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, "printVectorListDPair"> {
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 VecListTwoQAsmOperand : AsmOperandClass {
120 let Name = "VecListTwoQ";
121 let ParserMethod = "parseVectorList";
122 let RenderMethod = "addVecListOperands";
124 def VecListTwoQ : RegisterOperand<DPR, "printVectorListTwoSpaced"> {
125 let ParserMatchClass = VecListTwoQAsmOperand;
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 VecListTwoDAllLanesAsmOperand : AsmOperandClass {
157 let Name = "VecListTwoDAllLanes";
158 let ParserMethod = "parseVectorList";
159 let RenderMethod = "addVecListOperands";
161 def VecListTwoDAllLanes : RegisterOperand<DPR, "printVectorListTwoAllLanes"> {
162 let ParserMatchClass = VecListTwoDAllLanesAsmOperand;
164 // Register list of two D registers spaced by 2 (two sequential Q registers).
165 def VecListTwoQAllLanesAsmOperand : AsmOperandClass {
166 let Name = "VecListTwoQAllLanes";
167 let ParserMethod = "parseVectorList";
168 let RenderMethod = "addVecListOperands";
170 def VecListTwoQAllLanes : RegisterOperand<DPR,
171 "printVectorListTwoSpacedAllLanes"> {
172 let ParserMatchClass = VecListTwoQAllLanesAsmOperand;
174 // Register list of three D registers, with "all lanes" subscripting.
175 def VecListThreeDAllLanesAsmOperand : AsmOperandClass {
176 let Name = "VecListThreeDAllLanes";
177 let ParserMethod = "parseVectorList";
178 let RenderMethod = "addVecListOperands";
180 def VecListThreeDAllLanes : RegisterOperand<DPR,
181 "printVectorListThreeAllLanes"> {
182 let ParserMatchClass = VecListThreeDAllLanesAsmOperand;
184 // Register list of three D registers spaced by 2 (three sequential Q regs).
185 def VecListThreeQAllLanesAsmOperand : AsmOperandClass {
186 let Name = "VecListThreeQAllLanes";
187 let ParserMethod = "parseVectorList";
188 let RenderMethod = "addVecListOperands";
190 def VecListThreeQAllLanes : RegisterOperand<DPR,
191 "printVectorListThreeSpacedAllLanes"> {
192 let ParserMatchClass = VecListThreeQAllLanesAsmOperand;
194 // Register list of four D registers, with "all lanes" subscripting.
195 def VecListFourDAllLanesAsmOperand : AsmOperandClass {
196 let Name = "VecListFourDAllLanes";
197 let ParserMethod = "parseVectorList";
198 let RenderMethod = "addVecListOperands";
200 def VecListFourDAllLanes : RegisterOperand<DPR, "printVectorListFourAllLanes"> {
201 let ParserMatchClass = VecListFourDAllLanesAsmOperand;
203 // Register list of four D registers spaced by 2 (four sequential Q regs).
204 def VecListFourQAllLanesAsmOperand : AsmOperandClass {
205 let Name = "VecListFourQAllLanes";
206 let ParserMethod = "parseVectorList";
207 let RenderMethod = "addVecListOperands";
209 def VecListFourQAllLanes : RegisterOperand<DPR,
210 "printVectorListFourSpacedAllLanes"> {
211 let ParserMatchClass = VecListFourQAllLanesAsmOperand;
215 // Register list of one D register, with byte lane subscripting.
216 def VecListOneDByteIndexAsmOperand : AsmOperandClass {
217 let Name = "VecListOneDByteIndexed";
218 let ParserMethod = "parseVectorList";
219 let RenderMethod = "addVecListIndexedOperands";
221 def VecListOneDByteIndexed : Operand<i32> {
222 let ParserMatchClass = VecListOneDByteIndexAsmOperand;
223 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
225 // ...with half-word lane subscripting.
226 def VecListOneDHWordIndexAsmOperand : AsmOperandClass {
227 let Name = "VecListOneDHWordIndexed";
228 let ParserMethod = "parseVectorList";
229 let RenderMethod = "addVecListIndexedOperands";
231 def VecListOneDHWordIndexed : Operand<i32> {
232 let ParserMatchClass = VecListOneDHWordIndexAsmOperand;
233 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
235 // ...with word lane subscripting.
236 def VecListOneDWordIndexAsmOperand : AsmOperandClass {
237 let Name = "VecListOneDWordIndexed";
238 let ParserMethod = "parseVectorList";
239 let RenderMethod = "addVecListIndexedOperands";
241 def VecListOneDWordIndexed : Operand<i32> {
242 let ParserMatchClass = VecListOneDWordIndexAsmOperand;
243 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
246 // Register list of two D registers with byte lane subscripting.
247 def VecListTwoDByteIndexAsmOperand : AsmOperandClass {
248 let Name = "VecListTwoDByteIndexed";
249 let ParserMethod = "parseVectorList";
250 let RenderMethod = "addVecListIndexedOperands";
252 def VecListTwoDByteIndexed : Operand<i32> {
253 let ParserMatchClass = VecListTwoDByteIndexAsmOperand;
254 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
256 // ...with half-word lane subscripting.
257 def VecListTwoDHWordIndexAsmOperand : AsmOperandClass {
258 let Name = "VecListTwoDHWordIndexed";
259 let ParserMethod = "parseVectorList";
260 let RenderMethod = "addVecListIndexedOperands";
262 def VecListTwoDHWordIndexed : Operand<i32> {
263 let ParserMatchClass = VecListTwoDHWordIndexAsmOperand;
264 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
266 // ...with word lane subscripting.
267 def VecListTwoDWordIndexAsmOperand : AsmOperandClass {
268 let Name = "VecListTwoDWordIndexed";
269 let ParserMethod = "parseVectorList";
270 let RenderMethod = "addVecListIndexedOperands";
272 def VecListTwoDWordIndexed : Operand<i32> {
273 let ParserMatchClass = VecListTwoDWordIndexAsmOperand;
274 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
276 // Register list of two Q registers with half-word lane subscripting.
277 def VecListTwoQHWordIndexAsmOperand : AsmOperandClass {
278 let Name = "VecListTwoQHWordIndexed";
279 let ParserMethod = "parseVectorList";
280 let RenderMethod = "addVecListIndexedOperands";
282 def VecListTwoQHWordIndexed : Operand<i32> {
283 let ParserMatchClass = VecListTwoQHWordIndexAsmOperand;
284 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
286 // ...with word lane subscripting.
287 def VecListTwoQWordIndexAsmOperand : AsmOperandClass {
288 let Name = "VecListTwoQWordIndexed";
289 let ParserMethod = "parseVectorList";
290 let RenderMethod = "addVecListIndexedOperands";
292 def VecListTwoQWordIndexed : Operand<i32> {
293 let ParserMatchClass = VecListTwoQWordIndexAsmOperand;
294 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
298 // Register list of three D registers with byte lane subscripting.
299 def VecListThreeDByteIndexAsmOperand : AsmOperandClass {
300 let Name = "VecListThreeDByteIndexed";
301 let ParserMethod = "parseVectorList";
302 let RenderMethod = "addVecListIndexedOperands";
304 def VecListThreeDByteIndexed : Operand<i32> {
305 let ParserMatchClass = VecListThreeDByteIndexAsmOperand;
306 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
308 // ...with half-word lane subscripting.
309 def VecListThreeDHWordIndexAsmOperand : AsmOperandClass {
310 let Name = "VecListThreeDHWordIndexed";
311 let ParserMethod = "parseVectorList";
312 let RenderMethod = "addVecListIndexedOperands";
314 def VecListThreeDHWordIndexed : Operand<i32> {
315 let ParserMatchClass = VecListThreeDHWordIndexAsmOperand;
316 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
318 // ...with word lane subscripting.
319 def VecListThreeDWordIndexAsmOperand : AsmOperandClass {
320 let Name = "VecListThreeDWordIndexed";
321 let ParserMethod = "parseVectorList";
322 let RenderMethod = "addVecListIndexedOperands";
324 def VecListThreeDWordIndexed : Operand<i32> {
325 let ParserMatchClass = VecListThreeDWordIndexAsmOperand;
326 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
328 // Register list of three Q registers with half-word lane subscripting.
329 def VecListThreeQHWordIndexAsmOperand : AsmOperandClass {
330 let Name = "VecListThreeQHWordIndexed";
331 let ParserMethod = "parseVectorList";
332 let RenderMethod = "addVecListIndexedOperands";
334 def VecListThreeQHWordIndexed : Operand<i32> {
335 let ParserMatchClass = VecListThreeQHWordIndexAsmOperand;
336 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
338 // ...with word lane subscripting.
339 def VecListThreeQWordIndexAsmOperand : AsmOperandClass {
340 let Name = "VecListThreeQWordIndexed";
341 let ParserMethod = "parseVectorList";
342 let RenderMethod = "addVecListIndexedOperands";
344 def VecListThreeQWordIndexed : Operand<i32> {
345 let ParserMatchClass = VecListThreeQWordIndexAsmOperand;
346 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
349 // Register list of four D registers with byte lane subscripting.
350 def VecListFourDByteIndexAsmOperand : AsmOperandClass {
351 let Name = "VecListFourDByteIndexed";
352 let ParserMethod = "parseVectorList";
353 let RenderMethod = "addVecListIndexedOperands";
355 def VecListFourDByteIndexed : Operand<i32> {
356 let ParserMatchClass = VecListFourDByteIndexAsmOperand;
357 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
359 // ...with half-word lane subscripting.
360 def VecListFourDHWordIndexAsmOperand : AsmOperandClass {
361 let Name = "VecListFourDHWordIndexed";
362 let ParserMethod = "parseVectorList";
363 let RenderMethod = "addVecListIndexedOperands";
365 def VecListFourDHWordIndexed : Operand<i32> {
366 let ParserMatchClass = VecListFourDHWordIndexAsmOperand;
367 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
369 // ...with word lane subscripting.
370 def VecListFourDWordIndexAsmOperand : AsmOperandClass {
371 let Name = "VecListFourDWordIndexed";
372 let ParserMethod = "parseVectorList";
373 let RenderMethod = "addVecListIndexedOperands";
375 def VecListFourDWordIndexed : Operand<i32> {
376 let ParserMatchClass = VecListFourDWordIndexAsmOperand;
377 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
379 // Register list of four Q registers with half-word lane subscripting.
380 def VecListFourQHWordIndexAsmOperand : AsmOperandClass {
381 let Name = "VecListFourQHWordIndexed";
382 let ParserMethod = "parseVectorList";
383 let RenderMethod = "addVecListIndexedOperands";
385 def VecListFourQHWordIndexed : Operand<i32> {
386 let ParserMatchClass = VecListFourQHWordIndexAsmOperand;
387 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
389 // ...with word lane subscripting.
390 def VecListFourQWordIndexAsmOperand : AsmOperandClass {
391 let Name = "VecListFourQWordIndexed";
392 let ParserMethod = "parseVectorList";
393 let RenderMethod = "addVecListIndexedOperands";
395 def VecListFourQWordIndexed : Operand<i32> {
396 let ParserMatchClass = VecListFourQWordIndexAsmOperand;
397 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
401 //===----------------------------------------------------------------------===//
402 // NEON-specific DAG Nodes.
403 //===----------------------------------------------------------------------===//
405 def SDTARMVCMP : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisSameAs<1, 2>]>;
406 def SDTARMVCMPZ : SDTypeProfile<1, 1, []>;
408 def NEONvceq : SDNode<"ARMISD::VCEQ", SDTARMVCMP>;
409 def NEONvceqz : SDNode<"ARMISD::VCEQZ", SDTARMVCMPZ>;
410 def NEONvcge : SDNode<"ARMISD::VCGE", SDTARMVCMP>;
411 def NEONvcgez : SDNode<"ARMISD::VCGEZ", SDTARMVCMPZ>;
412 def NEONvclez : SDNode<"ARMISD::VCLEZ", SDTARMVCMPZ>;
413 def NEONvcgeu : SDNode<"ARMISD::VCGEU", SDTARMVCMP>;
414 def NEONvcgt : SDNode<"ARMISD::VCGT", SDTARMVCMP>;
415 def NEONvcgtz : SDNode<"ARMISD::VCGTZ", SDTARMVCMPZ>;
416 def NEONvcltz : SDNode<"ARMISD::VCLTZ", SDTARMVCMPZ>;
417 def NEONvcgtu : SDNode<"ARMISD::VCGTU", SDTARMVCMP>;
418 def NEONvtst : SDNode<"ARMISD::VTST", SDTARMVCMP>;
420 // Types for vector shift by immediates. The "SHX" version is for long and
421 // narrow operations where the source and destination vectors have different
422 // types. The "SHINS" version is for shift and insert operations.
423 def SDTARMVSH : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
425 def SDTARMVSHX : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
427 def SDTARMVSHINS : SDTypeProfile<1, 3, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
428 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
430 def NEONvshl : SDNode<"ARMISD::VSHL", SDTARMVSH>;
431 def NEONvshrs : SDNode<"ARMISD::VSHRs", SDTARMVSH>;
432 def NEONvshru : SDNode<"ARMISD::VSHRu", SDTARMVSH>;
433 def NEONvshlls : SDNode<"ARMISD::VSHLLs", SDTARMVSHX>;
434 def NEONvshllu : SDNode<"ARMISD::VSHLLu", SDTARMVSHX>;
435 def NEONvshlli : SDNode<"ARMISD::VSHLLi", SDTARMVSHX>;
436 def NEONvshrn : SDNode<"ARMISD::VSHRN", SDTARMVSHX>;
438 def NEONvrshrs : SDNode<"ARMISD::VRSHRs", SDTARMVSH>;
439 def NEONvrshru : SDNode<"ARMISD::VRSHRu", SDTARMVSH>;
440 def NEONvrshrn : SDNode<"ARMISD::VRSHRN", SDTARMVSHX>;
442 def NEONvqshls : SDNode<"ARMISD::VQSHLs", SDTARMVSH>;
443 def NEONvqshlu : SDNode<"ARMISD::VQSHLu", SDTARMVSH>;
444 def NEONvqshlsu : SDNode<"ARMISD::VQSHLsu", SDTARMVSH>;
445 def NEONvqshrns : SDNode<"ARMISD::VQSHRNs", SDTARMVSHX>;
446 def NEONvqshrnu : SDNode<"ARMISD::VQSHRNu", SDTARMVSHX>;
447 def NEONvqshrnsu : SDNode<"ARMISD::VQSHRNsu", SDTARMVSHX>;
449 def NEONvqrshrns : SDNode<"ARMISD::VQRSHRNs", SDTARMVSHX>;
450 def NEONvqrshrnu : SDNode<"ARMISD::VQRSHRNu", SDTARMVSHX>;
451 def NEONvqrshrnsu : SDNode<"ARMISD::VQRSHRNsu", SDTARMVSHX>;
453 def NEONvsli : SDNode<"ARMISD::VSLI", SDTARMVSHINS>;
454 def NEONvsri : SDNode<"ARMISD::VSRI", SDTARMVSHINS>;
456 def SDTARMVGETLN : SDTypeProfile<1, 2, [SDTCisVT<0, i32>, SDTCisInt<1>,
458 def NEONvgetlaneu : SDNode<"ARMISD::VGETLANEu", SDTARMVGETLN>;
459 def NEONvgetlanes : SDNode<"ARMISD::VGETLANEs", SDTARMVGETLN>;
461 def SDTARMVMOVIMM : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisVT<1, i32>]>;
462 def NEONvmovImm : SDNode<"ARMISD::VMOVIMM", SDTARMVMOVIMM>;
463 def NEONvmvnImm : SDNode<"ARMISD::VMVNIMM", SDTARMVMOVIMM>;
464 def NEONvmovFPImm : SDNode<"ARMISD::VMOVFPIMM", SDTARMVMOVIMM>;
466 def SDTARMVORRIMM : SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
468 def NEONvorrImm : SDNode<"ARMISD::VORRIMM", SDTARMVORRIMM>;
469 def NEONvbicImm : SDNode<"ARMISD::VBICIMM", SDTARMVORRIMM>;
471 def NEONvbsl : SDNode<"ARMISD::VBSL",
472 SDTypeProfile<1, 3, [SDTCisVec<0>,
475 SDTCisSameAs<0, 3>]>>;
477 def NEONvdup : SDNode<"ARMISD::VDUP", SDTypeProfile<1, 1, [SDTCisVec<0>]>>;
479 // VDUPLANE can produce a quad-register result from a double-register source,
480 // so the result is not constrained to match the source.
481 def NEONvduplane : SDNode<"ARMISD::VDUPLANE",
482 SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisVec<1>,
485 def SDTARMVEXT : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
486 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
487 def NEONvext : SDNode<"ARMISD::VEXT", SDTARMVEXT>;
489 def SDTARMVSHUF : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisSameAs<0, 1>]>;
490 def NEONvrev64 : SDNode<"ARMISD::VREV64", SDTARMVSHUF>;
491 def NEONvrev32 : SDNode<"ARMISD::VREV32", SDTARMVSHUF>;
492 def NEONvrev16 : SDNode<"ARMISD::VREV16", SDTARMVSHUF>;
494 def SDTARMVSHUF2 : SDTypeProfile<2, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
496 SDTCisSameAs<0, 3>]>;
497 def NEONzip : SDNode<"ARMISD::VZIP", SDTARMVSHUF2>;
498 def NEONuzp : SDNode<"ARMISD::VUZP", SDTARMVSHUF2>;
499 def NEONtrn : SDNode<"ARMISD::VTRN", SDTARMVSHUF2>;
501 def SDTARMVMULL : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
502 SDTCisSameAs<1, 2>]>;
503 def NEONvmulls : SDNode<"ARMISD::VMULLs", SDTARMVMULL>;
504 def NEONvmullu : SDNode<"ARMISD::VMULLu", SDTARMVMULL>;
506 def SDTARMFMAX : SDTypeProfile<1, 2, [SDTCisVT<0, f32>, SDTCisSameAs<0, 1>,
507 SDTCisSameAs<0, 2>]>;
508 def NEONfmax : SDNode<"ARMISD::FMAX", SDTARMFMAX>;
509 def NEONfmin : SDNode<"ARMISD::FMIN", SDTARMFMAX>;
511 def NEONimmAllZerosV: PatLeaf<(NEONvmovImm (i32 timm)), [{
512 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
513 unsigned EltBits = 0;
514 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
515 return (EltBits == 32 && EltVal == 0);
518 def NEONimmAllOnesV: PatLeaf<(NEONvmovImm (i32 timm)), [{
519 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
520 unsigned EltBits = 0;
521 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
522 return (EltBits == 8 && EltVal == 0xff);
525 //===----------------------------------------------------------------------===//
526 // NEON load / store instructions
527 //===----------------------------------------------------------------------===//
529 // Use VLDM to load a Q register as a D register pair.
530 // This is a pseudo instruction that is expanded to VLDMD after reg alloc.
532 : PseudoVFPLdStM<(outs QPR:$dst), (ins GPR:$Rn),
534 [(set QPR:$dst, (v2f64 (load GPR:$Rn)))]>;
536 // Use VSTM to store a Q register as a D register pair.
537 // This is a pseudo instruction that is expanded to VSTMD after reg alloc.
539 : PseudoVFPLdStM<(outs), (ins QPR:$src, GPR:$Rn),
541 [(store (v2f64 QPR:$src), GPR:$Rn)]>;
543 // Classes for VLD* pseudo-instructions with multi-register operands.
544 // These are expanded to real instructions after register allocation.
545 class VLDQPseudo<InstrItinClass itin>
546 : PseudoNLdSt<(outs QPR:$dst), (ins addrmode6:$addr), itin, "">;
547 class VLDQWBPseudo<InstrItinClass itin>
548 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
549 (ins addrmode6:$addr, am6offset:$offset), itin,
551 class VLDQWBfixedPseudo<InstrItinClass itin>
552 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
553 (ins addrmode6:$addr), itin,
555 class VLDQWBregisterPseudo<InstrItinClass itin>
556 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
557 (ins addrmode6:$addr, rGPR:$offset), itin,
560 class VLDQQPseudo<InstrItinClass itin>
561 : PseudoNLdSt<(outs QQPR:$dst), (ins addrmode6:$addr), itin, "">;
562 class VLDQQWBPseudo<InstrItinClass itin>
563 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
564 (ins addrmode6:$addr, am6offset:$offset), itin,
566 class VLDQQWBfixedPseudo<InstrItinClass itin>
567 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
568 (ins addrmode6:$addr), itin,
570 class VLDQQWBregisterPseudo<InstrItinClass itin>
571 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
572 (ins addrmode6:$addr, rGPR:$offset), itin,
576 class VLDQQQQPseudo<InstrItinClass itin>
577 : PseudoNLdSt<(outs QQQQPR:$dst), (ins addrmode6:$addr, QQQQPR:$src),itin,
579 class VLDQQQQWBPseudo<InstrItinClass itin>
580 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
581 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
582 "$addr.addr = $wb, $src = $dst">;
584 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
586 // VLD1 : Vector Load (multiple single elements)
587 class VLD1D<bits<4> op7_4, string Dt>
588 : NLdSt<0,0b10,0b0111,op7_4, (outs VecListOneD:$Vd),
589 (ins addrmode6:$Rn), IIC_VLD1,
590 "vld1", Dt, "$Vd, $Rn", "", []> {
593 let DecoderMethod = "DecodeVLDInstruction";
595 class VLD1Q<bits<4> op7_4, string Dt>
596 : NLdSt<0,0b10,0b1010,op7_4, (outs VecListDPair:$Vd),
597 (ins addrmode6:$Rn), IIC_VLD1x2,
598 "vld1", Dt, "$Vd, $Rn", "", []> {
600 let Inst{5-4} = Rn{5-4};
601 let DecoderMethod = "DecodeVLDInstruction";
604 def VLD1d8 : VLD1D<{0,0,0,?}, "8">;
605 def VLD1d16 : VLD1D<{0,1,0,?}, "16">;
606 def VLD1d32 : VLD1D<{1,0,0,?}, "32">;
607 def VLD1d64 : VLD1D<{1,1,0,?}, "64">;
609 def VLD1q8 : VLD1Q<{0,0,?,?}, "8">;
610 def VLD1q16 : VLD1Q<{0,1,?,?}, "16">;
611 def VLD1q32 : VLD1Q<{1,0,?,?}, "32">;
612 def VLD1q64 : VLD1Q<{1,1,?,?}, "64">;
614 // ...with address register writeback:
615 multiclass VLD1DWB<bits<4> op7_4, string Dt> {
616 def _fixed : NLdSt<0,0b10, 0b0111,op7_4, (outs VecListOneD:$Vd, GPR:$wb),
617 (ins addrmode6:$Rn), IIC_VLD1u,
618 "vld1", Dt, "$Vd, $Rn!",
619 "$Rn.addr = $wb", []> {
620 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
622 let DecoderMethod = "DecodeVLDInstruction";
623 let AsmMatchConverter = "cvtVLDwbFixed";
625 def _register : NLdSt<0,0b10,0b0111,op7_4, (outs VecListOneD:$Vd, GPR:$wb),
626 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1u,
627 "vld1", Dt, "$Vd, $Rn, $Rm",
628 "$Rn.addr = $wb", []> {
630 let DecoderMethod = "DecodeVLDInstruction";
631 let AsmMatchConverter = "cvtVLDwbRegister";
634 multiclass VLD1QWB<bits<4> op7_4, string Dt> {
635 def _fixed : NLdSt<0,0b10,0b1010,op7_4, (outs VecListDPair:$Vd, GPR:$wb),
636 (ins addrmode6:$Rn), IIC_VLD1x2u,
637 "vld1", Dt, "$Vd, $Rn!",
638 "$Rn.addr = $wb", []> {
639 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
640 let Inst{5-4} = Rn{5-4};
641 let DecoderMethod = "DecodeVLDInstruction";
642 let AsmMatchConverter = "cvtVLDwbFixed";
644 def _register : NLdSt<0,0b10,0b1010,op7_4, (outs VecListDPair:$Vd, GPR:$wb),
645 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1x2u,
646 "vld1", Dt, "$Vd, $Rn, $Rm",
647 "$Rn.addr = $wb", []> {
648 let Inst{5-4} = Rn{5-4};
649 let DecoderMethod = "DecodeVLDInstruction";
650 let AsmMatchConverter = "cvtVLDwbRegister";
654 defm VLD1d8wb : VLD1DWB<{0,0,0,?}, "8">;
655 defm VLD1d16wb : VLD1DWB<{0,1,0,?}, "16">;
656 defm VLD1d32wb : VLD1DWB<{1,0,0,?}, "32">;
657 defm VLD1d64wb : VLD1DWB<{1,1,0,?}, "64">;
658 defm VLD1q8wb : VLD1QWB<{0,0,?,?}, "8">;
659 defm VLD1q16wb : VLD1QWB<{0,1,?,?}, "16">;
660 defm VLD1q32wb : VLD1QWB<{1,0,?,?}, "32">;
661 defm VLD1q64wb : VLD1QWB<{1,1,?,?}, "64">;
663 // ...with 3 registers
664 class VLD1D3<bits<4> op7_4, string Dt>
665 : NLdSt<0,0b10,0b0110,op7_4, (outs VecListThreeD:$Vd),
666 (ins addrmode6:$Rn), IIC_VLD1x3, "vld1", Dt,
667 "$Vd, $Rn", "", []> {
670 let DecoderMethod = "DecodeVLDInstruction";
672 multiclass VLD1D3WB<bits<4> op7_4, string Dt> {
673 def _fixed : NLdSt<0,0b10,0b0110, op7_4, (outs VecListThreeD:$Vd, GPR:$wb),
674 (ins addrmode6:$Rn), IIC_VLD1x2u,
675 "vld1", Dt, "$Vd, $Rn!",
676 "$Rn.addr = $wb", []> {
677 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
679 let DecoderMethod = "DecodeVLDInstruction";
680 let AsmMatchConverter = "cvtVLDwbFixed";
682 def _register : NLdSt<0,0b10,0b0110,op7_4, (outs VecListThreeD:$Vd, GPR:$wb),
683 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1x2u,
684 "vld1", Dt, "$Vd, $Rn, $Rm",
685 "$Rn.addr = $wb", []> {
687 let DecoderMethod = "DecodeVLDInstruction";
688 let AsmMatchConverter = "cvtVLDwbRegister";
692 def VLD1d8T : VLD1D3<{0,0,0,?}, "8">;
693 def VLD1d16T : VLD1D3<{0,1,0,?}, "16">;
694 def VLD1d32T : VLD1D3<{1,0,0,?}, "32">;
695 def VLD1d64T : VLD1D3<{1,1,0,?}, "64">;
697 defm VLD1d8Twb : VLD1D3WB<{0,0,0,?}, "8">;
698 defm VLD1d16Twb : VLD1D3WB<{0,1,0,?}, "16">;
699 defm VLD1d32Twb : VLD1D3WB<{1,0,0,?}, "32">;
700 defm VLD1d64Twb : VLD1D3WB<{1,1,0,?}, "64">;
702 def VLD1d64TPseudo : VLDQQPseudo<IIC_VLD1x3>;
704 // ...with 4 registers
705 class VLD1D4<bits<4> op7_4, string Dt>
706 : NLdSt<0, 0b10, 0b0010, op7_4, (outs VecListFourD:$Vd),
707 (ins addrmode6:$Rn), IIC_VLD1x4, "vld1", Dt,
708 "$Vd, $Rn", "", []> {
710 let Inst{5-4} = Rn{5-4};
711 let DecoderMethod = "DecodeVLDInstruction";
713 multiclass VLD1D4WB<bits<4> op7_4, string Dt> {
714 def _fixed : NLdSt<0,0b10,0b0010, op7_4, (outs VecListFourD:$Vd, GPR:$wb),
715 (ins addrmode6:$Rn), IIC_VLD1x2u,
716 "vld1", Dt, "$Vd, $Rn!",
717 "$Rn.addr = $wb", []> {
718 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
719 let Inst{5-4} = Rn{5-4};
720 let DecoderMethod = "DecodeVLDInstruction";
721 let AsmMatchConverter = "cvtVLDwbFixed";
723 def _register : NLdSt<0,0b10,0b0010,op7_4, (outs VecListFourD:$Vd, GPR:$wb),
724 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1x2u,
725 "vld1", Dt, "$Vd, $Rn, $Rm",
726 "$Rn.addr = $wb", []> {
727 let Inst{5-4} = Rn{5-4};
728 let DecoderMethod = "DecodeVLDInstruction";
729 let AsmMatchConverter = "cvtVLDwbRegister";
733 def VLD1d8Q : VLD1D4<{0,0,?,?}, "8">;
734 def VLD1d16Q : VLD1D4<{0,1,?,?}, "16">;
735 def VLD1d32Q : VLD1D4<{1,0,?,?}, "32">;
736 def VLD1d64Q : VLD1D4<{1,1,?,?}, "64">;
738 defm VLD1d8Qwb : VLD1D4WB<{0,0,?,?}, "8">;
739 defm VLD1d16Qwb : VLD1D4WB<{0,1,?,?}, "16">;
740 defm VLD1d32Qwb : VLD1D4WB<{1,0,?,?}, "32">;
741 defm VLD1d64Qwb : VLD1D4WB<{1,1,?,?}, "64">;
743 def VLD1d64QPseudo : VLDQQPseudo<IIC_VLD1x4>;
745 // VLD2 : Vector Load (multiple 2-element structures)
746 class VLD2<bits<4> op11_8, bits<4> op7_4, string Dt, RegisterOperand VdTy,
748 : NLdSt<0, 0b10, op11_8, op7_4, (outs VdTy:$Vd),
749 (ins addrmode6:$Rn), itin,
750 "vld2", Dt, "$Vd, $Rn", "", []> {
752 let Inst{5-4} = Rn{5-4};
753 let DecoderMethod = "DecodeVLDInstruction";
756 def VLD2d8 : VLD2<0b1000, {0,0,?,?}, "8", VecListDPair, IIC_VLD2>;
757 def VLD2d16 : VLD2<0b1000, {0,1,?,?}, "16", VecListDPair, IIC_VLD2>;
758 def VLD2d32 : VLD2<0b1000, {1,0,?,?}, "32", VecListDPair, IIC_VLD2>;
760 def VLD2q8 : VLD2<0b0011, {0,0,?,?}, "8", VecListFourD, IIC_VLD2x2>;
761 def VLD2q16 : VLD2<0b0011, {0,1,?,?}, "16", VecListFourD, IIC_VLD2x2>;
762 def VLD2q32 : VLD2<0b0011, {1,0,?,?}, "32", VecListFourD, IIC_VLD2x2>;
764 def VLD2q8Pseudo : VLDQQPseudo<IIC_VLD2x2>;
765 def VLD2q16Pseudo : VLDQQPseudo<IIC_VLD2x2>;
766 def VLD2q32Pseudo : VLDQQPseudo<IIC_VLD2x2>;
768 // ...with address register writeback:
769 multiclass VLD2WB<bits<4> op11_8, bits<4> op7_4, string Dt,
770 RegisterOperand VdTy, InstrItinClass itin> {
771 def _fixed : NLdSt<0, 0b10, op11_8, op7_4, (outs VdTy:$Vd, GPR:$wb),
772 (ins addrmode6:$Rn), itin,
773 "vld2", Dt, "$Vd, $Rn!",
774 "$Rn.addr = $wb", []> {
775 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
776 let Inst{5-4} = Rn{5-4};
777 let DecoderMethod = "DecodeVLDInstruction";
778 let AsmMatchConverter = "cvtVLDwbFixed";
780 def _register : NLdSt<0, 0b10, op11_8, op7_4, (outs VdTy:$Vd, GPR:$wb),
781 (ins addrmode6:$Rn, rGPR:$Rm), itin,
782 "vld2", Dt, "$Vd, $Rn, $Rm",
783 "$Rn.addr = $wb", []> {
784 let Inst{5-4} = Rn{5-4};
785 let DecoderMethod = "DecodeVLDInstruction";
786 let AsmMatchConverter = "cvtVLDwbRegister";
790 defm VLD2d8wb : VLD2WB<0b1000, {0,0,?,?}, "8", VecListDPair, IIC_VLD2u>;
791 defm VLD2d16wb : VLD2WB<0b1000, {0,1,?,?}, "16", VecListDPair, IIC_VLD2u>;
792 defm VLD2d32wb : VLD2WB<0b1000, {1,0,?,?}, "32", VecListDPair, IIC_VLD2u>;
794 defm VLD2q8wb : VLD2WB<0b0011, {0,0,?,?}, "8", VecListFourD, IIC_VLD2x2u>;
795 defm VLD2q16wb : VLD2WB<0b0011, {0,1,?,?}, "16", VecListFourD, IIC_VLD2x2u>;
796 defm VLD2q32wb : VLD2WB<0b0011, {1,0,?,?}, "32", VecListFourD, IIC_VLD2x2u>;
798 def VLD2q8PseudoWB_fixed : VLDQQWBfixedPseudo<IIC_VLD2x2u>;
799 def VLD2q16PseudoWB_fixed : VLDQQWBfixedPseudo<IIC_VLD2x2u>;
800 def VLD2q32PseudoWB_fixed : VLDQQWBfixedPseudo<IIC_VLD2x2u>;
801 def VLD2q8PseudoWB_register : VLDQQWBregisterPseudo<IIC_VLD2x2u>;
802 def VLD2q16PseudoWB_register : VLDQQWBregisterPseudo<IIC_VLD2x2u>;
803 def VLD2q32PseudoWB_register : VLDQQWBregisterPseudo<IIC_VLD2x2u>;
805 // ...with double-spaced registers
806 def VLD2b8 : VLD2<0b1001, {0,0,?,?}, "8", VecListTwoQ, IIC_VLD2>;
807 def VLD2b16 : VLD2<0b1001, {0,1,?,?}, "16", VecListTwoQ, IIC_VLD2>;
808 def VLD2b32 : VLD2<0b1001, {1,0,?,?}, "32", VecListTwoQ, IIC_VLD2>;
809 defm VLD2b8wb : VLD2WB<0b1001, {0,0,?,?}, "8", VecListTwoQ, IIC_VLD2u>;
810 defm VLD2b16wb : VLD2WB<0b1001, {0,1,?,?}, "16", VecListTwoQ, IIC_VLD2u>;
811 defm VLD2b32wb : VLD2WB<0b1001, {1,0,?,?}, "32", VecListTwoQ, IIC_VLD2u>;
813 // VLD3 : Vector Load (multiple 3-element structures)
814 class VLD3D<bits<4> op11_8, bits<4> op7_4, string Dt>
815 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
816 (ins addrmode6:$Rn), IIC_VLD3,
817 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn", "", []> {
820 let DecoderMethod = "DecodeVLDInstruction";
823 def VLD3d8 : VLD3D<0b0100, {0,0,0,?}, "8">;
824 def VLD3d16 : VLD3D<0b0100, {0,1,0,?}, "16">;
825 def VLD3d32 : VLD3D<0b0100, {1,0,0,?}, "32">;
827 def VLD3d8Pseudo : VLDQQPseudo<IIC_VLD3>;
828 def VLD3d16Pseudo : VLDQQPseudo<IIC_VLD3>;
829 def VLD3d32Pseudo : VLDQQPseudo<IIC_VLD3>;
831 // ...with address register writeback:
832 class VLD3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
833 : NLdSt<0, 0b10, op11_8, op7_4,
834 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
835 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD3u,
836 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn$Rm",
837 "$Rn.addr = $wb", []> {
839 let DecoderMethod = "DecodeVLDInstruction";
842 def VLD3d8_UPD : VLD3DWB<0b0100, {0,0,0,?}, "8">;
843 def VLD3d16_UPD : VLD3DWB<0b0100, {0,1,0,?}, "16">;
844 def VLD3d32_UPD : VLD3DWB<0b0100, {1,0,0,?}, "32">;
846 def VLD3d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
847 def VLD3d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
848 def VLD3d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
850 // ...with double-spaced registers:
851 def VLD3q8 : VLD3D<0b0101, {0,0,0,?}, "8">;
852 def VLD3q16 : VLD3D<0b0101, {0,1,0,?}, "16">;
853 def VLD3q32 : VLD3D<0b0101, {1,0,0,?}, "32">;
854 def VLD3q8_UPD : VLD3DWB<0b0101, {0,0,0,?}, "8">;
855 def VLD3q16_UPD : VLD3DWB<0b0101, {0,1,0,?}, "16">;
856 def VLD3q32_UPD : VLD3DWB<0b0101, {1,0,0,?}, "32">;
858 def VLD3q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
859 def VLD3q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
860 def VLD3q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
862 // ...alternate versions to be allocated odd register numbers:
863 def VLD3q8oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
864 def VLD3q16oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
865 def VLD3q32oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
867 def VLD3q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
868 def VLD3q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
869 def VLD3q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
871 // VLD4 : Vector Load (multiple 4-element structures)
872 class VLD4D<bits<4> op11_8, bits<4> op7_4, string Dt>
873 : NLdSt<0, 0b10, op11_8, op7_4,
874 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
875 (ins addrmode6:$Rn), IIC_VLD4,
876 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn", "", []> {
878 let Inst{5-4} = Rn{5-4};
879 let DecoderMethod = "DecodeVLDInstruction";
882 def VLD4d8 : VLD4D<0b0000, {0,0,?,?}, "8">;
883 def VLD4d16 : VLD4D<0b0000, {0,1,?,?}, "16">;
884 def VLD4d32 : VLD4D<0b0000, {1,0,?,?}, "32">;
886 def VLD4d8Pseudo : VLDQQPseudo<IIC_VLD4>;
887 def VLD4d16Pseudo : VLDQQPseudo<IIC_VLD4>;
888 def VLD4d32Pseudo : VLDQQPseudo<IIC_VLD4>;
890 // ...with address register writeback:
891 class VLD4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
892 : NLdSt<0, 0b10, op11_8, op7_4,
893 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
894 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD4u,
895 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn$Rm",
896 "$Rn.addr = $wb", []> {
897 let Inst{5-4} = Rn{5-4};
898 let DecoderMethod = "DecodeVLDInstruction";
901 def VLD4d8_UPD : VLD4DWB<0b0000, {0,0,?,?}, "8">;
902 def VLD4d16_UPD : VLD4DWB<0b0000, {0,1,?,?}, "16">;
903 def VLD4d32_UPD : VLD4DWB<0b0000, {1,0,?,?}, "32">;
905 def VLD4d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
906 def VLD4d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
907 def VLD4d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
909 // ...with double-spaced registers:
910 def VLD4q8 : VLD4D<0b0001, {0,0,?,?}, "8">;
911 def VLD4q16 : VLD4D<0b0001, {0,1,?,?}, "16">;
912 def VLD4q32 : VLD4D<0b0001, {1,0,?,?}, "32">;
913 def VLD4q8_UPD : VLD4DWB<0b0001, {0,0,?,?}, "8">;
914 def VLD4q16_UPD : VLD4DWB<0b0001, {0,1,?,?}, "16">;
915 def VLD4q32_UPD : VLD4DWB<0b0001, {1,0,?,?}, "32">;
917 def VLD4q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
918 def VLD4q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
919 def VLD4q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
921 // ...alternate versions to be allocated odd register numbers:
922 def VLD4q8oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
923 def VLD4q16oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
924 def VLD4q32oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
926 def VLD4q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
927 def VLD4q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
928 def VLD4q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
930 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
932 // Classes for VLD*LN pseudo-instructions with multi-register operands.
933 // These are expanded to real instructions after register allocation.
