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3693 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
3695 // 128-bit vector types.
3696 def v16i8 : N2VQPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3697 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
3698 def v8i16 : N2VQPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3699 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
3700 def v4i32 : N2VQPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3701 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
3705 // Neon Pairwise long 2-register accumulate intrinsics,
3706 // element sizes of 8, 16 and 32 bits:
3707 multiclass N2VPLInt2_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3708 bits<5> op11_7, bit op4,
3709 string OpcodeStr, string Dt, SDPatternOperator IntOp> {
3710 // 64-bit vector types.
3711 def v8i8 : N2VDPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3712 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
3713 def v4i16 : N2VDPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3714 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
3715 def v2i32 : N2VDPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3716 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
3718 // 128-bit vector types.
3719 def v16i8 : N2VQPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3720 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
3721 def v8i16 : N2VQPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3722 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
3723 def v4i32 : N2VQPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3724 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
3728 // Neon 2-register vector shift by immediate,
3729 // with f of either N2RegVShLFrm or N2RegVShRFrm
3730 // element sizes of 8, 16, 32 and 64 bits:
3731 multiclass N2VShL_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3732 InstrItinClass itin, string OpcodeStr, string Dt,
3734 // 64-bit vector types.
3735 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3736 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
3737 let Inst{21-19} = 0b001; // imm6 = 001xxx
3739 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3740 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
3741 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3743 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3744 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
3745 let Inst{21} = 0b1; // imm6 = 1xxxxx
3747 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, N2RegVShLFrm, itin, i32imm,
3748 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
3751 // 128-bit vector types.
3752 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3753 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
3754 let Inst{21-19} = 0b001; // imm6 = 001xxx
3756 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3757 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
3758 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3760 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3761 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
3762 let Inst{21} = 0b1; // imm6 = 1xxxxx
3764 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, N2RegVShLFrm, itin, i32imm,
3765 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
3768 multiclass N2VShR_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3769 InstrItinClass itin, string OpcodeStr, string Dt,
3770 string baseOpc, SDNode OpNode> {
3771 // 64-bit vector types.
3772 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm8,
3773 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
3774 let Inst{21-19} = 0b001; // imm6 = 001xxx
3776 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm16,
3777 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
3778 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3780 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm32,
3781 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
3782 let Inst{21} = 0b1; // imm6 = 1xxxxx
3784 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, N2RegVShRFrm, itin, shr_imm64,
3785 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
3788 // 128-bit vector types.
3789 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm8,
3790 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
3791 let Inst{21-19} = 0b001; // imm6 = 001xxx
3793 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm16,
3794 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
3795 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3797 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm32,
3798 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
3799 let Inst{21} = 0b1; // imm6 = 1xxxxx
3801 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, N2RegVShRFrm, itin, shr_imm64,
3802 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
3806 // Neon Shift-Accumulate vector operations,
3807 // element sizes of 8, 16, 32 and 64 bits:
3808 multiclass N2VShAdd_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3809 string OpcodeStr, string Dt, SDNode ShOp> {
3810 // 64-bit vector types.
3811 def v8i8 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm8,
3812 OpcodeStr, !strconcat(Dt, "8"), v8i8, ShOp> {
3813 let Inst{21-19} = 0b001; // imm6 = 001xxx
3815 def v4i16 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm16,
3816 OpcodeStr, !strconcat(Dt, "16"), v4i16, ShOp> {
3817 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3819 def v2i32 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm32,
3820 OpcodeStr, !strconcat(Dt, "32"), v2i32, ShOp> {
3821 let Inst{21} = 0b1; // imm6 = 1xxxxx
3823 def v1i64 : N2VDShAdd<op24, op23, op11_8, 1, op4, shr_imm64,
3824 OpcodeStr, !strconcat(Dt, "64"), v1i64, ShOp>;
3827 // 128-bit vector types.
3828 def v16i8 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm8,
3829 OpcodeStr, !strconcat(Dt, "8"), v16i8, ShOp> {
3830 let Inst{21-19} = 0b001; // imm6 = 001xxx
3832 def v8i16 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm16,
3833 OpcodeStr, !strconcat(Dt, "16"), v8i16, ShOp> {
3834 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3836 def v4i32 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm32,
3837 OpcodeStr, !strconcat(Dt, "32"), v4i32, ShOp> {
3838 let Inst{21} = 0b1; // imm6 = 1xxxxx
3840 def v2i64 : N2VQShAdd<op24, op23, op11_8, 1, op4, shr_imm64,
3841 OpcodeStr, !strconcat(Dt, "64"), v2i64, ShOp>;
3845 // Neon Shift-Insert vector operations,
3846 // with f of either N2RegVShLFrm or N2RegVShRFrm
3847 // element sizes of 8, 16, 32 and 64 bits:
3848 multiclass N2VShInsL_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3850 // 64-bit vector types.
3851 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3852 N2RegVShLFrm, OpcodeStr, "8", v8i8, NEONvsli> {
3853 let Inst{21-19} = 0b001; // imm6 = 001xxx
3855 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3856 N2RegVShLFrm, OpcodeStr, "16", v4i16, NEONvsli> {
3857 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3859 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3860 N2RegVShLFrm, OpcodeStr, "32", v2i32, NEONvsli> {
3861 let Inst{21} = 0b1; // imm6 = 1xxxxx
3863 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4, i32imm,
3864 N2RegVShLFrm, OpcodeStr, "64", v1i64, NEONvsli>;
3867 // 128-bit vector types.
3868 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3869 N2RegVShLFrm, OpcodeStr, "8", v16i8, NEONvsli> {
3870 let Inst{21-19} = 0b001; // imm6 = 001xxx
3872 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3873 N2RegVShLFrm, OpcodeStr, "16", v8i16, NEONvsli> {
3874 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3876 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3877 N2RegVShLFrm, OpcodeStr, "32", v4i32, NEONvsli> {
3878 let Inst{21} = 0b1; // imm6 = 1xxxxx
3880 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4, i32imm,
3881 N2RegVShLFrm, OpcodeStr, "64", v2i64, NEONvsli>;
3884 multiclass N2VShInsR_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3886 // 64-bit vector types.
3887 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm8,
3888 N2RegVShRFrm, OpcodeStr, "8", v8i8, NEONvsri> {
3889 let Inst{21-19} = 0b001; // imm6 = 001xxx
3891 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm16,
3892 N2RegVShRFrm, OpcodeStr, "16", v4i16, NEONvsri> {
3893 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3895 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm32,
3896 N2RegVShRFrm, OpcodeStr, "32", v2i32, NEONvsri> {
3897 let Inst{21} = 0b1; // imm6 = 1xxxxx
3899 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4, shr_imm64,
3900 N2RegVShRFrm, OpcodeStr, "64", v1i64, NEONvsri>;
3903 // 128-bit vector types.
3904 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm8,
3905 N2RegVShRFrm, OpcodeStr, "8", v16i8, NEONvsri> {
3906 let Inst{21-19} = 0b001; // imm6 = 001xxx
3908 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm16,
3909 N2RegVShRFrm, OpcodeStr, "16", v8i16, NEONvsri> {
3910 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3912 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm32,
3913 N2RegVShRFrm, OpcodeStr, "32", v4i32, NEONvsri> {
3914 let Inst{21} = 0b1; // imm6 = 1xxxxx
3916 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4, shr_imm64,
3917 N2RegVShRFrm, OpcodeStr, "64", v2i64, NEONvsri>;
3921 // Neon Shift Long operations,
3922 // element sizes of 8, 16, 32 bits:
3923 multiclass N2VLSh_QHS<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
3924 bit op4, string OpcodeStr, string Dt, SDNode OpNode> {
3925 def v8i16 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3926 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, imm1_7, OpNode> {
3927 let Inst{21-19} = 0b001; // imm6 = 001xxx
3929 def v4i32 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3930 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, imm1_15, OpNode> {
3931 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3933 def v2i64 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3934 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, imm1_31, OpNode> {
3935 let Inst{21} = 0b1; // imm6 = 1xxxxx
3939 // Neon Shift Narrow operations,
3940 // element sizes of 16, 32, 64 bits:
3941 multiclass N2VNSh_HSD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
3942 bit op4, InstrItinClass itin, string OpcodeStr, string Dt,
3944 def v8i8 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3945 OpcodeStr, !strconcat(Dt, "16"),
3946 v8i8, v8i16, shr_imm8, OpNode> {
3947 let Inst{21-19} = 0b001; // imm6 = 001xxx
3949 def v4i16 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3950 OpcodeStr, !strconcat(Dt, "32"),
3951 v4i16, v4i32, shr_imm16, OpNode> {
3952 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3954 def v2i32 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3955 OpcodeStr, !strconcat(Dt, "64"),
3956 v2i32, v2i64, shr_imm32, OpNode> {
3957 let Inst{21} = 0b1; // imm6 = 1xxxxx
3961 //===----------------------------------------------------------------------===//
3962 // Instruction Definitions.
3963 //===----------------------------------------------------------------------===//
3965 // Vector Add Operations.
3967 // VADD : Vector Add (integer and floating-point)
3968 defm VADD : N3V_QHSD<0, 0, 0b1000, 0, IIC_VBINiD, IIC_VBINiQ, "vadd", "i",
3970 def VADDfd : N3VD<0, 0, 0b00, 0b1101, 0, IIC_VBIND, "vadd", "f32",
3971 v2f32, v2f32, fadd, 1>;
3972 def VADDfq : N3VQ<0, 0, 0b00, 0b1101, 0, IIC_VBINQ, "vadd", "f32",
3973 v4f32, v4f32, fadd, 1>;
3974 // VADDL : Vector Add Long (Q = D + D)
3975 defm VADDLs : N3VLExt_QHS<0,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
3976 "vaddl", "s", add, sext, 1>;
3977 defm VADDLu : N3VLExt_QHS<1,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
3978 "vaddl", "u", add, zext, 1>;
3979 // VADDW : Vector Add Wide (Q = Q + D)
3980 defm VADDWs : N3VW_QHS<0,1,0b0001,0, "vaddw", "s", add, sext, 0>;
3981 defm VADDWu : N3VW_QHS<1,1,0b0001,0, "vaddw", "u", add, zext, 0>;
3982 // VHADD : Vector Halving Add
3983 defm VHADDs : N3VInt_QHS<0, 0, 0b0000, 0, N3RegFrm,
3984 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3985 "vhadd", "s", int_arm_neon_vhadds, 1>;
3986 defm VHADDu : N3VInt_QHS<1, 0, 0b0000, 0, N3RegFrm,
3987 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3988 "vhadd", "u", int_arm_neon_vhaddu, 1>;
3989 // VRHADD : Vector Rounding Halving Add
3990 defm VRHADDs : N3VInt_QHS<0, 0, 0b0001, 0, N3RegFrm,
3991 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3992 "vrhadd", "s", int_arm_neon_vrhadds, 1>;
3993 defm VRHADDu : N3VInt_QHS<1, 0, 0b0001, 0, N3RegFrm,
3994 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3995 "vrhadd", "u", int_arm_neon_vrhaddu, 1>;
3996 // VQADD : Vector Saturating Add
3997 defm VQADDs : N3VInt_QHSD<0, 0, 0b0000, 1, N3RegFrm,
3998 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3999 "vqadd", "s", int_arm_neon_vqadds, 1>;
4000 defm VQADDu : N3VInt_QHSD<1, 0, 0b0000, 1, N3RegFrm,
4001 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
4002 "vqadd", "u", int_arm_neon_vqaddu, 1>;
4003 // VADDHN : Vector Add and Narrow Returning High Half (D = Q + Q)
4004 defm VADDHN : N3VNInt_HSD<0,1,0b0100,0, "vaddhn", "i",
4005 int_arm_neon_vaddhn, 1>;
4006 // VRADDHN : Vector Rounding Add and Narrow Returning High Half (D = Q + Q)
4007 defm VRADDHN : N3VNInt_HSD<1,1,0b0100,0, "vraddhn", "i",
4008 int_arm_neon_vraddhn, 1>;
4010 // Vector Multiply Operations.
4012 // VMUL : Vector Multiply (integer, polynomial and floating-point)
4013 defm VMUL : N3V_QHS<0, 0, 0b1001, 1, IIC_VMULi16D, IIC_VMULi32D,
4014 IIC_VMULi16Q, IIC_VMULi32Q, "vmul", "i", mul, 1>;
4015 def VMULpd : N3VDInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16D, "vmul",
4016 "p8", v8i8, v8i8, int_arm_neon_vmulp, 1>;
4017 def VMULpq : N3VQInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16Q, "vmul",
4018 "p8", v16i8, v16i8, int_arm_neon_vmulp, 1>;
4019 def VMULfd : N3VD<1, 0, 0b00, 0b1101, 1, IIC_VFMULD, "vmul", "f32",
4020 v2f32, v2f32, fmul, 1>;
4021 def VMULfq : N3VQ<1, 0, 0b00, 0b1101, 1, IIC_VFMULQ, "vmul", "f32",
4022 v4f32, v4f32, fmul, 1>;
4023 defm VMULsl : N3VSL_HS<0b1000, "vmul", mul>;
4024 def VMULslfd : N3VDSL<0b10, 0b1001, IIC_VBIND, "vmul", "f32", v2f32, fmul>;
4025 def VMULslfq : N3VQSL<0b10, 0b1001, IIC_VBINQ, "vmul", "f32", v4f32,
4028 def : Pat<(v8i16 (mul (v8i16 QPR:$src1),
4029 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
4030 (v8i16 (VMULslv8i16 (v8i16 QPR:$src1),
4031 (v4i16 (EXTRACT_SUBREG QPR:$src2,
4032 (DSubReg_i16_reg imm:$lane))),
4033 (SubReg_i16_lane imm:$lane)))>;
4034 def : Pat<(v4i32 (mul (v4i32 QPR:$src1),
4035 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
4036 (v4i32 (VMULslv4i32 (v4i32 QPR:$src1),
4037 (v2i32 (EXTRACT_SUBREG QPR:$src2,
4038 (DSubReg_i32_reg imm:$lane))),
4039 (SubReg_i32_lane imm:$lane)))>;
4040 def : Pat<(v4f32 (fmul (v4f32 QPR:$src1),
4041 (v4f32 (NEONvduplane (v4f32 QPR:$src2), imm:$lane)))),
4042 (v4f32 (VMULslfq (v4f32 QPR:$src1),
4043 (v2f32 (EXTRACT_SUBREG QPR:$src2,
4044 (DSubReg_i32_reg imm:$lane))),
4045 (SubReg_i32_lane imm:$lane)))>;
4047 // VQDMULH : Vector Saturating Doubling Multiply Returning High Half
4048 defm VQDMULH : N3VInt_HS<0, 0, 0b1011, 0, N3RegFrm, IIC_VMULi16D, IIC_VMULi32D,
4049 IIC_VMULi16Q, IIC_VMULi32Q,
4050 "vqdmulh", "s", int_arm_neon_vqdmulh, 1>;
4051 defm VQDMULHsl: N3VIntSL_HS<0b1100, IIC_VMULi16D, IIC_VMULi32D,
4052 IIC_VMULi16Q, IIC_VMULi32Q,
4053 "vqdmulh", "s", int_arm_neon_vqdmulh>;
4054 def : Pat<(v8i16 (int_arm_neon_vqdmulh (v8i16 QPR:$src1),
4055 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
4057 (v8i16 (VQDMULHslv8i16 (v8i16 QPR:$src1),
4058 (v4i16 (EXTRACT_SUBREG QPR:$src2,
4059 (DSubReg_i16_reg imm:$lane))),
4060 (SubReg_i16_lane imm:$lane)))>;
4061 def : Pat<(v4i32 (int_arm_neon_vqdmulh (v4i32 QPR:$src1),
4062 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
4064 (v4i32 (VQDMULHslv4i32 (v4i32 QPR:$src1),
4065 (v2i32 (EXTRACT_SUBREG QPR:$src2,
4066 (DSubReg_i32_reg imm:$lane))),
4067 (SubReg_i32_lane imm:$lane)))>;
4069 // VQRDMULH : Vector Rounding Saturating Doubling Multiply Returning High Half
4070 defm VQRDMULH : N3VInt_HS<1, 0, 0b1011, 0, N3RegFrm,
4071 IIC_VMULi16D,IIC_VMULi32D,IIC_VMULi16Q,IIC_VMULi32Q,
4072 "vqrdmulh", "s", int_arm_neon_vqrdmulh, 1>;
4073 defm VQRDMULHsl : N3VIntSL_HS<0b1101, IIC_VMULi16D, IIC_VMULi32D,
4074 IIC_VMULi16Q, IIC_VMULi32Q,
4075 "vqrdmulh", "s", int_arm_neon_vqrdmulh>;
4076 def : Pat<(v8i16 (int_arm_neon_vqrdmulh (v8i16 QPR:$src1),
4077 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
4079 (v8i16 (VQRDMULHslv8i16 (v8i16 QPR:$src1),
4080 (v4i16 (EXTRACT_SUBREG QPR:$src2,
4081 (DSubReg_i16_reg imm:$lane))),
4082 (SubReg_i16_lane imm:$lane)))>;
4083 def : Pat<(v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$src1),
4084 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
4086 (v4i32 (VQRDMULHslv4i32 (v4i32 QPR:$src1),
4087 (v2i32 (EXTRACT_SUBREG QPR:$src2,
4088 (DSubReg_i32_reg imm:$lane))),
4089 (SubReg_i32_lane imm:$lane)))>;
4091 // VMULL : Vector Multiply Long (integer and polynomial) (Q = D * D)
4092 defm VMULLs : N3VL_QHS<0,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
4093 "vmull", "s", NEONvmulls, 1>;
4094 defm VMULLu : N3VL_QHS<1,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
4095 "vmull", "u", NEONvmullu, 1>;
4096 def VMULLp : N3VLInt<0, 1, 0b00, 0b1110, 0, IIC_VMULi16D, "vmull", "p8",
4097 v8i16, v8i8, int_arm_neon_vmullp, 1>;
4098 defm VMULLsls : N3VLSL_HS<0, 0b1010, IIC_VMULi16D, "vmull", "s", NEONvmulls>;
4099 defm VMULLslu : N3VLSL_HS<1, 0b1010, IIC_VMULi16D, "vmull", "u", NEONvmullu>;
4101 // VQDMULL : Vector Saturating Doubling Multiply Long (Q = D * D)
4102 defm VQDMULL : N3VLInt_HS<0,1,0b1101,0, IIC_VMULi16D, IIC_VMULi32D,
4103 "vqdmull", "s", int_arm_neon_vqdmull, 1>;
4104 defm VQDMULLsl: N3VLIntSL_HS<0, 0b1011, IIC_VMULi16D,
4105 "vqdmull", "s", int_arm_neon_vqdmull>;
4107 // Vector Multiply-Accumulate and Multiply-Subtract Operations.
