1 //===- ARMInstrNEON.td - NEON support for ARM -----------------------------===//
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 VecListTwoDAsmOperand : AsmOperandClass {
93 let Name = "VecListTwoD";
94 let ParserMethod = "parseVectorList";
95 let RenderMethod = "addVecListOperands";
97 def VecListTwoD : RegisterOperand<DPR, "printVectorListTwo"> {
98 let ParserMatchClass = VecListTwoDAsmOperand;
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;
128 // Register list of one D register, with "all lanes" subscripting.
129 def VecListOneDAllLanesAsmOperand : AsmOperandClass {
130 let Name = "VecListOneDAllLanes";
131 let ParserMethod = "parseVectorList";
132 let RenderMethod = "addVecListOperands";
134 def VecListOneDAllLanes : RegisterOperand<DPR, "printVectorListOneAllLanes"> {
135 let ParserMatchClass = VecListOneDAllLanesAsmOperand;
137 // Register list of two D registers, with "all lanes" subscripting.
138 def VecListTwoDAllLanesAsmOperand : AsmOperandClass {
139 let Name = "VecListTwoDAllLanes";
140 let ParserMethod = "parseVectorList";
141 let RenderMethod = "addVecListOperands";
143 def VecListTwoDAllLanes : RegisterOperand<DPR, "printVectorListTwoAllLanes"> {
144 let ParserMatchClass = VecListTwoDAllLanesAsmOperand;
146 // Register list of two D registers spaced by 2 (two sequential Q registers).
147 def VecListTwoQAllLanesAsmOperand : AsmOperandClass {
148 let Name = "VecListTwoQAllLanes";
149 let ParserMethod = "parseVectorList";
150 let RenderMethod = "addVecListOperands";
152 def VecListTwoQAllLanes : RegisterOperand<DPR,
153 "printVectorListTwoSpacedAllLanes"> {
154 let ParserMatchClass = VecListTwoQAllLanesAsmOperand;
157 // Register list of one D register, with byte lane subscripting.
158 def VecListOneDByteIndexAsmOperand : AsmOperandClass {
159 let Name = "VecListOneDByteIndexed";
160 let ParserMethod = "parseVectorList";
161 let RenderMethod = "addVecListIndexedOperands";
163 def VecListOneDByteIndexed : Operand<i32> {
164 let ParserMatchClass = VecListOneDByteIndexAsmOperand;
165 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
167 // ...with half-word lane subscripting.
168 def VecListOneDHWordIndexAsmOperand : AsmOperandClass {
169 let Name = "VecListOneDHWordIndexed";
170 let ParserMethod = "parseVectorList";
171 let RenderMethod = "addVecListIndexedOperands";
173 def VecListOneDHWordIndexed : Operand<i32> {
174 let ParserMatchClass = VecListOneDHWordIndexAsmOperand;
175 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
177 // ...with word lane subscripting.
178 def VecListOneDWordIndexAsmOperand : AsmOperandClass {
179 let Name = "VecListOneDWordIndexed";
180 let ParserMethod = "parseVectorList";
181 let RenderMethod = "addVecListIndexedOperands";
183 def VecListOneDWordIndexed : Operand<i32> {
184 let ParserMatchClass = VecListOneDWordIndexAsmOperand;
185 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
187 // Register list of two D registers with byte lane subscripting.
188 def VecListTwoDByteIndexAsmOperand : AsmOperandClass {
189 let Name = "VecListTwoDByteIndexed";
190 let ParserMethod = "parseVectorList";
191 let RenderMethod = "addVecListIndexedOperands";
193 def VecListTwoDByteIndexed : Operand<i32> {
194 let ParserMatchClass = VecListTwoDByteIndexAsmOperand;
195 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
197 // ...with half-word lane subscripting.
198 def VecListTwoDHWordIndexAsmOperand : AsmOperandClass {
199 let Name = "VecListTwoDHWordIndexed";
200 let ParserMethod = "parseVectorList";
201 let RenderMethod = "addVecListIndexedOperands";
203 def VecListTwoDHWordIndexed : Operand<i32> {
204 let ParserMatchClass = VecListTwoDHWordIndexAsmOperand;
205 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
207 // ...with word lane subscripting.
208 def VecListTwoDWordIndexAsmOperand : AsmOperandClass {
209 let Name = "VecListTwoDWordIndexed";
210 let ParserMethod = "parseVectorList";
211 let RenderMethod = "addVecListIndexedOperands";
213 def VecListTwoDWordIndexed : Operand<i32> {
214 let ParserMatchClass = VecListTwoDWordIndexAsmOperand;
215 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
217 // Register list of two Q registers with half-word lane subscripting.
218 def VecListTwoQHWordIndexAsmOperand : AsmOperandClass {
219 let Name = "VecListTwoQHWordIndexed";
220 let ParserMethod = "parseVectorList";
221 let RenderMethod = "addVecListIndexedOperands";
223 def VecListTwoQHWordIndexed : Operand<i32> {
224 let ParserMatchClass = VecListTwoQHWordIndexAsmOperand;
225 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
227 // ...with word lane subscripting.
228 def VecListTwoQWordIndexAsmOperand : AsmOperandClass {
229 let Name = "VecListTwoQWordIndexed";
230 let ParserMethod = "parseVectorList";
231 let RenderMethod = "addVecListIndexedOperands";
233 def VecListTwoQWordIndexed : Operand<i32> {
234 let ParserMatchClass = VecListTwoQWordIndexAsmOperand;
235 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
238 //===----------------------------------------------------------------------===//
239 // NEON-specific DAG Nodes.
240 //===----------------------------------------------------------------------===//
242 def SDTARMVCMP : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisSameAs<1, 2>]>;
243 def SDTARMVCMPZ : SDTypeProfile<1, 1, []>;
245 def NEONvceq : SDNode<"ARMISD::VCEQ", SDTARMVCMP>;
246 def NEONvceqz : SDNode<"ARMISD::VCEQZ", SDTARMVCMPZ>;
247 def NEONvcge : SDNode<"ARMISD::VCGE", SDTARMVCMP>;
248 def NEONvcgez : SDNode<"ARMISD::VCGEZ", SDTARMVCMPZ>;
249 def NEONvclez : SDNode<"ARMISD::VCLEZ", SDTARMVCMPZ>;
250 def NEONvcgeu : SDNode<"ARMISD::VCGEU", SDTARMVCMP>;
251 def NEONvcgt : SDNode<"ARMISD::VCGT", SDTARMVCMP>;
252 def NEONvcgtz : SDNode<"ARMISD::VCGTZ", SDTARMVCMPZ>;
253 def NEONvcltz : SDNode<"ARMISD::VCLTZ", SDTARMVCMPZ>;
254 def NEONvcgtu : SDNode<"ARMISD::VCGTU", SDTARMVCMP>;
255 def NEONvtst : SDNode<"ARMISD::VTST", SDTARMVCMP>;
257 // Types for vector shift by immediates. The "SHX" version is for long and
258 // narrow operations where the source and destination vectors have different
259 // types. The "SHINS" version is for shift and insert operations.
260 def SDTARMVSH : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
262 def SDTARMVSHX : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
264 def SDTARMVSHINS : SDTypeProfile<1, 3, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
265 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
267 def NEONvshl : SDNode<"ARMISD::VSHL", SDTARMVSH>;
268 def NEONvshrs : SDNode<"ARMISD::VSHRs", SDTARMVSH>;
269 def NEONvshru : SDNode<"ARMISD::VSHRu", SDTARMVSH>;
270 def NEONvshlls : SDNode<"ARMISD::VSHLLs", SDTARMVSHX>;
271 def NEONvshllu : SDNode<"ARMISD::VSHLLu", SDTARMVSHX>;
272 def NEONvshlli : SDNode<"ARMISD::VSHLLi", SDTARMVSHX>;
273 def NEONvshrn : SDNode<"ARMISD::VSHRN", SDTARMVSHX>;
275 def NEONvrshrs : SDNode<"ARMISD::VRSHRs", SDTARMVSH>;
276 def NEONvrshru : SDNode<"ARMISD::VRSHRu", SDTARMVSH>;
277 def NEONvrshrn : SDNode<"ARMISD::VRSHRN", SDTARMVSHX>;
279 def NEONvqshls : SDNode<"ARMISD::VQSHLs", SDTARMVSH>;
280 def NEONvqshlu : SDNode<"ARMISD::VQSHLu", SDTARMVSH>;
281 def NEONvqshlsu : SDNode<"ARMISD::VQSHLsu", SDTARMVSH>;
282 def NEONvqshrns : SDNode<"ARMISD::VQSHRNs", SDTARMVSHX>;
283 def NEONvqshrnu : SDNode<"ARMISD::VQSHRNu", SDTARMVSHX>;
284 def NEONvqshrnsu : SDNode<"ARMISD::VQSHRNsu", SDTARMVSHX>;
286 def NEONvqrshrns : SDNode<"ARMISD::VQRSHRNs", SDTARMVSHX>;
287 def NEONvqrshrnu : SDNode<"ARMISD::VQRSHRNu", SDTARMVSHX>;
288 def NEONvqrshrnsu : SDNode<"ARMISD::VQRSHRNsu", SDTARMVSHX>;
290 def NEONvsli : SDNode<"ARMISD::VSLI", SDTARMVSHINS>;
291 def NEONvsri : SDNode<"ARMISD::VSRI", SDTARMVSHINS>;
293 def SDTARMVGETLN : SDTypeProfile<1, 2, [SDTCisVT<0, i32>, SDTCisInt<1>,
295 def NEONvgetlaneu : SDNode<"ARMISD::VGETLANEu", SDTARMVGETLN>;
296 def NEONvgetlanes : SDNode<"ARMISD::VGETLANEs", SDTARMVGETLN>;
298 def SDTARMVMOVIMM : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisVT<1, i32>]>;
299 def NEONvmovImm : SDNode<"ARMISD::VMOVIMM", SDTARMVMOVIMM>;
300 def NEONvmvnImm : SDNode<"ARMISD::VMVNIMM", SDTARMVMOVIMM>;
301 def NEONvmovFPImm : SDNode<"ARMISD::VMOVFPIMM", SDTARMVMOVIMM>;
303 def SDTARMVORRIMM : SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
305 def NEONvorrImm : SDNode<"ARMISD::VORRIMM", SDTARMVORRIMM>;
306 def NEONvbicImm : SDNode<"ARMISD::VBICIMM", SDTARMVORRIMM>;
308 def NEONvbsl : SDNode<"ARMISD::VBSL",
309 SDTypeProfile<1, 3, [SDTCisVec<0>,
312 SDTCisSameAs<0, 3>]>>;
314 def NEONvdup : SDNode<"ARMISD::VDUP", SDTypeProfile<1, 1, [SDTCisVec<0>]>>;
316 // VDUPLANE can produce a quad-register result from a double-register source,
317 // so the result is not constrained to match the source.
318 def NEONvduplane : SDNode<"ARMISD::VDUPLANE",
319 SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisVec<1>,
322 def SDTARMVEXT : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
323 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
324 def NEONvext : SDNode<"ARMISD::VEXT", SDTARMVEXT>;
326 def SDTARMVSHUF : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisSameAs<0, 1>]>;
327 def NEONvrev64 : SDNode<"ARMISD::VREV64", SDTARMVSHUF>;
328 def NEONvrev32 : SDNode<"ARMISD::VREV32", SDTARMVSHUF>;
329 def NEONvrev16 : SDNode<"ARMISD::VREV16", SDTARMVSHUF>;
331 def SDTARMVSHUF2 : SDTypeProfile<2, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
333 SDTCisSameAs<0, 3>]>;
334 def NEONzip : SDNode<"ARMISD::VZIP", SDTARMVSHUF2>;
335 def NEONuzp : SDNode<"ARMISD::VUZP", SDTARMVSHUF2>;
336 def NEONtrn : SDNode<"ARMISD::VTRN", SDTARMVSHUF2>;
338 def SDTARMVMULL : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
339 SDTCisSameAs<1, 2>]>;
340 def NEONvmulls : SDNode<"ARMISD::VMULLs", SDTARMVMULL>;
341 def NEONvmullu : SDNode<"ARMISD::VMULLu", SDTARMVMULL>;
343 def SDTARMFMAX : SDTypeProfile<1, 2, [SDTCisVT<0, f32>, SDTCisSameAs<0, 1>,
344 SDTCisSameAs<0, 2>]>;
345 def NEONfmax : SDNode<"ARMISD::FMAX", SDTARMFMAX>;
346 def NEONfmin : SDNode<"ARMISD::FMIN", SDTARMFMAX>;
348 def NEONimmAllZerosV: PatLeaf<(NEONvmovImm (i32 timm)), [{
349 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
350 unsigned EltBits = 0;
351 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
352 return (EltBits == 32 && EltVal == 0);
355 def NEONimmAllOnesV: PatLeaf<(NEONvmovImm (i32 timm)), [{
356 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
357 unsigned EltBits = 0;
358 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
359 return (EltBits == 8 && EltVal == 0xff);
362 //===----------------------------------------------------------------------===//
363 // NEON load / store instructions
364 //===----------------------------------------------------------------------===//
366 // Use VLDM to load a Q register as a D register pair.
367 // This is a pseudo instruction that is expanded to VLDMD after reg alloc.
369 : PseudoVFPLdStM<(outs QPR:$dst), (ins GPR:$Rn),
371 [(set QPR:$dst, (v2f64 (load GPR:$Rn)))]>;
373 // Use VSTM to store a Q register as a D register pair.
374 // This is a pseudo instruction that is expanded to VSTMD after reg alloc.
376 : PseudoVFPLdStM<(outs), (ins QPR:$src, GPR:$Rn),
378 [(store (v2f64 QPR:$src), GPR:$Rn)]>;
380 // Classes for VLD* pseudo-instructions with multi-register operands.
381 // These are expanded to real instructions after register allocation.
382 class VLDQPseudo<InstrItinClass itin>
383 : PseudoNLdSt<(outs QPR:$dst), (ins addrmode6:$addr), itin, "">;
384 class VLDQWBPseudo<InstrItinClass itin>
385 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
386 (ins addrmode6:$addr, am6offset:$offset), itin,
388 class VLDQWBfixedPseudo<InstrItinClass itin>
389 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
390 (ins addrmode6:$addr), itin,
392 class VLDQWBregisterPseudo<InstrItinClass itin>
393 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
394 (ins addrmode6:$addr, rGPR:$offset), itin,
397 class VLDQQPseudo<InstrItinClass itin>
398 : PseudoNLdSt<(outs QQPR:$dst), (ins addrmode6:$addr), itin, "">;
399 class VLDQQWBPseudo<InstrItinClass itin>
400 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
401 (ins addrmode6:$addr, am6offset:$offset), itin,
403 class VLDQQWBfixedPseudo<InstrItinClass itin>
404 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
405 (ins addrmode6:$addr), itin,
407 class VLDQQWBregisterPseudo<InstrItinClass itin>
408 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
409 (ins addrmode6:$addr, rGPR:$offset), itin,
413 class VLDQQQQPseudo<InstrItinClass itin>
414 : PseudoNLdSt<(outs QQQQPR:$dst), (ins addrmode6:$addr, QQQQPR:$src),itin,
416 class VLDQQQQWBPseudo<InstrItinClass itin>
417 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
418 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
419 "$addr.addr = $wb, $src = $dst">;
421 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
423 // VLD1 : Vector Load (multiple single elements)
424 class VLD1D<bits<4> op7_4, string Dt>
425 : NLdSt<0,0b10,0b0111,op7_4, (outs VecListOneD:$Vd),
426 (ins addrmode6:$Rn), IIC_VLD1,
427 "vld1", Dt, "$Vd, $Rn", "", []> {
430 let DecoderMethod = "DecodeVLDInstruction";
432 class VLD1Q<bits<4> op7_4, string Dt>
433 : NLdSt<0,0b10,0b1010,op7_4, (outs VecListTwoD:$Vd),
434 (ins addrmode6:$Rn), IIC_VLD1x2,
435 "vld1", Dt, "$Vd, $Rn", "", []> {
437 let Inst{5-4} = Rn{5-4};
438 let DecoderMethod = "DecodeVLDInstruction";
441 def VLD1d8 : VLD1D<{0,0,0,?}, "8">;
442 def VLD1d16 : VLD1D<{0,1,0,?}, "16">;
443 def VLD1d32 : VLD1D<{1,0,0,?}, "32">;
444 def VLD1d64 : VLD1D<{1,1,0,?}, "64">;
446 def VLD1q8 : VLD1Q<{0,0,?,?}, "8">;
447 def VLD1q16 : VLD1Q<{0,1,?,?}, "16">;
448 def VLD1q32 : VLD1Q<{1,0,?,?}, "32">;
449 def VLD1q64 : VLD1Q<{1,1,?,?}, "64">;
451 def VLD1q8Pseudo : VLDQPseudo<IIC_VLD1x2>;
452 def VLD1q16Pseudo : VLDQPseudo<IIC_VLD1x2>;
453 def VLD1q32Pseudo : VLDQPseudo<IIC_VLD1x2>;
454 def VLD1q64Pseudo : VLDQPseudo<IIC_VLD1x2>;
456 // ...with address register writeback:
457 multiclass VLD1DWB<bits<4> op7_4, string Dt> {
458 def _fixed : NLdSt<0,0b10, 0b0111,op7_4, (outs VecListOneD:$Vd, GPR:$wb),
459 (ins addrmode6:$Rn), IIC_VLD1u,
460 "vld1", Dt, "$Vd, $Rn!",
461 "$Rn.addr = $wb", []> {
462 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
464 let DecoderMethod = "DecodeVLDInstruction";
465 let AsmMatchConverter = "cvtVLDwbFixed";
467 def _register : NLdSt<0,0b10,0b0111,op7_4, (outs VecListOneD:$Vd, GPR:$wb),
468 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1u,
469 "vld1", Dt, "$Vd, $Rn, $Rm",
470 "$Rn.addr = $wb", []> {
472 let DecoderMethod = "DecodeVLDInstruction";
473 let AsmMatchConverter = "cvtVLDwbRegister";
476 multiclass VLD1QWB<bits<4> op7_4, string Dt> {
477 def _fixed : NLdSt<0,0b10,0b1010,op7_4, (outs VecListTwoD:$Vd, GPR:$wb),
478 (ins addrmode6:$Rn), IIC_VLD1x2u,
479 "vld1", Dt, "$Vd, $Rn!",
480 "$Rn.addr = $wb", []> {
481 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
482 let Inst{5-4} = Rn{5-4};
483 let DecoderMethod = "DecodeVLDInstruction";
484 let AsmMatchConverter = "cvtVLDwbFixed";
486 def _register : NLdSt<0,0b10,0b1010,op7_4, (outs VecListTwoD:$Vd, GPR:$wb),
487 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1x2u,
488 "vld1", Dt, "$Vd, $Rn, $Rm",
489 "$Rn.addr = $wb", []> {
490 let Inst{5-4} = Rn{5-4};
491 let DecoderMethod = "DecodeVLDInstruction";
492 let AsmMatchConverter = "cvtVLDwbRegister";
496 defm VLD1d8wb : VLD1DWB<{0,0,0,?}, "8">;
497 defm VLD1d16wb : VLD1DWB<{0,1,0,?}, "16">;
498 defm VLD1d32wb : VLD1DWB<{1,0,0,?}, "32">;
499 defm VLD1d64wb : VLD1DWB<{1,1,0,?}, "64">;
500 defm VLD1q8wb : VLD1QWB<{0,0,?,?}, "8">;
501 defm VLD1q16wb : VLD1QWB<{0,1,?,?}, "16">;
502 defm VLD1q32wb : VLD1QWB<{1,0,?,?}, "32">;
503 defm VLD1q64wb : VLD1QWB<{1,1,?,?}, "64">;
505 def VLD1q8PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD1x2u>;
506 def VLD1q16PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD1x2u>;
507 def VLD1q32PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD1x2u>;
508 def VLD1q64PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD1x2u>;
509 def VLD1q8PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD1x2u>;
510 def VLD1q16PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD1x2u>;
511 def VLD1q32PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD1x2u>;
512 def VLD1q64PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD1x2u>;
514 // ...with 3 registers
515 class VLD1D3<bits<4> op7_4, string Dt>
516 : NLdSt<0,0b10,0b0110,op7_4, (outs VecListThreeD:$Vd),
517 (ins addrmode6:$Rn), IIC_VLD1x3, "vld1", Dt,
518 "$Vd, $Rn", "", []> {
521 let DecoderMethod = "DecodeVLDInstruction";
523 multiclass VLD1D3WB<bits<4> op7_4, string Dt> {
524 def _fixed : NLdSt<0,0b10,0b0110, op7_4, (outs VecListThreeD:$Vd, GPR:$wb),
525 (ins addrmode6:$Rn), IIC_VLD1x2u,
526 "vld1", Dt, "$Vd, $Rn!",
527 "$Rn.addr = $wb", []> {
528 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
530 let DecoderMethod = "DecodeVLDInstruction";
531 let AsmMatchConverter = "cvtVLDwbFixed";
533 def _register : NLdSt<0,0b10,0b0110,op7_4, (outs VecListThreeD:$Vd, GPR:$wb),
534 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1x2u,
535 "vld1", Dt, "$Vd, $Rn, $Rm",
536 "$Rn.addr = $wb", []> {
538 let DecoderMethod = "DecodeVLDInstruction";
539 let AsmMatchConverter = "cvtVLDwbRegister";
543 def VLD1d8T : VLD1D3<{0,0,0,?}, "8">;
544 def VLD1d16T : VLD1D3<{0,1,0,?}, "16">;
545 def VLD1d32T : VLD1D3<{1,0,0,?}, "32">;
546 def VLD1d64T : VLD1D3<{1,1,0,?}, "64">;
548 defm VLD1d8Twb : VLD1D3WB<{0,0,0,?}, "8">;
549 defm VLD1d16Twb : VLD1D3WB<{0,1,0,?}, "16">;
550 defm VLD1d32Twb : VLD1D3WB<{1,0,0,?}, "32">;
551 defm VLD1d64Twb : VLD1D3WB<{1,1,0,?}, "64">;
553 def VLD1d64TPseudo : VLDQQPseudo<IIC_VLD1x3>;
555 // ...with 4 registers
556 class VLD1D4<bits<4> op7_4, string Dt>
557 : NLdSt<0, 0b10, 0b0010, op7_4, (outs VecListFourD:$Vd),
558 (ins addrmode6:$Rn), IIC_VLD1x4, "vld1", Dt,
559 "$Vd, $Rn", "", []> {
561 let Inst{5-4} = Rn{5-4};
562 let DecoderMethod = "DecodeVLDInstruction";
564 multiclass VLD1D4WB<bits<4> op7_4, string Dt> {
565 def _fixed : NLdSt<0,0b10,0b0010, op7_4, (outs VecListFourD:$Vd, GPR:$wb),
566 (ins addrmode6:$Rn), IIC_VLD1x2u,
567 "vld1", Dt, "$Vd, $Rn!",
568 "$Rn.addr = $wb", []> {
569 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
570 let Inst{5-4} = Rn{5-4};
571 let DecoderMethod = "DecodeVLDInstruction";
572 let AsmMatchConverter = "cvtVLDwbFixed";
574 def _register : NLdSt<0,0b10,0b0010,op7_4, (outs VecListFourD:$Vd, GPR:$wb),
575 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1x2u,
576 "vld1", Dt, "$Vd, $Rn, $Rm",
577 "$Rn.addr = $wb", []> {
578 let Inst{5-4} = Rn{5-4};
579 let DecoderMethod = "DecodeVLDInstruction";
580 let AsmMatchConverter = "cvtVLDwbRegister";
584 def VLD1d8Q : VLD1D4<{0,0,?,?}, "8">;
585 def VLD1d16Q : VLD1D4<{0,1,?,?}, "16">;
586 def VLD1d32Q : VLD1D4<{1,0,?,?}, "32">;
587 def VLD1d64Q : VLD1D4<{1,1,?,?}, "64">;
589 defm VLD1d8Qwb : VLD1D4WB<{0,0,?,?}, "8">;
590 defm VLD1d16Qwb : VLD1D4WB<{0,1,?,?}, "16">;
591 defm VLD1d32Qwb : VLD1D4WB<{1,0,?,?}, "32">;
592 defm VLD1d64Qwb : VLD1D4WB<{1,1,?,?}, "64">;
594 def VLD1d64QPseudo : VLDQQPseudo<IIC_VLD1x4>;
596 // VLD2 : Vector Load (multiple 2-element structures)
597 class VLD2<bits<4> op11_8, bits<4> op7_4, string Dt, RegisterOperand VdTy,
599 : NLdSt<0, 0b10, op11_8, op7_4, (outs VdTy:$Vd),
600 (ins addrmode6:$Rn), itin,
601 "vld2", Dt, "$Vd, $Rn", "", []> {
603 let Inst{5-4} = Rn{5-4};
604 let DecoderMethod = "DecodeVLDInstruction";
607 def VLD2d8 : VLD2<0b1000, {0,0,?,?}, "8", VecListTwoD, IIC_VLD2>;
608 def VLD2d16 : VLD2<0b1000, {0,1,?,?}, "16", VecListTwoD, IIC_VLD2>;
609 def VLD2d32 : VLD2<0b1000, {1,0,?,?}, "32", VecListTwoD, IIC_VLD2>;
611 def VLD2q8 : VLD2<0b0011, {0,0,?,?}, "8", VecListFourD, IIC_VLD2x2>;
612 def VLD2q16 : VLD2<0b0011, {0,1,?,?}, "16", VecListFourD, IIC_VLD2x2>;
613 def VLD2q32 : VLD2<0b0011, {1,0,?,?}, "32", VecListFourD, IIC_VLD2x2>;
615 def VLD2d8Pseudo : VLDQPseudo<IIC_VLD2>;
616 def VLD2d16Pseudo : VLDQPseudo<IIC_VLD2>;
617 def VLD2d32Pseudo : VLDQPseudo<IIC_VLD2>;
619 def VLD2q8Pseudo : VLDQQPseudo<IIC_VLD2x2>;
620 def VLD2q16Pseudo : VLDQQPseudo<IIC_VLD2x2>;
621 def VLD2q32Pseudo : VLDQQPseudo<IIC_VLD2x2>;
623 // ...with address register writeback:
624 multiclass VLD2WB<bits<4> op11_8, bits<4> op7_4, string Dt,
625 RegisterOperand VdTy, InstrItinClass itin> {
626 def _fixed : NLdSt<0, 0b10, op11_8, op7_4, (outs VdTy:$Vd, GPR:$wb),
627 (ins addrmode6:$Rn), itin,
628 "vld2", Dt, "$Vd, $Rn!",
629 "$Rn.addr = $wb", []> {
630 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
631 let Inst{5-4} = Rn{5-4};
632 let DecoderMethod = "DecodeVLDInstruction";
633 let AsmMatchConverter = "cvtVLDwbFixed";
635 def _register : NLdSt<0, 0b10, op11_8, op7_4, (outs VdTy:$Vd, GPR:$wb),
636 (ins addrmode6:$Rn, rGPR:$Rm), itin,
637 "vld2", Dt, "$Vd, $Rn, $Rm",
638 "$Rn.addr = $wb", []> {
639 let Inst{5-4} = Rn{5-4};
640 let DecoderMethod = "DecodeVLDInstruction";
641 let AsmMatchConverter = "cvtVLDwbRegister";
645 defm VLD2d8wb : VLD2WB<0b1000, {0,0,?,?}, "8", VecListTwoD, IIC_VLD2u>;
646 defm VLD2d16wb : VLD2WB<0b1000, {0,1,?,?}, "16", VecListTwoD, IIC_VLD2u>;
647 defm VLD2d32wb : VLD2WB<0b1000, {1,0,?,?}, "32", VecListTwoD, IIC_VLD2u>;
649 defm VLD2q8wb : VLD2WB<0b0011, {0,0,?,?}, "8", VecListFourD, IIC_VLD2x2u>;
650 defm VLD2q16wb : VLD2WB<0b0011, {0,1,?,?}, "16", VecListFourD, IIC_VLD2x2u>;
651 defm VLD2q32wb : VLD2WB<0b0011, {1,0,?,?}, "32", VecListFourD, IIC_VLD2x2u>;
653 def VLD2d8PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD2u>;
654 def VLD2d16PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD2u>;
655 def VLD2d32PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD2u>;
656 def VLD2d8PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD2u>;
657 def VLD2d16PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD2u>;
658 def VLD2d32PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD2u>;
660 def VLD2q8PseudoWB_fixed : VLDQQWBfixedPseudo<IIC_VLD2x2u>;
661 def VLD2q16PseudoWB_fixed : VLDQQWBfixedPseudo<IIC_VLD2x2u>;
662 def VLD2q32PseudoWB_fixed : VLDQQWBfixedPseudo<IIC_VLD2x2u>;
663 def VLD2q8PseudoWB_register : VLDQQWBregisterPseudo<IIC_VLD2x2u>;
664 def VLD2q16PseudoWB_register : VLDQQWBregisterPseudo<IIC_VLD2x2u>;
665 def VLD2q32PseudoWB_register : VLDQQWBregisterPseudo<IIC_VLD2x2u>;
667 // ...with double-spaced registers
668 def VLD2b8 : VLD2<0b1001, {0,0,?,?}, "8", VecListTwoQ, IIC_VLD2>;
669 def VLD2b16 : VLD2<0b1001, {0,1,?,?}, "16", VecListTwoQ, IIC_VLD2>;
670 def VLD2b32 : VLD2<0b1001, {1,0,?,?}, "32", VecListTwoQ, IIC_VLD2>;
671 defm VLD2b8wb : VLD2WB<0b1001, {0,0,?,?}, "8", VecListTwoQ, IIC_VLD2u>;
672 defm VLD2b16wb : VLD2WB<0b1001, {0,1,?,?}, "16", VecListTwoQ, IIC_VLD2u>;
673 defm VLD2b32wb : VLD2WB<0b1001, {1,0,?,?}, "32", VecListTwoQ, IIC_VLD2u>;
675 // VLD3 : Vector Load (multiple 3-element structures)
676 class VLD3D<bits<4> op11_8, bits<4> op7_4, string Dt>
677 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
678 (ins addrmode6:$Rn), IIC_VLD3,
679 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn", "", []> {
682 let DecoderMethod = "DecodeVLDInstruction";
685 def VLD3d8 : VLD3D<0b0100, {0,0,0,?}, "8">;
686 def VLD3d16 : VLD3D<0b0100, {0,1,0,?}, "16">;
687 def VLD3d32 : VLD3D<0b0100, {1,0,0,?}, "32">;
689 def VLD3d8Pseudo : VLDQQPseudo<IIC_VLD3>;
690 def VLD3d16Pseudo : VLDQQPseudo<IIC_VLD3>;
691 def VLD3d32Pseudo : VLDQQPseudo<IIC_VLD3>;
693 // ...with address register writeback:
694 class VLD3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
695 : NLdSt<0, 0b10, op11_8, op7_4,
696 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
697 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD3u,
698 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn$Rm",
699 "$Rn.addr = $wb", []> {
701 let DecoderMethod = "DecodeVLDInstruction";
704 def VLD3d8_UPD : VLD3DWB<0b0100, {0,0,0,?}, "8">;
705 def VLD3d16_UPD : VLD3DWB<0b0100, {0,1,0,?}, "16">;
706 def VLD3d32_UPD : VLD3DWB<0b0100, {1,0,0,?}, "32">;
708 def VLD3d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
709 def VLD3d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
710 def VLD3d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
712 // ...with double-spaced registers:
713 def VLD3q8 : VLD3D<0b0101, {0,0,0,?}, "8">;
714 def VLD3q16 : VLD3D<0b0101, {0,1,0,?}, "16">;
715 def VLD3q32 : VLD3D<0b0101, {1,0,0,?}, "32">;
716 def VLD3q8_UPD : VLD3DWB<0b0101, {0,0,0,?}, "8">;
717 def VLD3q16_UPD : VLD3DWB<0b0101, {0,1,0,?}, "16">;
718 def VLD3q32_UPD : VLD3DWB<0b0101, {1,0,0,?}, "32">;
720 def VLD3q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
721 def VLD3q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
722 def VLD3q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
724 // ...alternate versions to be allocated odd register numbers:
725 def VLD3q8oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
726 def VLD3q16oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
727 def VLD3q32oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
729 def VLD3q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
730 def VLD3q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
731 def VLD3q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
733 // VLD4 : Vector Load (multiple 4-element structures)
734 class VLD4D<bits<4> op11_8, bits<4> op7_4, string Dt>
735 : NLdSt<0, 0b10, op11_8, op7_4,
736 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
737 (ins addrmode6:$Rn), IIC_VLD4,
738 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn", "", []> {
740 let Inst{5-4} = Rn{5-4};
741 let DecoderMethod = "DecodeVLDInstruction";
744 def VLD4d8 : VLD4D<0b0000, {0,0,?,?}, "8">;
745 def VLD4d16 : VLD4D<0b0000, {0,1,?,?}, "16">;
746 def VLD4d32 : VLD4D<0b0000, {1,0,?,?}, "32">;
748 def VLD4d8Pseudo : VLDQQPseudo<IIC_VLD4>;
749 def VLD4d16Pseudo : VLDQQPseudo<IIC_VLD4>;
750 def VLD4d32Pseudo : VLDQQPseudo<IIC_VLD4>;
752 // ...with address register writeback:
753 class VLD4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
754 : NLdSt<0, 0b10, op11_8, op7_4,
755 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
756 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD4u,
757 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn$Rm",
758 "$Rn.addr = $wb", []> {
759 let Inst{5-4} = Rn{5-4};
760 let DecoderMethod = "DecodeVLDInstruction";
763 def VLD4d8_UPD : VLD4DWB<0b0000, {0,0,?,?}, "8">;
764 def VLD4d16_UPD : VLD4DWB<0b0000, {0,1,?,?}, "16">;
765 def VLD4d32_UPD : VLD4DWB<0b0000, {1,0,?,?}, "32">;
767 def VLD4d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
768 def VLD4d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
769 def VLD4d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
771 // ...with double-spaced registers:
772 def VLD4q8 : VLD4D<0b0001, {0,0,?,?}, "8">;
773 def VLD4q16 : VLD4D<0b0001, {0,1,?,?}, "16">;
774 def VLD4q32 : VLD4D<0b0001, {1,0,?,?}, "32">;
775 def VLD4q8_UPD : VLD4DWB<0b0001, {0,0,?,?}, "8">;
776 def VLD4q16_UPD : VLD4DWB<0b0001, {0,1,?,?}, "16">;
777 def VLD4q32_UPD : VLD4DWB<0b0001, {1,0,?,?}, "32">;
779 def VLD4q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
780 def VLD4q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
781 def VLD4q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
783 // ...alternate versions to be allocated odd register numbers:
784 def VLD4q8oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
785 def VLD4q16oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
786 def VLD4q32oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
788 def VLD4q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
789 def VLD4q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
790 def VLD4q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
792 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
794 // Classes for VLD*LN pseudo-instructions with multi-register operands.
795 // These are expanded to real instructions after register allocation.
