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;
147 // Register list of one D register, with byte lane subscripting.
148 def VecListOneDByteIndexAsmOperand : AsmOperandClass {
149 let Name = "VecListOneDByteIndexed";
150 let ParserMethod = "parseVectorList";
151 let RenderMethod = "addVecListIndexedOperands";
153 def VecListOneDByteIndexed : Operand<i32> {
154 let ParserMatchClass = VecListOneDByteIndexAsmOperand;
155 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
157 // ...with half-word lane subscripting.
158 def VecListOneDHWordIndexAsmOperand : AsmOperandClass {
159 let Name = "VecListOneDHWordIndexed";
160 let ParserMethod = "parseVectorList";
161 let RenderMethod = "addVecListIndexedOperands";
163 def VecListOneDHWordIndexed : Operand<i32> {
164 let ParserMatchClass = VecListOneDHWordIndexAsmOperand;
165 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
167 // ...with word lane subscripting.
168 def VecListOneDWordIndexAsmOperand : AsmOperandClass {
169 let Name = "VecListOneDWordIndexed";
170 let ParserMethod = "parseVectorList";
171 let RenderMethod = "addVecListIndexedOperands";
173 def VecListOneDWordIndexed : Operand<i32> {
174 let ParserMatchClass = VecListOneDWordIndexAsmOperand;
175 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
177 // Register list of two D registers, with byte lane subscripting.
178 def VecListTwoDByteIndexAsmOperand : AsmOperandClass {
179 let Name = "VecListTwoDByteIndexed";
180 let ParserMethod = "parseVectorList";
181 let RenderMethod = "addVecListIndexedOperands";
183 def VecListTwoDByteIndexed : Operand<i32> {
184 let ParserMatchClass = VecListTwoDByteIndexAsmOperand;
185 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
187 // ...with half-word lane subscripting.
188 def VecListTwoDHWordIndexAsmOperand : AsmOperandClass {
189 let Name = "VecListTwoDHWordIndexed";
190 let ParserMethod = "parseVectorList";
191 let RenderMethod = "addVecListIndexedOperands";
193 def VecListTwoDHWordIndexed : Operand<i32> {
194 let ParserMatchClass = VecListTwoDHWordIndexAsmOperand;
195 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
197 // ...with word lane subscripting.
198 def VecListTwoDWordIndexAsmOperand : AsmOperandClass {
199 let Name = "VecListTwoDWordIndexed";
200 let ParserMethod = "parseVectorList";
201 let RenderMethod = "addVecListIndexedOperands";
203 def VecListTwoDWordIndexed : Operand<i32> {
204 let ParserMatchClass = VecListTwoDWordIndexAsmOperand;
205 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
208 //===----------------------------------------------------------------------===//
209 // NEON-specific DAG Nodes.
210 //===----------------------------------------------------------------------===//
212 def SDTARMVCMP : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisSameAs<1, 2>]>;
213 def SDTARMVCMPZ : SDTypeProfile<1, 1, []>;
215 def NEONvceq : SDNode<"ARMISD::VCEQ", SDTARMVCMP>;
216 def NEONvceqz : SDNode<"ARMISD::VCEQZ", SDTARMVCMPZ>;
217 def NEONvcge : SDNode<"ARMISD::VCGE", SDTARMVCMP>;
218 def NEONvcgez : SDNode<"ARMISD::VCGEZ", SDTARMVCMPZ>;
219 def NEONvclez : SDNode<"ARMISD::VCLEZ", SDTARMVCMPZ>;
220 def NEONvcgeu : SDNode<"ARMISD::VCGEU", SDTARMVCMP>;
221 def NEONvcgt : SDNode<"ARMISD::VCGT", SDTARMVCMP>;
222 def NEONvcgtz : SDNode<"ARMISD::VCGTZ", SDTARMVCMPZ>;
223 def NEONvcltz : SDNode<"ARMISD::VCLTZ", SDTARMVCMPZ>;
224 def NEONvcgtu : SDNode<"ARMISD::VCGTU", SDTARMVCMP>;
225 def NEONvtst : SDNode<"ARMISD::VTST", SDTARMVCMP>;
227 // Types for vector shift by immediates. The "SHX" version is for long and
228 // narrow operations where the source and destination vectors have different
229 // types. The "SHINS" version is for shift and insert operations.
230 def SDTARMVSH : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
232 def SDTARMVSHX : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
234 def SDTARMVSHINS : SDTypeProfile<1, 3, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
235 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
237 def NEONvshl : SDNode<"ARMISD::VSHL", SDTARMVSH>;
238 def NEONvshrs : SDNode<"ARMISD::VSHRs", SDTARMVSH>;
239 def NEONvshru : SDNode<"ARMISD::VSHRu", SDTARMVSH>;
240 def NEONvshlls : SDNode<"ARMISD::VSHLLs", SDTARMVSHX>;
241 def NEONvshllu : SDNode<"ARMISD::VSHLLu", SDTARMVSHX>;
242 def NEONvshlli : SDNode<"ARMISD::VSHLLi", SDTARMVSHX>;
243 def NEONvshrn : SDNode<"ARMISD::VSHRN", SDTARMVSHX>;
245 def NEONvrshrs : SDNode<"ARMISD::VRSHRs", SDTARMVSH>;
246 def NEONvrshru : SDNode<"ARMISD::VRSHRu", SDTARMVSH>;
247 def NEONvrshrn : SDNode<"ARMISD::VRSHRN", SDTARMVSHX>;
249 def NEONvqshls : SDNode<"ARMISD::VQSHLs", SDTARMVSH>;
250 def NEONvqshlu : SDNode<"ARMISD::VQSHLu", SDTARMVSH>;
251 def NEONvqshlsu : SDNode<"ARMISD::VQSHLsu", SDTARMVSH>;
252 def NEONvqshrns : SDNode<"ARMISD::VQSHRNs", SDTARMVSHX>;
253 def NEONvqshrnu : SDNode<"ARMISD::VQSHRNu", SDTARMVSHX>;
254 def NEONvqshrnsu : SDNode<"ARMISD::VQSHRNsu", SDTARMVSHX>;
256 def NEONvqrshrns : SDNode<"ARMISD::VQRSHRNs", SDTARMVSHX>;
257 def NEONvqrshrnu : SDNode<"ARMISD::VQRSHRNu", SDTARMVSHX>;
258 def NEONvqrshrnsu : SDNode<"ARMISD::VQRSHRNsu", SDTARMVSHX>;
260 def NEONvsli : SDNode<"ARMISD::VSLI", SDTARMVSHINS>;
261 def NEONvsri : SDNode<"ARMISD::VSRI", SDTARMVSHINS>;
263 def SDTARMVGETLN : SDTypeProfile<1, 2, [SDTCisVT<0, i32>, SDTCisInt<1>,
265 def NEONvgetlaneu : SDNode<"ARMISD::VGETLANEu", SDTARMVGETLN>;
266 def NEONvgetlanes : SDNode<"ARMISD::VGETLANEs", SDTARMVGETLN>;
268 def SDTARMVMOVIMM : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisVT<1, i32>]>;
269 def NEONvmovImm : SDNode<"ARMISD::VMOVIMM", SDTARMVMOVIMM>;
270 def NEONvmvnImm : SDNode<"ARMISD::VMVNIMM", SDTARMVMOVIMM>;
271 def NEONvmovFPImm : SDNode<"ARMISD::VMOVFPIMM", SDTARMVMOVIMM>;
273 def SDTARMVORRIMM : SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
275 def NEONvorrImm : SDNode<"ARMISD::VORRIMM", SDTARMVORRIMM>;
276 def NEONvbicImm : SDNode<"ARMISD::VBICIMM", SDTARMVORRIMM>;
278 def NEONvbsl : SDNode<"ARMISD::VBSL",
279 SDTypeProfile<1, 3, [SDTCisVec<0>,
282 SDTCisSameAs<0, 3>]>>;
284 def NEONvdup : SDNode<"ARMISD::VDUP", SDTypeProfile<1, 1, [SDTCisVec<0>]>>;
286 // VDUPLANE can produce a quad-register result from a double-register source,
287 // so the result is not constrained to match the source.
288 def NEONvduplane : SDNode<"ARMISD::VDUPLANE",
289 SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisVec<1>,
292 def SDTARMVEXT : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
293 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
294 def NEONvext : SDNode<"ARMISD::VEXT", SDTARMVEXT>;
296 def SDTARMVSHUF : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisSameAs<0, 1>]>;
297 def NEONvrev64 : SDNode<"ARMISD::VREV64", SDTARMVSHUF>;
298 def NEONvrev32 : SDNode<"ARMISD::VREV32", SDTARMVSHUF>;
299 def NEONvrev16 : SDNode<"ARMISD::VREV16", SDTARMVSHUF>;
301 def SDTARMVSHUF2 : SDTypeProfile<2, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
303 SDTCisSameAs<0, 3>]>;
304 def NEONzip : SDNode<"ARMISD::VZIP", SDTARMVSHUF2>;
305 def NEONuzp : SDNode<"ARMISD::VUZP", SDTARMVSHUF2>;
306 def NEONtrn : SDNode<"ARMISD::VTRN", SDTARMVSHUF2>;
308 def SDTARMVMULL : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
309 SDTCisSameAs<1, 2>]>;
310 def NEONvmulls : SDNode<"ARMISD::VMULLs", SDTARMVMULL>;
311 def NEONvmullu : SDNode<"ARMISD::VMULLu", SDTARMVMULL>;
313 def SDTARMFMAX : SDTypeProfile<1, 2, [SDTCisVT<0, f32>, SDTCisSameAs<0, 1>,
314 SDTCisSameAs<0, 2>]>;
315 def NEONfmax : SDNode<"ARMISD::FMAX", SDTARMFMAX>;
316 def NEONfmin : SDNode<"ARMISD::FMIN", SDTARMFMAX>;
318 def NEONimmAllZerosV: PatLeaf<(NEONvmovImm (i32 timm)), [{
319 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
320 unsigned EltBits = 0;
321 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
322 return (EltBits == 32 && EltVal == 0);
325 def NEONimmAllOnesV: PatLeaf<(NEONvmovImm (i32 timm)), [{
326 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
327 unsigned EltBits = 0;
328 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
329 return (EltBits == 8 && EltVal == 0xff);
332 //===----------------------------------------------------------------------===//
333 // NEON load / store instructions
334 //===----------------------------------------------------------------------===//
336 // Use VLDM to load a Q register as a D register pair.
337 // This is a pseudo instruction that is expanded to VLDMD after reg alloc.
339 : PseudoVFPLdStM<(outs QPR:$dst), (ins GPR:$Rn),
341 [(set QPR:$dst, (v2f64 (load GPR:$Rn)))]>;
343 // Use VSTM to store a Q register as a D register pair.
344 // This is a pseudo instruction that is expanded to VSTMD after reg alloc.
346 : PseudoVFPLdStM<(outs), (ins QPR:$src, GPR:$Rn),
348 [(store (v2f64 QPR:$src), GPR:$Rn)]>;
350 // Classes for VLD* pseudo-instructions with multi-register operands.
351 // These are expanded to real instructions after register allocation.
352 class VLDQPseudo<InstrItinClass itin>
353 : PseudoNLdSt<(outs QPR:$dst), (ins addrmode6:$addr), itin, "">;
354 class VLDQWBPseudo<InstrItinClass itin>
355 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
356 (ins addrmode6:$addr, am6offset:$offset), itin,
358 class VLDQWBfixedPseudo<InstrItinClass itin>
359 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
360 (ins addrmode6:$addr), itin,
362 class VLDQWBregisterPseudo<InstrItinClass itin>
363 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
364 (ins addrmode6:$addr, rGPR:$offset), itin,
367 class VLDQQPseudo<InstrItinClass itin>
368 : PseudoNLdSt<(outs QQPR:$dst), (ins addrmode6:$addr), itin, "">;
369 class VLDQQWBPseudo<InstrItinClass itin>
370 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
371 (ins addrmode6:$addr, am6offset:$offset), itin,
373 class VLDQQWBfixedPseudo<InstrItinClass itin>
374 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
375 (ins addrmode6:$addr), itin,
377 class VLDQQWBregisterPseudo<InstrItinClass itin>
378 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
379 (ins addrmode6:$addr, rGPR:$offset), itin,
383 class VLDQQQQPseudo<InstrItinClass itin>
384 : PseudoNLdSt<(outs QQQQPR:$dst), (ins addrmode6:$addr, QQQQPR:$src),itin,
386 class VLDQQQQWBPseudo<InstrItinClass itin>
387 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
388 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
389 "$addr.addr = $wb, $src = $dst">;
391 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
393 // VLD1 : Vector Load (multiple single elements)
394 class VLD1D<bits<4> op7_4, string Dt>
395 : NLdSt<0,0b10,0b0111,op7_4, (outs VecListOneD:$Vd),
396 (ins addrmode6:$Rn), IIC_VLD1,
397 "vld1", Dt, "$Vd, $Rn", "", []> {
400 let DecoderMethod = "DecodeVLDInstruction";
402 class VLD1Q<bits<4> op7_4, string Dt>
403 : NLdSt<0,0b10,0b1010,op7_4, (outs VecListTwoD:$Vd),
404 (ins addrmode6:$Rn), IIC_VLD1x2,
405 "vld1", Dt, "$Vd, $Rn", "", []> {
407 let Inst{5-4} = Rn{5-4};
408 let DecoderMethod = "DecodeVLDInstruction";
411 def VLD1d8 : VLD1D<{0,0,0,?}, "8">;
412 def VLD1d16 : VLD1D<{0,1,0,?}, "16">;
413 def VLD1d32 : VLD1D<{1,0,0,?}, "32">;
414 def VLD1d64 : VLD1D<{1,1,0,?}, "64">;
416 def VLD1q8 : VLD1Q<{0,0,?,?}, "8">;
417 def VLD1q16 : VLD1Q<{0,1,?,?}, "16">;
418 def VLD1q32 : VLD1Q<{1,0,?,?}, "32">;
419 def VLD1q64 : VLD1Q<{1,1,?,?}, "64">;
421 def VLD1q8Pseudo : VLDQPseudo<IIC_VLD1x2>;
422 def VLD1q16Pseudo : VLDQPseudo<IIC_VLD1x2>;
423 def VLD1q32Pseudo : VLDQPseudo<IIC_VLD1x2>;
424 def VLD1q64Pseudo : VLDQPseudo<IIC_VLD1x2>;
426 // ...with address register writeback:
427 multiclass VLD1DWB<bits<4> op7_4, string Dt> {
428 def _fixed : NLdSt<0,0b10, 0b0111,op7_4, (outs VecListOneD:$Vd, GPR:$wb),
429 (ins addrmode6:$Rn), IIC_VLD1u,
430 "vld1", Dt, "$Vd, $Rn!",
431 "$Rn.addr = $wb", []> {
432 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
434 let DecoderMethod = "DecodeVLDInstruction";
435 let AsmMatchConverter = "cvtVLDwbFixed";
437 def _register : NLdSt<0,0b10,0b0111,op7_4, (outs VecListOneD:$Vd, GPR:$wb),
438 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1u,
439 "vld1", Dt, "$Vd, $Rn, $Rm",
440 "$Rn.addr = $wb", []> {
442 let DecoderMethod = "DecodeVLDInstruction";
443 let AsmMatchConverter = "cvtVLDwbRegister";
446 multiclass VLD1QWB<bits<4> op7_4, string Dt> {
447 def _fixed : NLdSt<0,0b10,0b1010,op7_4, (outs VecListTwoD:$Vd, GPR:$wb),
448 (ins addrmode6:$Rn), IIC_VLD1x2u,
449 "vld1", Dt, "$Vd, $Rn!",
450 "$Rn.addr = $wb", []> {
451 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
452 let Inst{5-4} = Rn{5-4};
453 let DecoderMethod = "DecodeVLDInstruction";
454 let AsmMatchConverter = "cvtVLDwbFixed";
456 def _register : NLdSt<0,0b10,0b1010,op7_4, (outs VecListTwoD:$Vd, GPR:$wb),
457 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1x2u,
458 "vld1", Dt, "$Vd, $Rn, $Rm",
459 "$Rn.addr = $wb", []> {
460 let Inst{5-4} = Rn{5-4};
461 let DecoderMethod = "DecodeVLDInstruction";
462 let AsmMatchConverter = "cvtVLDwbRegister";
466 defm VLD1d8wb : VLD1DWB<{0,0,0,?}, "8">;
467 defm VLD1d16wb : VLD1DWB<{0,1,0,?}, "16">;
468 defm VLD1d32wb : VLD1DWB<{1,0,0,?}, "32">;
469 defm VLD1d64wb : VLD1DWB<{1,1,0,?}, "64">;
470 defm VLD1q8wb : VLD1QWB<{0,0,?,?}, "8">;
471 defm VLD1q16wb : VLD1QWB<{0,1,?,?}, "16">;
472 defm VLD1q32wb : VLD1QWB<{1,0,?,?}, "32">;
473 defm VLD1q64wb : VLD1QWB<{1,1,?,?}, "64">;
475 def VLD1q8PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD1x2u>;
476 def VLD1q16PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD1x2u>;
477 def VLD1q32PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD1x2u>;
478 def VLD1q64PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD1x2u>;
479 def VLD1q8PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD1x2u>;
480 def VLD1q16PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD1x2u>;
481 def VLD1q32PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD1x2u>;
482 def VLD1q64PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD1x2u>;
484 // ...with 3 registers
485 class VLD1D3<bits<4> op7_4, string Dt>
486 : NLdSt<0,0b10,0b0110,op7_4, (outs VecListThreeD:$Vd),
487 (ins addrmode6:$Rn), IIC_VLD1x3, "vld1", Dt,
488 "$Vd, $Rn", "", []> {
491 let DecoderMethod = "DecodeVLDInstruction";
493 multiclass VLD1D3WB<bits<4> op7_4, string Dt> {
494 def _fixed : NLdSt<0,0b10,0b0110, op7_4, (outs VecListThreeD:$Vd, GPR:$wb),
495 (ins addrmode6:$Rn), IIC_VLD1x2u,
496 "vld1", Dt, "$Vd, $Rn!",
497 "$Rn.addr = $wb", []> {
498 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
500 let DecoderMethod = "DecodeVLDInstruction";
501 let AsmMatchConverter = "cvtVLDwbFixed";
503 def _register : NLdSt<0,0b10,0b0110,op7_4, (outs VecListThreeD:$Vd, GPR:$wb),
504 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1x2u,
505 "vld1", Dt, "$Vd, $Rn, $Rm",
506 "$Rn.addr = $wb", []> {
508 let DecoderMethod = "DecodeVLDInstruction";
509 let AsmMatchConverter = "cvtVLDwbRegister";
513 def VLD1d8T : VLD1D3<{0,0,0,?}, "8">;
514 def VLD1d16T : VLD1D3<{0,1,0,?}, "16">;
515 def VLD1d32T : VLD1D3<{1,0,0,?}, "32">;
516 def VLD1d64T : VLD1D3<{1,1,0,?}, "64">;
518 defm VLD1d8Twb : VLD1D3WB<{0,0,0,?}, "8">;
519 defm VLD1d16Twb : VLD1D3WB<{0,1,0,?}, "16">;
520 defm VLD1d32Twb : VLD1D3WB<{1,0,0,?}, "32">;
521 defm VLD1d64Twb : VLD1D3WB<{1,1,0,?}, "64">;
523 def VLD1d64TPseudo : VLDQQPseudo<IIC_VLD1x3>;
525 // ...with 4 registers
526 class VLD1D4<bits<4> op7_4, string Dt>
527 : NLdSt<0, 0b10, 0b0010, op7_4, (outs VecListFourD:$Vd),
528 (ins addrmode6:$Rn), IIC_VLD1x4, "vld1", Dt,
529 "$Vd, $Rn", "", []> {
531 let Inst{5-4} = Rn{5-4};
532 let DecoderMethod = "DecodeVLDInstruction";
534 multiclass VLD1D4WB<bits<4> op7_4, string Dt> {
535 def _fixed : NLdSt<0,0b10,0b0010, op7_4, (outs VecListFourD:$Vd, GPR:$wb),
536 (ins addrmode6:$Rn), IIC_VLD1x2u,
537 "vld1", Dt, "$Vd, $Rn!",
538 "$Rn.addr = $wb", []> {
539 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
540 let Inst{5-4} = Rn{5-4};
541 let DecoderMethod = "DecodeVLDInstruction";
542 let AsmMatchConverter = "cvtVLDwbFixed";
544 def _register : NLdSt<0,0b10,0b0010,op7_4, (outs VecListFourD:$Vd, GPR:$wb),
545 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1x2u,
546 "vld1", Dt, "$Vd, $Rn, $Rm",
547 "$Rn.addr = $wb", []> {
548 let Inst{5-4} = Rn{5-4};
549 let DecoderMethod = "DecodeVLDInstruction";
550 let AsmMatchConverter = "cvtVLDwbRegister";
554 def VLD1d8Q : VLD1D4<{0,0,?,?}, "8">;
555 def VLD1d16Q : VLD1D4<{0,1,?,?}, "16">;
556 def VLD1d32Q : VLD1D4<{1,0,?,?}, "32">;
557 def VLD1d64Q : VLD1D4<{1,1,?,?}, "64">;
559 defm VLD1d8Qwb : VLD1D4WB<{0,0,?,?}, "8">;
560 defm VLD1d16Qwb : VLD1D4WB<{0,1,?,?}, "16">;
561 defm VLD1d32Qwb : VLD1D4WB<{1,0,?,?}, "32">;
562 defm VLD1d64Qwb : VLD1D4WB<{1,1,?,?}, "64">;
564 def VLD1d64QPseudo : VLDQQPseudo<IIC_VLD1x4>;
566 // VLD2 : Vector Load (multiple 2-element structures)
567 class VLD2<bits<4> op11_8, bits<4> op7_4, string Dt, RegisterOperand VdTy,
569 : NLdSt<0, 0b10, op11_8, op7_4, (outs VdTy:$Vd),
570 (ins addrmode6:$Rn), itin,
571 "vld2", Dt, "$Vd, $Rn", "", []> {
573 let Inst{5-4} = Rn{5-4};
574 let DecoderMethod = "DecodeVLDInstruction";
577 def VLD2d8 : VLD2<0b1000, {0,0,?,?}, "8", VecListTwoD, IIC_VLD2>;
578 def VLD2d16 : VLD2<0b1000, {0,1,?,?}, "16", VecListTwoD, IIC_VLD2>;
579 def VLD2d32 : VLD2<0b1000, {1,0,?,?}, "32", VecListTwoD, IIC_VLD2>;
581 def VLD2q8 : VLD2<0b0011, {0,0,?,?}, "8", VecListFourD, IIC_VLD2x2>;
582 def VLD2q16 : VLD2<0b0011, {0,1,?,?}, "16", VecListFourD, IIC_VLD2x2>;
583 def VLD2q32 : VLD2<0b0011, {1,0,?,?}, "32", VecListFourD, IIC_VLD2x2>;
585 def VLD2d8Pseudo : VLDQPseudo<IIC_VLD2>;
586 def VLD2d16Pseudo : VLDQPseudo<IIC_VLD2>;
587 def VLD2d32Pseudo : VLDQPseudo<IIC_VLD2>;
589 def VLD2q8Pseudo : VLDQQPseudo<IIC_VLD2x2>;
590 def VLD2q16Pseudo : VLDQQPseudo<IIC_VLD2x2>;
591 def VLD2q32Pseudo : VLDQQPseudo<IIC_VLD2x2>;
593 // ...with address register writeback:
594 multiclass VLD2WB<bits<4> op11_8, bits<4> op7_4, string Dt,
595 RegisterOperand VdTy, InstrItinClass itin> {
596 def _fixed : NLdSt<0, 0b10, op11_8, op7_4, (outs VdTy:$Vd, GPR:$wb),
597 (ins addrmode6:$Rn), itin,
598 "vld2", Dt, "$Vd, $Rn!",
599 "$Rn.addr = $wb", []> {
600 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
601 let Inst{5-4} = Rn{5-4};
602 let DecoderMethod = "DecodeVLDInstruction";
603 let AsmMatchConverter = "cvtVLDwbFixed";
605 def _register : NLdSt<0, 0b10, op11_8, op7_4, (outs VdTy:$Vd, GPR:$wb),
606 (ins addrmode6:$Rn, rGPR:$Rm), itin,
607 "vld2", Dt, "$Vd, $Rn, $Rm",
608 "$Rn.addr = $wb", []> {
609 let Inst{5-4} = Rn{5-4};
610 let DecoderMethod = "DecodeVLDInstruction";
611 let AsmMatchConverter = "cvtVLDwbRegister";
615 defm VLD2d8wb : VLD2WB<0b1000, {0,0,?,?}, "8", VecListTwoD, IIC_VLD2u>;
616 defm VLD2d16wb : VLD2WB<0b1000, {0,1,?,?}, "16", VecListTwoD, IIC_VLD2u>;
617 defm VLD2d32wb : VLD2WB<0b1000, {1,0,?,?}, "32", VecListTwoD, IIC_VLD2u>;
619 defm VLD2q8wb : VLD2WB<0b0011, {0,0,?,?}, "8", VecListFourD, IIC_VLD2x2u>;
620 defm VLD2q16wb : VLD2WB<0b0011, {0,1,?,?}, "16", VecListFourD, IIC_VLD2x2u>;
621 defm VLD2q32wb : VLD2WB<0b0011, {1,0,?,?}, "32", VecListFourD, IIC_VLD2x2u>;
623 def VLD2d8PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD2u>;
624 def VLD2d16PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD2u>;
625 def VLD2d32PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD2u>;
626 def VLD2d8PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD2u>;
627 def VLD2d16PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD2u>;
628 def VLD2d32PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD2u>;
630 def VLD2q8PseudoWB_fixed : VLDQQWBfixedPseudo<IIC_VLD2x2u>;
631 def VLD2q16PseudoWB_fixed : VLDQQWBfixedPseudo<IIC_VLD2x2u>;
632 def VLD2q32PseudoWB_fixed : VLDQQWBfixedPseudo<IIC_VLD2x2u>;
633 def VLD2q8PseudoWB_register : VLDQQWBregisterPseudo<IIC_VLD2x2u>;
634 def VLD2q16PseudoWB_register : VLDQQWBregisterPseudo<IIC_VLD2x2u>;
635 def VLD2q32PseudoWB_register : VLDQQWBregisterPseudo<IIC_VLD2x2u>;
637 // ...with double-spaced registers
638 def VLD2b8 : VLD2<0b1001, {0,0,?,?}, "8", VecListTwoQ, IIC_VLD2>;
639 def VLD2b16 : VLD2<0b1001, {0,1,?,?}, "16", VecListTwoQ, IIC_VLD2>;
640 def VLD2b32 : VLD2<0b1001, {1,0,?,?}, "32", VecListTwoQ, IIC_VLD2>;
641 defm VLD2b8wb : VLD2WB<0b1001, {0,0,?,?}, "8", VecListTwoQ, IIC_VLD2u>;
642 defm VLD2b16wb : VLD2WB<0b1001, {0,1,?,?}, "16", VecListTwoQ, IIC_VLD2u>;
643 defm VLD2b32wb : VLD2WB<0b1001, {1,0,?,?}, "32", VecListTwoQ, IIC_VLD2u>;
645 // VLD3 : Vector Load (multiple 3-element structures)
646 class VLD3D<bits<4> op11_8, bits<4> op7_4, string Dt>
647 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
648 (ins addrmode6:$Rn), IIC_VLD3,
649 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn", "", []> {
652 let DecoderMethod = "DecodeVLDInstruction";
655 def VLD3d8 : VLD3D<0b0100, {0,0,0,?}, "8">;
656 def VLD3d16 : VLD3D<0b0100, {0,1,0,?}, "16">;
657 def VLD3d32 : VLD3D<0b0100, {1,0,0,?}, "32">;
659 def VLD3d8Pseudo : VLDQQPseudo<IIC_VLD3>;
660 def VLD3d16Pseudo : VLDQQPseudo<IIC_VLD3>;
661 def VLD3d32Pseudo : VLDQQPseudo<IIC_VLD3>;
663 // ...with address register writeback:
664 class VLD3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
665 : NLdSt<0, 0b10, op11_8, op7_4,
666 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
667 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD3u,
668 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn$Rm",
669 "$Rn.addr = $wb", []> {
671 let DecoderMethod = "DecodeVLDInstruction";
674 def VLD3d8_UPD : VLD3DWB<0b0100, {0,0,0,?}, "8">;
675 def VLD3d16_UPD : VLD3DWB<0b0100, {0,1,0,?}, "16">;
676 def VLD3d32_UPD : VLD3DWB<0b0100, {1,0,0,?}, "32">;
678 def VLD3d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
679 def VLD3d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
680 def VLD3d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
682 // ...with double-spaced registers:
683 def VLD3q8 : VLD3D<0b0101, {0,0,0,?}, "8">;
684 def VLD3q16 : VLD3D<0b0101, {0,1,0,?}, "16">;
685 def VLD3q32 : VLD3D<0b0101, {1,0,0,?}, "32">;
686 def VLD3q8_UPD : VLD3DWB<0b0101, {0,0,0,?}, "8">;
687 def VLD3q16_UPD : VLD3DWB<0b0101, {0,1,0,?}, "16">;
688 def VLD3q32_UPD : VLD3DWB<0b0101, {1,0,0,?}, "32">;
690 def VLD3q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
691 def VLD3q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
692 def VLD3q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
694 // ...alternate versions to be allocated odd register numbers:
695 def VLD3q8oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
696 def VLD3q16oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
697 def VLD3q32oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
699 def VLD3q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
700 def VLD3q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
701 def VLD3q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
703 // VLD4 : Vector Load (multiple 4-element structures)
704 class VLD4D<bits<4> op11_8, bits<4> op7_4, string Dt>
705 : NLdSt<0, 0b10, op11_8, op7_4,
706 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
707 (ins addrmode6:$Rn), IIC_VLD4,
708 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn", "", []> {
710 let Inst{5-4} = Rn{5-4};
711 let DecoderMethod = "DecodeVLDInstruction";
714 def VLD4d8 : VLD4D<0b0000, {0,0,?,?}, "8">;
715 def VLD4d16 : VLD4D<0b0000, {0,1,?,?}, "16">;
716 def VLD4d32 : VLD4D<0b0000, {1,0,?,?}, "32">;
718 def VLD4d8Pseudo : VLDQQPseudo<IIC_VLD4>;
719 def VLD4d16Pseudo : VLDQQPseudo<IIC_VLD4>;
720 def VLD4d32Pseudo : VLDQQPseudo<IIC_VLD4>;
722 // ...with address register writeback:
723 class VLD4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
724 : NLdSt<0, 0b10, op11_8, op7_4,
725 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
726 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD4u,
727 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn$Rm",
728 "$Rn.addr = $wb", []> {
729 let Inst{5-4} = Rn{5-4};
730 let DecoderMethod = "DecodeVLDInstruction";
733 def VLD4d8_UPD : VLD4DWB<0b0000, {0,0,?,?}, "8">;
734 def VLD4d16_UPD : VLD4DWB<0b0000, {0,1,?,?}, "16">;
735 def VLD4d32_UPD : VLD4DWB<0b0000, {1,0,?,?}, "32">;
737 def VLD4d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
738 def VLD4d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
739 def VLD4d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
741 // ...with double-spaced registers:
742 def VLD4q8 : VLD4D<0b0001, {0,0,?,?}, "8">;
743 def VLD4q16 : VLD4D<0b0001, {0,1,?,?}, "16">;
744 def VLD4q32 : VLD4D<0b0001, {1,0,?,?}, "32">;
745 def VLD4q8_UPD : VLD4DWB<0b0001, {0,0,?,?}, "8">;
746 def VLD4q16_UPD : VLD4DWB<0b0001, {0,1,?,?}, "16">;
747 def VLD4q32_UPD : VLD4DWB<0b0001, {1,0,?,?}, "32">;
749 def VLD4q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
750 def VLD4q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
751 def VLD4q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
753 // ...alternate versions to be allocated odd register numbers:
754 def VLD4q8oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
755 def VLD4q16oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
756 def VLD4q32oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
758 def VLD4q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
759 def VLD4q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
760 def VLD4q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
762 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
764 // Classes for VLD*LN pseudo-instructions with multi-register operands.
765 // These are expanded to real instructions after register allocation.