934 class VLDQLNPseudo<InstrItinClass itin>
935 : PseudoNLdSt<(outs QPR:$dst),
936 (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
937 itin, "$src = $dst">;
938 class VLDQLNWBPseudo<InstrItinClass itin>
939 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
940 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
941 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
942 class VLDQQLNPseudo<InstrItinClass itin>
943 : PseudoNLdSt<(outs QQPR:$dst),
944 (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
945 itin, "$src = $dst">;
946 class VLDQQLNWBPseudo<InstrItinClass itin>
947 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
948 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
949 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
950 class VLDQQQQLNPseudo<InstrItinClass itin>
951 : PseudoNLdSt<(outs QQQQPR:$dst),
952 (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
953 itin, "$src = $dst">;
954 class VLDQQQQLNWBPseudo<InstrItinClass itin>
955 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
956 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
957 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
959 // VLD1LN : Vector Load (single element to one lane)
960 class VLD1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
962 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd),
963 (ins addrmode6:$Rn, DPR:$src, nohash_imm:$lane),
964 IIC_VLD1ln, "vld1", Dt, "\\{$Vd[$lane]\\}, $Rn",
966 [(set DPR:$Vd, (vector_insert (Ty DPR:$src),
967 (i32 (LoadOp addrmode6:$Rn)),
970 let DecoderMethod = "DecodeVLD1LN";
972 class VLD1LN32<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
974 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd),
975 (ins addrmode6oneL32:$Rn, DPR:$src, nohash_imm:$lane),
976 IIC_VLD1ln, "vld1", Dt, "\\{$Vd[$lane]\\}, $Rn",
978 [(set DPR:$Vd, (vector_insert (Ty DPR:$src),
979 (i32 (LoadOp addrmode6oneL32:$Rn)),
982 let DecoderMethod = "DecodeVLD1LN";
984 class VLD1QLNPseudo<ValueType Ty, PatFrag LoadOp> : VLDQLNPseudo<IIC_VLD1ln> {
985 let Pattern = [(set QPR:$dst, (vector_insert (Ty QPR:$src),
986 (i32 (LoadOp addrmode6:$addr)),
990 def VLD1LNd8 : VLD1LN<0b0000, {?,?,?,0}, "8", v8i8, extloadi8> {
991 let Inst{7-5} = lane{2-0};
993 def VLD1LNd16 : VLD1LN<0b0100, {?,?,0,?}, "16", v4i16, extloadi16> {
994 let Inst{7-6} = lane{1-0};
995 let Inst{5-4} = Rn{5-4};
997 def VLD1LNd32 : VLD1LN32<0b1000, {?,0,?,?}, "32", v2i32, load> {
998 let Inst{7} = lane{0};
999 let Inst{5-4} = Rn{5-4};
1002 def VLD1LNq8Pseudo : VLD1QLNPseudo<v16i8, extloadi8>;
1003 def VLD1LNq16Pseudo : VLD1QLNPseudo<v8i16, extloadi16>;
1004 def VLD1LNq32Pseudo : VLD1QLNPseudo<v4i32, load>;
1006 def : Pat<(vector_insert (v2f32 DPR:$src),
1007 (f32 (load addrmode6:$addr)), imm:$lane),
1008 (VLD1LNd32 addrmode6:$addr, DPR:$src, imm:$lane)>;
1009 def : Pat<(vector_insert (v4f32 QPR:$src),
1010 (f32 (load addrmode6:$addr)), imm:$lane),
1011 (VLD1LNq32Pseudo addrmode6:$addr, QPR:$src, imm:$lane)>;
1013 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
1015 // ...with address register writeback:
1016 class VLD1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1017 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, GPR:$wb),
1018 (ins addrmode6:$Rn, am6offset:$Rm,
1019 DPR:$src, nohash_imm:$lane), IIC_VLD1lnu, "vld1", Dt,
1020 "\\{$Vd[$lane]\\}, $Rn$Rm",
1021 "$src = $Vd, $Rn.addr = $wb", []> {
1022 let DecoderMethod = "DecodeVLD1LN";
1025 def VLD1LNd8_UPD : VLD1LNWB<0b0000, {?,?,?,0}, "8"> {
1026 let Inst{7-5} = lane{2-0};
1028 def VLD1LNd16_UPD : VLD1LNWB<0b0100, {?,?,0,?}, "16"> {
1029 let Inst{7-6} = lane{1-0};
1030 let Inst{4} = Rn{4};
1032 def VLD1LNd32_UPD : VLD1LNWB<0b1000, {?,0,?,?}, "32"> {
1033 let Inst{7} = lane{0};
1034 let Inst{5} = Rn{4};
1035 let Inst{4} = Rn{4};
1038 def VLD1LNq8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
1039 def VLD1LNq16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
1040 def VLD1LNq32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
1042 // VLD2LN : Vector Load (single 2-element structure to one lane)
1043 class VLD2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1044 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2),
1045 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, nohash_imm:$lane),
1046 IIC_VLD2ln, "vld2", Dt, "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn",
1047 "$src1 = $Vd, $src2 = $dst2", []> {
1049 let Inst{4} = Rn{4};
1050 let DecoderMethod = "DecodeVLD2LN";
1053 def VLD2LNd8 : VLD2LN<0b0001, {?,?,?,?}, "8"> {
1054 let Inst{7-5} = lane{2-0};
1056 def VLD2LNd16 : VLD2LN<0b0101, {?,?,0,?}, "16"> {
1057 let Inst{7-6} = lane{1-0};
1059 def VLD2LNd32 : VLD2LN<0b1001, {?,0,0,?}, "32"> {
1060 let Inst{7} = lane{0};
1063 def VLD2LNd8Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
1064 def VLD2LNd16Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
1065 def VLD2LNd32Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
1067 // ...with double-spaced registers:
1068 def VLD2LNq16 : VLD2LN<0b0101, {?,?,1,?}, "16"> {
1069 let Inst{7-6} = lane{1-0};
1071 def VLD2LNq32 : VLD2LN<0b1001, {?,1,0,?}, "32"> {
1072 let Inst{7} = lane{0};
1075 def VLD2LNq16Pseudo : VLDQQLNPseudo<IIC_VLD2ln>;
1076 def VLD2LNq32Pseudo : VLDQQLNPseudo<IIC_VLD2ln>;
1078 // ...with address register writeback:
1079 class VLD2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1080 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
1081 (ins addrmode6:$Rn, am6offset:$Rm,
1082 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VLD2lnu, "vld2", Dt,
1083 "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn$Rm",
1084 "$src1 = $Vd, $src2 = $dst2, $Rn.addr = $wb", []> {
1085 let Inst{4} = Rn{4};
1086 let DecoderMethod = "DecodeVLD2LN";
1089 def VLD2LNd8_UPD : VLD2LNWB<0b0001, {?,?,?,?}, "8"> {
1090 let Inst{7-5} = lane{2-0};
1092 def VLD2LNd16_UPD : VLD2LNWB<0b0101, {?,?,0,?}, "16"> {
1093 let Inst{7-6} = lane{1-0};
1095 def VLD2LNd32_UPD : VLD2LNWB<0b1001, {?,0,0,?}, "32"> {
1096 let Inst{7} = lane{0};
1099 def VLD2LNd8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
1100 def VLD2LNd16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
1101 def VLD2LNd32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
1103 def VLD2LNq16_UPD : VLD2LNWB<0b0101, {?,?,1,?}, "16"> {
1104 let Inst{7-6} = lane{1-0};
1106 def VLD2LNq32_UPD : VLD2LNWB<0b1001, {?,1,0,?}, "32"> {
1107 let Inst{7} = lane{0};
1110 def VLD2LNq16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
1111 def VLD2LNq32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
1113 // VLD3LN : Vector Load (single 3-element structure to one lane)
1114 class VLD3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1115 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
1116 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3,
1117 nohash_imm:$lane), IIC_VLD3ln, "vld3", Dt,
1118 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn",
1119 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3", []> {
1121 let DecoderMethod = "DecodeVLD3LN";
1124 def VLD3LNd8 : VLD3LN<0b0010, {?,?,?,0}, "8"> {
1125 let Inst{7-5} = lane{2-0};
1127 def VLD3LNd16 : VLD3LN<0b0110, {?,?,0,0}, "16"> {
1128 let Inst{7-6} = lane{1-0};
1130 def VLD3LNd32 : VLD3LN<0b1010, {?,0,0,0}, "32"> {
1131 let Inst{7} = lane{0};
1134 def VLD3LNd8Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
1135 def VLD3LNd16Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
1136 def VLD3LNd32Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
1138 // ...with double-spaced registers:
1139 def VLD3LNq16 : VLD3LN<0b0110, {?,?,1,0}, "16"> {
1140 let Inst{7-6} = lane{1-0};
1142 def VLD3LNq32 : VLD3LN<0b1010, {?,1,0,0}, "32"> {
1143 let Inst{7} = lane{0};
1146 def VLD3LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
1147 def VLD3LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
1149 // ...with address register writeback:
1150 class VLD3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1151 : NLdStLn<1, 0b10, op11_8, op7_4,
1152 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
1153 (ins addrmode6:$Rn, am6offset:$Rm,
1154 DPR:$src1, DPR:$src2, DPR:$src3, nohash_imm:$lane),
1155 IIC_VLD3lnu, "vld3", Dt,
1156 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn$Rm",
1157 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $Rn.addr = $wb",
1159 let DecoderMethod = "DecodeVLD3LN";
1162 def VLD3LNd8_UPD : VLD3LNWB<0b0010, {?,?,?,0}, "8"> {
1163 let Inst{7-5} = lane{2-0};
1165 def VLD3LNd16_UPD : VLD3LNWB<0b0110, {?,?,0,0}, "16"> {
1166 let Inst{7-6} = lane{1-0};
1168 def VLD3LNd32_UPD : VLD3LNWB<0b1010, {?,0,0,0}, "32"> {
1169 let Inst{7} = lane{0};
1172 def VLD3LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
1173 def VLD3LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
1174 def VLD3LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
1176 def VLD3LNq16_UPD : VLD3LNWB<0b0110, {?,?,1,0}, "16"> {
1177 let Inst{7-6} = lane{1-0};
1179 def VLD3LNq32_UPD : VLD3LNWB<0b1010, {?,1,0,0}, "32"> {
1180 let Inst{7} = lane{0};
1183 def VLD3LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
1184 def VLD3LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
1186 // VLD4LN : Vector Load (single 4-element structure to one lane)
1187 class VLD4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1188 : NLdStLn<1, 0b10, op11_8, op7_4,
1189 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
1190 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
1191 nohash_imm:$lane), IIC_VLD4ln, "vld4", Dt,
1192 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn",
1193 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4", []> {
1195 let Inst{4} = Rn{4};
1196 let DecoderMethod = "DecodeVLD4LN";
1199 def VLD4LNd8 : VLD4LN<0b0011, {?,?,?,?}, "8"> {
1200 let Inst{7-5} = lane{2-0};
1202 def VLD4LNd16 : VLD4LN<0b0111, {?,?,0,?}, "16"> {
1203 let Inst{7-6} = lane{1-0};
1205 def VLD4LNd32 : VLD4LN<0b1011, {?,0,?,?}, "32"> {
1206 let Inst{7} = lane{0};
1207 let Inst{5} = Rn{5};
1210 def VLD4LNd8Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
1211 def VLD4LNd16Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
1212 def VLD4LNd32Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
1214 // ...with double-spaced registers:
1215 def VLD4LNq16 : VLD4LN<0b0111, {?,?,1,?}, "16"> {
1216 let Inst{7-6} = lane{1-0};
1218 def VLD4LNq32 : VLD4LN<0b1011, {?,1,?,?}, "32"> {
1219 let Inst{7} = lane{0};
1220 let Inst{5} = Rn{5};
1223 def VLD4LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>;
1224 def VLD4LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>;
1226 // ...with address register writeback:
1227 class VLD4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1228 : NLdStLn<1, 0b10, op11_8, op7_4,
1229 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
1230 (ins addrmode6:$Rn, am6offset:$Rm,
1231 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
1232 IIC_VLD4lnu, "vld4", Dt,
1233 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn$Rm",
1234 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4, $Rn.addr = $wb",
1236 let Inst{4} = Rn{4};
1237 let DecoderMethod = "DecodeVLD4LN" ;
1240 def VLD4LNd8_UPD : VLD4LNWB<0b0011, {?,?,?,?}, "8"> {
1241 let Inst{7-5} = lane{2-0};
1243 def VLD4LNd16_UPD : VLD4LNWB<0b0111, {?,?,0,?}, "16"> {
1244 let Inst{7-6} = lane{1-0};
1246 def VLD4LNd32_UPD : VLD4LNWB<0b1011, {?,0,?,?}, "32"> {
1247 let Inst{7} = lane{0};
1248 let Inst{5} = Rn{5};
1251 def VLD4LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
1252 def VLD4LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
1253 def VLD4LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
1255 def VLD4LNq16_UPD : VLD4LNWB<0b0111, {?,?,1,?}, "16"> {
1256 let Inst{7-6} = lane{1-0};
1258 def VLD4LNq32_UPD : VLD4LNWB<0b1011, {?,1,?,?}, "32"> {
1259 let Inst{7} = lane{0};
1260 let Inst{5} = Rn{5};
1263 def VLD4LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
1264 def VLD4LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
1266 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
1268 // VLD1DUP : Vector Load (single element to all lanes)
1269 class VLD1DUP<bits<4> op7_4, string Dt, ValueType Ty, PatFrag LoadOp>
1270 : NLdSt<1, 0b10, 0b1100, op7_4, (outs VecListOneDAllLanes:$Vd),
1271 (ins addrmode6dup:$Rn),
1272 IIC_VLD1dup, "vld1", Dt, "$Vd, $Rn", "",
1273 [(set VecListOneDAllLanes:$Vd,
1274 (Ty (NEONvdup (i32 (LoadOp addrmode6dup:$Rn)))))]> {
1276 let Inst{4} = Rn{4};
1277 let DecoderMethod = "DecodeVLD1DupInstruction";
1279 class VLD1QDUPPseudo<ValueType Ty, PatFrag LoadOp> : VLDQPseudo<IIC_VLD1dup> {
1280 let Pattern = [(set QPR:$dst,
1281 (Ty (NEONvdup (i32 (LoadOp addrmode6dup:$addr)))))];
1284 def VLD1DUPd8 : VLD1DUP<{0,0,0,?}, "8", v8i8, extloadi8>;
1285 def VLD1DUPd16 : VLD1DUP<{0,1,0,?}, "16", v4i16, extloadi16>;
1286 def VLD1DUPd32 : VLD1DUP<{1,0,0,?}, "32", v2i32, load>;
1288 def VLD1DUPq8Pseudo : VLD1QDUPPseudo<v16i8, extloadi8>;
1289 def VLD1DUPq16Pseudo : VLD1QDUPPseudo<v8i16, extloadi16>;
1290 def VLD1DUPq32Pseudo : VLD1QDUPPseudo<v4i32, load>;
1292 def : Pat<(v2f32 (NEONvdup (f32 (load addrmode6dup:$addr)))),
1293 (VLD1DUPd32 addrmode6:$addr)>;
1294 def : Pat<(v4f32 (NEONvdup (f32 (load addrmode6dup:$addr)))),
1295 (VLD1DUPq32Pseudo addrmode6:$addr)>;
1297 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
1299 class VLD1QDUP<bits<4> op7_4, string Dt>
1300 : NLdSt<1, 0b10, 0b1100, op7_4, (outs VecListTwoDAllLanes:$Vd),
1301 (ins addrmode6dup:$Rn), IIC_VLD1dup,
1302 "vld1", Dt, "$Vd, $Rn", "", []> {
1304 let Inst{4} = Rn{4};
1305 let DecoderMethod = "DecodeVLD1DupInstruction";
1308 def VLD1DUPq8 : VLD1QDUP<{0,0,1,0}, "8">;
1309 def VLD1DUPq16 : VLD1QDUP<{0,1,1,?}, "16">;
1310 def VLD1DUPq32 : VLD1QDUP<{1,0,1,?}, "32">;
1312 // ...with address register writeback:
1313 multiclass VLD1DUPWB<bits<4> op7_4, string Dt> {
1314 def _fixed : NLdSt<1, 0b10, 0b1100, op7_4,
1315 (outs VecListOneDAllLanes:$Vd, GPR:$wb),
1316 (ins addrmode6dup:$Rn), IIC_VLD1dupu,
1317 "vld1", Dt, "$Vd, $Rn!",
1318 "$Rn.addr = $wb", []> {
1319 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1320 let Inst{4} = Rn{4};
1321 let DecoderMethod = "DecodeVLD1DupInstruction";
1322 let AsmMatchConverter = "cvtVLDwbFixed";
1324 def _register : NLdSt<1, 0b10, 0b1100, op7_4,
1325 (outs VecListOneDAllLanes:$Vd, GPR:$wb),
1326 (ins addrmode6dup:$Rn, rGPR:$Rm), IIC_VLD1dupu,
1327 "vld1", Dt, "$Vd, $Rn, $Rm",
1328 "$Rn.addr = $wb", []> {
1329 let Inst{4} = Rn{4};
1330 let DecoderMethod = "DecodeVLD1DupInstruction";
1331 let AsmMatchConverter = "cvtVLDwbRegister";
1334 multiclass VLD1QDUPWB<bits<4> op7_4, string Dt> {
1335 def _fixed : NLdSt<1, 0b10, 0b1100, op7_4,
1336 (outs VecListTwoDAllLanes:$Vd, GPR:$wb),
1337 (ins addrmode6dup:$Rn), IIC_VLD1dupu,
1338 "vld1", Dt, "$Vd, $Rn!",
1339 "$Rn.addr = $wb", []> {
1340 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1341 let Inst{4} = Rn{4};
1342 let DecoderMethod = "DecodeVLD1DupInstruction";
1343 let AsmMatchConverter = "cvtVLDwbFixed";
1345 def _register : NLdSt<1, 0b10, 0b1100, op7_4,
1346 (outs VecListTwoDAllLanes:$Vd, GPR:$wb),
1347 (ins addrmode6dup:$Rn, rGPR:$Rm), IIC_VLD1dupu,
1348 "vld1", Dt, "$Vd, $Rn, $Rm",
1349 "$Rn.addr = $wb", []> {
1350 let Inst{4} = Rn{4};
1351 let DecoderMethod = "DecodeVLD1DupInstruction";
1352 let AsmMatchConverter = "cvtVLDwbRegister";
1356 defm VLD1DUPd8wb : VLD1DUPWB<{0,0,0,0}, "8">;
1357 defm VLD1DUPd16wb : VLD1DUPWB<{0,1,0,?}, "16">;
1358 defm VLD1DUPd32wb : VLD1DUPWB<{1,0,0,?}, "32">;
1360 defm VLD1DUPq8wb : VLD1QDUPWB<{0,0,1,0}, "8">;
1361 defm VLD1DUPq16wb : VLD1QDUPWB<{0,1,1,?}, "16">;
1362 defm VLD1DUPq32wb : VLD1QDUPWB<{1,0,1,?}, "32">;
1364 def VLD1DUPq8PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD1dupu>;
1365 def VLD1DUPq16PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD1dupu>;
1366 def VLD1DUPq32PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD1dupu>;
1367 def VLD1DUPq8PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD1dupu>;
1368 def VLD1DUPq16PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD1dupu>;
1369 def VLD1DUPq32PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD1dupu>;
1371 // VLD2DUP : Vector Load (single 2-element structure to all lanes)
1372 class VLD2DUP<bits<4> op7_4, string Dt, RegisterOperand VdTy>
1373 : NLdSt<1, 0b10, 0b1101, op7_4, (outs VdTy:$Vd),
1374 (ins addrmode6dup:$Rn), IIC_VLD2dup,
1375 "vld2", Dt, "$Vd, $Rn", "", []> {
1377 let Inst{4} = Rn{4};
1378 let DecoderMethod = "DecodeVLD2DupInstruction";
1381 def VLD2DUPd8 : VLD2DUP<{0,0,0,?}, "8", VecListTwoDAllLanes>;
1382 def VLD2DUPd16 : VLD2DUP<{0,1,0,?}, "16", VecListTwoDAllLanes>;
1383 def VLD2DUPd32 : VLD2DUP<{1,0,0,?}, "32", VecListTwoDAllLanes>;
1385 def VLD2DUPd8Pseudo : VLDQPseudo<IIC_VLD2dup>;
1386 def VLD2DUPd16Pseudo : VLDQPseudo<IIC_VLD2dup>;
1387 def VLD2DUPd32Pseudo : VLDQPseudo<IIC_VLD2dup>;
1389 // ...with double-spaced registers (not used for codegen):
1390 def VLD2DUPd8x2 : VLD2DUP<{0,0,1,?}, "8", VecListTwoQAllLanes>;
1391 def VLD2DUPd16x2 : VLD2DUP<{0,1,1,?}, "16", VecListTwoQAllLanes>;
1392 def VLD2DUPd32x2 : VLD2DUP<{1,0,1,?}, "32", VecListTwoQAllLanes>;
1394 // ...with address register writeback:
1395 multiclass VLD2DUPWB<bits<4> op7_4, string Dt, RegisterOperand VdTy> {
1396 def _fixed : NLdSt<1, 0b10, 0b1101, op7_4,
1397 (outs VdTy:$Vd, GPR:$wb),
1398 (ins addrmode6dup:$Rn), IIC_VLD2dupu,
1399 "vld2", Dt, "$Vd, $Rn!",
1400 "$Rn.addr = $wb", []> {
1401 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1402 let Inst{4} = Rn{4};
1403 let DecoderMethod = "DecodeVLD2DupInstruction";
1404 let AsmMatchConverter = "cvtVLDwbFixed";
1406 def _register : NLdSt<1, 0b10, 0b1101, op7_4,
1407 (outs VdTy:$Vd, GPR:$wb),
1408 (ins addrmode6dup:$Rn, rGPR:$Rm), IIC_VLD2dupu,
1409 "vld2", Dt, "$Vd, $Rn, $Rm",
1410 "$Rn.addr = $wb", []> {
1411 let Inst{4} = Rn{4};
1412 let DecoderMethod = "DecodeVLD2DupInstruction";
1413 let AsmMatchConverter = "cvtVLDwbRegister";
1417 defm VLD2DUPd8wb : VLD2DUPWB<{0,0,0,0}, "8", VecListTwoDAllLanes>;
1418 defm VLD2DUPd16wb : VLD2DUPWB<{0,1,0,?}, "16", VecListTwoDAllLanes>;
1419 defm VLD2DUPd32wb : VLD2DUPWB<{1,0,0,?}, "32", VecListTwoDAllLanes>;
1421 defm VLD2DUPd8x2wb : VLD2DUPWB<{0,0,1,0}, "8", VecListTwoQAllLanes>;
1422 defm VLD2DUPd16x2wb : VLD2DUPWB<{0,1,1,?}, "16", VecListTwoQAllLanes>;
1423 defm VLD2DUPd32x2wb : VLD2DUPWB<{1,0,1,?}, "32", VecListTwoQAllLanes>;
1425 def VLD2DUPd8PseudoWB_fixed : VLDQWBfixedPseudo <IIC_VLD2dupu>;
1426 def VLD2DUPd8PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD2dupu>;
1427 def VLD2DUPd16PseudoWB_fixed : VLDQWBfixedPseudo <IIC_VLD2dupu>;
1428 def VLD2DUPd16PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD2dupu>;
1429 def VLD2DUPd32PseudoWB_fixed : VLDQWBfixedPseudo <IIC_VLD2dupu>;
1430 def VLD2DUPd32PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD2dupu>;
1432 // VLD3DUP : Vector Load (single 3-element structure to all lanes)
1433 class VLD3DUP<bits<4> op7_4, string Dt>
1434 : NLdSt<1, 0b10, 0b1110, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
1435 (ins addrmode6dup:$Rn), IIC_VLD3dup,
1436 "vld3", Dt, "\\{$Vd[], $dst2[], $dst3[]\\}, $Rn", "", []> {
1439 let DecoderMethod = "DecodeVLD3DupInstruction";
1442 def VLD3DUPd8 : VLD3DUP<{0,0,0,?}, "8">;
1443 def VLD3DUPd16 : VLD3DUP<{0,1,0,?}, "16">;
1444 def VLD3DUPd32 : VLD3DUP<{1,0,0,?}, "32">;
1446 def VLD3DUPd8Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1447 def VLD3DUPd16Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1448 def VLD3DUPd32Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1450 // ...with double-spaced registers (not used for codegen):
1451 def VLD3DUPq8 : VLD3DUP<{0,0,1,?}, "8">;
1452 def VLD3DUPq16 : VLD3DUP<{0,1,1,?}, "16">;
1453 def VLD3DUPq32 : VLD3DUP<{1,0,1,?}, "32">;
1455 // ...with address register writeback:
1456 class VLD3DUPWB<bits<4> op7_4, string Dt>
1457 : NLdSt<1, 0b10, 0b1110, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
1458 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD3dupu,
1459 "vld3", Dt, "\\{$Vd[], $dst2[], $dst3[]\\}, $Rn$Rm",
1460 "$Rn.addr = $wb", []> {
1462 let DecoderMethod = "DecodeVLD3DupInstruction";
1465 def VLD3DUPd8_UPD : VLD3DUPWB<{0,0,0,0}, "8">;
1466 def VLD3DUPd16_UPD : VLD3DUPWB<{0,1,0,?}, "16">;
1467 def VLD3DUPd32_UPD : VLD3DUPWB<{1,0,0,?}, "32">;
1469 def VLD3DUPq8_UPD : VLD3DUPWB<{0,0,1,0}, "8">;
1470 def VLD3DUPq16_UPD : VLD3DUPWB<{0,1,1,?}, "16">;
1471 def VLD3DUPq32_UPD : VLD3DUPWB<{1,0,1,?}, "32">;
1473 def VLD3DUPd8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1474 def VLD3DUPd16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1475 def VLD3DUPd32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1477 // VLD4DUP : Vector Load (single 4-element structure to all lanes)
1478 class VLD4DUP<bits<4> op7_4, string Dt>
1479 : NLdSt<1, 0b10, 0b1111, op7_4,
1480 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
1481 (ins addrmode6dup:$Rn), IIC_VLD4dup,
1482 "vld4", Dt, "\\{$Vd[], $dst2[], $dst3[], $dst4[]\\}, $Rn", "", []> {
1484 let Inst{4} = Rn{4};
1485 let DecoderMethod = "DecodeVLD4DupInstruction";
1488 def VLD4DUPd8 : VLD4DUP<{0,0,0,?}, "8">;
1489 def VLD4DUPd16 : VLD4DUP<{0,1,0,?}, "16">;
1490 def VLD4DUPd32 : VLD4DUP<{1,?,0,?}, "32"> { let Inst{6} = Rn{5}; }
1492 def VLD4DUPd8Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1493 def VLD4DUPd16Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1494 def VLD4DUPd32Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1496 // ...with double-spaced registers (not used for codegen):
1497 def VLD4DUPq8 : VLD4DUP<{0,0,1,?}, "8">;
1498 def VLD4DUPq16 : VLD4DUP<{0,1,1,?}, "16">;
1499 def VLD4DUPq32 : VLD4DUP<{1,?,1,?}, "32"> { let Inst{6} = Rn{5}; }
1501 // ...with address register writeback:
1502 class VLD4DUPWB<bits<4> op7_4, string Dt>
1503 : NLdSt<1, 0b10, 0b1111, op7_4,
1504 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
1505 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD4dupu,
1506 "vld4", Dt, "\\{$Vd[], $dst2[], $dst3[], $dst4[]\\}, $Rn$Rm",
1507 "$Rn.addr = $wb", []> {
1508 let Inst{4} = Rn{4};
1509 let DecoderMethod = "DecodeVLD4DupInstruction";
1512 def VLD4DUPd8_UPD : VLD4DUPWB<{0,0,0,0}, "8">;
1513 def VLD4DUPd16_UPD : VLD4DUPWB<{0,1,0,?}, "16">;
1514 def VLD4DUPd32_UPD : VLD4DUPWB<{1,?,0,?}, "32"> { let Inst{6} = Rn{5}; }
1516 def VLD4DUPq8_UPD : VLD4DUPWB<{0,0,1,0}, "8">;
1517 def VLD4DUPq16_UPD : VLD4DUPWB<{0,1,1,?}, "16">;
1518 def VLD4DUPq32_UPD : VLD4DUPWB<{1,?,1,?}, "32"> { let Inst{6} = Rn{5}; }
1520 def VLD4DUPd8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1521 def VLD4DUPd16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1522 def VLD4DUPd32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1524 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
1526 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
1528 // Classes for VST* pseudo-instructions with multi-register operands.
1529 // These are expanded to real instructions after register allocation.
1530 class VSTQPseudo<InstrItinClass itin>
1531 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src), itin, "">;
1532 class VSTQWBPseudo<InstrItinClass itin>
1533 : PseudoNLdSt<(outs GPR:$wb),
1534 (ins addrmode6:$addr, am6offset:$offset, QPR:$src), itin,
1535 "$addr.addr = $wb">;
1536 class VSTQWBfixedPseudo<InstrItinClass itin>
1537 : PseudoNLdSt<(outs GPR:$wb),
1538 (ins addrmode6:$addr, QPR:$src), itin,
1539 "$addr.addr = $wb">;
1540 class VSTQWBregisterPseudo<InstrItinClass itin>
1541 : PseudoNLdSt<(outs GPR:$wb),
1542 (ins addrmode6:$addr, rGPR:$offset, QPR:$src), itin,
1543 "$addr.addr = $wb">;
1544 class VSTQQPseudo<InstrItinClass itin>
1545 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src), itin, "">;
1546 class VSTQQWBPseudo<InstrItinClass itin>
1547 : PseudoNLdSt<(outs GPR:$wb),
1548 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src), itin,
1549 "$addr.addr = $wb">;
1550 class VSTQQWBfixedPseudo<InstrItinClass itin>
1551 : PseudoNLdSt<(outs GPR:$wb),
1552 (ins addrmode6:$addr, QQPR:$src), itin,
1553 "$addr.addr = $wb">;
1554 class VSTQQWBregisterPseudo<InstrItinClass itin>
1555 : PseudoNLdSt<(outs GPR:$wb),
1556 (ins addrmode6:$addr, rGPR:$offset, QQPR:$src), itin,
1557 "$addr.addr = $wb">;
1559 class VSTQQQQPseudo<InstrItinClass itin>
1560 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src), itin, "">;
1561 class VSTQQQQWBPseudo<InstrItinClass itin>
1562 : PseudoNLdSt<(outs GPR:$wb),
1563 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
1564 "$addr.addr = $wb">;
1566 // VST1 : Vector Store (multiple single elements)
1567 class VST1D<bits<4> op7_4, string Dt>
1568 : NLdSt<0,0b00,0b0111,op7_4, (outs), (ins addrmode6:$Rn, VecListOneD:$Vd),
1569 IIC_VST1, "vst1", Dt, "$Vd, $Rn", "", []> {
1571 let Inst{4} = Rn{4};
1572 let DecoderMethod = "DecodeVSTInstruction";
1574 class VST1Q<bits<4> op7_4, string Dt>
1575 : NLdSt<0,0b00,0b1010,op7_4, (outs), (ins addrmode6:$Rn, VecListDPair:$Vd),
1576 IIC_VST1x2, "vst1", Dt, "$Vd, $Rn", "", []> {
1578 let Inst{5-4} = Rn{5-4};
1579 let DecoderMethod = "DecodeVSTInstruction";
1582 def VST1d8 : VST1D<{0,0,0,?}, "8">;
1583 def VST1d16 : VST1D<{0,1,0,?}, "16">;
1584 def VST1d32 : VST1D<{1,0,0,?}, "32">;
1585 def VST1d64 : VST1D<{1,1,0,?}, "64">;
1587 def VST1q8 : VST1Q<{0,0,?,?}, "8">;
1588 def VST1q16 : VST1Q<{0,1,?,?}, "16">;
1589 def VST1q32 : VST1Q<{1,0,?,?}, "32">;
1590 def VST1q64 : VST1Q<{1,1,?,?}, "64">;
1592 // ...with address register writeback:
1593 multiclass VST1DWB<bits<4> op7_4, string Dt> {
1594 def _fixed : NLdSt<0,0b00, 0b0111,op7_4, (outs GPR:$wb),
1595 (ins addrmode6:$Rn, VecListOneD:$Vd), IIC_VLD1u,
1596 "vst1", Dt, "$Vd, $Rn!",
1597 "$Rn.addr = $wb", []> {
1598 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1599 let Inst{4} = Rn{4};
1600 let DecoderMethod = "DecodeVSTInstruction";
1601 let AsmMatchConverter = "cvtVSTwbFixed";
1603 def _register : NLdSt<0,0b00,0b0111,op7_4, (outs GPR:$wb),
1604 (ins addrmode6:$Rn, rGPR:$Rm, VecListOneD:$Vd),
1606 "vst1", Dt, "$Vd, $Rn, $Rm",
1607 "$Rn.addr = $wb", []> {
1608 let Inst{4} = Rn{4};
1609 let DecoderMethod = "DecodeVSTInstruction";
1610 let AsmMatchConverter = "cvtVSTwbRegister";
1613 multiclass VST1QWB<bits<4> op7_4, string Dt> {
1614 def _fixed : NLdSt<0,0b00,0b1010,op7_4, (outs GPR:$wb),
1615 (ins addrmode6:$Rn, VecListDPair:$Vd), IIC_VLD1x2u,
1616 "vst1", Dt, "$Vd, $Rn!",
1617 "$Rn.addr = $wb", []> {
1618 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1619 let Inst{5-4} = Rn{5-4};
1620 let DecoderMethod = "DecodeVSTInstruction";
1621 let AsmMatchConverter = "cvtVSTwbFixed";
1623 def _register : NLdSt<0,0b00,0b1010,op7_4, (outs GPR:$wb),
1624 (ins addrmode6:$Rn, rGPR:$Rm, VecListDPair:$Vd),
1626 "vst1", Dt, "$Vd, $Rn, $Rm",
1627 "$Rn.addr = $wb", []> {
1628 let Inst{5-4} = Rn{5-4};
1629 let DecoderMethod = "DecodeVSTInstruction";
1630 let AsmMatchConverter = "cvtVSTwbRegister";
1634 defm VST1d8wb : VST1DWB<{0,0,0,?}, "8">;
1635 defm VST1d16wb : VST1DWB<{0,1,0,?}, "16">;
1636 defm VST1d32wb : VST1DWB<{1,0,0,?}, "32">;
1637 defm VST1d64wb : VST1DWB<{1,1,0,?}, "64">;
1639 defm VST1q8wb : VST1QWB<{0,0,?,?}, "8">;
1640 defm VST1q16wb : VST1QWB<{0,1,?,?}, "16">;
1641 defm VST1q32wb : VST1QWB<{1,0,?,?}, "32">;
1642 defm VST1q64wb : VST1QWB<{1,1,?,?}, "64">;
1644 // ...with 3 registers
1645 class VST1D3<bits<4> op7_4, string Dt>
1646 : NLdSt<0, 0b00, 0b0110, op7_4, (outs),
1647 (ins addrmode6:$Rn, VecListThreeD:$Vd),
1648 IIC_VST1x3, "vst1", Dt, "$Vd, $Rn", "", []> {
1650 let Inst{4} = Rn{4};
1651 let DecoderMethod = "DecodeVSTInstruction";
1653 multiclass VST1D3WB<bits<4> op7_4, string Dt> {
1654 def _fixed : NLdSt<0,0b00,0b0110,op7_4, (outs GPR:$wb),
1655 (ins addrmode6:$Rn, VecListThreeD:$Vd), IIC_VLD1x3u,
1656 "vst1", Dt, "$Vd, $Rn!",
1657 "$Rn.addr = $wb", []> {
1658 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1659 let Inst{5-4} = Rn{5-4};
1660 let DecoderMethod = "DecodeVSTInstruction";
1661 let AsmMatchConverter = "cvtVSTwbFixed";
1663 def _register : NLdSt<0,0b00,0b0110,op7_4, (outs GPR:$wb),
1664 (ins addrmode6:$Rn, rGPR:$Rm, VecListThreeD:$Vd),
1666 "vst1", Dt, "$Vd, $Rn, $Rm",
1667 "$Rn.addr = $wb", []> {
1668 let Inst{5-4} = Rn{5-4};
1669 let DecoderMethod = "DecodeVSTInstruction";
1670 let AsmMatchConverter = "cvtVSTwbRegister";
1674 def VST1d8T : VST1D3<{0,0,0,?}, "8">;
1675 def VST1d16T : VST1D3<{0,1,0,?}, "16">;
1676 def VST1d32T : VST1D3<{1,0,0,?}, "32">;
1677 def VST1d64T : VST1D3<{1,1,0,?}, "64">;
1679 defm VST1d8Twb : VST1D3WB<{0,0,0,?}, "8">;
1680 defm VST1d16Twb : VST1D3WB<{0,1,0,?}, "16">;
1681 defm VST1d32Twb : VST1D3WB<{1,0,0,?}, "32">;
1682 defm VST1d64Twb : VST1D3WB<{1,1,0,?}, "64">;
1684 def VST1d64TPseudo : VSTQQPseudo<IIC_VST1x3>;
1685 def VST1d64TPseudoWB_fixed : VSTQQWBPseudo<IIC_VST1x3u>;
1686 def VST1d64TPseudoWB_register : VSTQQWBPseudo<IIC_VST1x3u>;
1688 // ...with 4 registers
1689 class VST1D4<bits<4> op7_4, string Dt>
1690 : NLdSt<0, 0b00, 0b0010, op7_4, (outs),
1691 (ins addrmode6:$Rn, VecListFourD:$Vd),
1692 IIC_VST1x4, "vst1", Dt, "$Vd, $Rn", "",
1695 let Inst{5-4} = Rn{5-4};
1696 let DecoderMethod = "DecodeVSTInstruction";
1698 multiclass VST1D4WB<bits<4> op7_4, string Dt> {
1699 def _fixed : NLdSt<0,0b00,0b0010,op7_4, (outs GPR:$wb),
1700 (ins addrmode6:$Rn, VecListFourD:$Vd), IIC_VLD1x4u,
1701 "vst1", Dt, "$Vd, $Rn!",
1702 "$Rn.addr = $wb", []> {
1703 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1704 let Inst{5-4} = Rn{5-4};
1705 let DecoderMethod = "DecodeVSTInstruction";
1706 let AsmMatchConverter = "cvtVSTwbFixed";
1708 def _register : NLdSt<0,0b00,0b0010,op7_4, (outs GPR:$wb),
1709 (ins addrmode6:$Rn, rGPR:$Rm, VecListFourD:$Vd),
1711 "vst1", Dt, "$Vd, $Rn, $Rm",
1712 "$Rn.addr = $wb", []> {
1713 let Inst{5-4} = Rn{5-4};
1714 let DecoderMethod = "DecodeVSTInstruction";
1715 let AsmMatchConverter = "cvtVSTwbRegister";
1719 def VST1d8Q : VST1D4<{0,0,?,?}, "8">;
1720 def VST1d16Q : VST1D4<{0,1,?,?}, "16">;
1721 def VST1d32Q : VST1D4<{1,0,?,?}, "32">;
1722 def VST1d64Q : VST1D4<{1,1,?,?}, "64">;
1724 defm VST1d8Qwb : VST1D4WB<{0,0,?,?}, "8">;
1725 defm VST1d16Qwb : VST1D4WB<{0,1,?,?}, "16">;
1726 defm VST1d32Qwb : VST1D4WB<{1,0,?,?}, "32">;
1727 defm VST1d64Qwb : VST1D4WB<{1,1,?,?}, "64">;
1729 def VST1d64QPseudo : VSTQQPseudo<IIC_VST1x4>;
1730 def VST1d64QPseudoWB_fixed : VSTQQWBPseudo<IIC_VST1x4u>;
1731 def VST1d64QPseudoWB_register : VSTQQWBPseudo<IIC_VST1x4u>;
1733 // VST2 : Vector Store (multiple 2-element structures)
1734 class VST2<bits<4> op11_8, bits<4> op7_4, string Dt, RegisterOperand VdTy,
1735 InstrItinClass itin>
1736 : NLdSt<0, 0b00, op11_8, op7_4, (outs), (ins addrmode6:$Rn, VdTy:$Vd),
1737 itin, "vst2", Dt, "$Vd, $Rn", "", []> {
1739 let Inst{5-4} = Rn{5-4};
1740 let DecoderMethod = "DecodeVSTInstruction";
1743 def VST2d8 : VST2<0b1000, {0,0,?,?}, "8", VecListDPair, IIC_VST2>;
1744 def VST2d16 : VST2<0b1000, {0,1,?,?}, "16", VecListDPair, IIC_VST2>;
1745 def VST2d32 : VST2<0b1000, {1,0,?,?}, "32", VecListDPair, IIC_VST2>;
1747 def VST2q8 : VST2<0b0011, {0,0,?,?}, "8", VecListFourD, IIC_VST2x2>;
1748 def VST2q16 : VST2<0b0011, {0,1,?,?}, "16", VecListFourD, IIC_VST2x2>;
1749 def VST2q32 : VST2<0b0011, {1,0,?,?}, "32", VecListFourD, IIC_VST2x2>;
1751 def VST2q8Pseudo : VSTQQPseudo<IIC_VST2x2>;
1752 def VST2q16Pseudo : VSTQQPseudo<IIC_VST2x2>;
1753 def VST2q32Pseudo : VSTQQPseudo<IIC_VST2x2>;
1755 // ...with address register writeback:
1756 multiclass VST2DWB<bits<4> op11_8, bits<4> op7_4, string Dt,
1757 RegisterOperand VdTy> {
1758 def _fixed : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1759 (ins addrmode6:$Rn, VdTy:$Vd), IIC_VLD1u,
1760 "vst2", Dt, "$Vd, $Rn!",
1761 "$Rn.addr = $wb", []> {
1762 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1763 let Inst{5-4} = Rn{5-4};
1764 let DecoderMethod = "DecodeVSTInstruction";
1765 let AsmMatchConverter = "cvtVSTwbFixed";
1767 def _register : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1768 (ins addrmode6:$Rn, rGPR:$Rm, VdTy:$Vd), IIC_VLD1u,
1769 "vst2", Dt, "$Vd, $Rn, $Rm",
1770 "$Rn.addr = $wb", []> {
1771 let Inst{5-4} = Rn{5-4};
1772 let DecoderMethod = "DecodeVSTInstruction";
1773 let AsmMatchConverter = "cvtVSTwbRegister";
1776 multiclass VST2QWB<bits<4> op7_4, string Dt> {
1777 def _fixed : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
1778 (ins addrmode6:$Rn, VecListFourD:$Vd), IIC_VLD1u,
1779 "vst2", Dt, "$Vd, $Rn!",
1780 "$Rn.addr = $wb", []> {
1781 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1782 let Inst{5-4} = Rn{5-4};
1783 let DecoderMethod = "DecodeVSTInstruction";
1784 let AsmMatchConverter = "cvtVSTwbFixed";
1786 def _register : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
1787 (ins addrmode6:$Rn, rGPR:$Rm, VecListFourD:$Vd),
1789 "vst2", Dt, "$Vd, $Rn, $Rm",
1790 "$Rn.addr = $wb", []> {
1791 let Inst{5-4} = Rn{5-4};
1792 let DecoderMethod = "DecodeVSTInstruction";
1793 let AsmMatchConverter = "cvtVSTwbRegister";
1797 defm VST2d8wb : VST2DWB<0b1000, {0,0,?,?}, "8", VecListDPair>;
1798 defm VST2d16wb : VST2DWB<0b1000, {0,1,?,?}, "16", VecListDPair>;
1799 defm VST2d32wb : VST2DWB<0b1000, {1,0,?,?}, "32", VecListDPair>;
1801 defm VST2q8wb : VST2QWB<{0,0,?,?}, "8">;
1802 defm VST2q16wb : VST2QWB<{0,1,?,?}, "16">;
1803 defm VST2q32wb : VST2QWB<{1,0,?,?}, "32">;
1805 def VST2q8PseudoWB_fixed : VSTQQWBfixedPseudo<IIC_VST2x2u>;
1806 def VST2q16PseudoWB_fixed : VSTQQWBfixedPseudo<IIC_VST2x2u>;
1807 def VST2q32PseudoWB_fixed : VSTQQWBfixedPseudo<IIC_VST2x2u>;
1808 def VST2q8PseudoWB_register : VSTQQWBregisterPseudo<IIC_VST2x2u>;
1809 def VST2q16PseudoWB_register : VSTQQWBregisterPseudo<IIC_VST2x2u>;
1810 def VST2q32PseudoWB_register : VSTQQWBregisterPseudo<IIC_VST2x2u>;
1812 // ...with double-spaced registers
1813 def VST2b8 : VST2<0b1001, {0,0,?,?}, "8", VecListTwoQ, IIC_VST2>;
1814 def VST2b16 : VST2<0b1001, {0,1,?,?}, "16", VecListTwoQ, IIC_VST2>;
1815 def VST2b32 : VST2<0b1001, {1,0,?,?}, "32", VecListTwoQ, IIC_VST2>;
1816 defm VST2b8wb : VST2DWB<0b1001, {0,0,?,?}, "8", VecListTwoQ>;
1817 defm VST2b16wb : VST2DWB<0b1001, {0,1,?,?}, "16", VecListTwoQ>;
1818 defm VST2b32wb : VST2DWB<0b1001, {1,0,?,?}, "32", VecListTwoQ>;
1820 // VST3 : Vector Store (multiple 3-element structures)
1821 class VST3D<bits<4> op11_8, bits<4> op7_4, string Dt>
1822 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1823 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3,
1824 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn", "", []> {
1826 let Inst{4} = Rn{4};
1827 let DecoderMethod = "DecodeVSTInstruction";
1830 def VST3d8 : VST3D<0b0100, {0,0,0,?}, "8">;
1831 def VST3d16 : VST3D<0b0100, {0,1,0,?}, "16">;
1832 def VST3d32 : VST3D<0b0100, {1,0,0,?}, "32">;
1834 def VST3d8Pseudo : VSTQQPseudo<IIC_VST3>;
1835 def VST3d16Pseudo : VSTQQPseudo<IIC_VST3>;
1836 def VST3d32Pseudo : VSTQQPseudo<IIC_VST3>;
1838 // ...with address register writeback:
1839 class VST3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1840 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1841 (ins addrmode6:$Rn, am6offset:$Rm,
1842 DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3u,
1843 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn$Rm",
1844 "$Rn.addr = $wb", []> {
1845 let Inst{4} = Rn{4};
1846 let DecoderMethod = "DecodeVSTInstruction";
1849 def VST3d8_UPD : VST3DWB<0b0100, {0,0,0,?}, "8">;
1850 def VST3d16_UPD : VST3DWB<0b0100, {0,1,0,?}, "16">;
1851 def VST3d32_UPD : VST3DWB<0b0100, {1,0,0,?}, "32">;
1853 def VST3d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1854 def VST3d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1855 def VST3d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1857 // ...with double-spaced registers:
1858 def VST3q8 : VST3D<0b0101, {0,0,0,?}, "8">;
1859 def VST3q16 : VST3D<0b0101, {0,1,0,?}, "16">;
1860 def VST3q32 : VST3D<0b0101, {1,0,0,?}, "32">;
1861 def VST3q8_UPD : VST3DWB<0b0101, {0,0,0,?}, "8">;
1862 def VST3q16_UPD : VST3DWB<0b0101, {0,1,0,?}, "16">;
1863 def VST3q32_UPD : VST3DWB<0b0101, {1,0,0,?}, "32">;
1865 def VST3q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1866 def VST3q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1867 def VST3q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1869 // ...alternate versions to be allocated odd register numbers:
1870 def VST3q8oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1871 def VST3q16oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1872 def VST3q32oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1874 def VST3q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1875 def VST3q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1876 def VST3q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1878 // VST4 : Vector Store (multiple 4-element structures)
1879 class VST4D<bits<4> op11_8, bits<4> op7_4, string Dt>
1880 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1881 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
1882 IIC_VST4, "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn",
1885 let Inst{5-4} = Rn{5-4};
1886 let DecoderMethod = "DecodeVSTInstruction";
1889 def VST4d8 : VST4D<0b0000, {0,0,?,?}, "8">;
1890 def VST4d16 : VST4D<0b0000, {0,1,?,?}, "16">;
1891 def VST4d32 : VST4D<0b0000, {1,0,?,?}, "32">;
1893 def VST4d8Pseudo : VSTQQPseudo<IIC_VST4>;
1894 def VST4d16Pseudo : VSTQQPseudo<IIC_VST4>;
1895 def VST4d32Pseudo : VSTQQPseudo<IIC_VST4>;
1897 // ...with address register writeback:
1898 class VST4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1899 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1900 (ins addrmode6:$Rn, am6offset:$Rm,
1901 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST4u,
1902 "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
1903 "$Rn.addr = $wb", []> {
1904 let Inst{5-4} = Rn{5-4};
1905 let DecoderMethod = "DecodeVSTInstruction";
1908 def VST4d8_UPD : VST4DWB<0b0000, {0,0,?,?}, "8">;
1909 def VST4d16_UPD : VST4DWB<0b0000, {0,1,?,?}, "16">;
1910 def VST4d32_UPD : VST4DWB<0b0000, {1,0,?,?}, "32">;
1912 def VST4d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1913 def VST4d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1914 def VST4d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1916 // ...with double-spaced registers:
1917 def VST4q8 : VST4D<0b0001, {0,0,?,?}, "8">;
1918 def VST4q16 : VST4D<0b0001, {0,1,?,?}, "16">;
1919 def VST4q32 : VST4D<0b0001, {1,0,?,?}, "32">;
1920 def VST4q8_UPD : VST4DWB<0b0001, {0,0,?,?}, "8">;
1921 def VST4q16_UPD : VST4DWB<0b0001, {0,1,?,?}, "16">;
1922 def VST4q32_UPD : VST4DWB<0b0001, {1,0,?,?}, "32">;
1924 def VST4q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1925 def VST4q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1926 def VST4q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1928 // ...alternate versions to be allocated odd register numbers:
1929 def VST4q8oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1930 def VST4q16oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1931 def VST4q32oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1933 def VST4q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1934 def VST4q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1935 def VST4q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1937 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
1939 // Classes for VST*LN pseudo-instructions with multi-register operands.
