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);
152 // Register list of one D register, with byte lane subscripting.
153 def VecListTwoDByteIndexAsmOperand : AsmOperandClass {
154 let Name = "VecListTwoDByteIndexed";
155 let ParserMethod = "parseVectorList";
156 let RenderMethod = "addVecListIndexedOperands";
158 def VecListTwoDByteIndexed : Operand<i32> {
159 let ParserMatchClass = VecListTwoDByteIndexAsmOperand;
160 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
163 //===----------------------------------------------------------------------===//
164 // NEON-specific DAG Nodes.
165 //===----------------------------------------------------------------------===//
167 def SDTARMVCMP : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisSameAs<1, 2>]>;
168 def SDTARMVCMPZ : SDTypeProfile<1, 1, []>;
170 def NEONvceq : SDNode<"ARMISD::VCEQ", SDTARMVCMP>;
171 def NEONvceqz : SDNode<"ARMISD::VCEQZ", SDTARMVCMPZ>;
172 def NEONvcge : SDNode<"ARMISD::VCGE", SDTARMVCMP>;
173 def NEONvcgez : SDNode<"ARMISD::VCGEZ", SDTARMVCMPZ>;
174 def NEONvclez : SDNode<"ARMISD::VCLEZ", SDTARMVCMPZ>;
175 def NEONvcgeu : SDNode<"ARMISD::VCGEU", SDTARMVCMP>;
176 def NEONvcgt : SDNode<"ARMISD::VCGT", SDTARMVCMP>;
177 def NEONvcgtz : SDNode<"ARMISD::VCGTZ", SDTARMVCMPZ>;
178 def NEONvcltz : SDNode<"ARMISD::VCLTZ", SDTARMVCMPZ>;
179 def NEONvcgtu : SDNode<"ARMISD::VCGTU", SDTARMVCMP>;
180 def NEONvtst : SDNode<"ARMISD::VTST", SDTARMVCMP>;
182 // Types for vector shift by immediates. The "SHX" version is for long and
183 // narrow operations where the source and destination vectors have different
184 // types. The "SHINS" version is for shift and insert operations.
185 def SDTARMVSH : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
187 def SDTARMVSHX : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
189 def SDTARMVSHINS : SDTypeProfile<1, 3, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
190 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
192 def NEONvshl : SDNode<"ARMISD::VSHL", SDTARMVSH>;
193 def NEONvshrs : SDNode<"ARMISD::VSHRs", SDTARMVSH>;
194 def NEONvshru : SDNode<"ARMISD::VSHRu", SDTARMVSH>;
195 def NEONvshlls : SDNode<"ARMISD::VSHLLs", SDTARMVSHX>;
196 def NEONvshllu : SDNode<"ARMISD::VSHLLu", SDTARMVSHX>;
197 def NEONvshlli : SDNode<"ARMISD::VSHLLi", SDTARMVSHX>;
198 def NEONvshrn : SDNode<"ARMISD::VSHRN", SDTARMVSHX>;
200 def NEONvrshrs : SDNode<"ARMISD::VRSHRs", SDTARMVSH>;
201 def NEONvrshru : SDNode<"ARMISD::VRSHRu", SDTARMVSH>;
202 def NEONvrshrn : SDNode<"ARMISD::VRSHRN", SDTARMVSHX>;
204 def NEONvqshls : SDNode<"ARMISD::VQSHLs", SDTARMVSH>;
205 def NEONvqshlu : SDNode<"ARMISD::VQSHLu", SDTARMVSH>;
206 def NEONvqshlsu : SDNode<"ARMISD::VQSHLsu", SDTARMVSH>;
207 def NEONvqshrns : SDNode<"ARMISD::VQSHRNs", SDTARMVSHX>;
208 def NEONvqshrnu : SDNode<"ARMISD::VQSHRNu", SDTARMVSHX>;
209 def NEONvqshrnsu : SDNode<"ARMISD::VQSHRNsu", SDTARMVSHX>;
211 def NEONvqrshrns : SDNode<"ARMISD::VQRSHRNs", SDTARMVSHX>;
212 def NEONvqrshrnu : SDNode<"ARMISD::VQRSHRNu", SDTARMVSHX>;
213 def NEONvqrshrnsu : SDNode<"ARMISD::VQRSHRNsu", SDTARMVSHX>;
215 def NEONvsli : SDNode<"ARMISD::VSLI", SDTARMVSHINS>;
216 def NEONvsri : SDNode<"ARMISD::VSRI", SDTARMVSHINS>;
218 def SDTARMVGETLN : SDTypeProfile<1, 2, [SDTCisVT<0, i32>, SDTCisInt<1>,
220 def NEONvgetlaneu : SDNode<"ARMISD::VGETLANEu", SDTARMVGETLN>;
221 def NEONvgetlanes : SDNode<"ARMISD::VGETLANEs", SDTARMVGETLN>;
223 def SDTARMVMOVIMM : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisVT<1, i32>]>;
224 def NEONvmovImm : SDNode<"ARMISD::VMOVIMM", SDTARMVMOVIMM>;
225 def NEONvmvnImm : SDNode<"ARMISD::VMVNIMM", SDTARMVMOVIMM>;
226 def NEONvmovFPImm : SDNode<"ARMISD::VMOVFPIMM", SDTARMVMOVIMM>;
228 def SDTARMVORRIMM : SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
230 def NEONvorrImm : SDNode<"ARMISD::VORRIMM", SDTARMVORRIMM>;
231 def NEONvbicImm : SDNode<"ARMISD::VBICIMM", SDTARMVORRIMM>;
233 def NEONvbsl : SDNode<"ARMISD::VBSL",
234 SDTypeProfile<1, 3, [SDTCisVec<0>,
237 SDTCisSameAs<0, 3>]>>;
239 def NEONvdup : SDNode<"ARMISD::VDUP", SDTypeProfile<1, 1, [SDTCisVec<0>]>>;
241 // VDUPLANE can produce a quad-register result from a double-register source,
242 // so the result is not constrained to match the source.
243 def NEONvduplane : SDNode<"ARMISD::VDUPLANE",
244 SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisVec<1>,
247 def SDTARMVEXT : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
248 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
249 def NEONvext : SDNode<"ARMISD::VEXT", SDTARMVEXT>;
251 def SDTARMVSHUF : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisSameAs<0, 1>]>;
252 def NEONvrev64 : SDNode<"ARMISD::VREV64", SDTARMVSHUF>;
253 def NEONvrev32 : SDNode<"ARMISD::VREV32", SDTARMVSHUF>;
254 def NEONvrev16 : SDNode<"ARMISD::VREV16", SDTARMVSHUF>;
256 def SDTARMVSHUF2 : SDTypeProfile<2, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
258 SDTCisSameAs<0, 3>]>;
259 def NEONzip : SDNode<"ARMISD::VZIP", SDTARMVSHUF2>;
260 def NEONuzp : SDNode<"ARMISD::VUZP", SDTARMVSHUF2>;
261 def NEONtrn : SDNode<"ARMISD::VTRN", SDTARMVSHUF2>;
263 def SDTARMVMULL : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
264 SDTCisSameAs<1, 2>]>;
265 def NEONvmulls : SDNode<"ARMISD::VMULLs", SDTARMVMULL>;
266 def NEONvmullu : SDNode<"ARMISD::VMULLu", SDTARMVMULL>;
268 def SDTARMFMAX : SDTypeProfile<1, 2, [SDTCisVT<0, f32>, SDTCisSameAs<0, 1>,
269 SDTCisSameAs<0, 2>]>;
270 def NEONfmax : SDNode<"ARMISD::FMAX", SDTARMFMAX>;
271 def NEONfmin : SDNode<"ARMISD::FMIN", SDTARMFMAX>;
273 def NEONimmAllZerosV: PatLeaf<(NEONvmovImm (i32 timm)), [{
274 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
275 unsigned EltBits = 0;
276 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
277 return (EltBits == 32 && EltVal == 0);
280 def NEONimmAllOnesV: PatLeaf<(NEONvmovImm (i32 timm)), [{
281 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
282 unsigned EltBits = 0;
283 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
284 return (EltBits == 8 && EltVal == 0xff);
287 //===----------------------------------------------------------------------===//
288 // NEON load / store instructions
289 //===----------------------------------------------------------------------===//
291 // Use VLDM to load a Q register as a D register pair.
292 // This is a pseudo instruction that is expanded to VLDMD after reg alloc.
294 : PseudoVFPLdStM<(outs QPR:$dst), (ins GPR:$Rn),
296 [(set QPR:$dst, (v2f64 (load GPR:$Rn)))]>;
298 // Use VSTM to store a Q register as a D register pair.
299 // This is a pseudo instruction that is expanded to VSTMD after reg alloc.
301 : PseudoVFPLdStM<(outs), (ins QPR:$src, GPR:$Rn),
303 [(store (v2f64 QPR:$src), GPR:$Rn)]>;
305 // Classes for VLD* pseudo-instructions with multi-register operands.
306 // These are expanded to real instructions after register allocation.
307 class VLDQPseudo<InstrItinClass itin>
308 : PseudoNLdSt<(outs QPR:$dst), (ins addrmode6:$addr), itin, "">;
309 class VLDQWBPseudo<InstrItinClass itin>
310 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
311 (ins addrmode6:$addr, am6offset:$offset), itin,
313 class VLDQWBfixedPseudo<InstrItinClass itin>
314 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
315 (ins addrmode6:$addr), itin,
317 class VLDQWBregisterPseudo<InstrItinClass itin>
318 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
319 (ins addrmode6:$addr, rGPR:$offset), itin,
322 class VLDQQPseudo<InstrItinClass itin>
323 : PseudoNLdSt<(outs QQPR:$dst), (ins addrmode6:$addr), itin, "">;
324 class VLDQQWBPseudo<InstrItinClass itin>
325 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
326 (ins addrmode6:$addr, am6offset:$offset), itin,
328 class VLDQQWBfixedPseudo<InstrItinClass itin>
329 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
330 (ins addrmode6:$addr), itin,
332 class VLDQQWBregisterPseudo<InstrItinClass itin>
333 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
334 (ins addrmode6:$addr, rGPR:$offset), itin,
338 class VLDQQQQPseudo<InstrItinClass itin>
339 : PseudoNLdSt<(outs QQQQPR:$dst), (ins addrmode6:$addr, QQQQPR:$src),itin,
341 class VLDQQQQWBPseudo<InstrItinClass itin>
342 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
343 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
344 "$addr.addr = $wb, $src = $dst">;
346 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
348 // VLD1 : Vector Load (multiple single elements)
349 class VLD1D<bits<4> op7_4, string Dt>
350 : NLdSt<0,0b10,0b0111,op7_4, (outs VecListOneD:$Vd),
351 (ins addrmode6:$Rn), IIC_VLD1,
352 "vld1", Dt, "$Vd, $Rn", "", []> {
355 let DecoderMethod = "DecodeVLDInstruction";
357 class VLD1Q<bits<4> op7_4, string Dt>
358 : NLdSt<0,0b10,0b1010,op7_4, (outs VecListTwoD:$Vd),
359 (ins addrmode6:$Rn), IIC_VLD1x2,
360 "vld1", Dt, "$Vd, $Rn", "", []> {
362 let Inst{5-4} = Rn{5-4};
363 let DecoderMethod = "DecodeVLDInstruction";
366 def VLD1d8 : VLD1D<{0,0,0,?}, "8">;
367 def VLD1d16 : VLD1D<{0,1,0,?}, "16">;
368 def VLD1d32 : VLD1D<{1,0,0,?}, "32">;
369 def VLD1d64 : VLD1D<{1,1,0,?}, "64">;
371 def VLD1q8 : VLD1Q<{0,0,?,?}, "8">;
372 def VLD1q16 : VLD1Q<{0,1,?,?}, "16">;
373 def VLD1q32 : VLD1Q<{1,0,?,?}, "32">;
374 def VLD1q64 : VLD1Q<{1,1,?,?}, "64">;
376 def VLD1q8Pseudo : VLDQPseudo<IIC_VLD1x2>;
377 def VLD1q16Pseudo : VLDQPseudo<IIC_VLD1x2>;
378 def VLD1q32Pseudo : VLDQPseudo<IIC_VLD1x2>;
379 def VLD1q64Pseudo : VLDQPseudo<IIC_VLD1x2>;
381 // ...with address register writeback:
382 multiclass VLD1DWB<bits<4> op7_4, string Dt> {
383 def _fixed : NLdSt<0,0b10, 0b0111,op7_4, (outs VecListOneD:$Vd, GPR:$wb),
384 (ins addrmode6:$Rn), IIC_VLD1u,
385 "vld1", Dt, "$Vd, $Rn!",
386 "$Rn.addr = $wb", []> {
387 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
389 let DecoderMethod = "DecodeVLDInstruction";
390 let AsmMatchConverter = "cvtVLDwbFixed";
392 def _register : NLdSt<0,0b10,0b0111,op7_4, (outs VecListOneD:$Vd, GPR:$wb),
393 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1u,
394 "vld1", Dt, "$Vd, $Rn, $Rm",
395 "$Rn.addr = $wb", []> {
397 let DecoderMethod = "DecodeVLDInstruction";
398 let AsmMatchConverter = "cvtVLDwbRegister";
401 multiclass VLD1QWB<bits<4> op7_4, string Dt> {
402 def _fixed : NLdSt<0,0b10,0b1010,op7_4, (outs VecListTwoD:$Vd, GPR:$wb),
403 (ins addrmode6:$Rn), IIC_VLD1x2u,
404 "vld1", Dt, "$Vd, $Rn!",
405 "$Rn.addr = $wb", []> {
406 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
407 let Inst{5-4} = Rn{5-4};
408 let DecoderMethod = "DecodeVLDInstruction";
409 let AsmMatchConverter = "cvtVLDwbFixed";
411 def _register : NLdSt<0,0b10,0b1010,op7_4, (outs VecListTwoD:$Vd, GPR:$wb),
412 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1x2u,
413 "vld1", Dt, "$Vd, $Rn, $Rm",
414 "$Rn.addr = $wb", []> {
415 let Inst{5-4} = Rn{5-4};
416 let DecoderMethod = "DecodeVLDInstruction";
417 let AsmMatchConverter = "cvtVLDwbRegister";
421 defm VLD1d8wb : VLD1DWB<{0,0,0,?}, "8">;
422 defm VLD1d16wb : VLD1DWB<{0,1,0,?}, "16">;
423 defm VLD1d32wb : VLD1DWB<{1,0,0,?}, "32">;
424 defm VLD1d64wb : VLD1DWB<{1,1,0,?}, "64">;
425 defm VLD1q8wb : VLD1QWB<{0,0,?,?}, "8">;
426 defm VLD1q16wb : VLD1QWB<{0,1,?,?}, "16">;
427 defm VLD1q32wb : VLD1QWB<{1,0,?,?}, "32">;
428 defm VLD1q64wb : VLD1QWB<{1,1,?,?}, "64">;
430 def VLD1q8PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD1x2u>;
431 def VLD1q16PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD1x2u>;
432 def VLD1q32PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD1x2u>;
433 def VLD1q64PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD1x2u>;
434 def VLD1q8PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD1x2u>;
435 def VLD1q16PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD1x2u>;
436 def VLD1q32PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD1x2u>;
437 def VLD1q64PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD1x2u>;
439 // ...with 3 registers
440 class VLD1D3<bits<4> op7_4, string Dt>
441 : NLdSt<0,0b10,0b0110,op7_4, (outs VecListThreeD:$Vd),
442 (ins addrmode6:$Rn), IIC_VLD1x3, "vld1", Dt,
443 "$Vd, $Rn", "", []> {
446 let DecoderMethod = "DecodeVLDInstruction";
448 multiclass VLD1D3WB<bits<4> op7_4, string Dt> {
449 def _fixed : NLdSt<0,0b10,0b0110, op7_4, (outs VecListThreeD:$Vd, GPR:$wb),
450 (ins addrmode6:$Rn), IIC_VLD1x2u,
451 "vld1", Dt, "$Vd, $Rn!",
452 "$Rn.addr = $wb", []> {
453 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
455 let DecoderMethod = "DecodeVLDInstruction";
456 let AsmMatchConverter = "cvtVLDwbFixed";
458 def _register : NLdSt<0,0b10,0b0110,op7_4, (outs VecListThreeD:$Vd, GPR:$wb),
459 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1x2u,
460 "vld1", Dt, "$Vd, $Rn, $Rm",
461 "$Rn.addr = $wb", []> {
463 let DecoderMethod = "DecodeVLDInstruction";
464 let AsmMatchConverter = "cvtVLDwbRegister";
468 def VLD1d8T : VLD1D3<{0,0,0,?}, "8">;
469 def VLD1d16T : VLD1D3<{0,1,0,?}, "16">;
470 def VLD1d32T : VLD1D3<{1,0,0,?}, "32">;
471 def VLD1d64T : VLD1D3<{1,1,0,?}, "64">;
473 defm VLD1d8Twb : VLD1D3WB<{0,0,0,?}, "8">;
474 defm VLD1d16Twb : VLD1D3WB<{0,1,0,?}, "16">;
475 defm VLD1d32Twb : VLD1D3WB<{1,0,0,?}, "32">;
476 defm VLD1d64Twb : VLD1D3WB<{1,1,0,?}, "64">;
478 def VLD1d64TPseudo : VLDQQPseudo<IIC_VLD1x3>;
480 // ...with 4 registers
481 class VLD1D4<bits<4> op7_4, string Dt>
482 : NLdSt<0, 0b10, 0b0010, op7_4, (outs VecListFourD:$Vd),
483 (ins addrmode6:$Rn), IIC_VLD1x4, "vld1", Dt,
484 "$Vd, $Rn", "", []> {
486 let Inst{5-4} = Rn{5-4};
487 let DecoderMethod = "DecodeVLDInstruction";
489 multiclass VLD1D4WB<bits<4> op7_4, string Dt> {
490 def _fixed : NLdSt<0,0b10,0b0010, op7_4, (outs VecListFourD:$Vd, GPR:$wb),
491 (ins addrmode6:$Rn), IIC_VLD1x2u,
492 "vld1", Dt, "$Vd, $Rn!",
493 "$Rn.addr = $wb", []> {
494 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
495 let Inst{5-4} = Rn{5-4};
496 let DecoderMethod = "DecodeVLDInstruction";
497 let AsmMatchConverter = "cvtVLDwbFixed";
499 def _register : NLdSt<0,0b10,0b0010,op7_4, (outs VecListFourD:$Vd, GPR:$wb),
500 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1x2u,
501 "vld1", Dt, "$Vd, $Rn, $Rm",
502 "$Rn.addr = $wb", []> {
503 let Inst{5-4} = Rn{5-4};
504 let DecoderMethod = "DecodeVLDInstruction";
505 let AsmMatchConverter = "cvtVLDwbRegister";
509 def VLD1d8Q : VLD1D4<{0,0,?,?}, "8">;
510 def VLD1d16Q : VLD1D4<{0,1,?,?}, "16">;
511 def VLD1d32Q : VLD1D4<{1,0,?,?}, "32">;
512 def VLD1d64Q : VLD1D4<{1,1,?,?}, "64">;
514 defm VLD1d8Qwb : VLD1D4WB<{0,0,?,?}, "8">;
515 defm VLD1d16Qwb : VLD1D4WB<{0,1,?,?}, "16">;
516 defm VLD1d32Qwb : VLD1D4WB<{1,0,?,?}, "32">;
517 defm VLD1d64Qwb : VLD1D4WB<{1,1,?,?}, "64">;
519 def VLD1d64QPseudo : VLDQQPseudo<IIC_VLD1x4>;
521 // VLD2 : Vector Load (multiple 2-element structures)
522 class VLD2<bits<4> op11_8, bits<4> op7_4, string Dt, RegisterOperand VdTy,
524 : NLdSt<0, 0b10, op11_8, op7_4, (outs VdTy:$Vd),
525 (ins addrmode6:$Rn), itin,
526 "vld2", Dt, "$Vd, $Rn", "", []> {
528 let Inst{5-4} = Rn{5-4};
529 let DecoderMethod = "DecodeVLDInstruction";
532 def VLD2d8 : VLD2<0b1000, {0,0,?,?}, "8", VecListTwoD, IIC_VLD2>;
