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);
402 //===----------------------------------------------------------------------===//
403 // NEON-specific DAG Nodes.
404 //===----------------------------------------------------------------------===//
406 def SDTARMVCMP : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisSameAs<1, 2>]>;
407 def SDTARMVCMPZ : SDTypeProfile<1, 1, []>;
409 def NEONvceq : SDNode<"ARMISD::VCEQ", SDTARMVCMP>;
410 def NEONvceqz : SDNode<"ARMISD::VCEQZ", SDTARMVCMPZ>;
411 def NEONvcge : SDNode<"ARMISD::VCGE", SDTARMVCMP>;
412 def NEONvcgez : SDNode<"ARMISD::VCGEZ", SDTARMVCMPZ>;
413 def NEONvclez : SDNode<"ARMISD::VCLEZ", SDTARMVCMPZ>;
414 def NEONvcgeu : SDNode<"ARMISD::VCGEU", SDTARMVCMP>;
415 def NEONvcgt : SDNode<"ARMISD::VCGT", SDTARMVCMP>;
416 def NEONvcgtz : SDNode<"ARMISD::VCGTZ", SDTARMVCMPZ>;
417 def NEONvcltz : SDNode<"ARMISD::VCLTZ", SDTARMVCMPZ>;
418 def NEONvcgtu : SDNode<"ARMISD::VCGTU", SDTARMVCMP>;
419 def NEONvtst : SDNode<"ARMISD::VTST", SDTARMVCMP>;
421 // Types for vector shift by immediates. The "SHX" version is for long and
422 // narrow operations where the source and destination vectors have different
423 // types. The "SHINS" version is for shift and insert operations.
424 def SDTARMVSH : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
426 def SDTARMVSHX : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
428 def SDTARMVSHINS : SDTypeProfile<1, 3, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
429 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
431 def NEONvshl : SDNode<"ARMISD::VSHL", SDTARMVSH>;
432 def NEONvshrs : SDNode<"ARMISD::VSHRs", SDTARMVSH>;
433 def NEONvshru : SDNode<"ARMISD::VSHRu", SDTARMVSH>;
434 def NEONvshlls : SDNode<"ARMISD::VSHLLs", SDTARMVSHX>;
435 def NEONvshllu : SDNode<"ARMISD::VSHLLu", SDTARMVSHX>;
436 def NEONvshlli : SDNode<"ARMISD::VSHLLi", SDTARMVSHX>;
437 def NEONvshrn : SDNode<"ARMISD::VSHRN", SDTARMVSHX>;
439 def NEONvrshrs : SDNode<"ARMISD::VRSHRs", SDTARMVSH>;
440 def NEONvrshru : SDNode<"ARMISD::VRSHRu", SDTARMVSH>;
441 def NEONvrshrn : SDNode<"ARMISD::VRSHRN", SDTARMVSHX>;
443 def NEONvqshls : SDNode<"ARMISD::VQSHLs", SDTARMVSH>;
444 def NEONvqshlu : SDNode<"ARMISD::VQSHLu", SDTARMVSH>;
445 def NEONvqshlsu : SDNode<"ARMISD::VQSHLsu", SDTARMVSH>;
446 def NEONvqshrns : SDNode<"ARMISD::VQSHRNs", SDTARMVSHX>;
447 def NEONvqshrnu : SDNode<"ARMISD::VQSHRNu", SDTARMVSHX>;
448 def NEONvqshrnsu : SDNode<"ARMISD::VQSHRNsu", SDTARMVSHX>;
450 def NEONvqrshrns : SDNode<"ARMISD::VQRSHRNs", SDTARMVSHX>;
451 def NEONvqrshrnu : SDNode<"ARMISD::VQRSHRNu", SDTARMVSHX>;
452 def NEONvqrshrnsu : SDNode<"ARMISD::VQRSHRNsu", SDTARMVSHX>;
454 def NEONvsli : SDNode<"ARMISD::VSLI", SDTARMVSHINS>;
455 def NEONvsri : SDNode<"ARMISD::VSRI", SDTARMVSHINS>;
457 def SDTARMVGETLN : SDTypeProfile<1, 2, [SDTCisVT<0, i32>, SDTCisInt<1>,
459 def NEONvgetlaneu : SDNode<"ARMISD::VGETLANEu", SDTARMVGETLN>;
460 def NEONvgetlanes : SDNode<"ARMISD::VGETLANEs", SDTARMVGETLN>;
462 def SDTARMVMOVIMM : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisVT<1, i32>]>;
463 def NEONvmovImm : SDNode<"ARMISD::VMOVIMM", SDTARMVMOVIMM>;
464 def NEONvmvnImm : SDNode<"ARMISD::VMVNIMM", SDTARMVMOVIMM>;
465 def NEONvmovFPImm : SDNode<"ARMISD::VMOVFPIMM", SDTARMVMOVIMM>;
467 def SDTARMVORRIMM : SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
469 def NEONvorrImm : SDNode<"ARMISD::VORRIMM", SDTARMVORRIMM>;
470 def NEONvbicImm : SDNode<"ARMISD::VBICIMM", SDTARMVORRIMM>;
472 def NEONvbsl : SDNode<"ARMISD::VBSL",
473 SDTypeProfile<1, 3, [SDTCisVec<0>,
476 SDTCisSameAs<0, 3>]>>;
478 def NEONvdup : SDNode<"ARMISD::VDUP", SDTypeProfile<1, 1, [SDTCisVec<0>]>>;
480 // VDUPLANE can produce a quad-register result from a double-register source,
481 // so the result is not constrained to match the source.
482 def NEONvduplane : SDNode<"ARMISD::VDUPLANE",
483 SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisVec<1>,
486 def SDTARMVEXT : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
487 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
488 def NEONvext : SDNode<"ARMISD::VEXT", SDTARMVEXT>;
490 def SDTARMVSHUF : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisSameAs<0, 1>]>;
491 def NEONvrev64 : SDNode<"ARMISD::VREV64", SDTARMVSHUF>;
492 def NEONvrev32 : SDNode<"ARMISD::VREV32", SDTARMVSHUF>;
493 def NEONvrev16 : SDNode<"ARMISD::VREV16", SDTARMVSHUF>;
495 def SDTARMVSHUF2 : SDTypeProfile<2, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
497 SDTCisSameAs<0, 3>]>;
498 def NEONzip : SDNode<"ARMISD::VZIP", SDTARMVSHUF2>;
499 def NEONuzp : SDNode<"ARMISD::VUZP", SDTARMVSHUF2>;
500 def NEONtrn : SDNode<"ARMISD::VTRN", SDTARMVSHUF2>;
502 def SDTARMVMULL : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
503 SDTCisSameAs<1, 2>]>;
504 def NEONvmulls : SDNode<"ARMISD::VMULLs", SDTARMVMULL>;
505 def NEONvmullu : SDNode<"ARMISD::VMULLu", SDTARMVMULL>;
507 def SDTARMFMAX : SDTypeProfile<1, 2, [SDTCisVT<0, f32>, SDTCisSameAs<0, 1>,
508 SDTCisSameAs<0, 2>]>;
509 def NEONfmax : SDNode<"ARMISD::FMAX", SDTARMFMAX>;
510 def NEONfmin : SDNode<"ARMISD::FMIN", SDTARMFMAX>;
512 def NEONimmAllZerosV: PatLeaf<(NEONvmovImm (i32 timm)), [{
513 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
514 unsigned EltBits = 0;
515 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
516 return (EltBits == 32 && EltVal == 0);
519 def NEONimmAllOnesV: PatLeaf<(NEONvmovImm (i32 timm)), [{
520 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
521 unsigned EltBits = 0;
522 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
523 return (EltBits == 8 && EltVal == 0xff);
526 //===----------------------------------------------------------------------===//
527 // NEON load / store instructions
528 //===----------------------------------------------------------------------===//
530 // Use VLDM to load a Q register as a D register pair.
531 // This is a pseudo instruction that is expanded to VLDMD after reg alloc.
533 : PseudoVFPLdStM<(outs DPair:$dst), (ins GPR:$Rn),
535 [(set DPair:$dst, (v2f64 (load GPR:$Rn)))]>;
537 // Use VSTM to store a Q register as a D register pair.
538 // This is a pseudo instruction that is expanded to VSTMD after reg alloc.
540 : PseudoVFPLdStM<(outs), (ins DPair:$src, GPR:$Rn),
542 [(store (v2f64 DPair:$src), GPR:$Rn)]>;
544 // Classes for VLD* pseudo-instructions with multi-register operands.
545 // These are expanded to real instructions after register allocation.
546 class VLDQPseudo<InstrItinClass itin>
547 : PseudoNLdSt<(outs QPR:$dst), (ins addrmode6:$addr), itin, "">;
548 class VLDQWBPseudo<InstrItinClass itin>
549 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
550 (ins addrmode6:$addr, am6offset:$offset), itin,
552 class VLDQWBfixedPseudo<InstrItinClass itin>
553 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
554 (ins addrmode6:$addr), itin,
556 class VLDQWBregisterPseudo<InstrItinClass itin>
557 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
558 (ins addrmode6:$addr, rGPR:$offset), itin,
561 class VLDQQPseudo<InstrItinClass itin>
562 : PseudoNLdSt<(outs QQPR:$dst), (ins addrmode6:$addr), itin, "">;
563 class VLDQQWBPseudo<InstrItinClass itin>
564 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
565 (ins addrmode6:$addr, am6offset:$offset), itin,
567 class VLDQQWBfixedPseudo<InstrItinClass itin>
568 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
569 (ins addrmode6:$addr), itin,
571 class VLDQQWBregisterPseudo<InstrItinClass itin>
572 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
573 (ins addrmode6:$addr, rGPR:$offset), itin,
577 class VLDQQQQPseudo<InstrItinClass itin>
578 : PseudoNLdSt<(outs QQQQPR:$dst), (ins addrmode6:$addr, QQQQPR:$src),itin,
580 class VLDQQQQWBPseudo<InstrItinClass itin>
581 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
582 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
583 "$addr.addr = $wb, $src = $dst">;
585 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
587 // VLD1 : Vector Load (multiple single elements)
588 class VLD1D<bits<4> op7_4, string Dt>
589 : NLdSt<0,0b10,0b0111,op7_4, (outs VecListOneD:$Vd),
590 (ins addrmode6:$Rn), IIC_VLD1,
591 "vld1", Dt, "$Vd, $Rn", "", []> {
594 let DecoderMethod = "DecodeVLDInstruction";
596 class VLD1Q<bits<4> op7_4, string Dt>
597 : NLdSt<0,0b10,0b1010,op7_4, (outs VecListDPair:$Vd),
598 (ins addrmode6:$Rn), IIC_VLD1x2,
599 "vld1", Dt, "$Vd, $Rn", "", []> {
601 let Inst{5-4} = Rn{5-4};
602 let DecoderMethod = "DecodeVLDInstruction";
605 def VLD1d8 : VLD1D<{0,0,0,?}, "8">;
606 def VLD1d16 : VLD1D<{0,1,0,?}, "16">;
607 def VLD1d32 : VLD1D<{1,0,0,?}, "32">;
608 def VLD1d64 : VLD1D<{1,1,0,?}, "64">;
610 def VLD1q8 : VLD1Q<{0,0,?,?}, "8">;
611 def VLD1q16 : VLD1Q<{0,1,?,?}, "16">;
612 def VLD1q32 : VLD1Q<{1,0,?,?}, "32">;
613 def VLD1q64 : VLD1Q<{1,1,?,?}, "64">;
615 // ...with address register writeback:
616 multiclass VLD1DWB<bits<4> op7_4, string Dt> {
617 def _fixed : NLdSt<0,0b10, 0b0111,op7_4, (outs VecListOneD:$Vd, GPR:$wb),
618 (ins addrmode6:$Rn), IIC_VLD1u,
619 "vld1", Dt, "$Vd, $Rn!",
620 "$Rn.addr = $wb", []> {
621 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
623 let DecoderMethod = "DecodeVLDInstruction";
624 let AsmMatchConverter = "cvtVLDwbFixed";
626 def _register : NLdSt<0,0b10,0b0111,op7_4, (outs VecListOneD:$Vd, GPR:$wb),
627 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1u,
628 "vld1", Dt, "$Vd, $Rn, $Rm",
629 "$Rn.addr = $wb", []> {
631 let DecoderMethod = "DecodeVLDInstruction";
632 let AsmMatchConverter = "cvtVLDwbRegister";
635 multiclass VLD1QWB<bits<4> op7_4, string Dt> {
636 def _fixed : NLdSt<0,0b10,0b1010,op7_4, (outs VecListDPair:$Vd, GPR:$wb),
637 (ins addrmode6:$Rn), IIC_VLD1x2u,
638 "vld1", Dt, "$Vd, $Rn!",
639 "$Rn.addr = $wb", []> {
640 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
641 let Inst{5-4} = Rn{5-4};
642 let DecoderMethod = "DecodeVLDInstruction";
643 let AsmMatchConverter = "cvtVLDwbFixed";
645 def _register : NLdSt<0,0b10,0b1010,op7_4, (outs VecListDPair:$Vd, GPR:$wb),
646 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1x2u,
647 "vld1", Dt, "$Vd, $Rn, $Rm",
648 "$Rn.addr = $wb", []> {
649 let Inst{5-4} = Rn{5-4};
650 let DecoderMethod = "DecodeVLDInstruction";
651 let AsmMatchConverter = "cvtVLDwbRegister";
655 defm VLD1d8wb : VLD1DWB<{0,0,0,?}, "8">;
656 defm VLD1d16wb : VLD1DWB<{0,1,0,?}, "16">;
657 defm VLD1d32wb : VLD1DWB<{1,0,0,?}, "32">;
658 defm VLD1d64wb : VLD1DWB<{1,1,0,?}, "64">;
659 defm VLD1q8wb : VLD1QWB<{0,0,?,?}, "8">;
660 defm VLD1q16wb : VLD1QWB<{0,1,?,?}, "16">;
661 defm VLD1q32wb : VLD1QWB<{1,0,?,?}, "32">;
662 defm VLD1q64wb : VLD1QWB<{1,1,?,?}, "64">;
664 // ...with 3 registers
665 class VLD1D3<bits<4> op7_4, string Dt>
666 : NLdSt<0,0b10,0b0110,op7_4, (outs VecListThreeD:$Vd),
667 (ins addrmode6:$Rn), IIC_VLD1x3, "vld1", Dt,
668 "$Vd, $Rn", "", []> {
671 let DecoderMethod = "DecodeVLDInstruction";
673 multiclass VLD1D3WB<bits<4> op7_4, string Dt> {
674 def _fixed : NLdSt<0,0b10,0b0110, op7_4, (outs VecListThreeD:$Vd, GPR:$wb),
675 (ins addrmode6:$Rn), IIC_VLD1x2u,
676 "vld1", Dt, "$Vd, $Rn!",
677 "$Rn.addr = $wb", []> {
678 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
680 let DecoderMethod = "DecodeVLDInstruction";
681 let AsmMatchConverter = "cvtVLDwbFixed";
683 def _register : NLdSt<0,0b10,0b0110,op7_4, (outs VecListThreeD:$Vd, GPR:$wb),
684 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1x2u,
685 "vld1", Dt, "$Vd, $Rn, $Rm",
686 "$Rn.addr = $wb", []> {
688 let DecoderMethod = "DecodeVLDInstruction";
689 let AsmMatchConverter = "cvtVLDwbRegister";
693 def VLD1d8T : VLD1D3<{0,0,0,?}, "8">;
694 def VLD1d16T : VLD1D3<{0,1,0,?}, "16">;
695 def VLD1d32T : VLD1D3<{1,0,0,?}, "32">;
696 def VLD1d64T : VLD1D3<{1,1,0,?}, "64">;
698 defm VLD1d8Twb : VLD1D3WB<{0,0,0,?}, "8">;
699 defm VLD1d16Twb : VLD1D3WB<{0,1,0,?}, "16">;
700 defm VLD1d32Twb : VLD1D3WB<{1,0,0,?}, "32">;
701 defm VLD1d64Twb : VLD1D3WB<{1,1,0,?}, "64">;
703 def VLD1d64TPseudo : VLDQQPseudo<IIC_VLD1x3>;
705 // ...with 4 registers
706 class VLD1D4<bits<4> op7_4, string Dt>
707 : NLdSt<0, 0b10, 0b0010, op7_4, (outs VecListFourD:$Vd),
708 (ins addrmode6:$Rn), IIC_VLD1x4, "vld1", Dt,
709 "$Vd, $Rn", "", []> {
711 let Inst{5-4} = Rn{5-4};
712 let DecoderMethod = "DecodeVLDInstruction";
714 multiclass VLD1D4WB<bits<4> op7_4, string Dt> {
715 def _fixed : NLdSt<0,0b10,0b0010, op7_4, (outs VecListFourD:$Vd, GPR:$wb),
716 (ins addrmode6:$Rn), IIC_VLD1x2u,
717 "vld1", Dt, "$Vd, $Rn!",
718 "$Rn.addr = $wb", []> {
719 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
720 let Inst{5-4} = Rn{5-4};
721 let DecoderMethod = "DecodeVLDInstruction";
722 let AsmMatchConverter = "cvtVLDwbFixed";
724 def _register : NLdSt<0,0b10,0b0010,op7_4, (outs VecListFourD:$Vd, GPR:$wb),
725 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1x2u,
726 "vld1", Dt, "$Vd, $Rn, $Rm",
727 "$Rn.addr = $wb", []> {
728 let Inst{5-4} = Rn{5-4};
729 let DecoderMethod = "DecodeVLDInstruction";
730 let AsmMatchConverter = "cvtVLDwbRegister";
734 def VLD1d8Q : VLD1D4<{0,0,?,?}, "8">;
735 def VLD1d16Q : VLD1D4<{0,1,?,?}, "16">;
736 def VLD1d32Q : VLD1D4<{1,0,?,?}, "32">;
737 def VLD1d64Q : VLD1D4<{1,1,?,?}, "64">;
739 defm VLD1d8Qwb : VLD1D4WB<{0,0,?,?}, "8">;
740 defm VLD1d16Qwb : VLD1D4WB<{0,1,?,?}, "16">;
741 defm VLD1d32Qwb : VLD1D4WB<{1,0,?,?}, "32">;
742 defm VLD1d64Qwb : VLD1D4WB<{1,1,?,?}, "64">;
744 def VLD1d64QPseudo : VLDQQPseudo<IIC_VLD1x4>;
746 // VLD2 : Vector Load (multiple 2-element structures)
747 class VLD2<bits<4> op11_8, bits<4> op7_4, string Dt, RegisterOperand VdTy,
749 : NLdSt<0, 0b10, op11_8, op7_4, (outs VdTy:$Vd),
750 (ins addrmode6:$Rn), itin,
751 "vld2", Dt, "$Vd, $Rn", "", []> {
753 let Inst{5-4} = Rn{5-4};
754 let DecoderMethod = "DecodeVLDInstruction";
757 def VLD2d8 : VLD2<0b1000, {0,0,?,?}, "8", VecListDPair, IIC_VLD2>;
758 def VLD2d16 : VLD2<0b1000, {0,1,?,?}, "16", VecListDPair, IIC_VLD2>;
759 def VLD2d32 : VLD2<0b1000, {1,0,?,?}, "32", VecListDPair, IIC_VLD2>;
761 def VLD2q8 : VLD2<0b0011, {0,0,?,?}, "8", VecListFourD, IIC_VLD2x2>;
762 def VLD2q16 : VLD2<0b0011, {0,1,?,?}, "16", VecListFourD, IIC_VLD2x2>;
763 def VLD2q32 : VLD2<0b0011, {1,0,?,?}, "32", VecListFourD, IIC_VLD2x2>;
765 def VLD2q8Pseudo : VLDQQPseudo<IIC_VLD2x2>;
766 def VLD2q16Pseudo : VLDQQPseudo<IIC_VLD2x2>;
767 def VLD2q32Pseudo : VLDQQPseudo<IIC_VLD2x2>;
769 // ...with address register writeback:
770 multiclass VLD2WB<bits<4> op11_8, bits<4> op7_4, string Dt,
771 RegisterOperand VdTy, InstrItinClass itin> {
772 def _fixed : NLdSt<0, 0b10, op11_8, op7_4, (outs VdTy:$Vd, GPR:$wb),
773 (ins addrmode6:$Rn), itin,
774 "vld2", Dt, "$Vd, $Rn!",
775 "$Rn.addr = $wb", []> {
776 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
777 let Inst{5-4} = Rn{5-4};
778 let DecoderMethod = "DecodeVLDInstruction";
779 let AsmMatchConverter = "cvtVLDwbFixed";
781 def _register : NLdSt<0, 0b10, op11_8, op7_4, (outs VdTy:$Vd, GPR:$wb),
782 (ins addrmode6:$Rn, rGPR:$Rm), itin,
783 "vld2", Dt, "$Vd, $Rn, $Rm",
784 "$Rn.addr = $wb", []> {
785 let Inst{5-4} = Rn{5-4};
786 let DecoderMethod = "DecodeVLDInstruction";
787 let AsmMatchConverter = "cvtVLDwbRegister";
791 defm VLD2d8wb : VLD2WB<0b1000, {0,0,?,?}, "8", VecListDPair, IIC_VLD2u>;
792 defm VLD2d16wb : VLD2WB<0b1000, {0,1,?,?}, "16", VecListDPair, IIC_VLD2u>;
793 defm VLD2d32wb : VLD2WB<0b1000, {1,0,?,?}, "32", VecListDPair, IIC_VLD2u>;
795 defm VLD2q8wb : VLD2WB<0b0011, {0,0,?,?}, "8", VecListFourD, IIC_VLD2x2u>;
796 defm VLD2q16wb : VLD2WB<0b0011, {0,1,?,?}, "16", VecListFourD, IIC_VLD2x2u>;
797 defm VLD2q32wb : VLD2WB<0b0011, {1,0,?,?}, "32", VecListFourD, IIC_VLD2x2u>;
799 def VLD2q8PseudoWB_fixed : VLDQQWBfixedPseudo<IIC_VLD2x2u>;
800 def VLD2q16PseudoWB_fixed : VLDQQWBfixedPseudo<IIC_VLD2x2u>;
801 def VLD2q32PseudoWB_fixed : VLDQQWBfixedPseudo<IIC_VLD2x2u>;
802 def VLD2q8PseudoWB_register : VLDQQWBregisterPseudo<IIC_VLD2x2u>;
803 def VLD2q16PseudoWB_register : VLDQQWBregisterPseudo<IIC_VLD2x2u>;
804 def VLD2q32PseudoWB_register : VLDQQWBregisterPseudo<IIC_VLD2x2u>;
806 // ...with double-spaced registers
807 def VLD2b8 : VLD2<0b1001, {0,0,?,?}, "8", VecListDPairSpaced, IIC_VLD2>;
808 def VLD2b16 : VLD2<0b1001, {0,1,?,?}, "16", VecListDPairSpaced, IIC_VLD2>;
809 def VLD2b32 : VLD2<0b1001, {1,0,?,?}, "32", VecListDPairSpaced, IIC_VLD2>;
810 defm VLD2b8wb : VLD2WB<0b1001, {0,0,?,?}, "8", VecListDPairSpaced, IIC_VLD2u>;
811 defm VLD2b16wb : VLD2WB<0b1001, {0,1,?,?}, "16", VecListDPairSpaced, IIC_VLD2u>;
812 defm VLD2b32wb : VLD2WB<0b1001, {1,0,?,?}, "32", VecListDPairSpaced, IIC_VLD2u>;
814 // VLD3 : Vector Load (multiple 3-element structures)
815 class VLD3D<bits<4> op11_8, bits<4> op7_4, string Dt>
816 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
817 (ins addrmode6:$Rn), IIC_VLD3,
818 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn", "", []> {
821 let DecoderMethod = "DecodeVLDInstruction";
824 def VLD3d8 : VLD3D<0b0100, {0,0,0,?}, "8">;
825 def VLD3d16 : VLD3D<0b0100, {0,1,0,?}, "16">;
826 def VLD3d32 : VLD3D<0b0100, {1,0,0,?}, "32">;
828 def VLD3d8Pseudo : VLDQQPseudo<IIC_VLD3>;
829 def VLD3d16Pseudo : VLDQQPseudo<IIC_VLD3>;
830 def VLD3d32Pseudo : VLDQQPseudo<IIC_VLD3>;
832 // ...with address register writeback:
833 class VLD3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
834 : NLdSt<0, 0b10, op11_8, op7_4,
835 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
836 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD3u,
837 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn$Rm",
838 "$Rn.addr = $wb", []> {
840 let DecoderMethod = "DecodeVLDInstruction";
843 def VLD3d8_UPD : VLD3DWB<0b0100, {0,0,0,?}, "8">;
844 def VLD3d16_UPD : VLD3DWB<0b0100, {0,1,0,?}, "16">;
845 def VLD3d32_UPD : VLD3DWB<0b0100, {1,0,0,?}, "32">;
847 def VLD3d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
848 def VLD3d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
849 def VLD3d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
851 // ...with double-spaced registers:
852 def VLD3q8 : VLD3D<0b0101, {0,0,0,?}, "8">;
853 def VLD3q16 : VLD3D<0b0101, {0,1,0,?}, "16">;
854 def VLD3q32 : VLD3D<0b0101, {1,0,0,?}, "32">;
855 def VLD3q8_UPD : VLD3DWB<0b0101, {0,0,0,?}, "8">;
856 def VLD3q16_UPD : VLD3DWB<0b0101, {0,1,0,?}, "16">;
857 def VLD3q32_UPD : VLD3DWB<0b0101, {1,0,0,?}, "32">;
859 def VLD3q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
860 def VLD3q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
861 def VLD3q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
863 // ...alternate versions to be allocated odd register numbers:
864 def VLD3q8oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
865 def VLD3q16oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
866 def VLD3q32oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
868 def VLD3q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
869 def VLD3q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
870 def VLD3q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
872 // VLD4 : Vector Load (multiple 4-element structures)
873 class VLD4D<bits<4> op11_8, bits<4> op7_4, string Dt>
874 : NLdSt<0, 0b10, op11_8, op7_4,
875 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
876 (ins addrmode6:$Rn), IIC_VLD4,
877 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn", "", []> {
879 let Inst{5-4} = Rn{5-4};
880 let DecoderMethod = "DecodeVLDInstruction";
883 def VLD4d8 : VLD4D<0b0000, {0,0,?,?}, "8">;
884 def VLD4d16 : VLD4D<0b0000, {0,1,?,?}, "16">;
885 def VLD4d32 : VLD4D<0b0000, {1,0,?,?}, "32">;
887 def VLD4d8Pseudo : VLDQQPseudo<IIC_VLD4>;
888 def VLD4d16Pseudo : VLDQQPseudo<IIC_VLD4>;
889 def VLD4d32Pseudo : VLDQQPseudo<IIC_VLD4>;
891 // ...with address register writeback:
892 class VLD4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
893 : NLdSt<0, 0b10, op11_8, op7_4,
894 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
895 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD4u,
896 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn$Rm",
897 "$Rn.addr = $wb", []> {
898 let Inst{5-4} = Rn{5-4};
899 let DecoderMethod = "DecodeVLDInstruction";
902 def VLD4d8_UPD : VLD4DWB<0b0000, {0,0,?,?}, "8">;
903 def VLD4d16_UPD : VLD4DWB<0b0000, {0,1,?,?}, "16">;
904 def VLD4d32_UPD : VLD4DWB<0b0000, {1,0,?,?}, "32">;
906 def VLD4d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
907 def VLD4d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
908 def VLD4d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
910 // ...with double-spaced registers:
911 def VLD4q8 : VLD4D<0b0001, {0,0,?,?}, "8">;
912 def VLD4q16 : VLD4D<0b0001, {0,1,?,?}, "16">;
913 def VLD4q32 : VLD4D<0b0001, {1,0,?,?}, "32">;
914 def VLD4q8_UPD : VLD4DWB<0b0001, {0,0,?,?}, "8">;
915 def VLD4q16_UPD : VLD4DWB<0b0001, {0,1,?,?}, "16">;
916 def VLD4q32_UPD : VLD4DWB<0b0001, {1,0,?,?}, "32">;
918 def VLD4q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
919 def VLD4q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
920 def VLD4q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
922 // ...alternate versions to be allocated odd register numbers:
923 def VLD4q8oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
924 def VLD4q16oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
925 def VLD4q32oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
927 def VLD4q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
928 def VLD4q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
929 def VLD4q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
931 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
933 // Classes for VLD*LN pseudo-instructions with multi-register operands.
934 // These are expanded to real instructions after register allocation.