796 class VLDQLNPseudo<InstrItinClass itin>
797 : PseudoNLdSt<(outs QPR:$dst),
798 (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
799 itin, "$src = $dst">;
800 class VLDQLNWBPseudo<InstrItinClass itin>
801 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
802 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
803 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
804 class VLDQQLNPseudo<InstrItinClass itin>
805 : PseudoNLdSt<(outs QQPR:$dst),
806 (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
807 itin, "$src = $dst">;
808 class VLDQQLNWBPseudo<InstrItinClass itin>
809 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
810 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
811 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
812 class VLDQQQQLNPseudo<InstrItinClass itin>
813 : PseudoNLdSt<(outs QQQQPR:$dst),
814 (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
815 itin, "$src = $dst">;
816 class VLDQQQQLNWBPseudo<InstrItinClass itin>
817 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
818 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
819 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
821 // VLD1LN : Vector Load (single element to one lane)
822 class VLD1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
824 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd),
825 (ins addrmode6:$Rn, DPR:$src, nohash_imm:$lane),
826 IIC_VLD1ln, "vld1", Dt, "\\{$Vd[$lane]\\}, $Rn",
828 [(set DPR:$Vd, (vector_insert (Ty DPR:$src),
829 (i32 (LoadOp addrmode6:$Rn)),
832 let DecoderMethod = "DecodeVLD1LN";
834 class VLD1LN32<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
836 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd),
837 (ins addrmode6oneL32:$Rn, DPR:$src, nohash_imm:$lane),
838 IIC_VLD1ln, "vld1", Dt, "\\{$Vd[$lane]\\}, $Rn",
840 [(set DPR:$Vd, (vector_insert (Ty DPR:$src),
841 (i32 (LoadOp addrmode6oneL32:$Rn)),
844 let DecoderMethod = "DecodeVLD1LN";
846 class VLD1QLNPseudo<ValueType Ty, PatFrag LoadOp> : VLDQLNPseudo<IIC_VLD1ln> {
847 let Pattern = [(set QPR:$dst, (vector_insert (Ty QPR:$src),
848 (i32 (LoadOp addrmode6:$addr)),
852 def VLD1LNd8 : VLD1LN<0b0000, {?,?,?,0}, "8", v8i8, extloadi8> {
853 let Inst{7-5} = lane{2-0};
855 def VLD1LNd16 : VLD1LN<0b0100, {?,?,0,?}, "16", v4i16, extloadi16> {
856 let Inst{7-6} = lane{1-0};
857 let Inst{5-4} = Rn{5-4};
859 def VLD1LNd32 : VLD1LN32<0b1000, {?,0,?,?}, "32", v2i32, load> {
860 let Inst{7} = lane{0};
861 let Inst{5-4} = Rn{5-4};
864 def VLD1LNq8Pseudo : VLD1QLNPseudo<v16i8, extloadi8>;
865 def VLD1LNq16Pseudo : VLD1QLNPseudo<v8i16, extloadi16>;
866 def VLD1LNq32Pseudo : VLD1QLNPseudo<v4i32, load>;
868 def : Pat<(vector_insert (v2f32 DPR:$src),
869 (f32 (load addrmode6:$addr)), imm:$lane),
870 (VLD1LNd32 addrmode6:$addr, DPR:$src, imm:$lane)>;
871 def : Pat<(vector_insert (v4f32 QPR:$src),
872 (f32 (load addrmode6:$addr)), imm:$lane),
873 (VLD1LNq32Pseudo addrmode6:$addr, QPR:$src, imm:$lane)>;
875 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
877 // ...with address register writeback:
878 class VLD1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
879 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, GPR:$wb),
880 (ins addrmode6:$Rn, am6offset:$Rm,
881 DPR:$src, nohash_imm:$lane), IIC_VLD1lnu, "vld1", Dt,
882 "\\{$Vd[$lane]\\}, $Rn$Rm",
883 "$src = $Vd, $Rn.addr = $wb", []> {
884 let DecoderMethod = "DecodeVLD1LN";
887 def VLD1LNd8_UPD : VLD1LNWB<0b0000, {?,?,?,0}, "8"> {
888 let Inst{7-5} = lane{2-0};
890 def VLD1LNd16_UPD : VLD1LNWB<0b0100, {?,?,0,?}, "16"> {
891 let Inst{7-6} = lane{1-0};
894 def VLD1LNd32_UPD : VLD1LNWB<0b1000, {?,0,?,?}, "32"> {
895 let Inst{7} = lane{0};
900 def VLD1LNq8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
901 def VLD1LNq16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
902 def VLD1LNq32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
904 // VLD2LN : Vector Load (single 2-element structure to one lane)
905 class VLD2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
906 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2),
907 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, nohash_imm:$lane),
908 IIC_VLD2ln, "vld2", Dt, "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn",
909 "$src1 = $Vd, $src2 = $dst2", []> {
912 let DecoderMethod = "DecodeVLD2LN";
915 def VLD2LNd8 : VLD2LN<0b0001, {?,?,?,?}, "8"> {
916 let Inst{7-5} = lane{2-0};
918 def VLD2LNd16 : VLD2LN<0b0101, {?,?,0,?}, "16"> {
919 let Inst{7-6} = lane{1-0};
921 def VLD2LNd32 : VLD2LN<0b1001, {?,0,0,?}, "32"> {
922 let Inst{7} = lane{0};
925 def VLD2LNd8Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
926 def VLD2LNd16Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
927 def VLD2LNd32Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
929 // ...with double-spaced registers:
930 def VLD2LNq16 : VLD2LN<0b0101, {?,?,1,?}, "16"> {
931 let Inst{7-6} = lane{1-0};
933 def VLD2LNq32 : VLD2LN<0b1001, {?,1,0,?}, "32"> {
934 let Inst{7} = lane{0};
937 def VLD2LNq16Pseudo : VLDQQLNPseudo<IIC_VLD2ln>;
938 def VLD2LNq32Pseudo : VLDQQLNPseudo<IIC_VLD2ln>;
940 // ...with address register writeback:
941 class VLD2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
942 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
943 (ins addrmode6:$Rn, am6offset:$Rm,
944 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VLD2lnu, "vld2", Dt,
945 "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn$Rm",
946 "$src1 = $Vd, $src2 = $dst2, $Rn.addr = $wb", []> {
948 let DecoderMethod = "DecodeVLD2LN";
951 def VLD2LNd8_UPD : VLD2LNWB<0b0001, {?,?,?,?}, "8"> {
952 let Inst{7-5} = lane{2-0};
954 def VLD2LNd16_UPD : VLD2LNWB<0b0101, {?,?,0,?}, "16"> {
955 let Inst{7-6} = lane{1-0};
957 def VLD2LNd32_UPD : VLD2LNWB<0b1001, {?,0,0,?}, "32"> {
958 let Inst{7} = lane{0};
961 def VLD2LNd8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
962 def VLD2LNd16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
963 def VLD2LNd32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
965 def VLD2LNq16_UPD : VLD2LNWB<0b0101, {?,?,1,?}, "16"> {
966 let Inst{7-6} = lane{1-0};
968 def VLD2LNq32_UPD : VLD2LNWB<0b1001, {?,1,0,?}, "32"> {
969 let Inst{7} = lane{0};
972 def VLD2LNq16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
973 def VLD2LNq32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
975 // VLD3LN : Vector Load (single 3-element structure to one lane)
976 class VLD3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
977 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
978 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3,
979 nohash_imm:$lane), IIC_VLD3ln, "vld3", Dt,
980 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn",
981 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3", []> {
983 let DecoderMethod = "DecodeVLD3LN";
986 def VLD3LNd8 : VLD3LN<0b0010, {?,?,?,0}, "8"> {
987 let Inst{7-5} = lane{2-0};
989 def VLD3LNd16 : VLD3LN<0b0110, {?,?,0,0}, "16"> {
990 let Inst{7-6} = lane{1-0};
992 def VLD3LNd32 : VLD3LN<0b1010, {?,0,0,0}, "32"> {
993 let Inst{7} = lane{0};
996 def VLD3LNd8Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
997 def VLD3LNd16Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
998 def VLD3LNd32Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
1000 // ...with double-spaced registers:
1001 def VLD3LNq16 : VLD3LN<0b0110, {?,?,1,0}, "16"> {
1002 let Inst{7-6} = lane{1-0};
1004 def VLD3LNq32 : VLD3LN<0b1010, {?,1,0,0}, "32"> {
1005 let Inst{7} = lane{0};
1008 def VLD3LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
1009 def VLD3LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
1011 // ...with address register writeback:
1012 class VLD3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1013 : NLdStLn<1, 0b10, op11_8, op7_4,
1014 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
1015 (ins addrmode6:$Rn, am6offset:$Rm,
1016 DPR:$src1, DPR:$src2, DPR:$src3, nohash_imm:$lane),
1017 IIC_VLD3lnu, "vld3", Dt,
1018 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn$Rm",
1019 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $Rn.addr = $wb",
1021 let DecoderMethod = "DecodeVLD3LN";
1024 def VLD3LNd8_UPD : VLD3LNWB<0b0010, {?,?,?,0}, "8"> {
1025 let Inst{7-5} = lane{2-0};
1027 def VLD3LNd16_UPD : VLD3LNWB<0b0110, {?,?,0,0}, "16"> {
1028 let Inst{7-6} = lane{1-0};
1030 def VLD3LNd32_UPD : VLD3LNWB<0b1010, {?,0,0,0}, "32"> {
1031 let Inst{7} = lane{0};
1034 def VLD3LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
1035 def VLD3LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
1036 def VLD3LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
1038 def VLD3LNq16_UPD : VLD3LNWB<0b0110, {?,?,1,0}, "16"> {
1039 let Inst{7-6} = lane{1-0};
1041 def VLD3LNq32_UPD : VLD3LNWB<0b1010, {?,1,0,0}, "32"> {
1042 let Inst{7} = lane{0};
1045 def VLD3LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
1046 def VLD3LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
1048 // VLD4LN : Vector Load (single 4-element structure to one lane)
1049 class VLD4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1050 : NLdStLn<1, 0b10, op11_8, op7_4,
1051 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
1052 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
1053 nohash_imm:$lane), IIC_VLD4ln, "vld4", Dt,
1054 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn",
1055 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4", []> {
1057 let Inst{4} = Rn{4};
1058 let DecoderMethod = "DecodeVLD4LN";
1061 def VLD4LNd8 : VLD4LN<0b0011, {?,?,?,?}, "8"> {
1062 let Inst{7-5} = lane{2-0};
1064 def VLD4LNd16 : VLD4LN<0b0111, {?,?,0,?}, "16"> {
1065 let Inst{7-6} = lane{1-0};
1067 def VLD4LNd32 : VLD4LN<0b1011, {?,0,?,?}, "32"> {
1068 let Inst{7} = lane{0};
1069 let Inst{5} = Rn{5};
1072 def VLD4LNd8Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
1073 def VLD4LNd16Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
1074 def VLD4LNd32Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
1076 // ...with double-spaced registers:
1077 def VLD4LNq16 : VLD4LN<0b0111, {?,?,1,?}, "16"> {
1078 let Inst{7-6} = lane{1-0};
1080 def VLD4LNq32 : VLD4LN<0b1011, {?,1,?,?}, "32"> {
1081 let Inst{7} = lane{0};
1082 let Inst{5} = Rn{5};
1085 def VLD4LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>;
1086 def VLD4LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>;
1088 // ...with address register writeback:
1089 class VLD4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1090 : NLdStLn<1, 0b10, op11_8, op7_4,
1091 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
1092 (ins addrmode6:$Rn, am6offset:$Rm,
1093 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
1094 IIC_VLD4lnu, "vld4", Dt,
1095 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn$Rm",
1096 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4, $Rn.addr = $wb",
1098 let Inst{4} = Rn{4};
1099 let DecoderMethod = "DecodeVLD4LN" ;
1102 def VLD4LNd8_UPD : VLD4LNWB<0b0011, {?,?,?,?}, "8"> {
1103 let Inst{7-5} = lane{2-0};
1105 def VLD4LNd16_UPD : VLD4LNWB<0b0111, {?,?,0,?}, "16"> {
1106 let Inst{7-6} = lane{1-0};
1108 def VLD4LNd32_UPD : VLD4LNWB<0b1011, {?,0,?,?}, "32"> {
1109 let Inst{7} = lane{0};
1110 let Inst{5} = Rn{5};
1113 def VLD4LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
1114 def VLD4LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
1115 def VLD4LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
1117 def VLD4LNq16_UPD : VLD4LNWB<0b0111, {?,?,1,?}, "16"> {
1118 let Inst{7-6} = lane{1-0};
1120 def VLD4LNq32_UPD : VLD4LNWB<0b1011, {?,1,?,?}, "32"> {
1121 let Inst{7} = lane{0};
1122 let Inst{5} = Rn{5};
1125 def VLD4LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
1126 def VLD4LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
1128 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
1130 // VLD1DUP : Vector Load (single element to all lanes)
1131 class VLD1DUP<bits<4> op7_4, string Dt, ValueType Ty, PatFrag LoadOp>
1132 : NLdSt<1, 0b10, 0b1100, op7_4, (outs VecListOneDAllLanes:$Vd),
1133 (ins addrmode6dup:$Rn),
1134 IIC_VLD1dup, "vld1", Dt, "$Vd, $Rn", "",
1135 [(set VecListOneDAllLanes:$Vd,
1136 (Ty (NEONvdup (i32 (LoadOp addrmode6dup:$Rn)))))]> {
1138 let Inst{4} = Rn{4};
1139 let DecoderMethod = "DecodeVLD1DupInstruction";
1141 class VLD1QDUPPseudo<ValueType Ty, PatFrag LoadOp> : VLDQPseudo<IIC_VLD1dup> {
1142 let Pattern = [(set QPR:$dst,
1143 (Ty (NEONvdup (i32 (LoadOp addrmode6dup:$addr)))))];
1146 def VLD1DUPd8 : VLD1DUP<{0,0,0,?}, "8", v8i8, extloadi8>;
1147 def VLD1DUPd16 : VLD1DUP<{0,1,0,?}, "16", v4i16, extloadi16>;
1148 def VLD1DUPd32 : VLD1DUP<{1,0,0,?}, "32", v2i32, load>;
1150 def VLD1DUPq8Pseudo : VLD1QDUPPseudo<v16i8, extloadi8>;
1151 def VLD1DUPq16Pseudo : VLD1QDUPPseudo<v8i16, extloadi16>;
1152 def VLD1DUPq32Pseudo : VLD1QDUPPseudo<v4i32, load>;
1154 def : Pat<(v2f32 (NEONvdup (f32 (load addrmode6dup:$addr)))),
1155 (VLD1DUPd32 addrmode6:$addr)>;
1156 def : Pat<(v4f32 (NEONvdup (f32 (load addrmode6dup:$addr)))),
1157 (VLD1DUPq32Pseudo addrmode6:$addr)>;
1159 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
1161 class VLD1QDUP<bits<4> op7_4, string Dt>
1162 : NLdSt<1, 0b10, 0b1100, op7_4, (outs VecListTwoDAllLanes:$Vd),
1163 (ins addrmode6dup:$Rn), IIC_VLD1dup,
1164 "vld1", Dt, "$Vd, $Rn", "", []> {
1166 let Inst{4} = Rn{4};
1167 let DecoderMethod = "DecodeVLD1DupInstruction";
1170 def VLD1DUPq8 : VLD1QDUP<{0,0,1,0}, "8">;
1171 def VLD1DUPq16 : VLD1QDUP<{0,1,1,?}, "16">;
1172 def VLD1DUPq32 : VLD1QDUP<{1,0,1,?}, "32">;
1174 // ...with address register writeback:
1175 multiclass VLD1DUPWB<bits<4> op7_4, string Dt> {
1176 def _fixed : NLdSt<1, 0b10, 0b1100, op7_4,
1177 (outs VecListOneDAllLanes:$Vd, GPR:$wb),
1178 (ins addrmode6dup:$Rn), IIC_VLD1dupu,
1179 "vld1", Dt, "$Vd, $Rn!",
1180 "$Rn.addr = $wb", []> {
1181 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1182 let Inst{4} = Rn{4};
1183 let DecoderMethod = "DecodeVLD1DupInstruction";
1184 let AsmMatchConverter = "cvtVLDwbFixed";
1186 def _register : NLdSt<1, 0b10, 0b1100, op7_4,
1187 (outs VecListOneDAllLanes:$Vd, GPR:$wb),
1188 (ins addrmode6dup:$Rn, rGPR:$Rm), IIC_VLD1dupu,
1189 "vld1", Dt, "$Vd, $Rn, $Rm",
1190 "$Rn.addr = $wb", []> {
1191 let Inst{4} = Rn{4};
1192 let DecoderMethod = "DecodeVLD1DupInstruction";
1193 let AsmMatchConverter = "cvtVLDwbRegister";
1196 multiclass VLD1QDUPWB<bits<4> op7_4, string Dt> {
1197 def _fixed : NLdSt<1, 0b10, 0b1100, op7_4,
1198 (outs VecListTwoDAllLanes:$Vd, GPR:$wb),
1199 (ins addrmode6dup:$Rn), IIC_VLD1dupu,
1200 "vld1", Dt, "$Vd, $Rn!",
1201 "$Rn.addr = $wb", []> {
1202 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1203 let Inst{4} = Rn{4};
1204 let DecoderMethod = "DecodeVLD1DupInstruction";
1205 let AsmMatchConverter = "cvtVLDwbFixed";
1207 def _register : NLdSt<1, 0b10, 0b1100, op7_4,
1208 (outs VecListTwoDAllLanes:$Vd, GPR:$wb),
1209 (ins addrmode6dup:$Rn, rGPR:$Rm), IIC_VLD1dupu,
1210 "vld1", Dt, "$Vd, $Rn, $Rm",
1211 "$Rn.addr = $wb", []> {
1212 let Inst{4} = Rn{4};
1213 let DecoderMethod = "DecodeVLD1DupInstruction";
1214 let AsmMatchConverter = "cvtVLDwbRegister";
1218 defm VLD1DUPd8wb : VLD1DUPWB<{0,0,0,0}, "8">;
1219 defm VLD1DUPd16wb : VLD1DUPWB<{0,1,0,?}, "16">;
1220 defm VLD1DUPd32wb : VLD1DUPWB<{1,0,0,?}, "32">;
1222 defm VLD1DUPq8wb : VLD1QDUPWB<{0,0,1,0}, "8">;
1223 defm VLD1DUPq16wb : VLD1QDUPWB<{0,1,1,?}, "16">;
1224 defm VLD1DUPq32wb : VLD1QDUPWB<{1,0,1,?}, "32">;
1226 def VLD1DUPq8PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD1dupu>;
1227 def VLD1DUPq16PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD1dupu>;
1228 def VLD1DUPq32PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD1dupu>;
1229 def VLD1DUPq8PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD1dupu>;
1230 def VLD1DUPq16PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD1dupu>;
1231 def VLD1DUPq32PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD1dupu>;
1233 // VLD2DUP : Vector Load (single 2-element structure to all lanes)
1234 class VLD2DUP<bits<4> op7_4, string Dt, RegisterOperand VdTy>
1235 : NLdSt<1, 0b10, 0b1101, op7_4, (outs VdTy:$Vd),
1236 (ins addrmode6dup:$Rn), IIC_VLD2dup,
1237 "vld2", Dt, "$Vd, $Rn", "", []> {
1239 let Inst{4} = Rn{4};
1240 let DecoderMethod = "DecodeVLD2DupInstruction";
1243 def VLD2DUPd8 : VLD2DUP<{0,0,0,?}, "8", VecListTwoDAllLanes>;
1244 def VLD2DUPd16 : VLD2DUP<{0,1,0,?}, "16", VecListTwoDAllLanes>;
1245 def VLD2DUPd32 : VLD2DUP<{1,0,0,?}, "32", VecListTwoDAllLanes>;
1247 def VLD2DUPd8Pseudo : VLDQPseudo<IIC_VLD2dup>;
1248 def VLD2DUPd16Pseudo : VLDQPseudo<IIC_VLD2dup>;
1249 def VLD2DUPd32Pseudo : VLDQPseudo<IIC_VLD2dup>;
1251 // ...with double-spaced registers (not used for codegen):
1252 def VLD2DUPd8x2 : VLD2DUP<{0,0,1,?}, "8", VecListTwoQAllLanes>;
1253 def VLD2DUPd16x2 : VLD2DUP<{0,1,1,?}, "16", VecListTwoQAllLanes>;
1254 def VLD2DUPd32x2 : VLD2DUP<{1,0,1,?}, "32", VecListTwoQAllLanes>;
1256 // ...with address register writeback:
1257 class VLD2DUPWB<bits<4> op7_4, string Dt>
1258 : NLdSt<1, 0b10, 0b1101, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
1259 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD2dupu,
1260 "vld2", Dt, "\\{$Vd[], $dst2[]\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
1261 let Inst{4} = Rn{4};
1262 let DecoderMethod = "DecodeVLD2DupInstruction";
1265 def VLD2DUPd8_UPD : VLD2DUPWB<{0,0,0,0}, "8">;
1266 def VLD2DUPd16_UPD : VLD2DUPWB<{0,1,0,?}, "16">;
1267 def VLD2DUPd32_UPD : VLD2DUPWB<{1,0,0,?}, "32">;
1269 def VLD2DUPd8x2_UPD : VLD2DUPWB<{0,0,1,0}, "8">;
1270 def VLD2DUPd16x2_UPD : VLD2DUPWB<{0,1,1,?}, "16">;
1271 def VLD2DUPd32x2_UPD : VLD2DUPWB<{1,0,1,?}, "32">;
1273 def VLD2DUPd8Pseudo_UPD : VLDQWBPseudo<IIC_VLD2dupu>;
1274 def VLD2DUPd16Pseudo_UPD : VLDQWBPseudo<IIC_VLD2dupu>;
1275 def VLD2DUPd32Pseudo_UPD : VLDQWBPseudo<IIC_VLD2dupu>;
1277 // VLD3DUP : Vector Load (single 3-element structure to all lanes)
1278 class VLD3DUP<bits<4> op7_4, string Dt>
1279 : NLdSt<1, 0b10, 0b1110, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
1280 (ins addrmode6dup:$Rn), IIC_VLD3dup,
1281 "vld3", Dt, "\\{$Vd[], $dst2[], $dst3[]\\}, $Rn", "", []> {
1284 let DecoderMethod = "DecodeVLD3DupInstruction";
1287 def VLD3DUPd8 : VLD3DUP<{0,0,0,?}, "8">;
1288 def VLD3DUPd16 : VLD3DUP<{0,1,0,?}, "16">;
1289 def VLD3DUPd32 : VLD3DUP<{1,0,0,?}, "32">;
1291 def VLD3DUPd8Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1292 def VLD3DUPd16Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1293 def VLD3DUPd32Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1295 // ...with double-spaced registers (not used for codegen):
1296 def VLD3DUPd8x2 : VLD3DUP<{0,0,1,?}, "8">;
1297 def VLD3DUPd16x2 : VLD3DUP<{0,1,1,?}, "16">;
1298 def VLD3DUPd32x2 : VLD3DUP<{1,0,1,?}, "32">;
1300 // ...with address register writeback:
1301 class VLD3DUPWB<bits<4> op7_4, string Dt>
1302 : NLdSt<1, 0b10, 0b1110, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
1303 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD3dupu,
1304 "vld3", Dt, "\\{$Vd[], $dst2[], $dst3[]\\}, $Rn$Rm",
1305 "$Rn.addr = $wb", []> {
1307 let DecoderMethod = "DecodeVLD3DupInstruction";
1310 def VLD3DUPd8_UPD : VLD3DUPWB<{0,0,0,0}, "8">;
1311 def VLD3DUPd16_UPD : VLD3DUPWB<{0,1,0,?}, "16">;
1312 def VLD3DUPd32_UPD : VLD3DUPWB<{1,0,0,?}, "32">;
1314 def VLD3DUPd8x2_UPD : VLD3DUPWB<{0,0,1,0}, "8">;
1315 def VLD3DUPd16x2_UPD : VLD3DUPWB<{0,1,1,?}, "16">;
1316 def VLD3DUPd32x2_UPD : VLD3DUPWB<{1,0,1,?}, "32">;
1318 def VLD3DUPd8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1319 def VLD3DUPd16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1320 def VLD3DUPd32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1322 // VLD4DUP : Vector Load (single 4-element structure to all lanes)
1323 class VLD4DUP<bits<4> op7_4, string Dt>
1324 : NLdSt<1, 0b10, 0b1111, op7_4,
1325 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
1326 (ins addrmode6dup:$Rn), IIC_VLD4dup,
1327 "vld4", Dt, "\\{$Vd[], $dst2[], $dst3[], $dst4[]\\}, $Rn", "", []> {
1329 let Inst{4} = Rn{4};
1330 let DecoderMethod = "DecodeVLD4DupInstruction";
1333 def VLD4DUPd8 : VLD4DUP<{0,0,0,?}, "8">;
1334 def VLD4DUPd16 : VLD4DUP<{0,1,0,?}, "16">;
1335 def VLD4DUPd32 : VLD4DUP<{1,?,0,?}, "32"> { let Inst{6} = Rn{5}; }
1337 def VLD4DUPd8Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1338 def VLD4DUPd16Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1339 def VLD4DUPd32Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1341 // ...with double-spaced registers (not used for codegen):
1342 def VLD4DUPd8x2 : VLD4DUP<{0,0,1,?}, "8">;
1343 def VLD4DUPd16x2 : VLD4DUP<{0,1,1,?}, "16">;
1344 def VLD4DUPd32x2 : VLD4DUP<{1,?,1,?}, "32"> { let Inst{6} = Rn{5}; }
1346 // ...with address register writeback:
1347 class VLD4DUPWB<bits<4> op7_4, string Dt>
1348 : NLdSt<1, 0b10, 0b1111, op7_4,
1349 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
1350 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD4dupu,
1351 "vld4", Dt, "\\{$Vd[], $dst2[], $dst3[], $dst4[]\\}, $Rn$Rm",
1352 "$Rn.addr = $wb", []> {
1353 let Inst{4} = Rn{4};
1354 let DecoderMethod = "DecodeVLD4DupInstruction";
1357 def VLD4DUPd8_UPD : VLD4DUPWB<{0,0,0,0}, "8">;
1358 def VLD4DUPd16_UPD : VLD4DUPWB<{0,1,0,?}, "16">;
1359 def VLD4DUPd32_UPD : VLD4DUPWB<{1,?,0,?}, "32"> { let Inst{6} = Rn{5}; }
1361 def VLD4DUPd8x2_UPD : VLD4DUPWB<{0,0,1,0}, "8">;
1362 def VLD4DUPd16x2_UPD : VLD4DUPWB<{0,1,1,?}, "16">;
1363 def VLD4DUPd32x2_UPD : VLD4DUPWB<{1,?,1,?}, "32"> { let Inst{6} = Rn{5}; }
1365 def VLD4DUPd8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1366 def VLD4DUPd16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1367 def VLD4DUPd32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1369 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
1371 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
1373 // Classes for VST* pseudo-instructions with multi-register operands.
1374 // These are expanded to real instructions after register allocation.
1375 class VSTQPseudo<InstrItinClass itin>
1376 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src), itin, "">;
1377 class VSTQWBPseudo<InstrItinClass itin>
1378 : PseudoNLdSt<(outs GPR:$wb),
1379 (ins addrmode6:$addr, am6offset:$offset, QPR:$src), itin,
1380 "$addr.addr = $wb">;
1381 class VSTQWBfixedPseudo<InstrItinClass itin>
1382 : PseudoNLdSt<(outs GPR:$wb),
1383 (ins addrmode6:$addr, QPR:$src), itin,
1384 "$addr.addr = $wb">;
1385 class VSTQWBregisterPseudo<InstrItinClass itin>
1386 : PseudoNLdSt<(outs GPR:$wb),
1387 (ins addrmode6:$addr, rGPR:$offset, QPR:$src), itin,
1388 "$addr.addr = $wb">;
1389 class VSTQQPseudo<InstrItinClass itin>
1390 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src), itin, "">;
1391 class VSTQQWBPseudo<InstrItinClass itin>
1392 : PseudoNLdSt<(outs GPR:$wb),
1393 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src), itin,
1394 "$addr.addr = $wb">;
1395 class VSTQQQQPseudo<InstrItinClass itin>
1396 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src), itin, "">;
1397 class VSTQQQQWBPseudo<InstrItinClass itin>
1398 : PseudoNLdSt<(outs GPR:$wb),
1399 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
1400 "$addr.addr = $wb">;
1402 // VST1 : Vector Store (multiple single elements)
1403 class VST1D<bits<4> op7_4, string Dt>
1404 : NLdSt<0,0b00,0b0111,op7_4, (outs), (ins addrmode6:$Rn, VecListOneD:$Vd),
1405 IIC_VST1, "vst1", Dt, "$Vd, $Rn", "", []> {
1407 let Inst{4} = Rn{4};
1408 let DecoderMethod = "DecodeVSTInstruction";
1410 class VST1Q<bits<4> op7_4, string Dt>
1411 : NLdSt<0,0b00,0b1010,op7_4, (outs), (ins addrmode6:$Rn, VecListTwoD:$Vd),
1412 IIC_VST1x2, "vst1", Dt, "$Vd, $Rn", "", []> {
1414 let Inst{5-4} = Rn{5-4};
1415 let DecoderMethod = "DecodeVSTInstruction";
1418 def VST1d8 : VST1D<{0,0,0,?}, "8">;
1419 def VST1d16 : VST1D<{0,1,0,?}, "16">;
1420 def VST1d32 : VST1D<{1,0,0,?}, "32">;
1421 def VST1d64 : VST1D<{1,1,0,?}, "64">;
1423 def VST1q8 : VST1Q<{0,0,?,?}, "8">;
1424 def VST1q16 : VST1Q<{0,1,?,?}, "16">;
1425 def VST1q32 : VST1Q<{1,0,?,?}, "32">;
1426 def VST1q64 : VST1Q<{1,1,?,?}, "64">;
1428 def VST1q8Pseudo : VSTQPseudo<IIC_VST1x2>;
1429 def VST1q16Pseudo : VSTQPseudo<IIC_VST1x2>;
1430 def VST1q32Pseudo : VSTQPseudo<IIC_VST1x2>;
1431 def VST1q64Pseudo : VSTQPseudo<IIC_VST1x2>;
1433 // ...with address register writeback:
1434 multiclass VST1DWB<bits<4> op7_4, string Dt> {
1435 def _fixed : NLdSt<0,0b00, 0b0111,op7_4, (outs GPR:$wb),
1436 (ins addrmode6:$Rn, VecListOneD:$Vd), IIC_VLD1u,
1437 "vst1", Dt, "$Vd, $Rn!",
1438 "$Rn.addr = $wb", []> {
1439 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1440 let Inst{4} = Rn{4};
1441 let DecoderMethod = "DecodeVSTInstruction";
1442 let AsmMatchConverter = "cvtVSTwbFixed";
1444 def _register : NLdSt<0,0b00,0b0111,op7_4, (outs GPR:$wb),
1445 (ins addrmode6:$Rn, rGPR:$Rm, VecListOneD:$Vd),
1447 "vst1", Dt, "$Vd, $Rn, $Rm",
1448 "$Rn.addr = $wb", []> {
1449 let Inst{4} = Rn{4};
1450 let DecoderMethod = "DecodeVSTInstruction";
1451 let AsmMatchConverter = "cvtVSTwbRegister";
1454 multiclass VST1QWB<bits<4> op7_4, string Dt> {
1455 def _fixed : NLdSt<0,0b00,0b1010,op7_4, (outs GPR:$wb),
1456 (ins addrmode6:$Rn, VecListTwoD:$Vd), IIC_VLD1x2u,
1457 "vst1", Dt, "$Vd, $Rn!",
1458 "$Rn.addr = $wb", []> {
1459 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1460 let Inst{5-4} = Rn{5-4};
1461 let DecoderMethod = "DecodeVSTInstruction";
1462 let AsmMatchConverter = "cvtVSTwbFixed";
1464 def _register : NLdSt<0,0b00,0b1010,op7_4, (outs GPR:$wb),
1465 (ins addrmode6:$Rn, rGPR:$Rm, VecListTwoD:$Vd),
1467 "vst1", Dt, "$Vd, $Rn, $Rm",
1468 "$Rn.addr = $wb", []> {
1469 let Inst{5-4} = Rn{5-4};
1470 let DecoderMethod = "DecodeVSTInstruction";
1471 let AsmMatchConverter = "cvtVSTwbRegister";
1475 defm VST1d8wb : VST1DWB<{0,0,0,?}, "8">;
1476 defm VST1d16wb : VST1DWB<{0,1,0,?}, "16">;
1477 defm VST1d32wb : VST1DWB<{1,0,0,?}, "32">;
1478 defm VST1d64wb : VST1DWB<{1,1,0,?}, "64">;
1480 defm VST1q8wb : VST1QWB<{0,0,?,?}, "8">;
1481 defm VST1q16wb : VST1QWB<{0,1,?,?}, "16">;
1482 defm VST1q32wb : VST1QWB<{1,0,?,?}, "32">;
1483 defm VST1q64wb : VST1QWB<{1,1,?,?}, "64">;
1485 def VST1q8PseudoWB_fixed : VSTQWBfixedPseudo<IIC_VST1x2u>;
1486 def VST1q16PseudoWB_fixed : VSTQWBfixedPseudo<IIC_VST1x2u>;
1487 def VST1q32PseudoWB_fixed : VSTQWBfixedPseudo<IIC_VST1x2u>;
1488 def VST1q64PseudoWB_fixed : VSTQWBfixedPseudo<IIC_VST1x2u>;
1489 def VST1q8PseudoWB_register : VSTQWBregisterPseudo<IIC_VST1x2u>;
1490 def VST1q16PseudoWB_register : VSTQWBregisterPseudo<IIC_VST1x2u>;
1491 def VST1q32PseudoWB_register : VSTQWBregisterPseudo<IIC_VST1x2u>;
1492 def VST1q64PseudoWB_register : VSTQWBregisterPseudo<IIC_VST1x2u>;
1494 // ...with 3 registers
1495 class VST1D3<bits<4> op7_4, string Dt>
1496 : NLdSt<0, 0b00, 0b0110, op7_4, (outs),
1497 (ins addrmode6:$Rn, VecListThreeD:$Vd),
1498 IIC_VST1x3, "vst1", Dt, "$Vd, $Rn", "", []> {
1500 let Inst{4} = Rn{4};
1501 let DecoderMethod = "DecodeVSTInstruction";
1503 multiclass VST1D3WB<bits<4> op7_4, string Dt> {
1504 def _fixed : NLdSt<0,0b00,0b0110,op7_4, (outs GPR:$wb),
1505 (ins addrmode6:$Rn, VecListThreeD:$Vd), IIC_VLD1x3u,
1506 "vst1", Dt, "$Vd, $Rn!",
1507 "$Rn.addr = $wb", []> {
1508 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1509 let Inst{5-4} = Rn{5-4};
1510 let DecoderMethod = "DecodeVSTInstruction";
1511 let AsmMatchConverter = "cvtVSTwbFixed";
1513 def _register : NLdSt<0,0b00,0b0110,op7_4, (outs GPR:$wb),
1514 (ins addrmode6:$Rn, rGPR:$Rm, VecListThreeD:$Vd),
1516 "vst1", Dt, "$Vd, $Rn, $Rm",
1517 "$Rn.addr = $wb", []> {
1518 let Inst{5-4} = Rn{5-4};
1519 let DecoderMethod = "DecodeVSTInstruction";
1520 let AsmMatchConverter = "cvtVSTwbRegister";
1524 def VST1d8T : VST1D3<{0,0,0,?}, "8">;
1525 def VST1d16T : VST1D3<{0,1,0,?}, "16">;
1526 def VST1d32T : VST1D3<{1,0,0,?}, "32">;
1527 def VST1d64T : VST1D3<{1,1,0,?}, "64">;
1529 defm VST1d8Twb : VST1D3WB<{0,0,0,?}, "8">;
1530 defm VST1d16Twb : VST1D3WB<{0,1,0,?}, "16">;
1531 defm VST1d32Twb : VST1D3WB<{1,0,0,?}, "32">;
1532 defm VST1d64Twb : VST1D3WB<{1,1,0,?}, "64">;
1534 def VST1d64TPseudo : VSTQQPseudo<IIC_VST1x3>;
1535 def VST1d64TPseudoWB_fixed : VSTQQWBPseudo<IIC_VST1x3u>;
1536 def VST1d64TPseudoWB_register : VSTQQWBPseudo<IIC_VST1x3u>;
1538 // ...with 4 registers
1539 class VST1D4<bits<4> op7_4, string Dt>
1540 : NLdSt<0, 0b00, 0b0010, op7_4, (outs),
1541 (ins addrmode6:$Rn, VecListFourD:$Vd),
1542 IIC_VST1x4, "vst1", Dt, "$Vd, $Rn", "",
1545 let Inst{5-4} = Rn{5-4};
1546 let DecoderMethod = "DecodeVSTInstruction";
1548 multiclass VST1D4WB<bits<4> op7_4, string Dt> {
1549 def _fixed : NLdSt<0,0b00,0b0010,op7_4, (outs GPR:$wb),
1550 (ins addrmode6:$Rn, VecListFourD:$Vd), IIC_VLD1x4u,
1551 "vst1", Dt, "$Vd, $Rn!",
1552 "$Rn.addr = $wb", []> {
1553 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1554 let Inst{5-4} = Rn{5-4};
1555 let DecoderMethod = "DecodeVSTInstruction";
1556 let AsmMatchConverter = "cvtVSTwbFixed";
1558 def _register : NLdSt<0,0b00,0b0010,op7_4, (outs GPR:$wb),
1559 (ins addrmode6:$Rn, rGPR:$Rm, VecListFourD:$Vd),
1561 "vst1", Dt, "$Vd, $Rn, $Rm",
1562 "$Rn.addr = $wb", []> {
1563 let Inst{5-4} = Rn{5-4};
1564 let DecoderMethod = "DecodeVSTInstruction";
1565 let AsmMatchConverter = "cvtVSTwbRegister";
1569 def VST1d8Q : VST1D4<{0,0,?,?}, "8">;
1570 def VST1d16Q : VST1D4<{0,1,?,?}, "16">;
1571 def VST1d32Q : VST1D4<{1,0,?,?}, "32">;
1572 def VST1d64Q : VST1D4<{1,1,?,?}, "64">;
1574 defm VST1d8Qwb : VST1D4WB<{0,0,?,?}, "8">;
1575 defm VST1d16Qwb : VST1D4WB<{0,1,?,?}, "16">;
1576 defm VST1d32Qwb : VST1D4WB<{1,0,?,?}, "32">;
1577 defm VST1d64Qwb : VST1D4WB<{1,1,?,?}, "64">;
1579 def VST1d64QPseudo : VSTQQPseudo<IIC_VST1x4>;
1580 def VST1d64QPseudoWB_fixed : VSTQQWBPseudo<IIC_VST1x4u>;
1581 def VST1d64QPseudoWB_register : VSTQQWBPseudo<IIC_VST1x4u>;
1583 // VST2 : Vector Store (multiple 2-element structures)
1584 class VST2<bits<4> op11_8, bits<4> op7_4, string Dt, RegisterOperand VdTy,
1585 InstrItinClass itin>
1586 : NLdSt<0, 0b00, op11_8, op7_4, (outs), (ins addrmode6:$Rn, VdTy:$Vd),
1587 itin, "vst2", Dt, "$Vd, $Rn", "", []> {
1589 let Inst{5-4} = Rn{5-4};
1590 let DecoderMethod = "DecodeVSTInstruction";
1593 def VST2d8 : VST2<0b1000, {0,0,?,?}, "8", VecListTwoD, IIC_VST2>;
1594 def VST2d16 : VST2<0b1000, {0,1,?,?}, "16", VecListTwoD, IIC_VST2>;
1595 def VST2d32 : VST2<0b1000, {1,0,?,?}, "32", VecListTwoD, IIC_VST2>;
1597 def VST2q8 : VST2<0b0011, {0,0,?,?}, "8", VecListFourD, IIC_VST2x2>;
1598 def VST2q16 : VST2<0b0011, {0,1,?,?}, "16", VecListFourD, IIC_VST2x2>;
1599 def VST2q32 : VST2<0b0011, {1,0,?,?}, "32", VecListFourD, IIC_VST2x2>;
1601 def VST2d8Pseudo : VSTQPseudo<IIC_VST2>;
1602 def VST2d16Pseudo : VSTQPseudo<IIC_VST2>;
1603 def VST2d32Pseudo : VSTQPseudo<IIC_VST2>;
1605 def VST2q8Pseudo : VSTQQPseudo<IIC_VST2x2>;
1606 def VST2q16Pseudo : VSTQQPseudo<IIC_VST2x2>;
1607 def VST2q32Pseudo : VSTQQPseudo<IIC_VST2x2>;
1609 // ...with address register writeback:
1610 multiclass VST2DWB<bits<4> op11_8, bits<4> op7_4, string Dt,
1611 RegisterOperand VdTy> {
1612 def _fixed : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1613 (ins addrmode6:$Rn, VdTy:$Vd), IIC_VLD1u,
1614 "vst2", Dt, "$Vd, $Rn!",
1615 "$Rn.addr = $wb", []> {
1616 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1617 let Inst{5-4} = Rn{5-4};
1618 let DecoderMethod = "DecodeVSTInstruction";
1619 let AsmMatchConverter = "cvtVSTwbFixed";
1621 def _register : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1622 (ins addrmode6:$Rn, rGPR:$Rm, VdTy:$Vd), IIC_VLD1u,
1623 "vst2", Dt, "$Vd, $Rn, $Rm",
1624 "$Rn.addr = $wb", []> {
1625 let Inst{5-4} = Rn{5-4};
1626 let DecoderMethod = "DecodeVSTInstruction";
1627 let AsmMatchConverter = "cvtVSTwbRegister";
1630 multiclass VST2QWB<bits<4> op7_4, string Dt> {
1631 def _fixed : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
1632 (ins addrmode6:$Rn, VecListFourD:$Vd), IIC_VLD1u,
1633 "vst2", Dt, "$Vd, $Rn!",
1634 "$Rn.addr = $wb", []> {
1635 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1636 let Inst{5-4} = Rn{5-4};
1637 let DecoderMethod = "DecodeVSTInstruction";
1638 let AsmMatchConverter = "cvtVSTwbFixed";
1640 def _register : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
1641 (ins addrmode6:$Rn, rGPR:$Rm, VecListFourD:$Vd),
1643 "vst2", Dt, "$Vd, $Rn, $Rm",
1644 "$Rn.addr = $wb", []> {
1645 let Inst{5-4} = Rn{5-4};
1646 let DecoderMethod = "DecodeVSTInstruction";
1647 let AsmMatchConverter = "cvtVSTwbRegister";
1651 defm VST2d8wb : VST2DWB<0b1000, {0,0,?,?}, "8", VecListTwoD>;
1652 defm VST2d16wb : VST2DWB<0b1000, {0,1,?,?}, "16", VecListTwoD>;
1653 defm VST2d32wb : VST2DWB<0b1000, {1,0,?,?}, "32", VecListTwoD>;
1655 defm VST2q8wb : VST2QWB<{0,0,?,?}, "8">;
1656 defm VST2q16wb : VST2QWB<{0,1,?,?}, "16">;
1657 defm VST2q32wb : VST2QWB<{1,0,?,?}, "32">;
1659 def VST2d8PseudoWB_fixed : VSTQWBPseudo<IIC_VST2u>;
1660 def VST2d16PseudoWB_fixed : VSTQWBPseudo<IIC_VST2u>;
1661 def VST2d32PseudoWB_fixed : VSTQWBPseudo<IIC_VST2u>;
1662 def VST2d8PseudoWB_register : VSTQWBPseudo<IIC_VST2u>;
1663 def VST2d16PseudoWB_register : VSTQWBPseudo<IIC_VST2u>;
1664 def VST2d32PseudoWB_register : VSTQWBPseudo<IIC_VST2u>;
1666 def VST2q8PseudoWB_fixed : VSTQQWBPseudo<IIC_VST2x2u>;
1667 def VST2q16PseudoWB_fixed : VSTQQWBPseudo<IIC_VST2x2u>;
1668 def VST2q32PseudoWB_fixed : VSTQQWBPseudo<IIC_VST2x2u>;
1669 def VST2q8PseudoWB_register : VSTQQWBPseudo<IIC_VST2x2u>;
1670 def VST2q16PseudoWB_register : VSTQQWBPseudo<IIC_VST2x2u>;
1671 def VST2q32PseudoWB_register : VSTQQWBPseudo<IIC_VST2x2u>;
1673 // ...with double-spaced registers
1674 def VST2b8 : VST2<0b1001, {0,0,?,?}, "8", VecListTwoQ, IIC_VST2>;
1675 def VST2b16 : VST2<0b1001, {0,1,?,?}, "16", VecListTwoQ, IIC_VST2>;
1676 def VST2b32 : VST2<0b1001, {1,0,?,?}, "32", VecListTwoQ, IIC_VST2>;
1677 defm VST2b8wb : VST2DWB<0b1001, {0,0,?,?}, "8", VecListTwoQ>;
1678 defm VST2b16wb : VST2DWB<0b1001, {0,1,?,?}, "16", VecListTwoQ>;
1679 defm VST2b32wb : VST2DWB<0b1001, {1,0,?,?}, "32", VecListTwoQ>;
1681 // VST3 : Vector Store (multiple 3-element structures)
1682 class VST3D<bits<4> op11_8, bits<4> op7_4, string Dt>
1683 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1684 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3,
1685 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn", "", []> {
1687 let Inst{4} = Rn{4};
1688 let DecoderMethod = "DecodeVSTInstruction";
1691 def VST3d8 : VST3D<0b0100, {0,0,0,?}, "8">;
1692 def VST3d16 : VST3D<0b0100, {0,1,0,?}, "16">;
1693 def VST3d32 : VST3D<0b0100, {1,0,0,?}, "32">;
1695 def VST3d8Pseudo : VSTQQPseudo<IIC_VST3>;
1696 def VST3d16Pseudo : VSTQQPseudo<IIC_VST3>;
1697 def VST3d32Pseudo : VSTQQPseudo<IIC_VST3>;
1699 // ...with address register writeback:
1700 class VST3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1701 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1702 (ins addrmode6:$Rn, am6offset:$Rm,
1703 DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3u,
1704 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn$Rm",
1705 "$Rn.addr = $wb", []> {
1706 let Inst{4} = Rn{4};
1707 let DecoderMethod = "DecodeVSTInstruction";
1710 def VST3d8_UPD : VST3DWB<0b0100, {0,0,0,?}, "8">;
1711 def VST3d16_UPD : VST3DWB<0b0100, {0,1,0,?}, "16">;
1712 def VST3d32_UPD : VST3DWB<0b0100, {1,0,0,?}, "32">;
1714 def VST3d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1715 def VST3d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1716 def VST3d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1718 // ...with double-spaced registers:
1719 def VST3q8 : VST3D<0b0101, {0,0,0,?}, "8">;
1720 def VST3q16 : VST3D<0b0101, {0,1,0,?}, "16">;
1721 def VST3q32 : VST3D<0b0101, {1,0,0,?}, "32">;
1722 def VST3q8_UPD : VST3DWB<0b0101, {0,0,0,?}, "8">;
1723 def VST3q16_UPD : VST3DWB<0b0101, {0,1,0,?}, "16">;
1724 def VST3q32_UPD : VST3DWB<0b0101, {1,0,0,?}, "32">;
1726 def VST3q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1727 def VST3q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1728 def VST3q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1730 // ...alternate versions to be allocated odd register numbers:
1731 def VST3q8oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1732 def VST3q16oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1733 def VST3q32oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1735 def VST3q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1736 def VST3q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1737 def VST3q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1739 // VST4 : Vector Store (multiple 4-element structures)
1740 class VST4D<bits<4> op11_8, bits<4> op7_4, string Dt>
1741 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1742 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
1743 IIC_VST4, "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn",
1746 let Inst{5-4} = Rn{5-4};
1747 let DecoderMethod = "DecodeVSTInstruction";
1750 def VST4d8 : VST4D<0b0000, {0,0,?,?}, "8">;
1751 def VST4d16 : VST4D<0b0000, {0,1,?,?}, "16">;
1752 def VST4d32 : VST4D<0b0000, {1,0,?,?}, "32">;
1754 def VST4d8Pseudo : VSTQQPseudo<IIC_VST4>;
1755 def VST4d16Pseudo : VSTQQPseudo<IIC_VST4>;
1756 def VST4d32Pseudo : VSTQQPseudo<IIC_VST4>;
1758 // ...with address register writeback:
1759 class VST4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1760 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1761 (ins addrmode6:$Rn, am6offset:$Rm,
1762 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST4u,
1763 "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
1764 "$Rn.addr = $wb", []> {
1765 let Inst{5-4} = Rn{5-4};
1766 let DecoderMethod = "DecodeVSTInstruction";
1769 def VST4d8_UPD : VST4DWB<0b0000, {0,0,?,?}, "8">;
1770 def VST4d16_UPD : VST4DWB<0b0000, {0,1,?,?}, "16">;
1771 def VST4d32_UPD : VST4DWB<0b0000, {1,0,?,?}, "32">;
1773 def VST4d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1774 def VST4d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1775 def VST4d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1777 // ...with double-spaced registers:
1778 def VST4q8 : VST4D<0b0001, {0,0,?,?}, "8">;
1779 def VST4q16 : VST4D<0b0001, {0,1,?,?}, "16">;
1780 def VST4q32 : VST4D<0b0001, {1,0,?,?}, "32">;
1781 def VST4q8_UPD : VST4DWB<0b0001, {0,0,?,?}, "8">;
1782 def VST4q16_UPD : VST4DWB<0b0001, {0,1,?,?}, "16">;
1783 def VST4q32_UPD : VST4DWB<0b0001, {1,0,?,?}, "32">;
1785 def VST4q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1786 def VST4q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1787 def VST4q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1789 // ...alternate versions to be allocated odd register numbers:
1790 def VST4q8oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1791 def VST4q16oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1792 def VST4q32oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1794 def VST4q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1795 def VST4q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1796 def VST4q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1798 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
1800 // Classes for VST*LN pseudo-instructions with multi-register operands.