4109 // VMLA : Vector Multiply Accumulate (integer and floating-point)
4110 defm VMLA : N3VMulOp_QHS<0, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
4111 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
4112 def VMLAfd : N3VDMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACD, "vmla", "f32",
4113 v2f32, fmul_su, fadd_mlx>,
4114 Requires<[HasNEON, UseFPVMLx, DontUseFusedMAC]>;
4115 def VMLAfq : N3VQMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACQ, "vmla", "f32",
4116 v4f32, fmul_su, fadd_mlx>,
4117 Requires<[HasNEON, UseFPVMLx, DontUseFusedMAC]>;
4118 defm VMLAsl : N3VMulOpSL_HS<0b0000, IIC_VMACi16D, IIC_VMACi32D,
4119 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
4120 def VMLAslfd : N3VDMulOpSL<0b10, 0b0001, IIC_VMACD, "vmla", "f32",
4121 v2f32, fmul_su, fadd_mlx>,
4122 Requires<[HasNEON, UseFPVMLx]>;
4123 def VMLAslfq : N3VQMulOpSL<0b10, 0b0001, IIC_VMACQ, "vmla", "f32",
4124 v4f32, v2f32, fmul_su, fadd_mlx>,
4125 Requires<[HasNEON, UseFPVMLx]>;
4127 def : Pat<(v8i16 (add (v8i16 QPR:$src1),
4128 (mul (v8i16 QPR:$src2),
4129 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
4130 (v8i16 (VMLAslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
4131 (v4i16 (EXTRACT_SUBREG QPR:$src3,
4132 (DSubReg_i16_reg imm:$lane))),
4133 (SubReg_i16_lane imm:$lane)))>;
4135 def : Pat<(v4i32 (add (v4i32 QPR:$src1),
4136 (mul (v4i32 QPR:$src2),
4137 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
4138 (v4i32 (VMLAslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
4139 (v2i32 (EXTRACT_SUBREG QPR:$src3,
4140 (DSubReg_i32_reg imm:$lane))),
4141 (SubReg_i32_lane imm:$lane)))>;
4143 def : Pat<(v4f32 (fadd_mlx (v4f32 QPR:$src1),
4144 (fmul_su (v4f32 QPR:$src2),
4145 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
4146 (v4f32 (VMLAslfq (v4f32 QPR:$src1),
4148 (v2f32 (EXTRACT_SUBREG QPR:$src3,
4149 (DSubReg_i32_reg imm:$lane))),
4150 (SubReg_i32_lane imm:$lane)))>,
4151 Requires<[HasNEON, UseFPVMLx]>;
4153 // VMLAL : Vector Multiply Accumulate Long (Q += D * D)
4154 defm VMLALs : N3VLMulOp_QHS<0,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
4155 "vmlal", "s", NEONvmulls, add>;
4156 defm VMLALu : N3VLMulOp_QHS<1,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
4157 "vmlal", "u", NEONvmullu, add>;
4159 defm VMLALsls : N3VLMulOpSL_HS<0, 0b0010, "vmlal", "s", NEONvmulls, add>;
4160 defm VMLALslu : N3VLMulOpSL_HS<1, 0b0010, "vmlal", "u", NEONvmullu, add>;
4162 // VQDMLAL : Vector Saturating Doubling Multiply Accumulate Long (Q += D * D)
4163 defm VQDMLAL : N3VLInt3_HS<0, 1, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
4164 "vqdmlal", "s", int_arm_neon_vqdmlal>;
4165 defm VQDMLALsl: N3VLInt3SL_HS<0, 0b0011, "vqdmlal", "s", int_arm_neon_vqdmlal>;
4167 // VMLS : Vector Multiply Subtract (integer and floating-point)
4168 defm VMLS : N3VMulOp_QHS<1, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
4169 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
4170 def VMLSfd : N3VDMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACD, "vmls", "f32",
4171 v2f32, fmul_su, fsub_mlx>,
4172 Requires<[HasNEON, UseFPVMLx, DontUseFusedMAC]>;
4173 def VMLSfq : N3VQMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACQ, "vmls", "f32",
4174 v4f32, fmul_su, fsub_mlx>,
4175 Requires<[HasNEON, UseFPVMLx, DontUseFusedMAC]>;
4176 defm VMLSsl : N3VMulOpSL_HS<0b0100, IIC_VMACi16D, IIC_VMACi32D,
4177 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
4178 def VMLSslfd : N3VDMulOpSL<0b10, 0b0101, IIC_VMACD, "vmls", "f32",
4179 v2f32, fmul_su, fsub_mlx>,
4180 Requires<[HasNEON, UseFPVMLx]>;
4181 def VMLSslfq : N3VQMulOpSL<0b10, 0b0101, IIC_VMACQ, "vmls", "f32",
4182 v4f32, v2f32, fmul_su, fsub_mlx>,
4183 Requires<[HasNEON, UseFPVMLx]>;
4185 def : Pat<(v8i16 (sub (v8i16 QPR:$src1),
4186 (mul (v8i16 QPR:$src2),
4187 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
4188 (v8i16 (VMLSslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
4189 (v4i16 (EXTRACT_SUBREG QPR:$src3,
4190 (DSubReg_i16_reg imm:$lane))),
4191 (SubReg_i16_lane imm:$lane)))>;
4193 def : Pat<(v4i32 (sub (v4i32 QPR:$src1),
4194 (mul (v4i32 QPR:$src2),
4195 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
4196 (v4i32 (VMLSslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
4197 (v2i32 (EXTRACT_SUBREG QPR:$src3,
4198 (DSubReg_i32_reg imm:$lane))),
4199 (SubReg_i32_lane imm:$lane)))>;
4201 def : Pat<(v4f32 (fsub_mlx (v4f32 QPR:$src1),
4202 (fmul_su (v4f32 QPR:$src2),
4203 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
4204 (v4f32 (VMLSslfq (v4f32 QPR:$src1), (v4f32 QPR:$src2),
4205 (v2f32 (EXTRACT_SUBREG QPR:$src3,
4206 (DSubReg_i32_reg imm:$lane))),
4207 (SubReg_i32_lane imm:$lane)))>,
4208 Requires<[HasNEON, UseFPVMLx]>;
4210 // VMLSL : Vector Multiply Subtract Long (Q -= D * D)
4211 defm VMLSLs : N3VLMulOp_QHS<0,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
4212 "vmlsl", "s", NEONvmulls, sub>;
4213 defm VMLSLu : N3VLMulOp_QHS<1,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
4214 "vmlsl", "u", NEONvmullu, sub>;
4216 defm VMLSLsls : N3VLMulOpSL_HS<0, 0b0110, "vmlsl", "s", NEONvmulls, sub>;
4217 defm VMLSLslu : N3VLMulOpSL_HS<1, 0b0110, "vmlsl", "u", NEONvmullu, sub>;
4219 // VQDMLSL : Vector Saturating Doubling Multiply Subtract Long (Q -= D * D)
4220 defm VQDMLSL : N3VLInt3_HS<0, 1, 0b1011, 0, IIC_VMACi16D, IIC_VMACi32D,
4221 "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
4222 defm VQDMLSLsl: N3VLInt3SL_HS<0, 0b111, "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
4224 // Fused Vector Multiply-Accumulate and Fused Multiply-Subtract Operations.
4225 def VFMAfd : N3VDMulOp<0, 0, 0b00, 0b1100, 1, IIC_VFMACD, "vfma", "f32",
4226 v2f32, fmul_su, fadd_mlx>,
4227 Requires<[HasVFP4,UseFusedMAC]>;
4229 def VFMAfq : N3VQMulOp<0, 0, 0b00, 0b1100, 1, IIC_VFMACQ, "vfma", "f32",
4230 v4f32, fmul_su, fadd_mlx>,
4231 Requires<[HasVFP4,UseFusedMAC]>;
4233 // Fused Vector Multiply Subtract (floating-point)
4234 def VFMSfd : N3VDMulOp<0, 0, 0b10, 0b1100, 1, IIC_VFMACD, "vfms", "f32",
4235 v2f32, fmul_su, fsub_mlx>,
4236 Requires<[HasVFP4,UseFusedMAC]>;
4237 def VFMSfq : N3VQMulOp<0, 0, 0b10, 0b1100, 1, IIC_VFMACQ, "vfms", "f32",
4238 v4f32, fmul_su, fsub_mlx>,
4239 Requires<[HasVFP4,UseFusedMAC]>;
4241 // Match @llvm.fma.* intrinsics
4242 def : Pat<(v2f32 (fma DPR:$Vn, DPR:$Vm, DPR:$src1)),
4243 (VFMAfd DPR:$src1, DPR:$Vn, DPR:$Vm)>,
4244 Requires<[HasVFP4]>;
4245 def : Pat<(v4f32 (fma QPR:$Vn, QPR:$Vm, QPR:$src1)),
4246 (VFMAfq QPR:$src1, QPR:$Vn, QPR:$Vm)>,
4247 Requires<[HasVFP4]>;
4248 def : Pat<(v2f32 (fma (fneg DPR:$Vn), DPR:$Vm, DPR:$src1)),
4249 (VFMSfd DPR:$src1, DPR:$Vn, DPR:$Vm)>,
4250 Requires<[HasVFP4]>;
4251 def : Pat<(v4f32 (fma (fneg QPR:$Vn), QPR:$Vm, QPR:$src1)),
4252 (VFMSfq QPR:$src1, QPR:$Vn, QPR:$Vm)>,
4253 Requires<[HasVFP4]>;
4255 // Vector Subtract Operations.
4257 // VSUB : Vector Subtract (integer and floating-point)
4258 defm VSUB : N3V_QHSD<1, 0, 0b1000, 0, IIC_VSUBiD, IIC_VSUBiQ,
4259 "vsub", "i", sub, 0>;
4260 def VSUBfd : N3VD<0, 0, 0b10, 0b1101, 0, IIC_VBIND, "vsub", "f32",
4261 v2f32, v2f32, fsub, 0>;
4262 def VSUBfq : N3VQ<0, 0, 0b10, 0b1101, 0, IIC_VBINQ, "vsub", "f32",
4263 v4f32, v4f32, fsub, 0>;
4264 // VSUBL : Vector Subtract Long (Q = D - D)
4265 defm VSUBLs : N3VLExt_QHS<0,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
4266 "vsubl", "s", sub, sext, 0>;
4267 defm VSUBLu : N3VLExt_QHS<1,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
4268 "vsubl", "u", sub, zext, 0>;
4269 // VSUBW : Vector Subtract Wide (Q = Q - D)
4270 defm VSUBWs : N3VW_QHS<0,1,0b0011,0, "vsubw", "s", sub, sext, 0>;
4271 defm VSUBWu : N3VW_QHS<1,1,0b0011,0, "vsubw", "u", sub, zext, 0>;
4272 // VHSUB : Vector Halving Subtract
4273 defm VHSUBs : N3VInt_QHS<0, 0, 0b0010, 0, N3RegFrm,
4274 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4275 "vhsub", "s", int_arm_neon_vhsubs, 0>;
4276 defm VHSUBu : N3VInt_QHS<1, 0, 0b0010, 0, N3RegFrm,
4277 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4278 "vhsub", "u", int_arm_neon_vhsubu, 0>;
4279 // VQSUB : Vector Saturing Subtract
4280 defm VQSUBs : N3VInt_QHSD<0, 0, 0b0010, 1, N3RegFrm,
4281 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4282 "vqsub", "s", int_arm_neon_vqsubs, 0>;
4283 defm VQSUBu : N3VInt_QHSD<1, 0, 0b0010, 1, N3RegFrm,
4284 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4285 "vqsub", "u", int_arm_neon_vqsubu, 0>;
4286 // VSUBHN : Vector Subtract and Narrow Returning High Half (D = Q - Q)
4287 defm VSUBHN : N3VNInt_HSD<0,1,0b0110,0, "vsubhn", "i",
4288 int_arm_neon_vsubhn, 0>;
4289 // VRSUBHN : Vector Rounding Subtract and Narrow Returning High Half (D=Q-Q)
4290 defm VRSUBHN : N3VNInt_HSD<1,1,0b0110,0, "vrsubhn", "i",
4291 int_arm_neon_vrsubhn, 0>;
4293 // Vector Comparisons.
4295 // VCEQ : Vector Compare Equal
4296 defm VCEQ : N3V_QHS<1, 0, 0b1000, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
4297 IIC_VSUBi4Q, "vceq", "i", NEONvceq, 1>;
4298 def VCEQfd : N3VD<0,0,0b00,0b1110,0, IIC_VBIND, "vceq", "f32", v2i32, v2f32,
4300 def VCEQfq : N3VQ<0,0,0b00,0b1110,0, IIC_VBINQ, "vceq", "f32", v4i32, v4f32,
4303 let TwoOperandAliasConstraint = "$Vm = $Vd" in
4304 defm VCEQz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00010, 0, "vceq", "i",
4305 "$Vd, $Vm, #0", NEONvceqz>;
4307 // VCGE : Vector Compare Greater Than or Equal
4308 defm VCGEs : N3V_QHS<0, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
4309 IIC_VSUBi4Q, "vcge", "s", NEONvcge, 0>;
4310 defm VCGEu : N3V_QHS<1, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
4311 IIC_VSUBi4Q, "vcge", "u", NEONvcgeu, 0>;
4312 def VCGEfd : N3VD<1,0,0b00,0b1110,0, IIC_VBIND, "vcge", "f32", v2i32, v2f32,
4314 def VCGEfq : N3VQ<1,0,0b00,0b1110,0, IIC_VBINQ, "vcge", "f32", v4i32, v4f32,
4317 let TwoOperandAliasConstraint = "$Vm = $Vd" in {
4318 defm VCGEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00001, 0, "vcge", "s",
4319 "$Vd, $Vm, #0", NEONvcgez>;
4320 defm VCLEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00011, 0, "vcle", "s",
4321 "$Vd, $Vm, #0", NEONvclez>;
4324 // VCGT : Vector Compare Greater Than
4325 defm VCGTs : N3V_QHS<0, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
4326 IIC_VSUBi4Q, "vcgt", "s", NEONvcgt, 0>;
4327 defm VCGTu : N3V_QHS<1, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
4328 IIC_VSUBi4Q, "vcgt", "u", NEONvcgtu, 0>;
4329 def VCGTfd : N3VD<1,0,0b10,0b1110,0, IIC_VBIND, "vcgt", "f32", v2i32, v2f32,
4331 def VCGTfq : N3VQ<1,0,0b10,0b1110,0, IIC_VBINQ, "vcgt", "f32", v4i32, v4f32,
4334 let TwoOperandAliasConstraint = "$Vm = $Vd" in {
4335 defm VCGTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00000, 0, "vcgt", "s",
4336 "$Vd, $Vm, #0", NEONvcgtz>;
4337 defm VCLTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00100, 0, "vclt", "s",
4338 "$Vd, $Vm, #0", NEONvcltz>;
4341 // VACGE : Vector Absolute Compare Greater Than or Equal (aka VCAGE)
4342 def VACGEd : N3VDInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacge",
4343 "f32", v2i32, v2f32, int_arm_neon_vacged, 0>;
4344 def VACGEq : N3VQInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacge",
4345 "f32", v4i32, v4f32, int_arm_neon_vacgeq, 0>;
4346 // VACGT : Vector Absolute Compare Greater Than (aka VCAGT)
4347 def VACGTd : N3VDInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacgt",
4348 "f32", v2i32, v2f32, int_arm_neon_vacgtd, 0>;
4349 def VACGTq : N3VQInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacgt",
4350 "f32", v4i32, v4f32, int_arm_neon_vacgtq, 0>;
4351 // VTST : Vector Test Bits
4352 defm VTST : N3V_QHS<0, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
4353 IIC_VBINi4Q, "vtst", "", NEONvtst, 1>;
4355 def: NEONInstAlias<"vaclt${p}.f32 $Vd, $Vn, $Vm",
4356 (VACGTd DPR:$Vd, DPR:$Vm, DPR:$Vn, pred:$p)>;
4357 def: NEONInstAlias<"vaclt${p}.f32 $Vd, $Vn, $Vm",
4358 (VACGTq QPR:$Vd, QPR:$Vm, QPR:$Vn, pred:$p)>;
4359 def: NEONInstAlias<"vacle${p}.f32 $Vd, $Vn, $Vm",
4360 (VACGEd DPR:$Vd, DPR:$Vm, DPR:$Vn, pred:$p)>;
4361 def: NEONInstAlias<"vacle${p}.f32 $Vd, $Vn, $Vm",
4362 (VACGEq QPR:$Vd, QPR:$Vm, QPR:$Vn, pred:$p)>;
4364 def: NEONInstAlias<"vaclt${p}.f32 $Vd, $Vm",
4365 (VACGTd DPR:$Vd, DPR:$Vm, DPR:$Vd, pred:$p)>;
4366 def: NEONInstAlias<"vaclt${p}.f32 $Vd, $Vm",
4367 (VACGTq QPR:$Vd, QPR:$Vm, QPR:$Vd, pred:$p)>;
4368 def: NEONInstAlias<"vacle${p}.f32 $Vd, $Vm",
4369 (VACGEd DPR:$Vd, DPR:$Vm, DPR:$Vd, pred:$p)>;
4370 def: NEONInstAlias<"vacle${p}.f32 $Vd, $Vm",
4371 (VACGEq QPR:$Vd, QPR:$Vm, QPR:$Vd, pred:$p)>;
4373 // Vector Bitwise Operations.
4375 def vnotd : PatFrag<(ops node:$in),
4376 (xor node:$in, (bitconvert (v8i8 NEONimmAllOnesV)))>;
4377 def vnotq : PatFrag<(ops node:$in),
4378 (xor node:$in, (bitconvert (v16i8 NEONimmAllOnesV)))>;
4381 // VAND : Vector Bitwise AND
4382 def VANDd : N3VDX<0, 0, 0b00, 0b0001, 1, IIC_VBINiD, "vand",
4383 v2i32, v2i32, and, 1>;
4384 def VANDq : N3VQX<0, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "vand",
4385 v4i32, v4i32, and, 1>;
4387 // VEOR : Vector Bitwise Exclusive OR
4388 def VEORd : N3VDX<1, 0, 0b00, 0b0001, 1, IIC_VBINiD, "veor",
4389 v2i32, v2i32, xor, 1>;
4390 def VEORq : N3VQX<1, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "veor",
4391 v4i32, v4i32, xor, 1>;
4393 // VORR : Vector Bitwise OR
4394 def VORRd : N3VDX<0, 0, 0b10, 0b0001, 1, IIC_VBINiD, "vorr",
4395 v2i32, v2i32, or, 1>;
4396 def VORRq : N3VQX<0, 0, 0b10, 0b0001, 1, IIC_VBINiQ, "vorr",
4397 v4i32, v4i32, or, 1>;
4399 def VORRiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 0, 1,
4400 (outs DPR:$Vd), (ins nImmSplatI16:$SIMM, DPR:$src),
4402 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
4404 (v4i16 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
4405 let Inst{9} = SIMM{9};
4408 def VORRiv2i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 0, 0, 1,
4409 (outs DPR:$Vd), (ins nImmSplatI32:$SIMM, DPR:$src),
4411 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
4413 (v2i32 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
4414 let Inst{10-9} = SIMM{10-9};
4417 def VORRiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 0, 1,
4418 (outs QPR:$Vd), (ins nImmSplatI16:$SIMM, QPR:$src),
4420 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
4422 (v8i16 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
4423 let Inst{9} = SIMM{9};
4426 def VORRiv4i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 1, 0, 1,
4427 (outs QPR:$Vd), (ins nImmSplatI32:$SIMM, QPR:$src),
4429 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
4431 (v4i32 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
4432 let Inst{10-9} = SIMM{10-9};
4436 // VBIC : Vector Bitwise Bit Clear (AND NOT)
4437 let TwoOperandAliasConstraint = "$Vn = $Vd" in {
4438 def VBICd : N3VX<0, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
4439 (ins DPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VBINiD,
4440 "vbic", "$Vd, $Vn, $Vm", "",
4441 [(set DPR:$Vd, (v2i32 (and DPR:$Vn,
4442 (vnotd DPR:$Vm))))]>;
4443 def VBICq : N3VX<0, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
4444 (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINiQ,
4445 "vbic", "$Vd, $Vn, $Vm", "",
4446 [(set QPR:$Vd, (v4i32 (and QPR:$Vn,
4447 (vnotq QPR:$Vm))))]>;
4450 def VBICiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 1, 1,
4451 (outs DPR:$Vd), (ins nImmSplatI16:$SIMM, DPR:$src),
4453 "vbic", "i16", "$Vd, $SIMM", "$src = $Vd",
4455 (v4i16 (NEONvbicImm DPR:$src, timm:$SIMM)))]> {
4456 let Inst{9} = SIMM{9};
4459 def VBICiv2i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 0, 1, 1,
4460 (outs DPR:$Vd), (ins nImmSplatI32:$SIMM, DPR:$src),
4462 "vbic", "i32", "$Vd, $SIMM", "$src = $Vd",
4464 (v2i32 (NEONvbicImm DPR:$src, timm:$SIMM)))]> {
4465 let Inst{10-9} = SIMM{10-9};
4468 def VBICiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 1, 1,
4469 (outs QPR:$Vd), (ins nImmSplatI16:$SIMM, QPR:$src),
4471 "vbic", "i16", "$Vd, $SIMM", "$src = $Vd",
4473 (v8i16 (NEONvbicImm QPR:$src, timm:$SIMM)))]> {
4474 let Inst{9} = SIMM{9};
4477 def VBICiv4i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 1, 1, 1,
4478 (outs QPR:$Vd), (ins nImmSplatI32:$SIMM, QPR:$src),
4480 "vbic", "i32", "$Vd, $SIMM", "$src = $Vd",
4482 (v4i32 (NEONvbicImm QPR:$src, timm:$SIMM)))]> {
4483 let Inst{10-9} = SIMM{10-9};
4486 // VORN : Vector Bitwise OR NOT
4487 def VORNd : N3VX<0, 0, 0b11, 0b0001, 0, 1, (outs DPR:$Vd),
4488 (ins DPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VBINiD,
4489 "vorn", "$Vd, $Vn, $Vm", "",
4490 [(set DPR:$Vd, (v2i32 (or DPR:$Vn,
4491 (vnotd DPR:$Vm))))]>;
4492 def VORNq : N3VX<0, 0, 0b11, 0b0001, 1, 1, (outs QPR:$Vd),
4493 (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINiQ,
4494 "vorn", "$Vd, $Vn, $Vm", "",
4495 [(set QPR:$Vd, (v4i32 (or QPR:$Vn,
4496 (vnotq QPR:$Vm))))]>;
4498 // VMVN : Vector Bitwise NOT (Immediate)
4500 let isReMaterializable = 1 in {
4502 def VMVNv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 1, 1, (outs DPR:$Vd),
4503 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4504 "vmvn", "i16", "$Vd, $SIMM", "",
4505 [(set DPR:$Vd, (v4i16 (NEONvmvnImm timm:$SIMM)))]> {
4506 let Inst{9} = SIMM{9};
4509 def VMVNv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 1, 1, (outs QPR:$Vd),
4510 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4511 "vmvn", "i16", "$Vd, $SIMM", "",
4512 [(set QPR:$Vd, (v8i16 (NEONvmvnImm timm:$SIMM)))]> {
4513 let Inst{9} = SIMM{9};
4516 def VMVNv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 1, 1, (outs DPR:$Vd),
4517 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4518 "vmvn", "i32", "$Vd, $SIMM", "",
4519 [(set DPR:$Vd, (v2i32 (NEONvmvnImm timm:$SIMM)))]> {
4520 let Inst{11-8} = SIMM{11-8};
4523 def VMVNv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 1, 1, (outs QPR:$Vd),
4524 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4525 "vmvn", "i32", "$Vd, $SIMM", "",
4526 [(set QPR:$Vd, (v4i32 (NEONvmvnImm timm:$SIMM)))]> {
4527 let Inst{11-8} = SIMM{11-8};
4531 // VMVN : Vector Bitwise NOT
4532 def VMVNd : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 0, 0,
4533 (outs DPR:$Vd), (ins DPR:$Vm), IIC_VSUBiD,
4534 "vmvn", "$Vd, $Vm", "",
4535 [(set DPR:$Vd, (v2i32 (vnotd DPR:$Vm)))]>;
4536 def VMVNq : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 1, 0,
4537 (outs QPR:$Vd), (ins QPR:$Vm), IIC_VSUBiD,
4538 "vmvn", "$Vd, $Vm", "",
4539 [(set QPR:$Vd, (v4i32 (vnotq QPR:$Vm)))]>;
4540 def : Pat<(v2i32 (vnotd DPR:$src)), (VMVNd DPR:$src)>;
4541 def : Pat<(v4i32 (vnotq QPR:$src)), (VMVNq QPR:$src)>;
4543 // VBSL : Vector Bitwise Select
4544 def VBSLd : N3VX<1, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
4545 (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
4546 N3RegFrm, IIC_VCNTiD,
4547 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4549 (v2i32 (NEONvbsl DPR:$src1, DPR:$Vn, DPR:$Vm)))]>;
4550 def : Pat<(v8i8 (int_arm_neon_vbsl (v8i8 DPR:$src1),
4551 (v8i8 DPR:$Vn), (v8i8 DPR:$Vm))),
4552 (VBSLd DPR:$src1, DPR:$Vn, DPR:$Vm)>,
4553 Requires<[HasNEON]>;
4554 def : Pat<(v4i16 (int_arm_neon_vbsl (v4i16 DPR:$src1),
4555 (v4i16 DPR:$Vn), (v4i16 DPR:$Vm))),
4556 (VBSLd DPR:$src1, DPR:$Vn, DPR:$Vm)>,
4557 Requires<[HasNEON]>;
4558 def : Pat<(v2i32 (int_arm_neon_vbsl (v2i32 DPR:$src1),
4559 (v2i32 DPR:$Vn), (v2i32 DPR:$Vm))),
4560 (VBSLd DPR:$src1, DPR:$Vn, DPR:$Vm)>,
4561 Requires<[HasNEON]>;
4562 def : Pat<(v2f32 (int_arm_neon_vbsl (v2f32 DPR:$src1),
4563 (v2f32 DPR:$Vn), (v2f32 DPR:$Vm))),
4564 (VBSLd DPR:$src1, DPR:$Vn, DPR:$Vm)>,
4565 Requires<[HasNEON]>;
4566 def : Pat<(v1i64 (int_arm_neon_vbsl (v1i64 DPR:$src1),
4567 (v1i64 DPR:$Vn), (v1i64 DPR:$Vm))),
4568 (VBSLd DPR:$src1, DPR:$Vn, DPR:$Vm)>,
4569 Requires<[HasNEON]>;
4571 def : Pat<(v2i32 (or (and DPR:$Vn, DPR:$Vd),
4572 (and DPR:$Vm, (vnotd DPR:$Vd)))),
4573 (VBSLd DPR:$Vd, DPR:$Vn, DPR:$Vm)>,
4574 Requires<[HasNEON]>;
4576 def : Pat<(v1i64 (or (and DPR:$Vn, DPR:$Vd),
4577 (and DPR:$Vm, (vnotd DPR:$Vd)))),
4578 (VBSLd DPR:$Vd, DPR:$Vn, DPR:$Vm)>,
4579 Requires<[HasNEON]>;
4581 def VBSLq : N3VX<1, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
4582 (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
4583 N3RegFrm, IIC_VCNTiQ,
4584 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4586 (v4i32 (NEONvbsl QPR:$src1, QPR:$Vn, QPR:$Vm)))]>;
4588 def : Pat<(v16i8 (int_arm_neon_vbsl (v16i8 QPR:$src1),
4589 (v16i8 QPR:$Vn), (v16i8 QPR:$Vm))),
4590 (VBSLq QPR:$src1, QPR:$Vn, QPR:$Vm)>,
4591 Requires<[HasNEON]>;
4592 def : Pat<(v8i16 (int_arm_neon_vbsl (v8i16 QPR:$src1),
4593 (v8i16 QPR:$Vn), (v8i16 QPR:$Vm))),
4594 (VBSLq QPR:$src1, QPR:$Vn, QPR:$Vm)>,
4595 Requires<[HasNEON]>;
4596 def : Pat<(v4i32 (int_arm_neon_vbsl (v4i32 QPR:$src1),
4597 (v4i32 QPR:$Vn), (v4i32 QPR:$Vm))),
4598 (VBSLq QPR:$src1, QPR:$Vn, QPR:$Vm)>,
4599 Requires<[HasNEON]>;
4600 def : Pat<(v4f32 (int_arm_neon_vbsl (v4f32 QPR:$src1),
4601 (v4f32 QPR:$Vn), (v4f32 QPR:$Vm))),
4602 (VBSLq QPR:$src1, QPR:$Vn, QPR:$Vm)>,
4603 Requires<[HasNEON]>;
4604 def : Pat<(v2i64 (int_arm_neon_vbsl (v2i64 QPR:$src1),
4605 (v2i64 QPR:$Vn), (v2i64 QPR:$Vm))),
4606 (VBSLq QPR:$src1, QPR:$Vn, QPR:$Vm)>,
4607 Requires<[HasNEON]>;
4609 def : Pat<(v4i32 (or (and QPR:$Vn, QPR:$Vd),
4610 (and QPR:$Vm, (vnotq QPR:$Vd)))),
4611 (VBSLq QPR:$Vd, QPR:$Vn, QPR:$Vm)>,
4612 Requires<[HasNEON]>;
4613 def : Pat<(v2i64 (or (and QPR:$Vn, QPR:$Vd),
4614 (and QPR:$Vm, (vnotq QPR:$Vd)))),
4615 (VBSLq QPR:$Vd, QPR:$Vn, QPR:$Vm)>,
4616 Requires<[HasNEON]>;
4618 // VBIF : Vector Bitwise Insert if False
4619 // like VBSL but with: "vbif $dst, $src3, $src1", "$src2 = $dst",
4620 // FIXME: This instruction's encoding MAY NOT BE correct.