766 class VLDQLNPseudo<InstrItinClass itin>
767 : PseudoNLdSt<(outs QPR:$dst),
768 (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
769 itin, "$src = $dst">;
770 class VLDQLNWBPseudo<InstrItinClass itin>
771 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
772 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
773 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
774 class VLDQQLNPseudo<InstrItinClass itin>
775 : PseudoNLdSt<(outs QQPR:$dst),
776 (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
777 itin, "$src = $dst">;
778 class VLDQQLNWBPseudo<InstrItinClass itin>
779 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
780 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
781 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
782 class VLDQQQQLNPseudo<InstrItinClass itin>
783 : PseudoNLdSt<(outs QQQQPR:$dst),
784 (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
785 itin, "$src = $dst">;
786 class VLDQQQQLNWBPseudo<InstrItinClass itin>
787 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
788 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
789 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
791 // VLD1LN : Vector Load (single element to one lane)
792 class VLD1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
794 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd),
795 (ins addrmode6:$Rn, DPR:$src, nohash_imm:$lane),
796 IIC_VLD1ln, "vld1", Dt, "\\{$Vd[$lane]\\}, $Rn",
798 [(set DPR:$Vd, (vector_insert (Ty DPR:$src),
799 (i32 (LoadOp addrmode6:$Rn)),
802 let DecoderMethod = "DecodeVLD1LN";
804 class VLD1LN32<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
806 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd),
807 (ins addrmode6oneL32:$Rn, DPR:$src, nohash_imm:$lane),
808 IIC_VLD1ln, "vld1", Dt, "\\{$Vd[$lane]\\}, $Rn",
810 [(set DPR:$Vd, (vector_insert (Ty DPR:$src),
811 (i32 (LoadOp addrmode6oneL32:$Rn)),
814 let DecoderMethod = "DecodeVLD1LN";
816 class VLD1QLNPseudo<ValueType Ty, PatFrag LoadOp> : VLDQLNPseudo<IIC_VLD1ln> {
817 let Pattern = [(set QPR:$dst, (vector_insert (Ty QPR:$src),
818 (i32 (LoadOp addrmode6:$addr)),
822 def VLD1LNd8 : VLD1LN<0b0000, {?,?,?,0}, "8", v8i8, extloadi8> {
823 let Inst{7-5} = lane{2-0};
825 def VLD1LNd16 : VLD1LN<0b0100, {?,?,0,?}, "16", v4i16, extloadi16> {
826 let Inst{7-6} = lane{1-0};
827 let Inst{5-4} = Rn{5-4};
829 def VLD1LNd32 : VLD1LN32<0b1000, {?,0,?,?}, "32", v2i32, load> {
830 let Inst{7} = lane{0};
831 let Inst{5-4} = Rn{5-4};
834 def VLD1LNq8Pseudo : VLD1QLNPseudo<v16i8, extloadi8>;
835 def VLD1LNq16Pseudo : VLD1QLNPseudo<v8i16, extloadi16>;
836 def VLD1LNq32Pseudo : VLD1QLNPseudo<v4i32, load>;
838 def : Pat<(vector_insert (v2f32 DPR:$src),
839 (f32 (load addrmode6:$addr)), imm:$lane),
840 (VLD1LNd32 addrmode6:$addr, DPR:$src, imm:$lane)>;
841 def : Pat<(vector_insert (v4f32 QPR:$src),
842 (f32 (load addrmode6:$addr)), imm:$lane),
843 (VLD1LNq32Pseudo addrmode6:$addr, QPR:$src, imm:$lane)>;
845 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
847 // ...with address register writeback:
848 class VLD1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
849 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, GPR:$wb),
850 (ins addrmode6:$Rn, am6offset:$Rm,
851 DPR:$src, nohash_imm:$lane), IIC_VLD1lnu, "vld1", Dt,
852 "\\{$Vd[$lane]\\}, $Rn$Rm",
853 "$src = $Vd, $Rn.addr = $wb", []> {
854 let DecoderMethod = "DecodeVLD1LN";
857 def VLD1LNd8_UPD : VLD1LNWB<0b0000, {?,?,?,0}, "8"> {
858 let Inst{7-5} = lane{2-0};
860 def VLD1LNd16_UPD : VLD1LNWB<0b0100, {?,?,0,?}, "16"> {
861 let Inst{7-6} = lane{1-0};
864 def VLD1LNd32_UPD : VLD1LNWB<0b1000, {?,0,?,?}, "32"> {
865 let Inst{7} = lane{0};
870 def VLD1LNq8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
871 def VLD1LNq16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
872 def VLD1LNq32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
874 // VLD2LN : Vector Load (single 2-element structure to one lane)
875 class VLD2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
876 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2),
877 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, nohash_imm:$lane),
878 IIC_VLD2ln, "vld2", Dt, "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn",
879 "$src1 = $Vd, $src2 = $dst2", []> {
882 let DecoderMethod = "DecodeVLD2LN";
885 def VLD2LNd8 : VLD2LN<0b0001, {?,?,?,?}, "8"> {
886 let Inst{7-5} = lane{2-0};
888 def VLD2LNd16 : VLD2LN<0b0101, {?,?,0,?}, "16"> {
889 let Inst{7-6} = lane{1-0};
891 def VLD2LNd32 : VLD2LN<0b1001, {?,0,0,?}, "32"> {
892 let Inst{7} = lane{0};
895 def VLD2LNd8Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
896 def VLD2LNd16Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
897 def VLD2LNd32Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
899 // ...with double-spaced registers:
900 def VLD2LNq16 : VLD2LN<0b0101, {?,?,1,?}, "16"> {
901 let Inst{7-6} = lane{1-0};
903 def VLD2LNq32 : VLD2LN<0b1001, {?,1,0,?}, "32"> {
904 let Inst{7} = lane{0};
907 def VLD2LNq16Pseudo : VLDQQLNPseudo<IIC_VLD2ln>;
908 def VLD2LNq32Pseudo : VLDQQLNPseudo<IIC_VLD2ln>;
910 // ...with address register writeback:
911 class VLD2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
912 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
913 (ins addrmode6:$Rn, am6offset:$Rm,
914 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VLD2lnu, "vld2", Dt,
915 "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn$Rm",
916 "$src1 = $Vd, $src2 = $dst2, $Rn.addr = $wb", []> {
918 let DecoderMethod = "DecodeVLD2LN";
921 def VLD2LNd8_UPD : VLD2LNWB<0b0001, {?,?,?,?}, "8"> {
922 let Inst{7-5} = lane{2-0};
924 def VLD2LNd16_UPD : VLD2LNWB<0b0101, {?,?,0,?}, "16"> {
925 let Inst{7-6} = lane{1-0};
927 def VLD2LNd32_UPD : VLD2LNWB<0b1001, {?,0,0,?}, "32"> {
928 let Inst{7} = lane{0};
931 def VLD2LNd8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
932 def VLD2LNd16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
933 def VLD2LNd32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
935 def VLD2LNq16_UPD : VLD2LNWB<0b0101, {?,?,1,?}, "16"> {
936 let Inst{7-6} = lane{1-0};
938 def VLD2LNq32_UPD : VLD2LNWB<0b1001, {?,1,0,?}, "32"> {
939 let Inst{7} = lane{0};
942 def VLD2LNq16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
943 def VLD2LNq32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
945 // VLD3LN : Vector Load (single 3-element structure to one lane)
946 class VLD3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
947 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
948 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3,
949 nohash_imm:$lane), IIC_VLD3ln, "vld3", Dt,
950 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn",
951 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3", []> {
953 let DecoderMethod = "DecodeVLD3LN";
956 def VLD3LNd8 : VLD3LN<0b0010, {?,?,?,0}, "8"> {
957 let Inst{7-5} = lane{2-0};
959 def VLD3LNd16 : VLD3LN<0b0110, {?,?,0,0}, "16"> {
960 let Inst{7-6} = lane{1-0};
962 def VLD3LNd32 : VLD3LN<0b1010, {?,0,0,0}, "32"> {
963 let Inst{7} = lane{0};
966 def VLD3LNd8Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
967 def VLD3LNd16Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
968 def VLD3LNd32Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
970 // ...with double-spaced registers:
971 def VLD3LNq16 : VLD3LN<0b0110, {?,?,1,0}, "16"> {
972 let Inst{7-6} = lane{1-0};
974 def VLD3LNq32 : VLD3LN<0b1010, {?,1,0,0}, "32"> {
975 let Inst{7} = lane{0};
978 def VLD3LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
979 def VLD3LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
981 // ...with address register writeback:
982 class VLD3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
983 : NLdStLn<1, 0b10, op11_8, op7_4,
984 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
985 (ins addrmode6:$Rn, am6offset:$Rm,
986 DPR:$src1, DPR:$src2, DPR:$src3, nohash_imm:$lane),
987 IIC_VLD3lnu, "vld3", Dt,
988 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn$Rm",
989 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $Rn.addr = $wb",
991 let DecoderMethod = "DecodeVLD3LN";
994 def VLD3LNd8_UPD : VLD3LNWB<0b0010, {?,?,?,0}, "8"> {
995 let Inst{7-5} = lane{2-0};
997 def VLD3LNd16_UPD : VLD3LNWB<0b0110, {?,?,0,0}, "16"> {
998 let Inst{7-6} = lane{1-0};
1000 def VLD3LNd32_UPD : VLD3LNWB<0b1010, {?,0,0,0}, "32"> {
1001 let Inst{7} = lane{0};
1004 def VLD3LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
1005 def VLD3LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
1006 def VLD3LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
1008 def VLD3LNq16_UPD : VLD3LNWB<0b0110, {?,?,1,0}, "16"> {
1009 let Inst{7-6} = lane{1-0};
1011 def VLD3LNq32_UPD : VLD3LNWB<0b1010, {?,1,0,0}, "32"> {
1012 let Inst{7} = lane{0};
1015 def VLD3LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
1016 def VLD3LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
1018 // VLD4LN : Vector Load (single 4-element structure to one lane)
1019 class VLD4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1020 : NLdStLn<1, 0b10, op11_8, op7_4,
1021 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
1022 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
1023 nohash_imm:$lane), IIC_VLD4ln, "vld4", Dt,
1024 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn",
1025 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4", []> {
1027 let Inst{4} = Rn{4};
1028 let DecoderMethod = "DecodeVLD4LN";
1031 def VLD4LNd8 : VLD4LN<0b0011, {?,?,?,?}, "8"> {
1032 let Inst{7-5} = lane{2-0};
1034 def VLD4LNd16 : VLD4LN<0b0111, {?,?,0,?}, "16"> {
1035 let Inst{7-6} = lane{1-0};
1037 def VLD4LNd32 : VLD4LN<0b1011, {?,0,?,?}, "32"> {
1038 let Inst{7} = lane{0};
1039 let Inst{5} = Rn{5};
1042 def VLD4LNd8Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
1043 def VLD4LNd16Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
1044 def VLD4LNd32Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
1046 // ...with double-spaced registers:
1047 def VLD4LNq16 : VLD4LN<0b0111, {?,?,1,?}, "16"> {
1048 let Inst{7-6} = lane{1-0};
1050 def VLD4LNq32 : VLD4LN<0b1011, {?,1,?,?}, "32"> {
1051 let Inst{7} = lane{0};
1052 let Inst{5} = Rn{5};
1055 def VLD4LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>;
1056 def VLD4LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>;
1058 // ...with address register writeback:
1059 class VLD4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1060 : NLdStLn<1, 0b10, op11_8, op7_4,
1061 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
1062 (ins addrmode6:$Rn, am6offset:$Rm,
1063 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
1064 IIC_VLD4lnu, "vld4", Dt,
1065 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn$Rm",
1066 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4, $Rn.addr = $wb",
1068 let Inst{4} = Rn{4};
1069 let DecoderMethod = "DecodeVLD4LN" ;
1072 def VLD4LNd8_UPD : VLD4LNWB<0b0011, {?,?,?,?}, "8"> {
1073 let Inst{7-5} = lane{2-0};
1075 def VLD4LNd16_UPD : VLD4LNWB<0b0111, {?,?,0,?}, "16"> {
1076 let Inst{7-6} = lane{1-0};
1078 def VLD4LNd32_UPD : VLD4LNWB<0b1011, {?,0,?,?}, "32"> {
1079 let Inst{7} = lane{0};
1080 let Inst{5} = Rn{5};
1083 def VLD4LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
1084 def VLD4LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
1085 def VLD4LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
1087 def VLD4LNq16_UPD : VLD4LNWB<0b0111, {?,?,1,?}, "16"> {
1088 let Inst{7-6} = lane{1-0};
1090 def VLD4LNq32_UPD : VLD4LNWB<0b1011, {?,1,?,?}, "32"> {
1091 let Inst{7} = lane{0};
1092 let Inst{5} = Rn{5};
1095 def VLD4LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
1096 def VLD4LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
1098 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
1100 // VLD1DUP : Vector Load (single element to all lanes)
1101 class VLD1DUP<bits<4> op7_4, string Dt, ValueType Ty, PatFrag LoadOp>
1102 : NLdSt<1, 0b10, 0b1100, op7_4, (outs VecListOneDAllLanes:$Vd),
1103 (ins addrmode6dup:$Rn),
1104 IIC_VLD1dup, "vld1", Dt, "$Vd, $Rn", "",
1105 [(set VecListOneDAllLanes:$Vd,
1106 (Ty (NEONvdup (i32 (LoadOp addrmode6dup:$Rn)))))]> {
1108 let Inst{4} = Rn{4};
1109 let DecoderMethod = "DecodeVLD1DupInstruction";
1111 class VLD1QDUPPseudo<ValueType Ty, PatFrag LoadOp> : VLDQPseudo<IIC_VLD1dup> {
1112 let Pattern = [(set QPR:$dst,
1113 (Ty (NEONvdup (i32 (LoadOp addrmode6dup:$addr)))))];
1116 def VLD1DUPd8 : VLD1DUP<{0,0,0,?}, "8", v8i8, extloadi8>;
1117 def VLD1DUPd16 : VLD1DUP<{0,1,0,?}, "16", v4i16, extloadi16>;
1118 def VLD1DUPd32 : VLD1DUP<{1,0,0,?}, "32", v2i32, load>;
1120 def VLD1DUPq8Pseudo : VLD1QDUPPseudo<v16i8, extloadi8>;
1121 def VLD1DUPq16Pseudo : VLD1QDUPPseudo<v8i16, extloadi16>;
1122 def VLD1DUPq32Pseudo : VLD1QDUPPseudo<v4i32, load>;
1124 def : Pat<(v2f32 (NEONvdup (f32 (load addrmode6dup:$addr)))),
1125 (VLD1DUPd32 addrmode6:$addr)>;
1126 def : Pat<(v4f32 (NEONvdup (f32 (load addrmode6dup:$addr)))),
1127 (VLD1DUPq32Pseudo addrmode6:$addr)>;
1129 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
1131 class VLD1QDUP<bits<4> op7_4, string Dt>
1132 : NLdSt<1, 0b10, 0b1100, op7_4, (outs VecListTwoDAllLanes:$Vd),
1133 (ins addrmode6dup:$Rn), IIC_VLD1dup,
1134 "vld1", Dt, "$Vd, $Rn", "", []> {
1136 let Inst{4} = Rn{4};
1137 let DecoderMethod = "DecodeVLD1DupInstruction";
1140 def VLD1DUPq8 : VLD1QDUP<{0,0,1,0}, "8">;
1141 def VLD1DUPq16 : VLD1QDUP<{0,1,1,?}, "16">;
1142 def VLD1DUPq32 : VLD1QDUP<{1,0,1,?}, "32">;
1144 // ...with address register writeback:
1145 multiclass VLD1DUPWB<bits<4> op7_4, string Dt> {
1146 def _fixed : NLdSt<1, 0b10, 0b1100, op7_4,
1147 (outs VecListOneDAllLanes:$Vd, GPR:$wb),
1148 (ins addrmode6dup:$Rn), IIC_VLD1dupu,
1149 "vld1", Dt, "$Vd, $Rn!",
1150 "$Rn.addr = $wb", []> {
1151 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1152 let Inst{4} = Rn{4};
1153 let DecoderMethod = "DecodeVLD1DupInstruction";
1154 let AsmMatchConverter = "cvtVLDwbFixed";
1156 def _register : NLdSt<1, 0b10, 0b1100, op7_4,
1157 (outs VecListOneDAllLanes:$Vd, GPR:$wb),
1158 (ins addrmode6dup:$Rn, rGPR:$Rm), IIC_VLD1dupu,
1159 "vld1", Dt, "$Vd, $Rn, $Rm",
1160 "$Rn.addr = $wb", []> {
1161 let Inst{4} = Rn{4};
1162 let DecoderMethod = "DecodeVLD1DupInstruction";
1163 let AsmMatchConverter = "cvtVLDwbRegister";
1166 multiclass VLD1QDUPWB<bits<4> op7_4, string Dt> {
1167 def _fixed : NLdSt<1, 0b10, 0b1100, op7_4,
1168 (outs VecListTwoDAllLanes:$Vd, GPR:$wb),
1169 (ins addrmode6dup:$Rn), IIC_VLD1dupu,
1170 "vld1", Dt, "$Vd, $Rn!",
1171 "$Rn.addr = $wb", []> {
1172 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1173 let Inst{4} = Rn{4};
1174 let DecoderMethod = "DecodeVLD1DupInstruction";
1175 let AsmMatchConverter = "cvtVLDwbFixed";
1177 def _register : NLdSt<1, 0b10, 0b1100, op7_4,
1178 (outs VecListTwoDAllLanes:$Vd, GPR:$wb),
1179 (ins addrmode6dup:$Rn, rGPR:$Rm), IIC_VLD1dupu,
1180 "vld1", Dt, "$Vd, $Rn, $Rm",
1181 "$Rn.addr = $wb", []> {
1182 let Inst{4} = Rn{4};
1183 let DecoderMethod = "DecodeVLD1DupInstruction";
1184 let AsmMatchConverter = "cvtVLDwbRegister";
1188 defm VLD1DUPd8wb : VLD1DUPWB<{0,0,0,0}, "8">;
1189 defm VLD1DUPd16wb : VLD1DUPWB<{0,1,0,?}, "16">;
1190 defm VLD1DUPd32wb : VLD1DUPWB<{1,0,0,?}, "32">;
1192 defm VLD1DUPq8wb : VLD1QDUPWB<{0,0,1,0}, "8">;
1193 defm VLD1DUPq16wb : VLD1QDUPWB<{0,1,1,?}, "16">;
1194 defm VLD1DUPq32wb : VLD1QDUPWB<{1,0,1,?}, "32">;
1196 def VLD1DUPq8PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD1dupu>;
1197 def VLD1DUPq16PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD1dupu>;
1198 def VLD1DUPq32PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD1dupu>;
1199 def VLD1DUPq8PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD1dupu>;
1200 def VLD1DUPq16PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD1dupu>;
1201 def VLD1DUPq32PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD1dupu>;
1203 // VLD2DUP : Vector Load (single 2-element structure to all lanes)
1204 class VLD2DUP<bits<4> op7_4, string Dt>
1205 : NLdSt<1, 0b10, 0b1101, op7_4, (outs DPR:$Vd, DPR:$dst2),
1206 (ins addrmode6dup:$Rn), IIC_VLD2dup,
1207 "vld2", Dt, "\\{$Vd[], $dst2[]\\}, $Rn", "", []> {
1209 let Inst{4} = Rn{4};
1210 let DecoderMethod = "DecodeVLD2DupInstruction";
1213 def VLD2DUPd8 : VLD2DUP<{0,0,0,?}, "8">;
1214 def VLD2DUPd16 : VLD2DUP<{0,1,0,?}, "16">;
1215 def VLD2DUPd32 : VLD2DUP<{1,0,0,?}, "32">;
1217 def VLD2DUPd8Pseudo : VLDQPseudo<IIC_VLD2dup>;
1218 def VLD2DUPd16Pseudo : VLDQPseudo<IIC_VLD2dup>;
1219 def VLD2DUPd32Pseudo : VLDQPseudo<IIC_VLD2dup>;
1221 // ...with double-spaced registers (not used for codegen):
1222 def VLD2DUPd8x2 : VLD2DUP<{0,0,1,?}, "8">;
1223 def VLD2DUPd16x2 : VLD2DUP<{0,1,1,?}, "16">;
1224 def VLD2DUPd32x2 : VLD2DUP<{1,0,1,?}, "32">;
1226 // ...with address register writeback:
1227 class VLD2DUPWB<bits<4> op7_4, string Dt>
1228 : NLdSt<1, 0b10, 0b1101, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
1229 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD2dupu,
1230 "vld2", Dt, "\\{$Vd[], $dst2[]\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
1231 let Inst{4} = Rn{4};
1232 let DecoderMethod = "DecodeVLD2DupInstruction";
1235 def VLD2DUPd8_UPD : VLD2DUPWB<{0,0,0,0}, "8">;
1236 def VLD2DUPd16_UPD : VLD2DUPWB<{0,1,0,?}, "16">;
1237 def VLD2DUPd32_UPD : VLD2DUPWB<{1,0,0,?}, "32">;
1239 def VLD2DUPd8x2_UPD : VLD2DUPWB<{0,0,1,0}, "8">;
1240 def VLD2DUPd16x2_UPD : VLD2DUPWB<{0,1,1,?}, "16">;
1241 def VLD2DUPd32x2_UPD : VLD2DUPWB<{1,0,1,?}, "32">;
1243 def VLD2DUPd8Pseudo_UPD : VLDQWBPseudo<IIC_VLD2dupu>;
1244 def VLD2DUPd16Pseudo_UPD : VLDQWBPseudo<IIC_VLD2dupu>;
1245 def VLD2DUPd32Pseudo_UPD : VLDQWBPseudo<IIC_VLD2dupu>;
1247 // VLD3DUP : Vector Load (single 3-element structure to all lanes)
1248 class VLD3DUP<bits<4> op7_4, string Dt>
1249 : NLdSt<1, 0b10, 0b1110, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
1250 (ins addrmode6dup:$Rn), IIC_VLD3dup,
1251 "vld3", Dt, "\\{$Vd[], $dst2[], $dst3[]\\}, $Rn", "", []> {
1254 let DecoderMethod = "DecodeVLD3DupInstruction";
1257 def VLD3DUPd8 : VLD3DUP<{0,0,0,?}, "8">;
1258 def VLD3DUPd16 : VLD3DUP<{0,1,0,?}, "16">;
1259 def VLD3DUPd32 : VLD3DUP<{1,0,0,?}, "32">;
1261 def VLD3DUPd8Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1262 def VLD3DUPd16Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1263 def VLD3DUPd32Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1265 // ...with double-spaced registers (not used for codegen):
1266 def VLD3DUPd8x2 : VLD3DUP<{0,0,1,?}, "8">;
1267 def VLD3DUPd16x2 : VLD3DUP<{0,1,1,?}, "16">;
1268 def VLD3DUPd32x2 : VLD3DUP<{1,0,1,?}, "32">;
1270 // ...with address register writeback:
1271 class VLD3DUPWB<bits<4> op7_4, string Dt>
1272 : NLdSt<1, 0b10, 0b1110, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
1273 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD3dupu,
1274 "vld3", Dt, "\\{$Vd[], $dst2[], $dst3[]\\}, $Rn$Rm",
1275 "$Rn.addr = $wb", []> {
1277 let DecoderMethod = "DecodeVLD3DupInstruction";
1280 def VLD3DUPd8_UPD : VLD3DUPWB<{0,0,0,0}, "8">;
1281 def VLD3DUPd16_UPD : VLD3DUPWB<{0,1,0,?}, "16">;
1282 def VLD3DUPd32_UPD : VLD3DUPWB<{1,0,0,?}, "32">;
1284 def VLD3DUPd8x2_UPD : VLD3DUPWB<{0,0,1,0}, "8">;
1285 def VLD3DUPd16x2_UPD : VLD3DUPWB<{0,1,1,?}, "16">;
1286 def VLD3DUPd32x2_UPD : VLD3DUPWB<{1,0,1,?}, "32">;
1288 def VLD3DUPd8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1289 def VLD3DUPd16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1290 def VLD3DUPd32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1292 // VLD4DUP : Vector Load (single 4-element structure to all lanes)
1293 class VLD4DUP<bits<4> op7_4, string Dt>
1294 : NLdSt<1, 0b10, 0b1111, op7_4,
1295 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
1296 (ins addrmode6dup:$Rn), IIC_VLD4dup,
1297 "vld4", Dt, "\\{$Vd[], $dst2[], $dst3[], $dst4[]\\}, $Rn", "", []> {
1299 let Inst{4} = Rn{4};
1300 let DecoderMethod = "DecodeVLD4DupInstruction";
1303 def VLD4DUPd8 : VLD4DUP<{0,0,0,?}, "8">;
1304 def VLD4DUPd16 : VLD4DUP<{0,1,0,?}, "16">;
1305 def VLD4DUPd32 : VLD4DUP<{1,?,0,?}, "32"> { let Inst{6} = Rn{5}; }
1307 def VLD4DUPd8Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1308 def VLD4DUPd16Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1309 def VLD4DUPd32Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1311 // ...with double-spaced registers (not used for codegen):
1312 def VLD4DUPd8x2 : VLD4DUP<{0,0,1,?}, "8">;
1313 def VLD4DUPd16x2 : VLD4DUP<{0,1,1,?}, "16">;
1314 def VLD4DUPd32x2 : VLD4DUP<{1,?,1,?}, "32"> { let Inst{6} = Rn{5}; }
1316 // ...with address register writeback:
1317 class VLD4DUPWB<bits<4> op7_4, string Dt>
1318 : NLdSt<1, 0b10, 0b1111, op7_4,
1319 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
1320 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD4dupu,
1321 "vld4", Dt, "\\{$Vd[], $dst2[], $dst3[], $dst4[]\\}, $Rn$Rm",
1322 "$Rn.addr = $wb", []> {
1323 let Inst{4} = Rn{4};
1324 let DecoderMethod = "DecodeVLD4DupInstruction";
1327 def VLD4DUPd8_UPD : VLD4DUPWB<{0,0,0,0}, "8">;
1328 def VLD4DUPd16_UPD : VLD4DUPWB<{0,1,0,?}, "16">;
1329 def VLD4DUPd32_UPD : VLD4DUPWB<{1,?,0,?}, "32"> { let Inst{6} = Rn{5}; }
1331 def VLD4DUPd8x2_UPD : VLD4DUPWB<{0,0,1,0}, "8">;
1332 def VLD4DUPd16x2_UPD : VLD4DUPWB<{0,1,1,?}, "16">;
1333 def VLD4DUPd32x2_UPD : VLD4DUPWB<{1,?,1,?}, "32"> { let Inst{6} = Rn{5}; }
1335 def VLD4DUPd8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1336 def VLD4DUPd16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1337 def VLD4DUPd32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1339 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
1341 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
1343 // Classes for VST* pseudo-instructions with multi-register operands.
1344 // These are expanded to real instructions after register allocation.
1345 class VSTQPseudo<InstrItinClass itin>
1346 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src), itin, "">;
1347 class VSTQWBPseudo<InstrItinClass itin>
1348 : PseudoNLdSt<(outs GPR:$wb),
1349 (ins addrmode6:$addr, am6offset:$offset, QPR:$src), itin,
1350 "$addr.addr = $wb">;
1351 class VSTQWBfixedPseudo<InstrItinClass itin>
1352 : PseudoNLdSt<(outs GPR:$wb),
1353 (ins addrmode6:$addr, QPR:$src), itin,
1354 "$addr.addr = $wb">;
1355 class VSTQWBregisterPseudo<InstrItinClass itin>
1356 : PseudoNLdSt<(outs GPR:$wb),
1357 (ins addrmode6:$addr, rGPR:$offset, QPR:$src), itin,
1358 "$addr.addr = $wb">;
1359 class VSTQQPseudo<InstrItinClass itin>
1360 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src), itin, "">;
1361 class VSTQQWBPseudo<InstrItinClass itin>
1362 : PseudoNLdSt<(outs GPR:$wb),
1363 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src), itin,
1364 "$addr.addr = $wb">;
1365 class VSTQQQQPseudo<InstrItinClass itin>
1366 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src), itin, "">;
1367 class VSTQQQQWBPseudo<InstrItinClass itin>
1368 : PseudoNLdSt<(outs GPR:$wb),
1369 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
1370 "$addr.addr = $wb">;
1372 // VST1 : Vector Store (multiple single elements)
1373 class VST1D<bits<4> op7_4, string Dt>
1374 : NLdSt<0,0b00,0b0111,op7_4, (outs), (ins addrmode6:$Rn, VecListOneD:$Vd),
1375 IIC_VST1, "vst1", Dt, "$Vd, $Rn", "", []> {
1377 let Inst{4} = Rn{4};
1378 let DecoderMethod = "DecodeVSTInstruction";
1380 class VST1Q<bits<4> op7_4, string Dt>
1381 : NLdSt<0,0b00,0b1010,op7_4, (outs), (ins addrmode6:$Rn, VecListTwoD:$Vd),
1382 IIC_VST1x2, "vst1", Dt, "$Vd, $Rn", "", []> {
1384 let Inst{5-4} = Rn{5-4};
1385 let DecoderMethod = "DecodeVSTInstruction";
1388 def VST1d8 : VST1D<{0,0,0,?}, "8">;
1389 def VST1d16 : VST1D<{0,1,0,?}, "16">;
1390 def VST1d32 : VST1D<{1,0,0,?}, "32">;
1391 def VST1d64 : VST1D<{1,1,0,?}, "64">;
1393 def VST1q8 : VST1Q<{0,0,?,?}, "8">;
1394 def VST1q16 : VST1Q<{0,1,?,?}, "16">;
1395 def VST1q32 : VST1Q<{1,0,?,?}, "32">;
1396 def VST1q64 : VST1Q<{1,1,?,?}, "64">;
1398 def VST1q8Pseudo : VSTQPseudo<IIC_VST1x2>;
1399 def VST1q16Pseudo : VSTQPseudo<IIC_VST1x2>;
1400 def VST1q32Pseudo : VSTQPseudo<IIC_VST1x2>;
1401 def VST1q64Pseudo : VSTQPseudo<IIC_VST1x2>;
1403 // ...with address register writeback:
1404 multiclass VST1DWB<bits<4> op7_4, string Dt> {
1405 def _fixed : NLdSt<0,0b00, 0b0111,op7_4, (outs GPR:$wb),
1406 (ins addrmode6:$Rn, VecListOneD:$Vd), IIC_VLD1u,
1407 "vst1", Dt, "$Vd, $Rn!",
1408 "$Rn.addr = $wb", []> {
1409 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1410 let Inst{4} = Rn{4};
1411 let DecoderMethod = "DecodeVSTInstruction";
1412 let AsmMatchConverter = "cvtVSTwbFixed";
1414 def _register : NLdSt<0,0b00,0b0111,op7_4, (outs GPR:$wb),
1415 (ins addrmode6:$Rn, rGPR:$Rm, VecListOneD:$Vd),
1417 "vst1", Dt, "$Vd, $Rn, $Rm",
1418 "$Rn.addr = $wb", []> {
1419 let Inst{4} = Rn{4};
1420 let DecoderMethod = "DecodeVSTInstruction";
1421 let AsmMatchConverter = "cvtVSTwbRegister";
1424 multiclass VST1QWB<bits<4> op7_4, string Dt> {
1425 def _fixed : NLdSt<0,0b00,0b1010,op7_4, (outs GPR:$wb),
1426 (ins addrmode6:$Rn, VecListTwoD:$Vd), IIC_VLD1x2u,
1427 "vst1", Dt, "$Vd, $Rn!",
1428 "$Rn.addr = $wb", []> {
1429 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1430 let Inst{5-4} = Rn{5-4};
1431 let DecoderMethod = "DecodeVSTInstruction";
1432 let AsmMatchConverter = "cvtVSTwbFixed";
1434 def _register : NLdSt<0,0b00,0b1010,op7_4, (outs GPR:$wb),
1435 (ins addrmode6:$Rn, rGPR:$Rm, VecListTwoD:$Vd),
1437 "vst1", Dt, "$Vd, $Rn, $Rm",
1438 "$Rn.addr = $wb", []> {
1439 let Inst{5-4} = Rn{5-4};
1440 let DecoderMethod = "DecodeVSTInstruction";
1441 let AsmMatchConverter = "cvtVSTwbRegister";
1445 defm VST1d8wb : VST1DWB<{0,0,0,?}, "8">;
1446 defm VST1d16wb : VST1DWB<{0,1,0,?}, "16">;
1447 defm VST1d32wb : VST1DWB<{1,0,0,?}, "32">;
1448 defm VST1d64wb : VST1DWB<{1,1,0,?}, "64">;
1450 defm VST1q8wb : VST1QWB<{0,0,?,?}, "8">;
1451 defm VST1q16wb : VST1QWB<{0,1,?,?}, "16">;
1452 defm VST1q32wb : VST1QWB<{1,0,?,?}, "32">;
1453 defm VST1q64wb : VST1QWB<{1,1,?,?}, "64">;
1455 def VST1q8PseudoWB_fixed : VSTQWBfixedPseudo<IIC_VST1x2u>;
1456 def VST1q16PseudoWB_fixed : VSTQWBfixedPseudo<IIC_VST1x2u>;
1457 def VST1q32PseudoWB_fixed : VSTQWBfixedPseudo<IIC_VST1x2u>;
1458 def VST1q64PseudoWB_fixed : VSTQWBfixedPseudo<IIC_VST1x2u>;
1459 def VST1q8PseudoWB_register : VSTQWBregisterPseudo<IIC_VST1x2u>;
1460 def VST1q16PseudoWB_register : VSTQWBregisterPseudo<IIC_VST1x2u>;
1461 def VST1q32PseudoWB_register : VSTQWBregisterPseudo<IIC_VST1x2u>;
1462 def VST1q64PseudoWB_register : VSTQWBregisterPseudo<IIC_VST1x2u>;
1464 // ...with 3 registers
1465 class VST1D3<bits<4> op7_4, string Dt>
1466 : NLdSt<0, 0b00, 0b0110, op7_4, (outs),
1467 (ins addrmode6:$Rn, VecListThreeD:$Vd),
1468 IIC_VST1x3, "vst1", Dt, "$Vd, $Rn", "", []> {
1470 let Inst{4} = Rn{4};
1471 let DecoderMethod = "DecodeVSTInstruction";
1473 multiclass VST1D3WB<bits<4> op7_4, string Dt> {
1474 def _fixed : NLdSt<0,0b00,0b0110,op7_4, (outs GPR:$wb),
1475 (ins addrmode6:$Rn, VecListThreeD:$Vd), IIC_VLD1x3u,
1476 "vst1", Dt, "$Vd, $Rn!",
1477 "$Rn.addr = $wb", []> {
1478 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1479 let Inst{5-4} = Rn{5-4};
1480 let DecoderMethod = "DecodeVSTInstruction";
1481 let AsmMatchConverter = "cvtVSTwbFixed";
1483 def _register : NLdSt<0,0b00,0b0110,op7_4, (outs GPR:$wb),
1484 (ins addrmode6:$Rn, rGPR:$Rm, VecListThreeD:$Vd),
1486 "vst1", Dt, "$Vd, $Rn, $Rm",
1487 "$Rn.addr = $wb", []> {
1488 let Inst{5-4} = Rn{5-4};
1489 let DecoderMethod = "DecodeVSTInstruction";
1490 let AsmMatchConverter = "cvtVSTwbRegister";
1494 def VST1d8T : VST1D3<{0,0,0,?}, "8">;
1495 def VST1d16T : VST1D3<{0,1,0,?}, "16">;
1496 def VST1d32T : VST1D3<{1,0,0,?}, "32">;
1497 def VST1d64T : VST1D3<{1,1,0,?}, "64">;
1499 defm VST1d8Twb : VST1D3WB<{0,0,0,?}, "8">;
1500 defm VST1d16Twb : VST1D3WB<{0,1,0,?}, "16">;
1501 defm VST1d32Twb : VST1D3WB<{1,0,0,?}, "32">;
1502 defm VST1d64Twb : VST1D3WB<{1,1,0,?}, "64">;
1504 def VST1d64TPseudo : VSTQQPseudo<IIC_VST1x3>;
1505 def VST1d64TPseudoWB_fixed : VSTQQWBPseudo<IIC_VST1x3u>;
1506 def VST1d64TPseudoWB_register : VSTQQWBPseudo<IIC_VST1x3u>;
1508 // ...with 4 registers
1509 class VST1D4<bits<4> op7_4, string Dt>
1510 : NLdSt<0, 0b00, 0b0010, op7_4, (outs),
1511 (ins addrmode6:$Rn, VecListFourD:$Vd),
1512 IIC_VST1x4, "vst1", Dt, "$Vd, $Rn", "",
1515 let Inst{5-4} = Rn{5-4};
1516 let DecoderMethod = "DecodeVSTInstruction";
1518 multiclass VST1D4WB<bits<4> op7_4, string Dt> {
1519 def _fixed : NLdSt<0,0b00,0b0010,op7_4, (outs GPR:$wb),
1520 (ins addrmode6:$Rn, VecListFourD:$Vd), IIC_VLD1x4u,
1521 "vst1", Dt, "$Vd, $Rn!",
1522 "$Rn.addr = $wb", []> {
1523 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1524 let Inst{5-4} = Rn{5-4};
1525 let DecoderMethod = "DecodeVSTInstruction";
1526 let AsmMatchConverter = "cvtVSTwbFixed";
1528 def _register : NLdSt<0,0b00,0b0010,op7_4, (outs GPR:$wb),
1529 (ins addrmode6:$Rn, rGPR:$Rm, VecListFourD:$Vd),
1531 "vst1", Dt, "$Vd, $Rn, $Rm",
1532 "$Rn.addr = $wb", []> {
1533 let Inst{5-4} = Rn{5-4};
1534 let DecoderMethod = "DecodeVSTInstruction";
1535 let AsmMatchConverter = "cvtVSTwbRegister";
1539 def VST1d8Q : VST1D4<{0,0,?,?}, "8">;
1540 def VST1d16Q : VST1D4<{0,1,?,?}, "16">;
1541 def VST1d32Q : VST1D4<{1,0,?,?}, "32">;
1542 def VST1d64Q : VST1D4<{1,1,?,?}, "64">;
1544 defm VST1d8Qwb : VST1D4WB<{0,0,?,?}, "8">;
1545 defm VST1d16Qwb : VST1D4WB<{0,1,?,?}, "16">;
1546 defm VST1d32Qwb : VST1D4WB<{1,0,?,?}, "32">;
1547 defm VST1d64Qwb : VST1D4WB<{1,1,?,?}, "64">;
1549 def VST1d64QPseudo : VSTQQPseudo<IIC_VST1x4>;
1550 def VST1d64QPseudoWB_fixed : VSTQQWBPseudo<IIC_VST1x4u>;
1551 def VST1d64QPseudoWB_register : VSTQQWBPseudo<IIC_VST1x4u>;
1553 // VST2 : Vector Store (multiple 2-element structures)
1554 class VST2<bits<4> op11_8, bits<4> op7_4, string Dt, RegisterOperand VdTy,
1555 InstrItinClass itin>
1556 : NLdSt<0, 0b00, op11_8, op7_4, (outs), (ins addrmode6:$Rn, VdTy:$Vd),
1557 itin, "vst2", Dt, "$Vd, $Rn", "", []> {
1559 let Inst{5-4} = Rn{5-4};
1560 let DecoderMethod = "DecodeVSTInstruction";
1563 def VST2d8 : VST2<0b1000, {0,0,?,?}, "8", VecListTwoD, IIC_VST2>;
1564 def VST2d16 : VST2<0b1000, {0,1,?,?}, "16", VecListTwoD, IIC_VST2>;
1565 def VST2d32 : VST2<0b1000, {1,0,?,?}, "32", VecListTwoD, IIC_VST2>;
1567 def VST2q8 : VST2<0b0011, {0,0,?,?}, "8", VecListFourD, IIC_VST2x2>;
1568 def VST2q16 : VST2<0b0011, {0,1,?,?}, "16", VecListFourD, IIC_VST2x2>;
1569 def VST2q32 : VST2<0b0011, {1,0,?,?}, "32", VecListFourD, IIC_VST2x2>;
1571 def VST2d8Pseudo : VSTQPseudo<IIC_VST2>;
1572 def VST2d16Pseudo : VSTQPseudo<IIC_VST2>;
1573 def VST2d32Pseudo : VSTQPseudo<IIC_VST2>;
1575 def VST2q8Pseudo : VSTQQPseudo<IIC_VST2x2>;
1576 def VST2q16Pseudo : VSTQQPseudo<IIC_VST2x2>;
1577 def VST2q32Pseudo : VSTQQPseudo<IIC_VST2x2>;
1579 // ...with address register writeback:
1580 multiclass VST2DWB<bits<4> op11_8, bits<4> op7_4, string Dt,
1581 RegisterOperand VdTy> {
1582 def _fixed : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1583 (ins addrmode6:$Rn, VdTy:$Vd), IIC_VLD1u,
1584 "vst2", Dt, "$Vd, $Rn!",
1585 "$Rn.addr = $wb", []> {
1586 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1587 let Inst{5-4} = Rn{5-4};
1588 let DecoderMethod = "DecodeVSTInstruction";
1589 let AsmMatchConverter = "cvtVSTwbFixed";
1591 def _register : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1592 (ins addrmode6:$Rn, rGPR:$Rm, VdTy:$Vd), IIC_VLD1u,
1593 "vst2", Dt, "$Vd, $Rn, $Rm",
1594 "$Rn.addr = $wb", []> {
1595 let Inst{5-4} = Rn{5-4};
1596 let DecoderMethod = "DecodeVSTInstruction";
1597 let AsmMatchConverter = "cvtVSTwbRegister";
1600 multiclass VST2QWB<bits<4> op7_4, string Dt> {
1601 def _fixed : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
1602 (ins addrmode6:$Rn, VecListFourD:$Vd), IIC_VLD1u,
1603 "vst2", Dt, "$Vd, $Rn!",
1604 "$Rn.addr = $wb", []> {
1605 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1606 let Inst{5-4} = Rn{5-4};
1607 let DecoderMethod = "DecodeVSTInstruction";
1608 let AsmMatchConverter = "cvtVSTwbFixed";
1610 def _register : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
1611 (ins addrmode6:$Rn, rGPR:$Rm, VecListFourD:$Vd),
1613 "vst2", Dt, "$Vd, $Rn, $Rm",
1614 "$Rn.addr = $wb", []> {
1615 let Inst{5-4} = Rn{5-4};
1616 let DecoderMethod = "DecodeVSTInstruction";
1617 let AsmMatchConverter = "cvtVSTwbRegister";
1621 defm VST2d8wb : VST2DWB<0b1000, {0,0,?,?}, "8", VecListTwoD>;
1622 defm VST2d16wb : VST2DWB<0b1000, {0,1,?,?}, "16", VecListTwoD>;
1623 defm VST2d32wb : VST2DWB<0b1000, {1,0,?,?}, "32", VecListTwoD>;
1625 defm VST2q8wb : VST2QWB<{0,0,?,?}, "8">;
1626 defm VST2q16wb : VST2QWB<{0,1,?,?}, "16">;
1627 defm VST2q32wb : VST2QWB<{1,0,?,?}, "32">;
1629 def VST2d8PseudoWB_fixed : VSTQWBPseudo<IIC_VST2u>;
1630 def VST2d16PseudoWB_fixed : VSTQWBPseudo<IIC_VST2u>;
1631 def VST2d32PseudoWB_fixed : VSTQWBPseudo<IIC_VST2u>;
1632 def VST2d8PseudoWB_register : VSTQWBPseudo<IIC_VST2u>;
1633 def VST2d16PseudoWB_register : VSTQWBPseudo<IIC_VST2u>;
1634 def VST2d32PseudoWB_register : VSTQWBPseudo<IIC_VST2u>;
1636 def VST2q8PseudoWB_fixed : VSTQQWBPseudo<IIC_VST2x2u>;
1637 def VST2q16PseudoWB_fixed : VSTQQWBPseudo<IIC_VST2x2u>;
1638 def VST2q32PseudoWB_fixed : VSTQQWBPseudo<IIC_VST2x2u>;
1639 def VST2q8PseudoWB_register : VSTQQWBPseudo<IIC_VST2x2u>;
1640 def VST2q16PseudoWB_register : VSTQQWBPseudo<IIC_VST2x2u>;
1641 def VST2q32PseudoWB_register : VSTQQWBPseudo<IIC_VST2x2u>;
1643 // ...with double-spaced registers
1644 def VST2b8 : VST2<0b1001, {0,0,?,?}, "8", VecListTwoQ, IIC_VST2>;
1645 def VST2b16 : VST2<0b1001, {0,1,?,?}, "16", VecListTwoQ, IIC_VST2>;
1646 def VST2b32 : VST2<0b1001, {1,0,?,?}, "32", VecListTwoQ, IIC_VST2>;
1647 defm VST2b8wb : VST2DWB<0b1001, {0,0,?,?}, "8", VecListTwoQ>;
1648 defm VST2b16wb : VST2DWB<0b1001, {0,1,?,?}, "16", VecListTwoQ>;
1649 defm VST2b32wb : VST2DWB<0b1001, {1,0,?,?}, "32", VecListTwoQ>;
1651 // VST3 : Vector Store (multiple 3-element structures)
1652 class VST3D<bits<4> op11_8, bits<4> op7_4, string Dt>
1653 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1654 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3,
1655 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn", "", []> {
1657 let Inst{4} = Rn{4};
1658 let DecoderMethod = "DecodeVSTInstruction";
1661 def VST3d8 : VST3D<0b0100, {0,0,0,?}, "8">;
1662 def VST3d16 : VST3D<0b0100, {0,1,0,?}, "16">;
1663 def VST3d32 : VST3D<0b0100, {1,0,0,?}, "32">;
1665 def VST3d8Pseudo : VSTQQPseudo<IIC_VST3>;
1666 def VST3d16Pseudo : VSTQQPseudo<IIC_VST3>;
1667 def VST3d32Pseudo : VSTQQPseudo<IIC_VST3>;
1669 // ...with address register writeback:
1670 class VST3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1671 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1672 (ins addrmode6:$Rn, am6offset:$Rm,
1673 DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3u,
1674 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn$Rm",
1675 "$Rn.addr = $wb", []> {
1676 let Inst{4} = Rn{4};
1677 let DecoderMethod = "DecodeVSTInstruction";
1680 def VST3d8_UPD : VST3DWB<0b0100, {0,0,0,?}, "8">;
1681 def VST3d16_UPD : VST3DWB<0b0100, {0,1,0,?}, "16">;
1682 def VST3d32_UPD : VST3DWB<0b0100, {1,0,0,?}, "32">;
1684 def VST3d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1685 def VST3d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1686 def VST3d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1688 // ...with double-spaced registers:
1689 def VST3q8 : VST3D<0b0101, {0,0,0,?}, "8">;
1690 def VST3q16 : VST3D<0b0101, {0,1,0,?}, "16">;
1691 def VST3q32 : VST3D<0b0101, {1,0,0,?}, "32">;
1692 def VST3q8_UPD : VST3DWB<0b0101, {0,0,0,?}, "8">;
1693 def VST3q16_UPD : VST3DWB<0b0101, {0,1,0,?}, "16">;
1694 def VST3q32_UPD : VST3DWB<0b0101, {1,0,0,?}, "32">;
1696 def VST3q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1697 def VST3q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1698 def VST3q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1700 // ...alternate versions to be allocated odd register numbers:
1701 def VST3q8oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1702 def VST3q16oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1703 def VST3q32oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1705 def VST3q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1706 def VST3q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1707 def VST3q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1709 // VST4 : Vector Store (multiple 4-element structures)
1710 class VST4D<bits<4> op11_8, bits<4> op7_4, string Dt>
1711 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1712 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
1713 IIC_VST4, "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn",
1716 let Inst{5-4} = Rn{5-4};
1717 let DecoderMethod = "DecodeVSTInstruction";
1720 def VST4d8 : VST4D<0b0000, {0,0,?,?}, "8">;
1721 def VST4d16 : VST4D<0b0000, {0,1,?,?}, "16">;
1722 def VST4d32 : VST4D<0b0000, {1,0,?,?}, "32">;
1724 def VST4d8Pseudo : VSTQQPseudo<IIC_VST4>;
1725 def VST4d16Pseudo : VSTQQPseudo<IIC_VST4>;
1726 def VST4d32Pseudo : VSTQQPseudo<IIC_VST4>;
1728 // ...with address register writeback:
1729 class VST4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1730 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1731 (ins addrmode6:$Rn, am6offset:$Rm,
1732 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST4u,
1733 "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
1734 "$Rn.addr = $wb", []> {
1735 let Inst{5-4} = Rn{5-4};
1736 let DecoderMethod = "DecodeVSTInstruction";
1739 def VST4d8_UPD : VST4DWB<0b0000, {0,0,?,?}, "8">;
1740 def VST4d16_UPD : VST4DWB<0b0000, {0,1,?,?}, "16">;
1741 def VST4d32_UPD : VST4DWB<0b0000, {1,0,?,?}, "32">;
1743 def VST4d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1744 def VST4d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1745 def VST4d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1747 // ...with double-spaced registers:
1748 def VST4q8 : VST4D<0b0001, {0,0,?,?}, "8">;
1749 def VST4q16 : VST4D<0b0001, {0,1,?,?}, "16">;
1750 def VST4q32 : VST4D<0b0001, {1,0,?,?}, "32">;
1751 def VST4q8_UPD : VST4DWB<0b0001, {0,0,?,?}, "8">;
1752 def VST4q16_UPD : VST4DWB<0b0001, {0,1,?,?}, "16">;
1753 def VST4q32_UPD : VST4DWB<0b0001, {1,0,?,?}, "32">;
1755 def VST4q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1756 def VST4q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1757 def VST4q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1759 // ...alternate versions to be allocated odd register numbers:
1760 def VST4q8oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1761 def VST4q16oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1762 def VST4q32oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1764 def VST4q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1765 def VST4q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1766 def VST4q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1768 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
1770 // Classes for VST*LN pseudo-instructions with multi-register operands.