533 def VLD2d16 : VLD2<0b1000, {0,1,?,?}, "16", VecListTwoD, IIC_VLD2>;
534 def VLD2d32 : VLD2<0b1000, {1,0,?,?}, "32", VecListTwoD, IIC_VLD2>;
536 def VLD2q8 : VLD2<0b0011, {0,0,?,?}, "8", VecListFourD, IIC_VLD2x2>;
537 def VLD2q16 : VLD2<0b0011, {0,1,?,?}, "16", VecListFourD, IIC_VLD2x2>;
538 def VLD2q32 : VLD2<0b0011, {1,0,?,?}, "32", VecListFourD, IIC_VLD2x2>;
540 def VLD2d8Pseudo : VLDQPseudo<IIC_VLD2>;
541 def VLD2d16Pseudo : VLDQPseudo<IIC_VLD2>;
542 def VLD2d32Pseudo : VLDQPseudo<IIC_VLD2>;
544 def VLD2q8Pseudo : VLDQQPseudo<IIC_VLD2x2>;
545 def VLD2q16Pseudo : VLDQQPseudo<IIC_VLD2x2>;
546 def VLD2q32Pseudo : VLDQQPseudo<IIC_VLD2x2>;
548 // ...with address register writeback:
549 multiclass VLD2WB<bits<4> op11_8, bits<4> op7_4, string Dt,
550 RegisterOperand VdTy, InstrItinClass itin> {
551 def _fixed : NLdSt<0, 0b10, op11_8, op7_4, (outs VdTy:$Vd, GPR:$wb),
552 (ins addrmode6:$Rn), itin,
553 "vld2", Dt, "$Vd, $Rn!",
554 "$Rn.addr = $wb", []> {
555 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
556 let Inst{5-4} = Rn{5-4};
557 let DecoderMethod = "DecodeVLDInstruction";
558 let AsmMatchConverter = "cvtVLDwbFixed";
560 def _register : NLdSt<0, 0b10, op11_8, op7_4, (outs VdTy:$Vd, GPR:$wb),
561 (ins addrmode6:$Rn, rGPR:$Rm), itin,
562 "vld2", Dt, "$Vd, $Rn, $Rm",
563 "$Rn.addr = $wb", []> {
564 let Inst{5-4} = Rn{5-4};
565 let DecoderMethod = "DecodeVLDInstruction";
566 let AsmMatchConverter = "cvtVLDwbRegister";
570 defm VLD2d8wb : VLD2WB<0b1000, {0,0,?,?}, "8", VecListTwoD, IIC_VLD2u>;
571 defm VLD2d16wb : VLD2WB<0b1000, {0,1,?,?}, "16", VecListTwoD, IIC_VLD2u>;
572 defm VLD2d32wb : VLD2WB<0b1000, {1,0,?,?}, "32", VecListTwoD, IIC_VLD2u>;
574 defm VLD2q8wb : VLD2WB<0b0011, {0,0,?,?}, "8", VecListFourD, IIC_VLD2x2u>;
575 defm VLD2q16wb : VLD2WB<0b0011, {0,1,?,?}, "16", VecListFourD, IIC_VLD2x2u>;
576 defm VLD2q32wb : VLD2WB<0b0011, {1,0,?,?}, "32", VecListFourD, IIC_VLD2x2u>;
578 def VLD2d8PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD2u>;
579 def VLD2d16PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD2u>;
580 def VLD2d32PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD2u>;
581 def VLD2d8PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD2u>;
582 def VLD2d16PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD2u>;
583 def VLD2d32PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD2u>;
585 def VLD2q8PseudoWB_fixed : VLDQQWBfixedPseudo<IIC_VLD2x2u>;
586 def VLD2q16PseudoWB_fixed : VLDQQWBfixedPseudo<IIC_VLD2x2u>;
587 def VLD2q32PseudoWB_fixed : VLDQQWBfixedPseudo<IIC_VLD2x2u>;
588 def VLD2q8PseudoWB_register : VLDQQWBregisterPseudo<IIC_VLD2x2u>;
589 def VLD2q16PseudoWB_register : VLDQQWBregisterPseudo<IIC_VLD2x2u>;
590 def VLD2q32PseudoWB_register : VLDQQWBregisterPseudo<IIC_VLD2x2u>;
592 // ...with double-spaced registers
593 def VLD2b8 : VLD2<0b1001, {0,0,?,?}, "8", VecListTwoQ, IIC_VLD2>;
594 def VLD2b16 : VLD2<0b1001, {0,1,?,?}, "16", VecListTwoQ, IIC_VLD2>;
595 def VLD2b32 : VLD2<0b1001, {1,0,?,?}, "32", VecListTwoQ, IIC_VLD2>;
596 defm VLD2b8wb : VLD2WB<0b1001, {0,0,?,?}, "8", VecListTwoQ, IIC_VLD2u>;
597 defm VLD2b16wb : VLD2WB<0b1001, {0,1,?,?}, "16", VecListTwoQ, IIC_VLD2u>;
598 defm VLD2b32wb : VLD2WB<0b1001, {1,0,?,?}, "32", VecListTwoQ, IIC_VLD2u>;
600 // VLD3 : Vector Load (multiple 3-element structures)
601 class VLD3D<bits<4> op11_8, bits<4> op7_4, string Dt>
602 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
603 (ins addrmode6:$Rn), IIC_VLD3,
604 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn", "", []> {
607 let DecoderMethod = "DecodeVLDInstruction";
610 def VLD3d8 : VLD3D<0b0100, {0,0,0,?}, "8">;
611 def VLD3d16 : VLD3D<0b0100, {0,1,0,?}, "16">;
612 def VLD3d32 : VLD3D<0b0100, {1,0,0,?}, "32">;
614 def VLD3d8Pseudo : VLDQQPseudo<IIC_VLD3>;
615 def VLD3d16Pseudo : VLDQQPseudo<IIC_VLD3>;
616 def VLD3d32Pseudo : VLDQQPseudo<IIC_VLD3>;
618 // ...with address register writeback:
619 class VLD3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
620 : NLdSt<0, 0b10, op11_8, op7_4,
621 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
622 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD3u,
623 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn$Rm",
624 "$Rn.addr = $wb", []> {
626 let DecoderMethod = "DecodeVLDInstruction";
629 def VLD3d8_UPD : VLD3DWB<0b0100, {0,0,0,?}, "8">;
630 def VLD3d16_UPD : VLD3DWB<0b0100, {0,1,0,?}, "16">;
631 def VLD3d32_UPD : VLD3DWB<0b0100, {1,0,0,?}, "32">;
633 def VLD3d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
634 def VLD3d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
635 def VLD3d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
637 // ...with double-spaced registers:
638 def VLD3q8 : VLD3D<0b0101, {0,0,0,?}, "8">;
639 def VLD3q16 : VLD3D<0b0101, {0,1,0,?}, "16">;
640 def VLD3q32 : VLD3D<0b0101, {1,0,0,?}, "32">;
641 def VLD3q8_UPD : VLD3DWB<0b0101, {0,0,0,?}, "8">;
642 def VLD3q16_UPD : VLD3DWB<0b0101, {0,1,0,?}, "16">;
643 def VLD3q32_UPD : VLD3DWB<0b0101, {1,0,0,?}, "32">;
645 def VLD3q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
646 def VLD3q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
647 def VLD3q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
649 // ...alternate versions to be allocated odd register numbers:
650 def VLD3q8oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
651 def VLD3q16oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
652 def VLD3q32oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
654 def VLD3q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
655 def VLD3q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
656 def VLD3q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
658 // VLD4 : Vector Load (multiple 4-element structures)
659 class VLD4D<bits<4> op11_8, bits<4> op7_4, string Dt>
660 : NLdSt<0, 0b10, op11_8, op7_4,
661 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
662 (ins addrmode6:$Rn), IIC_VLD4,
663 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn", "", []> {
665 let Inst{5-4} = Rn{5-4};
666 let DecoderMethod = "DecodeVLDInstruction";
669 def VLD4d8 : VLD4D<0b0000, {0,0,?,?}, "8">;
670 def VLD4d16 : VLD4D<0b0000, {0,1,?,?}, "16">;
671 def VLD4d32 : VLD4D<0b0000, {1,0,?,?}, "32">;
673 def VLD4d8Pseudo : VLDQQPseudo<IIC_VLD4>;
674 def VLD4d16Pseudo : VLDQQPseudo<IIC_VLD4>;
675 def VLD4d32Pseudo : VLDQQPseudo<IIC_VLD4>;
677 // ...with address register writeback:
678 class VLD4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
679 : NLdSt<0, 0b10, op11_8, op7_4,
680 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
681 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD4u,
682 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn$Rm",
683 "$Rn.addr = $wb", []> {
684 let Inst{5-4} = Rn{5-4};
685 let DecoderMethod = "DecodeVLDInstruction";
688 def VLD4d8_UPD : VLD4DWB<0b0000, {0,0,?,?}, "8">;
689 def VLD4d16_UPD : VLD4DWB<0b0000, {0,1,?,?}, "16">;
690 def VLD4d32_UPD : VLD4DWB<0b0000, {1,0,?,?}, "32">;
692 def VLD4d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
693 def VLD4d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
694 def VLD4d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
696 // ...with double-spaced registers:
697 def VLD4q8 : VLD4D<0b0001, {0,0,?,?}, "8">;
698 def VLD4q16 : VLD4D<0b0001, {0,1,?,?}, "16">;
699 def VLD4q32 : VLD4D<0b0001, {1,0,?,?}, "32">;
700 def VLD4q8_UPD : VLD4DWB<0b0001, {0,0,?,?}, "8">;
701 def VLD4q16_UPD : VLD4DWB<0b0001, {0,1,?,?}, "16">;
702 def VLD4q32_UPD : VLD4DWB<0b0001, {1,0,?,?}, "32">;
704 def VLD4q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
705 def VLD4q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
706 def VLD4q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
708 // ...alternate versions to be allocated odd register numbers:
709 def VLD4q8oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
710 def VLD4q16oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
711 def VLD4q32oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
713 def VLD4q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
714 def VLD4q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
715 def VLD4q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
717 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
719 // Classes for VLD*LN pseudo-instructions with multi-register operands.
720 // These are expanded to real instructions after register allocation.
721 class VLDQLNPseudo<InstrItinClass itin>
722 : PseudoNLdSt<(outs QPR:$dst),
723 (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
724 itin, "$src = $dst">;
725 class VLDQLNWBPseudo<InstrItinClass itin>
726 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
727 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
728 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
729 class VLDQQLNPseudo<InstrItinClass itin>
730 : PseudoNLdSt<(outs QQPR:$dst),
731 (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
732 itin, "$src = $dst">;
733 class VLDQQLNWBPseudo<InstrItinClass itin>
734 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
735 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
736 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
737 class VLDQQQQLNPseudo<InstrItinClass itin>
738 : PseudoNLdSt<(outs QQQQPR:$dst),
739 (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
740 itin, "$src = $dst">;
741 class VLDQQQQLNWBPseudo<InstrItinClass itin>
742 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
743 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
744 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
746 // VLD1LN : Vector Load (single element to one lane)
747 class VLD1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
749 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd),
750 (ins addrmode6:$Rn, DPR:$src, nohash_imm:$lane),
751 IIC_VLD1ln, "vld1", Dt, "\\{$Vd[$lane]\\}, $Rn",
753 [(set DPR:$Vd, (vector_insert (Ty DPR:$src),
754 (i32 (LoadOp addrmode6:$Rn)),
757 let DecoderMethod = "DecodeVLD1LN";
759 class VLD1LN32<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
761 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd),
762 (ins addrmode6oneL32:$Rn, DPR:$src, nohash_imm:$lane),
763 IIC_VLD1ln, "vld1", Dt, "\\{$Vd[$lane]\\}, $Rn",
765 [(set DPR:$Vd, (vector_insert (Ty DPR:$src),
766 (i32 (LoadOp addrmode6oneL32:$Rn)),
769 let DecoderMethod = "DecodeVLD1LN";
771 class VLD1QLNPseudo<ValueType Ty, PatFrag LoadOp> : VLDQLNPseudo<IIC_VLD1ln> {
772 let Pattern = [(set QPR:$dst, (vector_insert (Ty QPR:$src),
773 (i32 (LoadOp addrmode6:$addr)),
777 def VLD1LNd8 : VLD1LN<0b0000, {?,?,?,0}, "8", v8i8, extloadi8> {
778 let Inst{7-5} = lane{2-0};
780 def VLD1LNd16 : VLD1LN<0b0100, {?,?,0,?}, "16", v4i16, extloadi16> {
781 let Inst{7-6} = lane{1-0};
784 def VLD1LNd32 : VLD1LN32<0b1000, {?,0,?,?}, "32", v2i32, load> {
785 let Inst{7} = lane{0};
790 def VLD1LNq8Pseudo : VLD1QLNPseudo<v16i8, extloadi8>;
791 def VLD1LNq16Pseudo : VLD1QLNPseudo<v8i16, extloadi16>;
792 def VLD1LNq32Pseudo : VLD1QLNPseudo<v4i32, load>;
794 def : Pat<(vector_insert (v2f32 DPR:$src),
795 (f32 (load addrmode6:$addr)), imm:$lane),
796 (VLD1LNd32 addrmode6:$addr, DPR:$src, imm:$lane)>;
797 def : Pat<(vector_insert (v4f32 QPR:$src),
798 (f32 (load addrmode6:$addr)), imm:$lane),
799 (VLD1LNq32Pseudo addrmode6:$addr, QPR:$src, imm:$lane)>;
801 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
803 // ...with address register writeback:
804 class VLD1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
805 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, GPR:$wb),
806 (ins addrmode6:$Rn, am6offset:$Rm,
807 DPR:$src, nohash_imm:$lane), IIC_VLD1lnu, "vld1", Dt,
808 "\\{$Vd[$lane]\\}, $Rn$Rm",
809 "$src = $Vd, $Rn.addr = $wb", []> {
810 let DecoderMethod = "DecodeVLD1LN";
813 def VLD1LNd8_UPD : VLD1LNWB<0b0000, {?,?,?,0}, "8"> {
814 let Inst{7-5} = lane{2-0};
816 def VLD1LNd16_UPD : VLD1LNWB<0b0100, {?,?,0,?}, "16"> {
817 let Inst{7-6} = lane{1-0};
820 def VLD1LNd32_UPD : VLD1LNWB<0b1000, {?,0,?,?}, "32"> {
821 let Inst{7} = lane{0};
826 def VLD1LNq8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
827 def VLD1LNq16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
828 def VLD1LNq32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
830 // VLD2LN : Vector Load (single 2-element structure to one lane)
831 class VLD2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
832 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2),
833 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, nohash_imm:$lane),
834 IIC_VLD2ln, "vld2", Dt, "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn",
835 "$src1 = $Vd, $src2 = $dst2", []> {
838 let DecoderMethod = "DecodeVLD2LN";
841 def VLD2LNd8 : VLD2LN<0b0001, {?,?,?,?}, "8"> {
842 let Inst{7-5} = lane{2-0};
844 def VLD2LNd16 : VLD2LN<0b0101, {?,?,0,?}, "16"> {
845 let Inst{7-6} = lane{1-0};
847 def VLD2LNd32 : VLD2LN<0b1001, {?,0,0,?}, "32"> {
848 let Inst{7} = lane{0};
851 def VLD2LNd8Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
852 def VLD2LNd16Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
853 def VLD2LNd32Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
855 // ...with double-spaced registers:
856 def VLD2LNq16 : VLD2LN<0b0101, {?,?,1,?}, "16"> {
857 let Inst{7-6} = lane{1-0};
859 def VLD2LNq32 : VLD2LN<0b1001, {?,1,0,?}, "32"> {
860 let Inst{7} = lane{0};
863 def VLD2LNq16Pseudo : VLDQQLNPseudo<IIC_VLD2ln>;
864 def VLD2LNq32Pseudo : VLDQQLNPseudo<IIC_VLD2ln>;
866 // ...with address register writeback:
867 class VLD2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
868 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
869 (ins addrmode6:$Rn, am6offset:$Rm,
870 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VLD2lnu, "vld2", Dt,
871 "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn$Rm",
872 "$src1 = $Vd, $src2 = $dst2, $Rn.addr = $wb", []> {
874 let DecoderMethod = "DecodeVLD2LN";
877 def VLD2LNd8_UPD : VLD2LNWB<0b0001, {?,?,?,?}, "8"> {
878 let Inst{7-5} = lane{2-0};
880 def VLD2LNd16_UPD : VLD2LNWB<0b0101, {?,?,0,?}, "16"> {
881 let Inst{7-6} = lane{1-0};
883 def VLD2LNd32_UPD : VLD2LNWB<0b1001, {?,0,0,?}, "32"> {
884 let Inst{7} = lane{0};
887 def VLD2LNd8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
888 def VLD2LNd16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
889 def VLD2LNd32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
891 def VLD2LNq16_UPD : VLD2LNWB<0b0101, {?,?,1,?}, "16"> {
892 let Inst{7-6} = lane{1-0};
894 def VLD2LNq32_UPD : VLD2LNWB<0b1001, {?,1,0,?}, "32"> {
895 let Inst{7} = lane{0};
898 def VLD2LNq16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
899 def VLD2LNq32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
901 // VLD3LN : Vector Load (single 3-element structure to one lane)
902 class VLD3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
903 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
904 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3,
905 nohash_imm:$lane), IIC_VLD3ln, "vld3", Dt,
906 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn",
