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 nImmVMOVF32 : Operand<i32> {
43 let PrintMethod = "printFPImmOperand";
44 let ParserMatchClass = FPImmOperand;
46 def nImmSplatI64AsmOperand : AsmOperandClass { let Name = "NEONi64splat"; }
47 def nImmSplatI64 : Operand<i32> {
48 let PrintMethod = "printNEONModImmOperand";
49 let ParserMatchClass = nImmSplatI64AsmOperand;
52 def VectorIndex8Operand : AsmOperandClass { let Name = "VectorIndex8"; }
53 def VectorIndex16Operand : AsmOperandClass { let Name = "VectorIndex16"; }
54 def VectorIndex32Operand : AsmOperandClass { let Name = "VectorIndex32"; }
55 def VectorIndex8 : Operand<i32>, ImmLeaf<i32, [{
56 return ((uint64_t)Imm) < 8;
58 let ParserMatchClass = VectorIndex8Operand;
59 let PrintMethod = "printVectorIndex";
60 let MIOperandInfo = (ops i32imm);
62 def VectorIndex16 : Operand<i32>, ImmLeaf<i32, [{
63 return ((uint64_t)Imm) < 4;
65 let ParserMatchClass = VectorIndex16Operand;
66 let PrintMethod = "printVectorIndex";
67 let MIOperandInfo = (ops i32imm);
69 def VectorIndex32 : Operand<i32>, ImmLeaf<i32, [{
70 return ((uint64_t)Imm) < 2;
72 let ParserMatchClass = VectorIndex32Operand;
73 let PrintMethod = "printVectorIndex";
74 let MIOperandInfo = (ops i32imm);
77 // Register list of one D register.
78 def VecListOneDAsmOperand : AsmOperandClass {
79 let Name = "VecListOneD";
80 let ParserMethod = "parseVectorList";
81 let RenderMethod = "addVecListOperands";
83 def VecListOneD : RegisterOperand<DPR, "printVectorListOne"> {
84 let ParserMatchClass = VecListOneDAsmOperand;
86 // Register list of two sequential D registers.
87 def VecListTwoDAsmOperand : AsmOperandClass {
88 let Name = "VecListTwoD";
89 let ParserMethod = "parseVectorList";
90 let RenderMethod = "addVecListOperands";
92 def VecListTwoD : RegisterOperand<DPR, "printVectorListTwo"> {
93 let ParserMatchClass = VecListTwoDAsmOperand;
95 // Register list of three sequential D registers.
96 def VecListThreeDAsmOperand : AsmOperandClass {
97 let Name = "VecListThreeD";
98 let ParserMethod = "parseVectorList";
99 let RenderMethod = "addVecListOperands";
101 def VecListThreeD : RegisterOperand<DPR, "printVectorListThree"> {
102 let ParserMatchClass = VecListThreeDAsmOperand;
104 // Register list of four sequential D registers.
105 def VecListFourDAsmOperand : AsmOperandClass {
106 let Name = "VecListFourD";
107 let ParserMethod = "parseVectorList";
108 let RenderMethod = "addVecListOperands";
110 def VecListFourD : RegisterOperand<DPR, "printVectorListFour"> {
111 let ParserMatchClass = VecListFourDAsmOperand;
113 // Register list of two D registers spaced by 2 (two sequential Q registers).
114 def VecListTwoQAsmOperand : AsmOperandClass {
115 let Name = "VecListTwoQ";
116 let ParserMethod = "parseVectorList";
117 let RenderMethod = "addVecListOperands";
119 def VecListTwoQ : RegisterOperand<DPR, "printVectorListTwoSpaced"> {
120 let ParserMatchClass = VecListTwoQAsmOperand;
123 // Register list of one D register, with "all lanes" subscripting.
124 def VecListOneDAllLanesAsmOperand : AsmOperandClass {
125 let Name = "VecListOneDAllLanes";
126 let ParserMethod = "parseVectorList";
127 let RenderMethod = "addVecListOperands";
129 def VecListOneDAllLanes : RegisterOperand<DPR, "printVectorListOneAllLanes"> {
130 let ParserMatchClass = VecListOneDAllLanesAsmOperand;
132 // Register list of two D registers, with "all lanes" subscripting.
133 def VecListTwoDAllLanesAsmOperand : AsmOperandClass {
134 let Name = "VecListTwoDAllLanes";
135 let ParserMethod = "parseVectorList";
136 let RenderMethod = "addVecListOperands";
138 def VecListTwoDAllLanes : RegisterOperand<DPR, "printVectorListTwoAllLanes"> {
139 let ParserMatchClass = VecListTwoDAllLanesAsmOperand;
142 // Register list of one D register, with byte lane subscripting.
143 def VecListOneDByteIndexAsmOperand : AsmOperandClass {
144 let Name = "VecListOneDByteIndexed";
145 let ParserMethod = "parseVectorList";
146 let RenderMethod = "addVecListIndexedOperands";
148 def VecListOneDByteIndexed : Operand<i32> {
149 let ParserMatchClass = VecListOneDByteIndexAsmOperand;
150 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
153 //===----------------------------------------------------------------------===//
154 // NEON-specific DAG Nodes.
155 //===----------------------------------------------------------------------===//
157 def SDTARMVCMP : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisSameAs<1, 2>]>;
158 def SDTARMVCMPZ : SDTypeProfile<1, 1, []>;
160 def NEONvceq : SDNode<"ARMISD::VCEQ", SDTARMVCMP>;
161 def NEONvceqz : SDNode<"ARMISD::VCEQZ", SDTARMVCMPZ>;
162 def NEONvcge : SDNode<"ARMISD::VCGE", SDTARMVCMP>;
163 def NEONvcgez : SDNode<"ARMISD::VCGEZ", SDTARMVCMPZ>;
164 def NEONvclez : SDNode<"ARMISD::VCLEZ", SDTARMVCMPZ>;
165 def NEONvcgeu : SDNode<"ARMISD::VCGEU", SDTARMVCMP>;
166 def NEONvcgt : SDNode<"ARMISD::VCGT", SDTARMVCMP>;
167 def NEONvcgtz : SDNode<"ARMISD::VCGTZ", SDTARMVCMPZ>;
168 def NEONvcltz : SDNode<"ARMISD::VCLTZ", SDTARMVCMPZ>;
169 def NEONvcgtu : SDNode<"ARMISD::VCGTU", SDTARMVCMP>;
170 def NEONvtst : SDNode<"ARMISD::VTST", SDTARMVCMP>;
172 // Types for vector shift by immediates. The "SHX" version is for long and
173 // narrow operations where the source and destination vectors have different
174 // types. The "SHINS" version is for shift and insert operations.
175 def SDTARMVSH : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
177 def SDTARMVSHX : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
179 def SDTARMVSHINS : SDTypeProfile<1, 3, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
180 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
182 def NEONvshl : SDNode<"ARMISD::VSHL", SDTARMVSH>;
183 def NEONvshrs : SDNode<"ARMISD::VSHRs", SDTARMVSH>;
184 def NEONvshru : SDNode<"ARMISD::VSHRu", SDTARMVSH>;
185 def NEONvshlls : SDNode<"ARMISD::VSHLLs", SDTARMVSHX>;
186 def NEONvshllu : SDNode<"ARMISD::VSHLLu", SDTARMVSHX>;
187 def NEONvshlli : SDNode<"ARMISD::VSHLLi", SDTARMVSHX>;
188 def NEONvshrn : SDNode<"ARMISD::VSHRN", SDTARMVSHX>;
190 def NEONvrshrs : SDNode<"ARMISD::VRSHRs", SDTARMVSH>;
191 def NEONvrshru : SDNode<"ARMISD::VRSHRu", SDTARMVSH>;
192 def NEONvrshrn : SDNode<"ARMISD::VRSHRN", SDTARMVSHX>;
194 def NEONvqshls : SDNode<"ARMISD::VQSHLs", SDTARMVSH>;
195 def NEONvqshlu : SDNode<"ARMISD::VQSHLu", SDTARMVSH>;
196 def NEONvqshlsu : SDNode<"ARMISD::VQSHLsu", SDTARMVSH>;
197 def NEONvqshrns : SDNode<"ARMISD::VQSHRNs", SDTARMVSHX>;
198 def NEONvqshrnu : SDNode<"ARMISD::VQSHRNu", SDTARMVSHX>;
199 def NEONvqshrnsu : SDNode<"ARMISD::VQSHRNsu", SDTARMVSHX>;
201 def NEONvqrshrns : SDNode<"ARMISD::VQRSHRNs", SDTARMVSHX>;
202 def NEONvqrshrnu : SDNode<"ARMISD::VQRSHRNu", SDTARMVSHX>;
203 def NEONvqrshrnsu : SDNode<"ARMISD::VQRSHRNsu", SDTARMVSHX>;
205 def NEONvsli : SDNode<"ARMISD::VSLI", SDTARMVSHINS>;
206 def NEONvsri : SDNode<"ARMISD::VSRI", SDTARMVSHINS>;
208 def SDTARMVGETLN : SDTypeProfile<1, 2, [SDTCisVT<0, i32>, SDTCisInt<1>,
210 def NEONvgetlaneu : SDNode<"ARMISD::VGETLANEu", SDTARMVGETLN>;
211 def NEONvgetlanes : SDNode<"ARMISD::VGETLANEs", SDTARMVGETLN>;
213 def SDTARMVMOVIMM : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisVT<1, i32>]>;
214 def NEONvmovImm : SDNode<"ARMISD::VMOVIMM", SDTARMVMOVIMM>;
215 def NEONvmvnImm : SDNode<"ARMISD::VMVNIMM", SDTARMVMOVIMM>;
216 def NEONvmovFPImm : SDNode<"ARMISD::VMOVFPIMM", SDTARMVMOVIMM>;
218 def SDTARMVORRIMM : SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
220 def NEONvorrImm : SDNode<"ARMISD::VORRIMM", SDTARMVORRIMM>;
221 def NEONvbicImm : SDNode<"ARMISD::VBICIMM", SDTARMVORRIMM>;
223 def NEONvbsl : SDNode<"ARMISD::VBSL",
224 SDTypeProfile<1, 3, [SDTCisVec<0>,
227 SDTCisSameAs<0, 3>]>>;
229 def NEONvdup : SDNode<"ARMISD::VDUP", SDTypeProfile<1, 1, [SDTCisVec<0>]>>;
231 // VDUPLANE can produce a quad-register result from a double-register source,
232 // so the result is not constrained to match the source.
233 def NEONvduplane : SDNode<"ARMISD::VDUPLANE",
234 SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisVec<1>,
237 def SDTARMVEXT : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
238 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
239 def NEONvext : SDNode<"ARMISD::VEXT", SDTARMVEXT>;
241 def SDTARMVSHUF : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisSameAs<0, 1>]>;
242 def NEONvrev64 : SDNode<"ARMISD::VREV64", SDTARMVSHUF>;
243 def NEONvrev32 : SDNode<"ARMISD::VREV32", SDTARMVSHUF>;
244 def NEONvrev16 : SDNode<"ARMISD::VREV16", SDTARMVSHUF>;
246 def SDTARMVSHUF2 : SDTypeProfile<2, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
248 SDTCisSameAs<0, 3>]>;
249 def NEONzip : SDNode<"ARMISD::VZIP", SDTARMVSHUF2>;
250 def NEONuzp : SDNode<"ARMISD::VUZP", SDTARMVSHUF2>;
251 def NEONtrn : SDNode<"ARMISD::VTRN", SDTARMVSHUF2>;
253 def SDTARMVMULL : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
254 SDTCisSameAs<1, 2>]>;
255 def NEONvmulls : SDNode<"ARMISD::VMULLs", SDTARMVMULL>;
256 def NEONvmullu : SDNode<"ARMISD::VMULLu", SDTARMVMULL>;
258 def SDTARMFMAX : SDTypeProfile<1, 2, [SDTCisVT<0, f32>, SDTCisSameAs<0, 1>,
259 SDTCisSameAs<0, 2>]>;
260 def NEONfmax : SDNode<"ARMISD::FMAX", SDTARMFMAX>;
261 def NEONfmin : SDNode<"ARMISD::FMIN", SDTARMFMAX>;
263 def NEONimmAllZerosV: PatLeaf<(NEONvmovImm (i32 timm)), [{
264 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
265 unsigned EltBits = 0;
266 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
267 return (EltBits == 32 && EltVal == 0);
270 def NEONimmAllOnesV: PatLeaf<(NEONvmovImm (i32 timm)), [{
271 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
272 unsigned EltBits = 0;
273 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
274 return (EltBits == 8 && EltVal == 0xff);
277 //===----------------------------------------------------------------------===//
278 // NEON load / store instructions
279 //===----------------------------------------------------------------------===//
281 // Use VLDM to load a Q register as a D register pair.
282 // This is a pseudo instruction that is expanded to VLDMD after reg alloc.
284 : PseudoVFPLdStM<(outs QPR:$dst), (ins GPR:$Rn),
286 [(set QPR:$dst, (v2f64 (load GPR:$Rn)))]>;
288 // Use VSTM to store a Q register as a D register pair.
289 // This is a pseudo instruction that is expanded to VSTMD after reg alloc.
291 : PseudoVFPLdStM<(outs), (ins QPR:$src, GPR:$Rn),
293 [(store (v2f64 QPR:$src), GPR:$Rn)]>;
295 // Classes for VLD* pseudo-instructions with multi-register operands.
296 // These are expanded to real instructions after register allocation.
297 class VLDQPseudo<InstrItinClass itin>
298 : PseudoNLdSt<(outs QPR:$dst), (ins addrmode6:$addr), itin, "">;
299 class VLDQWBPseudo<InstrItinClass itin>
300 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
301 (ins addrmode6:$addr, am6offset:$offset), itin,
303 class VLDQWBfixedPseudo<InstrItinClass itin>
304 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
305 (ins addrmode6:$addr), itin,
307 class VLDQWBregisterPseudo<InstrItinClass itin>
308 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
309 (ins addrmode6:$addr, rGPR:$offset), itin,
312 class VLDQQPseudo<InstrItinClass itin>
313 : PseudoNLdSt<(outs QQPR:$dst), (ins addrmode6:$addr), itin, "">;
314 class VLDQQWBPseudo<InstrItinClass itin>
315 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
316 (ins addrmode6:$addr, am6offset:$offset), itin,
318 class VLDQQWBfixedPseudo<InstrItinClass itin>
319 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
320 (ins addrmode6:$addr), itin,
322 class VLDQQWBregisterPseudo<InstrItinClass itin>
323 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
324 (ins addrmode6:$addr, rGPR:$offset), itin,
328 class VLDQQQQPseudo<InstrItinClass itin>
329 : PseudoNLdSt<(outs QQQQPR:$dst), (ins addrmode6:$addr, QQQQPR:$src),itin,
331 class VLDQQQQWBPseudo<InstrItinClass itin>
332 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
333 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
334 "$addr.addr = $wb, $src = $dst">;
336 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
338 // VLD1 : Vector Load (multiple single elements)
339 class VLD1D<bits<4> op7_4, string Dt>
340 : NLdSt<0,0b10,0b0111,op7_4, (outs VecListOneD:$Vd),
341 (ins addrmode6:$Rn), IIC_VLD1,
342 "vld1", Dt, "$Vd, $Rn", "", []> {
345 let DecoderMethod = "DecodeVLDInstruction";
347 class VLD1Q<bits<4> op7_4, string Dt>
348 : NLdSt<0,0b10,0b1010,op7_4, (outs VecListTwoD:$Vd),
349 (ins addrmode6:$Rn), IIC_VLD1x2,
350 "vld1", Dt, "$Vd, $Rn", "", []> {
352 let Inst{5-4} = Rn{5-4};
353 let DecoderMethod = "DecodeVLDInstruction";
356 def VLD1d8 : VLD1D<{0,0,0,?}, "8">;
357 def VLD1d16 : VLD1D<{0,1,0,?}, "16">;
358 def VLD1d32 : VLD1D<{1,0,0,?}, "32">;
359 def VLD1d64 : VLD1D<{1,1,0,?}, "64">;
361 def VLD1q8 : VLD1Q<{0,0,?,?}, "8">;
362 def VLD1q16 : VLD1Q<{0,1,?,?}, "16">;
363 def VLD1q32 : VLD1Q<{1,0,?,?}, "32">;
364 def VLD1q64 : VLD1Q<{1,1,?,?}, "64">;
366 def VLD1q8Pseudo : VLDQPseudo<IIC_VLD1x2>;
367 def VLD1q16Pseudo : VLDQPseudo<IIC_VLD1x2>;
368 def VLD1q32Pseudo : VLDQPseudo<IIC_VLD1x2>;
369 def VLD1q64Pseudo : VLDQPseudo<IIC_VLD1x2>;
371 // ...with address register writeback:
372 multiclass VLD1DWB<bits<4> op7_4, string Dt> {
373 def _fixed : NLdSt<0,0b10, 0b0111,op7_4, (outs VecListOneD:$Vd, GPR:$wb),
374 (ins addrmode6:$Rn), IIC_VLD1u,
375 "vld1", Dt, "$Vd, $Rn!",
376 "$Rn.addr = $wb", []> {
377 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
379 let DecoderMethod = "DecodeVLDInstruction";
380 let AsmMatchConverter = "cvtVLDwbFixed";
382 def _register : NLdSt<0,0b10,0b0111,op7_4, (outs VecListOneD:$Vd, GPR:$wb),
383 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1u,
384 "vld1", Dt, "$Vd, $Rn, $Rm",
385 "$Rn.addr = $wb", []> {
387 let DecoderMethod = "DecodeVLDInstruction";
388 let AsmMatchConverter = "cvtVLDwbRegister";
391 multiclass VLD1QWB<bits<4> op7_4, string Dt> {
392 def _fixed : NLdSt<0,0b10,0b1010,op7_4, (outs VecListTwoD:$Vd, GPR:$wb),
393 (ins addrmode6:$Rn), IIC_VLD1x2u,
394 "vld1", Dt, "$Vd, $Rn!",
395 "$Rn.addr = $wb", []> {
396 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
397 let Inst{5-4} = Rn{5-4};
398 let DecoderMethod = "DecodeVLDInstruction";
399 let AsmMatchConverter = "cvtVLDwbFixed";
401 def _register : NLdSt<0,0b10,0b1010,op7_4, (outs VecListTwoD:$Vd, GPR:$wb),
402 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1x2u,
403 "vld1", Dt, "$Vd, $Rn, $Rm",
404 "$Rn.addr = $wb", []> {
405 let Inst{5-4} = Rn{5-4};
406 let DecoderMethod = "DecodeVLDInstruction";
407 let AsmMatchConverter = "cvtVLDwbRegister";
411 defm VLD1d8wb : VLD1DWB<{0,0,0,?}, "8">;
412 defm VLD1d16wb : VLD1DWB<{0,1,0,?}, "16">;
413 defm VLD1d32wb : VLD1DWB<{1,0,0,?}, "32">;
414 defm VLD1d64wb : VLD1DWB<{1,1,0,?}, "64">;
415 defm VLD1q8wb : VLD1QWB<{0,0,?,?}, "8">;
416 defm VLD1q16wb : VLD1QWB<{0,1,?,?}, "16">;
417 defm VLD1q32wb : VLD1QWB<{1,0,?,?}, "32">;
418 defm VLD1q64wb : VLD1QWB<{1,1,?,?}, "64">;
420 def VLD1q8PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD1x2u>;
421 def VLD1q16PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD1x2u>;
422 def VLD1q32PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD1x2u>;
423 def VLD1q64PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD1x2u>;
424 def VLD1q8PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD1x2u>;
425 def VLD1q16PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD1x2u>;
426 def VLD1q32PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD1x2u>;
427 def VLD1q64PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD1x2u>;
429 // ...with 3 registers
430 class VLD1D3<bits<4> op7_4, string Dt>
431 : NLdSt<0,0b10,0b0110,op7_4, (outs VecListThreeD:$Vd),
432 (ins addrmode6:$Rn), IIC_VLD1x3, "vld1", Dt,
433 "$Vd, $Rn", "", []> {
436 let DecoderMethod = "DecodeVLDInstruction";
438 multiclass VLD1D3WB<bits<4> op7_4, string Dt> {
439 def _fixed : NLdSt<0,0b10,0b0110, op7_4, (outs VecListThreeD:$Vd, GPR:$wb),
440 (ins addrmode6:$Rn), IIC_VLD1x2u,
441 "vld1", Dt, "$Vd, $Rn!",
442 "$Rn.addr = $wb", []> {
443 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
445 let DecoderMethod = "DecodeVLDInstruction";
446 let AsmMatchConverter = "cvtVLDwbFixed";
448 def _register : NLdSt<0,0b10,0b0110,op7_4, (outs VecListThreeD:$Vd, GPR:$wb),
449 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1x2u,
450 "vld1", Dt, "$Vd, $Rn, $Rm",
451 "$Rn.addr = $wb", []> {
453 let DecoderMethod = "DecodeVLDInstruction";
454 let AsmMatchConverter = "cvtVLDwbRegister";
458 def VLD1d8T : VLD1D3<{0,0,0,?}, "8">;
459 def VLD1d16T : VLD1D3<{0,1,0,?}, "16">;
460 def VLD1d32T : VLD1D3<{1,0,0,?}, "32">;
461 def VLD1d64T : VLD1D3<{1,1,0,?}, "64">;
463 defm VLD1d8Twb : VLD1D3WB<{0,0,0,?}, "8">;
464 defm VLD1d16Twb : VLD1D3WB<{0,1,0,?}, "16">;
465 defm VLD1d32Twb : VLD1D3WB<{1,0,0,?}, "32">;
466 defm VLD1d64Twb : VLD1D3WB<{1,1,0,?}, "64">;
468 def VLD1d64TPseudo : VLDQQPseudo<IIC_VLD1x3>;
470 // ...with 4 registers
471 class VLD1D4<bits<4> op7_4, string Dt>
472 : NLdSt<0, 0b10, 0b0010, op7_4, (outs VecListFourD:$Vd),
473 (ins addrmode6:$Rn), IIC_VLD1x4, "vld1", Dt,
474 "$Vd, $Rn", "", []> {
476 let Inst{5-4} = Rn{5-4};
477 let DecoderMethod = "DecodeVLDInstruction";
479 multiclass VLD1D4WB<bits<4> op7_4, string Dt> {
480 def _fixed : NLdSt<0,0b10,0b0010, op7_4, (outs VecListFourD:$Vd, GPR:$wb),
481 (ins addrmode6:$Rn), IIC_VLD1x2u,
482 "vld1", Dt, "$Vd, $Rn!",
483 "$Rn.addr = $wb", []> {
484 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
485 let Inst{5-4} = Rn{5-4};
486 let DecoderMethod = "DecodeVLDInstruction";
487 let AsmMatchConverter = "cvtVLDwbFixed";
489 def _register : NLdSt<0,0b10,0b0010,op7_4, (outs VecListFourD:$Vd, GPR:$wb),
490 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1x2u,
491 "vld1", Dt, "$Vd, $Rn, $Rm",
492 "$Rn.addr = $wb", []> {
493 let Inst{5-4} = Rn{5-4};
494 let DecoderMethod = "DecodeVLDInstruction";
495 let AsmMatchConverter = "cvtVLDwbRegister";
499 def VLD1d8Q : VLD1D4<{0,0,?,?}, "8">;
500 def VLD1d16Q : VLD1D4<{0,1,?,?}, "16">;
501 def VLD1d32Q : VLD1D4<{1,0,?,?}, "32">;
502 def VLD1d64Q : VLD1D4<{1,1,?,?}, "64">;
504 defm VLD1d8Qwb : VLD1D4WB<{0,0,?,?}, "8">;
505 defm VLD1d16Qwb : VLD1D4WB<{0,1,?,?}, "16">;
506 defm VLD1d32Qwb : VLD1D4WB<{1,0,?,?}, "32">;
507 defm VLD1d64Qwb : VLD1D4WB<{1,1,?,?}, "64">;
509 def VLD1d64QPseudo : VLDQQPseudo<IIC_VLD1x4>;
511 // VLD2 : Vector Load (multiple 2-element structures)
512 class VLD2<bits<4> op11_8, bits<4> op7_4, string Dt, RegisterOperand VdTy,
514 : NLdSt<0, 0b10, op11_8, op7_4, (outs VdTy:$Vd),
515 (ins addrmode6:$Rn), itin,
516 "vld2", Dt, "$Vd, $Rn", "", []> {
518 let Inst{5-4} = Rn{5-4};
519 let DecoderMethod = "DecodeVLDInstruction";
522 def VLD2d8 : VLD2<0b1000, {0,0,?,?}, "8", VecListTwoD, IIC_VLD2>;
523 def VLD2d16 : VLD2<0b1000, {0,1,?,?}, "16", VecListTwoD, IIC_VLD2>;
524 def VLD2d32 : VLD2<0b1000, {1,0,?,?}, "32", VecListTwoD, IIC_VLD2>;
526 def VLD2q8 : VLD2<0b0011, {0,0,?,?}, "8", VecListFourD, IIC_VLD2x2>;
527 def VLD2q16 : VLD2<0b0011, {0,1,?,?}, "16", VecListFourD, IIC_VLD2x2>;
528 def VLD2q32 : VLD2<0b0011, {1,0,?,?}, "32", VecListFourD, IIC_VLD2x2>;
530 def VLD2d8Pseudo : VLDQPseudo<IIC_VLD2>;
531 def VLD2d16Pseudo : VLDQPseudo<IIC_VLD2>;
532 def VLD2d32Pseudo : VLDQPseudo<IIC_VLD2>;
534 def VLD2q8Pseudo : VLDQQPseudo<IIC_VLD2x2>;
535 def VLD2q16Pseudo : VLDQQPseudo<IIC_VLD2x2>;
536 def VLD2q32Pseudo : VLDQQPseudo<IIC_VLD2x2>;
538 // ...with address register writeback:
539 multiclass VLD2WB<bits<4> op11_8, bits<4> op7_4, string Dt,
540 RegisterOperand VdTy, InstrItinClass itin> {
541 def _fixed : NLdSt<0, 0b10, op11_8, op7_4, (outs VdTy:$Vd, GPR:$wb),
542 (ins addrmode6:$Rn), itin,
543 "vld2", Dt, "$Vd, $Rn!",
544 "$Rn.addr = $wb", []> {
545 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
546 let Inst{5-4} = Rn{5-4};
547 let DecoderMethod = "DecodeVLDInstruction";
548 let AsmMatchConverter = "cvtVLDwbFixed";
550 def _register : NLdSt<0, 0b10, op11_8, op7_4, (outs VdTy:$Vd, GPR:$wb),
551 (ins addrmode6:$Rn, rGPR:$Rm), itin,
552 "vld2", Dt, "$Vd, $Rn, $Rm",
553 "$Rn.addr = $wb", []> {
554 let Inst{5-4} = Rn{5-4};
555 let DecoderMethod = "DecodeVLDInstruction";
556 let AsmMatchConverter = "cvtVLDwbRegister";
560 defm VLD2d8wb : VLD2WB<0b1000, {0,0,?,?}, "8", VecListTwoD, IIC_VLD2u>;
561 defm VLD2d16wb : VLD2WB<0b1000, {0,1,?,?}, "16", VecListTwoD, IIC_VLD2u>;
562 defm VLD2d32wb : VLD2WB<0b1000, {1,0,?,?}, "32", VecListTwoD, IIC_VLD2u>;
564 defm VLD2q8wb : VLD2WB<0b0011, {0,0,?,?}, "8", VecListFourD, IIC_VLD2x2u>;
565 defm VLD2q16wb : VLD2WB<0b0011, {0,1,?,?}, "16", VecListFourD, IIC_VLD2x2u>;
566 defm VLD2q32wb : VLD2WB<0b0011, {1,0,?,?}, "32", VecListFourD, IIC_VLD2x2u>;
568 def VLD2d8PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD2u>;
569 def VLD2d16PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD2u>;
570 def VLD2d32PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD2u>;
571 def VLD2d8PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD2u>;
572 def VLD2d16PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD2u>;
573 def VLD2d32PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD2u>;
575 def VLD2q8PseudoWB_fixed : VLDQQWBfixedPseudo<IIC_VLD2x2u>;
576 def VLD2q16PseudoWB_fixed : VLDQQWBfixedPseudo<IIC_VLD2x2u>;
577 def VLD2q32PseudoWB_fixed : VLDQQWBfixedPseudo<IIC_VLD2x2u>;
578 def VLD2q8PseudoWB_register : VLDQQWBregisterPseudo<IIC_VLD2x2u>;
579 def VLD2q16PseudoWB_register : VLDQQWBregisterPseudo<IIC_VLD2x2u>;
580 def VLD2q32PseudoWB_register : VLDQQWBregisterPseudo<IIC_VLD2x2u>;
582 // ...with double-spaced registers
583 def VLD2b8 : VLD2<0b1001, {0,0,?,?}, "8", VecListTwoQ, IIC_VLD2>;
584 def VLD2b16 : VLD2<0b1001, {0,1,?,?}, "16", VecListTwoQ, IIC_VLD2>;
585 def VLD2b32 : VLD2<0b1001, {1,0,?,?}, "32", VecListTwoQ, IIC_VLD2>;
586 defm VLD2b8wb : VLD2WB<0b1001, {0,0,?,?}, "8", VecListTwoQ, IIC_VLD2u>;
587 defm VLD2b16wb : VLD2WB<0b1001, {0,1,?,?}, "16", VecListTwoQ, IIC_VLD2u>;
588 defm VLD2b32wb : VLD2WB<0b1001, {1,0,?,?}, "32", VecListTwoQ, IIC_VLD2u>;
590 // VLD3 : Vector Load (multiple 3-element structures)
591 class VLD3D<bits<4> op11_8, bits<4> op7_4, string Dt>
592 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
593 (ins addrmode6:$Rn), IIC_VLD3,
594 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn", "", []> {
597 let DecoderMethod = "DecodeVLDInstruction";
600 def VLD3d8 : VLD3D<0b0100, {0,0,0,?}, "8">;
601 def VLD3d16 : VLD3D<0b0100, {0,1,0,?}, "16">;
602 def VLD3d32 : VLD3D<0b0100, {1,0,0,?}, "32">;
604 def VLD3d8Pseudo : VLDQQPseudo<IIC_VLD3>;
605 def VLD3d16Pseudo : VLDQQPseudo<IIC_VLD3>;
606 def VLD3d32Pseudo : VLDQQPseudo<IIC_VLD3>;
608 // ...with address register writeback:
609 class VLD3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
610 : NLdSt<0, 0b10, op11_8, op7_4,
611 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
612 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD3u,
613 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn$Rm",
614 "$Rn.addr = $wb", []> {
616 let DecoderMethod = "DecodeVLDInstruction";
619 def VLD3d8_UPD : VLD3DWB<0b0100, {0,0,0,?}, "8">;
620 def VLD3d16_UPD : VLD3DWB<0b0100, {0,1,0,?}, "16">;
621 def VLD3d32_UPD : VLD3DWB<0b0100, {1,0,0,?}, "32">;
623 def VLD3d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
624 def VLD3d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
625 def VLD3d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
627 // ...with double-spaced registers:
628 def VLD3q8 : VLD3D<0b0101, {0,0,0,?}, "8">;
629 def VLD3q16 : VLD3D<0b0101, {0,1,0,?}, "16">;
630 def VLD3q32 : VLD3D<0b0101, {1,0,0,?}, "32">;
631 def VLD3q8_UPD : VLD3DWB<0b0101, {0,0,0,?}, "8">;
632 def VLD3q16_UPD : VLD3DWB<0b0101, {0,1,0,?}, "16">;
633 def VLD3q32_UPD : VLD3DWB<0b0101, {1,0,0,?}, "32">;
635 def VLD3q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
636 def VLD3q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
637 def VLD3q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
639 // ...alternate versions to be allocated odd register numbers:
640 def VLD3q8oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
641 def VLD3q16oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
642 def VLD3q32oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
644 def VLD3q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
645 def VLD3q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
646 def VLD3q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
648 // VLD4 : Vector Load (multiple 4-element structures)
649 class VLD4D<bits<4> op11_8, bits<4> op7_4, string Dt>
650 : NLdSt<0, 0b10, op11_8, op7_4,
651 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
652 (ins addrmode6:$Rn), IIC_VLD4,
653 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn", "", []> {
655 let Inst{5-4} = Rn{5-4};
656 let DecoderMethod = "DecodeVLDInstruction";
659 def VLD4d8 : VLD4D<0b0000, {0,0,?,?}, "8">;
660 def VLD4d16 : VLD4D<0b0000, {0,1,?,?}, "16">;
661 def VLD4d32 : VLD4D<0b0000, {1,0,?,?}, "32">;
663 def VLD4d8Pseudo : VLDQQPseudo<IIC_VLD4>;
664 def VLD4d16Pseudo : VLDQQPseudo<IIC_VLD4>;
665 def VLD4d32Pseudo : VLDQQPseudo<IIC_VLD4>;
667 // ...with address register writeback:
668 class VLD4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
669 : NLdSt<0, 0b10, op11_8, op7_4,
670 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
671 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD4u,
672 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn$Rm",
673 "$Rn.addr = $wb", []> {
674 let Inst{5-4} = Rn{5-4};
675 let DecoderMethod = "DecodeVLDInstruction";
678 def VLD4d8_UPD : VLD4DWB<0b0000, {0,0,?,?}, "8">;
679 def VLD4d16_UPD : VLD4DWB<0b0000, {0,1,?,?}, "16">;
680 def VLD4d32_UPD : VLD4DWB<0b0000, {1,0,?,?}, "32">;
682 def VLD4d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
683 def VLD4d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
684 def VLD4d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
686 // ...with double-spaced registers:
687 def VLD4q8 : VLD4D<0b0001, {0,0,?,?}, "8">;
688 def VLD4q16 : VLD4D<0b0001, {0,1,?,?}, "16">;
689 def VLD4q32 : VLD4D<0b0001, {1,0,?,?}, "32">;
690 def VLD4q8_UPD : VLD4DWB<0b0001, {0,0,?,?}, "8">;
691 def VLD4q16_UPD : VLD4DWB<0b0001, {0,1,?,?}, "16">;
692 def VLD4q32_UPD : VLD4DWB<0b0001, {1,0,?,?}, "32">;
694 def VLD4q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
695 def VLD4q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
696 def VLD4q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
698 // ...alternate versions to be allocated odd register numbers:
699 def VLD4q8oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
700 def VLD4q16oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
701 def VLD4q32oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
703 def VLD4q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
704 def VLD4q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
705 def VLD4q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
707 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
709 // Classes for VLD*LN pseudo-instructions with multi-register operands.
710 // These are expanded to real instructions after register allocation.