1940 // These are expanded to real instructions after register allocation.
1941 class VSTQLNPseudo<InstrItinClass itin>
1942 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
1944 class VSTQLNWBPseudo<InstrItinClass itin>
1945 : PseudoNLdSt<(outs GPR:$wb),
1946 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
1947 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1948 class VSTQQLNPseudo<InstrItinClass itin>
1949 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
1951 class VSTQQLNWBPseudo<InstrItinClass itin>
1952 : PseudoNLdSt<(outs GPR:$wb),
1953 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
1954 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1955 class VSTQQQQLNPseudo<InstrItinClass itin>
1956 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
1958 class VSTQQQQLNWBPseudo<InstrItinClass itin>
1959 : PseudoNLdSt<(outs GPR:$wb),
1960 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
1961 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1963 // VST1LN : Vector Store (single element from one lane)
1964 class VST1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1965 PatFrag StoreOp, SDNode ExtractOp>
1966 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1967 (ins addrmode6:$Rn, DPR:$Vd, nohash_imm:$lane),
1968 IIC_VST1ln, "vst1", Dt, "\\{$Vd[$lane]\\}, $Rn", "",
1969 [(StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane), addrmode6:$Rn)]> {
1971 let DecoderMethod = "DecodeVST1LN";
1973 class VST1LN32<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1974 PatFrag StoreOp, SDNode ExtractOp>
1975 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1976 (ins addrmode6oneL32:$Rn, DPR:$Vd, nohash_imm:$lane),
1977 IIC_VST1ln, "vst1", Dt, "\\{$Vd[$lane]\\}, $Rn", "",
1978 [(StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane), addrmode6oneL32:$Rn)]>{
1980 let DecoderMethod = "DecodeVST1LN";
1982 class VST1QLNPseudo<ValueType Ty, PatFrag StoreOp, SDNode ExtractOp>
1983 : VSTQLNPseudo<IIC_VST1ln> {
1984 let Pattern = [(StoreOp (ExtractOp (Ty QPR:$src), imm:$lane),
1988 def VST1LNd8 : VST1LN<0b0000, {?,?,?,0}, "8", v8i8, truncstorei8,
1990 let Inst{7-5} = lane{2-0};
1992 def VST1LNd16 : VST1LN<0b0100, {?,?,0,?}, "16", v4i16, truncstorei16,
1994 let Inst{7-6} = lane{1-0};
1995 let Inst{4} = Rn{5};
1998 def VST1LNd32 : VST1LN32<0b1000, {?,0,?,?}, "32", v2i32, store, extractelt> {
1999 let Inst{7} = lane{0};
2000 let Inst{5-4} = Rn{5-4};
2003 def VST1LNq8Pseudo : VST1QLNPseudo<v16i8, truncstorei8, NEONvgetlaneu>;
2004 def VST1LNq16Pseudo : VST1QLNPseudo<v8i16, truncstorei16, NEONvgetlaneu>;
2005 def VST1LNq32Pseudo : VST1QLNPseudo<v4i32, store, extractelt>;
2007 def : Pat<(store (extractelt (v2f32 DPR:$src), imm:$lane), addrmode6:$addr),
2008 (VST1LNd32 addrmode6:$addr, DPR:$src, imm:$lane)>;
2009 def : Pat<(store (extractelt (v4f32 QPR:$src), imm:$lane), addrmode6:$addr),
2010 (VST1LNq32Pseudo addrmode6:$addr, QPR:$src, imm:$lane)>;
2012 // ...with address register writeback:
2013 class VST1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
2014 PatFrag StoreOp, SDNode ExtractOp>
2015 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
2016 (ins addrmode6:$Rn, am6offset:$Rm,
2017 DPR:$Vd, nohash_imm:$lane), IIC_VST1lnu, "vst1", Dt,
2018 "\\{$Vd[$lane]\\}, $Rn$Rm",
2020 [(set GPR:$wb, (StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane),
2021 addrmode6:$Rn, am6offset:$Rm))]> {
2022 let DecoderMethod = "DecodeVST1LN";
2024 class VST1QLNWBPseudo<ValueType Ty, PatFrag StoreOp, SDNode ExtractOp>
2025 : VSTQLNWBPseudo<IIC_VST1lnu> {
2026 let Pattern = [(set GPR:$wb, (StoreOp (ExtractOp (Ty QPR:$src), imm:$lane),
2027 addrmode6:$addr, am6offset:$offset))];
2030 def VST1LNd8_UPD : VST1LNWB<0b0000, {?,?,?,0}, "8", v8i8, post_truncsti8,
2032 let Inst{7-5} = lane{2-0};
2034 def VST1LNd16_UPD : VST1LNWB<0b0100, {?,?,0,?}, "16", v4i16, post_truncsti16,
2036 let Inst{7-6} = lane{1-0};
2037 let Inst{4} = Rn{5};
2039 def VST1LNd32_UPD : VST1LNWB<0b1000, {?,0,?,?}, "32", v2i32, post_store,
2041 let Inst{7} = lane{0};
2042 let Inst{5-4} = Rn{5-4};
2045 def VST1LNq8Pseudo_UPD : VST1QLNWBPseudo<v16i8, post_truncsti8, NEONvgetlaneu>;
2046 def VST1LNq16Pseudo_UPD : VST1QLNWBPseudo<v8i16, post_truncsti16,NEONvgetlaneu>;
2047 def VST1LNq32Pseudo_UPD : VST1QLNWBPseudo<v4i32, post_store, extractelt>;
2049 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
2051 // VST2LN : Vector Store (single 2-element structure from one lane)
2052 class VST2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
2053 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
2054 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, nohash_imm:$lane),
2055 IIC_VST2ln, "vst2", Dt, "\\{$Vd[$lane], $src2[$lane]\\}, $Rn",
2058 let Inst{4} = Rn{4};
2059 let DecoderMethod = "DecodeVST2LN";
2062 def VST2LNd8 : VST2LN<0b0001, {?,?,?,?}, "8"> {
2063 let Inst{7-5} = lane{2-0};
2065 def VST2LNd16 : VST2LN<0b0101, {?,?,0,?}, "16"> {
2066 let Inst{7-6} = lane{1-0};
2068 def VST2LNd32 : VST2LN<0b1001, {?,0,0,?}, "32"> {
2069 let Inst{7} = lane{0};
2072 def VST2LNd8Pseudo : VSTQLNPseudo<IIC_VST2ln>;
2073 def VST2LNd16Pseudo : VSTQLNPseudo<IIC_VST2ln>;
2074 def VST2LNd32Pseudo : VSTQLNPseudo<IIC_VST2ln>;
2076 // ...with double-spaced registers:
2077 def VST2LNq16 : VST2LN<0b0101, {?,?,1,?}, "16"> {
2078 let Inst{7-6} = lane{1-0};
2079 let Inst{4} = Rn{4};
2081 def VST2LNq32 : VST2LN<0b1001, {?,1,0,?}, "32"> {
2082 let Inst{7} = lane{0};
2083 let Inst{4} = Rn{4};
2086 def VST2LNq16Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
2087 def VST2LNq32Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
2089 // ...with address register writeback:
2090 class VST2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
2091 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
2092 (ins addrmode6:$Rn, am6offset:$Rm,
2093 DPR:$Vd, DPR:$src2, nohash_imm:$lane), IIC_VST2lnu, "vst2", Dt,
2094 "\\{$Vd[$lane], $src2[$lane]\\}, $Rn$Rm",
2095 "$Rn.addr = $wb", []> {
2096 let Inst{4} = Rn{4};
2097 let DecoderMethod = "DecodeVST2LN";
2100 def VST2LNd8_UPD : VST2LNWB<0b0001, {?,?,?,?}, "8"> {
2101 let Inst{7-5} = lane{2-0};
2103 def VST2LNd16_UPD : VST2LNWB<0b0101, {?,?,0,?}, "16"> {
2104 let Inst{7-6} = lane{1-0};
2106 def VST2LNd32_UPD : VST2LNWB<0b1001, {?,0,0,?}, "32"> {
2107 let Inst{7} = lane{0};
2110 def VST2LNd8Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
2111 def VST2LNd16Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
2112 def VST2LNd32Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
2114 def VST2LNq16_UPD : VST2LNWB<0b0101, {?,?,1,?}, "16"> {
2115 let Inst{7-6} = lane{1-0};
2117 def VST2LNq32_UPD : VST2LNWB<0b1001, {?,1,0,?}, "32"> {
2118 let Inst{7} = lane{0};
2121 def VST2LNq16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
2122 def VST2LNq32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
2124 // VST3LN : Vector Store (single 3-element structure from one lane)
2125 class VST3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
2126 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
2127 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3,
2128 nohash_imm:$lane), IIC_VST3ln, "vst3", Dt,
2129 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn", "", []> {
2131 let DecoderMethod = "DecodeVST3LN";
2134 def VST3LNd8 : VST3LN<0b0010, {?,?,?,0}, "8"> {
2135 let Inst{7-5} = lane{2-0};
2137 def VST3LNd16 : VST3LN<0b0110, {?,?,0,0}, "16"> {
2138 let Inst{7-6} = lane{1-0};
2140 def VST3LNd32 : VST3LN<0b1010, {?,0,0,0}, "32"> {
2141 let Inst{7} = lane{0};
2144 def VST3LNd8Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
2145 def VST3LNd16Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
2146 def VST3LNd32Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
2148 // ...with double-spaced registers:
2149 def VST3LNq16 : VST3LN<0b0110, {?,?,1,0}, "16"> {
2150 let Inst{7-6} = lane{1-0};
2152 def VST3LNq32 : VST3LN<0b1010, {?,1,0,0}, "32"> {
2153 let Inst{7} = lane{0};
2156 def VST3LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
2157 def VST3LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
2159 // ...with address register writeback:
2160 class VST3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
2161 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
2162 (ins addrmode6:$Rn, am6offset:$Rm,
2163 DPR:$Vd, DPR:$src2, DPR:$src3, nohash_imm:$lane),
2164 IIC_VST3lnu, "vst3", Dt,
2165 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn$Rm",
2166 "$Rn.addr = $wb", []> {
2167 let DecoderMethod = "DecodeVST3LN";
2170 def VST3LNd8_UPD : VST3LNWB<0b0010, {?,?,?,0}, "8"> {
2171 let Inst{7-5} = lane{2-0};
2173 def VST3LNd16_UPD : VST3LNWB<0b0110, {?,?,0,0}, "16"> {
2174 let Inst{7-6} = lane{1-0};
2176 def VST3LNd32_UPD : VST3LNWB<0b1010, {?,0,0,0}, "32"> {
2177 let Inst{7} = lane{0};
2180 def VST3LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
2181 def VST3LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
2182 def VST3LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
2184 def VST3LNq16_UPD : VST3LNWB<0b0110, {?,?,1,0}, "16"> {
2185 let Inst{7-6} = lane{1-0};
2187 def VST3LNq32_UPD : VST3LNWB<0b1010, {?,1,0,0}, "32"> {
2188 let Inst{7} = lane{0};
2191 def VST3LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
2192 def VST3LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
2194 // VST4LN : Vector Store (single 4-element structure from one lane)
2195 class VST4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
2196 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
2197 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4,
2198 nohash_imm:$lane), IIC_VST4ln, "vst4", Dt,
2199 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn",
2202 let Inst{4} = Rn{4};
2203 let DecoderMethod = "DecodeVST4LN";
2206 def VST4LNd8 : VST4LN<0b0011, {?,?,?,?}, "8"> {
2207 let Inst{7-5} = lane{2-0};
2209 def VST4LNd16 : VST4LN<0b0111, {?,?,0,?}, "16"> {
2210 let Inst{7-6} = lane{1-0};
2212 def VST4LNd32 : VST4LN<0b1011, {?,0,?,?}, "32"> {
2213 let Inst{7} = lane{0};
2214 let Inst{5} = Rn{5};
2217 def VST4LNd8Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
2218 def VST4LNd16Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
2219 def VST4LNd32Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
2221 // ...with double-spaced registers:
2222 def VST4LNq16 : VST4LN<0b0111, {?,?,1,?}, "16"> {
2223 let Inst{7-6} = lane{1-0};
2225 def VST4LNq32 : VST4LN<0b1011, {?,1,?,?}, "32"> {
2226 let Inst{7} = lane{0};
2227 let Inst{5} = Rn{5};
2230 def VST4LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
2231 def VST4LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
2233 // ...with address register writeback:
2234 class VST4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
2235 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
2236 (ins addrmode6:$Rn, am6offset:$Rm,
2237 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
2238 IIC_VST4lnu, "vst4", Dt,
2239 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn$Rm",
2240 "$Rn.addr = $wb", []> {
2241 let Inst{4} = Rn{4};
2242 let DecoderMethod = "DecodeVST4LN";
2245 def VST4LNd8_UPD : VST4LNWB<0b0011, {?,?,?,?}, "8"> {
2246 let Inst{7-5} = lane{2-0};
2248 def VST4LNd16_UPD : VST4LNWB<0b0111, {?,?,0,?}, "16"> {
2249 let Inst{7-6} = lane{1-0};
2251 def VST4LNd32_UPD : VST4LNWB<0b1011, {?,0,?,?}, "32"> {
2252 let Inst{7} = lane{0};
2253 let Inst{5} = Rn{5};
2256 def VST4LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
2257 def VST4LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
2258 def VST4LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
2260 def VST4LNq16_UPD : VST4LNWB<0b0111, {?,?,1,?}, "16"> {
2261 let Inst{7-6} = lane{1-0};
2263 def VST4LNq32_UPD : VST4LNWB<0b1011, {?,1,?,?}, "32"> {
2264 let Inst{7} = lane{0};
2265 let Inst{5} = Rn{5};
2268 def VST4LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
2269 def VST4LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
2271 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
2274 //===----------------------------------------------------------------------===//
2275 // NEON pattern fragments
2276 //===----------------------------------------------------------------------===//
2278 // Extract D sub-registers of Q registers.
2279 def DSubReg_i8_reg : SDNodeXForm<imm, [{
2280 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
2281 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/8, MVT::i32);
2283 def DSubReg_i16_reg : SDNodeXForm<imm, [{
2284 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
2285 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/4, MVT::i32);
2287 def DSubReg_i32_reg : SDNodeXForm<imm, [{
2288 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
2289 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/2, MVT::i32);
2291 def DSubReg_f64_reg : SDNodeXForm<imm, [{
2292 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
2293 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue(), MVT::i32);
2296 // Extract S sub-registers of Q/D registers.
2297 def SSubReg_f32_reg : SDNodeXForm<imm, [{
2298 assert(ARM::ssub_3 == ARM::ssub_0+3 && "Unexpected subreg numbering");
2299 return CurDAG->getTargetConstant(ARM::ssub_0 + N->getZExtValue(), MVT::i32);
2302 // Translate lane numbers from Q registers to D subregs.
2303 def SubReg_i8_lane : SDNodeXForm<imm, [{
2304 return CurDAG->getTargetConstant(N->getZExtValue() & 7, MVT::i32);
2306 def SubReg_i16_lane : SDNodeXForm<imm, [{
2307 return CurDAG->getTargetConstant(N->getZExtValue() & 3, MVT::i32);
2309 def SubReg_i32_lane : SDNodeXForm<imm, [{
2310 return CurDAG->getTargetConstant(N->getZExtValue() & 1, MVT::i32);
2313 //===----------------------------------------------------------------------===//
2314 // Instruction Classes
2315 //===----------------------------------------------------------------------===//
2317 // Basic 2-register operations: double- and quad-register.
2318 class N2VD<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2319 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
2320 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
2321 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
2322 (ins DPR:$Vm), IIC_VUNAD, OpcodeStr, Dt,"$Vd, $Vm", "",
2323 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vm))))]>;
2324 class N2VQ<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2325 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
2326 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
2327 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
2328 (ins QPR:$Vm), IIC_VUNAQ, OpcodeStr, Dt,"$Vd, $Vm", "",
2329 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vm))))]>;
2331 // Basic 2-register intrinsics, both double- and quad-register.
2332 class N2VDInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2333 bits<2> op17_16, bits<5> op11_7, bit op4,
2334 InstrItinClass itin, string OpcodeStr, string Dt,
2335 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2336 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
2337 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2338 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm))))]>;
2339 class N2VQInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2340 bits<2> op17_16, bits<5> op11_7, bit op4,
2341 InstrItinClass itin, string OpcodeStr, string Dt,
2342 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2343 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
2344 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2345 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm))))]>;
2347 // Narrow 2-register operations.
2348 class N2VN<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2349 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2350 InstrItinClass itin, string OpcodeStr, string Dt,
2351 ValueType TyD, ValueType TyQ, SDNode OpNode>
2352 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$Vd),
2353 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2354 [(set DPR:$Vd, (TyD (OpNode (TyQ QPR:$Vm))))]>;
2356 // Narrow 2-register intrinsics.
2357 class N2VNInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2358 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2359 InstrItinClass itin, string OpcodeStr, string Dt,
2360 ValueType TyD, ValueType TyQ, Intrinsic IntOp>
2361 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$Vd),
2362 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2363 [(set DPR:$Vd, (TyD (IntOp (TyQ QPR:$Vm))))]>;
2365 // Long 2-register operations (currently only used for VMOVL).
2366 class N2VL<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2367 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2368 InstrItinClass itin, string OpcodeStr, string Dt,
2369 ValueType TyQ, ValueType TyD, SDNode OpNode>
2370 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$Vd),
2371 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2372 [(set QPR:$Vd, (TyQ (OpNode (TyD DPR:$Vm))))]>;
2374 // Long 2-register intrinsics.
2375 class N2VLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2376 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2377 InstrItinClass itin, string OpcodeStr, string Dt,
2378 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
2379 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$Vd),
2380 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2381 [(set QPR:$Vd, (TyQ (IntOp (TyD DPR:$Vm))))]>;
2383 // 2-register shuffles (VTRN/VZIP/VUZP), both double- and quad-register.
2384 class N2VDShuffle<bits<2> op19_18, bits<5> op11_7, string OpcodeStr, string Dt>
2385 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 0, 0, (outs DPR:$Vd, DPR:$Vm),
2386 (ins DPR:$src1, DPR:$src2), IIC_VPERMD,
2387 OpcodeStr, Dt, "$Vd, $Vm",
2388 "$src1 = $Vd, $src2 = $Vm", []>;
2389 class N2VQShuffle<bits<2> op19_18, bits<5> op11_7,
2390 InstrItinClass itin, string OpcodeStr, string Dt>
2391 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 1, 0, (outs QPR:$Vd, QPR:$Vm),
2392 (ins QPR:$src1, QPR:$src2), itin, OpcodeStr, Dt, "$Vd, $Vm",
2393 "$src1 = $Vd, $src2 = $Vm", []>;
2395 // Basic 3-register operations: double- and quad-register.
2396 class N3VD<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2397 InstrItinClass itin, string OpcodeStr, string Dt,
2398 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
2399 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2400 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2401 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2402 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
2403 let isCommutable = Commutable;
2405 // Same as N3VD but no data type.
2406 class N3VDX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2407 InstrItinClass itin, string OpcodeStr,
2408 ValueType ResTy, ValueType OpTy,
2409 SDNode OpNode, bit Commutable>
2410 : N3VX<op24, op23, op21_20, op11_8, 0, op4,
2411 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2412 OpcodeStr, "$Vd, $Vn, $Vm", "",
2413 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>{
2414 let isCommutable = Commutable;
2417 class N3VDSL<bits<2> op21_20, bits<4> op11_8,
2418 InstrItinClass itin, string OpcodeStr, string Dt,
2419 ValueType Ty, SDNode ShOp>
2420 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2421 (outs DPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2422 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2424 (Ty (ShOp (Ty DPR:$Vn),
2425 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),imm:$lane)))))]> {
2426 let isCommutable = 0;
2428 class N3VDSL16<bits<2> op21_20, bits<4> op11_8,
2429 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
2430 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2431 (outs DPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2432 NVMulSLFrm, IIC_VMULi16D, OpcodeStr, Dt,"$Vd, $Vn, $Vm$lane","",
2434 (Ty (ShOp (Ty DPR:$Vn),
2435 (Ty (NEONvduplane (Ty DPR_8:$Vm), imm:$lane)))))]> {
2436 let isCommutable = 0;
2439 class N3VQ<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2440 InstrItinClass itin, string OpcodeStr, string Dt,
2441 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
2442 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2443 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2444 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2445 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
2446 let isCommutable = Commutable;
2448 class N3VQX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2449 InstrItinClass itin, string OpcodeStr,
2450 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
2451 : N3VX<op24, op23, op21_20, op11_8, 1, op4,
2452 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2453 OpcodeStr, "$Vd, $Vn, $Vm", "",
2454 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]>{
2455 let isCommutable = Commutable;
2457 class N3VQSL<bits<2> op21_20, bits<4> op11_8,
2458 InstrItinClass itin, string OpcodeStr, string Dt,
2459 ValueType ResTy, ValueType OpTy, SDNode ShOp>
2460 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2461 (outs QPR:$Vd), (ins QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2462 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2463 [(set (ResTy QPR:$Vd),
2464 (ResTy (ShOp (ResTy QPR:$Vn),
2465 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2467 let isCommutable = 0;
2469 class N3VQSL16<bits<2> op21_20, bits<4> op11_8, string OpcodeStr, string Dt,
2470 ValueType ResTy, ValueType OpTy, SDNode ShOp>
2471 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2472 (outs QPR:$Vd), (ins QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2473 NVMulSLFrm, IIC_VMULi16Q, OpcodeStr, Dt,"$Vd, $Vn, $Vm$lane", "",
2474 [(set (ResTy QPR:$Vd),
2475 (ResTy (ShOp (ResTy QPR:$Vn),
2476 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2478 let isCommutable = 0;
2481 // Basic 3-register intrinsics, both double- and quad-register.
2482 class N3VDInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2483 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2484 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
2485 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2486 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), f, itin,
2487 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2488 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
2489 let isCommutable = Commutable;
2491 class N3VDIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2492 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
2493 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2494 (outs DPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2495 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2497 (Ty (IntOp (Ty DPR:$Vn),
2498 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),
2500 let isCommutable = 0;
2502 class N3VDIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2503 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
2504 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2505 (outs DPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2506 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2508 (Ty (IntOp (Ty DPR:$Vn),
2509 (Ty (NEONvduplane (Ty DPR_8:$Vm), imm:$lane)))))]> {
2510 let isCommutable = 0;
2512 class N3VDIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2513 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2514 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2515 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2516 (outs DPR:$Vd), (ins DPR:$Vm, DPR:$Vn), f, itin,
2517 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
2518 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (OpTy DPR:$Vn))))]> {
2519 let isCommutable = 0;
2522 class N3VQInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2523 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2524 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
2525 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2526 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), f, itin,
2527 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2528 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
2529 let isCommutable = Commutable;
2531 class N3VQIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2532 string OpcodeStr, string Dt,
2533 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2534 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2535 (outs QPR:$Vd), (ins QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2536 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2537 [(set (ResTy QPR:$Vd),
2538 (ResTy (IntOp (ResTy QPR:$Vn),
2539 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2541 let isCommutable = 0;
2543 class N3VQIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2544 string OpcodeStr, string Dt,
2545 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2546 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2547 (outs QPR:$Vd), (ins QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2548 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2549 [(set (ResTy QPR:$Vd),
2550 (ResTy (IntOp (ResTy QPR:$Vn),
2551 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2553 let isCommutable = 0;
2555 class N3VQIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2556 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2557 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2558 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2559 (outs QPR:$Vd), (ins QPR:$Vm, QPR:$Vn), f, itin,
2560 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
2561 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (OpTy QPR:$Vn))))]> {
2562 let isCommutable = 0;
2565 // Multiply-Add/Sub operations: double- and quad-register.
2566 class N3VDMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2567 InstrItinClass itin, string OpcodeStr, string Dt,
2568 ValueType Ty, SDPatternOperator MulOp, SDPatternOperator OpNode>
2569 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2570 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2571 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2572 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
2573 (Ty (MulOp DPR:$Vn, DPR:$Vm)))))]>;
2575 class N3VDMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2576 string OpcodeStr, string Dt,
2577 ValueType Ty, SDPatternOperator MulOp, SDPatternOperator ShOp>
2578 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2580 (ins DPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2582 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2584 (Ty (ShOp (Ty DPR:$src1),
2586 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),
2588 class N3VDMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2589 string OpcodeStr, string Dt,
2590 ValueType Ty, SDNode MulOp, SDNode ShOp>
2591 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2593 (ins DPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2595 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2597 (Ty (ShOp (Ty DPR:$src1),
2599 (Ty (NEONvduplane (Ty DPR_8:$Vm),
2602 class N3VQMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2603 InstrItinClass itin, string OpcodeStr, string Dt, ValueType Ty,
2604 SDPatternOperator MulOp, SDPatternOperator OpNode>
2605 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2606 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2607 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2608 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
2609 (Ty (MulOp QPR:$Vn, QPR:$Vm)))))]>;
2610 class N3VQMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2611 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2612 SDPatternOperator MulOp, SDPatternOperator ShOp>
2613 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2615 (ins QPR:$src1, QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2617 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2618 [(set (ResTy QPR:$Vd),
2619 (ResTy (ShOp (ResTy QPR:$src1),
2620 (ResTy (MulOp QPR:$Vn,
2621 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2623 class N3VQMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2624 string OpcodeStr, string Dt,
2625 ValueType ResTy, ValueType OpTy,
2626 SDNode MulOp, SDNode ShOp>
2627 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2629 (ins QPR:$src1, QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2631 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2632 [(set (ResTy QPR:$Vd),
2633 (ResTy (ShOp (ResTy QPR:$src1),
2634 (ResTy (MulOp QPR:$Vn,
2635 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2638 // Neon Intrinsic-Op instructions (VABA): double- and quad-register.
2639 class N3VDIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2640 InstrItinClass itin, string OpcodeStr, string Dt,
2641 ValueType Ty, Intrinsic IntOp, SDNode OpNode>
2642 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2643 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2644 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2645 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
2646 (Ty (IntOp (Ty DPR:$Vn), (Ty DPR:$Vm))))))]>;
2647 class N3VQIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2648 InstrItinClass itin, string OpcodeStr, string Dt,
2649 ValueType Ty, Intrinsic IntOp, SDNode OpNode>
2650 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2651 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2652 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2653 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
2654 (Ty (IntOp (Ty QPR:$Vn), (Ty QPR:$Vm))))))]>;
2656 // Neon 3-argument intrinsics, both double- and quad-register.
2657 // The destination register is also used as the first source operand register.
2658 class N3VDInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2659 InstrItinClass itin, string OpcodeStr, string Dt,
2660 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2661 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2662 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2663 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2664 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$src1),
2665 (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>;
2666 class N3VQInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2667 InstrItinClass itin, string OpcodeStr, string Dt,
2668 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2669 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2670 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2671 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2672 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$src1),
2673 (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]>;
2675 // Long Multiply-Add/Sub operations.
2676 class N3VLMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2677 InstrItinClass itin, string OpcodeStr, string Dt,
2678 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2679 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2680 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2681 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2682 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
2683 (TyQ (MulOp (TyD DPR:$Vn),
2684 (TyD DPR:$Vm)))))]>;
2685 class N3VLMulOpSL<bit op24, bits<2> op21_20, bits<4> op11_8,
2686 InstrItinClass itin, string OpcodeStr, string Dt,
2687 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2688 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$Vd),
2689 (ins QPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2691 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2693 (OpNode (TyQ QPR:$src1),
2694 (TyQ (MulOp (TyD DPR:$Vn),
2695 (TyD (NEONvduplane (TyD DPR_VFP2:$Vm),
2697 class N3VLMulOpSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2698 InstrItinClass itin, string OpcodeStr, string Dt,
2699 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2700 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$Vd),
2701 (ins QPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2703 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2705 (OpNode (TyQ QPR:$src1),
2706 (TyQ (MulOp (TyD DPR:$Vn),
2707 (TyD (NEONvduplane (TyD DPR_8:$Vm),
2710 // Long Intrinsic-Op vector operations with explicit extend (VABAL).
2711 class N3VLIntExtOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2712 InstrItinClass itin, string OpcodeStr, string Dt,
2713 ValueType TyQ, ValueType TyD, Intrinsic IntOp, SDNode ExtOp,
2715 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2716 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2717 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2718 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
2719 (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
2720 (TyD DPR:$Vm)))))))]>;
2722 // Neon Long 3-argument intrinsic. The destination register is
2723 // a quad-register and is also used as the first source operand register.
2724 class N3VLInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2725 InstrItinClass itin, string OpcodeStr, string Dt,
2726 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
2727 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2728 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2729 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2731 (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$Vn), (TyD DPR:$Vm))))]>;
2732 class N3VLInt3SL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2733 string OpcodeStr, string Dt,
2734 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2735 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2737 (ins QPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2739 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2740 [(set (ResTy QPR:$Vd),
2741 (ResTy (IntOp (ResTy QPR:$src1),
2743 (OpTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2745 class N3VLInt3SL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2746 InstrItinClass itin, string OpcodeStr, string Dt,
2747 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2748 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2750 (ins QPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2752 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2753 [(set (ResTy QPR:$Vd),
2754 (ResTy (IntOp (ResTy QPR:$src1),
2756 (OpTy (NEONvduplane (OpTy DPR_8:$Vm),
2759 // Narrowing 3-register intrinsics.
2760 class N3VNInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2761 string OpcodeStr, string Dt, ValueType TyD, ValueType TyQ,
2762 Intrinsic IntOp, bit Commutable>
2763 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2764 (outs DPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINi4D,
2765 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2766 [(set DPR:$Vd, (TyD (IntOp (TyQ QPR:$Vn), (TyQ QPR:$Vm))))]> {
2767 let isCommutable = Commutable;
2770 // Long 3-register operations.
2771 class N3VL<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2772 InstrItinClass itin, string OpcodeStr, string Dt,
2773 ValueType TyQ, ValueType TyD, SDNode OpNode, bit Commutable>
2774 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2775 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2776 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2777 [(set QPR:$Vd, (TyQ (OpNode (TyD DPR:$Vn), (TyD DPR:$Vm))))]> {
2778 let isCommutable = Commutable;
2780 class N3VLSL<bit op24, bits<2> op21_20, bits<4> op11_8,
2781 InstrItinClass itin, string OpcodeStr, string Dt,
2782 ValueType TyQ, ValueType TyD, SDNode OpNode>
2783 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2784 (outs QPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2785 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2787 (TyQ (OpNode (TyD DPR:$Vn),
2788 (TyD (NEONvduplane (TyD DPR_VFP2:$Vm),imm:$lane)))))]>;
2789 class N3VLSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2790 InstrItinClass itin, string OpcodeStr, string Dt,
2791 ValueType TyQ, ValueType TyD, SDNode OpNode>
2792 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2793 (outs QPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2794 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2796 (TyQ (OpNode (TyD DPR:$Vn),
2797 (TyD (NEONvduplane (TyD DPR_8:$Vm), imm:$lane)))))]>;
2799 // Long 3-register operations with explicitly extended operands.
2800 class N3VLExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2801 InstrItinClass itin, string OpcodeStr, string Dt,
2802 ValueType TyQ, ValueType TyD, SDNode OpNode, SDNode ExtOp,
2804 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2805 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2806 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2807 [(set QPR:$Vd, (OpNode (TyQ (ExtOp (TyD DPR:$Vn))),
2808 (TyQ (ExtOp (TyD DPR:$Vm)))))]> {
2809 let isCommutable = Commutable;
2812 // Long 3-register intrinsics with explicit extend (VABDL).
2813 class N3VLIntExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2814 InstrItinClass itin, string OpcodeStr, string Dt,
2815 ValueType TyQ, ValueType TyD, Intrinsic IntOp, SDNode ExtOp,
2817 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2818 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2819 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2820 [(set QPR:$Vd, (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
2821 (TyD DPR:$Vm))))))]> {
2822 let isCommutable = Commutable;
2825 // Long 3-register intrinsics.
2826 class N3VLInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2827 InstrItinClass itin, string OpcodeStr, string Dt,
2828 ValueType TyQ, ValueType TyD, Intrinsic IntOp, bit Commutable>
2829 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2830 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2831 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2832 [(set QPR:$Vd, (TyQ (IntOp (TyD DPR:$Vn), (TyD DPR:$Vm))))]> {
2833 let isCommutable = Commutable;
2835 class N3VLIntSL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2836 string OpcodeStr, string Dt,
2837 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2838 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2839 (outs QPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$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_VFP2:$Vm),
2845 class N3VLIntSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2846 InstrItinClass itin, string OpcodeStr, string Dt,
2847 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2848 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2849 (outs QPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2850 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2851 [(set (ResTy QPR:$Vd),
2852 (ResTy (IntOp (OpTy DPR:$Vn),
2853 (OpTy (NEONvduplane (OpTy DPR_8:$Vm),
2856 // Wide 3-register operations.
2857 class N3VW<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2858 string OpcodeStr, string Dt, ValueType TyQ, ValueType TyD,
2859 SDNode OpNode, SDNode ExtOp, bit Commutable>
2860 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2861 (outs QPR:$Vd), (ins QPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VSUBiD,
2862 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2863 [(set QPR:$Vd, (OpNode (TyQ QPR:$Vn),
2864 (TyQ (ExtOp (TyD DPR:$Vm)))))]> {
2865 let isCommutable = Commutable;
2868 // Pairwise long 2-register intrinsics, both double- and quad-register.
2869 class N2VDPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2870 bits<2> op17_16, bits<5> op11_7, bit op4,
2871 string OpcodeStr, string Dt,
2872 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2873 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
2874 (ins DPR:$Vm), IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
2875 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm))))]>;
2876 class N2VQPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2877 bits<2> op17_16, bits<5> op11_7, bit op4,
2878 string OpcodeStr, string Dt,
2879 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2880 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
2881 (ins QPR:$Vm), IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
2882 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm))))]>;
2884 // Pairwise long 2-register accumulate intrinsics,
2885 // both double- and quad-register.
2886 // The destination register is also used as the first source operand register.
2887 class N2VDPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2888 bits<2> op17_16, bits<5> op11_7, bit op4,
2889 string OpcodeStr, string Dt,
2890 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2891 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
2892 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vm), IIC_VPALiD,
2893 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
2894 [(set DPR:$Vd, (ResTy (IntOp (ResTy DPR:$src1), (OpTy DPR:$Vm))))]>;
2895 class N2VQPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2896 bits<2> op17_16, bits<5> op11_7, bit op4,
2897 string OpcodeStr, string Dt,
2898 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2899 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4,
2900 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vm), IIC_VPALiQ,
2901 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
2902 [(set QPR:$Vd, (ResTy (IntOp (ResTy QPR:$src1), (OpTy QPR:$Vm))))]>;
2904 // Shift by immediate,
2905 // both double- and quad-register.
2906 class N2VDSh<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, 0, op4,
2910 (outs DPR:$Vd), (ins DPR:$Vm, ImmTy:$SIMM), f, itin,
2911 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2912 [(set DPR:$Vd, (Ty (OpNode (Ty DPR:$Vm), (i32 imm:$SIMM))))]>;
2913 class N2VQSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2914 Format f, InstrItinClass itin, Operand ImmTy,
2915 string OpcodeStr, string Dt, ValueType Ty, SDNode OpNode>
2916 : N2VImm<op24, op23, op11_8, op7, 1, op4,
2917 (outs QPR:$Vd), (ins QPR:$Vm, ImmTy:$SIMM), f, itin,
2918 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2919 [(set QPR:$Vd, (Ty (OpNode (Ty QPR:$Vm), (i32 imm:$SIMM))))]>;
2921 // Long shift by immediate.
2922 class N2VLSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
2923 string OpcodeStr, string Dt,
2924 ValueType ResTy, ValueType OpTy, Operand ImmTy, SDNode OpNode>
2925 : N2VImm<op24, op23, op11_8, op7, op6, op4,
2926 (outs QPR:$Vd), (ins DPR:$Vm, ImmTy:$SIMM), N2RegVShLFrm,
2927 IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2928 [(set QPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vm),
2929 (i32 imm:$SIMM))))]>;
2931 // Narrow shift by immediate.
2932 class N2VNSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
2933 InstrItinClass itin, string OpcodeStr, string Dt,
2934 ValueType ResTy, ValueType OpTy, Operand ImmTy, SDNode OpNode>
2935 : N2VImm<op24, op23, op11_8, op7, op6, op4,
2936 (outs DPR:$Vd), (ins QPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, itin,
2937 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2938 [(set DPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vm),
2939 (i32 imm:$SIMM))))]>;
2941 // Shift right by immediate and accumulate,
2942 // both double- and quad-register.
2943 class N2VDShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2944 Operand ImmTy, string OpcodeStr, string Dt,
2945 ValueType Ty, SDNode ShOp>
2946 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
2947 (ins DPR:$src1, DPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, IIC_VPALiD,
2948 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2949 [(set DPR:$Vd, (Ty (add DPR:$src1,
2950 (Ty (ShOp DPR:$Vm, (i32 imm:$SIMM))))))]>;
2951 class N2VQShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2952 Operand ImmTy, string OpcodeStr, string Dt,
2953 ValueType Ty, SDNode ShOp>
2954 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
2955 (ins QPR:$src1, QPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, IIC_VPALiD,
2956 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2957 [(set QPR:$Vd, (Ty (add QPR:$src1,
2958 (Ty (ShOp QPR:$Vm, (i32 imm:$SIMM))))))]>;
2960 // Shift by immediate and insert,
2961 // both double- and quad-register.
2962 class N2VDShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2963 Operand ImmTy, Format f, string OpcodeStr, string Dt,
2964 ValueType Ty,SDNode ShOp>
2965 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
2966 (ins DPR:$src1, DPR:$Vm, ImmTy:$SIMM), f, IIC_VSHLiD,
2967 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2968 [(set DPR:$Vd, (Ty (ShOp DPR:$src1, DPR:$Vm, (i32 imm:$SIMM))))]>;
2969 class N2VQShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2970 Operand ImmTy, Format f, string OpcodeStr, string Dt,
2971 ValueType Ty,SDNode ShOp>
2972 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
2973 (ins QPR:$src1, QPR:$Vm, ImmTy:$SIMM), f, IIC_VSHLiQ,
2974 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2975 [(set QPR:$Vd, (Ty (ShOp QPR:$src1, QPR:$Vm, (i32 imm:$SIMM))))]>;
2977 // Convert, with fractional bits immediate,
2978 // both double- and quad-register.
2979 class N2VCvtD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2980 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2982 : N2VImm<op24, op23, op11_8, op7, 0, op4,
2983 (outs DPR:$Vd), (ins DPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
2984 IIC_VUNAD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2985 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (i32 imm:$SIMM))))]>;
2986 class N2VCvtQ<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2987 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2989 : N2VImm<op24, op23, op11_8, op7, 1, op4,
2990 (outs QPR:$Vd), (ins QPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
2991 IIC_VUNAQ, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2992 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (i32 imm:$SIMM))))]>;
2994 //===----------------------------------------------------------------------===//
2996 //===----------------------------------------------------------------------===//
2998 // Abbreviations used in multiclass suffixes:
2999 // Q = quarter int (8 bit) elements
3000 // H = half int (16 bit) elements
3001 // S = single int (32 bit) elements
3002 // D = double int (64 bit) elements
3004 // Neon 2-register vector operations and intrinsics.
3006 // Neon 2-register comparisons.
3007 // source operand element sizes of 8, 16 and 32 bits:
3008 multiclass N2V_QHS_cmp<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3009 bits<5> op11_7, bit op4, string opc, string Dt,
3010 string asm, SDNode OpNode> {
3011 // 64-bit vector types.
3012 def v8i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 0, op4,
3013 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
3014 opc, !strconcat(Dt, "8"), asm, "",
3015 [(set DPR:$Vd, (v8i8 (OpNode (v8i8 DPR:$Vm))))]>;
3016 def v4i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 0, op4,
3017 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
3018 opc, !strconcat(Dt, "16"), asm, "",
3019 [(set DPR:$Vd, (v4i16 (OpNode (v4i16 DPR:$Vm))))]>;
3020 def v2i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
3021 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
3022 opc, !strconcat(Dt, "32"), asm, "",
3023 [(set DPR:$Vd, (v2i32 (OpNode (v2i32 DPR:$Vm))))]>;
3024 def v2f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
3025 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
3026 opc, "f32", asm, "",
3027 [(set DPR:$Vd, (v2i32 (OpNode (v2f32 DPR:$Vm))))]> {
3028 let Inst{10} = 1; // overwrite F = 1
3031 // 128-bit vector types.
3032 def v16i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 1, op4,
3033 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
3034 opc, !strconcat(Dt, "8"), asm, "",
3035 [(set QPR:$Vd, (v16i8 (OpNode (v16i8 QPR:$Vm))))]>;
3036 def v8i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 1, op4,
3037 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
3038 opc, !strconcat(Dt, "16"), asm, "",
3039 [(set QPR:$Vd, (v8i16 (OpNode (v8i16 QPR:$Vm))))]>;
3040 def v4i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
3041 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
3042 opc, !strconcat(Dt, "32"), asm, "",
3043 [(set QPR:$Vd, (v4i32 (OpNode (v4i32 QPR:$Vm))))]>;
3044 def v4f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
3045 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
3046 opc, "f32", asm, "",
3047 [(set QPR:$Vd, (v4i32 (OpNode (v4f32 QPR:$Vm))))]> {
3048 let Inst{10} = 1; // overwrite F = 1
3053 // Neon 2-register vector intrinsics,
3054 // element sizes of 8, 16 and 32 bits:
3055 multiclass N2VInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3056 bits<5> op11_7, bit op4,
3057 InstrItinClass itinD, InstrItinClass itinQ,
3058 string OpcodeStr, string Dt, Intrinsic IntOp> {
3059 // 64-bit vector types.
3060 def v8i8 : N2VDInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3061 itinD, OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
3062 def v4i16 : N2VDInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3063 itinD, OpcodeStr, !strconcat(Dt, "16"),v4i16,v4i16,IntOp>;
3064 def v2i32 : N2VDInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3065 itinD, OpcodeStr, !strconcat(Dt, "32"),v2i32,v2i32,IntOp>;
3067 // 128-bit vector types.