1801 // These are expanded to real instructions after register allocation.
1802 class VSTQLNPseudo<InstrItinClass itin>
1803 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
1805 class VSTQLNWBPseudo<InstrItinClass itin>
1806 : PseudoNLdSt<(outs GPR:$wb),
1807 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
1808 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1809 class VSTQQLNPseudo<InstrItinClass itin>
1810 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
1812 class VSTQQLNWBPseudo<InstrItinClass itin>
1813 : PseudoNLdSt<(outs GPR:$wb),
1814 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
1815 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1816 class VSTQQQQLNPseudo<InstrItinClass itin>
1817 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
1819 class VSTQQQQLNWBPseudo<InstrItinClass itin>
1820 : PseudoNLdSt<(outs GPR:$wb),
1821 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
1822 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1824 // VST1LN : Vector Store (single element from one lane)
1825 class VST1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1826 PatFrag StoreOp, SDNode ExtractOp>
1827 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1828 (ins addrmode6:$Rn, DPR:$Vd, nohash_imm:$lane),
1829 IIC_VST1ln, "vst1", Dt, "\\{$Vd[$lane]\\}, $Rn", "",
1830 [(StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane), addrmode6:$Rn)]> {
1832 let DecoderMethod = "DecodeVST1LN";
1834 class VST1LN32<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1835 PatFrag StoreOp, SDNode ExtractOp>
1836 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1837 (ins addrmode6oneL32:$Rn, DPR:$Vd, nohash_imm:$lane),
1838 IIC_VST1ln, "vst1", Dt, "\\{$Vd[$lane]\\}, $Rn", "",
1839 [(StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane), addrmode6oneL32:$Rn)]>{
1841 let DecoderMethod = "DecodeVST1LN";
1843 class VST1QLNPseudo<ValueType Ty, PatFrag StoreOp, SDNode ExtractOp>
1844 : VSTQLNPseudo<IIC_VST1ln> {
1845 let Pattern = [(StoreOp (ExtractOp (Ty QPR:$src), imm:$lane),
1849 def VST1LNd8 : VST1LN<0b0000, {?,?,?,0}, "8", v8i8, truncstorei8,
1851 let Inst{7-5} = lane{2-0};
1853 def VST1LNd16 : VST1LN<0b0100, {?,?,0,?}, "16", v4i16, truncstorei16,
1855 let Inst{7-6} = lane{1-0};
1856 let Inst{4} = Rn{5};
1859 def VST1LNd32 : VST1LN32<0b1000, {?,0,?,?}, "32", v2i32, store, extractelt> {
1860 let Inst{7} = lane{0};
1861 let Inst{5-4} = Rn{5-4};
1864 def VST1LNq8Pseudo : VST1QLNPseudo<v16i8, truncstorei8, NEONvgetlaneu>;
1865 def VST1LNq16Pseudo : VST1QLNPseudo<v8i16, truncstorei16, NEONvgetlaneu>;
1866 def VST1LNq32Pseudo : VST1QLNPseudo<v4i32, store, extractelt>;
1868 def : Pat<(store (extractelt (v2f32 DPR:$src), imm:$lane), addrmode6:$addr),
1869 (VST1LNd32 addrmode6:$addr, DPR:$src, imm:$lane)>;
1870 def : Pat<(store (extractelt (v4f32 QPR:$src), imm:$lane), addrmode6:$addr),
1871 (VST1LNq32Pseudo addrmode6:$addr, QPR:$src, imm:$lane)>;
1873 // ...with address register writeback:
1874 class VST1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1875 PatFrag StoreOp, SDNode ExtractOp>
1876 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1877 (ins addrmode6:$Rn, am6offset:$Rm,
1878 DPR:$Vd, nohash_imm:$lane), IIC_VST1lnu, "vst1", Dt,
1879 "\\{$Vd[$lane]\\}, $Rn$Rm",
1881 [(set GPR:$wb, (StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane),
1882 addrmode6:$Rn, am6offset:$Rm))]> {
1883 let DecoderMethod = "DecodeVST1LN";
1885 class VST1QLNWBPseudo<ValueType Ty, PatFrag StoreOp, SDNode ExtractOp>
1886 : VSTQLNWBPseudo<IIC_VST1lnu> {
1887 let Pattern = [(set GPR:$wb, (StoreOp (ExtractOp (Ty QPR:$src), imm:$lane),
1888 addrmode6:$addr, am6offset:$offset))];
1891 def VST1LNd8_UPD : VST1LNWB<0b0000, {?,?,?,0}, "8", v8i8, post_truncsti8,
1893 let Inst{7-5} = lane{2-0};
1895 def VST1LNd16_UPD : VST1LNWB<0b0100, {?,?,0,?}, "16", v4i16, post_truncsti16,
1897 let Inst{7-6} = lane{1-0};
1898 let Inst{4} = Rn{5};
1900 def VST1LNd32_UPD : VST1LNWB<0b1000, {?,0,?,?}, "32", v2i32, post_store,
1902 let Inst{7} = lane{0};
1903 let Inst{5-4} = Rn{5-4};
1906 def VST1LNq8Pseudo_UPD : VST1QLNWBPseudo<v16i8, post_truncsti8, NEONvgetlaneu>;
1907 def VST1LNq16Pseudo_UPD : VST1QLNWBPseudo<v8i16, post_truncsti16,NEONvgetlaneu>;
1908 def VST1LNq32Pseudo_UPD : VST1QLNWBPseudo<v4i32, post_store, extractelt>;
1910 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
1912 // VST2LN : Vector Store (single 2-element structure from one lane)
1913 class VST2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1914 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1915 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, nohash_imm:$lane),
1916 IIC_VST2ln, "vst2", Dt, "\\{$Vd[$lane], $src2[$lane]\\}, $Rn",
1919 let Inst{4} = Rn{4};
1920 let DecoderMethod = "DecodeVST2LN";
1923 def VST2LNd8 : VST2LN<0b0001, {?,?,?,?}, "8"> {
1924 let Inst{7-5} = lane{2-0};
1926 def VST2LNd16 : VST2LN<0b0101, {?,?,0,?}, "16"> {
1927 let Inst{7-6} = lane{1-0};
1929 def VST2LNd32 : VST2LN<0b1001, {?,0,0,?}, "32"> {
1930 let Inst{7} = lane{0};
1933 def VST2LNd8Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1934 def VST2LNd16Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1935 def VST2LNd32Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1937 // ...with double-spaced registers:
1938 def VST2LNq16 : VST2LN<0b0101, {?,?,1,?}, "16"> {
1939 let Inst{7-6} = lane{1-0};
1940 let Inst{4} = Rn{4};
1942 def VST2LNq32 : VST2LN<0b1001, {?,1,0,?}, "32"> {
1943 let Inst{7} = lane{0};
1944 let Inst{4} = Rn{4};
1947 def VST2LNq16Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
1948 def VST2LNq32Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
1950 // ...with address register writeback:
1951 class VST2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1952 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1953 (ins addrmode6:$Rn, am6offset:$Rm,
1954 DPR:$Vd, DPR:$src2, nohash_imm:$lane), IIC_VST2lnu, "vst2", Dt,
1955 "\\{$Vd[$lane], $src2[$lane]\\}, $Rn$Rm",
1956 "$Rn.addr = $wb", []> {
1957 let Inst{4} = Rn{4};
1958 let DecoderMethod = "DecodeVST2LN";
1961 def VST2LNd8_UPD : VST2LNWB<0b0001, {?,?,?,?}, "8"> {
1962 let Inst{7-5} = lane{2-0};
1964 def VST2LNd16_UPD : VST2LNWB<0b0101, {?,?,0,?}, "16"> {
1965 let Inst{7-6} = lane{1-0};
1967 def VST2LNd32_UPD : VST2LNWB<0b1001, {?,0,0,?}, "32"> {
1968 let Inst{7} = lane{0};
1971 def VST2LNd8Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1972 def VST2LNd16Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1973 def VST2LNd32Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1975 def VST2LNq16_UPD : VST2LNWB<0b0101, {?,?,1,?}, "16"> {
1976 let Inst{7-6} = lane{1-0};
1978 def VST2LNq32_UPD : VST2LNWB<0b1001, {?,1,0,?}, "32"> {
1979 let Inst{7} = lane{0};
1982 def VST2LNq16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
1983 def VST2LNq32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
1985 // VST3LN : Vector Store (single 3-element structure from one lane)
1986 class VST3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1987 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1988 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3,
1989 nohash_imm:$lane), IIC_VST3ln, "vst3", Dt,
1990 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn", "", []> {
1992 let DecoderMethod = "DecodeVST3LN";
1995 def VST3LNd8 : VST3LN<0b0010, {?,?,?,0}, "8"> {
1996 let Inst{7-5} = lane{2-0};
1998 def VST3LNd16 : VST3LN<0b0110, {?,?,0,0}, "16"> {
1999 let Inst{7-6} = lane{1-0};
2001 def VST3LNd32 : VST3LN<0b1010, {?,0,0,0}, "32"> {
2002 let Inst{7} = lane{0};
2005 def VST3LNd8Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
2006 def VST3LNd16Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
2007 def VST3LNd32Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
2009 // ...with double-spaced registers:
2010 def VST3LNq16 : VST3LN<0b0110, {?,?,1,0}, "16"> {
2011 let Inst{7-6} = lane{1-0};
2013 def VST3LNq32 : VST3LN<0b1010, {?,1,0,0}, "32"> {
2014 let Inst{7} = lane{0};
2017 def VST3LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
2018 def VST3LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
2020 // ...with address register writeback:
2021 class VST3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
2022 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
2023 (ins addrmode6:$Rn, am6offset:$Rm,
2024 DPR:$Vd, DPR:$src2, DPR:$src3, nohash_imm:$lane),
2025 IIC_VST3lnu, "vst3", Dt,
2026 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn$Rm",
2027 "$Rn.addr = $wb", []> {
2028 let DecoderMethod = "DecodeVST3LN";
2031 def VST3LNd8_UPD : VST3LNWB<0b0010, {?,?,?,0}, "8"> {
2032 let Inst{7-5} = lane{2-0};
2034 def VST3LNd16_UPD : VST3LNWB<0b0110, {?,?,0,0}, "16"> {
2035 let Inst{7-6} = lane{1-0};
2037 def VST3LNd32_UPD : VST3LNWB<0b1010, {?,0,0,0}, "32"> {
2038 let Inst{7} = lane{0};
2041 def VST3LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
2042 def VST3LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
2043 def VST3LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
2045 def VST3LNq16_UPD : VST3LNWB<0b0110, {?,?,1,0}, "16"> {
2046 let Inst{7-6} = lane{1-0};
2048 def VST3LNq32_UPD : VST3LNWB<0b1010, {?,1,0,0}, "32"> {
2049 let Inst{7} = lane{0};
2052 def VST3LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
2053 def VST3LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
2055 // VST4LN : Vector Store (single 4-element structure from one lane)
2056 class VST4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
2057 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
2058 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4,
2059 nohash_imm:$lane), IIC_VST4ln, "vst4", Dt,
2060 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn",
2063 let Inst{4} = Rn{4};
2064 let DecoderMethod = "DecodeVST4LN";
2067 def VST4LNd8 : VST4LN<0b0011, {?,?,?,?}, "8"> {
2068 let Inst{7-5} = lane{2-0};
2070 def VST4LNd16 : VST4LN<0b0111, {?,?,0,?}, "16"> {
2071 let Inst{7-6} = lane{1-0};
2073 def VST4LNd32 : VST4LN<0b1011, {?,0,?,?}, "32"> {
2074 let Inst{7} = lane{0};
2075 let Inst{5} = Rn{5};
2078 def VST4LNd8Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
2079 def VST4LNd16Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
2080 def VST4LNd32Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
2082 // ...with double-spaced registers:
2083 def VST4LNq16 : VST4LN<0b0111, {?,?,1,?}, "16"> {
2084 let Inst{7-6} = lane{1-0};
2086 def VST4LNq32 : VST4LN<0b1011, {?,1,?,?}, "32"> {
2087 let Inst{7} = lane{0};
2088 let Inst{5} = Rn{5};
2091 def VST4LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
2092 def VST4LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
2094 // ...with address register writeback:
2095 class VST4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
2096 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
2097 (ins addrmode6:$Rn, am6offset:$Rm,
2098 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
2099 IIC_VST4lnu, "vst4", Dt,
2100 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn$Rm",
2101 "$Rn.addr = $wb", []> {
2102 let Inst{4} = Rn{4};
2103 let DecoderMethod = "DecodeVST4LN";
2106 def VST4LNd8_UPD : VST4LNWB<0b0011, {?,?,?,?}, "8"> {
2107 let Inst{7-5} = lane{2-0};
2109 def VST4LNd16_UPD : VST4LNWB<0b0111, {?,?,0,?}, "16"> {
2110 let Inst{7-6} = lane{1-0};
2112 def VST4LNd32_UPD : VST4LNWB<0b1011, {?,0,?,?}, "32"> {
2113 let Inst{7} = lane{0};
2114 let Inst{5} = Rn{5};
2117 def VST4LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
2118 def VST4LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
2119 def VST4LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
2121 def VST4LNq16_UPD : VST4LNWB<0b0111, {?,?,1,?}, "16"> {
2122 let Inst{7-6} = lane{1-0};
2124 def VST4LNq32_UPD : VST4LNWB<0b1011, {?,1,?,?}, "32"> {
2125 let Inst{7} = lane{0};
2126 let Inst{5} = Rn{5};
2129 def VST4LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
2130 def VST4LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
2132 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
2135 //===----------------------------------------------------------------------===//
2136 // NEON pattern fragments
2137 //===----------------------------------------------------------------------===//
2139 // Extract D sub-registers of Q registers.
2140 def DSubReg_i8_reg : SDNodeXForm<imm, [{
2141 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
2142 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/8, MVT::i32);
2144 def DSubReg_i16_reg : SDNodeXForm<imm, [{
2145 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
2146 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/4, MVT::i32);
2148 def DSubReg_i32_reg : SDNodeXForm<imm, [{
2149 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
2150 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/2, MVT::i32);
2152 def DSubReg_f64_reg : SDNodeXForm<imm, [{
2153 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
2154 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue(), MVT::i32);
2157 // Extract S sub-registers of Q/D registers.
2158 def SSubReg_f32_reg : SDNodeXForm<imm, [{
2159 assert(ARM::ssub_3 == ARM::ssub_0+3 && "Unexpected subreg numbering");
2160 return CurDAG->getTargetConstant(ARM::ssub_0 + N->getZExtValue(), MVT::i32);
2163 // Translate lane numbers from Q registers to D subregs.
2164 def SubReg_i8_lane : SDNodeXForm<imm, [{
2165 return CurDAG->getTargetConstant(N->getZExtValue() & 7, MVT::i32);
2167 def SubReg_i16_lane : SDNodeXForm<imm, [{
2168 return CurDAG->getTargetConstant(N->getZExtValue() & 3, MVT::i32);
2170 def SubReg_i32_lane : SDNodeXForm<imm, [{
2171 return CurDAG->getTargetConstant(N->getZExtValue() & 1, MVT::i32);
2174 //===----------------------------------------------------------------------===//
2175 // Instruction Classes
2176 //===----------------------------------------------------------------------===//
2178 // Basic 2-register operations: double- and quad-register.
2179 class N2VD<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2180 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
2181 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
2182 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
2183 (ins DPR:$Vm), IIC_VUNAD, OpcodeStr, Dt,"$Vd, $Vm", "",
2184 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vm))))]>;
2185 class N2VQ<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2186 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
2187 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
2188 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
2189 (ins QPR:$Vm), IIC_VUNAQ, OpcodeStr, Dt,"$Vd, $Vm", "",
2190 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vm))))]>;
2192 // Basic 2-register intrinsics, both double- and quad-register.
2193 class N2VDInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2194 bits<2> op17_16, bits<5> op11_7, bit op4,
2195 InstrItinClass itin, string OpcodeStr, string Dt,
2196 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2197 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
2198 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2199 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm))))]>;
2200 class N2VQInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2201 bits<2> op17_16, bits<5> op11_7, bit op4,
2202 InstrItinClass itin, string OpcodeStr, string Dt,
2203 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2204 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
2205 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2206 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm))))]>;
2208 // Narrow 2-register operations.
2209 class N2VN<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2210 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2211 InstrItinClass itin, string OpcodeStr, string Dt,
2212 ValueType TyD, ValueType TyQ, SDNode OpNode>
2213 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$Vd),
2214 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2215 [(set DPR:$Vd, (TyD (OpNode (TyQ QPR:$Vm))))]>;
2217 // Narrow 2-register intrinsics.
2218 class N2VNInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2219 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2220 InstrItinClass itin, string OpcodeStr, string Dt,
2221 ValueType TyD, ValueType TyQ, Intrinsic IntOp>
2222 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$Vd),
2223 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2224 [(set DPR:$Vd, (TyD (IntOp (TyQ QPR:$Vm))))]>;
2226 // Long 2-register operations (currently only used for VMOVL).
2227 class N2VL<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2228 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2229 InstrItinClass itin, string OpcodeStr, string Dt,
2230 ValueType TyQ, ValueType TyD, SDNode OpNode>
2231 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$Vd),
2232 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2233 [(set QPR:$Vd, (TyQ (OpNode (TyD DPR:$Vm))))]>;
2235 // Long 2-register intrinsics.
2236 class N2VLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2237 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2238 InstrItinClass itin, string OpcodeStr, string Dt,
2239 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
2240 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$Vd),
2241 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2242 [(set QPR:$Vd, (TyQ (IntOp (TyD DPR:$Vm))))]>;
2244 // 2-register shuffles (VTRN/VZIP/VUZP), both double- and quad-register.
2245 class N2VDShuffle<bits<2> op19_18, bits<5> op11_7, string OpcodeStr, string Dt>
2246 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 0, 0, (outs DPR:$Vd, DPR:$Vm),
2247 (ins DPR:$src1, DPR:$src2), IIC_VPERMD,
2248 OpcodeStr, Dt, "$Vd, $Vm",
2249 "$src1 = $Vd, $src2 = $Vm", []>;
2250 class N2VQShuffle<bits<2> op19_18, bits<5> op11_7,
2251 InstrItinClass itin, string OpcodeStr, string Dt>
2252 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 1, 0, (outs QPR:$Vd, QPR:$Vm),
2253 (ins QPR:$src1, QPR:$src2), itin, OpcodeStr, Dt, "$Vd, $Vm",
2254 "$src1 = $Vd, $src2 = $Vm", []>;
2256 // Basic 3-register operations: double- and quad-register.
2257 class N3VD<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2258 InstrItinClass itin, string OpcodeStr, string Dt,
2259 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
2260 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2261 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2262 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2263 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
2264 let isCommutable = Commutable;
2266 // Same as N3VD but no data type.
2267 class N3VDX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2268 InstrItinClass itin, string OpcodeStr,
2269 ValueType ResTy, ValueType OpTy,
2270 SDNode OpNode, bit Commutable>
2271 : N3VX<op24, op23, op21_20, op11_8, 0, op4,
2272 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2273 OpcodeStr, "$Vd, $Vn, $Vm", "",
2274 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>{
2275 let isCommutable = Commutable;
2278 class N3VDSL<bits<2> op21_20, bits<4> op11_8,
2279 InstrItinClass itin, string OpcodeStr, string Dt,
2280 ValueType Ty, SDNode ShOp>
2281 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2282 (outs DPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2283 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2285 (Ty (ShOp (Ty DPR:$Vn),
2286 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),imm:$lane)))))]> {
2287 let isCommutable = 0;
2289 class N3VDSL16<bits<2> op21_20, bits<4> op11_8,
2290 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
2291 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2292 (outs DPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2293 NVMulSLFrm, IIC_VMULi16D, OpcodeStr, Dt,"$Vd, $Vn, $Vm$lane","",
2295 (Ty (ShOp (Ty DPR:$Vn),
2296 (Ty (NEONvduplane (Ty DPR_8:$Vm), imm:$lane)))))]> {
2297 let isCommutable = 0;
2300 class N3VQ<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2301 InstrItinClass itin, string OpcodeStr, string Dt,
2302 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
2303 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2304 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2305 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2306 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
2307 let isCommutable = Commutable;
2309 class N3VQX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2310 InstrItinClass itin, string OpcodeStr,
2311 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
2312 : N3VX<op24, op23, op21_20, op11_8, 1, op4,
2313 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2314 OpcodeStr, "$Vd, $Vn, $Vm", "",
2315 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]>{
2316 let isCommutable = Commutable;
2318 class N3VQSL<bits<2> op21_20, bits<4> op11_8,
2319 InstrItinClass itin, string OpcodeStr, string Dt,
2320 ValueType ResTy, ValueType OpTy, SDNode ShOp>
2321 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2322 (outs QPR:$Vd), (ins QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2323 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2324 [(set (ResTy QPR:$Vd),
2325 (ResTy (ShOp (ResTy QPR:$Vn),
2326 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2328 let isCommutable = 0;
2330 class N3VQSL16<bits<2> op21_20, bits<4> op11_8, string OpcodeStr, string Dt,
2331 ValueType ResTy, ValueType OpTy, SDNode ShOp>
2332 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2333 (outs QPR:$Vd), (ins QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2334 NVMulSLFrm, IIC_VMULi16Q, OpcodeStr, Dt,"$Vd, $Vn, $Vm$lane", "",
2335 [(set (ResTy QPR:$Vd),
2336 (ResTy (ShOp (ResTy QPR:$Vn),
2337 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2339 let isCommutable = 0;
2342 // Basic 3-register intrinsics, both double- and quad-register.
2343 class N3VDInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2344 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2345 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
2346 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2347 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), f, itin,
2348 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2349 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
2350 let isCommutable = Commutable;
2352 class N3VDIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2353 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
2354 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2355 (outs DPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2356 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2358 (Ty (IntOp (Ty DPR:$Vn),
2359 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),
2361 let isCommutable = 0;
2363 class N3VDIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2364 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
2365 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2366 (outs DPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2367 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2369 (Ty (IntOp (Ty DPR:$Vn),
2370 (Ty (NEONvduplane (Ty DPR_8:$Vm), imm:$lane)))))]> {
2371 let isCommutable = 0;
2373 class N3VDIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2374 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2375 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2376 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2377 (outs DPR:$Vd), (ins DPR:$Vm, DPR:$Vn), f, itin,
2378 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
2379 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (OpTy DPR:$Vn))))]> {
2380 let isCommutable = 0;
2383 class N3VQInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2384 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2385 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
2386 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2387 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), f, itin,
2388 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2389 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
2390 let isCommutable = Commutable;
2392 class N3VQIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2393 string OpcodeStr, string Dt,
2394 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2395 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2396 (outs QPR:$Vd), (ins QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2397 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2398 [(set (ResTy QPR:$Vd),
2399 (ResTy (IntOp (ResTy QPR:$Vn),
2400 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2402 let isCommutable = 0;
2404 class N3VQIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2405 string OpcodeStr, string Dt,
2406 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2407 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2408 (outs QPR:$Vd), (ins QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2409 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2410 [(set (ResTy QPR:$Vd),
2411 (ResTy (IntOp (ResTy QPR:$Vn),
2412 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2414 let isCommutable = 0;
2416 class N3VQIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2417 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2418 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2419 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2420 (outs QPR:$Vd), (ins QPR:$Vm, QPR:$Vn), f, itin,
2421 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
2422 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (OpTy QPR:$Vn))))]> {
2423 let isCommutable = 0;
2426 // Multiply-Add/Sub operations: double- and quad-register.
2427 class N3VDMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2428 InstrItinClass itin, string OpcodeStr, string Dt,
2429 ValueType Ty, SDPatternOperator MulOp, SDPatternOperator OpNode>
2430 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2431 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2432 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2433 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
2434 (Ty (MulOp DPR:$Vn, DPR:$Vm)))))]>;
2436 class N3VDMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2437 string OpcodeStr, string Dt,
2438 ValueType Ty, SDPatternOperator MulOp, SDPatternOperator ShOp>
2439 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2441 (ins DPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2443 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2445 (Ty (ShOp (Ty DPR:$src1),
2447 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),
2449 class N3VDMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2450 string OpcodeStr, string Dt,
2451 ValueType Ty, SDNode MulOp, SDNode ShOp>
2452 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2454 (ins DPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2456 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2458 (Ty (ShOp (Ty DPR:$src1),
2460 (Ty (NEONvduplane (Ty DPR_8:$Vm),
2463 class N3VQMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2464 InstrItinClass itin, string OpcodeStr, string Dt, ValueType Ty,
2465 SDPatternOperator MulOp, SDPatternOperator OpNode>
2466 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2467 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2468 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2469 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
2470 (Ty (MulOp QPR:$Vn, QPR:$Vm)))))]>;
2471 class N3VQMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2472 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2473 SDPatternOperator MulOp, SDPatternOperator ShOp>
2474 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2476 (ins QPR:$src1, QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2478 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2479 [(set (ResTy QPR:$Vd),
2480 (ResTy (ShOp (ResTy QPR:$src1),
2481 (ResTy (MulOp QPR:$Vn,
2482 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2484 class N3VQMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2485 string OpcodeStr, string Dt,
2486 ValueType ResTy, ValueType OpTy,
2487 SDNode MulOp, SDNode ShOp>
2488 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2490 (ins QPR:$src1, QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2492 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2493 [(set (ResTy QPR:$Vd),
2494 (ResTy (ShOp (ResTy QPR:$src1),
2495 (ResTy (MulOp QPR:$Vn,
2496 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2499 // Neon Intrinsic-Op instructions (VABA): double- and quad-register.
2500 class N3VDIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2501 InstrItinClass itin, string OpcodeStr, string Dt,
2502 ValueType Ty, Intrinsic IntOp, SDNode OpNode>
2503 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2504 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2505 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2506 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
2507 (Ty (IntOp (Ty DPR:$Vn), (Ty DPR:$Vm))))))]>;
2508 class N3VQIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2509 InstrItinClass itin, string OpcodeStr, string Dt,
2510 ValueType Ty, Intrinsic IntOp, SDNode OpNode>
2511 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2512 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2513 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2514 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
2515 (Ty (IntOp (Ty QPR:$Vn), (Ty QPR:$Vm))))))]>;
2517 // Neon 3-argument intrinsics, both double- and quad-register.
2518 // The destination register is also used as the first source operand register.
2519 class N3VDInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2520 InstrItinClass itin, string OpcodeStr, string Dt,
2521 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2522 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2523 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2524 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2525 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$src1),
2526 (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>;
2527 class N3VQInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2528 InstrItinClass itin, string OpcodeStr, string Dt,
2529 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2530 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2531 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2532 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2533 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$src1),
2534 (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]>;
2536 // Long Multiply-Add/Sub operations.
2537 class N3VLMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2538 InstrItinClass itin, string OpcodeStr, string Dt,
2539 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2540 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2541 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2542 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2543 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
2544 (TyQ (MulOp (TyD DPR:$Vn),
2545 (TyD DPR:$Vm)))))]>;
2546 class N3VLMulOpSL<bit op24, bits<2> op21_20, bits<4> op11_8,
2547 InstrItinClass itin, string OpcodeStr, string Dt,
2548 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2549 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$Vd),
2550 (ins QPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2552 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2554 (OpNode (TyQ QPR:$src1),
2555 (TyQ (MulOp (TyD DPR:$Vn),
2556 (TyD (NEONvduplane (TyD DPR_VFP2:$Vm),
2558 class N3VLMulOpSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2559 InstrItinClass itin, string OpcodeStr, string Dt,
2560 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2561 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$Vd),
2562 (ins QPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2564 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2566 (OpNode (TyQ QPR:$src1),
2567 (TyQ (MulOp (TyD DPR:$Vn),
2568 (TyD (NEONvduplane (TyD DPR_8:$Vm),
2571 // Long Intrinsic-Op vector operations with explicit extend (VABAL).
2572 class N3VLIntExtOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2573 InstrItinClass itin, string OpcodeStr, string Dt,
2574 ValueType TyQ, ValueType TyD, Intrinsic IntOp, SDNode ExtOp,
2576 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2577 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2578 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2579 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
2580 (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
2581 (TyD DPR:$Vm)))))))]>;
2583 // Neon Long 3-argument intrinsic. The destination register is
2584 // a quad-register and is also used as the first source operand register.