935 class VLDQLNPseudo<InstrItinClass itin>
936 : PseudoNLdSt<(outs QPR:$dst),
937 (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
938 itin, "$src = $dst">;
939 class VLDQLNWBPseudo<InstrItinClass itin>
940 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
941 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
942 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
943 class VLDQQLNPseudo<InstrItinClass itin>
944 : PseudoNLdSt<(outs QQPR:$dst),
945 (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
946 itin, "$src = $dst">;
947 class VLDQQLNWBPseudo<InstrItinClass itin>
948 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
949 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
950 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
951 class VLDQQQQLNPseudo<InstrItinClass itin>
952 : PseudoNLdSt<(outs QQQQPR:$dst),
953 (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
954 itin, "$src = $dst">;
955 class VLDQQQQLNWBPseudo<InstrItinClass itin>
956 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
957 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
958 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
960 // VLD1LN : Vector Load (single element to one lane)
961 class VLD1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
963 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd),
964 (ins addrmode6:$Rn, DPR:$src, nohash_imm:$lane),
965 IIC_VLD1ln, "vld1", Dt, "\\{$Vd[$lane]\\}, $Rn",
967 [(set DPR:$Vd, (vector_insert (Ty DPR:$src),
968 (i32 (LoadOp addrmode6:$Rn)),
971 let DecoderMethod = "DecodeVLD1LN";
973 class VLD1LN32<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
975 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd),
976 (ins addrmode6oneL32:$Rn, DPR:$src, nohash_imm:$lane),
977 IIC_VLD1ln, "vld1", Dt, "\\{$Vd[$lane]\\}, $Rn",
979 [(set DPR:$Vd, (vector_insert (Ty DPR:$src),
980 (i32 (LoadOp addrmode6oneL32:$Rn)),
983 let DecoderMethod = "DecodeVLD1LN";
985 class VLD1QLNPseudo<ValueType Ty, PatFrag LoadOp> : VLDQLNPseudo<IIC_VLD1ln> {
986 let Pattern = [(set QPR:$dst, (vector_insert (Ty QPR:$src),
987 (i32 (LoadOp addrmode6:$addr)),
991 def VLD1LNd8 : VLD1LN<0b0000, {?,?,?,0}, "8", v8i8, extloadi8> {
992 let Inst{7-5} = lane{2-0};
994 def VLD1LNd16 : VLD1LN<0b0100, {?,?,0,?}, "16", v4i16, extloadi16> {
995 let Inst{7-6} = lane{1-0};
996 let Inst{5-4} = Rn{5-4};
998 def VLD1LNd32 : VLD1LN32<0b1000, {?,0,?,?}, "32", v2i32, load> {
999 let Inst{7} = lane{0};
1000 let Inst{5-4} = Rn{5-4};
1003 def VLD1LNq8Pseudo : VLD1QLNPseudo<v16i8, extloadi8>;
1004 def VLD1LNq16Pseudo : VLD1QLNPseudo<v8i16, extloadi16>;
1005 def VLD1LNq32Pseudo : VLD1QLNPseudo<v4i32, load>;
1007 def : Pat<(vector_insert (v2f32 DPR:$src),
1008 (f32 (load addrmode6:$addr)), imm:$lane),
1009 (VLD1LNd32 addrmode6:$addr, DPR:$src, imm:$lane)>;
1010 def : Pat<(vector_insert (v4f32 QPR:$src),
1011 (f32 (load addrmode6:$addr)), imm:$lane),
1012 (VLD1LNq32Pseudo addrmode6:$addr, QPR:$src, imm:$lane)>;
1014 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
1016 // ...with address register writeback:
1017 class VLD1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1018 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, GPR:$wb),
1019 (ins addrmode6:$Rn, am6offset:$Rm,
1020 DPR:$src, nohash_imm:$lane), IIC_VLD1lnu, "vld1", Dt,
1021 "\\{$Vd[$lane]\\}, $Rn$Rm",
1022 "$src = $Vd, $Rn.addr = $wb", []> {
1023 let DecoderMethod = "DecodeVLD1LN";
1026 def VLD1LNd8_UPD : VLD1LNWB<0b0000, {?,?,?,0}, "8"> {
1027 let Inst{7-5} = lane{2-0};
1029 def VLD1LNd16_UPD : VLD1LNWB<0b0100, {?,?,0,?}, "16"> {
1030 let Inst{7-6} = lane{1-0};
1031 let Inst{4} = Rn{4};
1033 def VLD1LNd32_UPD : VLD1LNWB<0b1000, {?,0,?,?}, "32"> {
1034 let Inst{7} = lane{0};
1035 let Inst{5} = Rn{4};
1036 let Inst{4} = Rn{4};
1039 def VLD1LNq8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
1040 def VLD1LNq16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
1041 def VLD1LNq32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
1043 // VLD2LN : Vector Load (single 2-element structure to one lane)
1044 class VLD2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1045 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2),
1046 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, nohash_imm:$lane),
1047 IIC_VLD2ln, "vld2", Dt, "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn",
1048 "$src1 = $Vd, $src2 = $dst2", []> {
1050 let Inst{4} = Rn{4};
1051 let DecoderMethod = "DecodeVLD2LN";
1054 def VLD2LNd8 : VLD2LN<0b0001, {?,?,?,?}, "8"> {
1055 let Inst{7-5} = lane{2-0};
1057 def VLD2LNd16 : VLD2LN<0b0101, {?,?,0,?}, "16"> {
1058 let Inst{7-6} = lane{1-0};
1060 def VLD2LNd32 : VLD2LN<0b1001, {?,0,0,?}, "32"> {
1061 let Inst{7} = lane{0};
1064 def VLD2LNd8Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
1065 def VLD2LNd16Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
1066 def VLD2LNd32Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
1068 // ...with double-spaced registers:
1069 def VLD2LNq16 : VLD2LN<0b0101, {?,?,1,?}, "16"> {
1070 let Inst{7-6} = lane{1-0};
1072 def VLD2LNq32 : VLD2LN<0b1001, {?,1,0,?}, "32"> {
1073 let Inst{7} = lane{0};
1076 def VLD2LNq16Pseudo : VLDQQLNPseudo<IIC_VLD2ln>;
1077 def VLD2LNq32Pseudo : VLDQQLNPseudo<IIC_VLD2ln>;
1079 // ...with address register writeback:
1080 class VLD2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1081 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
1082 (ins addrmode6:$Rn, am6offset:$Rm,
1083 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VLD2lnu, "vld2", Dt,
1084 "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn$Rm",
1085 "$src1 = $Vd, $src2 = $dst2, $Rn.addr = $wb", []> {
1086 let Inst{4} = Rn{4};
1087 let DecoderMethod = "DecodeVLD2LN";
1090 def VLD2LNd8_UPD : VLD2LNWB<0b0001, {?,?,?,?}, "8"> {
1091 let Inst{7-5} = lane{2-0};
1093 def VLD2LNd16_UPD : VLD2LNWB<0b0101, {?,?,0,?}, "16"> {
1094 let Inst{7-6} = lane{1-0};
1096 def VLD2LNd32_UPD : VLD2LNWB<0b1001, {?,0,0,?}, "32"> {
1097 let Inst{7} = lane{0};
1100 def VLD2LNd8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
1101 def VLD2LNd16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
1102 def VLD2LNd32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
1104 def VLD2LNq16_UPD : VLD2LNWB<0b0101, {?,?,1,?}, "16"> {
1105 let Inst{7-6} = lane{1-0};
1107 def VLD2LNq32_UPD : VLD2LNWB<0b1001, {?,1,0,?}, "32"> {
1108 let Inst{7} = lane{0};
1111 def VLD2LNq16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
1112 def VLD2LNq32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
1114 // VLD3LN : Vector Load (single 3-element structure to one lane)
1115 class VLD3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1116 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
1117 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3,
1118 nohash_imm:$lane), IIC_VLD3ln, "vld3", Dt,
1119 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn",
1120 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3", []> {
1122 let DecoderMethod = "DecodeVLD3LN";
1125 def VLD3LNd8 : VLD3LN<0b0010, {?,?,?,0}, "8"> {
1126 let Inst{7-5} = lane{2-0};
1128 def VLD3LNd16 : VLD3LN<0b0110, {?,?,0,0}, "16"> {
1129 let Inst{7-6} = lane{1-0};
1131 def VLD3LNd32 : VLD3LN<0b1010, {?,0,0,0}, "32"> {
1132 let Inst{7} = lane{0};
1135 def VLD3LNd8Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
1136 def VLD3LNd16Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
1137 def VLD3LNd32Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
1139 // ...with double-spaced registers:
1140 def VLD3LNq16 : VLD3LN<0b0110, {?,?,1,0}, "16"> {
1141 let Inst{7-6} = lane{1-0};
1143 def VLD3LNq32 : VLD3LN<0b1010, {?,1,0,0}, "32"> {
1144 let Inst{7} = lane{0};
1147 def VLD3LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
1148 def VLD3LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
1150 // ...with address register writeback:
1151 class VLD3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1152 : NLdStLn<1, 0b10, op11_8, op7_4,
1153 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
1154 (ins addrmode6:$Rn, am6offset:$Rm,
1155 DPR:$src1, DPR:$src2, DPR:$src3, nohash_imm:$lane),
1156 IIC_VLD3lnu, "vld3", Dt,
1157 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn$Rm",
1158 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $Rn.addr = $wb",
1160 let DecoderMethod = "DecodeVLD3LN";
1163 def VLD3LNd8_UPD : VLD3LNWB<0b0010, {?,?,?,0}, "8"> {
1164 let Inst{7-5} = lane{2-0};
1166 def VLD3LNd16_UPD : VLD3LNWB<0b0110, {?,?,0,0}, "16"> {
1167 let Inst{7-6} = lane{1-0};
1169 def VLD3LNd32_UPD : VLD3LNWB<0b1010, {?,0,0,0}, "32"> {
1170 let Inst{7} = lane{0};
1173 def VLD3LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
1174 def VLD3LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
1175 def VLD3LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
1177 def VLD3LNq16_UPD : VLD3LNWB<0b0110, {?,?,1,0}, "16"> {
1178 let Inst{7-6} = lane{1-0};
1180 def VLD3LNq32_UPD : VLD3LNWB<0b1010, {?,1,0,0}, "32"> {
1181 let Inst{7} = lane{0};
1184 def VLD3LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
1185 def VLD3LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
1187 // VLD4LN : Vector Load (single 4-element structure to one lane)
1188 class VLD4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1189 : NLdStLn<1, 0b10, op11_8, op7_4,
1190 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
1191 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
1192 nohash_imm:$lane), IIC_VLD4ln, "vld4", Dt,
1193 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn",
1194 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4", []> {
1196 let Inst{4} = Rn{4};
1197 let DecoderMethod = "DecodeVLD4LN";
1200 def VLD4LNd8 : VLD4LN<0b0011, {?,?,?,?}, "8"> {
1201 let Inst{7-5} = lane{2-0};
1203 def VLD4LNd16 : VLD4LN<0b0111, {?,?,0,?}, "16"> {
1204 let Inst{7-6} = lane{1-0};
1206 def VLD4LNd32 : VLD4LN<0b1011, {?,0,?,?}, "32"> {
1207 let Inst{7} = lane{0};
1208 let Inst{5} = Rn{5};
1211 def VLD4LNd8Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
1212 def VLD4LNd16Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
1213 def VLD4LNd32Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
1215 // ...with double-spaced registers:
1216 def VLD4LNq16 : VLD4LN<0b0111, {?,?,1,?}, "16"> {
1217 let Inst{7-6} = lane{1-0};
1219 def VLD4LNq32 : VLD4LN<0b1011, {?,1,?,?}, "32"> {
1220 let Inst{7} = lane{0};
1221 let Inst{5} = Rn{5};
1224 def VLD4LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>;
1225 def VLD4LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>;
1227 // ...with address register writeback:
1228 class VLD4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1229 : NLdStLn<1, 0b10, op11_8, op7_4,
1230 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
1231 (ins addrmode6:$Rn, am6offset:$Rm,
1232 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
1233 IIC_VLD4lnu, "vld4", Dt,
1234 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn$Rm",
1235 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4, $Rn.addr = $wb",
1237 let Inst{4} = Rn{4};
1238 let DecoderMethod = "DecodeVLD4LN" ;
1241 def VLD4LNd8_UPD : VLD4LNWB<0b0011, {?,?,?,?}, "8"> {
1242 let Inst{7-5} = lane{2-0};
1244 def VLD4LNd16_UPD : VLD4LNWB<0b0111, {?,?,0,?}, "16"> {
1245 let Inst{7-6} = lane{1-0};
1247 def VLD4LNd32_UPD : VLD4LNWB<0b1011, {?,0,?,?}, "32"> {
1248 let Inst{7} = lane{0};
1249 let Inst{5} = Rn{5};
1252 def VLD4LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
1253 def VLD4LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
1254 def VLD4LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
1256 def VLD4LNq16_UPD : VLD4LNWB<0b0111, {?,?,1,?}, "16"> {
1257 let Inst{7-6} = lane{1-0};
1259 def VLD4LNq32_UPD : VLD4LNWB<0b1011, {?,1,?,?}, "32"> {
1260 let Inst{7} = lane{0};
1261 let Inst{5} = Rn{5};
1264 def VLD4LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
1265 def VLD4LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
1267 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
1269 // VLD1DUP : Vector Load (single element to all lanes)
1270 class VLD1DUP<bits<4> op7_4, string Dt, ValueType Ty, PatFrag LoadOp>
1271 : NLdSt<1, 0b10, 0b1100, op7_4, (outs VecListOneDAllLanes:$Vd),
1272 (ins addrmode6dup:$Rn),
1273 IIC_VLD1dup, "vld1", Dt, "$Vd, $Rn", "",
1274 [(set VecListOneDAllLanes:$Vd,
1275 (Ty (NEONvdup (i32 (LoadOp addrmode6dup:$Rn)))))]> {
1277 let Inst{4} = Rn{4};
1278 let DecoderMethod = "DecodeVLD1DupInstruction";
1280 def VLD1DUPd8 : VLD1DUP<{0,0,0,?}, "8", v8i8, extloadi8>;
1281 def VLD1DUPd16 : VLD1DUP<{0,1,0,?}, "16", v4i16, extloadi16>;
1282 def VLD1DUPd32 : VLD1DUP<{1,0,0,?}, "32", v2i32, load>;
1284 def : Pat<(v2f32 (NEONvdup (f32 (load addrmode6dup:$addr)))),
1285 (VLD1DUPd32 addrmode6:$addr)>;
1287 class VLD1QDUP<bits<4> op7_4, string Dt, ValueType Ty, PatFrag LoadOp>
1288 : NLdSt<1, 0b10, 0b1100, op7_4, (outs VecListDPairAllLanes:$Vd),
1289 (ins addrmode6dup:$Rn), IIC_VLD1dup,
1290 "vld1", Dt, "$Vd, $Rn", "",
1291 [(set VecListDPairAllLanes:$Vd,
1292 (Ty (NEONvdup (i32 (LoadOp addrmode6dup:$Rn)))))]> {
1294 let Inst{4} = Rn{4};
1295 let DecoderMethod = "DecodeVLD1DupInstruction";
1298 def VLD1DUPq8 : VLD1QDUP<{0,0,1,0}, "8", v16i8, extloadi8>;
1299 def VLD1DUPq16 : VLD1QDUP<{0,1,1,?}, "16", v8i16, extloadi16>;
1300 def VLD1DUPq32 : VLD1QDUP<{1,0,1,?}, "32", v4i32, load>;
1302 def : Pat<(v4f32 (NEONvdup (f32 (load addrmode6dup:$addr)))),
1303 (VLD1DUPq32 addrmode6:$addr)>;
1305 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
1306 // ...with address register writeback:
1307 multiclass VLD1DUPWB<bits<4> op7_4, string Dt> {
1308 def _fixed : NLdSt<1, 0b10, 0b1100, op7_4,
1309 (outs VecListOneDAllLanes:$Vd, GPR:$wb),
1310 (ins addrmode6dup:$Rn), IIC_VLD1dupu,
1311 "vld1", Dt, "$Vd, $Rn!",
1312 "$Rn.addr = $wb", []> {
1313 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1314 let Inst{4} = Rn{4};
1315 let DecoderMethod = "DecodeVLD1DupInstruction";
1316 let AsmMatchConverter = "cvtVLDwbFixed";
1318 def _register : NLdSt<1, 0b10, 0b1100, op7_4,
1319 (outs VecListOneDAllLanes:$Vd, GPR:$wb),
1320 (ins addrmode6dup:$Rn, rGPR:$Rm), IIC_VLD1dupu,
1321 "vld1", Dt, "$Vd, $Rn, $Rm",
1322 "$Rn.addr = $wb", []> {
1323 let Inst{4} = Rn{4};
1324 let DecoderMethod = "DecodeVLD1DupInstruction";
1325 let AsmMatchConverter = "cvtVLDwbRegister";
1328 multiclass VLD1QDUPWB<bits<4> op7_4, string Dt> {
1329 def _fixed : NLdSt<1, 0b10, 0b1100, op7_4,
1330 (outs VecListDPairAllLanes:$Vd, GPR:$wb),
1331 (ins addrmode6dup:$Rn), IIC_VLD1dupu,
1332 "vld1", Dt, "$Vd, $Rn!",
1333 "$Rn.addr = $wb", []> {
1334 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1335 let Inst{4} = Rn{4};
1336 let DecoderMethod = "DecodeVLD1DupInstruction";
1337 let AsmMatchConverter = "cvtVLDwbFixed";
1339 def _register : NLdSt<1, 0b10, 0b1100, op7_4,
1340 (outs VecListDPairAllLanes:$Vd, GPR:$wb),
1341 (ins addrmode6dup:$Rn, rGPR:$Rm), IIC_VLD1dupu,
1342 "vld1", Dt, "$Vd, $Rn, $Rm",
1343 "$Rn.addr = $wb", []> {
1344 let Inst{4} = Rn{4};
1345 let DecoderMethod = "DecodeVLD1DupInstruction";
1346 let AsmMatchConverter = "cvtVLDwbRegister";
1350 defm VLD1DUPd8wb : VLD1DUPWB<{0,0,0,0}, "8">;
1351 defm VLD1DUPd16wb : VLD1DUPWB<{0,1,0,?}, "16">;
1352 defm VLD1DUPd32wb : VLD1DUPWB<{1,0,0,?}, "32">;
1354 defm VLD1DUPq8wb : VLD1QDUPWB<{0,0,1,0}, "8">;
1355 defm VLD1DUPq16wb : VLD1QDUPWB<{0,1,1,?}, "16">;
1356 defm VLD1DUPq32wb : VLD1QDUPWB<{1,0,1,?}, "32">;
1358 // VLD2DUP : Vector Load (single 2-element structure to all lanes)
1359 class VLD2DUP<bits<4> op7_4, string Dt, RegisterOperand VdTy>
1360 : NLdSt<1, 0b10, 0b1101, op7_4, (outs VdTy:$Vd),
1361 (ins addrmode6dup:$Rn), IIC_VLD2dup,
1362 "vld2", Dt, "$Vd, $Rn", "", []> {
1364 let Inst{4} = Rn{4};
1365 let DecoderMethod = "DecodeVLD2DupInstruction";
1368 def VLD2DUPd8 : VLD2DUP<{0,0,0,?}, "8", VecListDPairAllLanes>;
1369 def VLD2DUPd16 : VLD2DUP<{0,1,0,?}, "16", VecListDPairAllLanes>;
1370 def VLD2DUPd32 : VLD2DUP<{1,0,0,?}, "32", VecListDPairAllLanes>;
1372 // ...with double-spaced registers
1373 def VLD2DUPd8x2 : VLD2DUP<{0,0,1,?}, "8", VecListDPairSpacedAllLanes>;
1374 def VLD2DUPd16x2 : VLD2DUP<{0,1,1,?}, "16", VecListDPairSpacedAllLanes>;
1375 def VLD2DUPd32x2 : VLD2DUP<{1,0,1,?}, "32", VecListDPairSpacedAllLanes>;
1377 // ...with address register writeback:
1378 multiclass VLD2DUPWB<bits<4> op7_4, string Dt, RegisterOperand VdTy> {
1379 def _fixed : NLdSt<1, 0b10, 0b1101, op7_4,
1380 (outs VdTy:$Vd, GPR:$wb),
1381 (ins addrmode6dup:$Rn), IIC_VLD2dupu,
1382 "vld2", Dt, "$Vd, $Rn!",
1383 "$Rn.addr = $wb", []> {
1384 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1385 let Inst{4} = Rn{4};
1386 let DecoderMethod = "DecodeVLD2DupInstruction";
1387 let AsmMatchConverter = "cvtVLDwbFixed";
1389 def _register : NLdSt<1, 0b10, 0b1101, op7_4,
1390 (outs VdTy:$Vd, GPR:$wb),
1391 (ins addrmode6dup:$Rn, rGPR:$Rm), IIC_VLD2dupu,
1392 "vld2", Dt, "$Vd, $Rn, $Rm",
1393 "$Rn.addr = $wb", []> {
1394 let Inst{4} = Rn{4};
1395 let DecoderMethod = "DecodeVLD2DupInstruction";
1396 let AsmMatchConverter = "cvtVLDwbRegister";
1400 defm VLD2DUPd8wb : VLD2DUPWB<{0,0,0,0}, "8", VecListDPairAllLanes>;
1401 defm VLD2DUPd16wb : VLD2DUPWB<{0,1,0,?}, "16", VecListDPairAllLanes>;
1402 defm VLD2DUPd32wb : VLD2DUPWB<{1,0,0,?}, "32", VecListDPairAllLanes>;
1404 defm VLD2DUPd8x2wb : VLD2DUPWB<{0,0,1,0}, "8", VecListDPairSpacedAllLanes>;
1405 defm VLD2DUPd16x2wb : VLD2DUPWB<{0,1,1,?}, "16", VecListDPairSpacedAllLanes>;
1406 defm VLD2DUPd32x2wb : VLD2DUPWB<{1,0,1,?}, "32", VecListDPairSpacedAllLanes>;
1408 // VLD3DUP : Vector Load (single 3-element structure to all lanes)
1409 class VLD3DUP<bits<4> op7_4, string Dt>
1410 : NLdSt<1, 0b10, 0b1110, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
1411 (ins addrmode6dup:$Rn), IIC_VLD3dup,
1412 "vld3", Dt, "\\{$Vd[], $dst2[], $dst3[]\\}, $Rn", "", []> {
1415 let DecoderMethod = "DecodeVLD3DupInstruction";
1418 def VLD3DUPd8 : VLD3DUP<{0,0,0,?}, "8">;
1419 def VLD3DUPd16 : VLD3DUP<{0,1,0,?}, "16">;
1420 def VLD3DUPd32 : VLD3DUP<{1,0,0,?}, "32">;
1422 def VLD3DUPd8Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1423 def VLD3DUPd16Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1424 def VLD3DUPd32Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1426 // ...with double-spaced registers (not used for codegen):
1427 def VLD3DUPq8 : VLD3DUP<{0,0,1,?}, "8">;
1428 def VLD3DUPq16 : VLD3DUP<{0,1,1,?}, "16">;
1429 def VLD3DUPq32 : VLD3DUP<{1,0,1,?}, "32">;
1431 // ...with address register writeback:
1432 class VLD3DUPWB<bits<4> op7_4, string Dt>
1433 : NLdSt<1, 0b10, 0b1110, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
1434 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD3dupu,
1435 "vld3", Dt, "\\{$Vd[], $dst2[], $dst3[]\\}, $Rn$Rm",
1436 "$Rn.addr = $wb", []> {
1438 let DecoderMethod = "DecodeVLD3DupInstruction";
1441 def VLD3DUPd8_UPD : VLD3DUPWB<{0,0,0,0}, "8">;
1442 def VLD3DUPd16_UPD : VLD3DUPWB<{0,1,0,?}, "16">;
1443 def VLD3DUPd32_UPD : VLD3DUPWB<{1,0,0,?}, "32">;
1445 def VLD3DUPq8_UPD : VLD3DUPWB<{0,0,1,0}, "8">;
1446 def VLD3DUPq16_UPD : VLD3DUPWB<{0,1,1,?}, "16">;
1447 def VLD3DUPq32_UPD : VLD3DUPWB<{1,0,1,?}, "32">;
1449 def VLD3DUPd8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1450 def VLD3DUPd16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1451 def VLD3DUPd32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1453 // VLD4DUP : Vector Load (single 4-element structure to all lanes)
1454 class VLD4DUP<bits<4> op7_4, string Dt>
1455 : NLdSt<1, 0b10, 0b1111, op7_4,
1456 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
1457 (ins addrmode6dup:$Rn), IIC_VLD4dup,
1458 "vld4", Dt, "\\{$Vd[], $dst2[], $dst3[], $dst4[]\\}, $Rn", "", []> {
1460 let Inst{4} = Rn{4};
1461 let DecoderMethod = "DecodeVLD4DupInstruction";
1464 def VLD4DUPd8 : VLD4DUP<{0,0,0,?}, "8">;
1465 def VLD4DUPd16 : VLD4DUP<{0,1,0,?}, "16">;
1466 def VLD4DUPd32 : VLD4DUP<{1,?,0,?}, "32"> { let Inst{6} = Rn{5}; }
1468 def VLD4DUPd8Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1469 def VLD4DUPd16Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1470 def VLD4DUPd32Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1472 // ...with double-spaced registers (not used for codegen):
1473 def VLD4DUPq8 : VLD4DUP<{0,0,1,?}, "8">;
1474 def VLD4DUPq16 : VLD4DUP<{0,1,1,?}, "16">;
1475 def VLD4DUPq32 : VLD4DUP<{1,?,1,?}, "32"> { let Inst{6} = Rn{5}; }
1477 // ...with address register writeback:
1478 class VLD4DUPWB<bits<4> op7_4, string Dt>
1479 : NLdSt<1, 0b10, 0b1111, op7_4,
1480 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
1481 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD4dupu,
1482 "vld4", Dt, "\\{$Vd[], $dst2[], $dst3[], $dst4[]\\}, $Rn$Rm",
1483 "$Rn.addr = $wb", []> {
1484 let Inst{4} = Rn{4};
1485 let DecoderMethod = "DecodeVLD4DupInstruction";
1488 def VLD4DUPd8_UPD : VLD4DUPWB<{0,0,0,0}, "8">;
1489 def VLD4DUPd16_UPD : VLD4DUPWB<{0,1,0,?}, "16">;
1490 def VLD4DUPd32_UPD : VLD4DUPWB<{1,?,0,?}, "32"> { let Inst{6} = Rn{5}; }
1492 def VLD4DUPq8_UPD : VLD4DUPWB<{0,0,1,0}, "8">;
1493 def VLD4DUPq16_UPD : VLD4DUPWB<{0,1,1,?}, "16">;
1494 def VLD4DUPq32_UPD : VLD4DUPWB<{1,?,1,?}, "32"> { let Inst{6} = Rn{5}; }
1496 def VLD4DUPd8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1497 def VLD4DUPd16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1498 def VLD4DUPd32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1500 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
1502 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
1504 // Classes for VST* pseudo-instructions with multi-register operands.
1505 // These are expanded to real instructions after register allocation.
1506 class VSTQPseudo<InstrItinClass itin>
1507 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src), itin, "">;
1508 class VSTQWBPseudo<InstrItinClass itin>
1509 : PseudoNLdSt<(outs GPR:$wb),
1510 (ins addrmode6:$addr, am6offset:$offset, QPR:$src), itin,
1511 "$addr.addr = $wb">;
1512 class VSTQWBfixedPseudo<InstrItinClass itin>
1513 : PseudoNLdSt<(outs GPR:$wb),
1514 (ins addrmode6:$addr, QPR:$src), itin,
1515 "$addr.addr = $wb">;
1516 class VSTQWBregisterPseudo<InstrItinClass itin>
1517 : PseudoNLdSt<(outs GPR:$wb),
1518 (ins addrmode6:$addr, rGPR:$offset, QPR:$src), itin,
1519 "$addr.addr = $wb">;
1520 class VSTQQPseudo<InstrItinClass itin>
1521 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src), itin, "">;
1522 class VSTQQWBPseudo<InstrItinClass itin>
1523 : PseudoNLdSt<(outs GPR:$wb),
1524 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src), itin,
1525 "$addr.addr = $wb">;
1526 class VSTQQWBfixedPseudo<InstrItinClass itin>
1527 : PseudoNLdSt<(outs GPR:$wb),
1528 (ins addrmode6:$addr, QQPR:$src), itin,
1529 "$addr.addr = $wb">;
1530 class VSTQQWBregisterPseudo<InstrItinClass itin>
1531 : PseudoNLdSt<(outs GPR:$wb),
1532 (ins addrmode6:$addr, rGPR:$offset, QQPR:$src), itin,
1533 "$addr.addr = $wb">;
1535 class VSTQQQQPseudo<InstrItinClass itin>
1536 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src), itin, "">;
1537 class VSTQQQQWBPseudo<InstrItinClass itin>
1538 : PseudoNLdSt<(outs GPR:$wb),
1539 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
1540 "$addr.addr = $wb">;
1542 // VST1 : Vector Store (multiple single elements)
1543 class VST1D<bits<4> op7_4, string Dt>
1544 : NLdSt<0,0b00,0b0111,op7_4, (outs), (ins addrmode6:$Rn, VecListOneD:$Vd),
1545 IIC_VST1, "vst1", Dt, "$Vd, $Rn", "", []> {
1547 let Inst{4} = Rn{4};
1548 let DecoderMethod = "DecodeVSTInstruction";
1550 class VST1Q<bits<4> op7_4, string Dt>
1551 : NLdSt<0,0b00,0b1010,op7_4, (outs), (ins addrmode6:$Rn, VecListDPair:$Vd),
1552 IIC_VST1x2, "vst1", Dt, "$Vd, $Rn", "", []> {
1554 let Inst{5-4} = Rn{5-4};
1555 let DecoderMethod = "DecodeVSTInstruction";
1558 def VST1d8 : VST1D<{0,0,0,?}, "8">;
1559 def VST1d16 : VST1D<{0,1,0,?}, "16">;
1560 def VST1d32 : VST1D<{1,0,0,?}, "32">;
1561 def VST1d64 : VST1D<{1,1,0,?}, "64">;
1563 def VST1q8 : VST1Q<{0,0,?,?}, "8">;
1564 def VST1q16 : VST1Q<{0,1,?,?}, "16">;
1565 def VST1q32 : VST1Q<{1,0,?,?}, "32">;
1566 def VST1q64 : VST1Q<{1,1,?,?}, "64">;
1568 // ...with address register writeback:
1569 multiclass VST1DWB<bits<4> op7_4, string Dt> {
1570 def _fixed : NLdSt<0,0b00, 0b0111,op7_4, (outs GPR:$wb),
1571 (ins addrmode6:$Rn, VecListOneD:$Vd), IIC_VLD1u,
1572 "vst1", Dt, "$Vd, $Rn!",
1573 "$Rn.addr = $wb", []> {
1574 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1575 let Inst{4} = Rn{4};
1576 let DecoderMethod = "DecodeVSTInstruction";
1577 let AsmMatchConverter = "cvtVSTwbFixed";
1579 def _register : NLdSt<0,0b00,0b0111,op7_4, (outs GPR:$wb),
1580 (ins addrmode6:$Rn, rGPR:$Rm, VecListOneD:$Vd),
1582 "vst1", Dt, "$Vd, $Rn, $Rm",
1583 "$Rn.addr = $wb", []> {
1584 let Inst{4} = Rn{4};
1585 let DecoderMethod = "DecodeVSTInstruction";
1586 let AsmMatchConverter = "cvtVSTwbRegister";
1589 multiclass VST1QWB<bits<4> op7_4, string Dt> {
1590 def _fixed : NLdSt<0,0b00,0b1010,op7_4, (outs GPR:$wb),
1591 (ins addrmode6:$Rn, VecListDPair:$Vd), IIC_VLD1x2u,
1592 "vst1", Dt, "$Vd, $Rn!",
1593 "$Rn.addr = $wb", []> {
1594 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1595 let Inst{5-4} = Rn{5-4};
1596 let DecoderMethod = "DecodeVSTInstruction";
1597 let AsmMatchConverter = "cvtVSTwbFixed";
1599 def _register : NLdSt<0,0b00,0b1010,op7_4, (outs GPR:$wb),
1600 (ins addrmode6:$Rn, rGPR:$Rm, VecListDPair:$Vd),
1602 "vst1", Dt, "$Vd, $Rn, $Rm",
1603 "$Rn.addr = $wb", []> {
1604 let Inst{5-4} = Rn{5-4};
1605 let DecoderMethod = "DecodeVSTInstruction";
1606 let AsmMatchConverter = "cvtVSTwbRegister";
1610 defm VST1d8wb : VST1DWB<{0,0,0,?}, "8">;
1611 defm VST1d16wb : VST1DWB<{0,1,0,?}, "16">;
1612 defm VST1d32wb : VST1DWB<{1,0,0,?}, "32">;
1613 defm VST1d64wb : VST1DWB<{1,1,0,?}, "64">;
1615 defm VST1q8wb : VST1QWB<{0,0,?,?}, "8">;
1616 defm VST1q16wb : VST1QWB<{0,1,?,?}, "16">;
1617 defm VST1q32wb : VST1QWB<{1,0,?,?}, "32">;
1618 defm VST1q64wb : VST1QWB<{1,1,?,?}, "64">;
1620 // ...with 3 registers
1621 class VST1D3<bits<4> op7_4, string Dt>
1622 : NLdSt<0, 0b00, 0b0110, op7_4, (outs),
1623 (ins addrmode6:$Rn, VecListThreeD:$Vd),
1624 IIC_VST1x3, "vst1", Dt, "$Vd, $Rn", "", []> {
1626 let Inst{4} = Rn{4};
1627 let DecoderMethod = "DecodeVSTInstruction";
1629 multiclass VST1D3WB<bits<4> op7_4, string Dt> {
1630 def _fixed : NLdSt<0,0b00,0b0110,op7_4, (outs GPR:$wb),
1631 (ins addrmode6:$Rn, VecListThreeD:$Vd), IIC_VLD1x3u,
1632 "vst1", Dt, "$Vd, $Rn!",
1633 "$Rn.addr = $wb", []> {
1634 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1635 let Inst{5-4} = Rn{5-4};
1636 let DecoderMethod = "DecodeVSTInstruction";
1637 let AsmMatchConverter = "cvtVSTwbFixed";
1639 def _register : NLdSt<0,0b00,0b0110,op7_4, (outs GPR:$wb),
1640 (ins addrmode6:$Rn, rGPR:$Rm, VecListThreeD:$Vd),
1642 "vst1", Dt, "$Vd, $Rn, $Rm",
1643 "$Rn.addr = $wb", []> {
1644 let Inst{5-4} = Rn{5-4};
1645 let DecoderMethod = "DecodeVSTInstruction";
1646 let AsmMatchConverter = "cvtVSTwbRegister";
1650 def VST1d8T : VST1D3<{0,0,0,?}, "8">;
1651 def VST1d16T : VST1D3<{0,1,0,?}, "16">;
1652 def VST1d32T : VST1D3<{1,0,0,?}, "32">;
1653 def VST1d64T : VST1D3<{1,1,0,?}, "64">;
1655 defm VST1d8Twb : VST1D3WB<{0,0,0,?}, "8">;
1656 defm VST1d16Twb : VST1D3WB<{0,1,0,?}, "16">;
1657 defm VST1d32Twb : VST1D3WB<{1,0,0,?}, "32">;
1658 defm VST1d64Twb : VST1D3WB<{1,1,0,?}, "64">;
1660 def VST1d64TPseudo : VSTQQPseudo<IIC_VST1x3>;
1661 def VST1d64TPseudoWB_fixed : VSTQQWBPseudo<IIC_VST1x3u>;
1662 def VST1d64TPseudoWB_register : VSTQQWBPseudo<IIC_VST1x3u>;
1664 // ...with 4 registers
1665 class VST1D4<bits<4> op7_4, string Dt>
1666 : NLdSt<0, 0b00, 0b0010, op7_4, (outs),
1667 (ins addrmode6:$Rn, VecListFourD:$Vd),
1668 IIC_VST1x4, "vst1", Dt, "$Vd, $Rn", "",
1671 let Inst{5-4} = Rn{5-4};
1672 let DecoderMethod = "DecodeVSTInstruction";
1674 multiclass VST1D4WB<bits<4> op7_4, string Dt> {
1675 def _fixed : NLdSt<0,0b00,0b0010,op7_4, (outs GPR:$wb),
1676 (ins addrmode6:$Rn, VecListFourD:$Vd), IIC_VLD1x4u,
1677 "vst1", Dt, "$Vd, $Rn!",
1678 "$Rn.addr = $wb", []> {
1679 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1680 let Inst{5-4} = Rn{5-4};
1681 let DecoderMethod = "DecodeVSTInstruction";
1682 let AsmMatchConverter = "cvtVSTwbFixed";
1684 def _register : NLdSt<0,0b00,0b0010,op7_4, (outs GPR:$wb),
1685 (ins addrmode6:$Rn, rGPR:$Rm, VecListFourD:$Vd),
1687 "vst1", Dt, "$Vd, $Rn, $Rm",
1688 "$Rn.addr = $wb", []> {
1689 let Inst{5-4} = Rn{5-4};
1690 let DecoderMethod = "DecodeVSTInstruction";
1691 let AsmMatchConverter = "cvtVSTwbRegister";
1695 def VST1d8Q : VST1D4<{0,0,?,?}, "8">;
1696 def VST1d16Q : VST1D4<{0,1,?,?}, "16">;
1697 def VST1d32Q : VST1D4<{1,0,?,?}, "32">;
1698 def VST1d64Q : VST1D4<{1,1,?,?}, "64">;
1700 defm VST1d8Qwb : VST1D4WB<{0,0,?,?}, "8">;
1701 defm VST1d16Qwb : VST1D4WB<{0,1,?,?}, "16">;
1702 defm VST1d32Qwb : VST1D4WB<{1,0,?,?}, "32">;
1703 defm VST1d64Qwb : VST1D4WB<{1,1,?,?}, "64">;
1705 def VST1d64QPseudo : VSTQQPseudo<IIC_VST1x4>;
1706 def VST1d64QPseudoWB_fixed : VSTQQWBPseudo<IIC_VST1x4u>;
1707 def VST1d64QPseudoWB_register : VSTQQWBPseudo<IIC_VST1x4u>;
1709 // VST2 : Vector Store (multiple 2-element structures)
1710 class VST2<bits<4> op11_8, bits<4> op7_4, string Dt, RegisterOperand VdTy,
1711 InstrItinClass itin>
1712 : NLdSt<0, 0b00, op11_8, op7_4, (outs), (ins addrmode6:$Rn, VdTy:$Vd),
1713 itin, "vst2", Dt, "$Vd, $Rn", "", []> {
1715 let Inst{5-4} = Rn{5-4};
1716 let DecoderMethod = "DecodeVSTInstruction";
1719 def VST2d8 : VST2<0b1000, {0,0,?,?}, "8", VecListDPair, IIC_VST2>;
1720 def VST2d16 : VST2<0b1000, {0,1,?,?}, "16", VecListDPair, IIC_VST2>;
1721 def VST2d32 : VST2<0b1000, {1,0,?,?}, "32", VecListDPair, IIC_VST2>;
1723 def VST2q8 : VST2<0b0011, {0,0,?,?}, "8", VecListFourD, IIC_VST2x2>;
1724 def VST2q16 : VST2<0b0011, {0,1,?,?}, "16", VecListFourD, IIC_VST2x2>;
1725 def VST2q32 : VST2<0b0011, {1,0,?,?}, "32", VecListFourD, IIC_VST2x2>;
1727 def VST2q8Pseudo : VSTQQPseudo<IIC_VST2x2>;
1728 def VST2q16Pseudo : VSTQQPseudo<IIC_VST2x2>;
1729 def VST2q32Pseudo : VSTQQPseudo<IIC_VST2x2>;
1731 // ...with address register writeback:
1732 multiclass VST2DWB<bits<4> op11_8, bits<4> op7_4, string Dt,
1733 RegisterOperand VdTy> {
1734 def _fixed : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1735 (ins addrmode6:$Rn, VdTy:$Vd), IIC_VLD1u,
1736 "vst2", Dt, "$Vd, $Rn!",
1737 "$Rn.addr = $wb", []> {
1738 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1739 let Inst{5-4} = Rn{5-4};
1740 let DecoderMethod = "DecodeVSTInstruction";
1741 let AsmMatchConverter = "cvtVSTwbFixed";
1743 def _register : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1744 (ins addrmode6:$Rn, rGPR:$Rm, VdTy:$Vd), IIC_VLD1u,
1745 "vst2", Dt, "$Vd, $Rn, $Rm",
1746 "$Rn.addr = $wb", []> {
1747 let Inst{5-4} = Rn{5-4};
1748 let DecoderMethod = "DecodeVSTInstruction";
1749 let AsmMatchConverter = "cvtVSTwbRegister";
1752 multiclass VST2QWB<bits<4> op7_4, string Dt> {
1753 def _fixed : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
1754 (ins addrmode6:$Rn, VecListFourD:$Vd), IIC_VLD1u,
1755 "vst2", Dt, "$Vd, $Rn!",
1756 "$Rn.addr = $wb", []> {
1757 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1758 let Inst{5-4} = Rn{5-4};
1759 let DecoderMethod = "DecodeVSTInstruction";
1760 let AsmMatchConverter = "cvtVSTwbFixed";
1762 def _register : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
1763 (ins addrmode6:$Rn, rGPR:$Rm, VecListFourD:$Vd),
1765 "vst2", Dt, "$Vd, $Rn, $Rm",
1766 "$Rn.addr = $wb", []> {
1767 let Inst{5-4} = Rn{5-4};
1768 let DecoderMethod = "DecodeVSTInstruction";
1769 let AsmMatchConverter = "cvtVSTwbRegister";
1773 defm VST2d8wb : VST2DWB<0b1000, {0,0,?,?}, "8", VecListDPair>;
1774 defm VST2d16wb : VST2DWB<0b1000, {0,1,?,?}, "16", VecListDPair>;
1775 defm VST2d32wb : VST2DWB<0b1000, {1,0,?,?}, "32", VecListDPair>;
1777 defm VST2q8wb : VST2QWB<{0,0,?,?}, "8">;
1778 defm VST2q16wb : VST2QWB<{0,1,?,?}, "16">;
1779 defm VST2q32wb : VST2QWB<{1,0,?,?}, "32">;
1781 def VST2q8PseudoWB_fixed : VSTQQWBfixedPseudo<IIC_VST2x2u>;
1782 def VST2q16PseudoWB_fixed : VSTQQWBfixedPseudo<IIC_VST2x2u>;
1783 def VST2q32PseudoWB_fixed : VSTQQWBfixedPseudo<IIC_VST2x2u>;
1784 def VST2q8PseudoWB_register : VSTQQWBregisterPseudo<IIC_VST2x2u>;
1785 def VST2q16PseudoWB_register : VSTQQWBregisterPseudo<IIC_VST2x2u>;
1786 def VST2q32PseudoWB_register : VSTQQWBregisterPseudo<IIC_VST2x2u>;
1788 // ...with double-spaced registers
1789 def VST2b8 : VST2<0b1001, {0,0,?,?}, "8", VecListDPairSpaced, IIC_VST2>;
1790 def VST2b16 : VST2<0b1001, {0,1,?,?}, "16", VecListDPairSpaced, IIC_VST2>;
1791 def VST2b32 : VST2<0b1001, {1,0,?,?}, "32", VecListDPairSpaced, IIC_VST2>;
1792 defm VST2b8wb : VST2DWB<0b1001, {0,0,?,?}, "8", VecListDPairSpaced>;
1793 defm VST2b16wb : VST2DWB<0b1001, {0,1,?,?}, "16", VecListDPairSpaced>;
1794 defm VST2b32wb : VST2DWB<0b1001, {1,0,?,?}, "32", VecListDPairSpaced>;
1796 // VST3 : Vector Store (multiple 3-element structures)
1797 class VST3D<bits<4> op11_8, bits<4> op7_4, string Dt>
1798 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1799 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3,
1800 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn", "", []> {
1802 let Inst{4} = Rn{4};
1803 let DecoderMethod = "DecodeVSTInstruction";
1806 def VST3d8 : VST3D<0b0100, {0,0,0,?}, "8">;
1807 def VST3d16 : VST3D<0b0100, {0,1,0,?}, "16">;
1808 def VST3d32 : VST3D<0b0100, {1,0,0,?}, "32">;
1810 def VST3d8Pseudo : VSTQQPseudo<IIC_VST3>;
1811 def VST3d16Pseudo : VSTQQPseudo<IIC_VST3>;
1812 def VST3d32Pseudo : VSTQQPseudo<IIC_VST3>;
1814 // ...with address register writeback:
1815 class VST3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1816 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1817 (ins addrmode6:$Rn, am6offset:$Rm,
1818 DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3u,
1819 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn$Rm",
1820 "$Rn.addr = $wb", []> {
1821 let Inst{4} = Rn{4};
1822 let DecoderMethod = "DecodeVSTInstruction";
1825 def VST3d8_UPD : VST3DWB<0b0100, {0,0,0,?}, "8">;
1826 def VST3d16_UPD : VST3DWB<0b0100, {0,1,0,?}, "16">;
1827 def VST3d32_UPD : VST3DWB<0b0100, {1,0,0,?}, "32">;
1829 def VST3d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1830 def VST3d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1831 def VST3d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1833 // ...with double-spaced registers:
1834 def VST3q8 : VST3D<0b0101, {0,0,0,?}, "8">;
1835 def VST3q16 : VST3D<0b0101, {0,1,0,?}, "16">;
1836 def VST3q32 : VST3D<0b0101, {1,0,0,?}, "32">;
1837 def VST3q8_UPD : VST3DWB<0b0101, {0,0,0,?}, "8">;
1838 def VST3q16_UPD : VST3DWB<0b0101, {0,1,0,?}, "16">;
1839 def VST3q32_UPD : VST3DWB<0b0101, {1,0,0,?}, "32">;
1841 def VST3q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1842 def VST3q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1843 def VST3q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1845 // ...alternate versions to be allocated odd register numbers:
1846 def VST3q8oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1847 def VST3q16oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1848 def VST3q32oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1850 def VST3q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1851 def VST3q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1852 def VST3q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1854 // VST4 : Vector Store (multiple 4-element structures)
1855 class VST4D<bits<4> op11_8, bits<4> op7_4, string Dt>
1856 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1857 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
1858 IIC_VST4, "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn",
1861 let Inst{5-4} = Rn{5-4};
1862 let DecoderMethod = "DecodeVSTInstruction";
1865 def VST4d8 : VST4D<0b0000, {0,0,?,?}, "8">;
1866 def VST4d16 : VST4D<0b0000, {0,1,?,?}, "16">;
1867 def VST4d32 : VST4D<0b0000, {1,0,?,?}, "32">;
1869 def VST4d8Pseudo : VSTQQPseudo<IIC_VST4>;
1870 def VST4d16Pseudo : VSTQQPseudo<IIC_VST4>;
1871 def VST4d32Pseudo : VSTQQPseudo<IIC_VST4>;
1873 // ...with address register writeback:
1874 class VST4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1875 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1876 (ins addrmode6:$Rn, am6offset:$Rm,
1877 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST4u,
1878 "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
1879 "$Rn.addr = $wb", []> {
1880 let Inst{5-4} = Rn{5-4};
1881 let DecoderMethod = "DecodeVSTInstruction";
1884 def VST4d8_UPD : VST4DWB<0b0000, {0,0,?,?}, "8">;
1885 def VST4d16_UPD : VST4DWB<0b0000, {0,1,?,?}, "16">;
1886 def VST4d32_UPD : VST4DWB<0b0000, {1,0,?,?}, "32">;
1888 def VST4d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1889 def VST4d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1890 def VST4d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1892 // ...with double-spaced registers:
1893 def VST4q8 : VST4D<0b0001, {0,0,?,?}, "8">;
1894 def VST4q16 : VST4D<0b0001, {0,1,?,?}, "16">;
1895 def VST4q32 : VST4D<0b0001, {1,0,?,?}, "32">;
1896 def VST4q8_UPD : VST4DWB<0b0001, {0,0,?,?}, "8">;
1897 def VST4q16_UPD : VST4DWB<0b0001, {0,1,?,?}, "16">;
1898 def VST4q32_UPD : VST4DWB<0b0001, {1,0,?,?}, "32">;
1900 def VST4q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1901 def VST4q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1902 def VST4q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1904 // ...alternate versions to be allocated odd register numbers:
1905 def VST4q8oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1906 def VST4q16oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1907 def VST4q32oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1909 def VST4q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1910 def VST4q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1911 def VST4q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1913 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
1915 // Classes for VST*LN pseudo-instructions with multi-register operands.
1916 // These are expanded to real instructions after register allocation.