4621 def VBIFd : N3VX<1, 0, 0b11, 0b0001, 0, 1,
4622 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
4623 N3RegFrm, IIC_VBINiD,
4624 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4626 def VBIFq : N3VX<1, 0, 0b11, 0b0001, 1, 1,
4627 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
4628 N3RegFrm, IIC_VBINiQ,
4629 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4632 // VBIT : Vector Bitwise Insert if True
4633 // like VBSL but with: "vbit $dst, $src2, $src1", "$src3 = $dst",
4634 // FIXME: This instruction's encoding MAY NOT BE correct.
4635 def VBITd : N3VX<1, 0, 0b10, 0b0001, 0, 1,
4636 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
4637 N3RegFrm, IIC_VBINiD,
4638 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4640 def VBITq : N3VX<1, 0, 0b10, 0b0001, 1, 1,
4641 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
4642 N3RegFrm, IIC_VBINiQ,
4643 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4646 // VBIT/VBIF are not yet implemented. The TwoAddress pass will not go looking
4647 // for equivalent operations with different register constraints; it just
4650 // Vector Absolute Differences.
4652 // VABD : Vector Absolute Difference
4653 defm VABDs : N3VInt_QHS<0, 0, 0b0111, 0, N3RegFrm,
4654 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4655 "vabd", "s", int_arm_neon_vabds, 1>;
4656 defm VABDu : N3VInt_QHS<1, 0, 0b0111, 0, N3RegFrm,
4657 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4658 "vabd", "u", int_arm_neon_vabdu, 1>;
4659 def VABDfd : N3VDInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBIND,
4660 "vabd", "f32", v2f32, v2f32, int_arm_neon_vabds, 1>;
4661 def VABDfq : N3VQInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBINQ,
4662 "vabd", "f32", v4f32, v4f32, int_arm_neon_vabds, 1>;
4664 // VABDL : Vector Absolute Difference Long (Q = | D - D |)
4665 defm VABDLs : N3VLIntExt_QHS<0,1,0b0111,0, IIC_VSUBi4Q,
4666 "vabdl", "s", int_arm_neon_vabds, zext, 1>;
4667 defm VABDLu : N3VLIntExt_QHS<1,1,0b0111,0, IIC_VSUBi4Q,
4668 "vabdl", "u", int_arm_neon_vabdu, zext, 1>;
4670 // VABA : Vector Absolute Difference and Accumulate
4671 defm VABAs : N3VIntOp_QHS<0,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
4672 "vaba", "s", int_arm_neon_vabds, add>;
4673 defm VABAu : N3VIntOp_QHS<1,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
4674 "vaba", "u", int_arm_neon_vabdu, add>;
4676 // VABAL : Vector Absolute Difference and Accumulate Long (Q += | D - D |)
4677 defm VABALs : N3VLIntExtOp_QHS<0,1,0b0101,0, IIC_VABAD,
4678 "vabal", "s", int_arm_neon_vabds, zext, add>;
4679 defm VABALu : N3VLIntExtOp_QHS<1,1,0b0101,0, IIC_VABAD,
4680 "vabal", "u", int_arm_neon_vabdu, zext, add>;
4682 // Vector Maximum and Minimum.
4684 // VMAX : Vector Maximum
4685 defm VMAXs : N3VInt_QHS<0, 0, 0b0110, 0, N3RegFrm,
4686 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4687 "vmax", "s", int_arm_neon_vmaxs, 1>;
4688 defm VMAXu : N3VInt_QHS<1, 0, 0b0110, 0, N3RegFrm,
4689 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4690 "vmax", "u", int_arm_neon_vmaxu, 1>;
4691 def VMAXfd : N3VDInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBIND,
4693 v2f32, v2f32, int_arm_neon_vmaxs, 1>;
4694 def VMAXfq : N3VQInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBINQ,
4696 v4f32, v4f32, int_arm_neon_vmaxs, 1>;
4699 let PostEncoderMethod = "NEONThumb2V8PostEncoder", DecoderNamespace = "v8NEON" in {
4700 def VMAXNMND : N3VDIntnp<0b00110, 0b00, 0b1111, 0, 1,
4701 N3RegFrm, NoItinerary, "vmaxnm", "f32",
4702 v2f32, v2f32, int_arm_neon_vmaxnm, 1>,
4703 Requires<[HasV8, HasNEON]>;
4704 def VMAXNMNQ : N3VQIntnp<0b00110, 0b00, 0b1111, 1, 1,
4705 N3RegFrm, NoItinerary, "vmaxnm", "f32",
4706 v4f32, v4f32, int_arm_neon_vmaxnm, 1>,
4707 Requires<[HasV8, HasNEON]>;
4710 // VMIN : Vector Minimum
4711 defm VMINs : N3VInt_QHS<0, 0, 0b0110, 1, N3RegFrm,
4712 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4713 "vmin", "s", int_arm_neon_vmins, 1>;
4714 defm VMINu : N3VInt_QHS<1, 0, 0b0110, 1, N3RegFrm,
4715 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4716 "vmin", "u", int_arm_neon_vminu, 1>;
4717 def VMINfd : N3VDInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBIND,
4719 v2f32, v2f32, int_arm_neon_vmins, 1>;
4720 def VMINfq : N3VQInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBINQ,
4722 v4f32, v4f32, int_arm_neon_vmins, 1>;
4725 let PostEncoderMethod = "NEONThumb2V8PostEncoder", DecoderNamespace = "v8NEON" in {
4726 def VMINNMND : N3VDIntnp<0b00110, 0b10, 0b1111, 0, 1,
4727 N3RegFrm, NoItinerary, "vminnm", "f32",
4728 v2f32, v2f32, int_arm_neon_vminnm, 1>,
4729 Requires<[HasV8, HasNEON]>;
4730 def VMINNMNQ : N3VQIntnp<0b00110, 0b10, 0b1111, 1, 1,
4731 N3RegFrm, NoItinerary, "vminnm", "f32",
4732 v4f32, v4f32, int_arm_neon_vminnm, 1>,
4733 Requires<[HasV8, HasNEON]>;
4736 // Vector Pairwise Operations.
4738 // VPADD : Vector Pairwise Add
4739 def VPADDi8 : N3VDInt<0, 0, 0b00, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4741 v8i8, v8i8, int_arm_neon_vpadd, 0>;
4742 def VPADDi16 : N3VDInt<0, 0, 0b01, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4744 v4i16, v4i16, int_arm_neon_vpadd, 0>;
4745 def VPADDi32 : N3VDInt<0, 0, 0b10, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4747 v2i32, v2i32, int_arm_neon_vpadd, 0>;
4748 def VPADDf : N3VDInt<1, 0, 0b00, 0b1101, 0, N3RegFrm,
4749 IIC_VPBIND, "vpadd", "f32",
4750 v2f32, v2f32, int_arm_neon_vpadd, 0>;
4752 // VPADDL : Vector Pairwise Add Long
4753 defm VPADDLs : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00100, 0, "vpaddl", "s",
4754 int_arm_neon_vpaddls>;
4755 defm VPADDLu : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00101, 0, "vpaddl", "u",
4756 int_arm_neon_vpaddlu>;
4758 // VPADAL : Vector Pairwise Add and Accumulate Long
4759 defm VPADALs : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01100, 0, "vpadal", "s",
4760 int_arm_neon_vpadals>;
4761 defm VPADALu : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01101, 0, "vpadal", "u",
4762 int_arm_neon_vpadalu>;
4764 // VPMAX : Vector Pairwise Maximum
4765 def VPMAXs8 : N3VDInt<0, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4766 "s8", v8i8, v8i8, int_arm_neon_vpmaxs, 0>;
4767 def VPMAXs16 : N3VDInt<0, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4768 "s16", v4i16, v4i16, int_arm_neon_vpmaxs, 0>;
4769 def VPMAXs32 : N3VDInt<0, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4770 "s32", v2i32, v2i32, int_arm_neon_vpmaxs, 0>;
4771 def VPMAXu8 : N3VDInt<1, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4772 "u8", v8i8, v8i8, int_arm_neon_vpmaxu, 0>;
4773 def VPMAXu16 : N3VDInt<1, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4774 "u16", v4i16, v4i16, int_arm_neon_vpmaxu, 0>;
4775 def VPMAXu32 : N3VDInt<1, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4776 "u32", v2i32, v2i32, int_arm_neon_vpmaxu, 0>;
4777 def VPMAXf : N3VDInt<1, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmax",
4778 "f32", v2f32, v2f32, int_arm_neon_vpmaxs, 0>;
4780 // VPMIN : Vector Pairwise Minimum
4781 def VPMINs8 : N3VDInt<0, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4782 "s8", v8i8, v8i8, int_arm_neon_vpmins, 0>;
4783 def VPMINs16 : N3VDInt<0, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4784 "s16", v4i16, v4i16, int_arm_neon_vpmins, 0>;
4785 def VPMINs32 : N3VDInt<0, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4786 "s32", v2i32, v2i32, int_arm_neon_vpmins, 0>;
4787 def VPMINu8 : N3VDInt<1, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4788 "u8", v8i8, v8i8, int_arm_neon_vpminu, 0>;
4789 def VPMINu16 : N3VDInt<1, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4790 "u16", v4i16, v4i16, int_arm_neon_vpminu, 0>;
4791 def VPMINu32 : N3VDInt<1, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4792 "u32", v2i32, v2i32, int_arm_neon_vpminu, 0>;
4793 def VPMINf : N3VDInt<1, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmin",
4794 "f32", v2f32, v2f32, int_arm_neon_vpmins, 0>;
4796 // Vector Reciprocal and Reciprocal Square Root Estimate and Step.
4798 // VRECPE : Vector Reciprocal Estimate
4799 def VRECPEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
4800 IIC_VUNAD, "vrecpe", "u32",
4801 v2i32, v2i32, int_arm_neon_vrecpe>;
4802 def VRECPEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
4803 IIC_VUNAQ, "vrecpe", "u32",
4804 v4i32, v4i32, int_arm_neon_vrecpe>;
4805 def VRECPEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
4806 IIC_VUNAD, "vrecpe", "f32",
4807 v2f32, v2f32, int_arm_neon_vrecpe>;
4808 def VRECPEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
4809 IIC_VUNAQ, "vrecpe", "f32",
4810 v4f32, v4f32, int_arm_neon_vrecpe>;
4812 // VRECPS : Vector Reciprocal Step
4813 def VRECPSfd : N3VDInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
4814 IIC_VRECSD, "vrecps", "f32",
4815 v2f32, v2f32, int_arm_neon_vrecps, 1>;
4816 def VRECPSfq : N3VQInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
4817 IIC_VRECSQ, "vrecps", "f32",
4818 v4f32, v4f32, int_arm_neon_vrecps, 1>;
4820 // VRSQRTE : Vector Reciprocal Square Root Estimate
4821 def VRSQRTEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
4822 IIC_VUNAD, "vrsqrte", "u32",
4823 v2i32, v2i32, int_arm_neon_vrsqrte>;
4824 def VRSQRTEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
4825 IIC_VUNAQ, "vrsqrte", "u32",
4826 v4i32, v4i32, int_arm_neon_vrsqrte>;
4827 def VRSQRTEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
4828 IIC_VUNAD, "vrsqrte", "f32",
4829 v2f32, v2f32, int_arm_neon_vrsqrte>;
4830 def VRSQRTEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
4831 IIC_VUNAQ, "vrsqrte", "f32",
4832 v4f32, v4f32, int_arm_neon_vrsqrte>;
4834 // VRSQRTS : Vector Reciprocal Square Root Step
4835 def VRSQRTSfd : N3VDInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
4836 IIC_VRECSD, "vrsqrts", "f32",
4837 v2f32, v2f32, int_arm_neon_vrsqrts, 1>;
4838 def VRSQRTSfq : N3VQInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
4839 IIC_VRECSQ, "vrsqrts", "f32",
4840 v4f32, v4f32, int_arm_neon_vrsqrts, 1>;
4844 // VSHL : Vector Shift
4845 defm VSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 0, N3RegVShFrm,
4846 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
4847 "vshl", "s", int_arm_neon_vshifts>;
4848 defm VSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 0, N3RegVShFrm,
4849 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
4850 "vshl", "u", int_arm_neon_vshiftu>;
4852 // VSHL : Vector Shift Left (Immediate)
4853 defm VSHLi : N2VShL_QHSD<0, 1, 0b0101, 1, IIC_VSHLiD, "vshl", "i", NEONvshl>;
4855 // VSHR : Vector Shift Right (Immediate)
4856 defm VSHRs : N2VShR_QHSD<0, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "s", "VSHRs",
4858 defm VSHRu : N2VShR_QHSD<1, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "u", "VSHRu",
4861 // VSHLL : Vector Shift Left Long
4862 defm VSHLLs : N2VLSh_QHS<0, 1, 0b1010, 0, 0, 1, "vshll", "s", NEONvshlls>;
4863 defm VSHLLu : N2VLSh_QHS<1, 1, 0b1010, 0, 0, 1, "vshll", "u", NEONvshllu>;
4865 // VSHLL : Vector Shift Left Long (with maximum shift count)
4866 class N2VLShMax<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
4867 bit op6, bit op4, string OpcodeStr, string Dt, ValueType ResTy,
4868 ValueType OpTy, Operand ImmTy, SDNode OpNode>
4869 : N2VLSh<op24, op23, op11_8, op7, op6, op4, OpcodeStr, Dt,
4870 ResTy, OpTy, ImmTy, OpNode> {
4871 let Inst{21-16} = op21_16;
4872 let DecoderMethod = "DecodeVSHLMaxInstruction";
4874 def VSHLLi8 : N2VLShMax<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll", "i8",
4875 v8i16, v8i8, imm8, NEONvshlli>;
4876 def VSHLLi16 : N2VLShMax<1, 1, 0b110110, 0b0011, 0, 0, 0, "vshll", "i16",
4877 v4i32, v4i16, imm16, NEONvshlli>;
4878 def VSHLLi32 : N2VLShMax<1, 1, 0b111010, 0b0011, 0, 0, 0, "vshll", "i32",
4879 v2i64, v2i32, imm32, NEONvshlli>;
4881 // VSHRN : Vector Shift Right and Narrow
4882 defm VSHRN : N2VNSh_HSD<0,1,0b1000,0,0,1, IIC_VSHLiD, "vshrn", "i",
4885 // VRSHL : Vector Rounding Shift
4886 defm VRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 0, N3RegVShFrm,
4887 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4888 "vrshl", "s", int_arm_neon_vrshifts>;
4889 defm VRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 0, N3RegVShFrm,
4890 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4891 "vrshl", "u", int_arm_neon_vrshiftu>;
4892 // VRSHR : Vector Rounding Shift Right
4893 defm VRSHRs : N2VShR_QHSD<0,1,0b0010,1, IIC_VSHLi4D, "vrshr", "s", "VRSHRs",
4895 defm VRSHRu : N2VShR_QHSD<1,1,0b0010,1, IIC_VSHLi4D, "vrshr", "u", "VRSHRu",
4898 // VRSHRN : Vector Rounding Shift Right and Narrow
4899 defm VRSHRN : N2VNSh_HSD<0, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vrshrn", "i",
4902 // VQSHL : Vector Saturating Shift
4903 defm VQSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 1, N3RegVShFrm,
4904 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4905 "vqshl", "s", int_arm_neon_vqshifts>;
4906 defm VQSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 1, N3RegVShFrm,
4907 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4908 "vqshl", "u", int_arm_neon_vqshiftu>;
4909 // VQSHL : Vector Saturating Shift Left (Immediate)
4910 defm VQSHLsi : N2VShL_QHSD<0,1,0b0111,1, IIC_VSHLi4D, "vqshl", "s",NEONvqshls>;
4911 defm VQSHLui : N2VShL_QHSD<1,1,0b0111,1, IIC_VSHLi4D, "vqshl", "u",NEONvqshlu>;
4913 // VQSHLU : Vector Saturating Shift Left (Immediate, Unsigned)
4914 defm VQSHLsu : N2VShL_QHSD<1,1,0b0110,1, IIC_VSHLi4D,"vqshlu","s",NEONvqshlsu>;
4916 // VQSHRN : Vector Saturating Shift Right and Narrow
4917 defm VQSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "s",
4919 defm VQSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "u",
4922 // VQSHRUN : Vector Saturating Shift Right and Narrow (Unsigned)
4923 defm VQSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 0, 1, IIC_VSHLi4D, "vqshrun", "s",
4926 // VQRSHL : Vector Saturating Rounding Shift
4927 defm VQRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 1, N3RegVShFrm,
4928 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4929 "vqrshl", "s", int_arm_neon_vqrshifts>;
4930 defm VQRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 1, N3RegVShFrm,
4931 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4932 "vqrshl", "u", int_arm_neon_vqrshiftu>;
4934 // VQRSHRN : Vector Saturating Rounding Shift Right and Narrow
4935 defm VQRSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "s",
4937 defm VQRSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "u",
4940 // VQRSHRUN : Vector Saturating Rounding Shift Right and Narrow (Unsigned)
4941 defm VQRSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vqrshrun", "s",
4944 // VSRA : Vector Shift Right and Accumulate
4945 defm VSRAs : N2VShAdd_QHSD<0, 1, 0b0001, 1, "vsra", "s", NEONvshrs>;
4946 defm VSRAu : N2VShAdd_QHSD<1, 1, 0b0001, 1, "vsra", "u", NEONvshru>;
4947 // VRSRA : Vector Rounding Shift Right and Accumulate
4948 defm VRSRAs : N2VShAdd_QHSD<0, 1, 0b0011, 1, "vrsra", "s", NEONvrshrs>;
4949 defm VRSRAu : N2VShAdd_QHSD<1, 1, 0b0011, 1, "vrsra", "u", NEONvrshru>;
4951 // VSLI : Vector Shift Left and Insert
4952 defm VSLI : N2VShInsL_QHSD<1, 1, 0b0101, 1, "vsli">;
4954 // VSRI : Vector Shift Right and Insert
4955 defm VSRI : N2VShInsR_QHSD<1, 1, 0b0100, 1, "vsri">;
4957 // Vector Absolute and Saturating Absolute.