907 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3", []> {
909 let DecoderMethod = "DecodeVLD3LN";
912 def VLD3LNd8 : VLD3LN<0b0010, {?,?,?,0}, "8"> {
913 let Inst{7-5} = lane{2-0};
915 def VLD3LNd16 : VLD3LN<0b0110, {?,?,0,0}, "16"> {
916 let Inst{7-6} = lane{1-0};
918 def VLD3LNd32 : VLD3LN<0b1010, {?,0,0,0}, "32"> {
919 let Inst{7} = lane{0};
922 def VLD3LNd8Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
923 def VLD3LNd16Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
924 def VLD3LNd32Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
926 // ...with double-spaced registers:
927 def VLD3LNq16 : VLD3LN<0b0110, {?,?,1,0}, "16"> {
928 let Inst{7-6} = lane{1-0};
930 def VLD3LNq32 : VLD3LN<0b1010, {?,1,0,0}, "32"> {
931 let Inst{7} = lane{0};
934 def VLD3LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
935 def VLD3LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
937 // ...with address register writeback:
938 class VLD3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
939 : NLdStLn<1, 0b10, op11_8, op7_4,
940 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
941 (ins addrmode6:$Rn, am6offset:$Rm,
942 DPR:$src1, DPR:$src2, DPR:$src3, nohash_imm:$lane),
943 IIC_VLD3lnu, "vld3", Dt,
944 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn$Rm",
945 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $Rn.addr = $wb",
947 let DecoderMethod = "DecodeVLD3LN";
950 def VLD3LNd8_UPD : VLD3LNWB<0b0010, {?,?,?,0}, "8"> {
951 let Inst{7-5} = lane{2-0};
953 def VLD3LNd16_UPD : VLD3LNWB<0b0110, {?,?,0,0}, "16"> {
954 let Inst{7-6} = lane{1-0};
956 def VLD3LNd32_UPD : VLD3LNWB<0b1010, {?,0,0,0}, "32"> {
957 let Inst{7} = lane{0};
960 def VLD3LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
961 def VLD3LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
962 def VLD3LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
964 def VLD3LNq16_UPD : VLD3LNWB<0b0110, {?,?,1,0}, "16"> {
965 let Inst{7-6} = lane{1-0};
967 def VLD3LNq32_UPD : VLD3LNWB<0b1010, {?,1,0,0}, "32"> {
968 let Inst{7} = lane{0};
971 def VLD3LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
972 def VLD3LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
974 // VLD4LN : Vector Load (single 4-element structure to one lane)
975 class VLD4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
976 : NLdStLn<1, 0b10, op11_8, op7_4,
977 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
978 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
979 nohash_imm:$lane), IIC_VLD4ln, "vld4", Dt,
980 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn",
981 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4", []> {
984 let DecoderMethod = "DecodeVLD4LN";
987 def VLD4LNd8 : VLD4LN<0b0011, {?,?,?,?}, "8"> {
988 let Inst{7-5} = lane{2-0};
990 def VLD4LNd16 : VLD4LN<0b0111, {?,?,0,?}, "16"> {
991 let Inst{7-6} = lane{1-0};
993 def VLD4LNd32 : VLD4LN<0b1011, {?,0,?,?}, "32"> {
994 let Inst{7} = lane{0};
998 def VLD4LNd8Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
999 def VLD4LNd16Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
1000 def VLD4LNd32Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
1002 // ...with double-spaced registers:
1003 def VLD4LNq16 : VLD4LN<0b0111, {?,?,1,?}, "16"> {
1004 let Inst{7-6} = lane{1-0};
1006 def VLD4LNq32 : VLD4LN<0b1011, {?,1,?,?}, "32"> {
1007 let Inst{7} = lane{0};
1008 let Inst{5} = Rn{5};
1011 def VLD4LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>;
1012 def VLD4LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>;
1014 // ...with address register writeback:
1015 class VLD4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1016 : NLdStLn<1, 0b10, op11_8, op7_4,
1017 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
1018 (ins addrmode6:$Rn, am6offset:$Rm,
1019 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
1020 IIC_VLD4lnu, "vld4", Dt,
1021 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn$Rm",
1022 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4, $Rn.addr = $wb",
1024 let Inst{4} = Rn{4};
1025 let DecoderMethod = "DecodeVLD4LN" ;
1028 def VLD4LNd8_UPD : VLD4LNWB<0b0011, {?,?,?,?}, "8"> {
1029 let Inst{7-5} = lane{2-0};
1031 def VLD4LNd16_UPD : VLD4LNWB<0b0111, {?,?,0,?}, "16"> {
1032 let Inst{7-6} = lane{1-0};
1034 def VLD4LNd32_UPD : VLD4LNWB<0b1011, {?,0,?,?}, "32"> {
1035 let Inst{7} = lane{0};
1036 let Inst{5} = Rn{5};
1039 def VLD4LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
1040 def VLD4LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
1041 def VLD4LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
1043 def VLD4LNq16_UPD : VLD4LNWB<0b0111, {?,?,1,?}, "16"> {
1044 let Inst{7-6} = lane{1-0};
1046 def VLD4LNq32_UPD : VLD4LNWB<0b1011, {?,1,?,?}, "32"> {
1047 let Inst{7} = lane{0};
1048 let Inst{5} = Rn{5};
1051 def VLD4LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
1052 def VLD4LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
1054 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
1056 // VLD1DUP : Vector Load (single element to all lanes)
1057 class VLD1DUP<bits<4> op7_4, string Dt, ValueType Ty, PatFrag LoadOp>
1058 : NLdSt<1, 0b10, 0b1100, op7_4, (outs VecListOneDAllLanes:$Vd),
1059 (ins addrmode6dup:$Rn),
1060 IIC_VLD1dup, "vld1", Dt, "$Vd, $Rn", "",
1061 [(set VecListOneDAllLanes:$Vd,
1062 (Ty (NEONvdup (i32 (LoadOp addrmode6dup:$Rn)))))]> {
1064 let Inst{4} = Rn{4};
1065 let DecoderMethod = "DecodeVLD1DupInstruction";
1067 class VLD1QDUPPseudo<ValueType Ty, PatFrag LoadOp> : VLDQPseudo<IIC_VLD1dup> {
1068 let Pattern = [(set QPR:$dst,
1069 (Ty (NEONvdup (i32 (LoadOp addrmode6dup:$addr)))))];
1072 def VLD1DUPd8 : VLD1DUP<{0,0,0,?}, "8", v8i8, extloadi8>;
1073 def VLD1DUPd16 : VLD1DUP<{0,1,0,?}, "16", v4i16, extloadi16>;
1074 def VLD1DUPd32 : VLD1DUP<{1,0,0,?}, "32", v2i32, load>;
1076 def VLD1DUPq8Pseudo : VLD1QDUPPseudo<v16i8, extloadi8>;
1077 def VLD1DUPq16Pseudo : VLD1QDUPPseudo<v8i16, extloadi16>;
1078 def VLD1DUPq32Pseudo : VLD1QDUPPseudo<v4i32, load>;
1080 def : Pat<(v2f32 (NEONvdup (f32 (load addrmode6dup:$addr)))),
1081 (VLD1DUPd32 addrmode6:$addr)>;
1082 def : Pat<(v4f32 (NEONvdup (f32 (load addrmode6dup:$addr)))),
1083 (VLD1DUPq32Pseudo addrmode6:$addr)>;
1085 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
1087 class VLD1QDUP<bits<4> op7_4, string Dt>
1088 : NLdSt<1, 0b10, 0b1100, op7_4, (outs VecListTwoDAllLanes:$Vd),
1089 (ins addrmode6dup:$Rn), IIC_VLD1dup,
1090 "vld1", Dt, "$Vd, $Rn", "", []> {
1092 let Inst{4} = Rn{4};
1093 let DecoderMethod = "DecodeVLD1DupInstruction";
1096 def VLD1DUPq8 : VLD1QDUP<{0,0,1,0}, "8">;
1097 def VLD1DUPq16 : VLD1QDUP<{0,1,1,?}, "16">;
1098 def VLD1DUPq32 : VLD1QDUP<{1,0,1,?}, "32">;
1100 // ...with address register writeback:
1101 multiclass VLD1DUPWB<bits<4> op7_4, string Dt> {
1102 def _fixed : NLdSt<1, 0b10, 0b1100, op7_4,
1103 (outs VecListOneDAllLanes:$Vd, GPR:$wb),
1104 (ins addrmode6dup:$Rn), IIC_VLD1dupu,
1105 "vld1", Dt, "$Vd, $Rn!",
1106 "$Rn.addr = $wb", []> {
1107 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1108 let Inst{4} = Rn{4};
1109 let DecoderMethod = "DecodeVLD1DupInstruction";
1110 let AsmMatchConverter = "cvtVLDwbFixed";
1112 def _register : NLdSt<1, 0b10, 0b1100, op7_4,
1113 (outs VecListOneDAllLanes:$Vd, GPR:$wb),
1114 (ins addrmode6dup:$Rn, rGPR:$Rm), IIC_VLD1dupu,
1115 "vld1", Dt, "$Vd, $Rn, $Rm",
1116 "$Rn.addr = $wb", []> {
1117 let Inst{4} = Rn{4};
1118 let DecoderMethod = "DecodeVLD1DupInstruction";
1119 let AsmMatchConverter = "cvtVLDwbRegister";
1122 multiclass VLD1QDUPWB<bits<4> op7_4, string Dt> {
1123 def _fixed : NLdSt<1, 0b10, 0b1100, op7_4,
1124 (outs VecListTwoDAllLanes:$Vd, GPR:$wb),
1125 (ins addrmode6dup:$Rn), IIC_VLD1dupu,
1126 "vld1", Dt, "$Vd, $Rn!",
1127 "$Rn.addr = $wb", []> {
1128 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1129 let Inst{4} = Rn{4};
1130 let DecoderMethod = "DecodeVLD1DupInstruction";
1131 let AsmMatchConverter = "cvtVLDwbFixed";
1133 def _register : NLdSt<1, 0b10, 0b1100, op7_4,
1134 (outs VecListTwoDAllLanes:$Vd, GPR:$wb),
1135 (ins addrmode6dup:$Rn, rGPR:$Rm), IIC_VLD1dupu,
1136 "vld1", Dt, "$Vd, $Rn, $Rm",
1137 "$Rn.addr = $wb", []> {
1138 let Inst{4} = Rn{4};
1139 let DecoderMethod = "DecodeVLD1DupInstruction";
1140 let AsmMatchConverter = "cvtVLDwbRegister";
1144 defm VLD1DUPd8wb : VLD1DUPWB<{0,0,0,0}, "8">;
1145 defm VLD1DUPd16wb : VLD1DUPWB<{0,1,0,?}, "16">;
1146 defm VLD1DUPd32wb : VLD1DUPWB<{1,0,0,?}, "32">;
1148 defm VLD1DUPq8wb : VLD1QDUPWB<{0,0,1,0}, "8">;
1149 defm VLD1DUPq16wb : VLD1QDUPWB<{0,1,1,?}, "16">;
1150 defm VLD1DUPq32wb : VLD1QDUPWB<{1,0,1,?}, "32">;
1152 def VLD1DUPq8PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD1dupu>;
1153 def VLD1DUPq16PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD1dupu>;
1154 def VLD1DUPq32PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD1dupu>;
1155 def VLD1DUPq8PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD1dupu>;
1156 def VLD1DUPq16PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD1dupu>;
1157 def VLD1DUPq32PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD1dupu>;
1159 // VLD2DUP : Vector Load (single 2-element structure to all lanes)
1160 class VLD2DUP<bits<4> op7_4, string Dt>
1161 : NLdSt<1, 0b10, 0b1101, op7_4, (outs DPR:$Vd, DPR:$dst2),
1162 (ins addrmode6dup:$Rn), IIC_VLD2dup,
1163 "vld2", Dt, "\\{$Vd[], $dst2[]\\}, $Rn", "", []> {
1165 let Inst{4} = Rn{4};
1166 let DecoderMethod = "DecodeVLD2DupInstruction";
1169 def VLD2DUPd8 : VLD2DUP<{0,0,0,?}, "8">;
1170 def VLD2DUPd16 : VLD2DUP<{0,1,0,?}, "16">;
1171 def VLD2DUPd32 : VLD2DUP<{1,0,0,?}, "32">;
1173 def VLD2DUPd8Pseudo : VLDQPseudo<IIC_VLD2dup>;
1174 def VLD2DUPd16Pseudo : VLDQPseudo<IIC_VLD2dup>;
1175 def VLD2DUPd32Pseudo : VLDQPseudo<IIC_VLD2dup>;
1177 // ...with double-spaced registers (not used for codegen):
1178 def VLD2DUPd8x2 : VLD2DUP<{0,0,1,?}, "8">;
1179 def VLD2DUPd16x2 : VLD2DUP<{0,1,1,?}, "16">;
1180 def VLD2DUPd32x2 : VLD2DUP<{1,0,1,?}, "32">;
1182 // ...with address register writeback:
1183 class VLD2DUPWB<bits<4> op7_4, string Dt>
1184 : NLdSt<1, 0b10, 0b1101, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
1185 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD2dupu,
1186 "vld2", Dt, "\\{$Vd[], $dst2[]\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
1187 let Inst{4} = Rn{4};
1188 let DecoderMethod = "DecodeVLD2DupInstruction";
1191 def VLD2DUPd8_UPD : VLD2DUPWB<{0,0,0,0}, "8">;
1192 def VLD2DUPd16_UPD : VLD2DUPWB<{0,1,0,?}, "16">;
1193 def VLD2DUPd32_UPD : VLD2DUPWB<{1,0,0,?}, "32">;
1195 def VLD2DUPd8x2_UPD : VLD2DUPWB<{0,0,1,0}, "8">;
1196 def VLD2DUPd16x2_UPD : VLD2DUPWB<{0,1,1,?}, "16">;
1197 def VLD2DUPd32x2_UPD : VLD2DUPWB<{1,0,1,?}, "32">;
1199 def VLD2DUPd8Pseudo_UPD : VLDQWBPseudo<IIC_VLD2dupu>;
1200 def VLD2DUPd16Pseudo_UPD : VLDQWBPseudo<IIC_VLD2dupu>;
1201 def VLD2DUPd32Pseudo_UPD : VLDQWBPseudo<IIC_VLD2dupu>;
1203 // VLD3DUP : Vector Load (single 3-element structure to all lanes)
1204 class VLD3DUP<bits<4> op7_4, string Dt>
1205 : NLdSt<1, 0b10, 0b1110, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
1206 (ins addrmode6dup:$Rn), IIC_VLD3dup,
1207 "vld3", Dt, "\\{$Vd[], $dst2[], $dst3[]\\}, $Rn", "", []> {
1210 let DecoderMethod = "DecodeVLD3DupInstruction";
1213 def VLD3DUPd8 : VLD3DUP<{0,0,0,?}, "8">;
1214 def VLD3DUPd16 : VLD3DUP<{0,1,0,?}, "16">;
1215 def VLD3DUPd32 : VLD3DUP<{1,0,0,?}, "32">;
1217 def VLD3DUPd8Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1218 def VLD3DUPd16Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1219 def VLD3DUPd32Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1221 // ...with double-spaced registers (not used for codegen):
1222 def VLD3DUPd8x2 : VLD3DUP<{0,0,1,?}, "8">;
1223 def VLD3DUPd16x2 : VLD3DUP<{0,1,1,?}, "16">;
1224 def VLD3DUPd32x2 : VLD3DUP<{1,0,1,?}, "32">;
1226 // ...with address register writeback:
1227 class VLD3DUPWB<bits<4> op7_4, string Dt>
1228 : NLdSt<1, 0b10, 0b1110, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
1229 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD3dupu,
1230 "vld3", Dt, "\\{$Vd[], $dst2[], $dst3[]\\}, $Rn$Rm",
1231 "$Rn.addr = $wb", []> {
1233 let DecoderMethod = "DecodeVLD3DupInstruction";
1236 def VLD3DUPd8_UPD : VLD3DUPWB<{0,0,0,0}, "8">;
1237 def VLD3DUPd16_UPD : VLD3DUPWB<{0,1,0,?}, "16">;
1238 def VLD3DUPd32_UPD : VLD3DUPWB<{1,0,0,?}, "32">;
1240 def VLD3DUPd8x2_UPD : VLD3DUPWB<{0,0,1,0}, "8">;
1241 def VLD3DUPd16x2_UPD : VLD3DUPWB<{0,1,1,?}, "16">;
1242 def VLD3DUPd32x2_UPD : VLD3DUPWB<{1,0,1,?}, "32">;
1244 def VLD3DUPd8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1245 def VLD3DUPd16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1246 def VLD3DUPd32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1248 // VLD4DUP : Vector Load (single 4-element structure to all lanes)
1249 class VLD4DUP<bits<4> op7_4, string Dt>
1250 : NLdSt<1, 0b10, 0b1111, op7_4,
1251 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
1252 (ins addrmode6dup:$Rn), IIC_VLD4dup,
1253 "vld4", Dt, "\\{$Vd[], $dst2[], $dst3[], $dst4[]\\}, $Rn", "", []> {
1255 let Inst{4} = Rn{4};
1256 let DecoderMethod = "DecodeVLD4DupInstruction";
1259 def VLD4DUPd8 : VLD4DUP<{0,0,0,?}, "8">;
1260 def VLD4DUPd16 : VLD4DUP<{0,1,0,?}, "16">;
1261 def VLD4DUPd32 : VLD4DUP<{1,?,0,?}, "32"> { let Inst{6} = Rn{5}; }
1263 def VLD4DUPd8Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1264 def VLD4DUPd16Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1265 def VLD4DUPd32Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1267 // ...with double-spaced registers (not used for codegen):
1268 def VLD4DUPd8x2 : VLD4DUP<{0,0,1,?}, "8">;
1269 def VLD4DUPd16x2 : VLD4DUP<{0,1,1,?}, "16">;
1270 def VLD4DUPd32x2 : VLD4DUP<{1,?,1,?}, "32"> { let Inst{6} = Rn{5}; }
1272 // ...with address register writeback:
1273 class VLD4DUPWB<bits<4> op7_4, string Dt>
1274 : NLdSt<1, 0b10, 0b1111, op7_4,
1275 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
1276 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD4dupu,
1277 "vld4", Dt, "\\{$Vd[], $dst2[], $dst3[], $dst4[]\\}, $Rn$Rm",
1278 "$Rn.addr = $wb", []> {
1279 let Inst{4} = Rn{4};
1280 let DecoderMethod = "DecodeVLD4DupInstruction";
1283 def VLD4DUPd8_UPD : VLD4DUPWB<{0,0,0,0}, "8">;
1284 def VLD4DUPd16_UPD : VLD4DUPWB<{0,1,0,?}, "16">;
1285 def VLD4DUPd32_UPD : VLD4DUPWB<{1,?,0,?}, "32"> { let Inst{6} = Rn{5}; }
1287 def VLD4DUPd8x2_UPD : VLD4DUPWB<{0,0,1,0}, "8">;
1288 def VLD4DUPd16x2_UPD : VLD4DUPWB<{0,1,1,?}, "16">;
1289 def VLD4DUPd32x2_UPD : VLD4DUPWB<{1,?,1,?}, "32"> { let Inst{6} = Rn{5}; }
1291 def VLD4DUPd8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1292 def VLD4DUPd16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1293 def VLD4DUPd32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1295 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
1297 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
1299 // Classes for VST* pseudo-instructions with multi-register operands.
1300 // These are expanded to real instructions after register allocation.