1917 class VSTQLNPseudo<InstrItinClass itin>
1918 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
1920 class VSTQLNWBPseudo<InstrItinClass itin>
1921 : PseudoNLdSt<(outs GPR:$wb),
1922 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
1923 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1924 class VSTQQLNPseudo<InstrItinClass itin>
1925 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
1927 class VSTQQLNWBPseudo<InstrItinClass itin>
1928 : PseudoNLdSt<(outs GPR:$wb),
1929 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
1930 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1931 class VSTQQQQLNPseudo<InstrItinClass itin>
1932 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
1934 class VSTQQQQLNWBPseudo<InstrItinClass itin>
1935 : PseudoNLdSt<(outs GPR:$wb),
1936 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
1937 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1939 // VST1LN : Vector Store (single element from one lane)
1940 class VST1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1941 PatFrag StoreOp, SDNode ExtractOp, Operand AddrMode>
1942 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1943 (ins AddrMode:$Rn, DPR:$Vd, nohash_imm:$lane),
1944 IIC_VST1ln, "vst1", Dt, "\\{$Vd[$lane]\\}, $Rn", "",
1945 [(StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane), AddrMode:$Rn)]> {
1947 let DecoderMethod = "DecodeVST1LN";
1949 class VST1QLNPseudo<ValueType Ty, PatFrag StoreOp, SDNode ExtractOp>
1950 : VSTQLNPseudo<IIC_VST1ln> {
1951 let Pattern = [(StoreOp (ExtractOp (Ty QPR:$src), imm:$lane),
1955 def VST1LNd8 : VST1LN<0b0000, {?,?,?,0}, "8", v8i8, truncstorei8,
1956 NEONvgetlaneu, addrmode6> {
1957 let Inst{7-5} = lane{2-0};
1959 def VST1LNd16 : VST1LN<0b0100, {?,?,0,?}, "16", v4i16, truncstorei16,
1960 NEONvgetlaneu, addrmode6> {
1961 let Inst{7-6} = lane{1-0};
1962 let Inst{4} = Rn{5};
1965 def VST1LNd32 : VST1LN<0b1000, {?,0,?,?}, "32", v2i32, store, extractelt,
1967 let Inst{7} = lane{0};
1968 let Inst{5-4} = Rn{5-4};
1971 def VST1LNq8Pseudo : VST1QLNPseudo<v16i8, truncstorei8, NEONvgetlaneu>;
1972 def VST1LNq16Pseudo : VST1QLNPseudo<v8i16, truncstorei16, NEONvgetlaneu>;
1973 def VST1LNq32Pseudo : VST1QLNPseudo<v4i32, store, extractelt>;
1975 def : Pat<(store (extractelt (v2f32 DPR:$src), imm:$lane), addrmode6:$addr),
1976 (VST1LNd32 addrmode6:$addr, DPR:$src, imm:$lane)>;
1977 def : Pat<(store (extractelt (v4f32 QPR:$src), imm:$lane), addrmode6:$addr),
1978 (VST1LNq32Pseudo addrmode6:$addr, QPR:$src, imm:$lane)>;
1980 // ...with address register writeback:
1981 class VST1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1982 PatFrag StoreOp, SDNode ExtractOp, Operand AdrMode>
1983 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1984 (ins AdrMode:$Rn, am6offset:$Rm,
1985 DPR:$Vd, nohash_imm:$lane), IIC_VST1lnu, "vst1", Dt,
1986 "\\{$Vd[$lane]\\}, $Rn$Rm",
1988 [(set GPR:$wb, (StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane),
1989 AdrMode:$Rn, am6offset:$Rm))]> {
1990 let DecoderMethod = "DecodeVST1LN";
1992 class VST1QLNWBPseudo<ValueType Ty, PatFrag StoreOp, SDNode ExtractOp>
1993 : VSTQLNWBPseudo<IIC_VST1lnu> {
1994 let Pattern = [(set GPR:$wb, (StoreOp (ExtractOp (Ty QPR:$src), imm:$lane),
1995 addrmode6:$addr, am6offset:$offset))];
1998 def VST1LNd8_UPD : VST1LNWB<0b0000, {?,?,?,0}, "8", v8i8, post_truncsti8,
1999 NEONvgetlaneu, addrmode6> {
2000 let Inst{7-5} = lane{2-0};
2002 def VST1LNd16_UPD : VST1LNWB<0b0100, {?,?,0,?}, "16", v4i16, post_truncsti16,
2003 NEONvgetlaneu, addrmode6> {
2004 let Inst{7-6} = lane{1-0};
2005 let Inst{4} = Rn{5};
2007 def VST1LNd32_UPD : VST1LNWB<0b1000, {?,0,?,?}, "32", v2i32, post_store,
2008 extractelt, addrmode6oneL32> {
2009 let Inst{7} = lane{0};
2010 let Inst{5-4} = Rn{5-4};
2013 def VST1LNq8Pseudo_UPD : VST1QLNWBPseudo<v16i8, post_truncsti8, NEONvgetlaneu>;
2014 def VST1LNq16Pseudo_UPD : VST1QLNWBPseudo<v8i16, post_truncsti16,NEONvgetlaneu>;
2015 def VST1LNq32Pseudo_UPD : VST1QLNWBPseudo<v4i32, post_store, extractelt>;
2017 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
2019 // VST2LN : Vector Store (single 2-element structure from one lane)
2020 class VST2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
2021 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
2022 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, nohash_imm:$lane),
2023 IIC_VST2ln, "vst2", Dt, "\\{$Vd[$lane], $src2[$lane]\\}, $Rn",
2026 let Inst{4} = Rn{4};
2027 let DecoderMethod = "DecodeVST2LN";
2030 def VST2LNd8 : VST2LN<0b0001, {?,?,?,?}, "8"> {
2031 let Inst{7-5} = lane{2-0};
2033 def VST2LNd16 : VST2LN<0b0101, {?,?,0,?}, "16"> {
2034 let Inst{7-6} = lane{1-0};
2036 def VST2LNd32 : VST2LN<0b1001, {?,0,0,?}, "32"> {
2037 let Inst{7} = lane{0};
2040 def VST2LNd8Pseudo : VSTQLNPseudo<IIC_VST2ln>;
2041 def VST2LNd16Pseudo : VSTQLNPseudo<IIC_VST2ln>;
2042 def VST2LNd32Pseudo : VSTQLNPseudo<IIC_VST2ln>;
2044 // ...with double-spaced registers:
2045 def VST2LNq16 : VST2LN<0b0101, {?,?,1,?}, "16"> {
2046 let Inst{7-6} = lane{1-0};
2047 let Inst{4} = Rn{4};
2049 def VST2LNq32 : VST2LN<0b1001, {?,1,0,?}, "32"> {
2050 let Inst{7} = lane{0};
2051 let Inst{4} = Rn{4};
2054 def VST2LNq16Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
2055 def VST2LNq32Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
2057 // ...with address register writeback:
2058 class VST2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
2059 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
2060 (ins addrmode6:$Rn, am6offset:$Rm,
2061 DPR:$Vd, DPR:$src2, nohash_imm:$lane), IIC_VST2lnu, "vst2", Dt,
2062 "\\{$Vd[$lane], $src2[$lane]\\}, $Rn$Rm",
2063 "$Rn.addr = $wb", []> {
2064 let Inst{4} = Rn{4};
2065 let DecoderMethod = "DecodeVST2LN";
2068 def VST2LNd8_UPD : VST2LNWB<0b0001, {?,?,?,?}, "8"> {
2069 let Inst{7-5} = lane{2-0};
2071 def VST2LNd16_UPD : VST2LNWB<0b0101, {?,?,0,?}, "16"> {
2072 let Inst{7-6} = lane{1-0};
2074 def VST2LNd32_UPD : VST2LNWB<0b1001, {?,0,0,?}, "32"> {
2075 let Inst{7} = lane{0};
2078 def VST2LNd8Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
2079 def VST2LNd16Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
2080 def VST2LNd32Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
2082 def VST2LNq16_UPD : VST2LNWB<0b0101, {?,?,1,?}, "16"> {
2083 let Inst{7-6} = lane{1-0};
2085 def VST2LNq32_UPD : VST2LNWB<0b1001, {?,1,0,?}, "32"> {
2086 let Inst{7} = lane{0};
2089 def VST2LNq16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
2090 def VST2LNq32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
2092 // VST3LN : Vector Store (single 3-element structure from one lane)
2093 class VST3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
2094 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
2095 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3,
2096 nohash_imm:$lane), IIC_VST3ln, "vst3", Dt,
2097 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn", "", []> {
2099 let DecoderMethod = "DecodeVST3LN";
2102 def VST3LNd8 : VST3LN<0b0010, {?,?,?,0}, "8"> {
2103 let Inst{7-5} = lane{2-0};
2105 def VST3LNd16 : VST3LN<0b0110, {?,?,0,0}, "16"> {
2106 let Inst{7-6} = lane{1-0};
2108 def VST3LNd32 : VST3LN<0b1010, {?,0,0,0}, "32"> {
2109 let Inst{7} = lane{0};
2112 def VST3LNd8Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
2113 def VST3LNd16Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
2114 def VST3LNd32Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
2116 // ...with double-spaced registers:
2117 def VST3LNq16 : VST3LN<0b0110, {?,?,1,0}, "16"> {
2118 let Inst{7-6} = lane{1-0};
2120 def VST3LNq32 : VST3LN<0b1010, {?,1,0,0}, "32"> {
2121 let Inst{7} = lane{0};
2124 def VST3LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
2125 def VST3LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
2127 // ...with address register writeback:
2128 class VST3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
2129 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
2130 (ins addrmode6:$Rn, am6offset:$Rm,
2131 DPR:$Vd, DPR:$src2, DPR:$src3, nohash_imm:$lane),
2132 IIC_VST3lnu, "vst3", Dt,
2133 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn$Rm",
2134 "$Rn.addr = $wb", []> {
2135 let DecoderMethod = "DecodeVST3LN";
2138 def VST3LNd8_UPD : VST3LNWB<0b0010, {?,?,?,0}, "8"> {
2139 let Inst{7-5} = lane{2-0};
2141 def VST3LNd16_UPD : VST3LNWB<0b0110, {?,?,0,0}, "16"> {
2142 let Inst{7-6} = lane{1-0};
2144 def VST3LNd32_UPD : VST3LNWB<0b1010, {?,0,0,0}, "32"> {
2145 let Inst{7} = lane{0};
2148 def VST3LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
2149 def VST3LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
2150 def VST3LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
2152 def VST3LNq16_UPD : VST3LNWB<0b0110, {?,?,1,0}, "16"> {
2153 let Inst{7-6} = lane{1-0};
2155 def VST3LNq32_UPD : VST3LNWB<0b1010, {?,1,0,0}, "32"> {
2156 let Inst{7} = lane{0};
2159 def VST3LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
2160 def VST3LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
2162 // VST4LN : Vector Store (single 4-element structure from one lane)
2163 class VST4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
2164 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
2165 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4,
2166 nohash_imm:$lane), IIC_VST4ln, "vst4", Dt,
2167 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn",
2170 let Inst{4} = Rn{4};
2171 let DecoderMethod = "DecodeVST4LN";
2174 def VST4LNd8 : VST4LN<0b0011, {?,?,?,?}, "8"> {
2175 let Inst{7-5} = lane{2-0};
2177 def VST4LNd16 : VST4LN<0b0111, {?,?,0,?}, "16"> {
2178 let Inst{7-6} = lane{1-0};
2180 def VST4LNd32 : VST4LN<0b1011, {?,0,?,?}, "32"> {
2181 let Inst{7} = lane{0};
2182 let Inst{5} = Rn{5};
2185 def VST4LNd8Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
2186 def VST4LNd16Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
2187 def VST4LNd32Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
2189 // ...with double-spaced registers:
2190 def VST4LNq16 : VST4LN<0b0111, {?,?,1,?}, "16"> {
2191 let Inst{7-6} = lane{1-0};
2193 def VST4LNq32 : VST4LN<0b1011, {?,1,?,?}, "32"> {
2194 let Inst{7} = lane{0};
2195 let Inst{5} = Rn{5};
2198 def VST4LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
2199 def VST4LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
2201 // ...with address register writeback:
2202 class VST4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
2203 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
2204 (ins addrmode6:$Rn, am6offset:$Rm,
2205 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
2206 IIC_VST4lnu, "vst4", Dt,
2207 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn$Rm",
2208 "$Rn.addr = $wb", []> {
2209 let Inst{4} = Rn{4};
2210 let DecoderMethod = "DecodeVST4LN";
2213 def VST4LNd8_UPD : VST4LNWB<0b0011, {?,?,?,?}, "8"> {
2214 let Inst{7-5} = lane{2-0};
2216 def VST4LNd16_UPD : VST4LNWB<0b0111, {?,?,0,?}, "16"> {
2217 let Inst{7-6} = lane{1-0};
2219 def VST4LNd32_UPD : VST4LNWB<0b1011, {?,0,?,?}, "32"> {
2220 let Inst{7} = lane{0};
2221 let Inst{5} = Rn{5};
2224 def VST4LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
2225 def VST4LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
2226 def VST4LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
2228 def VST4LNq16_UPD : VST4LNWB<0b0111, {?,?,1,?}, "16"> {
2229 let Inst{7-6} = lane{1-0};
2231 def VST4LNq32_UPD : VST4LNWB<0b1011, {?,1,?,?}, "32"> {
2232 let Inst{7} = lane{0};
2233 let Inst{5} = Rn{5};
2236 def VST4LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
2237 def VST4LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
2239 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
2242 //===----------------------------------------------------------------------===//
2243 // NEON pattern fragments
2244 //===----------------------------------------------------------------------===//
2246 // Extract D sub-registers of Q registers.
2247 def DSubReg_i8_reg : SDNodeXForm<imm, [{
2248 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
2249 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/8, MVT::i32);
2251 def DSubReg_i16_reg : SDNodeXForm<imm, [{
2252 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
2253 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/4, MVT::i32);
2255 def DSubReg_i32_reg : SDNodeXForm<imm, [{
2256 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
2257 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/2, MVT::i32);
2259 def DSubReg_f64_reg : SDNodeXForm<imm, [{
2260 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
2261 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue(), MVT::i32);
2264 // Extract S sub-registers of Q/D registers.
2265 def SSubReg_f32_reg : SDNodeXForm<imm, [{
2266 assert(ARM::ssub_3 == ARM::ssub_0+3 && "Unexpected subreg numbering");
2267 return CurDAG->getTargetConstant(ARM::ssub_0 + N->getZExtValue(), MVT::i32);
2270 // Translate lane numbers from Q registers to D subregs.
2271 def SubReg_i8_lane : SDNodeXForm<imm, [{
2272 return CurDAG->getTargetConstant(N->getZExtValue() & 7, MVT::i32);
2274 def SubReg_i16_lane : SDNodeXForm<imm, [{
2275 return CurDAG->getTargetConstant(N->getZExtValue() & 3, MVT::i32);
2277 def SubReg_i32_lane : SDNodeXForm<imm, [{
2278 return CurDAG->getTargetConstant(N->getZExtValue() & 1, MVT::i32);
2281 //===----------------------------------------------------------------------===//
2282 // Instruction Classes
2283 //===----------------------------------------------------------------------===//
2285 // Basic 2-register operations: double- and quad-register.
2286 class N2VD<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2287 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
2288 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
2289 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
2290 (ins DPR:$Vm), IIC_VUNAD, OpcodeStr, Dt,"$Vd, $Vm", "",
2291 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vm))))]>;
2292 class N2VQ<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2293 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
2294 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
2295 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
2296 (ins QPR:$Vm), IIC_VUNAQ, OpcodeStr, Dt,"$Vd, $Vm", "",
2297 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vm))))]>;
2299 // Basic 2-register intrinsics, both double- and quad-register.
2300 class N2VDInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2301 bits<2> op17_16, bits<5> op11_7, bit op4,
2302 InstrItinClass itin, string OpcodeStr, string Dt,
2303 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2304 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
2305 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2306 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm))))]>;
2307 class N2VQInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2308 bits<2> op17_16, bits<5> op11_7, bit op4,
2309 InstrItinClass itin, string OpcodeStr, string Dt,
2310 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2311 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
2312 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2313 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm))))]>;
2315 // Narrow 2-register operations.
2316 class N2VN<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2317 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2318 InstrItinClass itin, string OpcodeStr, string Dt,
2319 ValueType TyD, ValueType TyQ, SDNode OpNode>
2320 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$Vd),
2321 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2322 [(set DPR:$Vd, (TyD (OpNode (TyQ QPR:$Vm))))]>;
2324 // Narrow 2-register intrinsics.
2325 class N2VNInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2326 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2327 InstrItinClass itin, string OpcodeStr, string Dt,
2328 ValueType TyD, ValueType TyQ, Intrinsic IntOp>
2329 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$Vd),
2330 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2331 [(set DPR:$Vd, (TyD (IntOp (TyQ QPR:$Vm))))]>;
2333 // Long 2-register operations (currently only used for VMOVL).
2334 class N2VL<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2335 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2336 InstrItinClass itin, string OpcodeStr, string Dt,
2337 ValueType TyQ, ValueType TyD, SDNode OpNode>
2338 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$Vd),
2339 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2340 [(set QPR:$Vd, (TyQ (OpNode (TyD DPR:$Vm))))]>;
2342 // Long 2-register intrinsics.
2343 class N2VLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2344 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2345 InstrItinClass itin, string OpcodeStr, string Dt,
2346 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
2347 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$Vd),
2348 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2349 [(set QPR:$Vd, (TyQ (IntOp (TyD DPR:$Vm))))]>;
2351 // 2-register shuffles (VTRN/VZIP/VUZP), both double- and quad-register.
2352 class N2VDShuffle<bits<2> op19_18, bits<5> op11_7, string OpcodeStr, string Dt>
2353 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 0, 0, (outs DPR:$Vd, DPR:$Vm),
2354 (ins DPR:$src1, DPR:$src2), IIC_VPERMD,
2355 OpcodeStr, Dt, "$Vd, $Vm",
2356 "$src1 = $Vd, $src2 = $Vm", []>;
2357 class N2VQShuffle<bits<2> op19_18, bits<5> op11_7,
2358 InstrItinClass itin, string OpcodeStr, string Dt>
2359 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 1, 0, (outs QPR:$Vd, QPR:$Vm),
2360 (ins QPR:$src1, QPR:$src2), itin, OpcodeStr, Dt, "$Vd, $Vm",
2361 "$src1 = $Vd, $src2 = $Vm", []>;
2363 // Basic 3-register operations: double- and quad-register.
2364 class N3VD<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2365 InstrItinClass itin, string OpcodeStr, string Dt,
2366 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
2367 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2368 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2369 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2370 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
2371 // All of these have a two-operand InstAlias.
2372 let TwoOperandAliasConstraint = "$Vn = $Vd";
2373 let isCommutable = Commutable;
2375 // Same as N3VD but no data type.
2376 class N3VDX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2377 InstrItinClass itin, string OpcodeStr,
2378 ValueType ResTy, ValueType OpTy,
2379 SDNode OpNode, bit Commutable>
2380 : N3VX<op24, op23, op21_20, op11_8, 0, op4,
2381 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2382 OpcodeStr, "$Vd, $Vn, $Vm", "",
2383 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>{
2384 // All of these have a two-operand InstAlias.
2385 let TwoOperandAliasConstraint = "$Vn = $Vd";
2386 let isCommutable = Commutable;
2389 class N3VDSL<bits<2> op21_20, bits<4> op11_8,
2390 InstrItinClass itin, string OpcodeStr, string Dt,
2391 ValueType Ty, SDNode ShOp>
2392 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2393 (outs DPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2394 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2396 (Ty (ShOp (Ty DPR:$Vn),
2397 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),imm:$lane)))))]> {
2398 // All of these have a two-operand InstAlias.
2399 let TwoOperandAliasConstraint = "$Vn = $Vd";
2400 let isCommutable = 0;
2402 class N3VDSL16<bits<2> op21_20, bits<4> op11_8,
2403 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
2404 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2405 (outs DPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2406 NVMulSLFrm, IIC_VMULi16D, OpcodeStr, Dt,"$Vd, $Vn, $Vm$lane","",
2408 (Ty (ShOp (Ty DPR:$Vn),
2409 (Ty (NEONvduplane (Ty DPR_8:$Vm), imm:$lane)))))]> {
2410 // All of these have a two-operand InstAlias.
2411 let TwoOperandAliasConstraint = "$Vn = $Vd";
2412 let isCommutable = 0;
2415 class N3VQ<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2416 InstrItinClass itin, string OpcodeStr, string Dt,
2417 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
2418 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2419 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2420 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2421 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
2422 // All of these have a two-operand InstAlias.
2423 let TwoOperandAliasConstraint = "$Vn = $Vd";
2424 let isCommutable = Commutable;
2426 class N3VQX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2427 InstrItinClass itin, string OpcodeStr,
2428 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
2429 : N3VX<op24, op23, op21_20, op11_8, 1, op4,
2430 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2431 OpcodeStr, "$Vd, $Vn, $Vm", "",
2432 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]>{
2433 // All of these have a two-operand InstAlias.
2434 let TwoOperandAliasConstraint = "$Vn = $Vd";
2435 let isCommutable = Commutable;
2437 class N3VQSL<bits<2> op21_20, bits<4> op11_8,
2438 InstrItinClass itin, string OpcodeStr, string Dt,
2439 ValueType ResTy, ValueType OpTy, SDNode ShOp>
2440 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2441 (outs QPR:$Vd), (ins QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2442 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2443 [(set (ResTy QPR:$Vd),
2444 (ResTy (ShOp (ResTy QPR:$Vn),
2445 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2447 // All of these have a two-operand InstAlias.
2448 let TwoOperandAliasConstraint = "$Vn = $Vd";
2449 let isCommutable = 0;
2451 class N3VQSL16<bits<2> op21_20, bits<4> op11_8, string OpcodeStr, string Dt,
2452 ValueType ResTy, ValueType OpTy, SDNode ShOp>
2453 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2454 (outs QPR:$Vd), (ins QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2455 NVMulSLFrm, IIC_VMULi16Q, OpcodeStr, Dt,"$Vd, $Vn, $Vm$lane", "",
2456 [(set (ResTy QPR:$Vd),
2457 (ResTy (ShOp (ResTy QPR:$Vn),
2458 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2460 // All of these have a two-operand InstAlias.
2461 let TwoOperandAliasConstraint = "$Vn = $Vd";
2462 let isCommutable = 0;
2465 // Basic 3-register intrinsics, both double- and quad-register.
2466 class N3VDInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2467 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2468 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
2469 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2470 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), f, itin,
2471 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2472 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
2473 // All of these have a two-operand InstAlias.
2474 let TwoOperandAliasConstraint = "$Vn = $Vd";
2475 let isCommutable = Commutable;
2477 class N3VDIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2478 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
2479 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2480 (outs DPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2481 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2483 (Ty (IntOp (Ty DPR:$Vn),
2484 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),
2486 let isCommutable = 0;
2488 class N3VDIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2489 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
2490 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2491 (outs DPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2492 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2494 (Ty (IntOp (Ty DPR:$Vn),
2495 (Ty (NEONvduplane (Ty DPR_8:$Vm), imm:$lane)))))]> {
2496 let isCommutable = 0;
2498 class N3VDIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2499 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2500 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2501 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2502 (outs DPR:$Vd), (ins DPR:$Vm, DPR:$Vn), f, itin,
2503 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
2504 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (OpTy DPR:$Vn))))]> {
2505 let TwoOperandAliasConstraint = "$Vm = $Vd";
2506 let isCommutable = 0;
2509 class N3VQInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2510 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2511 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
2512 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2513 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), f, itin,
2514 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2515 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
2516 // All of these have a two-operand InstAlias.
2517 let TwoOperandAliasConstraint = "$Vn = $Vd";
2518 let isCommutable = Commutable;
2520 class N3VQIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2521 string OpcodeStr, string Dt,
2522 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2523 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2524 (outs QPR:$Vd), (ins QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2525 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2526 [(set (ResTy QPR:$Vd),
2527 (ResTy (IntOp (ResTy QPR:$Vn),
2528 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2530 let isCommutable = 0;
2532 class N3VQIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2533 string OpcodeStr, string Dt,
2534 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2535 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2536 (outs QPR:$Vd), (ins QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2537 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2538 [(set (ResTy QPR:$Vd),
2539 (ResTy (IntOp (ResTy QPR:$Vn),
2540 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2542 let isCommutable = 0;
2544 class N3VQIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2545 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2546 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2547 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2548 (outs QPR:$Vd), (ins QPR:$Vm, QPR:$Vn), f, itin,
2549 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
2550 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (OpTy QPR:$Vn))))]> {
2551 let TwoOperandAliasConstraint = "$Vm = $Vd";
2552 let isCommutable = 0;
2555 // Multiply-Add/Sub operations: double- and quad-register.
2556 class N3VDMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2557 InstrItinClass itin, string OpcodeStr, string Dt,
2558 ValueType Ty, SDPatternOperator MulOp, SDPatternOperator OpNode>
2559 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2560 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2561 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2562 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
2563 (Ty (MulOp DPR:$Vn, DPR:$Vm)))))]>;
2565 class N3VDMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2566 string OpcodeStr, string Dt,
2567 ValueType Ty, SDPatternOperator MulOp, SDPatternOperator ShOp>
2568 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2570 (ins DPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2572 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2574 (Ty (ShOp (Ty DPR:$src1),
2576 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),
2578 class N3VDMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2579 string OpcodeStr, string Dt,
2580 ValueType Ty, SDNode MulOp, SDNode ShOp>
2581 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2583 (ins DPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2585 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2587 (Ty (ShOp (Ty DPR:$src1),
2589 (Ty (NEONvduplane (Ty DPR_8:$Vm),
2592 class N3VQMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2593 InstrItinClass itin, string OpcodeStr, string Dt, ValueType Ty,
2594 SDPatternOperator MulOp, SDPatternOperator OpNode>
2595 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2596 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2597 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2598 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
2599 (Ty (MulOp QPR:$Vn, QPR:$Vm)))))]>;
2600 class N3VQMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2601 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2602 SDPatternOperator MulOp, SDPatternOperator ShOp>
2603 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2605 (ins QPR:$src1, QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2607 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2608 [(set (ResTy QPR:$Vd),
2609 (ResTy (ShOp (ResTy QPR:$src1),
2610 (ResTy (MulOp QPR:$Vn,
2611 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2613 class N3VQMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2614 string OpcodeStr, string Dt,
2615 ValueType ResTy, ValueType OpTy,
2616 SDNode MulOp, SDNode ShOp>
2617 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2619 (ins QPR:$src1, QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2621 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2622 [(set (ResTy QPR:$Vd),
2623 (ResTy (ShOp (ResTy QPR:$src1),
2624 (ResTy (MulOp QPR:$Vn,
2625 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2628 // Neon Intrinsic-Op instructions (VABA): double- and quad-register.
2629 class N3VDIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2630 InstrItinClass itin, string OpcodeStr, string Dt,
2631 ValueType Ty, Intrinsic IntOp, SDNode OpNode>
2632 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2633 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2634 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2635 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
2636 (Ty (IntOp (Ty DPR:$Vn), (Ty DPR:$Vm))))))]>;
2637 class N3VQIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2638 InstrItinClass itin, string OpcodeStr, string Dt,
2639 ValueType Ty, Intrinsic IntOp, SDNode OpNode>
2640 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2641 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2642 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2643 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
2644 (Ty (IntOp (Ty QPR:$Vn), (Ty QPR:$Vm))))))]>;
2646 // Neon 3-argument intrinsics, both double- and quad-register.
2647 // The destination register is also used as the first source operand register.
2648 class N3VDInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2649 InstrItinClass itin, string OpcodeStr, string Dt,
2650 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2651 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2652 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2653 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2654 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$src1),
2655 (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>;
2656 class N3VQInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2657 InstrItinClass itin, string OpcodeStr, string Dt,
2658 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2659 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2660 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2661 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2662 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$src1),
2663 (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]>;
2665 // Long Multiply-Add/Sub operations.
2666 class N3VLMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2667 InstrItinClass itin, string OpcodeStr, string Dt,
2668 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2669 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2670 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2671 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2672 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
2673 (TyQ (MulOp (TyD DPR:$Vn),
2674 (TyD DPR:$Vm)))))]>;
2675 class N3VLMulOpSL<bit op24, bits<2> op21_20, bits<4> op11_8,
2676 InstrItinClass itin, string OpcodeStr, string Dt,
2677 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2678 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$Vd),
2679 (ins QPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2681 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2683 (OpNode (TyQ QPR:$src1),
2684 (TyQ (MulOp (TyD DPR:$Vn),
2685 (TyD (NEONvduplane (TyD DPR_VFP2:$Vm),
2687 class N3VLMulOpSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2688 InstrItinClass itin, string OpcodeStr, string Dt,
2689 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2690 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$Vd),
2691 (ins QPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2693 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2695 (OpNode (TyQ QPR:$src1),
2696 (TyQ (MulOp (TyD DPR:$Vn),
2697 (TyD (NEONvduplane (TyD DPR_8:$Vm),
2700 // Long Intrinsic-Op vector operations with explicit extend (VABAL).
2701 class N3VLIntExtOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2702 InstrItinClass itin, string OpcodeStr, string Dt,
2703 ValueType TyQ, ValueType TyD, Intrinsic IntOp, SDNode ExtOp,
2705 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2706 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2707 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2708 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
2709 (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
2710 (TyD DPR:$Vm)))))))]>;
2712 // Neon Long 3-argument intrinsic. The destination register is
2713 // a quad-register and is also used as the first source operand register.
2714 class N3VLInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2715 InstrItinClass itin, string OpcodeStr, string Dt,
2716 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
2717 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2718 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2719 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2721 (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$Vn), (TyD DPR:$Vm))))]>;
2722 class N3VLInt3SL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2723 string OpcodeStr, string Dt,
2724 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2725 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2727 (ins QPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2729 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2730 [(set (ResTy QPR:$Vd),
2731 (ResTy (IntOp (ResTy QPR:$src1),
2733 (OpTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2735 class N3VLInt3SL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2736 InstrItinClass itin, string OpcodeStr, string Dt,
2737 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2738 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2740 (ins QPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2742 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2743 [(set (ResTy QPR:$Vd),
2744 (ResTy (IntOp (ResTy QPR:$src1),
2746 (OpTy (NEONvduplane (OpTy DPR_8:$Vm),
2749 // Narrowing 3-register intrinsics.
2750 class N3VNInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2751 string OpcodeStr, string Dt, ValueType TyD, ValueType TyQ,
2752 Intrinsic IntOp, bit Commutable>
2753 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2754 (outs DPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINi4D,
2755 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2756 [(set DPR:$Vd, (TyD (IntOp (TyQ QPR:$Vn), (TyQ QPR:$Vm))))]> {
2757 let isCommutable = Commutable;
2760 // Long 3-register operations.
2761 class N3VL<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2762 InstrItinClass itin, string OpcodeStr, string Dt,
2763 ValueType TyQ, ValueType TyD, SDNode OpNode, bit Commutable>
2764 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2765 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2766 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2767 [(set QPR:$Vd, (TyQ (OpNode (TyD DPR:$Vn), (TyD DPR:$Vm))))]> {
2768 let isCommutable = Commutable;
2770 class N3VLSL<bit op24, bits<2> op21_20, bits<4> op11_8,
2771 InstrItinClass itin, string OpcodeStr, string Dt,
2772 ValueType TyQ, ValueType TyD, SDNode OpNode>
2773 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2774 (outs QPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2775 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2777 (TyQ (OpNode (TyD DPR:$Vn),
2778 (TyD (NEONvduplane (TyD DPR_VFP2:$Vm),imm:$lane)))))]>;
2779 class N3VLSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2780 InstrItinClass itin, string OpcodeStr, string Dt,
2781 ValueType TyQ, ValueType TyD, SDNode OpNode>
2782 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2783 (outs QPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2784 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2786 (TyQ (OpNode (TyD DPR:$Vn),
2787 (TyD (NEONvduplane (TyD DPR_8:$Vm), imm:$lane)))))]>;
2789 // Long 3-register operations with explicitly extended operands.
2790 class N3VLExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2791 InstrItinClass itin, string OpcodeStr, string Dt,
2792 ValueType TyQ, ValueType TyD, SDNode OpNode, SDNode ExtOp,
2794 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2795 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2796 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2797 [(set QPR:$Vd, (OpNode (TyQ (ExtOp (TyD DPR:$Vn))),
2798 (TyQ (ExtOp (TyD DPR:$Vm)))))]> {
2799 let isCommutable = Commutable;
2802 // Long 3-register intrinsics with explicit extend (VABDL).
2803 class N3VLIntExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2804 InstrItinClass itin, string OpcodeStr, string Dt,
2805 ValueType TyQ, ValueType TyD, Intrinsic IntOp, SDNode ExtOp,
2807 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2808 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2809 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2810 [(set QPR:$Vd, (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
2811 (TyD DPR:$Vm))))))]> {
2812 let isCommutable = Commutable;
2815 // Long 3-register intrinsics.
2816 class N3VLInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2817 InstrItinClass itin, string OpcodeStr, string Dt,
2818 ValueType TyQ, ValueType TyD, Intrinsic IntOp, bit Commutable>
2819 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2820 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2821 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2822 [(set QPR:$Vd, (TyQ (IntOp (TyD DPR:$Vn), (TyD DPR:$Vm))))]> {
2823 let isCommutable = Commutable;
2825 class N3VLIntSL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2826 string OpcodeStr, string Dt,
2827 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2828 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2829 (outs QPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2830 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2831 [(set (ResTy QPR:$Vd),
2832 (ResTy (IntOp (OpTy DPR:$Vn),
2833 (OpTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2835 class N3VLIntSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2836 InstrItinClass itin, string OpcodeStr, string Dt,
2837 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2838 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2839 (outs QPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2840 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2841 [(set (ResTy QPR:$Vd),
2842 (ResTy (IntOp (OpTy DPR:$Vn),
2843 (OpTy (NEONvduplane (OpTy DPR_8:$Vm),
2846 // Wide 3-register operations.
2847 class N3VW<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2848 string OpcodeStr, string Dt, ValueType TyQ, ValueType TyD,
2849 SDNode OpNode, SDNode ExtOp, bit Commutable>
2850 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2851 (outs QPR:$Vd), (ins QPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VSUBiD,
2852 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2853 [(set QPR:$Vd, (OpNode (TyQ QPR:$Vn),
2854 (TyQ (ExtOp (TyD DPR:$Vm)))))]> {
2855 // All of these have a two-operand InstAlias.
2856 let TwoOperandAliasConstraint = "$Vn = $Vd";
2857 let isCommutable = Commutable;
2860 // Pairwise long 2-register intrinsics, both double- and quad-register.
2861 class N2VDPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2862 bits<2> op17_16, bits<5> op11_7, bit op4,
2863 string OpcodeStr, string Dt,
2864 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2865 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
2866 (ins DPR:$Vm), IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
2867 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm))))]>;
2868 class N2VQPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2869 bits<2> op17_16, bits<5> op11_7, bit op4,
2870 string OpcodeStr, string Dt,
2871 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2872 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
2873 (ins QPR:$Vm), IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
2874 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm))))]>;
2876 // Pairwise long 2-register accumulate intrinsics,
2877 // both double- and quad-register.
2878 // The destination register is also used as the first source operand register.
2879 class N2VDPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2880 bits<2> op17_16, bits<5> op11_7, bit op4,
2881 string OpcodeStr, string Dt,
2882 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2883 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
2884 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vm), IIC_VPALiD,
2885 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
2886 [(set DPR:$Vd, (ResTy (IntOp (ResTy DPR:$src1), (OpTy DPR:$Vm))))]>;
2887 class N2VQPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2888 bits<2> op17_16, bits<5> op11_7, bit op4,
2889 string OpcodeStr, string Dt,
2890 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2891 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4,
2892 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vm), IIC_VPALiQ,
2893 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
2894 [(set QPR:$Vd, (ResTy (IntOp (ResTy QPR:$src1), (OpTy QPR:$Vm))))]>;
2896 // Shift by immediate,
2897 // both double- and quad-register.
2898 let TwoOperandAliasConstraint = "$Vm = $Vd" in {
2899 class N2VDSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2900 Format f, InstrItinClass itin, Operand ImmTy,
2901 string OpcodeStr, string Dt, ValueType Ty, SDNode OpNode>
2902 : N2VImm<op24, op23, op11_8, op7, 0, op4,
2903 (outs DPR:$Vd), (ins DPR:$Vm, ImmTy:$SIMM), f, itin,
2904 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2905 [(set DPR:$Vd, (Ty (OpNode (Ty DPR:$Vm), (i32 imm:$SIMM))))]>;
2906 class N2VQSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2907 Format f, InstrItinClass itin, Operand ImmTy,
2908 string OpcodeStr, string Dt, ValueType Ty, SDNode OpNode>
2909 : N2VImm<op24, op23, op11_8, op7, 1, op4,
2910 (outs QPR:$Vd), (ins QPR:$Vm, ImmTy:$SIMM), f, itin,
2911 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2912 [(set QPR:$Vd, (Ty (OpNode (Ty QPR:$Vm), (i32 imm:$SIMM))))]>;
2915 // Long shift by immediate.
2916 class N2VLSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
2917 string OpcodeStr, string Dt,
2918 ValueType ResTy, ValueType OpTy, Operand ImmTy, SDNode OpNode>
2919 : N2VImm<op24, op23, op11_8, op7, op6, op4,
2920 (outs QPR:$Vd), (ins DPR:$Vm, ImmTy:$SIMM), N2RegVShLFrm,
2921 IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2922 [(set QPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vm),
2923 (i32 imm:$SIMM))))]>;
2925 // Narrow shift by immediate.
2926 class N2VNSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
2927 InstrItinClass itin, string OpcodeStr, string Dt,
2928 ValueType ResTy, ValueType OpTy, Operand ImmTy, SDNode OpNode>
2929 : N2VImm<op24, op23, op11_8, op7, op6, op4,
2930 (outs DPR:$Vd), (ins QPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, itin,
2931 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2932 [(set DPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vm),
2933 (i32 imm:$SIMM))))]>;
2935 // Shift right by immediate and accumulate,
2936 // both double- and quad-register.
2937 class N2VDShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2938 Operand ImmTy, string OpcodeStr, string Dt,
2939 ValueType Ty, SDNode ShOp>
2940 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
2941 (ins DPR:$src1, DPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, IIC_VPALiD,
2942 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2943 [(set DPR:$Vd, (Ty (add DPR:$src1,
2944 (Ty (ShOp DPR:$Vm, (i32 imm:$SIMM))))))]>;
2945 class N2VQShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2946 Operand ImmTy, string OpcodeStr, string Dt,
2947 ValueType Ty, SDNode ShOp>
2948 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
2949 (ins QPR:$src1, QPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, IIC_VPALiD,
2950 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2951 [(set QPR:$Vd, (Ty (add QPR:$src1,
2952 (Ty (ShOp QPR:$Vm, (i32 imm:$SIMM))))))]>;
2954 // Shift by immediate and insert,
2955 // both double- and quad-register.
2956 class N2VDShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2957 Operand ImmTy, Format f, string OpcodeStr, string Dt,
2958 ValueType Ty,SDNode ShOp>
2959 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
2960 (ins DPR:$src1, DPR:$Vm, ImmTy:$SIMM), f, IIC_VSHLiD,
2961 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2962 [(set DPR:$Vd, (Ty (ShOp DPR:$src1, DPR:$Vm, (i32 imm:$SIMM))))]>;
2963 class N2VQShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2964 Operand ImmTy, Format f, string OpcodeStr, string Dt,
2965 ValueType Ty,SDNode ShOp>
2966 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
2967 (ins QPR:$src1, QPR:$Vm, ImmTy:$SIMM), f, IIC_VSHLiQ,
2968 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2969 [(set QPR:$Vd, (Ty (ShOp QPR:$src1, QPR:$Vm, (i32 imm:$SIMM))))]>;
2971 // Convert, with fractional bits immediate,
2972 // both double- and quad-register.
2973 class N2VCvtD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2974 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2976 : N2VImm<op24, op23, op11_8, op7, 0, op4,
2977 (outs DPR:$Vd), (ins DPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
2978 IIC_VUNAD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2979 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (i32 imm:$SIMM))))]>;
2980 class N2VCvtQ<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2981 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2983 : N2VImm<op24, op23, op11_8, op7, 1, op4,
2984 (outs QPR:$Vd), (ins QPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
2985 IIC_VUNAQ, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2986 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (i32 imm:$SIMM))))]>;
2988 //===----------------------------------------------------------------------===//
2990 //===----------------------------------------------------------------------===//
2992 // Abbreviations used in multiclass suffixes:
2993 // Q = quarter int (8 bit) elements
2994 // H = half int (16 bit) elements
2995 // S = single int (32 bit) elements
2996 // D = double int (64 bit) elements
2998 // Neon 2-register vector operations and intrinsics.
3000 // Neon 2-register comparisons.
3001 // source operand element sizes of 8, 16 and 32 bits:
3002 multiclass N2V_QHS_cmp<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3003 bits<5> op11_7, bit op4, string opc, string Dt,
3004 string asm, SDNode OpNode> {
3005 // 64-bit vector types.
3006 def v8i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 0, op4,
3007 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
3008 opc, !strconcat(Dt, "8"), asm, "",
3009 [(set DPR:$Vd, (v8i8 (OpNode (v8i8 DPR:$Vm))))]>;
3010 def v4i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 0, op4,
3011 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
3012 opc, !strconcat(Dt, "16"), asm, "",
3013 [(set DPR:$Vd, (v4i16 (OpNode (v4i16 DPR:$Vm))))]>;
3014 def v2i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
3015 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
3016 opc, !strconcat(Dt, "32"), asm, "",
3017 [(set DPR:$Vd, (v2i32 (OpNode (v2i32 DPR:$Vm))))]>;
3018 def v2f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
3019 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
3020 opc, "f32", asm, "",
3021 [(set DPR:$Vd, (v2i32 (OpNode (v2f32 DPR:$Vm))))]> {
3022 let Inst{10} = 1; // overwrite F = 1
3025 // 128-bit vector types.