2585 class N3VLInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2586 InstrItinClass itin, string OpcodeStr, string Dt,
2587 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
2588 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2589 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2590 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2592 (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$Vn), (TyD DPR:$Vm))))]>;
2593 class N3VLInt3SL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2594 string OpcodeStr, string Dt,
2595 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2596 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2598 (ins QPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2600 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2601 [(set (ResTy QPR:$Vd),
2602 (ResTy (IntOp (ResTy QPR:$src1),
2604 (OpTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2606 class N3VLInt3SL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2607 InstrItinClass itin, string OpcodeStr, string Dt,
2608 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2609 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2611 (ins QPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2613 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2614 [(set (ResTy QPR:$Vd),
2615 (ResTy (IntOp (ResTy QPR:$src1),
2617 (OpTy (NEONvduplane (OpTy DPR_8:$Vm),
2620 // Narrowing 3-register intrinsics.
2621 class N3VNInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2622 string OpcodeStr, string Dt, ValueType TyD, ValueType TyQ,
2623 Intrinsic IntOp, bit Commutable>
2624 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2625 (outs DPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINi4D,
2626 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2627 [(set DPR:$Vd, (TyD (IntOp (TyQ QPR:$Vn), (TyQ QPR:$Vm))))]> {
2628 let isCommutable = Commutable;
2631 // Long 3-register operations.
2632 class N3VL<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2633 InstrItinClass itin, string OpcodeStr, string Dt,
2634 ValueType TyQ, ValueType TyD, SDNode OpNode, bit Commutable>
2635 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2636 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2637 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2638 [(set QPR:$Vd, (TyQ (OpNode (TyD DPR:$Vn), (TyD DPR:$Vm))))]> {
2639 let isCommutable = Commutable;
2641 class N3VLSL<bit op24, bits<2> op21_20, bits<4> op11_8,
2642 InstrItinClass itin, string OpcodeStr, string Dt,
2643 ValueType TyQ, ValueType TyD, SDNode OpNode>
2644 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2645 (outs QPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2646 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2648 (TyQ (OpNode (TyD DPR:$Vn),
2649 (TyD (NEONvduplane (TyD DPR_VFP2:$Vm),imm:$lane)))))]>;
2650 class N3VLSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2651 InstrItinClass itin, string OpcodeStr, string Dt,
2652 ValueType TyQ, ValueType TyD, SDNode OpNode>
2653 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2654 (outs QPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2655 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2657 (TyQ (OpNode (TyD DPR:$Vn),
2658 (TyD (NEONvduplane (TyD DPR_8:$Vm), imm:$lane)))))]>;
2660 // Long 3-register operations with explicitly extended operands.
2661 class N3VLExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2662 InstrItinClass itin, string OpcodeStr, string Dt,
2663 ValueType TyQ, ValueType TyD, SDNode OpNode, SDNode ExtOp,
2665 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2666 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2667 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2668 [(set QPR:$Vd, (OpNode (TyQ (ExtOp (TyD DPR:$Vn))),
2669 (TyQ (ExtOp (TyD DPR:$Vm)))))]> {
2670 let isCommutable = Commutable;
2673 // Long 3-register intrinsics with explicit extend (VABDL).
2674 class N3VLIntExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2675 InstrItinClass itin, string OpcodeStr, string Dt,
2676 ValueType TyQ, ValueType TyD, Intrinsic IntOp, SDNode ExtOp,
2678 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2679 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2680 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2681 [(set QPR:$Vd, (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
2682 (TyD DPR:$Vm))))))]> {
2683 let isCommutable = Commutable;
2686 // Long 3-register intrinsics.
2687 class N3VLInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2688 InstrItinClass itin, string OpcodeStr, string Dt,
2689 ValueType TyQ, ValueType TyD, Intrinsic IntOp, bit Commutable>
2690 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2691 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2692 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2693 [(set QPR:$Vd, (TyQ (IntOp (TyD DPR:$Vn), (TyD DPR:$Vm))))]> {
2694 let isCommutable = Commutable;
2696 class N3VLIntSL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2697 string OpcodeStr, string Dt,
2698 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2699 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2700 (outs QPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2701 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2702 [(set (ResTy QPR:$Vd),
2703 (ResTy (IntOp (OpTy DPR:$Vn),
2704 (OpTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2706 class N3VLIntSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2707 InstrItinClass itin, string OpcodeStr, string Dt,
2708 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2709 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2710 (outs QPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2711 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2712 [(set (ResTy QPR:$Vd),
2713 (ResTy (IntOp (OpTy DPR:$Vn),
2714 (OpTy (NEONvduplane (OpTy DPR_8:$Vm),
2717 // Wide 3-register operations.
2718 class N3VW<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2719 string OpcodeStr, string Dt, ValueType TyQ, ValueType TyD,
2720 SDNode OpNode, SDNode ExtOp, bit Commutable>
2721 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2722 (outs QPR:$Vd), (ins QPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VSUBiD,
2723 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2724 [(set QPR:$Vd, (OpNode (TyQ QPR:$Vn),
2725 (TyQ (ExtOp (TyD DPR:$Vm)))))]> {
2726 let isCommutable = Commutable;
2729 // Pairwise long 2-register intrinsics, both double- and quad-register.
2730 class N2VDPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2731 bits<2> op17_16, bits<5> op11_7, bit op4,
2732 string OpcodeStr, string Dt,
2733 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2734 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
2735 (ins DPR:$Vm), IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
2736 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm))))]>;
2737 class N2VQPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2738 bits<2> op17_16, bits<5> op11_7, bit op4,
2739 string OpcodeStr, string Dt,
2740 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2741 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
2742 (ins QPR:$Vm), IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
2743 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm))))]>;
2745 // Pairwise long 2-register accumulate intrinsics,
2746 // both double- and quad-register.
2747 // The destination register is also used as the first source operand register.
2748 class N2VDPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2749 bits<2> op17_16, bits<5> op11_7, bit op4,
2750 string OpcodeStr, string Dt,
2751 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2752 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
2753 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vm), IIC_VPALiD,
2754 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
2755 [(set DPR:$Vd, (ResTy (IntOp (ResTy DPR:$src1), (OpTy DPR:$Vm))))]>;
2756 class N2VQPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2757 bits<2> op17_16, bits<5> op11_7, bit op4,
2758 string OpcodeStr, string Dt,
2759 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2760 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4,
2761 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vm), IIC_VPALiQ,
2762 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
2763 [(set QPR:$Vd, (ResTy (IntOp (ResTy QPR:$src1), (OpTy QPR:$Vm))))]>;
2765 // Shift by immediate,
2766 // both double- and quad-register.
2767 class N2VDSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2768 Format f, InstrItinClass itin, Operand ImmTy,
2769 string OpcodeStr, string Dt, ValueType Ty, SDNode OpNode>
2770 : N2VImm<op24, op23, op11_8, op7, 0, op4,
2771 (outs DPR:$Vd), (ins DPR:$Vm, ImmTy:$SIMM), f, itin,
2772 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2773 [(set DPR:$Vd, (Ty (OpNode (Ty DPR:$Vm), (i32 imm:$SIMM))))]>;
2774 class N2VQSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2775 Format f, InstrItinClass itin, Operand ImmTy,
2776 string OpcodeStr, string Dt, ValueType Ty, SDNode OpNode>
2777 : N2VImm<op24, op23, op11_8, op7, 1, op4,
2778 (outs QPR:$Vd), (ins QPR:$Vm, ImmTy:$SIMM), f, itin,
2779 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2780 [(set QPR:$Vd, (Ty (OpNode (Ty QPR:$Vm), (i32 imm:$SIMM))))]>;
2782 // Long shift by immediate.
2783 class N2VLSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
2784 string OpcodeStr, string Dt,
2785 ValueType ResTy, ValueType OpTy, Operand ImmTy, SDNode OpNode>
2786 : N2VImm<op24, op23, op11_8, op7, op6, op4,
2787 (outs QPR:$Vd), (ins DPR:$Vm, ImmTy:$SIMM), N2RegVShLFrm,
2788 IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2789 [(set QPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vm),
2790 (i32 imm:$SIMM))))]>;
2792 // Narrow shift by immediate.
2793 class N2VNSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
2794 InstrItinClass itin, string OpcodeStr, string Dt,
2795 ValueType ResTy, ValueType OpTy, Operand ImmTy, SDNode OpNode>
2796 : N2VImm<op24, op23, op11_8, op7, op6, op4,
2797 (outs DPR:$Vd), (ins QPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, itin,
2798 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2799 [(set DPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vm),
2800 (i32 imm:$SIMM))))]>;
2802 // Shift right by immediate and accumulate,
2803 // both double- and quad-register.
2804 class N2VDShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2805 Operand ImmTy, string OpcodeStr, string Dt,
2806 ValueType Ty, SDNode ShOp>
2807 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
2808 (ins DPR:$src1, DPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, IIC_VPALiD,
2809 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2810 [(set DPR:$Vd, (Ty (add DPR:$src1,
2811 (Ty (ShOp DPR:$Vm, (i32 imm:$SIMM))))))]>;
2812 class N2VQShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2813 Operand ImmTy, string OpcodeStr, string Dt,
2814 ValueType Ty, SDNode ShOp>
2815 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
2816 (ins QPR:$src1, QPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, IIC_VPALiD,
2817 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2818 [(set QPR:$Vd, (Ty (add QPR:$src1,
2819 (Ty (ShOp QPR:$Vm, (i32 imm:$SIMM))))))]>;
2821 // Shift by immediate and insert,
2822 // both double- and quad-register.
2823 class N2VDShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2824 Operand ImmTy, Format f, string OpcodeStr, string Dt,
2825 ValueType Ty,SDNode ShOp>
2826 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
2827 (ins DPR:$src1, DPR:$Vm, ImmTy:$SIMM), f, IIC_VSHLiD,
2828 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2829 [(set DPR:$Vd, (Ty (ShOp DPR:$src1, DPR:$Vm, (i32 imm:$SIMM))))]>;
2830 class N2VQShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2831 Operand ImmTy, Format f, string OpcodeStr, string Dt,
2832 ValueType Ty,SDNode ShOp>
2833 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
2834 (ins QPR:$src1, QPR:$Vm, ImmTy:$SIMM), f, IIC_VSHLiQ,
2835 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2836 [(set QPR:$Vd, (Ty (ShOp QPR:$src1, QPR:$Vm, (i32 imm:$SIMM))))]>;
2838 // Convert, with fractional bits immediate,
2839 // both double- and quad-register.
2840 class N2VCvtD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2841 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2843 : N2VImm<op24, op23, op11_8, op7, 0, op4,
2844 (outs DPR:$Vd), (ins DPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
2845 IIC_VUNAD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2846 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (i32 imm:$SIMM))))]>;
2847 class N2VCvtQ<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2848 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2850 : N2VImm<op24, op23, op11_8, op7, 1, op4,
2851 (outs QPR:$Vd), (ins QPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
2852 IIC_VUNAQ, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2853 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (i32 imm:$SIMM))))]>;
2855 //===----------------------------------------------------------------------===//
2857 //===----------------------------------------------------------------------===//
2859 // Abbreviations used in multiclass suffixes:
2860 // Q = quarter int (8 bit) elements
2861 // H = half int (16 bit) elements
2862 // S = single int (32 bit) elements
2863 // D = double int (64 bit) elements
2865 // Neon 2-register vector operations and intrinsics.
2867 // Neon 2-register comparisons.
2868 // source operand element sizes of 8, 16 and 32 bits:
2869 multiclass N2V_QHS_cmp<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2870 bits<5> op11_7, bit op4, string opc, string Dt,
2871 string asm, SDNode OpNode> {
2872 // 64-bit vector types.
2873 def v8i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 0, op4,
2874 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2875 opc, !strconcat(Dt, "8"), asm, "",
2876 [(set DPR:$Vd, (v8i8 (OpNode (v8i8 DPR:$Vm))))]>;
2877 def v4i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 0, op4,
2878 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2879 opc, !strconcat(Dt, "16"), asm, "",
2880 [(set DPR:$Vd, (v4i16 (OpNode (v4i16 DPR:$Vm))))]>;
2881 def v2i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
2882 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2883 opc, !strconcat(Dt, "32"), asm, "",
2884 [(set DPR:$Vd, (v2i32 (OpNode (v2i32 DPR:$Vm))))]>;
2885 def v2f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
2886 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2887 opc, "f32", asm, "",
2888 [(set DPR:$Vd, (v2i32 (OpNode (v2f32 DPR:$Vm))))]> {
2889 let Inst{10} = 1; // overwrite F = 1
2892 // 128-bit vector types.
2893 def v16i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 1, op4,
2894 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2895 opc, !strconcat(Dt, "8"), asm, "",
2896 [(set QPR:$Vd, (v16i8 (OpNode (v16i8 QPR:$Vm))))]>;
2897 def v8i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 1, op4,
2898 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2899 opc, !strconcat(Dt, "16"), asm, "",
2900 [(set QPR:$Vd, (v8i16 (OpNode (v8i16 QPR:$Vm))))]>;
2901 def v4i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
2902 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2903 opc, !strconcat(Dt, "32"), asm, "",
2904 [(set QPR:$Vd, (v4i32 (OpNode (v4i32 QPR:$Vm))))]>;
2905 def v4f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
2906 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2907 opc, "f32", asm, "",
2908 [(set QPR:$Vd, (v4i32 (OpNode (v4f32 QPR:$Vm))))]> {
2909 let Inst{10} = 1; // overwrite F = 1
2914 // Neon 2-register vector intrinsics,
2915 // element sizes of 8, 16 and 32 bits:
2916 multiclass N2VInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2917 bits<5> op11_7, bit op4,
2918 InstrItinClass itinD, InstrItinClass itinQ,
2919 string OpcodeStr, string Dt, Intrinsic IntOp> {
2920 // 64-bit vector types.
2921 def v8i8 : N2VDInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2922 itinD, OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
2923 def v4i16 : N2VDInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2924 itinD, OpcodeStr, !strconcat(Dt, "16"),v4i16,v4i16,IntOp>;
2925 def v2i32 : N2VDInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2926 itinD, OpcodeStr, !strconcat(Dt, "32"),v2i32,v2i32,IntOp>;
2928 // 128-bit vector types.
2929 def v16i8 : N2VQInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2930 itinQ, OpcodeStr, !strconcat(Dt, "8"), v16i8,v16i8,IntOp>;
2931 def v8i16 : N2VQInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2932 itinQ, OpcodeStr, !strconcat(Dt, "16"),v8i16,v8i16,IntOp>;
2933 def v4i32 : N2VQInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2934 itinQ, OpcodeStr, !strconcat(Dt, "32"),v4i32,v4i32,IntOp>;
2938 // Neon Narrowing 2-register vector operations,
2939 // source operand element sizes of 16, 32 and 64 bits:
2940 multiclass N2VN_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2941 bits<5> op11_7, bit op6, bit op4,
2942 InstrItinClass itin, string OpcodeStr, string Dt,
2944 def v8i8 : N2VN<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
2945 itin, OpcodeStr, !strconcat(Dt, "16"),
2946 v8i8, v8i16, OpNode>;
2947 def v4i16 : N2VN<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
2948 itin, OpcodeStr, !strconcat(Dt, "32"),
2949 v4i16, v4i32, OpNode>;
2950 def v2i32 : N2VN<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
2951 itin, OpcodeStr, !strconcat(Dt, "64"),
2952 v2i32, v2i64, OpNode>;
2955 // Neon Narrowing 2-register vector intrinsics,
2956 // source operand element sizes of 16, 32 and 64 bits:
2957 multiclass N2VNInt_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2958 bits<5> op11_7, bit op6, bit op4,
2959 InstrItinClass itin, string OpcodeStr, string Dt,
2961 def v8i8 : N2VNInt<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
2962 itin, OpcodeStr, !strconcat(Dt, "16"),
2963 v8i8, v8i16, IntOp>;
2964 def v4i16 : N2VNInt<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
2965 itin, OpcodeStr, !strconcat(Dt, "32"),
2966 v4i16, v4i32, IntOp>;
2967 def v2i32 : N2VNInt<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
2968 itin, OpcodeStr, !strconcat(Dt, "64"),
2969 v2i32, v2i64, IntOp>;
2973 // Neon Lengthening 2-register vector intrinsic (currently specific to VMOVL).
2974 // source operand element sizes of 16, 32 and 64 bits:
2975 multiclass N2VL_QHS<bits<2> op24_23, bits<5> op11_7, bit op6, bit op4,
2976 string OpcodeStr, string Dt, SDNode OpNode> {
2977 def v8i16 : N2VL<op24_23, 0b00, 0b10, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2978 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode>;
2979 def v4i32 : N2VL<op24_23, 0b01, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2980 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
2981 def v2i64 : N2VL<op24_23, 0b10, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2982 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
2986 // Neon 3-register vector operations.
2988 // First with only element sizes of 8, 16 and 32 bits:
2989 multiclass N3V_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2990 InstrItinClass itinD16, InstrItinClass itinD32,
2991 InstrItinClass itinQ16, InstrItinClass itinQ32,
2992 string OpcodeStr, string Dt,
2993 SDNode OpNode, bit Commutable = 0> {
2994 // 64-bit vector types.
2995 def v8i8 : N3VD<op24, op23, 0b00, op11_8, op4, itinD16,
2996 OpcodeStr, !strconcat(Dt, "8"),
2997 v8i8, v8i8, OpNode, Commutable>;
2998 def v4i16 : N3VD<op24, op23, 0b01, op11_8, op4, itinD16,
2999 OpcodeStr, !strconcat(Dt, "16"),
3000 v4i16, v4i16, OpNode, Commutable>;
3001 def v2i32 : N3VD<op24, op23, 0b10, op11_8, op4, itinD32,
3002 OpcodeStr, !strconcat(Dt, "32"),
3003 v2i32, v2i32, OpNode, Commutable>;
3005 // 128-bit vector types.
3006 def v16i8 : N3VQ<op24, op23, 0b00, op11_8, op4, itinQ16,
3007 OpcodeStr, !strconcat(Dt, "8"),
3008 v16i8, v16i8, OpNode, Commutable>;
3009 def v8i16 : N3VQ<op24, op23, 0b01, op11_8, op4, itinQ16,
3010 OpcodeStr, !strconcat(Dt, "16"),
3011 v8i16, v8i16, OpNode, Commutable>;
3012 def v4i32 : N3VQ<op24, op23, 0b10, op11_8, op4, itinQ32,
3013 OpcodeStr, !strconcat(Dt, "32"),
3014 v4i32, v4i32, OpNode, Commutable>;
3017 multiclass N3VSL_HS<bits<4> op11_8, string OpcodeStr, SDNode ShOp> {
3018 def v4i16 : N3VDSL16<0b01, op11_8, OpcodeStr, "i16", v4i16, ShOp>;
3019 def v2i32 : N3VDSL<0b10, op11_8, IIC_VMULi32D, OpcodeStr, "i32", v2i32, ShOp>;
3020 def v8i16 : N3VQSL16<0b01, op11_8, OpcodeStr, "i16", v8i16, v4i16, ShOp>;
3021 def v4i32 : N3VQSL<0b10, op11_8, IIC_VMULi32Q, OpcodeStr, "i32",
3022 v4i32, v2i32, ShOp>;
3025 // ....then also with element size 64 bits:
3026 multiclass N3V_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3027 InstrItinClass itinD, InstrItinClass itinQ,
3028 string OpcodeStr, string Dt,
3029 SDNode OpNode, bit Commutable = 0>
3030 : N3V_QHS<op24, op23, op11_8, op4, itinD, itinD, itinQ, itinQ,
3031 OpcodeStr, Dt, OpNode, Commutable> {
3032 def v1i64 : N3VD<op24, op23, 0b11, op11_8, op4, itinD,
3033 OpcodeStr, !strconcat(Dt, "64"),
3034 v1i64, v1i64, OpNode, Commutable>;
3035 def v2i64 : N3VQ<op24, op23, 0b11, op11_8, op4, itinQ,
3036 OpcodeStr, !strconcat(Dt, "64"),
3037 v2i64, v2i64, OpNode, Commutable>;
3041 // Neon 3-register vector intrinsics.
3043 // First with only element sizes of 16 and 32 bits:
3044 multiclass N3VInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3045 InstrItinClass itinD16, InstrItinClass itinD32,
3046 InstrItinClass itinQ16, InstrItinClass itinQ32,
3047 string OpcodeStr, string Dt,
3048 Intrinsic IntOp, bit Commutable = 0> {
3049 // 64-bit vector types.
3050 def v4i16 : N3VDInt<op24, op23, 0b01, op11_8, op4, f, itinD16,
3051 OpcodeStr, !strconcat(Dt, "16"),
3052 v4i16, v4i16, IntOp, Commutable>;
3053 def v2i32 : N3VDInt<op24, op23, 0b10, op11_8, op4, f, itinD32,
3054 OpcodeStr, !strconcat(Dt, "32"),
3055 v2i32, v2i32, IntOp, Commutable>;
3057 // 128-bit vector types.
3058 def v8i16 : N3VQInt<op24, op23, 0b01, op11_8, op4, f, itinQ16,
3059 OpcodeStr, !strconcat(Dt, "16"),
3060 v8i16, v8i16, IntOp, Commutable>;
3061 def v4i32 : N3VQInt<op24, op23, 0b10, op11_8, op4, f, itinQ32,
3062 OpcodeStr, !strconcat(Dt, "32"),
3063 v4i32, v4i32, IntOp, Commutable>;
3065 multiclass N3VInt_HSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3066 InstrItinClass itinD16, InstrItinClass itinD32,
3067 InstrItinClass itinQ16, InstrItinClass itinQ32,
3068 string OpcodeStr, string Dt,
3070 // 64-bit vector types.
3071 def v4i16 : N3VDIntSh<op24, op23, 0b01, op11_8, op4, f, itinD16,
3072 OpcodeStr, !strconcat(Dt, "16"),
3073 v4i16, v4i16, IntOp>;
3074 def v2i32 : N3VDIntSh<op24, op23, 0b10, op11_8, op4, f, itinD32,
3075 OpcodeStr, !strconcat(Dt, "32"),
3076 v2i32, v2i32, IntOp>;
3078 // 128-bit vector types.
3079 def v8i16 : N3VQIntSh<op24, op23, 0b01, op11_8, op4, f, itinQ16,
3080 OpcodeStr, !strconcat(Dt, "16"),
3081 v8i16, v8i16, IntOp>;
3082 def v4i32 : N3VQIntSh<op24, op23, 0b10, op11_8, op4, f, itinQ32,
3083 OpcodeStr, !strconcat(Dt, "32"),
3084 v4i32, v4i32, IntOp>;
3087 multiclass N3VIntSL_HS<bits<4> op11_8,
3088 InstrItinClass itinD16, InstrItinClass itinD32,
3089 InstrItinClass itinQ16, InstrItinClass itinQ32,
3090 string OpcodeStr, string Dt, Intrinsic IntOp> {
3091 def v4i16 : N3VDIntSL16<0b01, op11_8, itinD16,
3092 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp>;
3093 def v2i32 : N3VDIntSL<0b10, op11_8, itinD32,
3094 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp>;
3095 def v8i16 : N3VQIntSL16<0b01, op11_8, itinQ16,
3096 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16, IntOp>;
3097 def v4i32 : N3VQIntSL<0b10, op11_8, itinQ32,
3098 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32, IntOp>;
3101 // ....then also with element size of 8 bits:
3102 multiclass N3VInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3103 InstrItinClass itinD16, InstrItinClass itinD32,
3104 InstrItinClass itinQ16, InstrItinClass itinQ32,
3105 string OpcodeStr, string Dt,
3106 Intrinsic IntOp, bit Commutable = 0>
3107 : N3VInt_HS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
3108 OpcodeStr, Dt, IntOp, Commutable> {
3109 def v8i8 : N3VDInt<op24, op23, 0b00, op11_8, op4, f, itinD16,
3110 OpcodeStr, !strconcat(Dt, "8"),
3111 v8i8, v8i8, IntOp, Commutable>;
3112 def v16i8 : N3VQInt<op24, op23, 0b00, op11_8, op4, f, itinQ16,
3113 OpcodeStr, !strconcat(Dt, "8"),
3114 v16i8, v16i8, IntOp, Commutable>;
3116 multiclass N3VInt_QHSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3117 InstrItinClass itinD16, InstrItinClass itinD32,
3118 InstrItinClass itinQ16, InstrItinClass itinQ32,
3119 string OpcodeStr, string Dt,
3121 : N3VInt_HSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
3122 OpcodeStr, Dt, IntOp> {
3123 def v8i8 : N3VDIntSh<op24, op23, 0b00, op11_8, op4, f, itinD16,
3124 OpcodeStr, !strconcat(Dt, "8"),
3126 def v16i8 : N3VQIntSh<op24, op23, 0b00, op11_8, op4, f, itinQ16,
3127 OpcodeStr, !strconcat(Dt, "8"),
3128 v16i8, v16i8, IntOp>;
3132 // ....then also with element size of 64 bits:
3133 multiclass N3VInt_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3134 InstrItinClass itinD16, InstrItinClass itinD32,
3135 InstrItinClass itinQ16, InstrItinClass itinQ32,
3136 string OpcodeStr, string Dt,
3137 Intrinsic IntOp, bit Commutable = 0>
3138 : N3VInt_QHS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
3139 OpcodeStr, Dt, IntOp, Commutable> {
3140 def v1i64 : N3VDInt<op24, op23, 0b11, op11_8, op4, f, itinD32,
3141 OpcodeStr, !strconcat(Dt, "64"),
3142 v1i64, v1i64, IntOp, Commutable>;
3143 def v2i64 : N3VQInt<op24, op23, 0b11, op11_8, op4, f, itinQ32,
3144 OpcodeStr, !strconcat(Dt, "64"),
3145 v2i64, v2i64, IntOp, Commutable>;
3147 multiclass N3VInt_QHSDSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3148 InstrItinClass itinD16, InstrItinClass itinD32,
3149 InstrItinClass itinQ16, InstrItinClass itinQ32,
3150 string OpcodeStr, string Dt,
3152 : N3VInt_QHSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
3153 OpcodeStr, Dt, IntOp> {
3154 def v1i64 : N3VDIntSh<op24, op23, 0b11, op11_8, op4, f, itinD32,
3155 OpcodeStr, !strconcat(Dt, "64"),
3156 v1i64, v1i64, IntOp>;
3157 def v2i64 : N3VQIntSh<op24, op23, 0b11, op11_8, op4, f, itinQ32,
3158 OpcodeStr, !strconcat(Dt, "64"),
3159 v2i64, v2i64, IntOp>;
3162 // Neon Narrowing 3-register vector intrinsics,
3163 // source operand element sizes of 16, 32 and 64 bits:
3164 multiclass N3VNInt_HSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3165 string OpcodeStr, string Dt,
3166 Intrinsic IntOp, bit Commutable = 0> {
3167 def v8i8 : N3VNInt<op24, op23, 0b00, op11_8, op4,
3168 OpcodeStr, !strconcat(Dt, "16"),
3169 v8i8, v8i16, IntOp, Commutable>;
3170 def v4i16 : N3VNInt<op24, op23, 0b01, op11_8, op4,
3171 OpcodeStr, !strconcat(Dt, "32"),
3172 v4i16, v4i32, IntOp, Commutable>;
3173 def v2i32 : N3VNInt<op24, op23, 0b10, op11_8, op4,
3174 OpcodeStr, !strconcat(Dt, "64"),
3175 v2i32, v2i64, IntOp, Commutable>;
3179 // Neon Long 3-register vector operations.
3181 multiclass N3VL_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3182 InstrItinClass itin16, InstrItinClass itin32,
3183 string OpcodeStr, string Dt,
3184 SDNode OpNode, bit Commutable = 0> {
3185 def v8i16 : N3VL<op24, op23, 0b00, op11_8, op4, itin16,
3186 OpcodeStr, !strconcat(Dt, "8"),
3187 v8i16, v8i8, OpNode, Commutable>;
3188 def v4i32 : N3VL<op24, op23, 0b01, op11_8, op4, itin16,
3189 OpcodeStr, !strconcat(Dt, "16"),
3190 v4i32, v4i16, OpNode, Commutable>;
3191 def v2i64 : N3VL<op24, op23, 0b10, op11_8, op4, itin32,
3192 OpcodeStr, !strconcat(Dt, "32"),
3193 v2i64, v2i32, OpNode, Commutable>;
3196 multiclass N3VLSL_HS<bit op24, bits<4> op11_8,
3197 InstrItinClass itin, string OpcodeStr, string Dt,
3199 def v4i16 : N3VLSL16<op24, 0b01, op11_8, itin, OpcodeStr,
3200 !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
3201 def v2i32 : N3VLSL<op24, 0b10, op11_8, itin, OpcodeStr,
3202 !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
3205 multiclass N3VLExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3206 InstrItinClass itin16, InstrItinClass itin32,
3207 string OpcodeStr, string Dt,
3208 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
3209 def v8i16 : N3VLExt<op24, op23, 0b00, op11_8, op4, itin16,
3210 OpcodeStr, !strconcat(Dt, "8"),
3211 v8i16, v8i8, OpNode, ExtOp, Commutable>;
3212 def v4i32 : N3VLExt<op24, op23, 0b01, op11_8, op4, itin16,
3213 OpcodeStr, !strconcat(Dt, "16"),
3214 v4i32, v4i16, OpNode, ExtOp, Commutable>;
3215 def v2i64 : N3VLExt<op24, op23, 0b10, op11_8, op4, itin32,
3216 OpcodeStr, !strconcat(Dt, "32"),
3217 v2i64, v2i32, OpNode, ExtOp, Commutable>;
3220 // Neon Long 3-register vector intrinsics.
3222 // First with only element sizes of 16 and 32 bits:
3223 multiclass N3VLInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
3224 InstrItinClass itin16, InstrItinClass itin32,
3225 string OpcodeStr, string Dt,
3226 Intrinsic IntOp, bit Commutable = 0> {
3227 def v4i32 : N3VLInt<op24, op23, 0b01, op11_8, op4, itin16,
3228 OpcodeStr, !strconcat(Dt, "16"),
3229 v4i32, v4i16, IntOp, Commutable>;
3230 def v2i64 : N3VLInt<op24, op23, 0b10, op11_8, op4, itin32,
3231 OpcodeStr, !strconcat(Dt, "32"),
3232 v2i64, v2i32, IntOp, Commutable>;
3235 multiclass N3VLIntSL_HS<bit op24, bits<4> op11_8,
3236 InstrItinClass itin, string OpcodeStr, string Dt,
3238 def v4i16 : N3VLIntSL16<op24, 0b01, op11_8, itin,
3239 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
3240 def v2i32 : N3VLIntSL<op24, 0b10, op11_8, itin,
3241 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3244 // ....then also with element size of 8 bits:
3245 multiclass N3VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3246 InstrItinClass itin16, InstrItinClass itin32,
3247 string OpcodeStr, string Dt,
3248 Intrinsic IntOp, bit Commutable = 0>
3249 : N3VLInt_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt,
3250 IntOp, Commutable> {
3251 def v8i16 : N3VLInt<op24, op23, 0b00, op11_8, op4, itin16,
3252 OpcodeStr, !strconcat(Dt, "8"),
3253 v8i16, v8i8, IntOp, Commutable>;
3256 // ....with explicit extend (VABDL).
3257 multiclass N3VLIntExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3258 InstrItinClass itin, string OpcodeStr, string Dt,
3259 Intrinsic IntOp, SDNode ExtOp, bit Commutable = 0> {
3260 def v8i16 : N3VLIntExt<op24, op23, 0b00, op11_8, op4, itin,
3261 OpcodeStr, !strconcat(Dt, "8"),
3262 v8i16, v8i8, IntOp, ExtOp, Commutable>;
3263 def v4i32 : N3VLIntExt<op24, op23, 0b01, op11_8, op4, itin,
3264 OpcodeStr, !strconcat(Dt, "16"),
3265 v4i32, v4i16, IntOp, ExtOp, Commutable>;
3266 def v2i64 : N3VLIntExt<op24, op23, 0b10, op11_8, op4, itin,
3267 OpcodeStr, !strconcat(Dt, "32"),
3268 v2i64, v2i32, IntOp, ExtOp, Commutable>;
3272 // Neon Wide 3-register vector intrinsics,
3273 // source operand element sizes of 8, 16 and 32 bits:
3274 multiclass N3VW_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3275 string OpcodeStr, string Dt,
3276 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
3277 def v8i16 : N3VW<op24, op23, 0b00, op11_8, op4,
3278 OpcodeStr, !strconcat(Dt, "8"),
3279 v8i16, v8i8, OpNode, ExtOp, Commutable>;
3280 def v4i32 : N3VW<op24, op23, 0b01, op11_8, op4,
3281 OpcodeStr, !strconcat(Dt, "16"),
3282 v4i32, v4i16, OpNode, ExtOp, Commutable>;
3283 def v2i64 : N3VW<op24, op23, 0b10, op11_8, op4,
3284 OpcodeStr, !strconcat(Dt, "32"),
3285 v2i64, v2i32, OpNode, ExtOp, Commutable>;
3289 // Neon Multiply-Op vector operations,
3290 // element sizes of 8, 16 and 32 bits:
3291 multiclass N3VMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3292 InstrItinClass itinD16, InstrItinClass itinD32,
3293 InstrItinClass itinQ16, InstrItinClass itinQ32,
3294 string OpcodeStr, string Dt, SDNode OpNode> {
3295 // 64-bit vector types.
3296 def v8i8 : N3VDMulOp<op24, op23, 0b00, op11_8, op4, itinD16,
3297 OpcodeStr, !strconcat(Dt, "8"), v8i8, mul, OpNode>;
3298 def v4i16 : N3VDMulOp<op24, op23, 0b01, op11_8, op4, itinD16,
3299 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, OpNode>;
3300 def v2i32 : N3VDMulOp<op24, op23, 0b10, op11_8, op4, itinD32,
3301 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, OpNode>;
3303 // 128-bit vector types.
3304 def v16i8 : N3VQMulOp<op24, op23, 0b00, op11_8, op4, itinQ16,
3305 OpcodeStr, !strconcat(Dt, "8"), v16i8, mul, OpNode>;
3306 def v8i16 : N3VQMulOp<op24, op23, 0b01, op11_8, op4, itinQ16,
3307 OpcodeStr, !strconcat(Dt, "16"), v8i16, mul, OpNode>;
3308 def v4i32 : N3VQMulOp<op24, op23, 0b10, op11_8, op4, itinQ32,
3309 OpcodeStr, !strconcat(Dt, "32"), v4i32, mul, OpNode>;
3312 multiclass N3VMulOpSL_HS<bits<4> op11_8,
3313 InstrItinClass itinD16, InstrItinClass itinD32,
3314 InstrItinClass itinQ16, InstrItinClass itinQ32,
3315 string OpcodeStr, string Dt, SDNode ShOp> {
3316 def v4i16 : N3VDMulOpSL16<0b01, op11_8, itinD16,
3317 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, ShOp>;
3318 def v2i32 : N3VDMulOpSL<0b10, op11_8, itinD32,
3319 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, ShOp>;
3320 def v8i16 : N3VQMulOpSL16<0b01, op11_8, itinQ16,
3321 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16,
3323 def v4i32 : N3VQMulOpSL<0b10, op11_8, itinQ32,
3324 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32,
3328 // Neon Intrinsic-Op vector operations,
3329 // element sizes of 8, 16 and 32 bits:
3330 multiclass N3VIntOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3331 InstrItinClass itinD, InstrItinClass itinQ,
3332 string OpcodeStr, string Dt, Intrinsic IntOp,
3334 // 64-bit vector types.
3335 def v8i8 : N3VDIntOp<op24, op23, 0b00, op11_8, op4, itinD,
3336 OpcodeStr, !strconcat(Dt, "8"), v8i8, IntOp, OpNode>;
3337 def v4i16 : N3VDIntOp<op24, op23, 0b01, op11_8, op4, itinD,
3338 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp, OpNode>;
3339 def v2i32 : N3VDIntOp<op24, op23, 0b10, op11_8, op4, itinD,
3340 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp, OpNode>;
3342 // 128-bit vector types.
3343 def v16i8 : N3VQIntOp<op24, op23, 0b00, op11_8, op4, itinQ,
3344 OpcodeStr, !strconcat(Dt, "8"), v16i8, IntOp, OpNode>;
3345 def v8i16 : N3VQIntOp<op24, op23, 0b01, op11_8, op4, itinQ,
3346 OpcodeStr, !strconcat(Dt, "16"), v8i16, IntOp, OpNode>;
3347 def v4i32 : N3VQIntOp<op24, op23, 0b10, op11_8, op4, itinQ,
3348 OpcodeStr, !strconcat(Dt, "32"), v4i32, IntOp, OpNode>;
3351 // Neon 3-argument intrinsics,
3352 // element sizes of 8, 16 and 32 bits:
3353 multiclass N3VInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3354 InstrItinClass itinD, InstrItinClass itinQ,
3355 string OpcodeStr, string Dt, Intrinsic IntOp> {
3356 // 64-bit vector types.
3357 def v8i8 : N3VDInt3<op24, op23, 0b00, op11_8, op4, itinD,
3358 OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
3359 def v4i16 : N3VDInt3<op24, op23, 0b01, op11_8, op4, itinD,
3360 OpcodeStr, !strconcat(Dt, "16"), v4i16, v4i16, IntOp>;
3361 def v2i32 : N3VDInt3<op24, op23, 0b10, op11_8, op4, itinD,
3362 OpcodeStr, !strconcat(Dt, "32"), v2i32, v2i32, IntOp>;
3364 // 128-bit vector types.