3068 def v16i8 : N2VQInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3069 itinQ, OpcodeStr, !strconcat(Dt, "8"), v16i8,v16i8,IntOp>;
3070 def v8i16 : N2VQInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3071 itinQ, OpcodeStr, !strconcat(Dt, "16"),v8i16,v8i16,IntOp>;
3072 def v4i32 : N2VQInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3073 itinQ, OpcodeStr, !strconcat(Dt, "32"),v4i32,v4i32,IntOp>;
3077 // Neon Narrowing 2-register vector operations,
3078 // source operand element sizes of 16, 32 and 64 bits:
3079 multiclass N2VN_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3080 bits<5> op11_7, bit op6, bit op4,
3081 InstrItinClass itin, string OpcodeStr, string Dt,
3083 def v8i8 : N2VN<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
3084 itin, OpcodeStr, !strconcat(Dt, "16"),
3085 v8i8, v8i16, OpNode>;
3086 def v4i16 : N2VN<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
3087 itin, OpcodeStr, !strconcat(Dt, "32"),
3088 v4i16, v4i32, OpNode>;
3089 def v2i32 : N2VN<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
3090 itin, OpcodeStr, !strconcat(Dt, "64"),
3091 v2i32, v2i64, OpNode>;
3094 // Neon Narrowing 2-register vector intrinsics,
3095 // source operand element sizes of 16, 32 and 64 bits:
3096 multiclass N2VNInt_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3097 bits<5> op11_7, bit op6, bit op4,
3098 InstrItinClass itin, string OpcodeStr, string Dt,
3100 def v8i8 : N2VNInt<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
3101 itin, OpcodeStr, !strconcat(Dt, "16"),
3102 v8i8, v8i16, IntOp>;
3103 def v4i16 : N2VNInt<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
3104 itin, OpcodeStr, !strconcat(Dt, "32"),
3105 v4i16, v4i32, IntOp>;
3106 def v2i32 : N2VNInt<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
3107 itin, OpcodeStr, !strconcat(Dt, "64"),
3108 v2i32, v2i64, IntOp>;
3112 // Neon Lengthening 2-register vector intrinsic (currently specific to VMOVL).
3113 // source operand element sizes of 16, 32 and 64 bits:
3114 multiclass N2VL_QHS<bits<2> op24_23, bits<5> op11_7, bit op6, bit op4,
3115 string OpcodeStr, string Dt, SDNode OpNode> {
3116 def v8i16 : N2VL<op24_23, 0b00, 0b10, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
3117 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode>;
3118 def v4i32 : N2VL<op24_23, 0b01, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
3119 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
3120 def v2i64 : N2VL<op24_23, 0b10, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
3121 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
3125 // Neon 3-register vector operations.
3127 // First with only element sizes of 8, 16 and 32 bits:
3128 multiclass N3V_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3129 InstrItinClass itinD16, InstrItinClass itinD32,
3130 InstrItinClass itinQ16, InstrItinClass itinQ32,
3131 string OpcodeStr, string Dt,
3132 SDNode OpNode, bit Commutable = 0> {
3133 // 64-bit vector types.
3134 def v8i8 : N3VD<op24, op23, 0b00, op11_8, op4, itinD16,
3135 OpcodeStr, !strconcat(Dt, "8"),
3136 v8i8, v8i8, OpNode, Commutable>;
3137 def v4i16 : N3VD<op24, op23, 0b01, op11_8, op4, itinD16,
3138 OpcodeStr, !strconcat(Dt, "16"),
3139 v4i16, v4i16, OpNode, Commutable>;
3140 def v2i32 : N3VD<op24, op23, 0b10, op11_8, op4, itinD32,
3141 OpcodeStr, !strconcat(Dt, "32"),
3142 v2i32, v2i32, OpNode, Commutable>;
3144 // 128-bit vector types.
3145 def v16i8 : N3VQ<op24, op23, 0b00, op11_8, op4, itinQ16,
3146 OpcodeStr, !strconcat(Dt, "8"),
3147 v16i8, v16i8, OpNode, Commutable>;
3148 def v8i16 : N3VQ<op24, op23, 0b01, op11_8, op4, itinQ16,
3149 OpcodeStr, !strconcat(Dt, "16"),
3150 v8i16, v8i16, OpNode, Commutable>;
3151 def v4i32 : N3VQ<op24, op23, 0b10, op11_8, op4, itinQ32,
3152 OpcodeStr, !strconcat(Dt, "32"),
3153 v4i32, v4i32, OpNode, Commutable>;
3156 multiclass N3VSL_HS<bits<4> op11_8, string OpcodeStr, SDNode ShOp> {
3157 def v4i16 : N3VDSL16<0b01, op11_8, OpcodeStr, "i16", v4i16, ShOp>;
3158 def v2i32 : N3VDSL<0b10, op11_8, IIC_VMULi32D, OpcodeStr, "i32", v2i32, ShOp>;
3159 def v8i16 : N3VQSL16<0b01, op11_8, OpcodeStr, "i16", v8i16, v4i16, ShOp>;
3160 def v4i32 : N3VQSL<0b10, op11_8, IIC_VMULi32Q, OpcodeStr, "i32",
3161 v4i32, v2i32, ShOp>;
3164 // ....then also with element size 64 bits:
3165 multiclass N3V_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3166 InstrItinClass itinD, InstrItinClass itinQ,
3167 string OpcodeStr, string Dt,
3168 SDNode OpNode, bit Commutable = 0>
3169 : N3V_QHS<op24, op23, op11_8, op4, itinD, itinD, itinQ, itinQ,
3170 OpcodeStr, Dt, OpNode, Commutable> {
3171 def v1i64 : N3VD<op24, op23, 0b11, op11_8, op4, itinD,
3172 OpcodeStr, !strconcat(Dt, "64"),
3173 v1i64, v1i64, OpNode, Commutable>;
3174 def v2i64 : N3VQ<op24, op23, 0b11, op11_8, op4, itinQ,
3175 OpcodeStr, !strconcat(Dt, "64"),
3176 v2i64, v2i64, OpNode, Commutable>;
3180 // Neon 3-register vector intrinsics.
3182 // First with only element sizes of 16 and 32 bits:
3183 multiclass N3VInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3184 InstrItinClass itinD16, InstrItinClass itinD32,
3185 InstrItinClass itinQ16, InstrItinClass itinQ32,
3186 string OpcodeStr, string Dt,
3187 Intrinsic IntOp, bit Commutable = 0> {
3188 // 64-bit vector types.
3189 def v4i16 : N3VDInt<op24, op23, 0b01, op11_8, op4, f, itinD16,
3190 OpcodeStr, !strconcat(Dt, "16"),
3191 v4i16, v4i16, IntOp, Commutable>;
3192 def v2i32 : N3VDInt<op24, op23, 0b10, op11_8, op4, f, itinD32,
3193 OpcodeStr, !strconcat(Dt, "32"),
3194 v2i32, v2i32, IntOp, Commutable>;
3196 // 128-bit vector types.
3197 def v8i16 : N3VQInt<op24, op23, 0b01, op11_8, op4, f, itinQ16,
3198 OpcodeStr, !strconcat(Dt, "16"),
3199 v8i16, v8i16, IntOp, Commutable>;
3200 def v4i32 : N3VQInt<op24, op23, 0b10, op11_8, op4, f, itinQ32,
3201 OpcodeStr, !strconcat(Dt, "32"),
3202 v4i32, v4i32, IntOp, Commutable>;
3204 multiclass N3VInt_HSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3205 InstrItinClass itinD16, InstrItinClass itinD32,
3206 InstrItinClass itinQ16, InstrItinClass itinQ32,
3207 string OpcodeStr, string Dt,
3209 // 64-bit vector types.
3210 def v4i16 : N3VDIntSh<op24, op23, 0b01, op11_8, op4, f, itinD16,
3211 OpcodeStr, !strconcat(Dt, "16"),
3212 v4i16, v4i16, IntOp>;
3213 def v2i32 : N3VDIntSh<op24, op23, 0b10, op11_8, op4, f, itinD32,
3214 OpcodeStr, !strconcat(Dt, "32"),
3215 v2i32, v2i32, IntOp>;
3217 // 128-bit vector types.
3218 def v8i16 : N3VQIntSh<op24, op23, 0b01, op11_8, op4, f, itinQ16,
3219 OpcodeStr, !strconcat(Dt, "16"),
3220 v8i16, v8i16, IntOp>;
3221 def v4i32 : N3VQIntSh<op24, op23, 0b10, op11_8, op4, f, itinQ32,
3222 OpcodeStr, !strconcat(Dt, "32"),
3223 v4i32, v4i32, IntOp>;
3226 multiclass N3VIntSL_HS<bits<4> op11_8,
3227 InstrItinClass itinD16, InstrItinClass itinD32,
3228 InstrItinClass itinQ16, InstrItinClass itinQ32,
3229 string OpcodeStr, string Dt, Intrinsic IntOp> {
3230 def v4i16 : N3VDIntSL16<0b01, op11_8, itinD16,
3231 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp>;
3232 def v2i32 : N3VDIntSL<0b10, op11_8, itinD32,
3233 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp>;
3234 def v8i16 : N3VQIntSL16<0b01, op11_8, itinQ16,
3235 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16, IntOp>;
3236 def v4i32 : N3VQIntSL<0b10, op11_8, itinQ32,
3237 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32, IntOp>;
3240 // ....then also with element size of 8 bits:
3241 multiclass N3VInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3242 InstrItinClass itinD16, InstrItinClass itinD32,
3243 InstrItinClass itinQ16, InstrItinClass itinQ32,
3244 string OpcodeStr, string Dt,
3245 Intrinsic IntOp, bit Commutable = 0>
3246 : N3VInt_HS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
3247 OpcodeStr, Dt, IntOp, Commutable> {
3248 def v8i8 : N3VDInt<op24, op23, 0b00, op11_8, op4, f, itinD16,
3249 OpcodeStr, !strconcat(Dt, "8"),
3250 v8i8, v8i8, IntOp, Commutable>;
3251 def v16i8 : N3VQInt<op24, op23, 0b00, op11_8, op4, f, itinQ16,
3252 OpcodeStr, !strconcat(Dt, "8"),
3253 v16i8, v16i8, IntOp, Commutable>;
3255 multiclass N3VInt_QHSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3256 InstrItinClass itinD16, InstrItinClass itinD32,
3257 InstrItinClass itinQ16, InstrItinClass itinQ32,
3258 string OpcodeStr, string Dt,
3260 : N3VInt_HSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
3261 OpcodeStr, Dt, IntOp> {
3262 def v8i8 : N3VDIntSh<op24, op23, 0b00, op11_8, op4, f, itinD16,
3263 OpcodeStr, !strconcat(Dt, "8"),
3265 def v16i8 : N3VQIntSh<op24, op23, 0b00, op11_8, op4, f, itinQ16,
3266 OpcodeStr, !strconcat(Dt, "8"),
3267 v16i8, v16i8, IntOp>;
3271 // ....then also with element size of 64 bits:
3272 multiclass N3VInt_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3273 InstrItinClass itinD16, InstrItinClass itinD32,
3274 InstrItinClass itinQ16, InstrItinClass itinQ32,
3275 string OpcodeStr, string Dt,
3276 Intrinsic IntOp, bit Commutable = 0>
3277 : N3VInt_QHS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
3278 OpcodeStr, Dt, IntOp, Commutable> {
3279 def v1i64 : N3VDInt<op24, op23, 0b11, op11_8, op4, f, itinD32,
3280 OpcodeStr, !strconcat(Dt, "64"),
3281 v1i64, v1i64, IntOp, Commutable>;
3282 def v2i64 : N3VQInt<op24, op23, 0b11, op11_8, op4, f, itinQ32,
3283 OpcodeStr, !strconcat(Dt, "64"),
3284 v2i64, v2i64, IntOp, Commutable>;
3286 multiclass N3VInt_QHSDSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3287 InstrItinClass itinD16, InstrItinClass itinD32,
3288 InstrItinClass itinQ16, InstrItinClass itinQ32,
3289 string OpcodeStr, string Dt,
3291 : N3VInt_QHSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
3292 OpcodeStr, Dt, IntOp> {
3293 def v1i64 : N3VDIntSh<op24, op23, 0b11, op11_8, op4, f, itinD32,
3294 OpcodeStr, !strconcat(Dt, "64"),
3295 v1i64, v1i64, IntOp>;
3296 def v2i64 : N3VQIntSh<op24, op23, 0b11, op11_8, op4, f, itinQ32,
3297 OpcodeStr, !strconcat(Dt, "64"),
3298 v2i64, v2i64, IntOp>;
3301 // Neon Narrowing 3-register vector intrinsics,
3302 // source operand element sizes of 16, 32 and 64 bits:
3303 multiclass N3VNInt_HSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3304 string OpcodeStr, string Dt,
3305 Intrinsic IntOp, bit Commutable = 0> {
3306 def v8i8 : N3VNInt<op24, op23, 0b00, op11_8, op4,
3307 OpcodeStr, !strconcat(Dt, "16"),
3308 v8i8, v8i16, IntOp, Commutable>;
3309 def v4i16 : N3VNInt<op24, op23, 0b01, op11_8, op4,
3310 OpcodeStr, !strconcat(Dt, "32"),
3311 v4i16, v4i32, IntOp, Commutable>;
3312 def v2i32 : N3VNInt<op24, op23, 0b10, op11_8, op4,
3313 OpcodeStr, !strconcat(Dt, "64"),
3314 v2i32, v2i64, IntOp, Commutable>;
3318 // Neon Long 3-register vector operations.
3320 multiclass N3VL_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3321 InstrItinClass itin16, InstrItinClass itin32,
3322 string OpcodeStr, string Dt,
3323 SDNode OpNode, bit Commutable = 0> {
3324 def v8i16 : N3VL<op24, op23, 0b00, op11_8, op4, itin16,
3325 OpcodeStr, !strconcat(Dt, "8"),
3326 v8i16, v8i8, OpNode, Commutable>;
3327 def v4i32 : N3VL<op24, op23, 0b01, op11_8, op4, itin16,
3328 OpcodeStr, !strconcat(Dt, "16"),
3329 v4i32, v4i16, OpNode, Commutable>;
3330 def v2i64 : N3VL<op24, op23, 0b10, op11_8, op4, itin32,
3331 OpcodeStr, !strconcat(Dt, "32"),
3332 v2i64, v2i32, OpNode, Commutable>;
3335 multiclass N3VLSL_HS<bit op24, bits<4> op11_8,
3336 InstrItinClass itin, string OpcodeStr, string Dt,
3338 def v4i16 : N3VLSL16<op24, 0b01, op11_8, itin, OpcodeStr,
3339 !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
3340 def v2i32 : N3VLSL<op24, 0b10, op11_8, itin, OpcodeStr,
3341 !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
3344 multiclass N3VLExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3345 InstrItinClass itin16, InstrItinClass itin32,
3346 string OpcodeStr, string Dt,
3347 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
3348 def v8i16 : N3VLExt<op24, op23, 0b00, op11_8, op4, itin16,
3349 OpcodeStr, !strconcat(Dt, "8"),
3350 v8i16, v8i8, OpNode, ExtOp, Commutable>;
3351 def v4i32 : N3VLExt<op24, op23, 0b01, op11_8, op4, itin16,
3352 OpcodeStr, !strconcat(Dt, "16"),
3353 v4i32, v4i16, OpNode, ExtOp, Commutable>;
3354 def v2i64 : N3VLExt<op24, op23, 0b10, op11_8, op4, itin32,
3355 OpcodeStr, !strconcat(Dt, "32"),
3356 v2i64, v2i32, OpNode, ExtOp, Commutable>;
3359 // Neon Long 3-register vector intrinsics.
3361 // First with only element sizes of 16 and 32 bits:
3362 multiclass N3VLInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
3363 InstrItinClass itin16, InstrItinClass itin32,
3364 string OpcodeStr, string Dt,
3365 Intrinsic IntOp, bit Commutable = 0> {
3366 def v4i32 : N3VLInt<op24, op23, 0b01, op11_8, op4, itin16,
3367 OpcodeStr, !strconcat(Dt, "16"),
3368 v4i32, v4i16, IntOp, Commutable>;
3369 def v2i64 : N3VLInt<op24, op23, 0b10, op11_8, op4, itin32,
3370 OpcodeStr, !strconcat(Dt, "32"),
3371 v2i64, v2i32, IntOp, Commutable>;
3374 multiclass N3VLIntSL_HS<bit op24, bits<4> op11_8,
3375 InstrItinClass itin, string OpcodeStr, string Dt,
3377 def v4i16 : N3VLIntSL16<op24, 0b01, op11_8, itin,
3378 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
3379 def v2i32 : N3VLIntSL<op24, 0b10, op11_8, itin,
3380 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3383 // ....then also with element size of 8 bits:
3384 multiclass N3VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3385 InstrItinClass itin16, InstrItinClass itin32,
3386 string OpcodeStr, string Dt,
3387 Intrinsic IntOp, bit Commutable = 0>
3388 : N3VLInt_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt,
3389 IntOp, Commutable> {
3390 def v8i16 : N3VLInt<op24, op23, 0b00, op11_8, op4, itin16,
3391 OpcodeStr, !strconcat(Dt, "8"),
3392 v8i16, v8i8, IntOp, Commutable>;
3395 // ....with explicit extend (VABDL).
3396 multiclass N3VLIntExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3397 InstrItinClass itin, string OpcodeStr, string Dt,
3398 Intrinsic IntOp, SDNode ExtOp, bit Commutable = 0> {
3399 def v8i16 : N3VLIntExt<op24, op23, 0b00, op11_8, op4, itin,
3400 OpcodeStr, !strconcat(Dt, "8"),
3401 v8i16, v8i8, IntOp, ExtOp, Commutable>;
3402 def v4i32 : N3VLIntExt<op24, op23, 0b01, op11_8, op4, itin,
3403 OpcodeStr, !strconcat(Dt, "16"),
3404 v4i32, v4i16, IntOp, ExtOp, Commutable>;
3405 def v2i64 : N3VLIntExt<op24, op23, 0b10, op11_8, op4, itin,
3406 OpcodeStr, !strconcat(Dt, "32"),
3407 v2i64, v2i32, IntOp, ExtOp, Commutable>;
3411 // Neon Wide 3-register vector intrinsics,
3412 // source operand element sizes of 8, 16 and 32 bits:
3413 multiclass N3VW_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3414 string OpcodeStr, string Dt,
3415 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
3416 def v8i16 : N3VW<op24, op23, 0b00, op11_8, op4,
3417 OpcodeStr, !strconcat(Dt, "8"),
3418 v8i16, v8i8, OpNode, ExtOp, Commutable>;
3419 def v4i32 : N3VW<op24, op23, 0b01, op11_8, op4,
3420 OpcodeStr, !strconcat(Dt, "16"),
3421 v4i32, v4i16, OpNode, ExtOp, Commutable>;
3422 def v2i64 : N3VW<op24, op23, 0b10, op11_8, op4,
3423 OpcodeStr, !strconcat(Dt, "32"),
3424 v2i64, v2i32, OpNode, ExtOp, Commutable>;
3428 // Neon Multiply-Op vector operations,
3429 // element sizes of 8, 16 and 32 bits:
3430 multiclass N3VMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3431 InstrItinClass itinD16, InstrItinClass itinD32,
3432 InstrItinClass itinQ16, InstrItinClass itinQ32,
3433 string OpcodeStr, string Dt, SDNode OpNode> {
3434 // 64-bit vector types.
3435 def v8i8 : N3VDMulOp<op24, op23, 0b00, op11_8, op4, itinD16,
3436 OpcodeStr, !strconcat(Dt, "8"), v8i8, mul, OpNode>;
3437 def v4i16 : N3VDMulOp<op24, op23, 0b01, op11_8, op4, itinD16,
3438 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, OpNode>;
3439 def v2i32 : N3VDMulOp<op24, op23, 0b10, op11_8, op4, itinD32,
3440 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, OpNode>;
3442 // 128-bit vector types.
3443 def v16i8 : N3VQMulOp<op24, op23, 0b00, op11_8, op4, itinQ16,
3444 OpcodeStr, !strconcat(Dt, "8"), v16i8, mul, OpNode>;
3445 def v8i16 : N3VQMulOp<op24, op23, 0b01, op11_8, op4, itinQ16,
3446 OpcodeStr, !strconcat(Dt, "16"), v8i16, mul, OpNode>;
3447 def v4i32 : N3VQMulOp<op24, op23, 0b10, op11_8, op4, itinQ32,
3448 OpcodeStr, !strconcat(Dt, "32"), v4i32, mul, OpNode>;
3451 multiclass N3VMulOpSL_HS<bits<4> op11_8,
3452 InstrItinClass itinD16, InstrItinClass itinD32,
3453 InstrItinClass itinQ16, InstrItinClass itinQ32,
3454 string OpcodeStr, string Dt, SDNode ShOp> {
3455 def v4i16 : N3VDMulOpSL16<0b01, op11_8, itinD16,
3456 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, ShOp>;
3457 def v2i32 : N3VDMulOpSL<0b10, op11_8, itinD32,
3458 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, ShOp>;
3459 def v8i16 : N3VQMulOpSL16<0b01, op11_8, itinQ16,
3460 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16,
3462 def v4i32 : N3VQMulOpSL<0b10, op11_8, itinQ32,
3463 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32,
3467 // Neon Intrinsic-Op vector operations,
3468 // element sizes of 8, 16 and 32 bits:
3469 multiclass N3VIntOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3470 InstrItinClass itinD, InstrItinClass itinQ,
3471 string OpcodeStr, string Dt, Intrinsic IntOp,
3473 // 64-bit vector types.
3474 def v8i8 : N3VDIntOp<op24, op23, 0b00, op11_8, op4, itinD,
3475 OpcodeStr, !strconcat(Dt, "8"), v8i8, IntOp, OpNode>;
3476 def v4i16 : N3VDIntOp<op24, op23, 0b01, op11_8, op4, itinD,
3477 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp, OpNode>;
3478 def v2i32 : N3VDIntOp<op24, op23, 0b10, op11_8, op4, itinD,
3479 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp, OpNode>;
3481 // 128-bit vector types.
3482 def v16i8 : N3VQIntOp<op24, op23, 0b00, op11_8, op4, itinQ,
3483 OpcodeStr, !strconcat(Dt, "8"), v16i8, IntOp, OpNode>;
3484 def v8i16 : N3VQIntOp<op24, op23, 0b01, op11_8, op4, itinQ,
3485 OpcodeStr, !strconcat(Dt, "16"), v8i16, IntOp, OpNode>;
3486 def v4i32 : N3VQIntOp<op24, op23, 0b10, op11_8, op4, itinQ,
3487 OpcodeStr, !strconcat(Dt, "32"), v4i32, IntOp, OpNode>;
3490 // Neon 3-argument intrinsics,
3491 // element sizes of 8, 16 and 32 bits:
3492 multiclass N3VInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3493 InstrItinClass itinD, InstrItinClass itinQ,
3494 string OpcodeStr, string Dt, Intrinsic IntOp> {
3495 // 64-bit vector types.
3496 def v8i8 : N3VDInt3<op24, op23, 0b00, op11_8, op4, itinD,
3497 OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
3498 def v4i16 : N3VDInt3<op24, op23, 0b01, op11_8, op4, itinD,
3499 OpcodeStr, !strconcat(Dt, "16"), v4i16, v4i16, IntOp>;
3500 def v2i32 : N3VDInt3<op24, op23, 0b10, op11_8, op4, itinD,
3501 OpcodeStr, !strconcat(Dt, "32"), v2i32, v2i32, IntOp>;
3503 // 128-bit vector types.
3504 def v16i8 : N3VQInt3<op24, op23, 0b00, op11_8, op4, itinQ,
3505 OpcodeStr, !strconcat(Dt, "8"), v16i8, v16i8, IntOp>;
3506 def v8i16 : N3VQInt3<op24, op23, 0b01, op11_8, op4, itinQ,
3507 OpcodeStr, !strconcat(Dt, "16"), v8i16, v8i16, IntOp>;
3508 def v4i32 : N3VQInt3<op24, op23, 0b10, op11_8, op4, itinQ,
3509 OpcodeStr, !strconcat(Dt, "32"), v4i32, v4i32, IntOp>;
3513 // Neon Long Multiply-Op vector operations,
3514 // element sizes of 8, 16 and 32 bits:
3515 multiclass N3VLMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3516 InstrItinClass itin16, InstrItinClass itin32,
3517 string OpcodeStr, string Dt, SDNode MulOp,
3519 def v8i16 : N3VLMulOp<op24, op23, 0b00, op11_8, op4, itin16, OpcodeStr,
3520 !strconcat(Dt, "8"), v8i16, v8i8, MulOp, OpNode>;
3521 def v4i32 : N3VLMulOp<op24, op23, 0b01, op11_8, op4, itin16, OpcodeStr,
3522 !strconcat(Dt, "16"), v4i32, v4i16, MulOp, OpNode>;
3523 def v2i64 : N3VLMulOp<op24, op23, 0b10, op11_8, op4, itin32, OpcodeStr,
3524 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
3527 multiclass N3VLMulOpSL_HS<bit op24, bits<4> op11_8, string OpcodeStr,
3528 string Dt, SDNode MulOp, SDNode OpNode> {
3529 def v4i16 : N3VLMulOpSL16<op24, 0b01, op11_8, IIC_VMACi16D, OpcodeStr,
3530 !strconcat(Dt,"16"), v4i32, v4i16, MulOp, OpNode>;
3531 def v2i32 : N3VLMulOpSL<op24, 0b10, op11_8, IIC_VMACi32D, OpcodeStr,
3532 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
3536 // Neon Long 3-argument intrinsics.
3538 // First with only element sizes of 16 and 32 bits:
3539 multiclass N3VLInt3_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
3540 InstrItinClass itin16, InstrItinClass itin32,
3541 string OpcodeStr, string Dt, Intrinsic IntOp> {
3542 def v4i32 : N3VLInt3<op24, op23, 0b01, op11_8, op4, itin16,
3543 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
3544 def v2i64 : N3VLInt3<op24, op23, 0b10, op11_8, op4, itin32,
3545 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3548 multiclass N3VLInt3SL_HS<bit op24, bits<4> op11_8,
3549 string OpcodeStr, string Dt, Intrinsic IntOp> {
3550 def v4i16 : N3VLInt3SL16<op24, 0b01, op11_8, IIC_VMACi16D,
3551 OpcodeStr, !strconcat(Dt,"16"), v4i32, v4i16, IntOp>;
3552 def v2i32 : N3VLInt3SL<op24, 0b10, op11_8, IIC_VMACi32D,
3553 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3556 // ....then also with element size of 8 bits:
3557 multiclass N3VLInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3558 InstrItinClass itin16, InstrItinClass itin32,
3559 string OpcodeStr, string Dt, Intrinsic IntOp>
3560 : N3VLInt3_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt, IntOp> {
3561 def v8i16 : N3VLInt3<op24, op23, 0b00, op11_8, op4, itin16,
3562 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
3565 // ....with explicit extend (VABAL).
3566 multiclass N3VLIntExtOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3567 InstrItinClass itin, string OpcodeStr, string Dt,
3568 Intrinsic IntOp, SDNode ExtOp, SDNode OpNode> {
3569 def v8i16 : N3VLIntExtOp<op24, op23, 0b00, op11_8, op4, itin,
3570 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8,
3571 IntOp, ExtOp, OpNode>;
3572 def v4i32 : N3VLIntExtOp<op24, op23, 0b01, op11_8, op4, itin,
3573 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16,
3574 IntOp, ExtOp, OpNode>;
3575 def v2i64 : N3VLIntExtOp<op24, op23, 0b10, op11_8, op4, itin,
3576 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32,
3577 IntOp, ExtOp, OpNode>;
3581 // Neon Pairwise long 2-register intrinsics,
3582 // element sizes of 8, 16 and 32 bits:
3583 multiclass N2VPLInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3584 bits<5> op11_7, bit op4,
3585 string OpcodeStr, string Dt, Intrinsic IntOp> {
3586 // 64-bit vector types.
3587 def v8i8 : N2VDPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3588 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
3589 def v4i16 : N2VDPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3590 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
3591 def v2i32 : N2VDPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3592 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
3594 // 128-bit vector types.
3595 def v16i8 : N2VQPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3596 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
3597 def v8i16 : N2VQPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3598 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
3599 def v4i32 : N2VQPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3600 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
3604 // Neon Pairwise long 2-register accumulate intrinsics,
3605 // element sizes of 8, 16 and 32 bits:
3606 multiclass N2VPLInt2_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3607 bits<5> op11_7, bit op4,
3608 string OpcodeStr, string Dt, Intrinsic IntOp> {
3609 // 64-bit vector types.
3610 def v8i8 : N2VDPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3611 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
3612 def v4i16 : N2VDPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3613 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
3614 def v2i32 : N2VDPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3615 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
3617 // 128-bit vector types.
3618 def v16i8 : N2VQPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3619 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
3620 def v8i16 : N2VQPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3621 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
3622 def v4i32 : N2VQPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3623 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
3627 // Neon 2-register vector shift by immediate,
3628 // with f of either N2RegVShLFrm or N2RegVShRFrm
3629 // element sizes of 8, 16, 32 and 64 bits:
3630 multiclass N2VShL_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3631 InstrItinClass itin, string OpcodeStr, string Dt,
3633 // 64-bit vector types.
3634 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3635 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
3636 let Inst{21-19} = 0b001; // imm6 = 001xxx
3638 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3639 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
3640 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3642 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3643 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
3644 let Inst{21} = 0b1; // imm6 = 1xxxxx
3646 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, N2RegVShLFrm, itin, i32imm,
3647 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
3650 // 128-bit vector types.
3651 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3652 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
3653 let Inst{21-19} = 0b001; // imm6 = 001xxx
3655 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3656 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
3657 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3659 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3660 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
3661 let Inst{21} = 0b1; // imm6 = 1xxxxx
3663 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, N2RegVShLFrm, itin, i32imm,
3664 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
3667 multiclass N2VShR_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3668 InstrItinClass itin, string OpcodeStr, string Dt,
3670 // 64-bit vector types.
3671 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm8,
3672 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
3673 let Inst{21-19} = 0b001; // imm6 = 001xxx
3675 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm16,
3676 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
3677 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3679 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm32,
3680 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
3681 let Inst{21} = 0b1; // imm6 = 1xxxxx
3683 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, N2RegVShRFrm, itin, shr_imm64,
3684 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
3687 // 128-bit vector types.
3688 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm8,
3689 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
3690 let Inst{21-19} = 0b001; // imm6 = 001xxx
3692 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm16,
3693 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
3694 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3696 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm32,
3697 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
3698 let Inst{21} = 0b1; // imm6 = 1xxxxx
3700 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, N2RegVShRFrm, itin, shr_imm64,
3701 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
3705 // Neon Shift-Accumulate vector operations,
3706 // element sizes of 8, 16, 32 and 64 bits:
3707 multiclass N2VShAdd_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3708 string OpcodeStr, string Dt, SDNode ShOp> {
3709 // 64-bit vector types.
3710 def v8i8 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm8,
3711 OpcodeStr, !strconcat(Dt, "8"), v8i8, ShOp> {
3712 let Inst{21-19} = 0b001; // imm6 = 001xxx
3714 def v4i16 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm16,
3715 OpcodeStr, !strconcat(Dt, "16"), v4i16, ShOp> {
3716 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3718 def v2i32 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm32,
3719 OpcodeStr, !strconcat(Dt, "32"), v2i32, ShOp> {
3720 let Inst{21} = 0b1; // imm6 = 1xxxxx
3722 def v1i64 : N2VDShAdd<op24, op23, op11_8, 1, op4, shr_imm64,
3723 OpcodeStr, !strconcat(Dt, "64"), v1i64, ShOp>;
3726 // 128-bit vector types.
3727 def v16i8 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm8,
3728 OpcodeStr, !strconcat(Dt, "8"), v16i8, ShOp> {
3729 let Inst{21-19} = 0b001; // imm6 = 001xxx
3731 def v8i16 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm16,
3732 OpcodeStr, !strconcat(Dt, "16"), v8i16, ShOp> {
3733 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3735 def v4i32 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm32,
3736 OpcodeStr, !strconcat(Dt, "32"), v4i32, ShOp> {
3737 let Inst{21} = 0b1; // imm6 = 1xxxxx
3739 def v2i64 : N2VQShAdd<op24, op23, op11_8, 1, op4, shr_imm64,
3740 OpcodeStr, !strconcat(Dt, "64"), v2i64, ShOp>;
3744 // Neon Shift-Insert vector operations,
3745 // with f of either N2RegVShLFrm or N2RegVShRFrm
3746 // element sizes of 8, 16, 32 and 64 bits:
3747 multiclass N2VShInsL_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3749 // 64-bit vector types.
3750 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3751 N2RegVShLFrm, OpcodeStr, "8", v8i8, NEONvsli> {
3752 let Inst{21-19} = 0b001; // imm6 = 001xxx
3754 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3755 N2RegVShLFrm, OpcodeStr, "16", v4i16, NEONvsli> {
3756 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3758 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3759 N2RegVShLFrm, OpcodeStr, "32", v2i32, NEONvsli> {
3760 let Inst{21} = 0b1; // imm6 = 1xxxxx
3762 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4, i32imm,
3763 N2RegVShLFrm, OpcodeStr, "64", v1i64, NEONvsli>;
3766 // 128-bit vector types.
3767 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3768 N2RegVShLFrm, OpcodeStr, "8", v16i8, NEONvsli> {
3769 let Inst{21-19} = 0b001; // imm6 = 001xxx
3771 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3772 N2RegVShLFrm, OpcodeStr, "16", v8i16, NEONvsli> {
3773 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3775 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3776 N2RegVShLFrm, OpcodeStr, "32", v4i32, NEONvsli> {
3777 let Inst{21} = 0b1; // imm6 = 1xxxxx
3779 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4, i32imm,
3780 N2RegVShLFrm, OpcodeStr, "64", v2i64, NEONvsli>;
3783 multiclass N2VShInsR_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3785 // 64-bit vector types.
3786 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm8,
3787 N2RegVShRFrm, OpcodeStr, "8", v8i8, NEONvsri> {
3788 let Inst{21-19} = 0b001; // imm6 = 001xxx
3790 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm16,
3791 N2RegVShRFrm, OpcodeStr, "16", v4i16, NEONvsri> {
3792 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3794 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm32,
3795 N2RegVShRFrm, OpcodeStr, "32", v2i32, NEONvsri> {
3796 let Inst{21} = 0b1; // imm6 = 1xxxxx
3798 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4, shr_imm64,
3799 N2RegVShRFrm, OpcodeStr, "64", v1i64, NEONvsri>;
3802 // 128-bit vector types.
3803 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm8,
3804 N2RegVShRFrm, OpcodeStr, "8", v16i8, NEONvsri> {
3805 let Inst{21-19} = 0b001; // imm6 = 001xxx
3807 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm16,
3808 N2RegVShRFrm, OpcodeStr, "16", v8i16, NEONvsri> {
3809 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3811 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm32,
3812 N2RegVShRFrm, OpcodeStr, "32", v4i32, NEONvsri> {
3813 let Inst{21} = 0b1; // imm6 = 1xxxxx
3815 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4, shr_imm64,
3816 N2RegVShRFrm, OpcodeStr, "64", v2i64, NEONvsri>;
3820 // Neon Shift Long operations,
3821 // element sizes of 8, 16, 32 bits:
3822 multiclass N2VLSh_QHS<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
3823 bit op4, string OpcodeStr, string Dt, SDNode OpNode> {
3824 def v8i16 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3825 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, imm1_7, OpNode> {
3826 let Inst{21-19} = 0b001; // imm6 = 001xxx
3828 def v4i32 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3829 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, imm1_15, OpNode> {
3830 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3832 def v2i64 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3833 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, imm1_31, OpNode> {
3834 let Inst{21} = 0b1; // imm6 = 1xxxxx
3838 // Neon Shift Narrow operations,
3839 // element sizes of 16, 32, 64 bits:
3840 multiclass N2VNSh_HSD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
3841 bit op4, InstrItinClass itin, string OpcodeStr, string Dt,
3843 def v8i8 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3844 OpcodeStr, !strconcat(Dt, "16"),
3845 v8i8, v8i16, shr_imm8, OpNode> {
3846 let Inst{21-19} = 0b001; // imm6 = 001xxx
3848 def v4i16 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3849 OpcodeStr, !strconcat(Dt, "32"),
3850 v4i16, v4i32, shr_imm16, OpNode> {
3851 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3853 def v2i32 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3854 OpcodeStr, !strconcat(Dt, "64"),
3855 v2i32, v2i64, shr_imm32, OpNode> {
3856 let Inst{21} = 0b1; // imm6 = 1xxxxx
3860 //===----------------------------------------------------------------------===//
3861 // Instruction Definitions.
3862 //===----------------------------------------------------------------------===//
3864 // Vector Add Operations.
3866 // VADD : Vector Add (integer and floating-point)
3867 defm VADD : N3V_QHSD<0, 0, 0b1000, 0, IIC_VBINiD, IIC_VBINiQ, "vadd", "i",
3869 def VADDfd : N3VD<0, 0, 0b00, 0b1101, 0, IIC_VBIND, "vadd", "f32",
3870 v2f32, v2f32, fadd, 1>;
3871 def VADDfq : N3VQ<0, 0, 0b00, 0b1101, 0, IIC_VBINQ, "vadd", "f32",
3872 v4f32, v4f32, fadd, 1>;
3873 // VADDL : Vector Add Long (Q = D + D)
3874 defm VADDLs : N3VLExt_QHS<0,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
3875 "vaddl", "s", add, sext, 1>;
3876 defm VADDLu : N3VLExt_QHS<1,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
3877 "vaddl", "u", add, zext, 1>;
3878 // VADDW : Vector Add Wide (Q = Q + D)
3879 defm VADDWs : N3VW_QHS<0,1,0b0001,0, "vaddw", "s", add, sext, 0>;
3880 defm VADDWu : N3VW_QHS<1,1,0b0001,0, "vaddw", "u", add, zext, 0>;
3881 // VHADD : Vector Halving Add
3882 defm VHADDs : N3VInt_QHS<0, 0, 0b0000, 0, N3RegFrm,
3883 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3884 "vhadd", "s", int_arm_neon_vhadds, 1>;
3885 defm VHADDu : N3VInt_QHS<1, 0, 0b0000, 0, N3RegFrm,
3886 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3887 "vhadd", "u", int_arm_neon_vhaddu, 1>;
3888 // VRHADD : Vector Rounding Halving Add
3889 defm VRHADDs : N3VInt_QHS<0, 0, 0b0001, 0, N3RegFrm,
3890 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3891 "vrhadd", "s", int_arm_neon_vrhadds, 1>;
3892 defm VRHADDu : N3VInt_QHS<1, 0, 0b0001, 0, N3RegFrm,
3893 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3894 "vrhadd", "u", int_arm_neon_vrhaddu, 1>;
3895 // VQADD : Vector Saturating Add
3896 defm VQADDs : N3VInt_QHSD<0, 0, 0b0000, 1, N3RegFrm,
3897 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3898 "vqadd", "s", int_arm_neon_vqadds, 1>;
3899 defm VQADDu : N3VInt_QHSD<1, 0, 0b0000, 1, N3RegFrm,
3900 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3901 "vqadd", "u", int_arm_neon_vqaddu, 1>;
3902 // VADDHN : Vector Add and Narrow Returning High Half (D = Q + Q)
3903 defm VADDHN : N3VNInt_HSD<0,1,0b0100,0, "vaddhn", "i",
3904 int_arm_neon_vaddhn, 1>;
3905 // VRADDHN : Vector Rounding Add and Narrow Returning High Half (D = Q + Q)
3906 defm VRADDHN : N3VNInt_HSD<1,1,0b0100,0, "vraddhn", "i",
3907 int_arm_neon_vraddhn, 1>;
3909 // Vector Multiply Operations.
3911 // VMUL : Vector Multiply (integer, polynomial and floating-point)
3912 defm VMUL : N3V_QHS<0, 0, 0b1001, 1, IIC_VMULi16D, IIC_VMULi32D,
3913 IIC_VMULi16Q, IIC_VMULi32Q, "vmul", "i", mul, 1>;
3914 def VMULpd : N3VDInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16D, "vmul",
3915 "p8", v8i8, v8i8, int_arm_neon_vmulp, 1>;
3916 def VMULpq : N3VQInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16Q, "vmul",
3917 "p8", v16i8, v16i8, int_arm_neon_vmulp, 1>;
3918 def VMULfd : N3VD<1, 0, 0b00, 0b1101, 1, IIC_VFMULD, "vmul", "f32",
3919 v2f32, v2f32, fmul, 1>;
3920 def VMULfq : N3VQ<1, 0, 0b00, 0b1101, 1, IIC_VFMULQ, "vmul", "f32",
3921 v4f32, v4f32, fmul, 1>;
3922 defm VMULsl : N3VSL_HS<0b1000, "vmul", mul>;
3923 def VMULslfd : N3VDSL<0b10, 0b1001, IIC_VBIND, "vmul", "f32", v2f32, fmul>;
3924 def VMULslfq : N3VQSL<0b10, 0b1001, IIC_VBINQ, "vmul", "f32", v4f32,
3927 def : Pat<(v8i16 (mul (v8i16 QPR:$src1),
3928 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
3929 (v8i16 (VMULslv8i16 (v8i16 QPR:$src1),
3930 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3931 (DSubReg_i16_reg imm:$lane))),
3932 (SubReg_i16_lane imm:$lane)))>;
3933 def : Pat<(v4i32 (mul (v4i32 QPR:$src1),
3934 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
3935 (v4i32 (VMULslv4i32 (v4i32 QPR:$src1),
3936 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3937 (DSubReg_i32_reg imm:$lane))),
3938 (SubReg_i32_lane imm:$lane)))>;
3939 def : Pat<(v4f32 (fmul (v4f32 QPR:$src1),
3940 (v4f32 (NEONvduplane (v4f32 QPR:$src2), imm:$lane)))),
3941 (v4f32 (VMULslfq (v4f32 QPR:$src1),
3942 (v2f32 (EXTRACT_SUBREG QPR:$src2,
3943 (DSubReg_i32_reg imm:$lane))),
3944 (SubReg_i32_lane imm:$lane)))>;
3946 // VQDMULH : Vector Saturating Doubling Multiply Returning High Half
3947 defm VQDMULH : N3VInt_HS<0, 0, 0b1011, 0, N3RegFrm, IIC_VMULi16D, IIC_VMULi32D,
3948 IIC_VMULi16Q, IIC_VMULi32Q,
3949 "vqdmulh", "s", int_arm_neon_vqdmulh, 1>;
3950 defm VQDMULHsl: N3VIntSL_HS<0b1100, IIC_VMULi16D, IIC_VMULi32D,
3951 IIC_VMULi16Q, IIC_VMULi32Q,
3952 "vqdmulh", "s", int_arm_neon_vqdmulh>;
3953 def : Pat<(v8i16 (int_arm_neon_vqdmulh (v8i16 QPR:$src1),
3954 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
3956 (v8i16 (VQDMULHslv8i16 (v8i16 QPR:$src1),
3957 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3958 (DSubReg_i16_reg imm:$lane))),
3959 (SubReg_i16_lane imm:$lane)))>;
3960 def : Pat<(v4i32 (int_arm_neon_vqdmulh (v4i32 QPR:$src1),
3961 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
3963 (v4i32 (VQDMULHslv4i32 (v4i32 QPR:$src1),
3964 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3965 (DSubReg_i32_reg imm:$lane))),
3966 (SubReg_i32_lane imm:$lane)))>;
3968 // VQRDMULH : Vector Rounding Saturating Doubling Multiply Returning High Half
3969 defm VQRDMULH : N3VInt_HS<1, 0, 0b1011, 0, N3RegFrm,
3970 IIC_VMULi16D,IIC_VMULi32D,IIC_VMULi16Q,IIC_VMULi32Q,
3971 "vqrdmulh", "s", int_arm_neon_vqrdmulh, 1>;
3972 defm VQRDMULHsl : N3VIntSL_HS<0b1101, IIC_VMULi16D, IIC_VMULi32D,
3973 IIC_VMULi16Q, IIC_VMULi32Q,
3974 "vqrdmulh", "s", int_arm_neon_vqrdmulh>;
3975 def : Pat<(v8i16 (int_arm_neon_vqrdmulh (v8i16 QPR:$src1),
3976 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
3978 (v8i16 (VQRDMULHslv8i16 (v8i16 QPR:$src1),
3979 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3980 (DSubReg_i16_reg imm:$lane))),
3981 (SubReg_i16_lane imm:$lane)))>;
3982 def : Pat<(v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$src1),
3983 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
3985 (v4i32 (VQRDMULHslv4i32 (v4i32 QPR:$src1),
3986 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3987 (DSubReg_i32_reg imm:$lane))),
3988 (SubReg_i32_lane imm:$lane)))>;
3990 // VMULL : Vector Multiply Long (integer and polynomial) (Q = D * D)
3991 defm VMULLs : N3VL_QHS<0,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
3992 "vmull", "s", NEONvmulls, 1>;
3993 defm VMULLu : N3VL_QHS<1,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
3994 "vmull", "u", NEONvmullu, 1>;
3995 def VMULLp : N3VLInt<0, 1, 0b00, 0b1110, 0, IIC_VMULi16D, "vmull", "p8",
3996 v8i16, v8i8, int_arm_neon_vmullp, 1>;
3997 defm VMULLsls : N3VLSL_HS<0, 0b1010, IIC_VMULi16D, "vmull", "s", NEONvmulls>;
3998 defm VMULLslu : N3VLSL_HS<1, 0b1010, IIC_VMULi16D, "vmull", "u", NEONvmullu>;
4000 // VQDMULL : Vector Saturating Doubling Multiply Long (Q = D * D)
4001 defm VQDMULL : N3VLInt_HS<0,1,0b1101,0, IIC_VMULi16D, IIC_VMULi32D,
4002 "vqdmull", "s", int_arm_neon_vqdmull, 1>;
4003 defm VQDMULLsl: N3VLIntSL_HS<0, 0b1011, IIC_VMULi16D,
4004 "vqdmull", "s", int_arm_neon_vqdmull>;
4006 // Vector Multiply-Accumulate and Multiply-Subtract Operations.