3026 def v16i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 1, op4,
3027 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
3028 opc, !strconcat(Dt, "8"), asm, "",
3029 [(set QPR:$Vd, (v16i8 (OpNode (v16i8 QPR:$Vm))))]>;
3030 def v8i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 1, op4,
3031 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
3032 opc, !strconcat(Dt, "16"), asm, "",
3033 [(set QPR:$Vd, (v8i16 (OpNode (v8i16 QPR:$Vm))))]>;
3034 def v4i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
3035 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
3036 opc, !strconcat(Dt, "32"), asm, "",
3037 [(set QPR:$Vd, (v4i32 (OpNode (v4i32 QPR:$Vm))))]>;
3038 def v4f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
3039 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
3040 opc, "f32", asm, "",
3041 [(set QPR:$Vd, (v4i32 (OpNode (v4f32 QPR:$Vm))))]> {
3042 let Inst{10} = 1; // overwrite F = 1
3047 // Neon 2-register vector intrinsics,
3048 // element sizes of 8, 16 and 32 bits:
3049 multiclass N2VInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3050 bits<5> op11_7, bit op4,
3051 InstrItinClass itinD, InstrItinClass itinQ,
3052 string OpcodeStr, string Dt, Intrinsic IntOp> {
3053 // 64-bit vector types.
3054 def v8i8 : N2VDInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3055 itinD, OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
3056 def v4i16 : N2VDInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3057 itinD, OpcodeStr, !strconcat(Dt, "16"),v4i16,v4i16,IntOp>;
3058 def v2i32 : N2VDInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3059 itinD, OpcodeStr, !strconcat(Dt, "32"),v2i32,v2i32,IntOp>;
3061 // 128-bit vector types.
3062 def v16i8 : N2VQInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3063 itinQ, OpcodeStr, !strconcat(Dt, "8"), v16i8,v16i8,IntOp>;
3064 def v8i16 : N2VQInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3065 itinQ, OpcodeStr, !strconcat(Dt, "16"),v8i16,v8i16,IntOp>;
3066 def v4i32 : N2VQInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3067 itinQ, OpcodeStr, !strconcat(Dt, "32"),v4i32,v4i32,IntOp>;
3071 // Neon Narrowing 2-register vector operations,
3072 // source operand element sizes of 16, 32 and 64 bits:
3073 multiclass N2VN_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3074 bits<5> op11_7, bit op6, bit op4,
3075 InstrItinClass itin, string OpcodeStr, string Dt,
3077 def v8i8 : N2VN<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
3078 itin, OpcodeStr, !strconcat(Dt, "16"),
3079 v8i8, v8i16, OpNode>;
3080 def v4i16 : N2VN<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
3081 itin, OpcodeStr, !strconcat(Dt, "32"),
3082 v4i16, v4i32, OpNode>;
3083 def v2i32 : N2VN<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
3084 itin, OpcodeStr, !strconcat(Dt, "64"),
3085 v2i32, v2i64, OpNode>;
3088 // Neon Narrowing 2-register vector intrinsics,
3089 // source operand element sizes of 16, 32 and 64 bits:
3090 multiclass N2VNInt_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3091 bits<5> op11_7, bit op6, bit op4,
3092 InstrItinClass itin, string OpcodeStr, string Dt,
3094 def v8i8 : N2VNInt<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
3095 itin, OpcodeStr, !strconcat(Dt, "16"),
3096 v8i8, v8i16, IntOp>;
3097 def v4i16 : N2VNInt<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
3098 itin, OpcodeStr, !strconcat(Dt, "32"),
3099 v4i16, v4i32, IntOp>;
3100 def v2i32 : N2VNInt<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
3101 itin, OpcodeStr, !strconcat(Dt, "64"),
3102 v2i32, v2i64, IntOp>;
3106 // Neon Lengthening 2-register vector intrinsic (currently specific to VMOVL).
3107 // source operand element sizes of 16, 32 and 64 bits:
3108 multiclass N2VL_QHS<bits<2> op24_23, bits<5> op11_7, bit op6, bit op4,
3109 string OpcodeStr, string Dt, SDNode OpNode> {
3110 def v8i16 : N2VL<op24_23, 0b00, 0b10, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
3111 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode>;
3112 def v4i32 : N2VL<op24_23, 0b01, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
3113 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
3114 def v2i64 : N2VL<op24_23, 0b10, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
3115 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
3119 // Neon 3-register vector operations.
3121 // First with only element sizes of 8, 16 and 32 bits:
3122 multiclass N3V_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3123 InstrItinClass itinD16, InstrItinClass itinD32,
3124 InstrItinClass itinQ16, InstrItinClass itinQ32,
3125 string OpcodeStr, string Dt,
3126 SDNode OpNode, bit Commutable = 0> {
3127 // 64-bit vector types.
3128 def v8i8 : N3VD<op24, op23, 0b00, op11_8, op4, itinD16,
3129 OpcodeStr, !strconcat(Dt, "8"),
3130 v8i8, v8i8, OpNode, Commutable>;
3131 def v4i16 : N3VD<op24, op23, 0b01, op11_8, op4, itinD16,
3132 OpcodeStr, !strconcat(Dt, "16"),
3133 v4i16, v4i16, OpNode, Commutable>;
3134 def v2i32 : N3VD<op24, op23, 0b10, op11_8, op4, itinD32,
3135 OpcodeStr, !strconcat(Dt, "32"),
3136 v2i32, v2i32, OpNode, Commutable>;
3138 // 128-bit vector types.
3139 def v16i8 : N3VQ<op24, op23, 0b00, op11_8, op4, itinQ16,
3140 OpcodeStr, !strconcat(Dt, "8"),
3141 v16i8, v16i8, OpNode, Commutable>;
3142 def v8i16 : N3VQ<op24, op23, 0b01, op11_8, op4, itinQ16,
3143 OpcodeStr, !strconcat(Dt, "16"),
3144 v8i16, v8i16, OpNode, Commutable>;
3145 def v4i32 : N3VQ<op24, op23, 0b10, op11_8, op4, itinQ32,
3146 OpcodeStr, !strconcat(Dt, "32"),
3147 v4i32, v4i32, OpNode, Commutable>;
3150 multiclass N3VSL_HS<bits<4> op11_8, string OpcodeStr, SDNode ShOp> {
3151 def v4i16 : N3VDSL16<0b01, op11_8, OpcodeStr, "i16", v4i16, ShOp>;
3152 def v2i32 : N3VDSL<0b10, op11_8, IIC_VMULi32D, OpcodeStr, "i32", v2i32, ShOp>;
3153 def v8i16 : N3VQSL16<0b01, op11_8, OpcodeStr, "i16", v8i16, v4i16, ShOp>;
3154 def v4i32 : N3VQSL<0b10, op11_8, IIC_VMULi32Q, OpcodeStr, "i32",
3155 v4i32, v2i32, ShOp>;
3158 // ....then also with element size 64 bits:
3159 multiclass N3V_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3160 InstrItinClass itinD, InstrItinClass itinQ,
3161 string OpcodeStr, string Dt,
3162 SDNode OpNode, bit Commutable = 0>
3163 : N3V_QHS<op24, op23, op11_8, op4, itinD, itinD, itinQ, itinQ,
3164 OpcodeStr, Dt, OpNode, Commutable> {
3165 def v1i64 : N3VD<op24, op23, 0b11, op11_8, op4, itinD,
3166 OpcodeStr, !strconcat(Dt, "64"),
3167 v1i64, v1i64, OpNode, Commutable>;
3168 def v2i64 : N3VQ<op24, op23, 0b11, op11_8, op4, itinQ,
3169 OpcodeStr, !strconcat(Dt, "64"),
3170 v2i64, v2i64, OpNode, Commutable>;
3174 // Neon 3-register vector intrinsics.
3176 // First with only element sizes of 16 and 32 bits:
3177 multiclass N3VInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3178 InstrItinClass itinD16, InstrItinClass itinD32,
3179 InstrItinClass itinQ16, InstrItinClass itinQ32,
3180 string OpcodeStr, string Dt,
3181 Intrinsic IntOp, bit Commutable = 0> {
3182 // 64-bit vector types.
3183 def v4i16 : N3VDInt<op24, op23, 0b01, op11_8, op4, f, itinD16,
3184 OpcodeStr, !strconcat(Dt, "16"),
3185 v4i16, v4i16, IntOp, Commutable>;
3186 def v2i32 : N3VDInt<op24, op23, 0b10, op11_8, op4, f, itinD32,
3187 OpcodeStr, !strconcat(Dt, "32"),
3188 v2i32, v2i32, IntOp, Commutable>;
3190 // 128-bit vector types.
3191 def v8i16 : N3VQInt<op24, op23, 0b01, op11_8, op4, f, itinQ16,
3192 OpcodeStr, !strconcat(Dt, "16"),
3193 v8i16, v8i16, IntOp, Commutable>;
3194 def v4i32 : N3VQInt<op24, op23, 0b10, op11_8, op4, f, itinQ32,
3195 OpcodeStr, !strconcat(Dt, "32"),
3196 v4i32, v4i32, IntOp, Commutable>;
3198 multiclass N3VInt_HSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3199 InstrItinClass itinD16, InstrItinClass itinD32,
3200 InstrItinClass itinQ16, InstrItinClass itinQ32,
3201 string OpcodeStr, string Dt,
3203 // 64-bit vector types.
3204 def v4i16 : N3VDIntSh<op24, op23, 0b01, op11_8, op4, f, itinD16,
3205 OpcodeStr, !strconcat(Dt, "16"),
3206 v4i16, v4i16, IntOp>;
3207 def v2i32 : N3VDIntSh<op24, op23, 0b10, op11_8, op4, f, itinD32,
3208 OpcodeStr, !strconcat(Dt, "32"),
3209 v2i32, v2i32, IntOp>;
3211 // 128-bit vector types.
3212 def v8i16 : N3VQIntSh<op24, op23, 0b01, op11_8, op4, f, itinQ16,
3213 OpcodeStr, !strconcat(Dt, "16"),
3214 v8i16, v8i16, IntOp>;
3215 def v4i32 : N3VQIntSh<op24, op23, 0b10, op11_8, op4, f, itinQ32,
3216 OpcodeStr, !strconcat(Dt, "32"),
3217 v4i32, v4i32, IntOp>;
3220 multiclass N3VIntSL_HS<bits<4> op11_8,
3221 InstrItinClass itinD16, InstrItinClass itinD32,
3222 InstrItinClass itinQ16, InstrItinClass itinQ32,
3223 string OpcodeStr, string Dt, Intrinsic IntOp> {
3224 def v4i16 : N3VDIntSL16<0b01, op11_8, itinD16,
3225 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp>;
3226 def v2i32 : N3VDIntSL<0b10, op11_8, itinD32,
3227 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp>;
3228 def v8i16 : N3VQIntSL16<0b01, op11_8, itinQ16,
3229 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16, IntOp>;
3230 def v4i32 : N3VQIntSL<0b10, op11_8, itinQ32,
3231 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32, IntOp>;
3234 // ....then also with element size of 8 bits:
3235 multiclass N3VInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3236 InstrItinClass itinD16, InstrItinClass itinD32,
3237 InstrItinClass itinQ16, InstrItinClass itinQ32,
3238 string OpcodeStr, string Dt,
3239 Intrinsic IntOp, bit Commutable = 0>
3240 : N3VInt_HS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
3241 OpcodeStr, Dt, IntOp, Commutable> {
3242 def v8i8 : N3VDInt<op24, op23, 0b00, op11_8, op4, f, itinD16,
3243 OpcodeStr, !strconcat(Dt, "8"),
3244 v8i8, v8i8, IntOp, Commutable>;
3245 def v16i8 : N3VQInt<op24, op23, 0b00, op11_8, op4, f, itinQ16,
3246 OpcodeStr, !strconcat(Dt, "8"),
3247 v16i8, v16i8, IntOp, Commutable>;
3249 multiclass N3VInt_QHSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3250 InstrItinClass itinD16, InstrItinClass itinD32,
3251 InstrItinClass itinQ16, InstrItinClass itinQ32,
3252 string OpcodeStr, string Dt,
3254 : N3VInt_HSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
3255 OpcodeStr, Dt, IntOp> {
3256 def v8i8 : N3VDIntSh<op24, op23, 0b00, op11_8, op4, f, itinD16,
3257 OpcodeStr, !strconcat(Dt, "8"),
3259 def v16i8 : N3VQIntSh<op24, op23, 0b00, op11_8, op4, f, itinQ16,
3260 OpcodeStr, !strconcat(Dt, "8"),
3261 v16i8, v16i8, IntOp>;
3265 // ....then also with element size of 64 bits:
3266 multiclass N3VInt_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3267 InstrItinClass itinD16, InstrItinClass itinD32,
3268 InstrItinClass itinQ16, InstrItinClass itinQ32,
3269 string OpcodeStr, string Dt,
3270 Intrinsic IntOp, bit Commutable = 0>
3271 : N3VInt_QHS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
3272 OpcodeStr, Dt, IntOp, Commutable> {
3273 def v1i64 : N3VDInt<op24, op23, 0b11, op11_8, op4, f, itinD32,
3274 OpcodeStr, !strconcat(Dt, "64"),
3275 v1i64, v1i64, IntOp, Commutable>;
3276 def v2i64 : N3VQInt<op24, op23, 0b11, op11_8, op4, f, itinQ32,
3277 OpcodeStr, !strconcat(Dt, "64"),
3278 v2i64, v2i64, IntOp, Commutable>;
3280 multiclass N3VInt_QHSDSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3281 InstrItinClass itinD16, InstrItinClass itinD32,
3282 InstrItinClass itinQ16, InstrItinClass itinQ32,
3283 string OpcodeStr, string Dt,
3285 : N3VInt_QHSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
3286 OpcodeStr, Dt, IntOp> {
3287 def v1i64 : N3VDIntSh<op24, op23, 0b11, op11_8, op4, f, itinD32,
3288 OpcodeStr, !strconcat(Dt, "64"),
3289 v1i64, v1i64, IntOp>;
3290 def v2i64 : N3VQIntSh<op24, op23, 0b11, op11_8, op4, f, itinQ32,
3291 OpcodeStr, !strconcat(Dt, "64"),
3292 v2i64, v2i64, IntOp>;
3295 // Neon Narrowing 3-register vector intrinsics,
3296 // source operand element sizes of 16, 32 and 64 bits:
3297 multiclass N3VNInt_HSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3298 string OpcodeStr, string Dt,
3299 Intrinsic IntOp, bit Commutable = 0> {
3300 def v8i8 : N3VNInt<op24, op23, 0b00, op11_8, op4,
3301 OpcodeStr, !strconcat(Dt, "16"),
3302 v8i8, v8i16, IntOp, Commutable>;
3303 def v4i16 : N3VNInt<op24, op23, 0b01, op11_8, op4,
3304 OpcodeStr, !strconcat(Dt, "32"),
3305 v4i16, v4i32, IntOp, Commutable>;
3306 def v2i32 : N3VNInt<op24, op23, 0b10, op11_8, op4,
3307 OpcodeStr, !strconcat(Dt, "64"),
3308 v2i32, v2i64, IntOp, Commutable>;
3312 // Neon Long 3-register vector operations.
3314 multiclass N3VL_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3315 InstrItinClass itin16, InstrItinClass itin32,
3316 string OpcodeStr, string Dt,
3317 SDNode OpNode, bit Commutable = 0> {
3318 def v8i16 : N3VL<op24, op23, 0b00, op11_8, op4, itin16,
3319 OpcodeStr, !strconcat(Dt, "8"),
3320 v8i16, v8i8, OpNode, Commutable>;
3321 def v4i32 : N3VL<op24, op23, 0b01, op11_8, op4, itin16,
3322 OpcodeStr, !strconcat(Dt, "16"),
3323 v4i32, v4i16, OpNode, Commutable>;
3324 def v2i64 : N3VL<op24, op23, 0b10, op11_8, op4, itin32,
3325 OpcodeStr, !strconcat(Dt, "32"),
3326 v2i64, v2i32, OpNode, Commutable>;
3329 multiclass N3VLSL_HS<bit op24, bits<4> op11_8,
3330 InstrItinClass itin, string OpcodeStr, string Dt,
3332 def v4i16 : N3VLSL16<op24, 0b01, op11_8, itin, OpcodeStr,
3333 !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
3334 def v2i32 : N3VLSL<op24, 0b10, op11_8, itin, OpcodeStr,
3335 !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
3338 multiclass N3VLExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3339 InstrItinClass itin16, InstrItinClass itin32,
3340 string OpcodeStr, string Dt,
3341 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
3342 def v8i16 : N3VLExt<op24, op23, 0b00, op11_8, op4, itin16,
3343 OpcodeStr, !strconcat(Dt, "8"),
3344 v8i16, v8i8, OpNode, ExtOp, Commutable>;
3345 def v4i32 : N3VLExt<op24, op23, 0b01, op11_8, op4, itin16,
3346 OpcodeStr, !strconcat(Dt, "16"),
3347 v4i32, v4i16, OpNode, ExtOp, Commutable>;
3348 def v2i64 : N3VLExt<op24, op23, 0b10, op11_8, op4, itin32,
3349 OpcodeStr, !strconcat(Dt, "32"),
3350 v2i64, v2i32, OpNode, ExtOp, Commutable>;
3353 // Neon Long 3-register vector intrinsics.
3355 // First with only element sizes of 16 and 32 bits:
3356 multiclass N3VLInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
3357 InstrItinClass itin16, InstrItinClass itin32,
3358 string OpcodeStr, string Dt,
3359 Intrinsic IntOp, bit Commutable = 0> {
3360 def v4i32 : N3VLInt<op24, op23, 0b01, op11_8, op4, itin16,
3361 OpcodeStr, !strconcat(Dt, "16"),
3362 v4i32, v4i16, IntOp, Commutable>;
3363 def v2i64 : N3VLInt<op24, op23, 0b10, op11_8, op4, itin32,
3364 OpcodeStr, !strconcat(Dt, "32"),
3365 v2i64, v2i32, IntOp, Commutable>;
3368 multiclass N3VLIntSL_HS<bit op24, bits<4> op11_8,
3369 InstrItinClass itin, string OpcodeStr, string Dt,
3371 def v4i16 : N3VLIntSL16<op24, 0b01, op11_8, itin,
3372 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
3373 def v2i32 : N3VLIntSL<op24, 0b10, op11_8, itin,
3374 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3377 // ....then also with element size of 8 bits:
3378 multiclass N3VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3379 InstrItinClass itin16, InstrItinClass itin32,
3380 string OpcodeStr, string Dt,
3381 Intrinsic IntOp, bit Commutable = 0>
3382 : N3VLInt_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt,
3383 IntOp, Commutable> {
3384 def v8i16 : N3VLInt<op24, op23, 0b00, op11_8, op4, itin16,
3385 OpcodeStr, !strconcat(Dt, "8"),
3386 v8i16, v8i8, IntOp, Commutable>;
3389 // ....with explicit extend (VABDL).
3390 multiclass N3VLIntExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3391 InstrItinClass itin, string OpcodeStr, string Dt,
3392 Intrinsic IntOp, SDNode ExtOp, bit Commutable = 0> {
3393 def v8i16 : N3VLIntExt<op24, op23, 0b00, op11_8, op4, itin,
3394 OpcodeStr, !strconcat(Dt, "8"),
3395 v8i16, v8i8, IntOp, ExtOp, Commutable>;
3396 def v4i32 : N3VLIntExt<op24, op23, 0b01, op11_8, op4, itin,
3397 OpcodeStr, !strconcat(Dt, "16"),
3398 v4i32, v4i16, IntOp, ExtOp, Commutable>;
3399 def v2i64 : N3VLIntExt<op24, op23, 0b10, op11_8, op4, itin,
3400 OpcodeStr, !strconcat(Dt, "32"),
3401 v2i64, v2i32, IntOp, ExtOp, Commutable>;
3405 // Neon Wide 3-register vector intrinsics,
3406 // source operand element sizes of 8, 16 and 32 bits:
3407 multiclass N3VW_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3408 string OpcodeStr, string Dt,
3409 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
3410 def v8i16 : N3VW<op24, op23, 0b00, op11_8, op4,
3411 OpcodeStr, !strconcat(Dt, "8"),
3412 v8i16, v8i8, OpNode, ExtOp, Commutable>;
3413 def v4i32 : N3VW<op24, op23, 0b01, op11_8, op4,
3414 OpcodeStr, !strconcat(Dt, "16"),
3415 v4i32, v4i16, OpNode, ExtOp, Commutable>;
3416 def v2i64 : N3VW<op24, op23, 0b10, op11_8, op4,
3417 OpcodeStr, !strconcat(Dt, "32"),
3418 v2i64, v2i32, OpNode, ExtOp, Commutable>;
3422 // Neon Multiply-Op vector operations,
3423 // element sizes of 8, 16 and 32 bits:
3424 multiclass N3VMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3425 InstrItinClass itinD16, InstrItinClass itinD32,
3426 InstrItinClass itinQ16, InstrItinClass itinQ32,
3427 string OpcodeStr, string Dt, SDNode OpNode> {
3428 // 64-bit vector types.
3429 def v8i8 : N3VDMulOp<op24, op23, 0b00, op11_8, op4, itinD16,
3430 OpcodeStr, !strconcat(Dt, "8"), v8i8, mul, OpNode>;
3431 def v4i16 : N3VDMulOp<op24, op23, 0b01, op11_8, op4, itinD16,
3432 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, OpNode>;
3433 def v2i32 : N3VDMulOp<op24, op23, 0b10, op11_8, op4, itinD32,
3434 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, OpNode>;
3436 // 128-bit vector types.
3437 def v16i8 : N3VQMulOp<op24, op23, 0b00, op11_8, op4, itinQ16,
3438 OpcodeStr, !strconcat(Dt, "8"), v16i8, mul, OpNode>;
3439 def v8i16 : N3VQMulOp<op24, op23, 0b01, op11_8, op4, itinQ16,
3440 OpcodeStr, !strconcat(Dt, "16"), v8i16, mul, OpNode>;
3441 def v4i32 : N3VQMulOp<op24, op23, 0b10, op11_8, op4, itinQ32,
3442 OpcodeStr, !strconcat(Dt, "32"), v4i32, mul, OpNode>;
3445 multiclass N3VMulOpSL_HS<bits<4> op11_8,
3446 InstrItinClass itinD16, InstrItinClass itinD32,
3447 InstrItinClass itinQ16, InstrItinClass itinQ32,
3448 string OpcodeStr, string Dt, SDNode ShOp> {
3449 def v4i16 : N3VDMulOpSL16<0b01, op11_8, itinD16,
3450 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, ShOp>;
3451 def v2i32 : N3VDMulOpSL<0b10, op11_8, itinD32,
3452 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, ShOp>;
3453 def v8i16 : N3VQMulOpSL16<0b01, op11_8, itinQ16,
3454 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16,
3456 def v4i32 : N3VQMulOpSL<0b10, op11_8, itinQ32,
3457 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32,
3461 // Neon Intrinsic-Op vector operations,
3462 // element sizes of 8, 16 and 32 bits:
3463 multiclass N3VIntOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3464 InstrItinClass itinD, InstrItinClass itinQ,
3465 string OpcodeStr, string Dt, Intrinsic IntOp,
3467 // 64-bit vector types.
3468 def v8i8 : N3VDIntOp<op24, op23, 0b00, op11_8, op4, itinD,
3469 OpcodeStr, !strconcat(Dt, "8"), v8i8, IntOp, OpNode>;
3470 def v4i16 : N3VDIntOp<op24, op23, 0b01, op11_8, op4, itinD,
3471 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp, OpNode>;
3472 def v2i32 : N3VDIntOp<op24, op23, 0b10, op11_8, op4, itinD,
3473 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp, OpNode>;
3475 // 128-bit vector types.
3476 def v16i8 : N3VQIntOp<op24, op23, 0b00, op11_8, op4, itinQ,
3477 OpcodeStr, !strconcat(Dt, "8"), v16i8, IntOp, OpNode>;
3478 def v8i16 : N3VQIntOp<op24, op23, 0b01, op11_8, op4, itinQ,
3479 OpcodeStr, !strconcat(Dt, "16"), v8i16, IntOp, OpNode>;
3480 def v4i32 : N3VQIntOp<op24, op23, 0b10, op11_8, op4, itinQ,
3481 OpcodeStr, !strconcat(Dt, "32"), v4i32, IntOp, OpNode>;
3484 // Neon 3-argument intrinsics,
3485 // element sizes of 8, 16 and 32 bits:
3486 multiclass N3VInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3487 InstrItinClass itinD, InstrItinClass itinQ,
3488 string OpcodeStr, string Dt, Intrinsic IntOp> {
3489 // 64-bit vector types.
3490 def v8i8 : N3VDInt3<op24, op23, 0b00, op11_8, op4, itinD,
3491 OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
3492 def v4i16 : N3VDInt3<op24, op23, 0b01, op11_8, op4, itinD,
3493 OpcodeStr, !strconcat(Dt, "16"), v4i16, v4i16, IntOp>;
3494 def v2i32 : N3VDInt3<op24, op23, 0b10, op11_8, op4, itinD,
3495 OpcodeStr, !strconcat(Dt, "32"), v2i32, v2i32, IntOp>;
3497 // 128-bit vector types.
3498 def v16i8 : N3VQInt3<op24, op23, 0b00, op11_8, op4, itinQ,
3499 OpcodeStr, !strconcat(Dt, "8"), v16i8, v16i8, IntOp>;
3500 def v8i16 : N3VQInt3<op24, op23, 0b01, op11_8, op4, itinQ,
3501 OpcodeStr, !strconcat(Dt, "16"), v8i16, v8i16, IntOp>;
3502 def v4i32 : N3VQInt3<op24, op23, 0b10, op11_8, op4, itinQ,
3503 OpcodeStr, !strconcat(Dt, "32"), v4i32, v4i32, IntOp>;
3507 // Neon Long Multiply-Op vector operations,
3508 // element sizes of 8, 16 and 32 bits:
3509 multiclass N3VLMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3510 InstrItinClass itin16, InstrItinClass itin32,
3511 string OpcodeStr, string Dt, SDNode MulOp,
3513 def v8i16 : N3VLMulOp<op24, op23, 0b00, op11_8, op4, itin16, OpcodeStr,
3514 !strconcat(Dt, "8"), v8i16, v8i8, MulOp, OpNode>;
3515 def v4i32 : N3VLMulOp<op24, op23, 0b01, op11_8, op4, itin16, OpcodeStr,
3516 !strconcat(Dt, "16"), v4i32, v4i16, MulOp, OpNode>;
3517 def v2i64 : N3VLMulOp<op24, op23, 0b10, op11_8, op4, itin32, OpcodeStr,
3518 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
3521 multiclass N3VLMulOpSL_HS<bit op24, bits<4> op11_8, string OpcodeStr,
3522 string Dt, SDNode MulOp, SDNode OpNode> {
3523 def v4i16 : N3VLMulOpSL16<op24, 0b01, op11_8, IIC_VMACi16D, OpcodeStr,
3524 !strconcat(Dt,"16"), v4i32, v4i16, MulOp, OpNode>;
3525 def v2i32 : N3VLMulOpSL<op24, 0b10, op11_8, IIC_VMACi32D, OpcodeStr,
3526 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
3530 // Neon Long 3-argument intrinsics.
3532 // First with only element sizes of 16 and 32 bits:
3533 multiclass N3VLInt3_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
3534 InstrItinClass itin16, InstrItinClass itin32,
3535 string OpcodeStr, string Dt, Intrinsic IntOp> {
3536 def v4i32 : N3VLInt3<op24, op23, 0b01, op11_8, op4, itin16,
3537 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
3538 def v2i64 : N3VLInt3<op24, op23, 0b10, op11_8, op4, itin32,
3539 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3542 multiclass N3VLInt3SL_HS<bit op24, bits<4> op11_8,
3543 string OpcodeStr, string Dt, Intrinsic IntOp> {
3544 def v4i16 : N3VLInt3SL16<op24, 0b01, op11_8, IIC_VMACi16D,
3545 OpcodeStr, !strconcat(Dt,"16"), v4i32, v4i16, IntOp>;
3546 def v2i32 : N3VLInt3SL<op24, 0b10, op11_8, IIC_VMACi32D,
3547 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3550 // ....then also with element size of 8 bits:
3551 multiclass N3VLInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3552 InstrItinClass itin16, InstrItinClass itin32,
3553 string OpcodeStr, string Dt, Intrinsic IntOp>
3554 : N3VLInt3_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt, IntOp> {
3555 def v8i16 : N3VLInt3<op24, op23, 0b00, op11_8, op4, itin16,
3556 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
3559 // ....with explicit extend (VABAL).
3560 multiclass N3VLIntExtOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3561 InstrItinClass itin, string OpcodeStr, string Dt,
3562 Intrinsic IntOp, SDNode ExtOp, SDNode OpNode> {
3563 def v8i16 : N3VLIntExtOp<op24, op23, 0b00, op11_8, op4, itin,
3564 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8,
3565 IntOp, ExtOp, OpNode>;
3566 def v4i32 : N3VLIntExtOp<op24, op23, 0b01, op11_8, op4, itin,
3567 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16,
3568 IntOp, ExtOp, OpNode>;
3569 def v2i64 : N3VLIntExtOp<op24, op23, 0b10, op11_8, op4, itin,
3570 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32,
3571 IntOp, ExtOp, OpNode>;
3575 // Neon Pairwise long 2-register intrinsics,
3576 // element sizes of 8, 16 and 32 bits:
3577 multiclass N2VPLInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3578 bits<5> op11_7, bit op4,
3579 string OpcodeStr, string Dt, Intrinsic IntOp> {
3580 // 64-bit vector types.
3581 def v8i8 : N2VDPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3582 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
3583 def v4i16 : N2VDPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3584 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
3585 def v2i32 : N2VDPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3586 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
3588 // 128-bit vector types.
3589 def v16i8 : N2VQPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3590 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
3591 def v8i16 : N2VQPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3592 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
3593 def v4i32 : N2VQPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3594 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
3598 // Neon Pairwise long 2-register accumulate intrinsics,
3599 // element sizes of 8, 16 and 32 bits:
3600 multiclass N2VPLInt2_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3601 bits<5> op11_7, bit op4,
3602 string OpcodeStr, string Dt, Intrinsic IntOp> {
3603 // 64-bit vector types.
3604 def v8i8 : N2VDPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3605 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
3606 def v4i16 : N2VDPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3607 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
3608 def v2i32 : N2VDPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3609 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
3611 // 128-bit vector types.
3612 def v16i8 : N2VQPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3613 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
3614 def v8i16 : N2VQPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3615 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
3616 def v4i32 : N2VQPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3617 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
3621 // Neon 2-register vector shift by immediate,
3622 // with f of either N2RegVShLFrm or N2RegVShRFrm
3623 // element sizes of 8, 16, 32 and 64 bits:
3624 multiclass N2VShL_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3625 InstrItinClass itin, string OpcodeStr, string Dt,
3627 // 64-bit vector types.
3628 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3629 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
3630 let Inst{21-19} = 0b001; // imm6 = 001xxx
3632 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3633 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
3634 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3636 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3637 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
3638 let Inst{21} = 0b1; // imm6 = 1xxxxx
3640 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, N2RegVShLFrm, itin, i32imm,
3641 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
3644 // 128-bit vector types.
3645 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3646 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
3647 let Inst{21-19} = 0b001; // imm6 = 001xxx
3649 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3650 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
3651 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3653 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3654 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
3655 let Inst{21} = 0b1; // imm6 = 1xxxxx
3657 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, N2RegVShLFrm, itin, i32imm,
3658 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
3661 multiclass N2VShR_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3662 InstrItinClass itin, string OpcodeStr, string Dt,
3663 string baseOpc, SDNode OpNode> {
3664 // 64-bit vector types.
3665 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm8,
3666 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
3667 let Inst{21-19} = 0b001; // imm6 = 001xxx
3669 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm16,
3670 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
3671 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3673 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm32,
3674 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
3675 let Inst{21} = 0b1; // imm6 = 1xxxxx
3677 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, N2RegVShRFrm, itin, shr_imm64,
3678 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
3681 // 128-bit vector types.
3682 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm8,
3683 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
3684 let Inst{21-19} = 0b001; // imm6 = 001xxx
3686 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm16,
3687 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
3688 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3690 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm32,
3691 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
3692 let Inst{21} = 0b1; // imm6 = 1xxxxx
3694 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, N2RegVShRFrm, itin, shr_imm64,
3695 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
3699 // Neon Shift-Accumulate vector operations,
3700 // element sizes of 8, 16, 32 and 64 bits:
3701 multiclass N2VShAdd_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3702 string OpcodeStr, string Dt, SDNode ShOp> {
3703 // 64-bit vector types.
3704 def v8i8 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm8,
3705 OpcodeStr, !strconcat(Dt, "8"), v8i8, ShOp> {
3706 let Inst{21-19} = 0b001; // imm6 = 001xxx
3708 def v4i16 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm16,
3709 OpcodeStr, !strconcat(Dt, "16"), v4i16, ShOp> {
3710 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3712 def v2i32 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm32,
3713 OpcodeStr, !strconcat(Dt, "32"), v2i32, ShOp> {
3714 let Inst{21} = 0b1; // imm6 = 1xxxxx
3716 def v1i64 : N2VDShAdd<op24, op23, op11_8, 1, op4, shr_imm64,
3717 OpcodeStr, !strconcat(Dt, "64"), v1i64, ShOp>;
3720 // 128-bit vector types.
3721 def v16i8 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm8,
3722 OpcodeStr, !strconcat(Dt, "8"), v16i8, ShOp> {
3723 let Inst{21-19} = 0b001; // imm6 = 001xxx
3725 def v8i16 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm16,
3726 OpcodeStr, !strconcat(Dt, "16"), v8i16, ShOp> {
3727 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3729 def v4i32 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm32,
3730 OpcodeStr, !strconcat(Dt, "32"), v4i32, ShOp> {
3731 let Inst{21} = 0b1; // imm6 = 1xxxxx
3733 def v2i64 : N2VQShAdd<op24, op23, op11_8, 1, op4, shr_imm64,
3734 OpcodeStr, !strconcat(Dt, "64"), v2i64, ShOp>;
3738 // Neon Shift-Insert vector operations,
3739 // with f of either N2RegVShLFrm or N2RegVShRFrm
3740 // element sizes of 8, 16, 32 and 64 bits:
3741 multiclass N2VShInsL_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3743 // 64-bit vector types.
3744 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3745 N2RegVShLFrm, OpcodeStr, "8", v8i8, NEONvsli> {
3746 let Inst{21-19} = 0b001; // imm6 = 001xxx
3748 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3749 N2RegVShLFrm, OpcodeStr, "16", v4i16, NEONvsli> {
3750 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3752 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3753 N2RegVShLFrm, OpcodeStr, "32", v2i32, NEONvsli> {
3754 let Inst{21} = 0b1; // imm6 = 1xxxxx
3756 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4, i32imm,
3757 N2RegVShLFrm, OpcodeStr, "64", v1i64, NEONvsli>;
3760 // 128-bit vector types.
3761 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3762 N2RegVShLFrm, OpcodeStr, "8", v16i8, NEONvsli> {
3763 let Inst{21-19} = 0b001; // imm6 = 001xxx
3765 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3766 N2RegVShLFrm, OpcodeStr, "16", v8i16, NEONvsli> {
3767 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3769 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3770 N2RegVShLFrm, OpcodeStr, "32", v4i32, NEONvsli> {
3771 let Inst{21} = 0b1; // imm6 = 1xxxxx
3773 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4, i32imm,
3774 N2RegVShLFrm, OpcodeStr, "64", v2i64, NEONvsli>;
3777 multiclass N2VShInsR_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3779 // 64-bit vector types.
3780 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm8,
3781 N2RegVShRFrm, OpcodeStr, "8", v8i8, NEONvsri> {
3782 let Inst{21-19} = 0b001; // imm6 = 001xxx
3784 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm16,
3785 N2RegVShRFrm, OpcodeStr, "16", v4i16, NEONvsri> {
3786 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3788 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm32,
3789 N2RegVShRFrm, OpcodeStr, "32", v2i32, NEONvsri> {
3790 let Inst{21} = 0b1; // imm6 = 1xxxxx
3792 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4, shr_imm64,
3793 N2RegVShRFrm, OpcodeStr, "64", v1i64, NEONvsri>;
3796 // 128-bit vector types.
3797 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm8,
3798 N2RegVShRFrm, OpcodeStr, "8", v16i8, NEONvsri> {
3799 let Inst{21-19} = 0b001; // imm6 = 001xxx
3801 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm16,
3802 N2RegVShRFrm, OpcodeStr, "16", v8i16, NEONvsri> {
3803 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3805 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm32,
3806 N2RegVShRFrm, OpcodeStr, "32", v4i32, NEONvsri> {
3807 let Inst{21} = 0b1; // imm6 = 1xxxxx
3809 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4, shr_imm64,
3810 N2RegVShRFrm, OpcodeStr, "64", v2i64, NEONvsri>;
3814 // Neon Shift Long operations,
3815 // element sizes of 8, 16, 32 bits:
3816 multiclass N2VLSh_QHS<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
3817 bit op4, string OpcodeStr, string Dt, SDNode OpNode> {
3818 def v8i16 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3819 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, imm1_7, OpNode> {
3820 let Inst{21-19} = 0b001; // imm6 = 001xxx
3822 def v4i32 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3823 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, imm1_15, OpNode> {
3824 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3826 def v2i64 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3827 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, imm1_31, OpNode> {
3828 let Inst{21} = 0b1; // imm6 = 1xxxxx
3832 // Neon Shift Narrow operations,
3833 // element sizes of 16, 32, 64 bits:
3834 multiclass N2VNSh_HSD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
3835 bit op4, InstrItinClass itin, string OpcodeStr, string Dt,
3837 def v8i8 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3838 OpcodeStr, !strconcat(Dt, "16"),
3839 v8i8, v8i16, shr_imm8, OpNode> {
3840 let Inst{21-19} = 0b001; // imm6 = 001xxx
3842 def v4i16 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3843 OpcodeStr, !strconcat(Dt, "32"),
3844 v4i16, v4i32, shr_imm16, OpNode> {
3845 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3847 def v2i32 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3848 OpcodeStr, !strconcat(Dt, "64"),
3849 v2i32, v2i64, shr_imm32, OpNode> {
3850 let Inst{21} = 0b1; // imm6 = 1xxxxx
3854 //===----------------------------------------------------------------------===//
3855 // Instruction Definitions.
3856 //===----------------------------------------------------------------------===//
3858 // Vector Add Operations.
3860 // VADD : Vector Add (integer and floating-point)
3861 defm VADD : N3V_QHSD<0, 0, 0b1000, 0, IIC_VBINiD, IIC_VBINiQ, "vadd", "i",
3863 def VADDfd : N3VD<0, 0, 0b00, 0b1101, 0, IIC_VBIND, "vadd", "f32",
3864 v2f32, v2f32, fadd, 1>;
3865 def VADDfq : N3VQ<0, 0, 0b00, 0b1101, 0, IIC_VBINQ, "vadd", "f32",
3866 v4f32, v4f32, fadd, 1>;
3867 // VADDL : Vector Add Long (Q = D + D)
3868 defm VADDLs : N3VLExt_QHS<0,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
3869 "vaddl", "s", add, sext, 1>;
3870 defm VADDLu : N3VLExt_QHS<1,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
3871 "vaddl", "u", add, zext, 1>;
3872 // VADDW : Vector Add Wide (Q = Q + D)
3873 defm VADDWs : N3VW_QHS<0,1,0b0001,0, "vaddw", "s", add, sext, 0>;
3874 defm VADDWu : N3VW_QHS<1,1,0b0001,0, "vaddw", "u", add, zext, 0>;
3875 // VHADD : Vector Halving Add
3876 defm VHADDs : N3VInt_QHS<0, 0, 0b0000, 0, N3RegFrm,
3877 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3878 "vhadd", "s", int_arm_neon_vhadds, 1>;
3879 defm VHADDu : N3VInt_QHS<1, 0, 0b0000, 0, N3RegFrm,
3880 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3881 "vhadd", "u", int_arm_neon_vhaddu, 1>;
3882 // VRHADD : Vector Rounding Halving Add
3883 defm VRHADDs : N3VInt_QHS<0, 0, 0b0001, 0, N3RegFrm,
3884 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3885 "vrhadd", "s", int_arm_neon_vrhadds, 1>;
3886 defm VRHADDu : N3VInt_QHS<1, 0, 0b0001, 0, N3RegFrm,
3887 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3888 "vrhadd", "u", int_arm_neon_vrhaddu, 1>;
3889 // VQADD : Vector Saturating Add
3890 defm VQADDs : N3VInt_QHSD<0, 0, 0b0000, 1, N3RegFrm,
3891 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3892 "vqadd", "s", int_arm_neon_vqadds, 1>;
3893 defm VQADDu : N3VInt_QHSD<1, 0, 0b0000, 1, N3RegFrm,
3894 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3895 "vqadd", "u", int_arm_neon_vqaddu, 1>;
3896 // VADDHN : Vector Add and Narrow Returning High Half (D = Q + Q)
3897 defm VADDHN : N3VNInt_HSD<0,1,0b0100,0, "vaddhn", "i",
3898 int_arm_neon_vaddhn, 1>;
3899 // VRADDHN : Vector Rounding Add and Narrow Returning High Half (D = Q + Q)
3900 defm VRADDHN : N3VNInt_HSD<1,1,0b0100,0, "vraddhn", "i",
3901 int_arm_neon_vraddhn, 1>;
3903 // Vector Multiply Operations.