711 class VLDQLNPseudo<InstrItinClass itin>
712 : PseudoNLdSt<(outs QPR:$dst),
713 (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
714 itin, "$src = $dst">;
715 class VLDQLNWBPseudo<InstrItinClass itin>
716 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
717 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
718 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
719 class VLDQQLNPseudo<InstrItinClass itin>
720 : PseudoNLdSt<(outs QQPR:$dst),
721 (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
722 itin, "$src = $dst">;
723 class VLDQQLNWBPseudo<InstrItinClass itin>
724 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
725 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
726 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
727 class VLDQQQQLNPseudo<InstrItinClass itin>
728 : PseudoNLdSt<(outs QQQQPR:$dst),
729 (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
730 itin, "$src = $dst">;
731 class VLDQQQQLNWBPseudo<InstrItinClass itin>
732 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
733 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
734 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
736 // VLD1LN : Vector Load (single element to one lane)
737 class VLD1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
739 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd),
740 (ins addrmode6:$Rn, DPR:$src, nohash_imm:$lane),
741 IIC_VLD1ln, "vld1", Dt, "\\{$Vd[$lane]\\}, $Rn",
743 [(set DPR:$Vd, (vector_insert (Ty DPR:$src),
744 (i32 (LoadOp addrmode6:$Rn)),
747 let DecoderMethod = "DecodeVLD1LN";
749 class VLD1LN32<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
751 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd),
752 (ins addrmode6oneL32:$Rn, DPR:$src, nohash_imm:$lane),
753 IIC_VLD1ln, "vld1", Dt, "\\{$Vd[$lane]\\}, $Rn",
755 [(set DPR:$Vd, (vector_insert (Ty DPR:$src),
756 (i32 (LoadOp addrmode6oneL32:$Rn)),
759 let DecoderMethod = "DecodeVLD1LN";
761 class VLD1QLNPseudo<ValueType Ty, PatFrag LoadOp> : VLDQLNPseudo<IIC_VLD1ln> {
762 let Pattern = [(set QPR:$dst, (vector_insert (Ty QPR:$src),
763 (i32 (LoadOp addrmode6:$addr)),
767 def VLD1LNd8 : VLD1LN<0b0000, {?,?,?,0}, "8", v8i8, extloadi8> {
768 let Inst{7-5} = lane{2-0};
770 def VLD1LNd16 : VLD1LN<0b0100, {?,?,0,?}, "16", v4i16, extloadi16> {
771 let Inst{7-6} = lane{1-0};
774 def VLD1LNd32 : VLD1LN32<0b1000, {?,0,?,?}, "32", v2i32, load> {
775 let Inst{7} = lane{0};
780 def VLD1LNq8Pseudo : VLD1QLNPseudo<v16i8, extloadi8>;
781 def VLD1LNq16Pseudo : VLD1QLNPseudo<v8i16, extloadi16>;
782 def VLD1LNq32Pseudo : VLD1QLNPseudo<v4i32, load>;
784 def : Pat<(vector_insert (v2f32 DPR:$src),
785 (f32 (load addrmode6:$addr)), imm:$lane),
786 (VLD1LNd32 addrmode6:$addr, DPR:$src, imm:$lane)>;
787 def : Pat<(vector_insert (v4f32 QPR:$src),
788 (f32 (load addrmode6:$addr)), imm:$lane),
789 (VLD1LNq32Pseudo addrmode6:$addr, QPR:$src, imm:$lane)>;
791 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
793 // ...with address register writeback:
794 class VLD1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
795 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, GPR:$wb),
796 (ins addrmode6:$Rn, am6offset:$Rm,
797 DPR:$src, nohash_imm:$lane), IIC_VLD1lnu, "vld1", Dt,
798 "\\{$Vd[$lane]\\}, $Rn$Rm",
799 "$src = $Vd, $Rn.addr = $wb", []> {
800 let DecoderMethod = "DecodeVLD1LN";
803 def VLD1LNd8_UPD : VLD1LNWB<0b0000, {?,?,?,0}, "8"> {
804 let Inst{7-5} = lane{2-0};
806 def VLD1LNd16_UPD : VLD1LNWB<0b0100, {?,?,0,?}, "16"> {
807 let Inst{7-6} = lane{1-0};
810 def VLD1LNd32_UPD : VLD1LNWB<0b1000, {?,0,?,?}, "32"> {
811 let Inst{7} = lane{0};
816 def VLD1LNq8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
817 def VLD1LNq16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
818 def VLD1LNq32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
820 // VLD2LN : Vector Load (single 2-element structure to one lane)
821 class VLD2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
822 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2),
823 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, nohash_imm:$lane),
824 IIC_VLD2ln, "vld2", Dt, "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn",
825 "$src1 = $Vd, $src2 = $dst2", []> {
828 let DecoderMethod = "DecodeVLD2LN";
831 def VLD2LNd8 : VLD2LN<0b0001, {?,?,?,?}, "8"> {
832 let Inst{7-5} = lane{2-0};
834 def VLD2LNd16 : VLD2LN<0b0101, {?,?,0,?}, "16"> {
835 let Inst{7-6} = lane{1-0};
837 def VLD2LNd32 : VLD2LN<0b1001, {?,0,0,?}, "32"> {
838 let Inst{7} = lane{0};
841 def VLD2LNd8Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
842 def VLD2LNd16Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
843 def VLD2LNd32Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
845 // ...with double-spaced registers:
846 def VLD2LNq16 : VLD2LN<0b0101, {?,?,1,?}, "16"> {
847 let Inst{7-6} = lane{1-0};
849 def VLD2LNq32 : VLD2LN<0b1001, {?,1,0,?}, "32"> {
850 let Inst{7} = lane{0};
853 def VLD2LNq16Pseudo : VLDQQLNPseudo<IIC_VLD2ln>;
854 def VLD2LNq32Pseudo : VLDQQLNPseudo<IIC_VLD2ln>;
856 // ...with address register writeback:
857 class VLD2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
858 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
859 (ins addrmode6:$Rn, am6offset:$Rm,
860 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VLD2lnu, "vld2", Dt,
861 "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn$Rm",
862 "$src1 = $Vd, $src2 = $dst2, $Rn.addr = $wb", []> {
864 let DecoderMethod = "DecodeVLD2LN";
867 def VLD2LNd8_UPD : VLD2LNWB<0b0001, {?,?,?,?}, "8"> {
868 let Inst{7-5} = lane{2-0};
870 def VLD2LNd16_UPD : VLD2LNWB<0b0101, {?,?,0,?}, "16"> {
871 let Inst{7-6} = lane{1-0};
873 def VLD2LNd32_UPD : VLD2LNWB<0b1001, {?,0,0,?}, "32"> {
874 let Inst{7} = lane{0};
877 def VLD2LNd8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
878 def VLD2LNd16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
879 def VLD2LNd32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
881 def VLD2LNq16_UPD : VLD2LNWB<0b0101, {?,?,1,?}, "16"> {
882 let Inst{7-6} = lane{1-0};
884 def VLD2LNq32_UPD : VLD2LNWB<0b1001, {?,1,0,?}, "32"> {
885 let Inst{7} = lane{0};
888 def VLD2LNq16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
889 def VLD2LNq32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
891 // VLD3LN : Vector Load (single 3-element structure to one lane)
892 class VLD3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
893 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
894 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3,
895 nohash_imm:$lane), IIC_VLD3ln, "vld3", Dt,
896 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn",
897 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3", []> {
899 let DecoderMethod = "DecodeVLD3LN";
902 def VLD3LNd8 : VLD3LN<0b0010, {?,?,?,0}, "8"> {
903 let Inst{7-5} = lane{2-0};
905 def VLD3LNd16 : VLD3LN<0b0110, {?,?,0,0}, "16"> {
906 let Inst{7-6} = lane{1-0};
908 def VLD3LNd32 : VLD3LN<0b1010, {?,0,0,0}, "32"> {
909 let Inst{7} = lane{0};
912 def VLD3LNd8Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
913 def VLD3LNd16Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
914 def VLD3LNd32Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
916 // ...with double-spaced registers:
917 def VLD3LNq16 : VLD3LN<0b0110, {?,?,1,0}, "16"> {
918 let Inst{7-6} = lane{1-0};
920 def VLD3LNq32 : VLD3LN<0b1010, {?,1,0,0}, "32"> {
921 let Inst{7} = lane{0};
924 def VLD3LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
925 def VLD3LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
927 // ...with address register writeback:
928 class VLD3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
929 : NLdStLn<1, 0b10, op11_8, op7_4,
930 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
931 (ins addrmode6:$Rn, am6offset:$Rm,
932 DPR:$src1, DPR:$src2, DPR:$src3, nohash_imm:$lane),
933 IIC_VLD3lnu, "vld3", Dt,
934 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn$Rm",
935 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $Rn.addr = $wb",
937 let DecoderMethod = "DecodeVLD3LN";
940 def VLD3LNd8_UPD : VLD3LNWB<0b0010, {?,?,?,0}, "8"> {
941 let Inst{7-5} = lane{2-0};
943 def VLD3LNd16_UPD : VLD3LNWB<0b0110, {?,?,0,0}, "16"> {
944 let Inst{7-6} = lane{1-0};
946 def VLD3LNd32_UPD : VLD3LNWB<0b1010, {?,0,0,0}, "32"> {
947 let Inst{7} = lane{0};
950 def VLD3LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
951 def VLD3LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
952 def VLD3LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
954 def VLD3LNq16_UPD : VLD3LNWB<0b0110, {?,?,1,0}, "16"> {
955 let Inst{7-6} = lane{1-0};
957 def VLD3LNq32_UPD : VLD3LNWB<0b1010, {?,1,0,0}, "32"> {
958 let Inst{7} = lane{0};
961 def VLD3LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
962 def VLD3LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
964 // VLD4LN : Vector Load (single 4-element structure to one lane)
965 class VLD4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
966 : NLdStLn<1, 0b10, op11_8, op7_4,
967 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
968 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
969 nohash_imm:$lane), IIC_VLD4ln, "vld4", Dt,
970 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn",
971 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4", []> {
974 let DecoderMethod = "DecodeVLD4LN";
977 def VLD4LNd8 : VLD4LN<0b0011, {?,?,?,?}, "8"> {
978 let Inst{7-5} = lane{2-0};
980 def VLD4LNd16 : VLD4LN<0b0111, {?,?,0,?}, "16"> {
981 let Inst{7-6} = lane{1-0};
983 def VLD4LNd32 : VLD4LN<0b1011, {?,0,?,?}, "32"> {
984 let Inst{7} = lane{0};
988 def VLD4LNd8Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
989 def VLD4LNd16Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
990 def VLD4LNd32Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
992 // ...with double-spaced registers:
993 def VLD4LNq16 : VLD4LN<0b0111, {?,?,1,?}, "16"> {
994 let Inst{7-6} = lane{1-0};
996 def VLD4LNq32 : VLD4LN<0b1011, {?,1,?,?}, "32"> {
997 let Inst{7} = lane{0};
1001 def VLD4LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>;
1002 def VLD4LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>;
1004 // ...with address register writeback:
1005 class VLD4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1006 : NLdStLn<1, 0b10, op11_8, op7_4,
1007 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
1008 (ins addrmode6:$Rn, am6offset:$Rm,
1009 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
1010 IIC_VLD4lnu, "vld4", Dt,
1011 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn$Rm",
1012 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4, $Rn.addr = $wb",
1014 let Inst{4} = Rn{4};
1015 let DecoderMethod = "DecodeVLD4LN" ;
1018 def VLD4LNd8_UPD : VLD4LNWB<0b0011, {?,?,?,?}, "8"> {
1019 let Inst{7-5} = lane{2-0};
1021 def VLD4LNd16_UPD : VLD4LNWB<0b0111, {?,?,0,?}, "16"> {
1022 let Inst{7-6} = lane{1-0};
1024 def VLD4LNd32_UPD : VLD4LNWB<0b1011, {?,0,?,?}, "32"> {
1025 let Inst{7} = lane{0};
1026 let Inst{5} = Rn{5};
1029 def VLD4LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
1030 def VLD4LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
1031 def VLD4LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
1033 def VLD4LNq16_UPD : VLD4LNWB<0b0111, {?,?,1,?}, "16"> {
1034 let Inst{7-6} = lane{1-0};
1036 def VLD4LNq32_UPD : VLD4LNWB<0b1011, {?,1,?,?}, "32"> {
1037 let Inst{7} = lane{0};
1038 let Inst{5} = Rn{5};
1041 def VLD4LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
1042 def VLD4LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
1044 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
1046 // VLD1DUP : Vector Load (single element to all lanes)
1047 class VLD1DUP<bits<4> op7_4, string Dt, ValueType Ty, PatFrag LoadOp>
1048 : NLdSt<1, 0b10, 0b1100, op7_4, (outs VecListOneDAllLanes:$Vd),
1049 (ins addrmode6dup:$Rn),
1050 IIC_VLD1dup, "vld1", Dt, "$Vd, $Rn", "",
1051 [(set VecListOneDAllLanes:$Vd,
1052 (Ty (NEONvdup (i32 (LoadOp addrmode6dup:$Rn)))))]> {
1054 let Inst{4} = Rn{4};
1055 let DecoderMethod = "DecodeVLD1DupInstruction";
1057 class VLD1QDUPPseudo<ValueType Ty, PatFrag LoadOp> : VLDQPseudo<IIC_VLD1dup> {
1058 let Pattern = [(set QPR:$dst,
1059 (Ty (NEONvdup (i32 (LoadOp addrmode6dup:$addr)))))];
1062 def VLD1DUPd8 : VLD1DUP<{0,0,0,?}, "8", v8i8, extloadi8>;
1063 def VLD1DUPd16 : VLD1DUP<{0,1,0,?}, "16", v4i16, extloadi16>;
1064 def VLD1DUPd32 : VLD1DUP<{1,0,0,?}, "32", v2i32, load>;
1066 def VLD1DUPq8Pseudo : VLD1QDUPPseudo<v16i8, extloadi8>;
1067 def VLD1DUPq16Pseudo : VLD1QDUPPseudo<v8i16, extloadi16>;
1068 def VLD1DUPq32Pseudo : VLD1QDUPPseudo<v4i32, load>;
1070 def : Pat<(v2f32 (NEONvdup (f32 (load addrmode6dup:$addr)))),
1071 (VLD1DUPd32 addrmode6:$addr)>;
1072 def : Pat<(v4f32 (NEONvdup (f32 (load addrmode6dup:$addr)))),
1073 (VLD1DUPq32Pseudo addrmode6:$addr)>;
1075 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
1077 class VLD1QDUP<bits<4> op7_4, string Dt>
1078 : NLdSt<1, 0b10, 0b1100, op7_4, (outs VecListTwoDAllLanes:$Vd),
1079 (ins addrmode6dup:$Rn), IIC_VLD1dup,
1080 "vld1", Dt, "$Vd, $Rn", "", []> {
1082 let Inst{4} = Rn{4};
1083 let DecoderMethod = "DecodeVLD1DupInstruction";
1086 def VLD1DUPq8 : VLD1QDUP<{0,0,1,0}, "8">;
1087 def VLD1DUPq16 : VLD1QDUP<{0,1,1,?}, "16">;
1088 def VLD1DUPq32 : VLD1QDUP<{1,0,1,?}, "32">;
1090 // ...with address register writeback:
1091 multiclass VLD1DUPWB<bits<4> op7_4, string Dt> {
1092 def _fixed : NLdSt<1, 0b10, 0b1100, op7_4,
1093 (outs VecListOneDAllLanes:$Vd, GPR:$wb),
1094 (ins addrmode6dup:$Rn), IIC_VLD1dupu,
1095 "vld1", Dt, "$Vd, $Rn!",
1096 "$Rn.addr = $wb", []> {
1097 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1098 let Inst{4} = Rn{4};
1099 let DecoderMethod = "DecodeVLD1DupInstruction";
1100 let AsmMatchConverter = "cvtVLDwbFixed";
1102 def _register : NLdSt<1, 0b10, 0b1100, op7_4,
1103 (outs VecListOneDAllLanes:$Vd, GPR:$wb),
1104 (ins addrmode6dup:$Rn, rGPR:$Rm), IIC_VLD1dupu,
1105 "vld1", Dt, "$Vd, $Rn, $Rm",
1106 "$Rn.addr = $wb", []> {
1107 let Inst{4} = Rn{4};
1108 let DecoderMethod = "DecodeVLD1DupInstruction";
1109 let AsmMatchConverter = "cvtVLDwbRegister";
1112 multiclass VLD1QDUPWB<bits<4> op7_4, string Dt> {
1113 def _fixed : NLdSt<1, 0b10, 0b1100, op7_4,
1114 (outs VecListTwoDAllLanes:$Vd, GPR:$wb),
1115 (ins addrmode6dup:$Rn), IIC_VLD1dupu,
1116 "vld1", Dt, "$Vd, $Rn!",
1117 "$Rn.addr = $wb", []> {
1118 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1119 let Inst{4} = Rn{4};
1120 let DecoderMethod = "DecodeVLD1DupInstruction";
1121 let AsmMatchConverter = "cvtVLDwbFixed";
1123 def _register : NLdSt<1, 0b10, 0b1100, op7_4,
1124 (outs VecListTwoDAllLanes:$Vd, GPR:$wb),
1125 (ins addrmode6dup:$Rn, rGPR:$Rm), IIC_VLD1dupu,
1126 "vld1", Dt, "$Vd, $Rn, $Rm",
1127 "$Rn.addr = $wb", []> {
1128 let Inst{4} = Rn{4};
1129 let DecoderMethod = "DecodeVLD1DupInstruction";
1130 let AsmMatchConverter = "cvtVLDwbRegister";
1134 defm VLD1DUPd8wb : VLD1DUPWB<{0,0,0,0}, "8">;
1135 defm VLD1DUPd16wb : VLD1DUPWB<{0,1,0,?}, "16">;
1136 defm VLD1DUPd32wb : VLD1DUPWB<{1,0,0,?}, "32">;
1138 defm VLD1DUPq8wb : VLD1QDUPWB<{0,0,1,0}, "8">;
1139 defm VLD1DUPq16wb : VLD1QDUPWB<{0,1,1,?}, "16">;
1140 defm VLD1DUPq32wb : VLD1QDUPWB<{1,0,1,?}, "32">;
1142 def VLD1DUPq8PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD1dupu>;
1143 def VLD1DUPq16PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD1dupu>;
1144 def VLD1DUPq32PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD1dupu>;
1145 def VLD1DUPq8PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD1dupu>;
1146 def VLD1DUPq16PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD1dupu>;
1147 def VLD1DUPq32PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD1dupu>;
1149 // VLD2DUP : Vector Load (single 2-element structure to all lanes)
1150 class VLD2DUP<bits<4> op7_4, string Dt>
1151 : NLdSt<1, 0b10, 0b1101, op7_4, (outs DPR:$Vd, DPR:$dst2),
1152 (ins addrmode6dup:$Rn), IIC_VLD2dup,
1153 "vld2", Dt, "\\{$Vd[], $dst2[]\\}, $Rn", "", []> {
1155 let Inst{4} = Rn{4};
1156 let DecoderMethod = "DecodeVLD2DupInstruction";
1159 def VLD2DUPd8 : VLD2DUP<{0,0,0,?}, "8">;
1160 def VLD2DUPd16 : VLD2DUP<{0,1,0,?}, "16">;
1161 def VLD2DUPd32 : VLD2DUP<{1,0,0,?}, "32">;
1163 def VLD2DUPd8Pseudo : VLDQPseudo<IIC_VLD2dup>;
1164 def VLD2DUPd16Pseudo : VLDQPseudo<IIC_VLD2dup>;
1165 def VLD2DUPd32Pseudo : VLDQPseudo<IIC_VLD2dup>;
1167 // ...with double-spaced registers (not used for codegen):
1168 def VLD2DUPd8x2 : VLD2DUP<{0,0,1,?}, "8">;
1169 def VLD2DUPd16x2 : VLD2DUP<{0,1,1,?}, "16">;
1170 def VLD2DUPd32x2 : VLD2DUP<{1,0,1,?}, "32">;
1172 // ...with address register writeback:
1173 class VLD2DUPWB<bits<4> op7_4, string Dt>
1174 : NLdSt<1, 0b10, 0b1101, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
1175 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD2dupu,
1176 "vld2", Dt, "\\{$Vd[], $dst2[]\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
1177 let Inst{4} = Rn{4};
1178 let DecoderMethod = "DecodeVLD2DupInstruction";
1181 def VLD2DUPd8_UPD : VLD2DUPWB<{0,0,0,0}, "8">;
1182 def VLD2DUPd16_UPD : VLD2DUPWB<{0,1,0,?}, "16">;
1183 def VLD2DUPd32_UPD : VLD2DUPWB<{1,0,0,?}, "32">;
1185 def VLD2DUPd8x2_UPD : VLD2DUPWB<{0,0,1,0}, "8">;
1186 def VLD2DUPd16x2_UPD : VLD2DUPWB<{0,1,1,?}, "16">;
1187 def VLD2DUPd32x2_UPD : VLD2DUPWB<{1,0,1,?}, "32">;
1189 def VLD2DUPd8Pseudo_UPD : VLDQWBPseudo<IIC_VLD2dupu>;
1190 def VLD2DUPd16Pseudo_UPD : VLDQWBPseudo<IIC_VLD2dupu>;
1191 def VLD2DUPd32Pseudo_UPD : VLDQWBPseudo<IIC_VLD2dupu>;
1193 // VLD3DUP : Vector Load (single 3-element structure to all lanes)
1194 class VLD3DUP<bits<4> op7_4, string Dt>
1195 : NLdSt<1, 0b10, 0b1110, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
1196 (ins addrmode6dup:$Rn), IIC_VLD3dup,
1197 "vld3", Dt, "\\{$Vd[], $dst2[], $dst3[]\\}, $Rn", "", []> {
1200 let DecoderMethod = "DecodeVLD3DupInstruction";
1203 def VLD3DUPd8 : VLD3DUP<{0,0,0,?}, "8">;
1204 def VLD3DUPd16 : VLD3DUP<{0,1,0,?}, "16">;
1205 def VLD3DUPd32 : VLD3DUP<{1,0,0,?}, "32">;
1207 def VLD3DUPd8Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1208 def VLD3DUPd16Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1209 def VLD3DUPd32Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1211 // ...with double-spaced registers (not used for codegen):
1212 def VLD3DUPd8x2 : VLD3DUP<{0,0,1,?}, "8">;
1213 def VLD3DUPd16x2 : VLD3DUP<{0,1,1,?}, "16">;
1214 def VLD3DUPd32x2 : VLD3DUP<{1,0,1,?}, "32">;
1216 // ...with address register writeback:
1217 class VLD3DUPWB<bits<4> op7_4, string Dt>
1218 : NLdSt<1, 0b10, 0b1110, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
1219 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD3dupu,
1220 "vld3", Dt, "\\{$Vd[], $dst2[], $dst3[]\\}, $Rn$Rm",
1221 "$Rn.addr = $wb", []> {
1223 let DecoderMethod = "DecodeVLD3DupInstruction";
1226 def VLD3DUPd8_UPD : VLD3DUPWB<{0,0,0,0}, "8">;
1227 def VLD3DUPd16_UPD : VLD3DUPWB<{0,1,0,?}, "16">;
1228 def VLD3DUPd32_UPD : VLD3DUPWB<{1,0,0,?}, "32">;
1230 def VLD3DUPd8x2_UPD : VLD3DUPWB<{0,0,1,0}, "8">;
1231 def VLD3DUPd16x2_UPD : VLD3DUPWB<{0,1,1,?}, "16">;
1232 def VLD3DUPd32x2_UPD : VLD3DUPWB<{1,0,1,?}, "32">;
1234 def VLD3DUPd8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1235 def VLD3DUPd16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1236 def VLD3DUPd32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1238 // VLD4DUP : Vector Load (single 4-element structure to all lanes)
1239 class VLD4DUP<bits<4> op7_4, string Dt>
1240 : NLdSt<1, 0b10, 0b1111, op7_4,
1241 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
1242 (ins addrmode6dup:$Rn), IIC_VLD4dup,
1243 "vld4", Dt, "\\{$Vd[], $dst2[], $dst3[], $dst4[]\\}, $Rn", "", []> {
1245 let Inst{4} = Rn{4};
1246 let DecoderMethod = "DecodeVLD4DupInstruction";
1249 def VLD4DUPd8 : VLD4DUP<{0,0,0,?}, "8">;
1250 def VLD4DUPd16 : VLD4DUP<{0,1,0,?}, "16">;
1251 def VLD4DUPd32 : VLD4DUP<{1,?,0,?}, "32"> { let Inst{6} = Rn{5}; }
1253 def VLD4DUPd8Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1254 def VLD4DUPd16Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1255 def VLD4DUPd32Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1257 // ...with double-spaced registers (not used for codegen):
1258 def VLD4DUPd8x2 : VLD4DUP<{0,0,1,?}, "8">;
1259 def VLD4DUPd16x2 : VLD4DUP<{0,1,1,?}, "16">;
1260 def VLD4DUPd32x2 : VLD4DUP<{1,?,1,?}, "32"> { let Inst{6} = Rn{5}; }
1262 // ...with address register writeback:
1263 class VLD4DUPWB<bits<4> op7_4, string Dt>
1264 : NLdSt<1, 0b10, 0b1111, op7_4,
1265 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
1266 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD4dupu,
1267 "vld4", Dt, "\\{$Vd[], $dst2[], $dst3[], $dst4[]\\}, $Rn$Rm",
1268 "$Rn.addr = $wb", []> {
1269 let Inst{4} = Rn{4};
1270 let DecoderMethod = "DecodeVLD4DupInstruction";
1273 def VLD4DUPd8_UPD : VLD4DUPWB<{0,0,0,0}, "8">;
1274 def VLD4DUPd16_UPD : VLD4DUPWB<{0,1,0,?}, "16">;
1275 def VLD4DUPd32_UPD : VLD4DUPWB<{1,?,0,?}, "32"> { let Inst{6} = Rn{5}; }
1277 def VLD4DUPd8x2_UPD : VLD4DUPWB<{0,0,1,0}, "8">;
1278 def VLD4DUPd16x2_UPD : VLD4DUPWB<{0,1,1,?}, "16">;
1279 def VLD4DUPd32x2_UPD : VLD4DUPWB<{1,?,1,?}, "32"> { let Inst{6} = Rn{5}; }
1281 def VLD4DUPd8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1282 def VLD4DUPd16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1283 def VLD4DUPd32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1285 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
1287 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
1289 // Classes for VST* pseudo-instructions with multi-register operands.
1290 // These are expanded to real instructions after register allocation.
1291 class VSTQPseudo<InstrItinClass itin>
1292 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src), itin, "">;
1293 class VSTQWBPseudo<InstrItinClass itin>
1294 : PseudoNLdSt<(outs GPR:$wb),
1295 (ins addrmode6:$addr, am6offset:$offset, QPR:$src), itin,
1296 "$addr.addr = $wb">;
1297 class VSTQWBfixedPseudo<InstrItinClass itin>
1298 : PseudoNLdSt<(outs GPR:$wb),
1299 (ins addrmode6:$addr, QPR:$src), itin,
1300 "$addr.addr = $wb">;
1301 class VSTQWBregisterPseudo<InstrItinClass itin>
1302 : PseudoNLdSt<(outs GPR:$wb),
1303 (ins addrmode6:$addr, rGPR:$offset, QPR:$src), itin,
1304 "$addr.addr = $wb">;
1305 class VSTQQPseudo<InstrItinClass itin>
1306 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src), itin, "">;
1307 class VSTQQWBPseudo<InstrItinClass itin>
1308 : PseudoNLdSt<(outs GPR:$wb),
1309 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src), itin,
1310 "$addr.addr = $wb">;
1311 class VSTQQQQPseudo<InstrItinClass itin>
1312 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src), itin, "">;
1313 class VSTQQQQWBPseudo<InstrItinClass itin>
1314 : PseudoNLdSt<(outs GPR:$wb),
1315 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
1316 "$addr.addr = $wb">;
1318 // VST1 : Vector Store (multiple single elements)
1319 class VST1D<bits<4> op7_4, string Dt>
1320 : NLdSt<0,0b00,0b0111,op7_4, (outs), (ins addrmode6:$Rn, VecListOneD:$Vd),
1321 IIC_VST1, "vst1", Dt, "$Vd, $Rn", "", []> {
1323 let Inst{4} = Rn{4};
1324 let DecoderMethod = "DecodeVSTInstruction";
1326 class VST1Q<bits<4> op7_4, string Dt>
1327 : NLdSt<0,0b00,0b1010,op7_4, (outs), (ins addrmode6:$Rn, VecListTwoD:$Vd),
1328 IIC_VST1x2, "vst1", Dt, "$Vd, $Rn", "", []> {
1330 let Inst{5-4} = Rn{5-4};
1331 let DecoderMethod = "DecodeVSTInstruction";
1334 def VST1d8 : VST1D<{0,0,0,?}, "8">;
1335 def VST1d16 : VST1D<{0,1,0,?}, "16">;
1336 def VST1d32 : VST1D<{1,0,0,?}, "32">;
1337 def VST1d64 : VST1D<{1,1,0,?}, "64">;
1339 def VST1q8 : VST1Q<{0,0,?,?}, "8">;
1340 def VST1q16 : VST1Q<{0,1,?,?}, "16">;
1341 def VST1q32 : VST1Q<{1,0,?,?}, "32">;
1342 def VST1q64 : VST1Q<{1,1,?,?}, "64">;
1344 def VST1q8Pseudo : VSTQPseudo<IIC_VST1x2>;
1345 def VST1q16Pseudo : VSTQPseudo<IIC_VST1x2>;
1346 def VST1q32Pseudo : VSTQPseudo<IIC_VST1x2>;
1347 def VST1q64Pseudo : VSTQPseudo<IIC_VST1x2>;
1349 // ...with address register writeback:
1350 multiclass VST1DWB<bits<4> op7_4, string Dt> {
1351 def _fixed : NLdSt<0,0b00, 0b0111,op7_4, (outs GPR:$wb),
1352 (ins addrmode6:$Rn, VecListOneD:$Vd), IIC_VLD1u,
1353 "vst1", Dt, "$Vd, $Rn!",
1354 "$Rn.addr = $wb", []> {
1355 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1356 let Inst{4} = Rn{4};
1357 let DecoderMethod = "DecodeVSTInstruction";
1358 let AsmMatchConverter = "cvtVSTwbFixed";
1360 def _register : NLdSt<0,0b00,0b0111,op7_4, (outs GPR:$wb),
1361 (ins addrmode6:$Rn, rGPR:$Rm, VecListOneD:$Vd),
1363 "vst1", Dt, "$Vd, $Rn, $Rm",
1364 "$Rn.addr = $wb", []> {
1365 let Inst{4} = Rn{4};
1366 let DecoderMethod = "DecodeVSTInstruction";
1367 let AsmMatchConverter = "cvtVSTwbRegister";
1370 multiclass VST1QWB<bits<4> op7_4, string Dt> {
1371 def _fixed : NLdSt<0,0b00,0b1010,op7_4, (outs GPR:$wb),
1372 (ins addrmode6:$Rn, VecListTwoD:$Vd), IIC_VLD1x2u,
1373 "vst1", Dt, "$Vd, $Rn!",
1374 "$Rn.addr = $wb", []> {
1375 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1376 let Inst{5-4} = Rn{5-4};
1377 let DecoderMethod = "DecodeVSTInstruction";
1378 let AsmMatchConverter = "cvtVSTwbFixed";
1380 def _register : NLdSt<0,0b00,0b1010,op7_4, (outs GPR:$wb),
1381 (ins addrmode6:$Rn, rGPR:$Rm, VecListTwoD:$Vd),
1383 "vst1", Dt, "$Vd, $Rn, $Rm",
1384 "$Rn.addr = $wb", []> {
1385 let Inst{5-4} = Rn{5-4};
1386 let DecoderMethod = "DecodeVSTInstruction";
1387 let AsmMatchConverter = "cvtVSTwbRegister";
1391 defm VST1d8wb : VST1DWB<{0,0,0,?}, "8">;
1392 defm VST1d16wb : VST1DWB<{0,1,0,?}, "16">;
1393 defm VST1d32wb : VST1DWB<{1,0,0,?}, "32">;
1394 defm VST1d64wb : VST1DWB<{1,1,0,?}, "64">;
1396 defm VST1q8wb : VST1QWB<{0,0,?,?}, "8">;
1397 defm VST1q16wb : VST1QWB<{0,1,?,?}, "16">;
1398 defm VST1q32wb : VST1QWB<{1,0,?,?}, "32">;
1399 defm VST1q64wb : VST1QWB<{1,1,?,?}, "64">;
1401 def VST1q8PseudoWB_fixed : VSTQWBfixedPseudo<IIC_VST1x2u>;
1402 def VST1q16PseudoWB_fixed : VSTQWBfixedPseudo<IIC_VST1x2u>;
1403 def VST1q32PseudoWB_fixed : VSTQWBfixedPseudo<IIC_VST1x2u>;
1404 def VST1q64PseudoWB_fixed : VSTQWBfixedPseudo<IIC_VST1x2u>;
1405 def VST1q8PseudoWB_register : VSTQWBregisterPseudo<IIC_VST1x2u>;
1406 def VST1q16PseudoWB_register : VSTQWBregisterPseudo<IIC_VST1x2u>;
1407 def VST1q32PseudoWB_register : VSTQWBregisterPseudo<IIC_VST1x2u>;
1408 def VST1q64PseudoWB_register : VSTQWBregisterPseudo<IIC_VST1x2u>;
1410 // ...with 3 registers
1411 class VST1D3<bits<4> op7_4, string Dt>
1412 : NLdSt<0, 0b00, 0b0110, op7_4, (outs),
1413 (ins addrmode6:$Rn, VecListThreeD:$Vd),
1414 IIC_VST1x3, "vst1", Dt, "$Vd, $Rn", "", []> {
1416 let Inst{4} = Rn{4};
1417 let DecoderMethod = "DecodeVSTInstruction";
1419 multiclass VST1D3WB<bits<4> op7_4, string Dt> {
1420 def _fixed : NLdSt<0,0b00,0b0110,op7_4, (outs GPR:$wb),
1421 (ins addrmode6:$Rn, VecListThreeD:$Vd), IIC_VLD1x3u,
1422 "vst1", Dt, "$Vd, $Rn!",
1423 "$Rn.addr = $wb", []> {
1424 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1425 let Inst{5-4} = Rn{5-4};
1426 let DecoderMethod = "DecodeVSTInstruction";
1427 let AsmMatchConverter = "cvtVSTwbFixed";
1429 def _register : NLdSt<0,0b00,0b0110,op7_4, (outs GPR:$wb),
1430 (ins addrmode6:$Rn, rGPR:$Rm, VecListThreeD:$Vd),
1432 "vst1", Dt, "$Vd, $Rn, $Rm",
1433 "$Rn.addr = $wb", []> {
1434 let Inst{5-4} = Rn{5-4};
1435 let DecoderMethod = "DecodeVSTInstruction";
1436 let AsmMatchConverter = "cvtVSTwbRegister";
1440 def VST1d8T : VST1D3<{0,0,0,?}, "8">;
1441 def VST1d16T : VST1D3<{0,1,0,?}, "16">;
1442 def VST1d32T : VST1D3<{1,0,0,?}, "32">;
1443 def VST1d64T : VST1D3<{1,1,0,?}, "64">;
1445 defm VST1d8Twb : VST1D3WB<{0,0,0,?}, "8">;
1446 defm VST1d16Twb : VST1D3WB<{0,1,0,?}, "16">;
1447 defm VST1d32Twb : VST1D3WB<{1,0,0,?}, "32">;
1448 defm VST1d64Twb : VST1D3WB<{1,1,0,?}, "64">;
1450 def VST1d64TPseudo : VSTQQPseudo<IIC_VST1x3>;
1451 def VST1d64TPseudoWB_fixed : VSTQQWBPseudo<IIC_VST1x3u>;
1452 def VST1d64TPseudoWB_register : VSTQQWBPseudo<IIC_VST1x3u>;
1454 // ...with 4 registers
1455 class VST1D4<bits<4> op7_4, string Dt>
1456 : NLdSt<0, 0b00, 0b0010, op7_4, (outs),
1457 (ins addrmode6:$Rn, VecListFourD:$Vd),
1458 IIC_VST1x4, "vst1", Dt, "$Vd, $Rn", "",
1461 let Inst{5-4} = Rn{5-4};
1462 let DecoderMethod = "DecodeVSTInstruction";
1464 multiclass VST1D4WB<bits<4> op7_4, string Dt> {
1465 def _fixed : NLdSt<0,0b00,0b0010,op7_4, (outs GPR:$wb),
1466 (ins addrmode6:$Rn, VecListFourD:$Vd), IIC_VLD1x4u,
1467 "vst1", Dt, "$Vd, $Rn!",
1468 "$Rn.addr = $wb", []> {
1469 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1470 let Inst{5-4} = Rn{5-4};
1471 let DecoderMethod = "DecodeVSTInstruction";
1472 let AsmMatchConverter = "cvtVSTwbFixed";
1474 def _register : NLdSt<0,0b00,0b0010,op7_4, (outs GPR:$wb),
1475 (ins addrmode6:$Rn, rGPR:$Rm, VecListFourD:$Vd),
1477 "vst1", Dt, "$Vd, $Rn, $Rm",
1478 "$Rn.addr = $wb", []> {
1479 let Inst{5-4} = Rn{5-4};
1480 let DecoderMethod = "DecodeVSTInstruction";
1481 let AsmMatchConverter = "cvtVSTwbRegister";
1485 def VST1d8Q : VST1D4<{0,0,?,?}, "8">;
1486 def VST1d16Q : VST1D4<{0,1,?,?}, "16">;
1487 def VST1d32Q : VST1D4<{1,0,?,?}, "32">;
1488 def VST1d64Q : VST1D4<{1,1,?,?}, "64">;
1490 defm VST1d8Qwb : VST1D4WB<{0,0,?,?}, "8">;
1491 defm VST1d16Qwb : VST1D4WB<{0,1,?,?}, "16">;
1492 defm VST1d32Qwb : VST1D4WB<{1,0,?,?}, "32">;
1493 defm VST1d64Qwb : VST1D4WB<{1,1,?,?}, "64">;
1495 def VST1d64QPseudo : VSTQQPseudo<IIC_VST1x4>;
1496 def VST1d64QPseudoWB_fixed : VSTQQWBPseudo<IIC_VST1x4u>;
1497 def VST1d64QPseudoWB_register : VSTQQWBPseudo<IIC_VST1x4u>;
1499 // VST2 : Vector Store (multiple 2-element structures)
1500 class VST2D<bits<4> op11_8, bits<4> op7_4, string Dt, RegisterOperand VdTy>
1501 : NLdSt<0, 0b00, op11_8, op7_4, (outs), (ins addrmode6:$Rn, VdTy:$Vd),
1502 IIC_VST2, "vst2", Dt, "$Vd, $Rn", "", []> {
1504 let Inst{5-4} = Rn{5-4};
1505 let DecoderMethod = "DecodeVSTInstruction";
1507 class VST2Q<bits<4> op7_4, string Dt, RegisterOperand VdTy>
1508 : NLdSt<0, 0b00, 0b0011, op7_4, (outs), (ins addrmode6:$Rn, VdTy:$Vd),
1509 IIC_VST2x2, "vst2", Dt, "$Vd, $Rn", "", []> {
1511 let Inst{5-4} = Rn{5-4};
1512 let DecoderMethod = "DecodeVSTInstruction";
1515 def VST2d8 : VST2D<0b1000, {0,0,?,?}, "8", VecListTwoD>;
1516 def VST2d16 : VST2D<0b1000, {0,1,?,?}, "16", VecListTwoD>;
1517 def VST2d32 : VST2D<0b1000, {1,0,?,?}, "32", VecListTwoD>;
1519 def VST2q8 : VST2Q<{0,0,?,?}, "8", VecListFourD>;
1520 def VST2q16 : VST2Q<{0,1,?,?}, "16", VecListFourD>;
1521 def VST2q32 : VST2Q<{1,0,?,?}, "32", VecListFourD>;
1523 def VST2d8Pseudo : VSTQPseudo<IIC_VST2>;
1524 def VST2d16Pseudo : VSTQPseudo<IIC_VST2>;
1525 def VST2d32Pseudo : VSTQPseudo<IIC_VST2>;
1527 def VST2q8Pseudo : VSTQQPseudo<IIC_VST2x2>;
1528 def VST2q16Pseudo : VSTQQPseudo<IIC_VST2x2>;
1529 def VST2q32Pseudo : VSTQQPseudo<IIC_VST2x2>;
1531 // ...with address register writeback:
1532 class VST2DWB<bits<4> op11_8, bits<4> op7_4, string Dt, RegisterOperand VdTy>
1533 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1534 (ins addrmode6:$Rn, am6offset:$Rm, VdTy:$Vd),
1535 IIC_VST2u, "vst2", Dt, "$Vd, $Rn$Rm", "$Rn.addr = $wb", []> {
1536 let Inst{5-4} = Rn{5-4};
1537 let DecoderMethod = "DecodeVSTInstruction";
1539 class VST2QWB<bits<4> op7_4, string Dt>
1540 : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
1541 (ins addrmode6:$Rn, am6offset:$Rm, VecListFourD:$Vd), IIC_VST2x2u,
1542 "vst2", Dt, "$Vd, $Rn$Rm", "$Rn.addr = $wb", []> {
1543 let Inst{5-4} = Rn{5-4};
1544 let DecoderMethod = "DecodeVSTInstruction";
1547 def VST2d8_UPD : VST2DWB<0b1000, {0,0,?,?}, "8", VecListTwoD>;
1548 def VST2d16_UPD : VST2DWB<0b1000, {0,1,?,?}, "16", VecListTwoD>;
1549 def VST2d32_UPD : VST2DWB<0b1000, {1,0,?,?}, "32", VecListTwoD>;
1551 def VST2q8_UPD : VST2QWB<{0,0,?,?}, "8">;
1552 def VST2q16_UPD : VST2QWB<{0,1,?,?}, "16">;
1553 def VST2q32_UPD : VST2QWB<{1,0,?,?}, "32">;
1555 def VST2d8Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
1556 def VST2d16Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
1557 def VST2d32Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
1559 def VST2q8Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
1560 def VST2q16Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
1561 def VST2q32Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
1563 // ...with double-spaced registers
1564 def VST2b8 : VST2D<0b1001, {0,0,?,?}, "8", VecListTwoQ>;
1565 def VST2b16 : VST2D<0b1001, {0,1,?,?}, "16", VecListTwoQ>;
1566 def VST2b32 : VST2D<0b1001, {1,0,?,?}, "32", VecListTwoQ>;
1567 def VST2b8_UPD : VST2DWB<0b1001, {0,0,?,?}, "8", VecListTwoQ>;
1568 def VST2b16_UPD : VST2DWB<0b1001, {0,1,?,?}, "16", VecListTwoQ>;
1569 def VST2b32_UPD : VST2DWB<0b1001, {1,0,?,?}, "32", VecListTwoQ>;
1571 // VST3 : Vector Store (multiple 3-element structures)
1572 class VST3D<bits<4> op11_8, bits<4> op7_4, string Dt>
1573 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1574 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3,
1575 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn", "", []> {
1577 let Inst{4} = Rn{4};
1578 let DecoderMethod = "DecodeVSTInstruction";
1581 def VST3d8 : VST3D<0b0100, {0,0,0,?}, "8">;
1582 def VST3d16 : VST3D<0b0100, {0,1,0,?}, "16">;
1583 def VST3d32 : VST3D<0b0100, {1,0,0,?}, "32">;
1585 def VST3d8Pseudo : VSTQQPseudo<IIC_VST3>;
1586 def VST3d16Pseudo : VSTQQPseudo<IIC_VST3>;
1587 def VST3d32Pseudo : VSTQQPseudo<IIC_VST3>;
1589 // ...with address register writeback:
1590 class VST3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1591 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1592 (ins addrmode6:$Rn, am6offset:$Rm,
1593 DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3u,
1594 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn$Rm",
1595 "$Rn.addr = $wb", []> {
1596 let Inst{4} = Rn{4};
1597 let DecoderMethod = "DecodeVSTInstruction";
1600 def VST3d8_UPD : VST3DWB<0b0100, {0,0,0,?}, "8">;
1601 def VST3d16_UPD : VST3DWB<0b0100, {0,1,0,?}, "16">;
1602 def VST3d32_UPD : VST3DWB<0b0100, {1,0,0,?}, "32">;
1604 def VST3d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1605 def VST3d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1606 def VST3d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1608 // ...with double-spaced registers:
1609 def VST3q8 : VST3D<0b0101, {0,0,0,?}, "8">;
1610 def VST3q16 : VST3D<0b0101, {0,1,0,?}, "16">;
1611 def VST3q32 : VST3D<0b0101, {1,0,0,?}, "32">;
1612 def VST3q8_UPD : VST3DWB<0b0101, {0,0,0,?}, "8">;
1613 def VST3q16_UPD : VST3DWB<0b0101, {0,1,0,?}, "16">;
1614 def VST3q32_UPD : VST3DWB<0b0101, {1,0,0,?}, "32">;
1616 def VST3q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1617 def VST3q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1618 def VST3q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1620 // ...alternate versions to be allocated odd register numbers:
1621 def VST3q8oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1622 def VST3q16oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1623 def VST3q32oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1625 def VST3q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1626 def VST3q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1627 def VST3q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1629 // VST4 : Vector Store (multiple 4-element structures)
1630 class VST4D<bits<4> op11_8, bits<4> op7_4, string Dt>
1631 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1632 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
1633 IIC_VST4, "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn",
1636 let Inst{5-4} = Rn{5-4};
1637 let DecoderMethod = "DecodeVSTInstruction";
1640 def VST4d8 : VST4D<0b0000, {0,0,?,?}, "8">;
1641 def VST4d16 : VST4D<0b0000, {0,1,?,?}, "16">;
1642 def VST4d32 : VST4D<0b0000, {1,0,?,?}, "32">;
1644 def VST4d8Pseudo : VSTQQPseudo<IIC_VST4>;
1645 def VST4d16Pseudo : VSTQQPseudo<IIC_VST4>;
1646 def VST4d32Pseudo : VSTQQPseudo<IIC_VST4>;
1648 // ...with address register writeback:
1649 class VST4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1650 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1651 (ins addrmode6:$Rn, am6offset:$Rm,
1652 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST4u,
1653 "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
1654 "$Rn.addr = $wb", []> {
1655 let Inst{5-4} = Rn{5-4};
1656 let DecoderMethod = "DecodeVSTInstruction";
1659 def VST4d8_UPD : VST4DWB<0b0000, {0,0,?,?}, "8">;
1660 def VST4d16_UPD : VST4DWB<0b0000, {0,1,?,?}, "16">;
1661 def VST4d32_UPD : VST4DWB<0b0000, {1,0,?,?}, "32">;
1663 def VST4d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1664 def VST4d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1665 def VST4d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1667 // ...with double-spaced registers:
1668 def VST4q8 : VST4D<0b0001, {0,0,?,?}, "8">;
1669 def VST4q16 : VST4D<0b0001, {0,1,?,?}, "16">;
1670 def VST4q32 : VST4D<0b0001, {1,0,?,?}, "32">;
1671 def VST4q8_UPD : VST4DWB<0b0001, {0,0,?,?}, "8">;
1672 def VST4q16_UPD : VST4DWB<0b0001, {0,1,?,?}, "16">;
1673 def VST4q32_UPD : VST4DWB<0b0001, {1,0,?,?}, "32">;
1675 def VST4q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1676 def VST4q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1677 def VST4q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1679 // ...alternate versions to be allocated odd register numbers:
1680 def VST4q8oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1681 def VST4q16oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1682 def VST4q32oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1684 def VST4q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1685 def VST4q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1686 def VST4q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1688 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
1690 // Classes for VST*LN pseudo-instructions with multi-register operands.