3365 def v16i8 : N3VQInt3<op24, op23, 0b00, op11_8, op4, itinQ,
3366 OpcodeStr, !strconcat(Dt, "8"), v16i8, v16i8, IntOp>;
3367 def v8i16 : N3VQInt3<op24, op23, 0b01, op11_8, op4, itinQ,
3368 OpcodeStr, !strconcat(Dt, "16"), v8i16, v8i16, IntOp>;
3369 def v4i32 : N3VQInt3<op24, op23, 0b10, op11_8, op4, itinQ,
3370 OpcodeStr, !strconcat(Dt, "32"), v4i32, v4i32, IntOp>;
3374 // Neon Long Multiply-Op vector operations,
3375 // element sizes of 8, 16 and 32 bits:
3376 multiclass N3VLMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3377 InstrItinClass itin16, InstrItinClass itin32,
3378 string OpcodeStr, string Dt, SDNode MulOp,
3380 def v8i16 : N3VLMulOp<op24, op23, 0b00, op11_8, op4, itin16, OpcodeStr,
3381 !strconcat(Dt, "8"), v8i16, v8i8, MulOp, OpNode>;
3382 def v4i32 : N3VLMulOp<op24, op23, 0b01, op11_8, op4, itin16, OpcodeStr,
3383 !strconcat(Dt, "16"), v4i32, v4i16, MulOp, OpNode>;
3384 def v2i64 : N3VLMulOp<op24, op23, 0b10, op11_8, op4, itin32, OpcodeStr,
3385 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
3388 multiclass N3VLMulOpSL_HS<bit op24, bits<4> op11_8, string OpcodeStr,
3389 string Dt, SDNode MulOp, SDNode OpNode> {
3390 def v4i16 : N3VLMulOpSL16<op24, 0b01, op11_8, IIC_VMACi16D, OpcodeStr,
3391 !strconcat(Dt,"16"), v4i32, v4i16, MulOp, OpNode>;
3392 def v2i32 : N3VLMulOpSL<op24, 0b10, op11_8, IIC_VMACi32D, OpcodeStr,
3393 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
3397 // Neon Long 3-argument intrinsics.
3399 // First with only element sizes of 16 and 32 bits:
3400 multiclass N3VLInt3_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
3401 InstrItinClass itin16, InstrItinClass itin32,
3402 string OpcodeStr, string Dt, Intrinsic IntOp> {
3403 def v4i32 : N3VLInt3<op24, op23, 0b01, op11_8, op4, itin16,
3404 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
3405 def v2i64 : N3VLInt3<op24, op23, 0b10, op11_8, op4, itin32,
3406 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3409 multiclass N3VLInt3SL_HS<bit op24, bits<4> op11_8,
3410 string OpcodeStr, string Dt, Intrinsic IntOp> {
3411 def v4i16 : N3VLInt3SL16<op24, 0b01, op11_8, IIC_VMACi16D,
3412 OpcodeStr, !strconcat(Dt,"16"), v4i32, v4i16, IntOp>;
3413 def v2i32 : N3VLInt3SL<op24, 0b10, op11_8, IIC_VMACi32D,
3414 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3417 // ....then also with element size of 8 bits:
3418 multiclass N3VLInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3419 InstrItinClass itin16, InstrItinClass itin32,
3420 string OpcodeStr, string Dt, Intrinsic IntOp>
3421 : N3VLInt3_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt, IntOp> {
3422 def v8i16 : N3VLInt3<op24, op23, 0b00, op11_8, op4, itin16,
3423 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
3426 // ....with explicit extend (VABAL).
3427 multiclass N3VLIntExtOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3428 InstrItinClass itin, string OpcodeStr, string Dt,
3429 Intrinsic IntOp, SDNode ExtOp, SDNode OpNode> {
3430 def v8i16 : N3VLIntExtOp<op24, op23, 0b00, op11_8, op4, itin,
3431 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8,
3432 IntOp, ExtOp, OpNode>;
3433 def v4i32 : N3VLIntExtOp<op24, op23, 0b01, op11_8, op4, itin,
3434 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16,
3435 IntOp, ExtOp, OpNode>;
3436 def v2i64 : N3VLIntExtOp<op24, op23, 0b10, op11_8, op4, itin,
3437 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32,
3438 IntOp, ExtOp, OpNode>;
3442 // Neon Pairwise long 2-register intrinsics,
3443 // element sizes of 8, 16 and 32 bits:
3444 multiclass N2VPLInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3445 bits<5> op11_7, bit op4,
3446 string OpcodeStr, string Dt, Intrinsic IntOp> {
3447 // 64-bit vector types.
3448 def v8i8 : N2VDPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3449 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
3450 def v4i16 : N2VDPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3451 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
3452 def v2i32 : N2VDPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3453 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
3455 // 128-bit vector types.
3456 def v16i8 : N2VQPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3457 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
3458 def v8i16 : N2VQPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3459 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
3460 def v4i32 : N2VQPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3461 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
3465 // Neon Pairwise long 2-register accumulate intrinsics,
3466 // element sizes of 8, 16 and 32 bits:
3467 multiclass N2VPLInt2_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3468 bits<5> op11_7, bit op4,
3469 string OpcodeStr, string Dt, Intrinsic IntOp> {
3470 // 64-bit vector types.
3471 def v8i8 : N2VDPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3472 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
3473 def v4i16 : N2VDPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3474 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
3475 def v2i32 : N2VDPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3476 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
3478 // 128-bit vector types.
3479 def v16i8 : N2VQPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3480 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
3481 def v8i16 : N2VQPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3482 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
3483 def v4i32 : N2VQPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3484 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
3488 // Neon 2-register vector shift by immediate,
3489 // with f of either N2RegVShLFrm or N2RegVShRFrm
3490 // element sizes of 8, 16, 32 and 64 bits:
3491 multiclass N2VShL_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3492 InstrItinClass itin, string OpcodeStr, string Dt,
3494 // 64-bit vector types.
3495 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3496 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
3497 let Inst{21-19} = 0b001; // imm6 = 001xxx
3499 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3500 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
3501 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3503 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3504 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
3505 let Inst{21} = 0b1; // imm6 = 1xxxxx
3507 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, N2RegVShLFrm, itin, i32imm,
3508 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
3511 // 128-bit vector types.
3512 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3513 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
3514 let Inst{21-19} = 0b001; // imm6 = 001xxx
3516 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3517 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
3518 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3520 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3521 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
3522 let Inst{21} = 0b1; // imm6 = 1xxxxx
3524 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, N2RegVShLFrm, itin, i32imm,
3525 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
3528 multiclass N2VShR_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3529 InstrItinClass itin, string OpcodeStr, string Dt,
3531 // 64-bit vector types.
3532 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm8,
3533 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
3534 let Inst{21-19} = 0b001; // imm6 = 001xxx
3536 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm16,
3537 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
3538 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3540 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm32,
3541 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
3542 let Inst{21} = 0b1; // imm6 = 1xxxxx
3544 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, N2RegVShRFrm, itin, shr_imm64,
3545 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
3548 // 128-bit vector types.
3549 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm8,
3550 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
3551 let Inst{21-19} = 0b001; // imm6 = 001xxx
3553 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm16,
3554 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
3555 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3557 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm32,
3558 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
3559 let Inst{21} = 0b1; // imm6 = 1xxxxx
3561 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, N2RegVShRFrm, itin, shr_imm64,
3562 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
3566 // Neon Shift-Accumulate vector operations,
3567 // element sizes of 8, 16, 32 and 64 bits:
3568 multiclass N2VShAdd_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3569 string OpcodeStr, string Dt, SDNode ShOp> {
3570 // 64-bit vector types.
3571 def v8i8 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm8,
3572 OpcodeStr, !strconcat(Dt, "8"), v8i8, ShOp> {
3573 let Inst{21-19} = 0b001; // imm6 = 001xxx
3575 def v4i16 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm16,
3576 OpcodeStr, !strconcat(Dt, "16"), v4i16, ShOp> {
3577 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3579 def v2i32 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm32,
3580 OpcodeStr, !strconcat(Dt, "32"), v2i32, ShOp> {
3581 let Inst{21} = 0b1; // imm6 = 1xxxxx
3583 def v1i64 : N2VDShAdd<op24, op23, op11_8, 1, op4, shr_imm64,
3584 OpcodeStr, !strconcat(Dt, "64"), v1i64, ShOp>;
3587 // 128-bit vector types.
3588 def v16i8 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm8,
3589 OpcodeStr, !strconcat(Dt, "8"), v16i8, ShOp> {
3590 let Inst{21-19} = 0b001; // imm6 = 001xxx
3592 def v8i16 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm16,
3593 OpcodeStr, !strconcat(Dt, "16"), v8i16, ShOp> {
3594 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3596 def v4i32 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm32,
3597 OpcodeStr, !strconcat(Dt, "32"), v4i32, ShOp> {
3598 let Inst{21} = 0b1; // imm6 = 1xxxxx
3600 def v2i64 : N2VQShAdd<op24, op23, op11_8, 1, op4, shr_imm64,
3601 OpcodeStr, !strconcat(Dt, "64"), v2i64, ShOp>;
3605 // Neon Shift-Insert vector operations,
3606 // with f of either N2RegVShLFrm or N2RegVShRFrm
3607 // element sizes of 8, 16, 32 and 64 bits:
3608 multiclass N2VShInsL_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3610 // 64-bit vector types.
3611 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3612 N2RegVShLFrm, OpcodeStr, "8", v8i8, NEONvsli> {
3613 let Inst{21-19} = 0b001; // imm6 = 001xxx
3615 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3616 N2RegVShLFrm, OpcodeStr, "16", v4i16, NEONvsli> {
3617 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3619 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3620 N2RegVShLFrm, OpcodeStr, "32", v2i32, NEONvsli> {
3621 let Inst{21} = 0b1; // imm6 = 1xxxxx
3623 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4, i32imm,
3624 N2RegVShLFrm, OpcodeStr, "64", v1i64, NEONvsli>;
3627 // 128-bit vector types.
3628 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3629 N2RegVShLFrm, OpcodeStr, "8", v16i8, NEONvsli> {
3630 let Inst{21-19} = 0b001; // imm6 = 001xxx
3632 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3633 N2RegVShLFrm, OpcodeStr, "16", v8i16, NEONvsli> {
3634 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3636 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3637 N2RegVShLFrm, OpcodeStr, "32", v4i32, NEONvsli> {
3638 let Inst{21} = 0b1; // imm6 = 1xxxxx
3640 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4, i32imm,
3641 N2RegVShLFrm, OpcodeStr, "64", v2i64, NEONvsli>;
3644 multiclass N2VShInsR_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3646 // 64-bit vector types.
3647 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm8,
3648 N2RegVShRFrm, OpcodeStr, "8", v8i8, NEONvsri> {
3649 let Inst{21-19} = 0b001; // imm6 = 001xxx
3651 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm16,
3652 N2RegVShRFrm, OpcodeStr, "16", v4i16, NEONvsri> {
3653 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3655 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm32,
3656 N2RegVShRFrm, OpcodeStr, "32", v2i32, NEONvsri> {
3657 let Inst{21} = 0b1; // imm6 = 1xxxxx
3659 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4, shr_imm64,
3660 N2RegVShRFrm, OpcodeStr, "64", v1i64, NEONvsri>;
3663 // 128-bit vector types.
3664 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm8,
3665 N2RegVShRFrm, OpcodeStr, "8", v16i8, NEONvsri> {
3666 let Inst{21-19} = 0b001; // imm6 = 001xxx
3668 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm16,
3669 N2RegVShRFrm, OpcodeStr, "16", v8i16, NEONvsri> {
3670 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3672 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm32,
3673 N2RegVShRFrm, OpcodeStr, "32", v4i32, NEONvsri> {
3674 let Inst{21} = 0b1; // imm6 = 1xxxxx
3676 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4, shr_imm64,
3677 N2RegVShRFrm, OpcodeStr, "64", v2i64, NEONvsri>;
3681 // Neon Shift Long operations,
3682 // element sizes of 8, 16, 32 bits:
3683 multiclass N2VLSh_QHS<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
3684 bit op4, string OpcodeStr, string Dt, SDNode OpNode> {
3685 def v8i16 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3686 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, imm1_7, OpNode> {
3687 let Inst{21-19} = 0b001; // imm6 = 001xxx
3689 def v4i32 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3690 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, imm1_15, OpNode> {
3691 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3693 def v2i64 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3694 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, imm1_31, OpNode> {
3695 let Inst{21} = 0b1; // imm6 = 1xxxxx
3699 // Neon Shift Narrow operations,
3700 // element sizes of 16, 32, 64 bits:
3701 multiclass N2VNSh_HSD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
3702 bit op4, InstrItinClass itin, string OpcodeStr, string Dt,
3704 def v8i8 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3705 OpcodeStr, !strconcat(Dt, "16"),
3706 v8i8, v8i16, shr_imm8, OpNode> {
3707 let Inst{21-19} = 0b001; // imm6 = 001xxx
3709 def v4i16 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3710 OpcodeStr, !strconcat(Dt, "32"),
3711 v4i16, v4i32, shr_imm16, OpNode> {
3712 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3714 def v2i32 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3715 OpcodeStr, !strconcat(Dt, "64"),
3716 v2i32, v2i64, shr_imm32, OpNode> {
3717 let Inst{21} = 0b1; // imm6 = 1xxxxx
3721 //===----------------------------------------------------------------------===//
3722 // Instruction Definitions.
3723 //===----------------------------------------------------------------------===//
3725 // Vector Add Operations.
3727 // VADD : Vector Add (integer and floating-point)
3728 defm VADD : N3V_QHSD<0, 0, 0b1000, 0, IIC_VBINiD, IIC_VBINiQ, "vadd", "i",
3730 def VADDfd : N3VD<0, 0, 0b00, 0b1101, 0, IIC_VBIND, "vadd", "f32",
3731 v2f32, v2f32, fadd, 1>;
3732 def VADDfq : N3VQ<0, 0, 0b00, 0b1101, 0, IIC_VBINQ, "vadd", "f32",
3733 v4f32, v4f32, fadd, 1>;
3734 // VADDL : Vector Add Long (Q = D + D)
3735 defm VADDLs : N3VLExt_QHS<0,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
3736 "vaddl", "s", add, sext, 1>;
3737 defm VADDLu : N3VLExt_QHS<1,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
3738 "vaddl", "u", add, zext, 1>;
3739 // VADDW : Vector Add Wide (Q = Q + D)
3740 defm VADDWs : N3VW_QHS<0,1,0b0001,0, "vaddw", "s", add, sext, 0>;
3741 defm VADDWu : N3VW_QHS<1,1,0b0001,0, "vaddw", "u", add, zext, 0>;
3742 // VHADD : Vector Halving Add
3743 defm VHADDs : N3VInt_QHS<0, 0, 0b0000, 0, N3RegFrm,
3744 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3745 "vhadd", "s", int_arm_neon_vhadds, 1>;
3746 defm VHADDu : N3VInt_QHS<1, 0, 0b0000, 0, N3RegFrm,
3747 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3748 "vhadd", "u", int_arm_neon_vhaddu, 1>;
3749 // VRHADD : Vector Rounding Halving Add
3750 defm VRHADDs : N3VInt_QHS<0, 0, 0b0001, 0, N3RegFrm,
3751 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3752 "vrhadd", "s", int_arm_neon_vrhadds, 1>;
3753 defm VRHADDu : N3VInt_QHS<1, 0, 0b0001, 0, N3RegFrm,
3754 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3755 "vrhadd", "u", int_arm_neon_vrhaddu, 1>;
3756 // VQADD : Vector Saturating Add
3757 defm VQADDs : N3VInt_QHSD<0, 0, 0b0000, 1, N3RegFrm,
3758 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3759 "vqadd", "s", int_arm_neon_vqadds, 1>;
3760 defm VQADDu : N3VInt_QHSD<1, 0, 0b0000, 1, N3RegFrm,
3761 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3762 "vqadd", "u", int_arm_neon_vqaddu, 1>;
3763 // VADDHN : Vector Add and Narrow Returning High Half (D = Q + Q)
3764 defm VADDHN : N3VNInt_HSD<0,1,0b0100,0, "vaddhn", "i",
3765 int_arm_neon_vaddhn, 1>;
3766 // VRADDHN : Vector Rounding Add and Narrow Returning High Half (D = Q + Q)
3767 defm VRADDHN : N3VNInt_HSD<1,1,0b0100,0, "vraddhn", "i",
3768 int_arm_neon_vraddhn, 1>;
3770 // Vector Multiply Operations.
3772 // VMUL : Vector Multiply (integer, polynomial and floating-point)
3773 defm VMUL : N3V_QHS<0, 0, 0b1001, 1, IIC_VMULi16D, IIC_VMULi32D,
3774 IIC_VMULi16Q, IIC_VMULi32Q, "vmul", "i", mul, 1>;
3775 def VMULpd : N3VDInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16D, "vmul",
3776 "p8", v8i8, v8i8, int_arm_neon_vmulp, 1>;
3777 def VMULpq : N3VQInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16Q, "vmul",
3778 "p8", v16i8, v16i8, int_arm_neon_vmulp, 1>;
3779 def VMULfd : N3VD<1, 0, 0b00, 0b1101, 1, IIC_VFMULD, "vmul", "f32",
3780 v2f32, v2f32, fmul, 1>;
3781 def VMULfq : N3VQ<1, 0, 0b00, 0b1101, 1, IIC_VFMULQ, "vmul", "f32",
3782 v4f32, v4f32, fmul, 1>;
3783 defm VMULsl : N3VSL_HS<0b1000, "vmul", mul>;
3784 def VMULslfd : N3VDSL<0b10, 0b1001, IIC_VBIND, "vmul", "f32", v2f32, fmul>;
3785 def VMULslfq : N3VQSL<0b10, 0b1001, IIC_VBINQ, "vmul", "f32", v4f32,
3788 def : Pat<(v8i16 (mul (v8i16 QPR:$src1),
3789 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
3790 (v8i16 (VMULslv8i16 (v8i16 QPR:$src1),
3791 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3792 (DSubReg_i16_reg imm:$lane))),
3793 (SubReg_i16_lane imm:$lane)))>;
3794 def : Pat<(v4i32 (mul (v4i32 QPR:$src1),
3795 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
3796 (v4i32 (VMULslv4i32 (v4i32 QPR:$src1),
3797 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3798 (DSubReg_i32_reg imm:$lane))),
3799 (SubReg_i32_lane imm:$lane)))>;
3800 def : Pat<(v4f32 (fmul (v4f32 QPR:$src1),
3801 (v4f32 (NEONvduplane (v4f32 QPR:$src2), imm:$lane)))),
3802 (v4f32 (VMULslfq (v4f32 QPR:$src1),
3803 (v2f32 (EXTRACT_SUBREG QPR:$src2,
3804 (DSubReg_i32_reg imm:$lane))),
3805 (SubReg_i32_lane imm:$lane)))>;
3807 // VQDMULH : Vector Saturating Doubling Multiply Returning High Half
3808 defm VQDMULH : N3VInt_HS<0, 0, 0b1011, 0, N3RegFrm, IIC_VMULi16D, IIC_VMULi32D,
3809 IIC_VMULi16Q, IIC_VMULi32Q,
3810 "vqdmulh", "s", int_arm_neon_vqdmulh, 1>;
3811 defm VQDMULHsl: N3VIntSL_HS<0b1100, IIC_VMULi16D, IIC_VMULi32D,
3812 IIC_VMULi16Q, IIC_VMULi32Q,
3813 "vqdmulh", "s", int_arm_neon_vqdmulh>;
3814 def : Pat<(v8i16 (int_arm_neon_vqdmulh (v8i16 QPR:$src1),
3815 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
3817 (v8i16 (VQDMULHslv8i16 (v8i16 QPR:$src1),
3818 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3819 (DSubReg_i16_reg imm:$lane))),
3820 (SubReg_i16_lane imm:$lane)))>;
3821 def : Pat<(v4i32 (int_arm_neon_vqdmulh (v4i32 QPR:$src1),
3822 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
3824 (v4i32 (VQDMULHslv4i32 (v4i32 QPR:$src1),
3825 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3826 (DSubReg_i32_reg imm:$lane))),
3827 (SubReg_i32_lane imm:$lane)))>;
3829 // VQRDMULH : Vector Rounding Saturating Doubling Multiply Returning High Half
3830 defm VQRDMULH : N3VInt_HS<1, 0, 0b1011, 0, N3RegFrm,
3831 IIC_VMULi16D,IIC_VMULi32D,IIC_VMULi16Q,IIC_VMULi32Q,
3832 "vqrdmulh", "s", int_arm_neon_vqrdmulh, 1>;
3833 defm VQRDMULHsl : N3VIntSL_HS<0b1101, IIC_VMULi16D, IIC_VMULi32D,
3834 IIC_VMULi16Q, IIC_VMULi32Q,
3835 "vqrdmulh", "s", int_arm_neon_vqrdmulh>;
3836 def : Pat<(v8i16 (int_arm_neon_vqrdmulh (v8i16 QPR:$src1),
3837 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
3839 (v8i16 (VQRDMULHslv8i16 (v8i16 QPR:$src1),
3840 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3841 (DSubReg_i16_reg imm:$lane))),
3842 (SubReg_i16_lane imm:$lane)))>;
3843 def : Pat<(v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$src1),
3844 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
3846 (v4i32 (VQRDMULHslv4i32 (v4i32 QPR:$src1),
3847 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3848 (DSubReg_i32_reg imm:$lane))),
3849 (SubReg_i32_lane imm:$lane)))>;
3851 // VMULL : Vector Multiply Long (integer and polynomial) (Q = D * D)
3852 defm VMULLs : N3VL_QHS<0,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
3853 "vmull", "s", NEONvmulls, 1>;
3854 defm VMULLu : N3VL_QHS<1,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
3855 "vmull", "u", NEONvmullu, 1>;
3856 def VMULLp : N3VLInt<0, 1, 0b00, 0b1110, 0, IIC_VMULi16D, "vmull", "p8",
3857 v8i16, v8i8, int_arm_neon_vmullp, 1>;
3858 defm VMULLsls : N3VLSL_HS<0, 0b1010, IIC_VMULi16D, "vmull", "s", NEONvmulls>;
3859 defm VMULLslu : N3VLSL_HS<1, 0b1010, IIC_VMULi16D, "vmull", "u", NEONvmullu>;
3861 // VQDMULL : Vector Saturating Doubling Multiply Long (Q = D * D)
3862 defm VQDMULL : N3VLInt_HS<0,1,0b1101,0, IIC_VMULi16D, IIC_VMULi32D,
3863 "vqdmull", "s", int_arm_neon_vqdmull, 1>;
3864 defm VQDMULLsl: N3VLIntSL_HS<0, 0b1011, IIC_VMULi16D,
3865 "vqdmull", "s", int_arm_neon_vqdmull>;
3867 // Vector Multiply-Accumulate and Multiply-Subtract Operations.
3869 // VMLA : Vector Multiply Accumulate (integer and floating-point)
3870 defm VMLA : N3VMulOp_QHS<0, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3871 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
3872 def VMLAfd : N3VDMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACD, "vmla", "f32",
3873 v2f32, fmul_su, fadd_mlx>,
3874 Requires<[HasNEON, UseFPVMLx]>;
3875 def VMLAfq : N3VQMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACQ, "vmla", "f32",
3876 v4f32, fmul_su, fadd_mlx>,
3877 Requires<[HasNEON, UseFPVMLx]>;
3878 defm VMLAsl : N3VMulOpSL_HS<0b0000, IIC_VMACi16D, IIC_VMACi32D,
3879 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
3880 def VMLAslfd : N3VDMulOpSL<0b10, 0b0001, IIC_VMACD, "vmla", "f32",
3881 v2f32, fmul_su, fadd_mlx>,
3882 Requires<[HasNEON, UseFPVMLx]>;
3883 def VMLAslfq : N3VQMulOpSL<0b10, 0b0001, IIC_VMACQ, "vmla", "f32",
3884 v4f32, v2f32, fmul_su, fadd_mlx>,
3885 Requires<[HasNEON, UseFPVMLx]>;
3887 def : Pat<(v8i16 (add (v8i16 QPR:$src1),
3888 (mul (v8i16 QPR:$src2),
3889 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
3890 (v8i16 (VMLAslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
3891 (v4i16 (EXTRACT_SUBREG QPR:$src3,
3892 (DSubReg_i16_reg imm:$lane))),
3893 (SubReg_i16_lane imm:$lane)))>;
3895 def : Pat<(v4i32 (add (v4i32 QPR:$src1),
3896 (mul (v4i32 QPR:$src2),
3897 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
3898 (v4i32 (VMLAslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
3899 (v2i32 (EXTRACT_SUBREG QPR:$src3,
3900 (DSubReg_i32_reg imm:$lane))),
3901 (SubReg_i32_lane imm:$lane)))>;
3903 def : Pat<(v4f32 (fadd_mlx (v4f32 QPR:$src1),
3904 (fmul_su (v4f32 QPR:$src2),
3905 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
3906 (v4f32 (VMLAslfq (v4f32 QPR:$src1),
3908 (v2f32 (EXTRACT_SUBREG QPR:$src3,
3909 (DSubReg_i32_reg imm:$lane))),
3910 (SubReg_i32_lane imm:$lane)))>,
3911 Requires<[HasNEON, UseFPVMLx]>;
3913 // VMLAL : Vector Multiply Accumulate Long (Q += D * D)
3914 defm VMLALs : N3VLMulOp_QHS<0,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
3915 "vmlal", "s", NEONvmulls, add>;
3916 defm VMLALu : N3VLMulOp_QHS<1,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
3917 "vmlal", "u", NEONvmullu, add>;
3919 defm VMLALsls : N3VLMulOpSL_HS<0, 0b0010, "vmlal", "s", NEONvmulls, add>;
3920 defm VMLALslu : N3VLMulOpSL_HS<1, 0b0010, "vmlal", "u", NEONvmullu, add>;
3922 // VQDMLAL : Vector Saturating Doubling Multiply Accumulate Long (Q += D * D)
3923 defm VQDMLAL : N3VLInt3_HS<0, 1, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3924 "vqdmlal", "s", int_arm_neon_vqdmlal>;
3925 defm VQDMLALsl: N3VLInt3SL_HS<0, 0b0011, "vqdmlal", "s", int_arm_neon_vqdmlal>;
3927 // VMLS : Vector Multiply Subtract (integer and floating-point)
3928 defm VMLS : N3VMulOp_QHS<1, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3929 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
3930 def VMLSfd : N3VDMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACD, "vmls", "f32",
3931 v2f32, fmul_su, fsub_mlx>,
3932 Requires<[HasNEON, UseFPVMLx]>;
3933 def VMLSfq : N3VQMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACQ, "vmls", "f32",
3934 v4f32, fmul_su, fsub_mlx>,
3935 Requires<[HasNEON, UseFPVMLx]>;
3936 defm VMLSsl : N3VMulOpSL_HS<0b0100, IIC_VMACi16D, IIC_VMACi32D,
3937 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
3938 def VMLSslfd : N3VDMulOpSL<0b10, 0b0101, IIC_VMACD, "vmls", "f32",
3939 v2f32, fmul_su, fsub_mlx>,
3940 Requires<[HasNEON, UseFPVMLx]>;
3941 def VMLSslfq : N3VQMulOpSL<0b10, 0b0101, IIC_VMACQ, "vmls", "f32",
3942 v4f32, v2f32, fmul_su, fsub_mlx>,
3943 Requires<[HasNEON, UseFPVMLx]>;
3945 def : Pat<(v8i16 (sub (v8i16 QPR:$src1),
3946 (mul (v8i16 QPR:$src2),
3947 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
3948 (v8i16 (VMLSslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
3949 (v4i16 (EXTRACT_SUBREG QPR:$src3,
3950 (DSubReg_i16_reg imm:$lane))),
3951 (SubReg_i16_lane imm:$lane)))>;
3953 def : Pat<(v4i32 (sub (v4i32 QPR:$src1),
3954 (mul (v4i32 QPR:$src2),
3955 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
3956 (v4i32 (VMLSslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
3957 (v2i32 (EXTRACT_SUBREG QPR:$src3,
3958 (DSubReg_i32_reg imm:$lane))),
3959 (SubReg_i32_lane imm:$lane)))>;
3961 def : Pat<(v4f32 (fsub_mlx (v4f32 QPR:$src1),
3962 (fmul_su (v4f32 QPR:$src2),
3963 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
3964 (v4f32 (VMLSslfq (v4f32 QPR:$src1), (v4f32 QPR:$src2),
3965 (v2f32 (EXTRACT_SUBREG QPR:$src3,
3966 (DSubReg_i32_reg imm:$lane))),
3967 (SubReg_i32_lane imm:$lane)))>,
3968 Requires<[HasNEON, UseFPVMLx]>;
3970 // VMLSL : Vector Multiply Subtract Long (Q -= D * D)
3971 defm VMLSLs : N3VLMulOp_QHS<0,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
3972 "vmlsl", "s", NEONvmulls, sub>;
3973 defm VMLSLu : N3VLMulOp_QHS<1,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
3974 "vmlsl", "u", NEONvmullu, sub>;
3976 defm VMLSLsls : N3VLMulOpSL_HS<0, 0b0110, "vmlsl", "s", NEONvmulls, sub>;
3977 defm VMLSLslu : N3VLMulOpSL_HS<1, 0b0110, "vmlsl", "u", NEONvmullu, sub>;
3979 // VQDMLSL : Vector Saturating Doubling Multiply Subtract Long (Q -= D * D)
3980 defm VQDMLSL : N3VLInt3_HS<0, 1, 0b1011, 0, IIC_VMACi16D, IIC_VMACi32D,
3981 "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
3982 defm VQDMLSLsl: N3VLInt3SL_HS<0, 0b111, "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
3984 // Vector Subtract Operations.
3986 // VSUB : Vector Subtract (integer and floating-point)
3987 defm VSUB : N3V_QHSD<1, 0, 0b1000, 0, IIC_VSUBiD, IIC_VSUBiQ,
3988 "vsub", "i", sub, 0>;
3989 def VSUBfd : N3VD<0, 0, 0b10, 0b1101, 0, IIC_VBIND, "vsub", "f32",
3990 v2f32, v2f32, fsub, 0>;
3991 def VSUBfq : N3VQ<0, 0, 0b10, 0b1101, 0, IIC_VBINQ, "vsub", "f32",
3992 v4f32, v4f32, fsub, 0>;
3993 // VSUBL : Vector Subtract Long (Q = D - D)
3994 defm VSUBLs : N3VLExt_QHS<0,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
3995 "vsubl", "s", sub, sext, 0>;
3996 defm VSUBLu : N3VLExt_QHS<1,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
3997 "vsubl", "u", sub, zext, 0>;
3998 // VSUBW : Vector Subtract Wide (Q = Q - D)
3999 defm VSUBWs : N3VW_QHS<0,1,0b0011,0, "vsubw", "s", sub, sext, 0>;
4000 defm VSUBWu : N3VW_QHS<1,1,0b0011,0, "vsubw", "u", sub, zext, 0>;
4001 // VHSUB : Vector Halving Subtract
4002 defm VHSUBs : N3VInt_QHS<0, 0, 0b0010, 0, N3RegFrm,
4003 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4004 "vhsub", "s", int_arm_neon_vhsubs, 0>;
4005 defm VHSUBu : N3VInt_QHS<1, 0, 0b0010, 0, N3RegFrm,
4006 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4007 "vhsub", "u", int_arm_neon_vhsubu, 0>;
4008 // VQSUB : Vector Saturing Subtract
4009 defm VQSUBs : N3VInt_QHSD<0, 0, 0b0010, 1, N3RegFrm,
4010 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4011 "vqsub", "s", int_arm_neon_vqsubs, 0>;
4012 defm VQSUBu : N3VInt_QHSD<1, 0, 0b0010, 1, N3RegFrm,
4013 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4014 "vqsub", "u", int_arm_neon_vqsubu, 0>;
4015 // VSUBHN : Vector Subtract and Narrow Returning High Half (D = Q - Q)
4016 defm VSUBHN : N3VNInt_HSD<0,1,0b0110,0, "vsubhn", "i",
4017 int_arm_neon_vsubhn, 0>;
4018 // VRSUBHN : Vector Rounding Subtract and Narrow Returning High Half (D=Q-Q)
4019 defm VRSUBHN : N3VNInt_HSD<1,1,0b0110,0, "vrsubhn", "i",
4020 int_arm_neon_vrsubhn, 0>;
4022 // Vector Comparisons.
4024 // VCEQ : Vector Compare Equal
4025 defm VCEQ : N3V_QHS<1, 0, 0b1000, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
4026 IIC_VSUBi4Q, "vceq", "i", NEONvceq, 1>;
4027 def VCEQfd : N3VD<0,0,0b00,0b1110,0, IIC_VBIND, "vceq", "f32", v2i32, v2f32,
4029 def VCEQfq : N3VQ<0,0,0b00,0b1110,0, IIC_VBINQ, "vceq", "f32", v4i32, v4f32,
4032 defm VCEQz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00010, 0, "vceq", "i",
4033 "$Vd, $Vm, #0", NEONvceqz>;
4035 // VCGE : Vector Compare Greater Than or Equal
4036 defm VCGEs : N3V_QHS<0, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
4037 IIC_VSUBi4Q, "vcge", "s", NEONvcge, 0>;
4038 defm VCGEu : N3V_QHS<1, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
4039 IIC_VSUBi4Q, "vcge", "u", NEONvcgeu, 0>;
4040 def VCGEfd : N3VD<1,0,0b00,0b1110,0, IIC_VBIND, "vcge", "f32", v2i32, v2f32,
4042 def VCGEfq : N3VQ<1,0,0b00,0b1110,0, IIC_VBINQ, "vcge", "f32", v4i32, v4f32,
4045 defm VCGEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00001, 0, "vcge", "s",
4046 "$Vd, $Vm, #0", NEONvcgez>;
4047 defm VCLEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00011, 0, "vcle", "s",
4048 "$Vd, $Vm, #0", NEONvclez>;
4050 // VCGT : Vector Compare Greater Than
4051 defm VCGTs : N3V_QHS<0, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
4052 IIC_VSUBi4Q, "vcgt", "s", NEONvcgt, 0>;
4053 defm VCGTu : N3V_QHS<1, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
4054 IIC_VSUBi4Q, "vcgt", "u", NEONvcgtu, 0>;
4055 def VCGTfd : N3VD<1,0,0b10,0b1110,0, IIC_VBIND, "vcgt", "f32", v2i32, v2f32,
4057 def VCGTfq : N3VQ<1,0,0b10,0b1110,0, IIC_VBINQ, "vcgt", "f32", v4i32, v4f32,
4060 defm VCGTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00000, 0, "vcgt", "s",
4061 "$Vd, $Vm, #0", NEONvcgtz>;
4062 defm VCLTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00100, 0, "vclt", "s",
4063 "$Vd, $Vm, #0", NEONvcltz>;
4065 // VACGE : Vector Absolute Compare Greater Than or Equal (aka VCAGE)
4066 def VACGEd : N3VDInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacge",
4067 "f32", v2i32, v2f32, int_arm_neon_vacged, 0>;
4068 def VACGEq : N3VQInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacge",
4069 "f32", v4i32, v4f32, int_arm_neon_vacgeq, 0>;
4070 // VACGT : Vector Absolute Compare Greater Than (aka VCAGT)
4071 def VACGTd : N3VDInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacgt",
4072 "f32", v2i32, v2f32, int_arm_neon_vacgtd, 0>;
4073 def VACGTq : N3VQInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacgt",
4074 "f32", v4i32, v4f32, int_arm_neon_vacgtq, 0>;
4075 // VTST : Vector Test Bits
4076 defm VTST : N3V_QHS<0, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
4077 IIC_VBINi4Q, "vtst", "", NEONvtst, 1>;
4079 // Vector Bitwise Operations.