1771 // These are expanded to real instructions after register allocation.
1772 class VSTQLNPseudo<InstrItinClass itin>
1773 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
1775 class VSTQLNWBPseudo<InstrItinClass itin>
1776 : PseudoNLdSt<(outs GPR:$wb),
1777 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
1778 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1779 class VSTQQLNPseudo<InstrItinClass itin>
1780 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
1782 class VSTQQLNWBPseudo<InstrItinClass itin>
1783 : PseudoNLdSt<(outs GPR:$wb),
1784 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
1785 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1786 class VSTQQQQLNPseudo<InstrItinClass itin>
1787 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
1789 class VSTQQQQLNWBPseudo<InstrItinClass itin>
1790 : PseudoNLdSt<(outs GPR:$wb),
1791 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
1792 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1794 // VST1LN : Vector Store (single element from one lane)
1795 class VST1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1796 PatFrag StoreOp, SDNode ExtractOp>
1797 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1798 (ins addrmode6:$Rn, DPR:$Vd, nohash_imm:$lane),
1799 IIC_VST1ln, "vst1", Dt, "\\{$Vd[$lane]\\}, $Rn", "",
1800 [(StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane), addrmode6:$Rn)]> {
1802 let DecoderMethod = "DecodeVST1LN";
1804 class VST1LN32<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1805 PatFrag StoreOp, SDNode ExtractOp>
1806 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1807 (ins addrmode6oneL32:$Rn, DPR:$Vd, nohash_imm:$lane),
1808 IIC_VST1ln, "vst1", Dt, "\\{$Vd[$lane]\\}, $Rn", "",
1809 [(StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane), addrmode6oneL32:$Rn)]>{
1811 let DecoderMethod = "DecodeVST1LN";
1813 class VST1QLNPseudo<ValueType Ty, PatFrag StoreOp, SDNode ExtractOp>
1814 : VSTQLNPseudo<IIC_VST1ln> {
1815 let Pattern = [(StoreOp (ExtractOp (Ty QPR:$src), imm:$lane),
1819 def VST1LNd8 : VST1LN<0b0000, {?,?,?,0}, "8", v8i8, truncstorei8,
1821 let Inst{7-5} = lane{2-0};
1823 def VST1LNd16 : VST1LN<0b0100, {?,?,0,?}, "16", v4i16, truncstorei16,
1825 let Inst{7-6} = lane{1-0};
1826 let Inst{4} = Rn{5};
1829 def VST1LNd32 : VST1LN32<0b1000, {?,0,?,?}, "32", v2i32, store, extractelt> {
1830 let Inst{7} = lane{0};
1831 let Inst{5-4} = Rn{5-4};
1834 def VST1LNq8Pseudo : VST1QLNPseudo<v16i8, truncstorei8, NEONvgetlaneu>;
1835 def VST1LNq16Pseudo : VST1QLNPseudo<v8i16, truncstorei16, NEONvgetlaneu>;
1836 def VST1LNq32Pseudo : VST1QLNPseudo<v4i32, store, extractelt>;
1838 def : Pat<(store (extractelt (v2f32 DPR:$src), imm:$lane), addrmode6:$addr),
1839 (VST1LNd32 addrmode6:$addr, DPR:$src, imm:$lane)>;
1840 def : Pat<(store (extractelt (v4f32 QPR:$src), imm:$lane), addrmode6:$addr),
1841 (VST1LNq32Pseudo addrmode6:$addr, QPR:$src, imm:$lane)>;
1843 // ...with address register writeback:
1844 class VST1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1845 PatFrag StoreOp, SDNode ExtractOp>
1846 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1847 (ins addrmode6:$Rn, am6offset:$Rm,
1848 DPR:$Vd, nohash_imm:$lane), IIC_VST1lnu, "vst1", Dt,
1849 "\\{$Vd[$lane]\\}, $Rn$Rm",
1851 [(set GPR:$wb, (StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane),
1852 addrmode6:$Rn, am6offset:$Rm))]> {
1853 let DecoderMethod = "DecodeVST1LN";
1855 class VST1QLNWBPseudo<ValueType Ty, PatFrag StoreOp, SDNode ExtractOp>
1856 : VSTQLNWBPseudo<IIC_VST1lnu> {
1857 let Pattern = [(set GPR:$wb, (StoreOp (ExtractOp (Ty QPR:$src), imm:$lane),
1858 addrmode6:$addr, am6offset:$offset))];
1861 def VST1LNd8_UPD : VST1LNWB<0b0000, {?,?,?,0}, "8", v8i8, post_truncsti8,
1863 let Inst{7-5} = lane{2-0};
1865 def VST1LNd16_UPD : VST1LNWB<0b0100, {?,?,0,?}, "16", v4i16, post_truncsti16,
1867 let Inst{7-6} = lane{1-0};
1868 let Inst{4} = Rn{5};
1870 def VST1LNd32_UPD : VST1LNWB<0b1000, {?,0,?,?}, "32", v2i32, post_store,
1872 let Inst{7} = lane{0};
1873 let Inst{5-4} = Rn{5-4};
1876 def VST1LNq8Pseudo_UPD : VST1QLNWBPseudo<v16i8, post_truncsti8, NEONvgetlaneu>;
1877 def VST1LNq16Pseudo_UPD : VST1QLNWBPseudo<v8i16, post_truncsti16,NEONvgetlaneu>;
1878 def VST1LNq32Pseudo_UPD : VST1QLNWBPseudo<v4i32, post_store, extractelt>;
1880 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
1882 // VST2LN : Vector Store (single 2-element structure from one lane)
1883 class VST2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1884 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1885 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, nohash_imm:$lane),
1886 IIC_VST2ln, "vst2", Dt, "\\{$Vd[$lane], $src2[$lane]\\}, $Rn",
1889 let Inst{4} = Rn{4};
1890 let DecoderMethod = "DecodeVST2LN";
1893 def VST2LNd8 : VST2LN<0b0001, {?,?,?,?}, "8"> {
1894 let Inst{7-5} = lane{2-0};
1896 def VST2LNd16 : VST2LN<0b0101, {?,?,0,?}, "16"> {
1897 let Inst{7-6} = lane{1-0};
1899 def VST2LNd32 : VST2LN<0b1001, {?,0,0,?}, "32"> {
1900 let Inst{7} = lane{0};
1903 def VST2LNd8Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1904 def VST2LNd16Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1905 def VST2LNd32Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1907 // ...with double-spaced registers:
1908 def VST2LNq16 : VST2LN<0b0101, {?,?,1,?}, "16"> {
1909 let Inst{7-6} = lane{1-0};
1910 let Inst{4} = Rn{4};
1912 def VST2LNq32 : VST2LN<0b1001, {?,1,0,?}, "32"> {
1913 let Inst{7} = lane{0};
1914 let Inst{4} = Rn{4};
1917 def VST2LNq16Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
1918 def VST2LNq32Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
1920 // ...with address register writeback:
1921 class VST2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1922 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1923 (ins addrmode6:$Rn, am6offset:$Rm,
1924 DPR:$Vd, DPR:$src2, nohash_imm:$lane), IIC_VST2lnu, "vst2", Dt,
1925 "\\{$Vd[$lane], $src2[$lane]\\}, $Rn$Rm",
1926 "$Rn.addr = $wb", []> {
1927 let Inst{4} = Rn{4};
1928 let DecoderMethod = "DecodeVST2LN";
1931 def VST2LNd8_UPD : VST2LNWB<0b0001, {?,?,?,?}, "8"> {
1932 let Inst{7-5} = lane{2-0};
1934 def VST2LNd16_UPD : VST2LNWB<0b0101, {?,?,0,?}, "16"> {
1935 let Inst{7-6} = lane{1-0};
1937 def VST2LNd32_UPD : VST2LNWB<0b1001, {?,0,0,?}, "32"> {
1938 let Inst{7} = lane{0};
1941 def VST2LNd8Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1942 def VST2LNd16Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1943 def VST2LNd32Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1945 def VST2LNq16_UPD : VST2LNWB<0b0101, {?,?,1,?}, "16"> {
1946 let Inst{7-6} = lane{1-0};
1948 def VST2LNq32_UPD : VST2LNWB<0b1001, {?,1,0,?}, "32"> {
1949 let Inst{7} = lane{0};
1952 def VST2LNq16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
1953 def VST2LNq32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
1955 // VST3LN : Vector Store (single 3-element structure from one lane)
1956 class VST3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1957 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1958 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3,
1959 nohash_imm:$lane), IIC_VST3ln, "vst3", Dt,
1960 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn", "", []> {
1962 let DecoderMethod = "DecodeVST3LN";
1965 def VST3LNd8 : VST3LN<0b0010, {?,?,?,0}, "8"> {
1966 let Inst{7-5} = lane{2-0};
1968 def VST3LNd16 : VST3LN<0b0110, {?,?,0,0}, "16"> {
1969 let Inst{7-6} = lane{1-0};
1971 def VST3LNd32 : VST3LN<0b1010, {?,0,0,0}, "32"> {
1972 let Inst{7} = lane{0};
1975 def VST3LNd8Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1976 def VST3LNd16Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1977 def VST3LNd32Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1979 // ...with double-spaced registers:
1980 def VST3LNq16 : VST3LN<0b0110, {?,?,1,0}, "16"> {
1981 let Inst{7-6} = lane{1-0};
1983 def VST3LNq32 : VST3LN<0b1010, {?,1,0,0}, "32"> {
1984 let Inst{7} = lane{0};
1987 def VST3LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
1988 def VST3LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
1990 // ...with address register writeback:
1991 class VST3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1992 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1993 (ins addrmode6:$Rn, am6offset:$Rm,
1994 DPR:$Vd, DPR:$src2, DPR:$src3, nohash_imm:$lane),
1995 IIC_VST3lnu, "vst3", Dt,
1996 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn$Rm",
1997 "$Rn.addr = $wb", []> {
1998 let DecoderMethod = "DecodeVST3LN";
2001 def VST3LNd8_UPD : VST3LNWB<0b0010, {?,?,?,0}, "8"> {
2002 let Inst{7-5} = lane{2-0};
2004 def VST3LNd16_UPD : VST3LNWB<0b0110, {?,?,0,0}, "16"> {
2005 let Inst{7-6} = lane{1-0};
2007 def VST3LNd32_UPD : VST3LNWB<0b1010, {?,0,0,0}, "32"> {
2008 let Inst{7} = lane{0};
2011 def VST3LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
2012 def VST3LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
2013 def VST3LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
2015 def VST3LNq16_UPD : VST3LNWB<0b0110, {?,?,1,0}, "16"> {
2016 let Inst{7-6} = lane{1-0};
2018 def VST3LNq32_UPD : VST3LNWB<0b1010, {?,1,0,0}, "32"> {
2019 let Inst{7} = lane{0};
2022 def VST3LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
2023 def VST3LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
2025 // VST4LN : Vector Store (single 4-element structure from one lane)
2026 class VST4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
2027 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
2028 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4,
2029 nohash_imm:$lane), IIC_VST4ln, "vst4", Dt,
2030 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn",
2033 let Inst{4} = Rn{4};
2034 let DecoderMethod = "DecodeVST4LN";
2037 def VST4LNd8 : VST4LN<0b0011, {?,?,?,?}, "8"> {
2038 let Inst{7-5} = lane{2-0};
2040 def VST4LNd16 : VST4LN<0b0111, {?,?,0,?}, "16"> {
2041 let Inst{7-6} = lane{1-0};
2043 def VST4LNd32 : VST4LN<0b1011, {?,0,?,?}, "32"> {
2044 let Inst{7} = lane{0};
2045 let Inst{5} = Rn{5};
2048 def VST4LNd8Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
2049 def VST4LNd16Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
2050 def VST4LNd32Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
2052 // ...with double-spaced registers:
2053 def VST4LNq16 : VST4LN<0b0111, {?,?,1,?}, "16"> {
2054 let Inst{7-6} = lane{1-0};
2056 def VST4LNq32 : VST4LN<0b1011, {?,1,?,?}, "32"> {
2057 let Inst{7} = lane{0};
2058 let Inst{5} = Rn{5};
2061 def VST4LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
2062 def VST4LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
2064 // ...with address register writeback:
2065 class VST4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
2066 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
2067 (ins addrmode6:$Rn, am6offset:$Rm,
2068 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
2069 IIC_VST4lnu, "vst4", Dt,
2070 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn$Rm",
2071 "$Rn.addr = $wb", []> {
2072 let Inst{4} = Rn{4};
2073 let DecoderMethod = "DecodeVST4LN";
2076 def VST4LNd8_UPD : VST4LNWB<0b0011, {?,?,?,?}, "8"> {
2077 let Inst{7-5} = lane{2-0};
2079 def VST4LNd16_UPD : VST4LNWB<0b0111, {?,?,0,?}, "16"> {
2080 let Inst{7-6} = lane{1-0};
2082 def VST4LNd32_UPD : VST4LNWB<0b1011, {?,0,?,?}, "32"> {
2083 let Inst{7} = lane{0};
2084 let Inst{5} = Rn{5};
2087 def VST4LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
2088 def VST4LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
2089 def VST4LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
2091 def VST4LNq16_UPD : VST4LNWB<0b0111, {?,?,1,?}, "16"> {
2092 let Inst{7-6} = lane{1-0};
2094 def VST4LNq32_UPD : VST4LNWB<0b1011, {?,1,?,?}, "32"> {
2095 let Inst{7} = lane{0};
2096 let Inst{5} = Rn{5};
2099 def VST4LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
2100 def VST4LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
2102 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
2105 //===----------------------------------------------------------------------===//
2106 // NEON pattern fragments
2107 //===----------------------------------------------------------------------===//
2109 // Extract D sub-registers of Q registers.
2110 def DSubReg_i8_reg : SDNodeXForm<imm, [{
2111 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
2112 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/8, MVT::i32);
2114 def DSubReg_i16_reg : SDNodeXForm<imm, [{
2115 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
2116 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/4, MVT::i32);
2118 def DSubReg_i32_reg : SDNodeXForm<imm, [{
2119 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
2120 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/2, MVT::i32);
2122 def DSubReg_f64_reg : SDNodeXForm<imm, [{
2123 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
2124 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue(), MVT::i32);
2127 // Extract S sub-registers of Q/D registers.
2128 def SSubReg_f32_reg : SDNodeXForm<imm, [{
2129 assert(ARM::ssub_3 == ARM::ssub_0+3 && "Unexpected subreg numbering");
2130 return CurDAG->getTargetConstant(ARM::ssub_0 + N->getZExtValue(), MVT::i32);
2133 // Translate lane numbers from Q registers to D subregs.
2134 def SubReg_i8_lane : SDNodeXForm<imm, [{
2135 return CurDAG->getTargetConstant(N->getZExtValue() & 7, MVT::i32);
2137 def SubReg_i16_lane : SDNodeXForm<imm, [{
2138 return CurDAG->getTargetConstant(N->getZExtValue() & 3, MVT::i32);
2140 def SubReg_i32_lane : SDNodeXForm<imm, [{
2141 return CurDAG->getTargetConstant(N->getZExtValue() & 1, MVT::i32);
2144 //===----------------------------------------------------------------------===//
2145 // Instruction Classes
2146 //===----------------------------------------------------------------------===//
2148 // Basic 2-register operations: double- and quad-register.
2149 class N2VD<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2150 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
2151 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
2152 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
2153 (ins DPR:$Vm), IIC_VUNAD, OpcodeStr, Dt,"$Vd, $Vm", "",
2154 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vm))))]>;
2155 class N2VQ<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2156 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
2157 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
2158 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
2159 (ins QPR:$Vm), IIC_VUNAQ, OpcodeStr, Dt,"$Vd, $Vm", "",
2160 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vm))))]>;
2162 // Basic 2-register intrinsics, both double- and quad-register.
2163 class N2VDInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2164 bits<2> op17_16, bits<5> op11_7, bit op4,
2165 InstrItinClass itin, string OpcodeStr, string Dt,
2166 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2167 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
2168 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2169 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm))))]>;
2170 class N2VQInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2171 bits<2> op17_16, bits<5> op11_7, bit op4,
2172 InstrItinClass itin, string OpcodeStr, string Dt,
2173 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2174 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
2175 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2176 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm))))]>;
2178 // Narrow 2-register operations.
2179 class N2VN<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2180 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2181 InstrItinClass itin, string OpcodeStr, string Dt,
2182 ValueType TyD, ValueType TyQ, SDNode OpNode>
2183 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$Vd),
2184 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2185 [(set DPR:$Vd, (TyD (OpNode (TyQ QPR:$Vm))))]>;
2187 // Narrow 2-register intrinsics.
2188 class N2VNInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2189 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2190 InstrItinClass itin, string OpcodeStr, string Dt,
2191 ValueType TyD, ValueType TyQ, Intrinsic IntOp>
2192 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$Vd),
2193 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2194 [(set DPR:$Vd, (TyD (IntOp (TyQ QPR:$Vm))))]>;
2196 // Long 2-register operations (currently only used for VMOVL).
2197 class N2VL<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2198 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2199 InstrItinClass itin, string OpcodeStr, string Dt,
2200 ValueType TyQ, ValueType TyD, SDNode OpNode>
2201 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$Vd),
2202 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2203 [(set QPR:$Vd, (TyQ (OpNode (TyD DPR:$Vm))))]>;
2205 // Long 2-register intrinsics.
2206 class N2VLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2207 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2208 InstrItinClass itin, string OpcodeStr, string Dt,
2209 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
2210 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$Vd),
2211 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2212 [(set QPR:$Vd, (TyQ (IntOp (TyD DPR:$Vm))))]>;
2214 // 2-register shuffles (VTRN/VZIP/VUZP), both double- and quad-register.
2215 class N2VDShuffle<bits<2> op19_18, bits<5> op11_7, string OpcodeStr, string Dt>
2216 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 0, 0, (outs DPR:$Vd, DPR:$Vm),
2217 (ins DPR:$src1, DPR:$src2), IIC_VPERMD,
2218 OpcodeStr, Dt, "$Vd, $Vm",
2219 "$src1 = $Vd, $src2 = $Vm", []>;
2220 class N2VQShuffle<bits<2> op19_18, bits<5> op11_7,
2221 InstrItinClass itin, string OpcodeStr, string Dt>
2222 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 1, 0, (outs QPR:$Vd, QPR:$Vm),
2223 (ins QPR:$src1, QPR:$src2), itin, OpcodeStr, Dt, "$Vd, $Vm",
2224 "$src1 = $Vd, $src2 = $Vm", []>;
2226 // Basic 3-register operations: double- and quad-register.
2227 class N3VD<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2228 InstrItinClass itin, string OpcodeStr, string Dt,
2229 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
2230 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2231 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2232 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2233 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
2234 let isCommutable = Commutable;
2236 // Same as N3VD but no data type.
2237 class N3VDX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2238 InstrItinClass itin, string OpcodeStr,
2239 ValueType ResTy, ValueType OpTy,
2240 SDNode OpNode, bit Commutable>
2241 : N3VX<op24, op23, op21_20, op11_8, 0, op4,
2242 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2243 OpcodeStr, "$Vd, $Vn, $Vm", "",
2244 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>{
2245 let isCommutable = Commutable;
2248 class N3VDSL<bits<2> op21_20, bits<4> op11_8,
2249 InstrItinClass itin, string OpcodeStr, string Dt,
2250 ValueType Ty, SDNode ShOp>
2251 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2252 (outs DPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2253 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2255 (Ty (ShOp (Ty DPR:$Vn),
2256 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),imm:$lane)))))]> {
2257 let isCommutable = 0;
2259 class N3VDSL16<bits<2> op21_20, bits<4> op11_8,
2260 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
2261 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2262 (outs DPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2263 NVMulSLFrm, IIC_VMULi16D, OpcodeStr, Dt,"$Vd, $Vn, $Vm$lane","",
2265 (Ty (ShOp (Ty DPR:$Vn),
2266 (Ty (NEONvduplane (Ty DPR_8:$Vm), imm:$lane)))))]> {
2267 let isCommutable = 0;
2270 class N3VQ<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2271 InstrItinClass itin, string OpcodeStr, string Dt,
2272 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
2273 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2274 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2275 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2276 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
2277 let isCommutable = Commutable;
2279 class N3VQX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2280 InstrItinClass itin, string OpcodeStr,
2281 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
2282 : N3VX<op24, op23, op21_20, op11_8, 1, op4,
2283 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2284 OpcodeStr, "$Vd, $Vn, $Vm", "",
2285 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]>{
2286 let isCommutable = Commutable;
2288 class N3VQSL<bits<2> op21_20, bits<4> op11_8,
2289 InstrItinClass itin, string OpcodeStr, string Dt,
2290 ValueType ResTy, ValueType OpTy, SDNode ShOp>
2291 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2292 (outs QPR:$Vd), (ins QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2293 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2294 [(set (ResTy QPR:$Vd),
2295 (ResTy (ShOp (ResTy QPR:$Vn),
2296 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2298 let isCommutable = 0;
2300 class N3VQSL16<bits<2> op21_20, bits<4> op11_8, string OpcodeStr, string Dt,
2301 ValueType ResTy, ValueType OpTy, SDNode ShOp>
2302 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2303 (outs QPR:$Vd), (ins QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2304 NVMulSLFrm, IIC_VMULi16Q, OpcodeStr, Dt,"$Vd, $Vn, $Vm$lane", "",
2305 [(set (ResTy QPR:$Vd),
2306 (ResTy (ShOp (ResTy QPR:$Vn),
2307 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2309 let isCommutable = 0;
2312 // Basic 3-register intrinsics, both double- and quad-register.
2313 class N3VDInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2314 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2315 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
2316 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2317 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), f, itin,
2318 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2319 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
2320 let isCommutable = Commutable;
2322 class N3VDIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2323 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
2324 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2325 (outs DPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2326 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2328 (Ty (IntOp (Ty DPR:$Vn),
2329 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),
2331 let isCommutable = 0;
2333 class N3VDIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2334 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
2335 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2336 (outs DPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2337 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2339 (Ty (IntOp (Ty DPR:$Vn),
2340 (Ty (NEONvduplane (Ty DPR_8:$Vm), imm:$lane)))))]> {
2341 let isCommutable = 0;
2343 class N3VDIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2344 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2345 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2346 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2347 (outs DPR:$Vd), (ins DPR:$Vm, DPR:$Vn), f, itin,
2348 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
2349 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (OpTy DPR:$Vn))))]> {
2350 let isCommutable = 0;
2353 class N3VQInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2354 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2355 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
2356 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2357 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), f, itin,
2358 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2359 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
2360 let isCommutable = Commutable;
2362 class N3VQIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2363 string OpcodeStr, string Dt,
2364 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2365 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2366 (outs QPR:$Vd), (ins QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2367 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2368 [(set (ResTy QPR:$Vd),
2369 (ResTy (IntOp (ResTy QPR:$Vn),
2370 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2372 let isCommutable = 0;
2374 class N3VQIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2375 string OpcodeStr, string Dt,
2376 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2377 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2378 (outs QPR:$Vd), (ins QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2379 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2380 [(set (ResTy QPR:$Vd),
2381 (ResTy (IntOp (ResTy QPR:$Vn),
2382 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2384 let isCommutable = 0;
2386 class N3VQIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2387 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2388 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2389 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2390 (outs QPR:$Vd), (ins QPR:$Vm, QPR:$Vn), f, itin,
2391 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
2392 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (OpTy QPR:$Vn))))]> {
2393 let isCommutable = 0;
2396 // Multiply-Add/Sub operations: double- and quad-register.
2397 class N3VDMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2398 InstrItinClass itin, string OpcodeStr, string Dt,
2399 ValueType Ty, SDPatternOperator MulOp, SDPatternOperator OpNode>
2400 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2401 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2402 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2403 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
2404 (Ty (MulOp DPR:$Vn, DPR:$Vm)))))]>;
2406 class N3VDMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2407 string OpcodeStr, string Dt,
2408 ValueType Ty, SDPatternOperator MulOp, SDPatternOperator ShOp>
2409 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2411 (ins DPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2413 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2415 (Ty (ShOp (Ty DPR:$src1),
2417 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),
2419 class N3VDMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2420 string OpcodeStr, string Dt,
2421 ValueType Ty, SDNode MulOp, SDNode ShOp>
2422 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2424 (ins DPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2426 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2428 (Ty (ShOp (Ty DPR:$src1),
2430 (Ty (NEONvduplane (Ty DPR_8:$Vm),
2433 class N3VQMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2434 InstrItinClass itin, string OpcodeStr, string Dt, ValueType Ty,
2435 SDPatternOperator MulOp, SDPatternOperator OpNode>
2436 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2437 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2438 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2439 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
2440 (Ty (MulOp QPR:$Vn, QPR:$Vm)))))]>;
2441 class N3VQMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2442 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2443 SDPatternOperator MulOp, SDPatternOperator ShOp>
2444 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2446 (ins QPR:$src1, QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2448 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2449 [(set (ResTy QPR:$Vd),
2450 (ResTy (ShOp (ResTy QPR:$src1),
2451 (ResTy (MulOp QPR:$Vn,
2452 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2454 class N3VQMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2455 string OpcodeStr, string Dt,
2456 ValueType ResTy, ValueType OpTy,
2457 SDNode MulOp, SDNode ShOp>
2458 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2460 (ins QPR:$src1, QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2462 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2463 [(set (ResTy QPR:$Vd),
2464 (ResTy (ShOp (ResTy QPR:$src1),
2465 (ResTy (MulOp QPR:$Vn,
2466 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2469 // Neon Intrinsic-Op instructions (VABA): double- and quad-register.
2470 class N3VDIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2471 InstrItinClass itin, string OpcodeStr, string Dt,
2472 ValueType Ty, Intrinsic IntOp, SDNode OpNode>
2473 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2474 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2475 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2476 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
2477 (Ty (IntOp (Ty DPR:$Vn), (Ty DPR:$Vm))))))]>;
2478 class N3VQIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2479 InstrItinClass itin, string OpcodeStr, string Dt,
2480 ValueType Ty, Intrinsic IntOp, SDNode OpNode>
2481 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2482 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2483 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2484 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
2485 (Ty (IntOp (Ty QPR:$Vn), (Ty QPR:$Vm))))))]>;
2487 // Neon 3-argument intrinsics, both double- and quad-register.
2488 // The destination register is also used as the first source operand register.
2489 class N3VDInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2490 InstrItinClass itin, string OpcodeStr, string Dt,
2491 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2492 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2493 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2494 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2495 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$src1),
2496 (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>;
2497 class N3VQInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2498 InstrItinClass itin, string OpcodeStr, string Dt,
2499 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2500 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2501 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2502 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2503 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$src1),
2504 (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]>;
2506 // Long Multiply-Add/Sub operations.
2507 class N3VLMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2508 InstrItinClass itin, string OpcodeStr, string Dt,
2509 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2510 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2511 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2512 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2513 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
2514 (TyQ (MulOp (TyD DPR:$Vn),
2515 (TyD DPR:$Vm)))))]>;
2516 class N3VLMulOpSL<bit op24, bits<2> op21_20, bits<4> op11_8,
2517 InstrItinClass itin, string OpcodeStr, string Dt,
2518 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2519 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$Vd),
2520 (ins QPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2522 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2524 (OpNode (TyQ QPR:$src1),
2525 (TyQ (MulOp (TyD DPR:$Vn),
2526 (TyD (NEONvduplane (TyD DPR_VFP2:$Vm),
2528 class N3VLMulOpSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2529 InstrItinClass itin, string OpcodeStr, string Dt,
2530 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2531 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$Vd),
2532 (ins QPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2534 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2536 (OpNode (TyQ QPR:$src1),
2537 (TyQ (MulOp (TyD DPR:$Vn),
2538 (TyD (NEONvduplane (TyD DPR_8:$Vm),
2541 // Long Intrinsic-Op vector operations with explicit extend (VABAL).
2542 class N3VLIntExtOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2543 InstrItinClass itin, string OpcodeStr, string Dt,
2544 ValueType TyQ, ValueType TyD, Intrinsic IntOp, SDNode ExtOp,
2546 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2547 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2548 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2549 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
2550 (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
2551 (TyD DPR:$Vm)))))))]>;
2553 // Neon Long 3-argument intrinsic. The destination register is
2554 // a quad-register and is also used as the first source operand register.
2555 class N3VLInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2556 InstrItinClass itin, string OpcodeStr, string Dt,
2557 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
2558 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2559 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2560 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2562 (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$Vn), (TyD DPR:$Vm))))]>;
2563 class N3VLInt3SL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2564 string OpcodeStr, string Dt,
2565 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2566 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2568 (ins QPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2570 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2571 [(set (ResTy QPR:$Vd),
2572 (ResTy (IntOp (ResTy QPR:$src1),
2574 (OpTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2576 class N3VLInt3SL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2577 InstrItinClass itin, string OpcodeStr, string Dt,
2578 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2579 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2581 (ins QPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2583 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2584 [(set (ResTy QPR:$Vd),
2585 (ResTy (IntOp (ResTy QPR:$src1),
2587 (OpTy (NEONvduplane (OpTy DPR_8:$Vm),
2590 // Narrowing 3-register intrinsics.
2591 class N3VNInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2592 string OpcodeStr, string Dt, ValueType TyD, ValueType TyQ,
2593 Intrinsic IntOp, bit Commutable>
2594 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2595 (outs DPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINi4D,
2596 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2597 [(set DPR:$Vd, (TyD (IntOp (TyQ QPR:$Vn), (TyQ QPR:$Vm))))]> {
2598 let isCommutable = Commutable;
2601 // Long 3-register operations.