1691 // These are expanded to real instructions after register allocation.
1692 class VSTQLNPseudo<InstrItinClass itin>
1693 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
1695 class VSTQLNWBPseudo<InstrItinClass itin>
1696 : PseudoNLdSt<(outs GPR:$wb),
1697 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
1698 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1699 class VSTQQLNPseudo<InstrItinClass itin>
1700 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
1702 class VSTQQLNWBPseudo<InstrItinClass itin>
1703 : PseudoNLdSt<(outs GPR:$wb),
1704 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
1705 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1706 class VSTQQQQLNPseudo<InstrItinClass itin>
1707 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
1709 class VSTQQQQLNWBPseudo<InstrItinClass itin>
1710 : PseudoNLdSt<(outs GPR:$wb),
1711 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
1712 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1714 // VST1LN : Vector Store (single element from one lane)
1715 class VST1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1716 PatFrag StoreOp, SDNode ExtractOp>
1717 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1718 (ins addrmode6:$Rn, DPR:$Vd, nohash_imm:$lane),
1719 IIC_VST1ln, "vst1", Dt, "\\{$Vd[$lane]\\}, $Rn", "",
1720 [(StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane), addrmode6:$Rn)]> {
1722 let DecoderMethod = "DecodeVST1LN";
1724 class VST1LN32<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1725 PatFrag StoreOp, SDNode ExtractOp>
1726 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1727 (ins addrmode6oneL32:$Rn, DPR:$Vd, nohash_imm:$lane),
1728 IIC_VST1ln, "vst1", Dt, "\\{$Vd[$lane]\\}, $Rn", "",
1729 [(StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane), addrmode6oneL32:$Rn)]>{
1731 let DecoderMethod = "DecodeVST1LN";
1733 class VST1QLNPseudo<ValueType Ty, PatFrag StoreOp, SDNode ExtractOp>
1734 : VSTQLNPseudo<IIC_VST1ln> {
1735 let Pattern = [(StoreOp (ExtractOp (Ty QPR:$src), imm:$lane),
1739 def VST1LNd8 : VST1LN<0b0000, {?,?,?,0}, "8", v8i8, truncstorei8,
1741 let Inst{7-5} = lane{2-0};
1743 def VST1LNd16 : VST1LN<0b0100, {?,?,0,?}, "16", v4i16, truncstorei16,
1745 let Inst{7-6} = lane{1-0};
1746 let Inst{4} = Rn{5};
1749 def VST1LNd32 : VST1LN32<0b1000, {?,0,?,?}, "32", v2i32, store, extractelt> {
1750 let Inst{7} = lane{0};
1751 let Inst{5-4} = Rn{5-4};
1754 def VST1LNq8Pseudo : VST1QLNPseudo<v16i8, truncstorei8, NEONvgetlaneu>;
1755 def VST1LNq16Pseudo : VST1QLNPseudo<v8i16, truncstorei16, NEONvgetlaneu>;
1756 def VST1LNq32Pseudo : VST1QLNPseudo<v4i32, store, extractelt>;
1758 def : Pat<(store (extractelt (v2f32 DPR:$src), imm:$lane), addrmode6:$addr),
1759 (VST1LNd32 addrmode6:$addr, DPR:$src, imm:$lane)>;
1760 def : Pat<(store (extractelt (v4f32 QPR:$src), imm:$lane), addrmode6:$addr),
1761 (VST1LNq32Pseudo addrmode6:$addr, QPR:$src, imm:$lane)>;
1763 // ...with address register writeback:
1764 class VST1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1765 PatFrag StoreOp, SDNode ExtractOp>
1766 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1767 (ins addrmode6:$Rn, am6offset:$Rm,
1768 DPR:$Vd, nohash_imm:$lane), IIC_VST1lnu, "vst1", Dt,
1769 "\\{$Vd[$lane]\\}, $Rn$Rm",
1771 [(set GPR:$wb, (StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane),
1772 addrmode6:$Rn, am6offset:$Rm))]> {
1773 let DecoderMethod = "DecodeVST1LN";
1775 class VST1QLNWBPseudo<ValueType Ty, PatFrag StoreOp, SDNode ExtractOp>
1776 : VSTQLNWBPseudo<IIC_VST1lnu> {
1777 let Pattern = [(set GPR:$wb, (StoreOp (ExtractOp (Ty QPR:$src), imm:$lane),
1778 addrmode6:$addr, am6offset:$offset))];
1781 def VST1LNd8_UPD : VST1LNWB<0b0000, {?,?,?,0}, "8", v8i8, post_truncsti8,
1783 let Inst{7-5} = lane{2-0};
1785 def VST1LNd16_UPD : VST1LNWB<0b0100, {?,?,0,?}, "16", v4i16, post_truncsti16,
1787 let Inst{7-6} = lane{1-0};
1788 let Inst{4} = Rn{5};
1790 def VST1LNd32_UPD : VST1LNWB<0b1000, {?,0,?,?}, "32", v2i32, post_store,
1792 let Inst{7} = lane{0};
1793 let Inst{5-4} = Rn{5-4};
1796 def VST1LNq8Pseudo_UPD : VST1QLNWBPseudo<v16i8, post_truncsti8, NEONvgetlaneu>;
1797 def VST1LNq16Pseudo_UPD : VST1QLNWBPseudo<v8i16, post_truncsti16,NEONvgetlaneu>;
1798 def VST1LNq32Pseudo_UPD : VST1QLNWBPseudo<v4i32, post_store, extractelt>;
1800 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
1802 // VST2LN : Vector Store (single 2-element structure from one lane)
1803 class VST2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1804 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1805 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, nohash_imm:$lane),
1806 IIC_VST2ln, "vst2", Dt, "\\{$Vd[$lane], $src2[$lane]\\}, $Rn",
1809 let Inst{4} = Rn{4};
1810 let DecoderMethod = "DecodeVST2LN";
1813 def VST2LNd8 : VST2LN<0b0001, {?,?,?,?}, "8"> {
1814 let Inst{7-5} = lane{2-0};
1816 def VST2LNd16 : VST2LN<0b0101, {?,?,0,?}, "16"> {
1817 let Inst{7-6} = lane{1-0};
1819 def VST2LNd32 : VST2LN<0b1001, {?,0,0,?}, "32"> {
1820 let Inst{7} = lane{0};
1823 def VST2LNd8Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1824 def VST2LNd16Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1825 def VST2LNd32Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1827 // ...with double-spaced registers:
1828 def VST2LNq16 : VST2LN<0b0101, {?,?,1,?}, "16"> {
1829 let Inst{7-6} = lane{1-0};
1830 let Inst{4} = Rn{4};
1832 def VST2LNq32 : VST2LN<0b1001, {?,1,0,?}, "32"> {
1833 let Inst{7} = lane{0};
1834 let Inst{4} = Rn{4};
1837 def VST2LNq16Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
1838 def VST2LNq32Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
1840 // ...with address register writeback:
1841 class VST2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1842 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1843 (ins addrmode6:$addr, am6offset:$offset,
1844 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VST2lnu, "vst2", Dt,
1845 "\\{$src1[$lane], $src2[$lane]\\}, $addr$offset",
1846 "$addr.addr = $wb", []> {
1847 let Inst{4} = Rn{4};
1848 let DecoderMethod = "DecodeVST2LN";
1851 def VST2LNd8_UPD : VST2LNWB<0b0001, {?,?,?,?}, "8"> {
1852 let Inst{7-5} = lane{2-0};
1854 def VST2LNd16_UPD : VST2LNWB<0b0101, {?,?,0,?}, "16"> {
1855 let Inst{7-6} = lane{1-0};
1857 def VST2LNd32_UPD : VST2LNWB<0b1001, {?,0,0,?}, "32"> {
1858 let Inst{7} = lane{0};
1861 def VST2LNd8Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1862 def VST2LNd16Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1863 def VST2LNd32Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1865 def VST2LNq16_UPD : VST2LNWB<0b0101, {?,?,1,?}, "16"> {
1866 let Inst{7-6} = lane{1-0};
1868 def VST2LNq32_UPD : VST2LNWB<0b1001, {?,1,0,?}, "32"> {
1869 let Inst{7} = lane{0};
1872 def VST2LNq16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
1873 def VST2LNq32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
1875 // VST3LN : Vector Store (single 3-element structure from one lane)
1876 class VST3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1877 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1878 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3,
1879 nohash_imm:$lane), IIC_VST3ln, "vst3", Dt,
1880 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn", "", []> {
1882 let DecoderMethod = "DecodeVST3LN";
1885 def VST3LNd8 : VST3LN<0b0010, {?,?,?,0}, "8"> {
1886 let Inst{7-5} = lane{2-0};
1888 def VST3LNd16 : VST3LN<0b0110, {?,?,0,0}, "16"> {
1889 let Inst{7-6} = lane{1-0};
1891 def VST3LNd32 : VST3LN<0b1010, {?,0,0,0}, "32"> {
1892 let Inst{7} = lane{0};
1895 def VST3LNd8Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1896 def VST3LNd16Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1897 def VST3LNd32Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1899 // ...with double-spaced registers:
1900 def VST3LNq16 : VST3LN<0b0110, {?,?,1,0}, "16"> {
1901 let Inst{7-6} = lane{1-0};
1903 def VST3LNq32 : VST3LN<0b1010, {?,1,0,0}, "32"> {
1904 let Inst{7} = lane{0};
1907 def VST3LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
1908 def VST3LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
1910 // ...with address register writeback:
1911 class VST3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1912 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1913 (ins addrmode6:$Rn, am6offset:$Rm,
1914 DPR:$Vd, DPR:$src2, DPR:$src3, nohash_imm:$lane),
1915 IIC_VST3lnu, "vst3", Dt,
1916 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn$Rm",
1917 "$Rn.addr = $wb", []> {
1918 let DecoderMethod = "DecodeVST3LN";
1921 def VST3LNd8_UPD : VST3LNWB<0b0010, {?,?,?,0}, "8"> {
1922 let Inst{7-5} = lane{2-0};
1924 def VST3LNd16_UPD : VST3LNWB<0b0110, {?,?,0,0}, "16"> {
1925 let Inst{7-6} = lane{1-0};
1927 def VST3LNd32_UPD : VST3LNWB<0b1010, {?,0,0,0}, "32"> {
1928 let Inst{7} = lane{0};
1931 def VST3LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1932 def VST3LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1933 def VST3LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1935 def VST3LNq16_UPD : VST3LNWB<0b0110, {?,?,1,0}, "16"> {
1936 let Inst{7-6} = lane{1-0};
1938 def VST3LNq32_UPD : VST3LNWB<0b1010, {?,1,0,0}, "32"> {
1939 let Inst{7} = lane{0};
1942 def VST3LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
1943 def VST3LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
1945 // VST4LN : Vector Store (single 4-element structure from one lane)
1946 class VST4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1947 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1948 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4,
1949 nohash_imm:$lane), IIC_VST4ln, "vst4", Dt,
1950 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn",
1953 let Inst{4} = Rn{4};
1954 let DecoderMethod = "DecodeVST4LN";
1957 def VST4LNd8 : VST4LN<0b0011, {?,?,?,?}, "8"> {
1958 let Inst{7-5} = lane{2-0};
1960 def VST4LNd16 : VST4LN<0b0111, {?,?,0,?}, "16"> {
1961 let Inst{7-6} = lane{1-0};
1963 def VST4LNd32 : VST4LN<0b1011, {?,0,?,?}, "32"> {
1964 let Inst{7} = lane{0};
1965 let Inst{5} = Rn{5};
1968 def VST4LNd8Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1969 def VST4LNd16Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1970 def VST4LNd32Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1972 // ...with double-spaced registers:
1973 def VST4LNq16 : VST4LN<0b0111, {?,?,1,?}, "16"> {
1974 let Inst{7-6} = lane{1-0};
1976 def VST4LNq32 : VST4LN<0b1011, {?,1,?,?}, "32"> {
1977 let Inst{7} = lane{0};
1978 let Inst{5} = Rn{5};
1981 def VST4LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
1982 def VST4LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
1984 // ...with address register writeback:
1985 class VST4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1986 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1987 (ins addrmode6:$Rn, am6offset:$Rm,
1988 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
1989 IIC_VST4lnu, "vst4", Dt,
1990 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn$Rm",
1991 "$Rn.addr = $wb", []> {
1992 let Inst{4} = Rn{4};
1993 let DecoderMethod = "DecodeVST4LN";
1996 def VST4LNd8_UPD : VST4LNWB<0b0011, {?,?,?,?}, "8"> {
1997 let Inst{7-5} = lane{2-0};
1999 def VST4LNd16_UPD : VST4LNWB<0b0111, {?,?,0,?}, "16"> {
2000 let Inst{7-6} = lane{1-0};
2002 def VST4LNd32_UPD : VST4LNWB<0b1011, {?,0,?,?}, "32"> {
2003 let Inst{7} = lane{0};
2004 let Inst{5} = Rn{5};
2007 def VST4LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
2008 def VST4LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
2009 def VST4LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
2011 def VST4LNq16_UPD : VST4LNWB<0b0111, {?,?,1,?}, "16"> {
2012 let Inst{7-6} = lane{1-0};
2014 def VST4LNq32_UPD : VST4LNWB<0b1011, {?,1,?,?}, "32"> {
2015 let Inst{7} = lane{0};
2016 let Inst{5} = Rn{5};
2019 def VST4LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
2020 def VST4LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
2022 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
2025 //===----------------------------------------------------------------------===//
2026 // NEON pattern fragments
2027 //===----------------------------------------------------------------------===//
2029 // Extract D sub-registers of Q registers.
2030 def DSubReg_i8_reg : SDNodeXForm<imm, [{
2031 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
2032 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/8, MVT::i32);
2034 def DSubReg_i16_reg : SDNodeXForm<imm, [{
2035 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
2036 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/4, MVT::i32);
2038 def DSubReg_i32_reg : SDNodeXForm<imm, [{
2039 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
2040 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/2, MVT::i32);
2042 def DSubReg_f64_reg : SDNodeXForm<imm, [{
2043 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
2044 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue(), MVT::i32);
2047 // Extract S sub-registers of Q/D registers.
2048 def SSubReg_f32_reg : SDNodeXForm<imm, [{
2049 assert(ARM::ssub_3 == ARM::ssub_0+3 && "Unexpected subreg numbering");
2050 return CurDAG->getTargetConstant(ARM::ssub_0 + N->getZExtValue(), MVT::i32);
2053 // Translate lane numbers from Q registers to D subregs.
2054 def SubReg_i8_lane : SDNodeXForm<imm, [{
2055 return CurDAG->getTargetConstant(N->getZExtValue() & 7, MVT::i32);
2057 def SubReg_i16_lane : SDNodeXForm<imm, [{
2058 return CurDAG->getTargetConstant(N->getZExtValue() & 3, MVT::i32);
2060 def SubReg_i32_lane : SDNodeXForm<imm, [{
2061 return CurDAG->getTargetConstant(N->getZExtValue() & 1, MVT::i32);
2064 //===----------------------------------------------------------------------===//
2065 // Instruction Classes
2066 //===----------------------------------------------------------------------===//
2068 // Basic 2-register operations: double- and quad-register.
2069 class N2VD<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2070 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
2071 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
2072 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
2073 (ins DPR:$Vm), IIC_VUNAD, OpcodeStr, Dt,"$Vd, $Vm", "",
2074 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vm))))]>;
2075 class N2VQ<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2076 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
2077 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
2078 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
2079 (ins QPR:$Vm), IIC_VUNAQ, OpcodeStr, Dt,"$Vd, $Vm", "",
2080 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vm))))]>;
2082 // Basic 2-register intrinsics, both double- and quad-register.
2083 class N2VDInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2084 bits<2> op17_16, bits<5> op11_7, bit op4,
2085 InstrItinClass itin, string OpcodeStr, string Dt,
2086 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2087 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
2088 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2089 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm))))]>;
2090 class N2VQInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2091 bits<2> op17_16, bits<5> op11_7, bit op4,
2092 InstrItinClass itin, string OpcodeStr, string Dt,
2093 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2094 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
2095 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2096 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm))))]>;
2098 // Narrow 2-register operations.
2099 class N2VN<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2100 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2101 InstrItinClass itin, string OpcodeStr, string Dt,
2102 ValueType TyD, ValueType TyQ, SDNode OpNode>
2103 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$Vd),
2104 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2105 [(set DPR:$Vd, (TyD (OpNode (TyQ QPR:$Vm))))]>;
2107 // Narrow 2-register intrinsics.
2108 class N2VNInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2109 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2110 InstrItinClass itin, string OpcodeStr, string Dt,
2111 ValueType TyD, ValueType TyQ, Intrinsic IntOp>
2112 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$Vd),
2113 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2114 [(set DPR:$Vd, (TyD (IntOp (TyQ QPR:$Vm))))]>;
2116 // Long 2-register operations (currently only used for VMOVL).
2117 class N2VL<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2118 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2119 InstrItinClass itin, string OpcodeStr, string Dt,
2120 ValueType TyQ, ValueType TyD, SDNode OpNode>
2121 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$Vd),
2122 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2123 [(set QPR:$Vd, (TyQ (OpNode (TyD DPR:$Vm))))]>;
2125 // Long 2-register intrinsics.
2126 class N2VLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2127 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2128 InstrItinClass itin, string OpcodeStr, string Dt,
2129 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
2130 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$Vd),
2131 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2132 [(set QPR:$Vd, (TyQ (IntOp (TyD DPR:$Vm))))]>;
2134 // 2-register shuffles (VTRN/VZIP/VUZP), both double- and quad-register.
2135 class N2VDShuffle<bits<2> op19_18, bits<5> op11_7, string OpcodeStr, string Dt>
2136 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 0, 0, (outs DPR:$Vd, DPR:$Vm),
2137 (ins DPR:$src1, DPR:$src2), IIC_VPERMD,
2138 OpcodeStr, Dt, "$Vd, $Vm",
2139 "$src1 = $Vd, $src2 = $Vm", []>;
2140 class N2VQShuffle<bits<2> op19_18, bits<5> op11_7,
2141 InstrItinClass itin, string OpcodeStr, string Dt>
2142 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 1, 0, (outs QPR:$Vd, QPR:$Vm),
2143 (ins QPR:$src1, QPR:$src2), itin, OpcodeStr, Dt, "$Vd, $Vm",
2144 "$src1 = $Vd, $src2 = $Vm", []>;
2146 // Basic 3-register operations: double- and quad-register.
2147 class N3VD<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2148 InstrItinClass itin, string OpcodeStr, string Dt,
2149 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
2150 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2151 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2152 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2153 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
2154 let isCommutable = Commutable;
2156 // Same as N3VD but no data type.
2157 class N3VDX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2158 InstrItinClass itin, string OpcodeStr,
2159 ValueType ResTy, ValueType OpTy,
2160 SDNode OpNode, bit Commutable>
2161 : N3VX<op24, op23, op21_20, op11_8, 0, op4,
2162 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2163 OpcodeStr, "$Vd, $Vn, $Vm", "",
2164 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>{
2165 let isCommutable = Commutable;
2168 class N3VDSL<bits<2> op21_20, bits<4> op11_8,
2169 InstrItinClass itin, string OpcodeStr, string Dt,
2170 ValueType Ty, SDNode ShOp>
2171 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2172 (outs DPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2173 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2175 (Ty (ShOp (Ty DPR:$Vn),
2176 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),imm:$lane)))))]> {
2177 let isCommutable = 0;
2179 class N3VDSL16<bits<2> op21_20, bits<4> op11_8,
2180 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
2181 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2182 (outs DPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2183 NVMulSLFrm, IIC_VMULi16D, OpcodeStr, Dt,"$Vd, $Vn, $Vm$lane","",
2185 (Ty (ShOp (Ty DPR:$Vn),
2186 (Ty (NEONvduplane (Ty DPR_8:$Vm), imm:$lane)))))]> {
2187 let isCommutable = 0;
2190 class N3VQ<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2191 InstrItinClass itin, string OpcodeStr, string Dt,
2192 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
2193 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2194 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2195 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2196 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
2197 let isCommutable = Commutable;
2199 class N3VQX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2200 InstrItinClass itin, string OpcodeStr,
2201 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
2202 : N3VX<op24, op23, op21_20, op11_8, 1, op4,
2203 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2204 OpcodeStr, "$Vd, $Vn, $Vm", "",
2205 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]>{
2206 let isCommutable = Commutable;
2208 class N3VQSL<bits<2> op21_20, bits<4> op11_8,
2209 InstrItinClass itin, string OpcodeStr, string Dt,
2210 ValueType ResTy, ValueType OpTy, SDNode ShOp>
2211 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2212 (outs QPR:$Vd), (ins QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2213 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2214 [(set (ResTy QPR:$Vd),
2215 (ResTy (ShOp (ResTy QPR:$Vn),
2216 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2218 let isCommutable = 0;
2220 class N3VQSL16<bits<2> op21_20, bits<4> op11_8, string OpcodeStr, string Dt,
2221 ValueType ResTy, ValueType OpTy, SDNode ShOp>
2222 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2223 (outs QPR:$Vd), (ins QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2224 NVMulSLFrm, IIC_VMULi16Q, OpcodeStr, Dt,"$Vd, $Vn, $Vm$lane", "",
2225 [(set (ResTy QPR:$Vd),
2226 (ResTy (ShOp (ResTy QPR:$Vn),
2227 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2229 let isCommutable = 0;
2232 // Basic 3-register intrinsics, both double- and quad-register.
2233 class N3VDInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2234 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2235 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
2236 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2237 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), f, itin,
2238 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2239 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
2240 let isCommutable = Commutable;
2242 class N3VDIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2243 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
2244 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2245 (outs DPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2246 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2248 (Ty (IntOp (Ty DPR:$Vn),
2249 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),
2251 let isCommutable = 0;
2253 class N3VDIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2254 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
2255 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2256 (outs DPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2257 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2259 (Ty (IntOp (Ty DPR:$Vn),
2260 (Ty (NEONvduplane (Ty DPR_8:$Vm), imm:$lane)))))]> {
2261 let isCommutable = 0;
2263 class N3VDIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2264 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2265 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2266 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2267 (outs DPR:$Vd), (ins DPR:$Vm, DPR:$Vn), f, itin,
2268 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
2269 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (OpTy DPR:$Vn))))]> {
2270 let isCommutable = 0;
2273 class N3VQInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2274 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2275 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
2276 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2277 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), f, itin,
2278 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2279 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
2280 let isCommutable = Commutable;
2282 class N3VQIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2283 string OpcodeStr, string Dt,
2284 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2285 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2286 (outs QPR:$Vd), (ins QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2287 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2288 [(set (ResTy QPR:$Vd),
2289 (ResTy (IntOp (ResTy QPR:$Vn),
2290 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2292 let isCommutable = 0;
2294 class N3VQIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2295 string OpcodeStr, string Dt,
2296 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2297 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2298 (outs QPR:$Vd), (ins QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2299 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2300 [(set (ResTy QPR:$Vd),
2301 (ResTy (IntOp (ResTy QPR:$Vn),
2302 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2304 let isCommutable = 0;
2306 class N3VQIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2307 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2308 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2309 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2310 (outs QPR:$Vd), (ins QPR:$Vm, QPR:$Vn), f, itin,
2311 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
2312 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (OpTy QPR:$Vn))))]> {
2313 let isCommutable = 0;
2316 // Multiply-Add/Sub operations: double- and quad-register.
2317 class N3VDMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2318 InstrItinClass itin, string OpcodeStr, string Dt,
2319 ValueType Ty, SDPatternOperator MulOp, SDPatternOperator OpNode>
2320 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2321 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2322 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2323 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
2324 (Ty (MulOp DPR:$Vn, DPR:$Vm)))))]>;
2326 class N3VDMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2327 string OpcodeStr, string Dt,
2328 ValueType Ty, SDPatternOperator MulOp, SDPatternOperator ShOp>
2329 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2331 (ins DPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2333 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2335 (Ty (ShOp (Ty DPR:$src1),
2337 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),
2339 class N3VDMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2340 string OpcodeStr, string Dt,
2341 ValueType Ty, SDNode MulOp, SDNode ShOp>
2342 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2344 (ins DPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2346 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2348 (Ty (ShOp (Ty DPR:$src1),
2350 (Ty (NEONvduplane (Ty DPR_8:$Vm),
2353 class N3VQMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2354 InstrItinClass itin, string OpcodeStr, string Dt, ValueType Ty,
2355 SDPatternOperator MulOp, SDPatternOperator OpNode>
2356 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2357 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2358 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2359 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
2360 (Ty (MulOp QPR:$Vn, QPR:$Vm)))))]>;
2361 class N3VQMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2362 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2363 SDPatternOperator MulOp, SDPatternOperator ShOp>
2364 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2366 (ins QPR:$src1, QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2368 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2369 [(set (ResTy QPR:$Vd),
2370 (ResTy (ShOp (ResTy QPR:$src1),
2371 (ResTy (MulOp QPR:$Vn,
2372 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2374 class N3VQMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2375 string OpcodeStr, string Dt,
2376 ValueType ResTy, ValueType OpTy,
2377 SDNode MulOp, SDNode ShOp>
2378 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2380 (ins QPR:$src1, QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2382 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2383 [(set (ResTy QPR:$Vd),
2384 (ResTy (ShOp (ResTy QPR:$src1),
2385 (ResTy (MulOp QPR:$Vn,
2386 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2389 // Neon Intrinsic-Op instructions (VABA): double- and quad-register.