4008 // VMLA : Vector Multiply Accumulate (integer and floating-point)
4009 defm VMLA : N3VMulOp_QHS<0, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
4010 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
4011 def VMLAfd : N3VDMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACD, "vmla", "f32",
4012 v2f32, fmul_su, fadd_mlx>,
4013 Requires<[HasNEON, UseFPVMLx, NoNEON2]>;
4014 def VMLAfq : N3VQMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACQ, "vmla", "f32",
4015 v4f32, fmul_su, fadd_mlx>,
4016 Requires<[HasNEON, UseFPVMLx, NoNEON2]>;
4017 defm VMLAsl : N3VMulOpSL_HS<0b0000, IIC_VMACi16D, IIC_VMACi32D,
4018 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
4019 def VMLAslfd : N3VDMulOpSL<0b10, 0b0001, IIC_VMACD, "vmla", "f32",
4020 v2f32, fmul_su, fadd_mlx>,
4021 Requires<[HasNEON, UseFPVMLx]>;
4022 def VMLAslfq : N3VQMulOpSL<0b10, 0b0001, IIC_VMACQ, "vmla", "f32",
4023 v4f32, v2f32, fmul_su, fadd_mlx>,
4024 Requires<[HasNEON, UseFPVMLx]>;
4026 def : Pat<(v8i16 (add (v8i16 QPR:$src1),
4027 (mul (v8i16 QPR:$src2),
4028 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
4029 (v8i16 (VMLAslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
4030 (v4i16 (EXTRACT_SUBREG QPR:$src3,
4031 (DSubReg_i16_reg imm:$lane))),
4032 (SubReg_i16_lane imm:$lane)))>;
4034 def : Pat<(v4i32 (add (v4i32 QPR:$src1),
4035 (mul (v4i32 QPR:$src2),
4036 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
4037 (v4i32 (VMLAslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
4038 (v2i32 (EXTRACT_SUBREG QPR:$src3,
4039 (DSubReg_i32_reg imm:$lane))),
4040 (SubReg_i32_lane imm:$lane)))>;
4042 def : Pat<(v4f32 (fadd_mlx (v4f32 QPR:$src1),
4043 (fmul_su (v4f32 QPR:$src2),
4044 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
4045 (v4f32 (VMLAslfq (v4f32 QPR:$src1),
4047 (v2f32 (EXTRACT_SUBREG QPR:$src3,
4048 (DSubReg_i32_reg imm:$lane))),
4049 (SubReg_i32_lane imm:$lane)))>,
4050 Requires<[HasNEON, UseFPVMLx]>;
4052 // VMLAL : Vector Multiply Accumulate Long (Q += D * D)
4053 defm VMLALs : N3VLMulOp_QHS<0,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
4054 "vmlal", "s", NEONvmulls, add>;
4055 defm VMLALu : N3VLMulOp_QHS<1,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
4056 "vmlal", "u", NEONvmullu, add>;
4058 defm VMLALsls : N3VLMulOpSL_HS<0, 0b0010, "vmlal", "s", NEONvmulls, add>;
4059 defm VMLALslu : N3VLMulOpSL_HS<1, 0b0010, "vmlal", "u", NEONvmullu, add>;
4061 // VQDMLAL : Vector Saturating Doubling Multiply Accumulate Long (Q += D * D)
4062 defm VQDMLAL : N3VLInt3_HS<0, 1, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
4063 "vqdmlal", "s", int_arm_neon_vqdmlal>;
4064 defm VQDMLALsl: N3VLInt3SL_HS<0, 0b0011, "vqdmlal", "s", int_arm_neon_vqdmlal>;
4066 // VMLS : Vector Multiply Subtract (integer and floating-point)
4067 defm VMLS : N3VMulOp_QHS<1, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
4068 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
4069 def VMLSfd : N3VDMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACD, "vmls", "f32",
4070 v2f32, fmul_su, fsub_mlx>,
4071 Requires<[HasNEON, UseFPVMLx, NoNEON2]>;
4072 def VMLSfq : N3VQMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACQ, "vmls", "f32",
4073 v4f32, fmul_su, fsub_mlx>,
4074 Requires<[HasNEON, UseFPVMLx, NoNEON2]>;
4075 defm VMLSsl : N3VMulOpSL_HS<0b0100, IIC_VMACi16D, IIC_VMACi32D,
4076 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
4077 def VMLSslfd : N3VDMulOpSL<0b10, 0b0101, IIC_VMACD, "vmls", "f32",
4078 v2f32, fmul_su, fsub_mlx>,
4079 Requires<[HasNEON, UseFPVMLx]>;
4080 def VMLSslfq : N3VQMulOpSL<0b10, 0b0101, IIC_VMACQ, "vmls", "f32",
4081 v4f32, v2f32, fmul_su, fsub_mlx>,
4082 Requires<[HasNEON, UseFPVMLx]>;
4084 def : Pat<(v8i16 (sub (v8i16 QPR:$src1),
4085 (mul (v8i16 QPR:$src2),
4086 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
4087 (v8i16 (VMLSslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
4088 (v4i16 (EXTRACT_SUBREG QPR:$src3,
4089 (DSubReg_i16_reg imm:$lane))),
4090 (SubReg_i16_lane imm:$lane)))>;
4092 def : Pat<(v4i32 (sub (v4i32 QPR:$src1),
4093 (mul (v4i32 QPR:$src2),
4094 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
4095 (v4i32 (VMLSslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
4096 (v2i32 (EXTRACT_SUBREG QPR:$src3,
4097 (DSubReg_i32_reg imm:$lane))),
4098 (SubReg_i32_lane imm:$lane)))>;
4100 def : Pat<(v4f32 (fsub_mlx (v4f32 QPR:$src1),
4101 (fmul_su (v4f32 QPR:$src2),
4102 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
4103 (v4f32 (VMLSslfq (v4f32 QPR:$src1), (v4f32 QPR:$src2),
4104 (v2f32 (EXTRACT_SUBREG QPR:$src3,
4105 (DSubReg_i32_reg imm:$lane))),
4106 (SubReg_i32_lane imm:$lane)))>,
4107 Requires<[HasNEON, UseFPVMLx]>;
4109 // VMLSL : Vector Multiply Subtract Long (Q -= D * D)
4110 defm VMLSLs : N3VLMulOp_QHS<0,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
4111 "vmlsl", "s", NEONvmulls, sub>;
4112 defm VMLSLu : N3VLMulOp_QHS<1,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
4113 "vmlsl", "u", NEONvmullu, sub>;
4115 defm VMLSLsls : N3VLMulOpSL_HS<0, 0b0110, "vmlsl", "s", NEONvmulls, sub>;
4116 defm VMLSLslu : N3VLMulOpSL_HS<1, 0b0110, "vmlsl", "u", NEONvmullu, sub>;
4118 // VQDMLSL : Vector Saturating Doubling Multiply Subtract Long (Q -= D * D)
4119 defm VQDMLSL : N3VLInt3_HS<0, 1, 0b1011, 0, IIC_VMACi16D, IIC_VMACi32D,
4120 "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
4121 defm VQDMLSLsl: N3VLInt3SL_HS<0, 0b111, "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
4124 // Fused Vector Multiply-Accumulate and Fused Multiply-Subtract Operations.
4125 def VFMAfd : N3VDMulOp<0, 0, 0b00, 0b1100, 1, IIC_VFMACD, "vfma", "f32",
4126 v2f32, fmul_su, fadd_mlx>,
4127 Requires<[HasNEON2,FPContractions]>;
4129 def VFMAfq : N3VQMulOp<0, 0, 0b00, 0b1100, 1, IIC_VFMACQ, "vfma", "f32",
4130 v4f32, fmul_su, fadd_mlx>,
4131 Requires<[HasNEON2,FPContractions]>;
4133 // Fused Vector Multiply Subtract (floating-point)
4134 def VFMSfd : N3VDMulOp<0, 0, 0b10, 0b1100, 1, IIC_VFMACD, "vfms", "f32",
4135 v2f32, fmul_su, fsub_mlx>,
4136 Requires<[HasNEON2,FPContractions]>;
4137 def VFMSfq : N3VQMulOp<0, 0, 0b10, 0b1100, 1, IIC_VFMACQ, "vfms", "f32",
4138 v4f32, fmul_su, fsub_mlx>,
4139 Requires<[HasNEON2,FPContractions]>;
4141 // Vector Subtract Operations.
4143 // VSUB : Vector Subtract (integer and floating-point)
4144 defm VSUB : N3V_QHSD<1, 0, 0b1000, 0, IIC_VSUBiD, IIC_VSUBiQ,
4145 "vsub", "i", sub, 0>;
4146 def VSUBfd : N3VD<0, 0, 0b10, 0b1101, 0, IIC_VBIND, "vsub", "f32",
4147 v2f32, v2f32, fsub, 0>;
4148 def VSUBfq : N3VQ<0, 0, 0b10, 0b1101, 0, IIC_VBINQ, "vsub", "f32",
4149 v4f32, v4f32, fsub, 0>;
4150 // VSUBL : Vector Subtract Long (Q = D - D)
4151 defm VSUBLs : N3VLExt_QHS<0,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
4152 "vsubl", "s", sub, sext, 0>;
4153 defm VSUBLu : N3VLExt_QHS<1,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
4154 "vsubl", "u", sub, zext, 0>;
4155 // VSUBW : Vector Subtract Wide (Q = Q - D)
4156 defm VSUBWs : N3VW_QHS<0,1,0b0011,0, "vsubw", "s", sub, sext, 0>;
4157 defm VSUBWu : N3VW_QHS<1,1,0b0011,0, "vsubw", "u", sub, zext, 0>;
4158 // VHSUB : Vector Halving Subtract
4159 defm VHSUBs : N3VInt_QHS<0, 0, 0b0010, 0, N3RegFrm,
4160 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4161 "vhsub", "s", int_arm_neon_vhsubs, 0>;
4162 defm VHSUBu : N3VInt_QHS<1, 0, 0b0010, 0, N3RegFrm,
4163 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4164 "vhsub", "u", int_arm_neon_vhsubu, 0>;
4165 // VQSUB : Vector Saturing Subtract
4166 defm VQSUBs : N3VInt_QHSD<0, 0, 0b0010, 1, N3RegFrm,
4167 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4168 "vqsub", "s", int_arm_neon_vqsubs, 0>;
4169 defm VQSUBu : N3VInt_QHSD<1, 0, 0b0010, 1, N3RegFrm,
4170 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4171 "vqsub", "u", int_arm_neon_vqsubu, 0>;
4172 // VSUBHN : Vector Subtract and Narrow Returning High Half (D = Q - Q)
4173 defm VSUBHN : N3VNInt_HSD<0,1,0b0110,0, "vsubhn", "i",
4174 int_arm_neon_vsubhn, 0>;
4175 // VRSUBHN : Vector Rounding Subtract and Narrow Returning High Half (D=Q-Q)
4176 defm VRSUBHN : N3VNInt_HSD<1,1,0b0110,0, "vrsubhn", "i",
4177 int_arm_neon_vrsubhn, 0>;
4179 // Vector Comparisons.
4181 // VCEQ : Vector Compare Equal
4182 defm VCEQ : N3V_QHS<1, 0, 0b1000, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
4183 IIC_VSUBi4Q, "vceq", "i", NEONvceq, 1>;
4184 def VCEQfd : N3VD<0,0,0b00,0b1110,0, IIC_VBIND, "vceq", "f32", v2i32, v2f32,
4186 def VCEQfq : N3VQ<0,0,0b00,0b1110,0, IIC_VBINQ, "vceq", "f32", v4i32, v4f32,
4189 defm VCEQz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00010, 0, "vceq", "i",
4190 "$Vd, $Vm, #0", NEONvceqz>;
4192 // VCGE : Vector Compare Greater Than or Equal
4193 defm VCGEs : N3V_QHS<0, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
4194 IIC_VSUBi4Q, "vcge", "s", NEONvcge, 0>;
4195 defm VCGEu : N3V_QHS<1, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
4196 IIC_VSUBi4Q, "vcge", "u", NEONvcgeu, 0>;
4197 def VCGEfd : N3VD<1,0,0b00,0b1110,0, IIC_VBIND, "vcge", "f32", v2i32, v2f32,
4199 def VCGEfq : N3VQ<1,0,0b00,0b1110,0, IIC_VBINQ, "vcge", "f32", v4i32, v4f32,
4202 defm VCGEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00001, 0, "vcge", "s",
4203 "$Vd, $Vm, #0", NEONvcgez>;
4204 defm VCLEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00011, 0, "vcle", "s",
4205 "$Vd, $Vm, #0", NEONvclez>;
4207 // VCGT : Vector Compare Greater Than
4208 defm VCGTs : N3V_QHS<0, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
4209 IIC_VSUBi4Q, "vcgt", "s", NEONvcgt, 0>;
4210 defm VCGTu : N3V_QHS<1, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
4211 IIC_VSUBi4Q, "vcgt", "u", NEONvcgtu, 0>;
4212 def VCGTfd : N3VD<1,0,0b10,0b1110,0, IIC_VBIND, "vcgt", "f32", v2i32, v2f32,
4214 def VCGTfq : N3VQ<1,0,0b10,0b1110,0, IIC_VBINQ, "vcgt", "f32", v4i32, v4f32,
4217 defm VCGTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00000, 0, "vcgt", "s",
4218 "$Vd, $Vm, #0", NEONvcgtz>;
4219 defm VCLTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00100, 0, "vclt", "s",
4220 "$Vd, $Vm, #0", NEONvcltz>;
4222 // VACGE : Vector Absolute Compare Greater Than or Equal (aka VCAGE)
4223 def VACGEd : N3VDInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacge",
4224 "f32", v2i32, v2f32, int_arm_neon_vacged, 0>;
4225 def VACGEq : N3VQInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacge",
4226 "f32", v4i32, v4f32, int_arm_neon_vacgeq, 0>;
4227 // VACGT : Vector Absolute Compare Greater Than (aka VCAGT)
4228 def VACGTd : N3VDInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacgt",
4229 "f32", v2i32, v2f32, int_arm_neon_vacgtd, 0>;
4230 def VACGTq : N3VQInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacgt",
4231 "f32", v4i32, v4f32, int_arm_neon_vacgtq, 0>;
4232 // VTST : Vector Test Bits
4233 defm VTST : N3V_QHS<0, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
4234 IIC_VBINi4Q, "vtst", "", NEONvtst, 1>;
4236 // Vector Bitwise Operations.
4238 def vnotd : PatFrag<(ops node:$in),
4239 (xor node:$in, (bitconvert (v8i8 NEONimmAllOnesV)))>;
4240 def vnotq : PatFrag<(ops node:$in),
4241 (xor node:$in, (bitconvert (v16i8 NEONimmAllOnesV)))>;
4244 // VAND : Vector Bitwise AND
4245 def VANDd : N3VDX<0, 0, 0b00, 0b0001, 1, IIC_VBINiD, "vand",
4246 v2i32, v2i32, and, 1>;
4247 def VANDq : N3VQX<0, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "vand",
4248 v4i32, v4i32, and, 1>;
4250 // VEOR : Vector Bitwise Exclusive OR
4251 def VEORd : N3VDX<1, 0, 0b00, 0b0001, 1, IIC_VBINiD, "veor",
4252 v2i32, v2i32, xor, 1>;
4253 def VEORq : N3VQX<1, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "veor",
4254 v4i32, v4i32, xor, 1>;
4256 // VORR : Vector Bitwise OR
4257 def VORRd : N3VDX<0, 0, 0b10, 0b0001, 1, IIC_VBINiD, "vorr",
4258 v2i32, v2i32, or, 1>;
4259 def VORRq : N3VQX<0, 0, 0b10, 0b0001, 1, IIC_VBINiQ, "vorr",
4260 v4i32, v4i32, or, 1>;
4262 def VORRiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 0, 1,
4263 (outs DPR:$Vd), (ins nImmSplatI16:$SIMM, DPR:$src),
4265 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
4267 (v4i16 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
4268 let Inst{9} = SIMM{9};
4271 def VORRiv2i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 0, 0, 1,
4272 (outs DPR:$Vd), (ins nImmSplatI32:$SIMM, DPR:$src),
4274 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
4276 (v2i32 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
4277 let Inst{10-9} = SIMM{10-9};
4280 def VORRiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 0, 1,
4281 (outs QPR:$Vd), (ins nImmSplatI16:$SIMM, QPR:$src),
4283 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
4285 (v8i16 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
4286 let Inst{9} = SIMM{9};
4289 def VORRiv4i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 1, 0, 1,
4290 (outs QPR:$Vd), (ins nImmSplatI32:$SIMM, QPR:$src),
4292 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
4294 (v4i32 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
4295 let Inst{10-9} = SIMM{10-9};
4299 // VBIC : Vector Bitwise Bit Clear (AND NOT)
4300 def VBICd : N3VX<0, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
4301 (ins DPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VBINiD,
4302 "vbic", "$Vd, $Vn, $Vm", "",
4303 [(set DPR:$Vd, (v2i32 (and DPR:$Vn,
4304 (vnotd DPR:$Vm))))]>;
4305 def VBICq : N3VX<0, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
4306 (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINiQ,
4307 "vbic", "$Vd, $Vn, $Vm", "",
4308 [(set QPR:$Vd, (v4i32 (and QPR:$Vn,
4309 (vnotq QPR:$Vm))))]>;
4311 def VBICiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 1, 1,
4312 (outs DPR:$Vd), (ins nImmSplatI16:$SIMM, DPR:$src),
4314 "vbic", "i16", "$Vd, $SIMM", "$src = $Vd",
4316 (v4i16 (NEONvbicImm DPR:$src, timm:$SIMM)))]> {
4317 let Inst{9} = SIMM{9};
4320 def VBICiv2i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 0, 1, 1,
4321 (outs DPR:$Vd), (ins nImmSplatI32:$SIMM, DPR:$src),
4323 "vbic", "i32", "$Vd, $SIMM", "$src = $Vd",
4325 (v2i32 (NEONvbicImm DPR:$src, timm:$SIMM)))]> {
4326 let Inst{10-9} = SIMM{10-9};
4329 def VBICiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 1, 1,
4330 (outs QPR:$Vd), (ins nImmSplatI16:$SIMM, QPR:$src),
4332 "vbic", "i16", "$Vd, $SIMM", "$src = $Vd",
4334 (v8i16 (NEONvbicImm QPR:$src, timm:$SIMM)))]> {
4335 let Inst{9} = SIMM{9};
4338 def VBICiv4i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 1, 1, 1,
4339 (outs QPR:$Vd), (ins nImmSplatI32:$SIMM, QPR:$src),
4341 "vbic", "i32", "$Vd, $SIMM", "$src = $Vd",
4343 (v4i32 (NEONvbicImm QPR:$src, timm:$SIMM)))]> {
4344 let Inst{10-9} = SIMM{10-9};
4347 // VORN : Vector Bitwise OR NOT
4348 def VORNd : N3VX<0, 0, 0b11, 0b0001, 0, 1, (outs DPR:$Vd),
4349 (ins DPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VBINiD,
4350 "vorn", "$Vd, $Vn, $Vm", "",
4351 [(set DPR:$Vd, (v2i32 (or DPR:$Vn,
4352 (vnotd DPR:$Vm))))]>;
4353 def VORNq : N3VX<0, 0, 0b11, 0b0001, 1, 1, (outs QPR:$Vd),
4354 (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINiQ,
4355 "vorn", "$Vd, $Vn, $Vm", "",
4356 [(set QPR:$Vd, (v4i32 (or QPR:$Vn,
4357 (vnotq QPR:$Vm))))]>;
4359 // VMVN : Vector Bitwise NOT (Immediate)
4361 let isReMaterializable = 1 in {
4363 def VMVNv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 1, 1, (outs DPR:$Vd),
4364 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4365 "vmvn", "i16", "$Vd, $SIMM", "",
4366 [(set DPR:$Vd, (v4i16 (NEONvmvnImm timm:$SIMM)))]> {
4367 let Inst{9} = SIMM{9};
4370 def VMVNv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 1, 1, (outs QPR:$Vd),
4371 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4372 "vmvn", "i16", "$Vd, $SIMM", "",
4373 [(set QPR:$Vd, (v8i16 (NEONvmvnImm timm:$SIMM)))]> {
4374 let Inst{9} = SIMM{9};
4377 def VMVNv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 1, 1, (outs DPR:$Vd),
4378 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4379 "vmvn", "i32", "$Vd, $SIMM", "",
4380 [(set DPR:$Vd, (v2i32 (NEONvmvnImm timm:$SIMM)))]> {
4381 let Inst{11-8} = SIMM{11-8};
4384 def VMVNv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 1, 1, (outs QPR:$Vd),
4385 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4386 "vmvn", "i32", "$Vd, $SIMM", "",
4387 [(set QPR:$Vd, (v4i32 (NEONvmvnImm timm:$SIMM)))]> {
4388 let Inst{11-8} = SIMM{11-8};
4392 // VMVN : Vector Bitwise NOT
4393 def VMVNd : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 0, 0,
4394 (outs DPR:$Vd), (ins DPR:$Vm), IIC_VSUBiD,
4395 "vmvn", "$Vd, $Vm", "",
4396 [(set DPR:$Vd, (v2i32 (vnotd DPR:$Vm)))]>;
4397 def VMVNq : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 1, 0,
4398 (outs QPR:$Vd), (ins QPR:$Vm), IIC_VSUBiD,
4399 "vmvn", "$Vd, $Vm", "",
4400 [(set QPR:$Vd, (v4i32 (vnotq QPR:$Vm)))]>;
4401 def : Pat<(v2i32 (vnotd DPR:$src)), (VMVNd DPR:$src)>;
4402 def : Pat<(v4i32 (vnotq QPR:$src)), (VMVNq QPR:$src)>;
4404 // VBSL : Vector Bitwise Select
4405 def VBSLd : N3VX<1, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
4406 (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
4407 N3RegFrm, IIC_VCNTiD,
4408 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4410 (v2i32 (NEONvbsl DPR:$src1, DPR:$Vn, DPR:$Vm)))]>;
4412 def : Pat<(v2i32 (or (and DPR:$Vn, DPR:$Vd),
4413 (and DPR:$Vm, (vnotd DPR:$Vd)))),
4414 (VBSLd DPR:$Vd, DPR:$Vn, DPR:$Vm)>;
4416 def VBSLq : N3VX<1, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
4417 (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
4418 N3RegFrm, IIC_VCNTiQ,
4419 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4421 (v4i32 (NEONvbsl QPR:$src1, QPR:$Vn, QPR:$Vm)))]>;
4423 def : Pat<(v4i32 (or (and QPR:$Vn, QPR:$Vd),
4424 (and QPR:$Vm, (vnotq QPR:$Vd)))),
4425 (VBSLq QPR:$Vd, QPR:$Vn, QPR:$Vm)>;
4427 // VBIF : Vector Bitwise Insert if False
4428 // like VBSL but with: "vbif $dst, $src3, $src1", "$src2 = $dst",
4429 // FIXME: This instruction's encoding MAY NOT BE correct.
4430 def VBIFd : N3VX<1, 0, 0b11, 0b0001, 0, 1,
4431 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
4432 N3RegFrm, IIC_VBINiD,
4433 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4435 def VBIFq : N3VX<1, 0, 0b11, 0b0001, 1, 1,
4436 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
4437 N3RegFrm, IIC_VBINiQ,
4438 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4441 // VBIT : Vector Bitwise Insert if True
4442 // like VBSL but with: "vbit $dst, $src2, $src1", "$src3 = $dst",
4443 // FIXME: This instruction's encoding MAY NOT BE correct.
4444 def VBITd : N3VX<1, 0, 0b10, 0b0001, 0, 1,
4445 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
4446 N3RegFrm, IIC_VBINiD,
4447 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4449 def VBITq : N3VX<1, 0, 0b10, 0b0001, 1, 1,
4450 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
4451 N3RegFrm, IIC_VBINiQ,
4452 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4455 // VBIT/VBIF are not yet implemented. The TwoAddress pass will not go looking
4456 // for equivalent operations with different register constraints; it just
4459 // Vector Absolute Differences.
4461 // VABD : Vector Absolute Difference
4462 defm VABDs : N3VInt_QHS<0, 0, 0b0111, 0, N3RegFrm,
4463 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4464 "vabd", "s", int_arm_neon_vabds, 1>;
4465 defm VABDu : N3VInt_QHS<1, 0, 0b0111, 0, N3RegFrm,
4466 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4467 "vabd", "u", int_arm_neon_vabdu, 1>;
4468 def VABDfd : N3VDInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBIND,
4469 "vabd", "f32", v2f32, v2f32, int_arm_neon_vabds, 1>;
4470 def VABDfq : N3VQInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBINQ,
4471 "vabd", "f32", v4f32, v4f32, int_arm_neon_vabds, 1>;
4473 // VABDL : Vector Absolute Difference Long (Q = | D - D |)
4474 defm VABDLs : N3VLIntExt_QHS<0,1,0b0111,0, IIC_VSUBi4Q,
4475 "vabdl", "s", int_arm_neon_vabds, zext, 1>;
4476 defm VABDLu : N3VLIntExt_QHS<1,1,0b0111,0, IIC_VSUBi4Q,
4477 "vabdl", "u", int_arm_neon_vabdu, zext, 1>;
4479 // VABA : Vector Absolute Difference and Accumulate
4480 defm VABAs : N3VIntOp_QHS<0,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
4481 "vaba", "s", int_arm_neon_vabds, add>;
4482 defm VABAu : N3VIntOp_QHS<1,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
4483 "vaba", "u", int_arm_neon_vabdu, add>;
4485 // VABAL : Vector Absolute Difference and Accumulate Long (Q += | D - D |)
4486 defm VABALs : N3VLIntExtOp_QHS<0,1,0b0101,0, IIC_VABAD,
4487 "vabal", "s", int_arm_neon_vabds, zext, add>;
4488 defm VABALu : N3VLIntExtOp_QHS<1,1,0b0101,0, IIC_VABAD,
4489 "vabal", "u", int_arm_neon_vabdu, zext, add>;
4491 // Vector Maximum and Minimum.
4493 // VMAX : Vector Maximum
4494 defm VMAXs : N3VInt_QHS<0, 0, 0b0110, 0, N3RegFrm,
4495 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4496 "vmax", "s", int_arm_neon_vmaxs, 1>;
4497 defm VMAXu : N3VInt_QHS<1, 0, 0b0110, 0, N3RegFrm,
4498 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4499 "vmax", "u", int_arm_neon_vmaxu, 1>;
4500 def VMAXfd : N3VDInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBIND,
4502 v2f32, v2f32, int_arm_neon_vmaxs, 1>;
4503 def VMAXfq : N3VQInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBINQ,
4505 v4f32, v4f32, int_arm_neon_vmaxs, 1>;
4507 // VMIN : Vector Minimum
4508 defm VMINs : N3VInt_QHS<0, 0, 0b0110, 1, N3RegFrm,
4509 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4510 "vmin", "s", int_arm_neon_vmins, 1>;
4511 defm VMINu : N3VInt_QHS<1, 0, 0b0110, 1, N3RegFrm,
4512 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4513 "vmin", "u", int_arm_neon_vminu, 1>;
4514 def VMINfd : N3VDInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBIND,
4516 v2f32, v2f32, int_arm_neon_vmins, 1>;
4517 def VMINfq : N3VQInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBINQ,
4519 v4f32, v4f32, int_arm_neon_vmins, 1>;
4521 // Vector Pairwise Operations.
4523 // VPADD : Vector Pairwise Add
4524 def VPADDi8 : N3VDInt<0, 0, 0b00, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4526 v8i8, v8i8, int_arm_neon_vpadd, 0>;
4527 def VPADDi16 : N3VDInt<0, 0, 0b01, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4529 v4i16, v4i16, int_arm_neon_vpadd, 0>;
4530 def VPADDi32 : N3VDInt<0, 0, 0b10, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4532 v2i32, v2i32, int_arm_neon_vpadd, 0>;
4533 def VPADDf : N3VDInt<1, 0, 0b00, 0b1101, 0, N3RegFrm,
4534 IIC_VPBIND, "vpadd", "f32",
4535 v2f32, v2f32, int_arm_neon_vpadd, 0>;
4537 // VPADDL : Vector Pairwise Add Long
4538 defm VPADDLs : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00100, 0, "vpaddl", "s",
4539 int_arm_neon_vpaddls>;
4540 defm VPADDLu : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00101, 0, "vpaddl", "u",
4541 int_arm_neon_vpaddlu>;
4543 // VPADAL : Vector Pairwise Add and Accumulate Long
4544 defm VPADALs : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01100, 0, "vpadal", "s",
4545 int_arm_neon_vpadals>;
4546 defm VPADALu : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01101, 0, "vpadal", "u",
4547 int_arm_neon_vpadalu>;
4549 // VPMAX : Vector Pairwise Maximum
4550 def VPMAXs8 : N3VDInt<0, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4551 "s8", v8i8, v8i8, int_arm_neon_vpmaxs, 0>;
4552 def VPMAXs16 : N3VDInt<0, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4553 "s16", v4i16, v4i16, int_arm_neon_vpmaxs, 0>;
4554 def VPMAXs32 : N3VDInt<0, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4555 "s32", v2i32, v2i32, int_arm_neon_vpmaxs, 0>;
4556 def VPMAXu8 : N3VDInt<1, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4557 "u8", v8i8, v8i8, int_arm_neon_vpmaxu, 0>;
4558 def VPMAXu16 : N3VDInt<1, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4559 "u16", v4i16, v4i16, int_arm_neon_vpmaxu, 0>;
4560 def VPMAXu32 : N3VDInt<1, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4561 "u32", v2i32, v2i32, int_arm_neon_vpmaxu, 0>;
4562 def VPMAXf : N3VDInt<1, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmax",
4563 "f32", v2f32, v2f32, int_arm_neon_vpmaxs, 0>;
4565 // VPMIN : Vector Pairwise Minimum
4566 def VPMINs8 : N3VDInt<0, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4567 "s8", v8i8, v8i8, int_arm_neon_vpmins, 0>;
4568 def VPMINs16 : N3VDInt<0, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4569 "s16", v4i16, v4i16, int_arm_neon_vpmins, 0>;
4570 def VPMINs32 : N3VDInt<0, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4571 "s32", v2i32, v2i32, int_arm_neon_vpmins, 0>;
4572 def VPMINu8 : N3VDInt<1, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4573 "u8", v8i8, v8i8, int_arm_neon_vpminu, 0>;
4574 def VPMINu16 : N3VDInt<1, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4575 "u16", v4i16, v4i16, int_arm_neon_vpminu, 0>;
4576 def VPMINu32 : N3VDInt<1, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4577 "u32", v2i32, v2i32, int_arm_neon_vpminu, 0>;
4578 def VPMINf : N3VDInt<1, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmin",
4579 "f32", v2f32, v2f32, int_arm_neon_vpmins, 0>;
4581 // Vector Reciprocal and Reciprocal Square Root Estimate and Step.
4583 // VRECPE : Vector Reciprocal Estimate
4584 def VRECPEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
4585 IIC_VUNAD, "vrecpe", "u32",
4586 v2i32, v2i32, int_arm_neon_vrecpe>;
4587 def VRECPEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
4588 IIC_VUNAQ, "vrecpe", "u32",
4589 v4i32, v4i32, int_arm_neon_vrecpe>;
4590 def VRECPEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
4591 IIC_VUNAD, "vrecpe", "f32",
4592 v2f32, v2f32, int_arm_neon_vrecpe>;
4593 def VRECPEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
4594 IIC_VUNAQ, "vrecpe", "f32",
4595 v4f32, v4f32, int_arm_neon_vrecpe>;
4597 // VRECPS : Vector Reciprocal Step
4598 def VRECPSfd : N3VDInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
4599 IIC_VRECSD, "vrecps", "f32",
4600 v2f32, v2f32, int_arm_neon_vrecps, 1>;
4601 def VRECPSfq : N3VQInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
4602 IIC_VRECSQ, "vrecps", "f32",
4603 v4f32, v4f32, int_arm_neon_vrecps, 1>;
4605 // VRSQRTE : Vector Reciprocal Square Root Estimate
4606 def VRSQRTEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
4607 IIC_VUNAD, "vrsqrte", "u32",
4608 v2i32, v2i32, int_arm_neon_vrsqrte>;
4609 def VRSQRTEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
4610 IIC_VUNAQ, "vrsqrte", "u32",
4611 v4i32, v4i32, int_arm_neon_vrsqrte>;
4612 def VRSQRTEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
4613 IIC_VUNAD, "vrsqrte", "f32",
4614 v2f32, v2f32, int_arm_neon_vrsqrte>;
4615 def VRSQRTEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
4616 IIC_VUNAQ, "vrsqrte", "f32",
4617 v4f32, v4f32, int_arm_neon_vrsqrte>;
4619 // VRSQRTS : Vector Reciprocal Square Root Step
4620 def VRSQRTSfd : N3VDInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
4621 IIC_VRECSD, "vrsqrts", "f32",
4622 v2f32, v2f32, int_arm_neon_vrsqrts, 1>;
4623 def VRSQRTSfq : N3VQInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
4624 IIC_VRECSQ, "vrsqrts", "f32",
4625 v4f32, v4f32, int_arm_neon_vrsqrts, 1>;
4629 // VSHL : Vector Shift
4630 defm VSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 0, N3RegVShFrm,
4631 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
4632 "vshl", "s", int_arm_neon_vshifts>;
4633 defm VSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 0, N3RegVShFrm,
4634 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
4635 "vshl", "u", int_arm_neon_vshiftu>;
4637 // VSHL : Vector Shift Left (Immediate)
4638 defm VSHLi : N2VShL_QHSD<0, 1, 0b0101, 1, IIC_VSHLiD, "vshl", "i", NEONvshl>;
4640 // VSHR : Vector Shift Right (Immediate)
4641 defm VSHRs : N2VShR_QHSD<0, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "s",NEONvshrs>;
4642 defm VSHRu : N2VShR_QHSD<1, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "u",NEONvshru>;
4644 // VSHLL : Vector Shift Left Long
4645 defm VSHLLs : N2VLSh_QHS<0, 1, 0b1010, 0, 0, 1, "vshll", "s", NEONvshlls>;
4646 defm VSHLLu : N2VLSh_QHS<1, 1, 0b1010, 0, 0, 1, "vshll", "u", NEONvshllu>;
4648 // VSHLL : Vector Shift Left Long (with maximum shift count)
4649 class N2VLShMax<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
4650 bit op6, bit op4, string OpcodeStr, string Dt, ValueType ResTy,
4651 ValueType OpTy, Operand ImmTy, SDNode OpNode>
4652 : N2VLSh<op24, op23, op11_8, op7, op6, op4, OpcodeStr, Dt,
4653 ResTy, OpTy, ImmTy, OpNode> {
4654 let Inst{21-16} = op21_16;
4655 let DecoderMethod = "DecodeVSHLMaxInstruction";
4657 def VSHLLi8 : N2VLShMax<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll", "i8",
4658 v8i16, v8i8, imm8, NEONvshlli>;
4659 def VSHLLi16 : N2VLShMax<1, 1, 0b110110, 0b0011, 0, 0, 0, "vshll", "i16",
4660 v4i32, v4i16, imm16, NEONvshlli>;
4661 def VSHLLi32 : N2VLShMax<1, 1, 0b111010, 0b0011, 0, 0, 0, "vshll", "i32",
4662 v2i64, v2i32, imm32, NEONvshlli>;
4664 // VSHRN : Vector Shift Right and Narrow
4665 defm VSHRN : N2VNSh_HSD<0,1,0b1000,0,0,1, IIC_VSHLiD, "vshrn", "i",
4668 // VRSHL : Vector Rounding Shift
4669 defm VRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 0, N3RegVShFrm,
4670 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4671 "vrshl", "s", int_arm_neon_vrshifts>;
4672 defm VRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 0, N3RegVShFrm,
4673 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4674 "vrshl", "u", int_arm_neon_vrshiftu>;
4675 // VRSHR : Vector Rounding Shift Right
4676 defm VRSHRs : N2VShR_QHSD<0,1,0b0010,1, IIC_VSHLi4D, "vrshr", "s",NEONvrshrs>;
4677 defm VRSHRu : N2VShR_QHSD<1,1,0b0010,1, IIC_VSHLi4D, "vrshr", "u",NEONvrshru>;
4679 // VRSHRN : Vector Rounding Shift Right and Narrow
4680 defm VRSHRN : N2VNSh_HSD<0, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vrshrn", "i",
4683 // VQSHL : Vector Saturating Shift
4684 defm VQSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 1, N3RegVShFrm,
4685 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4686 "vqshl", "s", int_arm_neon_vqshifts>;
4687 defm VQSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 1, N3RegVShFrm,
4688 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4689 "vqshl", "u", int_arm_neon_vqshiftu>;
4690 // VQSHL : Vector Saturating Shift Left (Immediate)
4691 defm VQSHLsi : N2VShL_QHSD<0,1,0b0111,1, IIC_VSHLi4D, "vqshl", "s",NEONvqshls>;
4692 defm VQSHLui : N2VShL_QHSD<1,1,0b0111,1, IIC_VSHLi4D, "vqshl", "u",NEONvqshlu>;
4694 // VQSHLU : Vector Saturating Shift Left (Immediate, Unsigned)
4695 defm VQSHLsu : N2VShL_QHSD<1,1,0b0110,1, IIC_VSHLi4D,"vqshlu","s",NEONvqshlsu>;
4697 // VQSHRN : Vector Saturating Shift Right and Narrow
4698 defm VQSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "s",
4700 defm VQSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "u",
4703 // VQSHRUN : Vector Saturating Shift Right and Narrow (Unsigned)
4704 defm VQSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 0, 1, IIC_VSHLi4D, "vqshrun", "s",
4707 // VQRSHL : Vector Saturating Rounding Shift
4708 defm VQRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 1, N3RegVShFrm,
4709 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4710 "vqrshl", "s", int_arm_neon_vqrshifts>;
4711 defm VQRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 1, N3RegVShFrm,
4712 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4713 "vqrshl", "u", int_arm_neon_vqrshiftu>;
4715 // VQRSHRN : Vector Saturating Rounding Shift Right and Narrow
4716 defm VQRSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "s",
4718 defm VQRSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "u",
4721 // VQRSHRUN : Vector Saturating Rounding Shift Right and Narrow (Unsigned)
4722 defm VQRSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vqrshrun", "s",
4725 // VSRA : Vector Shift Right and Accumulate
4726 defm VSRAs : N2VShAdd_QHSD<0, 1, 0b0001, 1, "vsra", "s", NEONvshrs>;
4727 defm VSRAu : N2VShAdd_QHSD<1, 1, 0b0001, 1, "vsra", "u", NEONvshru>;
4728 // VRSRA : Vector Rounding Shift Right and Accumulate
4729 defm VRSRAs : N2VShAdd_QHSD<0, 1, 0b0011, 1, "vrsra", "s", NEONvrshrs>;
4730 defm VRSRAu : N2VShAdd_QHSD<1, 1, 0b0011, 1, "vrsra", "u", NEONvrshru>;
4732 // VSLI : Vector Shift Left and Insert
4733 defm VSLI : N2VShInsL_QHSD<1, 1, 0b0101, 1, "vsli">;
4735 // VSRI : Vector Shift Right and Insert
4736 defm VSRI : N2VShInsR_QHSD<1, 1, 0b0100, 1, "vsri">;
4738 // Vector Absolute and Saturating Absolute.
4740 // VABS : Vector Absolute Value
4741 defm VABS : N2VInt_QHS<0b11, 0b11, 0b01, 0b00110, 0,
4742 IIC_VUNAiD, IIC_VUNAiQ, "vabs", "s",
4744 def VABSfd : N2VDInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
4745 IIC_VUNAD, "vabs", "f32",
4746 v2f32, v2f32, int_arm_neon_vabs>;
4747 def VABSfq : N2VQInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
4748 IIC_VUNAQ, "vabs", "f32",
4749 v4f32, v4f32, int_arm_neon_vabs>;
4751 // VQABS : Vector Saturating Absolute Value
4752 defm VQABS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01110, 0,
4753 IIC_VQUNAiD, IIC_VQUNAiQ, "vqabs", "s",
4754 int_arm_neon_vqabs>;
4758 def vnegd : PatFrag<(ops node:$in),
4759 (sub (bitconvert (v2i32 NEONimmAllZerosV)), node:$in)>;
4760 def vnegq : PatFrag<(ops node:$in),
4761 (sub (bitconvert (v4i32 NEONimmAllZerosV)), node:$in)>;
4763 class VNEGD<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
4764 : N2V<0b11, 0b11, size, 0b01, 0b00111, 0, 0, (outs DPR:$Vd), (ins DPR:$Vm),
4765 IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
4766 [(set DPR:$Vd, (Ty (vnegd DPR:$Vm)))]>;
4767 class VNEGQ<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
4768 : N2V<0b11, 0b11, size, 0b01, 0b00111, 1, 0, (outs QPR:$Vd), (ins QPR:$Vm),
4769 IIC_VSHLiQ, OpcodeStr, Dt, "$Vd, $Vm", "",
4770 [(set QPR:$Vd, (Ty (vnegq QPR:$Vm)))]>;
4772 // VNEG : Vector Negate (integer)
4773 def VNEGs8d : VNEGD<0b00, "vneg", "s8", v8i8>;
4774 def VNEGs16d : VNEGD<0b01, "vneg", "s16", v4i16>;
4775 def VNEGs32d : VNEGD<0b10, "vneg", "s32", v2i32>;
4776 def VNEGs8q : VNEGQ<0b00, "vneg", "s8", v16i8>;
4777 def VNEGs16q : VNEGQ<0b01, "vneg", "s16", v8i16>;
4778 def VNEGs32q : VNEGQ<0b10, "vneg", "s32", v4i32>;
4780 // VNEG : Vector Negate (floating-point)
4781 def VNEGfd : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
4782 (outs DPR:$Vd), (ins DPR:$Vm), IIC_VUNAD,
4783 "vneg", "f32", "$Vd, $Vm", "",
4784 [(set DPR:$Vd, (v2f32 (fneg DPR:$Vm)))]>;
4785 def VNEGf32q : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 1, 0,
4786 (outs QPR:$Vd), (ins QPR:$Vm), IIC_VUNAQ,
4787 "vneg", "f32", "$Vd, $Vm", "",
4788 [(set QPR:$Vd, (v4f32 (fneg QPR:$Vm)))]>;
4790 def : Pat<(v8i8 (vnegd DPR:$src)), (VNEGs8d DPR:$src)>;
4791 def : Pat<(v4i16 (vnegd DPR:$src)), (VNEGs16d DPR:$src)>;
4792 def : Pat<(v2i32 (vnegd DPR:$src)), (VNEGs32d DPR:$src)>;
4793 def : Pat<(v16i8 (vnegq QPR:$src)), (VNEGs8q QPR:$src)>;
4794 def : Pat<(v8i16 (vnegq QPR:$src)), (VNEGs16q QPR:$src)>;
4795 def : Pat<(v4i32 (vnegq QPR:$src)), (VNEGs32q QPR:$src)>;
4797 // VQNEG : Vector Saturating Negate
4798 defm VQNEG : N2VInt_QHS<0b11, 0b11, 0b00, 0b01111, 0,
4799 IIC_VQUNAiD, IIC_VQUNAiQ, "vqneg", "s",
4800 int_arm_neon_vqneg>;
4802 // Vector Bit Counting Operations.