2602 class N3VL<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2603 InstrItinClass itin, string OpcodeStr, string Dt,
2604 ValueType TyQ, ValueType TyD, SDNode OpNode, bit Commutable>
2605 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2606 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2607 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2608 [(set QPR:$Vd, (TyQ (OpNode (TyD DPR:$Vn), (TyD DPR:$Vm))))]> {
2609 let isCommutable = Commutable;
2611 class N3VLSL<bit op24, bits<2> op21_20, bits<4> op11_8,
2612 InstrItinClass itin, string OpcodeStr, string Dt,
2613 ValueType TyQ, ValueType TyD, SDNode OpNode>
2614 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2615 (outs QPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2616 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2618 (TyQ (OpNode (TyD DPR:$Vn),
2619 (TyD (NEONvduplane (TyD DPR_VFP2:$Vm),imm:$lane)))))]>;
2620 class N3VLSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2621 InstrItinClass itin, string OpcodeStr, string Dt,
2622 ValueType TyQ, ValueType TyD, SDNode OpNode>
2623 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2624 (outs QPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2625 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2627 (TyQ (OpNode (TyD DPR:$Vn),
2628 (TyD (NEONvduplane (TyD DPR_8:$Vm), imm:$lane)))))]>;
2630 // Long 3-register operations with explicitly extended operands.
2631 class N3VLExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2632 InstrItinClass itin, string OpcodeStr, string Dt,
2633 ValueType TyQ, ValueType TyD, SDNode OpNode, SDNode ExtOp,
2635 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2636 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2637 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2638 [(set QPR:$Vd, (OpNode (TyQ (ExtOp (TyD DPR:$Vn))),
2639 (TyQ (ExtOp (TyD DPR:$Vm)))))]> {
2640 let isCommutable = Commutable;
2643 // Long 3-register intrinsics with explicit extend (VABDL).
2644 class N3VLIntExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2645 InstrItinClass itin, string OpcodeStr, string Dt,
2646 ValueType TyQ, ValueType TyD, Intrinsic IntOp, SDNode ExtOp,
2648 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2649 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2650 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2651 [(set QPR:$Vd, (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
2652 (TyD DPR:$Vm))))))]> {
2653 let isCommutable = Commutable;
2656 // Long 3-register intrinsics.
2657 class N3VLInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2658 InstrItinClass itin, string OpcodeStr, string Dt,
2659 ValueType TyQ, ValueType TyD, Intrinsic IntOp, bit Commutable>
2660 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2661 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2662 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2663 [(set QPR:$Vd, (TyQ (IntOp (TyD DPR:$Vn), (TyD DPR:$Vm))))]> {
2664 let isCommutable = Commutable;
2666 class N3VLIntSL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2667 string OpcodeStr, string Dt,
2668 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2669 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2670 (outs QPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2671 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2672 [(set (ResTy QPR:$Vd),
2673 (ResTy (IntOp (OpTy DPR:$Vn),
2674 (OpTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2676 class N3VLIntSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2677 InstrItinClass itin, string OpcodeStr, string Dt,
2678 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2679 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2680 (outs QPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2681 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2682 [(set (ResTy QPR:$Vd),
2683 (ResTy (IntOp (OpTy DPR:$Vn),
2684 (OpTy (NEONvduplane (OpTy DPR_8:$Vm),
2687 // Wide 3-register operations.
2688 class N3VW<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2689 string OpcodeStr, string Dt, ValueType TyQ, ValueType TyD,
2690 SDNode OpNode, SDNode ExtOp, bit Commutable>
2691 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2692 (outs QPR:$Vd), (ins QPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VSUBiD,
2693 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2694 [(set QPR:$Vd, (OpNode (TyQ QPR:$Vn),
2695 (TyQ (ExtOp (TyD DPR:$Vm)))))]> {
2696 let isCommutable = Commutable;
2699 // Pairwise long 2-register intrinsics, both double- and quad-register.
2700 class N2VDPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2701 bits<2> op17_16, bits<5> op11_7, bit op4,
2702 string OpcodeStr, string Dt,
2703 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2704 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
2705 (ins DPR:$Vm), IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
2706 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm))))]>;
2707 class N2VQPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2708 bits<2> op17_16, bits<5> op11_7, bit op4,
2709 string OpcodeStr, string Dt,
2710 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2711 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
2712 (ins QPR:$Vm), IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
2713 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm))))]>;
2715 // Pairwise long 2-register accumulate intrinsics,
2716 // both double- and quad-register.
2717 // The destination register is also used as the first source operand register.
2718 class N2VDPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2719 bits<2> op17_16, bits<5> op11_7, bit op4,
2720 string OpcodeStr, string Dt,
2721 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2722 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
2723 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vm), IIC_VPALiD,
2724 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
2725 [(set DPR:$Vd, (ResTy (IntOp (ResTy DPR:$src1), (OpTy DPR:$Vm))))]>;
2726 class N2VQPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2727 bits<2> op17_16, bits<5> op11_7, bit op4,
2728 string OpcodeStr, string Dt,
2729 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2730 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4,
2731 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vm), IIC_VPALiQ,
2732 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
2733 [(set QPR:$Vd, (ResTy (IntOp (ResTy QPR:$src1), (OpTy QPR:$Vm))))]>;
2735 // Shift by immediate,
2736 // both double- and quad-register.
2737 class N2VDSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2738 Format f, InstrItinClass itin, Operand ImmTy,
2739 string OpcodeStr, string Dt, ValueType Ty, SDNode OpNode>
2740 : N2VImm<op24, op23, op11_8, op7, 0, op4,
2741 (outs DPR:$Vd), (ins DPR:$Vm, ImmTy:$SIMM), f, itin,
2742 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2743 [(set DPR:$Vd, (Ty (OpNode (Ty DPR:$Vm), (i32 imm:$SIMM))))]>;
2744 class N2VQSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2745 Format f, InstrItinClass itin, Operand ImmTy,
2746 string OpcodeStr, string Dt, ValueType Ty, SDNode OpNode>
2747 : N2VImm<op24, op23, op11_8, op7, 1, op4,
2748 (outs QPR:$Vd), (ins QPR:$Vm, ImmTy:$SIMM), f, itin,
2749 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2750 [(set QPR:$Vd, (Ty (OpNode (Ty QPR:$Vm), (i32 imm:$SIMM))))]>;
2752 // Long shift by immediate.
2753 class N2VLSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
2754 string OpcodeStr, string Dt,
2755 ValueType ResTy, ValueType OpTy, Operand ImmTy, SDNode OpNode>
2756 : N2VImm<op24, op23, op11_8, op7, op6, op4,
2757 (outs QPR:$Vd), (ins DPR:$Vm, ImmTy:$SIMM), N2RegVShLFrm,
2758 IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2759 [(set QPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vm),
2760 (i32 imm:$SIMM))))]>;
2762 // Narrow shift by immediate.
2763 class N2VNSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
2764 InstrItinClass itin, string OpcodeStr, string Dt,
2765 ValueType ResTy, ValueType OpTy, Operand ImmTy, SDNode OpNode>
2766 : N2VImm<op24, op23, op11_8, op7, op6, op4,
2767 (outs DPR:$Vd), (ins QPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, itin,
2768 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2769 [(set DPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vm),
2770 (i32 imm:$SIMM))))]>;
2772 // Shift right by immediate and accumulate,
2773 // both double- and quad-register.
2774 class N2VDShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2775 Operand ImmTy, string OpcodeStr, string Dt,
2776 ValueType Ty, SDNode ShOp>
2777 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
2778 (ins DPR:$src1, DPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, IIC_VPALiD,
2779 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2780 [(set DPR:$Vd, (Ty (add DPR:$src1,
2781 (Ty (ShOp DPR:$Vm, (i32 imm:$SIMM))))))]>;
2782 class N2VQShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2783 Operand ImmTy, string OpcodeStr, string Dt,
2784 ValueType Ty, SDNode ShOp>
2785 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
2786 (ins QPR:$src1, QPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, IIC_VPALiD,
2787 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2788 [(set QPR:$Vd, (Ty (add QPR:$src1,
2789 (Ty (ShOp QPR:$Vm, (i32 imm:$SIMM))))))]>;
2791 // Shift by immediate and insert,
2792 // both double- and quad-register.
2793 class N2VDShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2794 Operand ImmTy, Format f, string OpcodeStr, string Dt,
2795 ValueType Ty,SDNode ShOp>
2796 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
2797 (ins DPR:$src1, DPR:$Vm, ImmTy:$SIMM), f, IIC_VSHLiD,
2798 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2799 [(set DPR:$Vd, (Ty (ShOp DPR:$src1, DPR:$Vm, (i32 imm:$SIMM))))]>;
2800 class N2VQShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2801 Operand ImmTy, Format f, string OpcodeStr, string Dt,
2802 ValueType Ty,SDNode ShOp>
2803 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
2804 (ins QPR:$src1, QPR:$Vm, ImmTy:$SIMM), f, IIC_VSHLiQ,
2805 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2806 [(set QPR:$Vd, (Ty (ShOp QPR:$src1, QPR:$Vm, (i32 imm:$SIMM))))]>;
2808 // Convert, with fractional bits immediate,
2809 // both double- and quad-register.
2810 class N2VCvtD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2811 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2813 : N2VImm<op24, op23, op11_8, op7, 0, op4,
2814 (outs DPR:$Vd), (ins DPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
2815 IIC_VUNAD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2816 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (i32 imm:$SIMM))))]>;
2817 class N2VCvtQ<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2818 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2820 : N2VImm<op24, op23, op11_8, op7, 1, op4,
2821 (outs QPR:$Vd), (ins QPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
2822 IIC_VUNAQ, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2823 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (i32 imm:$SIMM))))]>;
2825 //===----------------------------------------------------------------------===//
2827 //===----------------------------------------------------------------------===//
2829 // Abbreviations used in multiclass suffixes:
2830 // Q = quarter int (8 bit) elements
2831 // H = half int (16 bit) elements
2832 // S = single int (32 bit) elements
2833 // D = double int (64 bit) elements
2835 // Neon 2-register vector operations and intrinsics.
2837 // Neon 2-register comparisons.
2838 // source operand element sizes of 8, 16 and 32 bits:
2839 multiclass N2V_QHS_cmp<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2840 bits<5> op11_7, bit op4, string opc, string Dt,
2841 string asm, SDNode OpNode> {
2842 // 64-bit vector types.
2843 def v8i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 0, op4,
2844 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2845 opc, !strconcat(Dt, "8"), asm, "",
2846 [(set DPR:$Vd, (v8i8 (OpNode (v8i8 DPR:$Vm))))]>;
2847 def v4i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 0, op4,
2848 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2849 opc, !strconcat(Dt, "16"), asm, "",
2850 [(set DPR:$Vd, (v4i16 (OpNode (v4i16 DPR:$Vm))))]>;
2851 def v2i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
2852 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2853 opc, !strconcat(Dt, "32"), asm, "",
2854 [(set DPR:$Vd, (v2i32 (OpNode (v2i32 DPR:$Vm))))]>;
2855 def v2f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
2856 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2857 opc, "f32", asm, "",
2858 [(set DPR:$Vd, (v2i32 (OpNode (v2f32 DPR:$Vm))))]> {
2859 let Inst{10} = 1; // overwrite F = 1
2862 // 128-bit vector types.
2863 def v16i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 1, op4,
2864 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2865 opc, !strconcat(Dt, "8"), asm, "",
2866 [(set QPR:$Vd, (v16i8 (OpNode (v16i8 QPR:$Vm))))]>;
2867 def v8i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 1, op4,
2868 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2869 opc, !strconcat(Dt, "16"), asm, "",
2870 [(set QPR:$Vd, (v8i16 (OpNode (v8i16 QPR:$Vm))))]>;
2871 def v4i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
2872 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2873 opc, !strconcat(Dt, "32"), asm, "",
2874 [(set QPR:$Vd, (v4i32 (OpNode (v4i32 QPR:$Vm))))]>;
2875 def v4f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
2876 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2877 opc, "f32", asm, "",
2878 [(set QPR:$Vd, (v4i32 (OpNode (v4f32 QPR:$Vm))))]> {
2879 let Inst{10} = 1; // overwrite F = 1
2884 // Neon 2-register vector intrinsics,
2885 // element sizes of 8, 16 and 32 bits:
2886 multiclass N2VInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2887 bits<5> op11_7, bit op4,
2888 InstrItinClass itinD, InstrItinClass itinQ,
2889 string OpcodeStr, string Dt, Intrinsic IntOp> {
2890 // 64-bit vector types.
2891 def v8i8 : N2VDInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2892 itinD, OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
2893 def v4i16 : N2VDInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2894 itinD, OpcodeStr, !strconcat(Dt, "16"),v4i16,v4i16,IntOp>;
2895 def v2i32 : N2VDInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2896 itinD, OpcodeStr, !strconcat(Dt, "32"),v2i32,v2i32,IntOp>;
2898 // 128-bit vector types.
2899 def v16i8 : N2VQInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2900 itinQ, OpcodeStr, !strconcat(Dt, "8"), v16i8,v16i8,IntOp>;
2901 def v8i16 : N2VQInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2902 itinQ, OpcodeStr, !strconcat(Dt, "16"),v8i16,v8i16,IntOp>;
2903 def v4i32 : N2VQInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2904 itinQ, OpcodeStr, !strconcat(Dt, "32"),v4i32,v4i32,IntOp>;
2908 // Neon Narrowing 2-register vector operations,
2909 // source operand element sizes of 16, 32 and 64 bits:
2910 multiclass N2VN_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2911 bits<5> op11_7, bit op6, bit op4,
2912 InstrItinClass itin, string OpcodeStr, string Dt,
2914 def v8i8 : N2VN<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
2915 itin, OpcodeStr, !strconcat(Dt, "16"),
2916 v8i8, v8i16, OpNode>;
2917 def v4i16 : N2VN<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
2918 itin, OpcodeStr, !strconcat(Dt, "32"),
2919 v4i16, v4i32, OpNode>;
2920 def v2i32 : N2VN<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
2921 itin, OpcodeStr, !strconcat(Dt, "64"),
2922 v2i32, v2i64, OpNode>;
2925 // Neon Narrowing 2-register vector intrinsics,
2926 // source operand element sizes of 16, 32 and 64 bits:
2927 multiclass N2VNInt_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2928 bits<5> op11_7, bit op6, bit op4,
2929 InstrItinClass itin, string OpcodeStr, string Dt,
2931 def v8i8 : N2VNInt<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
2932 itin, OpcodeStr, !strconcat(Dt, "16"),
2933 v8i8, v8i16, IntOp>;
2934 def v4i16 : N2VNInt<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
2935 itin, OpcodeStr, !strconcat(Dt, "32"),
2936 v4i16, v4i32, IntOp>;
2937 def v2i32 : N2VNInt<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
2938 itin, OpcodeStr, !strconcat(Dt, "64"),
2939 v2i32, v2i64, IntOp>;
2943 // Neon Lengthening 2-register vector intrinsic (currently specific to VMOVL).
2944 // source operand element sizes of 16, 32 and 64 bits:
2945 multiclass N2VL_QHS<bits<2> op24_23, bits<5> op11_7, bit op6, bit op4,
2946 string OpcodeStr, string Dt, SDNode OpNode> {
2947 def v8i16 : N2VL<op24_23, 0b00, 0b10, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2948 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode>;
2949 def v4i32 : N2VL<op24_23, 0b01, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2950 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
2951 def v2i64 : N2VL<op24_23, 0b10, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2952 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
2956 // Neon 3-register vector operations.
2958 // First with only element sizes of 8, 16 and 32 bits:
2959 multiclass N3V_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2960 InstrItinClass itinD16, InstrItinClass itinD32,
2961 InstrItinClass itinQ16, InstrItinClass itinQ32,
2962 string OpcodeStr, string Dt,
2963 SDNode OpNode, bit Commutable = 0> {
2964 // 64-bit vector types.
2965 def v8i8 : N3VD<op24, op23, 0b00, op11_8, op4, itinD16,
2966 OpcodeStr, !strconcat(Dt, "8"),
2967 v8i8, v8i8, OpNode, Commutable>;
2968 def v4i16 : N3VD<op24, op23, 0b01, op11_8, op4, itinD16,
2969 OpcodeStr, !strconcat(Dt, "16"),
2970 v4i16, v4i16, OpNode, Commutable>;
2971 def v2i32 : N3VD<op24, op23, 0b10, op11_8, op4, itinD32,
2972 OpcodeStr, !strconcat(Dt, "32"),
2973 v2i32, v2i32, OpNode, Commutable>;
2975 // 128-bit vector types.
2976 def v16i8 : N3VQ<op24, op23, 0b00, op11_8, op4, itinQ16,
2977 OpcodeStr, !strconcat(Dt, "8"),
2978 v16i8, v16i8, OpNode, Commutable>;
2979 def v8i16 : N3VQ<op24, op23, 0b01, op11_8, op4, itinQ16,
2980 OpcodeStr, !strconcat(Dt, "16"),
2981 v8i16, v8i16, OpNode, Commutable>;
2982 def v4i32 : N3VQ<op24, op23, 0b10, op11_8, op4, itinQ32,
2983 OpcodeStr, !strconcat(Dt, "32"),
2984 v4i32, v4i32, OpNode, Commutable>;
2987 multiclass N3VSL_HS<bits<4> op11_8, string OpcodeStr, SDNode ShOp> {
2988 def v4i16 : N3VDSL16<0b01, op11_8, OpcodeStr, "i16", v4i16, ShOp>;
2989 def v2i32 : N3VDSL<0b10, op11_8, IIC_VMULi32D, OpcodeStr, "i32", v2i32, ShOp>;
2990 def v8i16 : N3VQSL16<0b01, op11_8, OpcodeStr, "i16", v8i16, v4i16, ShOp>;
2991 def v4i32 : N3VQSL<0b10, op11_8, IIC_VMULi32Q, OpcodeStr, "i32",
2992 v4i32, v2i32, ShOp>;
2995 // ....then also with element size 64 bits:
2996 multiclass N3V_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2997 InstrItinClass itinD, InstrItinClass itinQ,
2998 string OpcodeStr, string Dt,
2999 SDNode OpNode, bit Commutable = 0>
3000 : N3V_QHS<op24, op23, op11_8, op4, itinD, itinD, itinQ, itinQ,
3001 OpcodeStr, Dt, OpNode, Commutable> {
3002 def v1i64 : N3VD<op24, op23, 0b11, op11_8, op4, itinD,
3003 OpcodeStr, !strconcat(Dt, "64"),
3004 v1i64, v1i64, OpNode, Commutable>;
3005 def v2i64 : N3VQ<op24, op23, 0b11, op11_8, op4, itinQ,
3006 OpcodeStr, !strconcat(Dt, "64"),
3007 v2i64, v2i64, OpNode, Commutable>;
3011 // Neon 3-register vector intrinsics.
3013 // First with only element sizes of 16 and 32 bits:
3014 multiclass N3VInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3015 InstrItinClass itinD16, InstrItinClass itinD32,
3016 InstrItinClass itinQ16, InstrItinClass itinQ32,
3017 string OpcodeStr, string Dt,
3018 Intrinsic IntOp, bit Commutable = 0> {
3019 // 64-bit vector types.
3020 def v4i16 : N3VDInt<op24, op23, 0b01, op11_8, op4, f, itinD16,
3021 OpcodeStr, !strconcat(Dt, "16"),
3022 v4i16, v4i16, IntOp, Commutable>;
3023 def v2i32 : N3VDInt<op24, op23, 0b10, op11_8, op4, f, itinD32,
3024 OpcodeStr, !strconcat(Dt, "32"),
3025 v2i32, v2i32, IntOp, Commutable>;
3027 // 128-bit vector types.
3028 def v8i16 : N3VQInt<op24, op23, 0b01, op11_8, op4, f, itinQ16,
3029 OpcodeStr, !strconcat(Dt, "16"),
3030 v8i16, v8i16, IntOp, Commutable>;
3031 def v4i32 : N3VQInt<op24, op23, 0b10, op11_8, op4, f, itinQ32,
3032 OpcodeStr, !strconcat(Dt, "32"),
3033 v4i32, v4i32, IntOp, Commutable>;
3035 multiclass N3VInt_HSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3036 InstrItinClass itinD16, InstrItinClass itinD32,
3037 InstrItinClass itinQ16, InstrItinClass itinQ32,
3038 string OpcodeStr, string Dt,
3040 // 64-bit vector types.
3041 def v4i16 : N3VDIntSh<op24, op23, 0b01, op11_8, op4, f, itinD16,
3042 OpcodeStr, !strconcat(Dt, "16"),
3043 v4i16, v4i16, IntOp>;
3044 def v2i32 : N3VDIntSh<op24, op23, 0b10, op11_8, op4, f, itinD32,
3045 OpcodeStr, !strconcat(Dt, "32"),
3046 v2i32, v2i32, IntOp>;
3048 // 128-bit vector types.
3049 def v8i16 : N3VQIntSh<op24, op23, 0b01, op11_8, op4, f, itinQ16,
3050 OpcodeStr, !strconcat(Dt, "16"),
3051 v8i16, v8i16, IntOp>;
3052 def v4i32 : N3VQIntSh<op24, op23, 0b10, op11_8, op4, f, itinQ32,
3053 OpcodeStr, !strconcat(Dt, "32"),
3054 v4i32, v4i32, IntOp>;
3057 multiclass N3VIntSL_HS<bits<4> op11_8,
3058 InstrItinClass itinD16, InstrItinClass itinD32,
3059 InstrItinClass itinQ16, InstrItinClass itinQ32,
3060 string OpcodeStr, string Dt, Intrinsic IntOp> {
3061 def v4i16 : N3VDIntSL16<0b01, op11_8, itinD16,
3062 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp>;
3063 def v2i32 : N3VDIntSL<0b10, op11_8, itinD32,
3064 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp>;
3065 def v8i16 : N3VQIntSL16<0b01, op11_8, itinQ16,
3066 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16, IntOp>;
3067 def v4i32 : N3VQIntSL<0b10, op11_8, itinQ32,
3068 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32, IntOp>;
3071 // ....then also with element size of 8 bits:
3072 multiclass N3VInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3073 InstrItinClass itinD16, InstrItinClass itinD32,
3074 InstrItinClass itinQ16, InstrItinClass itinQ32,
3075 string OpcodeStr, string Dt,
3076 Intrinsic IntOp, bit Commutable = 0>
3077 : N3VInt_HS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
3078 OpcodeStr, Dt, IntOp, Commutable> {
3079 def v8i8 : N3VDInt<op24, op23, 0b00, op11_8, op4, f, itinD16,
3080 OpcodeStr, !strconcat(Dt, "8"),
3081 v8i8, v8i8, IntOp, Commutable>;
3082 def v16i8 : N3VQInt<op24, op23, 0b00, op11_8, op4, f, itinQ16,
3083 OpcodeStr, !strconcat(Dt, "8"),
3084 v16i8, v16i8, IntOp, Commutable>;
3086 multiclass N3VInt_QHSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3087 InstrItinClass itinD16, InstrItinClass itinD32,
3088 InstrItinClass itinQ16, InstrItinClass itinQ32,
3089 string OpcodeStr, string Dt,
3091 : N3VInt_HSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
3092 OpcodeStr, Dt, IntOp> {
3093 def v8i8 : N3VDIntSh<op24, op23, 0b00, op11_8, op4, f, itinD16,
3094 OpcodeStr, !strconcat(Dt, "8"),
3096 def v16i8 : N3VQIntSh<op24, op23, 0b00, op11_8, op4, f, itinQ16,
3097 OpcodeStr, !strconcat(Dt, "8"),
3098 v16i8, v16i8, IntOp>;
3102 // ....then also with element size of 64 bits:
3103 multiclass N3VInt_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3104 InstrItinClass itinD16, InstrItinClass itinD32,
3105 InstrItinClass itinQ16, InstrItinClass itinQ32,
3106 string OpcodeStr, string Dt,
3107 Intrinsic IntOp, bit Commutable = 0>
3108 : N3VInt_QHS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
3109 OpcodeStr, Dt, IntOp, Commutable> {
3110 def v1i64 : N3VDInt<op24, op23, 0b11, op11_8, op4, f, itinD32,
3111 OpcodeStr, !strconcat(Dt, "64"),
3112 v1i64, v1i64, IntOp, Commutable>;
3113 def v2i64 : N3VQInt<op24, op23, 0b11, op11_8, op4, f, itinQ32,
3114 OpcodeStr, !strconcat(Dt, "64"),
3115 v2i64, v2i64, IntOp, Commutable>;
3117 multiclass N3VInt_QHSDSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3118 InstrItinClass itinD16, InstrItinClass itinD32,
3119 InstrItinClass itinQ16, InstrItinClass itinQ32,
3120 string OpcodeStr, string Dt,
3122 : N3VInt_QHSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
3123 OpcodeStr, Dt, IntOp> {
3124 def v1i64 : N3VDIntSh<op24, op23, 0b11, op11_8, op4, f, itinD32,
3125 OpcodeStr, !strconcat(Dt, "64"),
3126 v1i64, v1i64, IntOp>;
3127 def v2i64 : N3VQIntSh<op24, op23, 0b11, op11_8, op4, f, itinQ32,
3128 OpcodeStr, !strconcat(Dt, "64"),
3129 v2i64, v2i64, IntOp>;
3132 // Neon Narrowing 3-register vector intrinsics,
3133 // source operand element sizes of 16, 32 and 64 bits:
3134 multiclass N3VNInt_HSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3135 string OpcodeStr, string Dt,
3136 Intrinsic IntOp, bit Commutable = 0> {
3137 def v8i8 : N3VNInt<op24, op23, 0b00, op11_8, op4,
3138 OpcodeStr, !strconcat(Dt, "16"),
3139 v8i8, v8i16, IntOp, Commutable>;
3140 def v4i16 : N3VNInt<op24, op23, 0b01, op11_8, op4,
3141 OpcodeStr, !strconcat(Dt, "32"),
3142 v4i16, v4i32, IntOp, Commutable>;
3143 def v2i32 : N3VNInt<op24, op23, 0b10, op11_8, op4,
3144 OpcodeStr, !strconcat(Dt, "64"),
3145 v2i32, v2i64, IntOp, Commutable>;
3149 // Neon Long 3-register vector operations.
3151 multiclass N3VL_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3152 InstrItinClass itin16, InstrItinClass itin32,
3153 string OpcodeStr, string Dt,
3154 SDNode OpNode, bit Commutable = 0> {
3155 def v8i16 : N3VL<op24, op23, 0b00, op11_8, op4, itin16,
3156 OpcodeStr, !strconcat(Dt, "8"),
3157 v8i16, v8i8, OpNode, Commutable>;
3158 def v4i32 : N3VL<op24, op23, 0b01, op11_8, op4, itin16,
3159 OpcodeStr, !strconcat(Dt, "16"),
3160 v4i32, v4i16, OpNode, Commutable>;
3161 def v2i64 : N3VL<op24, op23, 0b10, op11_8, op4, itin32,
3162 OpcodeStr, !strconcat(Dt, "32"),
3163 v2i64, v2i32, OpNode, Commutable>;
3166 multiclass N3VLSL_HS<bit op24, bits<4> op11_8,
3167 InstrItinClass itin, string OpcodeStr, string Dt,
3169 def v4i16 : N3VLSL16<op24, 0b01, op11_8, itin, OpcodeStr,
3170 !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
3171 def v2i32 : N3VLSL<op24, 0b10, op11_8, itin, OpcodeStr,
3172 !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
3175 multiclass N3VLExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3176 InstrItinClass itin16, InstrItinClass itin32,
3177 string OpcodeStr, string Dt,
3178 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
3179 def v8i16 : N3VLExt<op24, op23, 0b00, op11_8, op4, itin16,
3180 OpcodeStr, !strconcat(Dt, "8"),
3181 v8i16, v8i8, OpNode, ExtOp, Commutable>;
3182 def v4i32 : N3VLExt<op24, op23, 0b01, op11_8, op4, itin16,
3183 OpcodeStr, !strconcat(Dt, "16"),
3184 v4i32, v4i16, OpNode, ExtOp, Commutable>;
3185 def v2i64 : N3VLExt<op24, op23, 0b10, op11_8, op4, itin32,
3186 OpcodeStr, !strconcat(Dt, "32"),
3187 v2i64, v2i32, OpNode, ExtOp, Commutable>;
3190 // Neon Long 3-register vector intrinsics.
3192 // First with only element sizes of 16 and 32 bits:
3193 multiclass N3VLInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
3194 InstrItinClass itin16, InstrItinClass itin32,
3195 string OpcodeStr, string Dt,
3196 Intrinsic IntOp, bit Commutable = 0> {
3197 def v4i32 : N3VLInt<op24, op23, 0b01, op11_8, op4, itin16,
3198 OpcodeStr, !strconcat(Dt, "16"),
3199 v4i32, v4i16, IntOp, Commutable>;
3200 def v2i64 : N3VLInt<op24, op23, 0b10, op11_8, op4, itin32,
3201 OpcodeStr, !strconcat(Dt, "32"),
3202 v2i64, v2i32, IntOp, Commutable>;
3205 multiclass N3VLIntSL_HS<bit op24, bits<4> op11_8,
3206 InstrItinClass itin, string OpcodeStr, string Dt,
3208 def v4i16 : N3VLIntSL16<op24, 0b01, op11_8, itin,
3209 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
3210 def v2i32 : N3VLIntSL<op24, 0b10, op11_8, itin,
3211 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3214 // ....then also with element size of 8 bits:
3215 multiclass N3VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3216 InstrItinClass itin16, InstrItinClass itin32,
3217 string OpcodeStr, string Dt,
3218 Intrinsic IntOp, bit Commutable = 0>
3219 : N3VLInt_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt,
3220 IntOp, Commutable> {
3221 def v8i16 : N3VLInt<op24, op23, 0b00, op11_8, op4, itin16,
3222 OpcodeStr, !strconcat(Dt, "8"),
3223 v8i16, v8i8, IntOp, Commutable>;
3226 // ....with explicit extend (VABDL).
3227 multiclass N3VLIntExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3228 InstrItinClass itin, string OpcodeStr, string Dt,
3229 Intrinsic IntOp, SDNode ExtOp, bit Commutable = 0> {
3230 def v8i16 : N3VLIntExt<op24, op23, 0b00, op11_8, op4, itin,
3231 OpcodeStr, !strconcat(Dt, "8"),
3232 v8i16, v8i8, IntOp, ExtOp, Commutable>;
3233 def v4i32 : N3VLIntExt<op24, op23, 0b01, op11_8, op4, itin,
3234 OpcodeStr, !strconcat(Dt, "16"),
3235 v4i32, v4i16, IntOp, ExtOp, Commutable>;
3236 def v2i64 : N3VLIntExt<op24, op23, 0b10, op11_8, op4, itin,
3237 OpcodeStr, !strconcat(Dt, "32"),
3238 v2i64, v2i32, IntOp, ExtOp, Commutable>;
3242 // Neon Wide 3-register vector intrinsics,
3243 // source operand element sizes of 8, 16 and 32 bits:
3244 multiclass N3VW_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3245 string OpcodeStr, string Dt,
3246 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
3247 def v8i16 : N3VW<op24, op23, 0b00, op11_8, op4,
3248 OpcodeStr, !strconcat(Dt, "8"),
3249 v8i16, v8i8, OpNode, ExtOp, Commutable>;
3250 def v4i32 : N3VW<op24, op23, 0b01, op11_8, op4,
3251 OpcodeStr, !strconcat(Dt, "16"),
3252 v4i32, v4i16, OpNode, ExtOp, Commutable>;
3253 def v2i64 : N3VW<op24, op23, 0b10, op11_8, op4,
3254 OpcodeStr, !strconcat(Dt, "32"),
3255 v2i64, v2i32, OpNode, ExtOp, Commutable>;
3259 // Neon Multiply-Op vector operations,
3260 // element sizes of 8, 16 and 32 bits:
3261 multiclass N3VMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3262 InstrItinClass itinD16, InstrItinClass itinD32,
3263 InstrItinClass itinQ16, InstrItinClass itinQ32,
3264 string OpcodeStr, string Dt, SDNode OpNode> {
3265 // 64-bit vector types.
3266 def v8i8 : N3VDMulOp<op24, op23, 0b00, op11_8, op4, itinD16,
3267 OpcodeStr, !strconcat(Dt, "8"), v8i8, mul, OpNode>;
3268 def v4i16 : N3VDMulOp<op24, op23, 0b01, op11_8, op4, itinD16,
3269 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, OpNode>;
3270 def v2i32 : N3VDMulOp<op24, op23, 0b10, op11_8, op4, itinD32,
3271 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, OpNode>;
3273 // 128-bit vector types.
3274 def v16i8 : N3VQMulOp<op24, op23, 0b00, op11_8, op4, itinQ16,
3275 OpcodeStr, !strconcat(Dt, "8"), v16i8, mul, OpNode>;
3276 def v8i16 : N3VQMulOp<op24, op23, 0b01, op11_8, op4, itinQ16,
3277 OpcodeStr, !strconcat(Dt, "16"), v8i16, mul, OpNode>;
3278 def v4i32 : N3VQMulOp<op24, op23, 0b10, op11_8, op4, itinQ32,
3279 OpcodeStr, !strconcat(Dt, "32"), v4i32, mul, OpNode>;
3282 multiclass N3VMulOpSL_HS<bits<4> op11_8,
3283 InstrItinClass itinD16, InstrItinClass itinD32,
3284 InstrItinClass itinQ16, InstrItinClass itinQ32,
3285 string OpcodeStr, string Dt, SDNode ShOp> {
3286 def v4i16 : N3VDMulOpSL16<0b01, op11_8, itinD16,
3287 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, ShOp>;
3288 def v2i32 : N3VDMulOpSL<0b10, op11_8, itinD32,
3289 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, ShOp>;
3290 def v8i16 : N3VQMulOpSL16<0b01, op11_8, itinQ16,
3291 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16,
3293 def v4i32 : N3VQMulOpSL<0b10, op11_8, itinQ32,
3294 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32,
3298 // Neon Intrinsic-Op vector operations,
3299 // element sizes of 8, 16 and 32 bits:
3300 multiclass N3VIntOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3301 InstrItinClass itinD, InstrItinClass itinQ,
3302 string OpcodeStr, string Dt, Intrinsic IntOp,
3304 // 64-bit vector types.
3305 def v8i8 : N3VDIntOp<op24, op23, 0b00, op11_8, op4, itinD,
3306 OpcodeStr, !strconcat(Dt, "8"), v8i8, IntOp, OpNode>;
3307 def v4i16 : N3VDIntOp<op24, op23, 0b01, op11_8, op4, itinD,
3308 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp, OpNode>;
3309 def v2i32 : N3VDIntOp<op24, op23, 0b10, op11_8, op4, itinD,
3310 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp, OpNode>;
3312 // 128-bit vector types.
3313 def v16i8 : N3VQIntOp<op24, op23, 0b00, op11_8, op4, itinQ,
3314 OpcodeStr, !strconcat(Dt, "8"), v16i8, IntOp, OpNode>;
3315 def v8i16 : N3VQIntOp<op24, op23, 0b01, op11_8, op4, itinQ,
3316 OpcodeStr, !strconcat(Dt, "16"), v8i16, IntOp, OpNode>;
3317 def v4i32 : N3VQIntOp<op24, op23, 0b10, op11_8, op4, itinQ,
3318 OpcodeStr, !strconcat(Dt, "32"), v4i32, IntOp, OpNode>;
3321 // Neon 3-argument intrinsics,
3322 // element sizes of 8, 16 and 32 bits:
3323 multiclass N3VInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3324 InstrItinClass itinD, InstrItinClass itinQ,
3325 string OpcodeStr, string Dt, Intrinsic IntOp> {
3326 // 64-bit vector types.
3327 def v8i8 : N3VDInt3<op24, op23, 0b00, op11_8, op4, itinD,
3328 OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
3329 def v4i16 : N3VDInt3<op24, op23, 0b01, op11_8, op4, itinD,
3330 OpcodeStr, !strconcat(Dt, "16"), v4i16, v4i16, IntOp>;
3331 def v2i32 : N3VDInt3<op24, op23, 0b10, op11_8, op4, itinD,
3332 OpcodeStr, !strconcat(Dt, "32"), v2i32, v2i32, IntOp>;
3334 // 128-bit vector types.
3335 def v16i8 : N3VQInt3<op24, op23, 0b00, op11_8, op4, itinQ,
3336 OpcodeStr, !strconcat(Dt, "8"), v16i8, v16i8, IntOp>;
3337 def v8i16 : N3VQInt3<op24, op23, 0b01, op11_8, op4, itinQ,
3338 OpcodeStr, !strconcat(Dt, "16"), v8i16, v8i16, IntOp>;
3339 def v4i32 : N3VQInt3<op24, op23, 0b10, op11_8, op4, itinQ,
3340 OpcodeStr, !strconcat(Dt, "32"), v4i32, v4i32, IntOp>;
3344 // Neon Long Multiply-Op vector operations,
3345 // element sizes of 8, 16 and 32 bits:
3346 multiclass N3VLMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3347 InstrItinClass itin16, InstrItinClass itin32,
3348 string OpcodeStr, string Dt, SDNode MulOp,
3350 def v8i16 : N3VLMulOp<op24, op23, 0b00, op11_8, op4, itin16, OpcodeStr,
3351 !strconcat(Dt, "8"), v8i16, v8i8, MulOp, OpNode>;
3352 def v4i32 : N3VLMulOp<op24, op23, 0b01, op11_8, op4, itin16, OpcodeStr,
3353 !strconcat(Dt, "16"), v4i32, v4i16, MulOp, OpNode>;
3354 def v2i64 : N3VLMulOp<op24, op23, 0b10, op11_8, op4, itin32, OpcodeStr,
3355 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
3358 multiclass N3VLMulOpSL_HS<bit op24, bits<4> op11_8, string OpcodeStr,
3359 string Dt, SDNode MulOp, SDNode OpNode> {
3360 def v4i16 : N3VLMulOpSL16<op24, 0b01, op11_8, IIC_VMACi16D, OpcodeStr,
3361 !strconcat(Dt,"16"), v4i32, v4i16, MulOp, OpNode>;
3362 def v2i32 : N3VLMulOpSL<op24, 0b10, op11_8, IIC_VMACi32D, OpcodeStr,
3363 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
3367 // Neon Long 3-argument intrinsics.