2390 class N3VDIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2391 InstrItinClass itin, string OpcodeStr, string Dt,
2392 ValueType Ty, Intrinsic IntOp, SDNode OpNode>
2393 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2394 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2395 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2396 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
2397 (Ty (IntOp (Ty DPR:$Vn), (Ty DPR:$Vm))))))]>;
2398 class N3VQIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2399 InstrItinClass itin, string OpcodeStr, string Dt,
2400 ValueType Ty, Intrinsic IntOp, SDNode OpNode>
2401 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2402 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2403 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2404 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
2405 (Ty (IntOp (Ty QPR:$Vn), (Ty QPR:$Vm))))))]>;
2407 // Neon 3-argument intrinsics, both double- and quad-register.
2408 // The destination register is also used as the first source operand register.
2409 class N3VDInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2410 InstrItinClass itin, string OpcodeStr, string Dt,
2411 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2412 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2413 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2414 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2415 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$src1),
2416 (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>;
2417 class N3VQInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2418 InstrItinClass itin, string OpcodeStr, string Dt,
2419 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2420 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2421 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2422 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2423 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$src1),
2424 (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]>;
2426 // Long Multiply-Add/Sub operations.
2427 class N3VLMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2428 InstrItinClass itin, string OpcodeStr, string Dt,
2429 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2430 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2431 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2432 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2433 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
2434 (TyQ (MulOp (TyD DPR:$Vn),
2435 (TyD DPR:$Vm)))))]>;
2436 class N3VLMulOpSL<bit op24, bits<2> op21_20, bits<4> op11_8,
2437 InstrItinClass itin, string OpcodeStr, string Dt,
2438 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2439 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$Vd),
2440 (ins QPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2442 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2444 (OpNode (TyQ QPR:$src1),
2445 (TyQ (MulOp (TyD DPR:$Vn),
2446 (TyD (NEONvduplane (TyD DPR_VFP2:$Vm),
2448 class N3VLMulOpSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2449 InstrItinClass itin, string OpcodeStr, string Dt,
2450 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2451 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$Vd),
2452 (ins QPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2454 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2456 (OpNode (TyQ QPR:$src1),
2457 (TyQ (MulOp (TyD DPR:$Vn),
2458 (TyD (NEONvduplane (TyD DPR_8:$Vm),
2461 // Long Intrinsic-Op vector operations with explicit extend (VABAL).
2462 class N3VLIntExtOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2463 InstrItinClass itin, string OpcodeStr, string Dt,
2464 ValueType TyQ, ValueType TyD, Intrinsic IntOp, SDNode ExtOp,
2466 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2467 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2468 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2469 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
2470 (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
2471 (TyD DPR:$Vm)))))))]>;
2473 // Neon Long 3-argument intrinsic. The destination register is
2474 // a quad-register and is also used as the first source operand register.
2475 class N3VLInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2476 InstrItinClass itin, string OpcodeStr, string Dt,
2477 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
2478 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2479 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2480 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2482 (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$Vn), (TyD DPR:$Vm))))]>;
2483 class N3VLInt3SL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2484 string OpcodeStr, string Dt,
2485 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2486 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2488 (ins QPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2490 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2491 [(set (ResTy QPR:$Vd),
2492 (ResTy (IntOp (ResTy QPR:$src1),
2494 (OpTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2496 class N3VLInt3SL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2497 InstrItinClass itin, string OpcodeStr, string Dt,
2498 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2499 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2501 (ins QPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2503 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2504 [(set (ResTy QPR:$Vd),
2505 (ResTy (IntOp (ResTy QPR:$src1),
2507 (OpTy (NEONvduplane (OpTy DPR_8:$Vm),
2510 // Narrowing 3-register intrinsics.
2511 class N3VNInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2512 string OpcodeStr, string Dt, ValueType TyD, ValueType TyQ,
2513 Intrinsic IntOp, bit Commutable>
2514 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2515 (outs DPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINi4D,
2516 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2517 [(set DPR:$Vd, (TyD (IntOp (TyQ QPR:$Vn), (TyQ QPR:$Vm))))]> {
2518 let isCommutable = Commutable;
2521 // Long 3-register operations.
2522 class N3VL<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2523 InstrItinClass itin, string OpcodeStr, string Dt,
2524 ValueType TyQ, ValueType TyD, SDNode OpNode, bit Commutable>
2525 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2526 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2527 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2528 [(set QPR:$Vd, (TyQ (OpNode (TyD DPR:$Vn), (TyD DPR:$Vm))))]> {
2529 let isCommutable = Commutable;
2531 class N3VLSL<bit op24, bits<2> op21_20, bits<4> op11_8,
2532 InstrItinClass itin, string OpcodeStr, string Dt,
2533 ValueType TyQ, ValueType TyD, SDNode OpNode>
2534 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2535 (outs QPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2536 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2538 (TyQ (OpNode (TyD DPR:$Vn),
2539 (TyD (NEONvduplane (TyD DPR_VFP2:$Vm),imm:$lane)))))]>;
2540 class N3VLSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2541 InstrItinClass itin, string OpcodeStr, string Dt,
2542 ValueType TyQ, ValueType TyD, SDNode OpNode>
2543 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2544 (outs QPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2545 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2547 (TyQ (OpNode (TyD DPR:$Vn),
2548 (TyD (NEONvduplane (TyD DPR_8:$Vm), imm:$lane)))))]>;
2550 // Long 3-register operations with explicitly extended operands.
2551 class N3VLExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2552 InstrItinClass itin, string OpcodeStr, string Dt,
2553 ValueType TyQ, ValueType TyD, SDNode OpNode, SDNode ExtOp,
2555 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2556 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2557 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2558 [(set QPR:$Vd, (OpNode (TyQ (ExtOp (TyD DPR:$Vn))),
2559 (TyQ (ExtOp (TyD DPR:$Vm)))))]> {
2560 let isCommutable = Commutable;
2563 // Long 3-register intrinsics with explicit extend (VABDL).
2564 class N3VLIntExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2565 InstrItinClass itin, string OpcodeStr, string Dt,
2566 ValueType TyQ, ValueType TyD, Intrinsic IntOp, SDNode ExtOp,
2568 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2569 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2570 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2571 [(set QPR:$Vd, (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
2572 (TyD DPR:$Vm))))))]> {
2573 let isCommutable = Commutable;
2576 // Long 3-register intrinsics.
2577 class N3VLInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2578 InstrItinClass itin, string OpcodeStr, string Dt,
2579 ValueType TyQ, ValueType TyD, Intrinsic IntOp, bit Commutable>
2580 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2581 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2582 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2583 [(set QPR:$Vd, (TyQ (IntOp (TyD DPR:$Vn), (TyD DPR:$Vm))))]> {
2584 let isCommutable = Commutable;
2586 class N3VLIntSL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2587 string OpcodeStr, string Dt,
2588 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2589 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2590 (outs QPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2591 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2592 [(set (ResTy QPR:$Vd),
2593 (ResTy (IntOp (OpTy DPR:$Vn),
2594 (OpTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2596 class N3VLIntSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2597 InstrItinClass itin, string OpcodeStr, string Dt,
2598 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2599 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2600 (outs QPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2601 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2602 [(set (ResTy QPR:$Vd),
2603 (ResTy (IntOp (OpTy DPR:$Vn),
2604 (OpTy (NEONvduplane (OpTy DPR_8:$Vm),
2607 // Wide 3-register operations.
2608 class N3VW<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2609 string OpcodeStr, string Dt, ValueType TyQ, ValueType TyD,
2610 SDNode OpNode, SDNode ExtOp, bit Commutable>
2611 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2612 (outs QPR:$Vd), (ins QPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VSUBiD,
2613 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2614 [(set QPR:$Vd, (OpNode (TyQ QPR:$Vn),
2615 (TyQ (ExtOp (TyD DPR:$Vm)))))]> {
2616 let isCommutable = Commutable;
2619 // Pairwise long 2-register intrinsics, both double- and quad-register.
2620 class N2VDPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2621 bits<2> op17_16, bits<5> op11_7, bit op4,
2622 string OpcodeStr, string Dt,
2623 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2624 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
2625 (ins DPR:$Vm), IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
2626 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm))))]>;
2627 class N2VQPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2628 bits<2> op17_16, bits<5> op11_7, bit op4,
2629 string OpcodeStr, string Dt,
2630 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2631 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
2632 (ins QPR:$Vm), IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
2633 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm))))]>;
2635 // Pairwise long 2-register accumulate intrinsics,
2636 // both double- and quad-register.
2637 // The destination register is also used as the first source operand register.
2638 class N2VDPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2639 bits<2> op17_16, bits<5> op11_7, bit op4,
2640 string OpcodeStr, string Dt,
2641 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2642 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
2643 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vm), IIC_VPALiD,
2644 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
2645 [(set DPR:$Vd, (ResTy (IntOp (ResTy DPR:$src1), (OpTy DPR:$Vm))))]>;
2646 class N2VQPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2647 bits<2> op17_16, bits<5> op11_7, bit op4,
2648 string OpcodeStr, string Dt,
2649 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2650 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4,
2651 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vm), IIC_VPALiQ,
2652 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
2653 [(set QPR:$Vd, (ResTy (IntOp (ResTy QPR:$src1), (OpTy QPR:$Vm))))]>;
2655 // Shift by immediate,
2656 // both double- and quad-register.
2657 class N2VDSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2658 Format f, InstrItinClass itin, Operand ImmTy,
2659 string OpcodeStr, string Dt, ValueType Ty, SDNode OpNode>
2660 : N2VImm<op24, op23, op11_8, op7, 0, op4,
2661 (outs DPR:$Vd), (ins DPR:$Vm, ImmTy:$SIMM), f, itin,
2662 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2663 [(set DPR:$Vd, (Ty (OpNode (Ty DPR:$Vm), (i32 imm:$SIMM))))]>;
2664 class N2VQSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2665 Format f, InstrItinClass itin, Operand ImmTy,
2666 string OpcodeStr, string Dt, ValueType Ty, SDNode OpNode>
2667 : N2VImm<op24, op23, op11_8, op7, 1, op4,
2668 (outs QPR:$Vd), (ins QPR:$Vm, ImmTy:$SIMM), f, itin,
2669 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2670 [(set QPR:$Vd, (Ty (OpNode (Ty QPR:$Vm), (i32 imm:$SIMM))))]>;
2672 // Long shift by immediate.
2673 class N2VLSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
2674 string OpcodeStr, string Dt,
2675 ValueType ResTy, ValueType OpTy, Operand ImmTy, SDNode OpNode>
2676 : N2VImm<op24, op23, op11_8, op7, op6, op4,
2677 (outs QPR:$Vd), (ins DPR:$Vm, ImmTy:$SIMM), N2RegVShLFrm,
2678 IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2679 [(set QPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vm),
2680 (i32 imm:$SIMM))))]>;
2682 // Narrow shift by immediate.
2683 class N2VNSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
2684 InstrItinClass itin, string OpcodeStr, string Dt,
2685 ValueType ResTy, ValueType OpTy, Operand ImmTy, SDNode OpNode>
2686 : N2VImm<op24, op23, op11_8, op7, op6, op4,
2687 (outs DPR:$Vd), (ins QPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, itin,
2688 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2689 [(set DPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vm),
2690 (i32 imm:$SIMM))))]>;
2692 // Shift right by immediate and accumulate,
2693 // both double- and quad-register.
2694 class N2VDShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2695 Operand ImmTy, string OpcodeStr, string Dt,
2696 ValueType Ty, SDNode ShOp>
2697 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
2698 (ins DPR:$src1, DPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, IIC_VPALiD,
2699 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2700 [(set DPR:$Vd, (Ty (add DPR:$src1,
2701 (Ty (ShOp DPR:$Vm, (i32 imm:$SIMM))))))]>;
2702 class N2VQShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2703 Operand ImmTy, string OpcodeStr, string Dt,
2704 ValueType Ty, SDNode ShOp>
2705 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
2706 (ins QPR:$src1, QPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, IIC_VPALiD,
2707 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2708 [(set QPR:$Vd, (Ty (add QPR:$src1,
2709 (Ty (ShOp QPR:$Vm, (i32 imm:$SIMM))))))]>;
2711 // Shift by immediate and insert,
2712 // both double- and quad-register.
2713 class N2VDShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2714 Operand ImmTy, Format f, string OpcodeStr, string Dt,
2715 ValueType Ty,SDNode ShOp>
2716 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
2717 (ins DPR:$src1, DPR:$Vm, ImmTy:$SIMM), f, IIC_VSHLiD,
2718 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2719 [(set DPR:$Vd, (Ty (ShOp DPR:$src1, DPR:$Vm, (i32 imm:$SIMM))))]>;
2720 class N2VQShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2721 Operand ImmTy, Format f, string OpcodeStr, string Dt,
2722 ValueType Ty,SDNode ShOp>
2723 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
2724 (ins QPR:$src1, QPR:$Vm, ImmTy:$SIMM), f, IIC_VSHLiQ,
2725 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2726 [(set QPR:$Vd, (Ty (ShOp QPR:$src1, QPR:$Vm, (i32 imm:$SIMM))))]>;
2728 // Convert, with fractional bits immediate,
2729 // both double- and quad-register.
2730 class N2VCvtD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2731 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2733 : N2VImm<op24, op23, op11_8, op7, 0, op4,
2734 (outs DPR:$Vd), (ins DPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
2735 IIC_VUNAD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2736 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (i32 imm:$SIMM))))]>;
2737 class N2VCvtQ<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2738 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2740 : N2VImm<op24, op23, op11_8, op7, 1, op4,
2741 (outs QPR:$Vd), (ins QPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
2742 IIC_VUNAQ, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2743 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (i32 imm:$SIMM))))]>;
2745 //===----------------------------------------------------------------------===//
2747 //===----------------------------------------------------------------------===//
2749 // Abbreviations used in multiclass suffixes:
2750 // Q = quarter int (8 bit) elements
2751 // H = half int (16 bit) elements
2752 // S = single int (32 bit) elements
2753 // D = double int (64 bit) elements
2755 // Neon 2-register vector operations and intrinsics.
2757 // Neon 2-register comparisons.
2758 // source operand element sizes of 8, 16 and 32 bits:
2759 multiclass N2V_QHS_cmp<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2760 bits<5> op11_7, bit op4, string opc, string Dt,
2761 string asm, SDNode OpNode> {
2762 // 64-bit vector types.
2763 def v8i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 0, op4,
2764 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2765 opc, !strconcat(Dt, "8"), asm, "",
2766 [(set DPR:$Vd, (v8i8 (OpNode (v8i8 DPR:$Vm))))]>;
2767 def v4i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 0, op4,
2768 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2769 opc, !strconcat(Dt, "16"), asm, "",
2770 [(set DPR:$Vd, (v4i16 (OpNode (v4i16 DPR:$Vm))))]>;
2771 def v2i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
2772 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2773 opc, !strconcat(Dt, "32"), asm, "",
2774 [(set DPR:$Vd, (v2i32 (OpNode (v2i32 DPR:$Vm))))]>;
2775 def v2f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
2776 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2777 opc, "f32", asm, "",
2778 [(set DPR:$Vd, (v2i32 (OpNode (v2f32 DPR:$Vm))))]> {
2779 let Inst{10} = 1; // overwrite F = 1
2782 // 128-bit vector types.
2783 def v16i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 1, op4,
2784 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2785 opc, !strconcat(Dt, "8"), asm, "",
2786 [(set QPR:$Vd, (v16i8 (OpNode (v16i8 QPR:$Vm))))]>;
2787 def v8i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 1, op4,
2788 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2789 opc, !strconcat(Dt, "16"), asm, "",
2790 [(set QPR:$Vd, (v8i16 (OpNode (v8i16 QPR:$Vm))))]>;
2791 def v4i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
2792 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2793 opc, !strconcat(Dt, "32"), asm, "",
2794 [(set QPR:$Vd, (v4i32 (OpNode (v4i32 QPR:$Vm))))]>;
2795 def v4f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
2796 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2797 opc, "f32", asm, "",
2798 [(set QPR:$Vd, (v4i32 (OpNode (v4f32 QPR:$Vm))))]> {
2799 let Inst{10} = 1; // overwrite F = 1
2804 // Neon 2-register vector intrinsics,
2805 // element sizes of 8, 16 and 32 bits:
2806 multiclass N2VInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2807 bits<5> op11_7, bit op4,
2808 InstrItinClass itinD, InstrItinClass itinQ,
2809 string OpcodeStr, string Dt, Intrinsic IntOp> {
2810 // 64-bit vector types.
2811 def v8i8 : N2VDInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2812 itinD, OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
2813 def v4i16 : N2VDInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2814 itinD, OpcodeStr, !strconcat(Dt, "16"),v4i16,v4i16,IntOp>;
2815 def v2i32 : N2VDInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2816 itinD, OpcodeStr, !strconcat(Dt, "32"),v2i32,v2i32,IntOp>;
2818 // 128-bit vector types.
2819 def v16i8 : N2VQInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2820 itinQ, OpcodeStr, !strconcat(Dt, "8"), v16i8,v16i8,IntOp>;
2821 def v8i16 : N2VQInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2822 itinQ, OpcodeStr, !strconcat(Dt, "16"),v8i16,v8i16,IntOp>;
2823 def v4i32 : N2VQInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2824 itinQ, OpcodeStr, !strconcat(Dt, "32"),v4i32,v4i32,IntOp>;
2828 // Neon Narrowing 2-register vector operations,
2829 // source operand element sizes of 16, 32 and 64 bits:
2830 multiclass N2VN_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2831 bits<5> op11_7, bit op6, bit op4,
2832 InstrItinClass itin, string OpcodeStr, string Dt,
2834 def v8i8 : N2VN<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
2835 itin, OpcodeStr, !strconcat(Dt, "16"),
2836 v8i8, v8i16, OpNode>;
2837 def v4i16 : N2VN<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
2838 itin, OpcodeStr, !strconcat(Dt, "32"),
2839 v4i16, v4i32, OpNode>;
2840 def v2i32 : N2VN<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
2841 itin, OpcodeStr, !strconcat(Dt, "64"),
2842 v2i32, v2i64, OpNode>;
2845 // Neon Narrowing 2-register vector intrinsics,
2846 // source operand element sizes of 16, 32 and 64 bits:
2847 multiclass N2VNInt_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2848 bits<5> op11_7, bit op6, bit op4,
2849 InstrItinClass itin, string OpcodeStr, string Dt,
2851 def v8i8 : N2VNInt<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
2852 itin, OpcodeStr, !strconcat(Dt, "16"),
2853 v8i8, v8i16, IntOp>;
2854 def v4i16 : N2VNInt<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
2855 itin, OpcodeStr, !strconcat(Dt, "32"),
2856 v4i16, v4i32, IntOp>;
2857 def v2i32 : N2VNInt<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
2858 itin, OpcodeStr, !strconcat(Dt, "64"),
2859 v2i32, v2i64, IntOp>;
2863 // Neon Lengthening 2-register vector intrinsic (currently specific to VMOVL).
2864 // source operand element sizes of 16, 32 and 64 bits:
2865 multiclass N2VL_QHS<bits<2> op24_23, bits<5> op11_7, bit op6, bit op4,
2866 string OpcodeStr, string Dt, SDNode OpNode> {
2867 def v8i16 : N2VL<op24_23, 0b00, 0b10, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2868 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode>;
2869 def v4i32 : N2VL<op24_23, 0b01, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2870 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
2871 def v2i64 : N2VL<op24_23, 0b10, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2872 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
2876 // Neon 3-register vector operations.
2878 // First with only element sizes of 8, 16 and 32 bits:
2879 multiclass N3V_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2880 InstrItinClass itinD16, InstrItinClass itinD32,
2881 InstrItinClass itinQ16, InstrItinClass itinQ32,
2882 string OpcodeStr, string Dt,
2883 SDNode OpNode, bit Commutable = 0> {
2884 // 64-bit vector types.
2885 def v8i8 : N3VD<op24, op23, 0b00, op11_8, op4, itinD16,
2886 OpcodeStr, !strconcat(Dt, "8"),
2887 v8i8, v8i8, OpNode, Commutable>;
2888 def v4i16 : N3VD<op24, op23, 0b01, op11_8, op4, itinD16,
2889 OpcodeStr, !strconcat(Dt, "16"),
2890 v4i16, v4i16, OpNode, Commutable>;
2891 def v2i32 : N3VD<op24, op23, 0b10, op11_8, op4, itinD32,
2892 OpcodeStr, !strconcat(Dt, "32"),
2893 v2i32, v2i32, OpNode, Commutable>;
2895 // 128-bit vector types.
2896 def v16i8 : N3VQ<op24, op23, 0b00, op11_8, op4, itinQ16,
2897 OpcodeStr, !strconcat(Dt, "8"),
2898 v16i8, v16i8, OpNode, Commutable>;
2899 def v8i16 : N3VQ<op24, op23, 0b01, op11_8, op4, itinQ16,
2900 OpcodeStr, !strconcat(Dt, "16"),
2901 v8i16, v8i16, OpNode, Commutable>;
2902 def v4i32 : N3VQ<op24, op23, 0b10, op11_8, op4, itinQ32,
2903 OpcodeStr, !strconcat(Dt, "32"),
2904 v4i32, v4i32, OpNode, Commutable>;
2907 multiclass N3VSL_HS<bits<4> op11_8, string OpcodeStr, SDNode ShOp> {
2908 def v4i16 : N3VDSL16<0b01, op11_8, OpcodeStr, "i16", v4i16, ShOp>;
2909 def v2i32 : N3VDSL<0b10, op11_8, IIC_VMULi32D, OpcodeStr, "i32", v2i32, ShOp>;
2910 def v8i16 : N3VQSL16<0b01, op11_8, OpcodeStr, "i16", v8i16, v4i16, ShOp>;
2911 def v4i32 : N3VQSL<0b10, op11_8, IIC_VMULi32Q, OpcodeStr, "i32",
2912 v4i32, v2i32, ShOp>;
2915 // ....then also with element size 64 bits:
2916 multiclass N3V_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2917 InstrItinClass itinD, InstrItinClass itinQ,
2918 string OpcodeStr, string Dt,
2919 SDNode OpNode, bit Commutable = 0>
2920 : N3V_QHS<op24, op23, op11_8, op4, itinD, itinD, itinQ, itinQ,
2921 OpcodeStr, Dt, OpNode, Commutable> {
2922 def v1i64 : N3VD<op24, op23, 0b11, op11_8, op4, itinD,
2923 OpcodeStr, !strconcat(Dt, "64"),
2924 v1i64, v1i64, OpNode, Commutable>;
2925 def v2i64 : N3VQ<op24, op23, 0b11, op11_8, op4, itinQ,
2926 OpcodeStr, !strconcat(Dt, "64"),
2927 v2i64, v2i64, OpNode, Commutable>;
2931 // Neon 3-register vector intrinsics.
2933 // First with only element sizes of 16 and 32 bits:
2934 multiclass N3VInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2935 InstrItinClass itinD16, InstrItinClass itinD32,
2936 InstrItinClass itinQ16, InstrItinClass itinQ32,
2937 string OpcodeStr, string Dt,
2938 Intrinsic IntOp, bit Commutable = 0> {
2939 // 64-bit vector types.
2940 def v4i16 : N3VDInt<op24, op23, 0b01, op11_8, op4, f, itinD16,
2941 OpcodeStr, !strconcat(Dt, "16"),
2942 v4i16, v4i16, IntOp, Commutable>;
2943 def v2i32 : N3VDInt<op24, op23, 0b10, op11_8, op4, f, itinD32,
2944 OpcodeStr, !strconcat(Dt, "32"),
2945 v2i32, v2i32, IntOp, Commutable>;
2947 // 128-bit vector types.
2948 def v8i16 : N3VQInt<op24, op23, 0b01, op11_8, op4, f, itinQ16,
2949 OpcodeStr, !strconcat(Dt, "16"),
2950 v8i16, v8i16, IntOp, Commutable>;
2951 def v4i32 : N3VQInt<op24, op23, 0b10, op11_8, op4, f, itinQ32,
2952 OpcodeStr, !strconcat(Dt, "32"),
2953 v4i32, v4i32, IntOp, Commutable>;
2955 multiclass N3VInt_HSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2956 InstrItinClass itinD16, InstrItinClass itinD32,
2957 InstrItinClass itinQ16, InstrItinClass itinQ32,
2958 string OpcodeStr, string Dt,
2960 // 64-bit vector types.
2961 def v4i16 : N3VDIntSh<op24, op23, 0b01, op11_8, op4, f, itinD16,
2962 OpcodeStr, !strconcat(Dt, "16"),
2963 v4i16, v4i16, IntOp>;
2964 def v2i32 : N3VDIntSh<op24, op23, 0b10, op11_8, op4, f, itinD32,
2965 OpcodeStr, !strconcat(Dt, "32"),
2966 v2i32, v2i32, IntOp>;
2968 // 128-bit vector types.
2969 def v8i16 : N3VQIntSh<op24, op23, 0b01, op11_8, op4, f, itinQ16,
2970 OpcodeStr, !strconcat(Dt, "16"),
2971 v8i16, v8i16, IntOp>;
2972 def v4i32 : N3VQIntSh<op24, op23, 0b10, op11_8, op4, f, itinQ32,
2973 OpcodeStr, !strconcat(Dt, "32"),
2974 v4i32, v4i32, IntOp>;
2977 multiclass N3VIntSL_HS<bits<4> op11_8,
2978 InstrItinClass itinD16, InstrItinClass itinD32,
2979 InstrItinClass itinQ16, InstrItinClass itinQ32,
2980 string OpcodeStr, string Dt, Intrinsic IntOp> {
2981 def v4i16 : N3VDIntSL16<0b01, op11_8, itinD16,
2982 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp>;
2983 def v2i32 : N3VDIntSL<0b10, op11_8, itinD32,
2984 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp>;
2985 def v8i16 : N3VQIntSL16<0b01, op11_8, itinQ16,
2986 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16, IntOp>;
2987 def v4i32 : N3VQIntSL<0b10, op11_8, itinQ32,
2988 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32, IntOp>;
2991 // ....then also with element size of 8 bits:
2992 multiclass N3VInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2993 InstrItinClass itinD16, InstrItinClass itinD32,
2994 InstrItinClass itinQ16, InstrItinClass itinQ32,
2995 string OpcodeStr, string Dt,
2996 Intrinsic IntOp, bit Commutable = 0>
2997 : N3VInt_HS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2998 OpcodeStr, Dt, IntOp, Commutable> {
2999 def v8i8 : N3VDInt<op24, op23, 0b00, op11_8, op4, f, itinD16,
3000 OpcodeStr, !strconcat(Dt, "8"),
3001 v8i8, v8i8, IntOp, Commutable>;
3002 def v16i8 : N3VQInt<op24, op23, 0b00, op11_8, op4, f, itinQ16,
3003 OpcodeStr, !strconcat(Dt, "8"),
3004 v16i8, v16i8, IntOp, Commutable>;
3006 multiclass N3VInt_QHSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3007 InstrItinClass itinD16, InstrItinClass itinD32,
3008 InstrItinClass itinQ16, InstrItinClass itinQ32,
3009 string OpcodeStr, string Dt,
3011 : N3VInt_HSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
3012 OpcodeStr, Dt, IntOp> {
3013 def v8i8 : N3VDIntSh<op24, op23, 0b00, op11_8, op4, f, itinD16,
3014 OpcodeStr, !strconcat(Dt, "8"),
3016 def v16i8 : N3VQIntSh<op24, op23, 0b00, op11_8, op4, f, itinQ16,
3017 OpcodeStr, !strconcat(Dt, "8"),
3018 v16i8, v16i8, IntOp>;
3022 // ....then also with element size of 64 bits:
3023 multiclass N3VInt_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3024 InstrItinClass itinD16, InstrItinClass itinD32,
3025 InstrItinClass itinQ16, InstrItinClass itinQ32,
3026 string OpcodeStr, string Dt,
3027 Intrinsic IntOp, bit Commutable = 0>
3028 : N3VInt_QHS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
3029 OpcodeStr, Dt, IntOp, Commutable> {
3030 def v1i64 : N3VDInt<op24, op23, 0b11, op11_8, op4, f, itinD32,
3031 OpcodeStr, !strconcat(Dt, "64"),
3032 v1i64, v1i64, IntOp, Commutable>;
3033 def v2i64 : N3VQInt<op24, op23, 0b11, op11_8, op4, f, itinQ32,
3034 OpcodeStr, !strconcat(Dt, "64"),
3035 v2i64, v2i64, IntOp, Commutable>;
3037 multiclass N3VInt_QHSDSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3038 InstrItinClass itinD16, InstrItinClass itinD32,
3039 InstrItinClass itinQ16, InstrItinClass itinQ32,
3040 string OpcodeStr, string Dt,
3042 : N3VInt_QHSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
3043 OpcodeStr, Dt, IntOp> {
3044 def v1i64 : N3VDIntSh<op24, op23, 0b11, op11_8, op4, f, itinD32,
3045 OpcodeStr, !strconcat(Dt, "64"),
3046 v1i64, v1i64, IntOp>;
3047 def v2i64 : N3VQIntSh<op24, op23, 0b11, op11_8, op4, f, itinQ32,
3048 OpcodeStr, !strconcat(Dt, "64"),
3049 v2i64, v2i64, IntOp>;
3052 // Neon Narrowing 3-register vector intrinsics,
3053 // source operand element sizes of 16, 32 and 64 bits:
3054 multiclass N3VNInt_HSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3055 string OpcodeStr, string Dt,
3056 Intrinsic IntOp, bit Commutable = 0> {
3057 def v8i8 : N3VNInt<op24, op23, 0b00, op11_8, op4,
3058 OpcodeStr, !strconcat(Dt, "16"),
3059 v8i8, v8i16, IntOp, Commutable>;
3060 def v4i16 : N3VNInt<op24, op23, 0b01, op11_8, op4,
3061 OpcodeStr, !strconcat(Dt, "32"),
3062 v4i16, v4i32, IntOp, Commutable>;
3063 def v2i32 : N3VNInt<op24, op23, 0b10, op11_8, op4,
3064 OpcodeStr, !strconcat(Dt, "64"),
3065 v2i32, v2i64, IntOp, Commutable>;
3069 // Neon Long 3-register vector operations.
3071 multiclass N3VL_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3072 InstrItinClass itin16, InstrItinClass itin32,
3073 string OpcodeStr, string Dt,
3074 SDNode OpNode, bit Commutable = 0> {
3075 def v8i16 : N3VL<op24, op23, 0b00, op11_8, op4, itin16,
3076 OpcodeStr, !strconcat(Dt, "8"),
3077 v8i16, v8i8, OpNode, Commutable>;
3078 def v4i32 : N3VL<op24, op23, 0b01, op11_8, op4, itin16,
3079 OpcodeStr, !strconcat(Dt, "16"),
3080 v4i32, v4i16, OpNode, Commutable>;
3081 def v2i64 : N3VL<op24, op23, 0b10, op11_8, op4, itin32,
3082 OpcodeStr, !strconcat(Dt, "32"),
3083 v2i64, v2i32, OpNode, Commutable>;
3086 multiclass N3VLSL_HS<bit op24, bits<4> op11_8,
3087 InstrItinClass itin, string OpcodeStr, string Dt,
3089 def v4i16 : N3VLSL16<op24, 0b01, op11_8, itin, OpcodeStr,
3090 !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
3091 def v2i32 : N3VLSL<op24, 0b10, op11_8, itin, OpcodeStr,
3092 !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
3095 multiclass N3VLExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3096 InstrItinClass itin16, InstrItinClass itin32,
3097 string OpcodeStr, string Dt,
3098 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
3099 def v8i16 : N3VLExt<op24, op23, 0b00, op11_8, op4, itin16,
3100 OpcodeStr, !strconcat(Dt, "8"),
3101 v8i16, v8i8, OpNode, ExtOp, Commutable>;
3102 def v4i32 : N3VLExt<op24, op23, 0b01, op11_8, op4, itin16,
3103 OpcodeStr, !strconcat(Dt, "16"),
3104 v4i32, v4i16, OpNode, ExtOp, Commutable>;
3105 def v2i64 : N3VLExt<op24, op23, 0b10, op11_8, op4, itin32,
3106 OpcodeStr, !strconcat(Dt, "32"),
3107 v2i64, v2i32, OpNode, ExtOp, Commutable>;
3110 // Neon Long 3-register vector intrinsics.
3112 // First with only element sizes of 16 and 32 bits:
3113 multiclass N3VLInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
3114 InstrItinClass itin16, InstrItinClass itin32,
3115 string OpcodeStr, string Dt,
3116 Intrinsic IntOp, bit Commutable = 0> {
3117 def v4i32 : N3VLInt<op24, op23, 0b01, op11_8, op4, itin16,
3118 OpcodeStr, !strconcat(Dt, "16"),
3119 v4i32, v4i16, IntOp, Commutable>;
3120 def v2i64 : N3VLInt<op24, op23, 0b10, op11_8, op4, itin32,
3121 OpcodeStr, !strconcat(Dt, "32"),
3122 v2i64, v2i32, IntOp, Commutable>;
3125 multiclass N3VLIntSL_HS<bit op24, bits<4> op11_8,
3126 InstrItinClass itin, string OpcodeStr, string Dt,
3128 def v4i16 : N3VLIntSL16<op24, 0b01, op11_8, itin,
3129 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
3130 def v2i32 : N3VLIntSL<op24, 0b10, op11_8, itin,
3131 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3134 // ....then also with element size of 8 bits:
3135 multiclass N3VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3136 InstrItinClass itin16, InstrItinClass itin32,
3137 string OpcodeStr, string Dt,
3138 Intrinsic IntOp, bit Commutable = 0>
3139 : N3VLInt_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt,
3140 IntOp, Commutable> {
3141 def v8i16 : N3VLInt<op24, op23, 0b00, op11_8, op4, itin16,
3142 OpcodeStr, !strconcat(Dt, "8"),
3143 v8i16, v8i8, IntOp, Commutable>;
3146 // ....with explicit extend (VABDL).
3147 multiclass N3VLIntExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3148 InstrItinClass itin, string OpcodeStr, string Dt,
3149 Intrinsic IntOp, SDNode ExtOp, bit Commutable = 0> {
3150 def v8i16 : N3VLIntExt<op24, op23, 0b00, op11_8, op4, itin,
3151 OpcodeStr, !strconcat(Dt, "8"),
3152 v8i16, v8i8, IntOp, ExtOp, Commutable>;
3153 def v4i32 : N3VLIntExt<op24, op23, 0b01, op11_8, op4, itin,
3154 OpcodeStr, !strconcat(Dt, "16"),
3155 v4i32, v4i16, IntOp, ExtOp, Commutable>;
3156 def v2i64 : N3VLIntExt<op24, op23, 0b10, op11_8, op4, itin,
3157 OpcodeStr, !strconcat(Dt, "32"),
3158 v2i64, v2i32, IntOp, ExtOp, Commutable>;
3162 // Neon Wide 3-register vector intrinsics,
3163 // source operand element sizes of 8, 16 and 32 bits:
3164 multiclass N3VW_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3165 string OpcodeStr, string Dt,
3166 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
3167 def v8i16 : N3VW<op24, op23, 0b00, op11_8, op4,
3168 OpcodeStr, !strconcat(Dt, "8"),
3169 v8i16, v8i8, OpNode, ExtOp, Commutable>;
3170 def v4i32 : N3VW<op24, op23, 0b01, op11_8, op4,
3171 OpcodeStr, !strconcat(Dt, "16"),
3172 v4i32, v4i16, OpNode, ExtOp, Commutable>;
3173 def v2i64 : N3VW<op24, op23, 0b10, op11_8, op4,
3174 OpcodeStr, !strconcat(Dt, "32"),
3175 v2i64, v2i32, OpNode, ExtOp, Commutable>;
3179 // Neon Multiply-Op vector operations,
3180 // element sizes of 8, 16 and 32 bits:
3181 multiclass N3VMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3182 InstrItinClass itinD16, InstrItinClass itinD32,
3183 InstrItinClass itinQ16, InstrItinClass itinQ32,
3184 string OpcodeStr, string Dt, SDNode OpNode> {
3185 // 64-bit vector types.