4959 // VABS : Vector Absolute Value
4960 defm VABS : N2VInt_QHS<0b11, 0b11, 0b01, 0b00110, 0,
4961 IIC_VUNAiD, IIC_VUNAiQ, "vabs", "s",
4963 def VABSfd : N2VD<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
4965 v2f32, v2f32, fabs>;
4966 def VABSfq : N2VQ<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
4968 v4f32, v4f32, fabs>;
4970 def : Pat<(xor (v2i32 (bitconvert (v8i8 (NEONvshrs DPR:$src, (i32 7))))),
4971 (v2i32 (bitconvert (v8i8 (add DPR:$src,
4972 (NEONvshrs DPR:$src, (i32 7))))))),
4973 (VABSv8i8 DPR:$src)>;
4974 def : Pat<(xor (v2i32 (bitconvert (v4i16 (NEONvshrs DPR:$src, (i32 15))))),
4975 (v2i32 (bitconvert (v4i16 (add DPR:$src,
4976 (NEONvshrs DPR:$src, (i32 15))))))),
4977 (VABSv4i16 DPR:$src)>;
4978 def : Pat<(xor (v2i32 (NEONvshrs DPR:$src, (i32 31))),
4979 (v2i32 (add DPR:$src, (NEONvshrs DPR:$src, (i32 31))))),
4980 (VABSv2i32 DPR:$src)>;
4981 def : Pat<(xor (v4i32 (bitconvert (v16i8 (NEONvshrs QPR:$src, (i32 7))))),
4982 (v4i32 (bitconvert (v16i8 (add QPR:$src,
4983 (NEONvshrs QPR:$src, (i32 7))))))),
4984 (VABSv16i8 QPR:$src)>;
4985 def : Pat<(xor (v4i32 (bitconvert (v8i16 (NEONvshrs QPR:$src, (i32 15))))),
4986 (v4i32 (bitconvert (v8i16 (add QPR:$src,
4987 (NEONvshrs QPR:$src, (i32 15))))))),
4988 (VABSv8i16 QPR:$src)>;
4989 def : Pat<(xor (v4i32 (NEONvshrs QPR:$src, (i32 31))),
4990 (v4i32 (add QPR:$src, (NEONvshrs QPR:$src, (i32 31))))),
4991 (VABSv4i32 QPR:$src)>;
4993 def : Pat<(v2f32 (int_arm_neon_vabs (v2f32 DPR:$src))), (VABSfd DPR:$src)>;
4994 def : Pat<(v4f32 (int_arm_neon_vabs (v4f32 QPR:$src))), (VABSfq QPR:$src)>;
4996 // VQABS : Vector Saturating Absolute Value
4997 defm VQABS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01110, 0,
4998 IIC_VQUNAiD, IIC_VQUNAiQ, "vqabs", "s",
4999 int_arm_neon_vqabs>;
5003 def vnegd : PatFrag<(ops node:$in),
5004 (sub (bitconvert (v2i32 NEONimmAllZerosV)), node:$in)>;
5005 def vnegq : PatFrag<(ops node:$in),
5006 (sub (bitconvert (v4i32 NEONimmAllZerosV)), node:$in)>;
5008 class VNEGD<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
5009 : N2V<0b11, 0b11, size, 0b01, 0b00111, 0, 0, (outs DPR:$Vd), (ins DPR:$Vm),
5010 IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
5011 [(set DPR:$Vd, (Ty (vnegd DPR:$Vm)))]>;
5012 class VNEGQ<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
5013 : N2V<0b11, 0b11, size, 0b01, 0b00111, 1, 0, (outs QPR:$Vd), (ins QPR:$Vm),
5014 IIC_VSHLiQ, OpcodeStr, Dt, "$Vd, $Vm", "",
5015 [(set QPR:$Vd, (Ty (vnegq QPR:$Vm)))]>;
5017 // VNEG : Vector Negate (integer)
5018 def VNEGs8d : VNEGD<0b00, "vneg", "s8", v8i8>;
5019 def VNEGs16d : VNEGD<0b01, "vneg", "s16", v4i16>;
5020 def VNEGs32d : VNEGD<0b10, "vneg", "s32", v2i32>;
5021 def VNEGs8q : VNEGQ<0b00, "vneg", "s8", v16i8>;
5022 def VNEGs16q : VNEGQ<0b01, "vneg", "s16", v8i16>;
5023 def VNEGs32q : VNEGQ<0b10, "vneg", "s32", v4i32>;
5025 // VNEG : Vector Negate (floating-point)
5026 def VNEGfd : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
5027 (outs DPR:$Vd), (ins DPR:$Vm), IIC_VUNAD,
5028 "vneg", "f32", "$Vd, $Vm", "",
5029 [(set DPR:$Vd, (v2f32 (fneg DPR:$Vm)))]>;
5030 def VNEGf32q : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 1, 0,
5031 (outs QPR:$Vd), (ins QPR:$Vm), IIC_VUNAQ,
5032 "vneg", "f32", "$Vd, $Vm", "",
5033 [(set QPR:$Vd, (v4f32 (fneg QPR:$Vm)))]>;
5035 def : Pat<(v8i8 (vnegd DPR:$src)), (VNEGs8d DPR:$src)>;
5036 def : Pat<(v4i16 (vnegd DPR:$src)), (VNEGs16d DPR:$src)>;
5037 def : Pat<(v2i32 (vnegd DPR:$src)), (VNEGs32d DPR:$src)>;
5038 def : Pat<(v16i8 (vnegq QPR:$src)), (VNEGs8q QPR:$src)>;
5039 def : Pat<(v8i16 (vnegq QPR:$src)), (VNEGs16q QPR:$src)>;
5040 def : Pat<(v4i32 (vnegq QPR:$src)), (VNEGs32q QPR:$src)>;
5042 // VQNEG : Vector Saturating Negate
5043 defm VQNEG : N2VInt_QHS<0b11, 0b11, 0b00, 0b01111, 0,
5044 IIC_VQUNAiD, IIC_VQUNAiQ, "vqneg", "s",
5045 int_arm_neon_vqneg>;
5047 // Vector Bit Counting Operations.
5049 // VCLS : Vector Count Leading Sign Bits
5050 defm VCLS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01000, 0,
5051 IIC_VCNTiD, IIC_VCNTiQ, "vcls", "s",
5053 // VCLZ : Vector Count Leading Zeros
5054 defm VCLZ : N2VInt_QHS<0b11, 0b11, 0b00, 0b01001, 0,
5055 IIC_VCNTiD, IIC_VCNTiQ, "vclz", "i",
5057 // VCNT : Vector Count One Bits
5058 def VCNTd : N2VDInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
5059 IIC_VCNTiD, "vcnt", "8",
5061 def VCNTq : N2VQInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
5062 IIC_VCNTiQ, "vcnt", "8",
5063 v16i8, v16i8, ctpop>;
5066 def VSWPd : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 0, 0,
5067 (outs DPR:$Vd, DPR:$Vm), (ins DPR:$in1, DPR:$in2),
5068 NoItinerary, "vswp", "$Vd, $Vm", "$in1 = $Vd, $in2 = $Vm",
5070 def VSWPq : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 1, 0,
5071 (outs QPR:$Vd, QPR:$Vm), (ins QPR:$in1, QPR:$in2),
5072 NoItinerary, "vswp", "$Vd, $Vm", "$in1 = $Vd, $in2 = $Vm",
5075 // Vector Move Operations.
5077 // VMOV : Vector Move (Register)
5078 def : InstAlias<"vmov${p} $Vd, $Vm",
5079 (VORRd DPR:$Vd, DPR:$Vm, DPR:$Vm, pred:$p)>;
5080 def : InstAlias<"vmov${p} $Vd, $Vm",
5081 (VORRq QPR:$Vd, QPR:$Vm, QPR:$Vm, pred:$p)>;
5083 // VMOV : Vector Move (Immediate)
5085 let isReMaterializable = 1 in {
5086 def VMOVv8i8 : N1ModImm<1, 0b000, 0b1110, 0, 0, 0, 1, (outs DPR:$Vd),
5087 (ins nImmSplatI8:$SIMM), IIC_VMOVImm,
5088 "vmov", "i8", "$Vd, $SIMM", "",
5089 [(set DPR:$Vd, (v8i8 (NEONvmovImm timm:$SIMM)))]>;
5090 def VMOVv16i8 : N1ModImm<1, 0b000, 0b1110, 0, 1, 0, 1, (outs QPR:$Vd),
5091 (ins nImmSplatI8:$SIMM), IIC_VMOVImm,
5092 "vmov", "i8", "$Vd, $SIMM", "",
5093 [(set QPR:$Vd, (v16i8 (NEONvmovImm timm:$SIMM)))]>;
5095 def VMOVv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 0, 1, (outs DPR:$Vd),
5096 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
5097 "vmov", "i16", "$Vd, $SIMM", "",
5098 [(set DPR:$Vd, (v4i16 (NEONvmovImm timm:$SIMM)))]> {
5099 let Inst{9} = SIMM{9};
5102 def VMOVv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 0, 1, (outs QPR:$Vd),
5103 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
5104 "vmov", "i16", "$Vd, $SIMM", "",
5105 [(set QPR:$Vd, (v8i16 (NEONvmovImm timm:$SIMM)))]> {
5106 let Inst{9} = SIMM{9};
5109 def VMOVv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 0, 1, (outs DPR:$Vd),
5110 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
5111 "vmov", "i32", "$Vd, $SIMM", "",
5112 [(set DPR:$Vd, (v2i32 (NEONvmovImm timm:$SIMM)))]> {
5113 let Inst{11-8} = SIMM{11-8};
5116 def VMOVv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 0, 1, (outs QPR:$Vd),
5117 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
5118 "vmov", "i32", "$Vd, $SIMM", "",
5119 [(set QPR:$Vd, (v4i32 (NEONvmovImm timm:$SIMM)))]> {
5120 let Inst{11-8} = SIMM{11-8};
5123 def VMOVv1i64 : N1ModImm<1, 0b000, 0b1110, 0, 0, 1, 1, (outs DPR:$Vd),
5124 (ins nImmSplatI64:$SIMM), IIC_VMOVImm,
5125 "vmov", "i64", "$Vd, $SIMM", "",
5126 [(set DPR:$Vd, (v1i64 (NEONvmovImm timm:$SIMM)))]>;
5127 def VMOVv2i64 : N1ModImm<1, 0b000, 0b1110, 0, 1, 1, 1, (outs QPR:$Vd),
5128 (ins nImmSplatI64:$SIMM), IIC_VMOVImm,
5129 "vmov", "i64", "$Vd, $SIMM", "",
5130 [(set QPR:$Vd, (v2i64 (NEONvmovImm timm:$SIMM)))]>;
5132 def VMOVv2f32 : N1ModImm<1, 0b000, 0b1111, 0, 0, 0, 1, (outs DPR:$Vd),
5133 (ins nImmVMOVF32:$SIMM), IIC_VMOVImm,
5134 "vmov", "f32", "$Vd, $SIMM", "",
5135 [(set DPR:$Vd, (v2f32 (NEONvmovFPImm timm:$SIMM)))]>;
5136 def VMOVv4f32 : N1ModImm<1, 0b000, 0b1111, 0, 1, 0, 1, (outs QPR:$Vd),
5137 (ins nImmVMOVF32:$SIMM), IIC_VMOVImm,
5138 "vmov", "f32", "$Vd, $SIMM", "",
5139 [(set QPR:$Vd, (v4f32 (NEONvmovFPImm timm:$SIMM)))]>;
5140 } // isReMaterializable
5142 // VMOV : Vector Get Lane (move scalar to ARM core register)
5144 def VGETLNs8 : NVGetLane<{1,1,1,0,0,1,?,1}, 0b1011, {?,?},
5145 (outs GPR:$R), (ins DPR:$V, VectorIndex8:$lane),
5146 IIC_VMOVSI, "vmov", "s8", "$R, $V$lane",
5147 [(set GPR:$R, (NEONvgetlanes (v8i8 DPR:$V),
5149 let Inst{21} = lane{2};
5150 let Inst{6-5} = lane{1-0};
5152 def VGETLNs16 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, {?,1},
5153 (outs GPR:$R), (ins DPR:$V, VectorIndex16:$lane),
5154 IIC_VMOVSI, "vmov", "s16", "$R, $V$lane",
5155 [(set GPR:$R, (NEONvgetlanes (v4i16 DPR:$V),
5157 let Inst{21} = lane{1};
5158 let Inst{6} = lane{0};
5160 def VGETLNu8 : NVGetLane<{1,1,1,0,1,1,?,1}, 0b1011, {?,?},
5161 (outs GPR:$R), (ins DPR:$V, VectorIndex8:$lane),
5162 IIC_VMOVSI, "vmov", "u8", "$R, $V$lane",
5163 [(set GPR:$R, (NEONvgetlaneu (v8i8 DPR:$V),
5165 let Inst{21} = lane{2};
5166 let Inst{6-5} = lane{1-0};
5168 def VGETLNu16 : NVGetLane<{1,1,1,0,1,0,?,1}, 0b1011, {?,1},
5169 (outs GPR:$R), (ins DPR:$V, VectorIndex16:$lane),
5170 IIC_VMOVSI, "vmov", "u16", "$R, $V$lane",
5171 [(set GPR:$R, (NEONvgetlaneu (v4i16 DPR:$V),
5173 let Inst{21} = lane{1};
5174 let Inst{6} = lane{0};
5176 def VGETLNi32 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, 0b00,
5177 (outs GPR:$R), (ins DPR:$V, VectorIndex32:$lane),
5178 IIC_VMOVSI, "vmov", "32", "$R, $V$lane",
5179 [(set GPR:$R, (extractelt (v2i32 DPR:$V),
5181 Requires<[HasNEON, HasFastVGETLNi32]> {
5182 let Inst{21} = lane{0};
5184 // def VGETLNf32: see FMRDH and FMRDL in ARMInstrVFP.td
5185 def : Pat<(NEONvgetlanes (v16i8 QPR:$src), imm:$lane),
5186 (VGETLNs8 (v8i8 (EXTRACT_SUBREG QPR:$src,
5187 (DSubReg_i8_reg imm:$lane))),
5188 (SubReg_i8_lane imm:$lane))>;
5189 def : Pat<(NEONvgetlanes (v8i16 QPR:$src), imm:$lane),
5190 (VGETLNs16 (v4i16 (EXTRACT_SUBREG QPR:$src,
5191 (DSubReg_i16_reg imm:$lane))),
5192 (SubReg_i16_lane imm:$lane))>;
5193 def : Pat<(NEONvgetlaneu (v16i8 QPR:$src), imm:$lane),
5194 (VGETLNu8 (v8i8 (EXTRACT_SUBREG QPR:$src,
5195 (DSubReg_i8_reg imm:$lane))),
5196 (SubReg_i8_lane imm:$lane))>;
5197 def : Pat<(NEONvgetlaneu (v8i16 QPR:$src), imm:$lane),
5198 (VGETLNu16 (v4i16 (EXTRACT_SUBREG QPR:$src,
5199 (DSubReg_i16_reg imm:$lane))),
5200 (SubReg_i16_lane imm:$lane))>;
5201 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
5202 (VGETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src,
5203 (DSubReg_i32_reg imm:$lane))),
5204 (SubReg_i32_lane imm:$lane))>,
5205 Requires<[HasNEON, HasFastVGETLNi32]>;
5206 def : Pat<(extractelt (v2i32 DPR:$src), imm:$lane),
5208 (i32 (EXTRACT_SUBREG DPR:$src, (SSubReg_f32_reg imm:$lane))), GPR)>,
5209 Requires<[HasNEON, HasSlowVGETLNi32]>;
5210 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
5212 (i32 (EXTRACT_SUBREG QPR:$src, (SSubReg_f32_reg imm:$lane))), GPR)>,
5213 Requires<[HasNEON, HasSlowVGETLNi32]>;
5214 def : Pat<(extractelt (v2f32 DPR:$src1), imm:$src2),
5215 (EXTRACT_SUBREG (v2f32 (COPY_TO_REGCLASS (v2f32 DPR:$src1),DPR_VFP2)),
5216 (SSubReg_f32_reg imm:$src2))>;
5217 def : Pat<(extractelt (v4f32 QPR:$src1), imm:$src2),
5218 (EXTRACT_SUBREG (v4f32 (COPY_TO_REGCLASS (v4f32 QPR:$src1),QPR_VFP2)),
5219 (SSubReg_f32_reg imm:$src2))>;
5220 //def : Pat<(extractelt (v2i64 QPR:$src1), imm:$src2),
5221 // (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
5222 def : Pat<(extractelt (v2f64 QPR:$src1), imm:$src2),
5223 (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
5226 // VMOV : Vector Set Lane (move ARM core register to scalar)
5228 let Constraints = "$src1 = $V" in {
5229 def VSETLNi8 : NVSetLane<{1,1,1,0,0,1,?,0}, 0b1011, {?,?}, (outs DPR:$V),
5230 (ins DPR:$src1, GPR:$R, VectorIndex8:$lane),
5231 IIC_VMOVISL, "vmov", "8", "$V$lane, $R",
5232 [(set DPR:$V, (vector_insert (v8i8 DPR:$src1),
5233 GPR:$R, imm:$lane))]> {
5234 let Inst{21} = lane{2};
5235 let Inst{6-5} = lane{1-0};
5237 def VSETLNi16 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, {?,1}, (outs DPR:$V),
5238 (ins DPR:$src1, GPR:$R, VectorIndex16:$lane),
5239 IIC_VMOVISL, "vmov", "16", "$V$lane, $R",
5240 [(set DPR:$V, (vector_insert (v4i16 DPR:$src1),
5241 GPR:$R, imm:$lane))]> {
5242 let Inst{21} = lane{1};
5243 let Inst{6} = lane{0};
5245 def VSETLNi32 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, 0b00, (outs DPR:$V),
5246 (ins DPR:$src1, GPR:$R, VectorIndex32:$lane),
5247 IIC_VMOVISL, "vmov", "32", "$V$lane, $R",
5248 [(set DPR:$V, (insertelt (v2i32 DPR:$src1),
5249 GPR:$R, imm:$lane))]> {
5250 let Inst{21} = lane{0};
5253 def : Pat<(vector_insert (v16i8 QPR:$src1), GPR:$src2, imm:$lane),
5254 (v16i8 (INSERT_SUBREG QPR:$src1,
5255 (v8i8 (VSETLNi8 (v8i8 (EXTRACT_SUBREG QPR:$src1,
5256 (DSubReg_i8_reg imm:$lane))),
5257 GPR:$src2, (SubReg_i8_lane imm:$lane))),
5258 (DSubReg_i8_reg imm:$lane)))>;
5259 def : Pat<(vector_insert (v8i16 QPR:$src1), GPR:$src2, imm:$lane),
5260 (v8i16 (INSERT_SUBREG QPR:$src1,
5261 (v4i16 (VSETLNi16 (v4i16 (EXTRACT_SUBREG QPR:$src1,
5262 (DSubReg_i16_reg imm:$lane))),
5263 GPR:$src2, (SubReg_i16_lane imm:$lane))),
5264 (DSubReg_i16_reg imm:$lane)))>;
5265 def : Pat<(insertelt (v4i32 QPR:$src1), GPR:$src2, imm:$lane),
5266 (v4i32 (INSERT_SUBREG QPR:$src1,
5267 (v2i32 (VSETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src1,
5268 (DSubReg_i32_reg imm:$lane))),
5269 GPR:$src2, (SubReg_i32_lane imm:$lane))),
5270 (DSubReg_i32_reg imm:$lane)))>;
5272 def : Pat<(v2f32 (insertelt DPR:$src1, SPR:$src2, imm:$src3)),
5273 (INSERT_SUBREG (v2f32 (COPY_TO_REGCLASS DPR:$src1, DPR_VFP2)),
5274 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
5275 def : Pat<(v4f32 (insertelt QPR:$src1, SPR:$src2, imm:$src3)),
5276 (INSERT_SUBREG (v4f32 (COPY_TO_REGCLASS QPR:$src1, QPR_VFP2)),
5277 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
5279 //def : Pat<(v2i64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
5280 // (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
5281 def : Pat<(v2f64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
5282 (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
5284 def : Pat<(v2f32 (scalar_to_vector SPR:$src)),
5285 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
5286 def : Pat<(v2f64 (scalar_to_vector (f64 DPR:$src))),
5287 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
5288 def : Pat<(v4f32 (scalar_to_vector SPR:$src)),
5289 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
5291 def : Pat<(v8i8 (scalar_to_vector GPR:$src)),
5292 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
5293 def : Pat<(v4i16 (scalar_to_vector GPR:$src)),
5294 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
5295 def : Pat<(v2i32 (scalar_to_vector GPR:$src)),
5296 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
5298 def : Pat<(v16i8 (scalar_to_vector GPR:$src)),
5299 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
5300 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
5302 def : Pat<(v8i16 (scalar_to_vector GPR:$src)),
5303 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
5304 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
5306 def : Pat<(v4i32 (scalar_to_vector GPR:$src)),
5307 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
5308 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
5311 // VDUP : Vector Duplicate (from ARM core register to all elements)
5313 class VDUPD<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
5314 : NVDup<opcod1, 0b1011, opcod3, (outs DPR:$V), (ins GPR:$R),
5315 IIC_VMOVIS, "vdup", Dt, "$V, $R",
5316 [(set DPR:$V, (Ty (NEONvdup (i32 GPR:$R))))]>;
5317 class VDUPQ<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
5318 : NVDup<opcod1, 0b1011, opcod3, (outs QPR:$V), (ins GPR:$R),
5319 IIC_VMOVIS, "vdup", Dt, "$V, $R",
5320 [(set QPR:$V, (Ty (NEONvdup (i32 GPR:$R))))]>;
5322 def VDUP8d : VDUPD<0b11101100, 0b00, "8", v8i8>;
5323 def VDUP16d : VDUPD<0b11101000, 0b01, "16", v4i16>;
5324 def VDUP32d : VDUPD<0b11101000, 0b00, "32", v2i32>,
5325 Requires<[HasNEON, HasFastVDUP32]>;
5326 def VDUP8q : VDUPQ<0b11101110, 0b00, "8", v16i8>;
5327 def VDUP16q : VDUPQ<0b11101010, 0b01, "16", v8i16>;
5328 def VDUP32q : VDUPQ<0b11101010, 0b00, "32", v4i32>;
5330 // NEONvdup patterns for uarchs with fast VDUP.32.