3905 // VMUL : Vector Multiply (integer, polynomial and floating-point)
3906 defm VMUL : N3V_QHS<0, 0, 0b1001, 1, IIC_VMULi16D, IIC_VMULi32D,
3907 IIC_VMULi16Q, IIC_VMULi32Q, "vmul", "i", mul, 1>;
3908 def VMULpd : N3VDInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16D, "vmul",
3909 "p8", v8i8, v8i8, int_arm_neon_vmulp, 1>;
3910 def VMULpq : N3VQInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16Q, "vmul",
3911 "p8", v16i8, v16i8, int_arm_neon_vmulp, 1>;
3912 def VMULfd : N3VD<1, 0, 0b00, 0b1101, 1, IIC_VFMULD, "vmul", "f32",
3913 v2f32, v2f32, fmul, 1>;
3914 def VMULfq : N3VQ<1, 0, 0b00, 0b1101, 1, IIC_VFMULQ, "vmul", "f32",
3915 v4f32, v4f32, fmul, 1>;
3916 defm VMULsl : N3VSL_HS<0b1000, "vmul", mul>;
3917 def VMULslfd : N3VDSL<0b10, 0b1001, IIC_VBIND, "vmul", "f32", v2f32, fmul>;
3918 def VMULslfq : N3VQSL<0b10, 0b1001, IIC_VBINQ, "vmul", "f32", v4f32,
3921 def : Pat<(v8i16 (mul (v8i16 QPR:$src1),
3922 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
3923 (v8i16 (VMULslv8i16 (v8i16 QPR:$src1),
3924 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3925 (DSubReg_i16_reg imm:$lane))),
3926 (SubReg_i16_lane imm:$lane)))>;
3927 def : Pat<(v4i32 (mul (v4i32 QPR:$src1),
3928 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
3929 (v4i32 (VMULslv4i32 (v4i32 QPR:$src1),
3930 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3931 (DSubReg_i32_reg imm:$lane))),
3932 (SubReg_i32_lane imm:$lane)))>;
3933 def : Pat<(v4f32 (fmul (v4f32 QPR:$src1),
3934 (v4f32 (NEONvduplane (v4f32 QPR:$src2), imm:$lane)))),
3935 (v4f32 (VMULslfq (v4f32 QPR:$src1),
3936 (v2f32 (EXTRACT_SUBREG QPR:$src2,
3937 (DSubReg_i32_reg imm:$lane))),
3938 (SubReg_i32_lane imm:$lane)))>;
3940 // VQDMULH : Vector Saturating Doubling Multiply Returning High Half
3941 defm VQDMULH : N3VInt_HS<0, 0, 0b1011, 0, N3RegFrm, IIC_VMULi16D, IIC_VMULi32D,
3942 IIC_VMULi16Q, IIC_VMULi32Q,
3943 "vqdmulh", "s", int_arm_neon_vqdmulh, 1>;
3944 defm VQDMULHsl: N3VIntSL_HS<0b1100, IIC_VMULi16D, IIC_VMULi32D,
3945 IIC_VMULi16Q, IIC_VMULi32Q,
3946 "vqdmulh", "s", int_arm_neon_vqdmulh>;
3947 def : Pat<(v8i16 (int_arm_neon_vqdmulh (v8i16 QPR:$src1),
3948 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
3950 (v8i16 (VQDMULHslv8i16 (v8i16 QPR:$src1),
3951 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3952 (DSubReg_i16_reg imm:$lane))),
3953 (SubReg_i16_lane imm:$lane)))>;
3954 def : Pat<(v4i32 (int_arm_neon_vqdmulh (v4i32 QPR:$src1),
3955 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
3957 (v4i32 (VQDMULHslv4i32 (v4i32 QPR:$src1),
3958 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3959 (DSubReg_i32_reg imm:$lane))),
3960 (SubReg_i32_lane imm:$lane)))>;
3962 // VQRDMULH : Vector Rounding Saturating Doubling Multiply Returning High Half
3963 defm VQRDMULH : N3VInt_HS<1, 0, 0b1011, 0, N3RegFrm,
3964 IIC_VMULi16D,IIC_VMULi32D,IIC_VMULi16Q,IIC_VMULi32Q,
3965 "vqrdmulh", "s", int_arm_neon_vqrdmulh, 1>;
3966 defm VQRDMULHsl : N3VIntSL_HS<0b1101, IIC_VMULi16D, IIC_VMULi32D,
3967 IIC_VMULi16Q, IIC_VMULi32Q,
3968 "vqrdmulh", "s", int_arm_neon_vqrdmulh>;
3969 def : Pat<(v8i16 (int_arm_neon_vqrdmulh (v8i16 QPR:$src1),
3970 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
3972 (v8i16 (VQRDMULHslv8i16 (v8i16 QPR:$src1),
3973 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3974 (DSubReg_i16_reg imm:$lane))),
3975 (SubReg_i16_lane imm:$lane)))>;
3976 def : Pat<(v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$src1),
3977 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
3979 (v4i32 (VQRDMULHslv4i32 (v4i32 QPR:$src1),
3980 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3981 (DSubReg_i32_reg imm:$lane))),
3982 (SubReg_i32_lane imm:$lane)))>;
3984 // VMULL : Vector Multiply Long (integer and polynomial) (Q = D * D)
3985 defm VMULLs : N3VL_QHS<0,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
3986 "vmull", "s", NEONvmulls, 1>;
3987 defm VMULLu : N3VL_QHS<1,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
3988 "vmull", "u", NEONvmullu, 1>;
3989 def VMULLp : N3VLInt<0, 1, 0b00, 0b1110, 0, IIC_VMULi16D, "vmull", "p8",
3990 v8i16, v8i8, int_arm_neon_vmullp, 1>;
3991 defm VMULLsls : N3VLSL_HS<0, 0b1010, IIC_VMULi16D, "vmull", "s", NEONvmulls>;
3992 defm VMULLslu : N3VLSL_HS<1, 0b1010, IIC_VMULi16D, "vmull", "u", NEONvmullu>;
3994 // VQDMULL : Vector Saturating Doubling Multiply Long (Q = D * D)
3995 defm VQDMULL : N3VLInt_HS<0,1,0b1101,0, IIC_VMULi16D, IIC_VMULi32D,
3996 "vqdmull", "s", int_arm_neon_vqdmull, 1>;
3997 defm VQDMULLsl: N3VLIntSL_HS<0, 0b1011, IIC_VMULi16D,
3998 "vqdmull", "s", int_arm_neon_vqdmull>;
4000 // Vector Multiply-Accumulate and Multiply-Subtract Operations.
4002 // VMLA : Vector Multiply Accumulate (integer and floating-point)
4003 defm VMLA : N3VMulOp_QHS<0, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
4004 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
4005 def VMLAfd : N3VDMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACD, "vmla", "f32",
4006 v2f32, fmul_su, fadd_mlx>,
4007 Requires<[HasNEON, UseFPVMLx, DontUseFusedMAC]>;
4008 def VMLAfq : N3VQMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACQ, "vmla", "f32",
4009 v4f32, fmul_su, fadd_mlx>,
4010 Requires<[HasNEON, UseFPVMLx, DontUseFusedMAC]>;
4011 defm VMLAsl : N3VMulOpSL_HS<0b0000, IIC_VMACi16D, IIC_VMACi32D,
4012 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
4013 def VMLAslfd : N3VDMulOpSL<0b10, 0b0001, IIC_VMACD, "vmla", "f32",
4014 v2f32, fmul_su, fadd_mlx>,
4015 Requires<[HasNEON, UseFPVMLx]>;
4016 def VMLAslfq : N3VQMulOpSL<0b10, 0b0001, IIC_VMACQ, "vmla", "f32",
4017 v4f32, v2f32, fmul_su, fadd_mlx>,
4018 Requires<[HasNEON, UseFPVMLx]>;
4020 def : Pat<(v8i16 (add (v8i16 QPR:$src1),
4021 (mul (v8i16 QPR:$src2),
4022 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
4023 (v8i16 (VMLAslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
4024 (v4i16 (EXTRACT_SUBREG QPR:$src3,
4025 (DSubReg_i16_reg imm:$lane))),
4026 (SubReg_i16_lane imm:$lane)))>;
4028 def : Pat<(v4i32 (add (v4i32 QPR:$src1),
4029 (mul (v4i32 QPR:$src2),
4030 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
4031 (v4i32 (VMLAslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
4032 (v2i32 (EXTRACT_SUBREG QPR:$src3,
4033 (DSubReg_i32_reg imm:$lane))),
4034 (SubReg_i32_lane imm:$lane)))>;
4036 def : Pat<(v4f32 (fadd_mlx (v4f32 QPR:$src1),
4037 (fmul_su (v4f32 QPR:$src2),
4038 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
4039 (v4f32 (VMLAslfq (v4f32 QPR:$src1),
4041 (v2f32 (EXTRACT_SUBREG QPR:$src3,
4042 (DSubReg_i32_reg imm:$lane))),
4043 (SubReg_i32_lane imm:$lane)))>,
4044 Requires<[HasNEON, UseFPVMLx]>;
4046 // VMLAL : Vector Multiply Accumulate Long (Q += D * D)
4047 defm VMLALs : N3VLMulOp_QHS<0,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
4048 "vmlal", "s", NEONvmulls, add>;
4049 defm VMLALu : N3VLMulOp_QHS<1,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
4050 "vmlal", "u", NEONvmullu, add>;
4052 defm VMLALsls : N3VLMulOpSL_HS<0, 0b0010, "vmlal", "s", NEONvmulls, add>;
4053 defm VMLALslu : N3VLMulOpSL_HS<1, 0b0010, "vmlal", "u", NEONvmullu, add>;
4055 // VQDMLAL : Vector Saturating Doubling Multiply Accumulate Long (Q += D * D)
4056 defm VQDMLAL : N3VLInt3_HS<0, 1, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
4057 "vqdmlal", "s", int_arm_neon_vqdmlal>;
4058 defm VQDMLALsl: N3VLInt3SL_HS<0, 0b0011, "vqdmlal", "s", int_arm_neon_vqdmlal>;
4060 // VMLS : Vector Multiply Subtract (integer and floating-point)
4061 defm VMLS : N3VMulOp_QHS<1, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
4062 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
4063 def VMLSfd : N3VDMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACD, "vmls", "f32",
4064 v2f32, fmul_su, fsub_mlx>,
4065 Requires<[HasNEON, UseFPVMLx, DontUseFusedMAC]>;
4066 def VMLSfq : N3VQMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACQ, "vmls", "f32",
4067 v4f32, fmul_su, fsub_mlx>,
4068 Requires<[HasNEON, UseFPVMLx, DontUseFusedMAC]>;
4069 defm VMLSsl : N3VMulOpSL_HS<0b0100, IIC_VMACi16D, IIC_VMACi32D,
4070 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
4071 def VMLSslfd : N3VDMulOpSL<0b10, 0b0101, IIC_VMACD, "vmls", "f32",
4072 v2f32, fmul_su, fsub_mlx>,
4073 Requires<[HasNEON, UseFPVMLx]>;
4074 def VMLSslfq : N3VQMulOpSL<0b10, 0b0101, IIC_VMACQ, "vmls", "f32",
4075 v4f32, v2f32, fmul_su, fsub_mlx>,
4076 Requires<[HasNEON, UseFPVMLx]>;
4078 def : Pat<(v8i16 (sub (v8i16 QPR:$src1),
4079 (mul (v8i16 QPR:$src2),
4080 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
4081 (v8i16 (VMLSslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
4082 (v4i16 (EXTRACT_SUBREG QPR:$src3,
4083 (DSubReg_i16_reg imm:$lane))),
4084 (SubReg_i16_lane imm:$lane)))>;
4086 def : Pat<(v4i32 (sub (v4i32 QPR:$src1),
4087 (mul (v4i32 QPR:$src2),
4088 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
4089 (v4i32 (VMLSslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
4090 (v2i32 (EXTRACT_SUBREG QPR:$src3,
4091 (DSubReg_i32_reg imm:$lane))),
4092 (SubReg_i32_lane imm:$lane)))>;
4094 def : Pat<(v4f32 (fsub_mlx (v4f32 QPR:$src1),
4095 (fmul_su (v4f32 QPR:$src2),
4096 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
4097 (v4f32 (VMLSslfq (v4f32 QPR:$src1), (v4f32 QPR:$src2),
4098 (v2f32 (EXTRACT_SUBREG QPR:$src3,
4099 (DSubReg_i32_reg imm:$lane))),
4100 (SubReg_i32_lane imm:$lane)))>,
4101 Requires<[HasNEON, UseFPVMLx]>;
4103 // VMLSL : Vector Multiply Subtract Long (Q -= D * D)
4104 defm VMLSLs : N3VLMulOp_QHS<0,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
4105 "vmlsl", "s", NEONvmulls, sub>;
4106 defm VMLSLu : N3VLMulOp_QHS<1,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
4107 "vmlsl", "u", NEONvmullu, sub>;
4109 defm VMLSLsls : N3VLMulOpSL_HS<0, 0b0110, "vmlsl", "s", NEONvmulls, sub>;
4110 defm VMLSLslu : N3VLMulOpSL_HS<1, 0b0110, "vmlsl", "u", NEONvmullu, sub>;
4112 // VQDMLSL : Vector Saturating Doubling Multiply Subtract Long (Q -= D * D)
4113 defm VQDMLSL : N3VLInt3_HS<0, 1, 0b1011, 0, IIC_VMACi16D, IIC_VMACi32D,
4114 "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
4115 defm VQDMLSLsl: N3VLInt3SL_HS<0, 0b111, "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
4117 // Fused Vector Multiply-Accumulate and Fused Multiply-Subtract Operations.
4118 def VFMAfd : N3VDMulOp<0, 0, 0b00, 0b1100, 1, IIC_VFMACD, "vfma", "f32",
4119 v2f32, fmul_su, fadd_mlx>,
4120 Requires<[HasVFP4,UseFusedMAC]>;
4122 def VFMAfq : N3VQMulOp<0, 0, 0b00, 0b1100, 1, IIC_VFMACQ, "vfma", "f32",
4123 v4f32, fmul_su, fadd_mlx>,
4124 Requires<[HasVFP4,UseFusedMAC]>;
4126 // Fused Vector Multiply Subtract (floating-point)
4127 def VFMSfd : N3VDMulOp<0, 0, 0b10, 0b1100, 1, IIC_VFMACD, "vfms", "f32",
4128 v2f32, fmul_su, fsub_mlx>,
4129 Requires<[HasVFP4,UseFusedMAC]>;
4130 def VFMSfq : N3VQMulOp<0, 0, 0b10, 0b1100, 1, IIC_VFMACQ, "vfms", "f32",
4131 v4f32, fmul_su, fsub_mlx>,
4132 Requires<[HasVFP4,UseFusedMAC]>;
4134 // Match @llvm.fma.* intrinsics
4135 def : Pat<(v2f32 (fma DPR:$src1, DPR:$Vn, DPR:$Vm)),
4136 (VFMAfd DPR:$src1, DPR:$Vn, DPR:$Vm)>,
4137 Requires<[HasVFP4]>;
4138 def : Pat<(v4f32 (fma QPR:$src1, QPR:$Vn, QPR:$Vm)),
4139 (VFMAfq QPR:$src1, QPR:$Vn, QPR:$Vm)>,
4140 Requires<[HasVFP4]>;
4141 def : Pat<(v2f32 (fma (fneg DPR:$src1), DPR:$Vn, DPR:$Vm)),
4142 (VFMSfd DPR:$src1, DPR:$Vn, DPR:$Vm)>,
4143 Requires<[HasVFP4]>;
4144 def : Pat<(v4f32 (fma (fneg QPR:$src1), QPR:$Vn, QPR:$Vm)),
4145 (VFMSfq QPR:$src1, QPR:$Vn, QPR:$Vm)>,
4146 Requires<[HasVFP4]>;
4148 // Vector Subtract Operations.
4150 // VSUB : Vector Subtract (integer and floating-point)
4151 defm VSUB : N3V_QHSD<1, 0, 0b1000, 0, IIC_VSUBiD, IIC_VSUBiQ,
4152 "vsub", "i", sub, 0>;
4153 def VSUBfd : N3VD<0, 0, 0b10, 0b1101, 0, IIC_VBIND, "vsub", "f32",
4154 v2f32, v2f32, fsub, 0>;
4155 def VSUBfq : N3VQ<0, 0, 0b10, 0b1101, 0, IIC_VBINQ, "vsub", "f32",
4156 v4f32, v4f32, fsub, 0>;
4157 // VSUBL : Vector Subtract Long (Q = D - D)
4158 defm VSUBLs : N3VLExt_QHS<0,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
4159 "vsubl", "s", sub, sext, 0>;
4160 defm VSUBLu : N3VLExt_QHS<1,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
4161 "vsubl", "u", sub, zext, 0>;
4162 // VSUBW : Vector Subtract Wide (Q = Q - D)
4163 defm VSUBWs : N3VW_QHS<0,1,0b0011,0, "vsubw", "s", sub, sext, 0>;
4164 defm VSUBWu : N3VW_QHS<1,1,0b0011,0, "vsubw", "u", sub, zext, 0>;
4165 // VHSUB : Vector Halving Subtract
4166 defm VHSUBs : N3VInt_QHS<0, 0, 0b0010, 0, N3RegFrm,
4167 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4168 "vhsub", "s", int_arm_neon_vhsubs, 0>;
4169 defm VHSUBu : N3VInt_QHS<1, 0, 0b0010, 0, N3RegFrm,
4170 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4171 "vhsub", "u", int_arm_neon_vhsubu, 0>;
4172 // VQSUB : Vector Saturing Subtract
4173 defm VQSUBs : N3VInt_QHSD<0, 0, 0b0010, 1, N3RegFrm,
4174 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4175 "vqsub", "s", int_arm_neon_vqsubs, 0>;
4176 defm VQSUBu : N3VInt_QHSD<1, 0, 0b0010, 1, N3RegFrm,
4177 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4178 "vqsub", "u", int_arm_neon_vqsubu, 0>;
4179 // VSUBHN : Vector Subtract and Narrow Returning High Half (D = Q - Q)
4180 defm VSUBHN : N3VNInt_HSD<0,1,0b0110,0, "vsubhn", "i",
4181 int_arm_neon_vsubhn, 0>;
4182 // VRSUBHN : Vector Rounding Subtract and Narrow Returning High Half (D=Q-Q)
4183 defm VRSUBHN : N3VNInt_HSD<1,1,0b0110,0, "vrsubhn", "i",
4184 int_arm_neon_vrsubhn, 0>;
4186 // Vector Comparisons.
4188 // VCEQ : Vector Compare Equal
4189 defm VCEQ : N3V_QHS<1, 0, 0b1000, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
4190 IIC_VSUBi4Q, "vceq", "i", NEONvceq, 1>;
4191 def VCEQfd : N3VD<0,0,0b00,0b1110,0, IIC_VBIND, "vceq", "f32", v2i32, v2f32,
4193 def VCEQfq : N3VQ<0,0,0b00,0b1110,0, IIC_VBINQ, "vceq", "f32", v4i32, v4f32,
4196 defm VCEQz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00010, 0, "vceq", "i",
4197 "$Vd, $Vm, #0", NEONvceqz>;
4199 // VCGE : Vector Compare Greater Than or Equal
4200 defm VCGEs : N3V_QHS<0, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
4201 IIC_VSUBi4Q, "vcge", "s", NEONvcge, 0>;
4202 defm VCGEu : N3V_QHS<1, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
4203 IIC_VSUBi4Q, "vcge", "u", NEONvcgeu, 0>;
4204 def VCGEfd : N3VD<1,0,0b00,0b1110,0, IIC_VBIND, "vcge", "f32", v2i32, v2f32,
4206 def VCGEfq : N3VQ<1,0,0b00,0b1110,0, IIC_VBINQ, "vcge", "f32", v4i32, v4f32,
4209 defm VCGEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00001, 0, "vcge", "s",
4210 "$Vd, $Vm, #0", NEONvcgez>;
4211 defm VCLEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00011, 0, "vcle", "s",
4212 "$Vd, $Vm, #0", NEONvclez>;
4214 // VCGT : Vector Compare Greater Than
4215 defm VCGTs : N3V_QHS<0, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
4216 IIC_VSUBi4Q, "vcgt", "s", NEONvcgt, 0>;
4217 defm VCGTu : N3V_QHS<1, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
4218 IIC_VSUBi4Q, "vcgt", "u", NEONvcgtu, 0>;
4219 def VCGTfd : N3VD<1,0,0b10,0b1110,0, IIC_VBIND, "vcgt", "f32", v2i32, v2f32,
4221 def VCGTfq : N3VQ<1,0,0b10,0b1110,0, IIC_VBINQ, "vcgt", "f32", v4i32, v4f32,
4224 defm VCGTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00000, 0, "vcgt", "s",
4225 "$Vd, $Vm, #0", NEONvcgtz>;
4226 defm VCLTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00100, 0, "vclt", "s",
4227 "$Vd, $Vm, #0", NEONvcltz>;
4229 // VACGE : Vector Absolute Compare Greater Than or Equal (aka VCAGE)
4230 def VACGEd : N3VDInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacge",
4231 "f32", v2i32, v2f32, int_arm_neon_vacged, 0>;
4232 def VACGEq : N3VQInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacge",
4233 "f32", v4i32, v4f32, int_arm_neon_vacgeq, 0>;
4234 // VACGT : Vector Absolute Compare Greater Than (aka VCAGT)
4235 def VACGTd : N3VDInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacgt",
4236 "f32", v2i32, v2f32, int_arm_neon_vacgtd, 0>;
4237 def VACGTq : N3VQInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacgt",
4238 "f32", v4i32, v4f32, int_arm_neon_vacgtq, 0>;
4239 // VTST : Vector Test Bits
4240 defm VTST : N3V_QHS<0, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
4241 IIC_VBINi4Q, "vtst", "", NEONvtst, 1>;
4243 // Vector Bitwise Operations.
4245 def vnotd : PatFrag<(ops node:$in),
4246 (xor node:$in, (bitconvert (v8i8 NEONimmAllOnesV)))>;
4247 def vnotq : PatFrag<(ops node:$in),
4248 (xor node:$in, (bitconvert (v16i8 NEONimmAllOnesV)))>;
4251 // VAND : Vector Bitwise AND
4252 def VANDd : N3VDX<0, 0, 0b00, 0b0001, 1, IIC_VBINiD, "vand",
4253 v2i32, v2i32, and, 1>;
4254 def VANDq : N3VQX<0, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "vand",
4255 v4i32, v4i32, and, 1>;
4257 // VEOR : Vector Bitwise Exclusive OR
4258 def VEORd : N3VDX<1, 0, 0b00, 0b0001, 1, IIC_VBINiD, "veor",
4259 v2i32, v2i32, xor, 1>;
4260 def VEORq : N3VQX<1, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "veor",
4261 v4i32, v4i32, xor, 1>;
4263 // VORR : Vector Bitwise OR
4264 def VORRd : N3VDX<0, 0, 0b10, 0b0001, 1, IIC_VBINiD, "vorr",
4265 v2i32, v2i32, or, 1>;
4266 def VORRq : N3VQX<0, 0, 0b10, 0b0001, 1, IIC_VBINiQ, "vorr",
4267 v4i32, v4i32, or, 1>;
4269 def VORRiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 0, 1,
4270 (outs DPR:$Vd), (ins nImmSplatI16:$SIMM, DPR:$src),
4272 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
4274 (v4i16 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
4275 let Inst{9} = SIMM{9};
4278 def VORRiv2i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 0, 0, 1,
4279 (outs DPR:$Vd), (ins nImmSplatI32:$SIMM, DPR:$src),
4281 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
4283 (v2i32 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
4284 let Inst{10-9} = SIMM{10-9};
4287 def VORRiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 0, 1,
4288 (outs QPR:$Vd), (ins nImmSplatI16:$SIMM, QPR:$src),
4290 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
4292 (v8i16 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
4293 let Inst{9} = SIMM{9};
4296 def VORRiv4i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 1, 0, 1,
4297 (outs QPR:$Vd), (ins nImmSplatI32:$SIMM, QPR:$src),
4299 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
4301 (v4i32 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
4302 let Inst{10-9} = SIMM{10-9};
4306 // VBIC : Vector Bitwise Bit Clear (AND NOT)
4307 def VBICd : N3VX<0, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
4308 (ins DPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VBINiD,
4309 "vbic", "$Vd, $Vn, $Vm", "",
4310 [(set DPR:$Vd, (v2i32 (and DPR:$Vn,
4311 (vnotd DPR:$Vm))))]>;
4312 def VBICq : N3VX<0, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
4313 (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINiQ,
4314 "vbic", "$Vd, $Vn, $Vm", "",
4315 [(set QPR:$Vd, (v4i32 (and QPR:$Vn,
4316 (vnotq QPR:$Vm))))]>;
4318 def VBICiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 1, 1,
4319 (outs DPR:$Vd), (ins nImmSplatI16:$SIMM, DPR:$src),
4321 "vbic", "i16", "$Vd, $SIMM", "$src = $Vd",
4323 (v4i16 (NEONvbicImm DPR:$src, timm:$SIMM)))]> {
4324 let Inst{9} = SIMM{9};
4327 def VBICiv2i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 0, 1, 1,
4328 (outs DPR:$Vd), (ins nImmSplatI32:$SIMM, DPR:$src),
4330 "vbic", "i32", "$Vd, $SIMM", "$src = $Vd",
4332 (v2i32 (NEONvbicImm DPR:$src, timm:$SIMM)))]> {
4333 let Inst{10-9} = SIMM{10-9};
4336 def VBICiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 1, 1,
4337 (outs QPR:$Vd), (ins nImmSplatI16:$SIMM, QPR:$src),
4339 "vbic", "i16", "$Vd, $SIMM", "$src = $Vd",
4341 (v8i16 (NEONvbicImm QPR:$src, timm:$SIMM)))]> {
4342 let Inst{9} = SIMM{9};
4345 def VBICiv4i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 1, 1, 1,
4346 (outs QPR:$Vd), (ins nImmSplatI32:$SIMM, QPR:$src),
4348 "vbic", "i32", "$Vd, $SIMM", "$src = $Vd",
4350 (v4i32 (NEONvbicImm QPR:$src, timm:$SIMM)))]> {
4351 let Inst{10-9} = SIMM{10-9};
4354 // VORN : Vector Bitwise OR NOT
4355 def VORNd : N3VX<0, 0, 0b11, 0b0001, 0, 1, (outs DPR:$Vd),
4356 (ins DPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VBINiD,
4357 "vorn", "$Vd, $Vn, $Vm", "",
4358 [(set DPR:$Vd, (v2i32 (or DPR:$Vn,
4359 (vnotd DPR:$Vm))))]>;
4360 def VORNq : N3VX<0, 0, 0b11, 0b0001, 1, 1, (outs QPR:$Vd),
4361 (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINiQ,
4362 "vorn", "$Vd, $Vn, $Vm", "",
4363 [(set QPR:$Vd, (v4i32 (or QPR:$Vn,
4364 (vnotq QPR:$Vm))))]>;
4366 // VMVN : Vector Bitwise NOT (Immediate)
4368 let isReMaterializable = 1 in {
4370 def VMVNv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 1, 1, (outs DPR:$Vd),
4371 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4372 "vmvn", "i16", "$Vd, $SIMM", "",
4373 [(set DPR:$Vd, (v4i16 (NEONvmvnImm timm:$SIMM)))]> {
4374 let Inst{9} = SIMM{9};
4377 def VMVNv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 1, 1, (outs QPR:$Vd),
4378 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4379 "vmvn", "i16", "$Vd, $SIMM", "",
4380 [(set QPR:$Vd, (v8i16 (NEONvmvnImm timm:$SIMM)))]> {
4381 let Inst{9} = SIMM{9};
4384 def VMVNv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 1, 1, (outs DPR:$Vd),
4385 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4386 "vmvn", "i32", "$Vd, $SIMM", "",
4387 [(set DPR:$Vd, (v2i32 (NEONvmvnImm timm:$SIMM)))]> {
4388 let Inst{11-8} = SIMM{11-8};
4391 def VMVNv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 1, 1, (outs QPR:$Vd),
4392 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4393 "vmvn", "i32", "$Vd, $SIMM", "",
4394 [(set QPR:$Vd, (v4i32 (NEONvmvnImm timm:$SIMM)))]> {
4395 let Inst{11-8} = SIMM{11-8};
4399 // VMVN : Vector Bitwise NOT
4400 def VMVNd : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 0, 0,
4401 (outs DPR:$Vd), (ins DPR:$Vm), IIC_VSUBiD,
4402 "vmvn", "$Vd, $Vm", "",
4403 [(set DPR:$Vd, (v2i32 (vnotd DPR:$Vm)))]>;
4404 def VMVNq : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 1, 0,
4405 (outs QPR:$Vd), (ins QPR:$Vm), IIC_VSUBiD,
4406 "vmvn", "$Vd, $Vm", "",
4407 [(set QPR:$Vd, (v4i32 (vnotq QPR:$Vm)))]>;
4408 def : Pat<(v2i32 (vnotd DPR:$src)), (VMVNd DPR:$src)>;
4409 def : Pat<(v4i32 (vnotq QPR:$src)), (VMVNq QPR:$src)>;
4411 // VBSL : Vector Bitwise Select
4412 def VBSLd : N3VX<1, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
4413 (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
4414 N3RegFrm, IIC_VCNTiD,
4415 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4417 (v2i32 (NEONvbsl DPR:$src1, DPR:$Vn, DPR:$Vm)))]>;
4419 def : Pat<(v2i32 (or (and DPR:$Vn, DPR:$Vd),
4420 (and DPR:$Vm, (vnotd DPR:$Vd)))),
4421 (VBSLd DPR:$Vd, DPR:$Vn, DPR:$Vm)>;
4423 def VBSLq : N3VX<1, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
4424 (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
4425 N3RegFrm, IIC_VCNTiQ,
4426 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4428 (v4i32 (NEONvbsl QPR:$src1, QPR:$Vn, QPR:$Vm)))]>;
4430 def : Pat<(v4i32 (or (and QPR:$Vn, QPR:$Vd),
4431 (and QPR:$Vm, (vnotq QPR:$Vd)))),
4432 (VBSLq QPR:$Vd, QPR:$Vn, QPR:$Vm)>;
4434 // VBIF : Vector Bitwise Insert if False
4435 // like VBSL but with: "vbif $dst, $src3, $src1", "$src2 = $dst",
4436 // FIXME: This instruction's encoding MAY NOT BE correct.
4437 def VBIFd : N3VX<1, 0, 0b11, 0b0001, 0, 1,
4438 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
4439 N3RegFrm, IIC_VBINiD,
4440 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4442 def VBIFq : N3VX<1, 0, 0b11, 0b0001, 1, 1,
4443 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
4444 N3RegFrm, IIC_VBINiQ,
4445 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4448 // VBIT : Vector Bitwise Insert if True
4449 // like VBSL but with: "vbit $dst, $src2, $src1", "$src3 = $dst",
4450 // FIXME: This instruction's encoding MAY NOT BE correct.
4451 def VBITd : N3VX<1, 0, 0b10, 0b0001, 0, 1,
4452 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
4453 N3RegFrm, IIC_VBINiD,
4454 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4456 def VBITq : N3VX<1, 0, 0b10, 0b0001, 1, 1,
4457 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
4458 N3RegFrm, IIC_VBINiQ,
4459 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4462 // VBIT/VBIF are not yet implemented. The TwoAddress pass will not go looking
4463 // for equivalent operations with different register constraints; it just
4466 // Vector Absolute Differences.
4468 // VABD : Vector Absolute Difference
4469 defm VABDs : N3VInt_QHS<0, 0, 0b0111, 0, N3RegFrm,
4470 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4471 "vabd", "s", int_arm_neon_vabds, 1>;
4472 defm VABDu : N3VInt_QHS<1, 0, 0b0111, 0, N3RegFrm,
4473 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4474 "vabd", "u", int_arm_neon_vabdu, 1>;
4475 def VABDfd : N3VDInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBIND,
4476 "vabd", "f32", v2f32, v2f32, int_arm_neon_vabds, 1>;
4477 def VABDfq : N3VQInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBINQ,
4478 "vabd", "f32", v4f32, v4f32, int_arm_neon_vabds, 1>;
4480 // VABDL : Vector Absolute Difference Long (Q = | D - D |)
4481 defm VABDLs : N3VLIntExt_QHS<0,1,0b0111,0, IIC_VSUBi4Q,
4482 "vabdl", "s", int_arm_neon_vabds, zext, 1>;
4483 defm VABDLu : N3VLIntExt_QHS<1,1,0b0111,0, IIC_VSUBi4Q,
4484 "vabdl", "u", int_arm_neon_vabdu, zext, 1>;
4486 // VABA : Vector Absolute Difference and Accumulate
4487 defm VABAs : N3VIntOp_QHS<0,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
4488 "vaba", "s", int_arm_neon_vabds, add>;
4489 defm VABAu : N3VIntOp_QHS<1,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
4490 "vaba", "u", int_arm_neon_vabdu, add>;
4492 // VABAL : Vector Absolute Difference and Accumulate Long (Q += | D - D |)
4493 defm VABALs : N3VLIntExtOp_QHS<0,1,0b0101,0, IIC_VABAD,
4494 "vabal", "s", int_arm_neon_vabds, zext, add>;
4495 defm VABALu : N3VLIntExtOp_QHS<1,1,0b0101,0, IIC_VABAD,
4496 "vabal", "u", int_arm_neon_vabdu, zext, add>;
4498 // Vector Maximum and Minimum.
4500 // VMAX : Vector Maximum
4501 defm VMAXs : N3VInt_QHS<0, 0, 0b0110, 0, N3RegFrm,
4502 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4503 "vmax", "s", int_arm_neon_vmaxs, 1>;
4504 defm VMAXu : N3VInt_QHS<1, 0, 0b0110, 0, N3RegFrm,
4505 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4506 "vmax", "u", int_arm_neon_vmaxu, 1>;
4507 def VMAXfd : N3VDInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBIND,
4509 v2f32, v2f32, int_arm_neon_vmaxs, 1>;
4510 def VMAXfq : N3VQInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBINQ,
4512 v4f32, v4f32, int_arm_neon_vmaxs, 1>;
4514 // VMIN : Vector Minimum
4515 defm VMINs : N3VInt_QHS<0, 0, 0b0110, 1, N3RegFrm,
4516 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4517 "vmin", "s", int_arm_neon_vmins, 1>;
4518 defm VMINu : N3VInt_QHS<1, 0, 0b0110, 1, N3RegFrm,
4519 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4520 "vmin", "u", int_arm_neon_vminu, 1>;
4521 def VMINfd : N3VDInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBIND,
4523 v2f32, v2f32, int_arm_neon_vmins, 1>;
4524 def VMINfq : N3VQInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBINQ,
4526 v4f32, v4f32, int_arm_neon_vmins, 1>;
4528 // Vector Pairwise Operations.
4530 // VPADD : Vector Pairwise Add
4531 def VPADDi8 : N3VDInt<0, 0, 0b00, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4533 v8i8, v8i8, int_arm_neon_vpadd, 0>;
4534 def VPADDi16 : N3VDInt<0, 0, 0b01, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4536 v4i16, v4i16, int_arm_neon_vpadd, 0>;
4537 def VPADDi32 : N3VDInt<0, 0, 0b10, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4539 v2i32, v2i32, int_arm_neon_vpadd, 0>;
4540 def VPADDf : N3VDInt<1, 0, 0b00, 0b1101, 0, N3RegFrm,
4541 IIC_VPBIND, "vpadd", "f32",
4542 v2f32, v2f32, int_arm_neon_vpadd, 0>;
4544 // VPADDL : Vector Pairwise Add Long
4545 defm VPADDLs : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00100, 0, "vpaddl", "s",
4546 int_arm_neon_vpaddls>;
4547 defm VPADDLu : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00101, 0, "vpaddl", "u",
4548 int_arm_neon_vpaddlu>;
4550 // VPADAL : Vector Pairwise Add and Accumulate Long
4551 defm VPADALs : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01100, 0, "vpadal", "s",
4552 int_arm_neon_vpadals>;
4553 defm VPADALu : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01101, 0, "vpadal", "u",
4554 int_arm_neon_vpadalu>;
4556 // VPMAX : Vector Pairwise Maximum
4557 def VPMAXs8 : N3VDInt<0, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4558 "s8", v8i8, v8i8, int_arm_neon_vpmaxs, 0>;
4559 def VPMAXs16 : N3VDInt<0, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4560 "s16", v4i16, v4i16, int_arm_neon_vpmaxs, 0>;
4561 def VPMAXs32 : N3VDInt<0, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4562 "s32", v2i32, v2i32, int_arm_neon_vpmaxs, 0>;
4563 def VPMAXu8 : N3VDInt<1, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4564 "u8", v8i8, v8i8, int_arm_neon_vpmaxu, 0>;
4565 def VPMAXu16 : N3VDInt<1, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4566 "u16", v4i16, v4i16, int_arm_neon_vpmaxu, 0>;
4567 def VPMAXu32 : N3VDInt<1, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4568 "u32", v2i32, v2i32, int_arm_neon_vpmaxu, 0>;
4569 def VPMAXf : N3VDInt<1, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmax",
4570 "f32", v2f32, v2f32, int_arm_neon_vpmaxs, 0>;
4572 // VPMIN : Vector Pairwise Minimum
4573 def VPMINs8 : N3VDInt<0, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4574 "s8", v8i8, v8i8, int_arm_neon_vpmins, 0>;
4575 def VPMINs16 : N3VDInt<0, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4576 "s16", v4i16, v4i16, int_arm_neon_vpmins, 0>;
4577 def VPMINs32 : N3VDInt<0, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4578 "s32", v2i32, v2i32, int_arm_neon_vpmins, 0>;
4579 def VPMINu8 : N3VDInt<1, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4580 "u8", v8i8, v8i8, int_arm_neon_vpminu, 0>;
4581 def VPMINu16 : N3VDInt<1, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4582 "u16", v4i16, v4i16, int_arm_neon_vpminu, 0>;
4583 def VPMINu32 : N3VDInt<1, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4584 "u32", v2i32, v2i32, int_arm_neon_vpminu, 0>;
4585 def VPMINf : N3VDInt<1, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmin",
4586 "f32", v2f32, v2f32, int_arm_neon_vpmins, 0>;
4588 // Vector Reciprocal and Reciprocal Square Root Estimate and Step.