3369 // First with only element sizes of 16 and 32 bits:
3370 multiclass N3VLInt3_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
3371 InstrItinClass itin16, InstrItinClass itin32,
3372 string OpcodeStr, string Dt, Intrinsic IntOp> {
3373 def v4i32 : N3VLInt3<op24, op23, 0b01, op11_8, op4, itin16,
3374 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
3375 def v2i64 : N3VLInt3<op24, op23, 0b10, op11_8, op4, itin32,
3376 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3379 multiclass N3VLInt3SL_HS<bit op24, bits<4> op11_8,
3380 string OpcodeStr, string Dt, Intrinsic IntOp> {
3381 def v4i16 : N3VLInt3SL16<op24, 0b01, op11_8, IIC_VMACi16D,
3382 OpcodeStr, !strconcat(Dt,"16"), v4i32, v4i16, IntOp>;
3383 def v2i32 : N3VLInt3SL<op24, 0b10, op11_8, IIC_VMACi32D,
3384 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3387 // ....then also with element size of 8 bits:
3388 multiclass N3VLInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3389 InstrItinClass itin16, InstrItinClass itin32,
3390 string OpcodeStr, string Dt, Intrinsic IntOp>
3391 : N3VLInt3_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt, IntOp> {
3392 def v8i16 : N3VLInt3<op24, op23, 0b00, op11_8, op4, itin16,
3393 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
3396 // ....with explicit extend (VABAL).
3397 multiclass N3VLIntExtOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3398 InstrItinClass itin, string OpcodeStr, string Dt,
3399 Intrinsic IntOp, SDNode ExtOp, SDNode OpNode> {
3400 def v8i16 : N3VLIntExtOp<op24, op23, 0b00, op11_8, op4, itin,
3401 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8,
3402 IntOp, ExtOp, OpNode>;
3403 def v4i32 : N3VLIntExtOp<op24, op23, 0b01, op11_8, op4, itin,
3404 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16,
3405 IntOp, ExtOp, OpNode>;
3406 def v2i64 : N3VLIntExtOp<op24, op23, 0b10, op11_8, op4, itin,
3407 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32,
3408 IntOp, ExtOp, OpNode>;
3412 // Neon Pairwise long 2-register intrinsics,
3413 // element sizes of 8, 16 and 32 bits:
3414 multiclass N2VPLInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3415 bits<5> op11_7, bit op4,
3416 string OpcodeStr, string Dt, Intrinsic IntOp> {
3417 // 64-bit vector types.
3418 def v8i8 : N2VDPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3419 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
3420 def v4i16 : N2VDPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3421 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
3422 def v2i32 : N2VDPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3423 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
3425 // 128-bit vector types.
3426 def v16i8 : N2VQPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3427 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
3428 def v8i16 : N2VQPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3429 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
3430 def v4i32 : N2VQPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3431 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
3435 // Neon Pairwise long 2-register accumulate intrinsics,
3436 // element sizes of 8, 16 and 32 bits:
3437 multiclass N2VPLInt2_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3438 bits<5> op11_7, bit op4,
3439 string OpcodeStr, string Dt, Intrinsic IntOp> {
3440 // 64-bit vector types.
3441 def v8i8 : N2VDPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3442 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
3443 def v4i16 : N2VDPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3444 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
3445 def v2i32 : N2VDPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3446 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
3448 // 128-bit vector types.
3449 def v16i8 : N2VQPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3450 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
3451 def v8i16 : N2VQPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3452 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
3453 def v4i32 : N2VQPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3454 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
3458 // Neon 2-register vector shift by immediate,
3459 // with f of either N2RegVShLFrm or N2RegVShRFrm
3460 // element sizes of 8, 16, 32 and 64 bits:
3461 multiclass N2VShL_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3462 InstrItinClass itin, string OpcodeStr, string Dt,
3464 // 64-bit vector types.
3465 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3466 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
3467 let Inst{21-19} = 0b001; // imm6 = 001xxx
3469 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3470 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
3471 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3473 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3474 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
3475 let Inst{21} = 0b1; // imm6 = 1xxxxx
3477 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, N2RegVShLFrm, itin, i32imm,
3478 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
3481 // 128-bit vector types.
3482 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3483 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
3484 let Inst{21-19} = 0b001; // imm6 = 001xxx
3486 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3487 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
3488 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3490 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3491 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
3492 let Inst{21} = 0b1; // imm6 = 1xxxxx
3494 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, N2RegVShLFrm, itin, i32imm,
3495 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
3498 multiclass N2VShR_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3499 InstrItinClass itin, string OpcodeStr, string Dt,
3501 // 64-bit vector types.
3502 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm8,
3503 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
3504 let Inst{21-19} = 0b001; // imm6 = 001xxx
3506 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm16,
3507 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
3508 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3510 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm32,
3511 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
3512 let Inst{21} = 0b1; // imm6 = 1xxxxx
3514 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, N2RegVShRFrm, itin, shr_imm64,
3515 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
3518 // 128-bit vector types.
3519 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm8,
3520 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
3521 let Inst{21-19} = 0b001; // imm6 = 001xxx
3523 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm16,
3524 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
3525 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3527 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm32,
3528 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
3529 let Inst{21} = 0b1; // imm6 = 1xxxxx
3531 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, N2RegVShRFrm, itin, shr_imm64,
3532 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
3536 // Neon Shift-Accumulate vector operations,
3537 // element sizes of 8, 16, 32 and 64 bits:
3538 multiclass N2VShAdd_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3539 string OpcodeStr, string Dt, SDNode ShOp> {
3540 // 64-bit vector types.
3541 def v8i8 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm8,
3542 OpcodeStr, !strconcat(Dt, "8"), v8i8, ShOp> {
3543 let Inst{21-19} = 0b001; // imm6 = 001xxx
3545 def v4i16 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm16,
3546 OpcodeStr, !strconcat(Dt, "16"), v4i16, ShOp> {
3547 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3549 def v2i32 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm32,
3550 OpcodeStr, !strconcat(Dt, "32"), v2i32, ShOp> {
3551 let Inst{21} = 0b1; // imm6 = 1xxxxx
3553 def v1i64 : N2VDShAdd<op24, op23, op11_8, 1, op4, shr_imm64,
3554 OpcodeStr, !strconcat(Dt, "64"), v1i64, ShOp>;
3557 // 128-bit vector types.
3558 def v16i8 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm8,
3559 OpcodeStr, !strconcat(Dt, "8"), v16i8, ShOp> {
3560 let Inst{21-19} = 0b001; // imm6 = 001xxx
3562 def v8i16 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm16,
3563 OpcodeStr, !strconcat(Dt, "16"), v8i16, ShOp> {
3564 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3566 def v4i32 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm32,
3567 OpcodeStr, !strconcat(Dt, "32"), v4i32, ShOp> {
3568 let Inst{21} = 0b1; // imm6 = 1xxxxx
3570 def v2i64 : N2VQShAdd<op24, op23, op11_8, 1, op4, shr_imm64,
3571 OpcodeStr, !strconcat(Dt, "64"), v2i64, ShOp>;
3575 // Neon Shift-Insert vector operations,
3576 // with f of either N2RegVShLFrm or N2RegVShRFrm
3577 // element sizes of 8, 16, 32 and 64 bits:
3578 multiclass N2VShInsL_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3580 // 64-bit vector types.
3581 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3582 N2RegVShLFrm, OpcodeStr, "8", v8i8, NEONvsli> {
3583 let Inst{21-19} = 0b001; // imm6 = 001xxx
3585 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3586 N2RegVShLFrm, OpcodeStr, "16", v4i16, NEONvsli> {
3587 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3589 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3590 N2RegVShLFrm, OpcodeStr, "32", v2i32, NEONvsli> {
3591 let Inst{21} = 0b1; // imm6 = 1xxxxx
3593 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4, i32imm,
3594 N2RegVShLFrm, OpcodeStr, "64", v1i64, NEONvsli>;
3597 // 128-bit vector types.
3598 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3599 N2RegVShLFrm, OpcodeStr, "8", v16i8, NEONvsli> {
3600 let Inst{21-19} = 0b001; // imm6 = 001xxx
3602 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3603 N2RegVShLFrm, OpcodeStr, "16", v8i16, NEONvsli> {
3604 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3606 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3607 N2RegVShLFrm, OpcodeStr, "32", v4i32, NEONvsli> {
3608 let Inst{21} = 0b1; // imm6 = 1xxxxx
3610 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4, i32imm,
3611 N2RegVShLFrm, OpcodeStr, "64", v2i64, NEONvsli>;
3614 multiclass N2VShInsR_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3616 // 64-bit vector types.
3617 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm8,
3618 N2RegVShRFrm, OpcodeStr, "8", v8i8, NEONvsri> {
3619 let Inst{21-19} = 0b001; // imm6 = 001xxx
3621 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm16,
3622 N2RegVShRFrm, OpcodeStr, "16", v4i16, NEONvsri> {
3623 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3625 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm32,
3626 N2RegVShRFrm, OpcodeStr, "32", v2i32, NEONvsri> {
3627 let Inst{21} = 0b1; // imm6 = 1xxxxx
3629 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4, shr_imm64,
3630 N2RegVShRFrm, OpcodeStr, "64", v1i64, NEONvsri>;
3633 // 128-bit vector types.
3634 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm8,
3635 N2RegVShRFrm, OpcodeStr, "8", v16i8, NEONvsri> {
3636 let Inst{21-19} = 0b001; // imm6 = 001xxx
3638 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm16,
3639 N2RegVShRFrm, OpcodeStr, "16", v8i16, NEONvsri> {
3640 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3642 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm32,
3643 N2RegVShRFrm, OpcodeStr, "32", v4i32, NEONvsri> {
3644 let Inst{21} = 0b1; // imm6 = 1xxxxx
3646 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4, shr_imm64,
3647 N2RegVShRFrm, OpcodeStr, "64", v2i64, NEONvsri>;
3651 // Neon Shift Long operations,
3652 // element sizes of 8, 16, 32 bits:
3653 multiclass N2VLSh_QHS<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
3654 bit op4, string OpcodeStr, string Dt, SDNode OpNode> {
3655 def v8i16 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3656 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, imm1_7, OpNode> {
3657 let Inst{21-19} = 0b001; // imm6 = 001xxx
3659 def v4i32 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3660 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, imm1_15, OpNode> {
3661 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3663 def v2i64 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3664 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, imm1_31, OpNode> {
3665 let Inst{21} = 0b1; // imm6 = 1xxxxx
3669 // Neon Shift Narrow operations,
3670 // element sizes of 16, 32, 64 bits:
3671 multiclass N2VNSh_HSD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
3672 bit op4, InstrItinClass itin, string OpcodeStr, string Dt,
3674 def v8i8 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3675 OpcodeStr, !strconcat(Dt, "16"),
3676 v8i8, v8i16, shr_imm8, OpNode> {
3677 let Inst{21-19} = 0b001; // imm6 = 001xxx
3679 def v4i16 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3680 OpcodeStr, !strconcat(Dt, "32"),
3681 v4i16, v4i32, shr_imm16, OpNode> {
3682 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3684 def v2i32 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3685 OpcodeStr, !strconcat(Dt, "64"),
3686 v2i32, v2i64, shr_imm32, OpNode> {
3687 let Inst{21} = 0b1; // imm6 = 1xxxxx
3691 //===----------------------------------------------------------------------===//
3692 // Instruction Definitions.
3693 //===----------------------------------------------------------------------===//
3695 // Vector Add Operations.
3697 // VADD : Vector Add (integer and floating-point)
3698 defm VADD : N3V_QHSD<0, 0, 0b1000, 0, IIC_VBINiD, IIC_VBINiQ, "vadd", "i",
3700 def VADDfd : N3VD<0, 0, 0b00, 0b1101, 0, IIC_VBIND, "vadd", "f32",
3701 v2f32, v2f32, fadd, 1>;
3702 def VADDfq : N3VQ<0, 0, 0b00, 0b1101, 0, IIC_VBINQ, "vadd", "f32",
3703 v4f32, v4f32, fadd, 1>;
3704 // VADDL : Vector Add Long (Q = D + D)
3705 defm VADDLs : N3VLExt_QHS<0,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
3706 "vaddl", "s", add, sext, 1>;
3707 defm VADDLu : N3VLExt_QHS<1,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
3708 "vaddl", "u", add, zext, 1>;
3709 // VADDW : Vector Add Wide (Q = Q + D)
3710 defm VADDWs : N3VW_QHS<0,1,0b0001,0, "vaddw", "s", add, sext, 0>;
3711 defm VADDWu : N3VW_QHS<1,1,0b0001,0, "vaddw", "u", add, zext, 0>;
3712 // VHADD : Vector Halving Add
3713 defm VHADDs : N3VInt_QHS<0, 0, 0b0000, 0, N3RegFrm,
3714 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3715 "vhadd", "s", int_arm_neon_vhadds, 1>;
3716 defm VHADDu : N3VInt_QHS<1, 0, 0b0000, 0, N3RegFrm,
3717 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3718 "vhadd", "u", int_arm_neon_vhaddu, 1>;
3719 // VRHADD : Vector Rounding Halving Add
3720 defm VRHADDs : N3VInt_QHS<0, 0, 0b0001, 0, N3RegFrm,
3721 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3722 "vrhadd", "s", int_arm_neon_vrhadds, 1>;
3723 defm VRHADDu : N3VInt_QHS<1, 0, 0b0001, 0, N3RegFrm,
3724 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3725 "vrhadd", "u", int_arm_neon_vrhaddu, 1>;
3726 // VQADD : Vector Saturating Add
3727 defm VQADDs : N3VInt_QHSD<0, 0, 0b0000, 1, N3RegFrm,
3728 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3729 "vqadd", "s", int_arm_neon_vqadds, 1>;
3730 defm VQADDu : N3VInt_QHSD<1, 0, 0b0000, 1, N3RegFrm,
3731 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3732 "vqadd", "u", int_arm_neon_vqaddu, 1>;
3733 // VADDHN : Vector Add and Narrow Returning High Half (D = Q + Q)
3734 defm VADDHN : N3VNInt_HSD<0,1,0b0100,0, "vaddhn", "i",
3735 int_arm_neon_vaddhn, 1>;
3736 // VRADDHN : Vector Rounding Add and Narrow Returning High Half (D = Q + Q)
3737 defm VRADDHN : N3VNInt_HSD<1,1,0b0100,0, "vraddhn", "i",
3738 int_arm_neon_vraddhn, 1>;
3740 // Vector Multiply Operations.
3742 // VMUL : Vector Multiply (integer, polynomial and floating-point)
3743 defm VMUL : N3V_QHS<0, 0, 0b1001, 1, IIC_VMULi16D, IIC_VMULi32D,
3744 IIC_VMULi16Q, IIC_VMULi32Q, "vmul", "i", mul, 1>;
3745 def VMULpd : N3VDInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16D, "vmul",
3746 "p8", v8i8, v8i8, int_arm_neon_vmulp, 1>;
3747 def VMULpq : N3VQInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16Q, "vmul",
3748 "p8", v16i8, v16i8, int_arm_neon_vmulp, 1>;
3749 def VMULfd : N3VD<1, 0, 0b00, 0b1101, 1, IIC_VFMULD, "vmul", "f32",
3750 v2f32, v2f32, fmul, 1>;
3751 def VMULfq : N3VQ<1, 0, 0b00, 0b1101, 1, IIC_VFMULQ, "vmul", "f32",
3752 v4f32, v4f32, fmul, 1>;
3753 defm VMULsl : N3VSL_HS<0b1000, "vmul", mul>;
3754 def VMULslfd : N3VDSL<0b10, 0b1001, IIC_VBIND, "vmul", "f32", v2f32, fmul>;
3755 def VMULslfq : N3VQSL<0b10, 0b1001, IIC_VBINQ, "vmul", "f32", v4f32,
3758 def : Pat<(v8i16 (mul (v8i16 QPR:$src1),
3759 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
3760 (v8i16 (VMULslv8i16 (v8i16 QPR:$src1),
3761 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3762 (DSubReg_i16_reg imm:$lane))),
3763 (SubReg_i16_lane imm:$lane)))>;
3764 def : Pat<(v4i32 (mul (v4i32 QPR:$src1),
3765 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
3766 (v4i32 (VMULslv4i32 (v4i32 QPR:$src1),
3767 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3768 (DSubReg_i32_reg imm:$lane))),
3769 (SubReg_i32_lane imm:$lane)))>;
3770 def : Pat<(v4f32 (fmul (v4f32 QPR:$src1),
3771 (v4f32 (NEONvduplane (v4f32 QPR:$src2), imm:$lane)))),
3772 (v4f32 (VMULslfq (v4f32 QPR:$src1),
3773 (v2f32 (EXTRACT_SUBREG QPR:$src2,
3774 (DSubReg_i32_reg imm:$lane))),
3775 (SubReg_i32_lane imm:$lane)))>;
3777 // VQDMULH : Vector Saturating Doubling Multiply Returning High Half
3778 defm VQDMULH : N3VInt_HS<0, 0, 0b1011, 0, N3RegFrm, IIC_VMULi16D, IIC_VMULi32D,
3779 IIC_VMULi16Q, IIC_VMULi32Q,
3780 "vqdmulh", "s", int_arm_neon_vqdmulh, 1>;
3781 defm VQDMULHsl: N3VIntSL_HS<0b1100, IIC_VMULi16D, IIC_VMULi32D,
3782 IIC_VMULi16Q, IIC_VMULi32Q,
3783 "vqdmulh", "s", int_arm_neon_vqdmulh>;
3784 def : Pat<(v8i16 (int_arm_neon_vqdmulh (v8i16 QPR:$src1),
3785 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
3787 (v8i16 (VQDMULHslv8i16 (v8i16 QPR:$src1),
3788 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3789 (DSubReg_i16_reg imm:$lane))),
3790 (SubReg_i16_lane imm:$lane)))>;
3791 def : Pat<(v4i32 (int_arm_neon_vqdmulh (v4i32 QPR:$src1),
3792 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
3794 (v4i32 (VQDMULHslv4i32 (v4i32 QPR:$src1),
3795 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3796 (DSubReg_i32_reg imm:$lane))),
3797 (SubReg_i32_lane imm:$lane)))>;
3799 // VQRDMULH : Vector Rounding Saturating Doubling Multiply Returning High Half
3800 defm VQRDMULH : N3VInt_HS<1, 0, 0b1011, 0, N3RegFrm,
3801 IIC_VMULi16D,IIC_VMULi32D,IIC_VMULi16Q,IIC_VMULi32Q,
3802 "vqrdmulh", "s", int_arm_neon_vqrdmulh, 1>;
3803 defm VQRDMULHsl : N3VIntSL_HS<0b1101, IIC_VMULi16D, IIC_VMULi32D,
3804 IIC_VMULi16Q, IIC_VMULi32Q,
3805 "vqrdmulh", "s", int_arm_neon_vqrdmulh>;
3806 def : Pat<(v8i16 (int_arm_neon_vqrdmulh (v8i16 QPR:$src1),
3807 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
3809 (v8i16 (VQRDMULHslv8i16 (v8i16 QPR:$src1),
3810 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3811 (DSubReg_i16_reg imm:$lane))),
3812 (SubReg_i16_lane imm:$lane)))>;
3813 def : Pat<(v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$src1),
3814 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
3816 (v4i32 (VQRDMULHslv4i32 (v4i32 QPR:$src1),
3817 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3818 (DSubReg_i32_reg imm:$lane))),
3819 (SubReg_i32_lane imm:$lane)))>;
3821 // VMULL : Vector Multiply Long (integer and polynomial) (Q = D * D)
3822 defm VMULLs : N3VL_QHS<0,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
3823 "vmull", "s", NEONvmulls, 1>;
3824 defm VMULLu : N3VL_QHS<1,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
3825 "vmull", "u", NEONvmullu, 1>;
3826 def VMULLp : N3VLInt<0, 1, 0b00, 0b1110, 0, IIC_VMULi16D, "vmull", "p8",
3827 v8i16, v8i8, int_arm_neon_vmullp, 1>;
3828 defm VMULLsls : N3VLSL_HS<0, 0b1010, IIC_VMULi16D, "vmull", "s", NEONvmulls>;
3829 defm VMULLslu : N3VLSL_HS<1, 0b1010, IIC_VMULi16D, "vmull", "u", NEONvmullu>;
3831 // VQDMULL : Vector Saturating Doubling Multiply Long (Q = D * D)
3832 defm VQDMULL : N3VLInt_HS<0,1,0b1101,0, IIC_VMULi16D, IIC_VMULi32D,
3833 "vqdmull", "s", int_arm_neon_vqdmull, 1>;
3834 defm VQDMULLsl: N3VLIntSL_HS<0, 0b1011, IIC_VMULi16D,
3835 "vqdmull", "s", int_arm_neon_vqdmull>;
3837 // Vector Multiply-Accumulate and Multiply-Subtract Operations.
3839 // VMLA : Vector Multiply Accumulate (integer and floating-point)
3840 defm VMLA : N3VMulOp_QHS<0, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3841 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
3842 def VMLAfd : N3VDMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACD, "vmla", "f32",
3843 v2f32, fmul_su, fadd_mlx>,
3844 Requires<[HasNEON, UseFPVMLx]>;
3845 def VMLAfq : N3VQMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACQ, "vmla", "f32",
3846 v4f32, fmul_su, fadd_mlx>,
3847 Requires<[HasNEON, UseFPVMLx]>;
3848 defm VMLAsl : N3VMulOpSL_HS<0b0000, IIC_VMACi16D, IIC_VMACi32D,
3849 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
3850 def VMLAslfd : N3VDMulOpSL<0b10, 0b0001, IIC_VMACD, "vmla", "f32",
3851 v2f32, fmul_su, fadd_mlx>,
3852 Requires<[HasNEON, UseFPVMLx]>;
3853 def VMLAslfq : N3VQMulOpSL<0b10, 0b0001, IIC_VMACQ, "vmla", "f32",
3854 v4f32, v2f32, fmul_su, fadd_mlx>,
3855 Requires<[HasNEON, UseFPVMLx]>;
3857 def : Pat<(v8i16 (add (v8i16 QPR:$src1),
3858 (mul (v8i16 QPR:$src2),
3859 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
3860 (v8i16 (VMLAslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
3861 (v4i16 (EXTRACT_SUBREG QPR:$src3,
3862 (DSubReg_i16_reg imm:$lane))),
3863 (SubReg_i16_lane imm:$lane)))>;
3865 def : Pat<(v4i32 (add (v4i32 QPR:$src1),
3866 (mul (v4i32 QPR:$src2),
3867 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
3868 (v4i32 (VMLAslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
3869 (v2i32 (EXTRACT_SUBREG QPR:$src3,
3870 (DSubReg_i32_reg imm:$lane))),
3871 (SubReg_i32_lane imm:$lane)))>;
3873 def : Pat<(v4f32 (fadd_mlx (v4f32 QPR:$src1),
3874 (fmul_su (v4f32 QPR:$src2),
3875 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
3876 (v4f32 (VMLAslfq (v4f32 QPR:$src1),
3878 (v2f32 (EXTRACT_SUBREG QPR:$src3,
3879 (DSubReg_i32_reg imm:$lane))),
3880 (SubReg_i32_lane imm:$lane)))>,
3881 Requires<[HasNEON, UseFPVMLx]>;
3883 // VMLAL : Vector Multiply Accumulate Long (Q += D * D)
3884 defm VMLALs : N3VLMulOp_QHS<0,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
3885 "vmlal", "s", NEONvmulls, add>;
3886 defm VMLALu : N3VLMulOp_QHS<1,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
3887 "vmlal", "u", NEONvmullu, add>;
3889 defm VMLALsls : N3VLMulOpSL_HS<0, 0b0010, "vmlal", "s", NEONvmulls, add>;
3890 defm VMLALslu : N3VLMulOpSL_HS<1, 0b0010, "vmlal", "u", NEONvmullu, add>;
3892 // VQDMLAL : Vector Saturating Doubling Multiply Accumulate Long (Q += D * D)
3893 defm VQDMLAL : N3VLInt3_HS<0, 1, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3894 "vqdmlal", "s", int_arm_neon_vqdmlal>;
3895 defm VQDMLALsl: N3VLInt3SL_HS<0, 0b0011, "vqdmlal", "s", int_arm_neon_vqdmlal>;
3897 // VMLS : Vector Multiply Subtract (integer and floating-point)
3898 defm VMLS : N3VMulOp_QHS<1, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3899 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
3900 def VMLSfd : N3VDMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACD, "vmls", "f32",
3901 v2f32, fmul_su, fsub_mlx>,
3902 Requires<[HasNEON, UseFPVMLx]>;
3903 def VMLSfq : N3VQMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACQ, "vmls", "f32",
3904 v4f32, fmul_su, fsub_mlx>,
3905 Requires<[HasNEON, UseFPVMLx]>;
3906 defm VMLSsl : N3VMulOpSL_HS<0b0100, IIC_VMACi16D, IIC_VMACi32D,
3907 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
3908 def VMLSslfd : N3VDMulOpSL<0b10, 0b0101, IIC_VMACD, "vmls", "f32",
3909 v2f32, fmul_su, fsub_mlx>,
3910 Requires<[HasNEON, UseFPVMLx]>;
3911 def VMLSslfq : N3VQMulOpSL<0b10, 0b0101, IIC_VMACQ, "vmls", "f32",
3912 v4f32, v2f32, fmul_su, fsub_mlx>,
3913 Requires<[HasNEON, UseFPVMLx]>;
3915 def : Pat<(v8i16 (sub (v8i16 QPR:$src1),
3916 (mul (v8i16 QPR:$src2),
3917 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
3918 (v8i16 (VMLSslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
3919 (v4i16 (EXTRACT_SUBREG QPR:$src3,
3920 (DSubReg_i16_reg imm:$lane))),
3921 (SubReg_i16_lane imm:$lane)))>;
3923 def : Pat<(v4i32 (sub (v4i32 QPR:$src1),
3924 (mul (v4i32 QPR:$src2),
3925 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
3926 (v4i32 (VMLSslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
3927 (v2i32 (EXTRACT_SUBREG QPR:$src3,
3928 (DSubReg_i32_reg imm:$lane))),
3929 (SubReg_i32_lane imm:$lane)))>;
3931 def : Pat<(v4f32 (fsub_mlx (v4f32 QPR:$src1),
3932 (fmul_su (v4f32 QPR:$src2),
3933 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
3934 (v4f32 (VMLSslfq (v4f32 QPR:$src1), (v4f32 QPR:$src2),
3935 (v2f32 (EXTRACT_SUBREG QPR:$src3,
3936 (DSubReg_i32_reg imm:$lane))),
3937 (SubReg_i32_lane imm:$lane)))>,
3938 Requires<[HasNEON, UseFPVMLx]>;
3940 // VMLSL : Vector Multiply Subtract Long (Q -= D * D)
3941 defm VMLSLs : N3VLMulOp_QHS<0,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
3942 "vmlsl", "s", NEONvmulls, sub>;
3943 defm VMLSLu : N3VLMulOp_QHS<1,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
3944 "vmlsl", "u", NEONvmullu, sub>;
3946 defm VMLSLsls : N3VLMulOpSL_HS<0, 0b0110, "vmlsl", "s", NEONvmulls, sub>;
3947 defm VMLSLslu : N3VLMulOpSL_HS<1, 0b0110, "vmlsl", "u", NEONvmullu, sub>;
3949 // VQDMLSL : Vector Saturating Doubling Multiply Subtract Long (Q -= D * D)
3950 defm VQDMLSL : N3VLInt3_HS<0, 1, 0b1011, 0, IIC_VMACi16D, IIC_VMACi32D,
3951 "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
3952 defm VQDMLSLsl: N3VLInt3SL_HS<0, 0b111, "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
3954 // Vector Subtract Operations.
3956 // VSUB : Vector Subtract (integer and floating-point)
3957 defm VSUB : N3V_QHSD<1, 0, 0b1000, 0, IIC_VSUBiD, IIC_VSUBiQ,
3958 "vsub", "i", sub, 0>;
3959 def VSUBfd : N3VD<0, 0, 0b10, 0b1101, 0, IIC_VBIND, "vsub", "f32",
3960 v2f32, v2f32, fsub, 0>;
3961 def VSUBfq : N3VQ<0, 0, 0b10, 0b1101, 0, IIC_VBINQ, "vsub", "f32",
3962 v4f32, v4f32, fsub, 0>;
3963 // VSUBL : Vector Subtract Long (Q = D - D)
3964 defm VSUBLs : N3VLExt_QHS<0,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
3965 "vsubl", "s", sub, sext, 0>;
3966 defm VSUBLu : N3VLExt_QHS<1,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
3967 "vsubl", "u", sub, zext, 0>;
3968 // VSUBW : Vector Subtract Wide (Q = Q - D)
3969 defm VSUBWs : N3VW_QHS<0,1,0b0011,0, "vsubw", "s", sub, sext, 0>;
3970 defm VSUBWu : N3VW_QHS<1,1,0b0011,0, "vsubw", "u", sub, zext, 0>;
3971 // VHSUB : Vector Halving Subtract
3972 defm VHSUBs : N3VInt_QHS<0, 0, 0b0010, 0, N3RegFrm,
3973 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3974 "vhsub", "s", int_arm_neon_vhsubs, 0>;
3975 defm VHSUBu : N3VInt_QHS<1, 0, 0b0010, 0, N3RegFrm,
3976 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3977 "vhsub", "u", int_arm_neon_vhsubu, 0>;
3978 // VQSUB : Vector Saturing Subtract
3979 defm VQSUBs : N3VInt_QHSD<0, 0, 0b0010, 1, N3RegFrm,
3980 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3981 "vqsub", "s", int_arm_neon_vqsubs, 0>;
3982 defm VQSUBu : N3VInt_QHSD<1, 0, 0b0010, 1, N3RegFrm,
3983 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3984 "vqsub", "u", int_arm_neon_vqsubu, 0>;
3985 // VSUBHN : Vector Subtract and Narrow Returning High Half (D = Q - Q)
3986 defm VSUBHN : N3VNInt_HSD<0,1,0b0110,0, "vsubhn", "i",
3987 int_arm_neon_vsubhn, 0>;
3988 // VRSUBHN : Vector Rounding Subtract and Narrow Returning High Half (D=Q-Q)
3989 defm VRSUBHN : N3VNInt_HSD<1,1,0b0110,0, "vrsubhn", "i",
3990 int_arm_neon_vrsubhn, 0>;
3992 // Vector Comparisons.
3994 // VCEQ : Vector Compare Equal
3995 defm VCEQ : N3V_QHS<1, 0, 0b1000, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3996 IIC_VSUBi4Q, "vceq", "i", NEONvceq, 1>;
3997 def VCEQfd : N3VD<0,0,0b00,0b1110,0, IIC_VBIND, "vceq", "f32", v2i32, v2f32,
3999 def VCEQfq : N3VQ<0,0,0b00,0b1110,0, IIC_VBINQ, "vceq", "f32", v4i32, v4f32,
4002 defm VCEQz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00010, 0, "vceq", "i",
4003 "$Vd, $Vm, #0", NEONvceqz>;
4005 // VCGE : Vector Compare Greater Than or Equal
4006 defm VCGEs : N3V_QHS<0, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
4007 IIC_VSUBi4Q, "vcge", "s", NEONvcge, 0>;
4008 defm VCGEu : N3V_QHS<1, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
4009 IIC_VSUBi4Q, "vcge", "u", NEONvcgeu, 0>;
4010 def VCGEfd : N3VD<1,0,0b00,0b1110,0, IIC_VBIND, "vcge", "f32", v2i32, v2f32,
4012 def VCGEfq : N3VQ<1,0,0b00,0b1110,0, IIC_VBINQ, "vcge", "f32", v4i32, v4f32,
4015 defm VCGEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00001, 0, "vcge", "s",
4016 "$Vd, $Vm, #0", NEONvcgez>;
4017 defm VCLEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00011, 0, "vcle", "s",
4018 "$Vd, $Vm, #0", NEONvclez>;
4020 // VCGT : Vector Compare Greater Than
4021 defm VCGTs : N3V_QHS<0, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
4022 IIC_VSUBi4Q, "vcgt", "s", NEONvcgt, 0>;
4023 defm VCGTu : N3V_QHS<1, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
4024 IIC_VSUBi4Q, "vcgt", "u", NEONvcgtu, 0>;
4025 def VCGTfd : N3VD<1,0,0b10,0b1110,0, IIC_VBIND, "vcgt", "f32", v2i32, v2f32,
4027 def VCGTfq : N3VQ<1,0,0b10,0b1110,0, IIC_VBINQ, "vcgt", "f32", v4i32, v4f32,
4030 defm VCGTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00000, 0, "vcgt", "s",
4031 "$Vd, $Vm, #0", NEONvcgtz>;
4032 defm VCLTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00100, 0, "vclt", "s",
4033 "$Vd, $Vm, #0", NEONvcltz>;
4035 // VACGE : Vector Absolute Compare Greater Than or Equal (aka VCAGE)
4036 def VACGEd : N3VDInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacge",
4037 "f32", v2i32, v2f32, int_arm_neon_vacged, 0>;
4038 def VACGEq : N3VQInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacge",
4039 "f32", v4i32, v4f32, int_arm_neon_vacgeq, 0>;
4040 // VACGT : Vector Absolute Compare Greater Than (aka VCAGT)
4041 def VACGTd : N3VDInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacgt",
4042 "f32", v2i32, v2f32, int_arm_neon_vacgtd, 0>;
4043 def VACGTq : N3VQInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacgt",
4044 "f32", v4i32, v4f32, int_arm_neon_vacgtq, 0>;
4045 // VTST : Vector Test Bits
4046 defm VTST : N3V_QHS<0, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
4047 IIC_VBINi4Q, "vtst", "", NEONvtst, 1>;
4049 // Vector Bitwise Operations.