5331 def : Pat<(v2f32 (NEONvdup (f32 (bitconvert GPR:$R)))), (VDUP32d GPR:$R)>,
5332 Requires<[HasNEON,HasFastVDUP32]>;
5333 def : Pat<(v4f32 (NEONvdup (f32 (bitconvert GPR:$R)))), (VDUP32q GPR:$R)>;
5335 // NEONvdup patterns for uarchs with slow VDUP.32 - use VMOVDRR instead.
5336 def : Pat<(v2i32 (NEONvdup (i32 GPR:$R))), (VMOVDRR GPR:$R, GPR:$R)>,
5337 Requires<[HasNEON,HasSlowVDUP32]>;
5338 def : Pat<(v2f32 (NEONvdup (f32 (bitconvert GPR:$R)))), (VMOVDRR GPR:$R, GPR:$R)>,
5339 Requires<[HasNEON,HasSlowVDUP32]>;
5341 // VDUP : Vector Duplicate Lane (from scalar to all elements)
5343 class VDUPLND<bits<4> op19_16, string OpcodeStr, string Dt,
5344 ValueType Ty, Operand IdxTy>
5345 : NVDupLane<op19_16, 0, (outs DPR:$Vd), (ins DPR:$Vm, IdxTy:$lane),
5346 IIC_VMOVD, OpcodeStr, Dt, "$Vd, $Vm$lane",
5347 [(set DPR:$Vd, (Ty (NEONvduplane (Ty DPR:$Vm), imm:$lane)))]>;
5349 class VDUPLNQ<bits<4> op19_16, string OpcodeStr, string Dt,
5350 ValueType ResTy, ValueType OpTy, Operand IdxTy>
5351 : NVDupLane<op19_16, 1, (outs QPR:$Vd), (ins DPR:$Vm, IdxTy:$lane),
5352 IIC_VMOVQ, OpcodeStr, Dt, "$Vd, $Vm$lane",
5353 [(set QPR:$Vd, (ResTy (NEONvduplane (OpTy DPR:$Vm),
5354 VectorIndex32:$lane)))]>;
5356 // Inst{19-16} is partially specified depending on the element size.
5358 def VDUPLN8d : VDUPLND<{?,?,?,1}, "vdup", "8", v8i8, VectorIndex8> {
5360 let Inst{19-17} = lane{2-0};
5362 def VDUPLN16d : VDUPLND<{?,?,1,0}, "vdup", "16", v4i16, VectorIndex16> {
5364 let Inst{19-18} = lane{1-0};
5366 def VDUPLN32d : VDUPLND<{?,1,0,0}, "vdup", "32", v2i32, VectorIndex32> {
5368 let Inst{19} = lane{0};
5370 def VDUPLN8q : VDUPLNQ<{?,?,?,1}, "vdup", "8", v16i8, v8i8, VectorIndex8> {
5372 let Inst{19-17} = lane{2-0};
5374 def VDUPLN16q : VDUPLNQ<{?,?,1,0}, "vdup", "16", v8i16, v4i16, VectorIndex16> {
5376 let Inst{19-18} = lane{1-0};
5378 def VDUPLN32q : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4i32, v2i32, VectorIndex32> {
5380 let Inst{19} = lane{0};
5383 def : Pat<(v2f32 (NEONvduplane (v2f32 DPR:$Vm), imm:$lane)),
5384 (VDUPLN32d DPR:$Vm, imm:$lane)>;
5386 def : Pat<(v4f32 (NEONvduplane (v2f32 DPR:$Vm), imm:$lane)),
5387 (VDUPLN32q DPR:$Vm, imm:$lane)>;
5389 def : Pat<(v16i8 (NEONvduplane (v16i8 QPR:$src), imm:$lane)),
5390 (v16i8 (VDUPLN8q (v8i8 (EXTRACT_SUBREG QPR:$src,
5391 (DSubReg_i8_reg imm:$lane))),
5392 (SubReg_i8_lane imm:$lane)))>;
5393 def : Pat<(v8i16 (NEONvduplane (v8i16 QPR:$src), imm:$lane)),
5394 (v8i16 (VDUPLN16q (v4i16 (EXTRACT_SUBREG QPR:$src,
5395 (DSubReg_i16_reg imm:$lane))),
5396 (SubReg_i16_lane imm:$lane)))>;
5397 def : Pat<(v4i32 (NEONvduplane (v4i32 QPR:$src), imm:$lane)),
5398 (v4i32 (VDUPLN32q (v2i32 (EXTRACT_SUBREG QPR:$src,
5399 (DSubReg_i32_reg imm:$lane))),
5400 (SubReg_i32_lane imm:$lane)))>;
5401 def : Pat<(v4f32 (NEONvduplane (v4f32 QPR:$src), imm:$lane)),
5402 (v4f32 (VDUPLN32q (v2f32 (EXTRACT_SUBREG QPR:$src,
5403 (DSubReg_i32_reg imm:$lane))),
5404 (SubReg_i32_lane imm:$lane)))>;
5406 def VDUPfdf : PseudoNeonI<(outs DPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
5407 [(set DPR:$dst, (v2f32 (NEONvdup (f32 SPR:$src))))]>;
5408 def VDUPfqf : PseudoNeonI<(outs QPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
5409 [(set QPR:$dst, (v4f32 (NEONvdup (f32 SPR:$src))))]>;
5411 // VMOVN : Vector Narrowing Move
5412 defm VMOVN : N2VN_HSD<0b11,0b11,0b10,0b00100,0,0, IIC_VMOVN,
5413 "vmovn", "i", trunc>;
5414 // VQMOVN : Vector Saturating Narrowing Move
5415 defm VQMOVNs : N2VNInt_HSD<0b11,0b11,0b10,0b00101,0,0, IIC_VQUNAiD,
5416 "vqmovn", "s", int_arm_neon_vqmovns>;
5417 defm VQMOVNu : N2VNInt_HSD<0b11,0b11,0b10,0b00101,1,0, IIC_VQUNAiD,
5418 "vqmovn", "u", int_arm_neon_vqmovnu>;
5419 defm VQMOVNsu : N2VNInt_HSD<0b11,0b11,0b10,0b00100,1,0, IIC_VQUNAiD,
5420 "vqmovun", "s", int_arm_neon_vqmovnsu>;
5421 // VMOVL : Vector Lengthening Move
5422 defm VMOVLs : N2VL_QHS<0b01,0b10100,0,1, "vmovl", "s", sext>;
5423 defm VMOVLu : N2VL_QHS<0b11,0b10100,0,1, "vmovl", "u", zext>;
5424 def : Pat<(v8i16 (anyext (v8i8 DPR:$Vm))), (VMOVLuv8i16 DPR:$Vm)>;
5425 def : Pat<(v4i32 (anyext (v4i16 DPR:$Vm))), (VMOVLuv4i32 DPR:$Vm)>;
5426 def : Pat<(v2i64 (anyext (v2i32 DPR:$Vm))), (VMOVLuv2i64 DPR:$Vm)>;
5428 // Vector Conversions.
5430 // VCVT : Vector Convert Between Floating-Point and Integers
5431 def VCVTf2sd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
5432 v2i32, v2f32, fp_to_sint>;
5433 def VCVTf2ud : N2VD<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
5434 v2i32, v2f32, fp_to_uint>;
5435 def VCVTs2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
5436 v2f32, v2i32, sint_to_fp>;
5437 def VCVTu2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
5438 v2f32, v2i32, uint_to_fp>;
5440 def VCVTf2sq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
5441 v4i32, v4f32, fp_to_sint>;
5442 def VCVTf2uq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
5443 v4i32, v4f32, fp_to_uint>;
5444 def VCVTs2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
5445 v4f32, v4i32, sint_to_fp>;
5446 def VCVTu2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
5447 v4f32, v4i32, uint_to_fp>;
5450 multiclass VCVT_FPI<string op, bits<3> op10_8, SDPatternOperator IntS,
5451 SDPatternOperator IntU> {
5452 let PostEncoderMethod = "NEONThumb2V8PostEncoder", DecoderNamespace = "v8NEON" in {
5453 def SD : N2VDIntnp<0b11, op10_8, 0, NoItinerary, !strconcat("vcvt", op),
5454 "s32.f32", v2i32, v2f32, IntS>, Requires<[HasV8, HasNEON]>;
5455 def SQ : N2VQIntnp<0b11, op10_8, 0, NoItinerary, !strconcat("vcvt", op),
5456 "s32.f32", v4i32, v4f32, IntS>, Requires<[HasV8, HasNEON]>;
5457 def UD : N2VDIntnp<0b11, op10_8, 1, NoItinerary, !strconcat("vcvt", op),
5458 "u32.f32", v2i32, v2f32, IntU>, Requires<[HasV8, HasNEON]>;
5459 def UQ : N2VQIntnp<0b11, op10_8, 1, NoItinerary, !strconcat("vcvt", op),
5460 "u32.f32", v4i32, v4f32, IntU>, Requires<[HasV8, HasNEON]>;
5464 defm VCVTAN : VCVT_FPI<"a", 0b000, int_arm_neon_vcvtas, int_arm_neon_vcvtau>;
5465 defm VCVTNN : VCVT_FPI<"n", 0b001, int_arm_neon_vcvtns, int_arm_neon_vcvtnu>;
5466 defm VCVTPN : VCVT_FPI<"p", 0b010, int_arm_neon_vcvtps, int_arm_neon_vcvtpu>;
5467 defm VCVTMN : VCVT_FPI<"m", 0b011, int_arm_neon_vcvtms, int_arm_neon_vcvtmu>;
5469 // VCVT : Vector Convert Between Floating-Point and Fixed-Point.
5470 let DecoderMethod = "DecodeVCVTD" in {
5471 def VCVTf2xsd : N2VCvtD<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
5472 v2i32, v2f32, int_arm_neon_vcvtfp2fxs>;
5473 def VCVTf2xud : N2VCvtD<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
5474 v2i32, v2f32, int_arm_neon_vcvtfp2fxu>;
5475 def VCVTxs2fd : N2VCvtD<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
5476 v2f32, v2i32, int_arm_neon_vcvtfxs2fp>;
5477 def VCVTxu2fd : N2VCvtD<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
5478 v2f32, v2i32, int_arm_neon_vcvtfxu2fp>;
5481 let DecoderMethod = "DecodeVCVTQ" in {
5482 def VCVTf2xsq : N2VCvtQ<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
5483 v4i32, v4f32, int_arm_neon_vcvtfp2fxs>;
5484 def VCVTf2xuq : N2VCvtQ<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
5485 v4i32, v4f32, int_arm_neon_vcvtfp2fxu>;
5486 def VCVTxs2fq : N2VCvtQ<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
5487 v4f32, v4i32, int_arm_neon_vcvtfxs2fp>;
5488 def VCVTxu2fq : N2VCvtQ<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
5489 v4f32, v4i32, int_arm_neon_vcvtfxu2fp>;
5492 // VCVT : Vector Convert Between Half-Precision and Single-Precision.
5493 def VCVTf2h : N2VNInt<0b11, 0b11, 0b01, 0b10, 0b01100, 0, 0,
5494 IIC_VUNAQ, "vcvt", "f16.f32",
5495 v4i16, v4f32, int_arm_neon_vcvtfp2hf>,
5496 Requires<[HasNEON, HasFP16]>;
5497 def VCVTh2f : N2VLInt<0b11, 0b11, 0b01, 0b10, 0b01110, 0, 0,
5498 IIC_VUNAQ, "vcvt", "f32.f16",
5499 v4f32, v4i16, int_arm_neon_vcvthf2fp>,
5500 Requires<[HasNEON, HasFP16]>;
5504 // VREV64 : Vector Reverse elements within 64-bit doublewords
5506 class VREV64D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
5507 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 0, 0, (outs DPR:$Vd),
5508 (ins DPR:$Vm), IIC_VMOVD,
5509 OpcodeStr, Dt, "$Vd, $Vm", "",
5510 [(set DPR:$Vd, (Ty (NEONvrev64 (Ty DPR:$Vm))))]>;
5511 class VREV64Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
5512 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 1, 0, (outs QPR:$Vd),
5513 (ins QPR:$Vm), IIC_VMOVQ,
5514 OpcodeStr, Dt, "$Vd, $Vm", "",
5515 [(set QPR:$Vd, (Ty (NEONvrev64 (Ty QPR:$Vm))))]>;
5517 def VREV64d8 : VREV64D<0b00, "vrev64", "8", v8i8>;
5518 def VREV64d16 : VREV64D<0b01, "vrev64", "16", v4i16>;
5519 def VREV64d32 : VREV64D<0b10, "vrev64", "32", v2i32>;
5520 def : Pat<(v2f32 (NEONvrev64 (v2f32 DPR:$Vm))), (VREV64d32 DPR:$Vm)>;
5522 def VREV64q8 : VREV64Q<0b00, "vrev64", "8", v16i8>;
5523 def VREV64q16 : VREV64Q<0b01, "vrev64", "16", v8i16>;
5524 def VREV64q32 : VREV64Q<0b10, "vrev64", "32", v4i32>;
5525 def : Pat<(v4f32 (NEONvrev64 (v4f32 QPR:$Vm))), (VREV64q32 QPR:$Vm)>;
5527 // VREV32 : Vector Reverse elements within 32-bit words
5529 class VREV32D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
5530 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 0, 0, (outs DPR:$Vd),
5531 (ins DPR:$Vm), IIC_VMOVD,
5532 OpcodeStr, Dt, "$Vd, $Vm", "",
5533 [(set DPR:$Vd, (Ty (NEONvrev32 (Ty DPR:$Vm))))]>;
5534 class VREV32Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
5535 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 1, 0, (outs QPR:$Vd),
5536 (ins QPR:$Vm), IIC_VMOVQ,
5537 OpcodeStr, Dt, "$Vd, $Vm", "",
5538 [(set QPR:$Vd, (Ty (NEONvrev32 (Ty QPR:$Vm))))]>;
5540 def VREV32d8 : VREV32D<0b00, "vrev32", "8", v8i8>;
5541 def VREV32d16 : VREV32D<0b01, "vrev32", "16", v4i16>;
5543 def VREV32q8 : VREV32Q<0b00, "vrev32", "8", v16i8>;
5544 def VREV32q16 : VREV32Q<0b01, "vrev32", "16", v8i16>;
5546 // VREV16 : Vector Reverse elements within 16-bit halfwords
5548 class VREV16D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
5549 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 0, 0, (outs DPR:$Vd),
5550 (ins DPR:$Vm), IIC_VMOVD,
5551 OpcodeStr, Dt, "$Vd, $Vm", "",
5552 [(set DPR:$Vd, (Ty (NEONvrev16 (Ty DPR:$Vm))))]>;
5553 class VREV16Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
5554 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 1, 0, (outs QPR:$Vd),
5555 (ins QPR:$Vm), IIC_VMOVQ,
5556 OpcodeStr, Dt, "$Vd, $Vm", "",
5557 [(set QPR:$Vd, (Ty (NEONvrev16 (Ty QPR:$Vm))))]>;
5559 def VREV16d8 : VREV16D<0b00, "vrev16", "8", v8i8>;
5560 def VREV16q8 : VREV16Q<0b00, "vrev16", "8", v16i8>;
5562 // Other Vector Shuffles.
5564 // Aligned extractions: really just dropping registers
5566 class AlignedVEXTq<ValueType DestTy, ValueType SrcTy, SDNodeXForm LaneCVT>
5567 : Pat<(DestTy (vector_extract_subvec (SrcTy QPR:$src), (i32 imm:$start))),
5568 (EXTRACT_SUBREG (SrcTy QPR:$src), (LaneCVT imm:$start))>;
5570 def : AlignedVEXTq<v8i8, v16i8, DSubReg_i8_reg>;
5572 def : AlignedVEXTq<v4i16, v8i16, DSubReg_i16_reg>;
5574 def : AlignedVEXTq<v2i32, v4i32, DSubReg_i32_reg>;
5576 def : AlignedVEXTq<v1i64, v2i64, DSubReg_f64_reg>;
5578 def : AlignedVEXTq<v2f32, v4f32, DSubReg_i32_reg>;
5581 // VEXT : Vector Extract
5584 // All of these have a two-operand InstAlias.
5585 let TwoOperandAliasConstraint = "$Vn = $Vd" in {
5586 class VEXTd<string OpcodeStr, string Dt, ValueType Ty, Operand immTy>
5587 : N3V<0,1,0b11,{?,?,?,?},0,0, (outs DPR:$Vd),
5588 (ins DPR:$Vn, DPR:$Vm, immTy:$index), NVExtFrm,
5589 IIC_VEXTD, OpcodeStr, Dt, "$Vd, $Vn, $Vm, $index", "",
5590 [(set DPR:$Vd, (Ty (NEONvext (Ty DPR:$Vn),
5591 (Ty DPR:$Vm), imm:$index)))]> {
5594 let Inst{10-8} = index{2-0};
5597 class VEXTq<string OpcodeStr, string Dt, ValueType Ty, Operand immTy>
5598 : N3V<0,1,0b11,{?,?,?,?},1,0, (outs QPR:$Vd),
5599 (ins QPR:$Vn, QPR:$Vm, imm0_15:$index), NVExtFrm,
5600 IIC_VEXTQ, OpcodeStr, Dt, "$Vd, $Vn, $Vm, $index", "",
5601 [(set QPR:$Vd, (Ty (NEONvext (Ty QPR:$Vn),
5602 (Ty QPR:$Vm), imm:$index)))]> {
5604 let Inst{11-8} = index{3-0};
5608 def VEXTd8 : VEXTd<"vext", "8", v8i8, imm0_7> {
5609 let Inst{10-8} = index{2-0};
5611 def VEXTd16 : VEXTd<"vext", "16", v4i16, imm0_3> {
5612 let Inst{10-9} = index{1-0};
5615 def VEXTd32 : VEXTd<"vext", "32", v2i32, imm0_1> {
5616 let Inst{10} = index{0};
5617 let Inst{9-8} = 0b00;
5619 def : Pat<(v2f32 (NEONvext (v2f32 DPR:$Vn),
5622 (VEXTd32 DPR:$Vn, DPR:$Vm, imm:$index)>;
5624 def VEXTq8 : VEXTq<"vext", "8", v16i8, imm0_15> {
5625 let Inst{11-8} = index{3-0};
5627 def VEXTq16 : VEXTq<"vext", "16", v8i16, imm0_7> {
5628 let Inst{11-9} = index{2-0};
5631 def VEXTq32 : VEXTq<"vext", "32", v4i32, imm0_3> {
5632 let Inst{11-10} = index{1-0};
5633 let Inst{9-8} = 0b00;
5635 def VEXTq64 : VEXTq<"vext", "64", v2i64, imm0_1> {
5636 let Inst{11} = index{0};
5637 let Inst{10-8} = 0b000;
5639 def : Pat<(v4f32 (NEONvext (v4f32 QPR:$Vn),
5642 (VEXTq32 QPR:$Vn, QPR:$Vm, imm:$index)>;
5644 // VTRN : Vector Transpose
5646 def VTRNd8 : N2VDShuffle<0b00, 0b00001, "vtrn", "8">;
5647 def VTRNd16 : N2VDShuffle<0b01, 0b00001, "vtrn", "16">;
5648 def VTRNd32 : N2VDShuffle<0b10, 0b00001, "vtrn", "32">;
5650 def VTRNq8 : N2VQShuffle<0b00, 0b00001, IIC_VPERMQ, "vtrn", "8">;
5651 def VTRNq16 : N2VQShuffle<0b01, 0b00001, IIC_VPERMQ, "vtrn", "16">;
5652 def VTRNq32 : N2VQShuffle<0b10, 0b00001, IIC_VPERMQ, "vtrn", "32">;
5654 // VUZP : Vector Unzip (Deinterleave)
5656 def VUZPd8 : N2VDShuffle<0b00, 0b00010, "vuzp", "8">;
5657 def VUZPd16 : N2VDShuffle<0b01, 0b00010, "vuzp", "16">;
5658 // vuzp.32 Dd, Dm is a pseudo-instruction expanded to vtrn.32 Dd, Dm.
5659 def : NEONInstAlias<"vuzp${p}.32 $Dd, $Dm",
5660 (VTRNd32 DPR:$Dd, DPR:$Dm, pred:$p)>;
5662 def VUZPq8 : N2VQShuffle<0b00, 0b00010, IIC_VPERMQ3, "vuzp", "8">;
5663 def VUZPq16 : N2VQShuffle<0b01, 0b00010, IIC_VPERMQ3, "vuzp", "16">;
5664 def VUZPq32 : N2VQShuffle<0b10, 0b00010, IIC_VPERMQ3, "vuzp", "32">;
5666 // VZIP : Vector Zip (Interleave)
5668 def VZIPd8 : N2VDShuffle<0b00, 0b00011, "vzip", "8">;
5669 def VZIPd16 : N2VDShuffle<0b01, 0b00011, "vzip", "16">;
5670 // vzip.32 Dd, Dm is a pseudo-instruction expanded to vtrn.32 Dd, Dm.
5671 def : NEONInstAlias<"vzip${p}.32 $Dd, $Dm",
5672 (VTRNd32 DPR:$Dd, DPR:$Dm, pred:$p)>;
5674 def VZIPq8 : N2VQShuffle<0b00, 0b00011, IIC_VPERMQ3, "vzip", "8">;
5675 def VZIPq16 : N2VQShuffle<0b01, 0b00011, IIC_VPERMQ3, "vzip", "16">;
5676 def VZIPq32 : N2VQShuffle<0b10, 0b00011, IIC_VPERMQ3, "vzip", "32">;
5678 // Vector Table Lookup and Table Extension.