4590 // VRECPE : Vector Reciprocal Estimate
4591 def VRECPEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
4592 IIC_VUNAD, "vrecpe", "u32",
4593 v2i32, v2i32, int_arm_neon_vrecpe>;
4594 def VRECPEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
4595 IIC_VUNAQ, "vrecpe", "u32",
4596 v4i32, v4i32, int_arm_neon_vrecpe>;
4597 def VRECPEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
4598 IIC_VUNAD, "vrecpe", "f32",
4599 v2f32, v2f32, int_arm_neon_vrecpe>;
4600 def VRECPEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
4601 IIC_VUNAQ, "vrecpe", "f32",
4602 v4f32, v4f32, int_arm_neon_vrecpe>;
4604 // VRECPS : Vector Reciprocal Step
4605 def VRECPSfd : N3VDInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
4606 IIC_VRECSD, "vrecps", "f32",
4607 v2f32, v2f32, int_arm_neon_vrecps, 1>;
4608 def VRECPSfq : N3VQInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
4609 IIC_VRECSQ, "vrecps", "f32",
4610 v4f32, v4f32, int_arm_neon_vrecps, 1>;
4612 // VRSQRTE : Vector Reciprocal Square Root Estimate
4613 def VRSQRTEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
4614 IIC_VUNAD, "vrsqrte", "u32",
4615 v2i32, v2i32, int_arm_neon_vrsqrte>;
4616 def VRSQRTEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
4617 IIC_VUNAQ, "vrsqrte", "u32",
4618 v4i32, v4i32, int_arm_neon_vrsqrte>;
4619 def VRSQRTEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
4620 IIC_VUNAD, "vrsqrte", "f32",
4621 v2f32, v2f32, int_arm_neon_vrsqrte>;
4622 def VRSQRTEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
4623 IIC_VUNAQ, "vrsqrte", "f32",
4624 v4f32, v4f32, int_arm_neon_vrsqrte>;
4626 // VRSQRTS : Vector Reciprocal Square Root Step
4627 def VRSQRTSfd : N3VDInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
4628 IIC_VRECSD, "vrsqrts", "f32",
4629 v2f32, v2f32, int_arm_neon_vrsqrts, 1>;
4630 def VRSQRTSfq : N3VQInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
4631 IIC_VRECSQ, "vrsqrts", "f32",
4632 v4f32, v4f32, int_arm_neon_vrsqrts, 1>;
4636 // VSHL : Vector Shift
4637 defm VSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 0, N3RegVShFrm,
4638 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
4639 "vshl", "s", int_arm_neon_vshifts>;
4640 defm VSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 0, N3RegVShFrm,
4641 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
4642 "vshl", "u", int_arm_neon_vshiftu>;
4644 // VSHL : Vector Shift Left (Immediate)
4645 defm VSHLi : N2VShL_QHSD<0, 1, 0b0101, 1, IIC_VSHLiD, "vshl", "i", NEONvshl>;
4647 // VSHR : Vector Shift Right (Immediate)
4648 defm VSHRs : N2VShR_QHSD<0, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "s", "VSHRs",
4650 defm VSHRu : N2VShR_QHSD<1, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "u", "VSHRu",
4653 // VSHLL : Vector Shift Left Long
4654 defm VSHLLs : N2VLSh_QHS<0, 1, 0b1010, 0, 0, 1, "vshll", "s", NEONvshlls>;
4655 defm VSHLLu : N2VLSh_QHS<1, 1, 0b1010, 0, 0, 1, "vshll", "u", NEONvshllu>;
4657 // VSHLL : Vector Shift Left Long (with maximum shift count)
4658 class N2VLShMax<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
4659 bit op6, bit op4, string OpcodeStr, string Dt, ValueType ResTy,
4660 ValueType OpTy, Operand ImmTy, SDNode OpNode>
4661 : N2VLSh<op24, op23, op11_8, op7, op6, op4, OpcodeStr, Dt,
4662 ResTy, OpTy, ImmTy, OpNode> {
4663 let Inst{21-16} = op21_16;
4664 let DecoderMethod = "DecodeVSHLMaxInstruction";
4666 def VSHLLi8 : N2VLShMax<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll", "i8",
4667 v8i16, v8i8, imm8, NEONvshlli>;
4668 def VSHLLi16 : N2VLShMax<1, 1, 0b110110, 0b0011, 0, 0, 0, "vshll", "i16",
4669 v4i32, v4i16, imm16, NEONvshlli>;
4670 def VSHLLi32 : N2VLShMax<1, 1, 0b111010, 0b0011, 0, 0, 0, "vshll", "i32",
4671 v2i64, v2i32, imm32, NEONvshlli>;
4673 // VSHRN : Vector Shift Right and Narrow
4674 defm VSHRN : N2VNSh_HSD<0,1,0b1000,0,0,1, IIC_VSHLiD, "vshrn", "i",
4677 // VRSHL : Vector Rounding Shift
4678 defm VRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 0, N3RegVShFrm,
4679 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4680 "vrshl", "s", int_arm_neon_vrshifts>;
4681 defm VRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 0, N3RegVShFrm,
4682 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4683 "vrshl", "u", int_arm_neon_vrshiftu>;
4684 // VRSHR : Vector Rounding Shift Right
4685 defm VRSHRs : N2VShR_QHSD<0,1,0b0010,1, IIC_VSHLi4D, "vrshr", "s", "VRSHRs",
4687 defm VRSHRu : N2VShR_QHSD<1,1,0b0010,1, IIC_VSHLi4D, "vrshr", "u", "VRSHRu",
4690 // VRSHRN : Vector Rounding Shift Right and Narrow
4691 defm VRSHRN : N2VNSh_HSD<0, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vrshrn", "i",
4694 // VQSHL : Vector Saturating Shift
4695 defm VQSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 1, N3RegVShFrm,
4696 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4697 "vqshl", "s", int_arm_neon_vqshifts>;
4698 defm VQSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 1, N3RegVShFrm,
4699 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4700 "vqshl", "u", int_arm_neon_vqshiftu>;
4701 // VQSHL : Vector Saturating Shift Left (Immediate)
4702 defm VQSHLsi : N2VShL_QHSD<0,1,0b0111,1, IIC_VSHLi4D, "vqshl", "s",NEONvqshls>;
4703 defm VQSHLui : N2VShL_QHSD<1,1,0b0111,1, IIC_VSHLi4D, "vqshl", "u",NEONvqshlu>;
4705 // VQSHLU : Vector Saturating Shift Left (Immediate, Unsigned)
4706 defm VQSHLsu : N2VShL_QHSD<1,1,0b0110,1, IIC_VSHLi4D,"vqshlu","s",NEONvqshlsu>;
4708 // VQSHRN : Vector Saturating Shift Right and Narrow
4709 defm VQSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "s",
4711 defm VQSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "u",
4714 // VQSHRUN : Vector Saturating Shift Right and Narrow (Unsigned)
4715 defm VQSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 0, 1, IIC_VSHLi4D, "vqshrun", "s",
4718 // VQRSHL : Vector Saturating Rounding Shift
4719 defm VQRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 1, N3RegVShFrm,
4720 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4721 "vqrshl", "s", int_arm_neon_vqrshifts>;
4722 defm VQRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 1, N3RegVShFrm,
4723 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4724 "vqrshl", "u", int_arm_neon_vqrshiftu>;
4726 // VQRSHRN : Vector Saturating Rounding Shift Right and Narrow
4727 defm VQRSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "s",
4729 defm VQRSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "u",
4732 // VQRSHRUN : Vector Saturating Rounding Shift Right and Narrow (Unsigned)
4733 defm VQRSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vqrshrun", "s",
4736 // VSRA : Vector Shift Right and Accumulate
4737 defm VSRAs : N2VShAdd_QHSD<0, 1, 0b0001, 1, "vsra", "s", NEONvshrs>;
4738 defm VSRAu : N2VShAdd_QHSD<1, 1, 0b0001, 1, "vsra", "u", NEONvshru>;
4739 // VRSRA : Vector Rounding Shift Right and Accumulate
4740 defm VRSRAs : N2VShAdd_QHSD<0, 1, 0b0011, 1, "vrsra", "s", NEONvrshrs>;
4741 defm VRSRAu : N2VShAdd_QHSD<1, 1, 0b0011, 1, "vrsra", "u", NEONvrshru>;
4743 // VSLI : Vector Shift Left and Insert
4744 defm VSLI : N2VShInsL_QHSD<1, 1, 0b0101, 1, "vsli">;
4746 // VSRI : Vector Shift Right and Insert
4747 defm VSRI : N2VShInsR_QHSD<1, 1, 0b0100, 1, "vsri">;
4749 // Vector Absolute and Saturating Absolute.
4751 // VABS : Vector Absolute Value
4752 defm VABS : N2VInt_QHS<0b11, 0b11, 0b01, 0b00110, 0,
4753 IIC_VUNAiD, IIC_VUNAiQ, "vabs", "s",
4755 def VABSfd : N2VDInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
4756 IIC_VUNAD, "vabs", "f32",
4757 v2f32, v2f32, int_arm_neon_vabs>;
4758 def VABSfq : N2VQInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
4759 IIC_VUNAQ, "vabs", "f32",
4760 v4f32, v4f32, int_arm_neon_vabs>;
4762 // VQABS : Vector Saturating Absolute Value
4763 defm VQABS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01110, 0,
4764 IIC_VQUNAiD, IIC_VQUNAiQ, "vqabs", "s",
4765 int_arm_neon_vqabs>;
4769 def vnegd : PatFrag<(ops node:$in),
4770 (sub (bitconvert (v2i32 NEONimmAllZerosV)), node:$in)>;
4771 def vnegq : PatFrag<(ops node:$in),
4772 (sub (bitconvert (v4i32 NEONimmAllZerosV)), node:$in)>;
4774 class VNEGD<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
4775 : N2V<0b11, 0b11, size, 0b01, 0b00111, 0, 0, (outs DPR:$Vd), (ins DPR:$Vm),
4776 IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
4777 [(set DPR:$Vd, (Ty (vnegd DPR:$Vm)))]>;
4778 class VNEGQ<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
4779 : N2V<0b11, 0b11, size, 0b01, 0b00111, 1, 0, (outs QPR:$Vd), (ins QPR:$Vm),
4780 IIC_VSHLiQ, OpcodeStr, Dt, "$Vd, $Vm", "",
4781 [(set QPR:$Vd, (Ty (vnegq QPR:$Vm)))]>;
4783 // VNEG : Vector Negate (integer)
4784 def VNEGs8d : VNEGD<0b00, "vneg", "s8", v8i8>;
4785 def VNEGs16d : VNEGD<0b01, "vneg", "s16", v4i16>;
4786 def VNEGs32d : VNEGD<0b10, "vneg", "s32", v2i32>;
4787 def VNEGs8q : VNEGQ<0b00, "vneg", "s8", v16i8>;
4788 def VNEGs16q : VNEGQ<0b01, "vneg", "s16", v8i16>;
4789 def VNEGs32q : VNEGQ<0b10, "vneg", "s32", v4i32>;
4791 // VNEG : Vector Negate (floating-point)
4792 def VNEGfd : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
4793 (outs DPR:$Vd), (ins DPR:$Vm), IIC_VUNAD,
4794 "vneg", "f32", "$Vd, $Vm", "",
4795 [(set DPR:$Vd, (v2f32 (fneg DPR:$Vm)))]>;
4796 def VNEGf32q : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 1, 0,
4797 (outs QPR:$Vd), (ins QPR:$Vm), IIC_VUNAQ,
4798 "vneg", "f32", "$Vd, $Vm", "",
4799 [(set QPR:$Vd, (v4f32 (fneg QPR:$Vm)))]>;
4801 def : Pat<(v8i8 (vnegd DPR:$src)), (VNEGs8d DPR:$src)>;
4802 def : Pat<(v4i16 (vnegd DPR:$src)), (VNEGs16d DPR:$src)>;
4803 def : Pat<(v2i32 (vnegd DPR:$src)), (VNEGs32d DPR:$src)>;
4804 def : Pat<(v16i8 (vnegq QPR:$src)), (VNEGs8q QPR:$src)>;
4805 def : Pat<(v8i16 (vnegq QPR:$src)), (VNEGs16q QPR:$src)>;
4806 def : Pat<(v4i32 (vnegq QPR:$src)), (VNEGs32q QPR:$src)>;
4808 // VQNEG : Vector Saturating Negate
4809 defm VQNEG : N2VInt_QHS<0b11, 0b11, 0b00, 0b01111, 0,
4810 IIC_VQUNAiD, IIC_VQUNAiQ, "vqneg", "s",
4811 int_arm_neon_vqneg>;
4813 // Vector Bit Counting Operations.
4815 // VCLS : Vector Count Leading Sign Bits
4816 defm VCLS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01000, 0,
4817 IIC_VCNTiD, IIC_VCNTiQ, "vcls", "s",
4819 // VCLZ : Vector Count Leading Zeros
4820 defm VCLZ : N2VInt_QHS<0b11, 0b11, 0b00, 0b01001, 0,
4821 IIC_VCNTiD, IIC_VCNTiQ, "vclz", "i",
4823 // VCNT : Vector Count One Bits
4824 def VCNTd : N2VDInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
4825 IIC_VCNTiD, "vcnt", "8",
4826 v8i8, v8i8, int_arm_neon_vcnt>;
4827 def VCNTq : N2VQInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
4828 IIC_VCNTiQ, "vcnt", "8",
4829 v16i8, v16i8, int_arm_neon_vcnt>;
4832 def VSWPd : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 0, 0,
4833 (outs DPR:$Vd, DPR:$Vm), (ins DPR:$in1, DPR:$in2),
4834 NoItinerary, "vswp", "$Vd, $Vm", "$in1 = $Vd, $in2 = $Vm",
4836 def VSWPq : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 1, 0,
4837 (outs QPR:$Vd, QPR:$Vm), (ins QPR:$in1, QPR:$in2),
4838 NoItinerary, "vswp", "$Vd, $Vm", "$in1 = $Vd, $in2 = $Vm",
4841 // Vector Move Operations.
4843 // VMOV : Vector Move (Register)
4844 def : InstAlias<"vmov${p} $Vd, $Vm",
4845 (VORRd DPR:$Vd, DPR:$Vm, DPR:$Vm, pred:$p)>;
4846 def : InstAlias<"vmov${p} $Vd, $Vm",
4847 (VORRq QPR:$Vd, QPR:$Vm, QPR:$Vm, pred:$p)>;
4849 // VMOV : Vector Move (Immediate)
4851 let isReMaterializable = 1 in {
4852 def VMOVv8i8 : N1ModImm<1, 0b000, 0b1110, 0, 0, 0, 1, (outs DPR:$Vd),
4853 (ins nImmSplatI8:$SIMM), IIC_VMOVImm,
4854 "vmov", "i8", "$Vd, $SIMM", "",
4855 [(set DPR:$Vd, (v8i8 (NEONvmovImm timm:$SIMM)))]>;
4856 def VMOVv16i8 : N1ModImm<1, 0b000, 0b1110, 0, 1, 0, 1, (outs QPR:$Vd),
4857 (ins nImmSplatI8:$SIMM), IIC_VMOVImm,
4858 "vmov", "i8", "$Vd, $SIMM", "",
4859 [(set QPR:$Vd, (v16i8 (NEONvmovImm timm:$SIMM)))]>;
4861 def VMOVv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 0, 1, (outs DPR:$Vd),
4862 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4863 "vmov", "i16", "$Vd, $SIMM", "",
4864 [(set DPR:$Vd, (v4i16 (NEONvmovImm timm:$SIMM)))]> {
4865 let Inst{9} = SIMM{9};
4868 def VMOVv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 0, 1, (outs QPR:$Vd),
4869 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4870 "vmov", "i16", "$Vd, $SIMM", "",
4871 [(set QPR:$Vd, (v8i16 (NEONvmovImm timm:$SIMM)))]> {
4872 let Inst{9} = SIMM{9};
4875 def VMOVv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 0, 1, (outs DPR:$Vd),
4876 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4877 "vmov", "i32", "$Vd, $SIMM", "",
4878 [(set DPR:$Vd, (v2i32 (NEONvmovImm timm:$SIMM)))]> {
4879 let Inst{11-8} = SIMM{11-8};
4882 def VMOVv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 0, 1, (outs QPR:$Vd),
4883 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4884 "vmov", "i32", "$Vd, $SIMM", "",
4885 [(set QPR:$Vd, (v4i32 (NEONvmovImm timm:$SIMM)))]> {
4886 let Inst{11-8} = SIMM{11-8};
4889 def VMOVv1i64 : N1ModImm<1, 0b000, 0b1110, 0, 0, 1, 1, (outs DPR:$Vd),
4890 (ins nImmSplatI64:$SIMM), IIC_VMOVImm,
4891 "vmov", "i64", "$Vd, $SIMM", "",
4892 [(set DPR:$Vd, (v1i64 (NEONvmovImm timm:$SIMM)))]>;
4893 def VMOVv2i64 : N1ModImm<1, 0b000, 0b1110, 0, 1, 1, 1, (outs QPR:$Vd),
4894 (ins nImmSplatI64:$SIMM), IIC_VMOVImm,
4895 "vmov", "i64", "$Vd, $SIMM", "",
4896 [(set QPR:$Vd, (v2i64 (NEONvmovImm timm:$SIMM)))]>;
4898 def VMOVv2f32 : N1ModImm<1, 0b000, 0b1111, 0, 0, 0, 1, (outs DPR:$Vd),
4899 (ins nImmVMOVF32:$SIMM), IIC_VMOVImm,
4900 "vmov", "f32", "$Vd, $SIMM", "",
4901 [(set DPR:$Vd, (v2f32 (NEONvmovFPImm timm:$SIMM)))]>;
4902 def VMOVv4f32 : N1ModImm<1, 0b000, 0b1111, 0, 1, 0, 1, (outs QPR:$Vd),
4903 (ins nImmVMOVF32:$SIMM), IIC_VMOVImm,
4904 "vmov", "f32", "$Vd, $SIMM", "",
4905 [(set QPR:$Vd, (v4f32 (NEONvmovFPImm timm:$SIMM)))]>;
4906 } // isReMaterializable
4908 // VMOV : Vector Get Lane (move scalar to ARM core register)
4910 def VGETLNs8 : NVGetLane<{1,1,1,0,0,1,?,1}, 0b1011, {?,?},
4911 (outs GPR:$R), (ins DPR:$V, VectorIndex8:$lane),
4912 IIC_VMOVSI, "vmov", "s8", "$R, $V$lane",
4913 [(set GPR:$R, (NEONvgetlanes (v8i8 DPR:$V),
4915 let Inst{21} = lane{2};
4916 let Inst{6-5} = lane{1-0};
4918 def VGETLNs16 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, {?,1},
4919 (outs GPR:$R), (ins DPR:$V, VectorIndex16:$lane),
4920 IIC_VMOVSI, "vmov", "s16", "$R, $V$lane",
4921 [(set GPR:$R, (NEONvgetlanes (v4i16 DPR:$V),
4923 let Inst{21} = lane{1};
4924 let Inst{6} = lane{0};
4926 def VGETLNu8 : NVGetLane<{1,1,1,0,1,1,?,1}, 0b1011, {?,?},
4927 (outs GPR:$R), (ins DPR:$V, VectorIndex8:$lane),
4928 IIC_VMOVSI, "vmov", "u8", "$R, $V$lane",
4929 [(set GPR:$R, (NEONvgetlaneu (v8i8 DPR:$V),
4931 let Inst{21} = lane{2};
4932 let Inst{6-5} = lane{1-0};
4934 def VGETLNu16 : NVGetLane<{1,1,1,0,1,0,?,1}, 0b1011, {?,1},
4935 (outs GPR:$R), (ins DPR:$V, VectorIndex16:$lane),
4936 IIC_VMOVSI, "vmov", "u16", "$R, $V$lane",
4937 [(set GPR:$R, (NEONvgetlaneu (v4i16 DPR:$V),
4939 let Inst{21} = lane{1};
4940 let Inst{6} = lane{0};
4942 def VGETLNi32 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, 0b00,
4943 (outs GPR:$R), (ins DPR:$V, VectorIndex32:$lane),
4944 IIC_VMOVSI, "vmov", "32", "$R, $V$lane",
4945 [(set GPR:$R, (extractelt (v2i32 DPR:$V),
4947 let Inst{21} = lane{0};
4949 // def VGETLNf32: see FMRDH and FMRDL in ARMInstrVFP.td
4950 def : Pat<(NEONvgetlanes (v16i8 QPR:$src), imm:$lane),
4951 (VGETLNs8 (v8i8 (EXTRACT_SUBREG QPR:$src,
4952 (DSubReg_i8_reg imm:$lane))),
4953 (SubReg_i8_lane imm:$lane))>;
4954 def : Pat<(NEONvgetlanes (v8i16 QPR:$src), imm:$lane),
4955 (VGETLNs16 (v4i16 (EXTRACT_SUBREG QPR:$src,
4956 (DSubReg_i16_reg imm:$lane))),
4957 (SubReg_i16_lane imm:$lane))>;
4958 def : Pat<(NEONvgetlaneu (v16i8 QPR:$src), imm:$lane),
4959 (VGETLNu8 (v8i8 (EXTRACT_SUBREG QPR:$src,
4960 (DSubReg_i8_reg imm:$lane))),
4961 (SubReg_i8_lane imm:$lane))>;
4962 def : Pat<(NEONvgetlaneu (v8i16 QPR:$src), imm:$lane),
4963 (VGETLNu16 (v4i16 (EXTRACT_SUBREG QPR:$src,
4964 (DSubReg_i16_reg imm:$lane))),
4965 (SubReg_i16_lane imm:$lane))>;
4966 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
4967 (VGETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src,
4968 (DSubReg_i32_reg imm:$lane))),
4969 (SubReg_i32_lane imm:$lane))>;
4970 def : Pat<(extractelt (v2f32 DPR:$src1), imm:$src2),
4971 (EXTRACT_SUBREG (v2f32 (COPY_TO_REGCLASS (v2f32 DPR:$src1),DPR_VFP2)),
4972 (SSubReg_f32_reg imm:$src2))>;
4973 def : Pat<(extractelt (v4f32 QPR:$src1), imm:$src2),
4974 (EXTRACT_SUBREG (v4f32 (COPY_TO_REGCLASS (v4f32 QPR:$src1),QPR_VFP2)),
4975 (SSubReg_f32_reg imm:$src2))>;
4976 //def : Pat<(extractelt (v2i64 QPR:$src1), imm:$src2),
4977 // (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
4978 def : Pat<(extractelt (v2f64 QPR:$src1), imm:$src2),
4979 (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
4982 // VMOV : Vector Set Lane (move ARM core register to scalar)
4984 let Constraints = "$src1 = $V" in {
4985 def VSETLNi8 : NVSetLane<{1,1,1,0,0,1,?,0}, 0b1011, {?,?}, (outs DPR:$V),
4986 (ins DPR:$src1, GPR:$R, VectorIndex8:$lane),
4987 IIC_VMOVISL, "vmov", "8", "$V$lane, $R",
4988 [(set DPR:$V, (vector_insert (v8i8 DPR:$src1),
4989 GPR:$R, imm:$lane))]> {
4990 let Inst{21} = lane{2};
4991 let Inst{6-5} = lane{1-0};
4993 def VSETLNi16 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, {?,1}, (outs DPR:$V),
4994 (ins DPR:$src1, GPR:$R, VectorIndex16:$lane),
4995 IIC_VMOVISL, "vmov", "16", "$V$lane, $R",
4996 [(set DPR:$V, (vector_insert (v4i16 DPR:$src1),
4997 GPR:$R, imm:$lane))]> {
4998 let Inst{21} = lane{1};
4999 let Inst{6} = lane{0};
5001 def VSETLNi32 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, 0b00, (outs DPR:$V),
5002 (ins DPR:$src1, GPR:$R, VectorIndex32:$lane),
5003 IIC_VMOVISL, "vmov", "32", "$V$lane, $R",
5004 [(set DPR:$V, (insertelt (v2i32 DPR:$src1),
5005 GPR:$R, imm:$lane))]> {
5006 let Inst{21} = lane{0};
5009 def : Pat<(vector_insert (v16i8 QPR:$src1), GPR:$src2, imm:$lane),
5010 (v16i8 (INSERT_SUBREG QPR:$src1,
5011 (v8i8 (VSETLNi8 (v8i8 (EXTRACT_SUBREG QPR:$src1,
5012 (DSubReg_i8_reg imm:$lane))),
5013 GPR:$src2, (SubReg_i8_lane imm:$lane))),
5014 (DSubReg_i8_reg imm:$lane)))>;
5015 def : Pat<(vector_insert (v8i16 QPR:$src1), GPR:$src2, imm:$lane),
5016 (v8i16 (INSERT_SUBREG QPR:$src1,
5017 (v4i16 (VSETLNi16 (v4i16 (EXTRACT_SUBREG QPR:$src1,
5018 (DSubReg_i16_reg imm:$lane))),
5019 GPR:$src2, (SubReg_i16_lane imm:$lane))),
5020 (DSubReg_i16_reg imm:$lane)))>;
5021 def : Pat<(insertelt (v4i32 QPR:$src1), GPR:$src2, imm:$lane),
5022 (v4i32 (INSERT_SUBREG QPR:$src1,
5023 (v2i32 (VSETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src1,
5024 (DSubReg_i32_reg imm:$lane))),
5025 GPR:$src2, (SubReg_i32_lane imm:$lane))),
5026 (DSubReg_i32_reg imm:$lane)))>;
5028 def : Pat<(v2f32 (insertelt DPR:$src1, SPR:$src2, imm:$src3)),
5029 (INSERT_SUBREG (v2f32 (COPY_TO_REGCLASS DPR:$src1, DPR_VFP2)),
5030 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
5031 def : Pat<(v4f32 (insertelt QPR:$src1, SPR:$src2, imm:$src3)),
5032 (INSERT_SUBREG (v4f32 (COPY_TO_REGCLASS QPR:$src1, QPR_VFP2)),
5033 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
5035 //def : Pat<(v2i64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
5036 // (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
5037 def : Pat<(v2f64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
5038 (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
5040 def : Pat<(v2f32 (scalar_to_vector SPR:$src)),
5041 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
5042 def : Pat<(v2f64 (scalar_to_vector (f64 DPR:$src))),
5043 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
5044 def : Pat<(v4f32 (scalar_to_vector SPR:$src)),
5045 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
5047 def : Pat<(v8i8 (scalar_to_vector GPR:$src)),
5048 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
5049 def : Pat<(v4i16 (scalar_to_vector GPR:$src)),
5050 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
5051 def : Pat<(v2i32 (scalar_to_vector GPR:$src)),
5052 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
5054 def : Pat<(v16i8 (scalar_to_vector GPR:$src)),
5055 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
5056 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
5058 def : Pat<(v8i16 (scalar_to_vector GPR:$src)),
5059 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
5060 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
5062 def : Pat<(v4i32 (scalar_to_vector GPR:$src)),
5063 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
5064 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
5067 // VDUP : Vector Duplicate (from ARM core register to all elements)
5069 class VDUPD<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
5070 : NVDup<opcod1, 0b1011, opcod3, (outs DPR:$V), (ins GPR:$R),
5071 IIC_VMOVIS, "vdup", Dt, "$V, $R",
5072 [(set DPR:$V, (Ty (NEONvdup (i32 GPR:$R))))]>;
5073 class VDUPQ<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
5074 : NVDup<opcod1, 0b1011, opcod3, (outs QPR:$V), (ins GPR:$R),
5075 IIC_VMOVIS, "vdup", Dt, "$V, $R",
5076 [(set QPR:$V, (Ty (NEONvdup (i32 GPR:$R))))]>;
5078 def VDUP8d : VDUPD<0b11101100, 0b00, "8", v8i8>;
5079 def VDUP16d : VDUPD<0b11101000, 0b01, "16", v4i16>;
5080 def VDUP32d : VDUPD<0b11101000, 0b00, "32", v2i32>;
5081 def VDUP8q : VDUPQ<0b11101110, 0b00, "8", v16i8>;
5082 def VDUP16q : VDUPQ<0b11101010, 0b01, "16", v8i16>;
5083 def VDUP32q : VDUPQ<0b11101010, 0b00, "32", v4i32>;
5085 def : Pat<(v2f32 (NEONvdup (f32 (bitconvert GPR:$R)))), (VDUP32d GPR:$R)>;
5086 def : Pat<(v4f32 (NEONvdup (f32 (bitconvert GPR:$R)))), (VDUP32q GPR:$R)>;
5088 // VDUP : Vector Duplicate Lane (from scalar to all elements)
5090 class VDUPLND<bits<4> op19_16, string OpcodeStr, string Dt,
5091 ValueType Ty, Operand IdxTy>
5092 : NVDupLane<op19_16, 0, (outs DPR:$Vd), (ins DPR:$Vm, IdxTy:$lane),
5093 IIC_VMOVD, OpcodeStr, Dt, "$Vd, $Vm$lane",
5094 [(set DPR:$Vd, (Ty (NEONvduplane (Ty DPR:$Vm), imm:$lane)))]>;
5096 class VDUPLNQ<bits<4> op19_16, string OpcodeStr, string Dt,
5097 ValueType ResTy, ValueType OpTy, Operand IdxTy>
5098 : NVDupLane<op19_16, 1, (outs QPR:$Vd), (ins DPR:$Vm, IdxTy:$lane),
5099 IIC_VMOVQ, OpcodeStr, Dt, "$Vd, $Vm$lane",
5100 [(set QPR:$Vd, (ResTy (NEONvduplane (OpTy DPR:$Vm),
5101 VectorIndex32:$lane)))]>;
5103 // Inst{19-16} is partially specified depending on the element size.
5105 def VDUPLN8d : VDUPLND<{?,?,?,1}, "vdup", "8", v8i8, VectorIndex8> {
5107 let Inst{19-17} = lane{2-0};
5109 def VDUPLN16d : VDUPLND<{?,?,1,0}, "vdup", "16", v4i16, VectorIndex16> {
5111 let Inst{19-18} = lane{1-0};
5113 def VDUPLN32d : VDUPLND<{?,1,0,0}, "vdup", "32", v2i32, VectorIndex32> {
5115 let Inst{19} = lane{0};
5117 def VDUPLN8q : VDUPLNQ<{?,?,?,1}, "vdup", "8", v16i8, v8i8, VectorIndex8> {
5119 let Inst{19-17} = lane{2-0};
5121 def VDUPLN16q : VDUPLNQ<{?,?,1,0}, "vdup", "16", v8i16, v4i16, VectorIndex16> {
5123 let Inst{19-18} = lane{1-0};
5125 def VDUPLN32q : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4i32, v2i32, VectorIndex32> {
5127 let Inst{19} = lane{0};
5130 def : Pat<(v2f32 (NEONvduplane (v2f32 DPR:$Vm), imm:$lane)),
5131 (VDUPLN32d DPR:$Vm, imm:$lane)>;
5133 def : Pat<(v4f32 (NEONvduplane (v2f32 DPR:$Vm), imm:$lane)),
5134 (VDUPLN32q DPR:$Vm, imm:$lane)>;
5136 def : Pat<(v16i8 (NEONvduplane (v16i8 QPR:$src), imm:$lane)),
5137 (v16i8 (VDUPLN8q (v8i8 (EXTRACT_SUBREG QPR:$src,
5138 (DSubReg_i8_reg imm:$lane))),
5139 (SubReg_i8_lane imm:$lane)))>;
5140 def : Pat<(v8i16 (NEONvduplane (v8i16 QPR:$src), imm:$lane)),
5141 (v8i16 (VDUPLN16q (v4i16 (EXTRACT_SUBREG QPR:$src,
5142 (DSubReg_i16_reg imm:$lane))),
5143 (SubReg_i16_lane imm:$lane)))>;
5144 def : Pat<(v4i32 (NEONvduplane (v4i32 QPR:$src), imm:$lane)),
5145 (v4i32 (VDUPLN32q (v2i32 (EXTRACT_SUBREG QPR:$src,
5146 (DSubReg_i32_reg imm:$lane))),
5147 (SubReg_i32_lane imm:$lane)))>;
5148 def : Pat<(v4f32 (NEONvduplane (v4f32 QPR:$src), imm:$lane)),
5149 (v4f32 (VDUPLN32q (v2f32 (EXTRACT_SUBREG QPR:$src,
5150 (DSubReg_i32_reg imm:$lane))),
5151 (SubReg_i32_lane imm:$lane)))>;
5153 def VDUPfdf : PseudoNeonI<(outs DPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
5154 [(set DPR:$dst, (v2f32 (NEONvdup (f32 SPR:$src))))]>;
5155 def VDUPfqf : PseudoNeonI<(outs QPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
5156 [(set QPR:$dst, (v4f32 (NEONvdup (f32 SPR:$src))))]>;
5158 // VMOVN : Vector Narrowing Move
5159 defm VMOVN : N2VN_HSD<0b11,0b11,0b10,0b00100,0,0, IIC_VMOVN,
5160 "vmovn", "i", trunc>;
5161 // VQMOVN : Vector Saturating Narrowing Move
5162 defm VQMOVNs : N2VNInt_HSD<0b11,0b11,0b10,0b00101,0,0, IIC_VQUNAiD,
5163 "vqmovn", "s", int_arm_neon_vqmovns>;
5164 defm VQMOVNu : N2VNInt_HSD<0b11,0b11,0b10,0b00101,1,0, IIC_VQUNAiD,
5165 "vqmovn", "u", int_arm_neon_vqmovnu>;
5166 defm VQMOVNsu : N2VNInt_HSD<0b11,0b11,0b10,0b00100,1,0, IIC_VQUNAiD,
5167 "vqmovun", "s", int_arm_neon_vqmovnsu>;
5168 // VMOVL : Vector Lengthening Move
5169 defm VMOVLs : N2VL_QHS<0b01,0b10100,0,1, "vmovl", "s", sext>;
5170 defm VMOVLu : N2VL_QHS<0b11,0b10100,0,1, "vmovl", "u", zext>;
5171 def : Pat<(v8i16 (anyext (v8i8 DPR:$Vm))), (VMOVLuv8i16 DPR:$Vm)>;
5172 def : Pat<(v4i32 (anyext (v4i16 DPR:$Vm))), (VMOVLuv4i32 DPR:$Vm)>;
5173 def : Pat<(v2i64 (anyext (v2i32 DPR:$Vm))), (VMOVLuv2i64 DPR:$Vm)>;
5175 // Vector Conversions.
5177 // VCVT : Vector Convert Between Floating-Point and Integers
5178 def VCVTf2sd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
5179 v2i32, v2f32, fp_to_sint>;
5180 def VCVTf2ud : N2VD<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
5181 v2i32, v2f32, fp_to_uint>;
5182 def VCVTs2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
5183 v2f32, v2i32, sint_to_fp>;
5184 def VCVTu2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
5185 v2f32, v2i32, uint_to_fp>;
5187 def VCVTf2sq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
5188 v4i32, v4f32, fp_to_sint>;
5189 def VCVTf2uq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
5190 v4i32, v4f32, fp_to_uint>;
5191 def VCVTs2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
5192 v4f32, v4i32, sint_to_fp>;
5193 def VCVTu2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
5194 v4f32, v4i32, uint_to_fp>;
5196 // VCVT : Vector Convert Between Floating-Point and Fixed-Point.
5197 let DecoderMethod = "DecodeVCVTD" in {
5198 def VCVTf2xsd : N2VCvtD<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
5199 v2i32, v2f32, int_arm_neon_vcvtfp2fxs>;
5200 def VCVTf2xud : N2VCvtD<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
5201 v2i32, v2f32, int_arm_neon_vcvtfp2fxu>;
5202 def VCVTxs2fd : N2VCvtD<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
5203 v2f32, v2i32, int_arm_neon_vcvtfxs2fp>;
5204 def VCVTxu2fd : N2VCvtD<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
5205 v2f32, v2i32, int_arm_neon_vcvtfxu2fp>;
5208 let DecoderMethod = "DecodeVCVTQ" in {
5209 def VCVTf2xsq : N2VCvtQ<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
5210 v4i32, v4f32, int_arm_neon_vcvtfp2fxs>;
5211 def VCVTf2xuq : N2VCvtQ<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
5212 v4i32, v4f32, int_arm_neon_vcvtfp2fxu>;
5213 def VCVTxs2fq : N2VCvtQ<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
5214 v4f32, v4i32, int_arm_neon_vcvtfxs2fp>;
5215 def VCVTxu2fq : N2VCvtQ<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
5216 v4f32, v4i32, int_arm_neon_vcvtfxu2fp>;
5219 // VCVT : Vector Convert Between Half-Precision and Single-Precision.
5220 def VCVTf2h : N2VNInt<0b11, 0b11, 0b01, 0b10, 0b01100, 0, 0,
5221 IIC_VUNAQ, "vcvt", "f16.f32",
5222 v4i16, v4f32, int_arm_neon_vcvtfp2hf>,
5223 Requires<[HasNEON, HasFP16]>;
5224 def VCVTh2f : N2VLInt<0b11, 0b11, 0b01, 0b10, 0b01110, 0, 0,
5225 IIC_VUNAQ, "vcvt", "f32.f16",
5226 v4f32, v4i16, int_arm_neon_vcvthf2fp>,
5227 Requires<[HasNEON, HasFP16]>;
5231 // VREV64 : Vector Reverse elements within 64-bit doublewords
5233 class VREV64D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
5234 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 0, 0, (outs DPR:$Vd),
5235 (ins DPR:$Vm), IIC_VMOVD,
5236 OpcodeStr, Dt, "$Vd, $Vm", "",
5237 [(set DPR:$Vd, (Ty (NEONvrev64 (Ty DPR:$Vm))))]>;
5238 class VREV64Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
5239 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 1, 0, (outs QPR:$Vd),
5240 (ins QPR:$Vm), IIC_VMOVQ,
5241 OpcodeStr, Dt, "$Vd, $Vm", "",
5242 [(set QPR:$Vd, (Ty (NEONvrev64 (Ty QPR:$Vm))))]>;
5244 def VREV64d8 : VREV64D<0b00, "vrev64", "8", v8i8>;
5245 def VREV64d16 : VREV64D<0b01, "vrev64", "16", v4i16>;
5246 def VREV64d32 : VREV64D<0b10, "vrev64", "32", v2i32>;
5247 def : Pat<(v2f32 (NEONvrev64 (v2f32 DPR:$Vm))), (VREV64d32 DPR:$Vm)>;
5249 def VREV64q8 : VREV64Q<0b00, "vrev64", "8", v16i8>;
5250 def VREV64q16 : VREV64Q<0b01, "vrev64", "16", v8i16>;
5251 def VREV64q32 : VREV64Q<0b10, "vrev64", "32", v4i32>;
5252 def : Pat<(v4f32 (NEONvrev64 (v4f32 QPR:$Vm))), (VREV64q32 QPR:$Vm)>;
5254 // VREV32 : Vector Reverse elements within 32-bit words
5256 class VREV32D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
5257 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 0, 0, (outs DPR:$Vd),
5258 (ins DPR:$Vm), IIC_VMOVD,
5259 OpcodeStr, Dt, "$Vd, $Vm", "",
5260 [(set DPR:$Vd, (Ty (NEONvrev32 (Ty DPR:$Vm))))]>;
5261 class VREV32Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
5262 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 1, 0, (outs QPR:$Vd),
5263 (ins QPR:$Vm), IIC_VMOVQ,
5264 OpcodeStr, Dt, "$Vd, $Vm", "",
5265 [(set QPR:$Vd, (Ty (NEONvrev32 (Ty QPR:$Vm))))]>;
5267 def VREV32d8 : VREV32D<0b00, "vrev32", "8", v8i8>;
5268 def VREV32d16 : VREV32D<0b01, "vrev32", "16", v4i16>;
5270 def VREV32q8 : VREV32Q<0b00, "vrev32", "8", v16i8>;
5271 def VREV32q16 : VREV32Q<0b01, "vrev32", "16", v8i16>;
5273 // VREV16 : Vector Reverse elements within 16-bit halfwords
5275 class VREV16D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
5276 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 0, 0, (outs DPR:$Vd),
5277 (ins DPR:$Vm), IIC_VMOVD,
5278 OpcodeStr, Dt, "$Vd, $Vm", "",
5279 [(set DPR:$Vd, (Ty (NEONvrev16 (Ty DPR:$Vm))))]>;
5280 class VREV16Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
5281 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 1, 0, (outs QPR:$Vd),
5282 (ins QPR:$Vm), IIC_VMOVQ,
5283 OpcodeStr, Dt, "$Vd, $Vm", "",
5284 [(set QPR:$Vd, (Ty (NEONvrev16 (Ty QPR:$Vm))))]>;
5286 def VREV16d8 : VREV16D<0b00, "vrev16", "8", v8i8>;
5287 def VREV16q8 : VREV16Q<0b00, "vrev16", "8", v16i8>;
5289 // Other Vector Shuffles.