4804 // VCLS : Vector Count Leading Sign Bits
4805 defm VCLS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01000, 0,
4806 IIC_VCNTiD, IIC_VCNTiQ, "vcls", "s",
4808 // VCLZ : Vector Count Leading Zeros
4809 defm VCLZ : N2VInt_QHS<0b11, 0b11, 0b00, 0b01001, 0,
4810 IIC_VCNTiD, IIC_VCNTiQ, "vclz", "i",
4812 // VCNT : Vector Count One Bits
4813 def VCNTd : N2VDInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
4814 IIC_VCNTiD, "vcnt", "8",
4815 v8i8, v8i8, int_arm_neon_vcnt>;
4816 def VCNTq : N2VQInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
4817 IIC_VCNTiQ, "vcnt", "8",
4818 v16i8, v16i8, int_arm_neon_vcnt>;
4821 def VSWPd : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 0, 0,
4822 (outs DPR:$Vd, DPR:$Vd1), (ins DPR:$Vm, DPR:$Vm1),
4823 NoItinerary, "vswp", "$Vd, $Vd1", "$Vm = $Vd, $Vm1 = $Vd1",
4825 def VSWPq : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 1, 0,
4826 (outs QPR:$Vd, QPR:$Vd1), (ins QPR:$Vm, QPR:$Vm1),
4827 NoItinerary, "vswp", "$Vd, $Vd1", "$Vm = $Vd, $Vm1 = $Vd1",
4830 // Vector Move Operations.
4832 // VMOV : Vector Move (Register)
4833 def : InstAlias<"vmov${p} $Vd, $Vm",
4834 (VORRd DPR:$Vd, DPR:$Vm, DPR:$Vm, pred:$p)>;
4835 def : InstAlias<"vmov${p} $Vd, $Vm",
4836 (VORRq QPR:$Vd, QPR:$Vm, QPR:$Vm, pred:$p)>;
4838 // VMOV : Vector Move (Immediate)
4840 let isReMaterializable = 1 in {
4841 def VMOVv8i8 : N1ModImm<1, 0b000, 0b1110, 0, 0, 0, 1, (outs DPR:$Vd),
4842 (ins nImmSplatI8:$SIMM), IIC_VMOVImm,
4843 "vmov", "i8", "$Vd, $SIMM", "",
4844 [(set DPR:$Vd, (v8i8 (NEONvmovImm timm:$SIMM)))]>;
4845 def VMOVv16i8 : N1ModImm<1, 0b000, 0b1110, 0, 1, 0, 1, (outs QPR:$Vd),
4846 (ins nImmSplatI8:$SIMM), IIC_VMOVImm,
4847 "vmov", "i8", "$Vd, $SIMM", "",
4848 [(set QPR:$Vd, (v16i8 (NEONvmovImm timm:$SIMM)))]>;
4850 def VMOVv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 0, 1, (outs DPR:$Vd),
4851 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4852 "vmov", "i16", "$Vd, $SIMM", "",
4853 [(set DPR:$Vd, (v4i16 (NEONvmovImm timm:$SIMM)))]> {
4854 let Inst{9} = SIMM{9};
4857 def VMOVv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 0, 1, (outs QPR:$Vd),
4858 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4859 "vmov", "i16", "$Vd, $SIMM", "",
4860 [(set QPR:$Vd, (v8i16 (NEONvmovImm timm:$SIMM)))]> {
4861 let Inst{9} = SIMM{9};
4864 def VMOVv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 0, 1, (outs DPR:$Vd),
4865 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4866 "vmov", "i32", "$Vd, $SIMM", "",
4867 [(set DPR:$Vd, (v2i32 (NEONvmovImm timm:$SIMM)))]> {
4868 let Inst{11-8} = SIMM{11-8};
4871 def VMOVv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 0, 1, (outs QPR:$Vd),
4872 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4873 "vmov", "i32", "$Vd, $SIMM", "",
4874 [(set QPR:$Vd, (v4i32 (NEONvmovImm timm:$SIMM)))]> {
4875 let Inst{11-8} = SIMM{11-8};
4878 def VMOVv1i64 : N1ModImm<1, 0b000, 0b1110, 0, 0, 1, 1, (outs DPR:$Vd),
4879 (ins nImmSplatI64:$SIMM), IIC_VMOVImm,
4880 "vmov", "i64", "$Vd, $SIMM", "",
4881 [(set DPR:$Vd, (v1i64 (NEONvmovImm timm:$SIMM)))]>;
4882 def VMOVv2i64 : N1ModImm<1, 0b000, 0b1110, 0, 1, 1, 1, (outs QPR:$Vd),
4883 (ins nImmSplatI64:$SIMM), IIC_VMOVImm,
4884 "vmov", "i64", "$Vd, $SIMM", "",
4885 [(set QPR:$Vd, (v2i64 (NEONvmovImm timm:$SIMM)))]>;
4887 def VMOVv2f32 : N1ModImm<1, 0b000, 0b1111, 0, 0, 0, 1, (outs DPR:$Vd),
4888 (ins nImmVMOVF32:$SIMM), IIC_VMOVImm,
4889 "vmov", "f32", "$Vd, $SIMM", "",
4890 [(set DPR:$Vd, (v2f32 (NEONvmovFPImm timm:$SIMM)))]>;
4891 def VMOVv4f32 : N1ModImm<1, 0b000, 0b1111, 0, 1, 0, 1, (outs QPR:$Vd),
4892 (ins nImmVMOVF32:$SIMM), IIC_VMOVImm,
4893 "vmov", "f32", "$Vd, $SIMM", "",
4894 [(set QPR:$Vd, (v4f32 (NEONvmovFPImm timm:$SIMM)))]>;
4895 } // isReMaterializable
4897 // VMOV : Vector Get Lane (move scalar to ARM core register)
4899 def VGETLNs8 : NVGetLane<{1,1,1,0,0,1,?,1}, 0b1011, {?,?},
4900 (outs GPR:$R), (ins DPR:$V, VectorIndex8:$lane),
4901 IIC_VMOVSI, "vmov", "s8", "$R, $V$lane",
4902 [(set GPR:$R, (NEONvgetlanes (v8i8 DPR:$V),
4904 let Inst{21} = lane{2};
4905 let Inst{6-5} = lane{1-0};
4907 def VGETLNs16 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, {?,1},
4908 (outs GPR:$R), (ins DPR:$V, VectorIndex16:$lane),
4909 IIC_VMOVSI, "vmov", "s16", "$R, $V$lane",
4910 [(set GPR:$R, (NEONvgetlanes (v4i16 DPR:$V),
4912 let Inst{21} = lane{1};
4913 let Inst{6} = lane{0};
4915 def VGETLNu8 : NVGetLane<{1,1,1,0,1,1,?,1}, 0b1011, {?,?},
4916 (outs GPR:$R), (ins DPR:$V, VectorIndex8:$lane),
4917 IIC_VMOVSI, "vmov", "u8", "$R, $V$lane",
4918 [(set GPR:$R, (NEONvgetlaneu (v8i8 DPR:$V),
4920 let Inst{21} = lane{2};
4921 let Inst{6-5} = lane{1-0};
4923 def VGETLNu16 : NVGetLane<{1,1,1,0,1,0,?,1}, 0b1011, {?,1},
4924 (outs GPR:$R), (ins DPR:$V, VectorIndex16:$lane),
4925 IIC_VMOVSI, "vmov", "u16", "$R, $V$lane",
4926 [(set GPR:$R, (NEONvgetlaneu (v4i16 DPR:$V),
4928 let Inst{21} = lane{1};
4929 let Inst{6} = lane{0};
4931 def VGETLNi32 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, 0b00,
4932 (outs GPR:$R), (ins DPR:$V, VectorIndex32:$lane),
4933 IIC_VMOVSI, "vmov", "32", "$R, $V$lane",
4934 [(set GPR:$R, (extractelt (v2i32 DPR:$V),
4936 let Inst{21} = lane{0};
4938 // def VGETLNf32: see FMRDH and FMRDL in ARMInstrVFP.td
4939 def : Pat<(NEONvgetlanes (v16i8 QPR:$src), imm:$lane),
4940 (VGETLNs8 (v8i8 (EXTRACT_SUBREG QPR:$src,
4941 (DSubReg_i8_reg imm:$lane))),
4942 (SubReg_i8_lane imm:$lane))>;
4943 def : Pat<(NEONvgetlanes (v8i16 QPR:$src), imm:$lane),
4944 (VGETLNs16 (v4i16 (EXTRACT_SUBREG QPR:$src,
4945 (DSubReg_i16_reg imm:$lane))),
4946 (SubReg_i16_lane imm:$lane))>;
4947 def : Pat<(NEONvgetlaneu (v16i8 QPR:$src), imm:$lane),
4948 (VGETLNu8 (v8i8 (EXTRACT_SUBREG QPR:$src,
4949 (DSubReg_i8_reg imm:$lane))),
4950 (SubReg_i8_lane imm:$lane))>;
4951 def : Pat<(NEONvgetlaneu (v8i16 QPR:$src), imm:$lane),
4952 (VGETLNu16 (v4i16 (EXTRACT_SUBREG QPR:$src,
4953 (DSubReg_i16_reg imm:$lane))),
4954 (SubReg_i16_lane imm:$lane))>;
4955 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
4956 (VGETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src,
4957 (DSubReg_i32_reg imm:$lane))),
4958 (SubReg_i32_lane imm:$lane))>;
4959 def : Pat<(extractelt (v2f32 DPR:$src1), imm:$src2),
4960 (EXTRACT_SUBREG (v2f32 (COPY_TO_REGCLASS (v2f32 DPR:$src1),DPR_VFP2)),
4961 (SSubReg_f32_reg imm:$src2))>;
4962 def : Pat<(extractelt (v4f32 QPR:$src1), imm:$src2),
4963 (EXTRACT_SUBREG (v4f32 (COPY_TO_REGCLASS (v4f32 QPR:$src1),QPR_VFP2)),
4964 (SSubReg_f32_reg imm:$src2))>;
4965 //def : Pat<(extractelt (v2i64 QPR:$src1), imm:$src2),
4966 // (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
4967 def : Pat<(extractelt (v2f64 QPR:$src1), imm:$src2),
4968 (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
4971 // VMOV : Vector Set Lane (move ARM core register to scalar)
4973 let Constraints = "$src1 = $V" in {
4974 def VSETLNi8 : NVSetLane<{1,1,1,0,0,1,?,0}, 0b1011, {?,?}, (outs DPR:$V),
4975 (ins DPR:$src1, GPR:$R, VectorIndex8:$lane),
4976 IIC_VMOVISL, "vmov", "8", "$V$lane, $R",
4977 [(set DPR:$V, (vector_insert (v8i8 DPR:$src1),
4978 GPR:$R, imm:$lane))]> {
4979 let Inst{21} = lane{2};
4980 let Inst{6-5} = lane{1-0};
4982 def VSETLNi16 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, {?,1}, (outs DPR:$V),
4983 (ins DPR:$src1, GPR:$R, VectorIndex16:$lane),
4984 IIC_VMOVISL, "vmov", "16", "$V$lane, $R",
4985 [(set DPR:$V, (vector_insert (v4i16 DPR:$src1),
4986 GPR:$R, imm:$lane))]> {
4987 let Inst{21} = lane{1};
4988 let Inst{6} = lane{0};
4990 def VSETLNi32 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, 0b00, (outs DPR:$V),
4991 (ins DPR:$src1, GPR:$R, VectorIndex32:$lane),
4992 IIC_VMOVISL, "vmov", "32", "$V$lane, $R",
4993 [(set DPR:$V, (insertelt (v2i32 DPR:$src1),
4994 GPR:$R, imm:$lane))]> {
4995 let Inst{21} = lane{0};
4998 def : Pat<(vector_insert (v16i8 QPR:$src1), GPR:$src2, imm:$lane),
4999 (v16i8 (INSERT_SUBREG QPR:$src1,
5000 (v8i8 (VSETLNi8 (v8i8 (EXTRACT_SUBREG QPR:$src1,
5001 (DSubReg_i8_reg imm:$lane))),
5002 GPR:$src2, (SubReg_i8_lane imm:$lane))),
5003 (DSubReg_i8_reg imm:$lane)))>;
5004 def : Pat<(vector_insert (v8i16 QPR:$src1), GPR:$src2, imm:$lane),
5005 (v8i16 (INSERT_SUBREG QPR:$src1,
5006 (v4i16 (VSETLNi16 (v4i16 (EXTRACT_SUBREG QPR:$src1,
5007 (DSubReg_i16_reg imm:$lane))),
5008 GPR:$src2, (SubReg_i16_lane imm:$lane))),
5009 (DSubReg_i16_reg imm:$lane)))>;
5010 def : Pat<(insertelt (v4i32 QPR:$src1), GPR:$src2, imm:$lane),
5011 (v4i32 (INSERT_SUBREG QPR:$src1,
5012 (v2i32 (VSETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src1,
5013 (DSubReg_i32_reg imm:$lane))),
5014 GPR:$src2, (SubReg_i32_lane imm:$lane))),
5015 (DSubReg_i32_reg imm:$lane)))>;
5017 def : Pat<(v2f32 (insertelt DPR:$src1, SPR:$src2, imm:$src3)),
5018 (INSERT_SUBREG (v2f32 (COPY_TO_REGCLASS DPR:$src1, DPR_VFP2)),
5019 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
5020 def : Pat<(v4f32 (insertelt QPR:$src1, SPR:$src2, imm:$src3)),
5021 (INSERT_SUBREG (v4f32 (COPY_TO_REGCLASS QPR:$src1, QPR_VFP2)),
5022 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
5024 //def : Pat<(v2i64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
5025 // (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
5026 def : Pat<(v2f64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
5027 (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
5029 def : Pat<(v2f32 (scalar_to_vector SPR:$src)),
5030 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
5031 def : Pat<(v2f64 (scalar_to_vector (f64 DPR:$src))),
5032 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
5033 def : Pat<(v4f32 (scalar_to_vector SPR:$src)),
5034 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
5036 def : Pat<(v8i8 (scalar_to_vector GPR:$src)),
5037 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
5038 def : Pat<(v4i16 (scalar_to_vector GPR:$src)),
5039 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
5040 def : Pat<(v2i32 (scalar_to_vector GPR:$src)),
5041 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
5043 def : Pat<(v16i8 (scalar_to_vector GPR:$src)),
5044 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
5045 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
5047 def : Pat<(v8i16 (scalar_to_vector GPR:$src)),
5048 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
5049 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
5051 def : Pat<(v4i32 (scalar_to_vector GPR:$src)),
5052 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
5053 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
5056 // VDUP : Vector Duplicate (from ARM core register to all elements)
5058 class VDUPD<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
5059 : NVDup<opcod1, 0b1011, opcod3, (outs DPR:$V), (ins GPR:$R),
5060 IIC_VMOVIS, "vdup", Dt, "$V, $R",
5061 [(set DPR:$V, (Ty (NEONvdup (i32 GPR:$R))))]>;
5062 class VDUPQ<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
5063 : NVDup<opcod1, 0b1011, opcod3, (outs QPR:$V), (ins GPR:$R),
5064 IIC_VMOVIS, "vdup", Dt, "$V, $R",
5065 [(set QPR:$V, (Ty (NEONvdup (i32 GPR:$R))))]>;
5067 def VDUP8d : VDUPD<0b11101100, 0b00, "8", v8i8>;
5068 def VDUP16d : VDUPD<0b11101000, 0b01, "16", v4i16>;
5069 def VDUP32d : VDUPD<0b11101000, 0b00, "32", v2i32>;
5070 def VDUP8q : VDUPQ<0b11101110, 0b00, "8", v16i8>;
5071 def VDUP16q : VDUPQ<0b11101010, 0b01, "16", v8i16>;
5072 def VDUP32q : VDUPQ<0b11101010, 0b00, "32", v4i32>;
5074 def : Pat<(v2f32 (NEONvdup (f32 (bitconvert GPR:$R)))), (VDUP32d GPR:$R)>;
5075 def : Pat<(v4f32 (NEONvdup (f32 (bitconvert GPR:$R)))), (VDUP32q GPR:$R)>;
5077 // VDUP : Vector Duplicate Lane (from scalar to all elements)
5079 class VDUPLND<bits<4> op19_16, string OpcodeStr, string Dt,
5080 ValueType Ty, Operand IdxTy>
5081 : NVDupLane<op19_16, 0, (outs DPR:$Vd), (ins DPR:$Vm, IdxTy:$lane),
5082 IIC_VMOVD, OpcodeStr, Dt, "$Vd, $Vm$lane",
5083 [(set DPR:$Vd, (Ty (NEONvduplane (Ty DPR:$Vm), imm:$lane)))]>;
5085 class VDUPLNQ<bits<4> op19_16, string OpcodeStr, string Dt,
5086 ValueType ResTy, ValueType OpTy, Operand IdxTy>
5087 : NVDupLane<op19_16, 1, (outs QPR:$Vd), (ins DPR:$Vm, IdxTy:$lane),
5088 IIC_VMOVQ, OpcodeStr, Dt, "$Vd, $Vm$lane",
5089 [(set QPR:$Vd, (ResTy (NEONvduplane (OpTy DPR:$Vm),
5090 VectorIndex32:$lane)))]>;
5092 // Inst{19-16} is partially specified depending on the element size.
5094 def VDUPLN8d : VDUPLND<{?,?,?,1}, "vdup", "8", v8i8, VectorIndex8> {
5096 let Inst{19-17} = lane{2-0};
5098 def VDUPLN16d : VDUPLND<{?,?,1,0}, "vdup", "16", v4i16, VectorIndex16> {
5100 let Inst{19-18} = lane{1-0};
5102 def VDUPLN32d : VDUPLND<{?,1,0,0}, "vdup", "32", v2i32, VectorIndex32> {
5104 let Inst{19} = lane{0};
5106 def VDUPLN8q : VDUPLNQ<{?,?,?,1}, "vdup", "8", v16i8, v8i8, VectorIndex8> {
5108 let Inst{19-17} = lane{2-0};
5110 def VDUPLN16q : VDUPLNQ<{?,?,1,0}, "vdup", "16", v8i16, v4i16, VectorIndex16> {
5112 let Inst{19-18} = lane{1-0};
5114 def VDUPLN32q : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4i32, v2i32, VectorIndex32> {
5116 let Inst{19} = lane{0};
5119 def : Pat<(v2f32 (NEONvduplane (v2f32 DPR:$Vm), imm:$lane)),
5120 (VDUPLN32d DPR:$Vm, imm:$lane)>;
5122 def : Pat<(v4f32 (NEONvduplane (v2f32 DPR:$Vm), imm:$lane)),
5123 (VDUPLN32q DPR:$Vm, imm:$lane)>;
5125 def : Pat<(v16i8 (NEONvduplane (v16i8 QPR:$src), imm:$lane)),
5126 (v16i8 (VDUPLN8q (v8i8 (EXTRACT_SUBREG QPR:$src,
5127 (DSubReg_i8_reg imm:$lane))),
5128 (SubReg_i8_lane imm:$lane)))>;
5129 def : Pat<(v8i16 (NEONvduplane (v8i16 QPR:$src), imm:$lane)),
5130 (v8i16 (VDUPLN16q (v4i16 (EXTRACT_SUBREG QPR:$src,
5131 (DSubReg_i16_reg imm:$lane))),
5132 (SubReg_i16_lane imm:$lane)))>;
5133 def : Pat<(v4i32 (NEONvduplane (v4i32 QPR:$src), imm:$lane)),
5134 (v4i32 (VDUPLN32q (v2i32 (EXTRACT_SUBREG QPR:$src,
5135 (DSubReg_i32_reg imm:$lane))),
5136 (SubReg_i32_lane imm:$lane)))>;
5137 def : Pat<(v4f32 (NEONvduplane (v4f32 QPR:$src), imm:$lane)),
5138 (v4f32 (VDUPLN32q (v2f32 (EXTRACT_SUBREG QPR:$src,
5139 (DSubReg_i32_reg imm:$lane))),
5140 (SubReg_i32_lane imm:$lane)))>;
5142 def VDUPfdf : PseudoNeonI<(outs DPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
5143 [(set DPR:$dst, (v2f32 (NEONvdup (f32 SPR:$src))))]>;
5144 def VDUPfqf : PseudoNeonI<(outs QPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
5145 [(set QPR:$dst, (v4f32 (NEONvdup (f32 SPR:$src))))]>;
5147 // VMOVN : Vector Narrowing Move
5148 defm VMOVN : N2VN_HSD<0b11,0b11,0b10,0b00100,0,0, IIC_VMOVN,
5149 "vmovn", "i", trunc>;
5150 // VQMOVN : Vector Saturating Narrowing Move
5151 defm VQMOVNs : N2VNInt_HSD<0b11,0b11,0b10,0b00101,0,0, IIC_VQUNAiD,
5152 "vqmovn", "s", int_arm_neon_vqmovns>;
5153 defm VQMOVNu : N2VNInt_HSD<0b11,0b11,0b10,0b00101,1,0, IIC_VQUNAiD,
5154 "vqmovn", "u", int_arm_neon_vqmovnu>;
5155 defm VQMOVNsu : N2VNInt_HSD<0b11,0b11,0b10,0b00100,1,0, IIC_VQUNAiD,
5156 "vqmovun", "s", int_arm_neon_vqmovnsu>;
5157 // VMOVL : Vector Lengthening Move
5158 defm VMOVLs : N2VL_QHS<0b01,0b10100,0,1, "vmovl", "s", sext>;
5159 defm VMOVLu : N2VL_QHS<0b11,0b10100,0,1, "vmovl", "u", zext>;
5160 def : Pat<(v8i16 (anyext (v8i8 DPR:$Vm))), (VMOVLuv8i16 DPR:$Vm)>;
5161 def : Pat<(v4i32 (anyext (v4i16 DPR:$Vm))), (VMOVLuv4i32 DPR:$Vm)>;
5162 def : Pat<(v2i64 (anyext (v2i32 DPR:$Vm))), (VMOVLuv2i64 DPR:$Vm)>;
5164 // Vector Conversions.
5166 // VCVT : Vector Convert Between Floating-Point and Integers
5167 def VCVTf2sd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
5168 v2i32, v2f32, fp_to_sint>;
5169 def VCVTf2ud : N2VD<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
5170 v2i32, v2f32, fp_to_uint>;
5171 def VCVTs2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
5172 v2f32, v2i32, sint_to_fp>;
5173 def VCVTu2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
5174 v2f32, v2i32, uint_to_fp>;
5176 def VCVTf2sq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
5177 v4i32, v4f32, fp_to_sint>;
5178 def VCVTf2uq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
5179 v4i32, v4f32, fp_to_uint>;
5180 def VCVTs2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
5181 v4f32, v4i32, sint_to_fp>;
5182 def VCVTu2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
5183 v4f32, v4i32, uint_to_fp>;
5185 // VCVT : Vector Convert Between Floating-Point and Fixed-Point.
5186 let DecoderMethod = "DecodeVCVTD" in {
5187 def VCVTf2xsd : N2VCvtD<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
5188 v2i32, v2f32, int_arm_neon_vcvtfp2fxs>;
5189 def VCVTf2xud : N2VCvtD<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
5190 v2i32, v2f32, int_arm_neon_vcvtfp2fxu>;
5191 def VCVTxs2fd : N2VCvtD<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
5192 v2f32, v2i32, int_arm_neon_vcvtfxs2fp>;
5193 def VCVTxu2fd : N2VCvtD<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
5194 v2f32, v2i32, int_arm_neon_vcvtfxu2fp>;
5197 let DecoderMethod = "DecodeVCVTQ" in {
5198 def VCVTf2xsq : N2VCvtQ<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
5199 v4i32, v4f32, int_arm_neon_vcvtfp2fxs>;
5200 def VCVTf2xuq : N2VCvtQ<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
5201 v4i32, v4f32, int_arm_neon_vcvtfp2fxu>;
5202 def VCVTxs2fq : N2VCvtQ<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
5203 v4f32, v4i32, int_arm_neon_vcvtfxs2fp>;
5204 def VCVTxu2fq : N2VCvtQ<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
5205 v4f32, v4i32, int_arm_neon_vcvtfxu2fp>;
5208 // VCVT : Vector Convert Between Half-Precision and Single-Precision.
5209 def VCVTf2h : N2VNInt<0b11, 0b11, 0b01, 0b10, 0b01100, 0, 0,
5210 IIC_VUNAQ, "vcvt", "f16.f32",
5211 v4i16, v4f32, int_arm_neon_vcvtfp2hf>,
5212 Requires<[HasNEON, HasFP16]>;
5213 def VCVTh2f : N2VLInt<0b11, 0b11, 0b01, 0b10, 0b01110, 0, 0,
5214 IIC_VUNAQ, "vcvt", "f32.f16",
5215 v4f32, v4i16, int_arm_neon_vcvthf2fp>,
5216 Requires<[HasNEON, HasFP16]>;
5220 // VREV64 : Vector Reverse elements within 64-bit doublewords
5222 class VREV64D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
5223 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 0, 0, (outs DPR:$Vd),
5224 (ins DPR:$Vm), IIC_VMOVD,
5225 OpcodeStr, Dt, "$Vd, $Vm", "",
5226 [(set DPR:$Vd, (Ty (NEONvrev64 (Ty DPR:$Vm))))]>;
5227 class VREV64Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
5228 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 1, 0, (outs QPR:$Vd),
5229 (ins QPR:$Vm), IIC_VMOVQ,
5230 OpcodeStr, Dt, "$Vd, $Vm", "",
5231 [(set QPR:$Vd, (Ty (NEONvrev64 (Ty QPR:$Vm))))]>;
5233 def VREV64d8 : VREV64D<0b00, "vrev64", "8", v8i8>;
5234 def VREV64d16 : VREV64D<0b01, "vrev64", "16", v4i16>;
5235 def VREV64d32 : VREV64D<0b10, "vrev64", "32", v2i32>;
5236 def : Pat<(v2f32 (NEONvrev64 (v2f32 DPR:$Vm))), (VREV64d32 DPR:$Vm)>;
5238 def VREV64q8 : VREV64Q<0b00, "vrev64", "8", v16i8>;
5239 def VREV64q16 : VREV64Q<0b01, "vrev64", "16", v8i16>;
5240 def VREV64q32 : VREV64Q<0b10, "vrev64", "32", v4i32>;
5241 def : Pat<(v4f32 (NEONvrev64 (v4f32 QPR:$Vm))), (VREV64q32 QPR:$Vm)>;
5243 // VREV32 : Vector Reverse elements within 32-bit words
5245 class VREV32D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
5246 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 0, 0, (outs DPR:$Vd),
5247 (ins DPR:$Vm), IIC_VMOVD,
5248 OpcodeStr, Dt, "$Vd, $Vm", "",
5249 [(set DPR:$Vd, (Ty (NEONvrev32 (Ty DPR:$Vm))))]>;
5250 class VREV32Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
5251 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 1, 0, (outs QPR:$Vd),
5252 (ins QPR:$Vm), IIC_VMOVQ,
5253 OpcodeStr, Dt, "$Vd, $Vm", "",
5254 [(set QPR:$Vd, (Ty (NEONvrev32 (Ty QPR:$Vm))))]>;
5256 def VREV32d8 : VREV32D<0b00, "vrev32", "8", v8i8>;
5257 def VREV32d16 : VREV32D<0b01, "vrev32", "16", v4i16>;
5259 def VREV32q8 : VREV32Q<0b00, "vrev32", "8", v16i8>;
5260 def VREV32q16 : VREV32Q<0b01, "vrev32", "16", v8i16>;
5262 // VREV16 : Vector Reverse elements within 16-bit halfwords
5264 class VREV16D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
5265 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 0, 0, (outs DPR:$Vd),
5266 (ins DPR:$Vm), IIC_VMOVD,
5267 OpcodeStr, Dt, "$Vd, $Vm", "",
5268 [(set DPR:$Vd, (Ty (NEONvrev16 (Ty DPR:$Vm))))]>;
5269 class VREV16Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
5270 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 1, 0, (outs QPR:$Vd),
5271 (ins QPR:$Vm), IIC_VMOVQ,
5272 OpcodeStr, Dt, "$Vd, $Vm", "",
5273 [(set QPR:$Vd, (Ty (NEONvrev16 (Ty QPR:$Vm))))]>;
5275 def VREV16d8 : VREV16D<0b00, "vrev16", "8", v8i8>;
5276 def VREV16q8 : VREV16Q<0b00, "vrev16", "8", v16i8>;
5278 // Other Vector Shuffles.
5280 // Aligned extractions: really just dropping registers
5282 class AlignedVEXTq<ValueType DestTy, ValueType SrcTy, SDNodeXForm LaneCVT>
5283 : Pat<(DestTy (vector_extract_subvec (SrcTy QPR:$src), (i32 imm:$start))),
5284 (EXTRACT_SUBREG (SrcTy QPR:$src), (LaneCVT imm:$start))>;
5286 def : AlignedVEXTq<v8i8, v16i8, DSubReg_i8_reg>;
5288 def : AlignedVEXTq<v4i16, v8i16, DSubReg_i16_reg>;
5290 def : AlignedVEXTq<v2i32, v4i32, DSubReg_i32_reg>;
5292 def : AlignedVEXTq<v1i64, v2i64, DSubReg_f64_reg>;
5294 def : AlignedVEXTq<v2f32, v4f32, DSubReg_i32_reg>;
5297 // VEXT : Vector Extract
5299 class VEXTd<string OpcodeStr, string Dt, ValueType Ty, Operand immTy>
5300 : N3V<0,1,0b11,{?,?,?,?},0,0, (outs DPR:$Vd),
5301 (ins DPR:$Vn, DPR:$Vm, immTy:$index), NVExtFrm,
5302 IIC_VEXTD, OpcodeStr, Dt, "$Vd, $Vn, $Vm, $index", "",
5303 [(set DPR:$Vd, (Ty (NEONvext (Ty DPR:$Vn),
5304 (Ty DPR:$Vm), imm:$index)))]> {
5306 let Inst{11-8} = index{3-0};
5309 class VEXTq<string OpcodeStr, string Dt, ValueType Ty, Operand immTy>
5310 : N3V<0,1,0b11,{?,?,?,?},1,0, (outs QPR:$Vd),
5311 (ins QPR:$Vn, QPR:$Vm, imm0_15:$index), NVExtFrm,
5312 IIC_VEXTQ, OpcodeStr, Dt, "$Vd, $Vn, $Vm, $index", "",
5313 [(set QPR:$Vd, (Ty (NEONvext (Ty QPR:$Vn),
5314 (Ty QPR:$Vm), imm:$index)))]> {
5316 let Inst{11-8} = index{3-0};
5319 def VEXTd8 : VEXTd<"vext", "8", v8i8, imm0_7> {
5320 let Inst{11-8} = index{3-0};
5322 def VEXTd16 : VEXTd<"vext", "16", v4i16, imm0_3> {
5323 let Inst{11-9} = index{2-0};
5326 def VEXTd32 : VEXTd<"vext", "32", v2i32, imm0_1> {
5327 let Inst{11-10} = index{1-0};
5328 let Inst{9-8} = 0b00;
5330 def : Pat<(v2f32 (NEONvext (v2f32 DPR:$Vn),
5333 (VEXTd32 DPR:$Vn, DPR:$Vm, imm:$index)>;
5335 def VEXTq8 : VEXTq<"vext", "8", v16i8, imm0_15> {
5336 let Inst{11-8} = index{3-0};
5338 def VEXTq16 : VEXTq<"vext", "16", v8i16, imm0_7> {
5339 let Inst{11-9} = index{2-0};
5342 def VEXTq32 : VEXTq<"vext", "32", v4i32, imm0_3> {
5343 let Inst{11-10} = index{1-0};
5344 let Inst{9-8} = 0b00;
5346 def VEXTq64 : VEXTq<"vext", "64", v2i64, imm0_1> {
5347 let Inst{11} = index{0};
5348 let Inst{10-8} = 0b000;
5350 def : Pat<(v4f32 (NEONvext (v4f32 QPR:$Vn),
5353 (VEXTq32 QPR:$Vn, QPR:$Vm, imm:$index)>;
5355 // VTRN : Vector Transpose
5357 def VTRNd8 : N2VDShuffle<0b00, 0b00001, "vtrn", "8">;
5358 def VTRNd16 : N2VDShuffle<0b01, 0b00001, "vtrn", "16">;
5359 def VTRNd32 : N2VDShuffle<0b10, 0b00001, "vtrn", "32">;
5361 def VTRNq8 : N2VQShuffle<0b00, 0b00001, IIC_VPERMQ, "vtrn", "8">;
5362 def VTRNq16 : N2VQShuffle<0b01, 0b00001, IIC_VPERMQ, "vtrn", "16">;
5363 def VTRNq32 : N2VQShuffle<0b10, 0b00001, IIC_VPERMQ, "vtrn", "32">;
5365 // VUZP : Vector Unzip (Deinterleave)
5367 def VUZPd8 : N2VDShuffle<0b00, 0b00010, "vuzp", "8">;
5368 def VUZPd16 : N2VDShuffle<0b01, 0b00010, "vuzp", "16">;
5369 def VUZPd32 : N2VDShuffle<0b10, 0b00010, "vuzp", "32">;
5371 def VUZPq8 : N2VQShuffle<0b00, 0b00010, IIC_VPERMQ3, "vuzp", "8">;
5372 def VUZPq16 : N2VQShuffle<0b01, 0b00010, IIC_VPERMQ3, "vuzp", "16">;
5373 def VUZPq32 : N2VQShuffle<0b10, 0b00010, IIC_VPERMQ3, "vuzp", "32">;
5375 // VZIP : Vector Zip (Interleave)
5377 def VZIPd8 : N2VDShuffle<0b00, 0b00011, "vzip", "8">;
5378 def VZIPd16 : N2VDShuffle<0b01, 0b00011, "vzip", "16">;
5379 def VZIPd32 : N2VDShuffle<0b10, 0b00011, "vzip", "32">;
5381 def VZIPq8 : N2VQShuffle<0b00, 0b00011, IIC_VPERMQ3, "vzip", "8">;
5382 def VZIPq16 : N2VQShuffle<0b01, 0b00011, IIC_VPERMQ3, "vzip", "16">;
5383 def VZIPq32 : N2VQShuffle<0b10, 0b00011, IIC_VPERMQ3, "vzip", "32">;
5385 // Vector Table Lookup and Table Extension.
5387 // VTBL : Vector Table Lookup
5388 let DecoderMethod = "DecodeTBLInstruction" in {
5390 : N3V<1,1,0b11,0b1000,0,0, (outs DPR:$Vd),
5391 (ins VecListOneD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTB1,
5392 "vtbl", "8", "$Vd, $Vn, $Vm", "",
5393 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbl1 VecListOneD:$Vn, DPR:$Vm)))]>;
5394 let hasExtraSrcRegAllocReq = 1 in {
5396 : N3V<1,1,0b11,0b1001,0,0, (outs DPR:$Vd),
5397 (ins VecListDPair:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTB2,
5398 "vtbl", "8", "$Vd, $Vn, $Vm", "", []>;
5400 : N3V<1,1,0b11,0b1010,0,0, (outs DPR:$Vd),
5401 (ins VecListThreeD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTB3,
5402 "vtbl", "8", "$Vd, $Vn, $Vm", "", []>;
5404 : N3V<1,1,0b11,0b1011,0,0, (outs DPR:$Vd),
5405 (ins VecListFourD:$Vn, DPR:$Vm),
5407 "vtbl", "8", "$Vd, $Vn, $Vm", "", []>;
5408 } // hasExtraSrcRegAllocReq = 1
5411 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB3, "", []>;
5413 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB4, "", []>;
5415 // VTBX : Vector Table Extension
5417 : N3V<1,1,0b11,0b1000,1,0, (outs DPR:$Vd),
5418 (ins DPR:$orig, VecListOneD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTBX1,
5419 "vtbx", "8", "$Vd, $Vn, $Vm", "$orig = $Vd",
5420 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbx1
5421 DPR:$orig, VecListOneD:$Vn, DPR:$Vm)))]>;
5422 let hasExtraSrcRegAllocReq = 1 in {
5424 : N3V<1,1,0b11,0b1001,1,0, (outs DPR:$Vd),
5425 (ins DPR:$orig, VecListDPair:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTBX2,
5426 "vtbx", "8", "$Vd, $Vn, $Vm", "$orig = $Vd", []>;
5428 : N3V<1,1,0b11,0b1010,1,0, (outs DPR:$Vd),
5429 (ins DPR:$orig, VecListThreeD:$Vn, DPR:$Vm),
5430 NVTBLFrm, IIC_VTBX3,
5431 "vtbx", "8", "$Vd, $Vn, $Vm",
5434 : N3V<1,1,0b11,0b1011,1,0, (outs DPR:$Vd),
5435 (ins DPR:$orig, VecListFourD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTBX4,
5436 "vtbx", "8", "$Vd, $Vn, $Vm",
5438 } // hasExtraSrcRegAllocReq = 1
5441 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
5442 IIC_VTBX3, "$orig = $dst", []>;
5444 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
5445 IIC_VTBX4, "$orig = $dst", []>;
5446 } // DecoderMethod = "DecodeTBLInstruction"
5448 //===----------------------------------------------------------------------===//
5449 // NEON instructions for single-precision FP math
5450 //===----------------------------------------------------------------------===//
5452 class N2VSPat<SDNode OpNode, NeonI Inst>
5453 : NEONFPPat<(f32 (OpNode SPR:$a)),
5455 (v2f32 (COPY_TO_REGCLASS (Inst
5457 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5458 SPR:$a, ssub_0)), DPR_VFP2)), ssub_0)>;
5460 class N3VSPat<SDNode OpNode, NeonI Inst>
5461 : NEONFPPat<(f32 (OpNode SPR:$a, SPR:$b)),
5463 (v2f32 (COPY_TO_REGCLASS (Inst
5465 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5468 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5469 SPR:$b, ssub_0)), DPR_VFP2)), ssub_0)>;
5471 class N3VSMulOpPat<SDNode MulNode, SDNode OpNode, NeonI Inst>
5472 : NEONFPPat<(f32 (OpNode SPR:$acc, (f32 (MulNode SPR:$a, SPR:$b)))),
5474 (v2f32 (COPY_TO_REGCLASS (Inst
5476 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5479 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5482 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5483 SPR:$b, ssub_0)), DPR_VFP2)), ssub_0)>;
5485 def : N3VSPat<fadd, VADDfd>;
5486 def : N3VSPat<fsub, VSUBfd>;
5487 def : N3VSPat<fmul, VMULfd>;
5488 def : N3VSMulOpPat<fmul, fadd, VMLAfd>,
5489 Requires<[HasNEON, UseNEONForFP, UseFPVMLx, NoNEON2]>;
5490 def : N3VSMulOpPat<fmul, fsub, VMLSfd>,
5491 Requires<[HasNEON, UseNEONForFP, UseFPVMLx, NoNEON2]>;
5492 def : N3VSMulOpPat<fmul, fadd, VFMAfd>,
5493 Requires<[HasNEON2, UseNEONForFP,FPContractions]>;
5494 def : N3VSMulOpPat<fmul, fsub, VFMSfd>,
5495 Requires<[HasNEON2, UseNEONForFP,FPContractions]>;
5496 def : N2VSPat<fabs, VABSfd>;
5497 def : N2VSPat<fneg, VNEGfd>;
5498 def : N3VSPat<NEONfmax, VMAXfd>;
5499 def : N3VSPat<NEONfmin, VMINfd>;
5500 def : N2VSPat<arm_ftosi, VCVTf2sd>;
5501 def : N2VSPat<arm_ftoui, VCVTf2ud>;
5502 def : N2VSPat<arm_sitof, VCVTs2fd>;
5503 def : N2VSPat<arm_uitof, VCVTu2fd>;
5505 //===----------------------------------------------------------------------===//
5506 // Non-Instruction Patterns
5507 //===----------------------------------------------------------------------===//
5510 def : Pat<(v1i64 (bitconvert (v2i32 DPR:$src))), (v1i64 DPR:$src)>;
5511 def : Pat<(v1i64 (bitconvert (v4i16 DPR:$src))), (v1i64 DPR:$src)>;
5512 def : Pat<(v1i64 (bitconvert (v8i8 DPR:$src))), (v1i64 DPR:$src)>;
5513 def : Pat<(v1i64 (bitconvert (f64 DPR:$src))), (v1i64 DPR:$src)>;
5514 def : Pat<(v1i64 (bitconvert (v2f32 DPR:$src))), (v1i64 DPR:$src)>;
5515 def : Pat<(v2i32 (bitconvert (v1i64 DPR:$src))), (v2i32 DPR:$src)>;
5516 def : Pat<(v2i32 (bitconvert (v4i16 DPR:$src))), (v2i32 DPR:$src)>;
5517 def : Pat<(v2i32 (bitconvert (v8i8 DPR:$src))), (v2i32 DPR:$src)>;
5518 def : Pat<(v2i32 (bitconvert (f64 DPR:$src))), (v2i32 DPR:$src)>;
5519 def : Pat<(v2i32 (bitconvert (v2f32 DPR:$src))), (v2i32 DPR:$src)>;
5520 def : Pat<(v4i16 (bitconvert (v1i64 DPR:$src))), (v4i16 DPR:$src)>;
5521 def : Pat<(v4i16 (bitconvert (v2i32 DPR:$src))), (v4i16 DPR:$src)>;
5522 def : Pat<(v4i16 (bitconvert (v8i8 DPR:$src))), (v4i16 DPR:$src)>;
5523 def : Pat<(v4i16 (bitconvert (f64 DPR:$src))), (v4i16 DPR:$src)>;
5524 def : Pat<(v4i16 (bitconvert (v2f32 DPR:$src))), (v4i16 DPR:$src)>;
5525 def : Pat<(v8i8 (bitconvert (v1i64 DPR:$src))), (v8i8 DPR:$src)>;
5526 def : Pat<(v8i8 (bitconvert (v2i32 DPR:$src))), (v8i8 DPR:$src)>;
5527 def : Pat<(v8i8 (bitconvert (v4i16 DPR:$src))), (v8i8 DPR:$src)>;
5528 def : Pat<(v8i8 (bitconvert (f64 DPR:$src))), (v8i8 DPR:$src)>;
5529 def : Pat<(v8i8 (bitconvert (v2f32 DPR:$src))), (v8i8 DPR:$src)>;
5530 def : Pat<(f64 (bitconvert (v1i64 DPR:$src))), (f64 DPR:$src)>;
5531 def : Pat<(f64 (bitconvert (v2i32 DPR:$src))), (f64 DPR:$src)>;
5532 def : Pat<(f64 (bitconvert (v4i16 DPR:$src))), (f64 DPR:$src)>;
5533 def : Pat<(f64 (bitconvert (v8i8 DPR:$src))), (f64 DPR:$src)>;
5534 def : Pat<(f64 (bitconvert (v2f32 DPR:$src))), (f64 DPR:$src)>;
5535 def : Pat<(v2f32 (bitconvert (f64 DPR:$src))), (v2f32 DPR:$src)>;
5536 def : Pat<(v2f32 (bitconvert (v1i64 DPR:$src))), (v2f32 DPR:$src)>;
5537 def : Pat<(v2f32 (bitconvert (v2i32 DPR:$src))), (v2f32 DPR:$src)>;
5538 def : Pat<(v2f32 (bitconvert (v4i16 DPR:$src))), (v2f32 DPR:$src)>;
5539 def : Pat<(v2f32 (bitconvert (v8i8 DPR:$src))), (v2f32 DPR:$src)>;
5541 def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (v2i64 QPR:$src)>;
5542 def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (v2i64 QPR:$src)>;
5543 def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (v2i64 QPR:$src)>;
5544 def : Pat<(v2i64 (bitconvert (v2f64 QPR:$src))), (v2i64 QPR:$src)>;
5545 def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (v2i64 QPR:$src)>;
5546 def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (v4i32 QPR:$src)>;
5547 def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (v4i32 QPR:$src)>;
5548 def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (v4i32 QPR:$src)>;
5549 def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (v4i32 QPR:$src)>;
5550 def : Pat<(v4i32 (bitconvert (v4f32 QPR:$src))), (v4i32 QPR:$src)>;
5551 def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (v8i16 QPR:$src)>;
5552 def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (v8i16 QPR:$src)>;
5553 def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (v8i16 QPR:$src)>;
5554 def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (v8i16 QPR:$src)>;
5555 def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (v8i16 QPR:$src)>;
5556 def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (v16i8 QPR:$src)>;
5557 def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (v16i8 QPR:$src)>;
5558 def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (v16i8 QPR:$src)>;
5559 def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (v16i8 QPR:$src)>;
5560 def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (v16i8 QPR:$src)>;
5561 def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (v4f32 QPR:$src)>;
5562 def : Pat<(v4f32 (bitconvert (v4i32 QPR:$src))), (v4f32 QPR:$src)>;
5563 def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (v4f32 QPR:$src)>;
5564 def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (v4f32 QPR:$src)>;
5565 def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (v4f32 QPR:$src)>;
5566 def : Pat<(v2f64 (bitconvert (v2i64 QPR:$src))), (v2f64 QPR:$src)>;
5567 def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (v2f64 QPR:$src)>;
5568 def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 QPR:$src)>;
5569 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 QPR:$src)>;
5570 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 QPR:$src)>;
5572 // Vector lengthening move with load, matching extending loads.