1301 class VSTQPseudo<InstrItinClass itin>
1302 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src), itin, "">;
1303 class VSTQWBPseudo<InstrItinClass itin>
1304 : PseudoNLdSt<(outs GPR:$wb),
1305 (ins addrmode6:$addr, am6offset:$offset, QPR:$src), itin,
1306 "$addr.addr = $wb">;
1307 class VSTQWBfixedPseudo<InstrItinClass itin>
1308 : PseudoNLdSt<(outs GPR:$wb),
1309 (ins addrmode6:$addr, QPR:$src), itin,
1310 "$addr.addr = $wb">;
1311 class VSTQWBregisterPseudo<InstrItinClass itin>
1312 : PseudoNLdSt<(outs GPR:$wb),
1313 (ins addrmode6:$addr, rGPR:$offset, QPR:$src), itin,
1314 "$addr.addr = $wb">;
1315 class VSTQQPseudo<InstrItinClass itin>
1316 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src), itin, "">;
1317 class VSTQQWBPseudo<InstrItinClass itin>
1318 : PseudoNLdSt<(outs GPR:$wb),
1319 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src), itin,
1320 "$addr.addr = $wb">;
1321 class VSTQQQQPseudo<InstrItinClass itin>
1322 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src), itin, "">;
1323 class VSTQQQQWBPseudo<InstrItinClass itin>
1324 : PseudoNLdSt<(outs GPR:$wb),
1325 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
1326 "$addr.addr = $wb">;
1328 // VST1 : Vector Store (multiple single elements)
1329 class VST1D<bits<4> op7_4, string Dt>
1330 : NLdSt<0,0b00,0b0111,op7_4, (outs), (ins addrmode6:$Rn, VecListOneD:$Vd),
1331 IIC_VST1, "vst1", Dt, "$Vd, $Rn", "", []> {
1333 let Inst{4} = Rn{4};
1334 let DecoderMethod = "DecodeVSTInstruction";
1336 class VST1Q<bits<4> op7_4, string Dt>
1337 : NLdSt<0,0b00,0b1010,op7_4, (outs), (ins addrmode6:$Rn, VecListTwoD:$Vd),
1338 IIC_VST1x2, "vst1", Dt, "$Vd, $Rn", "", []> {
1340 let Inst{5-4} = Rn{5-4};
1341 let DecoderMethod = "DecodeVSTInstruction";
1344 def VST1d8 : VST1D<{0,0,0,?}, "8">;
1345 def VST1d16 : VST1D<{0,1,0,?}, "16">;
1346 def VST1d32 : VST1D<{1,0,0,?}, "32">;
1347 def VST1d64 : VST1D<{1,1,0,?}, "64">;
1349 def VST1q8 : VST1Q<{0,0,?,?}, "8">;
1350 def VST1q16 : VST1Q<{0,1,?,?}, "16">;
1351 def VST1q32 : VST1Q<{1,0,?,?}, "32">;
1352 def VST1q64 : VST1Q<{1,1,?,?}, "64">;
1354 def VST1q8Pseudo : VSTQPseudo<IIC_VST1x2>;
1355 def VST1q16Pseudo : VSTQPseudo<IIC_VST1x2>;
1356 def VST1q32Pseudo : VSTQPseudo<IIC_VST1x2>;
1357 def VST1q64Pseudo : VSTQPseudo<IIC_VST1x2>;
1359 // ...with address register writeback:
1360 multiclass VST1DWB<bits<4> op7_4, string Dt> {
1361 def _fixed : NLdSt<0,0b00, 0b0111,op7_4, (outs GPR:$wb),
1362 (ins addrmode6:$Rn, VecListOneD:$Vd), IIC_VLD1u,
1363 "vst1", Dt, "$Vd, $Rn!",
1364 "$Rn.addr = $wb", []> {
1365 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1366 let Inst{4} = Rn{4};
1367 let DecoderMethod = "DecodeVSTInstruction";
1368 let AsmMatchConverter = "cvtVSTwbFixed";
1370 def _register : NLdSt<0,0b00,0b0111,op7_4, (outs GPR:$wb),
1371 (ins addrmode6:$Rn, rGPR:$Rm, VecListOneD:$Vd),
1373 "vst1", Dt, "$Vd, $Rn, $Rm",
1374 "$Rn.addr = $wb", []> {
1375 let Inst{4} = Rn{4};
1376 let DecoderMethod = "DecodeVSTInstruction";
1377 let AsmMatchConverter = "cvtVSTwbRegister";
1380 multiclass VST1QWB<bits<4> op7_4, string Dt> {
1381 def _fixed : NLdSt<0,0b00,0b1010,op7_4, (outs GPR:$wb),
1382 (ins addrmode6:$Rn, VecListTwoD:$Vd), IIC_VLD1x2u,
1383 "vst1", Dt, "$Vd, $Rn!",
1384 "$Rn.addr = $wb", []> {
1385 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1386 let Inst{5-4} = Rn{5-4};
1387 let DecoderMethod = "DecodeVSTInstruction";
1388 let AsmMatchConverter = "cvtVSTwbFixed";
1390 def _register : NLdSt<0,0b00,0b1010,op7_4, (outs GPR:$wb),
1391 (ins addrmode6:$Rn, rGPR:$Rm, VecListTwoD:$Vd),
1393 "vst1", Dt, "$Vd, $Rn, $Rm",
1394 "$Rn.addr = $wb", []> {
1395 let Inst{5-4} = Rn{5-4};
1396 let DecoderMethod = "DecodeVSTInstruction";
1397 let AsmMatchConverter = "cvtVSTwbRegister";
1401 defm VST1d8wb : VST1DWB<{0,0,0,?}, "8">;
1402 defm VST1d16wb : VST1DWB<{0,1,0,?}, "16">;
1403 defm VST1d32wb : VST1DWB<{1,0,0,?}, "32">;
1404 defm VST1d64wb : VST1DWB<{1,1,0,?}, "64">;
1406 defm VST1q8wb : VST1QWB<{0,0,?,?}, "8">;
1407 defm VST1q16wb : VST1QWB<{0,1,?,?}, "16">;
1408 defm VST1q32wb : VST1QWB<{1,0,?,?}, "32">;
1409 defm VST1q64wb : VST1QWB<{1,1,?,?}, "64">;
1411 def VST1q8PseudoWB_fixed : VSTQWBfixedPseudo<IIC_VST1x2u>;
1412 def VST1q16PseudoWB_fixed : VSTQWBfixedPseudo<IIC_VST1x2u>;
1413 def VST1q32PseudoWB_fixed : VSTQWBfixedPseudo<IIC_VST1x2u>;
1414 def VST1q64PseudoWB_fixed : VSTQWBfixedPseudo<IIC_VST1x2u>;
1415 def VST1q8PseudoWB_register : VSTQWBregisterPseudo<IIC_VST1x2u>;
1416 def VST1q16PseudoWB_register : VSTQWBregisterPseudo<IIC_VST1x2u>;
1417 def VST1q32PseudoWB_register : VSTQWBregisterPseudo<IIC_VST1x2u>;
1418 def VST1q64PseudoWB_register : VSTQWBregisterPseudo<IIC_VST1x2u>;
1420 // ...with 3 registers
1421 class VST1D3<bits<4> op7_4, string Dt>
1422 : NLdSt<0, 0b00, 0b0110, op7_4, (outs),
1423 (ins addrmode6:$Rn, VecListThreeD:$Vd),
1424 IIC_VST1x3, "vst1", Dt, "$Vd, $Rn", "", []> {
1426 let Inst{4} = Rn{4};
1427 let DecoderMethod = "DecodeVSTInstruction";
1429 multiclass VST1D3WB<bits<4> op7_4, string Dt> {
1430 def _fixed : NLdSt<0,0b00,0b0110,op7_4, (outs GPR:$wb),
1431 (ins addrmode6:$Rn, VecListThreeD:$Vd), IIC_VLD1x3u,
1432 "vst1", Dt, "$Vd, $Rn!",
1433 "$Rn.addr = $wb", []> {
1434 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1435 let Inst{5-4} = Rn{5-4};
1436 let DecoderMethod = "DecodeVSTInstruction";
1437 let AsmMatchConverter = "cvtVSTwbFixed";
1439 def _register : NLdSt<0,0b00,0b0110,op7_4, (outs GPR:$wb),
1440 (ins addrmode6:$Rn, rGPR:$Rm, VecListThreeD:$Vd),
1442 "vst1", Dt, "$Vd, $Rn, $Rm",
1443 "$Rn.addr = $wb", []> {
1444 let Inst{5-4} = Rn{5-4};
1445 let DecoderMethod = "DecodeVSTInstruction";
1446 let AsmMatchConverter = "cvtVSTwbRegister";
1450 def VST1d8T : VST1D3<{0,0,0,?}, "8">;
1451 def VST1d16T : VST1D3<{0,1,0,?}, "16">;
1452 def VST1d32T : VST1D3<{1,0,0,?}, "32">;
1453 def VST1d64T : VST1D3<{1,1,0,?}, "64">;
1455 defm VST1d8Twb : VST1D3WB<{0,0,0,?}, "8">;
1456 defm VST1d16Twb : VST1D3WB<{0,1,0,?}, "16">;
1457 defm VST1d32Twb : VST1D3WB<{1,0,0,?}, "32">;
1458 defm VST1d64Twb : VST1D3WB<{1,1,0,?}, "64">;
1460 def VST1d64TPseudo : VSTQQPseudo<IIC_VST1x3>;
1461 def VST1d64TPseudoWB_fixed : VSTQQWBPseudo<IIC_VST1x3u>;
1462 def VST1d64TPseudoWB_register : VSTQQWBPseudo<IIC_VST1x3u>;
1464 // ...with 4 registers
1465 class VST1D4<bits<4> op7_4, string Dt>
1466 : NLdSt<0, 0b00, 0b0010, op7_4, (outs),
1467 (ins addrmode6:$Rn, VecListFourD:$Vd),
1468 IIC_VST1x4, "vst1", Dt, "$Vd, $Rn", "",
1471 let Inst{5-4} = Rn{5-4};
1472 let DecoderMethod = "DecodeVSTInstruction";
1474 multiclass VST1D4WB<bits<4> op7_4, string Dt> {
1475 def _fixed : NLdSt<0,0b00,0b0010,op7_4, (outs GPR:$wb),
1476 (ins addrmode6:$Rn, VecListFourD:$Vd), IIC_VLD1x4u,
1477 "vst1", Dt, "$Vd, $Rn!",
1478 "$Rn.addr = $wb", []> {
1479 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1480 let Inst{5-4} = Rn{5-4};
1481 let DecoderMethod = "DecodeVSTInstruction";
1482 let AsmMatchConverter = "cvtVSTwbFixed";
1484 def _register : NLdSt<0,0b00,0b0010,op7_4, (outs GPR:$wb),
1485 (ins addrmode6:$Rn, rGPR:$Rm, VecListFourD:$Vd),
1487 "vst1", Dt, "$Vd, $Rn, $Rm",
1488 "$Rn.addr = $wb", []> {
1489 let Inst{5-4} = Rn{5-4};
1490 let DecoderMethod = "DecodeVSTInstruction";
1491 let AsmMatchConverter = "cvtVSTwbRegister";
1495 def VST1d8Q : VST1D4<{0,0,?,?}, "8">;
1496 def VST1d16Q : VST1D4<{0,1,?,?}, "16">;
1497 def VST1d32Q : VST1D4<{1,0,?,?}, "32">;
1498 def VST1d64Q : VST1D4<{1,1,?,?}, "64">;
1500 defm VST1d8Qwb : VST1D4WB<{0,0,?,?}, "8">;
1501 defm VST1d16Qwb : VST1D4WB<{0,1,?,?}, "16">;
1502 defm VST1d32Qwb : VST1D4WB<{1,0,?,?}, "32">;
1503 defm VST1d64Qwb : VST1D4WB<{1,1,?,?}, "64">;
1505 def VST1d64QPseudo : VSTQQPseudo<IIC_VST1x4>;
1506 def VST1d64QPseudoWB_fixed : VSTQQWBPseudo<IIC_VST1x4u>;
1507 def VST1d64QPseudoWB_register : VSTQQWBPseudo<IIC_VST1x4u>;
1509 // VST2 : Vector Store (multiple 2-element structures)
1510 class VST2<bits<4> op11_8, bits<4> op7_4, string Dt, RegisterOperand VdTy,
1511 InstrItinClass itin>
1512 : NLdSt<0, 0b00, op11_8, op7_4, (outs), (ins addrmode6:$Rn, VdTy:$Vd),
1513 itin, "vst2", Dt, "$Vd, $Rn", "", []> {
1515 let Inst{5-4} = Rn{5-4};
1516 let DecoderMethod = "DecodeVSTInstruction";
1519 def VST2d8 : VST2<0b1000, {0,0,?,?}, "8", VecListTwoD, IIC_VST2>;
1520 def VST2d16 : VST2<0b1000, {0,1,?,?}, "16", VecListTwoD, IIC_VST2>;
1521 def VST2d32 : VST2<0b1000, {1,0,?,?}, "32", VecListTwoD, IIC_VST2>;
1523 def VST2q8 : VST2<0b0011, {0,0,?,?}, "8", VecListFourD, IIC_VST2x2>;
1524 def VST2q16 : VST2<0b0011, {0,1,?,?}, "16", VecListFourD, IIC_VST2x2>;
1525 def VST2q32 : VST2<0b0011, {1,0,?,?}, "32", VecListFourD, IIC_VST2x2>;
1527 def VST2d8Pseudo : VSTQPseudo<IIC_VST2>;
1528 def VST2d16Pseudo : VSTQPseudo<IIC_VST2>;
1529 def VST2d32Pseudo : VSTQPseudo<IIC_VST2>;
1531 def VST2q8Pseudo : VSTQQPseudo<IIC_VST2x2>;
1532 def VST2q16Pseudo : VSTQQPseudo<IIC_VST2x2>;
1533 def VST2q32Pseudo : VSTQQPseudo<IIC_VST2x2>;
1535 // ...with address register writeback:
1536 multiclass VST2DWB<bits<4> op11_8, bits<4> op7_4, string Dt,
1537 RegisterOperand VdTy> {
1538 def _fixed : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1539 (ins addrmode6:$Rn, VdTy:$Vd), IIC_VLD1u,
1540 "vst2", Dt, "$Vd, $Rn!",
1541 "$Rn.addr = $wb", []> {
1542 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1543 let Inst{5-4} = Rn{5-4};
1544 let DecoderMethod = "DecodeVSTInstruction";
1545 let AsmMatchConverter = "cvtVSTwbFixed";
1547 def _register : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1548 (ins addrmode6:$Rn, rGPR:$Rm, VdTy:$Vd), IIC_VLD1u,
1549 "vst2", Dt, "$Vd, $Rn, $Rm",
1550 "$Rn.addr = $wb", []> {
1551 let Inst{5-4} = Rn{5-4};
1552 let DecoderMethod = "DecodeVSTInstruction";
1553 let AsmMatchConverter = "cvtVSTwbRegister";
1556 multiclass VST2QWB<bits<4> op7_4, string Dt> {
1557 def _fixed : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
1558 (ins addrmode6:$Rn, VecListFourD:$Vd), IIC_VLD1u,
1559 "vst2", Dt, "$Vd, $Rn!",
1560 "$Rn.addr = $wb", []> {
1561 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1562 let Inst{5-4} = Rn{5-4};
1563 let DecoderMethod = "DecodeVSTInstruction";
1564 let AsmMatchConverter = "cvtVSTwbFixed";
1566 def _register : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
1567 (ins addrmode6:$Rn, rGPR:$Rm, VecListFourD:$Vd),
1569 "vst2", Dt, "$Vd, $Rn, $Rm",
1570 "$Rn.addr = $wb", []> {
1571 let Inst{5-4} = Rn{5-4};
1572 let DecoderMethod = "DecodeVSTInstruction";
1573 let AsmMatchConverter = "cvtVSTwbRegister";
1577 defm VST2d8wb : VST2DWB<0b1000, {0,0,?,?}, "8", VecListTwoD>;
1578 defm VST2d16wb : VST2DWB<0b1000, {0,1,?,?}, "16", VecListTwoD>;
1579 defm VST2d32wb : VST2DWB<0b1000, {1,0,?,?}, "32", VecListTwoD>;
1581 defm VST2q8wb : VST2QWB<{0,0,?,?}, "8">;
1582 defm VST2q16wb : VST2QWB<{0,1,?,?}, "16">;
1583 defm VST2q32wb : VST2QWB<{1,0,?,?}, "32">;
1585 def VST2d8PseudoWB_fixed : VSTQWBPseudo<IIC_VST2u>;
1586 def VST2d16PseudoWB_fixed : VSTQWBPseudo<IIC_VST2u>;
1587 def VST2d32PseudoWB_fixed : VSTQWBPseudo<IIC_VST2u>;
1588 def VST2d8PseudoWB_register : VSTQWBPseudo<IIC_VST2u>;
1589 def VST2d16PseudoWB_register : VSTQWBPseudo<IIC_VST2u>;
1590 def VST2d32PseudoWB_register : VSTQWBPseudo<IIC_VST2u>;
1592 def VST2q8PseudoWB_fixed : VSTQQWBPseudo<IIC_VST2x2u>;
1593 def VST2q16PseudoWB_fixed : VSTQQWBPseudo<IIC_VST2x2u>;
1594 def VST2q32PseudoWB_fixed : VSTQQWBPseudo<IIC_VST2x2u>;
1595 def VST2q8PseudoWB_register : VSTQQWBPseudo<IIC_VST2x2u>;
1596 def VST2q16PseudoWB_register : VSTQQWBPseudo<IIC_VST2x2u>;
1597 def VST2q32PseudoWB_register : VSTQQWBPseudo<IIC_VST2x2u>;
1599 // ...with double-spaced registers
1600 def VST2b8 : VST2<0b1001, {0,0,?,?}, "8", VecListTwoQ, IIC_VST2>;
1601 def VST2b16 : VST2<0b1001, {0,1,?,?}, "16", VecListTwoQ, IIC_VST2>;
1602 def VST2b32 : VST2<0b1001, {1,0,?,?}, "32", VecListTwoQ, IIC_VST2>;
1603 defm VST2b8wb : VST2DWB<0b1001, {0,0,?,?}, "8", VecListTwoQ>;
1604 defm VST2b16wb : VST2DWB<0b1001, {0,1,?,?}, "16", VecListTwoQ>;
1605 defm VST2b32wb : VST2DWB<0b1001, {1,0,?,?}, "32", VecListTwoQ>;
1607 // VST3 : Vector Store (multiple 3-element structures)
1608 class VST3D<bits<4> op11_8, bits<4> op7_4, string Dt>
1609 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1610 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3,
1611 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn", "", []> {
1613 let Inst{4} = Rn{4};
1614 let DecoderMethod = "DecodeVSTInstruction";
1617 def VST3d8 : VST3D<0b0100, {0,0,0,?}, "8">;
1618 def VST3d16 : VST3D<0b0100, {0,1,0,?}, "16">;
1619 def VST3d32 : VST3D<0b0100, {1,0,0,?}, "32">;
1621 def VST3d8Pseudo : VSTQQPseudo<IIC_VST3>;
1622 def VST3d16Pseudo : VSTQQPseudo<IIC_VST3>;
1623 def VST3d32Pseudo : VSTQQPseudo<IIC_VST3>;
1625 // ...with address register writeback:
1626 class VST3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1627 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1628 (ins addrmode6:$Rn, am6offset:$Rm,
1629 DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3u,
1630 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn$Rm",
1631 "$Rn.addr = $wb", []> {
1632 let Inst{4} = Rn{4};
1633 let DecoderMethod = "DecodeVSTInstruction";
1636 def VST3d8_UPD : VST3DWB<0b0100, {0,0,0,?}, "8">;
1637 def VST3d16_UPD : VST3DWB<0b0100, {0,1,0,?}, "16">;
1638 def VST3d32_UPD : VST3DWB<0b0100, {1,0,0,?}, "32">;
1640 def VST3d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1641 def VST3d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1642 def VST3d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1644 // ...with double-spaced registers:
1645 def VST3q8 : VST3D<0b0101, {0,0,0,?}, "8">;
1646 def VST3q16 : VST3D<0b0101, {0,1,0,?}, "16">;
1647 def VST3q32 : VST3D<0b0101, {1,0,0,?}, "32">;
1648 def VST3q8_UPD : VST3DWB<0b0101, {0,0,0,?}, "8">;
1649 def VST3q16_UPD : VST3DWB<0b0101, {0,1,0,?}, "16">;
1650 def VST3q32_UPD : VST3DWB<0b0101, {1,0,0,?}, "32">;
1652 def VST3q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1653 def VST3q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1654 def VST3q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1656 // ...alternate versions to be allocated odd register numbers:
1657 def VST3q8oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1658 def VST3q16oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1659 def VST3q32oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1661 def VST3q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1662 def VST3q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1663 def VST3q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1665 // VST4 : Vector Store (multiple 4-element structures)
1666 class VST4D<bits<4> op11_8, bits<4> op7_4, string Dt>
1667 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1668 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
1669 IIC_VST4, "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn",
1672 let Inst{5-4} = Rn{5-4};
1673 let DecoderMethod = "DecodeVSTInstruction";
1676 def VST4d8 : VST4D<0b0000, {0,0,?,?}, "8">;
1677 def VST4d16 : VST4D<0b0000, {0,1,?,?}, "16">;
1678 def VST4d32 : VST4D<0b0000, {1,0,?,?}, "32">;
1680 def VST4d8Pseudo : VSTQQPseudo<IIC_VST4>;
1681 def VST4d16Pseudo : VSTQQPseudo<IIC_VST4>;
1682 def VST4d32Pseudo : VSTQQPseudo<IIC_VST4>;
1684 // ...with address register writeback:
1685 class VST4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1686 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1687 (ins addrmode6:$Rn, am6offset:$Rm,
1688 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST4u,
1689 "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
1690 "$Rn.addr = $wb", []> {
1691 let Inst{5-4} = Rn{5-4};
1692 let DecoderMethod = "DecodeVSTInstruction";
1695 def VST4d8_UPD : VST4DWB<0b0000, {0,0,?,?}, "8">;
1696 def VST4d16_UPD : VST4DWB<0b0000, {0,1,?,?}, "16">;
1697 def VST4d32_UPD : VST4DWB<0b0000, {1,0,?,?}, "32">;
1699 def VST4d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1700 def VST4d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1701 def VST4d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1703 // ...with double-spaced registers:
1704 def VST4q8 : VST4D<0b0001, {0,0,?,?}, "8">;
1705 def VST4q16 : VST4D<0b0001, {0,1,?,?}, "16">;
1706 def VST4q32 : VST4D<0b0001, {1,0,?,?}, "32">;
1707 def VST4q8_UPD : VST4DWB<0b0001, {0,0,?,?}, "8">;
1708 def VST4q16_UPD : VST4DWB<0b0001, {0,1,?,?}, "16">;
1709 def VST4q32_UPD : VST4DWB<0b0001, {1,0,?,?}, "32">;
1711 def VST4q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1712 def VST4q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1713 def VST4q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1715 // ...alternate versions to be allocated odd register numbers:
1716 def VST4q8oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1717 def VST4q16oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1718 def VST4q32oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1720 def VST4q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1721 def VST4q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1722 def VST4q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1724 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
1726 // Classes for VST*LN pseudo-instructions with multi-register operands.
1727 // These are expanded to real instructions after register allocation.