4081 def vnotd : PatFrag<(ops node:$in),
4082 (xor node:$in, (bitconvert (v8i8 NEONimmAllOnesV)))>;
4083 def vnotq : PatFrag<(ops node:$in),
4084 (xor node:$in, (bitconvert (v16i8 NEONimmAllOnesV)))>;
4087 // VAND : Vector Bitwise AND
4088 def VANDd : N3VDX<0, 0, 0b00, 0b0001, 1, IIC_VBINiD, "vand",
4089 v2i32, v2i32, and, 1>;
4090 def VANDq : N3VQX<0, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "vand",
4091 v4i32, v4i32, and, 1>;
4093 // VEOR : Vector Bitwise Exclusive OR
4094 def VEORd : N3VDX<1, 0, 0b00, 0b0001, 1, IIC_VBINiD, "veor",
4095 v2i32, v2i32, xor, 1>;
4096 def VEORq : N3VQX<1, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "veor",
4097 v4i32, v4i32, xor, 1>;
4099 // VORR : Vector Bitwise OR
4100 def VORRd : N3VDX<0, 0, 0b10, 0b0001, 1, IIC_VBINiD, "vorr",
4101 v2i32, v2i32, or, 1>;
4102 def VORRq : N3VQX<0, 0, 0b10, 0b0001, 1, IIC_VBINiQ, "vorr",
4103 v4i32, v4i32, or, 1>;
4105 def VORRiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 0, 1,
4106 (outs DPR:$Vd), (ins nImmSplatI16:$SIMM, DPR:$src),
4108 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
4110 (v4i16 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
4111 let Inst{9} = SIMM{9};
4114 def VORRiv2i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 0, 0, 1,
4115 (outs DPR:$Vd), (ins nImmSplatI32:$SIMM, DPR:$src),
4117 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
4119 (v2i32 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
4120 let Inst{10-9} = SIMM{10-9};
4123 def VORRiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 0, 1,
4124 (outs QPR:$Vd), (ins nImmSplatI16:$SIMM, QPR:$src),
4126 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
4128 (v8i16 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
4129 let Inst{9} = SIMM{9};
4132 def VORRiv4i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 1, 0, 1,
4133 (outs QPR:$Vd), (ins nImmSplatI32:$SIMM, QPR:$src),
4135 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
4137 (v4i32 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
4138 let Inst{10-9} = SIMM{10-9};
4142 // VBIC : Vector Bitwise Bit Clear (AND NOT)
4143 def VBICd : N3VX<0, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
4144 (ins DPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VBINiD,
4145 "vbic", "$Vd, $Vn, $Vm", "",
4146 [(set DPR:$Vd, (v2i32 (and DPR:$Vn,
4147 (vnotd DPR:$Vm))))]>;
4148 def VBICq : N3VX<0, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
4149 (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINiQ,
4150 "vbic", "$Vd, $Vn, $Vm", "",
4151 [(set QPR:$Vd, (v4i32 (and QPR:$Vn,
4152 (vnotq QPR:$Vm))))]>;
4154 def VBICiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 1, 1,
4155 (outs DPR:$Vd), (ins nImmSplatI16:$SIMM, DPR:$src),
4157 "vbic", "i16", "$Vd, $SIMM", "$src = $Vd",
4159 (v4i16 (NEONvbicImm DPR:$src, timm:$SIMM)))]> {
4160 let Inst{9} = SIMM{9};
4163 def VBICiv2i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 0, 1, 1,
4164 (outs DPR:$Vd), (ins nImmSplatI32:$SIMM, DPR:$src),
4166 "vbic", "i32", "$Vd, $SIMM", "$src = $Vd",
4168 (v2i32 (NEONvbicImm DPR:$src, timm:$SIMM)))]> {
4169 let Inst{10-9} = SIMM{10-9};
4172 def VBICiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 1, 1,
4173 (outs QPR:$Vd), (ins nImmSplatI16:$SIMM, QPR:$src),
4175 "vbic", "i16", "$Vd, $SIMM", "$src = $Vd",
4177 (v8i16 (NEONvbicImm QPR:$src, timm:$SIMM)))]> {
4178 let Inst{9} = SIMM{9};
4181 def VBICiv4i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 1, 1, 1,
4182 (outs QPR:$Vd), (ins nImmSplatI32:$SIMM, QPR:$src),
4184 "vbic", "i32", "$Vd, $SIMM", "$src = $Vd",
4186 (v4i32 (NEONvbicImm QPR:$src, timm:$SIMM)))]> {
4187 let Inst{10-9} = SIMM{10-9};
4190 // VORN : Vector Bitwise OR NOT
4191 def VORNd : N3VX<0, 0, 0b11, 0b0001, 0, 1, (outs DPR:$Vd),
4192 (ins DPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VBINiD,
4193 "vorn", "$Vd, $Vn, $Vm", "",
4194 [(set DPR:$Vd, (v2i32 (or DPR:$Vn,
4195 (vnotd DPR:$Vm))))]>;
4196 def VORNq : N3VX<0, 0, 0b11, 0b0001, 1, 1, (outs QPR:$Vd),
4197 (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINiQ,
4198 "vorn", "$Vd, $Vn, $Vm", "",
4199 [(set QPR:$Vd, (v4i32 (or QPR:$Vn,
4200 (vnotq QPR:$Vm))))]>;
4202 // VMVN : Vector Bitwise NOT (Immediate)
4204 let isReMaterializable = 1 in {
4206 def VMVNv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 1, 1, (outs DPR:$Vd),
4207 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4208 "vmvn", "i16", "$Vd, $SIMM", "",
4209 [(set DPR:$Vd, (v4i16 (NEONvmvnImm timm:$SIMM)))]> {
4210 let Inst{9} = SIMM{9};
4213 def VMVNv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 1, 1, (outs QPR:$Vd),
4214 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4215 "vmvn", "i16", "$Vd, $SIMM", "",
4216 [(set QPR:$Vd, (v8i16 (NEONvmvnImm timm:$SIMM)))]> {
4217 let Inst{9} = SIMM{9};
4220 def VMVNv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 1, 1, (outs DPR:$Vd),
4221 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4222 "vmvn", "i32", "$Vd, $SIMM", "",
4223 [(set DPR:$Vd, (v2i32 (NEONvmvnImm timm:$SIMM)))]> {
4224 let Inst{11-8} = SIMM{11-8};
4227 def VMVNv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 1, 1, (outs QPR:$Vd),
4228 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4229 "vmvn", "i32", "$Vd, $SIMM", "",
4230 [(set QPR:$Vd, (v4i32 (NEONvmvnImm timm:$SIMM)))]> {
4231 let Inst{11-8} = SIMM{11-8};
4235 // VMVN : Vector Bitwise NOT
4236 def VMVNd : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 0, 0,
4237 (outs DPR:$Vd), (ins DPR:$Vm), IIC_VSUBiD,
4238 "vmvn", "$Vd, $Vm", "",
4239 [(set DPR:$Vd, (v2i32 (vnotd DPR:$Vm)))]>;
4240 def VMVNq : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 1, 0,
4241 (outs QPR:$Vd), (ins QPR:$Vm), IIC_VSUBiD,
4242 "vmvn", "$Vd, $Vm", "",
4243 [(set QPR:$Vd, (v4i32 (vnotq QPR:$Vm)))]>;
4244 def : Pat<(v2i32 (vnotd DPR:$src)), (VMVNd DPR:$src)>;
4245 def : Pat<(v4i32 (vnotq QPR:$src)), (VMVNq QPR:$src)>;
4247 // VBSL : Vector Bitwise Select
4248 def VBSLd : N3VX<1, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
4249 (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
4250 N3RegFrm, IIC_VCNTiD,
4251 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4253 (v2i32 (NEONvbsl DPR:$src1, DPR:$Vn, DPR:$Vm)))]>;
4255 def : Pat<(v2i32 (or (and DPR:$Vn, DPR:$Vd),
4256 (and DPR:$Vm, (vnotd DPR:$Vd)))),
4257 (VBSLd DPR:$Vd, DPR:$Vn, DPR:$Vm)>;
4259 def VBSLq : N3VX<1, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
4260 (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
4261 N3RegFrm, IIC_VCNTiQ,
4262 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4264 (v4i32 (NEONvbsl QPR:$src1, QPR:$Vn, QPR:$Vm)))]>;
4266 def : Pat<(v4i32 (or (and QPR:$Vn, QPR:$Vd),
4267 (and QPR:$Vm, (vnotq QPR:$Vd)))),
4268 (VBSLq QPR:$Vd, QPR:$Vn, QPR:$Vm)>;
4270 // VBIF : Vector Bitwise Insert if False
4271 // like VBSL but with: "vbif $dst, $src3, $src1", "$src2 = $dst",
4272 // FIXME: This instruction's encoding MAY NOT BE correct.
4273 def VBIFd : N3VX<1, 0, 0b11, 0b0001, 0, 1,
4274 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
4275 N3RegFrm, IIC_VBINiD,
4276 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4278 def VBIFq : N3VX<1, 0, 0b11, 0b0001, 1, 1,
4279 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
4280 N3RegFrm, IIC_VBINiQ,
4281 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4284 // VBIT : Vector Bitwise Insert if True
4285 // like VBSL but with: "vbit $dst, $src2, $src1", "$src3 = $dst",
4286 // FIXME: This instruction's encoding MAY NOT BE correct.
4287 def VBITd : N3VX<1, 0, 0b10, 0b0001, 0, 1,
4288 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
4289 N3RegFrm, IIC_VBINiD,
4290 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4292 def VBITq : N3VX<1, 0, 0b10, 0b0001, 1, 1,
4293 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
4294 N3RegFrm, IIC_VBINiQ,
4295 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4298 // VBIT/VBIF are not yet implemented. The TwoAddress pass will not go looking
4299 // for equivalent operations with different register constraints; it just
4302 // Vector Absolute Differences.
4304 // VABD : Vector Absolute Difference
4305 defm VABDs : N3VInt_QHS<0, 0, 0b0111, 0, N3RegFrm,
4306 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4307 "vabd", "s", int_arm_neon_vabds, 1>;
4308 defm VABDu : N3VInt_QHS<1, 0, 0b0111, 0, N3RegFrm,
4309 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4310 "vabd", "u", int_arm_neon_vabdu, 1>;
4311 def VABDfd : N3VDInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBIND,
4312 "vabd", "f32", v2f32, v2f32, int_arm_neon_vabds, 1>;
4313 def VABDfq : N3VQInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBINQ,
4314 "vabd", "f32", v4f32, v4f32, int_arm_neon_vabds, 1>;
4316 // VABDL : Vector Absolute Difference Long (Q = | D - D |)
4317 defm VABDLs : N3VLIntExt_QHS<0,1,0b0111,0, IIC_VSUBi4Q,
4318 "vabdl", "s", int_arm_neon_vabds, zext, 1>;
4319 defm VABDLu : N3VLIntExt_QHS<1,1,0b0111,0, IIC_VSUBi4Q,
4320 "vabdl", "u", int_arm_neon_vabdu, zext, 1>;
4322 // VABA : Vector Absolute Difference and Accumulate
4323 defm VABAs : N3VIntOp_QHS<0,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
4324 "vaba", "s", int_arm_neon_vabds, add>;
4325 defm VABAu : N3VIntOp_QHS<1,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
4326 "vaba", "u", int_arm_neon_vabdu, add>;
4328 // VABAL : Vector Absolute Difference and Accumulate Long (Q += | D - D |)
4329 defm VABALs : N3VLIntExtOp_QHS<0,1,0b0101,0, IIC_VABAD,
4330 "vabal", "s", int_arm_neon_vabds, zext, add>;
4331 defm VABALu : N3VLIntExtOp_QHS<1,1,0b0101,0, IIC_VABAD,
4332 "vabal", "u", int_arm_neon_vabdu, zext, add>;
4334 // Vector Maximum and Minimum.
4336 // VMAX : Vector Maximum
4337 defm VMAXs : N3VInt_QHS<0, 0, 0b0110, 0, N3RegFrm,
4338 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4339 "vmax", "s", int_arm_neon_vmaxs, 1>;
4340 defm VMAXu : N3VInt_QHS<1, 0, 0b0110, 0, N3RegFrm,
4341 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4342 "vmax", "u", int_arm_neon_vmaxu, 1>;
4343 def VMAXfd : N3VDInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBIND,
4345 v2f32, v2f32, int_arm_neon_vmaxs, 1>;
4346 def VMAXfq : N3VQInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBINQ,
4348 v4f32, v4f32, int_arm_neon_vmaxs, 1>;
4350 // VMIN : Vector Minimum
4351 defm VMINs : N3VInt_QHS<0, 0, 0b0110, 1, N3RegFrm,
4352 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4353 "vmin", "s", int_arm_neon_vmins, 1>;
4354 defm VMINu : N3VInt_QHS<1, 0, 0b0110, 1, N3RegFrm,
4355 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4356 "vmin", "u", int_arm_neon_vminu, 1>;
4357 def VMINfd : N3VDInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBIND,
4359 v2f32, v2f32, int_arm_neon_vmins, 1>;
4360 def VMINfq : N3VQInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBINQ,
4362 v4f32, v4f32, int_arm_neon_vmins, 1>;
4364 // Vector Pairwise Operations.
4366 // VPADD : Vector Pairwise Add
4367 def VPADDi8 : N3VDInt<0, 0, 0b00, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4369 v8i8, v8i8, int_arm_neon_vpadd, 0>;
4370 def VPADDi16 : N3VDInt<0, 0, 0b01, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4372 v4i16, v4i16, int_arm_neon_vpadd, 0>;
4373 def VPADDi32 : N3VDInt<0, 0, 0b10, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4375 v2i32, v2i32, int_arm_neon_vpadd, 0>;
4376 def VPADDf : N3VDInt<1, 0, 0b00, 0b1101, 0, N3RegFrm,
4377 IIC_VPBIND, "vpadd", "f32",
4378 v2f32, v2f32, int_arm_neon_vpadd, 0>;
4380 // VPADDL : Vector Pairwise Add Long
4381 defm VPADDLs : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00100, 0, "vpaddl", "s",
4382 int_arm_neon_vpaddls>;
4383 defm VPADDLu : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00101, 0, "vpaddl", "u",
4384 int_arm_neon_vpaddlu>;
4386 // VPADAL : Vector Pairwise Add and Accumulate Long
4387 defm VPADALs : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01100, 0, "vpadal", "s",
4388 int_arm_neon_vpadals>;
4389 defm VPADALu : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01101, 0, "vpadal", "u",
4390 int_arm_neon_vpadalu>;
4392 // VPMAX : Vector Pairwise Maximum
4393 def VPMAXs8 : N3VDInt<0, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4394 "s8", v8i8, v8i8, int_arm_neon_vpmaxs, 0>;
4395 def VPMAXs16 : N3VDInt<0, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4396 "s16", v4i16, v4i16, int_arm_neon_vpmaxs, 0>;
4397 def VPMAXs32 : N3VDInt<0, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4398 "s32", v2i32, v2i32, int_arm_neon_vpmaxs, 0>;
4399 def VPMAXu8 : N3VDInt<1, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4400 "u8", v8i8, v8i8, int_arm_neon_vpmaxu, 0>;
4401 def VPMAXu16 : N3VDInt<1, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4402 "u16", v4i16, v4i16, int_arm_neon_vpmaxu, 0>;
4403 def VPMAXu32 : N3VDInt<1, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4404 "u32", v2i32, v2i32, int_arm_neon_vpmaxu, 0>;
4405 def VPMAXf : N3VDInt<1, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmax",
4406 "f32", v2f32, v2f32, int_arm_neon_vpmaxs, 0>;
4408 // VPMIN : Vector Pairwise Minimum
4409 def VPMINs8 : N3VDInt<0, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4410 "s8", v8i8, v8i8, int_arm_neon_vpmins, 0>;
4411 def VPMINs16 : N3VDInt<0, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4412 "s16", v4i16, v4i16, int_arm_neon_vpmins, 0>;
4413 def VPMINs32 : N3VDInt<0, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4414 "s32", v2i32, v2i32, int_arm_neon_vpmins, 0>;
4415 def VPMINu8 : N3VDInt<1, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4416 "u8", v8i8, v8i8, int_arm_neon_vpminu, 0>;
4417 def VPMINu16 : N3VDInt<1, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4418 "u16", v4i16, v4i16, int_arm_neon_vpminu, 0>;
4419 def VPMINu32 : N3VDInt<1, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4420 "u32", v2i32, v2i32, int_arm_neon_vpminu, 0>;
4421 def VPMINf : N3VDInt<1, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmin",
4422 "f32", v2f32, v2f32, int_arm_neon_vpmins, 0>;
4424 // Vector Reciprocal and Reciprocal Square Root Estimate and Step.
4426 // VRECPE : Vector Reciprocal Estimate
4427 def VRECPEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
4428 IIC_VUNAD, "vrecpe", "u32",
4429 v2i32, v2i32, int_arm_neon_vrecpe>;
4430 def VRECPEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
4431 IIC_VUNAQ, "vrecpe", "u32",
4432 v4i32, v4i32, int_arm_neon_vrecpe>;
4433 def VRECPEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
4434 IIC_VUNAD, "vrecpe", "f32",
4435 v2f32, v2f32, int_arm_neon_vrecpe>;
4436 def VRECPEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
4437 IIC_VUNAQ, "vrecpe", "f32",
4438 v4f32, v4f32, int_arm_neon_vrecpe>;
4440 // VRECPS : Vector Reciprocal Step
4441 def VRECPSfd : N3VDInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
4442 IIC_VRECSD, "vrecps", "f32",
4443 v2f32, v2f32, int_arm_neon_vrecps, 1>;
4444 def VRECPSfq : N3VQInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
4445 IIC_VRECSQ, "vrecps", "f32",
4446 v4f32, v4f32, int_arm_neon_vrecps, 1>;
4448 // VRSQRTE : Vector Reciprocal Square Root Estimate
4449 def VRSQRTEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
4450 IIC_VUNAD, "vrsqrte", "u32",
4451 v2i32, v2i32, int_arm_neon_vrsqrte>;
4452 def VRSQRTEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
4453 IIC_VUNAQ, "vrsqrte", "u32",
4454 v4i32, v4i32, int_arm_neon_vrsqrte>;
4455 def VRSQRTEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
4456 IIC_VUNAD, "vrsqrte", "f32",
4457 v2f32, v2f32, int_arm_neon_vrsqrte>;
4458 def VRSQRTEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
4459 IIC_VUNAQ, "vrsqrte", "f32",
4460 v4f32, v4f32, int_arm_neon_vrsqrte>;
4462 // VRSQRTS : Vector Reciprocal Square Root Step
4463 def VRSQRTSfd : N3VDInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
4464 IIC_VRECSD, "vrsqrts", "f32",
4465 v2f32, v2f32, int_arm_neon_vrsqrts, 1>;
4466 def VRSQRTSfq : N3VQInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
4467 IIC_VRECSQ, "vrsqrts", "f32",
4468 v4f32, v4f32, int_arm_neon_vrsqrts, 1>;
4472 // VSHL : Vector Shift
4473 defm VSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 0, N3RegVShFrm,
4474 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
4475 "vshl", "s", int_arm_neon_vshifts>;
4476 defm VSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 0, N3RegVShFrm,
4477 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
4478 "vshl", "u", int_arm_neon_vshiftu>;
4480 // VSHL : Vector Shift Left (Immediate)
4481 defm VSHLi : N2VShL_QHSD<0, 1, 0b0101, 1, IIC_VSHLiD, "vshl", "i", NEONvshl>;
4483 // VSHR : Vector Shift Right (Immediate)
4484 defm VSHRs : N2VShR_QHSD<0, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "s",NEONvshrs>;
4485 defm VSHRu : N2VShR_QHSD<1, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "u",NEONvshru>;
4487 // VSHLL : Vector Shift Left Long
4488 defm VSHLLs : N2VLSh_QHS<0, 1, 0b1010, 0, 0, 1, "vshll", "s", NEONvshlls>;
4489 defm VSHLLu : N2VLSh_QHS<1, 1, 0b1010, 0, 0, 1, "vshll", "u", NEONvshllu>;
4491 // VSHLL : Vector Shift Left Long (with maximum shift count)
4492 class N2VLShMax<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
4493 bit op6, bit op4, string OpcodeStr, string Dt, ValueType ResTy,
4494 ValueType OpTy, Operand ImmTy, SDNode OpNode>
4495 : N2VLSh<op24, op23, op11_8, op7, op6, op4, OpcodeStr, Dt,
4496 ResTy, OpTy, ImmTy, OpNode> {
4497 let Inst{21-16} = op21_16;
4498 let DecoderMethod = "DecodeVSHLMaxInstruction";
4500 def VSHLLi8 : N2VLShMax<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll", "i8",
4501 v8i16, v8i8, imm8, NEONvshlli>;
4502 def VSHLLi16 : N2VLShMax<1, 1, 0b110110, 0b0011, 0, 0, 0, "vshll", "i16",
4503 v4i32, v4i16, imm16, NEONvshlli>;
4504 def VSHLLi32 : N2VLShMax<1, 1, 0b111010, 0b0011, 0, 0, 0, "vshll", "i32",
4505 v2i64, v2i32, imm32, NEONvshlli>;
4507 // VSHRN : Vector Shift Right and Narrow
4508 defm VSHRN : N2VNSh_HSD<0,1,0b1000,0,0,1, IIC_VSHLiD, "vshrn", "i",
4511 // VRSHL : Vector Rounding Shift
4512 defm VRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 0, N3RegVShFrm,
4513 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4514 "vrshl", "s", int_arm_neon_vrshifts>;
4515 defm VRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 0, N3RegVShFrm,
4516 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4517 "vrshl", "u", int_arm_neon_vrshiftu>;
4518 // VRSHR : Vector Rounding Shift Right
4519 defm VRSHRs : N2VShR_QHSD<0,1,0b0010,1, IIC_VSHLi4D, "vrshr", "s",NEONvrshrs>;
4520 defm VRSHRu : N2VShR_QHSD<1,1,0b0010,1, IIC_VSHLi4D, "vrshr", "u",NEONvrshru>;
4522 // VRSHRN : Vector Rounding Shift Right and Narrow
4523 defm VRSHRN : N2VNSh_HSD<0, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vrshrn", "i",
4526 // VQSHL : Vector Saturating Shift
4527 defm VQSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 1, N3RegVShFrm,
4528 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4529 "vqshl", "s", int_arm_neon_vqshifts>;
4530 defm VQSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 1, N3RegVShFrm,
4531 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4532 "vqshl", "u", int_arm_neon_vqshiftu>;
4533 // VQSHL : Vector Saturating Shift Left (Immediate)
4534 defm VQSHLsi : N2VShL_QHSD<0,1,0b0111,1, IIC_VSHLi4D, "vqshl", "s",NEONvqshls>;
4535 defm VQSHLui : N2VShL_QHSD<1,1,0b0111,1, IIC_VSHLi4D, "vqshl", "u",NEONvqshlu>;
4537 // VQSHLU : Vector Saturating Shift Left (Immediate, Unsigned)
4538 defm VQSHLsu : N2VShL_QHSD<1,1,0b0110,1, IIC_VSHLi4D,"vqshlu","s",NEONvqshlsu>;
4540 // VQSHRN : Vector Saturating Shift Right and Narrow
4541 defm VQSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "s",
4543 defm VQSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "u",
4546 // VQSHRUN : Vector Saturating Shift Right and Narrow (Unsigned)
4547 defm VQSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 0, 1, IIC_VSHLi4D, "vqshrun", "s",
4550 // VQRSHL : Vector Saturating Rounding Shift
4551 defm VQRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 1, N3RegVShFrm,
4552 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4553 "vqrshl", "s", int_arm_neon_vqrshifts>;
4554 defm VQRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 1, N3RegVShFrm,
4555 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4556 "vqrshl", "u", int_arm_neon_vqrshiftu>;
4558 // VQRSHRN : Vector Saturating Rounding Shift Right and Narrow
4559 defm VQRSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "s",
4561 defm VQRSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "u",
4564 // VQRSHRUN : Vector Saturating Rounding Shift Right and Narrow (Unsigned)
4565 defm VQRSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vqrshrun", "s",
4568 // VSRA : Vector Shift Right and Accumulate
4569 defm VSRAs : N2VShAdd_QHSD<0, 1, 0b0001, 1, "vsra", "s", NEONvshrs>;
4570 defm VSRAu : N2VShAdd_QHSD<1, 1, 0b0001, 1, "vsra", "u", NEONvshru>;
4571 // VRSRA : Vector Rounding Shift Right and Accumulate
4572 defm VRSRAs : N2VShAdd_QHSD<0, 1, 0b0011, 1, "vrsra", "s", NEONvrshrs>;
4573 defm VRSRAu : N2VShAdd_QHSD<1, 1, 0b0011, 1, "vrsra", "u", NEONvrshru>;
4575 // VSLI : Vector Shift Left and Insert
4576 defm VSLI : N2VShInsL_QHSD<1, 1, 0b0101, 1, "vsli">;
4578 // VSRI : Vector Shift Right and Insert
4579 defm VSRI : N2VShInsR_QHSD<1, 1, 0b0100, 1, "vsri">;
4581 // Vector Absolute and Saturating Absolute.
4583 // VABS : Vector Absolute Value
4584 defm VABS : N2VInt_QHS<0b11, 0b11, 0b01, 0b00110, 0,
4585 IIC_VUNAiD, IIC_VUNAiQ, "vabs", "s",
4587 def VABSfd : N2VDInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
4588 IIC_VUNAD, "vabs", "f32",
4589 v2f32, v2f32, int_arm_neon_vabs>;
4590 def VABSfq : N2VQInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
4591 IIC_VUNAQ, "vabs", "f32",
4592 v4f32, v4f32, int_arm_neon_vabs>;
4594 // VQABS : Vector Saturating Absolute Value
4595 defm VQABS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01110, 0,
4596 IIC_VQUNAiD, IIC_VQUNAiQ, "vqabs", "s",
4597 int_arm_neon_vqabs>;
4601 def vnegd : PatFrag<(ops node:$in),
4602 (sub (bitconvert (v2i32 NEONimmAllZerosV)), node:$in)>;
4603 def vnegq : PatFrag<(ops node:$in),
4604 (sub (bitconvert (v4i32 NEONimmAllZerosV)), node:$in)>;
4606 class VNEGD<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
4607 : N2V<0b11, 0b11, size, 0b01, 0b00111, 0, 0, (outs DPR:$Vd), (ins DPR:$Vm),
4608 IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
4609 [(set DPR:$Vd, (Ty (vnegd DPR:$Vm)))]>;
4610 class VNEGQ<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
4611 : N2V<0b11, 0b11, size, 0b01, 0b00111, 1, 0, (outs QPR:$Vd), (ins QPR:$Vm),
4612 IIC_VSHLiQ, OpcodeStr, Dt, "$Vd, $Vm", "",
4613 [(set QPR:$Vd, (Ty (vnegq QPR:$Vm)))]>;
4615 // VNEG : Vector Negate (integer)
4616 def VNEGs8d : VNEGD<0b00, "vneg", "s8", v8i8>;
4617 def VNEGs16d : VNEGD<0b01, "vneg", "s16", v4i16>;
4618 def VNEGs32d : VNEGD<0b10, "vneg", "s32", v2i32>;
4619 def VNEGs8q : VNEGQ<0b00, "vneg", "s8", v16i8>;
4620 def VNEGs16q : VNEGQ<0b01, "vneg", "s16", v8i16>;
4621 def VNEGs32q : VNEGQ<0b10, "vneg", "s32", v4i32>;
4623 // VNEG : Vector Negate (floating-point)
4624 def VNEGfd : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
4625 (outs DPR:$Vd), (ins DPR:$Vm), IIC_VUNAD,
4626 "vneg", "f32", "$Vd, $Vm", "",
4627 [(set DPR:$Vd, (v2f32 (fneg DPR:$Vm)))]>;
4628 def VNEGf32q : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 1, 0,
4629 (outs QPR:$Vd), (ins QPR:$Vm), IIC_VUNAQ,
4630 "vneg", "f32", "$Vd, $Vm", "",
4631 [(set QPR:$Vd, (v4f32 (fneg QPR:$Vm)))]>;
4633 def : Pat<(v8i8 (vnegd DPR:$src)), (VNEGs8d DPR:$src)>;
4634 def : Pat<(v4i16 (vnegd DPR:$src)), (VNEGs16d DPR:$src)>;
4635 def : Pat<(v2i32 (vnegd DPR:$src)), (VNEGs32d DPR:$src)>;
4636 def : Pat<(v16i8 (vnegq QPR:$src)), (VNEGs8q QPR:$src)>;
4637 def : Pat<(v8i16 (vnegq QPR:$src)), (VNEGs16q QPR:$src)>;
4638 def : Pat<(v4i32 (vnegq QPR:$src)), (VNEGs32q QPR:$src)>;
4640 // VQNEG : Vector Saturating Negate
4641 defm VQNEG : N2VInt_QHS<0b11, 0b11, 0b00, 0b01111, 0,
4642 IIC_VQUNAiD, IIC_VQUNAiQ, "vqneg", "s",
4643 int_arm_neon_vqneg>;
4645 // Vector Bit Counting Operations.
4647 // VCLS : Vector Count Leading Sign Bits
4648 defm VCLS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01000, 0,
4649 IIC_VCNTiD, IIC_VCNTiQ, "vcls", "s",
4651 // VCLZ : Vector Count Leading Zeros
4652 defm VCLZ : N2VInt_QHS<0b11, 0b11, 0b00, 0b01001, 0,
4653 IIC_VCNTiD, IIC_VCNTiQ, "vclz", "i",
4655 // VCNT : Vector Count One Bits
4656 def VCNTd : N2VDInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
4657 IIC_VCNTiD, "vcnt", "8",
4658 v8i8, v8i8, int_arm_neon_vcnt>;
4659 def VCNTq : N2VQInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
4660 IIC_VCNTiQ, "vcnt", "8",
4661 v16i8, v16i8, int_arm_neon_vcnt>;
4664 def VSWPd : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 0, 0,
4665 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
4666 "vswp", "$Vd, $Vm", "", []>;
4667 def VSWPq : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 1, 0,
4668 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
4669 "vswp", "$Vd, $Vm", "", []>;
4671 // Vector Move Operations.
4673 // VMOV : Vector Move (Register)
4674 def : InstAlias<"vmov${p} $Vd, $Vm",
4675 (VORRd DPR:$Vd, DPR:$Vm, DPR:$Vm, pred:$p)>;
4676 def : InstAlias<"vmov${p} $Vd, $Vm",
4677 (VORRq QPR:$Vd, QPR:$Vm, QPR:$Vm, pred:$p)>;
4679 // VMOV : Vector Move (Immediate)
4681 let isReMaterializable = 1 in {
4682 def VMOVv8i8 : N1ModImm<1, 0b000, 0b1110, 0, 0, 0, 1, (outs DPR:$Vd),
4683 (ins nImmSplatI8:$SIMM), IIC_VMOVImm,
4684 "vmov", "i8", "$Vd, $SIMM", "",
4685 [(set DPR:$Vd, (v8i8 (NEONvmovImm timm:$SIMM)))]>;
4686 def VMOVv16i8 : N1ModImm<1, 0b000, 0b1110, 0, 1, 0, 1, (outs QPR:$Vd),
4687 (ins nImmSplatI8:$SIMM), IIC_VMOVImm,
4688 "vmov", "i8", "$Vd, $SIMM", "",
4689 [(set QPR:$Vd, (v16i8 (NEONvmovImm timm:$SIMM)))]>;
4691 def VMOVv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 0, 1, (outs DPR:$Vd),
4692 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4693 "vmov", "i16", "$Vd, $SIMM", "",
4694 [(set DPR:$Vd, (v4i16 (NEONvmovImm timm:$SIMM)))]> {
4695 let Inst{9} = SIMM{9};
4698 def VMOVv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 0, 1, (outs QPR:$Vd),
4699 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4700 "vmov", "i16", "$Vd, $SIMM", "",
4701 [(set QPR:$Vd, (v8i16 (NEONvmovImm timm:$SIMM)))]> {
4702 let Inst{9} = SIMM{9};
4705 def VMOVv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 0, 1, (outs DPR:$Vd),
4706 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4707 "vmov", "i32", "$Vd, $SIMM", "",
4708 [(set DPR:$Vd, (v2i32 (NEONvmovImm timm:$SIMM)))]> {
4709 let Inst{11-8} = SIMM{11-8};
4712 def VMOVv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 0, 1, (outs QPR:$Vd),
4713 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4714 "vmov", "i32", "$Vd, $SIMM", "",
4715 [(set QPR:$Vd, (v4i32 (NEONvmovImm timm:$SIMM)))]> {
4716 let Inst{11-8} = SIMM{11-8};
4719 def VMOVv1i64 : N1ModImm<1, 0b000, 0b1110, 0, 0, 1, 1, (outs DPR:$Vd),
4720 (ins nImmSplatI64:$SIMM), IIC_VMOVImm,
4721 "vmov", "i64", "$Vd, $SIMM", "",
4722 [(set DPR:$Vd, (v1i64 (NEONvmovImm timm:$SIMM)))]>;
4723 def VMOVv2i64 : N1ModImm<1, 0b000, 0b1110, 0, 1, 1, 1, (outs QPR:$Vd),
4724 (ins nImmSplatI64:$SIMM), IIC_VMOVImm,
4725 "vmov", "i64", "$Vd, $SIMM", "",
4726 [(set QPR:$Vd, (v2i64 (NEONvmovImm timm:$SIMM)))]>;
4728 def VMOVv2f32 : N1ModImm<1, 0b000, 0b1111, 0, 0, 0, 1, (outs DPR:$Vd),
4729 (ins nImmVMOVF32:$SIMM), IIC_VMOVImm,
4730 "vmov", "f32", "$Vd, $SIMM", "",
4731 [(set DPR:$Vd, (v2f32 (NEONvmovFPImm timm:$SIMM)))]>;
4732 def VMOVv4f32 : N1ModImm<1, 0b000, 0b1111, 0, 1, 0, 1, (outs QPR:$Vd),
4733 (ins nImmVMOVF32:$SIMM), IIC_VMOVImm,
4734 "vmov", "f32", "$Vd, $SIMM", "",
4735 [(set QPR:$Vd, (v4f32 (NEONvmovFPImm timm:$SIMM)))]>;
4736 } // isReMaterializable
4738 // VMOV : Vector Get Lane (move scalar to ARM core register)
4740 def VGETLNs8 : NVGetLane<{1,1,1,0,0,1,?,1}, 0b1011, {?,?},
4741 (outs GPR:$R), (ins DPR:$V, VectorIndex8:$lane),
4742 IIC_VMOVSI, "vmov", "s8", "$R, $V$lane",
4743 [(set GPR:$R, (NEONvgetlanes (v8i8 DPR:$V),
4745 let Inst{21} = lane{2};
4746 let Inst{6-5} = lane{1-0};
4748 def VGETLNs16 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, {?,1},
4749 (outs GPR:$R), (ins DPR:$V, VectorIndex16:$lane),
4750 IIC_VMOVSI, "vmov", "s16", "$R, $V$lane",
4751 [(set GPR:$R, (NEONvgetlanes (v4i16 DPR:$V),
4753 let Inst{21} = lane{1};
4754 let Inst{6} = lane{0};
4756 def VGETLNu8 : NVGetLane<{1,1,1,0,1,1,?,1}, 0b1011, {?,?},
4757 (outs GPR:$R), (ins DPR:$V, VectorIndex8:$lane),
4758 IIC_VMOVSI, "vmov", "u8", "$R, $V$lane",
4759 [(set GPR:$R, (NEONvgetlaneu (v8i8 DPR:$V),
4761 let Inst{21} = lane{2};
4762 let Inst{6-5} = lane{1-0};
4764 def VGETLNu16 : NVGetLane<{1,1,1,0,1,0,?,1}, 0b1011, {?,1},
4765 (outs GPR:$R), (ins DPR:$V, VectorIndex16:$lane),
4766 IIC_VMOVSI, "vmov", "u16", "$R, $V$lane",
4767 [(set GPR:$R, (NEONvgetlaneu (v4i16 DPR:$V),
4769 let Inst{21} = lane{1};
4770 let Inst{6} = lane{0};
4772 def VGETLNi32 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, 0b00,
4773 (outs GPR:$R), (ins DPR:$V, VectorIndex32:$lane),
4774 IIC_VMOVSI, "vmov", "32", "$R, $V$lane",
4775 [(set GPR:$R, (extractelt (v2i32 DPR:$V),
4777 let Inst{21} = lane{0};
4779 // def VGETLNf32: see FMRDH and FMRDL in ARMInstrVFP.td
4780 def : Pat<(NEONvgetlanes (v16i8 QPR:$src), imm:$lane),
4781 (VGETLNs8 (v8i8 (EXTRACT_SUBREG QPR:$src,
4782 (DSubReg_i8_reg imm:$lane))),
4783 (SubReg_i8_lane imm:$lane))>;
4784 def : Pat<(NEONvgetlanes (v8i16 QPR:$src), imm:$lane),
4785 (VGETLNs16 (v4i16 (EXTRACT_SUBREG QPR:$src,
4786 (DSubReg_i16_reg imm:$lane))),
4787 (SubReg_i16_lane imm:$lane))>;
4788 def : Pat<(NEONvgetlaneu (v16i8 QPR:$src), imm:$lane),
4789 (VGETLNu8 (v8i8 (EXTRACT_SUBREG QPR:$src,
4790 (DSubReg_i8_reg imm:$lane))),
4791 (SubReg_i8_lane imm:$lane))>;
4792 def : Pat<(NEONvgetlaneu (v8i16 QPR:$src), imm:$lane),
4793 (VGETLNu16 (v4i16 (EXTRACT_SUBREG QPR:$src,
4794 (DSubReg_i16_reg imm:$lane))),
4795 (SubReg_i16_lane imm:$lane))>;
4796 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
4797 (VGETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src,
4798 (DSubReg_i32_reg imm:$lane))),
4799 (SubReg_i32_lane imm:$lane))>;
4800 def : Pat<(extractelt (v2f32 DPR:$src1), imm:$src2),
4801 (EXTRACT_SUBREG (v2f32 (COPY_TO_REGCLASS (v2f32 DPR:$src1),DPR_VFP2)),
4802 (SSubReg_f32_reg imm:$src2))>;
4803 def : Pat<(extractelt (v4f32 QPR:$src1), imm:$src2),
4804 (EXTRACT_SUBREG (v4f32 (COPY_TO_REGCLASS (v4f32 QPR:$src1),QPR_VFP2)),
4805 (SSubReg_f32_reg imm:$src2))>;
4806 //def : Pat<(extractelt (v2i64 QPR:$src1), imm:$src2),
4807 // (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
4808 def : Pat<(extractelt (v2f64 QPR:$src1), imm:$src2),
4809 (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
4812 // VMOV : Vector Set Lane (move ARM core register to scalar)
4814 let Constraints = "$src1 = $V" in {
4815 def VSETLNi8 : NVSetLane<{1,1,1,0,0,1,?,0}, 0b1011, {?,?}, (outs DPR:$V),
4816 (ins DPR:$src1, GPR:$R, VectorIndex8:$lane),
4817 IIC_VMOVISL, "vmov", "8", "$V$lane, $R",
4818 [(set DPR:$V, (vector_insert (v8i8 DPR:$src1),
4819 GPR:$R, imm:$lane))]> {
4820 let Inst{21} = lane{2};
4821 let Inst{6-5} = lane{1-0};
4823 def VSETLNi16 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, {?,1}, (outs DPR:$V),
4824 (ins DPR:$src1, GPR:$R, VectorIndex16:$lane),
4825 IIC_VMOVISL, "vmov", "16", "$V$lane, $R",
4826 [(set DPR:$V, (vector_insert (v4i16 DPR:$src1),
4827 GPR:$R, imm:$lane))]> {
4828 let Inst{21} = lane{1};
4829 let Inst{6} = lane{0};
4831 def VSETLNi32 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, 0b00, (outs DPR:$V),
4832 (ins DPR:$src1, GPR:$R, VectorIndex32:$lane),
4833 IIC_VMOVISL, "vmov", "32", "$V$lane, $R",
4834 [(set DPR:$V, (insertelt (v2i32 DPR:$src1),
4835 GPR:$R, imm:$lane))]> {
4836 let Inst{21} = lane{0};
4839 def : Pat<(vector_insert (v16i8 QPR:$src1), GPR:$src2, imm:$lane),
4840 (v16i8 (INSERT_SUBREG QPR:$src1,
4841 (v8i8 (VSETLNi8 (v8i8 (EXTRACT_SUBREG QPR:$src1,
4842 (DSubReg_i8_reg imm:$lane))),
4843 GPR:$src2, (SubReg_i8_lane imm:$lane))),
4844 (DSubReg_i8_reg imm:$lane)))>;
4845 def : Pat<(vector_insert (v8i16 QPR:$src1), GPR:$src2, imm:$lane),
4846 (v8i16 (INSERT_SUBREG QPR:$src1,
4847 (v4i16 (VSETLNi16 (v4i16 (EXTRACT_SUBREG QPR:$src1,
4848 (DSubReg_i16_reg imm:$lane))),
4849 GPR:$src2, (SubReg_i16_lane imm:$lane))),
4850 (DSubReg_i16_reg imm:$lane)))>;
4851 def : Pat<(insertelt (v4i32 QPR:$src1), GPR:$src2, imm:$lane),
4852 (v4i32 (INSERT_SUBREG QPR:$src1,
4853 (v2i32 (VSETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src1,
4854 (DSubReg_i32_reg imm:$lane))),
4855 GPR:$src2, (SubReg_i32_lane imm:$lane))),
4856 (DSubReg_i32_reg imm:$lane)))>;
4858 def : Pat<(v2f32 (insertelt DPR:$src1, SPR:$src2, imm:$src3)),
4859 (INSERT_SUBREG (v2f32 (COPY_TO_REGCLASS DPR:$src1, DPR_VFP2)),
4860 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
4861 def : Pat<(v4f32 (insertelt QPR:$src1, SPR:$src2, imm:$src3)),
4862 (INSERT_SUBREG (v4f32 (COPY_TO_REGCLASS QPR:$src1, QPR_VFP2)),
4863 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
4865 //def : Pat<(v2i64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
4866 // (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
4867 def : Pat<(v2f64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
4868 (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
4870 def : Pat<(v2f32 (scalar_to_vector SPR:$src)),
4871 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
4872 def : Pat<(v2f64 (scalar_to_vector (f64 DPR:$src))),
4873 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
4874 def : Pat<(v4f32 (scalar_to_vector SPR:$src)),
4875 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
4877 def : Pat<(v8i8 (scalar_to_vector GPR:$src)),
4878 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4879 def : Pat<(v4i16 (scalar_to_vector GPR:$src)),
4880 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4881 def : Pat<(v2i32 (scalar_to_vector GPR:$src)),
4882 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4884 def : Pat<(v16i8 (scalar_to_vector GPR:$src)),
4885 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4886 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4888 def : Pat<(v8i16 (scalar_to_vector GPR:$src)),
4889 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
4890 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4892 def : Pat<(v4i32 (scalar_to_vector GPR:$src)),
4893 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
4894 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4897 // VDUP : Vector Duplicate (from ARM core register to all elements)
4899 class VDUPD<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
4900 : NVDup<opcod1, 0b1011, opcod3, (outs DPR:$V), (ins GPR:$R),
4901 IIC_VMOVIS, "vdup", Dt, "$V, $R",
4902 [(set DPR:$V, (Ty (NEONvdup (i32 GPR:$R))))]>;
4903 class VDUPQ<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
4904 : NVDup<opcod1, 0b1011, opcod3, (outs QPR:$V), (ins GPR:$R),
4905 IIC_VMOVIS, "vdup", Dt, "$V, $R",
4906 [(set QPR:$V, (Ty (NEONvdup (i32 GPR:$R))))]>;
4908 def VDUP8d : VDUPD<0b11101100, 0b00, "8", v8i8>;
4909 def VDUP16d : VDUPD<0b11101000, 0b01, "16", v4i16>;
4910 def VDUP32d : VDUPD<0b11101000, 0b00, "32", v2i32>;
4911 def VDUP8q : VDUPQ<0b11101110, 0b00, "8", v16i8>;
4912 def VDUP16q : VDUPQ<0b11101010, 0b01, "16", v8i16>;
4913 def VDUP32q : VDUPQ<0b11101010, 0b00, "32", v4i32>;
4915 def : Pat<(v2f32 (NEONvdup (f32 (bitconvert GPR:$R)))), (VDUP32d GPR:$R)>;
4916 def : Pat<(v4f32 (NEONvdup (f32 (bitconvert GPR:$R)))), (VDUP32q GPR:$R)>;
4918 // VDUP : Vector Duplicate Lane (from scalar to all elements)
4920 class VDUPLND<bits<4> op19_16, string OpcodeStr, string Dt,
4921 ValueType Ty, Operand IdxTy>
4922 : NVDupLane<op19_16, 0, (outs DPR:$Vd), (ins DPR:$Vm, IdxTy:$lane),
4923 IIC_VMOVD, OpcodeStr, Dt, "$Vd, $Vm$lane",
4924 [(set DPR:$Vd, (Ty (NEONvduplane (Ty DPR:$Vm), imm:$lane)))]>;
4926 class VDUPLNQ<bits<4> op19_16, string OpcodeStr, string Dt,
4927 ValueType ResTy, ValueType OpTy, Operand IdxTy>
4928 : NVDupLane<op19_16, 1, (outs QPR:$Vd), (ins DPR:$Vm, IdxTy:$lane),
4929 IIC_VMOVQ, OpcodeStr, Dt, "$Vd, $Vm$lane",
4930 [(set QPR:$Vd, (ResTy (NEONvduplane (OpTy DPR:$Vm),
4931 VectorIndex32:$lane)))]>;
4933 // Inst{19-16} is partially specified depending on the element size.