5574 // extload, zextload and sextload for a standard lengthening load. Example:
5575 // Lengthen_Single<"8", "i16", "i8"> = Pat<(v8i16 (extloadvi8 addrmode5:$addr))
5576 // (VMOVLuv8i16 (VLDRD addrmode5:$addr))>;
5577 multiclass Lengthen_Single<string DestLanes, string DestTy, string SrcTy> {
5578 def _Any : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5579 (!cast<PatFrag>("extloadv" # SrcTy) addrmode5:$addr)),
5580 (!cast<Instruction>("VMOVLuv" # DestLanes # DestTy)
5581 (VLDRD addrmode5:$addr))>;
5582 def _Z : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5583 (!cast<PatFrag>("zextloadv" # SrcTy) addrmode5:$addr)),
5584 (!cast<Instruction>("VMOVLuv" # DestLanes # DestTy)
5585 (VLDRD addrmode5:$addr))>;
5586 def _S : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5587 (!cast<PatFrag>("sextloadv" # SrcTy) addrmode5:$addr)),
5588 (!cast<Instruction>("VMOVLsv" # DestLanes # DestTy)
5589 (VLDRD addrmode5:$addr))>;
5592 // extload, zextload and sextload for a lengthening load which only uses
5593 // half the lanes available. Example:
5594 // Lengthen_HalfSingle<"4", "i16", "8", "i16", "i8"> =
5595 // Pat<(v4i16 (extloadvi8 addrmode5:$addr))
5596 // (EXTRACT_SUBREG (VMOVLuv8i16 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
5597 // (VLDRS addrmode5:$addr),
5600 multiclass Lengthen_HalfSingle<string DestLanes, string DestTy, string SrcTy,
5601 string InsnLanes, string InsnTy> {
5602 def _Any : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5603 (!cast<PatFrag>("extloadv" # SrcTy) addrmode5:$addr)),
5604 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # InsnLanes # InsnTy)
5605 (INSERT_SUBREG (f64 (IMPLICIT_DEF)), (VLDRS addrmode5:$addr), ssub_0)),
5607 def _Z : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5608 (!cast<PatFrag>("zextloadv" # SrcTy) addrmode5:$addr)),
5609 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # InsnLanes # InsnTy)
5610 (INSERT_SUBREG (f64 (IMPLICIT_DEF)), (VLDRS addrmode5:$addr), ssub_0)),
5612 def _S : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5613 (!cast<PatFrag>("sextloadv" # SrcTy) addrmode5:$addr)),
5614 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLsv" # InsnLanes # InsnTy)
5615 (INSERT_SUBREG (f64 (IMPLICIT_DEF)), (VLDRS addrmode5:$addr), ssub_0)),
5619 // extload, zextload and sextload for a lengthening load followed by another
5620 // lengthening load, to quadruple the initial length.
5621 // Lengthen_Double<"4", "i32", "i8", "8", "i16", "4", "i32", qsub_0> =
5622 // Pat<(v4i32 (extloadvi8 addrmode5:$addr))
5623 // (EXTRACT_SUBREG (VMOVLuv4i32
5624 // (EXTRACT_SUBREG (VMOVLuv8i16 (INSERT_SUBREG (f64 (IMPLICIT_DEF)),
5625 // (VLDRS addrmode5:$addr),
5629 multiclass Lengthen_Double<string DestLanes, string DestTy, string SrcTy,
5630 string Insn1Lanes, string Insn1Ty, string Insn2Lanes,
5631 string Insn2Ty, SubRegIndex RegType> {
5632 def _Any : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5633 (!cast<PatFrag>("extloadv" # SrcTy) addrmode5:$addr)),
5634 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn2Lanes # Insn2Ty)
5635 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn1Lanes # Insn1Ty)
5636 (INSERT_SUBREG (f64 (IMPLICIT_DEF)), (VLDRS addrmode5:$addr),
5639 def _Z : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5640 (!cast<PatFrag>("zextloadv" # SrcTy) addrmode5:$addr)),
5641 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn2Lanes # Insn2Ty)
5642 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLuv" # Insn1Lanes # Insn1Ty)
5643 (INSERT_SUBREG (f64 (IMPLICIT_DEF)), (VLDRS addrmode5:$addr),
5646 def _S : Pat<(!cast<ValueType>("v" # DestLanes # DestTy)
5647 (!cast<PatFrag>("sextloadv" # SrcTy) addrmode5:$addr)),
5648 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLsv" # Insn2Lanes # Insn2Ty)
5649 (EXTRACT_SUBREG (!cast<Instruction>("VMOVLsv" # Insn1Lanes # Insn1Ty)
5650 (INSERT_SUBREG (f64 (IMPLICIT_DEF)), (VLDRS addrmode5:$addr),
5655 defm : Lengthen_Single<"8", "i16", "i8">; // v8i8 -> v8i16
5656 defm : Lengthen_Single<"4", "i32", "i16">; // v4i16 -> v4i32
5657 defm : Lengthen_Single<"2", "i64", "i32">; // v2i32 -> v2i64
5659 defm : Lengthen_HalfSingle<"4", "i16", "i8", "8", "i16">; // v4i8 -> v4i16
5660 defm : Lengthen_HalfSingle<"2", "i16", "i8", "8", "i16">; // v2i8 -> v2i16
5661 defm : Lengthen_HalfSingle<"2", "i32", "i16", "4", "i32">; // v2i16 -> v2i32
5663 // Double lengthening - v4i8 -> v4i16 -> v4i32
5664 defm : Lengthen_Double<"4", "i32", "i8", "8", "i16", "4", "i32", qsub_0>;
5665 // v2i8 -> v2i16 -> v2i32
5666 defm : Lengthen_Double<"2", "i32", "i8", "8", "i16", "4", "i32", dsub_0>;
5667 // v2i16 -> v2i32 -> v2i64
5668 defm : Lengthen_Double<"2", "i64", "i16", "4", "i32", "2", "i64", qsub_0>;
5670 // Triple lengthening - v2i8 -> v2i16 -> v2i32 -> v2i64
5671 def : Pat<(v2i64 (extloadvi8 addrmode5:$addr)),
5672 (VMOVLuv2i64 (EXTRACT_SUBREG (VMOVLuv4i32 (EXTRACT_SUBREG (VMOVLuv8i16
5673 (INSERT_SUBREG (f64 (IMPLICIT_DEF)), (VLDRS addrmode5:$addr), ssub_0)),
5674 dsub_0)), dsub_0))>;
5675 def : Pat<(v2i64 (zextloadvi8 addrmode5:$addr)),
5676 (VMOVLuv2i64 (EXTRACT_SUBREG (VMOVLuv4i32 (EXTRACT_SUBREG (VMOVLuv8i16
5677 (INSERT_SUBREG (f64 (IMPLICIT_DEF)), (VLDRS addrmode5:$addr), ssub_0)),
5678 dsub_0)), dsub_0))>;
5679 def : Pat<(v2i64 (sextloadvi8 addrmode5:$addr)),
5680 (VMOVLsv2i64 (EXTRACT_SUBREG (VMOVLsv4i32 (EXTRACT_SUBREG (VMOVLsv8i16
5681 (INSERT_SUBREG (f64 (IMPLICIT_DEF)), (VLDRS addrmode5:$addr), ssub_0)),
5682 dsub_0)), dsub_0))>;
5684 //===----------------------------------------------------------------------===//
5685 // Assembler aliases
5688 def : VFP2InstAlias<"fmdhr${p} $Dd, $Rn",
5689 (VSETLNi32 DPR:$Dd, GPR:$Rn, 1, pred:$p)>;
5690 def : VFP2InstAlias<"fmdlr${p} $Dd, $Rn",
5691 (VSETLNi32 DPR:$Dd, GPR:$Rn, 0, pred:$p)>;
5694 // VADD two-operand aliases.
5695 def : NEONInstAlias<"vadd${p}.i8 $Vdn, $Vm",
5696 (VADDv16i8 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5697 def : NEONInstAlias<"vadd${p}.i16 $Vdn, $Vm",
5698 (VADDv8i16 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5699 def : NEONInstAlias<"vadd${p}.i32 $Vdn, $Vm",
5700 (VADDv4i32 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5701 def : NEONInstAlias<"vadd${p}.i64 $Vdn, $Vm",
5702 (VADDv2i64 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5704 def : NEONInstAlias<"vadd${p}.i8 $Vdn, $Vm",
5705 (VADDv8i8 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5706 def : NEONInstAlias<"vadd${p}.i16 $Vdn, $Vm",
5707 (VADDv4i16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5708 def : NEONInstAlias<"vadd${p}.i32 $Vdn, $Vm",
5709 (VADDv2i32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5710 def : NEONInstAlias<"vadd${p}.i64 $Vdn, $Vm",
5711 (VADDv1i64 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5713 def : NEONInstAlias<"vadd${p}.f32 $Vdn, $Vm",
5714 (VADDfd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5715 def : NEONInstAlias<"vadd${p}.f32 $Vdn, $Vm",
5716 (VADDfq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5718 // VSUB two-operand aliases.
5719 def : NEONInstAlias<"vsub${p}.i8 $Vdn, $Vm",
5720 (VSUBv16i8 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5721 def : NEONInstAlias<"vsub${p}.i16 $Vdn, $Vm",
5722 (VSUBv8i16 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5723 def : NEONInstAlias<"vsub${p}.i32 $Vdn, $Vm",
5724 (VSUBv4i32 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5725 def : NEONInstAlias<"vsub${p}.i64 $Vdn, $Vm",
5726 (VSUBv2i64 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5728 def : NEONInstAlias<"vsub${p}.i8 $Vdn, $Vm",
5729 (VSUBv8i8 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5730 def : NEONInstAlias<"vsub${p}.i16 $Vdn, $Vm",
5731 (VSUBv4i16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5732 def : NEONInstAlias<"vsub${p}.i32 $Vdn, $Vm",
5733 (VSUBv2i32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5734 def : NEONInstAlias<"vsub${p}.i64 $Vdn, $Vm",
5735 (VSUBv1i64 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5737 def : NEONInstAlias<"vsub${p}.f32 $Vdn, $Vm",
5738 (VSUBfd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5739 def : NEONInstAlias<"vsub${p}.f32 $Vdn, $Vm",
5740 (VSUBfq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5742 // VADDW two-operand aliases.
5743 def : NEONInstAlias<"vaddw${p}.s8 $Vdn, $Vm",
5744 (VADDWsv8i16 QPR:$Vdn, QPR:$Vdn, DPR:$Vm, pred:$p)>;
5745 def : NEONInstAlias<"vaddw${p}.s16 $Vdn, $Vm",
5746 (VADDWsv4i32 QPR:$Vdn, QPR:$Vdn, DPR:$Vm, pred:$p)>;
5747 def : NEONInstAlias<"vaddw${p}.s32 $Vdn, $Vm",
5748 (VADDWsv2i64 QPR:$Vdn, QPR:$Vdn, DPR:$Vm, pred:$p)>;
5749 def : NEONInstAlias<"vaddw${p}.u8 $Vdn, $Vm",
5750 (VADDWuv8i16 QPR:$Vdn, QPR:$Vdn, DPR:$Vm, pred:$p)>;
5751 def : NEONInstAlias<"vaddw${p}.u16 $Vdn, $Vm",
5752 (VADDWuv4i32 QPR:$Vdn, QPR:$Vdn, DPR:$Vm, pred:$p)>;
5753 def : NEONInstAlias<"vaddw${p}.u32 $Vdn, $Vm",
5754 (VADDWuv2i64 QPR:$Vdn, QPR:$Vdn, DPR:$Vm, pred:$p)>;
5756 // VAND/VBIC/VEOR/VORR accept but do not require a type suffix.
5757 defm : NEONDTAnyInstAlias<"vand${p}", "$Vd, $Vn, $Vm",
5758 (VANDd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
5759 defm : NEONDTAnyInstAlias<"vand${p}", "$Vd, $Vn, $Vm",
5760 (VANDq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
5761 defm : NEONDTAnyInstAlias<"vbic${p}", "$Vd, $Vn, $Vm",
5762 (VBICd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
5763 defm : NEONDTAnyInstAlias<"vbic${p}", "$Vd, $Vn, $Vm",
5764 (VBICq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
5765 defm : NEONDTAnyInstAlias<"veor${p}", "$Vd, $Vn, $Vm",
5766 (VEORd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
5767 defm : NEONDTAnyInstAlias<"veor${p}", "$Vd, $Vn, $Vm",
5768 (VEORq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
5769 defm : NEONDTAnyInstAlias<"vorr${p}", "$Vd, $Vn, $Vm",
5770 (VORRd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
5771 defm : NEONDTAnyInstAlias<"vorr${p}", "$Vd, $Vn, $Vm",
5772 (VORRq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
5773 // ... two-operand aliases
5774 def : NEONInstAlias<"vand${p} $Vdn, $Vm",
5775 (VANDd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5776 def : NEONInstAlias<"vand${p} $Vdn, $Vm",
5777 (VANDq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5778 def : NEONInstAlias<"vbic${p} $Vdn, $Vm",
5779 (VBICd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5780 def : NEONInstAlias<"vbic${p} $Vdn, $Vm",
5781 (VBICq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5782 def : NEONInstAlias<"veor${p} $Vdn, $Vm",
5783 (VEORd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5784 def : NEONInstAlias<"veor${p} $Vdn, $Vm",
5785 (VEORq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5786 def : NEONInstAlias<"vorr${p} $Vdn, $Vm",
5787 (VORRd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5788 def : NEONInstAlias<"vorr${p} $Vdn, $Vm",
5789 (VORRq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5791 defm : NEONDTAnyInstAlias<"vand${p}", "$Vdn, $Vm",
5792 (VANDd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5793 defm : NEONDTAnyInstAlias<"vand${p}", "$Vdn, $Vm",
5794 (VANDq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5795 defm : NEONDTAnyInstAlias<"veor${p}", "$Vdn, $Vm",
5796 (VEORd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5797 defm : NEONDTAnyInstAlias<"veor${p}", "$Vdn, $Vm",
5798 (VEORq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5799 defm : NEONDTAnyInstAlias<"vorr${p}", "$Vdn, $Vm",
5800 (VORRd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5801 defm : NEONDTAnyInstAlias<"vorr${p}", "$Vdn, $Vm",
5802 (VORRq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5804 // VMUL two-operand aliases.
5805 def : NEONInstAlias<"vmul${p}.p8 $Qdn, $Qm",
5806 (VMULpq QPR:$Qdn, QPR:$Qdn, QPR:$Qm, pred:$p)>;
5807 def : NEONInstAlias<"vmul${p}.i8 $Qdn, $Qm",
5808 (VMULv16i8 QPR:$Qdn, QPR:$Qdn, QPR:$Qm, pred:$p)>;
5809 def : NEONInstAlias<"vmul${p}.i16 $Qdn, $Qm",
5810 (VMULv8i16 QPR:$Qdn, QPR:$Qdn, QPR:$Qm, pred:$p)>;
5811 def : NEONInstAlias<"vmul${p}.i32 $Qdn, $Qm",
5812 (VMULv4i32 QPR:$Qdn, QPR:$Qdn, QPR:$Qm, pred:$p)>;
5814 def : NEONInstAlias<"vmul${p}.p8 $Ddn, $Dm",
5815 (VMULpd DPR:$Ddn, DPR:$Ddn, DPR:$Dm, pred:$p)>;
5816 def : NEONInstAlias<"vmul${p}.i8 $Ddn, $Dm",
5817 (VMULv8i8 DPR:$Ddn, DPR:$Ddn, DPR:$Dm, pred:$p)>;
5818 def : NEONInstAlias<"vmul${p}.i16 $Ddn, $Dm",
5819 (VMULv4i16 DPR:$Ddn, DPR:$Ddn, DPR:$Dm, pred:$p)>;
5820 def : NEONInstAlias<"vmul${p}.i32 $Ddn, $Dm",
5821 (VMULv2i32 DPR:$Ddn, DPR:$Ddn, DPR:$Dm, pred:$p)>;
5823 def : NEONInstAlias<"vmul${p}.f32 $Qdn, $Qm",
5824 (VMULfq QPR:$Qdn, QPR:$Qdn, QPR:$Qm, pred:$p)>;
5825 def : NEONInstAlias<"vmul${p}.f32 $Ddn, $Dm",
5826 (VMULfd DPR:$Ddn, DPR:$Ddn, DPR:$Dm, pred:$p)>;
5828 def : NEONInstAlias<"vmul${p}.i16 $Ddn, $Dm$lane",
5829 (VMULslv4i16 DPR:$Ddn, DPR:$Ddn, DPR_8:$Dm,
5830 VectorIndex16:$lane, pred:$p)>;
5831 def : NEONInstAlias<"vmul${p}.i16 $Qdn, $Dm$lane",
5832 (VMULslv8i16 QPR:$Qdn, QPR:$Qdn, DPR_8:$Dm,
5833 VectorIndex16:$lane, pred:$p)>;
5835 def : NEONInstAlias<"vmul${p}.i32 $Ddn, $Dm$lane",
5836 (VMULslv2i32 DPR:$Ddn, DPR:$Ddn, DPR_VFP2:$Dm,
5837 VectorIndex32:$lane, pred:$p)>;
5838 def : NEONInstAlias<"vmul${p}.i32 $Qdn, $Dm$lane",
5839 (VMULslv4i32 QPR:$Qdn, QPR:$Qdn, DPR_VFP2:$Dm,
5840 VectorIndex32:$lane, pred:$p)>;
5842 def : NEONInstAlias<"vmul${p}.f32 $Ddn, $Dm$lane",
5843 (VMULslfd DPR:$Ddn, DPR:$Ddn, DPR_VFP2:$Dm,
5844 VectorIndex32:$lane, pred:$p)>;
5845 def : NEONInstAlias<"vmul${p}.f32 $Qdn, $Dm$lane",
5846 (VMULslfq QPR:$Qdn, QPR:$Qdn, DPR_VFP2:$Dm,
5847 VectorIndex32:$lane, pred:$p)>;
5849 // VQADD (register) two-operand aliases.
5850 def : NEONInstAlias<"vqadd${p}.s8 $Vdn, $Vm",
5851 (VQADDsv8i8 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5852 def : NEONInstAlias<"vqadd${p}.s16 $Vdn, $Vm",
5853 (VQADDsv4i16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5854 def : NEONInstAlias<"vqadd${p}.s32 $Vdn, $Vm",
5855 (VQADDsv2i32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5856 def : NEONInstAlias<"vqadd${p}.s64 $Vdn, $Vm",
5857 (VQADDsv1i64 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5858 def : NEONInstAlias<"vqadd${p}.u8 $Vdn, $Vm",
5859 (VQADDuv8i8 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5860 def : NEONInstAlias<"vqadd${p}.u16 $Vdn, $Vm",
5861 (VQADDuv4i16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5862 def : NEONInstAlias<"vqadd${p}.u32 $Vdn, $Vm",
5863 (VQADDuv2i32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5864 def : NEONInstAlias<"vqadd${p}.u64 $Vdn, $Vm",
5865 (VQADDuv1i64 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5867 def : NEONInstAlias<"vqadd${p}.s8 $Vdn, $Vm",
5868 (VQADDsv16i8 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5869 def : NEONInstAlias<"vqadd${p}.s16 $Vdn, $Vm",
5870 (VQADDsv8i16 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5871 def : NEONInstAlias<"vqadd${p}.s32 $Vdn, $Vm",
5872 (VQADDsv4i32 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5873 def : NEONInstAlias<"vqadd${p}.s64 $Vdn, $Vm",
5874 (VQADDsv2i64 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5875 def : NEONInstAlias<"vqadd${p}.u8 $Vdn, $Vm",
5876 (VQADDuv16i8 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5877 def : NEONInstAlias<"vqadd${p}.u16 $Vdn, $Vm",
5878 (VQADDuv8i16 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5879 def : NEONInstAlias<"vqadd${p}.u32 $Vdn, $Vm",
5880 (VQADDuv4i32 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5881 def : NEONInstAlias<"vqadd${p}.u64 $Vdn, $Vm",
5882 (VQADDuv2i64 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5884 // VSHL (immediate) two-operand aliases.
5885 def : NEONInstAlias<"vshl${p}.i8 $Vdn, $imm",
5886 (VSHLiv8i8 DPR:$Vdn, DPR:$Vdn, imm0_7:$imm, pred:$p)>;
5887 def : NEONInstAlias<"vshl${p}.i16 $Vdn, $imm",
5888 (VSHLiv4i16 DPR:$Vdn, DPR:$Vdn, imm0_15:$imm, pred:$p)>;
5889 def : NEONInstAlias<"vshl${p}.i32 $Vdn, $imm",
5890 (VSHLiv2i32 DPR:$Vdn, DPR:$Vdn, imm0_31:$imm, pred:$p)>;
5891 def : NEONInstAlias<"vshl${p}.i64 $Vdn, $imm",
5892 (VSHLiv1i64 DPR:$Vdn, DPR:$Vdn, imm0_63:$imm, pred:$p)>;
5894 def : NEONInstAlias<"vshl${p}.i8 $Vdn, $imm",
5895 (VSHLiv16i8 QPR:$Vdn, QPR:$Vdn, imm0_7:$imm, pred:$p)>;
5896 def : NEONInstAlias<"vshl${p}.i16 $Vdn, $imm",
5897 (VSHLiv8i16 QPR:$Vdn, QPR:$Vdn, imm0_15:$imm, pred:$p)>;
5898 def : NEONInstAlias<"vshl${p}.i32 $Vdn, $imm",
5899 (VSHLiv4i32 QPR:$Vdn, QPR:$Vdn, imm0_31:$imm, pred:$p)>;
5900 def : NEONInstAlias<"vshl${p}.i64 $Vdn, $imm",
5901 (VSHLiv2i64 QPR:$Vdn, QPR:$Vdn, imm0_63:$imm, pred:$p)>;
5903 // VSHL (register) two-operand aliases.
5904 def : NEONInstAlias<"vshl${p}.s8 $Vdn, $Vm",
5905 (VSHLsv8i8 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5906 def : NEONInstAlias<"vshl${p}.s16 $Vdn, $Vm",
5907 (VSHLsv4i16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5908 def : NEONInstAlias<"vshl${p}.s32 $Vdn, $Vm",
5909 (VSHLsv2i32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5910 def : NEONInstAlias<"vshl${p}.s64 $Vdn, $Vm",
5911 (VSHLsv1i64 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5912 def : NEONInstAlias<"vshl${p}.u8 $Vdn, $Vm",
5913 (VSHLuv8i8 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5914 def : NEONInstAlias<"vshl${p}.u16 $Vdn, $Vm",
5915 (VSHLuv4i16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5916 def : NEONInstAlias<"vshl${p}.u32 $Vdn, $Vm",
5917 (VSHLuv2i32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5918 def : NEONInstAlias<"vshl${p}.u64 $Vdn, $Vm",
5919 (VSHLuv1i64 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5921 def : NEONInstAlias<"vshl${p}.s8 $Vdn, $Vm",
5922 (VSHLsv16i8 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5923 def : NEONInstAlias<"vshl${p}.s16 $Vdn, $Vm",
5924 (VSHLsv8i16 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5925 def : NEONInstAlias<"vshl${p}.s32 $Vdn, $Vm",
5926 (VSHLsv4i32 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5927 def : NEONInstAlias<"vshl${p}.s64 $Vdn, $Vm",
5928 (VSHLsv2i64 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5929 def : NEONInstAlias<"vshl${p}.u8 $Vdn, $Vm",
5930 (VSHLuv16i8 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5931 def : NEONInstAlias<"vshl${p}.u16 $Vdn, $Vm",
5932 (VSHLuv8i16 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5933 def : NEONInstAlias<"vshl${p}.u32 $Vdn, $Vm",
5934 (VSHLuv4i32 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5935 def : NEONInstAlias<"vshl${p}.u64 $Vdn, $Vm",
5936 (VSHLuv2i64 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5938 // VSHL (immediate) two-operand aliases.
5939 def : NEONInstAlias<"vshr${p}.s8 $Vdn, $imm",
5940 (VSHRsv8i8 DPR:$Vdn, DPR:$Vdn, shr_imm8:$imm, pred:$p)>;
5941 def : NEONInstAlias<"vshr${p}.s16 $Vdn, $imm",
5942 (VSHRsv4i16 DPR:$Vdn, DPR:$Vdn, shr_imm16:$imm, pred:$p)>;
5943 def : NEONInstAlias<"vshr${p}.s32 $Vdn, $imm",
5944 (VSHRsv2i32 DPR:$Vdn, DPR:$Vdn, shr_imm32:$imm, pred:$p)>;
5945 def : NEONInstAlias<"vshr${p}.s64 $Vdn, $imm",
5946 (VSHRsv1i64 DPR:$Vdn, DPR:$Vdn, shr_imm64:$imm, pred:$p)>;
5948 def : NEONInstAlias<"vshr${p}.s8 $Vdn, $imm",
5949 (VSHRsv16i8 QPR:$Vdn, QPR:$Vdn, shr_imm8:$imm, pred:$p)>;
5950 def : NEONInstAlias<"vshr${p}.s16 $Vdn, $imm",
5951 (VSHRsv8i16 QPR:$Vdn, QPR:$Vdn, shr_imm16:$imm, pred:$p)>;
5952 def : NEONInstAlias<"vshr${p}.s32 $Vdn, $imm",
5953 (VSHRsv4i32 QPR:$Vdn, QPR:$Vdn, shr_imm32:$imm, pred:$p)>;
5954 def : NEONInstAlias<"vshr${p}.s64 $Vdn, $imm",
5955 (VSHRsv2i64 QPR:$Vdn, QPR:$Vdn, shr_imm64:$imm, pred:$p)>;
5957 def : NEONInstAlias<"vshr${p}.u8 $Vdn, $imm",
5958 (VSHRuv8i8 DPR:$Vdn, DPR:$Vdn, shr_imm8:$imm, pred:$p)>;
5959 def : NEONInstAlias<"vshr${p}.u16 $Vdn, $imm",
5960 (VSHRuv4i16 DPR:$Vdn, DPR:$Vdn, shr_imm16:$imm, pred:$p)>;
5961 def : NEONInstAlias<"vshr${p}.u32 $Vdn, $imm",
5962 (VSHRuv2i32 DPR:$Vdn, DPR:$Vdn, shr_imm32:$imm, pred:$p)>;
5963 def : NEONInstAlias<"vshr${p}.u64 $Vdn, $imm",
5964 (VSHRuv1i64 DPR:$Vdn, DPR:$Vdn, shr_imm64:$imm, pred:$p)>;
5966 def : NEONInstAlias<"vshr${p}.u8 $Vdn, $imm",
5967 (VSHRuv16i8 QPR:$Vdn, QPR:$Vdn, shr_imm8:$imm, pred:$p)>;
5968 def : NEONInstAlias<"vshr${p}.u16 $Vdn, $imm",
5969 (VSHRuv8i16 QPR:$Vdn, QPR:$Vdn, shr_imm16:$imm, pred:$p)>;
5970 def : NEONInstAlias<"vshr${p}.u32 $Vdn, $imm",
5971 (VSHRuv4i32 QPR:$Vdn, QPR:$Vdn, shr_imm32:$imm, pred:$p)>;
5972 def : NEONInstAlias<"vshr${p}.u64 $Vdn, $imm",
5973 (VSHRuv2i64 QPR:$Vdn, QPR:$Vdn, shr_imm64:$imm, pred:$p)>;
5975 // VLD1 single-lane pseudo-instructions. These need special handling for
5976 // the lane index that an InstAlias can't handle, so we use these instead.
5977 def VLD1LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vld1${p}", ".8", "$list, $addr",
5978 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5979 def VLD1LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vld1${p}", ".16", "$list, $addr",
5980 (ins VecListOneDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5981 def VLD1LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vld1${p}", ".32", "$list, $addr",
5982 (ins VecListOneDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5984 def VLD1LNdWB_fixed_Asm_8 :
5985 NEONDataTypeAsmPseudoInst<"vld1${p}", ".8", "$list, $addr!",
5986 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5987 def VLD1LNdWB_fixed_Asm_16 :
5988 NEONDataTypeAsmPseudoInst<"vld1${p}", ".16", "$list, $addr!",
5989 (ins VecListOneDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5990 def VLD1LNdWB_fixed_Asm_32 :
5991 NEONDataTypeAsmPseudoInst<"vld1${p}", ".32", "$list, $addr!",
5992 (ins VecListOneDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5993 def VLD1LNdWB_register_Asm_8 :
5994 NEONDataTypeAsmPseudoInst<"vld1${p}", ".8", "$list, $addr, $Rm",
5995 (ins VecListOneDByteIndexed:$list, addrmode6:$addr,
5996 rGPR:$Rm, pred:$p)>;
5997 def VLD1LNdWB_register_Asm_16 :
5998 NEONDataTypeAsmPseudoInst<"vld1${p}", ".16", "$list, $addr, $Rm",
5999 (ins VecListOneDHWordIndexed:$list, addrmode6:$addr,
6000 rGPR:$Rm, pred:$p)>;
6001 def VLD1LNdWB_register_Asm_32 :
6002 NEONDataTypeAsmPseudoInst<"vld1${p}", ".32", "$list, $addr, $Rm",
6003 (ins VecListOneDWordIndexed:$list, addrmode6:$addr,
6004 rGPR:$Rm, pred:$p)>;
6007 // VST1 single-lane pseudo-instructions. These need special handling for
6008 // the lane index that an InstAlias can't handle, so we use these instead.
6009 def VST1LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vst1${p}", ".8", "$list, $addr",
6010 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6011 def VST1LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vst1${p}", ".16", "$list, $addr",
6012 (ins VecListOneDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6013 def VST1LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vst1${p}", ".32", "$list, $addr",
6014 (ins VecListOneDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6016 def VST1LNdWB_fixed_Asm_8 :
6017 NEONDataTypeAsmPseudoInst<"vst1${p}", ".8", "$list, $addr!",
6018 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6019 def VST1LNdWB_fixed_Asm_16 :
6020 NEONDataTypeAsmPseudoInst<"vst1${p}", ".16", "$list, $addr!",
6021 (ins VecListOneDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6022 def VST1LNdWB_fixed_Asm_32 :
6023 NEONDataTypeAsmPseudoInst<"vst1${p}", ".32", "$list, $addr!",
6024 (ins VecListOneDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6025 def VST1LNdWB_register_Asm_8 :
6026 NEONDataTypeAsmPseudoInst<"vst1${p}", ".8", "$list, $addr, $Rm",
6027 (ins VecListOneDByteIndexed:$list, addrmode6:$addr,
6028 rGPR:$Rm, pred:$p)>;
6029 def VST1LNdWB_register_Asm_16 :
6030 NEONDataTypeAsmPseudoInst<"vst1${p}", ".16", "$list, $addr, $Rm",
6031 (ins VecListOneDHWordIndexed:$list, addrmode6:$addr,
6032 rGPR:$Rm, pred:$p)>;
6033 def VST1LNdWB_register_Asm_32 :
6034 NEONDataTypeAsmPseudoInst<"vst1${p}", ".32", "$list, $addr, $Rm",
6035 (ins VecListOneDWordIndexed:$list, addrmode6:$addr,
6036 rGPR:$Rm, pred:$p)>;
6038 // VLD2 single-lane pseudo-instructions. These need special handling for
6039 // the lane index that an InstAlias can't handle, so we use these instead.
6040 def VLD2LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vld2${p}", ".8", "$list, $addr",
6041 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6042 def VLD2LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vld2${p}", ".16", "$list, $addr",
6043 (ins VecListTwoDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6044 def VLD2LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vld2${p}", ".32", "$list, $addr",
6045 (ins VecListTwoDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6046 def VLD2LNqAsm_16 : NEONDataTypeAsmPseudoInst<"vld2${p}", ".16", "$list, $addr",
6047 (ins VecListTwoQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6048 def VLD2LNqAsm_32 : NEONDataTypeAsmPseudoInst<"vld2${p}", ".32", "$list, $addr",
6049 (ins VecListTwoQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6051 def VLD2LNdWB_fixed_Asm_8 :
6052 NEONDataTypeAsmPseudoInst<"vld2${p}", ".8", "$list, $addr!",
6053 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6054 def VLD2LNdWB_fixed_Asm_16 :
6055 NEONDataTypeAsmPseudoInst<"vld2${p}", ".16", "$list, $addr!",
6056 (ins VecListTwoDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6057 def VLD2LNdWB_fixed_Asm_32 :
6058 NEONDataTypeAsmPseudoInst<"vld2${p}", ".32", "$list, $addr!",
6059 (ins VecListTwoDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6060 def VLD2LNqWB_fixed_Asm_16 :
6061 NEONDataTypeAsmPseudoInst<"vld2${p}", ".16", "$list, $addr!",
6062 (ins VecListTwoQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6063 def VLD2LNqWB_fixed_Asm_32 :
6064 NEONDataTypeAsmPseudoInst<"vld2${p}", ".32", "$list, $addr!",
6065 (ins VecListTwoQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6066 def VLD2LNdWB_register_Asm_8 :
6067 NEONDataTypeAsmPseudoInst<"vld2${p}", ".8", "$list, $addr, $Rm",
6068 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr,
6069 rGPR:$Rm, pred:$p)>;
6070 def VLD2LNdWB_register_Asm_16 :
6071 NEONDataTypeAsmPseudoInst<"vld2${p}", ".16", "$list, $addr, $Rm",
6072 (ins VecListTwoDHWordIndexed:$list, addrmode6:$addr,
6073 rGPR:$Rm, pred:$p)>;
6074 def VLD2LNdWB_register_Asm_32 :
6075 NEONDataTypeAsmPseudoInst<"vld2${p}", ".32", "$list, $addr, $Rm",
6076 (ins VecListTwoDWordIndexed:$list, addrmode6:$addr,
6077 rGPR:$Rm, pred:$p)>;
6078 def VLD2LNqWB_register_Asm_16 :
6079 NEONDataTypeAsmPseudoInst<"vld2${p}", ".16", "$list, $addr, $Rm",
6080 (ins VecListTwoQHWordIndexed:$list, addrmode6:$addr,
6081 rGPR:$Rm, pred:$p)>;
6082 def VLD2LNqWB_register_Asm_32 :
6083 NEONDataTypeAsmPseudoInst<"vld2${p}", ".32", "$list, $addr, $Rm",
6084 (ins VecListTwoQWordIndexed:$list, addrmode6:$addr,
6085 rGPR:$Rm, pred:$p)>;
6088 // VST2 single-lane pseudo-instructions. These need special handling for
6089 // the lane index that an InstAlias can't handle, so we use these instead.
6090 def VST2LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vst2${p}", ".8", "$list, $addr",
6091 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6092 def VST2LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vst2${p}", ".16", "$list, $addr",
6093 (ins VecListTwoDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6094 def VST2LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vst2${p}", ".32", "$list, $addr",
6095 (ins VecListTwoDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6096 def VST2LNqAsm_16 : NEONDataTypeAsmPseudoInst<"vst2${p}", ".16", "$list, $addr",
6097 (ins VecListTwoQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6098 def VST2LNqAsm_32 : NEONDataTypeAsmPseudoInst<"vst2${p}", ".32", "$list, $addr",
6099 (ins VecListTwoQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6101 def VST2LNdWB_fixed_Asm_8 :
6102 NEONDataTypeAsmPseudoInst<"vst2${p}", ".8", "$list, $addr!",
6103 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6104 def VST2LNdWB_fixed_Asm_16 :
6105 NEONDataTypeAsmPseudoInst<"vst2${p}", ".16", "$list, $addr!",
6106 (ins VecListTwoDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6107 def VST2LNdWB_fixed_Asm_32 :
6108 NEONDataTypeAsmPseudoInst<"vst2${p}", ".32", "$list, $addr!",
6109 (ins VecListTwoDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6110 def VST2LNqWB_fixed_Asm_16 :
6111 NEONDataTypeAsmPseudoInst<"vst2${p}", ".16", "$list, $addr!",
6112 (ins VecListTwoQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6113 def VST2LNqWB_fixed_Asm_32 :
6114 NEONDataTypeAsmPseudoInst<"vst2${p}", ".32", "$list, $addr!",
6115 (ins VecListTwoQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6116 def VST2LNdWB_register_Asm_8 :
6117 NEONDataTypeAsmPseudoInst<"vst2${p}", ".8", "$list, $addr, $Rm",
6118 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr,
6119 rGPR:$Rm, pred:$p)>;
6120 def VST2LNdWB_register_Asm_16 :
6121 NEONDataTypeAsmPseudoInst<"vst2${p}", ".16","$list, $addr, $Rm",
6122 (ins VecListTwoDHWordIndexed:$list, addrmode6:$addr,
6123 rGPR:$Rm, pred:$p)>;
6124 def VST2LNdWB_register_Asm_32 :
6125 NEONDataTypeAsmPseudoInst<"vst2${p}", ".32", "$list, $addr, $Rm",
6126 (ins VecListTwoDWordIndexed:$list, addrmode6:$addr,
6127 rGPR:$Rm, pred:$p)>;
6128 def VST2LNqWB_register_Asm_16 :
6129 NEONDataTypeAsmPseudoInst<"vst2${p}", ".16","$list, $addr, $Rm",
6130 (ins VecListTwoQHWordIndexed:$list, addrmode6:$addr,
6131 rGPR:$Rm, pred:$p)>;
6132 def VST2LNqWB_register_Asm_32 :
6133 NEONDataTypeAsmPseudoInst<"vst2${p}", ".32", "$list, $addr, $Rm",
6134 (ins VecListTwoQWordIndexed:$list, addrmode6:$addr,
6135 rGPR:$Rm, pred:$p)>;
6137 // VLD3 all-lanes pseudo-instructions. These need special handling for
6138 // the lane index that an InstAlias can't handle, so we use these instead.
6139 def VLD3DUPdAsm_8 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr",
6140 (ins VecListThreeDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6141 def VLD3DUPdAsm_16 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr",
6142 (ins VecListThreeDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6143 def VLD3DUPdAsm_32 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr",
6144 (ins VecListThreeDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6145 def VLD3DUPqAsm_8 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr",
6146 (ins VecListThreeQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6147 def VLD3DUPqAsm_16 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr",
6148 (ins VecListThreeQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6149 def VLD3DUPqAsm_32 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr",
6150 (ins VecListThreeQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6152 def VLD3DUPdWB_fixed_Asm_8 :
6153 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr!",
6154 (ins VecListThreeDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6155 def VLD3DUPdWB_fixed_Asm_16 :
6156 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr!",
6157 (ins VecListThreeDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6158 def VLD3DUPdWB_fixed_Asm_32 :
6159 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr!",
6160 (ins VecListThreeDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6161 def VLD3DUPqWB_fixed_Asm_8 :
6162 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr!",
6163 (ins VecListThreeQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6164 def VLD3DUPqWB_fixed_Asm_16 :
6165 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr!",
6166 (ins VecListThreeQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6167 def VLD3DUPqWB_fixed_Asm_32 :
6168 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr!",
6169 (ins VecListThreeQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6170 def VLD3DUPdWB_register_Asm_8 :
6171 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr, $Rm",
6172 (ins VecListThreeDAllLanes:$list, addrmode6:$addr,
6173 rGPR:$Rm, pred:$p)>;
6174 def VLD3DUPdWB_register_Asm_16 :
6175 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr, $Rm",
6176 (ins VecListThreeDAllLanes:$list, addrmode6:$addr,
6177 rGPR:$Rm, pred:$p)>;
6178 def VLD3DUPdWB_register_Asm_32 :
6179 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr, $Rm",
6180 (ins VecListThreeDAllLanes:$list, addrmode6:$addr,
6181 rGPR:$Rm, pred:$p)>;
6182 def VLD3DUPqWB_register_Asm_8 :
6183 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr, $Rm",
6184 (ins VecListThreeQAllLanes:$list, addrmode6:$addr,
6185 rGPR:$Rm, pred:$p)>;
6186 def VLD3DUPqWB_register_Asm_16 :
6187 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr, $Rm",
6188 (ins VecListThreeQAllLanes:$list, addrmode6:$addr,
6189 rGPR:$Rm, pred:$p)>;
6190 def VLD3DUPqWB_register_Asm_32 :
6191 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr, $Rm",
6192 (ins VecListThreeQAllLanes:$list, addrmode6:$addr,
6193 rGPR:$Rm, pred:$p)>;
6196 // VLD3 single-lane pseudo-instructions. These need special handling for
6197 // the lane index that an InstAlias can't handle, so we use these instead.