5291 // Aligned extractions: really just dropping registers
5293 class AlignedVEXTq<ValueType DestTy, ValueType SrcTy, SDNodeXForm LaneCVT>
5294 : Pat<(DestTy (vector_extract_subvec (SrcTy QPR:$src), (i32 imm:$start))),
5295 (EXTRACT_SUBREG (SrcTy QPR:$src), (LaneCVT imm:$start))>;
5297 def : AlignedVEXTq<v8i8, v16i8, DSubReg_i8_reg>;
5299 def : AlignedVEXTq<v4i16, v8i16, DSubReg_i16_reg>;
5301 def : AlignedVEXTq<v2i32, v4i32, DSubReg_i32_reg>;
5303 def : AlignedVEXTq<v1i64, v2i64, DSubReg_f64_reg>;
5305 def : AlignedVEXTq<v2f32, v4f32, DSubReg_i32_reg>;
5308 // VEXT : Vector Extract
5311 // All of these have a two-operand InstAlias.
5312 let TwoOperandAliasConstraint = "$Vn = $Vd" in {
5313 class VEXTd<string OpcodeStr, string Dt, ValueType Ty, Operand immTy>
5314 : N3V<0,1,0b11,{?,?,?,?},0,0, (outs DPR:$Vd),
5315 (ins DPR:$Vn, DPR:$Vm, immTy:$index), NVExtFrm,
5316 IIC_VEXTD, OpcodeStr, Dt, "$Vd, $Vn, $Vm, $index", "",
5317 [(set DPR:$Vd, (Ty (NEONvext (Ty DPR:$Vn),
5318 (Ty DPR:$Vm), imm:$index)))]> {
5320 let Inst{11-8} = index{3-0};
5323 class VEXTq<string OpcodeStr, string Dt, ValueType Ty, Operand immTy>
5324 : N3V<0,1,0b11,{?,?,?,?},1,0, (outs QPR:$Vd),
5325 (ins QPR:$Vn, QPR:$Vm, imm0_15:$index), NVExtFrm,
5326 IIC_VEXTQ, OpcodeStr, Dt, "$Vd, $Vn, $Vm, $index", "",
5327 [(set QPR:$Vd, (Ty (NEONvext (Ty QPR:$Vn),
5328 (Ty QPR:$Vm), imm:$index)))]> {
5330 let Inst{11-8} = index{3-0};
5334 def VEXTd8 : VEXTd<"vext", "8", v8i8, imm0_7> {
5335 let Inst{11-8} = index{3-0};
5337 def VEXTd16 : VEXTd<"vext", "16", v4i16, imm0_3> {
5338 let Inst{11-9} = index{2-0};
5341 def VEXTd32 : VEXTd<"vext", "32", v2i32, imm0_1> {
5342 let Inst{11-10} = index{1-0};
5343 let Inst{9-8} = 0b00;
5345 def : Pat<(v2f32 (NEONvext (v2f32 DPR:$Vn),
5348 (VEXTd32 DPR:$Vn, DPR:$Vm, imm:$index)>;
5350 def VEXTq8 : VEXTq<"vext", "8", v16i8, imm0_15> {
5351 let Inst{11-8} = index{3-0};
5353 def VEXTq16 : VEXTq<"vext", "16", v8i16, imm0_7> {
5354 let Inst{11-9} = index{2-0};
5357 def VEXTq32 : VEXTq<"vext", "32", v4i32, imm0_3> {
5358 let Inst{11-10} = index{1-0};
5359 let Inst{9-8} = 0b00;
5361 def VEXTq64 : VEXTq<"vext", "64", v2i64, imm0_1> {
5362 let Inst{11} = index{0};
5363 let Inst{10-8} = 0b000;
5365 def : Pat<(v4f32 (NEONvext (v4f32 QPR:$Vn),
5368 (VEXTq32 QPR:$Vn, QPR:$Vm, imm:$index)>;
5370 // VTRN : Vector Transpose
5372 def VTRNd8 : N2VDShuffle<0b00, 0b00001, "vtrn", "8">;
5373 def VTRNd16 : N2VDShuffle<0b01, 0b00001, "vtrn", "16">;
5374 def VTRNd32 : N2VDShuffle<0b10, 0b00001, "vtrn", "32">;
5376 def VTRNq8 : N2VQShuffle<0b00, 0b00001, IIC_VPERMQ, "vtrn", "8">;
5377 def VTRNq16 : N2VQShuffle<0b01, 0b00001, IIC_VPERMQ, "vtrn", "16">;
5378 def VTRNq32 : N2VQShuffle<0b10, 0b00001, IIC_VPERMQ, "vtrn", "32">;
5380 // VUZP : Vector Unzip (Deinterleave)
5382 def VUZPd8 : N2VDShuffle<0b00, 0b00010, "vuzp", "8">;
5383 def VUZPd16 : N2VDShuffle<0b01, 0b00010, "vuzp", "16">;
5384 // vuzp.32 Dd, Dm is a pseudo-instruction expanded to vtrn.32 Dd, Dm.
5385 def : NEONInstAlias<"vuzp${p}.32 $Dd, $Dm",
5386 (VTRNd32 DPR:$Dd, DPR:$Dm, pred:$p)>;
5388 def VUZPq8 : N2VQShuffle<0b00, 0b00010, IIC_VPERMQ3, "vuzp", "8">;
5389 def VUZPq16 : N2VQShuffle<0b01, 0b00010, IIC_VPERMQ3, "vuzp", "16">;
5390 def VUZPq32 : N2VQShuffle<0b10, 0b00010, IIC_VPERMQ3, "vuzp", "32">;
5392 // VZIP : Vector Zip (Interleave)
5394 def VZIPd8 : N2VDShuffle<0b00, 0b00011, "vzip", "8">;
5395 def VZIPd16 : N2VDShuffle<0b01, 0b00011, "vzip", "16">;
5396 // vzip.32 Dd, Dm is a pseudo-instruction expanded to vtrn.32 Dd, Dm.
5397 def : NEONInstAlias<"vzip${p}.32 $Dd, $Dm",
5398 (VTRNd32 DPR:$Dd, DPR:$Dm, pred:$p)>;
5400 def VZIPq8 : N2VQShuffle<0b00, 0b00011, IIC_VPERMQ3, "vzip", "8">;
5401 def VZIPq16 : N2VQShuffle<0b01, 0b00011, IIC_VPERMQ3, "vzip", "16">;
5402 def VZIPq32 : N2VQShuffle<0b10, 0b00011, IIC_VPERMQ3, "vzip", "32">;
5404 // Vector Table Lookup and Table Extension.
5406 // VTBL : Vector Table Lookup
5407 let DecoderMethod = "DecodeTBLInstruction" in {
5409 : N3V<1,1,0b11,0b1000,0,0, (outs DPR:$Vd),
5410 (ins VecListOneD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTB1,
5411 "vtbl", "8", "$Vd, $Vn, $Vm", "",
5412 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbl1 VecListOneD:$Vn, DPR:$Vm)))]>;
5413 let hasExtraSrcRegAllocReq = 1 in {
5415 : N3V<1,1,0b11,0b1001,0,0, (outs DPR:$Vd),
5416 (ins VecListDPair:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTB2,
5417 "vtbl", "8", "$Vd, $Vn, $Vm", "", []>;
5419 : N3V<1,1,0b11,0b1010,0,0, (outs DPR:$Vd),
5420 (ins VecListThreeD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTB3,
5421 "vtbl", "8", "$Vd, $Vn, $Vm", "", []>;
5423 : N3V<1,1,0b11,0b1011,0,0, (outs DPR:$Vd),
5424 (ins VecListFourD:$Vn, DPR:$Vm),
5426 "vtbl", "8", "$Vd, $Vn, $Vm", "", []>;
5427 } // hasExtraSrcRegAllocReq = 1
5430 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB3, "", []>;
5432 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB4, "", []>;
5434 // VTBX : Vector Table Extension
5436 : N3V<1,1,0b11,0b1000,1,0, (outs DPR:$Vd),
5437 (ins DPR:$orig, VecListOneD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTBX1,
5438 "vtbx", "8", "$Vd, $Vn, $Vm", "$orig = $Vd",
5439 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbx1
5440 DPR:$orig, VecListOneD:$Vn, DPR:$Vm)))]>;
5441 let hasExtraSrcRegAllocReq = 1 in {
5443 : N3V<1,1,0b11,0b1001,1,0, (outs DPR:$Vd),
5444 (ins DPR:$orig, VecListDPair:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTBX2,
5445 "vtbx", "8", "$Vd, $Vn, $Vm", "$orig = $Vd", []>;
5447 : N3V<1,1,0b11,0b1010,1,0, (outs DPR:$Vd),
5448 (ins DPR:$orig, VecListThreeD:$Vn, DPR:$Vm),
5449 NVTBLFrm, IIC_VTBX3,
5450 "vtbx", "8", "$Vd, $Vn, $Vm",
5453 : N3V<1,1,0b11,0b1011,1,0, (outs DPR:$Vd),
5454 (ins DPR:$orig, VecListFourD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTBX4,
5455 "vtbx", "8", "$Vd, $Vn, $Vm",
5457 } // hasExtraSrcRegAllocReq = 1
5460 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
5461 IIC_VTBX3, "$orig = $dst", []>;
5463 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
5464 IIC_VTBX4, "$orig = $dst", []>;
5465 } // DecoderMethod = "DecodeTBLInstruction"
5467 //===----------------------------------------------------------------------===//
5468 // NEON instructions for single-precision FP math
5469 //===----------------------------------------------------------------------===//
5471 class N2VSPat<SDNode OpNode, NeonI Inst>
5472 : NEONFPPat<(f32 (OpNode SPR:$a)),
5474 (v2f32 (COPY_TO_REGCLASS (Inst
5476 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5477 SPR:$a, ssub_0)), DPR_VFP2)), ssub_0)>;
5479 class N3VSPat<SDNode OpNode, NeonI Inst>
5480 : NEONFPPat<(f32 (OpNode SPR:$a, SPR:$b)),
5482 (v2f32 (COPY_TO_REGCLASS (Inst
5484 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5487 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5488 SPR:$b, ssub_0)), DPR_VFP2)), ssub_0)>;
5490 class N3VSMulOpPat<SDNode MulNode, SDNode OpNode, NeonI Inst>
5491 : NEONFPPat<(f32 (OpNode SPR:$acc, (f32 (MulNode SPR:$a, SPR:$b)))),
5493 (v2f32 (COPY_TO_REGCLASS (Inst
5495 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5498 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5501 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5502 SPR:$b, ssub_0)), DPR_VFP2)), ssub_0)>;
5504 def : N3VSPat<fadd, VADDfd>;
5505 def : N3VSPat<fsub, VSUBfd>;
5506 def : N3VSPat<fmul, VMULfd>;
5507 def : N3VSMulOpPat<fmul, fadd, VMLAfd>,
5508 Requires<[HasNEON, UseNEONForFP, UseFPVMLx, DontUseFusedMAC]>;
5509 def : N3VSMulOpPat<fmul, fsub, VMLSfd>,
5510 Requires<[HasNEON, UseNEONForFP, UseFPVMLx, DontUseFusedMAC]>;
5511 def : N3VSMulOpPat<fmul, fadd, VFMAfd>,
5512 Requires<[HasVFP4, UseNEONForFP, UseFusedMAC]>;
5513 def : N3VSMulOpPat<fmul, fsub, VFMSfd>,
5514 Requires<[HasVFP4, UseNEONForFP, UseFusedMAC]>;
5515 def : N2VSPat<fabs, VABSfd>;
5516 def : N2VSPat<fneg, VNEGfd>;
5517 def : N3VSPat<NEONfmax, VMAXfd>;
5518 def : N3VSPat<NEONfmin, VMINfd>;
5519 def : N2VSPat<arm_ftosi, VCVTf2sd>;
5520 def : N2VSPat<arm_ftoui, VCVTf2ud>;
5521 def : N2VSPat<arm_sitof, VCVTs2fd>;
5522 def : N2VSPat<arm_uitof, VCVTu2fd>;
5524 //===----------------------------------------------------------------------===//
5525 // Non-Instruction Patterns
5526 //===----------------------------------------------------------------------===//
5529 def : Pat<(v1i64 (bitconvert (v2i32 DPR:$src))), (v1i64 DPR:$src)>;
5530 def : Pat<(v1i64 (bitconvert (v4i16 DPR:$src))), (v1i64 DPR:$src)>;
5531 def : Pat<(v1i64 (bitconvert (v8i8 DPR:$src))), (v1i64 DPR:$src)>;
5532 def : Pat<(v1i64 (bitconvert (f64 DPR:$src))), (v1i64 DPR:$src)>;
5533 def : Pat<(v1i64 (bitconvert (v2f32 DPR:$src))), (v1i64 DPR:$src)>;
5534 def : Pat<(v2i32 (bitconvert (v1i64 DPR:$src))), (v2i32 DPR:$src)>;
5535 def : Pat<(v2i32 (bitconvert (v4i16 DPR:$src))), (v2i32 DPR:$src)>;
5536 def : Pat<(v2i32 (bitconvert (v8i8 DPR:$src))), (v2i32 DPR:$src)>;
5537 def : Pat<(v2i32 (bitconvert (f64 DPR:$src))), (v2i32 DPR:$src)>;
5538 def : Pat<(v2i32 (bitconvert (v2f32 DPR:$src))), (v2i32 DPR:$src)>;
5539 def : Pat<(v4i16 (bitconvert (v1i64 DPR:$src))), (v4i16 DPR:$src)>;
5540 def : Pat<(v4i16 (bitconvert (v2i32 DPR:$src))), (v4i16 DPR:$src)>;
5541 def : Pat<(v4i16 (bitconvert (v8i8 DPR:$src))), (v4i16 DPR:$src)>;
5542 def : Pat<(v4i16 (bitconvert (f64 DPR:$src))), (v4i16 DPR:$src)>;
5543 def : Pat<(v4i16 (bitconvert (v2f32 DPR:$src))), (v4i16 DPR:$src)>;
5544 def : Pat<(v8i8 (bitconvert (v1i64 DPR:$src))), (v8i8 DPR:$src)>;
5545 def : Pat<(v8i8 (bitconvert (v2i32 DPR:$src))), (v8i8 DPR:$src)>;
5546 def : Pat<(v8i8 (bitconvert (v4i16 DPR:$src))), (v8i8 DPR:$src)>;
5547 def : Pat<(v8i8 (bitconvert (f64 DPR:$src))), (v8i8 DPR:$src)>;
5548 def : Pat<(v8i8 (bitconvert (v2f32 DPR:$src))), (v8i8 DPR:$src)>;
5549 def : Pat<(f64 (bitconvert (v1i64 DPR:$src))), (f64 DPR:$src)>;
5550 def : Pat<(f64 (bitconvert (v2i32 DPR:$src))), (f64 DPR:$src)>;
5551 def : Pat<(f64 (bitconvert (v4i16 DPR:$src))), (f64 DPR:$src)>;
5552 def : Pat<(f64 (bitconvert (v8i8 DPR:$src))), (f64 DPR:$src)>;
5553 def : Pat<(f64 (bitconvert (v2f32 DPR:$src))), (f64 DPR:$src)>;
5554 def : Pat<(v2f32 (bitconvert (f64 DPR:$src))), (v2f32 DPR:$src)>;
5555 def : Pat<(v2f32 (bitconvert (v1i64 DPR:$src))), (v2f32 DPR:$src)>;
5556 def : Pat<(v2f32 (bitconvert (v2i32 DPR:$src))), (v2f32 DPR:$src)>;
5557 def : Pat<(v2f32 (bitconvert (v4i16 DPR:$src))), (v2f32 DPR:$src)>;
5558 def : Pat<(v2f32 (bitconvert (v8i8 DPR:$src))), (v2f32 DPR:$src)>;
5560 def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (v2i64 QPR:$src)>;
5561 def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (v2i64 QPR:$src)>;
5562 def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (v2i64 QPR:$src)>;
5563 def : Pat<(v2i64 (bitconvert (v2f64 QPR:$src))), (v2i64 QPR:$src)>;
5564 def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (v2i64 QPR:$src)>;
5565 def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (v4i32 QPR:$src)>;
5566 def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (v4i32 QPR:$src)>;
5567 def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (v4i32 QPR:$src)>;
5568 def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (v4i32 QPR:$src)>;
5569 def : Pat<(v4i32 (bitconvert (v4f32 QPR:$src))), (v4i32 QPR:$src)>;
5570 def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (v8i16 QPR:$src)>;
5571 def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (v8i16 QPR:$src)>;
5572 def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (v8i16 QPR:$src)>;
5573 def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (v8i16 QPR:$src)>;
5574 def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (v8i16 QPR:$src)>;
5575 def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (v16i8 QPR:$src)>;
5576 def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (v16i8 QPR:$src)>;
5577 def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (v16i8 QPR:$src)>;
5578 def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (v16i8 QPR:$src)>;
5579 def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (v16i8 QPR:$src)>;
5580 def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (v4f32 QPR:$src)>;
5581 def : Pat<(v4f32 (bitconvert (v4i32 QPR:$src))), (v4f32 QPR:$src)>;
5582 def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (v4f32 QPR:$src)>;
5583 def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (v4f32 QPR:$src)>;
5584 def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (v4f32 QPR:$src)>;
5585 def : Pat<(v2f64 (bitconvert (v2i64 QPR:$src))), (v2f64 QPR:$src)>;
5586 def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (v2f64 QPR:$src)>;
5587 def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 QPR:$src)>;
5588 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 QPR:$src)>;
5589 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 QPR:$src)>;
5591 // Vector lengthening move with load, matching extending loads.
5593 // extload, zextload and sextload for a standard lengthening load. Example:
5594 // Lengthen_Single<"8", "i16", "i8"> = Pat<(v8i16 (extloadvi8 addrmode5:$addr))
5595 // (VMOVLuv8i16 (VLDRD addrmode5:$addr))>;
5596 multiclass Lengthen_Single<string DestLanes, string DestTy, string SrcTy> {
5597 def _Any : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5598 (!cast<PatFrag>("extloadv" # SrcTy) addrmode5:$addr)),
5599 (!cast<Instruction>("VMOVLuv" # DestLanes # DestTy)
5600 (VLDRD addrmode5:$addr))>;
5601 def _Z : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5602 (!cast<PatFrag>("zextloadv" # SrcTy) addrmode5:$addr)),
5603 (!cast<Instruction>("VMOVLuv" # DestLanes # DestTy)
5604 (VLDRD addrmode5:$addr))>;
5605 def _S : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5606 (!cast<PatFrag>("sextloadv" # SrcTy) addrmode5:$addr)),
5607 (!cast<Instruction>("VMOVLsv" # DestLanes # DestTy)
5608 (VLDRD addrmode5:$addr))>;
5611 // extload, zextload and sextload for a lengthening load which only uses
5612 // half the lanes available. Example:
5613 // Lengthen_HalfSingle<"4", "i16", "8", "i16", "i8"> =
5614 // Pat<(v4i16 (extloadvi8 addrmode5:$addr))
5615 // (EXTRACT_SUBREG (VMOVLuv8i16 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
5616 // (VLDRS addrmode5:$addr),
5619 multiclass Lengthen_HalfSingle<string DestLanes, string DestTy, string SrcTy,
5620 string InsnLanes, string InsnTy> {
5621 def _Any : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5622 (!cast<PatFrag>("extloadv" # SrcTy) addrmode5:$addr)),
5623 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # InsnLanes # InsnTy)
5624 (INSERT_SUBREG (f64 (IMPLICIT_DEF)), (VLDRS addrmode5:$addr), ssub_0)),
5626 def _Z : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5627 (!cast<PatFrag>("zextloadv" # SrcTy) addrmode5:$addr)),
5628 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # InsnLanes # InsnTy)
5629 (INSERT_SUBREG (f64 (IMPLICIT_DEF)), (VLDRS addrmode5:$addr), ssub_0)),
5631 def _S : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5632 (!cast<PatFrag>("sextloadv" # SrcTy) addrmode5:$addr)),
5633 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLsv" # InsnLanes # InsnTy)
5634 (INSERT_SUBREG (f64 (IMPLICIT_DEF)), (VLDRS addrmode5:$addr), ssub_0)),
5638 // extload, zextload and sextload for a lengthening load followed by another
5639 // lengthening load, to quadruple the initial length.
5641 // Lengthen_Double<"4", "i32", "i8", "8", "i16", "4", "i32", qsub_0> =
5642 // Pat<(v4i32 (extloadvi8 addrmode5:$addr))
5643 // (EXTRACT_SUBREG (VMOVLuv4i32
5644 // (EXTRACT_SUBREG (VMOVLuv8i16 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
5645 // (VLDRS addrmode5:$addr),
5649 multiclass Lengthen_Double<string DestLanes, string DestTy, string SrcTy,
5650 string Insn1Lanes, string Insn1Ty, string Insn2Lanes,
5652 def _Any : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5653 (!cast<PatFrag>("extloadv" # SrcTy) addrmode5:$addr)),
5654 (!cast<Instruction>("VMOVLuv" # Insn2Lanes # Insn2Ty)
5655 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn1Lanes # Insn1Ty)
5656 (INSERT_SUBREG (f64 (IMPLICIT_DEF)), (VLDRS addrmode5:$addr),
5657 ssub_0)), dsub_0))>;
5658 def _Z : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5659 (!cast<PatFrag>("zextloadv" # SrcTy) addrmode5:$addr)),
5660 (!cast<Instruction>("VMOVLuv" # Insn2Lanes # Insn2Ty)
5661 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn1Lanes # Insn1Ty)
5662 (INSERT_SUBREG (f64 (IMPLICIT_DEF)), (VLDRS addrmode5:$addr),
5663 ssub_0)), dsub_0))>;
5664 def _S : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5665 (!cast<PatFrag>("sextloadv" # SrcTy) addrmode5:$addr)),
5666 (!cast<Instruction>("VMOVLsv" # Insn2Lanes # Insn2Ty)
5667 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLsv" # Insn1Lanes # Insn1Ty)
5668 (INSERT_SUBREG (f64 (IMPLICIT_DEF)), (VLDRS addrmode5:$addr),
5669 ssub_0)), dsub_0))>;
5672 // extload, zextload and sextload for a lengthening load followed by another
5673 // lengthening load, to quadruple the initial length, but which ends up only
5674 // requiring half the available lanes (a 64-bit outcome instead of a 128-bit).
5676 // Lengthen_HalfDouble<"2", "i32", "i8", "8", "i16", "4", "i32"> =
5677 // Pat<(v4i32 (extloadvi8 addrmode5:$addr))
5678 // (EXTRACT_SUBREG (VMOVLuv4i32
5679 // (EXTRACT_SUBREG (VMOVLuv8i16 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
5680 // (VLDRS addrmode5:$addr),
5684 multiclass Lengthen_HalfDouble<string DestLanes, string DestTy, string SrcTy,
5685 string Insn1Lanes, string Insn1Ty, string Insn2Lanes,
5687 def _Any : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5688 (!cast<PatFrag>("extloadv" # SrcTy) addrmode5:$addr)),
5689 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn2Lanes # Insn2Ty)
5690 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn1Lanes # Insn1Ty)
5691 (INSERT_SUBREG (f64 (IMPLICIT_DEF)), (VLDRS addrmode5:$addr),
5694 def _Z : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5695 (!cast<PatFrag>("zextloadv" # SrcTy) addrmode5:$addr)),
5696 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn2Lanes # Insn2Ty)
5697 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn1Lanes # Insn1Ty)
5698 (INSERT_SUBREG (f64 (IMPLICIT_DEF)), (VLDRS addrmode5:$addr),
5701 def _S : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5702 (!cast<PatFrag>("sextloadv" # SrcTy) addrmode5:$addr)),
5703 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLsv" # Insn2Lanes # Insn2Ty)
5704 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLsv" # Insn1Lanes # Insn1Ty)
5705 (INSERT_SUBREG (f64 (IMPLICIT_DEF)), (VLDRS addrmode5:$addr),
5710 defm : Lengthen_Single<"8", "i16", "i8">; // v8i8 -> v8i16
5711 defm : Lengthen_Single<"4", "i32", "i16">; // v4i16 -> v4i32
5712 defm : Lengthen_Single<"2", "i64", "i32">; // v2i32 -> v2i64
5714 defm : Lengthen_HalfSingle<"4", "i16", "i8", "8", "i16">; // v4i8 -> v4i16
5715 defm : Lengthen_HalfSingle<"2", "i16", "i8", "8", "i16">; // v2i8 -> v2i16
5716 defm : Lengthen_HalfSingle<"2", "i32", "i16", "4", "i32">; // v2i16 -> v2i32
5718 // Double lengthening - v4i8 -> v4i16 -> v4i32
5719 defm : Lengthen_Double<"4", "i32", "i8", "8", "i16", "4", "i32">;
5720 // v2i8 -> v2i16 -> v2i32
5721 defm : Lengthen_HalfDouble<"2", "i32", "i8", "8", "i16", "4", "i32">;
5722 // v2i16 -> v2i32 -> v2i64
5723 defm : Lengthen_Double<"2", "i64", "i16", "4", "i32", "2", "i64">;
5725 // Triple lengthening - v2i8 -> v2i16 -> v2i32 -> v2i64
5726 def : Pat<(v2i64 (extloadvi8 addrmode5:$addr)),
5727 (VMOVLuv2i64 (EXTRACT_SUBREG (VMOVLuv4i32 (EXTRACT_SUBREG (VMOVLuv8i16
5728 (INSERT_SUBREG (f64 (IMPLICIT_DEF)), (VLDRS addrmode5:$addr), ssub_0)),
5729 dsub_0)), dsub_0))>;
5730 def : Pat<(v2i64 (zextloadvi8 addrmode5:$addr)),
5731 (VMOVLuv2i64 (EXTRACT_SUBREG (VMOVLuv4i32 (EXTRACT_SUBREG (VMOVLuv8i16
5732 (INSERT_SUBREG (f64 (IMPLICIT_DEF)), (VLDRS addrmode5:$addr), ssub_0)),
5733 dsub_0)), dsub_0))>;
5734 def : Pat<(v2i64 (sextloadvi8 addrmode5:$addr)),
5735 (VMOVLsv2i64 (EXTRACT_SUBREG (VMOVLsv4i32 (EXTRACT_SUBREG (VMOVLsv8i16
5736 (INSERT_SUBREG (f64 (IMPLICIT_DEF)), (VLDRS addrmode5:$addr), ssub_0)),
5737 dsub_0)), dsub_0))>;
5739 //===----------------------------------------------------------------------===//
5740 // Assembler aliases
5743 def : VFP2InstAlias<"fmdhr${p} $Dd, $Rn",
5744 (VSETLNi32 DPR:$Dd, GPR:$Rn, 1, pred:$p)>;
5745 def : VFP2InstAlias<"fmdlr${p} $Dd, $Rn",
5746 (VSETLNi32 DPR:$Dd, GPR:$Rn, 0, pred:$p)>;
5748 // VAND/VBIC/VEOR/VORR accept but do not require a type suffix.
5749 defm : NEONDTAnyInstAlias<"vand${p}", "$Vd, $Vn, $Vm",
5750 (VANDd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
5751 defm : NEONDTAnyInstAlias<"vand${p}", "$Vd, $Vn, $Vm",
5752 (VANDq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
5753 defm : NEONDTAnyInstAlias<"vbic${p}", "$Vd, $Vn, $Vm",
5754 (VBICd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
5755 defm : NEONDTAnyInstAlias<"vbic${p}", "$Vd, $Vn, $Vm",
5756 (VBICq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
5757 defm : NEONDTAnyInstAlias<"veor${p}", "$Vd, $Vn, $Vm",
5758 (VEORd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
5759 defm : NEONDTAnyInstAlias<"veor${p}", "$Vd, $Vn, $Vm",
5760 (VEORq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
5761 defm : NEONDTAnyInstAlias<"vorr${p}", "$Vd, $Vn, $Vm",
5762 (VORRd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
5763 defm : NEONDTAnyInstAlias<"vorr${p}", "$Vd, $Vn, $Vm",
5764 (VORRq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
5765 // ... two-operand aliases
5766 defm : NEONDTAnyInstAlias<"vand${p}", "$Vdn, $Vm",
5767 (VANDd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5768 defm : NEONDTAnyInstAlias<"vand${p}", "$Vdn, $Vm",
5769 (VANDq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5770 defm : NEONDTAnyInstAlias<"veor${p}", "$Vdn, $Vm",
5771 (VEORd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5772 defm : NEONDTAnyInstAlias<"veor${p}", "$Vdn, $Vm",
5773 (VEORq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5774 defm : NEONDTAnyInstAlias<"vorr${p}", "$Vdn, $Vm",
5775 (VORRd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5776 defm : NEONDTAnyInstAlias<"vorr${p}", "$Vdn, $Vm",
5777 (VORRq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5779 // VLD1 single-lane pseudo-instructions. These need special handling for
5780 // the lane index that an InstAlias can't handle, so we use these instead.
5781 def VLD1LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vld1${p}", ".8", "$list, $addr",
5782 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5783 def VLD1LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vld1${p}", ".16", "$list, $addr",
5784 (ins VecListOneDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5785 def VLD1LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vld1${p}", ".32", "$list, $addr",
5786 (ins VecListOneDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5788 def VLD1LNdWB_fixed_Asm_8 :
5789 NEONDataTypeAsmPseudoInst<"vld1${p}", ".8", "$list, $addr!",
5790 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5791 def VLD1LNdWB_fixed_Asm_16 :
5792 NEONDataTypeAsmPseudoInst<"vld1${p}", ".16", "$list, $addr!",
5793 (ins VecListOneDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5794 def VLD1LNdWB_fixed_Asm_32 :
5795 NEONDataTypeAsmPseudoInst<"vld1${p}", ".32", "$list, $addr!",
5796 (ins VecListOneDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5797 def VLD1LNdWB_register_Asm_8 :
5798 NEONDataTypeAsmPseudoInst<"vld1${p}", ".8", "$list, $addr, $Rm",
5799 (ins VecListOneDByteIndexed:$list, addrmode6:$addr,
5800 rGPR:$Rm, pred:$p)>;
5801 def VLD1LNdWB_register_Asm_16 :
5802 NEONDataTypeAsmPseudoInst<"vld1${p}", ".16", "$list, $addr, $Rm",
5803 (ins VecListOneDHWordIndexed:$list, addrmode6:$addr,
5804 rGPR:$Rm, pred:$p)>;
5805 def VLD1LNdWB_register_Asm_32 :
5806 NEONDataTypeAsmPseudoInst<"vld1${p}", ".32", "$list, $addr, $Rm",
5807 (ins VecListOneDWordIndexed:$list, addrmode6:$addr,
5808 rGPR:$Rm, pred:$p)>;
5811 // VST1 single-lane pseudo-instructions. These need special handling for
5812 // the lane index that an InstAlias can't handle, so we use these instead.
5813 def VST1LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vst1${p}", ".8", "$list, $addr",
5814 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5815 def VST1LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vst1${p}", ".16", "$list, $addr",
5816 (ins VecListOneDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5817 def VST1LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vst1${p}", ".32", "$list, $addr",
5818 (ins VecListOneDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5820 def VST1LNdWB_fixed_Asm_8 :
5821 NEONDataTypeAsmPseudoInst<"vst1${p}", ".8", "$list, $addr!",
5822 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5823 def VST1LNdWB_fixed_Asm_16 :
5824 NEONDataTypeAsmPseudoInst<"vst1${p}", ".16", "$list, $addr!",
5825 (ins VecListOneDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5826 def VST1LNdWB_fixed_Asm_32 :
5827 NEONDataTypeAsmPseudoInst<"vst1${p}", ".32", "$list, $addr!",
5828 (ins VecListOneDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5829 def VST1LNdWB_register_Asm_8 :
5830 NEONDataTypeAsmPseudoInst<"vst1${p}", ".8", "$list, $addr, $Rm",
5831 (ins VecListOneDByteIndexed:$list, addrmode6:$addr,
5832 rGPR:$Rm, pred:$p)>;
5833 def VST1LNdWB_register_Asm_16 :
5834 NEONDataTypeAsmPseudoInst<"vst1${p}", ".16", "$list, $addr, $Rm",
5835 (ins VecListOneDHWordIndexed:$list, addrmode6:$addr,
5836 rGPR:$Rm, pred:$p)>;
5837 def VST1LNdWB_register_Asm_32 :
5838 NEONDataTypeAsmPseudoInst<"vst1${p}", ".32", "$list, $addr, $Rm",
5839 (ins VecListOneDWordIndexed:$list, addrmode6:$addr,
5840 rGPR:$Rm, pred:$p)>;
5842 // VLD2 single-lane pseudo-instructions. These need special handling for
5843 // the lane index that an InstAlias can't handle, so we use these instead.
5844 def VLD2LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vld2${p}", ".8", "$list, $addr",
5845 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5846 def VLD2LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vld2${p}", ".16", "$list, $addr",
5847 (ins VecListTwoDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5848 def VLD2LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vld2${p}", ".32", "$list, $addr",
5849 (ins VecListTwoDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5850 def VLD2LNqAsm_16 : NEONDataTypeAsmPseudoInst<"vld2${p}", ".16", "$list, $addr",
5851 (ins VecListTwoQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5852 def VLD2LNqAsm_32 : NEONDataTypeAsmPseudoInst<"vld2${p}", ".32", "$list, $addr",
5853 (ins VecListTwoQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5855 def VLD2LNdWB_fixed_Asm_8 :
5856 NEONDataTypeAsmPseudoInst<"vld2${p}", ".8", "$list, $addr!",
5857 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5858 def VLD2LNdWB_fixed_Asm_16 :
5859 NEONDataTypeAsmPseudoInst<"vld2${p}", ".16", "$list, $addr!",
5860 (ins VecListTwoDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5861 def VLD2LNdWB_fixed_Asm_32 :
5862 NEONDataTypeAsmPseudoInst<"vld2${p}", ".32", "$list, $addr!",
5863 (ins VecListTwoDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5864 def VLD2LNqWB_fixed_Asm_16 :
5865 NEONDataTypeAsmPseudoInst<"vld2${p}", ".16", "$list, $addr!",
5866 (ins VecListTwoQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5867 def VLD2LNqWB_fixed_Asm_32 :
5868 NEONDataTypeAsmPseudoInst<"vld2${p}", ".32", "$list, $addr!",
5869 (ins VecListTwoQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5870 def VLD2LNdWB_register_Asm_8 :
5871 NEONDataTypeAsmPseudoInst<"vld2${p}", ".8", "$list, $addr, $Rm",
5872 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr,
5873 rGPR:$Rm, pred:$p)>;
5874 def VLD2LNdWB_register_Asm_16 :
5875 NEONDataTypeAsmPseudoInst<"vld2${p}", ".16", "$list, $addr, $Rm",
5876 (ins VecListTwoDHWordIndexed:$list, addrmode6:$addr,
5877 rGPR:$Rm, pred:$p)>;
5878 def VLD2LNdWB_register_Asm_32 :
5879 NEONDataTypeAsmPseudoInst<"vld2${p}", ".32", "$list, $addr, $Rm",
5880 (ins VecListTwoDWordIndexed:$list, addrmode6:$addr,
5881 rGPR:$Rm, pred:$p)>;
5882 def VLD2LNqWB_register_Asm_16 :
5883 NEONDataTypeAsmPseudoInst<"vld2${p}", ".16", "$list, $addr, $Rm",
5884 (ins VecListTwoQHWordIndexed:$list, addrmode6:$addr,
5885 rGPR:$Rm, pred:$p)>;
5886 def VLD2LNqWB_register_Asm_32 :
5887 NEONDataTypeAsmPseudoInst<"vld2${p}", ".32", "$list, $addr, $Rm",
5888 (ins VecListTwoQWordIndexed:$list, addrmode6:$addr,
5889 rGPR:$Rm, pred:$p)>;
5892 // VST2 single-lane pseudo-instructions. These need special handling for
5893 // the lane index that an InstAlias can't handle, so we use these instead.
5894 def VST2LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vst2${p}", ".8", "$list, $addr",
5895 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5896 def VST2LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vst2${p}", ".16", "$list, $addr",
5897 (ins VecListTwoDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5898 def VST2LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vst2${p}", ".32", "$list, $addr",
5899 (ins VecListTwoDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5900 def VST2LNqAsm_16 : NEONDataTypeAsmPseudoInst<"vst2${p}", ".16", "$list, $addr",
5901 (ins VecListTwoQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5902 def VST2LNqAsm_32 : NEONDataTypeAsmPseudoInst<"vst2${p}", ".32", "$list, $addr",
5903 (ins VecListTwoQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5905 def VST2LNdWB_fixed_Asm_8 :
5906 NEONDataTypeAsmPseudoInst<"vst2${p}", ".8", "$list, $addr!",
5907 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5908 def VST2LNdWB_fixed_Asm_16 :
5909 NEONDataTypeAsmPseudoInst<"vst2${p}", ".16", "$list, $addr!",
5910 (ins VecListTwoDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5911 def VST2LNdWB_fixed_Asm_32 :
5912 NEONDataTypeAsmPseudoInst<"vst2${p}", ".32", "$list, $addr!",
5913 (ins VecListTwoDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5914 def VST2LNqWB_fixed_Asm_16 :
5915 NEONDataTypeAsmPseudoInst<"vst2${p}", ".16", "$list, $addr!",
5916 (ins VecListTwoQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5917 def VST2LNqWB_fixed_Asm_32 :
5918 NEONDataTypeAsmPseudoInst<"vst2${p}", ".32", "$list, $addr!",
5919 (ins VecListTwoQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5920 def VST2LNdWB_register_Asm_8 :
5921 NEONDataTypeAsmPseudoInst<"vst2${p}", ".8", "$list, $addr, $Rm",
5922 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr,
5923 rGPR:$Rm, pred:$p)>;
5924 def VST2LNdWB_register_Asm_16 :
5925 NEONDataTypeAsmPseudoInst<"vst2${p}", ".16","$list, $addr, $Rm",
5926 (ins VecListTwoDHWordIndexed:$list, addrmode6:$addr,
5927 rGPR:$Rm, pred:$p)>;
5928 def VST2LNdWB_register_Asm_32 :
5929 NEONDataTypeAsmPseudoInst<"vst2${p}", ".32", "$list, $addr, $Rm",
5930 (ins VecListTwoDWordIndexed:$list, addrmode6:$addr,
5931 rGPR:$Rm, pred:$p)>;
5932 def VST2LNqWB_register_Asm_16 :
5933 NEONDataTypeAsmPseudoInst<"vst2${p}", ".16","$list, $addr, $Rm",
5934 (ins VecListTwoQHWordIndexed:$list, addrmode6:$addr,
5935 rGPR:$Rm, pred:$p)>;
5936 def VST2LNqWB_register_Asm_32 :
5937 NEONDataTypeAsmPseudoInst<"vst2${p}", ".32", "$list, $addr, $Rm",
5938 (ins VecListTwoQWordIndexed:$list, addrmode6:$addr,
5939 rGPR:$Rm, pred:$p)>;
5941 // VLD3 all-lanes pseudo-instructions. These need special handling for
5942 // the lane index that an InstAlias can't handle, so we use these instead.