5680 // VTBL : Vector Table Lookup
5681 let DecoderMethod = "DecodeTBLInstruction" in {
5683 : N3V<1,1,0b11,0b1000,0,0, (outs DPR:$Vd),
5684 (ins VecListOneD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTB1,
5685 "vtbl", "8", "$Vd, $Vn, $Vm", "",
5686 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbl1 VecListOneD:$Vn, DPR:$Vm)))]>;
5687 let hasExtraSrcRegAllocReq = 1 in {
5689 : N3V<1,1,0b11,0b1001,0,0, (outs DPR:$Vd),
5690 (ins VecListDPair:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTB2,
5691 "vtbl", "8", "$Vd, $Vn, $Vm", "", []>;
5693 : N3V<1,1,0b11,0b1010,0,0, (outs DPR:$Vd),
5694 (ins VecListThreeD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTB3,
5695 "vtbl", "8", "$Vd, $Vn, $Vm", "", []>;
5697 : N3V<1,1,0b11,0b1011,0,0, (outs DPR:$Vd),
5698 (ins VecListFourD:$Vn, DPR:$Vm),
5700 "vtbl", "8", "$Vd, $Vn, $Vm", "", []>;
5701 } // hasExtraSrcRegAllocReq = 1
5704 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB3, "", []>;
5706 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB4, "", []>;
5708 // VTBX : Vector Table Extension
5710 : N3V<1,1,0b11,0b1000,1,0, (outs DPR:$Vd),
5711 (ins DPR:$orig, VecListOneD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTBX1,
5712 "vtbx", "8", "$Vd, $Vn, $Vm", "$orig = $Vd",
5713 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbx1
5714 DPR:$orig, VecListOneD:$Vn, DPR:$Vm)))]>;
5715 let hasExtraSrcRegAllocReq = 1 in {
5717 : N3V<1,1,0b11,0b1001,1,0, (outs DPR:$Vd),
5718 (ins DPR:$orig, VecListDPair:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTBX2,
5719 "vtbx", "8", "$Vd, $Vn, $Vm", "$orig = $Vd", []>;
5721 : N3V<1,1,0b11,0b1010,1,0, (outs DPR:$Vd),
5722 (ins DPR:$orig, VecListThreeD:$Vn, DPR:$Vm),
5723 NVTBLFrm, IIC_VTBX3,
5724 "vtbx", "8", "$Vd, $Vn, $Vm",
5727 : N3V<1,1,0b11,0b1011,1,0, (outs DPR:$Vd),
5728 (ins DPR:$orig, VecListFourD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTBX4,
5729 "vtbx", "8", "$Vd, $Vn, $Vm",
5731 } // hasExtraSrcRegAllocReq = 1
5734 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
5735 IIC_VTBX3, "$orig = $dst", []>;
5737 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
5738 IIC_VTBX4, "$orig = $dst", []>;
5739 } // DecoderMethod = "DecodeTBLInstruction"
5741 //===----------------------------------------------------------------------===//
5742 // NEON instructions for single-precision FP math
5743 //===----------------------------------------------------------------------===//
5745 class N2VSPat<SDNode OpNode, NeonI Inst>
5746 : NEONFPPat<(f32 (OpNode SPR:$a)),
5748 (v2f32 (COPY_TO_REGCLASS (Inst
5750 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5751 SPR:$a, ssub_0)), DPR_VFP2)), ssub_0)>;
5753 class N3VSPat<SDNode OpNode, NeonI Inst>
5754 : NEONFPPat<(f32 (OpNode SPR:$a, SPR:$b)),
5756 (v2f32 (COPY_TO_REGCLASS (Inst
5758 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5761 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5762 SPR:$b, ssub_0)), DPR_VFP2)), ssub_0)>;
5764 class N3VSMulOpPat<SDNode MulNode, SDNode OpNode, NeonI Inst>
5765 : NEONFPPat<(f32 (OpNode SPR:$acc, (f32 (MulNode SPR:$a, SPR:$b)))),
5767 (v2f32 (COPY_TO_REGCLASS (Inst
5769 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5772 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5775 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5776 SPR:$b, ssub_0)), DPR_VFP2)), ssub_0)>;
5778 def : N3VSPat<fadd, VADDfd>;
5779 def : N3VSPat<fsub, VSUBfd>;
5780 def : N3VSPat<fmul, VMULfd>;
5781 def : N3VSMulOpPat<fmul, fadd, VMLAfd>,
5782 Requires<[HasNEON, UseNEONForFP, UseFPVMLx, DontUseFusedMAC]>;
5783 def : N3VSMulOpPat<fmul, fsub, VMLSfd>,
5784 Requires<[HasNEON, UseNEONForFP, UseFPVMLx, DontUseFusedMAC]>;
5785 def : N3VSMulOpPat<fmul, fadd, VFMAfd>,
5786 Requires<[HasVFP4, UseNEONForFP, UseFusedMAC]>;
5787 def : N3VSMulOpPat<fmul, fsub, VFMSfd>,
5788 Requires<[HasVFP4, UseNEONForFP, UseFusedMAC]>;
5789 def : N2VSPat<fabs, VABSfd>;
5790 def : N2VSPat<fneg, VNEGfd>;
5791 def : N3VSPat<NEONfmax, VMAXfd>;
5792 def : N3VSPat<NEONfmin, VMINfd>;
5793 def : N2VSPat<arm_ftosi, VCVTf2sd>;
5794 def : N2VSPat<arm_ftoui, VCVTf2ud>;
5795 def : N2VSPat<arm_sitof, VCVTs2fd>;
5796 def : N2VSPat<arm_uitof, VCVTu2fd>;
5798 // Prefer VMOVDRR for i32 -> f32 bitcasts, it can write all DPR registers.
5799 def : Pat<(f32 (bitconvert GPR:$a)),
5800 (EXTRACT_SUBREG (VMOVDRR GPR:$a, GPR:$a), ssub_0)>,
5801 Requires<[HasNEON, DontUseVMOVSR]>;
5803 //===----------------------------------------------------------------------===//
5804 // Non-Instruction Patterns
5805 //===----------------------------------------------------------------------===//
5808 def : Pat<(v1i64 (bitconvert (v2i32 DPR:$src))), (v1i64 DPR:$src)>;
5809 def : Pat<(v1i64 (bitconvert (v4i16 DPR:$src))), (v1i64 DPR:$src)>;
5810 def : Pat<(v1i64 (bitconvert (v8i8 DPR:$src))), (v1i64 DPR:$src)>;
5811 def : Pat<(v1i64 (bitconvert (f64 DPR:$src))), (v1i64 DPR:$src)>;
5812 def : Pat<(v1i64 (bitconvert (v2f32 DPR:$src))), (v1i64 DPR:$src)>;
5813 def : Pat<(v2i32 (bitconvert (v1i64 DPR:$src))), (v2i32 DPR:$src)>;
5814 def : Pat<(v2i32 (bitconvert (v4i16 DPR:$src))), (v2i32 DPR:$src)>;
5815 def : Pat<(v2i32 (bitconvert (v8i8 DPR:$src))), (v2i32 DPR:$src)>;
5816 def : Pat<(v2i32 (bitconvert (f64 DPR:$src))), (v2i32 DPR:$src)>;
5817 def : Pat<(v2i32 (bitconvert (v2f32 DPR:$src))), (v2i32 DPR:$src)>;
5818 def : Pat<(v4i16 (bitconvert (v1i64 DPR:$src))), (v4i16 DPR:$src)>;
5819 def : Pat<(v4i16 (bitconvert (v2i32 DPR:$src))), (v4i16 DPR:$src)>;
5820 def : Pat<(v4i16 (bitconvert (v8i8 DPR:$src))), (v4i16 DPR:$src)>;
5821 def : Pat<(v4i16 (bitconvert (f64 DPR:$src))), (v4i16 DPR:$src)>;
5822 def : Pat<(v4i16 (bitconvert (v2f32 DPR:$src))), (v4i16 DPR:$src)>;
5823 def : Pat<(v8i8 (bitconvert (v1i64 DPR:$src))), (v8i8 DPR:$src)>;
5824 def : Pat<(v8i8 (bitconvert (v2i32 DPR:$src))), (v8i8 DPR:$src)>;
5825 def : Pat<(v8i8 (bitconvert (v4i16 DPR:$src))), (v8i8 DPR:$src)>;
5826 def : Pat<(v8i8 (bitconvert (f64 DPR:$src))), (v8i8 DPR:$src)>;
5827 def : Pat<(v8i8 (bitconvert (v2f32 DPR:$src))), (v8i8 DPR:$src)>;
5828 def : Pat<(f64 (bitconvert (v1i64 DPR:$src))), (f64 DPR:$src)>;
5829 def : Pat<(f64 (bitconvert (v2i32 DPR:$src))), (f64 DPR:$src)>;
5830 def : Pat<(f64 (bitconvert (v4i16 DPR:$src))), (f64 DPR:$src)>;
5831 def : Pat<(f64 (bitconvert (v8i8 DPR:$src))), (f64 DPR:$src)>;
5832 def : Pat<(f64 (bitconvert (v2f32 DPR:$src))), (f64 DPR:$src)>;
5833 def : Pat<(v2f32 (bitconvert (f64 DPR:$src))), (v2f32 DPR:$src)>;
5834 def : Pat<(v2f32 (bitconvert (v1i64 DPR:$src))), (v2f32 DPR:$src)>;
5835 def : Pat<(v2f32 (bitconvert (v2i32 DPR:$src))), (v2f32 DPR:$src)>;
5836 def : Pat<(v2f32 (bitconvert (v4i16 DPR:$src))), (v2f32 DPR:$src)>;
5837 def : Pat<(v2f32 (bitconvert (v8i8 DPR:$src))), (v2f32 DPR:$src)>;
5839 def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (v2i64 QPR:$src)>;
5840 def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (v2i64 QPR:$src)>;
5841 def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (v2i64 QPR:$src)>;
5842 def : Pat<(v2i64 (bitconvert (v2f64 QPR:$src))), (v2i64 QPR:$src)>;
5843 def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (v2i64 QPR:$src)>;
5844 def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (v4i32 QPR:$src)>;
5845 def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (v4i32 QPR:$src)>;
5846 def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (v4i32 QPR:$src)>;
5847 def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (v4i32 QPR:$src)>;
5848 def : Pat<(v4i32 (bitconvert (v4f32 QPR:$src))), (v4i32 QPR:$src)>;
5849 def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (v8i16 QPR:$src)>;
5850 def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (v8i16 QPR:$src)>;
5851 def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (v8i16 QPR:$src)>;
5852 def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (v8i16 QPR:$src)>;
5853 def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (v8i16 QPR:$src)>;
5854 def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (v16i8 QPR:$src)>;
5855 def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (v16i8 QPR:$src)>;
5856 def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (v16i8 QPR:$src)>;
5857 def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (v16i8 QPR:$src)>;
5858 def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (v16i8 QPR:$src)>;
5859 def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (v4f32 QPR:$src)>;
5860 def : Pat<(v4f32 (bitconvert (v4i32 QPR:$src))), (v4f32 QPR:$src)>;
5861 def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (v4f32 QPR:$src)>;
5862 def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (v4f32 QPR:$src)>;
5863 def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (v4f32 QPR:$src)>;
5864 def : Pat<(v2f64 (bitconvert (v2i64 QPR:$src))), (v2f64 QPR:$src)>;
5865 def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (v2f64 QPR:$src)>;
5866 def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 QPR:$src)>;
5867 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 QPR:$src)>;
5868 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 QPR:$src)>;
5870 // Fold extracting an element out of a v2i32 into a vfp register.
5871 def : Pat<(f32 (bitconvert (i32 (extractelt (v2i32 DPR:$src), imm:$lane)))),
5872 (f32 (EXTRACT_SUBREG DPR:$src, (SSubReg_f32_reg imm:$lane)))>;
5874 // Vector lengthening move with load, matching extending loads.
5876 // extload, zextload and sextload for a standard lengthening load. Example:
5877 // Lengthen_Single<"8", "i16", "8"> =
5878 // Pat<(v8i16 (extloadvi8 addrmode6:$addr))
5879 // (VMOVLuv8i16 (VLD1d8 addrmode6:$addr,
5880 // (f64 (IMPLICIT_DEF)), (i32 0)))>;
5881 multiclass Lengthen_Single<string DestLanes, string DestTy, string SrcTy> {
5882 let AddedComplexity = 10 in {
5883 def _Any : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5884 (!cast<PatFrag>("extloadvi" # SrcTy) addrmode6:$addr)),
5885 (!cast<Instruction>("VMOVLuv" # DestLanes # DestTy)
5886 (!cast<Instruction>("VLD1d" # SrcTy) addrmode6:$addr))>;
5888 def _Z : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5889 (!cast<PatFrag>("zextloadvi" # SrcTy) addrmode6:$addr)),
5890 (!cast<Instruction>("VMOVLuv" # DestLanes # DestTy)
5891 (!cast<Instruction>("VLD1d" # SrcTy) addrmode6:$addr))>;
5893 def _S : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5894 (!cast<PatFrag>("sextloadvi" # SrcTy) addrmode6:$addr)),
5895 (!cast<Instruction>("VMOVLsv" # DestLanes # DestTy)
5896 (!cast<Instruction>("VLD1d" # SrcTy) addrmode6:$addr))>;
5900 // extload, zextload and sextload for a lengthening load which only uses
5901 // half the lanes available. Example:
5902 // Lengthen_HalfSingle<"4", "i16", "8", "i16", "i8"> =
5903 // Pat<(v4i16 (extloadvi8 addrmode6oneL32:$addr)),
5904 // (EXTRACT_SUBREG (VMOVLuv8i16 (VLD1LNd32 addrmode6oneL32:$addr,
5905 // (f64 (IMPLICIT_DEF)), (i32 0))),
5907 multiclass Lengthen_HalfSingle<string DestLanes, string DestTy, string SrcTy,
5908 string InsnLanes, string InsnTy> {
5909 def _Any : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5910 (!cast<PatFrag>("extloadv" # SrcTy) addrmode6oneL32:$addr)),
5911 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # InsnLanes # InsnTy)
5912 (VLD1LNd32 addrmode6oneL32:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
5914 def _Z : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5915 (!cast<PatFrag>("zextloadv" # SrcTy) addrmode6oneL32:$addr)),
5916 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # InsnLanes # InsnTy)
5917 (VLD1LNd32 addrmode6oneL32:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
5919 def _S : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5920 (!cast<PatFrag>("sextloadv" # SrcTy) addrmode6oneL32:$addr)),
5921 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLsv" # InsnLanes # InsnTy)
5922 (VLD1LNd32 addrmode6oneL32:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
5926 // extload, zextload and sextload for a lengthening load followed by another
5927 // lengthening load, to quadruple the initial length.
5929 // Lengthen_Double<"4", "i32", "i8", "8", "i16", "4", "i32"> =
5930 // Pat<(v4i32 (extloadvi8 addrmode6oneL32:$addr))
5931 // (EXTRACT_SUBREG (VMOVLuv4i32
5932 // (EXTRACT_SUBREG (VMOVLuv8i16 (VLD1LNd32 addrmode6oneL32:$addr,
5933 // (f64 (IMPLICIT_DEF)),
5937 multiclass Lengthen_Double<string DestLanes, string DestTy, string SrcTy,
5938 string Insn1Lanes, string Insn1Ty, string Insn2Lanes,
5940 def _Any : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5941 (!cast<PatFrag>("extloadv" # SrcTy) addrmode6oneL32:$addr)),
5942 (!cast<Instruction>("VMOVLuv" # Insn2Lanes # Insn2Ty)
5943 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn1Lanes # Insn1Ty)
5944 (VLD1LNd32 addrmode6oneL32:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
5946 def _Z : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5947 (!cast<PatFrag>("zextloadv" # SrcTy) addrmode6oneL32:$addr)),
5948 (!cast<Instruction>("VMOVLuv" # Insn2Lanes # Insn2Ty)
5949 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn1Lanes # Insn1Ty)
5950 (VLD1LNd32 addrmode6oneL32:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
5952 def _S : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5953 (!cast<PatFrag>("sextloadv" # SrcTy) addrmode6oneL32:$addr)),
5954 (!cast<Instruction>("VMOVLsv" # Insn2Lanes # Insn2Ty)
5955 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLsv" # Insn1Lanes # Insn1Ty)
5956 (VLD1LNd32 addrmode6oneL32:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
5960 // extload, zextload and sextload for a lengthening load followed by another
5961 // lengthening load, to quadruple the initial length, but which ends up only
5962 // requiring half the available lanes (a 64-bit outcome instead of a 128-bit).
5964 // Lengthen_HalfDouble<"2", "i32", "i8", "8", "i16", "4", "i32"> =
5965 // Pat<(v2i32 (extloadvi8 addrmode6:$addr))
5966 // (EXTRACT_SUBREG (VMOVLuv4i32
5967 // (EXTRACT_SUBREG (VMOVLuv8i16 (VLD1LNd16 addrmode6:$addr,
5968 // (f64 (IMPLICIT_DEF)), (i32 0))),
5971 multiclass Lengthen_HalfDouble<string DestLanes, string DestTy, string SrcTy,
5972 string Insn1Lanes, string Insn1Ty, string Insn2Lanes,
5974 def _Any : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5975 (!cast<PatFrag>("extloadv" # SrcTy) addrmode6:$addr)),
5976 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn2Lanes # Insn2Ty)
5977 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn1Lanes # Insn1Ty)
5978 (VLD1LNd16 addrmode6:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
5981 def _Z : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5982 (!cast<PatFrag>("zextloadv" # SrcTy) addrmode6:$addr)),
5983 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn2Lanes # Insn2Ty)
5984 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn1Lanes # Insn1Ty)
5985 (VLD1LNd16 addrmode6:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
5988 def _S : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5989 (!cast<PatFrag>("sextloadv" # SrcTy) addrmode6:$addr)),
5990 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLsv" # Insn2Lanes # Insn2Ty)
5991 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLsv" # Insn1Lanes # Insn1Ty)
5992 (VLD1LNd16 addrmode6:$addr, (f64 (IMPLICIT_DEF)), (i32 0))),
5997 defm : Lengthen_Single<"8", "i16", "8">; // v8i8 -> v8i16
5998 defm : Lengthen_Single<"4", "i32", "16">; // v4i16 -> v4i32
5999 defm : Lengthen_Single<"2", "i64", "32">; // v2i32 -> v2i64
6001 defm : Lengthen_HalfSingle<"4", "i16", "i8", "8", "i16">; // v4i8 -> v4i16
6002 defm : Lengthen_HalfSingle<"2", "i32", "i16", "4", "i32">; // v2i16 -> v2i32
6004 // Double lengthening - v4i8 -> v4i16 -> v4i32
6005 defm : Lengthen_Double<"4", "i32", "i8", "8", "i16", "4", "i32">;
6006 // v2i8 -> v2i16 -> v2i32
6007 defm : Lengthen_HalfDouble<"2", "i32", "i8", "8", "i16", "4", "i32">;
6008 // v2i16 -> v2i32 -> v2i64
6009 defm : Lengthen_Double<"2", "i64", "i16", "4", "i32", "2", "i64">;
6011 // Triple lengthening - v2i8 -> v2i16 -> v2i32 -> v2i64
6012 def : Pat<(v2i64 (extloadvi8 addrmode6:$addr)),
6013 (VMOVLuv2i64 (EXTRACT_SUBREG (VMOVLuv4i32 (EXTRACT_SUBREG (VMOVLuv8i16
6014 (VLD1LNd16 addrmode6:$addr,
6015 (f64 (IMPLICIT_DEF)), (i32 0))), dsub_0)), dsub_0))>;
6016 def : Pat<(v2i64 (zextloadvi8 addrmode6:$addr)),
6017 (VMOVLuv2i64 (EXTRACT_SUBREG (VMOVLuv4i32 (EXTRACT_SUBREG (VMOVLuv8i16
6018 (VLD1LNd16 addrmode6:$addr,
6019 (f64 (IMPLICIT_DEF)), (i32 0))), dsub_0)), dsub_0))>;
6020 def : Pat<(v2i64 (sextloadvi8 addrmode6:$addr)),
6021 (VMOVLsv2i64 (EXTRACT_SUBREG (VMOVLsv4i32 (EXTRACT_SUBREG (VMOVLsv8i16
6022 (VLD1LNd16 addrmode6:$addr,
6023 (f64 (IMPLICIT_DEF)), (i32 0))), dsub_0)), dsub_0))>;
6025 //===----------------------------------------------------------------------===//
6026 // Assembler aliases
6029 def : VFP2InstAlias<"fmdhr${p} $Dd, $Rn",
6030 (VSETLNi32 DPR:$Dd, GPR:$Rn, 1, pred:$p)>;
6031 def : VFP2InstAlias<"fmdlr${p} $Dd, $Rn",
6032 (VSETLNi32 DPR:$Dd, GPR:$Rn, 0, pred:$p)>;
6034 // VAND/VBIC/VEOR/VORR accept but do not require a type suffix.