3186 def v8i8 : N3VDMulOp<op24, op23, 0b00, op11_8, op4, itinD16,
3187 OpcodeStr, !strconcat(Dt, "8"), v8i8, mul, OpNode>;
3188 def v4i16 : N3VDMulOp<op24, op23, 0b01, op11_8, op4, itinD16,
3189 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, OpNode>;
3190 def v2i32 : N3VDMulOp<op24, op23, 0b10, op11_8, op4, itinD32,
3191 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, OpNode>;
3193 // 128-bit vector types.
3194 def v16i8 : N3VQMulOp<op24, op23, 0b00, op11_8, op4, itinQ16,
3195 OpcodeStr, !strconcat(Dt, "8"), v16i8, mul, OpNode>;
3196 def v8i16 : N3VQMulOp<op24, op23, 0b01, op11_8, op4, itinQ16,
3197 OpcodeStr, !strconcat(Dt, "16"), v8i16, mul, OpNode>;
3198 def v4i32 : N3VQMulOp<op24, op23, 0b10, op11_8, op4, itinQ32,
3199 OpcodeStr, !strconcat(Dt, "32"), v4i32, mul, OpNode>;
3202 multiclass N3VMulOpSL_HS<bits<4> op11_8,
3203 InstrItinClass itinD16, InstrItinClass itinD32,
3204 InstrItinClass itinQ16, InstrItinClass itinQ32,
3205 string OpcodeStr, string Dt, SDNode ShOp> {
3206 def v4i16 : N3VDMulOpSL16<0b01, op11_8, itinD16,
3207 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, ShOp>;
3208 def v2i32 : N3VDMulOpSL<0b10, op11_8, itinD32,
3209 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, ShOp>;
3210 def v8i16 : N3VQMulOpSL16<0b01, op11_8, itinQ16,
3211 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16,
3213 def v4i32 : N3VQMulOpSL<0b10, op11_8, itinQ32,
3214 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32,
3218 // Neon Intrinsic-Op vector operations,
3219 // element sizes of 8, 16 and 32 bits:
3220 multiclass N3VIntOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3221 InstrItinClass itinD, InstrItinClass itinQ,
3222 string OpcodeStr, string Dt, Intrinsic IntOp,
3224 // 64-bit vector types.
3225 def v8i8 : N3VDIntOp<op24, op23, 0b00, op11_8, op4, itinD,
3226 OpcodeStr, !strconcat(Dt, "8"), v8i8, IntOp, OpNode>;
3227 def v4i16 : N3VDIntOp<op24, op23, 0b01, op11_8, op4, itinD,
3228 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp, OpNode>;
3229 def v2i32 : N3VDIntOp<op24, op23, 0b10, op11_8, op4, itinD,
3230 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp, OpNode>;
3232 // 128-bit vector types.
3233 def v16i8 : N3VQIntOp<op24, op23, 0b00, op11_8, op4, itinQ,
3234 OpcodeStr, !strconcat(Dt, "8"), v16i8, IntOp, OpNode>;
3235 def v8i16 : N3VQIntOp<op24, op23, 0b01, op11_8, op4, itinQ,
3236 OpcodeStr, !strconcat(Dt, "16"), v8i16, IntOp, OpNode>;
3237 def v4i32 : N3VQIntOp<op24, op23, 0b10, op11_8, op4, itinQ,
3238 OpcodeStr, !strconcat(Dt, "32"), v4i32, IntOp, OpNode>;
3241 // Neon 3-argument intrinsics,
3242 // element sizes of 8, 16 and 32 bits:
3243 multiclass N3VInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3244 InstrItinClass itinD, InstrItinClass itinQ,
3245 string OpcodeStr, string Dt, Intrinsic IntOp> {
3246 // 64-bit vector types.
3247 def v8i8 : N3VDInt3<op24, op23, 0b00, op11_8, op4, itinD,
3248 OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
3249 def v4i16 : N3VDInt3<op24, op23, 0b01, op11_8, op4, itinD,
3250 OpcodeStr, !strconcat(Dt, "16"), v4i16, v4i16, IntOp>;
3251 def v2i32 : N3VDInt3<op24, op23, 0b10, op11_8, op4, itinD,
3252 OpcodeStr, !strconcat(Dt, "32"), v2i32, v2i32, IntOp>;
3254 // 128-bit vector types.
3255 def v16i8 : N3VQInt3<op24, op23, 0b00, op11_8, op4, itinQ,
3256 OpcodeStr, !strconcat(Dt, "8"), v16i8, v16i8, IntOp>;
3257 def v8i16 : N3VQInt3<op24, op23, 0b01, op11_8, op4, itinQ,
3258 OpcodeStr, !strconcat(Dt, "16"), v8i16, v8i16, IntOp>;
3259 def v4i32 : N3VQInt3<op24, op23, 0b10, op11_8, op4, itinQ,
3260 OpcodeStr, !strconcat(Dt, "32"), v4i32, v4i32, IntOp>;
3264 // Neon Long Multiply-Op vector operations,
3265 // element sizes of 8, 16 and 32 bits:
3266 multiclass N3VLMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3267 InstrItinClass itin16, InstrItinClass itin32,
3268 string OpcodeStr, string Dt, SDNode MulOp,
3270 def v8i16 : N3VLMulOp<op24, op23, 0b00, op11_8, op4, itin16, OpcodeStr,
3271 !strconcat(Dt, "8"), v8i16, v8i8, MulOp, OpNode>;
3272 def v4i32 : N3VLMulOp<op24, op23, 0b01, op11_8, op4, itin16, OpcodeStr,
3273 !strconcat(Dt, "16"), v4i32, v4i16, MulOp, OpNode>;
3274 def v2i64 : N3VLMulOp<op24, op23, 0b10, op11_8, op4, itin32, OpcodeStr,
3275 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
3278 multiclass N3VLMulOpSL_HS<bit op24, bits<4> op11_8, string OpcodeStr,
3279 string Dt, SDNode MulOp, SDNode OpNode> {
3280 def v4i16 : N3VLMulOpSL16<op24, 0b01, op11_8, IIC_VMACi16D, OpcodeStr,
3281 !strconcat(Dt,"16"), v4i32, v4i16, MulOp, OpNode>;
3282 def v2i32 : N3VLMulOpSL<op24, 0b10, op11_8, IIC_VMACi32D, OpcodeStr,
3283 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
3287 // Neon Long 3-argument intrinsics.
3289 // First with only element sizes of 16 and 32 bits:
3290 multiclass N3VLInt3_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
3291 InstrItinClass itin16, InstrItinClass itin32,
3292 string OpcodeStr, string Dt, Intrinsic IntOp> {
3293 def v4i32 : N3VLInt3<op24, op23, 0b01, op11_8, op4, itin16,
3294 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
3295 def v2i64 : N3VLInt3<op24, op23, 0b10, op11_8, op4, itin32,
3296 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3299 multiclass N3VLInt3SL_HS<bit op24, bits<4> op11_8,
3300 string OpcodeStr, string Dt, Intrinsic IntOp> {
3301 def v4i16 : N3VLInt3SL16<op24, 0b01, op11_8, IIC_VMACi16D,
3302 OpcodeStr, !strconcat(Dt,"16"), v4i32, v4i16, IntOp>;
3303 def v2i32 : N3VLInt3SL<op24, 0b10, op11_8, IIC_VMACi32D,
3304 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3307 // ....then also with element size of 8 bits:
3308 multiclass N3VLInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3309 InstrItinClass itin16, InstrItinClass itin32,
3310 string OpcodeStr, string Dt, Intrinsic IntOp>
3311 : N3VLInt3_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt, IntOp> {
3312 def v8i16 : N3VLInt3<op24, op23, 0b00, op11_8, op4, itin16,
3313 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
3316 // ....with explicit extend (VABAL).
3317 multiclass N3VLIntExtOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3318 InstrItinClass itin, string OpcodeStr, string Dt,
3319 Intrinsic IntOp, SDNode ExtOp, SDNode OpNode> {
3320 def v8i16 : N3VLIntExtOp<op24, op23, 0b00, op11_8, op4, itin,
3321 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8,
3322 IntOp, ExtOp, OpNode>;
3323 def v4i32 : N3VLIntExtOp<op24, op23, 0b01, op11_8, op4, itin,
3324 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16,
3325 IntOp, ExtOp, OpNode>;
3326 def v2i64 : N3VLIntExtOp<op24, op23, 0b10, op11_8, op4, itin,
3327 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32,
3328 IntOp, ExtOp, OpNode>;
3332 // Neon Pairwise long 2-register intrinsics,
3333 // element sizes of 8, 16 and 32 bits:
3334 multiclass N2VPLInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3335 bits<5> op11_7, bit op4,
3336 string OpcodeStr, string Dt, Intrinsic IntOp> {
3337 // 64-bit vector types.
3338 def v8i8 : N2VDPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3339 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
3340 def v4i16 : N2VDPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3341 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
3342 def v2i32 : N2VDPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3343 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
3345 // 128-bit vector types.
3346 def v16i8 : N2VQPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3347 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
3348 def v8i16 : N2VQPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3349 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
3350 def v4i32 : N2VQPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3351 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
3355 // Neon Pairwise long 2-register accumulate intrinsics,
3356 // element sizes of 8, 16 and 32 bits:
3357 multiclass N2VPLInt2_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3358 bits<5> op11_7, bit op4,
3359 string OpcodeStr, string Dt, Intrinsic IntOp> {
3360 // 64-bit vector types.
3361 def v8i8 : N2VDPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3362 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
3363 def v4i16 : N2VDPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3364 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
3365 def v2i32 : N2VDPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3366 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
3368 // 128-bit vector types.
3369 def v16i8 : N2VQPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3370 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
3371 def v8i16 : N2VQPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3372 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
3373 def v4i32 : N2VQPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3374 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
3378 // Neon 2-register vector shift by immediate,
3379 // with f of either N2RegVShLFrm or N2RegVShRFrm
3380 // element sizes of 8, 16, 32 and 64 bits:
3381 multiclass N2VShL_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3382 InstrItinClass itin, string OpcodeStr, string Dt,
3384 // 64-bit vector types.
3385 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3386 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
3387 let Inst{21-19} = 0b001; // imm6 = 001xxx
3389 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3390 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
3391 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3393 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3394 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
3395 let Inst{21} = 0b1; // imm6 = 1xxxxx
3397 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, N2RegVShLFrm, itin, i32imm,
3398 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
3401 // 128-bit vector types.
3402 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3403 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
3404 let Inst{21-19} = 0b001; // imm6 = 001xxx
3406 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3407 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
3408 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3410 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3411 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
3412 let Inst{21} = 0b1; // imm6 = 1xxxxx
3414 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, N2RegVShLFrm, itin, i32imm,
3415 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
3418 multiclass N2VShR_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3419 InstrItinClass itin, string OpcodeStr, string Dt,
3421 // 64-bit vector types.
3422 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm8,
3423 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
3424 let Inst{21-19} = 0b001; // imm6 = 001xxx
3426 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm16,
3427 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
3428 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3430 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm32,
3431 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
3432 let Inst{21} = 0b1; // imm6 = 1xxxxx
3434 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, N2RegVShRFrm, itin, shr_imm64,
3435 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
3438 // 128-bit vector types.
3439 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm8,
3440 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
3441 let Inst{21-19} = 0b001; // imm6 = 001xxx
3443 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm16,
3444 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
3445 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3447 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm32,
3448 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
3449 let Inst{21} = 0b1; // imm6 = 1xxxxx
3451 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, N2RegVShRFrm, itin, shr_imm64,
3452 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
3456 // Neon Shift-Accumulate vector operations,
3457 // element sizes of 8, 16, 32 and 64 bits:
3458 multiclass N2VShAdd_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3459 string OpcodeStr, string Dt, SDNode ShOp> {
3460 // 64-bit vector types.
3461 def v8i8 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm8,
3462 OpcodeStr, !strconcat(Dt, "8"), v8i8, ShOp> {
3463 let Inst{21-19} = 0b001; // imm6 = 001xxx
3465 def v4i16 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm16,
3466 OpcodeStr, !strconcat(Dt, "16"), v4i16, ShOp> {
3467 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3469 def v2i32 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm32,
3470 OpcodeStr, !strconcat(Dt, "32"), v2i32, ShOp> {
3471 let Inst{21} = 0b1; // imm6 = 1xxxxx
3473 def v1i64 : N2VDShAdd<op24, op23, op11_8, 1, op4, shr_imm64,
3474 OpcodeStr, !strconcat(Dt, "64"), v1i64, ShOp>;
3477 // 128-bit vector types.
3478 def v16i8 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm8,
3479 OpcodeStr, !strconcat(Dt, "8"), v16i8, ShOp> {
3480 let Inst{21-19} = 0b001; // imm6 = 001xxx
3482 def v8i16 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm16,
3483 OpcodeStr, !strconcat(Dt, "16"), v8i16, ShOp> {
3484 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3486 def v4i32 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm32,
3487 OpcodeStr, !strconcat(Dt, "32"), v4i32, ShOp> {
3488 let Inst{21} = 0b1; // imm6 = 1xxxxx
3490 def v2i64 : N2VQShAdd<op24, op23, op11_8, 1, op4, shr_imm64,
3491 OpcodeStr, !strconcat(Dt, "64"), v2i64, ShOp>;
3495 // Neon Shift-Insert vector operations,
3496 // with f of either N2RegVShLFrm or N2RegVShRFrm
3497 // element sizes of 8, 16, 32 and 64 bits:
3498 multiclass N2VShInsL_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3500 // 64-bit vector types.
3501 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3502 N2RegVShLFrm, OpcodeStr, "8", v8i8, NEONvsli> {
3503 let Inst{21-19} = 0b001; // imm6 = 001xxx
3505 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3506 N2RegVShLFrm, OpcodeStr, "16", v4i16, NEONvsli> {
3507 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3509 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3510 N2RegVShLFrm, OpcodeStr, "32", v2i32, NEONvsli> {
3511 let Inst{21} = 0b1; // imm6 = 1xxxxx
3513 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4, i32imm,
3514 N2RegVShLFrm, OpcodeStr, "64", v1i64, NEONvsli>;
3517 // 128-bit vector types.
3518 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3519 N2RegVShLFrm, OpcodeStr, "8", v16i8, NEONvsli> {
3520 let Inst{21-19} = 0b001; // imm6 = 001xxx
3522 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3523 N2RegVShLFrm, OpcodeStr, "16", v8i16, NEONvsli> {
3524 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3526 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3527 N2RegVShLFrm, OpcodeStr, "32", v4i32, NEONvsli> {
3528 let Inst{21} = 0b1; // imm6 = 1xxxxx
3530 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4, i32imm,
3531 N2RegVShLFrm, OpcodeStr, "64", v2i64, NEONvsli>;
3534 multiclass N2VShInsR_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3536 // 64-bit vector types.
3537 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm8,
3538 N2RegVShRFrm, OpcodeStr, "8", v8i8, NEONvsri> {
3539 let Inst{21-19} = 0b001; // imm6 = 001xxx
3541 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm16,
3542 N2RegVShRFrm, OpcodeStr, "16", v4i16, NEONvsri> {
3543 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3545 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm32,
3546 N2RegVShRFrm, OpcodeStr, "32", v2i32, NEONvsri> {
3547 let Inst{21} = 0b1; // imm6 = 1xxxxx
3549 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4, shr_imm64,
3550 N2RegVShRFrm, OpcodeStr, "64", v1i64, NEONvsri>;
3553 // 128-bit vector types.
3554 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm8,
3555 N2RegVShRFrm, OpcodeStr, "8", v16i8, NEONvsri> {
3556 let Inst{21-19} = 0b001; // imm6 = 001xxx
3558 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm16,
3559 N2RegVShRFrm, OpcodeStr, "16", v8i16, NEONvsri> {
3560 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3562 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm32,
3563 N2RegVShRFrm, OpcodeStr, "32", v4i32, NEONvsri> {
3564 let Inst{21} = 0b1; // imm6 = 1xxxxx
3566 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4, shr_imm64,
3567 N2RegVShRFrm, OpcodeStr, "64", v2i64, NEONvsri>;
3571 // Neon Shift Long operations,
3572 // element sizes of 8, 16, 32 bits:
3573 multiclass N2VLSh_QHS<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
3574 bit op4, string OpcodeStr, string Dt, SDNode OpNode> {
3575 def v8i16 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3576 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, imm1_7, OpNode> {
3577 let Inst{21-19} = 0b001; // imm6 = 001xxx
3579 def v4i32 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3580 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, imm1_15, OpNode> {
3581 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3583 def v2i64 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3584 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, imm1_31, OpNode> {
3585 let Inst{21} = 0b1; // imm6 = 1xxxxx
3589 // Neon Shift Narrow operations,
3590 // element sizes of 16, 32, 64 bits:
3591 multiclass N2VNSh_HSD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
3592 bit op4, InstrItinClass itin, string OpcodeStr, string Dt,
3594 def v8i8 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3595 OpcodeStr, !strconcat(Dt, "16"),
3596 v8i8, v8i16, shr_imm8, OpNode> {
3597 let Inst{21-19} = 0b001; // imm6 = 001xxx
3599 def v4i16 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3600 OpcodeStr, !strconcat(Dt, "32"),
3601 v4i16, v4i32, shr_imm16, OpNode> {
3602 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3604 def v2i32 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3605 OpcodeStr, !strconcat(Dt, "64"),
3606 v2i32, v2i64, shr_imm32, OpNode> {
3607 let Inst{21} = 0b1; // imm6 = 1xxxxx
3611 //===----------------------------------------------------------------------===//
3612 // Instruction Definitions.
3613 //===----------------------------------------------------------------------===//
3615 // Vector Add Operations.
3617 // VADD : Vector Add (integer and floating-point)
3618 defm VADD : N3V_QHSD<0, 0, 0b1000, 0, IIC_VBINiD, IIC_VBINiQ, "vadd", "i",
3620 def VADDfd : N3VD<0, 0, 0b00, 0b1101, 0, IIC_VBIND, "vadd", "f32",
3621 v2f32, v2f32, fadd, 1>;
3622 def VADDfq : N3VQ<0, 0, 0b00, 0b1101, 0, IIC_VBINQ, "vadd", "f32",
3623 v4f32, v4f32, fadd, 1>;
3624 // VADDL : Vector Add Long (Q = D + D)
3625 defm VADDLs : N3VLExt_QHS<0,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
3626 "vaddl", "s", add, sext, 1>;
3627 defm VADDLu : N3VLExt_QHS<1,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
3628 "vaddl", "u", add, zext, 1>;
3629 // VADDW : Vector Add Wide (Q = Q + D)
3630 defm VADDWs : N3VW_QHS<0,1,0b0001,0, "vaddw", "s", add, sext, 0>;
3631 defm VADDWu : N3VW_QHS<1,1,0b0001,0, "vaddw", "u", add, zext, 0>;
3632 // VHADD : Vector Halving Add
3633 defm VHADDs : N3VInt_QHS<0, 0, 0b0000, 0, N3RegFrm,
3634 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3635 "vhadd", "s", int_arm_neon_vhadds, 1>;
3636 defm VHADDu : N3VInt_QHS<1, 0, 0b0000, 0, N3RegFrm,
3637 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3638 "vhadd", "u", int_arm_neon_vhaddu, 1>;
3639 // VRHADD : Vector Rounding Halving Add
3640 defm VRHADDs : N3VInt_QHS<0, 0, 0b0001, 0, N3RegFrm,
3641 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3642 "vrhadd", "s", int_arm_neon_vrhadds, 1>;
3643 defm VRHADDu : N3VInt_QHS<1, 0, 0b0001, 0, N3RegFrm,
3644 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3645 "vrhadd", "u", int_arm_neon_vrhaddu, 1>;
3646 // VQADD : Vector Saturating Add
3647 defm VQADDs : N3VInt_QHSD<0, 0, 0b0000, 1, N3RegFrm,
3648 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3649 "vqadd", "s", int_arm_neon_vqadds, 1>;
3650 defm VQADDu : N3VInt_QHSD<1, 0, 0b0000, 1, N3RegFrm,
3651 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3652 "vqadd", "u", int_arm_neon_vqaddu, 1>;
3653 // VADDHN : Vector Add and Narrow Returning High Half (D = Q + Q)
3654 defm VADDHN : N3VNInt_HSD<0,1,0b0100,0, "vaddhn", "i",
3655 int_arm_neon_vaddhn, 1>;
3656 // VRADDHN : Vector Rounding Add and Narrow Returning High Half (D = Q + Q)
3657 defm VRADDHN : N3VNInt_HSD<1,1,0b0100,0, "vraddhn", "i",
3658 int_arm_neon_vraddhn, 1>;
3660 // Vector Multiply Operations.
3662 // VMUL : Vector Multiply (integer, polynomial and floating-point)
3663 defm VMUL : N3V_QHS<0, 0, 0b1001, 1, IIC_VMULi16D, IIC_VMULi32D,
3664 IIC_VMULi16Q, IIC_VMULi32Q, "vmul", "i", mul, 1>;
3665 def VMULpd : N3VDInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16D, "vmul",
3666 "p8", v8i8, v8i8, int_arm_neon_vmulp, 1>;
3667 def VMULpq : N3VQInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16Q, "vmul",
3668 "p8", v16i8, v16i8, int_arm_neon_vmulp, 1>;
3669 def VMULfd : N3VD<1, 0, 0b00, 0b1101, 1, IIC_VFMULD, "vmul", "f32",
3670 v2f32, v2f32, fmul, 1>;
3671 def VMULfq : N3VQ<1, 0, 0b00, 0b1101, 1, IIC_VFMULQ, "vmul", "f32",
3672 v4f32, v4f32, fmul, 1>;
3673 defm VMULsl : N3VSL_HS<0b1000, "vmul", mul>;
3674 def VMULslfd : N3VDSL<0b10, 0b1001, IIC_VBIND, "vmul", "f32", v2f32, fmul>;
3675 def VMULslfq : N3VQSL<0b10, 0b1001, IIC_VBINQ, "vmul", "f32", v4f32,
3678 def : Pat<(v8i16 (mul (v8i16 QPR:$src1),
3679 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
3680 (v8i16 (VMULslv8i16 (v8i16 QPR:$src1),
3681 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3682 (DSubReg_i16_reg imm:$lane))),
3683 (SubReg_i16_lane imm:$lane)))>;
3684 def : Pat<(v4i32 (mul (v4i32 QPR:$src1),
3685 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
3686 (v4i32 (VMULslv4i32 (v4i32 QPR:$src1),
3687 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3688 (DSubReg_i32_reg imm:$lane))),
3689 (SubReg_i32_lane imm:$lane)))>;
3690 def : Pat<(v4f32 (fmul (v4f32 QPR:$src1),
3691 (v4f32 (NEONvduplane (v4f32 QPR:$src2), imm:$lane)))),
3692 (v4f32 (VMULslfq (v4f32 QPR:$src1),
3693 (v2f32 (EXTRACT_SUBREG QPR:$src2,
3694 (DSubReg_i32_reg imm:$lane))),
3695 (SubReg_i32_lane imm:$lane)))>;
3697 // VQDMULH : Vector Saturating Doubling Multiply Returning High Half
3698 defm VQDMULH : N3VInt_HS<0, 0, 0b1011, 0, N3RegFrm, IIC_VMULi16D, IIC_VMULi32D,
3699 IIC_VMULi16Q, IIC_VMULi32Q,
3700 "vqdmulh", "s", int_arm_neon_vqdmulh, 1>;
3701 defm VQDMULHsl: N3VIntSL_HS<0b1100, IIC_VMULi16D, IIC_VMULi32D,
3702 IIC_VMULi16Q, IIC_VMULi32Q,
3703 "vqdmulh", "s", int_arm_neon_vqdmulh>;
3704 def : Pat<(v8i16 (int_arm_neon_vqdmulh (v8i16 QPR:$src1),
3705 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
3707 (v8i16 (VQDMULHslv8i16 (v8i16 QPR:$src1),
3708 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3709 (DSubReg_i16_reg imm:$lane))),
3710 (SubReg_i16_lane imm:$lane)))>;
3711 def : Pat<(v4i32 (int_arm_neon_vqdmulh (v4i32 QPR:$src1),
3712 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
3714 (v4i32 (VQDMULHslv4i32 (v4i32 QPR:$src1),
3715 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3716 (DSubReg_i32_reg imm:$lane))),
3717 (SubReg_i32_lane imm:$lane)))>;
3719 // VQRDMULH : Vector Rounding Saturating Doubling Multiply Returning High Half
3720 defm VQRDMULH : N3VInt_HS<1, 0, 0b1011, 0, N3RegFrm,
3721 IIC_VMULi16D,IIC_VMULi32D,IIC_VMULi16Q,IIC_VMULi32Q,
3722 "vqrdmulh", "s", int_arm_neon_vqrdmulh, 1>;
3723 defm VQRDMULHsl : N3VIntSL_HS<0b1101, IIC_VMULi16D, IIC_VMULi32D,
3724 IIC_VMULi16Q, IIC_VMULi32Q,
3725 "vqrdmulh", "s", int_arm_neon_vqrdmulh>;
3726 def : Pat<(v8i16 (int_arm_neon_vqrdmulh (v8i16 QPR:$src1),
3727 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
3729 (v8i16 (VQRDMULHslv8i16 (v8i16 QPR:$src1),
3730 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3731 (DSubReg_i16_reg imm:$lane))),
3732 (SubReg_i16_lane imm:$lane)))>;
3733 def : Pat<(v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$src1),
3734 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
3736 (v4i32 (VQRDMULHslv4i32 (v4i32 QPR:$src1),
3737 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3738 (DSubReg_i32_reg imm:$lane))),
3739 (SubReg_i32_lane imm:$lane)))>;
3741 // VMULL : Vector Multiply Long (integer and polynomial) (Q = D * D)
3742 defm VMULLs : N3VL_QHS<0,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
3743 "vmull", "s", NEONvmulls, 1>;
3744 defm VMULLu : N3VL_QHS<1,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
3745 "vmull", "u", NEONvmullu, 1>;
3746 def VMULLp : N3VLInt<0, 1, 0b00, 0b1110, 0, IIC_VMULi16D, "vmull", "p8",
3747 v8i16, v8i8, int_arm_neon_vmullp, 1>;
3748 defm VMULLsls : N3VLSL_HS<0, 0b1010, IIC_VMULi16D, "vmull", "s", NEONvmulls>;
3749 defm VMULLslu : N3VLSL_HS<1, 0b1010, IIC_VMULi16D, "vmull", "u", NEONvmullu>;
3751 // VQDMULL : Vector Saturating Doubling Multiply Long (Q = D * D)
3752 defm VQDMULL : N3VLInt_HS<0,1,0b1101,0, IIC_VMULi16D, IIC_VMULi32D,
3753 "vqdmull", "s", int_arm_neon_vqdmull, 1>;
3754 defm VQDMULLsl: N3VLIntSL_HS<0, 0b1011, IIC_VMULi16D,
3755 "vqdmull", "s", int_arm_neon_vqdmull>;
3757 // Vector Multiply-Accumulate and Multiply-Subtract Operations.
3759 // VMLA : Vector Multiply Accumulate (integer and floating-point)
3760 defm VMLA : N3VMulOp_QHS<0, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3761 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
3762 def VMLAfd : N3VDMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACD, "vmla", "f32",
3763 v2f32, fmul_su, fadd_mlx>,
3764 Requires<[HasNEON, UseFPVMLx]>;
3765 def VMLAfq : N3VQMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACQ, "vmla", "f32",
3766 v4f32, fmul_su, fadd_mlx>,
3767 Requires<[HasNEON, UseFPVMLx]>;
3768 defm VMLAsl : N3VMulOpSL_HS<0b0000, IIC_VMACi16D, IIC_VMACi32D,
3769 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
3770 def VMLAslfd : N3VDMulOpSL<0b10, 0b0001, IIC_VMACD, "vmla", "f32",
3771 v2f32, fmul_su, fadd_mlx>,
3772 Requires<[HasNEON, UseFPVMLx]>;
3773 def VMLAslfq : N3VQMulOpSL<0b10, 0b0001, IIC_VMACQ, "vmla", "f32",
3774 v4f32, v2f32, fmul_su, fadd_mlx>,
3775 Requires<[HasNEON, UseFPVMLx]>;
3777 def : Pat<(v8i16 (add (v8i16 QPR:$src1),
3778 (mul (v8i16 QPR:$src2),
3779 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
3780 (v8i16 (VMLAslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
3781 (v4i16 (EXTRACT_SUBREG QPR:$src3,
3782 (DSubReg_i16_reg imm:$lane))),
3783 (SubReg_i16_lane imm:$lane)))>;
3785 def : Pat<(v4i32 (add (v4i32 QPR:$src1),
3786 (mul (v4i32 QPR:$src2),
3787 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
3788 (v4i32 (VMLAslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
3789 (v2i32 (EXTRACT_SUBREG QPR:$src3,
3790 (DSubReg_i32_reg imm:$lane))),
3791 (SubReg_i32_lane imm:$lane)))>;
3793 def : Pat<(v4f32 (fadd_mlx (v4f32 QPR:$src1),
3794 (fmul_su (v4f32 QPR:$src2),
3795 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
3796 (v4f32 (VMLAslfq (v4f32 QPR:$src1),
3798 (v2f32 (EXTRACT_SUBREG QPR:$src3,
3799 (DSubReg_i32_reg imm:$lane))),
3800 (SubReg_i32_lane imm:$lane)))>,
3801 Requires<[HasNEON, UseFPVMLx]>;
3803 // VMLAL : Vector Multiply Accumulate Long (Q += D * D)
3804 defm VMLALs : N3VLMulOp_QHS<0,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
3805 "vmlal", "s", NEONvmulls, add>;
3806 defm VMLALu : N3VLMulOp_QHS<1,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
3807 "vmlal", "u", NEONvmullu, add>;
3809 defm VMLALsls : N3VLMulOpSL_HS<0, 0b0010, "vmlal", "s", NEONvmulls, add>;
3810 defm VMLALslu : N3VLMulOpSL_HS<1, 0b0010, "vmlal", "u", NEONvmullu, add>;
3812 // VQDMLAL : Vector Saturating Doubling Multiply Accumulate Long (Q += D * D)
3813 defm VQDMLAL : N3VLInt3_HS<0, 1, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3814 "vqdmlal", "s", int_arm_neon_vqdmlal>;
3815 defm VQDMLALsl: N3VLInt3SL_HS<0, 0b0011, "vqdmlal", "s", int_arm_neon_vqdmlal>;
3817 // VMLS : Vector Multiply Subtract (integer and floating-point)
3818 defm VMLS : N3VMulOp_QHS<1, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3819 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
3820 def VMLSfd : N3VDMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACD, "vmls", "f32",
3821 v2f32, fmul_su, fsub_mlx>,
3822 Requires<[HasNEON, UseFPVMLx]>;
3823 def VMLSfq : N3VQMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACQ, "vmls", "f32",
3824 v4f32, fmul_su, fsub_mlx>,
3825 Requires<[HasNEON, UseFPVMLx]>;
3826 defm VMLSsl : N3VMulOpSL_HS<0b0100, IIC_VMACi16D, IIC_VMACi32D,
3827 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
3828 def VMLSslfd : N3VDMulOpSL<0b10, 0b0101, IIC_VMACD, "vmls", "f32",
3829 v2f32, fmul_su, fsub_mlx>,
3830 Requires<[HasNEON, UseFPVMLx]>;
3831 def VMLSslfq : N3VQMulOpSL<0b10, 0b0101, IIC_VMACQ, "vmls", "f32",
3832 v4f32, v2f32, fmul_su, fsub_mlx>,
3833 Requires<[HasNEON, UseFPVMLx]>;
3835 def : Pat<(v8i16 (sub (v8i16 QPR:$src1),
3836 (mul (v8i16 QPR:$src2),
3837 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
3838 (v8i16 (VMLSslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
3839 (v4i16 (EXTRACT_SUBREG QPR:$src3,
3840 (DSubReg_i16_reg imm:$lane))),
3841 (SubReg_i16_lane imm:$lane)))>;
3843 def : Pat<(v4i32 (sub (v4i32 QPR:$src1),
3844 (mul (v4i32 QPR:$src2),
3845 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
3846 (v4i32 (VMLSslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
3847 (v2i32 (EXTRACT_SUBREG QPR:$src3,
3848 (DSubReg_i32_reg imm:$lane))),
3849 (SubReg_i32_lane imm:$lane)))>;
3851 def : Pat<(v4f32 (fsub_mlx (v4f32 QPR:$src1),
3852 (fmul_su (v4f32 QPR:$src2),
3853 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
3854 (v4f32 (VMLSslfq (v4f32 QPR:$src1), (v4f32 QPR:$src2),
3855 (v2f32 (EXTRACT_SUBREG QPR:$src3,
3856 (DSubReg_i32_reg imm:$lane))),
3857 (SubReg_i32_lane imm:$lane)))>,
3858 Requires<[HasNEON, UseFPVMLx]>;
3860 // VMLSL : Vector Multiply Subtract Long (Q -= D * D)
3861 defm VMLSLs : N3VLMulOp_QHS<0,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
3862 "vmlsl", "s", NEONvmulls, sub>;
3863 defm VMLSLu : N3VLMulOp_QHS<1,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
3864 "vmlsl", "u", NEONvmullu, sub>;
3866 defm VMLSLsls : N3VLMulOpSL_HS<0, 0b0110, "vmlsl", "s", NEONvmulls, sub>;
3867 defm VMLSLslu : N3VLMulOpSL_HS<1, 0b0110, "vmlsl", "u", NEONvmullu, sub>;
3869 // VQDMLSL : Vector Saturating Doubling Multiply Subtract Long (Q -= D * D)
3870 defm VQDMLSL : N3VLInt3_HS<0, 1, 0b1011, 0, IIC_VMACi16D, IIC_VMACi32D,
3871 "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
3872 defm VQDMLSLsl: N3VLInt3SL_HS<0, 0b111, "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
3874 // Vector Subtract Operations.
3876 // VSUB : Vector Subtract (integer and floating-point)
3877 defm VSUB : N3V_QHSD<1, 0, 0b1000, 0, IIC_VSUBiD, IIC_VSUBiQ,
3878 "vsub", "i", sub, 0>;
3879 def VSUBfd : N3VD<0, 0, 0b10, 0b1101, 0, IIC_VBIND, "vsub", "f32",
3880 v2f32, v2f32, fsub, 0>;
3881 def VSUBfq : N3VQ<0, 0, 0b10, 0b1101, 0, IIC_VBINQ, "vsub", "f32",
3882 v4f32, v4f32, fsub, 0>;
3883 // VSUBL : Vector Subtract Long (Q = D - D)
3884 defm VSUBLs : N3VLExt_QHS<0,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
3885 "vsubl", "s", sub, sext, 0>;
3886 defm VSUBLu : N3VLExt_QHS<1,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
3887 "vsubl", "u", sub, zext, 0>;
3888 // VSUBW : Vector Subtract Wide (Q = Q - D)
3889 defm VSUBWs : N3VW_QHS<0,1,0b0011,0, "vsubw", "s", sub, sext, 0>;
3890 defm VSUBWu : N3VW_QHS<1,1,0b0011,0, "vsubw", "u", sub, zext, 0>;
3891 // VHSUB : Vector Halving Subtract
3892 defm VHSUBs : N3VInt_QHS<0, 0, 0b0010, 0, N3RegFrm,
3893 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3894 "vhsub", "s", int_arm_neon_vhsubs, 0>;
3895 defm VHSUBu : N3VInt_QHS<1, 0, 0b0010, 0, N3RegFrm,
3896 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3897 "vhsub", "u", int_arm_neon_vhsubu, 0>;
3898 // VQSUB : Vector Saturing Subtract
3899 defm VQSUBs : N3VInt_QHSD<0, 0, 0b0010, 1, N3RegFrm,
3900 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3901 "vqsub", "s", int_arm_neon_vqsubs, 0>;
3902 defm VQSUBu : N3VInt_QHSD<1, 0, 0b0010, 1, N3RegFrm,
3903 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3904 "vqsub", "u", int_arm_neon_vqsubu, 0>;
3905 // VSUBHN : Vector Subtract and Narrow Returning High Half (D = Q - Q)
3906 defm VSUBHN : N3VNInt_HSD<0,1,0b0110,0, "vsubhn", "i",
3907 int_arm_neon_vsubhn, 0>;
3908 // VRSUBHN : Vector Rounding Subtract and Narrow Returning High Half (D=Q-Q)
3909 defm VRSUBHN : N3VNInt_HSD<1,1,0b0110,0, "vrsubhn", "i",
3910 int_arm_neon_vrsubhn, 0>;
3912 // Vector Comparisons.