4051 def vnotd : PatFrag<(ops node:$in),
4052 (xor node:$in, (bitconvert (v8i8 NEONimmAllOnesV)))>;
4053 def vnotq : PatFrag<(ops node:$in),
4054 (xor node:$in, (bitconvert (v16i8 NEONimmAllOnesV)))>;
4057 // VAND : Vector Bitwise AND
4058 def VANDd : N3VDX<0, 0, 0b00, 0b0001, 1, IIC_VBINiD, "vand",
4059 v2i32, v2i32, and, 1>;
4060 def VANDq : N3VQX<0, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "vand",
4061 v4i32, v4i32, and, 1>;
4063 // VEOR : Vector Bitwise Exclusive OR
4064 def VEORd : N3VDX<1, 0, 0b00, 0b0001, 1, IIC_VBINiD, "veor",
4065 v2i32, v2i32, xor, 1>;
4066 def VEORq : N3VQX<1, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "veor",
4067 v4i32, v4i32, xor, 1>;
4069 // VORR : Vector Bitwise OR
4070 def VORRd : N3VDX<0, 0, 0b10, 0b0001, 1, IIC_VBINiD, "vorr",
4071 v2i32, v2i32, or, 1>;
4072 def VORRq : N3VQX<0, 0, 0b10, 0b0001, 1, IIC_VBINiQ, "vorr",
4073 v4i32, v4i32, or, 1>;
4075 def VORRiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 0, 1,
4076 (outs DPR:$Vd), (ins nImmSplatI16:$SIMM, DPR:$src),
4078 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
4080 (v4i16 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
4081 let Inst{9} = SIMM{9};
4084 def VORRiv2i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 0, 0, 1,
4085 (outs DPR:$Vd), (ins nImmSplatI32:$SIMM, DPR:$src),
4087 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
4089 (v2i32 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
4090 let Inst{10-9} = SIMM{10-9};
4093 def VORRiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 0, 1,
4094 (outs QPR:$Vd), (ins nImmSplatI16:$SIMM, QPR:$src),
4096 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
4098 (v8i16 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
4099 let Inst{9} = SIMM{9};
4102 def VORRiv4i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 1, 0, 1,
4103 (outs QPR:$Vd), (ins nImmSplatI32:$SIMM, QPR:$src),
4105 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
4107 (v4i32 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
4108 let Inst{10-9} = SIMM{10-9};
4112 // VBIC : Vector Bitwise Bit Clear (AND NOT)
4113 def VBICd : N3VX<0, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
4114 (ins DPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VBINiD,
4115 "vbic", "$Vd, $Vn, $Vm", "",
4116 [(set DPR:$Vd, (v2i32 (and DPR:$Vn,
4117 (vnotd DPR:$Vm))))]>;
4118 def VBICq : N3VX<0, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
4119 (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINiQ,
4120 "vbic", "$Vd, $Vn, $Vm", "",
4121 [(set QPR:$Vd, (v4i32 (and QPR:$Vn,
4122 (vnotq QPR:$Vm))))]>;
4124 def VBICiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 1, 1,
4125 (outs DPR:$Vd), (ins nImmSplatI16:$SIMM, DPR:$src),
4127 "vbic", "i16", "$Vd, $SIMM", "$src = $Vd",
4129 (v4i16 (NEONvbicImm DPR:$src, timm:$SIMM)))]> {
4130 let Inst{9} = SIMM{9};
4133 def VBICiv2i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 0, 1, 1,
4134 (outs DPR:$Vd), (ins nImmSplatI32:$SIMM, DPR:$src),
4136 "vbic", "i32", "$Vd, $SIMM", "$src = $Vd",
4138 (v2i32 (NEONvbicImm DPR:$src, timm:$SIMM)))]> {
4139 let Inst{10-9} = SIMM{10-9};
4142 def VBICiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 1, 1,
4143 (outs QPR:$Vd), (ins nImmSplatI16:$SIMM, QPR:$src),
4145 "vbic", "i16", "$Vd, $SIMM", "$src = $Vd",
4147 (v8i16 (NEONvbicImm QPR:$src, timm:$SIMM)))]> {
4148 let Inst{9} = SIMM{9};
4151 def VBICiv4i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 1, 1, 1,
4152 (outs QPR:$Vd), (ins nImmSplatI32:$SIMM, QPR:$src),
4154 "vbic", "i32", "$Vd, $SIMM", "$src = $Vd",
4156 (v4i32 (NEONvbicImm QPR:$src, timm:$SIMM)))]> {
4157 let Inst{10-9} = SIMM{10-9};
4160 // VORN : Vector Bitwise OR NOT
4161 def VORNd : N3VX<0, 0, 0b11, 0b0001, 0, 1, (outs DPR:$Vd),
4162 (ins DPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VBINiD,
4163 "vorn", "$Vd, $Vn, $Vm", "",
4164 [(set DPR:$Vd, (v2i32 (or DPR:$Vn,
4165 (vnotd DPR:$Vm))))]>;
4166 def VORNq : N3VX<0, 0, 0b11, 0b0001, 1, 1, (outs QPR:$Vd),
4167 (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINiQ,
4168 "vorn", "$Vd, $Vn, $Vm", "",
4169 [(set QPR:$Vd, (v4i32 (or QPR:$Vn,
4170 (vnotq QPR:$Vm))))]>;
4172 // VMVN : Vector Bitwise NOT (Immediate)
4174 let isReMaterializable = 1 in {
4176 def VMVNv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 1, 1, (outs DPR:$Vd),
4177 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4178 "vmvn", "i16", "$Vd, $SIMM", "",
4179 [(set DPR:$Vd, (v4i16 (NEONvmvnImm timm:$SIMM)))]> {
4180 let Inst{9} = SIMM{9};
4183 def VMVNv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 1, 1, (outs QPR:$Vd),
4184 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4185 "vmvn", "i16", "$Vd, $SIMM", "",
4186 [(set QPR:$Vd, (v8i16 (NEONvmvnImm timm:$SIMM)))]> {
4187 let Inst{9} = SIMM{9};
4190 def VMVNv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 1, 1, (outs DPR:$Vd),
4191 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4192 "vmvn", "i32", "$Vd, $SIMM", "",
4193 [(set DPR:$Vd, (v2i32 (NEONvmvnImm timm:$SIMM)))]> {
4194 let Inst{11-8} = SIMM{11-8};
4197 def VMVNv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 1, 1, (outs QPR:$Vd),
4198 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4199 "vmvn", "i32", "$Vd, $SIMM", "",
4200 [(set QPR:$Vd, (v4i32 (NEONvmvnImm timm:$SIMM)))]> {
4201 let Inst{11-8} = SIMM{11-8};
4205 // VMVN : Vector Bitwise NOT
4206 def VMVNd : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 0, 0,
4207 (outs DPR:$Vd), (ins DPR:$Vm), IIC_VSUBiD,
4208 "vmvn", "$Vd, $Vm", "",
4209 [(set DPR:$Vd, (v2i32 (vnotd DPR:$Vm)))]>;
4210 def VMVNq : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 1, 0,
4211 (outs QPR:$Vd), (ins QPR:$Vm), IIC_VSUBiD,
4212 "vmvn", "$Vd, $Vm", "",
4213 [(set QPR:$Vd, (v4i32 (vnotq QPR:$Vm)))]>;
4214 def : Pat<(v2i32 (vnotd DPR:$src)), (VMVNd DPR:$src)>;
4215 def : Pat<(v4i32 (vnotq QPR:$src)), (VMVNq QPR:$src)>;
4217 // VBSL : Vector Bitwise Select
4218 def VBSLd : N3VX<1, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
4219 (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
4220 N3RegFrm, IIC_VCNTiD,
4221 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4223 (v2i32 (NEONvbsl DPR:$src1, DPR:$Vn, DPR:$Vm)))]>;
4225 def : Pat<(v2i32 (or (and DPR:$Vn, DPR:$Vd),
4226 (and DPR:$Vm, (vnotd DPR:$Vd)))),
4227 (VBSLd DPR:$Vd, DPR:$Vn, DPR:$Vm)>;
4229 def VBSLq : N3VX<1, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
4230 (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
4231 N3RegFrm, IIC_VCNTiQ,
4232 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4234 (v4i32 (NEONvbsl QPR:$src1, QPR:$Vn, QPR:$Vm)))]>;
4236 def : Pat<(v4i32 (or (and QPR:$Vn, QPR:$Vd),
4237 (and QPR:$Vm, (vnotq QPR:$Vd)))),
4238 (VBSLq QPR:$Vd, QPR:$Vn, QPR:$Vm)>;
4240 // VBIF : Vector Bitwise Insert if False
4241 // like VBSL but with: "vbif $dst, $src3, $src1", "$src2 = $dst",
4242 // FIXME: This instruction's encoding MAY NOT BE correct.
4243 def VBIFd : N3VX<1, 0, 0b11, 0b0001, 0, 1,
4244 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
4245 N3RegFrm, IIC_VBINiD,
4246 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4248 def VBIFq : N3VX<1, 0, 0b11, 0b0001, 1, 1,
4249 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
4250 N3RegFrm, IIC_VBINiQ,
4251 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4254 // VBIT : Vector Bitwise Insert if True
4255 // like VBSL but with: "vbit $dst, $src2, $src1", "$src3 = $dst",
4256 // FIXME: This instruction's encoding MAY NOT BE correct.
4257 def VBITd : N3VX<1, 0, 0b10, 0b0001, 0, 1,
4258 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
4259 N3RegFrm, IIC_VBINiD,
4260 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4262 def VBITq : N3VX<1, 0, 0b10, 0b0001, 1, 1,
4263 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
4264 N3RegFrm, IIC_VBINiQ,
4265 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4268 // VBIT/VBIF are not yet implemented. The TwoAddress pass will not go looking
4269 // for equivalent operations with different register constraints; it just
4272 // Vector Absolute Differences.
4274 // VABD : Vector Absolute Difference
4275 defm VABDs : N3VInt_QHS<0, 0, 0b0111, 0, N3RegFrm,
4276 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4277 "vabd", "s", int_arm_neon_vabds, 1>;
4278 defm VABDu : N3VInt_QHS<1, 0, 0b0111, 0, N3RegFrm,
4279 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4280 "vabd", "u", int_arm_neon_vabdu, 1>;
4281 def VABDfd : N3VDInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBIND,
4282 "vabd", "f32", v2f32, v2f32, int_arm_neon_vabds, 1>;
4283 def VABDfq : N3VQInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBINQ,
4284 "vabd", "f32", v4f32, v4f32, int_arm_neon_vabds, 1>;
4286 // VABDL : Vector Absolute Difference Long (Q = | D - D |)
4287 defm VABDLs : N3VLIntExt_QHS<0,1,0b0111,0, IIC_VSUBi4Q,
4288 "vabdl", "s", int_arm_neon_vabds, zext, 1>;
4289 defm VABDLu : N3VLIntExt_QHS<1,1,0b0111,0, IIC_VSUBi4Q,
4290 "vabdl", "u", int_arm_neon_vabdu, zext, 1>;
4292 // VABA : Vector Absolute Difference and Accumulate
4293 defm VABAs : N3VIntOp_QHS<0,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
4294 "vaba", "s", int_arm_neon_vabds, add>;
4295 defm VABAu : N3VIntOp_QHS<1,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
4296 "vaba", "u", int_arm_neon_vabdu, add>;
4298 // VABAL : Vector Absolute Difference and Accumulate Long (Q += | D - D |)
4299 defm VABALs : N3VLIntExtOp_QHS<0,1,0b0101,0, IIC_VABAD,
4300 "vabal", "s", int_arm_neon_vabds, zext, add>;
4301 defm VABALu : N3VLIntExtOp_QHS<1,1,0b0101,0, IIC_VABAD,
4302 "vabal", "u", int_arm_neon_vabdu, zext, add>;
4304 // Vector Maximum and Minimum.
4306 // VMAX : Vector Maximum
4307 defm VMAXs : N3VInt_QHS<0, 0, 0b0110, 0, N3RegFrm,
4308 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4309 "vmax", "s", int_arm_neon_vmaxs, 1>;
4310 defm VMAXu : N3VInt_QHS<1, 0, 0b0110, 0, N3RegFrm,
4311 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4312 "vmax", "u", int_arm_neon_vmaxu, 1>;
4313 def VMAXfd : N3VDInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBIND,
4315 v2f32, v2f32, int_arm_neon_vmaxs, 1>;
4316 def VMAXfq : N3VQInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBINQ,
4318 v4f32, v4f32, int_arm_neon_vmaxs, 1>;
4320 // VMIN : Vector Minimum
4321 defm VMINs : N3VInt_QHS<0, 0, 0b0110, 1, N3RegFrm,
4322 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4323 "vmin", "s", int_arm_neon_vmins, 1>;
4324 defm VMINu : N3VInt_QHS<1, 0, 0b0110, 1, N3RegFrm,
4325 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4326 "vmin", "u", int_arm_neon_vminu, 1>;
4327 def VMINfd : N3VDInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBIND,
4329 v2f32, v2f32, int_arm_neon_vmins, 1>;
4330 def VMINfq : N3VQInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBINQ,
4332 v4f32, v4f32, int_arm_neon_vmins, 1>;
4334 // Vector Pairwise Operations.
4336 // VPADD : Vector Pairwise Add
4337 def VPADDi8 : N3VDInt<0, 0, 0b00, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4339 v8i8, v8i8, int_arm_neon_vpadd, 0>;
4340 def VPADDi16 : N3VDInt<0, 0, 0b01, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4342 v4i16, v4i16, int_arm_neon_vpadd, 0>;
4343 def VPADDi32 : N3VDInt<0, 0, 0b10, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4345 v2i32, v2i32, int_arm_neon_vpadd, 0>;
4346 def VPADDf : N3VDInt<1, 0, 0b00, 0b1101, 0, N3RegFrm,
4347 IIC_VPBIND, "vpadd", "f32",
4348 v2f32, v2f32, int_arm_neon_vpadd, 0>;
4350 // VPADDL : Vector Pairwise Add Long
4351 defm VPADDLs : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00100, 0, "vpaddl", "s",
4352 int_arm_neon_vpaddls>;
4353 defm VPADDLu : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00101, 0, "vpaddl", "u",
4354 int_arm_neon_vpaddlu>;
4356 // VPADAL : Vector Pairwise Add and Accumulate Long
4357 defm VPADALs : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01100, 0, "vpadal", "s",
4358 int_arm_neon_vpadals>;
4359 defm VPADALu : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01101, 0, "vpadal", "u",
4360 int_arm_neon_vpadalu>;
4362 // VPMAX : Vector Pairwise Maximum
4363 def VPMAXs8 : N3VDInt<0, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4364 "s8", v8i8, v8i8, int_arm_neon_vpmaxs, 0>;
4365 def VPMAXs16 : N3VDInt<0, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4366 "s16", v4i16, v4i16, int_arm_neon_vpmaxs, 0>;
4367 def VPMAXs32 : N3VDInt<0, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4368 "s32", v2i32, v2i32, int_arm_neon_vpmaxs, 0>;
4369 def VPMAXu8 : N3VDInt<1, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4370 "u8", v8i8, v8i8, int_arm_neon_vpmaxu, 0>;
4371 def VPMAXu16 : N3VDInt<1, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4372 "u16", v4i16, v4i16, int_arm_neon_vpmaxu, 0>;
4373 def VPMAXu32 : N3VDInt<1, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4374 "u32", v2i32, v2i32, int_arm_neon_vpmaxu, 0>;
4375 def VPMAXf : N3VDInt<1, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmax",
4376 "f32", v2f32, v2f32, int_arm_neon_vpmaxs, 0>;
4378 // VPMIN : Vector Pairwise Minimum
4379 def VPMINs8 : N3VDInt<0, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4380 "s8", v8i8, v8i8, int_arm_neon_vpmins, 0>;
4381 def VPMINs16 : N3VDInt<0, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4382 "s16", v4i16, v4i16, int_arm_neon_vpmins, 0>;
4383 def VPMINs32 : N3VDInt<0, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4384 "s32", v2i32, v2i32, int_arm_neon_vpmins, 0>;
4385 def VPMINu8 : N3VDInt<1, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4386 "u8", v8i8, v8i8, int_arm_neon_vpminu, 0>;
4387 def VPMINu16 : N3VDInt<1, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4388 "u16", v4i16, v4i16, int_arm_neon_vpminu, 0>;
4389 def VPMINu32 : N3VDInt<1, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4390 "u32", v2i32, v2i32, int_arm_neon_vpminu, 0>;
4391 def VPMINf : N3VDInt<1, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmin",
4392 "f32", v2f32, v2f32, int_arm_neon_vpmins, 0>;
4394 // Vector Reciprocal and Reciprocal Square Root Estimate and Step.
4396 // VRECPE : Vector Reciprocal Estimate
4397 def VRECPEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
4398 IIC_VUNAD, "vrecpe", "u32",
4399 v2i32, v2i32, int_arm_neon_vrecpe>;
4400 def VRECPEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
4401 IIC_VUNAQ, "vrecpe", "u32",
4402 v4i32, v4i32, int_arm_neon_vrecpe>;
4403 def VRECPEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
4404 IIC_VUNAD, "vrecpe", "f32",
4405 v2f32, v2f32, int_arm_neon_vrecpe>;
4406 def VRECPEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
4407 IIC_VUNAQ, "vrecpe", "f32",
4408 v4f32, v4f32, int_arm_neon_vrecpe>;
4410 // VRECPS : Vector Reciprocal Step
4411 def VRECPSfd : N3VDInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
4412 IIC_VRECSD, "vrecps", "f32",
4413 v2f32, v2f32, int_arm_neon_vrecps, 1>;
4414 def VRECPSfq : N3VQInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
4415 IIC_VRECSQ, "vrecps", "f32",
4416 v4f32, v4f32, int_arm_neon_vrecps, 1>;
4418 // VRSQRTE : Vector Reciprocal Square Root Estimate
4419 def VRSQRTEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
4420 IIC_VUNAD, "vrsqrte", "u32",
4421 v2i32, v2i32, int_arm_neon_vrsqrte>;
4422 def VRSQRTEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
4423 IIC_VUNAQ, "vrsqrte", "u32",
4424 v4i32, v4i32, int_arm_neon_vrsqrte>;
4425 def VRSQRTEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
4426 IIC_VUNAD, "vrsqrte", "f32",
4427 v2f32, v2f32, int_arm_neon_vrsqrte>;
4428 def VRSQRTEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
4429 IIC_VUNAQ, "vrsqrte", "f32",
4430 v4f32, v4f32, int_arm_neon_vrsqrte>;
4432 // VRSQRTS : Vector Reciprocal Square Root Step
4433 def VRSQRTSfd : N3VDInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
4434 IIC_VRECSD, "vrsqrts", "f32",
4435 v2f32, v2f32, int_arm_neon_vrsqrts, 1>;
4436 def VRSQRTSfq : N3VQInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
4437 IIC_VRECSQ, "vrsqrts", "f32",
4438 v4f32, v4f32, int_arm_neon_vrsqrts, 1>;
4442 // VSHL : Vector Shift
4443 defm VSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 0, N3RegVShFrm,
4444 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
4445 "vshl", "s", int_arm_neon_vshifts>;
4446 defm VSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 0, N3RegVShFrm,
4447 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
4448 "vshl", "u", int_arm_neon_vshiftu>;
4450 // VSHL : Vector Shift Left (Immediate)
4451 defm VSHLi : N2VShL_QHSD<0, 1, 0b0101, 1, IIC_VSHLiD, "vshl", "i", NEONvshl>;
4453 // VSHR : Vector Shift Right (Immediate)
4454 defm VSHRs : N2VShR_QHSD<0, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "s",NEONvshrs>;
4455 defm VSHRu : N2VShR_QHSD<1, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "u",NEONvshru>;
4457 // VSHLL : Vector Shift Left Long
4458 defm VSHLLs : N2VLSh_QHS<0, 1, 0b1010, 0, 0, 1, "vshll", "s", NEONvshlls>;
4459 defm VSHLLu : N2VLSh_QHS<1, 1, 0b1010, 0, 0, 1, "vshll", "u", NEONvshllu>;
4461 // VSHLL : Vector Shift Left Long (with maximum shift count)
4462 class N2VLShMax<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
4463 bit op6, bit op4, string OpcodeStr, string Dt, ValueType ResTy,
4464 ValueType OpTy, Operand ImmTy, SDNode OpNode>
4465 : N2VLSh<op24, op23, op11_8, op7, op6, op4, OpcodeStr, Dt,
4466 ResTy, OpTy, ImmTy, OpNode> {
4467 let Inst{21-16} = op21_16;
4468 let DecoderMethod = "DecodeVSHLMaxInstruction";
4470 def VSHLLi8 : N2VLShMax<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll", "i8",
4471 v8i16, v8i8, imm8, NEONvshlli>;
4472 def VSHLLi16 : N2VLShMax<1, 1, 0b110110, 0b0011, 0, 0, 0, "vshll", "i16",
4473 v4i32, v4i16, imm16, NEONvshlli>;
4474 def VSHLLi32 : N2VLShMax<1, 1, 0b111010, 0b0011, 0, 0, 0, "vshll", "i32",
4475 v2i64, v2i32, imm32, NEONvshlli>;
4477 // VSHRN : Vector Shift Right and Narrow
4478 defm VSHRN : N2VNSh_HSD<0,1,0b1000,0,0,1, IIC_VSHLiD, "vshrn", "i",
4481 // VRSHL : Vector Rounding Shift
4482 defm VRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 0, N3RegVShFrm,
4483 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4484 "vrshl", "s", int_arm_neon_vrshifts>;
4485 defm VRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 0, N3RegVShFrm,
4486 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4487 "vrshl", "u", int_arm_neon_vrshiftu>;
4488 // VRSHR : Vector Rounding Shift Right
4489 defm VRSHRs : N2VShR_QHSD<0,1,0b0010,1, IIC_VSHLi4D, "vrshr", "s",NEONvrshrs>;
4490 defm VRSHRu : N2VShR_QHSD<1,1,0b0010,1, IIC_VSHLi4D, "vrshr", "u",NEONvrshru>;
4492 // VRSHRN : Vector Rounding Shift Right and Narrow
4493 defm VRSHRN : N2VNSh_HSD<0, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vrshrn", "i",
4496 // VQSHL : Vector Saturating Shift
4497 defm VQSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 1, N3RegVShFrm,
4498 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4499 "vqshl", "s", int_arm_neon_vqshifts>;
4500 defm VQSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 1, N3RegVShFrm,
4501 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4502 "vqshl", "u", int_arm_neon_vqshiftu>;
4503 // VQSHL : Vector Saturating Shift Left (Immediate)
4504 defm VQSHLsi : N2VShL_QHSD<0,1,0b0111,1, IIC_VSHLi4D, "vqshl", "s",NEONvqshls>;
4505 defm VQSHLui : N2VShL_QHSD<1,1,0b0111,1, IIC_VSHLi4D, "vqshl", "u",NEONvqshlu>;
4507 // VQSHLU : Vector Saturating Shift Left (Immediate, Unsigned)
4508 defm VQSHLsu : N2VShL_QHSD<1,1,0b0110,1, IIC_VSHLi4D,"vqshlu","s",NEONvqshlsu>;
4510 // VQSHRN : Vector Saturating Shift Right and Narrow
4511 defm VQSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "s",
4513 defm VQSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "u",
4516 // VQSHRUN : Vector Saturating Shift Right and Narrow (Unsigned)
4517 defm VQSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 0, 1, IIC_VSHLi4D, "vqshrun", "s",
4520 // VQRSHL : Vector Saturating Rounding Shift
4521 defm VQRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 1, N3RegVShFrm,
4522 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4523 "vqrshl", "s", int_arm_neon_vqrshifts>;
4524 defm VQRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 1, N3RegVShFrm,
4525 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4526 "vqrshl", "u", int_arm_neon_vqrshiftu>;
4528 // VQRSHRN : Vector Saturating Rounding Shift Right and Narrow
4529 defm VQRSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "s",
4531 defm VQRSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "u",
4534 // VQRSHRUN : Vector Saturating Rounding Shift Right and Narrow (Unsigned)
4535 defm VQRSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vqrshrun", "s",
4538 // VSRA : Vector Shift Right and Accumulate
4539 defm VSRAs : N2VShAdd_QHSD<0, 1, 0b0001, 1, "vsra", "s", NEONvshrs>;
4540 defm VSRAu : N2VShAdd_QHSD<1, 1, 0b0001, 1, "vsra", "u", NEONvshru>;
4541 // VRSRA : Vector Rounding Shift Right and Accumulate
4542 defm VRSRAs : N2VShAdd_QHSD<0, 1, 0b0011, 1, "vrsra", "s", NEONvrshrs>;
4543 defm VRSRAu : N2VShAdd_QHSD<1, 1, 0b0011, 1, "vrsra", "u", NEONvrshru>;
4545 // VSLI : Vector Shift Left and Insert
4546 defm VSLI : N2VShInsL_QHSD<1, 1, 0b0101, 1, "vsli">;
4548 // VSRI : Vector Shift Right and Insert
4549 defm VSRI : N2VShInsR_QHSD<1, 1, 0b0100, 1, "vsri">;
4551 // Vector Absolute and Saturating Absolute.
4553 // VABS : Vector Absolute Value
4554 defm VABS : N2VInt_QHS<0b11, 0b11, 0b01, 0b00110, 0,
4555 IIC_VUNAiD, IIC_VUNAiQ, "vabs", "s",
4557 def VABSfd : N2VDInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
4558 IIC_VUNAD, "vabs", "f32",
4559 v2f32, v2f32, int_arm_neon_vabs>;
4560 def VABSfq : N2VQInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
4561 IIC_VUNAQ, "vabs", "f32",
4562 v4f32, v4f32, int_arm_neon_vabs>;
4564 // VQABS : Vector Saturating Absolute Value
4565 defm VQABS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01110, 0,
4566 IIC_VQUNAiD, IIC_VQUNAiQ, "vqabs", "s",
4567 int_arm_neon_vqabs>;
4571 def vnegd : PatFrag<(ops node:$in),
4572 (sub (bitconvert (v2i32 NEONimmAllZerosV)), node:$in)>;
4573 def vnegq : PatFrag<(ops node:$in),
4574 (sub (bitconvert (v4i32 NEONimmAllZerosV)), node:$in)>;
4576 class VNEGD<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
4577 : N2V<0b11, 0b11, size, 0b01, 0b00111, 0, 0, (outs DPR:$Vd), (ins DPR:$Vm),
4578 IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
4579 [(set DPR:$Vd, (Ty (vnegd DPR:$Vm)))]>;
4580 class VNEGQ<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
4581 : N2V<0b11, 0b11, size, 0b01, 0b00111, 1, 0, (outs QPR:$Vd), (ins QPR:$Vm),
4582 IIC_VSHLiQ, OpcodeStr, Dt, "$Vd, $Vm", "",
4583 [(set QPR:$Vd, (Ty (vnegq QPR:$Vm)))]>;
4585 // VNEG : Vector Negate (integer)
4586 def VNEGs8d : VNEGD<0b00, "vneg", "s8", v8i8>;
4587 def VNEGs16d : VNEGD<0b01, "vneg", "s16", v4i16>;
4588 def VNEGs32d : VNEGD<0b10, "vneg", "s32", v2i32>;
4589 def VNEGs8q : VNEGQ<0b00, "vneg", "s8", v16i8>;
4590 def VNEGs16q : VNEGQ<0b01, "vneg", "s16", v8i16>;
4591 def VNEGs32q : VNEGQ<0b10, "vneg", "s32", v4i32>;
4593 // VNEG : Vector Negate (floating-point)
4594 def VNEGfd : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
4595 (outs DPR:$Vd), (ins DPR:$Vm), IIC_VUNAD,
4596 "vneg", "f32", "$Vd, $Vm", "",
4597 [(set DPR:$Vd, (v2f32 (fneg DPR:$Vm)))]>;
4598 def VNEGf32q : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 1, 0,
4599 (outs QPR:$Vd), (ins QPR:$Vm), IIC_VUNAQ,
4600 "vneg", "f32", "$Vd, $Vm", "",
4601 [(set QPR:$Vd, (v4f32 (fneg QPR:$Vm)))]>;
4603 def : Pat<(v8i8 (vnegd DPR:$src)), (VNEGs8d DPR:$src)>;
4604 def : Pat<(v4i16 (vnegd DPR:$src)), (VNEGs16d DPR:$src)>;
4605 def : Pat<(v2i32 (vnegd DPR:$src)), (VNEGs32d DPR:$src)>;
4606 def : Pat<(v16i8 (vnegq QPR:$src)), (VNEGs8q QPR:$src)>;
4607 def : Pat<(v8i16 (vnegq QPR:$src)), (VNEGs16q QPR:$src)>;
4608 def : Pat<(v4i32 (vnegq QPR:$src)), (VNEGs32q QPR:$src)>;
4610 // VQNEG : Vector Saturating Negate
4611 defm VQNEG : N2VInt_QHS<0b11, 0b11, 0b00, 0b01111, 0,
4612 IIC_VQUNAiD, IIC_VQUNAiQ, "vqneg", "s",
4613 int_arm_neon_vqneg>;
4615 // Vector Bit Counting Operations.
4617 // VCLS : Vector Count Leading Sign Bits
4618 defm VCLS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01000, 0,
4619 IIC_VCNTiD, IIC_VCNTiQ, "vcls", "s",
4621 // VCLZ : Vector Count Leading Zeros
4622 defm VCLZ : N2VInt_QHS<0b11, 0b11, 0b00, 0b01001, 0,
4623 IIC_VCNTiD, IIC_VCNTiQ, "vclz", "i",
4625 // VCNT : Vector Count One Bits
4626 def VCNTd : N2VDInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
4627 IIC_VCNTiD, "vcnt", "8",
4628 v8i8, v8i8, int_arm_neon_vcnt>;
4629 def VCNTq : N2VQInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
4630 IIC_VCNTiQ, "vcnt", "8",
4631 v16i8, v16i8, int_arm_neon_vcnt>;
4634 def VSWPd : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 0, 0,
4635 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
4636 "vswp", "$Vd, $Vm", "", []>;
4637 def VSWPq : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 1, 0,
4638 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
4639 "vswp", "$Vd, $Vm", "", []>;
4641 // Vector Move Operations.
4643 // VMOV : Vector Move (Register)
4644 def : InstAlias<"vmov${p} $Vd, $Vm",
4645 (VORRd DPR:$Vd, DPR:$Vm, DPR:$Vm, pred:$p)>;
4646 def : InstAlias<"vmov${p} $Vd, $Vm",
4647 (VORRq QPR:$Vd, QPR:$Vm, QPR:$Vm, pred:$p)>;
4649 // VMOV : Vector Move (Immediate)
4651 let isReMaterializable = 1 in {
4652 def VMOVv8i8 : N1ModImm<1, 0b000, 0b1110, 0, 0, 0, 1, (outs DPR:$Vd),
4653 (ins nImmSplatI8:$SIMM), IIC_VMOVImm,
4654 "vmov", "i8", "$Vd, $SIMM", "",
4655 [(set DPR:$Vd, (v8i8 (NEONvmovImm timm:$SIMM)))]>;
4656 def VMOVv16i8 : N1ModImm<1, 0b000, 0b1110, 0, 1, 0, 1, (outs QPR:$Vd),
4657 (ins nImmSplatI8:$SIMM), IIC_VMOVImm,
4658 "vmov", "i8", "$Vd, $SIMM", "",
4659 [(set QPR:$Vd, (v16i8 (NEONvmovImm timm:$SIMM)))]>;
4661 def VMOVv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 0, 1, (outs DPR:$Vd),
4662 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4663 "vmov", "i16", "$Vd, $SIMM", "",
4664 [(set DPR:$Vd, (v4i16 (NEONvmovImm timm:$SIMM)))]> {
4665 let Inst{9} = SIMM{9};
4668 def VMOVv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 0, 1, (outs QPR:$Vd),
4669 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4670 "vmov", "i16", "$Vd, $SIMM", "",
4671 [(set QPR:$Vd, (v8i16 (NEONvmovImm timm:$SIMM)))]> {
4672 let Inst{9} = SIMM{9};
4675 def VMOVv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 0, 1, (outs DPR:$Vd),
4676 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4677 "vmov", "i32", "$Vd, $SIMM", "",
4678 [(set DPR:$Vd, (v2i32 (NEONvmovImm timm:$SIMM)))]> {
4679 let Inst{11-8} = SIMM{11-8};
4682 def VMOVv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 0, 1, (outs QPR:$Vd),
4683 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4684 "vmov", "i32", "$Vd, $SIMM", "",
4685 [(set QPR:$Vd, (v4i32 (NEONvmovImm timm:$SIMM)))]> {
4686 let Inst{11-8} = SIMM{11-8};
4689 def VMOVv1i64 : N1ModImm<1, 0b000, 0b1110, 0, 0, 1, 1, (outs DPR:$Vd),
4690 (ins nImmSplatI64:$SIMM), IIC_VMOVImm,
4691 "vmov", "i64", "$Vd, $SIMM", "",
4692 [(set DPR:$Vd, (v1i64 (NEONvmovImm timm:$SIMM)))]>;
4693 def VMOVv2i64 : N1ModImm<1, 0b000, 0b1110, 0, 1, 1, 1, (outs QPR:$Vd),
4694 (ins nImmSplatI64:$SIMM), IIC_VMOVImm,
4695 "vmov", "i64", "$Vd, $SIMM", "",
4696 [(set QPR:$Vd, (v2i64 (NEONvmovImm timm:$SIMM)))]>;
4698 def VMOVv2f32 : N1ModImm<1, 0b000, 0b1111, 0, 0, 0, 1, (outs DPR:$Vd),
4699 (ins nImmVMOVF32:$SIMM), IIC_VMOVImm,
4700 "vmov", "f32", "$Vd, $SIMM", "",
4701 [(set DPR:$Vd, (v2f32 (NEONvmovFPImm timm:$SIMM)))]>;
4702 def VMOVv4f32 : N1ModImm<1, 0b000, 0b1111, 0, 1, 0, 1, (outs QPR:$Vd),
4703 (ins nImmVMOVF32:$SIMM), IIC_VMOVImm,
4704 "vmov", "f32", "$Vd, $SIMM", "",
4705 [(set QPR:$Vd, (v4f32 (NEONvmovFPImm timm:$SIMM)))]>;
4706 } // isReMaterializable
4708 // VMOV : Vector Get Lane (move scalar to ARM core register)
4710 def VGETLNs8 : NVGetLane<{1,1,1,0,0,1,?,1}, 0b1011, {?,?},
4711 (outs GPR:$R), (ins DPR:$V, VectorIndex8:$lane),
4712 IIC_VMOVSI, "vmov", "s8", "$R, $V$lane",
4713 [(set GPR:$R, (NEONvgetlanes (v8i8 DPR:$V),
4715 let Inst{21} = lane{2};
4716 let Inst{6-5} = lane{1-0};
4718 def VGETLNs16 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, {?,1},
4719 (outs GPR:$R), (ins DPR:$V, VectorIndex16:$lane),
4720 IIC_VMOVSI, "vmov", "s16", "$R, $V$lane",
4721 [(set GPR:$R, (NEONvgetlanes (v4i16 DPR:$V),
4723 let Inst{21} = lane{1};
4724 let Inst{6} = lane{0};
4726 def VGETLNu8 : NVGetLane<{1,1,1,0,1,1,?,1}, 0b1011, {?,?},
4727 (outs GPR:$R), (ins DPR:$V, VectorIndex8:$lane),
4728 IIC_VMOVSI, "vmov", "u8", "$R, $V$lane",
4729 [(set GPR:$R, (NEONvgetlaneu (v8i8 DPR:$V),
4731 let Inst{21} = lane{2};
4732 let Inst{6-5} = lane{1-0};
4734 def VGETLNu16 : NVGetLane<{1,1,1,0,1,0,?,1}, 0b1011, {?,1},
4735 (outs GPR:$R), (ins DPR:$V, VectorIndex16:$lane),
4736 IIC_VMOVSI, "vmov", "u16", "$R, $V$lane",
4737 [(set GPR:$R, (NEONvgetlaneu (v4i16 DPR:$V),
4739 let Inst{21} = lane{1};
4740 let Inst{6} = lane{0};
4742 def VGETLNi32 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, 0b00,
4743 (outs GPR:$R), (ins DPR:$V, VectorIndex32:$lane),
4744 IIC_VMOVSI, "vmov", "32", "$R, $V$lane",
4745 [(set GPR:$R, (extractelt (v2i32 DPR:$V),
4747 let Inst{21} = lane{0};
4749 // def VGETLNf32: see FMRDH and FMRDL in ARMInstrVFP.td
4750 def : Pat<(NEONvgetlanes (v16i8 QPR:$src), imm:$lane),
4751 (VGETLNs8 (v8i8 (EXTRACT_SUBREG QPR:$src,
4752 (DSubReg_i8_reg imm:$lane))),
4753 (SubReg_i8_lane imm:$lane))>;
4754 def : Pat<(NEONvgetlanes (v8i16 QPR:$src), imm:$lane),
4755 (VGETLNs16 (v4i16 (EXTRACT_SUBREG QPR:$src,
4756 (DSubReg_i16_reg imm:$lane))),
4757 (SubReg_i16_lane imm:$lane))>;
4758 def : Pat<(NEONvgetlaneu (v16i8 QPR:$src), imm:$lane),
4759 (VGETLNu8 (v8i8 (EXTRACT_SUBREG QPR:$src,
4760 (DSubReg_i8_reg imm:$lane))),
4761 (SubReg_i8_lane imm:$lane))>;
4762 def : Pat<(NEONvgetlaneu (v8i16 QPR:$src), imm:$lane),
4763 (VGETLNu16 (v4i16 (EXTRACT_SUBREG QPR:$src,
4764 (DSubReg_i16_reg imm:$lane))),
4765 (SubReg_i16_lane imm:$lane))>;
4766 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
4767 (VGETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src,
4768 (DSubReg_i32_reg imm:$lane))),
4769 (SubReg_i32_lane imm:$lane))>;
4770 def : Pat<(extractelt (v2f32 DPR:$src1), imm:$src2),
4771 (EXTRACT_SUBREG (v2f32 (COPY_TO_REGCLASS (v2f32 DPR:$src1),DPR_VFP2)),
4772 (SSubReg_f32_reg imm:$src2))>;
4773 def : Pat<(extractelt (v4f32 QPR:$src1), imm:$src2),
4774 (EXTRACT_SUBREG (v4f32 (COPY_TO_REGCLASS (v4f32 QPR:$src1),QPR_VFP2)),
4775 (SSubReg_f32_reg imm:$src2))>;
4776 //def : Pat<(extractelt (v2i64 QPR:$src1), imm:$src2),
4777 // (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
4778 def : Pat<(extractelt (v2f64 QPR:$src1), imm:$src2),
4779 (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
4782 // VMOV : Vector Set Lane (move ARM core register to scalar)
4784 let Constraints = "$src1 = $V" in {
4785 def VSETLNi8 : NVSetLane<{1,1,1,0,0,1,?,0}, 0b1011, {?,?}, (outs DPR:$V),
4786 (ins DPR:$src1, GPR:$R, VectorIndex8:$lane),
4787 IIC_VMOVISL, "vmov", "8", "$V$lane, $R",
4788 [(set DPR:$V, (vector_insert (v8i8 DPR:$src1),
4789 GPR:$R, imm:$lane))]> {
4790 let Inst{21} = lane{2};
4791 let Inst{6-5} = lane{1-0};
4793 def VSETLNi16 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, {?,1}, (outs DPR:$V),
4794 (ins DPR:$src1, GPR:$R, VectorIndex16:$lane),
4795 IIC_VMOVISL, "vmov", "16", "$V$lane, $R",
4796 [(set DPR:$V, (vector_insert (v4i16 DPR:$src1),
4797 GPR:$R, imm:$lane))]> {
4798 let Inst{21} = lane{1};
4799 let Inst{6} = lane{0};
4801 def VSETLNi32 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, 0b00, (outs DPR:$V),
4802 (ins DPR:$src1, GPR:$R, VectorIndex32:$lane),
4803 IIC_VMOVISL, "vmov", "32", "$V$lane, $R",
4804 [(set DPR:$V, (insertelt (v2i32 DPR:$src1),
4805 GPR:$R, imm:$lane))]> {
4806 let Inst{21} = lane{0};
4809 def : Pat<(vector_insert (v16i8 QPR:$src1), GPR:$src2, imm:$lane),
4810 (v16i8 (INSERT_SUBREG QPR:$src1,
4811 (v8i8 (VSETLNi8 (v8i8 (EXTRACT_SUBREG QPR:$src1,
4812 (DSubReg_i8_reg imm:$lane))),
4813 GPR:$src2, (SubReg_i8_lane imm:$lane))),
4814 (DSubReg_i8_reg imm:$lane)))>;
4815 def : Pat<(vector_insert (v8i16 QPR:$src1), GPR:$src2, imm:$lane),
4816 (v8i16 (INSERT_SUBREG QPR:$src1,
4817 (v4i16 (VSETLNi16 (v4i16 (EXTRACT_SUBREG QPR:$src1,
4818 (DSubReg_i16_reg imm:$lane))),
4819 GPR:$src2, (SubReg_i16_lane imm:$lane))),
4820 (DSubReg_i16_reg imm:$lane)))>;
4821 def : Pat<(insertelt (v4i32 QPR:$src1), GPR:$src2, imm:$lane),
4822 (v4i32 (INSERT_SUBREG QPR:$src1,
4823 (v2i32 (VSETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src1,
4824 (DSubReg_i32_reg imm:$lane))),
4825 GPR:$src2, (SubReg_i32_lane imm:$lane))),
4826 (DSubReg_i32_reg imm:$lane)))>;
4828 def : Pat<(v2f32 (insertelt DPR:$src1, SPR:$src2, imm:$src3)),
4829 (INSERT_SUBREG (v2f32 (COPY_TO_REGCLASS DPR:$src1, DPR_VFP2)),
4830 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
4831 def : Pat<(v4f32 (insertelt QPR:$src1, SPR:$src2, imm:$src3)),
4832 (INSERT_SUBREG (v4f32 (COPY_TO_REGCLASS QPR:$src1, QPR_VFP2)),
4833 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
4835 //def : Pat<(v2i64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
4836 // (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
4837 def : Pat<(v2f64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
4838 (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
4840 def : Pat<(v2f32 (scalar_to_vector SPR:$src)),
4841 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
4842 def : Pat<(v2f64 (scalar_to_vector (f64 DPR:$src))),
4843 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
4844 def : Pat<(v4f32 (scalar_to_vector SPR:$src)),
4845 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
4847 def : Pat<(v8i8 (scalar_to_vector GPR:$src)),
4848 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4849 def : Pat<(v4i16 (scalar_to_vector GPR:$src)),
4850 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4851 def : Pat<(v2i32 (scalar_to_vector GPR:$src)),
4852 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4854 def : Pat<(v16i8 (scalar_to_vector GPR:$src)),
4855 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4856 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4858 def : Pat<(v8i16 (scalar_to_vector GPR:$src)),
4859 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
4860 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4862 def : Pat<(v4i32 (scalar_to_vector GPR:$src)),
4863 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
4864 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4867 // VDUP : Vector Duplicate (from ARM core register to all elements)
4869 class VDUPD<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
4870 : NVDup<opcod1, 0b1011, opcod3, (outs DPR:$V), (ins GPR:$R),
4871 IIC_VMOVIS, "vdup", Dt, "$V, $R",
4872 [(set DPR:$V, (Ty (NEONvdup (i32 GPR:$R))))]>;
4873 class VDUPQ<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
4874 : NVDup<opcod1, 0b1011, opcod3, (outs QPR:$V), (ins GPR:$R),
4875 IIC_VMOVIS, "vdup", Dt, "$V, $R",
4876 [(set QPR:$V, (Ty (NEONvdup (i32 GPR:$R))))]>;
4878 def VDUP8d : VDUPD<0b11101100, 0b00, "8", v8i8>;
4879 def VDUP16d : VDUPD<0b11101000, 0b01, "16", v4i16>;
4880 def VDUP32d : VDUPD<0b11101000, 0b00, "32", v2i32>;
4881 def VDUP8q : VDUPQ<0b11101110, 0b00, "8", v16i8>;
4882 def VDUP16q : VDUPQ<0b11101010, 0b01, "16", v8i16>;
4883 def VDUP32q : VDUPQ<0b11101010, 0b00, "32", v4i32>;
4885 def : Pat<(v2f32 (NEONvdup (f32 (bitconvert GPR:$R)))), (VDUP32d GPR:$R)>;
4886 def : Pat<(v4f32 (NEONvdup (f32 (bitconvert GPR:$R)))), (VDUP32q GPR:$R)>;
4888 // VDUP : Vector Duplicate Lane (from scalar to all elements)
4890 class VDUPLND<bits<4> op19_16, string OpcodeStr, string Dt,
4891 ValueType Ty, Operand IdxTy>
4892 : NVDupLane<op19_16, 0, (outs DPR:$Vd), (ins DPR:$Vm, IdxTy:$lane),
4893 IIC_VMOVD, OpcodeStr, Dt, "$Vd, $Vm$lane",
4894 [(set DPR:$Vd, (Ty (NEONvduplane (Ty DPR:$Vm), imm:$lane)))]>;
4896 class VDUPLNQ<bits<4> op19_16, string OpcodeStr, string Dt,
4897 ValueType ResTy, ValueType OpTy, Operand IdxTy>
4898 : NVDupLane<op19_16, 1, (outs QPR:$Vd), (ins DPR:$Vm, IdxTy:$lane),
4899 IIC_VMOVQ, OpcodeStr, Dt, "$Vd, $Vm$lane",
4900 [(set QPR:$Vd, (ResTy (NEONvduplane (OpTy DPR:$Vm),
4901 VectorIndex32:$lane)))]>;
4903 // Inst{19-16} is partially specified depending on the element size.