4935 def VDUPLN8d : VDUPLND<{?,?,?,1}, "vdup", "8", v8i8, VectorIndex8> {
4937 let Inst{19-17} = lane{2-0};
4939 def VDUPLN16d : VDUPLND<{?,?,1,0}, "vdup", "16", v4i16, VectorIndex16> {
4941 let Inst{19-18} = lane{1-0};
4943 def VDUPLN32d : VDUPLND<{?,1,0,0}, "vdup", "32", v2i32, VectorIndex32> {
4945 let Inst{19} = lane{0};
4947 def VDUPLN8q : VDUPLNQ<{?,?,?,1}, "vdup", "8", v16i8, v8i8, VectorIndex8> {
4949 let Inst{19-17} = lane{2-0};
4951 def VDUPLN16q : VDUPLNQ<{?,?,1,0}, "vdup", "16", v8i16, v4i16, VectorIndex16> {
4953 let Inst{19-18} = lane{1-0};
4955 def VDUPLN32q : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4i32, v2i32, VectorIndex32> {
4957 let Inst{19} = lane{0};
4960 def : Pat<(v2f32 (NEONvduplane (v2f32 DPR:$Vm), imm:$lane)),
4961 (VDUPLN32d DPR:$Vm, imm:$lane)>;
4963 def : Pat<(v4f32 (NEONvduplane (v2f32 DPR:$Vm), imm:$lane)),
4964 (VDUPLN32q DPR:$Vm, imm:$lane)>;
4966 def : Pat<(v16i8 (NEONvduplane (v16i8 QPR:$src), imm:$lane)),
4967 (v16i8 (VDUPLN8q (v8i8 (EXTRACT_SUBREG QPR:$src,
4968 (DSubReg_i8_reg imm:$lane))),
4969 (SubReg_i8_lane imm:$lane)))>;
4970 def : Pat<(v8i16 (NEONvduplane (v8i16 QPR:$src), imm:$lane)),
4971 (v8i16 (VDUPLN16q (v4i16 (EXTRACT_SUBREG QPR:$src,
4972 (DSubReg_i16_reg imm:$lane))),
4973 (SubReg_i16_lane imm:$lane)))>;
4974 def : Pat<(v4i32 (NEONvduplane (v4i32 QPR:$src), imm:$lane)),
4975 (v4i32 (VDUPLN32q (v2i32 (EXTRACT_SUBREG QPR:$src,
4976 (DSubReg_i32_reg imm:$lane))),
4977 (SubReg_i32_lane imm:$lane)))>;
4978 def : Pat<(v4f32 (NEONvduplane (v4f32 QPR:$src), imm:$lane)),
4979 (v4f32 (VDUPLN32q (v2f32 (EXTRACT_SUBREG QPR:$src,
4980 (DSubReg_i32_reg imm:$lane))),
4981 (SubReg_i32_lane imm:$lane)))>;
4983 def VDUPfdf : PseudoNeonI<(outs DPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
4984 [(set DPR:$dst, (v2f32 (NEONvdup (f32 SPR:$src))))]>;
4985 def VDUPfqf : PseudoNeonI<(outs QPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
4986 [(set QPR:$dst, (v4f32 (NEONvdup (f32 SPR:$src))))]>;
4988 // VMOVN : Vector Narrowing Move
4989 defm VMOVN : N2VN_HSD<0b11,0b11,0b10,0b00100,0,0, IIC_VMOVN,
4990 "vmovn", "i", trunc>;
4991 // VQMOVN : Vector Saturating Narrowing Move
4992 defm VQMOVNs : N2VNInt_HSD<0b11,0b11,0b10,0b00101,0,0, IIC_VQUNAiD,
4993 "vqmovn", "s", int_arm_neon_vqmovns>;
4994 defm VQMOVNu : N2VNInt_HSD<0b11,0b11,0b10,0b00101,1,0, IIC_VQUNAiD,
4995 "vqmovn", "u", int_arm_neon_vqmovnu>;
4996 defm VQMOVNsu : N2VNInt_HSD<0b11,0b11,0b10,0b00100,1,0, IIC_VQUNAiD,
4997 "vqmovun", "s", int_arm_neon_vqmovnsu>;
4998 // VMOVL : Vector Lengthening Move
4999 defm VMOVLs : N2VL_QHS<0b01,0b10100,0,1, "vmovl", "s", sext>;
5000 defm VMOVLu : N2VL_QHS<0b11,0b10100,0,1, "vmovl", "u", zext>;
5002 // Vector Conversions.
5004 // VCVT : Vector Convert Between Floating-Point and Integers
5005 def VCVTf2sd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
5006 v2i32, v2f32, fp_to_sint>;
5007 def VCVTf2ud : N2VD<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
5008 v2i32, v2f32, fp_to_uint>;
5009 def VCVTs2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
5010 v2f32, v2i32, sint_to_fp>;
5011 def VCVTu2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
5012 v2f32, v2i32, uint_to_fp>;
5014 def VCVTf2sq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
5015 v4i32, v4f32, fp_to_sint>;
5016 def VCVTf2uq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
5017 v4i32, v4f32, fp_to_uint>;
5018 def VCVTs2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
5019 v4f32, v4i32, sint_to_fp>;
5020 def VCVTu2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
5021 v4f32, v4i32, uint_to_fp>;
5023 // VCVT : Vector Convert Between Floating-Point and Fixed-Point.
5024 let DecoderMethod = "DecodeVCVTD" in {
5025 def VCVTf2xsd : N2VCvtD<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
5026 v2i32, v2f32, int_arm_neon_vcvtfp2fxs>;
5027 def VCVTf2xud : N2VCvtD<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
5028 v2i32, v2f32, int_arm_neon_vcvtfp2fxu>;
5029 def VCVTxs2fd : N2VCvtD<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
5030 v2f32, v2i32, int_arm_neon_vcvtfxs2fp>;
5031 def VCVTxu2fd : N2VCvtD<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
5032 v2f32, v2i32, int_arm_neon_vcvtfxu2fp>;
5035 let DecoderMethod = "DecodeVCVTQ" in {
5036 def VCVTf2xsq : N2VCvtQ<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
5037 v4i32, v4f32, int_arm_neon_vcvtfp2fxs>;
5038 def VCVTf2xuq : N2VCvtQ<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
5039 v4i32, v4f32, int_arm_neon_vcvtfp2fxu>;
5040 def VCVTxs2fq : N2VCvtQ<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
5041 v4f32, v4i32, int_arm_neon_vcvtfxs2fp>;
5042 def VCVTxu2fq : N2VCvtQ<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
5043 v4f32, v4i32, int_arm_neon_vcvtfxu2fp>;
5046 // VCVT : Vector Convert Between Half-Precision and Single-Precision.
5047 def VCVTf2h : N2VNInt<0b11, 0b11, 0b01, 0b10, 0b01100, 0, 0,
5048 IIC_VUNAQ, "vcvt", "f16.f32",
5049 v4i16, v4f32, int_arm_neon_vcvtfp2hf>,
5050 Requires<[HasNEON, HasFP16]>;
5051 def VCVTh2f : N2VLInt<0b11, 0b11, 0b01, 0b10, 0b01110, 0, 0,
5052 IIC_VUNAQ, "vcvt", "f32.f16",
5053 v4f32, v4i16, int_arm_neon_vcvthf2fp>,
5054 Requires<[HasNEON, HasFP16]>;
5058 // VREV64 : Vector Reverse elements within 64-bit doublewords
5060 class VREV64D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
5061 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 0, 0, (outs DPR:$Vd),
5062 (ins DPR:$Vm), IIC_VMOVD,
5063 OpcodeStr, Dt, "$Vd, $Vm", "",
5064 [(set DPR:$Vd, (Ty (NEONvrev64 (Ty DPR:$Vm))))]>;
5065 class VREV64Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
5066 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 1, 0, (outs QPR:$Vd),
5067 (ins QPR:$Vm), IIC_VMOVQ,
5068 OpcodeStr, Dt, "$Vd, $Vm", "",
5069 [(set QPR:$Vd, (Ty (NEONvrev64 (Ty QPR:$Vm))))]>;
5071 def VREV64d8 : VREV64D<0b00, "vrev64", "8", v8i8>;
5072 def VREV64d16 : VREV64D<0b01, "vrev64", "16", v4i16>;
5073 def VREV64d32 : VREV64D<0b10, "vrev64", "32", v2i32>;
5074 def : Pat<(v2f32 (NEONvrev64 (v2f32 DPR:$Vm))), (VREV64d32 DPR:$Vm)>;
5076 def VREV64q8 : VREV64Q<0b00, "vrev64", "8", v16i8>;
5077 def VREV64q16 : VREV64Q<0b01, "vrev64", "16", v8i16>;
5078 def VREV64q32 : VREV64Q<0b10, "vrev64", "32", v4i32>;
5079 def : Pat<(v4f32 (NEONvrev64 (v4f32 QPR:$Vm))), (VREV64q32 QPR:$Vm)>;
5081 // VREV32 : Vector Reverse elements within 32-bit words
5083 class VREV32D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
5084 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 0, 0, (outs DPR:$Vd),
5085 (ins DPR:$Vm), IIC_VMOVD,
5086 OpcodeStr, Dt, "$Vd, $Vm", "",
5087 [(set DPR:$Vd, (Ty (NEONvrev32 (Ty DPR:$Vm))))]>;
5088 class VREV32Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
5089 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 1, 0, (outs QPR:$Vd),
5090 (ins QPR:$Vm), IIC_VMOVQ,
5091 OpcodeStr, Dt, "$Vd, $Vm", "",
5092 [(set QPR:$Vd, (Ty (NEONvrev32 (Ty QPR:$Vm))))]>;
5094 def VREV32d8 : VREV32D<0b00, "vrev32", "8", v8i8>;
5095 def VREV32d16 : VREV32D<0b01, "vrev32", "16", v4i16>;
5097 def VREV32q8 : VREV32Q<0b00, "vrev32", "8", v16i8>;
5098 def VREV32q16 : VREV32Q<0b01, "vrev32", "16", v8i16>;
5100 // VREV16 : Vector Reverse elements within 16-bit halfwords
5102 class VREV16D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
5103 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 0, 0, (outs DPR:$Vd),
5104 (ins DPR:$Vm), IIC_VMOVD,
5105 OpcodeStr, Dt, "$Vd, $Vm", "",
5106 [(set DPR:$Vd, (Ty (NEONvrev16 (Ty DPR:$Vm))))]>;
5107 class VREV16Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
5108 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 1, 0, (outs QPR:$Vd),
5109 (ins QPR:$Vm), IIC_VMOVQ,
5110 OpcodeStr, Dt, "$Vd, $Vm", "",
5111 [(set QPR:$Vd, (Ty (NEONvrev16 (Ty QPR:$Vm))))]>;
5113 def VREV16d8 : VREV16D<0b00, "vrev16", "8", v8i8>;
5114 def VREV16q8 : VREV16Q<0b00, "vrev16", "8", v16i8>;
5116 // Other Vector Shuffles.
5118 // Aligned extractions: really just dropping registers
5120 class AlignedVEXTq<ValueType DestTy, ValueType SrcTy, SDNodeXForm LaneCVT>
5121 : Pat<(DestTy (vector_extract_subvec (SrcTy QPR:$src), (i32 imm:$start))),
5122 (EXTRACT_SUBREG (SrcTy QPR:$src), (LaneCVT imm:$start))>;
5124 def : AlignedVEXTq<v8i8, v16i8, DSubReg_i8_reg>;
5126 def : AlignedVEXTq<v4i16, v8i16, DSubReg_i16_reg>;
5128 def : AlignedVEXTq<v2i32, v4i32, DSubReg_i32_reg>;
5130 def : AlignedVEXTq<v1i64, v2i64, DSubReg_f64_reg>;
5132 def : AlignedVEXTq<v2f32, v4f32, DSubReg_i32_reg>;
5135 // VEXT : Vector Extract
5137 class VEXTd<string OpcodeStr, string Dt, ValueType Ty, Operand immTy>
5138 : N3V<0,1,0b11,{?,?,?,?},0,0, (outs DPR:$Vd),
5139 (ins DPR:$Vn, DPR:$Vm, immTy:$index), NVExtFrm,
5140 IIC_VEXTD, OpcodeStr, Dt, "$Vd, $Vn, $Vm, $index", "",
5141 [(set DPR:$Vd, (Ty (NEONvext (Ty DPR:$Vn),
5142 (Ty DPR:$Vm), imm:$index)))]> {
5144 let Inst{11-8} = index{3-0};
5147 class VEXTq<string OpcodeStr, string Dt, ValueType Ty, Operand immTy>
5148 : N3V<0,1,0b11,{?,?,?,?},1,0, (outs QPR:$Vd),
5149 (ins QPR:$Vn, QPR:$Vm, imm0_15:$index), NVExtFrm,
5150 IIC_VEXTQ, OpcodeStr, Dt, "$Vd, $Vn, $Vm, $index", "",
5151 [(set QPR:$Vd, (Ty (NEONvext (Ty QPR:$Vn),
5152 (Ty QPR:$Vm), imm:$index)))]> {
5154 let Inst{11-8} = index{3-0};
5157 def VEXTd8 : VEXTd<"vext", "8", v8i8, imm0_7> {
5158 let Inst{11-8} = index{3-0};
5160 def VEXTd16 : VEXTd<"vext", "16", v4i16, imm0_3> {
5161 let Inst{11-9} = index{2-0};
5164 def VEXTd32 : VEXTd<"vext", "32", v2i32, imm0_1> {
5165 let Inst{11-10} = index{1-0};
5166 let Inst{9-8} = 0b00;
5168 def : Pat<(v2f32 (NEONvext (v2f32 DPR:$Vn),
5171 (VEXTd32 DPR:$Vn, DPR:$Vm, imm:$index)>;
5173 def VEXTq8 : VEXTq<"vext", "8", v16i8, imm0_15> {
5174 let Inst{11-8} = index{3-0};
5176 def VEXTq16 : VEXTq<"vext", "16", v8i16, imm0_7> {
5177 let Inst{11-9} = index{2-0};
5180 def VEXTq32 : VEXTq<"vext", "32", v4i32, imm0_3> {
5181 let Inst{11-10} = index{1-0};
5182 let Inst{9-8} = 0b00;
5184 def VEXTq64 : VEXTq<"vext", "64", v2i64, imm0_1> {
5185 let Inst{11} = index{0};
5186 let Inst{10-8} = 0b000;
5188 def : Pat<(v4f32 (NEONvext (v4f32 QPR:$Vn),
5191 (VEXTq32 QPR:$Vn, QPR:$Vm, imm:$index)>;
5193 // VTRN : Vector Transpose
5195 def VTRNd8 : N2VDShuffle<0b00, 0b00001, "vtrn", "8">;
5196 def VTRNd16 : N2VDShuffle<0b01, 0b00001, "vtrn", "16">;
5197 def VTRNd32 : N2VDShuffle<0b10, 0b00001, "vtrn", "32">;
5199 def VTRNq8 : N2VQShuffle<0b00, 0b00001, IIC_VPERMQ, "vtrn", "8">;
5200 def VTRNq16 : N2VQShuffle<0b01, 0b00001, IIC_VPERMQ, "vtrn", "16">;
5201 def VTRNq32 : N2VQShuffle<0b10, 0b00001, IIC_VPERMQ, "vtrn", "32">;
5203 // VUZP : Vector Unzip (Deinterleave)
5205 def VUZPd8 : N2VDShuffle<0b00, 0b00010, "vuzp", "8">;
5206 def VUZPd16 : N2VDShuffle<0b01, 0b00010, "vuzp", "16">;
5207 def VUZPd32 : N2VDShuffle<0b10, 0b00010, "vuzp", "32">;
5209 def VUZPq8 : N2VQShuffle<0b00, 0b00010, IIC_VPERMQ3, "vuzp", "8">;
5210 def VUZPq16 : N2VQShuffle<0b01, 0b00010, IIC_VPERMQ3, "vuzp", "16">;
5211 def VUZPq32 : N2VQShuffle<0b10, 0b00010, IIC_VPERMQ3, "vuzp", "32">;
5213 // VZIP : Vector Zip (Interleave)
5215 def VZIPd8 : N2VDShuffle<0b00, 0b00011, "vzip", "8">;
5216 def VZIPd16 : N2VDShuffle<0b01, 0b00011, "vzip", "16">;
5217 def VZIPd32 : N2VDShuffle<0b10, 0b00011, "vzip", "32">;
5219 def VZIPq8 : N2VQShuffle<0b00, 0b00011, IIC_VPERMQ3, "vzip", "8">;
5220 def VZIPq16 : N2VQShuffle<0b01, 0b00011, IIC_VPERMQ3, "vzip", "16">;
5221 def VZIPq32 : N2VQShuffle<0b10, 0b00011, IIC_VPERMQ3, "vzip", "32">;
5223 // Vector Table Lookup and Table Extension.
5225 // VTBL : Vector Table Lookup
5226 let DecoderMethod = "DecodeTBLInstruction" in {
5228 : N3V<1,1,0b11,0b1000,0,0, (outs DPR:$Vd),
5229 (ins VecListOneD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTB1,
5230 "vtbl", "8", "$Vd, $Vn, $Vm", "",
5231 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbl1 VecListOneD:$Vn, DPR:$Vm)))]>;
5232 let hasExtraSrcRegAllocReq = 1 in {
5234 : N3V<1,1,0b11,0b1001,0,0, (outs DPR:$Vd),
5235 (ins VecListTwoD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTB2,
5236 "vtbl", "8", "$Vd, $Vn, $Vm", "", []>;
5238 : N3V<1,1,0b11,0b1010,0,0, (outs DPR:$Vd),
5239 (ins VecListThreeD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTB3,
5240 "vtbl", "8", "$Vd, $Vn, $Vm", "", []>;
5242 : N3V<1,1,0b11,0b1011,0,0, (outs DPR:$Vd),
5243 (ins VecListFourD:$Vn, DPR:$Vm),
5245 "vtbl", "8", "$Vd, $Vn, $Vm", "", []>;
5246 } // hasExtraSrcRegAllocReq = 1
5249 : PseudoNeonI<(outs DPR:$dst), (ins QPR:$tbl, DPR:$src), IIC_VTB2, "", []>;
5251 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB3, "", []>;
5253 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB4, "", []>;
5255 // VTBX : Vector Table Extension
5257 : N3V<1,1,0b11,0b1000,1,0, (outs DPR:$Vd),
5258 (ins DPR:$orig, VecListOneD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTBX1,
5259 "vtbx", "8", "$Vd, $Vn, $Vm", "$orig = $Vd",
5260 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbx1
5261 DPR:$orig, VecListOneD:$Vn, DPR:$Vm)))]>;
5262 let hasExtraSrcRegAllocReq = 1 in {
5264 : N3V<1,1,0b11,0b1001,1,0, (outs DPR:$Vd),
5265 (ins DPR:$orig, VecListTwoD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTBX2,
5266 "vtbx", "8", "$Vd, $Vn, $Vm", "$orig = $Vd", []>;
5268 : N3V<1,1,0b11,0b1010,1,0, (outs DPR:$Vd),
5269 (ins DPR:$orig, VecListThreeD:$Vn, DPR:$Vm),
5270 NVTBLFrm, IIC_VTBX3,
5271 "vtbx", "8", "$Vd, $Vn, $Vm",
5274 : N3V<1,1,0b11,0b1011,1,0, (outs DPR:$Vd),
5275 (ins DPR:$orig, VecListFourD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTBX4,
5276 "vtbx", "8", "$Vd, $Vn, $Vm",
5278 } // hasExtraSrcRegAllocReq = 1
5281 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QPR:$tbl, DPR:$src),
5282 IIC_VTBX2, "$orig = $dst", []>;
5284 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
5285 IIC_VTBX3, "$orig = $dst", []>;
5287 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
5288 IIC_VTBX4, "$orig = $dst", []>;
5289 } // DecoderMethod = "DecodeTBLInstruction"
5291 //===----------------------------------------------------------------------===//
5292 // NEON instructions for single-precision FP math
5293 //===----------------------------------------------------------------------===//
5295 class N2VSPat<SDNode OpNode, NeonI Inst>
5296 : NEONFPPat<(f32 (OpNode SPR:$a)),
5298 (v2f32 (COPY_TO_REGCLASS (Inst
5300 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5301 SPR:$a, ssub_0)), DPR_VFP2)), ssub_0)>;
5303 class N3VSPat<SDNode OpNode, NeonI Inst>
5304 : NEONFPPat<(f32 (OpNode SPR:$a, SPR:$b)),
5306 (v2f32 (COPY_TO_REGCLASS (Inst
5308 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5311 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5312 SPR:$b, ssub_0)), DPR_VFP2)), ssub_0)>;
5314 class N3VSMulOpPat<SDNode MulNode, SDNode OpNode, NeonI Inst>
5315 : NEONFPPat<(f32 (OpNode SPR:$acc, (f32 (MulNode SPR:$a, SPR:$b)))),
5317 (v2f32 (COPY_TO_REGCLASS (Inst
5319 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5322 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5325 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5326 SPR:$b, ssub_0)), DPR_VFP2)), ssub_0)>;
5328 def : N3VSPat<fadd, VADDfd>;
5329 def : N3VSPat<fsub, VSUBfd>;
5330 def : N3VSPat<fmul, VMULfd>;
5331 def : N3VSMulOpPat<fmul, fadd, VMLAfd>,
5332 Requires<[HasNEON, UseNEONForFP, UseFPVMLx]>;
5333 def : N3VSMulOpPat<fmul, fsub, VMLSfd>,
5334 Requires<[HasNEON, UseNEONForFP, UseFPVMLx]>;
5335 def : N2VSPat<fabs, VABSfd>;
5336 def : N2VSPat<fneg, VNEGfd>;
5337 def : N3VSPat<NEONfmax, VMAXfd>;
5338 def : N3VSPat<NEONfmin, VMINfd>;
5339 def : N2VSPat<arm_ftosi, VCVTf2sd>;
5340 def : N2VSPat<arm_ftoui, VCVTf2ud>;
5341 def : N2VSPat<arm_sitof, VCVTs2fd>;
5342 def : N2VSPat<arm_uitof, VCVTu2fd>;
5344 //===----------------------------------------------------------------------===//
5345 // Non-Instruction Patterns
5346 //===----------------------------------------------------------------------===//
5349 def : Pat<(v1i64 (bitconvert (v2i32 DPR:$src))), (v1i64 DPR:$src)>;
5350 def : Pat<(v1i64 (bitconvert (v4i16 DPR:$src))), (v1i64 DPR:$src)>;
5351 def : Pat<(v1i64 (bitconvert (v8i8 DPR:$src))), (v1i64 DPR:$src)>;
5352 def : Pat<(v1i64 (bitconvert (f64 DPR:$src))), (v1i64 DPR:$src)>;
5353 def : Pat<(v1i64 (bitconvert (v2f32 DPR:$src))), (v1i64 DPR:$src)>;
5354 def : Pat<(v2i32 (bitconvert (v1i64 DPR:$src))), (v2i32 DPR:$src)>;
5355 def : Pat<(v2i32 (bitconvert (v4i16 DPR:$src))), (v2i32 DPR:$src)>;
5356 def : Pat<(v2i32 (bitconvert (v8i8 DPR:$src))), (v2i32 DPR:$src)>;
5357 def : Pat<(v2i32 (bitconvert (f64 DPR:$src))), (v2i32 DPR:$src)>;
5358 def : Pat<(v2i32 (bitconvert (v2f32 DPR:$src))), (v2i32 DPR:$src)>;
5359 def : Pat<(v4i16 (bitconvert (v1i64 DPR:$src))), (v4i16 DPR:$src)>;
5360 def : Pat<(v4i16 (bitconvert (v2i32 DPR:$src))), (v4i16 DPR:$src)>;
5361 def : Pat<(v4i16 (bitconvert (v8i8 DPR:$src))), (v4i16 DPR:$src)>;
5362 def : Pat<(v4i16 (bitconvert (f64 DPR:$src))), (v4i16 DPR:$src)>;
5363 def : Pat<(v4i16 (bitconvert (v2f32 DPR:$src))), (v4i16 DPR:$src)>;
5364 def : Pat<(v8i8 (bitconvert (v1i64 DPR:$src))), (v8i8 DPR:$src)>;
5365 def : Pat<(v8i8 (bitconvert (v2i32 DPR:$src))), (v8i8 DPR:$src)>;
5366 def : Pat<(v8i8 (bitconvert (v4i16 DPR:$src))), (v8i8 DPR:$src)>;
5367 def : Pat<(v8i8 (bitconvert (f64 DPR:$src))), (v8i8 DPR:$src)>;
5368 def : Pat<(v8i8 (bitconvert (v2f32 DPR:$src))), (v8i8 DPR:$src)>;
5369 def : Pat<(f64 (bitconvert (v1i64 DPR:$src))), (f64 DPR:$src)>;
5370 def : Pat<(f64 (bitconvert (v2i32 DPR:$src))), (f64 DPR:$src)>;
5371 def : Pat<(f64 (bitconvert (v4i16 DPR:$src))), (f64 DPR:$src)>;
5372 def : Pat<(f64 (bitconvert (v8i8 DPR:$src))), (f64 DPR:$src)>;
5373 def : Pat<(f64 (bitconvert (v2f32 DPR:$src))), (f64 DPR:$src)>;
5374 def : Pat<(v2f32 (bitconvert (f64 DPR:$src))), (v2f32 DPR:$src)>;
5375 def : Pat<(v2f32 (bitconvert (v1i64 DPR:$src))), (v2f32 DPR:$src)>;
5376 def : Pat<(v2f32 (bitconvert (v2i32 DPR:$src))), (v2f32 DPR:$src)>;
5377 def : Pat<(v2f32 (bitconvert (v4i16 DPR:$src))), (v2f32 DPR:$src)>;
5378 def : Pat<(v2f32 (bitconvert (v8i8 DPR:$src))), (v2f32 DPR:$src)>;
5380 def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (v2i64 QPR:$src)>;
5381 def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (v2i64 QPR:$src)>;
5382 def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (v2i64 QPR:$src)>;
5383 def : Pat<(v2i64 (bitconvert (v2f64 QPR:$src))), (v2i64 QPR:$src)>;
5384 def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (v2i64 QPR:$src)>;
5385 def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (v4i32 QPR:$src)>;
5386 def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (v4i32 QPR:$src)>;
5387 def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (v4i32 QPR:$src)>;
5388 def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (v4i32 QPR:$src)>;
5389 def : Pat<(v4i32 (bitconvert (v4f32 QPR:$src))), (v4i32 QPR:$src)>;
5390 def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (v8i16 QPR:$src)>;
5391 def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (v8i16 QPR:$src)>;
5392 def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (v8i16 QPR:$src)>;
5393 def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (v8i16 QPR:$src)>;
5394 def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (v8i16 QPR:$src)>;
5395 def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (v16i8 QPR:$src)>;
5396 def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (v16i8 QPR:$src)>;
5397 def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (v16i8 QPR:$src)>;
5398 def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (v16i8 QPR:$src)>;
5399 def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (v16i8 QPR:$src)>;
5400 def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (v4f32 QPR:$src)>;
5401 def : Pat<(v4f32 (bitconvert (v4i32 QPR:$src))), (v4f32 QPR:$src)>;
5402 def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (v4f32 QPR:$src)>;
5403 def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (v4f32 QPR:$src)>;
5404 def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (v4f32 QPR:$src)>;
5405 def : Pat<(v2f64 (bitconvert (v2i64 QPR:$src))), (v2f64 QPR:$src)>;
5406 def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (v2f64 QPR:$src)>;
5407 def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 QPR:$src)>;
5408 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 QPR:$src)>;
5409 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 QPR:$src)>;
5412 //===----------------------------------------------------------------------===//
5413 // Assembler aliases
5416 def : VFP2InstAlias<"fmdhr${p} $Dd, $Rn",
5417 (VSETLNi32 DPR:$Dd, GPR:$Rn, 1, pred:$p)>;
5418 def : VFP2InstAlias<"fmdlr${p} $Dd, $Rn",
5419 (VSETLNi32 DPR:$Dd, GPR:$Rn, 0, pred:$p)>;
5422 // VADD two-operand aliases.
5423 def : NEONInstAlias<"vadd${p}.i8 $Vdn, $Vm",
5424 (VADDv16i8 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5425 def : NEONInstAlias<"vadd${p}.i16 $Vdn, $Vm",
5426 (VADDv8i16 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5427 def : NEONInstAlias<"vadd${p}.i32 $Vdn, $Vm",
5428 (VADDv4i32 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5429 def : NEONInstAlias<"vadd${p}.i64 $Vdn, $Vm",
5430 (VADDv2i64 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5432 def : NEONInstAlias<"vadd${p}.i8 $Vdn, $Vm",
5433 (VADDv8i8 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5434 def : NEONInstAlias<"vadd${p}.i16 $Vdn, $Vm",
5435 (VADDv4i16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5436 def : NEONInstAlias<"vadd${p}.i32 $Vdn, $Vm",
5437 (VADDv2i32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5438 def : NEONInstAlias<"vadd${p}.i64 $Vdn, $Vm",
5439 (VADDv1i64 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5441 def : NEONInstAlias<"vadd${p}.f32 $Vdn, $Vm",
5442 (VADDfd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5443 def : NEONInstAlias<"vadd${p}.f32 $Vdn, $Vm",
5444 (VADDfq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5446 // VSUB two-operand aliases.
5447 def : NEONInstAlias<"vsub${p}.i8 $Vdn, $Vm",
5448 (VSUBv16i8 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5449 def : NEONInstAlias<"vsub${p}.i16 $Vdn, $Vm",
5450 (VSUBv8i16 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5451 def : NEONInstAlias<"vsub${p}.i32 $Vdn, $Vm",
5452 (VSUBv4i32 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5453 def : NEONInstAlias<"vsub${p}.i64 $Vdn, $Vm",
5454 (VSUBv2i64 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5456 def : NEONInstAlias<"vsub${p}.i8 $Vdn, $Vm",
5457 (VSUBv8i8 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5458 def : NEONInstAlias<"vsub${p}.i16 $Vdn, $Vm",
5459 (VSUBv4i16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5460 def : NEONInstAlias<"vsub${p}.i32 $Vdn, $Vm",
5461 (VSUBv2i32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5462 def : NEONInstAlias<"vsub${p}.i64 $Vdn, $Vm",
5463 (VSUBv1i64 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5465 def : NEONInstAlias<"vsub${p}.f32 $Vdn, $Vm",
5466 (VSUBfd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5467 def : NEONInstAlias<"vsub${p}.f32 $Vdn, $Vm",
5468 (VSUBfq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5470 // VADDW two-operand aliases.
5471 def : NEONInstAlias<"vaddw${p}.s8 $Vdn, $Vm",
5472 (VADDWsv8i16 QPR:$Vdn, QPR:$Vdn, DPR:$Vm, pred:$p)>;
5473 def : NEONInstAlias<"vaddw${p}.s16 $Vdn, $Vm",
5474 (VADDWsv4i32 QPR:$Vdn, QPR:$Vdn, DPR:$Vm, pred:$p)>;
5475 def : NEONInstAlias<"vaddw${p}.s32 $Vdn, $Vm",
5476 (VADDWsv2i64 QPR:$Vdn, QPR:$Vdn, DPR:$Vm, pred:$p)>;
5477 def : NEONInstAlias<"vaddw${p}.u8 $Vdn, $Vm",
5478 (VADDWuv8i16 QPR:$Vdn, QPR:$Vdn, DPR:$Vm, pred:$p)>;
5479 def : NEONInstAlias<"vaddw${p}.u16 $Vdn, $Vm",
5480 (VADDWuv4i32 QPR:$Vdn, QPR:$Vdn, DPR:$Vm, pred:$p)>;
5481 def : NEONInstAlias<"vaddw${p}.u32 $Vdn, $Vm",
5482 (VADDWuv2i64 QPR:$Vdn, QPR:$Vdn, DPR:$Vm, pred:$p)>;
5484 // VAND/VBIC/VEOR/VORR accept but do not require a type suffix.