3914 // VCEQ : Vector Compare Equal
3915 defm VCEQ : N3V_QHS<1, 0, 0b1000, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3916 IIC_VSUBi4Q, "vceq", "i", NEONvceq, 1>;
3917 def VCEQfd : N3VD<0,0,0b00,0b1110,0, IIC_VBIND, "vceq", "f32", v2i32, v2f32,
3919 def VCEQfq : N3VQ<0,0,0b00,0b1110,0, IIC_VBINQ, "vceq", "f32", v4i32, v4f32,
3922 defm VCEQz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00010, 0, "vceq", "i",
3923 "$Vd, $Vm, #0", NEONvceqz>;
3925 // VCGE : Vector Compare Greater Than or Equal
3926 defm VCGEs : N3V_QHS<0, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3927 IIC_VSUBi4Q, "vcge", "s", NEONvcge, 0>;
3928 defm VCGEu : N3V_QHS<1, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3929 IIC_VSUBi4Q, "vcge", "u", NEONvcgeu, 0>;
3930 def VCGEfd : N3VD<1,0,0b00,0b1110,0, IIC_VBIND, "vcge", "f32", v2i32, v2f32,
3932 def VCGEfq : N3VQ<1,0,0b00,0b1110,0, IIC_VBINQ, "vcge", "f32", v4i32, v4f32,
3935 defm VCGEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00001, 0, "vcge", "s",
3936 "$Vd, $Vm, #0", NEONvcgez>;
3937 defm VCLEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00011, 0, "vcle", "s",
3938 "$Vd, $Vm, #0", NEONvclez>;
3940 // VCGT : Vector Compare Greater Than
3941 defm VCGTs : N3V_QHS<0, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3942 IIC_VSUBi4Q, "vcgt", "s", NEONvcgt, 0>;
3943 defm VCGTu : N3V_QHS<1, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3944 IIC_VSUBi4Q, "vcgt", "u", NEONvcgtu, 0>;
3945 def VCGTfd : N3VD<1,0,0b10,0b1110,0, IIC_VBIND, "vcgt", "f32", v2i32, v2f32,
3947 def VCGTfq : N3VQ<1,0,0b10,0b1110,0, IIC_VBINQ, "vcgt", "f32", v4i32, v4f32,
3950 defm VCGTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00000, 0, "vcgt", "s",
3951 "$Vd, $Vm, #0", NEONvcgtz>;
3952 defm VCLTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00100, 0, "vclt", "s",
3953 "$Vd, $Vm, #0", NEONvcltz>;
3955 // VACGE : Vector Absolute Compare Greater Than or Equal (aka VCAGE)
3956 def VACGEd : N3VDInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacge",
3957 "f32", v2i32, v2f32, int_arm_neon_vacged, 0>;
3958 def VACGEq : N3VQInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacge",
3959 "f32", v4i32, v4f32, int_arm_neon_vacgeq, 0>;
3960 // VACGT : Vector Absolute Compare Greater Than (aka VCAGT)
3961 def VACGTd : N3VDInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacgt",
3962 "f32", v2i32, v2f32, int_arm_neon_vacgtd, 0>;
3963 def VACGTq : N3VQInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacgt",
3964 "f32", v4i32, v4f32, int_arm_neon_vacgtq, 0>;
3965 // VTST : Vector Test Bits
3966 defm VTST : N3V_QHS<0, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
3967 IIC_VBINi4Q, "vtst", "", NEONvtst, 1>;
3969 // Vector Bitwise Operations.
3971 def vnotd : PatFrag<(ops node:$in),
3972 (xor node:$in, (bitconvert (v8i8 NEONimmAllOnesV)))>;
3973 def vnotq : PatFrag<(ops node:$in),
3974 (xor node:$in, (bitconvert (v16i8 NEONimmAllOnesV)))>;
3977 // VAND : Vector Bitwise AND
3978 def VANDd : N3VDX<0, 0, 0b00, 0b0001, 1, IIC_VBINiD, "vand",
3979 v2i32, v2i32, and, 1>;
3980 def VANDq : N3VQX<0, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "vand",
3981 v4i32, v4i32, and, 1>;
3983 // VEOR : Vector Bitwise Exclusive OR
3984 def VEORd : N3VDX<1, 0, 0b00, 0b0001, 1, IIC_VBINiD, "veor",
3985 v2i32, v2i32, xor, 1>;
3986 def VEORq : N3VQX<1, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "veor",
3987 v4i32, v4i32, xor, 1>;
3989 // VORR : Vector Bitwise OR
3990 def VORRd : N3VDX<0, 0, 0b10, 0b0001, 1, IIC_VBINiD, "vorr",
3991 v2i32, v2i32, or, 1>;
3992 def VORRq : N3VQX<0, 0, 0b10, 0b0001, 1, IIC_VBINiQ, "vorr",
3993 v4i32, v4i32, or, 1>;
3995 def VORRiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 0, 1,
3996 (outs DPR:$Vd), (ins nImmSplatI16:$SIMM, DPR:$src),
3998 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
4000 (v4i16 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
4001 let Inst{9} = SIMM{9};
4004 def VORRiv2i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 0, 0, 1,
4005 (outs DPR:$Vd), (ins nImmSplatI32:$SIMM, DPR:$src),
4007 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
4009 (v2i32 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
4010 let Inst{10-9} = SIMM{10-9};
4013 def VORRiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 0, 1,
4014 (outs QPR:$Vd), (ins nImmSplatI16:$SIMM, QPR:$src),
4016 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
4018 (v8i16 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
4019 let Inst{9} = SIMM{9};
4022 def VORRiv4i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 1, 0, 1,
4023 (outs QPR:$Vd), (ins nImmSplatI32:$SIMM, QPR:$src),
4025 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
4027 (v4i32 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
4028 let Inst{10-9} = SIMM{10-9};
4032 // VBIC : Vector Bitwise Bit Clear (AND NOT)
4033 def VBICd : N3VX<0, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
4034 (ins DPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VBINiD,
4035 "vbic", "$Vd, $Vn, $Vm", "",
4036 [(set DPR:$Vd, (v2i32 (and DPR:$Vn,
4037 (vnotd DPR:$Vm))))]>;
4038 def VBICq : N3VX<0, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
4039 (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINiQ,
4040 "vbic", "$Vd, $Vn, $Vm", "",
4041 [(set QPR:$Vd, (v4i32 (and QPR:$Vn,
4042 (vnotq QPR:$Vm))))]>;
4044 def VBICiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 1, 1,
4045 (outs DPR:$Vd), (ins nImmSplatI16:$SIMM, DPR:$src),
4047 "vbic", "i16", "$Vd, $SIMM", "$src = $Vd",
4049 (v4i16 (NEONvbicImm DPR:$src, timm:$SIMM)))]> {
4050 let Inst{9} = SIMM{9};
4053 def VBICiv2i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 0, 1, 1,
4054 (outs DPR:$Vd), (ins nImmSplatI32:$SIMM, DPR:$src),
4056 "vbic", "i32", "$Vd, $SIMM", "$src = $Vd",
4058 (v2i32 (NEONvbicImm DPR:$src, timm:$SIMM)))]> {
4059 let Inst{10-9} = SIMM{10-9};
4062 def VBICiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 1, 1,
4063 (outs QPR:$Vd), (ins nImmSplatI16:$SIMM, QPR:$src),
4065 "vbic", "i16", "$Vd, $SIMM", "$src = $Vd",
4067 (v8i16 (NEONvbicImm QPR:$src, timm:$SIMM)))]> {
4068 let Inst{9} = SIMM{9};
4071 def VBICiv4i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 1, 1, 1,
4072 (outs QPR:$Vd), (ins nImmSplatI32:$SIMM, QPR:$src),
4074 "vbic", "i32", "$Vd, $SIMM", "$src = $Vd",
4076 (v4i32 (NEONvbicImm QPR:$src, timm:$SIMM)))]> {
4077 let Inst{10-9} = SIMM{10-9};
4080 // VORN : Vector Bitwise OR NOT
4081 def VORNd : N3VX<0, 0, 0b11, 0b0001, 0, 1, (outs DPR:$Vd),
4082 (ins DPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VBINiD,
4083 "vorn", "$Vd, $Vn, $Vm", "",
4084 [(set DPR:$Vd, (v2i32 (or DPR:$Vn,
4085 (vnotd DPR:$Vm))))]>;
4086 def VORNq : N3VX<0, 0, 0b11, 0b0001, 1, 1, (outs QPR:$Vd),
4087 (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINiQ,
4088 "vorn", "$Vd, $Vn, $Vm", "",
4089 [(set QPR:$Vd, (v4i32 (or QPR:$Vn,
4090 (vnotq QPR:$Vm))))]>;
4092 // VMVN : Vector Bitwise NOT (Immediate)
4094 let isReMaterializable = 1 in {
4096 def VMVNv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 1, 1, (outs DPR:$Vd),
4097 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4098 "vmvn", "i16", "$Vd, $SIMM", "",
4099 [(set DPR:$Vd, (v4i16 (NEONvmvnImm timm:$SIMM)))]> {
4100 let Inst{9} = SIMM{9};
4103 def VMVNv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 1, 1, (outs QPR:$Vd),
4104 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4105 "vmvn", "i16", "$Vd, $SIMM", "",
4106 [(set QPR:$Vd, (v8i16 (NEONvmvnImm timm:$SIMM)))]> {
4107 let Inst{9} = SIMM{9};
4110 def VMVNv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 1, 1, (outs DPR:$Vd),
4111 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4112 "vmvn", "i32", "$Vd, $SIMM", "",
4113 [(set DPR:$Vd, (v2i32 (NEONvmvnImm timm:$SIMM)))]> {
4114 let Inst{11-8} = SIMM{11-8};
4117 def VMVNv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 1, 1, (outs QPR:$Vd),
4118 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4119 "vmvn", "i32", "$Vd, $SIMM", "",
4120 [(set QPR:$Vd, (v4i32 (NEONvmvnImm timm:$SIMM)))]> {
4121 let Inst{11-8} = SIMM{11-8};
4125 // VMVN : Vector Bitwise NOT
4126 def VMVNd : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 0, 0,
4127 (outs DPR:$Vd), (ins DPR:$Vm), IIC_VSUBiD,
4128 "vmvn", "$Vd, $Vm", "",
4129 [(set DPR:$Vd, (v2i32 (vnotd DPR:$Vm)))]>;
4130 def VMVNq : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 1, 0,
4131 (outs QPR:$Vd), (ins QPR:$Vm), IIC_VSUBiD,
4132 "vmvn", "$Vd, $Vm", "",
4133 [(set QPR:$Vd, (v4i32 (vnotq QPR:$Vm)))]>;
4134 def : Pat<(v2i32 (vnotd DPR:$src)), (VMVNd DPR:$src)>;
4135 def : Pat<(v4i32 (vnotq QPR:$src)), (VMVNq QPR:$src)>;
4137 // VBSL : Vector Bitwise Select
4138 def VBSLd : N3VX<1, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
4139 (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
4140 N3RegFrm, IIC_VCNTiD,
4141 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4143 (v2i32 (NEONvbsl DPR:$src1, DPR:$Vn, DPR:$Vm)))]>;
4145 def : Pat<(v2i32 (or (and DPR:$Vn, DPR:$Vd),
4146 (and DPR:$Vm, (vnotd DPR:$Vd)))),
4147 (VBSLd DPR:$Vd, DPR:$Vn, DPR:$Vm)>;
4149 def VBSLq : N3VX<1, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
4150 (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
4151 N3RegFrm, IIC_VCNTiQ,
4152 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4154 (v4i32 (NEONvbsl QPR:$src1, QPR:$Vn, QPR:$Vm)))]>;
4156 def : Pat<(v4i32 (or (and QPR:$Vn, QPR:$Vd),
4157 (and QPR:$Vm, (vnotq QPR:$Vd)))),
4158 (VBSLq QPR:$Vd, QPR:$Vn, QPR:$Vm)>;
4160 // VBIF : Vector Bitwise Insert if False
4161 // like VBSL but with: "vbif $dst, $src3, $src1", "$src2 = $dst",
4162 // FIXME: This instruction's encoding MAY NOT BE correct.
4163 def VBIFd : N3VX<1, 0, 0b11, 0b0001, 0, 1,
4164 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
4165 N3RegFrm, IIC_VBINiD,
4166 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4168 def VBIFq : N3VX<1, 0, 0b11, 0b0001, 1, 1,
4169 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
4170 N3RegFrm, IIC_VBINiQ,
4171 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4174 // VBIT : Vector Bitwise Insert if True
4175 // like VBSL but with: "vbit $dst, $src2, $src1", "$src3 = $dst",
4176 // FIXME: This instruction's encoding MAY NOT BE correct.
4177 def VBITd : N3VX<1, 0, 0b10, 0b0001, 0, 1,
4178 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
4179 N3RegFrm, IIC_VBINiD,
4180 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4182 def VBITq : N3VX<1, 0, 0b10, 0b0001, 1, 1,
4183 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
4184 N3RegFrm, IIC_VBINiQ,
4185 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4188 // VBIT/VBIF are not yet implemented. The TwoAddress pass will not go looking
4189 // for equivalent operations with different register constraints; it just
4192 // Vector Absolute Differences.
4194 // VABD : Vector Absolute Difference
4195 defm VABDs : N3VInt_QHS<0, 0, 0b0111, 0, N3RegFrm,
4196 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4197 "vabd", "s", int_arm_neon_vabds, 1>;
4198 defm VABDu : N3VInt_QHS<1, 0, 0b0111, 0, N3RegFrm,
4199 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4200 "vabd", "u", int_arm_neon_vabdu, 1>;
4201 def VABDfd : N3VDInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBIND,
4202 "vabd", "f32", v2f32, v2f32, int_arm_neon_vabds, 1>;
4203 def VABDfq : N3VQInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBINQ,
4204 "vabd", "f32", v4f32, v4f32, int_arm_neon_vabds, 1>;
4206 // VABDL : Vector Absolute Difference Long (Q = | D - D |)
4207 defm VABDLs : N3VLIntExt_QHS<0,1,0b0111,0, IIC_VSUBi4Q,
4208 "vabdl", "s", int_arm_neon_vabds, zext, 1>;
4209 defm VABDLu : N3VLIntExt_QHS<1,1,0b0111,0, IIC_VSUBi4Q,
4210 "vabdl", "u", int_arm_neon_vabdu, zext, 1>;
4212 // VABA : Vector Absolute Difference and Accumulate
4213 defm VABAs : N3VIntOp_QHS<0,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
4214 "vaba", "s", int_arm_neon_vabds, add>;
4215 defm VABAu : N3VIntOp_QHS<1,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
4216 "vaba", "u", int_arm_neon_vabdu, add>;
4218 // VABAL : Vector Absolute Difference and Accumulate Long (Q += | D - D |)
4219 defm VABALs : N3VLIntExtOp_QHS<0,1,0b0101,0, IIC_VABAD,
4220 "vabal", "s", int_arm_neon_vabds, zext, add>;
4221 defm VABALu : N3VLIntExtOp_QHS<1,1,0b0101,0, IIC_VABAD,
4222 "vabal", "u", int_arm_neon_vabdu, zext, add>;
4224 // Vector Maximum and Minimum.
4226 // VMAX : Vector Maximum
4227 defm VMAXs : N3VInt_QHS<0, 0, 0b0110, 0, N3RegFrm,
4228 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4229 "vmax", "s", int_arm_neon_vmaxs, 1>;
4230 defm VMAXu : N3VInt_QHS<1, 0, 0b0110, 0, N3RegFrm,
4231 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4232 "vmax", "u", int_arm_neon_vmaxu, 1>;
4233 def VMAXfd : N3VDInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBIND,
4235 v2f32, v2f32, int_arm_neon_vmaxs, 1>;
4236 def VMAXfq : N3VQInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBINQ,
4238 v4f32, v4f32, int_arm_neon_vmaxs, 1>;
4240 // VMIN : Vector Minimum
4241 defm VMINs : N3VInt_QHS<0, 0, 0b0110, 1, N3RegFrm,
4242 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4243 "vmin", "s", int_arm_neon_vmins, 1>;
4244 defm VMINu : N3VInt_QHS<1, 0, 0b0110, 1, N3RegFrm,
4245 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4246 "vmin", "u", int_arm_neon_vminu, 1>;
4247 def VMINfd : N3VDInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBIND,
4249 v2f32, v2f32, int_arm_neon_vmins, 1>;
4250 def VMINfq : N3VQInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBINQ,
4252 v4f32, v4f32, int_arm_neon_vmins, 1>;
4254 // Vector Pairwise Operations.
4256 // VPADD : Vector Pairwise Add
4257 def VPADDi8 : N3VDInt<0, 0, 0b00, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4259 v8i8, v8i8, int_arm_neon_vpadd, 0>;
4260 def VPADDi16 : N3VDInt<0, 0, 0b01, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4262 v4i16, v4i16, int_arm_neon_vpadd, 0>;
4263 def VPADDi32 : N3VDInt<0, 0, 0b10, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4265 v2i32, v2i32, int_arm_neon_vpadd, 0>;
4266 def VPADDf : N3VDInt<1, 0, 0b00, 0b1101, 0, N3RegFrm,
4267 IIC_VPBIND, "vpadd", "f32",
4268 v2f32, v2f32, int_arm_neon_vpadd, 0>;
4270 // VPADDL : Vector Pairwise Add Long
4271 defm VPADDLs : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00100, 0, "vpaddl", "s",
4272 int_arm_neon_vpaddls>;
4273 defm VPADDLu : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00101, 0, "vpaddl", "u",
4274 int_arm_neon_vpaddlu>;
4276 // VPADAL : Vector Pairwise Add and Accumulate Long
4277 defm VPADALs : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01100, 0, "vpadal", "s",
4278 int_arm_neon_vpadals>;
4279 defm VPADALu : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01101, 0, "vpadal", "u",
4280 int_arm_neon_vpadalu>;
4282 // VPMAX : Vector Pairwise Maximum
4283 def VPMAXs8 : N3VDInt<0, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4284 "s8", v8i8, v8i8, int_arm_neon_vpmaxs, 0>;
4285 def VPMAXs16 : N3VDInt<0, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4286 "s16", v4i16, v4i16, int_arm_neon_vpmaxs, 0>;
4287 def VPMAXs32 : N3VDInt<0, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4288 "s32", v2i32, v2i32, int_arm_neon_vpmaxs, 0>;
4289 def VPMAXu8 : N3VDInt<1, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4290 "u8", v8i8, v8i8, int_arm_neon_vpmaxu, 0>;
4291 def VPMAXu16 : N3VDInt<1, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4292 "u16", v4i16, v4i16, int_arm_neon_vpmaxu, 0>;
4293 def VPMAXu32 : N3VDInt<1, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4294 "u32", v2i32, v2i32, int_arm_neon_vpmaxu, 0>;
4295 def VPMAXf : N3VDInt<1, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmax",
4296 "f32", v2f32, v2f32, int_arm_neon_vpmaxs, 0>;
4298 // VPMIN : Vector Pairwise Minimum
4299 def VPMINs8 : N3VDInt<0, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4300 "s8", v8i8, v8i8, int_arm_neon_vpmins, 0>;
4301 def VPMINs16 : N3VDInt<0, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4302 "s16", v4i16, v4i16, int_arm_neon_vpmins, 0>;
4303 def VPMINs32 : N3VDInt<0, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4304 "s32", v2i32, v2i32, int_arm_neon_vpmins, 0>;
4305 def VPMINu8 : N3VDInt<1, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4306 "u8", v8i8, v8i8, int_arm_neon_vpminu, 0>;
4307 def VPMINu16 : N3VDInt<1, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4308 "u16", v4i16, v4i16, int_arm_neon_vpminu, 0>;
4309 def VPMINu32 : N3VDInt<1, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4310 "u32", v2i32, v2i32, int_arm_neon_vpminu, 0>;
4311 def VPMINf : N3VDInt<1, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmin",
4312 "f32", v2f32, v2f32, int_arm_neon_vpmins, 0>;
4314 // Vector Reciprocal and Reciprocal Square Root Estimate and Step.
4316 // VRECPE : Vector Reciprocal Estimate
4317 def VRECPEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
4318 IIC_VUNAD, "vrecpe", "u32",
4319 v2i32, v2i32, int_arm_neon_vrecpe>;
4320 def VRECPEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
4321 IIC_VUNAQ, "vrecpe", "u32",
4322 v4i32, v4i32, int_arm_neon_vrecpe>;
4323 def VRECPEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
4324 IIC_VUNAD, "vrecpe", "f32",
4325 v2f32, v2f32, int_arm_neon_vrecpe>;
4326 def VRECPEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
4327 IIC_VUNAQ, "vrecpe", "f32",
4328 v4f32, v4f32, int_arm_neon_vrecpe>;
4330 // VRECPS : Vector Reciprocal Step
4331 def VRECPSfd : N3VDInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
4332 IIC_VRECSD, "vrecps", "f32",
4333 v2f32, v2f32, int_arm_neon_vrecps, 1>;
4334 def VRECPSfq : N3VQInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
4335 IIC_VRECSQ, "vrecps", "f32",
4336 v4f32, v4f32, int_arm_neon_vrecps, 1>;
4338 // VRSQRTE : Vector Reciprocal Square Root Estimate
4339 def VRSQRTEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
4340 IIC_VUNAD, "vrsqrte", "u32",
4341 v2i32, v2i32, int_arm_neon_vrsqrte>;
4342 def VRSQRTEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
4343 IIC_VUNAQ, "vrsqrte", "u32",
4344 v4i32, v4i32, int_arm_neon_vrsqrte>;
4345 def VRSQRTEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
4346 IIC_VUNAD, "vrsqrte", "f32",
4347 v2f32, v2f32, int_arm_neon_vrsqrte>;
4348 def VRSQRTEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
4349 IIC_VUNAQ, "vrsqrte", "f32",
4350 v4f32, v4f32, int_arm_neon_vrsqrte>;
4352 // VRSQRTS : Vector Reciprocal Square Root Step
4353 def VRSQRTSfd : N3VDInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
4354 IIC_VRECSD, "vrsqrts", "f32",
4355 v2f32, v2f32, int_arm_neon_vrsqrts, 1>;
4356 def VRSQRTSfq : N3VQInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
4357 IIC_VRECSQ, "vrsqrts", "f32",
4358 v4f32, v4f32, int_arm_neon_vrsqrts, 1>;
4362 // VSHL : Vector Shift
4363 defm VSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 0, N3RegVShFrm,
4364 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
4365 "vshl", "s", int_arm_neon_vshifts>;
4366 defm VSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 0, N3RegVShFrm,
4367 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
4368 "vshl", "u", int_arm_neon_vshiftu>;
4370 // VSHL : Vector Shift Left (Immediate)
4371 defm VSHLi : N2VShL_QHSD<0, 1, 0b0101, 1, IIC_VSHLiD, "vshl", "i", NEONvshl>;
4373 // VSHR : Vector Shift Right (Immediate)
4374 defm VSHRs : N2VShR_QHSD<0, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "s",NEONvshrs>;
4375 defm VSHRu : N2VShR_QHSD<1, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "u",NEONvshru>;
4377 // VSHLL : Vector Shift Left Long
4378 defm VSHLLs : N2VLSh_QHS<0, 1, 0b1010, 0, 0, 1, "vshll", "s", NEONvshlls>;
4379 defm VSHLLu : N2VLSh_QHS<1, 1, 0b1010, 0, 0, 1, "vshll", "u", NEONvshllu>;
4381 // VSHLL : Vector Shift Left Long (with maximum shift count)
4382 class N2VLShMax<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
4383 bit op6, bit op4, string OpcodeStr, string Dt, ValueType ResTy,
4384 ValueType OpTy, Operand ImmTy, SDNode OpNode>
4385 : N2VLSh<op24, op23, op11_8, op7, op6, op4, OpcodeStr, Dt,
4386 ResTy, OpTy, ImmTy, OpNode> {
4387 let Inst{21-16} = op21_16;
4388 let DecoderMethod = "DecodeVSHLMaxInstruction";
4390 def VSHLLi8 : N2VLShMax<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll", "i8",
4391 v8i16, v8i8, imm8, NEONvshlli>;
4392 def VSHLLi16 : N2VLShMax<1, 1, 0b110110, 0b0011, 0, 0, 0, "vshll", "i16",
4393 v4i32, v4i16, imm16, NEONvshlli>;
4394 def VSHLLi32 : N2VLShMax<1, 1, 0b111010, 0b0011, 0, 0, 0, "vshll", "i32",
4395 v2i64, v2i32, imm32, NEONvshlli>;
4397 // VSHRN : Vector Shift Right and Narrow
4398 defm VSHRN : N2VNSh_HSD<0,1,0b1000,0,0,1, IIC_VSHLiD, "vshrn", "i",
4401 // VRSHL : Vector Rounding Shift
4402 defm VRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 0, N3RegVShFrm,
4403 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4404 "vrshl", "s", int_arm_neon_vrshifts>;
4405 defm VRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 0, N3RegVShFrm,
4406 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4407 "vrshl", "u", int_arm_neon_vrshiftu>;
4408 // VRSHR : Vector Rounding Shift Right
4409 defm VRSHRs : N2VShR_QHSD<0,1,0b0010,1, IIC_VSHLi4D, "vrshr", "s",NEONvrshrs>;
4410 defm VRSHRu : N2VShR_QHSD<1,1,0b0010,1, IIC_VSHLi4D, "vrshr", "u",NEONvrshru>;
4412 // VRSHRN : Vector Rounding Shift Right and Narrow
4413 defm VRSHRN : N2VNSh_HSD<0, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vrshrn", "i",
4416 // VQSHL : Vector Saturating Shift
4417 defm VQSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 1, N3RegVShFrm,
4418 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4419 "vqshl", "s", int_arm_neon_vqshifts>;
4420 defm VQSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 1, N3RegVShFrm,
4421 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4422 "vqshl", "u", int_arm_neon_vqshiftu>;
4423 // VQSHL : Vector Saturating Shift Left (Immediate)
4424 defm VQSHLsi : N2VShL_QHSD<0,1,0b0111,1, IIC_VSHLi4D, "vqshl", "s",NEONvqshls>;
4425 defm VQSHLui : N2VShL_QHSD<1,1,0b0111,1, IIC_VSHLi4D, "vqshl", "u",NEONvqshlu>;
4427 // VQSHLU : Vector Saturating Shift Left (Immediate, Unsigned)
4428 defm VQSHLsu : N2VShL_QHSD<1,1,0b0110,1, IIC_VSHLi4D,"vqshlu","s",NEONvqshlsu>;
4430 // VQSHRN : Vector Saturating Shift Right and Narrow
4431 defm VQSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "s",
4433 defm VQSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "u",
4436 // VQSHRUN : Vector Saturating Shift Right and Narrow (Unsigned)
4437 defm VQSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 0, 1, IIC_VSHLi4D, "vqshrun", "s",
4440 // VQRSHL : Vector Saturating Rounding Shift
4441 defm VQRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 1, N3RegVShFrm,
4442 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4443 "vqrshl", "s", int_arm_neon_vqrshifts>;
4444 defm VQRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 1, N3RegVShFrm,
4445 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4446 "vqrshl", "u", int_arm_neon_vqrshiftu>;
4448 // VQRSHRN : Vector Saturating Rounding Shift Right and Narrow
4449 defm VQRSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "s",
4451 defm VQRSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "u",
4454 // VQRSHRUN : Vector Saturating Rounding Shift Right and Narrow (Unsigned)
4455 defm VQRSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vqrshrun", "s",
4458 // VSRA : Vector Shift Right and Accumulate
4459 defm VSRAs : N2VShAdd_QHSD<0, 1, 0b0001, 1, "vsra", "s", NEONvshrs>;
4460 defm VSRAu : N2VShAdd_QHSD<1, 1, 0b0001, 1, "vsra", "u", NEONvshru>;
4461 // VRSRA : Vector Rounding Shift Right and Accumulate
4462 defm VRSRAs : N2VShAdd_QHSD<0, 1, 0b0011, 1, "vrsra", "s", NEONvrshrs>;
4463 defm VRSRAu : N2VShAdd_QHSD<1, 1, 0b0011, 1, "vrsra", "u", NEONvrshru>;
4465 // VSLI : Vector Shift Left and Insert
4466 defm VSLI : N2VShInsL_QHSD<1, 1, 0b0101, 1, "vsli">;
4468 // VSRI : Vector Shift Right and Insert
4469 defm VSRI : N2VShInsR_QHSD<1, 1, 0b0100, 1, "vsri">;
4471 // Vector Absolute and Saturating Absolute.
4473 // VABS : Vector Absolute Value
4474 defm VABS : N2VInt_QHS<0b11, 0b11, 0b01, 0b00110, 0,
4475 IIC_VUNAiD, IIC_VUNAiQ, "vabs", "s",
4477 def VABSfd : N2VDInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
4478 IIC_VUNAD, "vabs", "f32",
4479 v2f32, v2f32, int_arm_neon_vabs>;
4480 def VABSfq : N2VQInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
4481 IIC_VUNAQ, "vabs", "f32",
4482 v4f32, v4f32, int_arm_neon_vabs>;
4484 // VQABS : Vector Saturating Absolute Value
4485 defm VQABS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01110, 0,
4486 IIC_VQUNAiD, IIC_VQUNAiQ, "vqabs", "s",
4487 int_arm_neon_vqabs>;
4491 def vnegd : PatFrag<(ops node:$in),
4492 (sub (bitconvert (v2i32 NEONimmAllZerosV)), node:$in)>;
4493 def vnegq : PatFrag<(ops node:$in),
4494 (sub (bitconvert (v4i32 NEONimmAllZerosV)), node:$in)>;
4496 class VNEGD<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
4497 : N2V<0b11, 0b11, size, 0b01, 0b00111, 0, 0, (outs DPR:$Vd), (ins DPR:$Vm),
4498 IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
4499 [(set DPR:$Vd, (Ty (vnegd DPR:$Vm)))]>;
4500 class VNEGQ<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
4501 : N2V<0b11, 0b11, size, 0b01, 0b00111, 1, 0, (outs QPR:$Vd), (ins QPR:$Vm),
4502 IIC_VSHLiQ, OpcodeStr, Dt, "$Vd, $Vm", "",
4503 [(set QPR:$Vd, (Ty (vnegq QPR:$Vm)))]>;
4505 // VNEG : Vector Negate (integer)
4506 def VNEGs8d : VNEGD<0b00, "vneg", "s8", v8i8>;
4507 def VNEGs16d : VNEGD<0b01, "vneg", "s16", v4i16>;
4508 def VNEGs32d : VNEGD<0b10, "vneg", "s32", v2i32>;
4509 def VNEGs8q : VNEGQ<0b00, "vneg", "s8", v16i8>;
4510 def VNEGs16q : VNEGQ<0b01, "vneg", "s16", v8i16>;
4511 def VNEGs32q : VNEGQ<0b10, "vneg", "s32", v4i32>;
4513 // VNEG : Vector Negate (floating-point)
4514 def VNEGfd : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
4515 (outs DPR:$Vd), (ins DPR:$Vm), IIC_VUNAD,
4516 "vneg", "f32", "$Vd, $Vm", "",
4517 [(set DPR:$Vd, (v2f32 (fneg DPR:$Vm)))]>;
4518 def VNEGf32q : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 1, 0,
4519 (outs QPR:$Vd), (ins QPR:$Vm), IIC_VUNAQ,
4520 "vneg", "f32", "$Vd, $Vm", "",
4521 [(set QPR:$Vd, (v4f32 (fneg QPR:$Vm)))]>;
4523 def : Pat<(v8i8 (vnegd DPR:$src)), (VNEGs8d DPR:$src)>;
4524 def : Pat<(v4i16 (vnegd DPR:$src)), (VNEGs16d DPR:$src)>;
4525 def : Pat<(v2i32 (vnegd DPR:$src)), (VNEGs32d DPR:$src)>;
4526 def : Pat<(v16i8 (vnegq QPR:$src)), (VNEGs8q QPR:$src)>;
4527 def : Pat<(v8i16 (vnegq QPR:$src)), (VNEGs16q QPR:$src)>;
4528 def : Pat<(v4i32 (vnegq QPR:$src)), (VNEGs32q QPR:$src)>;
4530 // VQNEG : Vector Saturating Negate
4531 defm VQNEG : N2VInt_QHS<0b11, 0b11, 0b00, 0b01111, 0,
4532 IIC_VQUNAiD, IIC_VQUNAiQ, "vqneg", "s",
4533 int_arm_neon_vqneg>;
4535 // Vector Bit Counting Operations.
4537 // VCLS : Vector Count Leading Sign Bits
4538 defm VCLS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01000, 0,
4539 IIC_VCNTiD, IIC_VCNTiQ, "vcls", "s",
4541 // VCLZ : Vector Count Leading Zeros
4542 defm VCLZ : N2VInt_QHS<0b11, 0b11, 0b00, 0b01001, 0,
4543 IIC_VCNTiD, IIC_VCNTiQ, "vclz", "i",
4545 // VCNT : Vector Count One Bits
4546 def VCNTd : N2VDInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
4547 IIC_VCNTiD, "vcnt", "8",
4548 v8i8, v8i8, int_arm_neon_vcnt>;
4549 def VCNTq : N2VQInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
4550 IIC_VCNTiQ, "vcnt", "8",
4551 v16i8, v16i8, int_arm_neon_vcnt>;
4554 def VSWPd : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 0, 0,
4555 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
4556 "vswp", "$Vd, $Vm", "", []>;
4557 def VSWPq : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 1, 0,
4558 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
4559 "vswp", "$Vd, $Vm", "", []>;
4561 // Vector Move Operations.