6035 defm : NEONDTAnyInstAlias<"vand${p}", "$Vd, $Vn, $Vm",
6036 (VANDd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
6037 defm : NEONDTAnyInstAlias<"vand${p}", "$Vd, $Vn, $Vm",
6038 (VANDq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
6039 defm : NEONDTAnyInstAlias<"vbic${p}", "$Vd, $Vn, $Vm",
6040 (VBICd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
6041 defm : NEONDTAnyInstAlias<"vbic${p}", "$Vd, $Vn, $Vm",
6042 (VBICq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
6043 defm : NEONDTAnyInstAlias<"veor${p}", "$Vd, $Vn, $Vm",
6044 (VEORd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
6045 defm : NEONDTAnyInstAlias<"veor${p}", "$Vd, $Vn, $Vm",
6046 (VEORq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
6047 defm : NEONDTAnyInstAlias<"vorr${p}", "$Vd, $Vn, $Vm",
6048 (VORRd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
6049 defm : NEONDTAnyInstAlias<"vorr${p}", "$Vd, $Vn, $Vm",
6050 (VORRq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
6051 // ... two-operand aliases
6052 defm : NEONDTAnyInstAlias<"vand${p}", "$Vdn, $Vm",
6053 (VANDd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
6054 defm : NEONDTAnyInstAlias<"vand${p}", "$Vdn, $Vm",
6055 (VANDq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
6056 defm : NEONDTAnyInstAlias<"veor${p}", "$Vdn, $Vm",
6057 (VEORd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
6058 defm : NEONDTAnyInstAlias<"veor${p}", "$Vdn, $Vm",
6059 (VEORq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
6060 defm : NEONDTAnyInstAlias<"vorr${p}", "$Vdn, $Vm",
6061 (VORRd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
6062 defm : NEONDTAnyInstAlias<"vorr${p}", "$Vdn, $Vm",
6063 (VORRq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
6065 // VLD1 single-lane pseudo-instructions. These need special handling for
6066 // the lane index that an InstAlias can't handle, so we use these instead.
6067 def VLD1LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vld1${p}", ".8", "$list, $addr",
6068 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6069 def VLD1LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vld1${p}", ".16", "$list, $addr",
6070 (ins VecListOneDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6071 def VLD1LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vld1${p}", ".32", "$list, $addr",
6072 (ins VecListOneDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6074 def VLD1LNdWB_fixed_Asm_8 :
6075 NEONDataTypeAsmPseudoInst<"vld1${p}", ".8", "$list, $addr!",
6076 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6077 def VLD1LNdWB_fixed_Asm_16 :
6078 NEONDataTypeAsmPseudoInst<"vld1${p}", ".16", "$list, $addr!",
6079 (ins VecListOneDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6080 def VLD1LNdWB_fixed_Asm_32 :
6081 NEONDataTypeAsmPseudoInst<"vld1${p}", ".32", "$list, $addr!",
6082 (ins VecListOneDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6083 def VLD1LNdWB_register_Asm_8 :
6084 NEONDataTypeAsmPseudoInst<"vld1${p}", ".8", "$list, $addr, $Rm",
6085 (ins VecListOneDByteIndexed:$list, addrmode6:$addr,
6086 rGPR:$Rm, pred:$p)>;
6087 def VLD1LNdWB_register_Asm_16 :
6088 NEONDataTypeAsmPseudoInst<"vld1${p}", ".16", "$list, $addr, $Rm",
6089 (ins VecListOneDHWordIndexed:$list, addrmode6:$addr,
6090 rGPR:$Rm, pred:$p)>;
6091 def VLD1LNdWB_register_Asm_32 :
6092 NEONDataTypeAsmPseudoInst<"vld1${p}", ".32", "$list, $addr, $Rm",
6093 (ins VecListOneDWordIndexed:$list, addrmode6:$addr,
6094 rGPR:$Rm, pred:$p)>;
6097 // VST1 single-lane pseudo-instructions. These need special handling for
6098 // the lane index that an InstAlias can't handle, so we use these instead.
6099 def VST1LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vst1${p}", ".8", "$list, $addr",
6100 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6101 def VST1LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vst1${p}", ".16", "$list, $addr",
6102 (ins VecListOneDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6103 def VST1LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vst1${p}", ".32", "$list, $addr",
6104 (ins VecListOneDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6106 def VST1LNdWB_fixed_Asm_8 :
6107 NEONDataTypeAsmPseudoInst<"vst1${p}", ".8", "$list, $addr!",
6108 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6109 def VST1LNdWB_fixed_Asm_16 :
6110 NEONDataTypeAsmPseudoInst<"vst1${p}", ".16", "$list, $addr!",
6111 (ins VecListOneDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6112 def VST1LNdWB_fixed_Asm_32 :
6113 NEONDataTypeAsmPseudoInst<"vst1${p}", ".32", "$list, $addr!",
6114 (ins VecListOneDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6115 def VST1LNdWB_register_Asm_8 :
6116 NEONDataTypeAsmPseudoInst<"vst1${p}", ".8", "$list, $addr, $Rm",
6117 (ins VecListOneDByteIndexed:$list, addrmode6:$addr,
6118 rGPR:$Rm, pred:$p)>;
6119 def VST1LNdWB_register_Asm_16 :
6120 NEONDataTypeAsmPseudoInst<"vst1${p}", ".16", "$list, $addr, $Rm",
6121 (ins VecListOneDHWordIndexed:$list, addrmode6:$addr,
6122 rGPR:$Rm, pred:$p)>;
6123 def VST1LNdWB_register_Asm_32 :
6124 NEONDataTypeAsmPseudoInst<"vst1${p}", ".32", "$list, $addr, $Rm",
6125 (ins VecListOneDWordIndexed:$list, addrmode6:$addr,
6126 rGPR:$Rm, pred:$p)>;
6128 // VLD2 single-lane pseudo-instructions. These need special handling for
6129 // the lane index that an InstAlias can't handle, so we use these instead.
6130 def VLD2LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vld2${p}", ".8", "$list, $addr",
6131 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6132 def VLD2LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vld2${p}", ".16", "$list, $addr",
6133 (ins VecListTwoDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6134 def VLD2LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vld2${p}", ".32", "$list, $addr",
6135 (ins VecListTwoDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6136 def VLD2LNqAsm_16 : NEONDataTypeAsmPseudoInst<"vld2${p}", ".16", "$list, $addr",
6137 (ins VecListTwoQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6138 def VLD2LNqAsm_32 : NEONDataTypeAsmPseudoInst<"vld2${p}", ".32", "$list, $addr",
6139 (ins VecListTwoQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6141 def VLD2LNdWB_fixed_Asm_8 :
6142 NEONDataTypeAsmPseudoInst<"vld2${p}", ".8", "$list, $addr!",
6143 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6144 def VLD2LNdWB_fixed_Asm_16 :
6145 NEONDataTypeAsmPseudoInst<"vld2${p}", ".16", "$list, $addr!",
6146 (ins VecListTwoDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6147 def VLD2LNdWB_fixed_Asm_32 :
6148 NEONDataTypeAsmPseudoInst<"vld2${p}", ".32", "$list, $addr!",
6149 (ins VecListTwoDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6150 def VLD2LNqWB_fixed_Asm_16 :
6151 NEONDataTypeAsmPseudoInst<"vld2${p}", ".16", "$list, $addr!",
6152 (ins VecListTwoQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6153 def VLD2LNqWB_fixed_Asm_32 :
6154 NEONDataTypeAsmPseudoInst<"vld2${p}", ".32", "$list, $addr!",
6155 (ins VecListTwoQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6156 def VLD2LNdWB_register_Asm_8 :
6157 NEONDataTypeAsmPseudoInst<"vld2${p}", ".8", "$list, $addr, $Rm",
6158 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr,
6159 rGPR:$Rm, pred:$p)>;
6160 def VLD2LNdWB_register_Asm_16 :
6161 NEONDataTypeAsmPseudoInst<"vld2${p}", ".16", "$list, $addr, $Rm",
6162 (ins VecListTwoDHWordIndexed:$list, addrmode6:$addr,
6163 rGPR:$Rm, pred:$p)>;
6164 def VLD2LNdWB_register_Asm_32 :
6165 NEONDataTypeAsmPseudoInst<"vld2${p}", ".32", "$list, $addr, $Rm",
6166 (ins VecListTwoDWordIndexed:$list, addrmode6:$addr,
6167 rGPR:$Rm, pred:$p)>;
6168 def VLD2LNqWB_register_Asm_16 :
6169 NEONDataTypeAsmPseudoInst<"vld2${p}", ".16", "$list, $addr, $Rm",
6170 (ins VecListTwoQHWordIndexed:$list, addrmode6:$addr,
6171 rGPR:$Rm, pred:$p)>;
6172 def VLD2LNqWB_register_Asm_32 :
6173 NEONDataTypeAsmPseudoInst<"vld2${p}", ".32", "$list, $addr, $Rm",
6174 (ins VecListTwoQWordIndexed:$list, addrmode6:$addr,
6175 rGPR:$Rm, pred:$p)>;
6178 // VST2 single-lane pseudo-instructions. These need special handling for
6179 // the lane index that an InstAlias can't handle, so we use these instead.
6180 def VST2LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vst2${p}", ".8", "$list, $addr",
6181 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6182 def VST2LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vst2${p}", ".16", "$list, $addr",
6183 (ins VecListTwoDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6184 def VST2LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vst2${p}", ".32", "$list, $addr",
6185 (ins VecListTwoDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6186 def VST2LNqAsm_16 : NEONDataTypeAsmPseudoInst<"vst2${p}", ".16", "$list, $addr",
6187 (ins VecListTwoQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6188 def VST2LNqAsm_32 : NEONDataTypeAsmPseudoInst<"vst2${p}", ".32", "$list, $addr",
6189 (ins VecListTwoQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6191 def VST2LNdWB_fixed_Asm_8 :
6192 NEONDataTypeAsmPseudoInst<"vst2${p}", ".8", "$list, $addr!",
6193 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6194 def VST2LNdWB_fixed_Asm_16 :
6195 NEONDataTypeAsmPseudoInst<"vst2${p}", ".16", "$list, $addr!",
6196 (ins VecListTwoDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6197 def VST2LNdWB_fixed_Asm_32 :
6198 NEONDataTypeAsmPseudoInst<"vst2${p}", ".32", "$list, $addr!",
6199 (ins VecListTwoDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6200 def VST2LNqWB_fixed_Asm_16 :
6201 NEONDataTypeAsmPseudoInst<"vst2${p}", ".16", "$list, $addr!",
6202 (ins VecListTwoQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6203 def VST2LNqWB_fixed_Asm_32 :
6204 NEONDataTypeAsmPseudoInst<"vst2${p}", ".32", "$list, $addr!",
6205 (ins VecListTwoQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6206 def VST2LNdWB_register_Asm_8 :
6207 NEONDataTypeAsmPseudoInst<"vst2${p}", ".8", "$list, $addr, $Rm",
6208 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr,
6209 rGPR:$Rm, pred:$p)>;
6210 def VST2LNdWB_register_Asm_16 :
6211 NEONDataTypeAsmPseudoInst<"vst2${p}", ".16","$list, $addr, $Rm",
6212 (ins VecListTwoDHWordIndexed:$list, addrmode6:$addr,
6213 rGPR:$Rm, pred:$p)>;
6214 def VST2LNdWB_register_Asm_32 :
6215 NEONDataTypeAsmPseudoInst<"vst2${p}", ".32", "$list, $addr, $Rm",
6216 (ins VecListTwoDWordIndexed:$list, addrmode6:$addr,
6217 rGPR:$Rm, pred:$p)>;
6218 def VST2LNqWB_register_Asm_16 :
6219 NEONDataTypeAsmPseudoInst<"vst2${p}", ".16","$list, $addr, $Rm",
6220 (ins VecListTwoQHWordIndexed:$list, addrmode6:$addr,
6221 rGPR:$Rm, pred:$p)>;
6222 def VST2LNqWB_register_Asm_32 :
6223 NEONDataTypeAsmPseudoInst<"vst2${p}", ".32", "$list, $addr, $Rm",
6224 (ins VecListTwoQWordIndexed:$list, addrmode6:$addr,
6225 rGPR:$Rm, pred:$p)>;
6227 // VLD3 all-lanes pseudo-instructions. These need special handling for
6228 // the lane index that an InstAlias can't handle, so we use these instead.
6229 def VLD3DUPdAsm_8 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr",
6230 (ins VecListThreeDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6231 def VLD3DUPdAsm_16: NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr",
6232 (ins VecListThreeDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6233 def VLD3DUPdAsm_32: NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr",
6234 (ins VecListThreeDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6235 def VLD3DUPqAsm_8 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr",
6236 (ins VecListThreeQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6237 def VLD3DUPqAsm_16: NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr",
6238 (ins VecListThreeQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6239 def VLD3DUPqAsm_32: NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr",
6240 (ins VecListThreeQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6242 def VLD3DUPdWB_fixed_Asm_8 :
6243 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr!",
6244 (ins VecListThreeDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6245 def VLD3DUPdWB_fixed_Asm_16 :
6246 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr!",
6247 (ins VecListThreeDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6248 def VLD3DUPdWB_fixed_Asm_32 :
6249 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr!",
6250 (ins VecListThreeDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6251 def VLD3DUPqWB_fixed_Asm_8 :
6252 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr!",
6253 (ins VecListThreeQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6254 def VLD3DUPqWB_fixed_Asm_16 :
6255 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr!",
6256 (ins VecListThreeQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6257 def VLD3DUPqWB_fixed_Asm_32 :
6258 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr!",
6259 (ins VecListThreeQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6260 def VLD3DUPdWB_register_Asm_8 :
6261 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr, $Rm",
6262 (ins VecListThreeDAllLanes:$list, addrmode6:$addr,
6263 rGPR:$Rm, pred:$p)>;
6264 def VLD3DUPdWB_register_Asm_16 :
6265 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr, $Rm",
6266 (ins VecListThreeDAllLanes:$list, addrmode6:$addr,
6267 rGPR:$Rm, pred:$p)>;
6268 def VLD3DUPdWB_register_Asm_32 :
6269 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr, $Rm",
6270 (ins VecListThreeDAllLanes:$list, addrmode6:$addr,
6271 rGPR:$Rm, pred:$p)>;
6272 def VLD3DUPqWB_register_Asm_8 :
6273 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr, $Rm",
6274 (ins VecListThreeQAllLanes:$list, addrmode6:$addr,
6275 rGPR:$Rm, pred:$p)>;
6276 def VLD3DUPqWB_register_Asm_16 :
6277 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr, $Rm",
6278 (ins VecListThreeQAllLanes:$list, addrmode6:$addr,
6279 rGPR:$Rm, pred:$p)>;
6280 def VLD3DUPqWB_register_Asm_32 :
6281 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr, $Rm",
6282 (ins VecListThreeQAllLanes:$list, addrmode6:$addr,
6283 rGPR:$Rm, pred:$p)>;
6286 // VLD3 single-lane pseudo-instructions. These need special handling for
6287 // the lane index that an InstAlias can't handle, so we use these instead.
6288 def VLD3LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr",
6289 (ins VecListThreeDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6290 def VLD3LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr",
6291 (ins VecListThreeDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6292 def VLD3LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr",
6293 (ins VecListThreeDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6294 def VLD3LNqAsm_16 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr",
6295 (ins VecListThreeQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6296 def VLD3LNqAsm_32 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr",
6297 (ins VecListThreeQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6299 def VLD3LNdWB_fixed_Asm_8 :
6300 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr!",
6301 (ins VecListThreeDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6302 def VLD3LNdWB_fixed_Asm_16 :
6303 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr!",
6304 (ins VecListThreeDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6305 def VLD3LNdWB_fixed_Asm_32 :
6306 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr!",
6307 (ins VecListThreeDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6308 def VLD3LNqWB_fixed_Asm_16 :
6309 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr!",
6310 (ins VecListThreeQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6311 def VLD3LNqWB_fixed_Asm_32 :
6312 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr!",
6313 (ins VecListThreeQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6314 def VLD3LNdWB_register_Asm_8 :
6315 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr, $Rm",
6316 (ins VecListThreeDByteIndexed:$list, addrmode6:$addr,
6317 rGPR:$Rm, pred:$p)>;
6318 def VLD3LNdWB_register_Asm_16 :
6319 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr, $Rm",
6320 (ins VecListThreeDHWordIndexed:$list, addrmode6:$addr,
6321 rGPR:$Rm, pred:$p)>;
6322 def VLD3LNdWB_register_Asm_32 :
6323 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr, $Rm",
6324 (ins VecListThreeDWordIndexed:$list, addrmode6:$addr,
6325 rGPR:$Rm, pred:$p)>;
6326 def VLD3LNqWB_register_Asm_16 :
6327 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr, $Rm",
6328 (ins VecListThreeQHWordIndexed:$list, addrmode6:$addr,
6329 rGPR:$Rm, pred:$p)>;
6330 def VLD3LNqWB_register_Asm_32 :
6331 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr, $Rm",
6332 (ins VecListThreeQWordIndexed:$list, addrmode6:$addr,
6333 rGPR:$Rm, pred:$p)>;
6335 // VLD3 multiple structure pseudo-instructions. These need special handling for
6336 // the vector operands that the normal instructions don't yet model.
6337 // FIXME: Remove these when the register classes and instructions are updated.
6338 def VLD3dAsm_8 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr",
6339 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6340 def VLD3dAsm_16 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr",
6341 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6342 def VLD3dAsm_32 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr",
6343 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6344 def VLD3qAsm_8 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr",
6345 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6346 def VLD3qAsm_16 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr",
6347 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6348 def VLD3qAsm_32 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr",
6349 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6351 def VLD3dWB_fixed_Asm_8 :
6352 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr!",
6353 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6354 def VLD3dWB_fixed_Asm_16 :
6355 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr!",
6356 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6357 def VLD3dWB_fixed_Asm_32 :
6358 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr!",
6359 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6360 def VLD3qWB_fixed_Asm_8 :
6361 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr!",
6362 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6363 def VLD3qWB_fixed_Asm_16 :
6364 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr!",
6365 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6366 def VLD3qWB_fixed_Asm_32 :
6367 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr!",
6368 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6369 def VLD3dWB_register_Asm_8 :
6370 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr, $Rm",
6371 (ins VecListThreeD:$list, addrmode6:$addr,
6372 rGPR:$Rm, pred:$p)>;
6373 def VLD3dWB_register_Asm_16 :
6374 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr, $Rm",
6375 (ins VecListThreeD:$list, addrmode6:$addr,
6376 rGPR:$Rm, pred:$p)>;
6377 def VLD3dWB_register_Asm_32 :
6378 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr, $Rm",
6379 (ins VecListThreeD:$list, addrmode6:$addr,
6380 rGPR:$Rm, pred:$p)>;
6381 def VLD3qWB_register_Asm_8 :
6382 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr, $Rm",
6383 (ins VecListThreeQ:$list, addrmode6:$addr,
6384 rGPR:$Rm, pred:$p)>;
6385 def VLD3qWB_register_Asm_16 :
6386 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr, $Rm",
6387 (ins VecListThreeQ:$list, addrmode6:$addr,
6388 rGPR:$Rm, pred:$p)>;
6389 def VLD3qWB_register_Asm_32 :
6390 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr, $Rm",
6391 (ins VecListThreeQ:$list, addrmode6:$addr,
6392 rGPR:$Rm, pred:$p)>;
6394 // VST3 single-lane pseudo-instructions. These need special handling for
6395 // the lane index that an InstAlias can't handle, so we use these instead.
6396 def VST3LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr",
6397 (ins VecListThreeDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6398 def VST3LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr",
6399 (ins VecListThreeDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6400 def VST3LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr",
6401 (ins VecListThreeDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6402 def VST3LNqAsm_16 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr",
6403 (ins VecListThreeQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6404 def VST3LNqAsm_32 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr",
6405 (ins VecListThreeQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6407 def VST3LNdWB_fixed_Asm_8 :
6408 NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr!",
6409 (ins VecListThreeDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6410 def VST3LNdWB_fixed_Asm_16 :
6411 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr!",
6412 (ins VecListThreeDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6413 def VST3LNdWB_fixed_Asm_32 :
6414 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr!",
6415 (ins VecListThreeDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6416 def VST3LNqWB_fixed_Asm_16 :
6417 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr!",
6418 (ins VecListThreeQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6419 def VST3LNqWB_fixed_Asm_32 :
6420 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr!",
6421 (ins VecListThreeQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6422 def VST3LNdWB_register_Asm_8 :
6423 NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr, $Rm",
6424 (ins VecListThreeDByteIndexed:$list, addrmode6:$addr,
6425 rGPR:$Rm, pred:$p)>;
6426 def VST3LNdWB_register_Asm_16 :
6427 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr, $Rm",
6428 (ins VecListThreeDHWordIndexed:$list, addrmode6:$addr,
6429 rGPR:$Rm, pred:$p)>;
6430 def VST3LNdWB_register_Asm_32 :
6431 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr, $Rm",
6432 (ins VecListThreeDWordIndexed:$list, addrmode6:$addr,
6433 rGPR:$Rm, pred:$p)>;
6434 def VST3LNqWB_register_Asm_16 :
6435 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr, $Rm",
6436 (ins VecListThreeQHWordIndexed:$list, addrmode6:$addr,
6437 rGPR:$Rm, pred:$p)>;
6438 def VST3LNqWB_register_Asm_32 :
6439 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr, $Rm",
6440 (ins VecListThreeQWordIndexed:$list, addrmode6:$addr,
6441 rGPR:$Rm, pred:$p)>;
6444 // VST3 multiple structure pseudo-instructions. These need special handling for
6445 // the vector operands that the normal instructions don't yet model.
6446 // FIXME: Remove these when the register classes and instructions are updated.
6447 def VST3dAsm_8 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr",
6448 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6449 def VST3dAsm_16 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr",
6450 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6451 def VST3dAsm_32 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr",
6452 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6453 def VST3qAsm_8 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr",
6454 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6455 def VST3qAsm_16 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr",
6456 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6457 def VST3qAsm_32 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr",
6458 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6460 def VST3dWB_fixed_Asm_8 :
6461 NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr!",
6462 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6463 def VST3dWB_fixed_Asm_16 :
6464 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr!",
6465 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6466 def VST3dWB_fixed_Asm_32 :
6467 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr!",
6468 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6469 def VST3qWB_fixed_Asm_8 :
6470 NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr!",
6471 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6472 def VST3qWB_fixed_Asm_16 :
6473 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr!",
6474 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6475 def VST3qWB_fixed_Asm_32 :
6476 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr!",
6477 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6478 def VST3dWB_register_Asm_8 :
6479 NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr, $Rm",
6480 (ins VecListThreeD:$list, addrmode6:$addr,
6481 rGPR:$Rm, pred:$p)>;
6482 def VST3dWB_register_Asm_16 :
6483 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr, $Rm",
6484 (ins VecListThreeD:$list, addrmode6:$addr,
6485 rGPR:$Rm, pred:$p)>;
6486 def VST3dWB_register_Asm_32 :
6487 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr, $Rm",
6488 (ins VecListThreeD:$list, addrmode6:$addr,
6489 rGPR:$Rm, pred:$p)>;
6490 def VST3qWB_register_Asm_8 :
6491 NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr, $Rm",
6492 (ins VecListThreeQ:$list, addrmode6:$addr,
6493 rGPR:$Rm, pred:$p)>;
6494 def VST3qWB_register_Asm_16 :
6495 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr, $Rm",
6496 (ins VecListThreeQ:$list, addrmode6:$addr,
6497 rGPR:$Rm, pred:$p)>;
6498 def VST3qWB_register_Asm_32 :
6499 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr, $Rm",
6500 (ins VecListThreeQ:$list, addrmode6:$addr,
6501 rGPR:$Rm, pred:$p)>;
6503 // VLD4 all-lanes pseudo-instructions. These need special handling for
6504 // the lane index that an InstAlias can't handle, so we use these instead.