5485 defm : VFPDTAnyInstAlias<"vand${p}", "$Vd, $Vn, $Vm",
5486 (VANDd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
5487 defm : VFPDTAnyInstAlias<"vand${p}", "$Vd, $Vn, $Vm",
5488 (VANDq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
5489 defm : VFPDTAnyInstAlias<"vbic${p}", "$Vd, $Vn, $Vm",
5490 (VBICd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
5491 defm : VFPDTAnyInstAlias<"vbic${p}", "$Vd, $Vn, $Vm",
5492 (VBICq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
5493 defm : VFPDTAnyInstAlias<"veor${p}", "$Vd, $Vn, $Vm",
5494 (VEORd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
5495 defm : VFPDTAnyInstAlias<"veor${p}", "$Vd, $Vn, $Vm",
5496 (VEORq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
5497 defm : VFPDTAnyInstAlias<"vorr${p}", "$Vd, $Vn, $Vm",
5498 (VORRd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
5499 defm : VFPDTAnyInstAlias<"vorr${p}", "$Vd, $Vn, $Vm",
5500 (VORRq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
5501 // ... two-operand aliases
5502 def : NEONInstAlias<"vand${p} $Vdn, $Vm",
5503 (VANDd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5504 def : NEONInstAlias<"vand${p} $Vdn, $Vm",
5505 (VANDq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5506 def : NEONInstAlias<"vbic${p} $Vdn, $Vm",
5507 (VBICd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5508 def : NEONInstAlias<"vbic${p} $Vdn, $Vm",
5509 (VBICq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5510 def : NEONInstAlias<"veor${p} $Vdn, $Vm",
5511 (VEORd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5512 def : NEONInstAlias<"veor${p} $Vdn, $Vm",
5513 (VEORq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5514 def : NEONInstAlias<"vorr${p} $Vdn, $Vm",
5515 (VORRd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5516 def : NEONInstAlias<"vorr${p} $Vdn, $Vm",
5517 (VORRq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5519 defm : VFPDTAnyInstAlias<"vand${p}", "$Vdn, $Vm",
5520 (VANDd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5521 defm : VFPDTAnyInstAlias<"vand${p}", "$Vdn, $Vm",
5522 (VANDq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5523 defm : VFPDTAnyInstAlias<"veor${p}", "$Vdn, $Vm",
5524 (VEORd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5525 defm : VFPDTAnyInstAlias<"veor${p}", "$Vdn, $Vm",
5526 (VEORq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5527 defm : VFPDTAnyInstAlias<"vorr${p}", "$Vdn, $Vm",
5528 (VORRd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5529 defm : VFPDTAnyInstAlias<"vorr${p}", "$Vdn, $Vm",
5530 (VORRq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5532 // VMUL two-operand aliases.
5533 def : NEONInstAlias<"vmul${p}.p8 $Qdn, $Qm",
5534 (VMULpq QPR:$Qdn, QPR:$Qdn, QPR:$Qm, pred:$p)>;
5535 def : NEONInstAlias<"vmul${p}.i8 $Qdn, $Qm",
5536 (VMULv16i8 QPR:$Qdn, QPR:$Qdn, QPR:$Qm, pred:$p)>;
5537 def : NEONInstAlias<"vmul${p}.i16 $Qdn, $Qm",
5538 (VMULv8i16 QPR:$Qdn, QPR:$Qdn, QPR:$Qm, pred:$p)>;
5539 def : NEONInstAlias<"vmul${p}.i32 $Qdn, $Qm",
5540 (VMULv4i32 QPR:$Qdn, QPR:$Qdn, QPR:$Qm, pred:$p)>;
5542 def : NEONInstAlias<"vmul${p}.p8 $Ddn, $Dm",
5543 (VMULpd DPR:$Ddn, DPR:$Ddn, DPR:$Dm, pred:$p)>;
5544 def : NEONInstAlias<"vmul${p}.i8 $Ddn, $Dm",
5545 (VMULv8i8 DPR:$Ddn, DPR:$Ddn, DPR:$Dm, pred:$p)>;
5546 def : NEONInstAlias<"vmul${p}.i16 $Ddn, $Dm",
5547 (VMULv4i16 DPR:$Ddn, DPR:$Ddn, DPR:$Dm, pred:$p)>;
5548 def : NEONInstAlias<"vmul${p}.i32 $Ddn, $Dm",
5549 (VMULv2i32 DPR:$Ddn, DPR:$Ddn, DPR:$Dm, pred:$p)>;
5551 def : NEONInstAlias<"vmul${p}.f32 $Qdn, $Qm",
5552 (VMULfq QPR:$Qdn, QPR:$Qdn, QPR:$Qm, pred:$p)>;
5553 def : NEONInstAlias<"vmul${p}.f32 $Ddn, $Dm",
5554 (VMULfd DPR:$Ddn, DPR:$Ddn, DPR:$Dm, pred:$p)>;
5556 def : NEONInstAlias<"vmul${p}.i16 $Ddn, $Dm$lane",
5557 (VMULslv4i16 DPR:$Ddn, DPR:$Ddn, DPR_8:$Dm,
5558 VectorIndex16:$lane, pred:$p)>;
5559 def : NEONInstAlias<"vmul${p}.i16 $Qdn, $Dm$lane",
5560 (VMULslv8i16 QPR:$Qdn, QPR:$Qdn, DPR_8:$Dm,
5561 VectorIndex16:$lane, pred:$p)>;
5563 def : NEONInstAlias<"vmul${p}.i32 $Ddn, $Dm$lane",
5564 (VMULslv2i32 DPR:$Ddn, DPR:$Ddn, DPR_VFP2:$Dm,
5565 VectorIndex32:$lane, pred:$p)>;
5566 def : NEONInstAlias<"vmul${p}.i32 $Qdn, $Dm$lane",
5567 (VMULslv4i32 QPR:$Qdn, QPR:$Qdn, DPR_VFP2:$Dm,
5568 VectorIndex32:$lane, pred:$p)>;
5570 def : NEONInstAlias<"vmul${p}.f32 $Ddn, $Dm$lane",
5571 (VMULslfd DPR:$Ddn, DPR:$Ddn, DPR_VFP2:$Dm,
5572 VectorIndex32:$lane, pred:$p)>;
5573 def : NEONInstAlias<"vmul${p}.f32 $Qdn, $Dm$lane",
5574 (VMULslfq QPR:$Qdn, QPR:$Qdn, DPR_VFP2:$Dm,
5575 VectorIndex32:$lane, pred:$p)>;
5577 // VQADD (register) two-operand aliases.
5578 def : NEONInstAlias<"vqadd${p}.s8 $Vdn, $Vm",
5579 (VQADDsv8i8 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5580 def : NEONInstAlias<"vqadd${p}.s16 $Vdn, $Vm",
5581 (VQADDsv4i16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5582 def : NEONInstAlias<"vqadd${p}.s32 $Vdn, $Vm",
5583 (VQADDsv2i32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5584 def : NEONInstAlias<"vqadd${p}.s64 $Vdn, $Vm",
5585 (VQADDsv1i64 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5586 def : NEONInstAlias<"vqadd${p}.u8 $Vdn, $Vm",
5587 (VQADDuv8i8 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5588 def : NEONInstAlias<"vqadd${p}.u16 $Vdn, $Vm",
5589 (VQADDuv4i16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5590 def : NEONInstAlias<"vqadd${p}.u32 $Vdn, $Vm",
5591 (VQADDuv2i32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5592 def : NEONInstAlias<"vqadd${p}.u64 $Vdn, $Vm",
5593 (VQADDuv1i64 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5595 def : NEONInstAlias<"vqadd${p}.s8 $Vdn, $Vm",
5596 (VQADDsv16i8 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5597 def : NEONInstAlias<"vqadd${p}.s16 $Vdn, $Vm",
5598 (VQADDsv8i16 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5599 def : NEONInstAlias<"vqadd${p}.s32 $Vdn, $Vm",
5600 (VQADDsv4i32 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5601 def : NEONInstAlias<"vqadd${p}.s64 $Vdn, $Vm",
5602 (VQADDsv2i64 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5603 def : NEONInstAlias<"vqadd${p}.u8 $Vdn, $Vm",
5604 (VQADDuv16i8 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5605 def : NEONInstAlias<"vqadd${p}.u16 $Vdn, $Vm",
5606 (VQADDuv8i16 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5607 def : NEONInstAlias<"vqadd${p}.u32 $Vdn, $Vm",
5608 (VQADDuv4i32 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5609 def : NEONInstAlias<"vqadd${p}.u64 $Vdn, $Vm",
5610 (VQADDuv2i64 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5612 // VSHL (immediate) two-operand aliases.
5613 def : NEONInstAlias<"vshl${p}.i8 $Vdn, $imm",
5614 (VSHLiv8i8 DPR:$Vdn, DPR:$Vdn, imm0_7:$imm, pred:$p)>;
5615 def : NEONInstAlias<"vshl${p}.i16 $Vdn, $imm",
5616 (VSHLiv4i16 DPR:$Vdn, DPR:$Vdn, imm0_15:$imm, pred:$p)>;
5617 def : NEONInstAlias<"vshl${p}.i32 $Vdn, $imm",
5618 (VSHLiv2i32 DPR:$Vdn, DPR:$Vdn, imm0_31:$imm, pred:$p)>;
5619 def : NEONInstAlias<"vshl${p}.i64 $Vdn, $imm",
5620 (VSHLiv1i64 DPR:$Vdn, DPR:$Vdn, imm0_63:$imm, pred:$p)>;
5622 def : NEONInstAlias<"vshl${p}.i8 $Vdn, $imm",
5623 (VSHLiv16i8 QPR:$Vdn, QPR:$Vdn, imm0_7:$imm, pred:$p)>;
5624 def : NEONInstAlias<"vshl${p}.i16 $Vdn, $imm",
5625 (VSHLiv8i16 QPR:$Vdn, QPR:$Vdn, imm0_15:$imm, pred:$p)>;
5626 def : NEONInstAlias<"vshl${p}.i32 $Vdn, $imm",
5627 (VSHLiv4i32 QPR:$Vdn, QPR:$Vdn, imm0_31:$imm, pred:$p)>;
5628 def : NEONInstAlias<"vshl${p}.i64 $Vdn, $imm",
5629 (VSHLiv2i64 QPR:$Vdn, QPR:$Vdn, imm0_63:$imm, pred:$p)>;
5631 // VSHL (register) two-operand aliases.
5632 def : NEONInstAlias<"vshl${p}.s8 $Vdn, $Vm",
5633 (VSHLsv8i8 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5634 def : NEONInstAlias<"vshl${p}.s16 $Vdn, $Vm",
5635 (VSHLsv4i16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5636 def : NEONInstAlias<"vshl${p}.s32 $Vdn, $Vm",
5637 (VSHLsv2i32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5638 def : NEONInstAlias<"vshl${p}.s64 $Vdn, $Vm",
5639 (VSHLsv1i64 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5640 def : NEONInstAlias<"vshl${p}.u8 $Vdn, $Vm",
5641 (VSHLuv8i8 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5642 def : NEONInstAlias<"vshl${p}.u16 $Vdn, $Vm",
5643 (VSHLuv4i16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5644 def : NEONInstAlias<"vshl${p}.u32 $Vdn, $Vm",
5645 (VSHLuv2i32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5646 def : NEONInstAlias<"vshl${p}.u64 $Vdn, $Vm",
5647 (VSHLuv1i64 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5649 def : NEONInstAlias<"vshl${p}.s8 $Vdn, $Vm",
5650 (VSHLsv16i8 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5651 def : NEONInstAlias<"vshl${p}.s16 $Vdn, $Vm",
5652 (VSHLsv8i16 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5653 def : NEONInstAlias<"vshl${p}.s32 $Vdn, $Vm",
5654 (VSHLsv4i32 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5655 def : NEONInstAlias<"vshl${p}.s64 $Vdn, $Vm",
5656 (VSHLsv2i64 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5657 def : NEONInstAlias<"vshl${p}.u8 $Vdn, $Vm",
5658 (VSHLuv16i8 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5659 def : NEONInstAlias<"vshl${p}.u16 $Vdn, $Vm",
5660 (VSHLuv8i16 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5661 def : NEONInstAlias<"vshl${p}.u32 $Vdn, $Vm",
5662 (VSHLuv4i32 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5663 def : NEONInstAlias<"vshl${p}.u64 $Vdn, $Vm",
5664 (VSHLuv2i64 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5666 // VSHL (immediate) two-operand aliases.
5667 def : NEONInstAlias<"vshr${p}.s8 $Vdn, $imm",
5668 (VSHRsv8i8 DPR:$Vdn, DPR:$Vdn, shr_imm8:$imm, pred:$p)>;
5669 def : NEONInstAlias<"vshr${p}.s16 $Vdn, $imm",
5670 (VSHRsv4i16 DPR:$Vdn, DPR:$Vdn, shr_imm16:$imm, pred:$p)>;
5671 def : NEONInstAlias<"vshr${p}.s32 $Vdn, $imm",
5672 (VSHRsv2i32 DPR:$Vdn, DPR:$Vdn, shr_imm32:$imm, pred:$p)>;
5673 def : NEONInstAlias<"vshr${p}.s64 $Vdn, $imm",
5674 (VSHRsv1i64 DPR:$Vdn, DPR:$Vdn, shr_imm64:$imm, pred:$p)>;
5676 def : NEONInstAlias<"vshr${p}.s8 $Vdn, $imm",
5677 (VSHRsv16i8 QPR:$Vdn, QPR:$Vdn, shr_imm8:$imm, pred:$p)>;
5678 def : NEONInstAlias<"vshr${p}.s16 $Vdn, $imm",
5679 (VSHRsv8i16 QPR:$Vdn, QPR:$Vdn, shr_imm16:$imm, pred:$p)>;
5680 def : NEONInstAlias<"vshr${p}.s32 $Vdn, $imm",
5681 (VSHRsv4i32 QPR:$Vdn, QPR:$Vdn, shr_imm32:$imm, pred:$p)>;
5682 def : NEONInstAlias<"vshr${p}.s64 $Vdn, $imm",
5683 (VSHRsv2i64 QPR:$Vdn, QPR:$Vdn, shr_imm64:$imm, pred:$p)>;
5685 def : NEONInstAlias<"vshr${p}.u8 $Vdn, $imm",
5686 (VSHRuv8i8 DPR:$Vdn, DPR:$Vdn, shr_imm8:$imm, pred:$p)>;
5687 def : NEONInstAlias<"vshr${p}.u16 $Vdn, $imm",
5688 (VSHRuv4i16 DPR:$Vdn, DPR:$Vdn, shr_imm16:$imm, pred:$p)>;
5689 def : NEONInstAlias<"vshr${p}.u32 $Vdn, $imm",
5690 (VSHRuv2i32 DPR:$Vdn, DPR:$Vdn, shr_imm32:$imm, pred:$p)>;
5691 def : NEONInstAlias<"vshr${p}.u64 $Vdn, $imm",
5692 (VSHRuv1i64 DPR:$Vdn, DPR:$Vdn, shr_imm64:$imm, pred:$p)>;
5694 def : NEONInstAlias<"vshr${p}.u8 $Vdn, $imm",
5695 (VSHRuv16i8 QPR:$Vdn, QPR:$Vdn, shr_imm8:$imm, pred:$p)>;
5696 def : NEONInstAlias<"vshr${p}.u16 $Vdn, $imm",
5697 (VSHRuv8i16 QPR:$Vdn, QPR:$Vdn, shr_imm16:$imm, pred:$p)>;
5698 def : NEONInstAlias<"vshr${p}.u32 $Vdn, $imm",
5699 (VSHRuv4i32 QPR:$Vdn, QPR:$Vdn, shr_imm32:$imm, pred:$p)>;
5700 def : NEONInstAlias<"vshr${p}.u64 $Vdn, $imm",
5701 (VSHRuv2i64 QPR:$Vdn, QPR:$Vdn, shr_imm64:$imm, pred:$p)>;
5703 // VLD1 single-lane pseudo-instructions. These need special handling for
5704 // the lane index that an InstAlias can't handle, so we use these instead.
5705 defm VLD1LNdAsm : NEONDT8AsmPseudoInst<"vld1${p}", "$list, $addr",
5706 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5707 defm VLD1LNdAsm : NEONDT16AsmPseudoInst<"vld1${p}", "$list, $addr",
5708 (ins VecListOneDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5709 defm VLD1LNdAsm : NEONDT32AsmPseudoInst<"vld1${p}", "$list, $addr",
5710 (ins VecListOneDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5712 defm VLD1LNdWB_fixed_Asm : NEONDT8AsmPseudoInst<"vld1${p}", "$list, $addr!",
5713 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5714 defm VLD1LNdWB_fixed_Asm : NEONDT16AsmPseudoInst<"vld1${p}", "$list, $addr!",
5715 (ins VecListOneDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5716 defm VLD1LNdWB_fixed_Asm : NEONDT32AsmPseudoInst<"vld1${p}", "$list, $addr!",
5717 (ins VecListOneDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5718 defm VLD1LNdWB_register_Asm :
5719 NEONDT8AsmPseudoInst<"vld1${p}", "$list, $addr, $Rm",
5720 (ins VecListOneDByteIndexed:$list, addrmode6:$addr,
5721 rGPR:$Rm, pred:$p)>;
5722 defm VLD1LNdWB_register_Asm :
5723 NEONDT16AsmPseudoInst<"vld1${p}", "$list, $addr, $Rm",
5724 (ins VecListOneDHWordIndexed:$list, addrmode6:$addr,
5725 rGPR:$Rm, pred:$p)>;
5726 defm VLD1LNdWB_register_Asm :
5727 NEONDT32AsmPseudoInst<"vld1${p}", "$list, $addr, $Rm",
5728 (ins VecListOneDWordIndexed:$list, addrmode6:$addr,
5729 rGPR:$Rm, pred:$p)>;
5732 // VST1 single-lane pseudo-instructions. These need special handling for
5733 // the lane index that an InstAlias can't handle, so we use these instead.
5734 defm VST1LNdAsm : NEONDT8AsmPseudoInst<"vst1${p}", "$list, $addr",
5735 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5736 defm VST1LNdAsm : NEONDT16AsmPseudoInst<"vst1${p}", "$list, $addr",
5737 (ins VecListOneDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5738 defm VST1LNdAsm : NEONDT32AsmPseudoInst<"vst1${p}", "$list, $addr",
5739 (ins VecListOneDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5741 defm VST1LNdWB_fixed_Asm : NEONDT8AsmPseudoInst<"vst1${p}", "$list, $addr!",
5742 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5743 defm VST1LNdWB_fixed_Asm : NEONDT16AsmPseudoInst<"vst1${p}", "$list, $addr!",
5744 (ins VecListOneDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5745 defm VST1LNdWB_fixed_Asm : NEONDT32AsmPseudoInst<"vst1${p}", "$list, $addr!",
5746 (ins VecListOneDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5747 defm VST1LNdWB_register_Asm :
5748 NEONDT8AsmPseudoInst<"vst1${p}", "$list, $addr, $Rm",
5749 (ins VecListOneDByteIndexed:$list, addrmode6:$addr,
5750 rGPR:$Rm, pred:$p)>;
5751 defm VST1LNdWB_register_Asm :
5752 NEONDT16AsmPseudoInst<"vst1${p}", "$list, $addr, $Rm",
5753 (ins VecListOneDHWordIndexed:$list, addrmode6:$addr,
5754 rGPR:$Rm, pred:$p)>;
5755 defm VST1LNdWB_register_Asm :
5756 NEONDT32AsmPseudoInst<"vst1${p}", "$list, $addr, $Rm",
5757 (ins VecListOneDWordIndexed:$list, addrmode6:$addr,
5758 rGPR:$Rm, pred:$p)>;
5760 // VLD2 single-lane pseudo-instructions. These need special handling for
5761 // the lane index that an InstAlias can't handle, so we use these instead.
5762 defm VLD2LNdAsm : NEONDT8AsmPseudoInst<"vld2${p}", "$list, $addr",
5763 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5764 defm VLD2LNdAsm : NEONDT16AsmPseudoInst<"vld2${p}", "$list, $addr",
5765 (ins VecListTwoDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5766 defm VLD2LNdAsm : NEONDT32AsmPseudoInst<"vld2${p}", "$list, $addr",
5767 (ins VecListTwoDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5768 defm VLD2LNqAsm : NEONDT16AsmPseudoInst<"vld2${p}", "$list, $addr",
5769 (ins VecListTwoQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5770 defm VLD2LNqAsm : NEONDT32AsmPseudoInst<"vld2${p}", "$list, $addr",
5771 (ins VecListTwoQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5773 defm VLD2LNdWB_fixed_Asm : NEONDT8AsmPseudoInst<"vld2${p}", "$list, $addr!",
5774 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5775 defm VLD2LNdWB_fixed_Asm : NEONDT16AsmPseudoInst<"vld2${p}", "$list, $addr!",
5776 (ins VecListTwoDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5777 defm VLD2LNdWB_fixed_Asm : NEONDT32AsmPseudoInst<"vld2${p}", "$list, $addr!",
5778 (ins VecListTwoDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5779 defm VLD2LNqWB_fixed_Asm : NEONDT16AsmPseudoInst<"vld2${p}", "$list, $addr!",
5780 (ins VecListTwoQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5781 defm VLD2LNqWB_fixed_Asm : NEONDT32AsmPseudoInst<"vld2${p}", "$list, $addr!",
5782 (ins VecListTwoQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5783 defm VLD2LNdWB_register_Asm :
5784 NEONDT8AsmPseudoInst<"vld2${p}", "$list, $addr, $Rm",
5785 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr,
5786 rGPR:$Rm, pred:$p)>;
5787 defm VLD2LNdWB_register_Asm :
5788 NEONDT16AsmPseudoInst<"vld2${p}", "$list, $addr, $Rm",
5789 (ins VecListTwoDHWordIndexed:$list, addrmode6:$addr,
5790 rGPR:$Rm, pred:$p)>;
5791 defm VLD2LNdWB_register_Asm :
5792 NEONDT32AsmPseudoInst<"vld2${p}", "$list, $addr, $Rm",
5793 (ins VecListTwoDWordIndexed:$list, addrmode6:$addr,
5794 rGPR:$Rm, pred:$p)>;
5795 defm VLD2LNqWB_register_Asm :
5796 NEONDT16AsmPseudoInst<"vld2${p}", "$list, $addr, $Rm",
5797 (ins VecListTwoQHWordIndexed:$list, addrmode6:$addr,
5798 rGPR:$Rm, pred:$p)>;
5799 defm VLD2LNqWB_register_Asm :
5800 NEONDT32AsmPseudoInst<"vld2${p}", "$list, $addr, $Rm",
5801 (ins VecListTwoQWordIndexed:$list, addrmode6:$addr,
5802 rGPR:$Rm, pred:$p)>;
5805 // VST2 single-lane pseudo-instructions. These need special handling for
5806 // the lane index that an InstAlias can't handle, so we use these instead.
5807 defm VST2LNdAsm : NEONDT8AsmPseudoInst<"vst2${p}", "$list, $addr",
5808 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5809 defm VST2LNdAsm : NEONDT16AsmPseudoInst<"vst2${p}", "$list, $addr",
5810 (ins VecListTwoDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5811 defm VST2LNdAsm : NEONDT32AsmPseudoInst<"vst2${p}", "$list, $addr",
5812 (ins VecListTwoDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5813 defm VST2LNqAsm : NEONDT16AsmPseudoInst<"vst2${p}", "$list, $addr",
5814 (ins VecListTwoQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5815 defm VST2LNqAsm : NEONDT32AsmPseudoInst<"vst2${p}", "$list, $addr",
5816 (ins VecListTwoQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5818 defm VST2LNdWB_fixed_Asm : NEONDT8AsmPseudoInst<"vst2${p}", "$list, $addr!",
5819 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5820 defm VST2LNdWB_fixed_Asm : NEONDT16AsmPseudoInst<"vst2${p}", "$list, $addr!",
5821 (ins VecListTwoDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5822 defm VST2LNdWB_fixed_Asm : NEONDT32AsmPseudoInst<"vst2${p}", "$list, $addr!",
5823 (ins VecListTwoDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5824 defm VST2LNqWB_fixed_Asm : NEONDT16AsmPseudoInst<"vst2${p}", "$list, $addr!",
5825 (ins VecListTwoQHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5826 defm VST2LNqWB_fixed_Asm : NEONDT32AsmPseudoInst<"vst2${p}", "$list, $addr!",
5827 (ins VecListTwoQWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5828 defm VST2LNdWB_register_Asm :
5829 NEONDT8AsmPseudoInst<"vst2${p}", "$list, $addr, $Rm",
5830 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr,
5831 rGPR:$Rm, pred:$p)>;
5832 defm VST2LNdWB_register_Asm :
5833 NEONDT16AsmPseudoInst<"vst2${p}", "$list, $addr, $Rm",
5834 (ins VecListTwoDHWordIndexed:$list, addrmode6:$addr,
5835 rGPR:$Rm, pred:$p)>;
5836 defm VST2LNdWB_register_Asm :
5837 NEONDT32AsmPseudoInst<"vst2${p}", "$list, $addr, $Rm",
5838 (ins VecListTwoDWordIndexed:$list, addrmode6:$addr,
5839 rGPR:$Rm, pred:$p)>;
5840 defm VST2LNqWB_register_Asm :
5841 NEONDT16AsmPseudoInst<"vst2${p}", "$list, $addr, $Rm",
5842 (ins VecListTwoQHWordIndexed:$list, addrmode6:$addr,
5843 rGPR:$Rm, pred:$p)>;
5844 defm VST2LNqWB_register_Asm :
5845 NEONDT32AsmPseudoInst<"vst2${p}", "$list, $addr, $Rm",
5846 (ins VecListTwoQWordIndexed:$list, addrmode6:$addr,
5847 rGPR:$Rm, pred:$p)>;
5849 // VMOV takes an optional datatype suffix
5850 defm : VFPDTAnyInstAlias<"vmov${p}", "$Vd, $Vm",
5851 (VORRd DPR:$Vd, DPR:$Vm, DPR:$Vm, pred:$p)>;
5852 defm : VFPDTAnyInstAlias<"vmov${p}", "$Vd, $Vm",
5853 (VORRq QPR:$Vd, QPR:$Vm, QPR:$Vm, pred:$p)>;
5855 // VCLT (register) is an assembler alias for VCGT w/ the operands reversed.
5856 // D-register versions.
5857 def : NEONInstAlias<"vcle${p}.s8 $Dd, $Dn, $Dm",
5858 (VCGEsv8i8 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
5859 def : NEONInstAlias<"vcle${p}.s16 $Dd, $Dn, $Dm",
5860 (VCGEsv4i16 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
5861 def : NEONInstAlias<"vcle${p}.s32 $Dd, $Dn, $Dm",
5862 (VCGEsv2i32 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
5863 def : NEONInstAlias<"vcle${p}.u8 $Dd, $Dn, $Dm",
5864 (VCGEuv8i8 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
5865 def : NEONInstAlias<"vcle${p}.u16 $Dd, $Dn, $Dm",
5866 (VCGEuv4i16 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
5867 def : NEONInstAlias<"vcle${p}.u32 $Dd, $Dn, $Dm",
5868 (VCGEuv2i32 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
5869 def : NEONInstAlias<"vcle${p}.f32 $Dd, $Dn, $Dm",
5870 (VCGEfd DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
5871 // Q-register versions.
5872 def : NEONInstAlias<"vcle${p}.s8 $Qd, $Qn, $Qm",
5873 (VCGEsv16i8 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
5874 def : NEONInstAlias<"vcle${p}.s16 $Qd, $Qn, $Qm",
5875 (VCGEsv8i16 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
5876 def : NEONInstAlias<"vcle${p}.s32 $Qd, $Qn, $Qm",
5877 (VCGEsv4i32 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
5878 def : NEONInstAlias<"vcle${p}.u8 $Qd, $Qn, $Qm",
5879 (VCGEuv16i8 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
5880 def : NEONInstAlias<"vcle${p}.u16 $Qd, $Qn, $Qm",
5881 (VCGEuv8i16 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
5882 def : NEONInstAlias<"vcle${p}.u32 $Qd, $Qn, $Qm",
5883 (VCGEuv4i32 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
5884 def : NEONInstAlias<"vcle${p}.f32 $Qd, $Qn, $Qm",
5885 (VCGEfq QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
5887 // VCLT (register) is an assembler alias for VCGT w/ the operands reversed.
5888 // D-register versions.
5889 def : NEONInstAlias<"vclt${p}.s8 $Dd, $Dn, $Dm",
5890 (VCGTsv8i8 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
5891 def : NEONInstAlias<"vclt${p}.s16 $Dd, $Dn, $Dm",
5892 (VCGTsv4i16 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
5893 def : NEONInstAlias<"vclt${p}.s32 $Dd, $Dn, $Dm",
5894 (VCGTsv2i32 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
5895 def : NEONInstAlias<"vclt${p}.u8 $Dd, $Dn, $Dm",
5896 (VCGTuv8i8 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
5897 def : NEONInstAlias<"vclt${p}.u16 $Dd, $Dn, $Dm",
5898 (VCGTuv4i16 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
5899 def : NEONInstAlias<"vclt${p}.u32 $Dd, $Dn, $Dm",
5900 (VCGTuv2i32 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
5901 def : NEONInstAlias<"vclt${p}.f32 $Dd, $Dn, $Dm",
5902 (VCGTfd DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
5903 // Q-register versions.
5904 def : NEONInstAlias<"vclt${p}.s8 $Qd, $Qn, $Qm",
5905 (VCGTsv16i8 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
5906 def : NEONInstAlias<"vclt${p}.s16 $Qd, $Qn, $Qm",
5907 (VCGTsv8i16 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
5908 def : NEONInstAlias<"vclt${p}.s32 $Qd, $Qn, $Qm",
5909 (VCGTsv4i32 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
5910 def : NEONInstAlias<"vclt${p}.u8 $Qd, $Qn, $Qm",
5911 (VCGTuv16i8 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
5912 def : NEONInstAlias<"vclt${p}.u16 $Qd, $Qn, $Qm",
5913 (VCGTuv8i16 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
5914 def : NEONInstAlias<"vclt${p}.u32 $Qd, $Qn, $Qm",
5915 (VCGTuv4i32 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
5916 def : NEONInstAlias<"vclt${p}.f32 $Qd, $Qn, $Qm",
5917 (VCGTfq QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
5919 // Two-operand variants for VEXT
5920 def : NEONInstAlias<"vext${p}.8 $Vdn, $Vm, $imm",
5921 (VEXTd8 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, imm0_7:$imm, pred:$p)>;
5922 def : NEONInstAlias<"vext${p}.16 $Vdn, $Vm, $imm",
5923 (VEXTd16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, imm0_3:$imm, pred:$p)>;
5924 def : NEONInstAlias<"vext${p}.32 $Vdn, $Vm, $imm",
5925 (VEXTd32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, imm0_1:$imm, pred:$p)>;
5927 def : NEONInstAlias<"vext${p}.8 $Vdn, $Vm, $imm",
5928 (VEXTq8 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, imm0_15:$imm, pred:$p)>;
5929 def : NEONInstAlias<"vext${p}.16 $Vdn, $Vm, $imm",
5930 (VEXTq16 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, imm0_7:$imm, pred:$p)>;
5931 def : NEONInstAlias<"vext${p}.32 $Vdn, $Vm, $imm",
5932 (VEXTq32 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, imm0_3:$imm, pred:$p)>;
5933 def : NEONInstAlias<"vext${p}.64 $Vdn, $Vm, $imm",
5934 (VEXTq64 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, imm0_1:$imm, pred:$p)>;
5936 // Two-operand variants for VQDMULH
5937 def : NEONInstAlias<"vqdmulh${p}.s16 $Vdn, $Vm",
5938 (VQDMULHv4i16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5939 def : NEONInstAlias<"vqdmulh${p}.s32 $Vdn, $Vm",
5940 (VQDMULHv2i32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5942 def : NEONInstAlias<"vqdmulh${p}.s16 $Vdn, $Vm",
5943 (VQDMULHv8i16 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5944 def : NEONInstAlias<"vqdmulh${p}.s32 $Vdn, $Vm",
5945 (VQDMULHv4i32 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5947 // Two-operand variants for VMAX.
5948 def : NEONInstAlias<"vmax${p}.s8 $Vdn, $Vm",
5949 (VMAXsv8i8 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5950 def : NEONInstAlias<"vmax${p}.s16 $Vdn, $Vm",
5951 (VMAXsv4i16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5952 def : NEONInstAlias<"vmax${p}.s32 $Vdn, $Vm",
5953 (VMAXsv2i32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5954 def : NEONInstAlias<"vmax${p}.u8 $Vdn, $Vm",
5955 (VMAXuv8i8 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5956 def : NEONInstAlias<"vmax${p}.u16 $Vdn, $Vm",
5957 (VMAXuv4i16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5958 def : NEONInstAlias<"vmax${p}.u32 $Vdn, $Vm",
5959 (VMAXuv2i32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5960 def : NEONInstAlias<"vmax${p}.f32 $Vdn, $Vm",
5961 (VMAXfd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5963 def : NEONInstAlias<"vmax${p}.s8 $Vdn, $Vm",
5964 (VMAXsv16i8 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5965 def : NEONInstAlias<"vmax${p}.s16 $Vdn, $Vm",
5966 (VMAXsv8i16 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5967 def : NEONInstAlias<"vmax${p}.s32 $Vdn, $Vm",
5968 (VMAXsv4i32 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5969 def : NEONInstAlias<"vmax${p}.u8 $Vdn, $Vm",
5970 (VMAXuv16i8 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5971 def : NEONInstAlias<"vmax${p}.u16 $Vdn, $Vm",
5972 (VMAXuv8i16 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5973 def : NEONInstAlias<"vmax${p}.u32 $Vdn, $Vm",
5974 (VMAXuv4i32 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5975 def : NEONInstAlias<"vmax${p}.f32 $Vdn, $Vm",
5976 (VMAXfq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5978 // Two-operand variants for VMIN.
5979 def : NEONInstAlias<"vmin${p}.s8 $Vdn, $Vm",
5980 (VMINsv8i8 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5981 def : NEONInstAlias<"vmin${p}.s16 $Vdn, $Vm",
5982 (VMINsv4i16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5983 def : NEONInstAlias<"vmin${p}.s32 $Vdn, $Vm",
5984 (VMINsv2i32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5985 def : NEONInstAlias<"vmin${p}.u8 $Vdn, $Vm",
5986 (VMINuv8i8 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5987 def : NEONInstAlias<"vmin${p}.u16 $Vdn, $Vm",
5988 (VMINuv4i16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5989 def : NEONInstAlias<"vmin${p}.u32 $Vdn, $Vm",
5990 (VMINuv2i32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5991 def : NEONInstAlias<"vmin${p}.f32 $Vdn, $Vm",
5992 (VMINfd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5994 def : NEONInstAlias<"vmin${p}.s8 $Vdn, $Vm",
5995 (VMINsv16i8 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5996 def : NEONInstAlias<"vmin${p}.s16 $Vdn, $Vm",
5997 (VMINsv8i16 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5998 def : NEONInstAlias<"vmin${p}.s32 $Vdn, $Vm",
5999 (VMINsv4i32 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
6000 def : NEONInstAlias<"vmin${p}.u8 $Vdn, $Vm",
6001 (VMINuv16i8 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
6002 def : NEONInstAlias<"vmin${p}.u16 $Vdn, $Vm",
6003 (VMINuv8i16 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
6004 def : NEONInstAlias<"vmin${p}.u32 $Vdn, $Vm",
6005 (VMINuv4i32 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
6006 def : NEONInstAlias<"vmin${p}.f32 $Vdn, $Vm",
6007 (VMINfq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
6009 // Two-operand variants for VPADD.
6010 def : NEONInstAlias<"vpadd${p}.i8 $Vdn, $Vm",
6011 (VPADDi8 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
6012 def : NEONInstAlias<"vpadd${p}.i16 $Vdn, $Vm",
6013 (VPADDi16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
6014 def : NEONInstAlias<"vpadd${p}.i32 $Vdn, $Vm",
6015 (VPADDi32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
6016 def : NEONInstAlias<"vpadd${p}.f32 $Vdn, $Vm",
6017 (VPADDf DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
6019 // "vmov Rd, #-imm" can be handled via "vmvn".
6020 def : NEONInstAlias<"vmov${p}.i32 $Vd, $imm",
6021 (VMVNv2i32 DPR:$Vd, nImmVMOVI32Neg:$imm, pred:$p)>;
6022 def : NEONInstAlias<"vmov${p}.i32 $Vd, $imm",
6023 (VMVNv4i32 QPR:$Vd, nImmVMOVI32Neg:$imm, pred:$p)>;
6024 def : NEONInstAlias<"vmvn${p}.i32 $Vd, $imm",
6025 (VMOVv2i32 DPR:$Vd, nImmVMOVI32Neg:$imm, pred:$p)>;
6026 def : NEONInstAlias<"vmvn${p}.i32 $Vd, $imm",
6027 (VMOVv4i32 QPR:$Vd, nImmVMOVI32Neg:$imm, pred:$p)>;
6029 // 'gas' compatibility aliases for quad-word instructions. Strictly speaking,
6030 // these should restrict to just the Q register variants, but the register
6031 // classes are enough to match correctly regardless, so we keep it simple
6032 // and just use MnemonicAlias.
6033 def : NEONMnemonicAlias<"vbicq", "vbic">;
6034 def : NEONMnemonicAlias<"vandq", "vand">;
6035 def : NEONMnemonicAlias<"veorq", "veor">;
6036 def : NEONMnemonicAlias<"vorrq", "vorr">;
6038 def : NEONMnemonicAlias<"vmovq", "vmov">;
6039 def : NEONMnemonicAlias<"vmvnq", "vmvn">;
6040 // Explicit versions for floating point so that the FPImm variants get
6041 // handled early. The parser gets confused otherwise.
6042 def : NEONMnemonicAlias<"vmovq.f32", "vmov.f32">;
6043 def : NEONMnemonicAlias<"vmovq.f64", "vmov.f64">;
6045 def : NEONMnemonicAlias<"vaddq", "vadd">;
6046 def : NEONMnemonicAlias<"vsubq", "vsub">;
6048 def : NEONMnemonicAlias<"vminq", "vmin">;
6049 def : NEONMnemonicAlias<"vmaxq", "vmax">;
6051 def : NEONMnemonicAlias<"vmulq", "vmul">;
6053 def : NEONMnemonicAlias<"vabsq", "vabs">;
6055 def : NEONMnemonicAlias<"vshlq", "vshl">;
6056 def : NEONMnemonicAlias<"vshrq", "vshr">;
6058 def : NEONMnemonicAlias<"vcvtq", "vcvt">;
6060 def : NEONMnemonicAlias<"vcleq", "vcle">;
6061 def : NEONMnemonicAlias<"vceqq", "vceq">;