4563 // VMOV : Vector Move (Register)
4564 def : InstAlias<"vmov${p} $Vd, $Vm",
4565 (VORRd DPR:$Vd, DPR:$Vm, DPR:$Vm, pred:$p)>;
4566 def : InstAlias<"vmov${p} $Vd, $Vm",
4567 (VORRq QPR:$Vd, QPR:$Vm, QPR:$Vm, pred:$p)>;
4569 // VMOV : Vector Move (Immediate)
4571 let isReMaterializable = 1 in {
4572 def VMOVv8i8 : N1ModImm<1, 0b000, 0b1110, 0, 0, 0, 1, (outs DPR:$Vd),
4573 (ins nImmSplatI8:$SIMM), IIC_VMOVImm,
4574 "vmov", "i8", "$Vd, $SIMM", "",
4575 [(set DPR:$Vd, (v8i8 (NEONvmovImm timm:$SIMM)))]>;
4576 def VMOVv16i8 : N1ModImm<1, 0b000, 0b1110, 0, 1, 0, 1, (outs QPR:$Vd),
4577 (ins nImmSplatI8:$SIMM), IIC_VMOVImm,
4578 "vmov", "i8", "$Vd, $SIMM", "",
4579 [(set QPR:$Vd, (v16i8 (NEONvmovImm timm:$SIMM)))]>;
4581 def VMOVv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 0, 1, (outs DPR:$Vd),
4582 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4583 "vmov", "i16", "$Vd, $SIMM", "",
4584 [(set DPR:$Vd, (v4i16 (NEONvmovImm timm:$SIMM)))]> {
4585 let Inst{9} = SIMM{9};
4588 def VMOVv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 0, 1, (outs QPR:$Vd),
4589 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4590 "vmov", "i16", "$Vd, $SIMM", "",
4591 [(set QPR:$Vd, (v8i16 (NEONvmovImm timm:$SIMM)))]> {
4592 let Inst{9} = SIMM{9};
4595 def VMOVv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 0, 1, (outs DPR:$Vd),
4596 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4597 "vmov", "i32", "$Vd, $SIMM", "",
4598 [(set DPR:$Vd, (v2i32 (NEONvmovImm timm:$SIMM)))]> {
4599 let Inst{11-8} = SIMM{11-8};
4602 def VMOVv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 0, 1, (outs QPR:$Vd),
4603 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4604 "vmov", "i32", "$Vd, $SIMM", "",
4605 [(set QPR:$Vd, (v4i32 (NEONvmovImm timm:$SIMM)))]> {
4606 let Inst{11-8} = SIMM{11-8};
4609 def VMOVv1i64 : N1ModImm<1, 0b000, 0b1110, 0, 0, 1, 1, (outs DPR:$Vd),
4610 (ins nImmSplatI64:$SIMM), IIC_VMOVImm,
4611 "vmov", "i64", "$Vd, $SIMM", "",
4612 [(set DPR:$Vd, (v1i64 (NEONvmovImm timm:$SIMM)))]>;
4613 def VMOVv2i64 : N1ModImm<1, 0b000, 0b1110, 0, 1, 1, 1, (outs QPR:$Vd),
4614 (ins nImmSplatI64:$SIMM), IIC_VMOVImm,
4615 "vmov", "i64", "$Vd, $SIMM", "",
4616 [(set QPR:$Vd, (v2i64 (NEONvmovImm timm:$SIMM)))]>;
4618 def VMOVv2f32 : N1ModImm<1, 0b000, 0b1111, 0, 0, 0, 1, (outs DPR:$Vd),
4619 (ins nImmVMOVF32:$SIMM), IIC_VMOVImm,
4620 "vmov", "f32", "$Vd, $SIMM", "",
4621 [(set DPR:$Vd, (v2f32 (NEONvmovFPImm timm:$SIMM)))]>;
4622 def VMOVv4f32 : N1ModImm<1, 0b000, 0b1111, 0, 1, 0, 1, (outs QPR:$Vd),
4623 (ins nImmVMOVF32:$SIMM), IIC_VMOVImm,
4624 "vmov", "f32", "$Vd, $SIMM", "",
4625 [(set QPR:$Vd, (v4f32 (NEONvmovFPImm timm:$SIMM)))]>;
4626 } // isReMaterializable
4628 // VMOV : Vector Get Lane (move scalar to ARM core register)
4630 def VGETLNs8 : NVGetLane<{1,1,1,0,0,1,?,1}, 0b1011, {?,?},
4631 (outs GPR:$R), (ins DPR:$V, VectorIndex8:$lane),
4632 IIC_VMOVSI, "vmov", "s8", "$R, $V$lane",
4633 [(set GPR:$R, (NEONvgetlanes (v8i8 DPR:$V),
4635 let Inst{21} = lane{2};
4636 let Inst{6-5} = lane{1-0};
4638 def VGETLNs16 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, {?,1},
4639 (outs GPR:$R), (ins DPR:$V, VectorIndex16:$lane),
4640 IIC_VMOVSI, "vmov", "s16", "$R, $V$lane",
4641 [(set GPR:$R, (NEONvgetlanes (v4i16 DPR:$V),
4643 let Inst{21} = lane{1};
4644 let Inst{6} = lane{0};
4646 def VGETLNu8 : NVGetLane<{1,1,1,0,1,1,?,1}, 0b1011, {?,?},
4647 (outs GPR:$R), (ins DPR:$V, VectorIndex8:$lane),
4648 IIC_VMOVSI, "vmov", "u8", "$R, $V$lane",
4649 [(set GPR:$R, (NEONvgetlaneu (v8i8 DPR:$V),
4651 let Inst{21} = lane{2};
4652 let Inst{6-5} = lane{1-0};
4654 def VGETLNu16 : NVGetLane<{1,1,1,0,1,0,?,1}, 0b1011, {?,1},
4655 (outs GPR:$R), (ins DPR:$V, VectorIndex16:$lane),
4656 IIC_VMOVSI, "vmov", "u16", "$R, $V$lane",
4657 [(set GPR:$R, (NEONvgetlaneu (v4i16 DPR:$V),
4659 let Inst{21} = lane{1};
4660 let Inst{6} = lane{0};
4662 def VGETLNi32 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, 0b00,
4663 (outs GPR:$R), (ins DPR:$V, VectorIndex32:$lane),
4664 IIC_VMOVSI, "vmov", "32", "$R, $V$lane",
4665 [(set GPR:$R, (extractelt (v2i32 DPR:$V),
4667 let Inst{21} = lane{0};
4669 // def VGETLNf32: see FMRDH and FMRDL in ARMInstrVFP.td
4670 def : Pat<(NEONvgetlanes (v16i8 QPR:$src), imm:$lane),
4671 (VGETLNs8 (v8i8 (EXTRACT_SUBREG QPR:$src,
4672 (DSubReg_i8_reg imm:$lane))),
4673 (SubReg_i8_lane imm:$lane))>;
4674 def : Pat<(NEONvgetlanes (v8i16 QPR:$src), imm:$lane),
4675 (VGETLNs16 (v4i16 (EXTRACT_SUBREG QPR:$src,
4676 (DSubReg_i16_reg imm:$lane))),
4677 (SubReg_i16_lane imm:$lane))>;
4678 def : Pat<(NEONvgetlaneu (v16i8 QPR:$src), imm:$lane),
4679 (VGETLNu8 (v8i8 (EXTRACT_SUBREG QPR:$src,
4680 (DSubReg_i8_reg imm:$lane))),
4681 (SubReg_i8_lane imm:$lane))>;
4682 def : Pat<(NEONvgetlaneu (v8i16 QPR:$src), imm:$lane),
4683 (VGETLNu16 (v4i16 (EXTRACT_SUBREG QPR:$src,
4684 (DSubReg_i16_reg imm:$lane))),
4685 (SubReg_i16_lane imm:$lane))>;
4686 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
4687 (VGETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src,
4688 (DSubReg_i32_reg imm:$lane))),
4689 (SubReg_i32_lane imm:$lane))>;
4690 def : Pat<(extractelt (v2f32 DPR:$src1), imm:$src2),
4691 (EXTRACT_SUBREG (v2f32 (COPY_TO_REGCLASS (v2f32 DPR:$src1),DPR_VFP2)),
4692 (SSubReg_f32_reg imm:$src2))>;
4693 def : Pat<(extractelt (v4f32 QPR:$src1), imm:$src2),
4694 (EXTRACT_SUBREG (v4f32 (COPY_TO_REGCLASS (v4f32 QPR:$src1),QPR_VFP2)),
4695 (SSubReg_f32_reg imm:$src2))>;
4696 //def : Pat<(extractelt (v2i64 QPR:$src1), imm:$src2),
4697 // (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
4698 def : Pat<(extractelt (v2f64 QPR:$src1), imm:$src2),
4699 (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
4702 // VMOV : Vector Set Lane (move ARM core register to scalar)
4704 let Constraints = "$src1 = $V" in {
4705 def VSETLNi8 : NVSetLane<{1,1,1,0,0,1,?,0}, 0b1011, {?,?}, (outs DPR:$V),
4706 (ins DPR:$src1, GPR:$R, VectorIndex8:$lane),
4707 IIC_VMOVISL, "vmov", "8", "$V$lane, $R",
4708 [(set DPR:$V, (vector_insert (v8i8 DPR:$src1),
4709 GPR:$R, imm:$lane))]> {
4710 let Inst{21} = lane{2};
4711 let Inst{6-5} = lane{1-0};
4713 def VSETLNi16 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, {?,1}, (outs DPR:$V),
4714 (ins DPR:$src1, GPR:$R, VectorIndex16:$lane),
4715 IIC_VMOVISL, "vmov", "16", "$V$lane, $R",
4716 [(set DPR:$V, (vector_insert (v4i16 DPR:$src1),
4717 GPR:$R, imm:$lane))]> {
4718 let Inst{21} = lane{1};
4719 let Inst{6} = lane{0};
4721 def VSETLNi32 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, 0b00, (outs DPR:$V),
4722 (ins DPR:$src1, GPR:$R, VectorIndex32:$lane),
4723 IIC_VMOVISL, "vmov", "32", "$V$lane, $R",
4724 [(set DPR:$V, (insertelt (v2i32 DPR:$src1),
4725 GPR:$R, imm:$lane))]> {
4726 let Inst{21} = lane{0};
4729 def : Pat<(vector_insert (v16i8 QPR:$src1), GPR:$src2, imm:$lane),
4730 (v16i8 (INSERT_SUBREG QPR:$src1,
4731 (v8i8 (VSETLNi8 (v8i8 (EXTRACT_SUBREG QPR:$src1,
4732 (DSubReg_i8_reg imm:$lane))),
4733 GPR:$src2, (SubReg_i8_lane imm:$lane))),
4734 (DSubReg_i8_reg imm:$lane)))>;
4735 def : Pat<(vector_insert (v8i16 QPR:$src1), GPR:$src2, imm:$lane),
4736 (v8i16 (INSERT_SUBREG QPR:$src1,
4737 (v4i16 (VSETLNi16 (v4i16 (EXTRACT_SUBREG QPR:$src1,
4738 (DSubReg_i16_reg imm:$lane))),
4739 GPR:$src2, (SubReg_i16_lane imm:$lane))),
4740 (DSubReg_i16_reg imm:$lane)))>;
4741 def : Pat<(insertelt (v4i32 QPR:$src1), GPR:$src2, imm:$lane),
4742 (v4i32 (INSERT_SUBREG QPR:$src1,
4743 (v2i32 (VSETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src1,
4744 (DSubReg_i32_reg imm:$lane))),
4745 GPR:$src2, (SubReg_i32_lane imm:$lane))),
4746 (DSubReg_i32_reg imm:$lane)))>;
4748 def : Pat<(v2f32 (insertelt DPR:$src1, SPR:$src2, imm:$src3)),
4749 (INSERT_SUBREG (v2f32 (COPY_TO_REGCLASS DPR:$src1, DPR_VFP2)),
4750 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
4751 def : Pat<(v4f32 (insertelt QPR:$src1, SPR:$src2, imm:$src3)),
4752 (INSERT_SUBREG (v4f32 (COPY_TO_REGCLASS QPR:$src1, QPR_VFP2)),
4753 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
4755 //def : Pat<(v2i64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
4756 // (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
4757 def : Pat<(v2f64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
4758 (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
4760 def : Pat<(v2f32 (scalar_to_vector SPR:$src)),
4761 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
4762 def : Pat<(v2f64 (scalar_to_vector (f64 DPR:$src))),
4763 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
4764 def : Pat<(v4f32 (scalar_to_vector SPR:$src)),
4765 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
4767 def : Pat<(v8i8 (scalar_to_vector GPR:$src)),
4768 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4769 def : Pat<(v4i16 (scalar_to_vector GPR:$src)),
4770 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4771 def : Pat<(v2i32 (scalar_to_vector GPR:$src)),
4772 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4774 def : Pat<(v16i8 (scalar_to_vector GPR:$src)),
4775 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4776 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4778 def : Pat<(v8i16 (scalar_to_vector GPR:$src)),
4779 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
4780 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4782 def : Pat<(v4i32 (scalar_to_vector GPR:$src)),
4783 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
4784 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4787 // VDUP : Vector Duplicate (from ARM core register to all elements)
4789 class VDUPD<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
4790 : NVDup<opcod1, 0b1011, opcod3, (outs DPR:$V), (ins GPR:$R),
4791 IIC_VMOVIS, "vdup", Dt, "$V, $R",
4792 [(set DPR:$V, (Ty (NEONvdup (i32 GPR:$R))))]>;
4793 class VDUPQ<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
4794 : NVDup<opcod1, 0b1011, opcod3, (outs QPR:$V), (ins GPR:$R),
4795 IIC_VMOVIS, "vdup", Dt, "$V, $R",
4796 [(set QPR:$V, (Ty (NEONvdup (i32 GPR:$R))))]>;
4798 def VDUP8d : VDUPD<0b11101100, 0b00, "8", v8i8>;
4799 def VDUP16d : VDUPD<0b11101000, 0b01, "16", v4i16>;
4800 def VDUP32d : VDUPD<0b11101000, 0b00, "32", v2i32>;
4801 def VDUP8q : VDUPQ<0b11101110, 0b00, "8", v16i8>;
4802 def VDUP16q : VDUPQ<0b11101010, 0b01, "16", v8i16>;
4803 def VDUP32q : VDUPQ<0b11101010, 0b00, "32", v4i32>;
4805 def : Pat<(v2f32 (NEONvdup (f32 (bitconvert GPR:$R)))), (VDUP32d GPR:$R)>;
4806 def : Pat<(v4f32 (NEONvdup (f32 (bitconvert GPR:$R)))), (VDUP32q GPR:$R)>;
4808 // VDUP : Vector Duplicate Lane (from scalar to all elements)
4810 class VDUPLND<bits<4> op19_16, string OpcodeStr, string Dt,
4811 ValueType Ty, Operand IdxTy>
4812 : NVDupLane<op19_16, 0, (outs DPR:$Vd), (ins DPR:$Vm, IdxTy:$lane),
4813 IIC_VMOVD, OpcodeStr, Dt, "$Vd, $Vm$lane",
4814 [(set DPR:$Vd, (Ty (NEONvduplane (Ty DPR:$Vm), imm:$lane)))]>;
4816 class VDUPLNQ<bits<4> op19_16, string OpcodeStr, string Dt,
4817 ValueType ResTy, ValueType OpTy, Operand IdxTy>
4818 : NVDupLane<op19_16, 1, (outs QPR:$Vd), (ins DPR:$Vm, IdxTy:$lane),
4819 IIC_VMOVQ, OpcodeStr, Dt, "$Vd, $Vm$lane",
4820 [(set QPR:$Vd, (ResTy (NEONvduplane (OpTy DPR:$Vm),
4821 VectorIndex32:$lane)))]>;
4823 // Inst{19-16} is partially specified depending on the element size.
4825 def VDUPLN8d : VDUPLND<{?,?,?,1}, "vdup", "8", v8i8, VectorIndex8> {
4827 let Inst{19-17} = lane{2-0};
4829 def VDUPLN16d : VDUPLND<{?,?,1,0}, "vdup", "16", v4i16, VectorIndex16> {
4831 let Inst{19-18} = lane{1-0};
4833 def VDUPLN32d : VDUPLND<{?,1,0,0}, "vdup", "32", v2i32, VectorIndex32> {
4835 let Inst{19} = lane{0};
4837 def VDUPLN8q : VDUPLNQ<{?,?,?,1}, "vdup", "8", v16i8, v8i8, VectorIndex8> {
4839 let Inst{19-17} = lane{2-0};
4841 def VDUPLN16q : VDUPLNQ<{?,?,1,0}, "vdup", "16", v8i16, v4i16, VectorIndex16> {
4843 let Inst{19-18} = lane{1-0};
4845 def VDUPLN32q : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4i32, v2i32, VectorIndex32> {
4847 let Inst{19} = lane{0};
4850 def : Pat<(v2f32 (NEONvduplane (v2f32 DPR:$Vm), imm:$lane)),
4851 (VDUPLN32d DPR:$Vm, imm:$lane)>;
4853 def : Pat<(v4f32 (NEONvduplane (v2f32 DPR:$Vm), imm:$lane)),
4854 (VDUPLN32q DPR:$Vm, imm:$lane)>;
4856 def : Pat<(v16i8 (NEONvduplane (v16i8 QPR:$src), imm:$lane)),
4857 (v16i8 (VDUPLN8q (v8i8 (EXTRACT_SUBREG QPR:$src,
4858 (DSubReg_i8_reg imm:$lane))),
4859 (SubReg_i8_lane imm:$lane)))>;
4860 def : Pat<(v8i16 (NEONvduplane (v8i16 QPR:$src), imm:$lane)),
4861 (v8i16 (VDUPLN16q (v4i16 (EXTRACT_SUBREG QPR:$src,
4862 (DSubReg_i16_reg imm:$lane))),
4863 (SubReg_i16_lane imm:$lane)))>;
4864 def : Pat<(v4i32 (NEONvduplane (v4i32 QPR:$src), imm:$lane)),
4865 (v4i32 (VDUPLN32q (v2i32 (EXTRACT_SUBREG QPR:$src,
4866 (DSubReg_i32_reg imm:$lane))),
4867 (SubReg_i32_lane imm:$lane)))>;
4868 def : Pat<(v4f32 (NEONvduplane (v4f32 QPR:$src), imm:$lane)),
4869 (v4f32 (VDUPLN32q (v2f32 (EXTRACT_SUBREG QPR:$src,
4870 (DSubReg_i32_reg imm:$lane))),
4871 (SubReg_i32_lane imm:$lane)))>;
4873 def VDUPfdf : PseudoNeonI<(outs DPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
4874 [(set DPR:$dst, (v2f32 (NEONvdup (f32 SPR:$src))))]>;
4875 def VDUPfqf : PseudoNeonI<(outs QPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
4876 [(set QPR:$dst, (v4f32 (NEONvdup (f32 SPR:$src))))]>;
4878 // VMOVN : Vector Narrowing Move
4879 defm VMOVN : N2VN_HSD<0b11,0b11,0b10,0b00100,0,0, IIC_VMOVN,
4880 "vmovn", "i", trunc>;
4881 // VQMOVN : Vector Saturating Narrowing Move
4882 defm VQMOVNs : N2VNInt_HSD<0b11,0b11,0b10,0b00101,0,0, IIC_VQUNAiD,
4883 "vqmovn", "s", int_arm_neon_vqmovns>;
4884 defm VQMOVNu : N2VNInt_HSD<0b11,0b11,0b10,0b00101,1,0, IIC_VQUNAiD,
4885 "vqmovn", "u", int_arm_neon_vqmovnu>;
4886 defm VQMOVNsu : N2VNInt_HSD<0b11,0b11,0b10,0b00100,1,0, IIC_VQUNAiD,
4887 "vqmovun", "s", int_arm_neon_vqmovnsu>;
4888 // VMOVL : Vector Lengthening Move
4889 defm VMOVLs : N2VL_QHS<0b01,0b10100,0,1, "vmovl", "s", sext>;
4890 defm VMOVLu : N2VL_QHS<0b11,0b10100,0,1, "vmovl", "u", zext>;
4892 // Vector Conversions.
4894 // VCVT : Vector Convert Between Floating-Point and Integers
4895 def VCVTf2sd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
4896 v2i32, v2f32, fp_to_sint>;
4897 def VCVTf2ud : N2VD<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
4898 v2i32, v2f32, fp_to_uint>;
4899 def VCVTs2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
4900 v2f32, v2i32, sint_to_fp>;
4901 def VCVTu2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
4902 v2f32, v2i32, uint_to_fp>;
4904 def VCVTf2sq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
4905 v4i32, v4f32, fp_to_sint>;
4906 def VCVTf2uq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
4907 v4i32, v4f32, fp_to_uint>;
4908 def VCVTs2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
4909 v4f32, v4i32, sint_to_fp>;
4910 def VCVTu2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
4911 v4f32, v4i32, uint_to_fp>;
4913 // VCVT : Vector Convert Between Floating-Point and Fixed-Point.
4914 let DecoderMethod = "DecodeVCVTD" in {
4915 def VCVTf2xsd : N2VCvtD<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
4916 v2i32, v2f32, int_arm_neon_vcvtfp2fxs>;
4917 def VCVTf2xud : N2VCvtD<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
4918 v2i32, v2f32, int_arm_neon_vcvtfp2fxu>;
4919 def VCVTxs2fd : N2VCvtD<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
4920 v2f32, v2i32, int_arm_neon_vcvtfxs2fp>;
4921 def VCVTxu2fd : N2VCvtD<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
4922 v2f32, v2i32, int_arm_neon_vcvtfxu2fp>;
4925 let DecoderMethod = "DecodeVCVTQ" in {
4926 def VCVTf2xsq : N2VCvtQ<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
4927 v4i32, v4f32, int_arm_neon_vcvtfp2fxs>;
4928 def VCVTf2xuq : N2VCvtQ<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
4929 v4i32, v4f32, int_arm_neon_vcvtfp2fxu>;
4930 def VCVTxs2fq : N2VCvtQ<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
4931 v4f32, v4i32, int_arm_neon_vcvtfxs2fp>;
4932 def VCVTxu2fq : N2VCvtQ<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
4933 v4f32, v4i32, int_arm_neon_vcvtfxu2fp>;
4936 // VCVT : Vector Convert Between Half-Precision and Single-Precision.
4937 def VCVTf2h : N2VNInt<0b11, 0b11, 0b01, 0b10, 0b01100, 0, 0,
4938 IIC_VUNAQ, "vcvt", "f16.f32",
4939 v4i16, v4f32, int_arm_neon_vcvtfp2hf>,
4940 Requires<[HasNEON, HasFP16]>;
4941 def VCVTh2f : N2VLInt<0b11, 0b11, 0b01, 0b10, 0b01110, 0, 0,
4942 IIC_VUNAQ, "vcvt", "f32.f16",
4943 v4f32, v4i16, int_arm_neon_vcvthf2fp>,
4944 Requires<[HasNEON, HasFP16]>;
4948 // VREV64 : Vector Reverse elements within 64-bit doublewords
4950 class VREV64D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4951 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 0, 0, (outs DPR:$Vd),
4952 (ins DPR:$Vm), IIC_VMOVD,
4953 OpcodeStr, Dt, "$Vd, $Vm", "",
4954 [(set DPR:$Vd, (Ty (NEONvrev64 (Ty DPR:$Vm))))]>;
4955 class VREV64Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4956 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 1, 0, (outs QPR:$Vd),
4957 (ins QPR:$Vm), IIC_VMOVQ,
4958 OpcodeStr, Dt, "$Vd, $Vm", "",
4959 [(set QPR:$Vd, (Ty (NEONvrev64 (Ty QPR:$Vm))))]>;
4961 def VREV64d8 : VREV64D<0b00, "vrev64", "8", v8i8>;
4962 def VREV64d16 : VREV64D<0b01, "vrev64", "16", v4i16>;
4963 def VREV64d32 : VREV64D<0b10, "vrev64", "32", v2i32>;
4964 def : Pat<(v2f32 (NEONvrev64 (v2f32 DPR:$Vm))), (VREV64d32 DPR:$Vm)>;
4966 def VREV64q8 : VREV64Q<0b00, "vrev64", "8", v16i8>;
4967 def VREV64q16 : VREV64Q<0b01, "vrev64", "16", v8i16>;
4968 def VREV64q32 : VREV64Q<0b10, "vrev64", "32", v4i32>;
4969 def : Pat<(v4f32 (NEONvrev64 (v4f32 QPR:$Vm))), (VREV64q32 QPR:$Vm)>;
4971 // VREV32 : Vector Reverse elements within 32-bit words
4973 class VREV32D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4974 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 0, 0, (outs DPR:$Vd),
4975 (ins DPR:$Vm), IIC_VMOVD,
4976 OpcodeStr, Dt, "$Vd, $Vm", "",
4977 [(set DPR:$Vd, (Ty (NEONvrev32 (Ty DPR:$Vm))))]>;
4978 class VREV32Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4979 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 1, 0, (outs QPR:$Vd),
4980 (ins QPR:$Vm), IIC_VMOVQ,
4981 OpcodeStr, Dt, "$Vd, $Vm", "",
4982 [(set QPR:$Vd, (Ty (NEONvrev32 (Ty QPR:$Vm))))]>;
4984 def VREV32d8 : VREV32D<0b00, "vrev32", "8", v8i8>;
4985 def VREV32d16 : VREV32D<0b01, "vrev32", "16", v4i16>;
4987 def VREV32q8 : VREV32Q<0b00, "vrev32", "8", v16i8>;
4988 def VREV32q16 : VREV32Q<0b01, "vrev32", "16", v8i16>;
4990 // VREV16 : Vector Reverse elements within 16-bit halfwords
4992 class VREV16D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4993 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 0, 0, (outs DPR:$Vd),
4994 (ins DPR:$Vm), IIC_VMOVD,
4995 OpcodeStr, Dt, "$Vd, $Vm", "",
4996 [(set DPR:$Vd, (Ty (NEONvrev16 (Ty DPR:$Vm))))]>;
4997 class VREV16Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4998 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 1, 0, (outs QPR:$Vd),
4999 (ins QPR:$Vm), IIC_VMOVQ,
5000 OpcodeStr, Dt, "$Vd, $Vm", "",
5001 [(set QPR:$Vd, (Ty (NEONvrev16 (Ty QPR:$Vm))))]>;
5003 def VREV16d8 : VREV16D<0b00, "vrev16", "8", v8i8>;
5004 def VREV16q8 : VREV16Q<0b00, "vrev16", "8", v16i8>;
5006 // Other Vector Shuffles.
5008 // Aligned extractions: really just dropping registers
5010 class AlignedVEXTq<ValueType DestTy, ValueType SrcTy, SDNodeXForm LaneCVT>
5011 : Pat<(DestTy (vector_extract_subvec (SrcTy QPR:$src), (i32 imm:$start))),
5012 (EXTRACT_SUBREG (SrcTy QPR:$src), (LaneCVT imm:$start))>;
5014 def : AlignedVEXTq<v8i8, v16i8, DSubReg_i8_reg>;
5016 def : AlignedVEXTq<v4i16, v8i16, DSubReg_i16_reg>;
5018 def : AlignedVEXTq<v2i32, v4i32, DSubReg_i32_reg>;
5020 def : AlignedVEXTq<v1i64, v2i64, DSubReg_f64_reg>;
5022 def : AlignedVEXTq<v2f32, v4f32, DSubReg_i32_reg>;
5025 // VEXT : Vector Extract
5027 class VEXTd<string OpcodeStr, string Dt, ValueType Ty, Operand immTy>
5028 : N3V<0,1,0b11,{?,?,?,?},0,0, (outs DPR:$Vd),
5029 (ins DPR:$Vn, DPR:$Vm, immTy:$index), NVExtFrm,
5030 IIC_VEXTD, OpcodeStr, Dt, "$Vd, $Vn, $Vm, $index", "",
5031 [(set DPR:$Vd, (Ty (NEONvext (Ty DPR:$Vn),
5032 (Ty DPR:$Vm), imm:$index)))]> {
5034 let Inst{11-8} = index{3-0};
5037 class VEXTq<string OpcodeStr, string Dt, ValueType Ty, Operand immTy>
5038 : N3V<0,1,0b11,{?,?,?,?},1,0, (outs QPR:$Vd),
5039 (ins QPR:$Vn, QPR:$Vm, imm0_15:$index), NVExtFrm,
5040 IIC_VEXTQ, OpcodeStr, Dt, "$Vd, $Vn, $Vm, $index", "",
5041 [(set QPR:$Vd, (Ty (NEONvext (Ty QPR:$Vn),
5042 (Ty QPR:$Vm), imm:$index)))]> {
5044 let Inst{11-8} = index{3-0};
5047 def VEXTd8 : VEXTd<"vext", "8", v8i8, imm0_7> {
5048 let Inst{11-8} = index{3-0};
5050 def VEXTd16 : VEXTd<"vext", "16", v4i16, imm0_3> {
5051 let Inst{11-9} = index{2-0};
5054 def VEXTd32 : VEXTd<"vext", "32", v2i32, imm0_1> {
5055 let Inst{11-10} = index{1-0};
5056 let Inst{9-8} = 0b00;
5058 def : Pat<(v2f32 (NEONvext (v2f32 DPR:$Vn),
5061 (VEXTd32 DPR:$Vn, DPR:$Vm, imm:$index)>;
5063 def VEXTq8 : VEXTq<"vext", "8", v16i8, imm0_15> {
5064 let Inst{11-8} = index{3-0};
5066 def VEXTq16 : VEXTq<"vext", "16", v8i16, imm0_7> {
5067 let Inst{11-9} = index{2-0};
5070 def VEXTq32 : VEXTq<"vext", "32", v4i32, imm0_3> {
5071 let Inst{11-10} = index{1-0};
5072 let Inst{9-8} = 0b00;
5074 def VEXTq64 : VEXTq<"vext", "64", v2i64, imm0_1> {
5075 let Inst{11} = index{0};
5076 let Inst{10-8} = 0b000;
5078 def : Pat<(v4f32 (NEONvext (v4f32 QPR:$Vn),
5081 (VEXTq32 QPR:$Vn, QPR:$Vm, imm:$index)>;
5083 // VTRN : Vector Transpose
5085 def VTRNd8 : N2VDShuffle<0b00, 0b00001, "vtrn", "8">;
5086 def VTRNd16 : N2VDShuffle<0b01, 0b00001, "vtrn", "16">;
5087 def VTRNd32 : N2VDShuffle<0b10, 0b00001, "vtrn", "32">;
5089 def VTRNq8 : N2VQShuffle<0b00, 0b00001, IIC_VPERMQ, "vtrn", "8">;
5090 def VTRNq16 : N2VQShuffle<0b01, 0b00001, IIC_VPERMQ, "vtrn", "16">;
5091 def VTRNq32 : N2VQShuffle<0b10, 0b00001, IIC_VPERMQ, "vtrn", "32">;
5093 // VUZP : Vector Unzip (Deinterleave)
5095 def VUZPd8 : N2VDShuffle<0b00, 0b00010, "vuzp", "8">;
5096 def VUZPd16 : N2VDShuffle<0b01, 0b00010, "vuzp", "16">;
5097 def VUZPd32 : N2VDShuffle<0b10, 0b00010, "vuzp", "32">;
5099 def VUZPq8 : N2VQShuffle<0b00, 0b00010, IIC_VPERMQ3, "vuzp", "8">;
5100 def VUZPq16 : N2VQShuffle<0b01, 0b00010, IIC_VPERMQ3, "vuzp", "16">;
5101 def VUZPq32 : N2VQShuffle<0b10, 0b00010, IIC_VPERMQ3, "vuzp", "32">;
5103 // VZIP : Vector Zip (Interleave)
5105 def VZIPd8 : N2VDShuffle<0b00, 0b00011, "vzip", "8">;
5106 def VZIPd16 : N2VDShuffle<0b01, 0b00011, "vzip", "16">;
5107 def VZIPd32 : N2VDShuffle<0b10, 0b00011, "vzip", "32">;
5109 def VZIPq8 : N2VQShuffle<0b00, 0b00011, IIC_VPERMQ3, "vzip", "8">;
5110 def VZIPq16 : N2VQShuffle<0b01, 0b00011, IIC_VPERMQ3, "vzip", "16">;
5111 def VZIPq32 : N2VQShuffle<0b10, 0b00011, IIC_VPERMQ3, "vzip", "32">;
5113 // Vector Table Lookup and Table Extension.