4905 def VDUPLN8d : VDUPLND<{?,?,?,1}, "vdup", "8", v8i8, VectorIndex8> {
4907 let Inst{19-17} = lane{2-0};
4909 def VDUPLN16d : VDUPLND<{?,?,1,0}, "vdup", "16", v4i16, VectorIndex16> {
4911 let Inst{19-18} = lane{1-0};
4913 def VDUPLN32d : VDUPLND<{?,1,0,0}, "vdup", "32", v2i32, VectorIndex32> {
4915 let Inst{19} = lane{0};
4917 def VDUPLN8q : VDUPLNQ<{?,?,?,1}, "vdup", "8", v16i8, v8i8, VectorIndex8> {
4919 let Inst{19-17} = lane{2-0};
4921 def VDUPLN16q : VDUPLNQ<{?,?,1,0}, "vdup", "16", v8i16, v4i16, VectorIndex16> {
4923 let Inst{19-18} = lane{1-0};
4925 def VDUPLN32q : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4i32, v2i32, VectorIndex32> {
4927 let Inst{19} = lane{0};
4930 def : Pat<(v2f32 (NEONvduplane (v2f32 DPR:$Vm), imm:$lane)),
4931 (VDUPLN32d DPR:$Vm, imm:$lane)>;
4933 def : Pat<(v4f32 (NEONvduplane (v2f32 DPR:$Vm), imm:$lane)),
4934 (VDUPLN32q DPR:$Vm, imm:$lane)>;
4936 def : Pat<(v16i8 (NEONvduplane (v16i8 QPR:$src), imm:$lane)),
4937 (v16i8 (VDUPLN8q (v8i8 (EXTRACT_SUBREG QPR:$src,
4938 (DSubReg_i8_reg imm:$lane))),
4939 (SubReg_i8_lane imm:$lane)))>;
4940 def : Pat<(v8i16 (NEONvduplane (v8i16 QPR:$src), imm:$lane)),
4941 (v8i16 (VDUPLN16q (v4i16 (EXTRACT_SUBREG QPR:$src,
4942 (DSubReg_i16_reg imm:$lane))),
4943 (SubReg_i16_lane imm:$lane)))>;
4944 def : Pat<(v4i32 (NEONvduplane (v4i32 QPR:$src), imm:$lane)),
4945 (v4i32 (VDUPLN32q (v2i32 (EXTRACT_SUBREG QPR:$src,
4946 (DSubReg_i32_reg imm:$lane))),
4947 (SubReg_i32_lane imm:$lane)))>;
4948 def : Pat<(v4f32 (NEONvduplane (v4f32 QPR:$src), imm:$lane)),
4949 (v4f32 (VDUPLN32q (v2f32 (EXTRACT_SUBREG QPR:$src,
4950 (DSubReg_i32_reg imm:$lane))),
4951 (SubReg_i32_lane imm:$lane)))>;
4953 def VDUPfdf : PseudoNeonI<(outs DPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
4954 [(set DPR:$dst, (v2f32 (NEONvdup (f32 SPR:$src))))]>;
4955 def VDUPfqf : PseudoNeonI<(outs QPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
4956 [(set QPR:$dst, (v4f32 (NEONvdup (f32 SPR:$src))))]>;
4958 // VMOVN : Vector Narrowing Move
4959 defm VMOVN : N2VN_HSD<0b11,0b11,0b10,0b00100,0,0, IIC_VMOVN,
4960 "vmovn", "i", trunc>;
4961 // VQMOVN : Vector Saturating Narrowing Move
4962 defm VQMOVNs : N2VNInt_HSD<0b11,0b11,0b10,0b00101,0,0, IIC_VQUNAiD,
4963 "vqmovn", "s", int_arm_neon_vqmovns>;
4964 defm VQMOVNu : N2VNInt_HSD<0b11,0b11,0b10,0b00101,1,0, IIC_VQUNAiD,
4965 "vqmovn", "u", int_arm_neon_vqmovnu>;
4966 defm VQMOVNsu : N2VNInt_HSD<0b11,0b11,0b10,0b00100,1,0, IIC_VQUNAiD,
4967 "vqmovun", "s", int_arm_neon_vqmovnsu>;
4968 // VMOVL : Vector Lengthening Move
4969 defm VMOVLs : N2VL_QHS<0b01,0b10100,0,1, "vmovl", "s", sext>;
4970 defm VMOVLu : N2VL_QHS<0b11,0b10100,0,1, "vmovl", "u", zext>;
4972 // Vector Conversions.
4974 // VCVT : Vector Convert Between Floating-Point and Integers
4975 def VCVTf2sd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
4976 v2i32, v2f32, fp_to_sint>;
4977 def VCVTf2ud : N2VD<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
4978 v2i32, v2f32, fp_to_uint>;
4979 def VCVTs2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
4980 v2f32, v2i32, sint_to_fp>;
4981 def VCVTu2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
4982 v2f32, v2i32, uint_to_fp>;
4984 def VCVTf2sq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
4985 v4i32, v4f32, fp_to_sint>;
4986 def VCVTf2uq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
4987 v4i32, v4f32, fp_to_uint>;
4988 def VCVTs2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
4989 v4f32, v4i32, sint_to_fp>;
4990 def VCVTu2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
4991 v4f32, v4i32, uint_to_fp>;
4993 // VCVT : Vector Convert Between Floating-Point and Fixed-Point.
4994 let DecoderMethod = "DecodeVCVTD" in {
4995 def VCVTf2xsd : N2VCvtD<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
4996 v2i32, v2f32, int_arm_neon_vcvtfp2fxs>;
4997 def VCVTf2xud : N2VCvtD<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
4998 v2i32, v2f32, int_arm_neon_vcvtfp2fxu>;
4999 def VCVTxs2fd : N2VCvtD<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
5000 v2f32, v2i32, int_arm_neon_vcvtfxs2fp>;
5001 def VCVTxu2fd : N2VCvtD<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
5002 v2f32, v2i32, int_arm_neon_vcvtfxu2fp>;
5005 let DecoderMethod = "DecodeVCVTQ" in {
5006 def VCVTf2xsq : N2VCvtQ<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
5007 v4i32, v4f32, int_arm_neon_vcvtfp2fxs>;
5008 def VCVTf2xuq : N2VCvtQ<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
5009 v4i32, v4f32, int_arm_neon_vcvtfp2fxu>;
5010 def VCVTxs2fq : N2VCvtQ<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
5011 v4f32, v4i32, int_arm_neon_vcvtfxs2fp>;
5012 def VCVTxu2fq : N2VCvtQ<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
5013 v4f32, v4i32, int_arm_neon_vcvtfxu2fp>;
5016 // VCVT : Vector Convert Between Half-Precision and Single-Precision.
5017 def VCVTf2h : N2VNInt<0b11, 0b11, 0b01, 0b10, 0b01100, 0, 0,
5018 IIC_VUNAQ, "vcvt", "f16.f32",
5019 v4i16, v4f32, int_arm_neon_vcvtfp2hf>,
5020 Requires<[HasNEON, HasFP16]>;
5021 def VCVTh2f : N2VLInt<0b11, 0b11, 0b01, 0b10, 0b01110, 0, 0,
5022 IIC_VUNAQ, "vcvt", "f32.f16",
5023 v4f32, v4i16, int_arm_neon_vcvthf2fp>,
5024 Requires<[HasNEON, HasFP16]>;
5028 // VREV64 : Vector Reverse elements within 64-bit doublewords
5030 class VREV64D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
5031 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 0, 0, (outs DPR:$Vd),
5032 (ins DPR:$Vm), IIC_VMOVD,
5033 OpcodeStr, Dt, "$Vd, $Vm", "",
5034 [(set DPR:$Vd, (Ty (NEONvrev64 (Ty DPR:$Vm))))]>;
5035 class VREV64Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
5036 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 1, 0, (outs QPR:$Vd),
5037 (ins QPR:$Vm), IIC_VMOVQ,
5038 OpcodeStr, Dt, "$Vd, $Vm", "",
5039 [(set QPR:$Vd, (Ty (NEONvrev64 (Ty QPR:$Vm))))]>;
5041 def VREV64d8 : VREV64D<0b00, "vrev64", "8", v8i8>;
5042 def VREV64d16 : VREV64D<0b01, "vrev64", "16", v4i16>;
5043 def VREV64d32 : VREV64D<0b10, "vrev64", "32", v2i32>;
5044 def : Pat<(v2f32 (NEONvrev64 (v2f32 DPR:$Vm))), (VREV64d32 DPR:$Vm)>;
5046 def VREV64q8 : VREV64Q<0b00, "vrev64", "8", v16i8>;
5047 def VREV64q16 : VREV64Q<0b01, "vrev64", "16", v8i16>;
5048 def VREV64q32 : VREV64Q<0b10, "vrev64", "32", v4i32>;
5049 def : Pat<(v4f32 (NEONvrev64 (v4f32 QPR:$Vm))), (VREV64q32 QPR:$Vm)>;
5051 // VREV32 : Vector Reverse elements within 32-bit words
5053 class VREV32D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
5054 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 0, 0, (outs DPR:$Vd),
5055 (ins DPR:$Vm), IIC_VMOVD,
5056 OpcodeStr, Dt, "$Vd, $Vm", "",
5057 [(set DPR:$Vd, (Ty (NEONvrev32 (Ty DPR:$Vm))))]>;
5058 class VREV32Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
5059 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 1, 0, (outs QPR:$Vd),
5060 (ins QPR:$Vm), IIC_VMOVQ,
5061 OpcodeStr, Dt, "$Vd, $Vm", "",
5062 [(set QPR:$Vd, (Ty (NEONvrev32 (Ty QPR:$Vm))))]>;
5064 def VREV32d8 : VREV32D<0b00, "vrev32", "8", v8i8>;
5065 def VREV32d16 : VREV32D<0b01, "vrev32", "16", v4i16>;
5067 def VREV32q8 : VREV32Q<0b00, "vrev32", "8", v16i8>;
5068 def VREV32q16 : VREV32Q<0b01, "vrev32", "16", v8i16>;
5070 // VREV16 : Vector Reverse elements within 16-bit halfwords
5072 class VREV16D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
5073 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 0, 0, (outs DPR:$Vd),
5074 (ins DPR:$Vm), IIC_VMOVD,
5075 OpcodeStr, Dt, "$Vd, $Vm", "",
5076 [(set DPR:$Vd, (Ty (NEONvrev16 (Ty DPR:$Vm))))]>;
5077 class VREV16Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
5078 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 1, 0, (outs QPR:$Vd),
5079 (ins QPR:$Vm), IIC_VMOVQ,
5080 OpcodeStr, Dt, "$Vd, $Vm", "",
5081 [(set QPR:$Vd, (Ty (NEONvrev16 (Ty QPR:$Vm))))]>;
5083 def VREV16d8 : VREV16D<0b00, "vrev16", "8", v8i8>;
5084 def VREV16q8 : VREV16Q<0b00, "vrev16", "8", v16i8>;
5086 // Other Vector Shuffles.
5088 // Aligned extractions: really just dropping registers
5090 class AlignedVEXTq<ValueType DestTy, ValueType SrcTy, SDNodeXForm LaneCVT>
5091 : Pat<(DestTy (vector_extract_subvec (SrcTy QPR:$src), (i32 imm:$start))),
5092 (EXTRACT_SUBREG (SrcTy QPR:$src), (LaneCVT imm:$start))>;
5094 def : AlignedVEXTq<v8i8, v16i8, DSubReg_i8_reg>;
5096 def : AlignedVEXTq<v4i16, v8i16, DSubReg_i16_reg>;
5098 def : AlignedVEXTq<v2i32, v4i32, DSubReg_i32_reg>;
5100 def : AlignedVEXTq<v1i64, v2i64, DSubReg_f64_reg>;
5102 def : AlignedVEXTq<v2f32, v4f32, DSubReg_i32_reg>;
5105 // VEXT : Vector Extract
5107 class VEXTd<string OpcodeStr, string Dt, ValueType Ty, Operand immTy>
5108 : N3V<0,1,0b11,{?,?,?,?},0,0, (outs DPR:$Vd),
5109 (ins DPR:$Vn, DPR:$Vm, immTy:$index), NVExtFrm,
5110 IIC_VEXTD, OpcodeStr, Dt, "$Vd, $Vn, $Vm, $index", "",
5111 [(set DPR:$Vd, (Ty (NEONvext (Ty DPR:$Vn),
5112 (Ty DPR:$Vm), imm:$index)))]> {
5114 let Inst{11-8} = index{3-0};
5117 class VEXTq<string OpcodeStr, string Dt, ValueType Ty, Operand immTy>
5118 : N3V<0,1,0b11,{?,?,?,?},1,0, (outs QPR:$Vd),
5119 (ins QPR:$Vn, QPR:$Vm, imm0_15:$index), NVExtFrm,
5120 IIC_VEXTQ, OpcodeStr, Dt, "$Vd, $Vn, $Vm, $index", "",
5121 [(set QPR:$Vd, (Ty (NEONvext (Ty QPR:$Vn),
5122 (Ty QPR:$Vm), imm:$index)))]> {
5124 let Inst{11-8} = index{3-0};
5127 def VEXTd8 : VEXTd<"vext", "8", v8i8, imm0_7> {
5128 let Inst{11-8} = index{3-0};
5130 def VEXTd16 : VEXTd<"vext", "16", v4i16, imm0_3> {
5131 let Inst{11-9} = index{2-0};
5134 def VEXTd32 : VEXTd<"vext", "32", v2i32, imm0_1> {
5135 let Inst{11-10} = index{1-0};
5136 let Inst{9-8} = 0b00;
5138 def : Pat<(v2f32 (NEONvext (v2f32 DPR:$Vn),
5141 (VEXTd32 DPR:$Vn, DPR:$Vm, imm:$index)>;
5143 def VEXTq8 : VEXTq<"vext", "8", v16i8, imm0_15> {
5144 let Inst{11-8} = index{3-0};
5146 def VEXTq16 : VEXTq<"vext", "16", v8i16, imm0_7> {
5147 let Inst{11-9} = index{2-0};
5150 def VEXTq32 : VEXTq<"vext", "32", v4i32, imm0_3> {
5151 let Inst{11-10} = index{1-0};
5152 let Inst{9-8} = 0b00;
5154 def VEXTq64 : VEXTq<"vext", "64", v2i64, imm0_1> {
5155 let Inst{11} = index{0};
5156 let Inst{10-8} = 0b000;
5158 def : Pat<(v4f32 (NEONvext (v4f32 QPR:$Vn),
5161 (VEXTq32 QPR:$Vn, QPR:$Vm, imm:$index)>;
5163 // VTRN : Vector Transpose
5165 def VTRNd8 : N2VDShuffle<0b00, 0b00001, "vtrn", "8">;
5166 def VTRNd16 : N2VDShuffle<0b01, 0b00001, "vtrn", "16">;
5167 def VTRNd32 : N2VDShuffle<0b10, 0b00001, "vtrn", "32">;
5169 def VTRNq8 : N2VQShuffle<0b00, 0b00001, IIC_VPERMQ, "vtrn", "8">;
5170 def VTRNq16 : N2VQShuffle<0b01, 0b00001, IIC_VPERMQ, "vtrn", "16">;
5171 def VTRNq32 : N2VQShuffle<0b10, 0b00001, IIC_VPERMQ, "vtrn", "32">;
5173 // VUZP : Vector Unzip (Deinterleave)
5175 def VUZPd8 : N2VDShuffle<0b00, 0b00010, "vuzp", "8">;
5176 def VUZPd16 : N2VDShuffle<0b01, 0b00010, "vuzp", "16">;
5177 def VUZPd32 : N2VDShuffle<0b10, 0b00010, "vuzp", "32">;
5179 def VUZPq8 : N2VQShuffle<0b00, 0b00010, IIC_VPERMQ3, "vuzp", "8">;
5180 def VUZPq16 : N2VQShuffle<0b01, 0b00010, IIC_VPERMQ3, "vuzp", "16">;
5181 def VUZPq32 : N2VQShuffle<0b10, 0b00010, IIC_VPERMQ3, "vuzp", "32">;
5183 // VZIP : Vector Zip (Interleave)
5185 def VZIPd8 : N2VDShuffle<0b00, 0b00011, "vzip", "8">;
5186 def VZIPd16 : N2VDShuffle<0b01, 0b00011, "vzip", "16">;
5187 def VZIPd32 : N2VDShuffle<0b10, 0b00011, "vzip", "32">;
5189 def VZIPq8 : N2VQShuffle<0b00, 0b00011, IIC_VPERMQ3, "vzip", "8">;
5190 def VZIPq16 : N2VQShuffle<0b01, 0b00011, IIC_VPERMQ3, "vzip", "16">;
5191 def VZIPq32 : N2VQShuffle<0b10, 0b00011, IIC_VPERMQ3, "vzip", "32">;
5193 // Vector Table Lookup and Table Extension.
5195 // VTBL : Vector Table Lookup
5196 let DecoderMethod = "DecodeTBLInstruction" in {
5198 : N3V<1,1,0b11,0b1000,0,0, (outs DPR:$Vd),
5199 (ins VecListOneD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTB1,
5200 "vtbl", "8", "$Vd, $Vn, $Vm", "",
5201 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbl1 VecListOneD:$Vn, DPR:$Vm)))]>;
5202 let hasExtraSrcRegAllocReq = 1 in {
5204 : N3V<1,1,0b11,0b1001,0,0, (outs DPR:$Vd),
5205 (ins VecListTwoD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTB2,
5206 "vtbl", "8", "$Vd, $Vn, $Vm", "", []>;
5208 : N3V<1,1,0b11,0b1010,0,0, (outs DPR:$Vd),
5209 (ins VecListThreeD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTB3,
5210 "vtbl", "8", "$Vd, $Vn, $Vm", "", []>;
5212 : N3V<1,1,0b11,0b1011,0,0, (outs DPR:$Vd),
5213 (ins VecListFourD:$Vn, DPR:$Vm),
5215 "vtbl", "8", "$Vd, $Vn, $Vm", "", []>;
5216 } // hasExtraSrcRegAllocReq = 1
5219 : PseudoNeonI<(outs DPR:$dst), (ins QPR:$tbl, DPR:$src), IIC_VTB2, "", []>;
5221 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB3, "", []>;
5223 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB4, "", []>;
5225 // VTBX : Vector Table Extension
5227 : N3V<1,1,0b11,0b1000,1,0, (outs DPR:$Vd),
5228 (ins DPR:$orig, VecListOneD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTBX1,
5229 "vtbx", "8", "$Vd, $Vn, $Vm", "$orig = $Vd",
5230 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbx1
5231 DPR:$orig, VecListOneD:$Vn, DPR:$Vm)))]>;
5232 let hasExtraSrcRegAllocReq = 1 in {
5234 : N3V<1,1,0b11,0b1001,1,0, (outs DPR:$Vd),
5235 (ins DPR:$orig, VecListTwoD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTBX2,
5236 "vtbx", "8", "$Vd, $Vn, $Vm", "$orig = $Vd", []>;
5238 : N3V<1,1,0b11,0b1010,1,0, (outs DPR:$Vd),
5239 (ins DPR:$orig, VecListThreeD:$Vn, DPR:$Vm),
5240 NVTBLFrm, IIC_VTBX3,
5241 "vtbx", "8", "$Vd, $Vn, $Vm",
5244 : N3V<1,1,0b11,0b1011,1,0, (outs DPR:$Vd),
5245 (ins DPR:$orig, VecListFourD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTBX4,
5246 "vtbx", "8", "$Vd, $Vn, $Vm",
5248 } // hasExtraSrcRegAllocReq = 1
5251 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QPR:$tbl, DPR:$src),
5252 IIC_VTBX2, "$orig = $dst", []>;
5254 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
5255 IIC_VTBX3, "$orig = $dst", []>;
5257 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
5258 IIC_VTBX4, "$orig = $dst", []>;
5259 } // DecoderMethod = "DecodeTBLInstruction"
5261 //===----------------------------------------------------------------------===//
5262 // NEON instructions for single-precision FP math
5263 //===----------------------------------------------------------------------===//
5265 class N2VSPat<SDNode OpNode, NeonI Inst>
5266 : NEONFPPat<(f32 (OpNode SPR:$a)),
5268 (v2f32 (COPY_TO_REGCLASS (Inst
5270 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5271 SPR:$a, ssub_0)), DPR_VFP2)), ssub_0)>;
5273 class N3VSPat<SDNode OpNode, NeonI Inst>
5274 : NEONFPPat<(f32 (OpNode SPR:$a, SPR:$b)),
5276 (v2f32 (COPY_TO_REGCLASS (Inst
5278 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5281 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5282 SPR:$b, ssub_0)), DPR_VFP2)), ssub_0)>;
5284 class N3VSMulOpPat<SDNode MulNode, SDNode OpNode, NeonI Inst>
5285 : NEONFPPat<(f32 (OpNode SPR:$acc, (f32 (MulNode SPR:$a, SPR:$b)))),
5287 (v2f32 (COPY_TO_REGCLASS (Inst
5289 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5292 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5295 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5296 SPR:$b, ssub_0)), DPR_VFP2)), ssub_0)>;
5298 def : N3VSPat<fadd, VADDfd>;
5299 def : N3VSPat<fsub, VSUBfd>;
5300 def : N3VSPat<fmul, VMULfd>;
5301 def : N3VSMulOpPat<fmul, fadd, VMLAfd>,
5302 Requires<[HasNEON, UseNEONForFP, UseFPVMLx]>;
5303 def : N3VSMulOpPat<fmul, fsub, VMLSfd>,
5304 Requires<[HasNEON, UseNEONForFP, UseFPVMLx]>;
5305 def : N2VSPat<fabs, VABSfd>;
5306 def : N2VSPat<fneg, VNEGfd>;
5307 def : N3VSPat<NEONfmax, VMAXfd>;
5308 def : N3VSPat<NEONfmin, VMINfd>;
5309 def : N2VSPat<arm_ftosi, VCVTf2sd>;
5310 def : N2VSPat<arm_ftoui, VCVTf2ud>;
5311 def : N2VSPat<arm_sitof, VCVTs2fd>;
5312 def : N2VSPat<arm_uitof, VCVTu2fd>;
5314 //===----------------------------------------------------------------------===//
5315 // Non-Instruction Patterns
5316 //===----------------------------------------------------------------------===//
5319 def : Pat<(v1i64 (bitconvert (v2i32 DPR:$src))), (v1i64 DPR:$src)>;
5320 def : Pat<(v1i64 (bitconvert (v4i16 DPR:$src))), (v1i64 DPR:$src)>;
5321 def : Pat<(v1i64 (bitconvert (v8i8 DPR:$src))), (v1i64 DPR:$src)>;
5322 def : Pat<(v1i64 (bitconvert (f64 DPR:$src))), (v1i64 DPR:$src)>;
5323 def : Pat<(v1i64 (bitconvert (v2f32 DPR:$src))), (v1i64 DPR:$src)>;
5324 def : Pat<(v2i32 (bitconvert (v1i64 DPR:$src))), (v2i32 DPR:$src)>;
5325 def : Pat<(v2i32 (bitconvert (v4i16 DPR:$src))), (v2i32 DPR:$src)>;
5326 def : Pat<(v2i32 (bitconvert (v8i8 DPR:$src))), (v2i32 DPR:$src)>;
5327 def : Pat<(v2i32 (bitconvert (f64 DPR:$src))), (v2i32 DPR:$src)>;
5328 def : Pat<(v2i32 (bitconvert (v2f32 DPR:$src))), (v2i32 DPR:$src)>;
5329 def : Pat<(v4i16 (bitconvert (v1i64 DPR:$src))), (v4i16 DPR:$src)>;
5330 def : Pat<(v4i16 (bitconvert (v2i32 DPR:$src))), (v4i16 DPR:$src)>;
5331 def : Pat<(v4i16 (bitconvert (v8i8 DPR:$src))), (v4i16 DPR:$src)>;
5332 def : Pat<(v4i16 (bitconvert (f64 DPR:$src))), (v4i16 DPR:$src)>;
5333 def : Pat<(v4i16 (bitconvert (v2f32 DPR:$src))), (v4i16 DPR:$src)>;
5334 def : Pat<(v8i8 (bitconvert (v1i64 DPR:$src))), (v8i8 DPR:$src)>;
5335 def : Pat<(v8i8 (bitconvert (v2i32 DPR:$src))), (v8i8 DPR:$src)>;
5336 def : Pat<(v8i8 (bitconvert (v4i16 DPR:$src))), (v8i8 DPR:$src)>;
5337 def : Pat<(v8i8 (bitconvert (f64 DPR:$src))), (v8i8 DPR:$src)>;
5338 def : Pat<(v8i8 (bitconvert (v2f32 DPR:$src))), (v8i8 DPR:$src)>;
5339 def : Pat<(f64 (bitconvert (v1i64 DPR:$src))), (f64 DPR:$src)>;
5340 def : Pat<(f64 (bitconvert (v2i32 DPR:$src))), (f64 DPR:$src)>;
5341 def : Pat<(f64 (bitconvert (v4i16 DPR:$src))), (f64 DPR:$src)>;
5342 def : Pat<(f64 (bitconvert (v8i8 DPR:$src))), (f64 DPR:$src)>;
5343 def : Pat<(f64 (bitconvert (v2f32 DPR:$src))), (f64 DPR:$src)>;
5344 def : Pat<(v2f32 (bitconvert (f64 DPR:$src))), (v2f32 DPR:$src)>;
5345 def : Pat<(v2f32 (bitconvert (v1i64 DPR:$src))), (v2f32 DPR:$src)>;
5346 def : Pat<(v2f32 (bitconvert (v2i32 DPR:$src))), (v2f32 DPR:$src)>;
5347 def : Pat<(v2f32 (bitconvert (v4i16 DPR:$src))), (v2f32 DPR:$src)>;
5348 def : Pat<(v2f32 (bitconvert (v8i8 DPR:$src))), (v2f32 DPR:$src)>;
5350 def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (v2i64 QPR:$src)>;
5351 def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (v2i64 QPR:$src)>;
5352 def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (v2i64 QPR:$src)>;
5353 def : Pat<(v2i64 (bitconvert (v2f64 QPR:$src))), (v2i64 QPR:$src)>;
5354 def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (v2i64 QPR:$src)>;
5355 def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (v4i32 QPR:$src)>;
5356 def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (v4i32 QPR:$src)>;
5357 def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (v4i32 QPR:$src)>;
5358 def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (v4i32 QPR:$src)>;
5359 def : Pat<(v4i32 (bitconvert (v4f32 QPR:$src))), (v4i32 QPR:$src)>;
5360 def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (v8i16 QPR:$src)>;
5361 def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (v8i16 QPR:$src)>;
5362 def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (v8i16 QPR:$src)>;
5363 def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (v8i16 QPR:$src)>;
5364 def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (v8i16 QPR:$src)>;
5365 def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (v16i8 QPR:$src)>;
5366 def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (v16i8 QPR:$src)>;
5367 def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (v16i8 QPR:$src)>;
5368 def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (v16i8 QPR:$src)>;
5369 def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (v16i8 QPR:$src)>;
5370 def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (v4f32 QPR:$src)>;
5371 def : Pat<(v4f32 (bitconvert (v4i32 QPR:$src))), (v4f32 QPR:$src)>;
5372 def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (v4f32 QPR:$src)>;
5373 def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (v4f32 QPR:$src)>;
5374 def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (v4f32 QPR:$src)>;
5375 def : Pat<(v2f64 (bitconvert (v2i64 QPR:$src))), (v2f64 QPR:$src)>;
5376 def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (v2f64 QPR:$src)>;
5377 def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 QPR:$src)>;
5378 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 QPR:$src)>;
5379 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 QPR:$src)>;
5382 //===----------------------------------------------------------------------===//
5383 // Assembler aliases
5386 def : VFP2InstAlias<"fmdhr${p} $Dd, $Rn",
5387 (VSETLNi32 DPR:$Dd, GPR:$Rn, 1, pred:$p)>;
5388 def : VFP2InstAlias<"fmdlr${p} $Dd, $Rn",
5389 (VSETLNi32 DPR:$Dd, GPR:$Rn, 0, pred:$p)>;
5392 // VADD two-operand aliases.
5393 def : NEONInstAlias<"vadd${p}.i8 $Vdn, $Vm",
5394 (VADDv16i8 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5395 def : NEONInstAlias<"vadd${p}.i16 $Vdn, $Vm",
5396 (VADDv8i16 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5397 def : NEONInstAlias<"vadd${p}.i32 $Vdn, $Vm",
5398 (VADDv4i32 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5399 def : NEONInstAlias<"vadd${p}.i64 $Vdn, $Vm",
5400 (VADDv2i64 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5402 def : NEONInstAlias<"vadd${p}.i8 $Vdn, $Vm",
5403 (VADDv8i8 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5404 def : NEONInstAlias<"vadd${p}.i16 $Vdn, $Vm",
5405 (VADDv4i16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5406 def : NEONInstAlias<"vadd${p}.i32 $Vdn, $Vm",
5407 (VADDv2i32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5408 def : NEONInstAlias<"vadd${p}.i64 $Vdn, $Vm",
5409 (VADDv1i64 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5411 def : NEONInstAlias<"vadd${p}.f32 $Vdn, $Vm",
5412 (VADDfd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5413 def : NEONInstAlias<"vadd${p}.f32 $Vdn, $Vm",
5414 (VADDfq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5416 // VSUB two-operand aliases.
5417 def : NEONInstAlias<"vsub${p}.i8 $Vdn, $Vm",
5418 (VSUBv16i8 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5419 def : NEONInstAlias<"vsub${p}.i16 $Vdn, $Vm",
5420 (VSUBv8i16 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5421 def : NEONInstAlias<"vsub${p}.i32 $Vdn, $Vm",
5422 (VSUBv4i32 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5423 def : NEONInstAlias<"vsub${p}.i64 $Vdn, $Vm",
5424 (VSUBv2i64 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5426 def : NEONInstAlias<"vsub${p}.i8 $Vdn, $Vm",
5427 (VSUBv8i8 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5428 def : NEONInstAlias<"vsub${p}.i16 $Vdn, $Vm",
5429 (VSUBv4i16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5430 def : NEONInstAlias<"vsub${p}.i32 $Vdn, $Vm",
5431 (VSUBv2i32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5432 def : NEONInstAlias<"vsub${p}.i64 $Vdn, $Vm",
5433 (VSUBv1i64 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5435 def : NEONInstAlias<"vsub${p}.f32 $Vdn, $Vm",
5436 (VSUBfd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5437 def : NEONInstAlias<"vsub${p}.f32 $Vdn, $Vm",
5438 (VSUBfq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5440 // VADDW two-operand aliases.