6505 def VLD4DUPdAsm_8 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr",
6506 (ins VecListFourDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6507 def VLD4DUPdAsm_16: NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr",
6508 (ins VecListFourDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6509 def VLD4DUPdAsm_32: NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr",
6510 (ins VecListFourDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6511 def VLD4DUPqAsm_8 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr",
6512 (ins VecListFourQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6513 def VLD4DUPqAsm_16: NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr",
6514 (ins VecListFourQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6515 def VLD4DUPqAsm_32: NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr",
6516 (ins VecListFourQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6518 def VLD4DUPdWB_fixed_Asm_8 :
6519 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr!",
6520 (ins VecListFourDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6521 def VLD4DUPdWB_fixed_Asm_16 :
6522 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr!",
6523 (ins VecListFourDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6524 def VLD4DUPdWB_fixed_Asm_32 :
6525 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr!",
6526 (ins VecListFourDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6527 def VLD4DUPqWB_fixed_Asm_8 :
6528 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr!",
6529 (ins VecListFourQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6530 def VLD4DUPqWB_fixed_Asm_16 :
6531 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr!",
6532 (ins VecListFourQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6533 def VLD4DUPqWB_fixed_Asm_32 :
6534 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr!",
6535 (ins VecListFourQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6536 def VLD4DUPdWB_register_Asm_8 :
6537 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr, $Rm",
6538 (ins VecListFourDAllLanes:$list, addrmode6:$addr,
6539 rGPR:$Rm, pred:$p)>;
6540 def VLD4DUPdWB_register_Asm_16 :
6541 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr, $Rm",
6542 (ins VecListFourDAllLanes:$list, addrmode6:$addr,
6543 rGPR:$Rm, pred:$p)>;
6544 def VLD4DUPdWB_register_Asm_32 :
6545 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr, $Rm",
6546 (ins VecListFourDAllLanes:$list, addrmode6:$addr,
6547 rGPR:$Rm, pred:$p)>;
6548 def VLD4DUPqWB_register_Asm_8 :
6549 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr, $Rm",
6550 (ins VecListFourQAllLanes:$list, addrmode6:$addr,
6551 rGPR:$Rm, pred:$p)>;
6552 def VLD4DUPqWB_register_Asm_16 :
6553 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr, $Rm",
6554 (ins VecListFourQAllLanes:$list, addrmode6:$addr,
6555 rGPR:$Rm, pred:$p)>;
6556 def VLD4DUPqWB_register_Asm_32 :
6557 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr, $Rm",
6558 (ins VecListFourQAllLanes:$list, addrmode6:$addr,
6559 rGPR:$Rm, pred:$p)>;
6562 // VLD4 single-lane pseudo-instructions. These need special handling for
6563 // the lane index that an InstAlias can't handle, so we use these instead.
6564 def VLD4LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr",
6565 (ins VecListFourDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6566 def VLD4LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr",
6567 (ins VecListFourDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6568 def VLD4LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr",
6569 (ins VecListFourDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6570 def VLD4LNqAsm_16 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr",
6571 (ins VecListFourQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6572 def VLD4LNqAsm_32 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr",
6573 (ins VecListFourQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6575 def VLD4LNdWB_fixed_Asm_8 :
6576 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr!",
6577 (ins VecListFourDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6578 def VLD4LNdWB_fixed_Asm_16 :
6579 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr!",
6580 (ins VecListFourDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6581 def VLD4LNdWB_fixed_Asm_32 :
6582 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr!",
6583 (ins VecListFourDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6584 def VLD4LNqWB_fixed_Asm_16 :
6585 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr!",
6586 (ins VecListFourQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6587 def VLD4LNqWB_fixed_Asm_32 :
6588 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr!",
6589 (ins VecListFourQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6590 def VLD4LNdWB_register_Asm_8 :
6591 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr, $Rm",
6592 (ins VecListFourDByteIndexed:$list, addrmode6:$addr,
6593 rGPR:$Rm, pred:$p)>;
6594 def VLD4LNdWB_register_Asm_16 :
6595 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr, $Rm",
6596 (ins VecListFourDHWordIndexed:$list, addrmode6:$addr,
6597 rGPR:$Rm, pred:$p)>;
6598 def VLD4LNdWB_register_Asm_32 :
6599 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr, $Rm",
6600 (ins VecListFourDWordIndexed:$list, addrmode6:$addr,
6601 rGPR:$Rm, pred:$p)>;
6602 def VLD4LNqWB_register_Asm_16 :
6603 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr, $Rm",
6604 (ins VecListFourQHWordIndexed:$list, addrmode6:$addr,
6605 rGPR:$Rm, pred:$p)>;
6606 def VLD4LNqWB_register_Asm_32 :
6607 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr, $Rm",
6608 (ins VecListFourQWordIndexed:$list, addrmode6:$addr,
6609 rGPR:$Rm, pred:$p)>;
6613 // VLD4 multiple structure pseudo-instructions. These need special handling for
6614 // the vector operands that the normal instructions don't yet model.
6615 // FIXME: Remove these when the register classes and instructions are updated.
6616 def VLD4dAsm_8 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr",
6617 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6618 def VLD4dAsm_16 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr",
6619 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6620 def VLD4dAsm_32 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr",
6621 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6622 def VLD4qAsm_8 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr",
6623 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6624 def VLD4qAsm_16 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr",
6625 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6626 def VLD4qAsm_32 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr",
6627 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6629 def VLD4dWB_fixed_Asm_8 :
6630 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr!",
6631 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6632 def VLD4dWB_fixed_Asm_16 :
6633 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr!",
6634 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6635 def VLD4dWB_fixed_Asm_32 :
6636 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr!",
6637 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6638 def VLD4qWB_fixed_Asm_8 :
6639 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr!",
6640 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6641 def VLD4qWB_fixed_Asm_16 :
6642 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr!",
6643 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6644 def VLD4qWB_fixed_Asm_32 :
6645 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr!",
6646 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6647 def VLD4dWB_register_Asm_8 :
6648 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr, $Rm",
6649 (ins VecListFourD:$list, addrmode6:$addr,
6650 rGPR:$Rm, pred:$p)>;
6651 def VLD4dWB_register_Asm_16 :
6652 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr, $Rm",
6653 (ins VecListFourD:$list, addrmode6:$addr,
6654 rGPR:$Rm, pred:$p)>;
6655 def VLD4dWB_register_Asm_32 :
6656 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr, $Rm",
6657 (ins VecListFourD:$list, addrmode6:$addr,
6658 rGPR:$Rm, pred:$p)>;
6659 def VLD4qWB_register_Asm_8 :
6660 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr, $Rm",
6661 (ins VecListFourQ:$list, addrmode6:$addr,
6662 rGPR:$Rm, pred:$p)>;
6663 def VLD4qWB_register_Asm_16 :
6664 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr, $Rm",
6665 (ins VecListFourQ:$list, addrmode6:$addr,
6666 rGPR:$Rm, pred:$p)>;
6667 def VLD4qWB_register_Asm_32 :
6668 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr, $Rm",
6669 (ins VecListFourQ:$list, addrmode6:$addr,
6670 rGPR:$Rm, pred:$p)>;
6672 // VST4 single-lane pseudo-instructions. These need special handling for
6673 // the lane index that an InstAlias can't handle, so we use these instead.
6674 def VST4LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr",
6675 (ins VecListFourDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6676 def VST4LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr",
6677 (ins VecListFourDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6678 def VST4LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr",
6679 (ins VecListFourDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6680 def VST4LNqAsm_16 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr",
6681 (ins VecListFourQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6682 def VST4LNqAsm_32 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr",
6683 (ins VecListFourQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6685 def VST4LNdWB_fixed_Asm_8 :
6686 NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr!",
6687 (ins VecListFourDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6688 def VST4LNdWB_fixed_Asm_16 :
6689 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr!",
6690 (ins VecListFourDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6691 def VST4LNdWB_fixed_Asm_32 :
6692 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr!",
6693 (ins VecListFourDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6694 def VST4LNqWB_fixed_Asm_16 :
6695 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr!",
6696 (ins VecListFourQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6697 def VST4LNqWB_fixed_Asm_32 :
6698 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr!",
6699 (ins VecListFourQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6700 def VST4LNdWB_register_Asm_8 :
6701 NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr, $Rm",
6702 (ins VecListFourDByteIndexed:$list, addrmode6:$addr,
6703 rGPR:$Rm, pred:$p)>;
6704 def VST4LNdWB_register_Asm_16 :
6705 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr, $Rm",
6706 (ins VecListFourDHWordIndexed:$list, addrmode6:$addr,
6707 rGPR:$Rm, pred:$p)>;
6708 def VST4LNdWB_register_Asm_32 :
6709 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr, $Rm",
6710 (ins VecListFourDWordIndexed:$list, addrmode6:$addr,
6711 rGPR:$Rm, pred:$p)>;
6712 def VST4LNqWB_register_Asm_16 :
6713 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr, $Rm",
6714 (ins VecListFourQHWordIndexed:$list, addrmode6:$addr,
6715 rGPR:$Rm, pred:$p)>;
6716 def VST4LNqWB_register_Asm_32 :
6717 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr, $Rm",
6718 (ins VecListFourQWordIndexed:$list, addrmode6:$addr,
6719 rGPR:$Rm, pred:$p)>;
6722 // VST4 multiple structure pseudo-instructions. These need special handling for
6723 // the vector operands that the normal instructions don't yet model.
6724 // FIXME: Remove these when the register classes and instructions are updated.
6725 def VST4dAsm_8 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr",
6726 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6727 def VST4dAsm_16 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr",
6728 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6729 def VST4dAsm_32 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr",
6730 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6731 def VST4qAsm_8 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr",
6732 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6733 def VST4qAsm_16 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr",
6734 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6735 def VST4qAsm_32 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr",
6736 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6738 def VST4dWB_fixed_Asm_8 :
6739 NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr!",
6740 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6741 def VST4dWB_fixed_Asm_16 :
6742 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr!",
6743 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6744 def VST4dWB_fixed_Asm_32 :
6745 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr!",
6746 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6747 def VST4qWB_fixed_Asm_8 :
6748 NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr!",
6749 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6750 def VST4qWB_fixed_Asm_16 :
6751 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr!",
6752 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6753 def VST4qWB_fixed_Asm_32 :
6754 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr!",
6755 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6756 def VST4dWB_register_Asm_8 :
6757 NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr, $Rm",
6758 (ins VecListFourD:$list, addrmode6:$addr,
6759 rGPR:$Rm, pred:$p)>;
6760 def VST4dWB_register_Asm_16 :
6761 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr, $Rm",
6762 (ins VecListFourD:$list, addrmode6:$addr,
6763 rGPR:$Rm, pred:$p)>;
6764 def VST4dWB_register_Asm_32 :
6765 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr, $Rm",
6766 (ins VecListFourD:$list, addrmode6:$addr,
6767 rGPR:$Rm, pred:$p)>;
6768 def VST4qWB_register_Asm_8 :
6769 NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr, $Rm",
6770 (ins VecListFourQ:$list, addrmode6:$addr,
6771 rGPR:$Rm, pred:$p)>;
6772 def VST4qWB_register_Asm_16 :
6773 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr, $Rm",
6774 (ins VecListFourQ:$list, addrmode6:$addr,
6775 rGPR:$Rm, pred:$p)>;
6776 def VST4qWB_register_Asm_32 :
6777 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr, $Rm",
6778 (ins VecListFourQ:$list, addrmode6:$addr,
6779 rGPR:$Rm, pred:$p)>;
6781 // VMOV/VMVN takes an optional datatype suffix
6782 defm : NEONDTAnyInstAlias<"vmov${p}", "$Vd, $Vm",
6783 (VORRd DPR:$Vd, DPR:$Vm, DPR:$Vm, pred:$p)>;
6784 defm : NEONDTAnyInstAlias<"vmov${p}", "$Vd, $Vm",
6785 (VORRq QPR:$Vd, QPR:$Vm, QPR:$Vm, pred:$p)>;
6787 defm : NEONDTAnyInstAlias<"vmvn${p}", "$Vd, $Vm",
6788 (VMVNd DPR:$Vd, DPR:$Vm, pred:$p)>;
6789 defm : NEONDTAnyInstAlias<"vmvn${p}", "$Vd, $Vm",
6790 (VMVNq QPR:$Vd, QPR:$Vm, pred:$p)>;
6792 // VCLT (register) is an assembler alias for VCGT w/ the operands reversed.
6793 // D-register versions.
6794 def : NEONInstAlias<"vcle${p}.s8 $Dd, $Dn, $Dm",
6795 (VCGEsv8i8 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6796 def : NEONInstAlias<"vcle${p}.s16 $Dd, $Dn, $Dm",
6797 (VCGEsv4i16 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6798 def : NEONInstAlias<"vcle${p}.s32 $Dd, $Dn, $Dm",
6799 (VCGEsv2i32 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6800 def : NEONInstAlias<"vcle${p}.u8 $Dd, $Dn, $Dm",
6801 (VCGEuv8i8 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6802 def : NEONInstAlias<"vcle${p}.u16 $Dd, $Dn, $Dm",
6803 (VCGEuv4i16 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6804 def : NEONInstAlias<"vcle${p}.u32 $Dd, $Dn, $Dm",
6805 (VCGEuv2i32 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6806 def : NEONInstAlias<"vcle${p}.f32 $Dd, $Dn, $Dm",
6807 (VCGEfd DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6808 // Q-register versions.
6809 def : NEONInstAlias<"vcle${p}.s8 $Qd, $Qn, $Qm",
6810 (VCGEsv16i8 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6811 def : NEONInstAlias<"vcle${p}.s16 $Qd, $Qn, $Qm",
6812 (VCGEsv8i16 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6813 def : NEONInstAlias<"vcle${p}.s32 $Qd, $Qn, $Qm",
6814 (VCGEsv4i32 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6815 def : NEONInstAlias<"vcle${p}.u8 $Qd, $Qn, $Qm",
6816 (VCGEuv16i8 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6817 def : NEONInstAlias<"vcle${p}.u16 $Qd, $Qn, $Qm",
6818 (VCGEuv8i16 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6819 def : NEONInstAlias<"vcle${p}.u32 $Qd, $Qn, $Qm",
6820 (VCGEuv4i32 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6821 def : NEONInstAlias<"vcle${p}.f32 $Qd, $Qn, $Qm",
6822 (VCGEfq QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6824 // VCLT (register) is an assembler alias for VCGT w/ the operands reversed.
6825 // D-register versions.
6826 def : NEONInstAlias<"vclt${p}.s8 $Dd, $Dn, $Dm",
6827 (VCGTsv8i8 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6828 def : NEONInstAlias<"vclt${p}.s16 $Dd, $Dn, $Dm",
6829 (VCGTsv4i16 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6830 def : NEONInstAlias<"vclt${p}.s32 $Dd, $Dn, $Dm",
6831 (VCGTsv2i32 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6832 def : NEONInstAlias<"vclt${p}.u8 $Dd, $Dn, $Dm",
6833 (VCGTuv8i8 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6834 def : NEONInstAlias<"vclt${p}.u16 $Dd, $Dn, $Dm",
6835 (VCGTuv4i16 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6836 def : NEONInstAlias<"vclt${p}.u32 $Dd, $Dn, $Dm",
6837 (VCGTuv2i32 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6838 def : NEONInstAlias<"vclt${p}.f32 $Dd, $Dn, $Dm",
6839 (VCGTfd DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6840 // Q-register versions.
6841 def : NEONInstAlias<"vclt${p}.s8 $Qd, $Qn, $Qm",
6842 (VCGTsv16i8 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6843 def : NEONInstAlias<"vclt${p}.s16 $Qd, $Qn, $Qm",
6844 (VCGTsv8i16 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6845 def : NEONInstAlias<"vclt${p}.s32 $Qd, $Qn, $Qm",
6846 (VCGTsv4i32 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6847 def : NEONInstAlias<"vclt${p}.u8 $Qd, $Qn, $Qm",
6848 (VCGTuv16i8 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6849 def : NEONInstAlias<"vclt${p}.u16 $Qd, $Qn, $Qm",
6850 (VCGTuv8i16 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6851 def : NEONInstAlias<"vclt${p}.u32 $Qd, $Qn, $Qm",
6852 (VCGTuv4i32 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6853 def : NEONInstAlias<"vclt${p}.f32 $Qd, $Qn, $Qm",
6854 (VCGTfq QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6856 // VSWP allows, but does not require, a type suffix.
6857 defm : NEONDTAnyInstAlias<"vswp${p}", "$Vd, $Vm",
6858 (VSWPd DPR:$Vd, DPR:$Vm, pred:$p)>;
6859 defm : NEONDTAnyInstAlias<"vswp${p}", "$Vd, $Vm",
6860 (VSWPq QPR:$Vd, QPR:$Vm, pred:$p)>;
6862 // VBIF, VBIT, and VBSL allow, but do not require, a type suffix.
6863 defm : NEONDTAnyInstAlias<"vbif${p}", "$Vd, $Vn, $Vm",
6864 (VBIFd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
6865 defm : NEONDTAnyInstAlias<"vbit${p}", "$Vd, $Vn, $Vm",
6866 (VBITd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
6867 defm : NEONDTAnyInstAlias<"vbsl${p}", "$Vd, $Vn, $Vm",
6868 (VBSLd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
6869 defm : NEONDTAnyInstAlias<"vbif${p}", "$Vd, $Vn, $Vm",
6870 (VBIFq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
6871 defm : NEONDTAnyInstAlias<"vbit${p}", "$Vd, $Vn, $Vm",
6872 (VBITq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
6873 defm : NEONDTAnyInstAlias<"vbsl${p}", "$Vd, $Vn, $Vm",
6874 (VBSLq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
6876 // "vmov Rd, #-imm" can be handled via "vmvn".
6877 def : NEONInstAlias<"vmov${p}.i32 $Vd, $imm",
6878 (VMVNv2i32 DPR:$Vd, nImmVMOVI32Neg:$imm, pred:$p)>;
6879 def : NEONInstAlias<"vmov${p}.i32 $Vd, $imm",
6880 (VMVNv4i32 QPR:$Vd, nImmVMOVI32Neg:$imm, pred:$p)>;
6881 def : NEONInstAlias<"vmvn${p}.i32 $Vd, $imm",
6882 (VMOVv2i32 DPR:$Vd, nImmVMOVI32Neg:$imm, pred:$p)>;
6883 def : NEONInstAlias<"vmvn${p}.i32 $Vd, $imm",
6884 (VMOVv4i32 QPR:$Vd, nImmVMOVI32Neg:$imm, pred:$p)>;
6886 // 'gas' compatibility aliases for quad-word instructions. Strictly speaking,
6887 // these should restrict to just the Q register variants, but the register
6888 // classes are enough to match correctly regardless, so we keep it simple
6889 // and just use MnemonicAlias.
6890 def : NEONMnemonicAlias<"vbicq", "vbic">;
6891 def : NEONMnemonicAlias<"vandq", "vand">;
6892 def : NEONMnemonicAlias<"veorq", "veor">;
6893 def : NEONMnemonicAlias<"vorrq", "vorr">;
6895 def : NEONMnemonicAlias<"vmovq", "vmov">;
6896 def : NEONMnemonicAlias<"vmvnq", "vmvn">;
6897 // Explicit versions for floating point so that the FPImm variants get
6898 // handled early. The parser gets confused otherwise.
6899 def : NEONMnemonicAlias<"vmovq.f32", "vmov.f32">;
6900 def : NEONMnemonicAlias<"vmovq.f64", "vmov.f64">;
6902 def : NEONMnemonicAlias<"vaddq", "vadd">;
6903 def : NEONMnemonicAlias<"vsubq", "vsub">;
6905 def : NEONMnemonicAlias<"vminq", "vmin">;
6906 def : NEONMnemonicAlias<"vmaxq", "vmax">;
6908 def : NEONMnemonicAlias<"vmulq", "vmul">;
6910 def : NEONMnemonicAlias<"vabsq", "vabs">;
6912 def : NEONMnemonicAlias<"vshlq", "vshl">;
6913 def : NEONMnemonicAlias<"vshrq", "vshr">;
6915 def : NEONMnemonicAlias<"vcvtq", "vcvt">;
6917 def : NEONMnemonicAlias<"vcleq", "vcle">;
6918 def : NEONMnemonicAlias<"vceqq", "vceq">;
6920 def : NEONMnemonicAlias<"vzipq", "vzip">;
6921 def : NEONMnemonicAlias<"vswpq", "vswp">;
6923 def : NEONMnemonicAlias<"vrecpeq.f32", "vrecpe.f32">;
6924 def : NEONMnemonicAlias<"vrecpeq.u32", "vrecpe.u32">;
6927 // Alias for loading floating point immediates that aren't representable
6928 // using the vmov.f32 encoding but the bitpattern is representable using
6929 // the .i32 encoding.
6930 def : NEONInstAlias<"vmov${p}.f32 $Vd, $imm",
6931 (VMOVv4i32 QPR:$Vd, nImmVMOVI32:$imm, pred:$p)>;
6932 def : NEONInstAlias<"vmov${p}.f32 $Vd, $imm",
6933 (VMOVv2i32 DPR:$Vd, nImmVMOVI32:$imm, pred:$p)>;