5115 // VTBL : Vector Table Lookup
5116 let DecoderMethod = "DecodeTBLInstruction" in {
5118 : N3V<1,1,0b11,0b1000,0,0, (outs DPR:$Vd),
5119 (ins VecListOneD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTB1,
5120 "vtbl", "8", "$Vd, $Vn, $Vm", "",
5121 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbl1 VecListOneD:$Vn, DPR:$Vm)))]>;
5122 let hasExtraSrcRegAllocReq = 1 in {
5124 : N3V<1,1,0b11,0b1001,0,0, (outs DPR:$Vd),
5125 (ins DPR:$Vn, DPR:$tbl2, DPR:$Vm), NVTBLFrm, IIC_VTB2,
5126 "vtbl", "8", "$Vd, \\{$Vn, $tbl2\\}, $Vm", "", []>;
5128 : N3V<1,1,0b11,0b1010,0,0, (outs DPR:$Vd),
5129 (ins DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$Vm), NVTBLFrm, IIC_VTB3,
5130 "vtbl", "8", "$Vd, \\{$Vn, $tbl2, $tbl3\\}, $Vm", "", []>;
5132 : N3V<1,1,0b11,0b1011,0,0, (outs DPR:$Vd),
5133 (ins DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$Vm),
5135 "vtbl", "8", "$Vd, \\{$Vn, $tbl2, $tbl3, $tbl4\\}, $Vm", "", []>;
5136 } // hasExtraSrcRegAllocReq = 1
5139 : PseudoNeonI<(outs DPR:$dst), (ins QPR:$tbl, DPR:$src), IIC_VTB2, "", []>;
5141 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB3, "", []>;
5143 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB4, "", []>;
5145 // VTBX : Vector Table Extension
5147 : N3V<1,1,0b11,0b1000,1,0, (outs DPR:$Vd),
5148 (ins DPR:$orig, VecListOneD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTBX1,
5149 "vtbx", "8", "$Vd, $Vn, $Vm", "$orig = $Vd",
5150 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbx1
5151 DPR:$orig, VecListOneD:$Vn, DPR:$Vm)))]>;
5152 let hasExtraSrcRegAllocReq = 1 in {
5154 : N3V<1,1,0b11,0b1001,1,0, (outs DPR:$Vd),
5155 (ins DPR:$orig, DPR:$Vn, DPR:$tbl2, DPR:$Vm), NVTBLFrm, IIC_VTBX2,
5156 "vtbx", "8", "$Vd, \\{$Vn, $tbl2\\}, $Vm", "$orig = $Vd", []>;
5158 : N3V<1,1,0b11,0b1010,1,0, (outs DPR:$Vd),
5159 (ins DPR:$orig, DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$Vm),
5160 NVTBLFrm, IIC_VTBX3,
5161 "vtbx", "8", "$Vd, \\{$Vn, $tbl2, $tbl3\\}, $Vm",
5164 : N3V<1,1,0b11,0b1011,1,0, (outs DPR:$Vd), (ins DPR:$orig, DPR:$Vn,
5165 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$Vm), NVTBLFrm, IIC_VTBX4,
5166 "vtbx", "8", "$Vd, \\{$Vn, $tbl2, $tbl3, $tbl4\\}, $Vm",
5168 } // hasExtraSrcRegAllocReq = 1
5171 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QPR:$tbl, DPR:$src),
5172 IIC_VTBX2, "$orig = $dst", []>;
5174 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
5175 IIC_VTBX3, "$orig = $dst", []>;
5177 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
5178 IIC_VTBX4, "$orig = $dst", []>;
5179 } // DecoderMethod = "DecodeTBLInstruction"
5181 //===----------------------------------------------------------------------===//
5182 // NEON instructions for single-precision FP math
5183 //===----------------------------------------------------------------------===//
5185 class N2VSPat<SDNode OpNode, NeonI Inst>
5186 : NEONFPPat<(f32 (OpNode SPR:$a)),
5188 (v2f32 (COPY_TO_REGCLASS (Inst
5190 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5191 SPR:$a, ssub_0)), DPR_VFP2)), ssub_0)>;
5193 class N3VSPat<SDNode OpNode, NeonI Inst>
5194 : NEONFPPat<(f32 (OpNode SPR:$a, SPR:$b)),
5196 (v2f32 (COPY_TO_REGCLASS (Inst
5198 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5201 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5202 SPR:$b, ssub_0)), DPR_VFP2)), ssub_0)>;
5204 class N3VSMulOpPat<SDNode MulNode, SDNode OpNode, NeonI Inst>
5205 : NEONFPPat<(f32 (OpNode SPR:$acc, (f32 (MulNode SPR:$a, SPR:$b)))),
5207 (v2f32 (COPY_TO_REGCLASS (Inst
5209 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5212 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5215 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5216 SPR:$b, ssub_0)), DPR_VFP2)), ssub_0)>;
5218 def : N3VSPat<fadd, VADDfd>;
5219 def : N3VSPat<fsub, VSUBfd>;
5220 def : N3VSPat<fmul, VMULfd>;
5221 def : N3VSMulOpPat<fmul, fadd, VMLAfd>,
5222 Requires<[HasNEON, UseNEONForFP, UseFPVMLx]>;
5223 def : N3VSMulOpPat<fmul, fsub, VMLSfd>,
5224 Requires<[HasNEON, UseNEONForFP, UseFPVMLx]>;
5225 def : N2VSPat<fabs, VABSfd>;
5226 def : N2VSPat<fneg, VNEGfd>;
5227 def : N3VSPat<NEONfmax, VMAXfd>;
5228 def : N3VSPat<NEONfmin, VMINfd>;
5229 def : N2VSPat<arm_ftosi, VCVTf2sd>;
5230 def : N2VSPat<arm_ftoui, VCVTf2ud>;
5231 def : N2VSPat<arm_sitof, VCVTs2fd>;
5232 def : N2VSPat<arm_uitof, VCVTu2fd>;
5234 //===----------------------------------------------------------------------===//
5235 // Non-Instruction Patterns
5236 //===----------------------------------------------------------------------===//
5239 def : Pat<(v1i64 (bitconvert (v2i32 DPR:$src))), (v1i64 DPR:$src)>;
5240 def : Pat<(v1i64 (bitconvert (v4i16 DPR:$src))), (v1i64 DPR:$src)>;
5241 def : Pat<(v1i64 (bitconvert (v8i8 DPR:$src))), (v1i64 DPR:$src)>;
5242 def : Pat<(v1i64 (bitconvert (f64 DPR:$src))), (v1i64 DPR:$src)>;
5243 def : Pat<(v1i64 (bitconvert (v2f32 DPR:$src))), (v1i64 DPR:$src)>;
5244 def : Pat<(v2i32 (bitconvert (v1i64 DPR:$src))), (v2i32 DPR:$src)>;
5245 def : Pat<(v2i32 (bitconvert (v4i16 DPR:$src))), (v2i32 DPR:$src)>;
5246 def : Pat<(v2i32 (bitconvert (v8i8 DPR:$src))), (v2i32 DPR:$src)>;
5247 def : Pat<(v2i32 (bitconvert (f64 DPR:$src))), (v2i32 DPR:$src)>;
5248 def : Pat<(v2i32 (bitconvert (v2f32 DPR:$src))), (v2i32 DPR:$src)>;
5249 def : Pat<(v4i16 (bitconvert (v1i64 DPR:$src))), (v4i16 DPR:$src)>;
5250 def : Pat<(v4i16 (bitconvert (v2i32 DPR:$src))), (v4i16 DPR:$src)>;
5251 def : Pat<(v4i16 (bitconvert (v8i8 DPR:$src))), (v4i16 DPR:$src)>;
5252 def : Pat<(v4i16 (bitconvert (f64 DPR:$src))), (v4i16 DPR:$src)>;
5253 def : Pat<(v4i16 (bitconvert (v2f32 DPR:$src))), (v4i16 DPR:$src)>;
5254 def : Pat<(v8i8 (bitconvert (v1i64 DPR:$src))), (v8i8 DPR:$src)>;
5255 def : Pat<(v8i8 (bitconvert (v2i32 DPR:$src))), (v8i8 DPR:$src)>;
5256 def : Pat<(v8i8 (bitconvert (v4i16 DPR:$src))), (v8i8 DPR:$src)>;
5257 def : Pat<(v8i8 (bitconvert (f64 DPR:$src))), (v8i8 DPR:$src)>;
5258 def : Pat<(v8i8 (bitconvert (v2f32 DPR:$src))), (v8i8 DPR:$src)>;
5259 def : Pat<(f64 (bitconvert (v1i64 DPR:$src))), (f64 DPR:$src)>;
5260 def : Pat<(f64 (bitconvert (v2i32 DPR:$src))), (f64 DPR:$src)>;
5261 def : Pat<(f64 (bitconvert (v4i16 DPR:$src))), (f64 DPR:$src)>;
5262 def : Pat<(f64 (bitconvert (v8i8 DPR:$src))), (f64 DPR:$src)>;
5263 def : Pat<(f64 (bitconvert (v2f32 DPR:$src))), (f64 DPR:$src)>;
5264 def : Pat<(v2f32 (bitconvert (f64 DPR:$src))), (v2f32 DPR:$src)>;
5265 def : Pat<(v2f32 (bitconvert (v1i64 DPR:$src))), (v2f32 DPR:$src)>;
5266 def : Pat<(v2f32 (bitconvert (v2i32 DPR:$src))), (v2f32 DPR:$src)>;
5267 def : Pat<(v2f32 (bitconvert (v4i16 DPR:$src))), (v2f32 DPR:$src)>;
5268 def : Pat<(v2f32 (bitconvert (v8i8 DPR:$src))), (v2f32 DPR:$src)>;
5270 def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (v2i64 QPR:$src)>;
5271 def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (v2i64 QPR:$src)>;
5272 def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (v2i64 QPR:$src)>;
5273 def : Pat<(v2i64 (bitconvert (v2f64 QPR:$src))), (v2i64 QPR:$src)>;
5274 def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (v2i64 QPR:$src)>;
5275 def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (v4i32 QPR:$src)>;
5276 def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (v4i32 QPR:$src)>;
5277 def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (v4i32 QPR:$src)>;
5278 def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (v4i32 QPR:$src)>;
5279 def : Pat<(v4i32 (bitconvert (v4f32 QPR:$src))), (v4i32 QPR:$src)>;
5280 def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (v8i16 QPR:$src)>;
5281 def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (v8i16 QPR:$src)>;
5282 def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (v8i16 QPR:$src)>;
5283 def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (v8i16 QPR:$src)>;
5284 def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (v8i16 QPR:$src)>;
5285 def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (v16i8 QPR:$src)>;
5286 def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (v16i8 QPR:$src)>;
5287 def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (v16i8 QPR:$src)>;
5288 def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (v16i8 QPR:$src)>;
5289 def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (v16i8 QPR:$src)>;
5290 def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (v4f32 QPR:$src)>;
5291 def : Pat<(v4f32 (bitconvert (v4i32 QPR:$src))), (v4f32 QPR:$src)>;
5292 def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (v4f32 QPR:$src)>;
5293 def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (v4f32 QPR:$src)>;
5294 def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (v4f32 QPR:$src)>;
5295 def : Pat<(v2f64 (bitconvert (v2i64 QPR:$src))), (v2f64 QPR:$src)>;
5296 def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (v2f64 QPR:$src)>;
5297 def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 QPR:$src)>;
5298 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 QPR:$src)>;
5299 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 QPR:$src)>;
5302 //===----------------------------------------------------------------------===//
5303 // Assembler aliases
5306 def : VFP2InstAlias<"fmdhr${p} $Dd, $Rn",
5307 (VSETLNi32 DPR:$Dd, GPR:$Rn, 1, pred:$p)>;
5308 def : VFP2InstAlias<"fmdlr${p} $Dd, $Rn",
5309 (VSETLNi32 DPR:$Dd, GPR:$Rn, 0, pred:$p)>;
5312 // VADD two-operand aliases.
5313 def : NEONInstAlias<"vadd${p}.i8 $Vdn, $Vm",
5314 (VADDv16i8 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5315 def : NEONInstAlias<"vadd${p}.i16 $Vdn, $Vm",
5316 (VADDv8i16 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5317 def : NEONInstAlias<"vadd${p}.i32 $Vdn, $Vm",
5318 (VADDv4i32 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5319 def : NEONInstAlias<"vadd${p}.i64 $Vdn, $Vm",
5320 (VADDv2i64 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5322 def : NEONInstAlias<"vadd${p}.i8 $Vdn, $Vm",
5323 (VADDv8i8 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5324 def : NEONInstAlias<"vadd${p}.i16 $Vdn, $Vm",
5325 (VADDv4i16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5326 def : NEONInstAlias<"vadd${p}.i32 $Vdn, $Vm",
5327 (VADDv2i32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5328 def : NEONInstAlias<"vadd${p}.i64 $Vdn, $Vm",
5329 (VADDv1i64 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5331 def : NEONInstAlias<"vadd${p}.f32 $Vdn, $Vm",
5332 (VADDfd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5333 def : NEONInstAlias<"vadd${p}.f32 $Vdn, $Vm",
5334 (VADDfq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5336 // VSUB two-operand aliases.
5337 def : NEONInstAlias<"vsub${p}.i8 $Vdn, $Vm",
5338 (VSUBv16i8 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5339 def : NEONInstAlias<"vsub${p}.i16 $Vdn, $Vm",
5340 (VSUBv8i16 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5341 def : NEONInstAlias<"vsub${p}.i32 $Vdn, $Vm",
5342 (VSUBv4i32 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5343 def : NEONInstAlias<"vsub${p}.i64 $Vdn, $Vm",
5344 (VSUBv2i64 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5346 def : NEONInstAlias<"vsub${p}.i8 $Vdn, $Vm",
5347 (VSUBv8i8 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5348 def : NEONInstAlias<"vsub${p}.i16 $Vdn, $Vm",
5349 (VSUBv4i16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5350 def : NEONInstAlias<"vsub${p}.i32 $Vdn, $Vm",
5351 (VSUBv2i32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5352 def : NEONInstAlias<"vsub${p}.i64 $Vdn, $Vm",
5353 (VSUBv1i64 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5355 def : NEONInstAlias<"vsub${p}.f32 $Vdn, $Vm",
5356 (VSUBfd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5357 def : NEONInstAlias<"vsub${p}.f32 $Vdn, $Vm",
5358 (VSUBfq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5360 // VADDW two-operand aliases.
5361 def : NEONInstAlias<"vaddw${p}.s8 $Vdn, $Vm",
5362 (VADDWsv8i16 QPR:$Vdn, QPR:$Vdn, DPR:$Vm, pred:$p)>;
5363 def : NEONInstAlias<"vaddw${p}.s16 $Vdn, $Vm",
5364 (VADDWsv4i32 QPR:$Vdn, QPR:$Vdn, DPR:$Vm, pred:$p)>;
5365 def : NEONInstAlias<"vaddw${p}.s32 $Vdn, $Vm",
5366 (VADDWsv2i64 QPR:$Vdn, QPR:$Vdn, DPR:$Vm, pred:$p)>;
5367 def : NEONInstAlias<"vaddw${p}.u8 $Vdn, $Vm",
5368 (VADDWuv8i16 QPR:$Vdn, QPR:$Vdn, DPR:$Vm, pred:$p)>;
5369 def : NEONInstAlias<"vaddw${p}.u16 $Vdn, $Vm",
5370 (VADDWuv4i32 QPR:$Vdn, QPR:$Vdn, DPR:$Vm, pred:$p)>;
5371 def : NEONInstAlias<"vaddw${p}.u32 $Vdn, $Vm",
5372 (VADDWuv2i64 QPR:$Vdn, QPR:$Vdn, DPR:$Vm, pred:$p)>;
5374 // VAND/VBIC/VEOR/VORR accept but do not require a type suffix.
5375 defm : VFPDTAnyInstAlias<"vand${p}", "$Vd, $Vn, $Vm",
5376 (VANDd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
5377 defm : VFPDTAnyInstAlias<"vand${p}", "$Vd, $Vn, $Vm",
5378 (VANDq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
5379 defm : VFPDTAnyInstAlias<"vbic${p}", "$Vd, $Vn, $Vm",
5380 (VBICd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
5381 defm : VFPDTAnyInstAlias<"vbic${p}", "$Vd, $Vn, $Vm",
5382 (VBICq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
5383 defm : VFPDTAnyInstAlias<"veor${p}", "$Vd, $Vn, $Vm",
5384 (VEORd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
5385 defm : VFPDTAnyInstAlias<"veor${p}", "$Vd, $Vn, $Vm",
5386 (VEORq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
5387 defm : VFPDTAnyInstAlias<"vorr${p}", "$Vd, $Vn, $Vm",
5388 (VORRd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
5389 defm : VFPDTAnyInstAlias<"vorr${p}", "$Vd, $Vn, $Vm",
5390 (VORRq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
5391 // ... two-operand aliases
5392 def : NEONInstAlias<"vand${p} $Vdn, $Vm",
5393 (VANDd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5394 def : NEONInstAlias<"vand${p} $Vdn, $Vm",
5395 (VANDq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5396 def : NEONInstAlias<"vbic${p} $Vdn, $Vm",
5397 (VBICd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5398 def : NEONInstAlias<"vbic${p} $Vdn, $Vm",
5399 (VBICq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5400 def : NEONInstAlias<"veor${p} $Vdn, $Vm",
5401 (VEORd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5402 def : NEONInstAlias<"veor${p} $Vdn, $Vm",
5403 (VEORq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5404 def : NEONInstAlias<"vorr${p} $Vdn, $Vm",
5405 (VORRd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5406 def : NEONInstAlias<"vorr${p} $Vdn, $Vm",
5407 (VORRq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5409 defm : VFPDTAnyInstAlias<"vand${p}", "$Vdn, $Vm",
5410 (VANDd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5411 defm : VFPDTAnyInstAlias<"vand${p}", "$Vdn, $Vm",
5412 (VANDq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5413 defm : VFPDTAnyInstAlias<"veor${p}", "$Vdn, $Vm",
5414 (VEORd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5415 defm : VFPDTAnyInstAlias<"veor${p}", "$Vdn, $Vm",
5416 (VEORq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5417 defm : VFPDTAnyInstAlias<"vorr${p}", "$Vdn, $Vm",
5418 (VORRd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5419 defm : VFPDTAnyInstAlias<"vorr${p}", "$Vdn, $Vm",
5420 (VORRq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5422 // VMUL two-operand aliases.
5423 def : NEONInstAlias<"vmul${p}.p8 $Qdn, $Qm",
5424 (VMULpq QPR:$Qdn, QPR:$Qdn, QPR:$Qm, pred:$p)>;
5425 def : NEONInstAlias<"vmul${p}.i8 $Qdn, $Qm",
5426 (VMULv16i8 QPR:$Qdn, QPR:$Qdn, QPR:$Qm, pred:$p)>;
5427 def : NEONInstAlias<"vmul${p}.i16 $Qdn, $Qm",
5428 (VMULv8i16 QPR:$Qdn, QPR:$Qdn, QPR:$Qm, pred:$p)>;
5429 def : NEONInstAlias<"vmul${p}.i32 $Qdn, $Qm",
5430 (VMULv4i32 QPR:$Qdn, QPR:$Qdn, QPR:$Qm, pred:$p)>;
5432 def : NEONInstAlias<"vmul${p}.p8 $Ddn, $Dm",
5433 (VMULpd DPR:$Ddn, DPR:$Ddn, DPR:$Dm, pred:$p)>;
5434 def : NEONInstAlias<"vmul${p}.i8 $Ddn, $Dm",
5435 (VMULv8i8 DPR:$Ddn, DPR:$Ddn, DPR:$Dm, pred:$p)>;
5436 def : NEONInstAlias<"vmul${p}.i16 $Ddn, $Dm",
5437 (VMULv4i16 DPR:$Ddn, DPR:$Ddn, DPR:$Dm, pred:$p)>;
5438 def : NEONInstAlias<"vmul${p}.i32 $Ddn, $Dm",
5439 (VMULv2i32 DPR:$Ddn, DPR:$Ddn, DPR:$Dm, pred:$p)>;
5441 def : NEONInstAlias<"vmul${p}.f32 $Qdn, $Qm",
5442 (VMULfq QPR:$Qdn, QPR:$Qdn, QPR:$Qm, pred:$p)>;
5443 def : NEONInstAlias<"vmul${p}.f32 $Ddn, $Dm",
5444 (VMULfd DPR:$Ddn, DPR:$Ddn, DPR:$Dm, pred:$p)>;
5446 def : NEONInstAlias<"vmul${p}.i16 $Ddn, $Dm$lane",
5447 (VMULslv4i16 DPR:$Ddn, DPR:$Ddn, DPR_8:$Dm,
5448 VectorIndex16:$lane, pred:$p)>;
5449 def : NEONInstAlias<"vmul${p}.i16 $Qdn, $Dm$lane",
5450 (VMULslv8i16 QPR:$Qdn, QPR:$Qdn, DPR_8:$Dm,
5451 VectorIndex16:$lane, pred:$p)>;
5453 def : NEONInstAlias<"vmul${p}.i32 $Ddn, $Dm$lane",
5454 (VMULslv2i32 DPR:$Ddn, DPR:$Ddn, DPR_VFP2:$Dm,
5455 VectorIndex32:$lane, pred:$p)>;
5456 def : NEONInstAlias<"vmul${p}.i32 $Qdn, $Dm$lane",
5457 (VMULslv4i32 QPR:$Qdn, QPR:$Qdn, DPR_VFP2:$Dm,
5458 VectorIndex32:$lane, pred:$p)>;
5460 def : NEONInstAlias<"vmul${p}.f32 $Ddn, $Dm$lane",
5461 (VMULslfd DPR:$Ddn, DPR:$Ddn, DPR_VFP2:$Dm,
5462 VectorIndex32:$lane, pred:$p)>;
5463 def : NEONInstAlias<"vmul${p}.f32 $Qdn, $Dm$lane",
5464 (VMULslfq QPR:$Qdn, QPR:$Qdn, DPR_VFP2:$Dm,
5465 VectorIndex32:$lane, pred:$p)>;
5467 // VQADD (register) two-operand aliases.
5468 def : NEONInstAlias<"vqadd${p}.s8 $Vdn, $Vm",
5469 (VQADDsv8i8 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5470 def : NEONInstAlias<"vqadd${p}.s16 $Vdn, $Vm",
5471 (VQADDsv4i16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5472 def : NEONInstAlias<"vqadd${p}.s32 $Vdn, $Vm",
5473 (VQADDsv2i32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5474 def : NEONInstAlias<"vqadd${p}.s64 $Vdn, $Vm",
5475 (VQADDsv1i64 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5476 def : NEONInstAlias<"vqadd${p}.u8 $Vdn, $Vm",
5477 (VQADDuv8i8 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5478 def : NEONInstAlias<"vqadd${p}.u16 $Vdn, $Vm",
5479 (VQADDuv4i16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5480 def : NEONInstAlias<"vqadd${p}.u32 $Vdn, $Vm",
5481 (VQADDuv2i32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5482 def : NEONInstAlias<"vqadd${p}.u64 $Vdn, $Vm",
5483 (VQADDuv1i64 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5485 def : NEONInstAlias<"vqadd${p}.s8 $Vdn, $Vm",
5486 (VQADDsv16i8 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5487 def : NEONInstAlias<"vqadd${p}.s16 $Vdn, $Vm",
5488 (VQADDsv8i16 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5489 def : NEONInstAlias<"vqadd${p}.s32 $Vdn, $Vm",
5490 (VQADDsv4i32 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5491 def : NEONInstAlias<"vqadd${p}.s64 $Vdn, $Vm",
5492 (VQADDsv2i64 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5493 def : NEONInstAlias<"vqadd${p}.u8 $Vdn, $Vm",
5494 (VQADDuv16i8 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5495 def : NEONInstAlias<"vqadd${p}.u16 $Vdn, $Vm",
5496 (VQADDuv8i16 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5497 def : NEONInstAlias<"vqadd${p}.u32 $Vdn, $Vm",
5498 (VQADDuv4i32 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5499 def : NEONInstAlias<"vqadd${p}.u64 $Vdn, $Vm",
5500 (VQADDuv2i64 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5502 // VSHL (immediate) two-operand aliases.
5503 def : NEONInstAlias<"vshl${p}.i8 $Vdn, $imm",
5504 (VSHLiv8i8 DPR:$Vdn, DPR:$Vdn, imm0_7:$imm, pred:$p)>;
5505 def : NEONInstAlias<"vshl${p}.i16 $Vdn, $imm",
5506 (VSHLiv4i16 DPR:$Vdn, DPR:$Vdn, imm0_15:$imm, pred:$p)>;
5507 def : NEONInstAlias<"vshl${p}.i32 $Vdn, $imm",
5508 (VSHLiv2i32 DPR:$Vdn, DPR:$Vdn, imm0_31:$imm, pred:$p)>;
5509 def : NEONInstAlias<"vshl${p}.i64 $Vdn, $imm",
5510 (VSHLiv1i64 DPR:$Vdn, DPR:$Vdn, imm0_63:$imm, pred:$p)>;
5512 def : NEONInstAlias<"vshl${p}.i8 $Vdn, $imm",
5513 (VSHLiv16i8 QPR:$Vdn, QPR:$Vdn, imm0_7:$imm, pred:$p)>;
5514 def : NEONInstAlias<"vshl${p}.i16 $Vdn, $imm",
5515 (VSHLiv8i16 QPR:$Vdn, QPR:$Vdn, imm0_15:$imm, pred:$p)>;
5516 def : NEONInstAlias<"vshl${p}.i32 $Vdn, $imm",
5517 (VSHLiv4i32 QPR:$Vdn, QPR:$Vdn, imm0_31:$imm, pred:$p)>;
5518 def : NEONInstAlias<"vshl${p}.i64 $Vdn, $imm",
5519 (VSHLiv2i64 QPR:$Vdn, QPR:$Vdn, imm0_63:$imm, pred:$p)>;
5521 // VSHL (register) two-operand aliases.
5522 def : NEONInstAlias<"vshl${p}.s8 $Vdn, $Vm",
5523 (VSHLsv8i8 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5524 def : NEONInstAlias<"vshl${p}.s16 $Vdn, $Vm",
5525 (VSHLsv4i16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5526 def : NEONInstAlias<"vshl${p}.s32 $Vdn, $Vm",
5527 (VSHLsv2i32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5528 def : NEONInstAlias<"vshl${p}.s64 $Vdn, $Vm",
5529 (VSHLsv1i64 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5530 def : NEONInstAlias<"vshl${p}.u8 $Vdn, $Vm",
5531 (VSHLuv8i8 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5532 def : NEONInstAlias<"vshl${p}.u16 $Vdn, $Vm",
5533 (VSHLuv4i16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5534 def : NEONInstAlias<"vshl${p}.u32 $Vdn, $Vm",
5535 (VSHLuv2i32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5536 def : NEONInstAlias<"vshl${p}.u64 $Vdn, $Vm",
5537 (VSHLuv1i64 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5539 def : NEONInstAlias<"vshl${p}.s8 $Vdn, $Vm",
5540 (VSHLsv16i8 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5541 def : NEONInstAlias<"vshl${p}.s16 $Vdn, $Vm",
5542 (VSHLsv8i16 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5543 def : NEONInstAlias<"vshl${p}.s32 $Vdn, $Vm",
5544 (VSHLsv4i32 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5545 def : NEONInstAlias<"vshl${p}.s64 $Vdn, $Vm",
5546 (VSHLsv2i64 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5547 def : NEONInstAlias<"vshl${p}.u8 $Vdn, $Vm",
5548 (VSHLuv16i8 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5549 def : NEONInstAlias<"vshl${p}.u16 $Vdn, $Vm",
5550 (VSHLuv8i16 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5551 def : NEONInstAlias<"vshl${p}.u32 $Vdn, $Vm",
5552 (VSHLuv4i32 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5553 def : NEONInstAlias<"vshl${p}.u64 $Vdn, $Vm",
5554 (VSHLuv2i64 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5556 // VSHL (immediate) two-operand aliases.
5557 def : NEONInstAlias<"vshr${p}.s8 $Vdn, $imm",
5558 (VSHRsv8i8 DPR:$Vdn, DPR:$Vdn, shr_imm8:$imm, pred:$p)>;
5559 def : NEONInstAlias<"vshr${p}.s16 $Vdn, $imm",
5560 (VSHRsv4i16 DPR:$Vdn, DPR:$Vdn, shr_imm16:$imm, pred:$p)>;
5561 def : NEONInstAlias<"vshr${p}.s32 $Vdn, $imm",
5562 (VSHRsv2i32 DPR:$Vdn, DPR:$Vdn, shr_imm32:$imm, pred:$p)>;
5563 def : NEONInstAlias<"vshr${p}.s64 $Vdn, $imm",
5564 (VSHRsv1i64 DPR:$Vdn, DPR:$Vdn, shr_imm64:$imm, pred:$p)>;
5566 def : NEONInstAlias<"vshr${p}.s8 $Vdn, $imm",
5567 (VSHRsv16i8 QPR:$Vdn, QPR:$Vdn, shr_imm8:$imm, pred:$p)>;
5568 def : NEONInstAlias<"vshr${p}.s16 $Vdn, $imm",
5569 (VSHRsv8i16 QPR:$Vdn, QPR:$Vdn, shr_imm16:$imm, pred:$p)>;
5570 def : NEONInstAlias<"vshr${p}.s32 $Vdn, $imm",
5571 (VSHRsv4i32 QPR:$Vdn, QPR:$Vdn, shr_imm32:$imm, pred:$p)>;
5572 def : NEONInstAlias<"vshr${p}.s64 $Vdn, $imm",
5573 (VSHRsv2i64 QPR:$Vdn, QPR:$Vdn, shr_imm64:$imm, pred:$p)>;
5575 def : NEONInstAlias<"vshr${p}.u8 $Vdn, $imm",
5576 (VSHRuv8i8 DPR:$Vdn, DPR:$Vdn, shr_imm8:$imm, pred:$p)>;
5577 def : NEONInstAlias<"vshr${p}.u16 $Vdn, $imm",
5578 (VSHRuv4i16 DPR:$Vdn, DPR:$Vdn, shr_imm16:$imm, pred:$p)>;
5579 def : NEONInstAlias<"vshr${p}.u32 $Vdn, $imm",
5580 (VSHRuv2i32 DPR:$Vdn, DPR:$Vdn, shr_imm32:$imm, pred:$p)>;
5581 def : NEONInstAlias<"vshr${p}.u64 $Vdn, $imm",
5582 (VSHRuv1i64 DPR:$Vdn, DPR:$Vdn, shr_imm64:$imm, pred:$p)>;
5584 def : NEONInstAlias<"vshr${p}.u8 $Vdn, $imm",
5585 (VSHRuv16i8 QPR:$Vdn, QPR:$Vdn, shr_imm8:$imm, pred:$p)>;
5586 def : NEONInstAlias<"vshr${p}.u16 $Vdn, $imm",
5587 (VSHRuv8i16 QPR:$Vdn, QPR:$Vdn, shr_imm16:$imm, pred:$p)>;
5588 def : NEONInstAlias<"vshr${p}.u32 $Vdn, $imm",
5589 (VSHRuv4i32 QPR:$Vdn, QPR:$Vdn, shr_imm32:$imm, pred:$p)>;
5590 def : NEONInstAlias<"vshr${p}.u64 $Vdn, $imm",
5591 (VSHRuv2i64 QPR:$Vdn, QPR:$Vdn, shr_imm64:$imm, pred:$p)>;
5593 // VLD1 single-lane pseudo-instructions. These need special handling for
5594 // the lane index that an InstAlias can't handle, so we use these instead.
5595 defm VLD1LNdAsm : NEONDT8AsmPseudoInst<"vld1${p}", "$list, $addr",
5596 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5597 defm VLD1LNdAsm : NEONDT16AsmPseudoInst<"vld1${p}", "$list, $addr",
5598 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5599 defm VLD1LNdAsm : NEONDT32AsmPseudoInst<"vld1${p}", "$list, $addr",
5600 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5602 defm VLD1LNdWB_fixed_Asm : NEONDT8AsmPseudoInst<"vld1${p}", "$list, $addr!",
5603 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5604 defm VLD1LNdWB_fixed_Asm : NEONDT16AsmPseudoInst<"vld1${p}", "$list, $addr!",
5605 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5606 defm VLD1LNdWB_fixed_Asm : NEONDT32AsmPseudoInst<"vld1${p}", "$list, $addr!",
5607 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5608 defm VLD1LNdWB_register_Asm :
5609 NEONDT8AsmPseudoInst<"vld1${p}", "$list, $addr, $Rm",
5610 (ins VecListOneDByteIndexed:$list, addrmode6:$addr,
5611 rGPR:$Rm, pred:$p)>;
5612 defm VLD1LNdWB_register_Asm :
5613 NEONDT16AsmPseudoInst<"vld1${p}", "$list, $addr, $Rm",
5614 (ins VecListOneDByteIndexed:$list, addrmode6:$addr,
5615 rGPR:$Rm, pred:$p)>;
5616 defm VLD1LNdWB_register_Asm :
5617 NEONDT32AsmPseudoInst<"vld1${p}", "$list, $addr, $Rm",
5618 (ins VecListOneDByteIndexed:$list, addrmode6:$addr,
5619 rGPR:$Rm, pred:$p)>;
5622 // VST1 single-lane pseudo-instructions. These need special handling for
5623 // the lane index that an InstAlias can't handle, so we use these instead.
5624 defm VST1LNdAsm : NEONDT8AsmPseudoInst<"vst1${p}", "$list, $addr",
5625 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5626 defm VST1LNdAsm : NEONDT16AsmPseudoInst<"vst1${p}", "$list, $addr",
5627 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5628 defm VST1LNdAsm : NEONDT32AsmPseudoInst<"vst1${p}", "$list, $addr",
5629 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5631 defm VST1LNdWB_fixed_Asm : NEONDT8AsmPseudoInst<"vst1${p}", "$list, $addr!",
5632 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5633 defm VST1LNdWB_fixed_Asm : NEONDT16AsmPseudoInst<"vst1${p}", "$list, $addr!",
5634 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5635 defm VST1LNdWB_fixed_Asm : NEONDT32AsmPseudoInst<"vst1${p}", "$list, $addr!",
5636 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5637 defm VST1LNdWB_register_Asm :
5638 NEONDT8AsmPseudoInst<"vst1${p}", "$list, $addr, $Rm",
5639 (ins VecListOneDByteIndexed:$list, addrmode6:$addr,
5640 rGPR:$Rm, pred:$p)>;
5641 defm VST1LNdWB_register_Asm :
5642 NEONDT16AsmPseudoInst<"vst1${p}", "$list, $addr, $Rm",
5643 (ins VecListOneDByteIndexed:$list, addrmode6:$addr,
5644 rGPR:$Rm, pred:$p)>;
5645 defm VST1LNdWB_register_Asm :
5646 NEONDT32AsmPseudoInst<"vst1${p}", "$list, $addr, $Rm",
5647 (ins VecListOneDByteIndexed:$list, addrmode6:$addr,
5648 rGPR:$Rm, pred:$p)>;
5650 // VMOV takes an optional datatype suffix
5651 defm : VFPDTAnyInstAlias<"vmov${p}", "$Vd, $Vm",
5652 (VORRd DPR:$Vd, DPR:$Vm, DPR:$Vm, pred:$p)>;
5653 defm : VFPDTAnyInstAlias<"vmov${p}", "$Vd, $Vm",
5654 (VORRq QPR:$Vd, QPR:$Vm, QPR:$Vm, pred:$p)>;
5656 // VCLT (register) is an assembler alias for VCGT w/ the operands reversed.
5657 // D-register versions.
5658 def : NEONInstAlias<"vclt${p}.s8 $Dd, $Dn, $Dm",
5659 (VCGTsv8i8 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
5660 def : NEONInstAlias<"vclt${p}.s16 $Dd, $Dn, $Dm",
5661 (VCGTsv4i16 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
5662 def : NEONInstAlias<"vclt${p}.s32 $Dd, $Dn, $Dm",
5663 (VCGTsv2i32 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
5664 def : NEONInstAlias<"vclt${p}.u8 $Dd, $Dn, $Dm",
5665 (VCGTuv8i8 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
5666 def : NEONInstAlias<"vclt${p}.u16 $Dd, $Dn, $Dm",
5667 (VCGTuv4i16 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
5668 def : NEONInstAlias<"vclt${p}.u32 $Dd, $Dn, $Dm",
5669 (VCGTuv2i32 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
5670 def : NEONInstAlias<"vclt${p}.f32 $Dd, $Dn, $Dm",
5671 (VCGTfd DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
5672 // Q-register versions.
5673 def : NEONInstAlias<"vclt${p}.s8 $Qd, $Qn, $Qm",
5674 (VCGTsv16i8 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
5675 def : NEONInstAlias<"vclt${p}.s16 $Qd, $Qn, $Qm",
5676 (VCGTsv8i16 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
5677 def : NEONInstAlias<"vclt${p}.s32 $Qd, $Qn, $Qm",
5678 (VCGTsv4i32 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
5679 def : NEONInstAlias<"vclt${p}.u8 $Qd, $Qn, $Qm",
5680 (VCGTuv16i8 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
5681 def : NEONInstAlias<"vclt${p}.u16 $Qd, $Qn, $Qm",
5682 (VCGTuv8i16 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
5683 def : NEONInstAlias<"vclt${p}.u32 $Qd, $Qn, $Qm",
5684 (VCGTuv4i32 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
5685 def : NEONInstAlias<"vclt${p}.f32 $Qd, $Qn, $Qm",
5686 (VCGTfq QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
5688 // Two-operand variants for VEXT
5689 def : NEONInstAlias<"vext${p}.8 $Vdn, $Vm, $imm",
5690 (VEXTd8 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, imm0_7:$imm, pred:$p)>;
5691 def : NEONInstAlias<"vext${p}.16 $Vdn, $Vm, $imm",
5692 (VEXTd16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, imm0_3:$imm, pred:$p)>;
5693 def : NEONInstAlias<"vext${p}.32 $Vdn, $Vm, $imm",
5694 (VEXTd32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, imm0_1:$imm, pred:$p)>;
5696 def : NEONInstAlias<"vext${p}.8 $Vdn, $Vm, $imm",
5697 (VEXTq8 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, imm0_15:$imm, pred:$p)>;
5698 def : NEONInstAlias<"vext${p}.16 $Vdn, $Vm, $imm",
5699 (VEXTq16 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, imm0_7:$imm, pred:$p)>;
5700 def : NEONInstAlias<"vext${p}.32 $Vdn, $Vm, $imm",
5701 (VEXTq32 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, imm0_3:$imm, pred:$p)>;
5702 def : NEONInstAlias<"vext${p}.64 $Vdn, $Vm, $imm",
5703 (VEXTq64 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, imm0_1:$imm, pred:$p)>;
5705 // Two-operand variants for VQDMULH
5706 def : NEONInstAlias<"vqdmulh${p}.s16 $Vdn, $Vm",
5707 (VQDMULHv4i16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5708 def : NEONInstAlias<"vqdmulh${p}.s32 $Vdn, $Vm",
5709 (VQDMULHv2i32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5711 def : NEONInstAlias<"vqdmulh${p}.s16 $Vdn, $Vm",
5712 (VQDMULHv8i16 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5713 def : NEONInstAlias<"vqdmulh${p}.s32 $Vdn, $Vm",
5714 (VQDMULHv4i32 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5716 // 'gas' compatibility aliases for quad-word instructions. Strictly speaking,
5717 // these should restrict to just the Q register variants, but the register
5718 // classes are enough to match correctly regardless, so we keep it simple
5719 // and just use MnemonicAlias.
5720 def : NEONMnemonicAlias<"vbicq", "vbic">;
5721 def : NEONMnemonicAlias<"vandq", "vand">;
5722 def : NEONMnemonicAlias<"veorq", "veor">;
5723 def : NEONMnemonicAlias<"vorrq", "vorr">;
5725 def : NEONMnemonicAlias<"vmovq", "vmov">;
5726 def : NEONMnemonicAlias<"vmvnq", "vmvn">;
5728 def : NEONMnemonicAlias<"vaddq", "vadd">;
5729 def : NEONMnemonicAlias<"vsubq", "vsub">;
5731 def : NEONMnemonicAlias<"vminq", "vmin">;
5732 def : NEONMnemonicAlias<"vmaxq", "vmax">;
5734 def : NEONMnemonicAlias<"vmulq", "vmul">;
5736 def : NEONMnemonicAlias<"vabsq", "vabs">;
5738 def : NEONMnemonicAlias<"vshlq", "vshl">;
5739 def : NEONMnemonicAlias<"vshrq", "vshr">;
5741 def : NEONMnemonicAlias<"vcvtq", "vcvt">;
5743 def : NEONMnemonicAlias<"vcleq", "vcle">;
5744 def : NEONMnemonicAlias<"vceqq", "vceq">;