1728 class VSTQLNPseudo<InstrItinClass itin>
1729 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
1731 class VSTQLNWBPseudo<InstrItinClass itin>
1732 : PseudoNLdSt<(outs GPR:$wb),
1733 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
1734 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1735 class VSTQQLNPseudo<InstrItinClass itin>
1736 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
1738 class VSTQQLNWBPseudo<InstrItinClass itin>
1739 : PseudoNLdSt<(outs GPR:$wb),
1740 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
1741 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1742 class VSTQQQQLNPseudo<InstrItinClass itin>
1743 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
1745 class VSTQQQQLNWBPseudo<InstrItinClass itin>
1746 : PseudoNLdSt<(outs GPR:$wb),
1747 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
1748 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1750 // VST1LN : Vector Store (single element from one lane)
1751 class VST1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1752 PatFrag StoreOp, SDNode ExtractOp>
1753 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1754 (ins addrmode6:$Rn, DPR:$Vd, nohash_imm:$lane),
1755 IIC_VST1ln, "vst1", Dt, "\\{$Vd[$lane]\\}, $Rn", "",
1756 [(StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane), addrmode6:$Rn)]> {
1758 let DecoderMethod = "DecodeVST1LN";
1760 class VST1LN32<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1761 PatFrag StoreOp, SDNode ExtractOp>
1762 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1763 (ins addrmode6oneL32:$Rn, DPR:$Vd, nohash_imm:$lane),
1764 IIC_VST1ln, "vst1", Dt, "\\{$Vd[$lane]\\}, $Rn", "",
1765 [(StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane), addrmode6oneL32:$Rn)]>{
1767 let DecoderMethod = "DecodeVST1LN";
1769 class VST1QLNPseudo<ValueType Ty, PatFrag StoreOp, SDNode ExtractOp>
1770 : VSTQLNPseudo<IIC_VST1ln> {
1771 let Pattern = [(StoreOp (ExtractOp (Ty QPR:$src), imm:$lane),
1775 def VST1LNd8 : VST1LN<0b0000, {?,?,?,0}, "8", v8i8, truncstorei8,
1777 let Inst{7-5} = lane{2-0};
1779 def VST1LNd16 : VST1LN<0b0100, {?,?,0,?}, "16", v4i16, truncstorei16,
1781 let Inst{7-6} = lane{1-0};
1782 let Inst{4} = Rn{5};
1785 def VST1LNd32 : VST1LN32<0b1000, {?,0,?,?}, "32", v2i32, store, extractelt> {
1786 let Inst{7} = lane{0};
1787 let Inst{5-4} = Rn{5-4};
1790 def VST1LNq8Pseudo : VST1QLNPseudo<v16i8, truncstorei8, NEONvgetlaneu>;
1791 def VST1LNq16Pseudo : VST1QLNPseudo<v8i16, truncstorei16, NEONvgetlaneu>;
1792 def VST1LNq32Pseudo : VST1QLNPseudo<v4i32, store, extractelt>;
1794 def : Pat<(store (extractelt (v2f32 DPR:$src), imm:$lane), addrmode6:$addr),
1795 (VST1LNd32 addrmode6:$addr, DPR:$src, imm:$lane)>;
1796 def : Pat<(store (extractelt (v4f32 QPR:$src), imm:$lane), addrmode6:$addr),
1797 (VST1LNq32Pseudo addrmode6:$addr, QPR:$src, imm:$lane)>;
1799 // ...with address register writeback:
1800 class VST1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1801 PatFrag StoreOp, SDNode ExtractOp>
1802 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1803 (ins addrmode6:$Rn, am6offset:$Rm,
1804 DPR:$Vd, nohash_imm:$lane), IIC_VST1lnu, "vst1", Dt,
1805 "\\{$Vd[$lane]\\}, $Rn$Rm",
1807 [(set GPR:$wb, (StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane),
1808 addrmode6:$Rn, am6offset:$Rm))]> {
1809 let DecoderMethod = "DecodeVST1LN";
1811 class VST1QLNWBPseudo<ValueType Ty, PatFrag StoreOp, SDNode ExtractOp>
1812 : VSTQLNWBPseudo<IIC_VST1lnu> {
1813 let Pattern = [(set GPR:$wb, (StoreOp (ExtractOp (Ty QPR:$src), imm:$lane),
1814 addrmode6:$addr, am6offset:$offset))];
1817 def VST1LNd8_UPD : VST1LNWB<0b0000, {?,?,?,0}, "8", v8i8, post_truncsti8,
1819 let Inst{7-5} = lane{2-0};
1821 def VST1LNd16_UPD : VST1LNWB<0b0100, {?,?,0,?}, "16", v4i16, post_truncsti16,
1823 let Inst{7-6} = lane{1-0};
1824 let Inst{4} = Rn{5};
1826 def VST1LNd32_UPD : VST1LNWB<0b1000, {?,0,?,?}, "32", v2i32, post_store,
1828 let Inst{7} = lane{0};
1829 let Inst{5-4} = Rn{5-4};
1832 def VST1LNq8Pseudo_UPD : VST1QLNWBPseudo<v16i8, post_truncsti8, NEONvgetlaneu>;
1833 def VST1LNq16Pseudo_UPD : VST1QLNWBPseudo<v8i16, post_truncsti16,NEONvgetlaneu>;
1834 def VST1LNq32Pseudo_UPD : VST1QLNWBPseudo<v4i32, post_store, extractelt>;
1836 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
1838 // VST2LN : Vector Store (single 2-element structure from one lane)
1839 class VST2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1840 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1841 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, nohash_imm:$lane),
1842 IIC_VST2ln, "vst2", Dt, "\\{$Vd[$lane], $src2[$lane]\\}, $Rn",
1845 let Inst{4} = Rn{4};
1846 let DecoderMethod = "DecodeVST2LN";
1849 def VST2LNd8 : VST2LN<0b0001, {?,?,?,?}, "8"> {
1850 let Inst{7-5} = lane{2-0};
1852 def VST2LNd16 : VST2LN<0b0101, {?,?,0,?}, "16"> {
1853 let Inst{7-6} = lane{1-0};
1855 def VST2LNd32 : VST2LN<0b1001, {?,0,0,?}, "32"> {
1856 let Inst{7} = lane{0};
1859 def VST2LNd8Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1860 def VST2LNd16Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1861 def VST2LNd32Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1863 // ...with double-spaced registers:
1864 def VST2LNq16 : VST2LN<0b0101, {?,?,1,?}, "16"> {
1865 let Inst{7-6} = lane{1-0};
1866 let Inst{4} = Rn{4};
1868 def VST2LNq32 : VST2LN<0b1001, {?,1,0,?}, "32"> {
1869 let Inst{7} = lane{0};
1870 let Inst{4} = Rn{4};
1873 def VST2LNq16Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
1874 def VST2LNq32Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
1876 // ...with address register writeback:
1877 class VST2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1878 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1879 (ins addrmode6:$Rn, am6offset:$Rm,
1880 DPR:$Vd, DPR:$src2, nohash_imm:$lane), IIC_VST2lnu, "vst2", Dt,
1881 "\\{$Vd[$lane], $src2[$lane]\\}, $Rn$Rm",
1882 "$Rn.addr = $wb", []> {
1883 let Inst{4} = Rn{4};
1884 let DecoderMethod = "DecodeVST2LN";
1887 def VST2LNd8_UPD : VST2LNWB<0b0001, {?,?,?,?}, "8"> {
1888 let Inst{7-5} = lane{2-0};
1890 def VST2LNd16_UPD : VST2LNWB<0b0101, {?,?,0,?}, "16"> {
1891 let Inst{7-6} = lane{1-0};
1893 def VST2LNd32_UPD : VST2LNWB<0b1001, {?,0,0,?}, "32"> {
1894 let Inst{7} = lane{0};
1897 def VST2LNd8Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1898 def VST2LNd16Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1899 def VST2LNd32Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1901 def VST2LNq16_UPD : VST2LNWB<0b0101, {?,?,1,?}, "16"> {
1902 let Inst{7-6} = lane{1-0};
1904 def VST2LNq32_UPD : VST2LNWB<0b1001, {?,1,0,?}, "32"> {
1905 let Inst{7} = lane{0};
1908 def VST2LNq16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
1909 def VST2LNq32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
1911 // VST3LN : Vector Store (single 3-element structure from one lane)
1912 class VST3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1913 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1914 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3,
1915 nohash_imm:$lane), IIC_VST3ln, "vst3", Dt,
1916 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn", "", []> {
1918 let DecoderMethod = "DecodeVST3LN";
1921 def VST3LNd8 : VST3LN<0b0010, {?,?,?,0}, "8"> {
1922 let Inst{7-5} = lane{2-0};
1924 def VST3LNd16 : VST3LN<0b0110, {?,?,0,0}, "16"> {
1925 let Inst{7-6} = lane{1-0};
1927 def VST3LNd32 : VST3LN<0b1010, {?,0,0,0}, "32"> {
1928 let Inst{7} = lane{0};
1931 def VST3LNd8Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1932 def VST3LNd16Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1933 def VST3LNd32Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1935 // ...with double-spaced registers:
1936 def VST3LNq16 : VST3LN<0b0110, {?,?,1,0}, "16"> {
1937 let Inst{7-6} = lane{1-0};
1939 def VST3LNq32 : VST3LN<0b1010, {?,1,0,0}, "32"> {
1940 let Inst{7} = lane{0};
1943 def VST3LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
1944 def VST3LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
1946 // ...with address register writeback:
1947 class VST3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1948 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1949 (ins addrmode6:$Rn, am6offset:$Rm,
1950 DPR:$Vd, DPR:$src2, DPR:$src3, nohash_imm:$lane),
1951 IIC_VST3lnu, "vst3", Dt,
1952 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn$Rm",
1953 "$Rn.addr = $wb", []> {
1954 let DecoderMethod = "DecodeVST3LN";
1957 def VST3LNd8_UPD : VST3LNWB<0b0010, {?,?,?,0}, "8"> {
1958 let Inst{7-5} = lane{2-0};
1960 def VST3LNd16_UPD : VST3LNWB<0b0110, {?,?,0,0}, "16"> {
1961 let Inst{7-6} = lane{1-0};
1963 def VST3LNd32_UPD : VST3LNWB<0b1010, {?,0,0,0}, "32"> {
1964 let Inst{7} = lane{0};
1967 def VST3LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1968 def VST3LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1969 def VST3LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1971 def VST3LNq16_UPD : VST3LNWB<0b0110, {?,?,1,0}, "16"> {
1972 let Inst{7-6} = lane{1-0};
1974 def VST3LNq32_UPD : VST3LNWB<0b1010, {?,1,0,0}, "32"> {
1975 let Inst{7} = lane{0};
1978 def VST3LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
1979 def VST3LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
1981 // VST4LN : Vector Store (single 4-element structure from one lane)
1982 class VST4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1983 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1984 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4,
1985 nohash_imm:$lane), IIC_VST4ln, "vst4", Dt,
1986 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn",
1989 let Inst{4} = Rn{4};
1990 let DecoderMethod = "DecodeVST4LN";
1993 def VST4LNd8 : VST4LN<0b0011, {?,?,?,?}, "8"> {
1994 let Inst{7-5} = lane{2-0};
1996 def VST4LNd16 : VST4LN<0b0111, {?,?,0,?}, "16"> {
1997 let Inst{7-6} = lane{1-0};
1999 def VST4LNd32 : VST4LN<0b1011, {?,0,?,?}, "32"> {
2000 let Inst{7} = lane{0};
2001 let Inst{5} = Rn{5};
2004 def VST4LNd8Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
2005 def VST4LNd16Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
2006 def VST4LNd32Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
2008 // ...with double-spaced registers:
2009 def VST4LNq16 : VST4LN<0b0111, {?,?,1,?}, "16"> {
2010 let Inst{7-6} = lane{1-0};
2012 def VST4LNq32 : VST4LN<0b1011, {?,1,?,?}, "32"> {
2013 let Inst{7} = lane{0};
2014 let Inst{5} = Rn{5};
2017 def VST4LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
2018 def VST4LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
2020 // ...with address register writeback:
2021 class VST4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
2022 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
2023 (ins addrmode6:$Rn, am6offset:$Rm,
2024 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
2025 IIC_VST4lnu, "vst4", Dt,
2026 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn$Rm",
2027 "$Rn.addr = $wb", []> {
2028 let Inst{4} = Rn{4};
2029 let DecoderMethod = "DecodeVST4LN";
2032 def VST4LNd8_UPD : VST4LNWB<0b0011, {?,?,?,?}, "8"> {
2033 let Inst{7-5} = lane{2-0};
2035 def VST4LNd16_UPD : VST4LNWB<0b0111, {?,?,0,?}, "16"> {
2036 let Inst{7-6} = lane{1-0};
2038 def VST4LNd32_UPD : VST4LNWB<0b1011, {?,0,?,?}, "32"> {
2039 let Inst{7} = lane{0};
2040 let Inst{5} = Rn{5};
2043 def VST4LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
2044 def VST4LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
2045 def VST4LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
2047 def VST4LNq16_UPD : VST4LNWB<0b0111, {?,?,1,?}, "16"> {
2048 let Inst{7-6} = lane{1-0};
2050 def VST4LNq32_UPD : VST4LNWB<0b1011, {?,1,?,?}, "32"> {
2051 let Inst{7} = lane{0};
2052 let Inst{5} = Rn{5};
2055 def VST4LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
2056 def VST4LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
2058 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
2061 //===----------------------------------------------------------------------===//
2062 // NEON pattern fragments
2063 //===----------------------------------------------------------------------===//
2065 // Extract D sub-registers of Q registers.
2066 def DSubReg_i8_reg : SDNodeXForm<imm, [{
2067 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
2068 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/8, MVT::i32);
2070 def DSubReg_i16_reg : SDNodeXForm<imm, [{
2071 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
2072 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/4, MVT::i32);
2074 def DSubReg_i32_reg : SDNodeXForm<imm, [{
2075 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
2076 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/2, MVT::i32);
2078 def DSubReg_f64_reg : SDNodeXForm<imm, [{
2079 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
2080 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue(), MVT::i32);
2083 // Extract S sub-registers of Q/D registers.
2084 def SSubReg_f32_reg : SDNodeXForm<imm, [{
2085 assert(ARM::ssub_3 == ARM::ssub_0+3 && "Unexpected subreg numbering");
2086 return CurDAG->getTargetConstant(ARM::ssub_0 + N->getZExtValue(), MVT::i32);
2089 // Translate lane numbers from Q registers to D subregs.
2090 def SubReg_i8_lane : SDNodeXForm<imm, [{
2091 return CurDAG->getTargetConstant(N->getZExtValue() & 7, MVT::i32);
2093 def SubReg_i16_lane : SDNodeXForm<imm, [{
2094 return CurDAG->getTargetConstant(N->getZExtValue() & 3, MVT::i32);
2096 def SubReg_i32_lane : SDNodeXForm<imm, [{
2097 return CurDAG->getTargetConstant(N->getZExtValue() & 1, MVT::i32);
2100 //===----------------------------------------------------------------------===//
2101 // Instruction Classes
2102 //===----------------------------------------------------------------------===//
2104 // Basic 2-register operations: double- and quad-register.
2105 class N2VD<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2106 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
2107 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
2108 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
2109 (ins DPR:$Vm), IIC_VUNAD, OpcodeStr, Dt,"$Vd, $Vm", "",
2110 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vm))))]>;
2111 class N2VQ<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2112 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
2113 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
2114 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
2115 (ins QPR:$Vm), IIC_VUNAQ, OpcodeStr, Dt,"$Vd, $Vm", "",
2116 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vm))))]>;
2118 // Basic 2-register intrinsics, both double- and quad-register.
2119 class N2VDInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2120 bits<2> op17_16, bits<5> op11_7, bit op4,
2121 InstrItinClass itin, string OpcodeStr, string Dt,
2122 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2123 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
2124 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2125 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm))))]>;
2126 class N2VQInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2127 bits<2> op17_16, bits<5> op11_7, bit op4,
2128 InstrItinClass itin, string OpcodeStr, string Dt,
2129 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2130 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
2131 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2132 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm))))]>;
2134 // Narrow 2-register operations.
2135 class N2VN<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2136 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2137 InstrItinClass itin, string OpcodeStr, string Dt,
2138 ValueType TyD, ValueType TyQ, SDNode OpNode>
2139 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$Vd),
2140 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2141 [(set DPR:$Vd, (TyD (OpNode (TyQ QPR:$Vm))))]>;
2143 // Narrow 2-register intrinsics.
2144 class N2VNInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2145 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2146 InstrItinClass itin, string OpcodeStr, string Dt,
2147 ValueType TyD, ValueType TyQ, Intrinsic IntOp>
2148 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$Vd),
2149 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2150 [(set DPR:$Vd, (TyD (IntOp (TyQ QPR:$Vm))))]>;
2152 // Long 2-register operations (currently only used for VMOVL).
2153 class N2VL<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2154 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2155 InstrItinClass itin, string OpcodeStr, string Dt,
2156 ValueType TyQ, ValueType TyD, SDNode OpNode>
2157 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$Vd),
2158 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2159 [(set QPR:$Vd, (TyQ (OpNode (TyD DPR:$Vm))))]>;
2161 // Long 2-register intrinsics.
2162 class N2VLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2163 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2164 InstrItinClass itin, string OpcodeStr, string Dt,
2165 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
2166 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$Vd),
2167 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2168 [(set QPR:$Vd, (TyQ (IntOp (TyD DPR:$Vm))))]>;
2170 // 2-register shuffles (VTRN/VZIP/VUZP), both double- and quad-register.
2171 class N2VDShuffle<bits<2> op19_18, bits<5> op11_7, string OpcodeStr, string Dt>
2172 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 0, 0, (outs DPR:$Vd, DPR:$Vm),
2173 (ins DPR:$src1, DPR:$src2), IIC_VPERMD,
2174 OpcodeStr, Dt, "$Vd, $Vm",
2175 "$src1 = $Vd, $src2 = $Vm", []>;
2176 class N2VQShuffle<bits<2> op19_18, bits<5> op11_7,
2177 InstrItinClass itin, string OpcodeStr, string Dt>
2178 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 1, 0, (outs QPR:$Vd, QPR:$Vm),
2179 (ins QPR:$src1, QPR:$src2), itin, OpcodeStr, Dt, "$Vd, $Vm",
2180 "$src1 = $Vd, $src2 = $Vm", []>;
2182 // Basic 3-register operations: double- and quad-register.
2183 class N3VD<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2184 InstrItinClass itin, string OpcodeStr, string Dt,
2185 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
2186 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2187 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2188 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2189 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
2190 let isCommutable = Commutable;
2192 // Same as N3VD but no data type.
2193 class N3VDX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2194 InstrItinClass itin, string OpcodeStr,
2195 ValueType ResTy, ValueType OpTy,
2196 SDNode OpNode, bit Commutable>
2197 : N3VX<op24, op23, op21_20, op11_8, 0, op4,
2198 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2199 OpcodeStr, "$Vd, $Vn, $Vm", "",
2200 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>{
2201 let isCommutable = Commutable;
2204 class N3VDSL<bits<2> op21_20, bits<4> op11_8,
2205 InstrItinClass itin, string OpcodeStr, string Dt,
2206 ValueType Ty, SDNode ShOp>
2207 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2208 (outs DPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2209 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2211 (Ty (ShOp (Ty DPR:$Vn),
2212 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),imm:$lane)))))]> {
2213 let isCommutable = 0;
2215 class N3VDSL16<bits<2> op21_20, bits<4> op11_8,
2216 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
2217 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2218 (outs DPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2219 NVMulSLFrm, IIC_VMULi16D, OpcodeStr, Dt,"$Vd, $Vn, $Vm$lane","",
2221 (Ty (ShOp (Ty DPR:$Vn),
2222 (Ty (NEONvduplane (Ty DPR_8:$Vm), imm:$lane)))))]> {
2223 let isCommutable = 0;
2226 class N3VQ<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2227 InstrItinClass itin, string OpcodeStr, string Dt,
2228 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
2229 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2230 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2231 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2232 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
2233 let isCommutable = Commutable;
2235 class N3VQX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2236 InstrItinClass itin, string OpcodeStr,
2237 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
2238 : N3VX<op24, op23, op21_20, op11_8, 1, op4,
2239 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2240 OpcodeStr, "$Vd, $Vn, $Vm", "",
2241 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]>{
2242 let isCommutable = Commutable;
2244 class N3VQSL<bits<2> op21_20, bits<4> op11_8,
2245 InstrItinClass itin, string OpcodeStr, string Dt,
2246 ValueType ResTy, ValueType OpTy, SDNode ShOp>
2247 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2248 (outs QPR:$Vd), (ins QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2249 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2250 [(set (ResTy QPR:$Vd),
2251 (ResTy (ShOp (ResTy QPR:$Vn),
2252 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2254 let isCommutable = 0;
2256 class N3VQSL16<bits<2> op21_20, bits<4> op11_8, string OpcodeStr, string Dt,
2257 ValueType ResTy, ValueType OpTy, SDNode ShOp>
2258 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2259 (outs QPR:$Vd), (ins QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2260 NVMulSLFrm, IIC_VMULi16Q, OpcodeStr, Dt,"$Vd, $Vn, $Vm$lane", "",
2261 [(set (ResTy QPR:$Vd),
2262 (ResTy (ShOp (ResTy QPR:$Vn),
2263 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2265 let isCommutable = 0;
2268 // Basic 3-register intrinsics, both double- and quad-register.
2269 class N3VDInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2270 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2271 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
2272 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2273 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), f, itin,
2274 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2275 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
2276 let isCommutable = Commutable;
2278 class N3VDIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2279 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
2280 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2281 (outs DPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2282 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2284 (Ty (IntOp (Ty DPR:$Vn),
2285 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),
2287 let isCommutable = 0;
2289 class N3VDIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2290 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
2291 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2292 (outs DPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2293 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2295 (Ty (IntOp (Ty DPR:$Vn),
2296 (Ty (NEONvduplane (Ty DPR_8:$Vm), imm:$lane)))))]> {
2297 let isCommutable = 0;
2299 class N3VDIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2300 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2301 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2302 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2303 (outs DPR:$Vd), (ins DPR:$Vm, DPR:$Vn), f, itin,
2304 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
2305 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (OpTy DPR:$Vn))))]> {
2306 let isCommutable = 0;
2309 class N3VQInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2310 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2311 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
2312 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2313 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), f, itin,
2314 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2315 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
2316 let isCommutable = Commutable;
2318 class N3VQIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2319 string OpcodeStr, string Dt,
2320 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2321 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2322 (outs QPR:$Vd), (ins QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2323 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2324 [(set (ResTy QPR:$Vd),
2325 (ResTy (IntOp (ResTy QPR:$Vn),
2326 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2328 let isCommutable = 0;
2330 class N3VQIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2331 string OpcodeStr, string Dt,
2332 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2333 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2334 (outs QPR:$Vd), (ins QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2335 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2336 [(set (ResTy QPR:$Vd),
2337 (ResTy (IntOp (ResTy QPR:$Vn),
2338 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2340 let isCommutable = 0;
2342 class N3VQIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2343 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2344 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2345 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2346 (outs QPR:$Vd), (ins QPR:$Vm, QPR:$Vn), f, itin,
2347 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
2348 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (OpTy QPR:$Vn))))]> {
2349 let isCommutable = 0;
2352 // Multiply-Add/Sub operations: double- and quad-register.