6198 def VLD3LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr",
6199 (ins VecListThreeDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6200 def VLD3LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr",
6201 (ins VecListThreeDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6202 def VLD3LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr",
6203 (ins VecListThreeDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6204 def VLD3LNqAsm_16 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr",
6205 (ins VecListThreeQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6206 def VLD3LNqAsm_32 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr",
6207 (ins VecListThreeQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6209 def VLD3LNdWB_fixed_Asm_8 :
6210 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr!",
6211 (ins VecListThreeDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6212 def VLD3LNdWB_fixed_Asm_16 :
6213 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr!",
6214 (ins VecListThreeDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6215 def VLD3LNdWB_fixed_Asm_32 :
6216 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr!",
6217 (ins VecListThreeDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6218 def VLD3LNqWB_fixed_Asm_16 :
6219 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr!",
6220 (ins VecListThreeQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6221 def VLD3LNqWB_fixed_Asm_32 :
6222 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr!",
6223 (ins VecListThreeQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6224 def VLD3LNdWB_register_Asm_8 :
6225 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr, $Rm",
6226 (ins VecListThreeDByteIndexed:$list, addrmode6:$addr,
6227 rGPR:$Rm, pred:$p)>;
6228 def VLD3LNdWB_register_Asm_16 :
6229 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr, $Rm",
6230 (ins VecListThreeDHWordIndexed:$list, addrmode6:$addr,
6231 rGPR:$Rm, pred:$p)>;
6232 def VLD3LNdWB_register_Asm_32 :
6233 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr, $Rm",
6234 (ins VecListThreeDWordIndexed:$list, addrmode6:$addr,
6235 rGPR:$Rm, pred:$p)>;
6236 def VLD3LNqWB_register_Asm_16 :
6237 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr, $Rm",
6238 (ins VecListThreeQHWordIndexed:$list, addrmode6:$addr,
6239 rGPR:$Rm, pred:$p)>;
6240 def VLD3LNqWB_register_Asm_32 :
6241 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr, $Rm",
6242 (ins VecListThreeQWordIndexed:$list, addrmode6:$addr,
6243 rGPR:$Rm, pred:$p)>;
6245 // VLD3 multiple structure pseudo-instructions. These need special handling for
6246 // the vector operands that the normal instructions don't yet model.
6247 // FIXME: Remove these when the register classes and instructions are updated.
6248 def VLD3dAsm_8 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr",
6249 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6250 def VLD3dAsm_16 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr",
6251 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6252 def VLD3dAsm_32 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr",
6253 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6254 def VLD3qAsm_8 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr",
6255 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6256 def VLD3qAsm_16 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr",
6257 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6258 def VLD3qAsm_32 : NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr",
6259 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6261 def VLD3dWB_fixed_Asm_8 :
6262 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr!",
6263 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6264 def VLD3dWB_fixed_Asm_16 :
6265 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr!",
6266 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6267 def VLD3dWB_fixed_Asm_32 :
6268 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr!",
6269 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6270 def VLD3qWB_fixed_Asm_8 :
6271 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr!",
6272 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6273 def VLD3qWB_fixed_Asm_16 :
6274 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr!",
6275 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6276 def VLD3qWB_fixed_Asm_32 :
6277 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr!",
6278 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6279 def VLD3dWB_register_Asm_8 :
6280 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr, $Rm",
6281 (ins VecListThreeD:$list, addrmode6:$addr,
6282 rGPR:$Rm, pred:$p)>;
6283 def VLD3dWB_register_Asm_16 :
6284 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr, $Rm",
6285 (ins VecListThreeD:$list, addrmode6:$addr,
6286 rGPR:$Rm, pred:$p)>;
6287 def VLD3dWB_register_Asm_32 :
6288 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr, $Rm",
6289 (ins VecListThreeD:$list, addrmode6:$addr,
6290 rGPR:$Rm, pred:$p)>;
6291 def VLD3qWB_register_Asm_8 :
6292 NEONDataTypeAsmPseudoInst<"vld3${p}", ".8", "$list, $addr, $Rm",
6293 (ins VecListThreeQ:$list, addrmode6:$addr,
6294 rGPR:$Rm, pred:$p)>;
6295 def VLD3qWB_register_Asm_16 :
6296 NEONDataTypeAsmPseudoInst<"vld3${p}", ".16", "$list, $addr, $Rm",
6297 (ins VecListThreeQ:$list, addrmode6:$addr,
6298 rGPR:$Rm, pred:$p)>;
6299 def VLD3qWB_register_Asm_32 :
6300 NEONDataTypeAsmPseudoInst<"vld3${p}", ".32", "$list, $addr, $Rm",
6301 (ins VecListThreeQ:$list, addrmode6:$addr,
6302 rGPR:$Rm, pred:$p)>;
6304 // VST3 single-lane pseudo-instructions. These need special handling for
6305 // the lane index that an InstAlias can't handle, so we use these instead.
6306 def VST3LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr",
6307 (ins VecListThreeDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6308 def VST3LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr",
6309 (ins VecListThreeDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6310 def VST3LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr",
6311 (ins VecListThreeDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6312 def VST3LNqAsm_16 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr",
6313 (ins VecListThreeQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6314 def VST3LNqAsm_32 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr",
6315 (ins VecListThreeQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6317 def VST3LNdWB_fixed_Asm_8 :
6318 NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr!",
6319 (ins VecListThreeDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6320 def VST3LNdWB_fixed_Asm_16 :
6321 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr!",
6322 (ins VecListThreeDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6323 def VST3LNdWB_fixed_Asm_32 :
6324 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr!",
6325 (ins VecListThreeDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6326 def VST3LNqWB_fixed_Asm_16 :
6327 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr!",
6328 (ins VecListThreeQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6329 def VST3LNqWB_fixed_Asm_32 :
6330 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr!",
6331 (ins VecListThreeQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6332 def VST3LNdWB_register_Asm_8 :
6333 NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr, $Rm",
6334 (ins VecListThreeDByteIndexed:$list, addrmode6:$addr,
6335 rGPR:$Rm, pred:$p)>;
6336 def VST3LNdWB_register_Asm_16 :
6337 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr, $Rm",
6338 (ins VecListThreeDHWordIndexed:$list, addrmode6:$addr,
6339 rGPR:$Rm, pred:$p)>;
6340 def VST3LNdWB_register_Asm_32 :
6341 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr, $Rm",
6342 (ins VecListThreeDWordIndexed:$list, addrmode6:$addr,
6343 rGPR:$Rm, pred:$p)>;
6344 def VST3LNqWB_register_Asm_16 :
6345 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr, $Rm",
6346 (ins VecListThreeQHWordIndexed:$list, addrmode6:$addr,
6347 rGPR:$Rm, pred:$p)>;
6348 def VST3LNqWB_register_Asm_32 :
6349 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr, $Rm",
6350 (ins VecListThreeQWordIndexed:$list, addrmode6:$addr,
6351 rGPR:$Rm, pred:$p)>;
6354 // VST3 multiple structure pseudo-instructions. These need special handling for
6355 // the vector operands that the normal instructions don't yet model.
6356 // FIXME: Remove these when the register classes and instructions are updated.
6357 def VST3dAsm_8 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr",
6358 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6359 def VST3dAsm_16 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr",
6360 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6361 def VST3dAsm_32 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr",
6362 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6363 def VST3qAsm_8 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr",
6364 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6365 def VST3qAsm_16 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr",
6366 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6367 def VST3qAsm_32 : NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr",
6368 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6370 def VST3dWB_fixed_Asm_8 :
6371 NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr!",
6372 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6373 def VST3dWB_fixed_Asm_16 :
6374 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr!",
6375 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6376 def VST3dWB_fixed_Asm_32 :
6377 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr!",
6378 (ins VecListThreeD:$list, addrmode6:$addr, pred:$p)>;
6379 def VST3qWB_fixed_Asm_8 :
6380 NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr!",
6381 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6382 def VST3qWB_fixed_Asm_16 :
6383 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr!",
6384 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6385 def VST3qWB_fixed_Asm_32 :
6386 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr!",
6387 (ins VecListThreeQ:$list, addrmode6:$addr, pred:$p)>;
6388 def VST3dWB_register_Asm_8 :
6389 NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr, $Rm",
6390 (ins VecListThreeD:$list, addrmode6:$addr,
6391 rGPR:$Rm, pred:$p)>;
6392 def VST3dWB_register_Asm_16 :
6393 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr, $Rm",
6394 (ins VecListThreeD:$list, addrmode6:$addr,
6395 rGPR:$Rm, pred:$p)>;
6396 def VST3dWB_register_Asm_32 :
6397 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr, $Rm",
6398 (ins VecListThreeD:$list, addrmode6:$addr,
6399 rGPR:$Rm, pred:$p)>;
6400 def VST3qWB_register_Asm_8 :
6401 NEONDataTypeAsmPseudoInst<"vst3${p}", ".8", "$list, $addr, $Rm",
6402 (ins VecListThreeQ:$list, addrmode6:$addr,
6403 rGPR:$Rm, pred:$p)>;
6404 def VST3qWB_register_Asm_16 :
6405 NEONDataTypeAsmPseudoInst<"vst3${p}", ".16", "$list, $addr, $Rm",
6406 (ins VecListThreeQ:$list, addrmode6:$addr,
6407 rGPR:$Rm, pred:$p)>;
6408 def VST3qWB_register_Asm_32 :
6409 NEONDataTypeAsmPseudoInst<"vst3${p}", ".32", "$list, $addr, $Rm",
6410 (ins VecListThreeQ:$list, addrmode6:$addr,
6411 rGPR:$Rm, pred:$p)>;
6413 // VLD4 all-lanes pseudo-instructions. These need special handling for
6414 // the lane index that an InstAlias can't handle, so we use these instead.
6415 def VLD4DUPdAsm_8 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr",
6416 (ins VecListFourDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6417 def VLD4DUPdAsm_16 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr",
6418 (ins VecListFourDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6419 def VLD4DUPdAsm_32 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr",
6420 (ins VecListFourDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6421 def VLD4DUPqAsm_8 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr",
6422 (ins VecListFourQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6423 def VLD4DUPqAsm_16 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr",
6424 (ins VecListFourQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6425 def VLD4DUPqAsm_32 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr",
6426 (ins VecListFourQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6428 def VLD4DUPdWB_fixed_Asm_8 :
6429 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr!",
6430 (ins VecListFourDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6431 def VLD4DUPdWB_fixed_Asm_16 :
6432 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr!",
6433 (ins VecListFourDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6434 def VLD4DUPdWB_fixed_Asm_32 :
6435 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr!",
6436 (ins VecListFourDAllLanes:$list, addrmode6:$addr, pred:$p)>;
6437 def VLD4DUPqWB_fixed_Asm_8 :
6438 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr!",
6439 (ins VecListFourQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6440 def VLD4DUPqWB_fixed_Asm_16 :
6441 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr!",
6442 (ins VecListFourQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6443 def VLD4DUPqWB_fixed_Asm_32 :
6444 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr!",
6445 (ins VecListFourQAllLanes:$list, addrmode6:$addr, pred:$p)>;
6446 def VLD4DUPdWB_register_Asm_8 :
6447 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr, $Rm",
6448 (ins VecListFourDAllLanes:$list, addrmode6:$addr,
6449 rGPR:$Rm, pred:$p)>;
6450 def VLD4DUPdWB_register_Asm_16 :
6451 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr, $Rm",
6452 (ins VecListFourDAllLanes:$list, addrmode6:$addr,
6453 rGPR:$Rm, pred:$p)>;
6454 def VLD4DUPdWB_register_Asm_32 :
6455 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr, $Rm",
6456 (ins VecListFourDAllLanes:$list, addrmode6:$addr,
6457 rGPR:$Rm, pred:$p)>;
6458 def VLD4DUPqWB_register_Asm_8 :
6459 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr, $Rm",
6460 (ins VecListFourQAllLanes:$list, addrmode6:$addr,
6461 rGPR:$Rm, pred:$p)>;
6462 def VLD4DUPqWB_register_Asm_16 :
6463 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr, $Rm",
6464 (ins VecListFourQAllLanes:$list, addrmode6:$addr,
6465 rGPR:$Rm, pred:$p)>;
6466 def VLD4DUPqWB_register_Asm_32 :
6467 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr, $Rm",
6468 (ins VecListFourQAllLanes:$list, addrmode6:$addr,
6469 rGPR:$Rm, pred:$p)>;
6472 // VLD4 single-lane pseudo-instructions. These need special handling for
6473 // the lane index that an InstAlias can't handle, so we use these instead.
6474 def VLD4LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr",
6475 (ins VecListFourDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6476 def VLD4LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr",
6477 (ins VecListFourDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6478 def VLD4LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr",
6479 (ins VecListFourDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6480 def VLD4LNqAsm_16 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr",
6481 (ins VecListFourQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6482 def VLD4LNqAsm_32 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr",
6483 (ins VecListFourQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6485 def VLD4LNdWB_fixed_Asm_8 :
6486 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr!",
6487 (ins VecListFourDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6488 def VLD4LNdWB_fixed_Asm_16 :
6489 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr!",
6490 (ins VecListFourDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6491 def VLD4LNdWB_fixed_Asm_32 :
6492 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr!",
6493 (ins VecListFourDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6494 def VLD4LNqWB_fixed_Asm_16 :
6495 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr!",
6496 (ins VecListFourQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6497 def VLD4LNqWB_fixed_Asm_32 :
6498 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr!",
6499 (ins VecListFourQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6500 def VLD4LNdWB_register_Asm_8 :
6501 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr, $Rm",
6502 (ins VecListFourDByteIndexed:$list, addrmode6:$addr,
6503 rGPR:$Rm, pred:$p)>;
6504 def VLD4LNdWB_register_Asm_16 :
6505 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr, $Rm",
6506 (ins VecListFourDHWordIndexed:$list, addrmode6:$addr,
6507 rGPR:$Rm, pred:$p)>;
6508 def VLD4LNdWB_register_Asm_32 :
6509 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr, $Rm",
6510 (ins VecListFourDWordIndexed:$list, addrmode6:$addr,
6511 rGPR:$Rm, pred:$p)>;
6512 def VLD4LNqWB_register_Asm_16 :
6513 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr, $Rm",
6514 (ins VecListFourQHWordIndexed:$list, addrmode6:$addr,
6515 rGPR:$Rm, pred:$p)>;
6516 def VLD4LNqWB_register_Asm_32 :
6517 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr, $Rm",
6518 (ins VecListFourQWordIndexed:$list, addrmode6:$addr,
6519 rGPR:$Rm, pred:$p)>;
6523 // VLD4 multiple structure pseudo-instructions. These need special handling for
6524 // the vector operands that the normal instructions don't yet model.
6525 // FIXME: Remove these when the register classes and instructions are updated.
6526 def VLD4dAsm_8 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr",
6527 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6528 def VLD4dAsm_16 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr",
6529 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6530 def VLD4dAsm_32 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr",
6531 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6532 def VLD4qAsm_8 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr",
6533 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6534 def VLD4qAsm_16 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr",
6535 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6536 def VLD4qAsm_32 : NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr",
6537 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6539 def VLD4dWB_fixed_Asm_8 :
6540 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr!",
6541 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6542 def VLD4dWB_fixed_Asm_16 :
6543 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr!",
6544 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6545 def VLD4dWB_fixed_Asm_32 :
6546 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr!",
6547 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6548 def VLD4qWB_fixed_Asm_8 :
6549 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr!",
6550 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6551 def VLD4qWB_fixed_Asm_16 :
6552 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr!",
6553 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6554 def VLD4qWB_fixed_Asm_32 :
6555 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr!",
6556 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6557 def VLD4dWB_register_Asm_8 :
6558 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr, $Rm",
6559 (ins VecListFourD:$list, addrmode6:$addr,
6560 rGPR:$Rm, pred:$p)>;
6561 def VLD4dWB_register_Asm_16 :
6562 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr, $Rm",
6563 (ins VecListFourD:$list, addrmode6:$addr,
6564 rGPR:$Rm, pred:$p)>;
6565 def VLD4dWB_register_Asm_32 :
6566 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr, $Rm",
6567 (ins VecListFourD:$list, addrmode6:$addr,
6568 rGPR:$Rm, pred:$p)>;
6569 def VLD4qWB_register_Asm_8 :
6570 NEONDataTypeAsmPseudoInst<"vld4${p}", ".8", "$list, $addr, $Rm",
6571 (ins VecListFourQ:$list, addrmode6:$addr,
6572 rGPR:$Rm, pred:$p)>;
6573 def VLD4qWB_register_Asm_16 :
6574 NEONDataTypeAsmPseudoInst<"vld4${p}", ".16", "$list, $addr, $Rm",
6575 (ins VecListFourQ:$list, addrmode6:$addr,
6576 rGPR:$Rm, pred:$p)>;
6577 def VLD4qWB_register_Asm_32 :
6578 NEONDataTypeAsmPseudoInst<"vld4${p}", ".32", "$list, $addr, $Rm",
6579 (ins VecListFourQ:$list, addrmode6:$addr,
6580 rGPR:$Rm, pred:$p)>;
6582 // VST4 single-lane pseudo-instructions. These need special handling for
6583 // the lane index that an InstAlias can't handle, so we use these instead.
6584 def VST4LNdAsm_8 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr",
6585 (ins VecListFourDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6586 def VST4LNdAsm_16 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr",
6587 (ins VecListFourDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6588 def VST4LNdAsm_32 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr",
6589 (ins VecListFourDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6590 def VST4LNqAsm_16 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr",
6591 (ins VecListFourQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6592 def VST4LNqAsm_32 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr",
6593 (ins VecListFourQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6595 def VST4LNdWB_fixed_Asm_8 :
6596 NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr!",
6597 (ins VecListFourDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
6598 def VST4LNdWB_fixed_Asm_16 :
6599 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr!",
6600 (ins VecListFourDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6601 def VST4LNdWB_fixed_Asm_32 :
6602 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr!",
6603 (ins VecListFourDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6604 def VST4LNqWB_fixed_Asm_16 :
6605 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr!",
6606 (ins VecListFourQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6607 def VST4LNqWB_fixed_Asm_32 :
6608 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr!",
6609 (ins VecListFourQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
6610 def VST4LNdWB_register_Asm_8 :
6611 NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr, $Rm",
6612 (ins VecListFourDByteIndexed:$list, addrmode6:$addr,
6613 rGPR:$Rm, pred:$p)>;
6614 def VST4LNdWB_register_Asm_16 :
6615 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr, $Rm",
6616 (ins VecListFourDHWordIndexed:$list, addrmode6:$addr,
6617 rGPR:$Rm, pred:$p)>;
6618 def VST4LNdWB_register_Asm_32 :
6619 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr, $Rm",
6620 (ins VecListFourDWordIndexed:$list, addrmode6:$addr,
6621 rGPR:$Rm, pred:$p)>;
6622 def VST4LNqWB_register_Asm_16 :
6623 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr, $Rm",
6624 (ins VecListFourQHWordIndexed:$list, addrmode6:$addr,
6625 rGPR:$Rm, pred:$p)>;
6626 def VST4LNqWB_register_Asm_32 :
6627 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr, $Rm",
6628 (ins VecListFourQWordIndexed:$list, addrmode6:$addr,
6629 rGPR:$Rm, pred:$p)>;
6632 // VST4 multiple structure pseudo-instructions. These need special handling for
6633 // the vector operands that the normal instructions don't yet model.
6634 // FIXME: Remove these when the register classes and instructions are updated.
6635 def VST4dAsm_8 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr",
6636 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6637 def VST4dAsm_16 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr",
6638 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6639 def VST4dAsm_32 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr",
6640 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6641 def VST4qAsm_8 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr",
6642 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6643 def VST4qAsm_16 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr",
6644 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6645 def VST4qAsm_32 : NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr",
6646 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6648 def VST4dWB_fixed_Asm_8 :
6649 NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr!",
6650 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6651 def VST4dWB_fixed_Asm_16 :
6652 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr!",
6653 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6654 def VST4dWB_fixed_Asm_32 :
6655 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr!",
6656 (ins VecListFourD:$list, addrmode6:$addr, pred:$p)>;
6657 def VST4qWB_fixed_Asm_8 :
6658 NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr!",
6659 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6660 def VST4qWB_fixed_Asm_16 :
6661 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr!",
6662 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6663 def VST4qWB_fixed_Asm_32 :
6664 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr!",
6665 (ins VecListFourQ:$list, addrmode6:$addr, pred:$p)>;
6666 def VST4dWB_register_Asm_8 :
6667 NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr, $Rm",
6668 (ins VecListFourD:$list, addrmode6:$addr,
6669 rGPR:$Rm, pred:$p)>;
6670 def VST4dWB_register_Asm_16 :
6671 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr, $Rm",
6672 (ins VecListFourD:$list, addrmode6:$addr,
6673 rGPR:$Rm, pred:$p)>;
6674 def VST4dWB_register_Asm_32 :
6675 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr, $Rm",
6676 (ins VecListFourD:$list, addrmode6:$addr,
6677 rGPR:$Rm, pred:$p)>;
6678 def VST4qWB_register_Asm_8 :
6679 NEONDataTypeAsmPseudoInst<"vst4${p}", ".8", "$list, $addr, $Rm",
6680 (ins VecListFourQ:$list, addrmode6:$addr,
6681 rGPR:$Rm, pred:$p)>;
6682 def VST4qWB_register_Asm_16 :
6683 NEONDataTypeAsmPseudoInst<"vst4${p}", ".16", "$list, $addr, $Rm",
6684 (ins VecListFourQ:$list, addrmode6:$addr,
6685 rGPR:$Rm, pred:$p)>;
6686 def VST4qWB_register_Asm_32 :
6687 NEONDataTypeAsmPseudoInst<"vst4${p}", ".32", "$list, $addr, $Rm",
6688 (ins VecListFourQ:$list, addrmode6:$addr,
6689 rGPR:$Rm, pred:$p)>;
6691 // VMOV takes an optional datatype suffix
6692 defm : NEONDTAnyInstAlias<"vmov${p}", "$Vd, $Vm",
6693 (VORRd DPR:$Vd, DPR:$Vm, DPR:$Vm, pred:$p)>;
6694 defm : NEONDTAnyInstAlias<"vmov${p}", "$Vd, $Vm",
6695 (VORRq QPR:$Vd, QPR:$Vm, QPR:$Vm, pred:$p)>;
6697 // VCLT (register) is an assembler alias for VCGT w/ the operands reversed.
6698 // D-register versions.
6699 def : NEONInstAlias<"vcle${p}.s8 $Dd, $Dn, $Dm",
6700 (VCGEsv8i8 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6701 def : NEONInstAlias<"vcle${p}.s16 $Dd, $Dn, $Dm",
6702 (VCGEsv4i16 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6703 def : NEONInstAlias<"vcle${p}.s32 $Dd, $Dn, $Dm",
6704 (VCGEsv2i32 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6705 def : NEONInstAlias<"vcle${p}.u8 $Dd, $Dn, $Dm",
6706 (VCGEuv8i8 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6707 def : NEONInstAlias<"vcle${p}.u16 $Dd, $Dn, $Dm",
6708 (VCGEuv4i16 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6709 def : NEONInstAlias<"vcle${p}.u32 $Dd, $Dn, $Dm",
6710 (VCGEuv2i32 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6711 def : NEONInstAlias<"vcle${p}.f32 $Dd, $Dn, $Dm",
6712 (VCGEfd DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6713 // Q-register versions.
6714 def : NEONInstAlias<"vcle${p}.s8 $Qd, $Qn, $Qm",
6715 (VCGEsv16i8 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6716 def : NEONInstAlias<"vcle${p}.s16 $Qd, $Qn, $Qm",
6717 (VCGEsv8i16 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6718 def : NEONInstAlias<"vcle${p}.s32 $Qd, $Qn, $Qm",
6719 (VCGEsv4i32 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6720 def : NEONInstAlias<"vcle${p}.u8 $Qd, $Qn, $Qm",
6721 (VCGEuv16i8 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6722 def : NEONInstAlias<"vcle${p}.u16 $Qd, $Qn, $Qm",
6723 (VCGEuv8i16 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6724 def : NEONInstAlias<"vcle${p}.u32 $Qd, $Qn, $Qm",
6725 (VCGEuv4i32 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6726 def : NEONInstAlias<"vcle${p}.f32 $Qd, $Qn, $Qm",
6727 (VCGEfq QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6729 // VCLT (register) is an assembler alias for VCGT w/ the operands reversed.
6730 // D-register versions.
6731 def : NEONInstAlias<"vclt${p}.s8 $Dd, $Dn, $Dm",
6732 (VCGTsv8i8 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6733 def : NEONInstAlias<"vclt${p}.s16 $Dd, $Dn, $Dm",
6734 (VCGTsv4i16 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6735 def : NEONInstAlias<"vclt${p}.s32 $Dd, $Dn, $Dm",
6736 (VCGTsv2i32 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6737 def : NEONInstAlias<"vclt${p}.u8 $Dd, $Dn, $Dm",
6738 (VCGTuv8i8 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6739 def : NEONInstAlias<"vclt${p}.u16 $Dd, $Dn, $Dm",
6740 (VCGTuv4i16 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6741 def : NEONInstAlias<"vclt${p}.u32 $Dd, $Dn, $Dm",
6742 (VCGTuv2i32 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6743 def : NEONInstAlias<"vclt${p}.f32 $Dd, $Dn, $Dm",
6744 (VCGTfd DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
6745 // Q-register versions.
6746 def : NEONInstAlias<"vclt${p}.s8 $Qd, $Qn, $Qm",
6747 (VCGTsv16i8 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6748 def : NEONInstAlias<"vclt${p}.s16 $Qd, $Qn, $Qm",
6749 (VCGTsv8i16 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6750 def : NEONInstAlias<"vclt${p}.s32 $Qd, $Qn, $Qm",
6751 (VCGTsv4i32 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6752 def : NEONInstAlias<"vclt${p}.u8 $Qd, $Qn, $Qm",
6753 (VCGTuv16i8 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6754 def : NEONInstAlias<"vclt${p}.u16 $Qd, $Qn, $Qm",
6755 (VCGTuv8i16 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6756 def : NEONInstAlias<"vclt${p}.u32 $Qd, $Qn, $Qm",
6757 (VCGTuv4i32 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6758 def : NEONInstAlias<"vclt${p}.f32 $Qd, $Qn, $Qm",
6759 (VCGTfq QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
6761 // Two-operand variants for VEXT
6762 def : NEONInstAlias<"vext${p}.8 $Vdn, $Vm, $imm",
6763 (VEXTd8 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, imm0_7:$imm, pred:$p)>;
6764 def : NEONInstAlias<"vext${p}.16 $Vdn, $Vm, $imm",
6765 (VEXTd16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, imm0_3:$imm, pred:$p)>;
6766 def : NEONInstAlias<"vext${p}.32 $Vdn, $Vm, $imm",
6767 (VEXTd32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, imm0_1:$imm, pred:$p)>;
6769 def : NEONInstAlias<"vext${p}.8 $Vdn, $Vm, $imm",
6770 (VEXTq8 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, imm0_15:$imm, pred:$p)>;
6771 def : NEONInstAlias<"vext${p}.16 $Vdn, $Vm, $imm",
6772 (VEXTq16 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, imm0_7:$imm, pred:$p)>;
6773 def : NEONInstAlias<"vext${p}.32 $Vdn, $Vm, $imm",
6774 (VEXTq32 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, imm0_3:$imm, pred:$p)>;
6775 def : NEONInstAlias<"vext${p}.64 $Vdn, $Vm, $imm",
6776 (VEXTq64 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, imm0_1:$imm, pred:$p)>;
6778 // Two-operand variants for VQDMULH
6779 def : NEONInstAlias<"vqdmulh${p}.s16 $Vdn, $Vm",
6780 (VQDMULHv4i16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
6781 def : NEONInstAlias<"vqdmulh${p}.s32 $Vdn, $Vm",
6782 (VQDMULHv2i32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
6784 def : NEONInstAlias<"vqdmulh${p}.s16 $Vdn, $Vm",
6785 (VQDMULHv8i16 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
6786 def : NEONInstAlias<"vqdmulh${p}.s32 $Vdn, $Vm",
6787 (VQDMULHv4i32 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
6789 // Two-operand variants for VMAX.
6790 def : NEONInstAlias<"vmax${p}.s8 $Vdn, $Vm",
6791 (VMAXsv8i8 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
6792 def : NEONInstAlias<"vmax${p}.s16 $Vdn, $Vm",
6793 (VMAXsv4i16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
6794 def : NEONInstAlias<"vmax${p}.s32 $Vdn, $Vm",
6795 (VMAXsv2i32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
6796 def : NEONInstAlias<"vmax${p}.u8 $Vdn, $Vm",
6797 (VMAXuv8i8 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
6798 def : NEONInstAlias<"vmax${p}.u16 $Vdn, $Vm",
6799 (VMAXuv4i16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
6800 def : NEONInstAlias<"vmax${p}.u32 $Vdn, $Vm",
6801 (VMAXuv2i32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
6802 def : NEONInstAlias<"vmax${p}.f32 $Vdn, $Vm",
6803 (VMAXfd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
6805 def : NEONInstAlias<"vmax${p}.s8 $Vdn, $Vm",
6806 (VMAXsv16i8 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
6807 def : NEONInstAlias<"vmax${p}.s16 $Vdn, $Vm",
6808 (VMAXsv8i16 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
6809 def : NEONInstAlias<"vmax${p}.s32 $Vdn, $Vm",
6810 (VMAXsv4i32 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
6811 def : NEONInstAlias<"vmax${p}.u8 $Vdn, $Vm",
6812 (VMAXuv16i8 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
6813 def : NEONInstAlias<"vmax${p}.u16 $Vdn, $Vm",
6814 (VMAXuv8i16 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
6815 def : NEONInstAlias<"vmax${p}.u32 $Vdn, $Vm",
6816 (VMAXuv4i32 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
6817 def : NEONInstAlias<"vmax${p}.f32 $Vdn, $Vm",
6818 (VMAXfq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
6820 // Two-operand variants for VMIN.
6821 def : NEONInstAlias<"vmin${p}.s8 $Vdn, $Vm",
6822 (VMINsv8i8 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
6823 def : NEONInstAlias<"vmin${p}.s16 $Vdn, $Vm",
6824 (VMINsv4i16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
6825 def : NEONInstAlias<"vmin${p}.s32 $Vdn, $Vm",
6826 (VMINsv2i32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
6827 def : NEONInstAlias<"vmin${p}.u8 $Vdn, $Vm",
6828 (VMINuv8i8 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
6829 def : NEONInstAlias<"vmin${p}.u16 $Vdn, $Vm",
6830 (VMINuv4i16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
6831 def : NEONInstAlias<"vmin${p}.u32 $Vdn, $Vm",
6832 (VMINuv2i32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
6833 def : NEONInstAlias<"vmin${p}.f32 $Vdn, $Vm",
6834 (VMINfd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
6836 def : NEONInstAlias<"vmin${p}.s8 $Vdn, $Vm",
6837 (VMINsv16i8 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
6838 def : NEONInstAlias<"vmin${p}.s16 $Vdn, $Vm",
6839 (VMINsv8i16 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
6840 def : NEONInstAlias<"vmin${p}.s32 $Vdn, $Vm",
6841 (VMINsv4i32 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
6842 def : NEONInstAlias<"vmin${p}.u8 $Vdn, $Vm",
6843 (VMINuv16i8 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
6844 def : NEONInstAlias<"vmin${p}.u16 $Vdn, $Vm",
6845 (VMINuv8i16 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
6846 def : NEONInstAlias<"vmin${p}.u32 $Vdn, $Vm",
6847 (VMINuv4i32 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
6848 def : NEONInstAlias<"vmin${p}.f32 $Vdn, $Vm",
6849 (VMINfq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
6851 // Two-operand variants for VPADD.
6852 def : NEONInstAlias<"vpadd${p}.i8 $Vdn, $Vm",
6853 (VPADDi8 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
6854 def : NEONInstAlias<"vpadd${p}.i16 $Vdn, $Vm",
6855 (VPADDi16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
6856 def : NEONInstAlias<"vpadd${p}.i32 $Vdn, $Vm",
6857 (VPADDi32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
6858 def : NEONInstAlias<"vpadd${p}.f32 $Vdn, $Vm",
6859 (VPADDf DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
6861 // Two-operand variants for VSRA.
6863 def : NEONInstAlias<"vsra${p}.s8 $Vdm, $imm",
6864 (VSRAsv8i8 DPR:$Vdm, DPR:$Vdm, shr_imm8:$imm, pred:$p)>;
6865 def : NEONInstAlias<"vsra${p}.s16 $Vdm, $imm",
6866 (VSRAsv4i16 DPR:$Vdm, DPR:$Vdm, shr_imm16:$imm, pred:$p)>;
6867 def : NEONInstAlias<"vsra${p}.s32 $Vdm, $imm",
6868 (VSRAsv2i32 DPR:$Vdm, DPR:$Vdm, shr_imm32:$imm, pred:$p)>;
6869 def : NEONInstAlias<"vsra${p}.s64 $Vdm, $imm",
6870 (VSRAsv1i64 DPR:$Vdm, DPR:$Vdm, shr_imm64:$imm, pred:$p)>;
6872 def : NEONInstAlias<"vsra${p}.s8 $Vdm, $imm",
6873 (VSRAsv16i8 QPR:$Vdm, QPR:$Vdm, shr_imm8:$imm, pred:$p)>;
6874 def : NEONInstAlias<"vsra${p}.s16 $Vdm, $imm",
6875 (VSRAsv8i16 QPR:$Vdm, QPR:$Vdm, shr_imm16:$imm, pred:$p)>;
6876 def : NEONInstAlias<"vsra${p}.s32 $Vdm, $imm",
6877 (VSRAsv4i32 QPR:$Vdm, QPR:$Vdm, shr_imm32:$imm, pred:$p)>;
6878 def : NEONInstAlias<"vsra${p}.s64 $Vdm, $imm",
6879 (VSRAsv2i64 QPR:$Vdm, QPR:$Vdm, shr_imm64:$imm, pred:$p)>;
6882 def : NEONInstAlias<"vsra${p}.u8 $Vdm, $imm",
6883 (VSRAuv8i8 DPR:$Vdm, DPR:$Vdm, shr_imm8:$imm, pred:$p)>;
6884 def : NEONInstAlias<"vsra${p}.u16 $Vdm, $imm",
6885 (VSRAuv4i16 DPR:$Vdm, DPR:$Vdm, shr_imm16:$imm, pred:$p)>;
6886 def : NEONInstAlias<"vsra${p}.u32 $Vdm, $imm",
6887 (VSRAuv2i32 DPR:$Vdm, DPR:$Vdm, shr_imm32:$imm, pred:$p)>;
6888 def : NEONInstAlias<"vsra${p}.u64 $Vdm, $imm",
6889 (VSRAuv1i64 DPR:$Vdm, DPR:$Vdm, shr_imm64:$imm, pred:$p)>;
6891 def : NEONInstAlias<"vsra${p}.u8 $Vdm, $imm",
6892 (VSRAuv16i8 QPR:$Vdm, QPR:$Vdm, shr_imm8:$imm, pred:$p)>;
6893 def : NEONInstAlias<"vsra${p}.u16 $Vdm, $imm",
6894 (VSRAuv8i16 QPR:$Vdm, QPR:$Vdm, shr_imm16:$imm, pred:$p)>;
6895 def : NEONInstAlias<"vsra${p}.u32 $Vdm, $imm",
6896 (VSRAuv4i32 QPR:$Vdm, QPR:$Vdm, shr_imm32:$imm, pred:$p)>;
6897 def : NEONInstAlias<"vsra${p}.u64 $Vdm, $imm",
6898 (VSRAuv2i64 QPR:$Vdm, QPR:$Vdm, shr_imm64:$imm, pred:$p)>;
6900 // Two-operand variants for VSRI.
6901 def : NEONInstAlias<"vsri${p}.8 $Vdm, $imm",
6902 (VSRIv8i8 DPR:$Vdm, DPR:$Vdm, shr_imm8:$imm, pred:$p)>;
6903 def : NEONInstAlias<"vsri${p}.16 $Vdm, $imm",
6904 (VSRIv4i16 DPR:$Vdm, DPR:$Vdm, shr_imm16:$imm, pred:$p)>;
6905 def : NEONInstAlias<"vsri${p}.32 $Vdm, $imm",
6906 (VSRIv2i32 DPR:$Vdm, DPR:$Vdm, shr_imm32:$imm, pred:$p)>;
6907 def : NEONInstAlias<"vsri${p}.64 $Vdm, $imm",
6908 (VSRIv1i64 DPR:$Vdm, DPR:$Vdm, shr_imm64:$imm, pred:$p)>;
6910 def : NEONInstAlias<"vsri${p}.8 $Vdm, $imm",
6911 (VSRIv16i8 QPR:$Vdm, QPR:$Vdm, shr_imm8:$imm, pred:$p)>;
6912 def : NEONInstAlias<"vsri${p}.16 $Vdm, $imm",
6913 (VSRIv8i16 QPR:$Vdm, QPR:$Vdm, shr_imm16:$imm, pred:$p)>;
6914 def : NEONInstAlias<"vsri${p}.32 $Vdm, $imm",
6915 (VSRIv4i32 QPR:$Vdm, QPR:$Vdm, shr_imm32:$imm, pred:$p)>;
6916 def : NEONInstAlias<"vsri${p}.64 $Vdm, $imm",
6917 (VSRIv2i64 QPR:$Vdm, QPR:$Vdm, shr_imm64:$imm, pred:$p)>;
6919 // Two-operand variants for VSLI.
6920 def : NEONInstAlias<"vsli${p}.8 $Vdm, $imm",
6921 (VSLIv8i8 DPR:$Vdm, DPR:$Vdm, shr_imm8:$imm, pred:$p)>;
6922 def : NEONInstAlias<"vsli${p}.16 $Vdm, $imm",
6923 (VSLIv4i16 DPR:$Vdm, DPR:$Vdm, shr_imm16:$imm, pred:$p)>;
6924 def : NEONInstAlias<"vsli${p}.32 $Vdm, $imm",
6925 (VSLIv2i32 DPR:$Vdm, DPR:$Vdm, shr_imm32:$imm, pred:$p)>;
6926 def : NEONInstAlias<"vsli${p}.64 $Vdm, $imm",
6927 (VSLIv1i64 DPR:$Vdm, DPR:$Vdm, shr_imm64:$imm, pred:$p)>;
6929 def : NEONInstAlias<"vsli${p}.8 $Vdm, $imm",
6930 (VSLIv16i8 QPR:$Vdm, QPR:$Vdm, shr_imm8:$imm, pred:$p)>;
6931 def : NEONInstAlias<"vsli${p}.16 $Vdm, $imm",
6932 (VSLIv8i16 QPR:$Vdm, QPR:$Vdm, shr_imm16:$imm, pred:$p)>;
6933 def : NEONInstAlias<"vsli${p}.32 $Vdm, $imm",
6934 (VSLIv4i32 QPR:$Vdm, QPR:$Vdm, shr_imm32:$imm, pred:$p)>;
6935 def : NEONInstAlias<"vsli${p}.64 $Vdm, $imm",
6936 (VSLIv2i64 QPR:$Vdm, QPR:$Vdm, shr_imm64:$imm, pred:$p)>;
6938 // VSWP allows, but does not require, a type suffix.
6939 defm : NEONDTAnyInstAlias<"vswp${p}", "$Vd, $Vm",
6940 (VSWPd DPR:$Vd, DPR:$Vm, pred:$p)>;
6941 defm : NEONDTAnyInstAlias<"vswp${p}", "$Vd, $Vm",
6942 (VSWPq QPR:$Vd, QPR:$Vm, pred:$p)>;
6944 // VBIF, VBIT, and VBSL allow, but do not require, a type suffix.
6945 defm : NEONDTAnyInstAlias<"vbif${p}", "$Vd, $Vn, $Vm",
6946 (VBIFd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
6947 defm : NEONDTAnyInstAlias<"vbit${p}", "$Vd, $Vn, $Vm",
6948 (VBITd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
6949 defm : NEONDTAnyInstAlias<"vbsl${p}", "$Vd, $Vn, $Vm",
6950 (VBSLd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
6951 defm : NEONDTAnyInstAlias<"vbif${p}", "$Vd, $Vn, $Vm",
6952 (VBIFq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
6953 defm : NEONDTAnyInstAlias<"vbit${p}", "$Vd, $Vn, $Vm",
6954 (VBITq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
6955 defm : NEONDTAnyInstAlias<"vbsl${p}", "$Vd, $Vn, $Vm",
6956 (VBSLq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
6958 // "vmov Rd, #-imm" can be handled via "vmvn".
6959 def : NEONInstAlias<"vmov${p}.i32 $Vd, $imm",
6960 (VMVNv2i32 DPR:$Vd, nImmVMOVI32Neg:$imm, pred:$p)>;
6961 def : NEONInstAlias<"vmov${p}.i32 $Vd, $imm",
6962 (VMVNv4i32 QPR:$Vd, nImmVMOVI32Neg:$imm, pred:$p)>;
6963 def : NEONInstAlias<"vmvn${p}.i32 $Vd, $imm",
6964 (VMOVv2i32 DPR:$Vd, nImmVMOVI32Neg:$imm, pred:$p)>;
6965 def : NEONInstAlias<"vmvn${p}.i32 $Vd, $imm",
6966 (VMOVv4i32 QPR:$Vd, nImmVMOVI32Neg:$imm, pred:$p)>;
6968 // 'gas' compatibility aliases for quad-word instructions. Strictly speaking,
6969 // these should restrict to just the Q register variants, but the register
6970 // classes are enough to match correctly regardless, so we keep it simple
6971 // and just use MnemonicAlias.
6972 def : NEONMnemonicAlias<"vbicq", "vbic">;
6973 def : NEONMnemonicAlias<"vandq", "vand">;
6974 def : NEONMnemonicAlias<"veorq", "veor">;
6975 def : NEONMnemonicAlias<"vorrq", "vorr">;
6977 def : NEONMnemonicAlias<"vmovq", "vmov">;
6978 def : NEONMnemonicAlias<"vmvnq", "vmvn">;
6979 // Explicit versions for floating point so that the FPImm variants get
6980 // handled early. The parser gets confused otherwise.
6981 def : NEONMnemonicAlias<"vmovq.f32", "vmov.f32">;
6982 def : NEONMnemonicAlias<"vmovq.f64", "vmov.f64">;
6984 def : NEONMnemonicAlias<"vaddq", "vadd">;
6985 def : NEONMnemonicAlias<"vsubq", "vsub">;
6987 def : NEONMnemonicAlias<"vminq", "vmin">;
6988 def : NEONMnemonicAlias<"vmaxq", "vmax">;
6990 def : NEONMnemonicAlias<"vmulq", "vmul">;
6992 def : NEONMnemonicAlias<"vabsq", "vabs">;
6994 def : NEONMnemonicAlias<"vshlq", "vshl">;
6995 def : NEONMnemonicAlias<"vshrq", "vshr">;
6997 def : NEONMnemonicAlias<"vcvtq", "vcvt">;
6999 def : NEONMnemonicAlias<"vcleq", "vcle">;
7000 def : NEONMnemonicAlias<"vceqq", "vceq">;
7002 def : NEONMnemonicAlias<"vzipq", "vzip">;
7003 def : NEONMnemonicAlias<"vswpq", "vswp">;
7005 def : NEONMnemonicAlias<"vrecpeq.f32", "vrecpe.f32">;
7006 def : NEONMnemonicAlias<"vrecpeq.u32", "vrecpe.u32">;
7009 // Alias for loading floating point immediates that aren't representable
7010 // using the vmov.f32 encoding but the bitpattern is representable using
7011 // the .i32 encoding.
7012 def : NEONInstAlias<"vmov${p}.f32 $Vd, $imm",
7013 (VMOVv4i32 QPR:$Vd, nImmVMOVI32:$imm, pred:$p)>;
7014 def : NEONInstAlias<"vmov${p}.f32 $Vd, $imm",
7015 (VMOVv2i32 DPR:$Vd, nImmVMOVI32:$imm, pred:$p)>;