5441 def : NEONInstAlias<"vaddw${p}.s8 $Vdn, $Vm",
5442 (VADDWsv8i16 QPR:$Vdn, QPR:$Vdn, DPR:$Vm, pred:$p)>;
5443 def : NEONInstAlias<"vaddw${p}.s16 $Vdn, $Vm",
5444 (VADDWsv4i32 QPR:$Vdn, QPR:$Vdn, DPR:$Vm, pred:$p)>;
5445 def : NEONInstAlias<"vaddw${p}.s32 $Vdn, $Vm",
5446 (VADDWsv2i64 QPR:$Vdn, QPR:$Vdn, DPR:$Vm, pred:$p)>;
5447 def : NEONInstAlias<"vaddw${p}.u8 $Vdn, $Vm",
5448 (VADDWuv8i16 QPR:$Vdn, QPR:$Vdn, DPR:$Vm, pred:$p)>;
5449 def : NEONInstAlias<"vaddw${p}.u16 $Vdn, $Vm",
5450 (VADDWuv4i32 QPR:$Vdn, QPR:$Vdn, DPR:$Vm, pred:$p)>;
5451 def : NEONInstAlias<"vaddw${p}.u32 $Vdn, $Vm",
5452 (VADDWuv2i64 QPR:$Vdn, QPR:$Vdn, DPR:$Vm, pred:$p)>;
5454 // VAND/VBIC/VEOR/VORR accept but do not require a type suffix.
5455 defm : VFPDTAnyInstAlias<"vand${p}", "$Vd, $Vn, $Vm",
5456 (VANDd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
5457 defm : VFPDTAnyInstAlias<"vand${p}", "$Vd, $Vn, $Vm",
5458 (VANDq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
5459 defm : VFPDTAnyInstAlias<"vbic${p}", "$Vd, $Vn, $Vm",
5460 (VBICd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
5461 defm : VFPDTAnyInstAlias<"vbic${p}", "$Vd, $Vn, $Vm",
5462 (VBICq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
5463 defm : VFPDTAnyInstAlias<"veor${p}", "$Vd, $Vn, $Vm",
5464 (VEORd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
5465 defm : VFPDTAnyInstAlias<"veor${p}", "$Vd, $Vn, $Vm",
5466 (VEORq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
5467 defm : VFPDTAnyInstAlias<"vorr${p}", "$Vd, $Vn, $Vm",
5468 (VORRd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
5469 defm : VFPDTAnyInstAlias<"vorr${p}", "$Vd, $Vn, $Vm",
5470 (VORRq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
5471 // ... two-operand aliases
5472 def : NEONInstAlias<"vand${p} $Vdn, $Vm",
5473 (VANDd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5474 def : NEONInstAlias<"vand${p} $Vdn, $Vm",
5475 (VANDq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5476 def : NEONInstAlias<"vbic${p} $Vdn, $Vm",
5477 (VBICd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5478 def : NEONInstAlias<"vbic${p} $Vdn, $Vm",
5479 (VBICq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5480 def : NEONInstAlias<"veor${p} $Vdn, $Vm",
5481 (VEORd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5482 def : NEONInstAlias<"veor${p} $Vdn, $Vm",
5483 (VEORq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5484 def : NEONInstAlias<"vorr${p} $Vdn, $Vm",
5485 (VORRd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5486 def : NEONInstAlias<"vorr${p} $Vdn, $Vm",
5487 (VORRq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5489 defm : VFPDTAnyInstAlias<"vand${p}", "$Vdn, $Vm",
5490 (VANDd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5491 defm : VFPDTAnyInstAlias<"vand${p}", "$Vdn, $Vm",
5492 (VANDq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5493 defm : VFPDTAnyInstAlias<"veor${p}", "$Vdn, $Vm",
5494 (VEORd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5495 defm : VFPDTAnyInstAlias<"veor${p}", "$Vdn, $Vm",
5496 (VEORq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5497 defm : VFPDTAnyInstAlias<"vorr${p}", "$Vdn, $Vm",
5498 (VORRd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5499 defm : VFPDTAnyInstAlias<"vorr${p}", "$Vdn, $Vm",
5500 (VORRq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5502 // VMUL two-operand aliases.
5503 def : NEONInstAlias<"vmul${p}.p8 $Qdn, $Qm",
5504 (VMULpq QPR:$Qdn, QPR:$Qdn, QPR:$Qm, pred:$p)>;
5505 def : NEONInstAlias<"vmul${p}.i8 $Qdn, $Qm",
5506 (VMULv16i8 QPR:$Qdn, QPR:$Qdn, QPR:$Qm, pred:$p)>;
5507 def : NEONInstAlias<"vmul${p}.i16 $Qdn, $Qm",
5508 (VMULv8i16 QPR:$Qdn, QPR:$Qdn, QPR:$Qm, pred:$p)>;
5509 def : NEONInstAlias<"vmul${p}.i32 $Qdn, $Qm",
5510 (VMULv4i32 QPR:$Qdn, QPR:$Qdn, QPR:$Qm, pred:$p)>;
5512 def : NEONInstAlias<"vmul${p}.p8 $Ddn, $Dm",
5513 (VMULpd DPR:$Ddn, DPR:$Ddn, DPR:$Dm, pred:$p)>;
5514 def : NEONInstAlias<"vmul${p}.i8 $Ddn, $Dm",
5515 (VMULv8i8 DPR:$Ddn, DPR:$Ddn, DPR:$Dm, pred:$p)>;
5516 def : NEONInstAlias<"vmul${p}.i16 $Ddn, $Dm",
5517 (VMULv4i16 DPR:$Ddn, DPR:$Ddn, DPR:$Dm, pred:$p)>;
5518 def : NEONInstAlias<"vmul${p}.i32 $Ddn, $Dm",
5519 (VMULv2i32 DPR:$Ddn, DPR:$Ddn, DPR:$Dm, pred:$p)>;
5521 def : NEONInstAlias<"vmul${p}.f32 $Qdn, $Qm",
5522 (VMULfq QPR:$Qdn, QPR:$Qdn, QPR:$Qm, pred:$p)>;
5523 def : NEONInstAlias<"vmul${p}.f32 $Ddn, $Dm",
5524 (VMULfd DPR:$Ddn, DPR:$Ddn, DPR:$Dm, pred:$p)>;
5526 def : NEONInstAlias<"vmul${p}.i16 $Ddn, $Dm$lane",
5527 (VMULslv4i16 DPR:$Ddn, DPR:$Ddn, DPR_8:$Dm,
5528 VectorIndex16:$lane, pred:$p)>;
5529 def : NEONInstAlias<"vmul${p}.i16 $Qdn, $Dm$lane",
5530 (VMULslv8i16 QPR:$Qdn, QPR:$Qdn, DPR_8:$Dm,
5531 VectorIndex16:$lane, pred:$p)>;
5533 def : NEONInstAlias<"vmul${p}.i32 $Ddn, $Dm$lane",
5534 (VMULslv2i32 DPR:$Ddn, DPR:$Ddn, DPR_VFP2:$Dm,
5535 VectorIndex32:$lane, pred:$p)>;
5536 def : NEONInstAlias<"vmul${p}.i32 $Qdn, $Dm$lane",
5537 (VMULslv4i32 QPR:$Qdn, QPR:$Qdn, DPR_VFP2:$Dm,
5538 VectorIndex32:$lane, pred:$p)>;
5540 def : NEONInstAlias<"vmul${p}.f32 $Ddn, $Dm$lane",
5541 (VMULslfd DPR:$Ddn, DPR:$Ddn, DPR_VFP2:$Dm,
5542 VectorIndex32:$lane, pred:$p)>;
5543 def : NEONInstAlias<"vmul${p}.f32 $Qdn, $Dm$lane",
5544 (VMULslfq QPR:$Qdn, QPR:$Qdn, DPR_VFP2:$Dm,
5545 VectorIndex32:$lane, pred:$p)>;
5547 // VQADD (register) two-operand aliases.
5548 def : NEONInstAlias<"vqadd${p}.s8 $Vdn, $Vm",
5549 (VQADDsv8i8 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5550 def : NEONInstAlias<"vqadd${p}.s16 $Vdn, $Vm",
5551 (VQADDsv4i16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5552 def : NEONInstAlias<"vqadd${p}.s32 $Vdn, $Vm",
5553 (VQADDsv2i32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5554 def : NEONInstAlias<"vqadd${p}.s64 $Vdn, $Vm",
5555 (VQADDsv1i64 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5556 def : NEONInstAlias<"vqadd${p}.u8 $Vdn, $Vm",
5557 (VQADDuv8i8 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5558 def : NEONInstAlias<"vqadd${p}.u16 $Vdn, $Vm",
5559 (VQADDuv4i16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5560 def : NEONInstAlias<"vqadd${p}.u32 $Vdn, $Vm",
5561 (VQADDuv2i32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5562 def : NEONInstAlias<"vqadd${p}.u64 $Vdn, $Vm",
5563 (VQADDuv1i64 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5565 def : NEONInstAlias<"vqadd${p}.s8 $Vdn, $Vm",
5566 (VQADDsv16i8 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5567 def : NEONInstAlias<"vqadd${p}.s16 $Vdn, $Vm",
5568 (VQADDsv8i16 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5569 def : NEONInstAlias<"vqadd${p}.s32 $Vdn, $Vm",
5570 (VQADDsv4i32 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5571 def : NEONInstAlias<"vqadd${p}.s64 $Vdn, $Vm",
5572 (VQADDsv2i64 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5573 def : NEONInstAlias<"vqadd${p}.u8 $Vdn, $Vm",
5574 (VQADDuv16i8 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5575 def : NEONInstAlias<"vqadd${p}.u16 $Vdn, $Vm",
5576 (VQADDuv8i16 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5577 def : NEONInstAlias<"vqadd${p}.u32 $Vdn, $Vm",
5578 (VQADDuv4i32 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5579 def : NEONInstAlias<"vqadd${p}.u64 $Vdn, $Vm",
5580 (VQADDuv2i64 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5582 // VSHL (immediate) two-operand aliases.
5583 def : NEONInstAlias<"vshl${p}.i8 $Vdn, $imm",
5584 (VSHLiv8i8 DPR:$Vdn, DPR:$Vdn, imm0_7:$imm, pred:$p)>;
5585 def : NEONInstAlias<"vshl${p}.i16 $Vdn, $imm",
5586 (VSHLiv4i16 DPR:$Vdn, DPR:$Vdn, imm0_15:$imm, pred:$p)>;
5587 def : NEONInstAlias<"vshl${p}.i32 $Vdn, $imm",
5588 (VSHLiv2i32 DPR:$Vdn, DPR:$Vdn, imm0_31:$imm, pred:$p)>;
5589 def : NEONInstAlias<"vshl${p}.i64 $Vdn, $imm",
5590 (VSHLiv1i64 DPR:$Vdn, DPR:$Vdn, imm0_63:$imm, pred:$p)>;
5592 def : NEONInstAlias<"vshl${p}.i8 $Vdn, $imm",
5593 (VSHLiv16i8 QPR:$Vdn, QPR:$Vdn, imm0_7:$imm, pred:$p)>;
5594 def : NEONInstAlias<"vshl${p}.i16 $Vdn, $imm",
5595 (VSHLiv8i16 QPR:$Vdn, QPR:$Vdn, imm0_15:$imm, pred:$p)>;
5596 def : NEONInstAlias<"vshl${p}.i32 $Vdn, $imm",
5597 (VSHLiv4i32 QPR:$Vdn, QPR:$Vdn, imm0_31:$imm, pred:$p)>;
5598 def : NEONInstAlias<"vshl${p}.i64 $Vdn, $imm",
5599 (VSHLiv2i64 QPR:$Vdn, QPR:$Vdn, imm0_63:$imm, pred:$p)>;
5601 // VSHL (register) two-operand aliases.
5602 def : NEONInstAlias<"vshl${p}.s8 $Vdn, $Vm",
5603 (VSHLsv8i8 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5604 def : NEONInstAlias<"vshl${p}.s16 $Vdn, $Vm",
5605 (VSHLsv4i16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5606 def : NEONInstAlias<"vshl${p}.s32 $Vdn, $Vm",
5607 (VSHLsv2i32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5608 def : NEONInstAlias<"vshl${p}.s64 $Vdn, $Vm",
5609 (VSHLsv1i64 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5610 def : NEONInstAlias<"vshl${p}.u8 $Vdn, $Vm",
5611 (VSHLuv8i8 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5612 def : NEONInstAlias<"vshl${p}.u16 $Vdn, $Vm",
5613 (VSHLuv4i16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5614 def : NEONInstAlias<"vshl${p}.u32 $Vdn, $Vm",
5615 (VSHLuv2i32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5616 def : NEONInstAlias<"vshl${p}.u64 $Vdn, $Vm",
5617 (VSHLuv1i64 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5619 def : NEONInstAlias<"vshl${p}.s8 $Vdn, $Vm",
5620 (VSHLsv16i8 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5621 def : NEONInstAlias<"vshl${p}.s16 $Vdn, $Vm",
5622 (VSHLsv8i16 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5623 def : NEONInstAlias<"vshl${p}.s32 $Vdn, $Vm",
5624 (VSHLsv4i32 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5625 def : NEONInstAlias<"vshl${p}.s64 $Vdn, $Vm",
5626 (VSHLsv2i64 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5627 def : NEONInstAlias<"vshl${p}.u8 $Vdn, $Vm",
5628 (VSHLuv16i8 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5629 def : NEONInstAlias<"vshl${p}.u16 $Vdn, $Vm",
5630 (VSHLuv8i16 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5631 def : NEONInstAlias<"vshl${p}.u32 $Vdn, $Vm",
5632 (VSHLuv4i32 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5633 def : NEONInstAlias<"vshl${p}.u64 $Vdn, $Vm",
5634 (VSHLuv2i64 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5636 // VSHL (immediate) two-operand aliases.
5637 def : NEONInstAlias<"vshr${p}.s8 $Vdn, $imm",
5638 (VSHRsv8i8 DPR:$Vdn, DPR:$Vdn, shr_imm8:$imm, pred:$p)>;
5639 def : NEONInstAlias<"vshr${p}.s16 $Vdn, $imm",
5640 (VSHRsv4i16 DPR:$Vdn, DPR:$Vdn, shr_imm16:$imm, pred:$p)>;
5641 def : NEONInstAlias<"vshr${p}.s32 $Vdn, $imm",
5642 (VSHRsv2i32 DPR:$Vdn, DPR:$Vdn, shr_imm32:$imm, pred:$p)>;
5643 def : NEONInstAlias<"vshr${p}.s64 $Vdn, $imm",
5644 (VSHRsv1i64 DPR:$Vdn, DPR:$Vdn, shr_imm64:$imm, pred:$p)>;
5646 def : NEONInstAlias<"vshr${p}.s8 $Vdn, $imm",
5647 (VSHRsv16i8 QPR:$Vdn, QPR:$Vdn, shr_imm8:$imm, pred:$p)>;
5648 def : NEONInstAlias<"vshr${p}.s16 $Vdn, $imm",
5649 (VSHRsv8i16 QPR:$Vdn, QPR:$Vdn, shr_imm16:$imm, pred:$p)>;
5650 def : NEONInstAlias<"vshr${p}.s32 $Vdn, $imm",
5651 (VSHRsv4i32 QPR:$Vdn, QPR:$Vdn, shr_imm32:$imm, pred:$p)>;
5652 def : NEONInstAlias<"vshr${p}.s64 $Vdn, $imm",
5653 (VSHRsv2i64 QPR:$Vdn, QPR:$Vdn, shr_imm64:$imm, pred:$p)>;
5655 def : NEONInstAlias<"vshr${p}.u8 $Vdn, $imm",
5656 (VSHRuv8i8 DPR:$Vdn, DPR:$Vdn, shr_imm8:$imm, pred:$p)>;
5657 def : NEONInstAlias<"vshr${p}.u16 $Vdn, $imm",
5658 (VSHRuv4i16 DPR:$Vdn, DPR:$Vdn, shr_imm16:$imm, pred:$p)>;
5659 def : NEONInstAlias<"vshr${p}.u32 $Vdn, $imm",
5660 (VSHRuv2i32 DPR:$Vdn, DPR:$Vdn, shr_imm32:$imm, pred:$p)>;
5661 def : NEONInstAlias<"vshr${p}.u64 $Vdn, $imm",
5662 (VSHRuv1i64 DPR:$Vdn, DPR:$Vdn, shr_imm64:$imm, pred:$p)>;
5664 def : NEONInstAlias<"vshr${p}.u8 $Vdn, $imm",
5665 (VSHRuv16i8 QPR:$Vdn, QPR:$Vdn, shr_imm8:$imm, pred:$p)>;
5666 def : NEONInstAlias<"vshr${p}.u16 $Vdn, $imm",
5667 (VSHRuv8i16 QPR:$Vdn, QPR:$Vdn, shr_imm16:$imm, pred:$p)>;
5668 def : NEONInstAlias<"vshr${p}.u32 $Vdn, $imm",
5669 (VSHRuv4i32 QPR:$Vdn, QPR:$Vdn, shr_imm32:$imm, pred:$p)>;
5670 def : NEONInstAlias<"vshr${p}.u64 $Vdn, $imm",
5671 (VSHRuv2i64 QPR:$Vdn, QPR:$Vdn, shr_imm64:$imm, pred:$p)>;
5673 // VLD1 single-lane pseudo-instructions. These need special handling for
5674 // the lane index that an InstAlias can't handle, so we use these instead.
5675 defm VLD1LNdAsm : NEONDT8AsmPseudoInst<"vld1${p}", "$list, $addr",
5676 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5677 defm VLD1LNdAsm : NEONDT16AsmPseudoInst<"vld1${p}", "$list, $addr",
5678 (ins VecListOneDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5679 defm VLD1LNdAsm : NEONDT32AsmPseudoInst<"vld1${p}", "$list, $addr",
5680 (ins VecListOneDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5682 defm VLD1LNdWB_fixed_Asm : NEONDT8AsmPseudoInst<"vld1${p}", "$list, $addr!",
5683 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5684 defm VLD1LNdWB_fixed_Asm : NEONDT16AsmPseudoInst<"vld1${p}", "$list, $addr!",
5685 (ins VecListOneDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5686 defm VLD1LNdWB_fixed_Asm : NEONDT32AsmPseudoInst<"vld1${p}", "$list, $addr!",
5687 (ins VecListOneDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5688 defm VLD1LNdWB_register_Asm :
5689 NEONDT8AsmPseudoInst<"vld1${p}", "$list, $addr, $Rm",
5690 (ins VecListOneDByteIndexed:$list, addrmode6:$addr,
5691 rGPR:$Rm, pred:$p)>;
5692 defm VLD1LNdWB_register_Asm :
5693 NEONDT16AsmPseudoInst<"vld1${p}", "$list, $addr, $Rm",
5694 (ins VecListOneDHWordIndexed:$list, addrmode6:$addr,
5695 rGPR:$Rm, pred:$p)>;
5696 defm VLD1LNdWB_register_Asm :
5697 NEONDT32AsmPseudoInst<"vld1${p}", "$list, $addr, $Rm",
5698 (ins VecListOneDWordIndexed:$list, addrmode6:$addr,
5699 rGPR:$Rm, pred:$p)>;
5702 // VST1 single-lane pseudo-instructions. These need special handling for
5703 // the lane index that an InstAlias can't handle, so we use these instead.
5704 defm VST1LNdAsm : NEONDT8AsmPseudoInst<"vst1${p}", "$list, $addr",
5705 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5706 defm VST1LNdAsm : NEONDT16AsmPseudoInst<"vst1${p}", "$list, $addr",
5707 (ins VecListOneDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5708 defm VST1LNdAsm : NEONDT32AsmPseudoInst<"vst1${p}", "$list, $addr",
5709 (ins VecListOneDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5711 defm VST1LNdWB_fixed_Asm : NEONDT8AsmPseudoInst<"vst1${p}", "$list, $addr!",
5712 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5713 defm VST1LNdWB_fixed_Asm : NEONDT16AsmPseudoInst<"vst1${p}", "$list, $addr!",
5714 (ins VecListOneDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5715 defm VST1LNdWB_fixed_Asm : NEONDT32AsmPseudoInst<"vst1${p}", "$list, $addr!",
5716 (ins VecListOneDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5717 defm VST1LNdWB_register_Asm :
5718 NEONDT8AsmPseudoInst<"vst1${p}", "$list, $addr, $Rm",
5719 (ins VecListOneDByteIndexed:$list, addrmode6:$addr,
5720 rGPR:$Rm, pred:$p)>;
5721 defm VST1LNdWB_register_Asm :
5722 NEONDT16AsmPseudoInst<"vst1${p}", "$list, $addr, $Rm",
5723 (ins VecListOneDHWordIndexed:$list, addrmode6:$addr,
5724 rGPR:$Rm, pred:$p)>;
5725 defm VST1LNdWB_register_Asm :
5726 NEONDT32AsmPseudoInst<"vst1${p}", "$list, $addr, $Rm",
5727 (ins VecListOneDWordIndexed:$list, addrmode6:$addr,
5728 rGPR:$Rm, pred:$p)>;
5730 // VLD2 single-lane pseudo-instructions. These need special handling for
5731 // the lane index that an InstAlias can't handle, so we use these instead.
5732 defm VLD2LNdAsm : NEONDT8AsmPseudoInst<"vld2${p}", "$list, $addr",
5733 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5734 defm VLD2LNdAsm : NEONDT16AsmPseudoInst<"vld2${p}", "$list, $addr",
5735 (ins VecListTwoDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5736 defm VLD2LNdAsm : NEONDT32AsmPseudoInst<"vld2${p}", "$list, $addr",
5737 (ins VecListTwoDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5739 defm VLD2LNdWB_fixed_Asm : NEONDT8AsmPseudoInst<"vld2${p}", "$list, $addr!",
5740 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5741 defm VLD2LNdWB_fixed_Asm : NEONDT16AsmPseudoInst<"vld2${p}", "$list, $addr!",
5742 (ins VecListTwoDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5743 defm VLD2LNdWB_fixed_Asm : NEONDT32AsmPseudoInst<"vld2${p}", "$list, $addr!",
5744 (ins VecListTwoDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5745 defm VLD2LNdWB_register_Asm :
5746 NEONDT8AsmPseudoInst<"vld2${p}", "$list, $addr, $Rm",
5747 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr,
5748 rGPR:$Rm, pred:$p)>;
5749 defm VLD2LNdWB_register_Asm :
5750 NEONDT16AsmPseudoInst<"vld2${p}", "$list, $addr, $Rm",
5751 (ins VecListTwoDHWordIndexed:$list, addrmode6:$addr,
5752 rGPR:$Rm, pred:$p)>;
5753 defm VLD2LNdWB_register_Asm :
5754 NEONDT32AsmPseudoInst<"vld2${p}", "$list, $addr, $Rm",
5755 (ins VecListTwoDWordIndexed:$list, addrmode6:$addr,
5756 rGPR:$Rm, pred:$p)>;
5759 // VST2 single-lane pseudo-instructions. These need special handling for
5760 // the lane index that an InstAlias can't handle, so we use these instead.
5761 defm VST2LNdAsm : NEONDT8AsmPseudoInst<"vst2${p}", "$list, $addr",
5762 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5763 defm VST2LNdAsm : NEONDT16AsmPseudoInst<"vst2${p}", "$list, $addr",
5764 (ins VecListTwoDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5765 defm VST2LNdAsm : NEONDT32AsmPseudoInst<"vst2${p}", "$list, $addr",
5766 (ins VecListTwoDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5768 defm VST2LNdWB_fixed_Asm : NEONDT8AsmPseudoInst<"vst2${p}", "$list, $addr!",
5769 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5770 defm VST2LNdWB_fixed_Asm : NEONDT16AsmPseudoInst<"vst2${p}", "$list, $addr!",
5771 (ins VecListTwoDHWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5772 defm VST2LNdWB_fixed_Asm : NEONDT32AsmPseudoInst<"vst2${p}", "$list, $addr!",
5773 (ins VecListTwoDWordIndexed:$list, addrmode6:$addr, pred:$p)>;
5774 defm VST2LNdWB_register_Asm :
5775 NEONDT8AsmPseudoInst<"vst2${p}", "$list, $addr, $Rm",
5776 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr,
5777 rGPR:$Rm, pred:$p)>;
5778 defm VST2LNdWB_register_Asm :
5779 NEONDT16AsmPseudoInst<"vst2${p}", "$list, $addr, $Rm",
5780 (ins VecListTwoDHWordIndexed:$list, addrmode6:$addr,
5781 rGPR:$Rm, pred:$p)>;
5782 defm VST2LNdWB_register_Asm :
5783 NEONDT32AsmPseudoInst<"vst2${p}", "$list, $addr, $Rm",
5784 (ins VecListTwoDWordIndexed:$list, addrmode6:$addr,
5785 rGPR:$Rm, pred:$p)>;
5787 // VMOV takes an optional datatype suffix
5788 defm : VFPDTAnyInstAlias<"vmov${p}", "$Vd, $Vm",
5789 (VORRd DPR:$Vd, DPR:$Vm, DPR:$Vm, pred:$p)>;
5790 defm : VFPDTAnyInstAlias<"vmov${p}", "$Vd, $Vm",
5791 (VORRq QPR:$Vd, QPR:$Vm, QPR:$Vm, pred:$p)>;
5793 // VCLT (register) is an assembler alias for VCGT w/ the operands reversed.
5794 // D-register versions.
5795 def : NEONInstAlias<"vcle${p}.s8 $Dd, $Dn, $Dm",
5796 (VCGEsv8i8 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
5797 def : NEONInstAlias<"vcle${p}.s16 $Dd, $Dn, $Dm",
5798 (VCGEsv4i16 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
5799 def : NEONInstAlias<"vcle${p}.s32 $Dd, $Dn, $Dm",
5800 (VCGEsv2i32 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
5801 def : NEONInstAlias<"vcle${p}.u8 $Dd, $Dn, $Dm",
5802 (VCGEuv8i8 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
5803 def : NEONInstAlias<"vcle${p}.u16 $Dd, $Dn, $Dm",
5804 (VCGEuv4i16 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
5805 def : NEONInstAlias<"vcle${p}.u32 $Dd, $Dn, $Dm",
5806 (VCGEuv2i32 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
5807 def : NEONInstAlias<"vcle${p}.f32 $Dd, $Dn, $Dm",
5808 (VCGEfd DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
5809 // Q-register versions.
5810 def : NEONInstAlias<"vcle${p}.s8 $Qd, $Qn, $Qm",
5811 (VCGEsv16i8 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
5812 def : NEONInstAlias<"vcle${p}.s16 $Qd, $Qn, $Qm",
5813 (VCGEsv8i16 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
5814 def : NEONInstAlias<"vcle${p}.s32 $Qd, $Qn, $Qm",
5815 (VCGEsv4i32 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
5816 def : NEONInstAlias<"vcle${p}.u8 $Qd, $Qn, $Qm",
5817 (VCGEuv16i8 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
5818 def : NEONInstAlias<"vcle${p}.u16 $Qd, $Qn, $Qm",
5819 (VCGEuv8i16 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
5820 def : NEONInstAlias<"vcle${p}.u32 $Qd, $Qn, $Qm",
5821 (VCGEuv4i32 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
5822 def : NEONInstAlias<"vcle${p}.f32 $Qd, $Qn, $Qm",
5823 (VCGEfq QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
5825 // VCLT (register) is an assembler alias for VCGT w/ the operands reversed.
5826 // D-register versions.
5827 def : NEONInstAlias<"vclt${p}.s8 $Dd, $Dn, $Dm",
5828 (VCGTsv8i8 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
5829 def : NEONInstAlias<"vclt${p}.s16 $Dd, $Dn, $Dm",
5830 (VCGTsv4i16 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
5831 def : NEONInstAlias<"vclt${p}.s32 $Dd, $Dn, $Dm",
5832 (VCGTsv2i32 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
5833 def : NEONInstAlias<"vclt${p}.u8 $Dd, $Dn, $Dm",
5834 (VCGTuv8i8 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
5835 def : NEONInstAlias<"vclt${p}.u16 $Dd, $Dn, $Dm",
5836 (VCGTuv4i16 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
5837 def : NEONInstAlias<"vclt${p}.u32 $Dd, $Dn, $Dm",
5838 (VCGTuv2i32 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
5839 def : NEONInstAlias<"vclt${p}.f32 $Dd, $Dn, $Dm",
5840 (VCGTfd DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
5841 // Q-register versions.
5842 def : NEONInstAlias<"vclt${p}.s8 $Qd, $Qn, $Qm",
5843 (VCGTsv16i8 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
5844 def : NEONInstAlias<"vclt${p}.s16 $Qd, $Qn, $Qm",
5845 (VCGTsv8i16 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
5846 def : NEONInstAlias<"vclt${p}.s32 $Qd, $Qn, $Qm",
5847 (VCGTsv4i32 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
5848 def : NEONInstAlias<"vclt${p}.u8 $Qd, $Qn, $Qm",
5849 (VCGTuv16i8 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
5850 def : NEONInstAlias<"vclt${p}.u16 $Qd, $Qn, $Qm",
5851 (VCGTuv8i16 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
5852 def : NEONInstAlias<"vclt${p}.u32 $Qd, $Qn, $Qm",
5853 (VCGTuv4i32 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
5854 def : NEONInstAlias<"vclt${p}.f32 $Qd, $Qn, $Qm",
5855 (VCGTfq QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
5857 // Two-operand variants for VEXT
5858 def : NEONInstAlias<"vext${p}.8 $Vdn, $Vm, $imm",
5859 (VEXTd8 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, imm0_7:$imm, pred:$p)>;
5860 def : NEONInstAlias<"vext${p}.16 $Vdn, $Vm, $imm",
5861 (VEXTd16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, imm0_3:$imm, pred:$p)>;
5862 def : NEONInstAlias<"vext${p}.32 $Vdn, $Vm, $imm",
5863 (VEXTd32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, imm0_1:$imm, pred:$p)>;
5865 def : NEONInstAlias<"vext${p}.8 $Vdn, $Vm, $imm",
5866 (VEXTq8 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, imm0_15:$imm, pred:$p)>;
5867 def : NEONInstAlias<"vext${p}.16 $Vdn, $Vm, $imm",
5868 (VEXTq16 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, imm0_7:$imm, pred:$p)>;
5869 def : NEONInstAlias<"vext${p}.32 $Vdn, $Vm, $imm",
5870 (VEXTq32 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, imm0_3:$imm, pred:$p)>;
5871 def : NEONInstAlias<"vext${p}.64 $Vdn, $Vm, $imm",
5872 (VEXTq64 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, imm0_1:$imm, pred:$p)>;
5874 // Two-operand variants for VQDMULH
5875 def : NEONInstAlias<"vqdmulh${p}.s16 $Vdn, $Vm",
5876 (VQDMULHv4i16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5877 def : NEONInstAlias<"vqdmulh${p}.s32 $Vdn, $Vm",
5878 (VQDMULHv2i32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5880 def : NEONInstAlias<"vqdmulh${p}.s16 $Vdn, $Vm",
5881 (VQDMULHv8i16 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5882 def : NEONInstAlias<"vqdmulh${p}.s32 $Vdn, $Vm",
5883 (VQDMULHv4i32 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5885 // Two-operand variants for VMAX.
5886 def : NEONInstAlias<"vmax${p}.s8 $Vdn, $Vm",
5887 (VMAXsv8i8 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5888 def : NEONInstAlias<"vmax${p}.s16 $Vdn, $Vm",
5889 (VMAXsv4i16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5890 def : NEONInstAlias<"vmax${p}.s32 $Vdn, $Vm",
5891 (VMAXsv2i32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5892 def : NEONInstAlias<"vmax${p}.u8 $Vdn, $Vm",
5893 (VMAXuv8i8 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5894 def : NEONInstAlias<"vmax${p}.u16 $Vdn, $Vm",
5895 (VMAXuv4i16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5896 def : NEONInstAlias<"vmax${p}.u32 $Vdn, $Vm",
5897 (VMAXuv2i32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5898 def : NEONInstAlias<"vmax${p}.f32 $Vdn, $Vm",
5899 (VMAXfd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5901 def : NEONInstAlias<"vmax${p}.s8 $Vdn, $Vm",
5902 (VMAXsv16i8 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5903 def : NEONInstAlias<"vmax${p}.s16 $Vdn, $Vm",
5904 (VMAXsv8i16 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5905 def : NEONInstAlias<"vmax${p}.s32 $Vdn, $Vm",
5906 (VMAXsv4i32 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5907 def : NEONInstAlias<"vmax${p}.u8 $Vdn, $Vm",
5908 (VMAXuv16i8 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5909 def : NEONInstAlias<"vmax${p}.u16 $Vdn, $Vm",
5910 (VMAXuv8i16 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5911 def : NEONInstAlias<"vmax${p}.u32 $Vdn, $Vm",
5912 (VMAXuv4i32 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5913 def : NEONInstAlias<"vmax${p}.f32 $Vdn, $Vm",
5914 (VMAXfq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5916 // Two-operand variants for VMIN.
5917 def : NEONInstAlias<"vmin${p}.s8 $Vdn, $Vm",
5918 (VMINsv8i8 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5919 def : NEONInstAlias<"vmin${p}.s16 $Vdn, $Vm",
5920 (VMINsv4i16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5921 def : NEONInstAlias<"vmin${p}.s32 $Vdn, $Vm",
5922 (VMINsv2i32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5923 def : NEONInstAlias<"vmin${p}.u8 $Vdn, $Vm",
5924 (VMINuv8i8 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5925 def : NEONInstAlias<"vmin${p}.u16 $Vdn, $Vm",
5926 (VMINuv4i16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5927 def : NEONInstAlias<"vmin${p}.u32 $Vdn, $Vm",
5928 (VMINuv2i32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5929 def : NEONInstAlias<"vmin${p}.f32 $Vdn, $Vm",
5930 (VMINfd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5932 def : NEONInstAlias<"vmin${p}.s8 $Vdn, $Vm",
5933 (VMINsv16i8 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5934 def : NEONInstAlias<"vmin${p}.s16 $Vdn, $Vm",
5935 (VMINsv8i16 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5936 def : NEONInstAlias<"vmin${p}.s32 $Vdn, $Vm",
5937 (VMINsv4i32 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5938 def : NEONInstAlias<"vmin${p}.u8 $Vdn, $Vm",
5939 (VMINuv16i8 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5940 def : NEONInstAlias<"vmin${p}.u16 $Vdn, $Vm",
5941 (VMINuv8i16 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5942 def : NEONInstAlias<"vmin${p}.u32 $Vdn, $Vm",
5943 (VMINuv4i32 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5944 def : NEONInstAlias<"vmin${p}.f32 $Vdn, $Vm",
5945 (VMINfq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5947 // Two-operand variants for VPADD.
5948 def : NEONInstAlias<"vpadd${p}.i8 $Vdn, $Vm",
5949 (VPADDi8 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5950 def : NEONInstAlias<"vpadd${p}.i16 $Vdn, $Vm",
5951 (VPADDi16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5952 def : NEONInstAlias<"vpadd${p}.i32 $Vdn, $Vm",
5953 (VPADDi32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5954 def : NEONInstAlias<"vpadd${p}.f32 $Vdn, $Vm",
5955 (VPADDf DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5957 // "vmov Rd, #-imm" can be handled via "vmvn".
5958 def : NEONInstAlias<"vmov${p}.i32 $Vd, $imm",
5959 (VMVNv2i32 DPR:$Vd, nImmVMOVI32Neg:$imm, pred:$p)>;
5960 def : NEONInstAlias<"vmov${p}.i32 $Vd, $imm",
5961 (VMVNv4i32 QPR:$Vd, nImmVMOVI32Neg:$imm, pred:$p)>;
5962 def : NEONInstAlias<"vmvn${p}.i32 $Vd, $imm",
5963 (VMOVv2i32 DPR:$Vd, nImmVMOVI32Neg:$imm, pred:$p)>;
5964 def : NEONInstAlias<"vmvn${p}.i32 $Vd, $imm",
5965 (VMOVv4i32 QPR:$Vd, nImmVMOVI32Neg:$imm, pred:$p)>;
5967 // 'gas' compatibility aliases for quad-word instructions. Strictly speaking,
5968 // these should restrict to just the Q register variants, but the register
5969 // classes are enough to match correctly regardless, so we keep it simple
5970 // and just use MnemonicAlias.
5971 def : NEONMnemonicAlias<"vbicq", "vbic">;
5972 def : NEONMnemonicAlias<"vandq", "vand">;
5973 def : NEONMnemonicAlias<"veorq", "veor">;
5974 def : NEONMnemonicAlias<"vorrq", "vorr">;
5976 def : NEONMnemonicAlias<"vmovq", "vmov">;
5977 def : NEONMnemonicAlias<"vmvnq", "vmvn">;
5978 // Explicit versions for floating point so that the FPImm variants get
5979 // handled early. The parser gets confused otherwise.
5980 def : NEONMnemonicAlias<"vmovq.f32", "vmov.f32">;
5981 def : NEONMnemonicAlias<"vmovq.f64", "vmov.f64">;
5983 def : NEONMnemonicAlias<"vaddq", "vadd">;
5984 def : NEONMnemonicAlias<"vsubq", "vsub">;
5986 def : NEONMnemonicAlias<"vminq", "vmin">;
5987 def : NEONMnemonicAlias<"vmaxq", "vmax">;
5989 def : NEONMnemonicAlias<"vmulq", "vmul">;
5991 def : NEONMnemonicAlias<"vabsq", "vabs">;
5993 def : NEONMnemonicAlias<"vshlq", "vshl">;
5994 def : NEONMnemonicAlias<"vshrq", "vshr">;
5996 def : NEONMnemonicAlias<"vcvtq", "vcvt">;
5998 def : NEONMnemonicAlias<"vcleq", "vcle">;
5999 def : NEONMnemonicAlias<"vceqq", "vceq">;