2353 class N3VDMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2354 InstrItinClass itin, string OpcodeStr, string Dt,
2355 ValueType Ty, SDPatternOperator MulOp, SDPatternOperator OpNode>
2356 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2357 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2358 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2359 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
2360 (Ty (MulOp DPR:$Vn, DPR:$Vm)))))]>;
2362 class N3VDMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2363 string OpcodeStr, string Dt,
2364 ValueType Ty, SDPatternOperator MulOp, SDPatternOperator ShOp>
2365 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2367 (ins DPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2369 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2371 (Ty (ShOp (Ty DPR:$src1),
2373 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),
2375 class N3VDMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2376 string OpcodeStr, string Dt,
2377 ValueType Ty, SDNode MulOp, SDNode ShOp>
2378 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2380 (ins DPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2382 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2384 (Ty (ShOp (Ty DPR:$src1),
2386 (Ty (NEONvduplane (Ty DPR_8:$Vm),
2389 class N3VQMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2390 InstrItinClass itin, string OpcodeStr, string Dt, ValueType Ty,
2391 SDPatternOperator MulOp, SDPatternOperator OpNode>
2392 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2393 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2394 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2395 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
2396 (Ty (MulOp QPR:$Vn, QPR:$Vm)))))]>;
2397 class N3VQMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2398 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2399 SDPatternOperator MulOp, SDPatternOperator ShOp>
2400 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2402 (ins QPR:$src1, QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2404 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2405 [(set (ResTy QPR:$Vd),
2406 (ResTy (ShOp (ResTy QPR:$src1),
2407 (ResTy (MulOp QPR:$Vn,
2408 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2410 class N3VQMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2411 string OpcodeStr, string Dt,
2412 ValueType ResTy, ValueType OpTy,
2413 SDNode MulOp, SDNode ShOp>
2414 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2416 (ins QPR:$src1, QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2418 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2419 [(set (ResTy QPR:$Vd),
2420 (ResTy (ShOp (ResTy QPR:$src1),
2421 (ResTy (MulOp QPR:$Vn,
2422 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2425 // Neon Intrinsic-Op instructions (VABA): double- and quad-register.
2426 class N3VDIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2427 InstrItinClass itin, string OpcodeStr, string Dt,
2428 ValueType Ty, Intrinsic IntOp, SDNode OpNode>
2429 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2430 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2431 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2432 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
2433 (Ty (IntOp (Ty DPR:$Vn), (Ty DPR:$Vm))))))]>;
2434 class N3VQIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2435 InstrItinClass itin, string OpcodeStr, string Dt,
2436 ValueType Ty, Intrinsic IntOp, SDNode OpNode>
2437 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2438 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2439 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2440 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
2441 (Ty (IntOp (Ty QPR:$Vn), (Ty QPR:$Vm))))))]>;
2443 // Neon 3-argument intrinsics, both double- and quad-register.
2444 // The destination register is also used as the first source operand register.
2445 class N3VDInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2446 InstrItinClass itin, string OpcodeStr, string Dt,
2447 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2448 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2449 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2450 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2451 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$src1),
2452 (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>;
2453 class N3VQInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2454 InstrItinClass itin, string OpcodeStr, string Dt,
2455 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2456 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2457 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2458 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2459 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$src1),
2460 (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]>;
2462 // Long Multiply-Add/Sub operations.
2463 class N3VLMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2464 InstrItinClass itin, string OpcodeStr, string Dt,
2465 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2466 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2467 (outs QPR:$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:$src1),
2470 (TyQ (MulOp (TyD DPR:$Vn),
2471 (TyD DPR:$Vm)))))]>;
2472 class N3VLMulOpSL<bit op24, bits<2> op21_20, bits<4> op11_8,
2473 InstrItinClass itin, string OpcodeStr, string Dt,
2474 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2475 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$Vd),
2476 (ins QPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2478 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2480 (OpNode (TyQ QPR:$src1),
2481 (TyQ (MulOp (TyD DPR:$Vn),
2482 (TyD (NEONvduplane (TyD DPR_VFP2:$Vm),
2484 class N3VLMulOpSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2485 InstrItinClass itin, string OpcodeStr, string Dt,
2486 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2487 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$Vd),
2488 (ins QPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2490 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2492 (OpNode (TyQ QPR:$src1),
2493 (TyQ (MulOp (TyD DPR:$Vn),
2494 (TyD (NEONvduplane (TyD DPR_8:$Vm),
2497 // Long Intrinsic-Op vector operations with explicit extend (VABAL).
2498 class N3VLIntExtOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2499 InstrItinClass itin, string OpcodeStr, string Dt,
2500 ValueType TyQ, ValueType TyD, Intrinsic IntOp, SDNode ExtOp,
2502 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2503 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2504 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2505 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
2506 (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
2507 (TyD DPR:$Vm)))))))]>;
2509 // Neon Long 3-argument intrinsic. The destination register is
2510 // a quad-register and is also used as the first source operand register.
2511 class N3VLInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2512 InstrItinClass itin, string OpcodeStr, string Dt,
2513 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
2514 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2515 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2516 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2518 (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$Vn), (TyD DPR:$Vm))))]>;
2519 class N3VLInt3SL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2520 string OpcodeStr, string Dt,
2521 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2522 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2524 (ins QPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2526 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2527 [(set (ResTy QPR:$Vd),
2528 (ResTy (IntOp (ResTy QPR:$src1),
2530 (OpTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2532 class N3VLInt3SL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2533 InstrItinClass itin, string OpcodeStr, string Dt,
2534 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2535 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2537 (ins QPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2539 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2540 [(set (ResTy QPR:$Vd),
2541 (ResTy (IntOp (ResTy QPR:$src1),
2543 (OpTy (NEONvduplane (OpTy DPR_8:$Vm),
2546 // Narrowing 3-register intrinsics.
2547 class N3VNInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2548 string OpcodeStr, string Dt, ValueType TyD, ValueType TyQ,
2549 Intrinsic IntOp, bit Commutable>
2550 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2551 (outs DPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINi4D,
2552 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2553 [(set DPR:$Vd, (TyD (IntOp (TyQ QPR:$Vn), (TyQ QPR:$Vm))))]> {
2554 let isCommutable = Commutable;
2557 // Long 3-register operations.
2558 class N3VL<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2559 InstrItinClass itin, string OpcodeStr, string Dt,
2560 ValueType TyQ, ValueType TyD, SDNode OpNode, bit Commutable>
2561 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2562 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2563 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2564 [(set QPR:$Vd, (TyQ (OpNode (TyD DPR:$Vn), (TyD DPR:$Vm))))]> {
2565 let isCommutable = Commutable;
2567 class N3VLSL<bit op24, bits<2> op21_20, bits<4> op11_8,
2568 InstrItinClass itin, string OpcodeStr, string Dt,
2569 ValueType TyQ, ValueType TyD, SDNode OpNode>
2570 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2571 (outs QPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2572 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2574 (TyQ (OpNode (TyD DPR:$Vn),
2575 (TyD (NEONvduplane (TyD DPR_VFP2:$Vm),imm:$lane)))))]>;
2576 class N3VLSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2577 InstrItinClass itin, string OpcodeStr, string Dt,
2578 ValueType TyQ, ValueType TyD, SDNode OpNode>
2579 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2580 (outs QPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2581 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2583 (TyQ (OpNode (TyD DPR:$Vn),
2584 (TyD (NEONvduplane (TyD DPR_8:$Vm), imm:$lane)))))]>;
2586 // Long 3-register operations with explicitly extended operands.
2587 class N3VLExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2588 InstrItinClass itin, string OpcodeStr, string Dt,
2589 ValueType TyQ, ValueType TyD, SDNode OpNode, SDNode ExtOp,
2591 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2592 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2593 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2594 [(set QPR:$Vd, (OpNode (TyQ (ExtOp (TyD DPR:$Vn))),
2595 (TyQ (ExtOp (TyD DPR:$Vm)))))]> {
2596 let isCommutable = Commutable;
2599 // Long 3-register intrinsics with explicit extend (VABDL).
2600 class N3VLIntExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2601 InstrItinClass itin, string OpcodeStr, string Dt,
2602 ValueType TyQ, ValueType TyD, Intrinsic IntOp, SDNode ExtOp,
2604 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2605 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2606 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2607 [(set QPR:$Vd, (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
2608 (TyD DPR:$Vm))))))]> {
2609 let isCommutable = Commutable;
2612 // Long 3-register intrinsics.
2613 class N3VLInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2614 InstrItinClass itin, string OpcodeStr, string Dt,
2615 ValueType TyQ, ValueType TyD, Intrinsic IntOp, bit Commutable>
2616 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2617 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2618 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2619 [(set QPR:$Vd, (TyQ (IntOp (TyD DPR:$Vn), (TyD DPR:$Vm))))]> {
2620 let isCommutable = Commutable;
2622 class N3VLIntSL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2623 string OpcodeStr, string Dt,
2624 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2625 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2626 (outs QPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2627 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2628 [(set (ResTy QPR:$Vd),
2629 (ResTy (IntOp (OpTy DPR:$Vn),
2630 (OpTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2632 class N3VLIntSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2633 InstrItinClass itin, string OpcodeStr, string Dt,
2634 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2635 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2636 (outs QPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2637 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2638 [(set (ResTy QPR:$Vd),
2639 (ResTy (IntOp (OpTy DPR:$Vn),
2640 (OpTy (NEONvduplane (OpTy DPR_8:$Vm),
2643 // Wide 3-register operations.
2644 class N3VW<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2645 string OpcodeStr, string Dt, ValueType TyQ, ValueType TyD,
2646 SDNode OpNode, SDNode ExtOp, bit Commutable>
2647 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2648 (outs QPR:$Vd), (ins QPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VSUBiD,
2649 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2650 [(set QPR:$Vd, (OpNode (TyQ QPR:$Vn),
2651 (TyQ (ExtOp (TyD DPR:$Vm)))))]> {
2652 let isCommutable = Commutable;
2655 // Pairwise long 2-register intrinsics, both double- and quad-register.
2656 class N2VDPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2657 bits<2> op17_16, bits<5> op11_7, bit op4,
2658 string OpcodeStr, string Dt,
2659 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2660 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
2661 (ins DPR:$Vm), IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
2662 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm))))]>;
2663 class N2VQPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2664 bits<2> op17_16, bits<5> op11_7, bit op4,
2665 string OpcodeStr, string Dt,
2666 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2667 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
2668 (ins QPR:$Vm), IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
2669 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm))))]>;
2671 // Pairwise long 2-register accumulate intrinsics,
2672 // both double- and quad-register.
2673 // The destination register is also used as the first source operand register.
2674 class N2VDPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2675 bits<2> op17_16, bits<5> op11_7, bit op4,
2676 string OpcodeStr, string Dt,
2677 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2678 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
2679 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vm), IIC_VPALiD,
2680 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
2681 [(set DPR:$Vd, (ResTy (IntOp (ResTy DPR:$src1), (OpTy DPR:$Vm))))]>;
2682 class N2VQPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2683 bits<2> op17_16, bits<5> op11_7, bit op4,
2684 string OpcodeStr, string Dt,
2685 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2686 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4,
2687 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vm), IIC_VPALiQ,
2688 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
2689 [(set QPR:$Vd, (ResTy (IntOp (ResTy QPR:$src1), (OpTy QPR:$Vm))))]>;
2691 // Shift by immediate,
2692 // both double- and quad-register.
2693 class N2VDSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2694 Format f, InstrItinClass itin, Operand ImmTy,
2695 string OpcodeStr, string Dt, ValueType Ty, SDNode OpNode>
2696 : N2VImm<op24, op23, op11_8, op7, 0, op4,
2697 (outs DPR:$Vd), (ins DPR:$Vm, ImmTy:$SIMM), f, itin,
2698 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2699 [(set DPR:$Vd, (Ty (OpNode (Ty DPR:$Vm), (i32 imm:$SIMM))))]>;
2700 class N2VQSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2701 Format f, InstrItinClass itin, Operand ImmTy,
2702 string OpcodeStr, string Dt, ValueType Ty, SDNode OpNode>
2703 : N2VImm<op24, op23, op11_8, op7, 1, op4,
2704 (outs QPR:$Vd), (ins QPR:$Vm, ImmTy:$SIMM), f, itin,
2705 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2706 [(set QPR:$Vd, (Ty (OpNode (Ty QPR:$Vm), (i32 imm:$SIMM))))]>;
2708 // Long shift by immediate.
2709 class N2VLSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
2710 string OpcodeStr, string Dt,
2711 ValueType ResTy, ValueType OpTy, Operand ImmTy, SDNode OpNode>
2712 : N2VImm<op24, op23, op11_8, op7, op6, op4,
2713 (outs QPR:$Vd), (ins DPR:$Vm, ImmTy:$SIMM), N2RegVShLFrm,
2714 IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2715 [(set QPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vm),
2716 (i32 imm:$SIMM))))]>;
2718 // Narrow shift by immediate.
2719 class N2VNSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
2720 InstrItinClass itin, string OpcodeStr, string Dt,
2721 ValueType ResTy, ValueType OpTy, Operand ImmTy, SDNode OpNode>
2722 : N2VImm<op24, op23, op11_8, op7, op6, op4,
2723 (outs DPR:$Vd), (ins QPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, itin,
2724 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2725 [(set DPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vm),
2726 (i32 imm:$SIMM))))]>;
2728 // Shift right by immediate and accumulate,
2729 // both double- and quad-register.
2730 class N2VDShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2731 Operand ImmTy, string OpcodeStr, string Dt,
2732 ValueType Ty, SDNode ShOp>
2733 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
2734 (ins DPR:$src1, DPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, IIC_VPALiD,
2735 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2736 [(set DPR:$Vd, (Ty (add DPR:$src1,
2737 (Ty (ShOp DPR:$Vm, (i32 imm:$SIMM))))))]>;
2738 class N2VQShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2739 Operand ImmTy, string OpcodeStr, string Dt,
2740 ValueType Ty, SDNode ShOp>
2741 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
2742 (ins QPR:$src1, QPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, IIC_VPALiD,
2743 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2744 [(set QPR:$Vd, (Ty (add QPR:$src1,
2745 (Ty (ShOp QPR:$Vm, (i32 imm:$SIMM))))))]>;
2747 // Shift by immediate and insert,
2748 // both double- and quad-register.
2749 class N2VDShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2750 Operand ImmTy, Format f, string OpcodeStr, string Dt,
2751 ValueType Ty,SDNode ShOp>
2752 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
2753 (ins DPR:$src1, DPR:$Vm, ImmTy:$SIMM), f, IIC_VSHLiD,
2754 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2755 [(set DPR:$Vd, (Ty (ShOp DPR:$src1, DPR:$Vm, (i32 imm:$SIMM))))]>;
2756 class N2VQShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2757 Operand ImmTy, Format f, string OpcodeStr, string Dt,
2758 ValueType Ty,SDNode ShOp>
2759 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
2760 (ins QPR:$src1, QPR:$Vm, ImmTy:$SIMM), f, IIC_VSHLiQ,
2761 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2762 [(set QPR:$Vd, (Ty (ShOp QPR:$src1, QPR:$Vm, (i32 imm:$SIMM))))]>;
2764 // Convert, with fractional bits immediate,
2765 // both double- and quad-register.
2766 class N2VCvtD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2767 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2769 : N2VImm<op24, op23, op11_8, op7, 0, op4,
2770 (outs DPR:$Vd), (ins DPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
2771 IIC_VUNAD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2772 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (i32 imm:$SIMM))))]>;
2773 class N2VCvtQ<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2774 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2776 : N2VImm<op24, op23, op11_8, op7, 1, op4,
2777 (outs QPR:$Vd), (ins QPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
2778 IIC_VUNAQ, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2779 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (i32 imm:$SIMM))))]>;
2781 //===----------------------------------------------------------------------===//
2783 //===----------------------------------------------------------------------===//
2785 // Abbreviations used in multiclass suffixes:
2786 // Q = quarter int (8 bit) elements
2787 // H = half int (16 bit) elements
2788 // S = single int (32 bit) elements
2789 // D = double int (64 bit) elements
2791 // Neon 2-register vector operations and intrinsics.
2793 // Neon 2-register comparisons.
2794 // source operand element sizes of 8, 16 and 32 bits:
2795 multiclass N2V_QHS_cmp<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2796 bits<5> op11_7, bit op4, string opc, string Dt,
2797 string asm, SDNode OpNode> {
2798 // 64-bit vector types.
2799 def v8i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 0, op4,
2800 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2801 opc, !strconcat(Dt, "8"), asm, "",
2802 [(set DPR:$Vd, (v8i8 (OpNode (v8i8 DPR:$Vm))))]>;
2803 def v4i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 0, op4,
2804 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2805 opc, !strconcat(Dt, "16"), asm, "",
2806 [(set DPR:$Vd, (v4i16 (OpNode (v4i16 DPR:$Vm))))]>;
2807 def v2i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
2808 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2809 opc, !strconcat(Dt, "32"), asm, "",
2810 [(set DPR:$Vd, (v2i32 (OpNode (v2i32 DPR:$Vm))))]>;
2811 def v2f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
2812 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2813 opc, "f32", asm, "",
2814 [(set DPR:$Vd, (v2i32 (OpNode (v2f32 DPR:$Vm))))]> {
2815 let Inst{10} = 1; // overwrite F = 1
2818 // 128-bit vector types.
2819 def v16i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 1, op4,
2820 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2821 opc, !strconcat(Dt, "8"), asm, "",
2822 [(set QPR:$Vd, (v16i8 (OpNode (v16i8 QPR:$Vm))))]>;
2823 def v8i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 1, op4,
2824 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2825 opc, !strconcat(Dt, "16"), asm, "",
2826 [(set QPR:$Vd, (v8i16 (OpNode (v8i16 QPR:$Vm))))]>;
2827 def v4i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
2828 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2829 opc, !strconcat(Dt, "32"), asm, "",
2830 [(set QPR:$Vd, (v4i32 (OpNode (v4i32 QPR:$Vm))))]>;
2831 def v4f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
2832 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2833 opc, "f32", asm, "",
2834 [(set QPR:$Vd, (v4i32 (OpNode (v4f32 QPR:$Vm))))]> {
2835 let Inst{10} = 1; // overwrite F = 1
2840 // Neon 2-register vector intrinsics,
2841 // element sizes of 8, 16 and 32 bits:
2842 multiclass N2VInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2843 bits<5> op11_7, bit op4,
2844 InstrItinClass itinD, InstrItinClass itinQ,
2845 string OpcodeStr, string Dt, Intrinsic IntOp> {
2846 // 64-bit vector types.
2847 def v8i8 : N2VDInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2848 itinD, OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
2849 def v4i16 : N2VDInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2850 itinD, OpcodeStr, !strconcat(Dt, "16"),v4i16,v4i16,IntOp>;
2851 def v2i32 : N2VDInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2852 itinD, OpcodeStr, !strconcat(Dt, "32"),v2i32,v2i32,IntOp>;
2854 // 128-bit vector types.