5943 def VLD3DUPdAsm_8 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr",
5944 (ins VecListThreeDAllLanes:$list, addrmode6:$addr, pred:$p)>;
5945 def VLD3DUPdAsm_16 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr",
5946 (ins VecListThreeDAllLanes:$list, addrmode6:$addr, pred:$p)>;
5947 def VLD3DUPdAsm_32 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr",
5948 (ins VecListThreeDAllLanes:$list, addrmode6:$addr, pred:$p)>;
5949 def VLD3DUPqAsm_8 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr",
5950 (ins VecListThreeQAllLanes:$list, addrmode6:$addr, pred:$p)>;
5951 def VLD3DUPqAsm_16 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr",
5952 (ins VecListThreeQAllLanes:$list, addrmode6:$addr, pred:$p)>;
5953 def VLD3DUPqAsm_32 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr",
5954 (ins VecListThreeQAllLanes:$list, addrmode6:$addr, pred:$p)>;
5956 def VLD3DUPdWB_fixed_Asm_8 :
5957 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr!",
5958 (ins VecListThreeDAllLanes:$list, addrmode6:$addr, pred:$p)>;
5959 def VLD3DUPdWB_fixed_Asm_16 :
5960 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr!",
5961 (ins VecListThreeDAllLanes:$list, addrmode6:$addr, pred:$p)>;
5962 def VLD3DUPdWB_fixed_Asm_32 :
5963 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr!",
5964 (ins VecListThreeDAllLanes:$list, addrmode6:$addr, pred:$p)>;
5965 def VLD3DUPqWB_fixed_Asm_8 :
5966 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr!",
5967 (ins VecListThreeQAllLanes:$list, addrmode6:$addr, pred:$p)>;
5968 def VLD3DUPqWB_fixed_Asm_16 :
5969 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr!",
5970 (ins VecListThreeQAllLanes:$list, addrmode6:$addr, pred:$p)>;
5971 def VLD3DUPqWB_fixed_Asm_32 :
5972 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr!",
5973 (ins VecListThreeQAllLanes:$list, addrmode6:$addr, pred:$p)>;
5974 def VLD3DUPdWB_register_Asm_8 :
5975 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr, $Rm",
5976 (ins VecListThreeDAllLanes:$list, addrmode6:$addr,
5977 rGPR:$Rm, pred:$p)>;
5978 def VLD3DUPdWB_register_Asm_16 :
5979 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr, $Rm",
5980 (ins VecListThreeDAllLanes:$list, addrmode6:$addr,
5981 rGPR:$Rm, pred:$p)>;
5982 def VLD3DUPdWB_register_Asm_32 :
5983 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr, $Rm",
5984 (ins VecListThreeDAllLanes:$list, addrmode6:$addr,
5985 rGPR:$Rm, pred:$p)>;
5986 def VLD3DUPqWB_register_Asm_8 :
5987 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr, $Rm",
5988 (ins VecListThreeQAllLanes:$list, addrmode6:$addr,
5989 rGPR:$Rm, pred:$p)>;
5990 def VLD3DUPqWB_register_Asm_16 :
5991 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr, $Rm",
5992 (ins VecListThreeQAllLanes:$list, addrmode6:$addr,
5993 rGPR:$Rm, pred:$p)>;
5994 def VLD3DUPqWB_register_Asm_32 :
5995 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr, $Rm",
5996 (ins VecListThreeQAllLanes:$list, addrmode6:$addr,
5997 rGPR:$Rm, pred:$p)>;
6000 // VLD3 single-lane pseudo-instructions. These need special handling for
6001 // the lane index that an InstAlias can't handle, so we use these instead.
6002 def VLD3LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr",
6003 (ins VecListThreeDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6004 def VLD3LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr",
6005 (ins VecListThreeDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6006 def VLD3LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr",
6007 (ins VecListThreeDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6008 def VLD3LNqAsm_16 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr",
6009 (ins VecListThreeQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6010 def VLD3LNqAsm_32 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr",
6011 (ins VecListThreeQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6013 def VLD3LNdWB_fixed_Asm_8 :
6014 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr!",
6015 (ins VecListThreeDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6016 def VLD3LNdWB_fixed_Asm_16 :
6017 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr!",
6018 (ins VecListThreeDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6019 def VLD3LNdWB_fixed_Asm_32 :
6020 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr!",
6021 (ins VecListThreeDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6022 def VLD3LNqWB_fixed_Asm_16 :
6023 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr!",
6024 (ins VecListThreeQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6025 def VLD3LNqWB_fixed_Asm_32 :
6026 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr!",
6027 (ins VecListThreeQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6028 def VLD3LNdWB_register_Asm_8 :
6029 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr, $Rm",
6030 (ins VecListThreeDByteIndexed:$list, addrmode6:$addr,
6031 rGPR:$Rm, pred:$p)>;
6032 def VLD3LNdWB_register_Asm_16 :
6033 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr, $Rm",
6034 (ins VecListThreeDHWordIndexed:$list, addrmode6:$addr,
6035 rGPR:$Rm, pred:$p)>;
6036 def VLD3LNdWB_register_Asm_32 :
6037 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr, $Rm",
6038 (ins VecListThreeDWordIndexed:$list, addrmode6:$addr,
6039 rGPR:$Rm, pred:$p)>;
6040 def VLD3LNqWB_register_Asm_16 :
6041 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr, $Rm",
6042 (ins VecListThreeQHWordIndexed:$list, addrmode6:$addr,
6043 rGPR:$Rm, pred:$p)>;
6044 def VLD3LNqWB_register_Asm_32 :
6045 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr, $Rm",
6046 (ins VecListThreeQWordIndexed:$list, addrmode6:$addr,
6047 rGPR:$Rm, pred:$p)>;
6049 // VLD3 multiple structure pseudo-instructions. These need special handling for
6050 // the vector operands that the normal instructions don't yet model.
6051 // FIXME: Remove these when the register classes and instructions are updated.
6052 def VLD3dAsm_8 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr",
6053 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6054 def VLD3dAsm_16 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr",
6055 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6056 def VLD3dAsm_32 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr",
6057 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6058 def VLD3qAsm_8 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr",
6059 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6060 def VLD3qAsm_16 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr",
6061 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6062 def VLD3qAsm_32 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr",
6063 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6065 def VLD3dWB_fixed_Asm_8 :
6066 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr!",
6067 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6068 def VLD3dWB_fixed_Asm_16 :
6069 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr!",
6070 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6071 def VLD3dWB_fixed_Asm_32 :
6072 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr!",
6073 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6074 def VLD3qWB_fixed_Asm_8 :
6075 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr!",
6076 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6077 def VLD3qWB_fixed_Asm_16 :
6078 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr!",
6079 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6080 def VLD3qWB_fixed_Asm_32 :
6081 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr!",
6082 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6083 def VLD3dWB_register_Asm_8 :
6084 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr, $Rm",
6085 (ins VecListThreeD:$list, addrmode6:$addr,
6086 rGPR:$Rm, pred:$p)>;
6087 def VLD3dWB_register_Asm_16 :
6088 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr, $Rm",
6089 (ins VecListThreeD:$list, addrmode6:$addr,
6090 rGPR:$Rm, pred:$p)>;
6091 def VLD3dWB_register_Asm_32 :
6092 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr, $Rm",
6093 (ins VecListThreeD:$list, addrmode6:$addr,
6094 rGPR:$Rm, pred:$p)>;
6095 def VLD3qWB_register_Asm_8 :
6096 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr, $Rm",
6097 (ins VecListThreeQ:$list, addrmode6:$addr,
6098 rGPR:$Rm, pred:$p)>;
6099 def VLD3qWB_register_Asm_16 :
6100 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr, $Rm",
6101 (ins VecListThreeQ:$list, addrmode6:$addr,
6102 rGPR:$Rm, pred:$p)>;
6103 def VLD3qWB_register_Asm_32 :
6104 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr, $Rm",
6105 (ins VecListThreeQ:$list, addrmode6:$addr,
6106 rGPR:$Rm, pred:$p)>;
6108 // VST3 single-lane pseudo-instructions. These need special handling for
6109 // the lane index that an InstAlias can't handle, so we use these instead.
6110 def VST3LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr",
6111 (ins VecListThreeDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6112 def VST3LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr",
6113 (ins VecListThreeDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6114 def VST3LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr",
6115 (ins VecListThreeDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6116 def VST3LNqAsm_16 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr",
6117 (ins VecListThreeQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6118 def VST3LNqAsm_32 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr",
6119 (ins VecListThreeQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6121 def VST3LNdWB_fixed_Asm_8 :
6122 NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr!",
6123 (ins VecListThreeDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6124 def VST3LNdWB_fixed_Asm_16 :
6125 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr!",
6126 (ins VecListThreeDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6127 def VST3LNdWB_fixed_Asm_32 :
6128 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr!",
6129 (ins VecListThreeDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6130 def VST3LNqWB_fixed_Asm_16 :
6131 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr!",
6132 (ins VecListThreeQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6133 def VST3LNqWB_fixed_Asm_32 :
6134 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr!",
6135 (ins VecListThreeQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6136 def VST3LNdWB_register_Asm_8 :
6137 NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr, $Rm",
6138 (ins VecListThreeDByteIndexed:$list, addrmode6:$addr,
6139 rGPR:$Rm, pred:$p)>;
6140 def VST3LNdWB_register_Asm_16 :
6141 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr, $Rm",
6142 (ins VecListThreeDHWordIndexed:$list, addrmode6:$addr,
6143 rGPR:$Rm, pred:$p)>;
6144 def VST3LNdWB_register_Asm_32 :
6145 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr, $Rm",
6146 (ins VecListThreeDWordIndexed:$list, addrmode6:$addr,
6147 rGPR:$Rm, pred:$p)>;
6148 def VST3LNqWB_register_Asm_16 :
6149 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr, $Rm",
6150 (ins VecListThreeQHWordIndexed:$list, addrmode6:$addr,
6151 rGPR:$Rm, pred:$p)>;
6152 def VST3LNqWB_register_Asm_32 :
6153 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr, $Rm",
6154 (ins VecListThreeQWordIndexed:$list, addrmode6:$addr,
6155 rGPR:$Rm, pred:$p)>;
6158 // VST3 multiple structure pseudo-instructions. These need special handling for
6159 // the vector operands that the normal instructions don't yet model.
6160 // FIXME: Remove these when the register classes and instructions are updated.
6161 def VST3dAsm_8 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr",
6162 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6163 def VST3dAsm_16 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr",
6164 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6165 def VST3dAsm_32 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr",
6166 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6167 def VST3qAsm_8 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr",
6168 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6169 def VST3qAsm_16 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr",
6170 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6171 def VST3qAsm_32 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr",
6172 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6174 def VST3dWB_fixed_Asm_8 :
6175 NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr!",
6176 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6177 def VST3dWB_fixed_Asm_16 :
6178 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr!",
6179 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6180 def VST3dWB_fixed_Asm_32 :
6181 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr!",
6182 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6183 def VST3qWB_fixed_Asm_8 :
6184 NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr!",
6185 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6186 def VST3qWB_fixed_Asm_16 :
6187 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr!",
6188 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6189 def VST3qWB_fixed_Asm_32 :
6190 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr!",
6191 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6192 def VST3dWB_register_Asm_8 :
6193 NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr, $Rm",
6194 (ins VecListThreeD:$list, addrmode6:$addr,
6195 rGPR:$Rm, pred:$p)>;
6196 def VST3dWB_register_Asm_16 :
6197 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr, $Rm",
6198 (ins VecListThreeD:$list, addrmode6:$addr,
6199 rGPR:$Rm, pred:$p)>;
6200 def VST3dWB_register_Asm_32 :
6201 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr, $Rm",
6202 (ins VecListThreeD:$list, addrmode6:$addr,
6203 rGPR:$Rm, pred:$p)>;
6204 def VST3qWB_register_Asm_8 :
6205 NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr, $Rm",
6206 (ins VecListThreeQ:$list, addrmode6:$addr,
6207 rGPR:$Rm, pred:$p)>;
6208 def VST3qWB_register_Asm_16 :
6209 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr, $Rm",
6210 (ins VecListThreeQ:$list, addrmode6:$addr,
6211 rGPR:$Rm, pred:$p)>;
6212 def VST3qWB_register_Asm_32 :
6213 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr, $Rm",
6214 (ins VecListThreeQ:$list, addrmode6:$addr,
6215 rGPR:$Rm, pred:$p)>;
6217 // VLD4 all-lanes pseudo-instructions. These need special handling for
6218 // the lane index that an InstAlias can't handle, so we use these instead.
6219 def VLD4DUPdAsm_8 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr",
6220 (ins VecListFourDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6221 def VLD4DUPdAsm_16 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr",
6222 (ins VecListFourDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6223 def VLD4DUPdAsm_32 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr",
6224 (ins VecListFourDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6225 def VLD4DUPqAsm_8 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr",
6226 (ins VecListFourQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6227 def VLD4DUPqAsm_16 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr",
6228 (ins VecListFourQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6229 def VLD4DUPqAsm_32 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr",
6230 (ins VecListFourQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6232 def VLD4DUPdWB_fixed_Asm_8 :
6233 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr!",
6234 (ins VecListFourDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6235 def VLD4DUPdWB_fixed_Asm_16 :
6236 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr!",
6237 (ins VecListFourDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6238 def VLD4DUPdWB_fixed_Asm_32 :
6239 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr!",
6240 (ins VecListFourDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6241 def VLD4DUPqWB_fixed_Asm_8 :
6242 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr!",
6243 (ins VecListFourQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6244 def VLD4DUPqWB_fixed_Asm_16 :
6245 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr!",
6246 (ins VecListFourQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6247 def VLD4DUPqWB_fixed_Asm_32 :
6248 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr!",
6249 (ins VecListFourQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6250 def VLD4DUPdWB_register_Asm_8 :
6251 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr, $Rm",
6252 (ins VecListFourDAllLanes:$list, addrmode6:$addr,
6253 rGPR:$Rm, pred:$p)>;
6254 def VLD4DUPdWB_register_Asm_16 :
6255 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr, $Rm",
6256 (ins VecListFourDAllLanes:$list, addrmode6:$addr,
6257 rGPR:$Rm, pred:$p)>;
6258 def VLD4DUPdWB_register_Asm_32 :
6259 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr, $Rm",
6260 (ins VecListFourDAllLanes:$list, addrmode6:$addr,
6261 rGPR:$Rm, pred:$p)>;
6262 def VLD4DUPqWB_register_Asm_8 :
6263 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr, $Rm",
6264 (ins VecListFourQAllLanes:$list, addrmode6:$addr,
6265 rGPR:$Rm, pred:$p)>;
6266 def VLD4DUPqWB_register_Asm_16 :
6267 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr, $Rm",
6268 (ins VecListFourQAllLanes:$list, addrmode6:$addr,
6269 rGPR:$Rm, pred:$p)>;
6270 def VLD4DUPqWB_register_Asm_32 :
6271 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr, $Rm",
6272 (ins VecListFourQAllLanes:$list, addrmode6:$addr,
6273 rGPR:$Rm, pred:$p)>;
6276 // VLD4 single-lane pseudo-instructions. These need special handling for
6277 // the lane index that an InstAlias can't handle, so we use these instead.
6278 def VLD4LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr",
6279 (ins VecListFourDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6280 def VLD4LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr",
6281 (ins VecListFourDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6282 def VLD4LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr",
6283 (ins VecListFourDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6284 def VLD4LNqAsm_16 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr",
6285 (ins VecListFourQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6286 def VLD4LNqAsm_32 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr",
6287 (ins VecListFourQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6289 def VLD4LNdWB_fixed_Asm_8 :
6290 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr!",
6291 (ins VecListFourDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6292 def VLD4LNdWB_fixed_Asm_16 :
6293 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr!",
6294 (ins VecListFourDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6295 def VLD4LNdWB_fixed_Asm_32 :
6296 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr!",
6297 (ins VecListFourDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6298 def VLD4LNqWB_fixed_Asm_16 :
6299 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr!",
6300 (ins VecListFourQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6301 def VLD4LNqWB_fixed_Asm_32 :
6302 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr!",
6303 (ins VecListFourQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6304 def VLD4LNdWB_register_Asm_8 :
6305 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr, $Rm",
6306 (ins VecListFourDByteIndexed:$list, addrmode6:$addr,
6307 rGPR:$Rm, pred:$p)>;
6308 def VLD4LNdWB_register_Asm_16 :
6309 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr, $Rm",
6310 (ins VecListFourDHWordIndexed:$list, addrmode6:$addr,
6311 rGPR:$Rm, pred:$p)>;
6312 def VLD4LNdWB_register_Asm_32 :
6313 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr, $Rm",
6314 (ins VecListFourDWordIndexed:$list, addrmode6:$addr,
6315 rGPR:$Rm, pred:$p)>;
6316 def VLD4LNqWB_register_Asm_16 :
6317 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr, $Rm",
6318 (ins VecListFourQHWordIndexed:$list, addrmode6:$addr,
6319 rGPR:$Rm, pred:$p)>;
6320 def VLD4LNqWB_register_Asm_32 :
6321 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr, $Rm",
6322 (ins VecListFourQWordIndexed:$list, addrmode6:$addr,
6323 rGPR:$Rm, pred:$p)>;
6327 // VLD4 multiple structure pseudo-instructions. These need special handling for
6328 // the vector operands that the normal instructions don't yet model.
6329 // FIXME: Remove these when the register classes and instructions are updated.
6330 def VLD4dAsm_8 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr",
6331 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6332 def VLD4dAsm_16 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr",
6333 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6334 def VLD4dAsm_32 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr",
6335 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6336 def VLD4qAsm_8 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr",
6337 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6338 def VLD4qAsm_16 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr",
6339 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6340 def VLD4qAsm_32 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr",
6341 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6343 def VLD4dWB_fixed_Asm_8 :
6344 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr!",
6345 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6346 def VLD4dWB_fixed_Asm_16 :
6347 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr!",
6348 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6349 def VLD4dWB_fixed_Asm_32 :
6350 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr!",
6351 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6352 def VLD4qWB_fixed_Asm_8 :
6353 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr!",
6354 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6355 def VLD4qWB_fixed_Asm_16 :
6356 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr!",
6357 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6358 def VLD4qWB_fixed_Asm_32 :
6359 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr!",
6360 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6361 def VLD4dWB_register_Asm_8 :
6362 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr, $Rm",
6363 (ins VecListFourD:$list, addrmode6:$addr,
6364 rGPR:$Rm, pred:$p)>;
6365 def VLD4dWB_register_Asm_16 :
6366 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr, $Rm",
6367 (ins VecListFourD:$list, addrmode6:$addr,
6368 rGPR:$Rm, pred:$p)>;
6369 def VLD4dWB_register_Asm_32 :
6370 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr, $Rm",
6371 (ins VecListFourD:$list, addrmode6:$addr,
6372 rGPR:$Rm, pred:$p)>;
6373 def VLD4qWB_register_Asm_8 :
6374 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr, $Rm",
6375 (ins VecListFourQ:$list, addrmode6:$addr,
6376 rGPR:$Rm, pred:$p)>;
6377 def VLD4qWB_register_Asm_16 :
6378 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr, $Rm",
6379 (ins VecListFourQ:$list, addrmode6:$addr,
6380 rGPR:$Rm, pred:$p)>;
6381 def VLD4qWB_register_Asm_32 :
6382 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr, $Rm",
6383 (ins VecListFourQ:$list, addrmode6:$addr,
6384 rGPR:$Rm, pred:$p)>;
6386 // VST4 single-lane pseudo-instructions. These need special handling for
6387 // the lane index that an InstAlias can't handle, so we use these instead.
6388 def VST4LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr",
6389 (ins VecListFourDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6390 def VST4LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr",
6391 (ins VecListFourDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6392 def VST4LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr",
6393 (ins VecListFourDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6394 def VST4LNqAsm_16 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr",
6395 (ins VecListFourQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6396 def VST4LNqAsm_32 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr",
6397 (ins VecListFourQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6399 def VST4LNdWB_fixed_Asm_8 :
6400 NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr!",
6401 (ins VecListFourDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6402 def VST4LNdWB_fixed_Asm_16 :
6403 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr!",
6404 (ins VecListFourDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6405 def VST4LNdWB_fixed_Asm_32 :
6406 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr!",
6407 (ins VecListFourDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6408 def VST4LNqWB_fixed_Asm_16 :
6409 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr!",
6410 (ins VecListFourQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6411 def VST4LNqWB_fixed_Asm_32 :
6412 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr!",
6413 (ins VecListFourQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6414 def VST4LNdWB_register_Asm_8 :
6415 NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr, $Rm",
6416 (ins VecListFourDByteIndexed:$list, addrmode6:$addr,
6417 rGPR:$Rm, pred:$p)>;
6418 def VST4LNdWB_register_Asm_16 :
6419 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr, $Rm",
6420 (ins VecListFourDHWordIndexed:$list, addrmode6:$addr,
6421 rGPR:$Rm, pred:$p)>;
6422 def VST4LNdWB_register_Asm_32 :
6423 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr, $Rm",
6424 (ins VecListFourDWordIndexed:$list, addrmode6:$addr,
6425 rGPR:$Rm, pred:$p)>;
6426 def VST4LNqWB_register_Asm_16 :
6427 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr, $Rm",
6428 (ins VecListFourQHWordIndexed:$list, addrmode6:$addr,
6429 rGPR:$Rm, pred:$p)>;
6430 def VST4LNqWB_register_Asm_32 :
6431 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr, $Rm",
6432 (ins VecListFourQWordIndexed:$list, addrmode6:$addr,
6433 rGPR:$Rm, pred:$p)>;
6436 // VST4 multiple structure pseudo-instructions. These need special handling for
6437 // the vector operands that the normal instructions don't yet model.
6438 // FIXME: Remove these when the register classes and instructions are updated.
6439 def VST4dAsm_8 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr",
6440 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6441 def VST4dAsm_16 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr",
6442 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6443 def VST4dAsm_32 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr",
6444 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6445 def VST4qAsm_8 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr",
6446 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6447 def VST4qAsm_16 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr",
6448 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6449 def VST4qAsm_32 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr",
6450 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6452 def VST4dWB_fixed_Asm_8 :
6453 NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr!",
6454 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6455 def VST4dWB_fixed_Asm_16 :
6456 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr!",
6457 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6458 def VST4dWB_fixed_Asm_32 :
6459 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr!",
6460 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6461 def VST4qWB_fixed_Asm_8 :
6462 NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr!",
6463 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6464 def VST4qWB_fixed_Asm_16 :
6465 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr!",
6466 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6467 def VST4qWB_fixed_Asm_32 :
6468 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr!",
6469 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6470 def VST4dWB_register_Asm_8 :
6471 NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr, $Rm",
6472 (ins VecListFourD:$list, addrmode6:$addr,
6473 rGPR:$Rm, pred:$p)>;
6474 def VST4dWB_register_Asm_16 :
6475 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr, $Rm",
6476 (ins VecListFourD:$list, addrmode6:$addr,
6477 rGPR:$Rm, pred:$p)>;
6478 def VST4dWB_register_Asm_32 :
6479 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr, $Rm",
6480 (ins VecListFourD:$list, addrmode6:$addr,
6481 rGPR:$Rm, pred:$p)>;
6482 def VST4qWB_register_Asm_8 :
6483 NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr, $Rm",
6484 (ins VecListFourQ:$list, addrmode6:$addr,
6485 rGPR:$Rm, pred:$p)>;
6486 def VST4qWB_register_Asm_16 :
6487 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr, $Rm",
6488 (ins VecListFourQ:$list, addrmode6:$addr,
6489 rGPR:$Rm, pred:$p)>;
6490 def VST4qWB_register_Asm_32 :
6491 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr, $Rm",
6492 (ins VecListFourQ:$list, addrmode6:$addr,
6493 rGPR:$Rm, pred:$p)>;
6495 // VMOV takes an optional datatype suffix
6496 defm : NEONDTAnyInstAlias<"vmov${p}", "$Vd, $Vm",
6497 (VORRd DPR:$Vd, DPR:$Vm, DPR:$Vm, pred:$p)>;
6498 defm : NEONDTAnyInstAlias<"vmov${p}", "$Vd, $Vm",
6499 (VORRq QPR:$Vd, QPR:$Vm, QPR:$Vm, pred:$p)>;
6501 // VCLT (register) is an assembler alias for VCGT w/ the operands reversed.
6502 // D-register versions.
6503 def : NEONInstAlias<"vcle${p}.s8 $Dd, $Dn, $Dm",
6504 (VCGEsv8i8 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6505 def : NEONInstAlias<"vcle${p}.s16 $Dd, $Dn, $Dm",
6506 (VCGEsv4i16 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6507 def : NEONInstAlias<"vcle${p}.s32 $Dd, $Dn, $Dm",
6508 (VCGEsv2i32 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6509 def : NEONInstAlias<"vcle${p}.u8 $Dd, $Dn, $Dm",
6510 (VCGEuv8i8 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6511 def : NEONInstAlias<"vcle${p}.u16 $Dd, $Dn, $Dm",
6512 (VCGEuv4i16 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6513 def : NEONInstAlias<"vcle${p}.u32 $Dd, $Dn, $Dm",
6514 (VCGEuv2i32 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6515 def : NEONInstAlias<"vcle${p}.f32 $Dd, $Dn, $Dm",
6516 (VCGEfd DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6517 // Q-register versions.
6518 def : NEONInstAlias<"vcle${p}.s8 $Qd, $Qn, $Qm",
6519 (VCGEsv16i8 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6520 def : NEONInstAlias<"vcle${p}.s16 $Qd, $Qn, $Qm",
6521 (VCGEsv8i16 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6522 def : NEONInstAlias<"vcle${p}.s32 $Qd, $Qn, $Qm",
6523 (VCGEsv4i32 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6524 def : NEONInstAlias<"vcle${p}.u8 $Qd, $Qn, $Qm",
6525 (VCGEuv16i8 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6526 def : NEONInstAlias<"vcle${p}.u16 $Qd, $Qn, $Qm",
6527 (VCGEuv8i16 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6528 def : NEONInstAlias<"vcle${p}.u32 $Qd, $Qn, $Qm",
6529 (VCGEuv4i32 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6530 def : NEONInstAlias<"vcle${p}.f32 $Qd, $Qn, $Qm",
6531 (VCGEfq QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6533 // VCLT (register) is an assembler alias for VCGT w/ the operands reversed.
6534 // D-register versions.
6535 def : NEONInstAlias<"vclt${p}.s8 $Dd, $Dn, $Dm",
6536 (VCGTsv8i8 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6537 def : NEONInstAlias<"vclt${p}.s16 $Dd, $Dn, $Dm",
6538 (VCGTsv4i16 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6539 def : NEONInstAlias<"vclt${p}.s32 $Dd, $Dn, $Dm",
6540 (VCGTsv2i32 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6541 def : NEONInstAlias<"vclt${p}.u8 $Dd, $Dn, $Dm",
6542 (VCGTuv8i8 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6543 def : NEONInstAlias<"vclt${p}.u16 $Dd, $Dn, $Dm",
6544 (VCGTuv4i16 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6545 def : NEONInstAlias<"vclt${p}.u32 $Dd, $Dn, $Dm",
6546 (VCGTuv2i32 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6547 def : NEONInstAlias<"vclt${p}.f32 $Dd, $Dn, $Dm",
6548 (VCGTfd DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6549 // Q-register versions.
6550 def : NEONInstAlias<"vclt${p}.s8 $Qd, $Qn, $Qm",
6551 (VCGTsv16i8 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6552 def : NEONInstAlias<"vclt${p}.s16 $Qd, $Qn, $Qm",
6553 (VCGTsv8i16 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6554 def : NEONInstAlias<"vclt${p}.s32 $Qd, $Qn, $Qm",
6555 (VCGTsv4i32 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6556 def : NEONInstAlias<"vclt${p}.u8 $Qd, $Qn, $Qm",
6557 (VCGTuv16i8 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6558 def : NEONInstAlias<"vclt${p}.u16 $Qd, $Qn, $Qm",
6559 (VCGTuv8i16 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6560 def : NEONInstAlias<"vclt${p}.u32 $Qd, $Qn, $Qm",
6561 (VCGTuv4i32 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6562 def : NEONInstAlias<"vclt${p}.f32 $Qd, $Qn, $Qm",
6563 (VCGTfq QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6565 // Two-operand variants for VQDMULH
6566 def : NEONInstAlias<"vqdmulh${p}.s16 $Vdn, $Vm",
6567 (VQDMULHv4i16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
6568 def : NEONInstAlias<"vqdmulh${p}.s32 $Vdn, $Vm",
6569 (VQDMULHv2i32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
6571 def : NEONInstAlias<"vqdmulh${p}.s16 $Vdn, $Vm",
6572 (VQDMULHv8i16 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
6573 def : NEONInstAlias<"vqdmulh${p}.s32 $Vdn, $Vm",
6574 (VQDMULHv4i32 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
6576 // Two-operand variants for VSRA.
6578 def : NEONInstAlias<"vsra${p}.s8 $Vdm, $imm",
6579 (VSRAsv8i8 DPR:$Vdm, DPR:$Vdm, shr_imm8:$imm, pred:$p)>;
6580 def : NEONInstAlias<"vsra${p}.s16 $Vdm, $imm",
6581 (VSRAsv4i16 DPR:$Vdm, DPR:$Vdm, shr_imm16:$imm, pred:$p)>;
6582 def : NEONInstAlias<"vsra${p}.s32 $Vdm, $imm",
6583 (VSRAsv2i32 DPR:$Vdm, DPR:$Vdm, shr_imm32:$imm, pred:$p)>;
6584 def : NEONInstAlias<"vsra${p}.s64 $Vdm, $imm",
6585 (VSRAsv1i64 DPR:$Vdm, DPR:$Vdm, shr_imm64:$imm, pred:$p)>;
6587 def : NEONInstAlias<"vsra${p}.s8 $Vdm, $imm",
6588 (VSRAsv16i8 QPR:$Vdm, QPR:$Vdm, shr_imm8:$imm, pred:$p)>;
6589 def : NEONInstAlias<"vsra${p}.s16 $Vdm, $imm",
6590 (VSRAsv8i16 QPR:$Vdm, QPR:$Vdm, shr_imm16:$imm, pred:$p)>;
6591 def : NEONInstAlias<"vsra${p}.s32 $Vdm, $imm",
6592 (VSRAsv4i32 QPR:$Vdm, QPR:$Vdm, shr_imm32:$imm, pred:$p)>;
6593 def : NEONInstAlias<"vsra${p}.s64 $Vdm, $imm",
6594 (VSRAsv2i64 QPR:$Vdm, QPR:$Vdm, shr_imm64:$imm, pred:$p)>;
6597 def : NEONInstAlias<"vsra${p}.u8 $Vdm, $imm",
6598 (VSRAuv8i8 DPR:$Vdm, DPR:$Vdm, shr_imm8:$imm, pred:$p)>;
6599 def : NEONInstAlias<"vsra${p}.u16 $Vdm, $imm",
6600 (VSRAuv4i16 DPR:$Vdm, DPR:$Vdm, shr_imm16:$imm, pred:$p)>;
6601 def : NEONInstAlias<"vsra${p}.u32 $Vdm, $imm",
6602 (VSRAuv2i32 DPR:$Vdm, DPR:$Vdm, shr_imm32:$imm, pred:$p)>;
6603 def : NEONInstAlias<"vsra${p}.u64 $Vdm, $imm",
6604 (VSRAuv1i64 DPR:$Vdm, DPR:$Vdm, shr_imm64:$imm, pred:$p)>;
6606 def : NEONInstAlias<"vsra${p}.u8 $Vdm, $imm",
6607 (VSRAuv16i8 QPR:$Vdm, QPR:$Vdm, shr_imm8:$imm, pred:$p)>;
6608 def : NEONInstAlias<"vsra${p}.u16 $Vdm, $imm",
6609 (VSRAuv8i16 QPR:$Vdm, QPR:$Vdm, shr_imm16:$imm, pred:$p)>;
6610 def : NEONInstAlias<"vsra${p}.u32 $Vdm, $imm",
6611 (VSRAuv4i32 QPR:$Vdm, QPR:$Vdm, shr_imm32:$imm, pred:$p)>;
6612 def : NEONInstAlias<"vsra${p}.u64 $Vdm, $imm",
6613 (VSRAuv2i64 QPR:$Vdm, QPR:$Vdm, shr_imm64:$imm, pred:$p)>;
6615 // Two-operand variants for VSRI.
6616 def : NEONInstAlias<"vsri${p}.8 $Vdm, $imm",
6617 (VSRIv8i8 DPR:$Vdm, DPR:$Vdm, shr_imm8:$imm, pred:$p)>;
6618 def : NEONInstAlias<"vsri${p}.16 $Vdm, $imm",
6619 (VSRIv4i16 DPR:$Vdm, DPR:$Vdm, shr_imm16:$imm, pred:$p)>;
6620 def : NEONInstAlias<"vsri${p}.32 $Vdm, $imm",
6621 (VSRIv2i32 DPR:$Vdm, DPR:$Vdm, shr_imm32:$imm, pred:$p)>;
6622 def : NEONInstAlias<"vsri${p}.64 $Vdm, $imm",
6623 (VSRIv1i64 DPR:$Vdm, DPR:$Vdm, shr_imm64:$imm, pred:$p)>;
6625 def : NEONInstAlias<"vsri${p}.8 $Vdm, $imm",
6626 (VSRIv16i8 QPR:$Vdm, QPR:$Vdm, shr_imm8:$imm, pred:$p)>;
6627 def : NEONInstAlias<"vsri${p}.16 $Vdm, $imm",
6628 (VSRIv8i16 QPR:$Vdm, QPR:$Vdm, shr_imm16:$imm, pred:$p)>;
6629 def : NEONInstAlias<"vsri${p}.32 $Vdm, $imm",
6630 (VSRIv4i32 QPR:$Vdm, QPR:$Vdm, shr_imm32:$imm, pred:$p)>;
6631 def : NEONInstAlias<"vsri${p}.64 $Vdm, $imm",
6632 (VSRIv2i64 QPR:$Vdm, QPR:$Vdm, shr_imm64:$imm, pred:$p)>;
6634 // Two-operand variants for VSLI.
6635 def : NEONInstAlias<"vsli${p}.8 $Vdm, $imm",
6636 (VSLIv8i8 DPR:$Vdm, DPR:$Vdm, shr_imm8:$imm, pred:$p)>;
6637 def : NEONInstAlias<"vsli${p}.16 $Vdm, $imm",
6638 (VSLIv4i16 DPR:$Vdm, DPR:$Vdm, shr_imm16:$imm, pred:$p)>;
6639 def : NEONInstAlias<"vsli${p}.32 $Vdm, $imm",
6640 (VSLIv2i32 DPR:$Vdm, DPR:$Vdm, shr_imm32:$imm, pred:$p)>;
6641 def : NEONInstAlias<"vsli${p}.64 $Vdm, $imm",
6642 (VSLIv1i64 DPR:$Vdm, DPR:$Vdm, shr_imm64:$imm, pred:$p)>;
6644 def : NEONInstAlias<"vsli${p}.8 $Vdm, $imm",
6645 (VSLIv16i8 QPR:$Vdm, QPR:$Vdm, shr_imm8:$imm, pred:$p)>;
6646 def : NEONInstAlias<"vsli${p}.16 $Vdm, $imm",
6647 (VSLIv8i16 QPR:$Vdm, QPR:$Vdm, shr_imm16:$imm, pred:$p)>;
6648 def : NEONInstAlias<"vsli${p}.32 $Vdm, $imm",
6649 (VSLIv4i32 QPR:$Vdm, QPR:$Vdm, shr_imm32:$imm, pred:$p)>;
6650 def : NEONInstAlias<"vsli${p}.64 $Vdm, $imm",
6651 (VSLIv2i64 QPR:$Vdm, QPR:$Vdm, shr_imm64:$imm, pred:$p)>;
6653 // VSWP allows, but does not require, a type suffix.
6654 defm : NEONDTAnyInstAlias<"vswp${p}", "$Vd, $Vm",
6655 (VSWPd DPR:$Vd, DPR:$Vm, pred:$p)>;
6656 defm : NEONDTAnyInstAlias<"vswp${p}", "$Vd, $Vm",
6657 (VSWPq QPR:$Vd, QPR:$Vm, pred:$p)>;
6659 // VBIF, VBIT, and VBSL allow, but do not require, a type suffix.
6660 defm : NEONDTAnyInstAlias<"vbif${p}", "$Vd, $Vn, $Vm",
6661 (VBIFd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
6662 defm : NEONDTAnyInstAlias<"vbit${p}", "$Vd, $Vn, $Vm",
6663 (VBITd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
6664 defm : NEONDTAnyInstAlias<"vbsl${p}", "$Vd, $Vn, $Vm",
6665 (VBSLd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
6666 defm : NEONDTAnyInstAlias<"vbif${p}", "$Vd, $Vn, $Vm",
6667 (VBIFq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
6668 defm : NEONDTAnyInstAlias<"vbit${p}", "$Vd, $Vn, $Vm",
6669 (VBITq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
6670 defm : NEONDTAnyInstAlias<"vbsl${p}", "$Vd, $Vn, $Vm",
6671 (VBSLq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
6673 // "vmov Rd, #-imm" can be handled via "vmvn".
6674 def : NEONInstAlias<"vmov${p}.i32 $Vd, $imm",
6675 (VMVNv2i32 DPR:$Vd, nImmVMOVI32Neg:$imm, pred:$p)>;
6676 def : NEONInstAlias<"vmov${p}.i32 $Vd, $imm",
6677 (VMVNv4i32 QPR:$Vd, nImmVMOVI32Neg:$imm, pred:$p)>;
6678 def : NEONInstAlias<"vmvn${p}.i32 $Vd, $imm",
6679 (VMOVv2i32 DPR:$Vd, nImmVMOVI32Neg:$imm, pred:$p)>;
6680 def : NEONInstAlias<"vmvn${p}.i32 $Vd, $imm",
6681 (VMOVv4i32 QPR:$Vd, nImmVMOVI32Neg:$imm, pred:$p)>;
6683 // 'gas' compatibility aliases for quad-word instructions. Strictly speaking,
6684 // these should restrict to just the Q register variants, but the register
6685 // classes are enough to match correctly regardless, so we keep it simple
6686 // and just use MnemonicAlias.
6687 def : NEONMnemonicAlias<"vbicq", "vbic">;
6688 def : NEONMnemonicAlias<"vandq", "vand">;
6689 def : NEONMnemonicAlias<"veorq", "veor">;
6690 def : NEONMnemonicAlias<"vorrq", "vorr">;
6692 def : NEONMnemonicAlias<"vmovq", "vmov">;
6693 def : NEONMnemonicAlias<"vmvnq", "vmvn">;
6694 // Explicit versions for floating point so that the FPImm variants get
6695 // handled early. The parser gets confused otherwise.
6696 def : NEONMnemonicAlias<"vmovq.f32", "vmov.f32">;
6697 def : NEONMnemonicAlias<"vmovq.f64", "vmov.f64">;
6699 def : NEONMnemonicAlias<"vaddq", "vadd">;
6700 def : NEONMnemonicAlias<"vsubq", "vsub">;
6702 def : NEONMnemonicAlias<"vminq", "vmin">;
6703 def : NEONMnemonicAlias<"vmaxq", "vmax">;
6705 def : NEONMnemonicAlias<"vmulq", "vmul">;
6707 def : NEONMnemonicAlias<"vabsq", "vabs">;
6709 def : NEONMnemonicAlias<"vshlq", "vshl">;
6710 def : NEONMnemonicAlias<"vshrq", "vshr">;
6712 def : NEONMnemonicAlias<"vcvtq", "vcvt">;
6714 def : NEONMnemonicAlias<"vcleq", "vcle">;
6715 def : NEONMnemonicAlias<"vceqq", "vceq">;
6717 def : NEONMnemonicAlias<"vzipq", "vzip">;
6718 def : NEONMnemonicAlias<"vswpq", "vswp">;
6720 def : NEONMnemonicAlias<"vrecpeq.f32", "vrecpe.f32">;
6721 def : NEONMnemonicAlias<"vrecpeq.u32", "vrecpe.u32">;
6724 // Alias for loading floating point immediates that aren't representable
6725 // using the vmov.f32 encoding but the bitpattern is representable using
6726 // the .i32 encoding.
6727 def : NEONInstAlias<"vmov${p}.f32 $Vd, $imm",
6728 (VMOVv4i32 QPR:$Vd, nImmVMOVI32:$imm, pred:$p)>;
6729 def : NEONInstAlias<"vmov${p}.f32 $Vd, $imm",
6730 (VMOVv2i32 DPR:$Vd, nImmVMOVI32:$imm, pred:$p)>;