2855 def v16i8 : N2VQInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2856 itinQ, OpcodeStr, !strconcat(Dt, "8"), v16i8,v16i8,IntOp>;
2857 def v8i16 : N2VQInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2858 itinQ, OpcodeStr, !strconcat(Dt, "16"),v8i16,v8i16,IntOp>;
2859 def v4i32 : N2VQInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2860 itinQ, OpcodeStr, !strconcat(Dt, "32"),v4i32,v4i32,IntOp>;
2864 // Neon Narrowing 2-register vector operations,
2865 // source operand element sizes of 16, 32 and 64 bits:
2866 multiclass N2VN_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2867 bits<5> op11_7, bit op6, bit op4,
2868 InstrItinClass itin, string OpcodeStr, string Dt,
2870 def v8i8 : N2VN<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
2871 itin, OpcodeStr, !strconcat(Dt, "16"),
2872 v8i8, v8i16, OpNode>;
2873 def v4i16 : N2VN<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
2874 itin, OpcodeStr, !strconcat(Dt, "32"),
2875 v4i16, v4i32, OpNode>;
2876 def v2i32 : N2VN<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
2877 itin, OpcodeStr, !strconcat(Dt, "64"),
2878 v2i32, v2i64, OpNode>;
2881 // Neon Narrowing 2-register vector intrinsics,
2882 // source operand element sizes of 16, 32 and 64 bits:
2883 multiclass N2VNInt_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2884 bits<5> op11_7, bit op6, bit op4,
2885 InstrItinClass itin, string OpcodeStr, string Dt,
2887 def v8i8 : N2VNInt<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
2888 itin, OpcodeStr, !strconcat(Dt, "16"),
2889 v8i8, v8i16, IntOp>;
2890 def v4i16 : N2VNInt<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
2891 itin, OpcodeStr, !strconcat(Dt, "32"),
2892 v4i16, v4i32, IntOp>;
2893 def v2i32 : N2VNInt<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
2894 itin, OpcodeStr, !strconcat(Dt, "64"),
2895 v2i32, v2i64, IntOp>;
2899 // Neon Lengthening 2-register vector intrinsic (currently specific to VMOVL).
2900 // source operand element sizes of 16, 32 and 64 bits:
2901 multiclass N2VL_QHS<bits<2> op24_23, bits<5> op11_7, bit op6, bit op4,
2902 string OpcodeStr, string Dt, SDNode OpNode> {
2903 def v8i16 : N2VL<op24_23, 0b00, 0b10, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2904 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode>;
2905 def v4i32 : N2VL<op24_23, 0b01, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2906 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
2907 def v2i64 : N2VL<op24_23, 0b10, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2908 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
2912 // Neon 3-register vector operations.
2914 // First with only element sizes of 8, 16 and 32 bits:
2915 multiclass N3V_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2916 InstrItinClass itinD16, InstrItinClass itinD32,
2917 InstrItinClass itinQ16, InstrItinClass itinQ32,
2918 string OpcodeStr, string Dt,
2919 SDNode OpNode, bit Commutable = 0> {
2920 // 64-bit vector types.
2921 def v8i8 : N3VD<op24, op23, 0b00, op11_8, op4, itinD16,
2922 OpcodeStr, !strconcat(Dt, "8"),
2923 v8i8, v8i8, OpNode, Commutable>;
2924 def v4i16 : N3VD<op24, op23, 0b01, op11_8, op4, itinD16,
2925 OpcodeStr, !strconcat(Dt, "16"),
2926 v4i16, v4i16, OpNode, Commutable>;
2927 def v2i32 : N3VD<op24, op23, 0b10, op11_8, op4, itinD32,
2928 OpcodeStr, !strconcat(Dt, "32"),
2929 v2i32, v2i32, OpNode, Commutable>;
2931 // 128-bit vector types.
2932 def v16i8 : N3VQ<op24, op23, 0b00, op11_8, op4, itinQ16,
2933 OpcodeStr, !strconcat(Dt, "8"),
2934 v16i8, v16i8, OpNode, Commutable>;
2935 def v8i16 : N3VQ<op24, op23, 0b01, op11_8, op4, itinQ16,
2936 OpcodeStr, !strconcat(Dt, "16"),
2937 v8i16, v8i16, OpNode, Commutable>;
2938 def v4i32 : N3VQ<op24, op23, 0b10, op11_8, op4, itinQ32,
2939 OpcodeStr, !strconcat(Dt, "32"),
2940 v4i32, v4i32, OpNode, Commutable>;
2943 multiclass N3VSL_HS<bits<4> op11_8, string OpcodeStr, SDNode ShOp> {
2944 def v4i16 : N3VDSL16<0b01, op11_8, OpcodeStr, "i16", v4i16, ShOp>;
2945 def v2i32 : N3VDSL<0b10, op11_8, IIC_VMULi32D, OpcodeStr, "i32", v2i32, ShOp>;
2946 def v8i16 : N3VQSL16<0b01, op11_8, OpcodeStr, "i16", v8i16, v4i16, ShOp>;
2947 def v4i32 : N3VQSL<0b10, op11_8, IIC_VMULi32Q, OpcodeStr, "i32",
2948 v4i32, v2i32, ShOp>;
2951 // ....then also with element size 64 bits:
2952 multiclass N3V_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2953 InstrItinClass itinD, InstrItinClass itinQ,
2954 string OpcodeStr, string Dt,
2955 SDNode OpNode, bit Commutable = 0>
2956 : N3V_QHS<op24, op23, op11_8, op4, itinD, itinD, itinQ, itinQ,
2957 OpcodeStr, Dt, OpNode, Commutable> {
2958 def v1i64 : N3VD<op24, op23, 0b11, op11_8, op4, itinD,
2959 OpcodeStr, !strconcat(Dt, "64"),
2960 v1i64, v1i64, OpNode, Commutable>;
2961 def v2i64 : N3VQ<op24, op23, 0b11, op11_8, op4, itinQ,
2962 OpcodeStr, !strconcat(Dt, "64"),
2963 v2i64, v2i64, OpNode, Commutable>;
2967 // Neon 3-register vector intrinsics.
2969 // First with only element sizes of 16 and 32 bits:
2970 multiclass N3VInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2971 InstrItinClass itinD16, InstrItinClass itinD32,
2972 InstrItinClass itinQ16, InstrItinClass itinQ32,
2973 string OpcodeStr, string Dt,
2974 Intrinsic IntOp, bit Commutable = 0> {
2975 // 64-bit vector types.
2976 def v4i16 : N3VDInt<op24, op23, 0b01, op11_8, op4, f, itinD16,
2977 OpcodeStr, !strconcat(Dt, "16"),
2978 v4i16, v4i16, IntOp, Commutable>;
2979 def v2i32 : N3VDInt<op24, op23, 0b10, op11_8, op4, f, itinD32,
2980 OpcodeStr, !strconcat(Dt, "32"),
2981 v2i32, v2i32, IntOp, Commutable>;
2983 // 128-bit vector types.
2984 def v8i16 : N3VQInt<op24, op23, 0b01, op11_8, op4, f, itinQ16,
2985 OpcodeStr, !strconcat(Dt, "16"),
2986 v8i16, v8i16, IntOp, Commutable>;
2987 def v4i32 : N3VQInt<op24, op23, 0b10, op11_8, op4, f, itinQ32,
2988 OpcodeStr, !strconcat(Dt, "32"),
2989 v4i32, v4i32, IntOp, Commutable>;
2991 multiclass N3VInt_HSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2992 InstrItinClass itinD16, InstrItinClass itinD32,
2993 InstrItinClass itinQ16, InstrItinClass itinQ32,
2994 string OpcodeStr, string Dt,
2996 // 64-bit vector types.
2997 def v4i16 : N3VDIntSh<op24, op23, 0b01, op11_8, op4, f, itinD16,
2998 OpcodeStr, !strconcat(Dt, "16"),
2999 v4i16, v4i16, IntOp>;
3000 def v2i32 : N3VDIntSh<op24, op23, 0b10, op11_8, op4, f, itinD32,
3001 OpcodeStr, !strconcat(Dt, "32"),
3002 v2i32, v2i32, IntOp>;
3004 // 128-bit vector types.
3005 def v8i16 : N3VQIntSh<op24, op23, 0b01, op11_8, op4, f, itinQ16,
3006 OpcodeStr, !strconcat(Dt, "16"),
3007 v8i16, v8i16, IntOp>;
3008 def v4i32 : N3VQIntSh<op24, op23, 0b10, op11_8, op4, f, itinQ32,
3009 OpcodeStr, !strconcat(Dt, "32"),
3010 v4i32, v4i32, IntOp>;
3013 multiclass N3VIntSL_HS<bits<4> op11_8,
3014 InstrItinClass itinD16, InstrItinClass itinD32,
3015 InstrItinClass itinQ16, InstrItinClass itinQ32,
3016 string OpcodeStr, string Dt, Intrinsic IntOp> {
3017 def v4i16 : N3VDIntSL16<0b01, op11_8, itinD16,
3018 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp>;
3019 def v2i32 : N3VDIntSL<0b10, op11_8, itinD32,
3020 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp>;
3021 def v8i16 : N3VQIntSL16<0b01, op11_8, itinQ16,
3022 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16, IntOp>;
3023 def v4i32 : N3VQIntSL<0b10, op11_8, itinQ32,
3024 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32, IntOp>;
3027 // ....then also with element size of 8 bits:
3028 multiclass N3VInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3029 InstrItinClass itinD16, InstrItinClass itinD32,
3030 InstrItinClass itinQ16, InstrItinClass itinQ32,
3031 string OpcodeStr, string Dt,
3032 Intrinsic IntOp, bit Commutable = 0>
3033 : N3VInt_HS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
3034 OpcodeStr, Dt, IntOp, Commutable> {
3035 def v8i8 : N3VDInt<op24, op23, 0b00, op11_8, op4, f, itinD16,
3036 OpcodeStr, !strconcat(Dt, "8"),
3037 v8i8, v8i8, IntOp, Commutable>;
3038 def v16i8 : N3VQInt<op24, op23, 0b00, op11_8, op4, f, itinQ16,
3039 OpcodeStr, !strconcat(Dt, "8"),
3040 v16i8, v16i8, IntOp, Commutable>;
3042 multiclass N3VInt_QHSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3043 InstrItinClass itinD16, InstrItinClass itinD32,
3044 InstrItinClass itinQ16, InstrItinClass itinQ32,
3045 string OpcodeStr, string Dt,
3047 : N3VInt_HSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
3048 OpcodeStr, Dt, IntOp> {
3049 def v8i8 : N3VDIntSh<op24, op23, 0b00, op11_8, op4, f, itinD16,
3050 OpcodeStr, !strconcat(Dt, "8"),
3052 def v16i8 : N3VQIntSh<op24, op23, 0b00, op11_8, op4, f, itinQ16,
3053 OpcodeStr, !strconcat(Dt, "8"),
3054 v16i8, v16i8, IntOp>;
3058 // ....then also with element size of 64 bits:
3059 multiclass N3VInt_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3060 InstrItinClass itinD16, InstrItinClass itinD32,
3061 InstrItinClass itinQ16, InstrItinClass itinQ32,
3062 string OpcodeStr, string Dt,
3063 Intrinsic IntOp, bit Commutable = 0>
3064 : N3VInt_QHS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
3065 OpcodeStr, Dt, IntOp, Commutable> {
3066 def v1i64 : N3VDInt<op24, op23, 0b11, op11_8, op4, f, itinD32,
3067 OpcodeStr, !strconcat(Dt, "64"),
3068 v1i64, v1i64, IntOp, Commutable>;
3069 def v2i64 : N3VQInt<op24, op23, 0b11, op11_8, op4, f, itinQ32,
3070 OpcodeStr, !strconcat(Dt, "64"),
3071 v2i64, v2i64, IntOp, Commutable>;
3073 multiclass N3VInt_QHSDSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3074 InstrItinClass itinD16, InstrItinClass itinD32,
3075 InstrItinClass itinQ16, InstrItinClass itinQ32,
3076 string OpcodeStr, string Dt,
3078 : N3VInt_QHSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
3079 OpcodeStr, Dt, IntOp> {
3080 def v1i64 : N3VDIntSh<op24, op23, 0b11, op11_8, op4, f, itinD32,
3081 OpcodeStr, !strconcat(Dt, "64"),
3082 v1i64, v1i64, IntOp>;
3083 def v2i64 : N3VQIntSh<op24, op23, 0b11, op11_8, op4, f, itinQ32,
3084 OpcodeStr, !strconcat(Dt, "64"),
3085 v2i64, v2i64, IntOp>;
3088 // Neon Narrowing 3-register vector intrinsics,
3089 // source operand element sizes of 16, 32 and 64 bits:
3090 multiclass N3VNInt_HSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3091 string OpcodeStr, string Dt,
3092 Intrinsic IntOp, bit Commutable = 0> {
3093 def v8i8 : N3VNInt<op24, op23, 0b00, op11_8, op4,
3094 OpcodeStr, !strconcat(Dt, "16"),
3095 v8i8, v8i16, IntOp, Commutable>;
3096 def v4i16 : N3VNInt<op24, op23, 0b01, op11_8, op4,
3097 OpcodeStr, !strconcat(Dt, "32"),
3098 v4i16, v4i32, IntOp, Commutable>;
3099 def v2i32 : N3VNInt<op24, op23, 0b10, op11_8, op4,
3100 OpcodeStr, !strconcat(Dt, "64"),
3101 v2i32, v2i64, IntOp, Commutable>;
3105 // Neon Long 3-register vector operations.
3107 multiclass N3VL_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3108 InstrItinClass itin16, InstrItinClass itin32,
3109 string OpcodeStr, string Dt,
3110 SDNode OpNode, bit Commutable = 0> {
3111 def v8i16 : N3VL<op24, op23, 0b00, op11_8, op4, itin16,
3112 OpcodeStr, !strconcat(Dt, "8"),
3113 v8i16, v8i8, OpNode, Commutable>;
3114 def v4i32 : N3VL<op24, op23, 0b01, op11_8, op4, itin16,
3115 OpcodeStr, !strconcat(Dt, "16"),
3116 v4i32, v4i16, OpNode, Commutable>;
3117 def v2i64 : N3VL<op24, op23, 0b10, op11_8, op4, itin32,
3118 OpcodeStr, !strconcat(Dt, "32"),
3119 v2i64, v2i32, OpNode, Commutable>;
3122 multiclass N3VLSL_HS<bit op24, bits<4> op11_8,
3123 InstrItinClass itin, string OpcodeStr, string Dt,
3125 def v4i16 : N3VLSL16<op24, 0b01, op11_8, itin, OpcodeStr,
3126 !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
3127 def v2i32 : N3VLSL<op24, 0b10, op11_8, itin, OpcodeStr,
3128 !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
3131 multiclass N3VLExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3132 InstrItinClass itin16, InstrItinClass itin32,
3133 string OpcodeStr, string Dt,
3134 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
3135 def v8i16 : N3VLExt<op24, op23, 0b00, op11_8, op4, itin16,
3136 OpcodeStr, !strconcat(Dt, "8"),
3137 v8i16, v8i8, OpNode, ExtOp, Commutable>;
3138 def v4i32 : N3VLExt<op24, op23, 0b01, op11_8, op4, itin16,
3139 OpcodeStr, !strconcat(Dt, "16"),
3140 v4i32, v4i16, OpNode, ExtOp, Commutable>;
3141 def v2i64 : N3VLExt<op24, op23, 0b10, op11_8, op4, itin32,
3142 OpcodeStr, !strconcat(Dt, "32"),
3143 v2i64, v2i32, OpNode, ExtOp, Commutable>;
3146 // Neon Long 3-register vector intrinsics.
3148 // First with only element sizes of 16 and 32 bits:
3149 multiclass N3VLInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
3150 InstrItinClass itin16, InstrItinClass itin32,
3151 string OpcodeStr, string Dt,
3152 Intrinsic IntOp, bit Commutable = 0> {
3153 def v4i32 : N3VLInt<op24, op23, 0b01, op11_8, op4, itin16,
3154 OpcodeStr, !strconcat(Dt, "16"),
3155 v4i32, v4i16, IntOp, Commutable>;
3156 def v2i64 : N3VLInt<op24, op23, 0b10, op11_8, op4, itin32,
3157 OpcodeStr, !strconcat(Dt, "32"),
3158 v2i64, v2i32, IntOp, Commutable>;
3161 multiclass N3VLIntSL_HS<bit op24, bits<4> op11_8,
3162 InstrItinClass itin, string OpcodeStr, string Dt,
3164 def v4i16 : N3VLIntSL16<op24, 0b01, op11_8, itin,
3165 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
3166 def v2i32 : N3VLIntSL<op24, 0b10, op11_8, itin,
3167 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3170 // ....then also with element size of 8 bits:
3171 multiclass N3VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3172 InstrItinClass itin16, InstrItinClass itin32,
3173 string OpcodeStr, string Dt,
3174 Intrinsic IntOp, bit Commutable = 0>
3175 : N3VLInt_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt,
3176 IntOp, Commutable> {
3177 def v8i16 : N3VLInt<op24, op23, 0b00, op11_8, op4, itin16,
3178 OpcodeStr, !strconcat(Dt, "8"),
3179 v8i16, v8i8, IntOp, Commutable>;
3182 // ....with explicit extend (VABDL).
3183 multiclass N3VLIntExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3184 InstrItinClass itin, string OpcodeStr, string Dt,
3185 Intrinsic IntOp, SDNode ExtOp, bit Commutable = 0> {
3186 def v8i16 : N3VLIntExt<op24, op23, 0b00, op11_8, op4, itin,
3187 OpcodeStr, !strconcat(Dt, "8"),
3188 v8i16, v8i8, IntOp, ExtOp, Commutable>;
3189 def v4i32 : N3VLIntExt<op24, op23, 0b01, op11_8, op4, itin,
3190 OpcodeStr, !strconcat(Dt, "16"),
3191 v4i32, v4i16, IntOp, ExtOp, Commutable>;
3192 def v2i64 : N3VLIntExt<op24, op23, 0b10, op11_8, op4, itin,
3193 OpcodeStr, !strconcat(Dt, "32"),
3194 v2i64, v2i32, IntOp, ExtOp, Commutable>;
3198 // Neon Wide 3-register vector intrinsics,
3199 // source operand element sizes of 8, 16 and 32 bits:
3200 multiclass N3VW_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3201 string OpcodeStr, string Dt,
3202 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
3203 def v8i16 : N3VW<op24, op23, 0b00, op11_8, op4,
3204 OpcodeStr, !strconcat(Dt, "8"),
3205 v8i16, v8i8, OpNode, ExtOp, Commutable>;
3206 def v4i32 : N3VW<op24, op23, 0b01, op11_8, op4,
3207 OpcodeStr, !strconcat(Dt, "16"),
3208 v4i32, v4i16, OpNode, ExtOp, Commutable>;
3209 def v2i64 : N3VW<op24, op23, 0b10, op11_8, op4,
3210 OpcodeStr, !strconcat(Dt, "32"),
3211 v2i64, v2i32, OpNode, ExtOp, Commutable>;
3215 // Neon Multiply-Op vector operations,
3216 // element sizes of 8, 16 and 32 bits:
3217 multiclass N3VMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3218 InstrItinClass itinD16, InstrItinClass itinD32,
3219 InstrItinClass itinQ16, InstrItinClass itinQ32,
3220 string OpcodeStr, string Dt, SDNode OpNode> {
3221 // 64-bit vector types.
3222 def v8i8 : N3VDMulOp<op24, op23, 0b00, op11_8, op4, itinD16,
3223 OpcodeStr, !strconcat(Dt, "8"), v8i8, mul, OpNode>;
3224 def v4i16 : N3VDMulOp<op24, op23, 0b01, op11_8, op4, itinD16,
3225 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, OpNode>;
3226 def v2i32 : N3VDMulOp<op24, op23, 0b10, op11_8, op4, itinD32,
3227 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, OpNode>;
3229 // 128-bit vector types.
3230 def v16i8 : N3VQMulOp<op24, op23, 0b00, op11_8, op4, itinQ16,
3231 OpcodeStr, !strconcat(Dt, "8"), v16i8, mul, OpNode>;
3232 def v8i16 : N3VQMulOp<op24, op23, 0b01, op11_8, op4, itinQ16,
3233 OpcodeStr, !strconcat(Dt, "16"), v8i16, mul, OpNode>;
3234 def v4i32 : N3VQMulOp<op24, op23, 0b10, op11_8, op4, itinQ32,
3235 OpcodeStr, !strconcat(Dt, "32"), v4i32, mul, OpNode>;
3238 multiclass N3VMulOpSL_HS<bits<4> op11_8,
3239 InstrItinClass itinD16, InstrItinClass itinD32,
3240 InstrItinClass itinQ16, InstrItinClass itinQ32,
3241 string OpcodeStr, string Dt, SDNode ShOp> {
3242 def v4i16 : N3VDMulOpSL16<0b01, op11_8, itinD16,
3243 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, ShOp>;
3244 def v2i32 : N3VDMulOpSL<0b10, op11_8, itinD32,
3245 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, ShOp>;
3246 def v8i16 : N3VQMulOpSL16<0b01, op11_8, itinQ16,
3247 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16,
3249 def v4i32 : N3VQMulOpSL<0b10, op11_8, itinQ32,
3250 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32,
3254 // Neon Intrinsic-Op vector operations,
3255 // element sizes of 8, 16 and 32 bits:
3256 multiclass N3VIntOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3257 InstrItinClass itinD, InstrItinClass itinQ,
3258 string OpcodeStr, string Dt, Intrinsic IntOp,
3260 // 64-bit vector types.
3261 def v8i8 : N3VDIntOp<op24, op23, 0b00, op11_8, op4, itinD,
3262 OpcodeStr, !strconcat(Dt, "8"), v8i8, IntOp, OpNode>;
3263 def v4i16 : N3VDIntOp<op24, op23, 0b01, op11_8, op4, itinD,
3264 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp, OpNode>;
3265 def v2i32 : N3VDIntOp<op24, op23, 0b10, op11_8, op4, itinD,
3266 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp, OpNode>;
3268 // 128-bit vector types.
3269 def v16i8 : N3VQIntOp<op24, op23, 0b00, op11_8, op4, itinQ,
3270 OpcodeStr, !strconcat(Dt, "8"), v16i8, IntOp, OpNode>;
3271 def v8i16 : N3VQIntOp<op24, op23, 0b01, op11_8, op4, itinQ,
3272 OpcodeStr, !strconcat(Dt, "16"), v8i16, IntOp, OpNode>;
3273 def v4i32 : N3VQIntOp<op24, op23, 0b10, op11_8, op4, itinQ,
3274 OpcodeStr, !strconcat(Dt, "32"), v4i32, IntOp, OpNode>;
3277 // Neon 3-argument intrinsics,
3278 // element sizes of 8, 16 and 32 bits:
3279 multiclass N3VInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3280 InstrItinClass itinD, InstrItinClass itinQ,
3281 string OpcodeStr, string Dt, Intrinsic IntOp> {
3282 // 64-bit vector types.
3283 def v8i8 : N3VDInt3<op24, op23, 0b00, op11_8, op4, itinD,
3284 OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
3285 def v4i16 : N3VDInt3<op24, op23, 0b01, op11_8, op4, itinD,
3286 OpcodeStr, !strconcat(Dt, "16"), v4i16, v4i16, IntOp>;
3287 def v2i32 : N3VDInt3<op24, op23, 0b10, op11_8, op4, itinD,
3288 OpcodeStr, !strconcat(Dt, "32"), v2i32, v2i32, IntOp>;
3290 // 128-bit vector types.
3291 def v16i8 : N3VQInt3<op24, op23, 0b00, op11_8, op4, itinQ,
3292 OpcodeStr, !strconcat(Dt, "8"), v16i8, v16i8, IntOp>;
3293 def v8i16 : N3VQInt3<op24, op23, 0b01, op11_8, op4, itinQ,
3294 OpcodeStr, !strconcat(Dt, "16"), v8i16, v8i16, IntOp>;
3295 def v4i32 : N3VQInt3<op24, op23, 0b10, op11_8, op4, itinQ,
3296 OpcodeStr, !strconcat(Dt, "32"), v4i32, v4i32, IntOp>;
3300 // Neon Long Multiply-Op vector operations,
3301 // element sizes of 8, 16 and 32 bits:
3302 multiclass N3VLMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3303 InstrItinClass itin16, InstrItinClass itin32,
3304 string OpcodeStr, string Dt, SDNode MulOp,
3306 def v8i16 : N3VLMulOp<op24, op23, 0b00, op11_8, op4, itin16, OpcodeStr,
3307 !strconcat(Dt, "8"), v8i16, v8i8, MulOp, OpNode>;
3308 def v4i32 : N3VLMulOp<op24, op23, 0b01, op11_8, op4, itin16, OpcodeStr,
3309 !strconcat(Dt, "16"), v4i32, v4i16, MulOp, OpNode>;
3310 def v2i64 : N3VLMulOp<op24, op23, 0b10, op11_8, op4, itin32, OpcodeStr,
3311 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
3314 multiclass N3VLMulOpSL_HS<bit op24, bits<4> op11_8, string OpcodeStr,
3315 string Dt, SDNode MulOp, SDNode OpNode> {
3316 def v4i16 : N3VLMulOpSL16<op24, 0b01, op11_8, IIC_VMACi16D, OpcodeStr,
3317 !strconcat(Dt,"16"), v4i32, v4i16, MulOp, OpNode>;
3318 def v2i32 : N3VLMulOpSL<op24, 0b10, op11_8, IIC_VMACi32D, OpcodeStr,
3319 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
3323 // Neon Long 3-argument intrinsics.
3325 // First with only element sizes of 16 and 32 bits:
3326 multiclass N3VLInt3_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
3327 InstrItinClass itin16, InstrItinClass itin32,
3328 string OpcodeStr, string Dt, Intrinsic IntOp> {
3329 def v4i32 : N3VLInt3<op24, op23, 0b01, op11_8, op4, itin16,
3330 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
3331 def v2i64 : N3VLInt3<op24, op23, 0b10, op11_8, op4, itin32,
3332 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3335 multiclass N3VLInt3SL_HS<bit op24, bits<4> op11_8,
3336 string OpcodeStr, string Dt, Intrinsic IntOp> {
3337 def v4i16 : N3VLInt3SL16<op24, 0b01, op11_8, IIC_VMACi16D,
3338 OpcodeStr, !strconcat(Dt,"16"), v4i32, v4i16, IntOp>;
3339 def v2i32 : N3VLInt3SL<op24, 0b10, op11_8, IIC_VMACi32D,
3340 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3343 // ....then also with element size of 8 bits:
3344 multiclass N3VLInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3345 InstrItinClass itin16, InstrItinClass itin32,
3346 string OpcodeStr, string Dt, Intrinsic IntOp>
3347 : N3VLInt3_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt, IntOp> {
3348 def v8i16 : N3VLInt3<op24, op23, 0b00, op11_8, op4, itin16,
3349 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
3352 // ....with explicit extend (VABAL).
3353 multiclass N3VLIntExtOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3354 InstrItinClass itin, string OpcodeStr, string Dt,
3355 Intrinsic IntOp, SDNode ExtOp, SDNode OpNode> {
3356 def v8i16 : N3VLIntExtOp<op24, op23, 0b00, op11_8, op4, itin,
3357 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8,
3358 IntOp, ExtOp, OpNode>;
3359 def v4i32 : N3VLIntExtOp<op24, op23, 0b01, op11_8, op4, itin,
3360 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16,
3361 IntOp, ExtOp, OpNode>;
3362 def v2i64 : N3VLIntExtOp<op24, op23, 0b10, op11_8, op4, itin,
3363 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32,
3364 IntOp, ExtOp, OpNode>;
3368 // Neon Pairwise long 2-register intrinsics,
3369 // element sizes of 8, 16 and 32 bits:
3370 multiclass N2VPLInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3371 bits<5> op11_7, bit op4,
3372 string OpcodeStr, string Dt, Intrinsic IntOp> {
3373 // 64-bit vector types.
3374 def v8i8 : N2VDPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3375 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
3376 def v4i16 : N2VDPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3377 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
3378 def v2i32 : N2VDPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3379 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
3381 // 128-bit vector types.
3382 def v16i8 : N2VQPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3383 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
3384 def v8i16 : N2VQPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3385 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
3386 def v4i32 : N2VQPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3387 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
3391 // Neon Pairwise long 2-register accumulate intrinsics,
3392 // element sizes of 8, 16 and 32 bits:
3393 multiclass N2VPLInt2_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3394 bits<5> op11_7, bit op4,
3395 string OpcodeStr, string Dt, Intrinsic IntOp> {
3396 // 64-bit vector types.
3397 def v8i8 : N2VDPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3398 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
3399 def v4i16 : N2VDPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3400 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
3401 def v2i32 : N2VDPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3402 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
3404 // 128-bit vector types.
3405 def v16i8 : N2VQPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3406 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
3407 def v8i16 : N2VQPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3408 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
3409 def v4i32 : N2VQPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3410 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
3414 // Neon 2-register vector shift by immediate,
3415 // with f of either N2RegVShLFrm or N2RegVShRFrm
3416 // element sizes of 8, 16, 32 and 64 bits:
3417 multiclass N2VShL_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3418 InstrItinClass itin, string OpcodeStr, string Dt,
3420 // 64-bit vector types.
3421 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3422 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
3423 let Inst{21-19} = 0b001; // imm6 = 001xxx
3425 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3426 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
3427 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3429 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3430 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
3431 let Inst{21} = 0b1; // imm6 = 1xxxxx
3433 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, N2RegVShLFrm, itin, i32imm,
3434 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
3437 // 128-bit vector types.
3438 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3439 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
3440 let Inst{21-19} = 0b001; // imm6 = 001xxx
3442 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3443 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
3444 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3446 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3447 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
3448 let Inst{21} = 0b1; // imm6 = 1xxxxx
3450 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, N2RegVShLFrm, itin, i32imm,
3451 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
3454 multiclass N2VShR_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3455 InstrItinClass itin, string OpcodeStr, string Dt,
3457 // 64-bit vector types.
3458 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm8,
3459 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
3460 let Inst{21-19} = 0b001; // imm6 = 001xxx
3462 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm16,
3463 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
3464 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3466 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm32,
3467 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
3468 let Inst{21} = 0b1; // imm6 = 1xxxxx
3470 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, N2RegVShRFrm, itin, shr_imm64,
3471 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
3474 // 128-bit vector types.
3475 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm8,
3476 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
3477 let Inst{21-19} = 0b001; // imm6 = 001xxx
3479 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm16,
3480 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
3481 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3483 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm32,
3484 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
3485 let Inst{21} = 0b1; // imm6 = 1xxxxx
3487 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, N2RegVShRFrm, itin, shr_imm64,
3488 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
3492 // Neon Shift-Accumulate vector operations,
3493 // element sizes of 8, 16, 32 and 64 bits:
3494 multiclass N2VShAdd_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3495 string OpcodeStr, string Dt, SDNode ShOp> {
3496 // 64-bit vector types.
3497 def v8i8 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm8,
3498 OpcodeStr, !strconcat(Dt, "8"), v8i8, ShOp> {
3499 let Inst{21-19} = 0b001; // imm6 = 001xxx
3501 def v4i16 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm16,
3502 OpcodeStr, !strconcat(Dt, "16"), v4i16, ShOp> {
3503 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3505 def v2i32 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm32,
3506 OpcodeStr, !strconcat(Dt, "32"), v2i32, ShOp> {
3507 let Inst{21} = 0b1; // imm6 = 1xxxxx
3509 def v1i64 : N2VDShAdd<op24, op23, op11_8, 1, op4, shr_imm64,
3510 OpcodeStr, !strconcat(Dt, "64"), v1i64, ShOp>;
3513 // 128-bit vector types.
3514 def v16i8 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm8,
3515 OpcodeStr, !strconcat(Dt, "8"), v16i8, ShOp> {
3516 let Inst{21-19} = 0b001; // imm6 = 001xxx
3518 def v8i16 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm16,
3519 OpcodeStr, !strconcat(Dt, "16"), v8i16, ShOp> {
3520 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3522 def v4i32 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm32,
3523 OpcodeStr, !strconcat(Dt, "32"), v4i32, ShOp> {
3524 let Inst{21} = 0b1; // imm6 = 1xxxxx
3526 def v2i64 : N2VQShAdd<op24, op23, op11_8, 1, op4, shr_imm64,
3527 OpcodeStr, !strconcat(Dt, "64"), v2i64, ShOp>;
3531 // Neon Shift-Insert vector operations,
3532 // with f of either N2RegVShLFrm or N2RegVShRFrm
3533 // element sizes of 8, 16, 32 and 64 bits:
3534 multiclass N2VShInsL_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, i32imm,
3538 N2RegVShLFrm, OpcodeStr, "8", v8i8, NEONvsli> {
3539 let Inst{21-19} = 0b001; // imm6 = 001xxx
3541 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3542 N2RegVShLFrm, OpcodeStr, "16", v4i16, NEONvsli> {
3543 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3545 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3546 N2RegVShLFrm, OpcodeStr, "32", v2i32, NEONvsli> {
3547 let Inst{21} = 0b1; // imm6 = 1xxxxx
3549 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4, i32imm,
3550 N2RegVShLFrm, OpcodeStr, "64", v1i64, NEONvsli>;
3553 // 128-bit vector types.
3554 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3555 N2RegVShLFrm, OpcodeStr, "8", v16i8, NEONvsli> {
3556 let Inst{21-19} = 0b001; // imm6 = 001xxx
3558 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3559 N2RegVShLFrm, OpcodeStr, "16", v8i16, NEONvsli> {
3560 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3562 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3563 N2RegVShLFrm, OpcodeStr, "32", v4i32, NEONvsli> {
3564 let Inst{21} = 0b1; // imm6 = 1xxxxx
3566 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4, i32imm,
3567 N2RegVShLFrm, OpcodeStr, "64", v2i64, NEONvsli>;
3570 multiclass N2VShInsR_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3572 // 64-bit vector types.
3573 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm8,
3574 N2RegVShRFrm, OpcodeStr, "8", v8i8, NEONvsri> {
3575 let Inst{21-19} = 0b001; // imm6 = 001xxx
3577 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm16,
3578 N2RegVShRFrm, OpcodeStr, "16", v4i16, NEONvsri> {
3579 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3581 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm32,
3582 N2RegVShRFrm, OpcodeStr, "32", v2i32, NEONvsri> {
3583 let Inst{21} = 0b1; // imm6 = 1xxxxx
3585 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4, shr_imm64,
3586 N2RegVShRFrm, OpcodeStr, "64", v1i64, NEONvsri>;
3589 // 128-bit vector types.
3590 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm8,
3591 N2RegVShRFrm, OpcodeStr, "8", v16i8, NEONvsri> {
3592 let Inst{21-19} = 0b001; // imm6 = 001xxx
3594 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm16,
3595 N2RegVShRFrm, OpcodeStr, "16", v8i16, NEONvsri> {
3596 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3598 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm32,
3599 N2RegVShRFrm, OpcodeStr, "32", v4i32, NEONvsri> {
3600 let Inst{21} = 0b1; // imm6 = 1xxxxx
3602 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4, shr_imm64,
3603 N2RegVShRFrm, OpcodeStr, "64", v2i64, NEONvsri>;
3607 // Neon Shift Long operations,
3608 // element sizes of 8, 16, 32 bits:
3609 multiclass N2VLSh_QHS<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
3610 bit op4, string OpcodeStr, string Dt, SDNode OpNode> {
3611 def v8i16 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3612 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, imm1_7, OpNode> {
3613 let Inst{21-19} = 0b001; // imm6 = 001xxx
3615 def v4i32 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3616 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, imm1_15, OpNode> {
3617 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3619 def v2i64 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3620 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, imm1_31, OpNode> {
3621 let Inst{21} = 0b1; // imm6 = 1xxxxx
3625 // Neon Shift Narrow operations,
3626 // element sizes of 16, 32, 64 bits:
3627 multiclass N2VNSh_HSD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
3628 bit op4, InstrItinClass itin, string OpcodeStr, string Dt,
3630 def v8i8 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3631 OpcodeStr, !strconcat(Dt, "16"),
3632 v8i8, v8i16, shr_imm8, OpNode> {
3633 let Inst{21-19} = 0b001; // imm6 = 001xxx
3635 def v4i16 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3636 OpcodeStr, !strconcat(Dt, "32"),
3637 v4i16, v4i32, shr_imm16, OpNode> {
3638 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3640 def v2i32 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3641 OpcodeStr, !strconcat(Dt, "64"),
3642 v2i32, v2i64, shr_imm32, OpNode> {
3643 let Inst{21} = 0b1; // imm6 = 1xxxxx
3647 //===----------------------------------------------------------------------===//
3648 // Instruction Definitions.
3649 //===----------------------------------------------------------------------===//
3651 // Vector Add Operations.
3653 // VADD : Vector Add (integer and floating-point)
3654 defm VADD : N3V_QHSD<0, 0, 0b1000, 0, IIC_VBINiD, IIC_VBINiQ, "vadd", "i",
3656 def VADDfd : N3VD<0, 0, 0b00, 0b1101, 0, IIC_VBIND, "vadd", "f32",
3657 v2f32, v2f32, fadd, 1>;
3658 def VADDfq : N3VQ<0, 0, 0b00, 0b1101, 0, IIC_VBINQ, "vadd", "f32",
3659 v4f32, v4f32, fadd, 1>;
3660 // VADDL : Vector Add Long (Q = D + D)
3661 defm VADDLs : N3VLExt_QHS<0,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
3662 "vaddl", "s", add, sext, 1>;
3663 defm VADDLu : N3VLExt_QHS<1,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
3664 "vaddl", "u", add, zext, 1>;
3665 // VADDW : Vector Add Wide (Q = Q + D)
3666 defm VADDWs : N3VW_QHS<0,1,0b0001,0, "vaddw", "s", add, sext, 0>;
3667 defm VADDWu : N3VW_QHS<1,1,0b0001,0, "vaddw", "u", add, zext, 0>;
3668 // VHADD : Vector Halving Add
3669 defm VHADDs : N3VInt_QHS<0, 0, 0b0000, 0, N3RegFrm,
3670 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3671 "vhadd", "s", int_arm_neon_vhadds, 1>;
3672 defm VHADDu : N3VInt_QHS<1, 0, 0b0000, 0, N3RegFrm,
3673 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3674 "vhadd", "u", int_arm_neon_vhaddu, 1>;
3675 // VRHADD : Vector Rounding Halving Add
3676 defm VRHADDs : N3VInt_QHS<0, 0, 0b0001, 0, N3RegFrm,
3677 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3678 "vrhadd", "s", int_arm_neon_vrhadds, 1>;
3679 defm VRHADDu : N3VInt_QHS<1, 0, 0b0001, 0, N3RegFrm,
3680 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3681 "vrhadd", "u", int_arm_neon_vrhaddu, 1>;
3682 // VQADD : Vector Saturating Add
3683 defm VQADDs : N3VInt_QHSD<0, 0, 0b0000, 1, N3RegFrm,
3684 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3685 "vqadd", "s", int_arm_neon_vqadds, 1>;
3686 defm VQADDu : N3VInt_QHSD<1, 0, 0b0000, 1, N3RegFrm,
3687 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3688 "vqadd", "u", int_arm_neon_vqaddu, 1>;
3689 // VADDHN : Vector Add and Narrow Returning High Half (D = Q + Q)
3690 defm VADDHN : N3VNInt_HSD<0,1,0b0100,0, "vaddhn", "i",
3691 int_arm_neon_vaddhn, 1>;
3692 // VRADDHN : Vector Rounding Add and Narrow Returning High Half (D = Q + Q)
3693 defm VRADDHN : N3VNInt_HSD<1,1,0b0100,0, "vraddhn", "i",
3694 int_arm_neon_vraddhn, 1>;
3696 // Vector Multiply Operations.
3698 // VMUL : Vector Multiply (integer, polynomial and floating-point)
3699 defm VMUL : N3V_QHS<0, 0, 0b1001, 1, IIC_VMULi16D, IIC_VMULi32D,
3700 IIC_VMULi16Q, IIC_VMULi32Q, "vmul", "i", mul, 1>;
3701 def VMULpd : N3VDInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16D, "vmul",
3702 "p8", v8i8, v8i8, int_arm_neon_vmulp, 1>;
3703 def VMULpq : N3VQInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16Q, "vmul",
3704 "p8", v16i8, v16i8, int_arm_neon_vmulp, 1>;
3705 def VMULfd : N3VD<1, 0, 0b00, 0b1101, 1, IIC_VFMULD, "vmul", "f32",
3706 v2f32, v2f32, fmul, 1>;
3707 def VMULfq : N3VQ<1, 0, 0b00, 0b1101, 1, IIC_VFMULQ, "vmul", "f32",
3708 v4f32, v4f32, fmul, 1>;
3709 defm VMULsl : N3VSL_HS<0b1000, "vmul", mul>;
3710 def VMULslfd : N3VDSL<0b10, 0b1001, IIC_VBIND, "vmul", "f32", v2f32, fmul>;
3711 def VMULslfq : N3VQSL<0b10, 0b1001, IIC_VBINQ, "vmul", "f32", v4f32,
3714 def : Pat<(v8i16 (mul (v8i16 QPR:$src1),
3715 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
3716 (v8i16 (VMULslv8i16 (v8i16 QPR:$src1),
3717 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3718 (DSubReg_i16_reg imm:$lane))),
3719 (SubReg_i16_lane imm:$lane)))>;
3720 def : Pat<(v4i32 (mul (v4i32 QPR:$src1),
3721 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
3722 (v4i32 (VMULslv4i32 (v4i32 QPR:$src1),
3723 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3724 (DSubReg_i32_reg imm:$lane))),
3725 (SubReg_i32_lane imm:$lane)))>;
3726 def : Pat<(v4f32 (fmul (v4f32 QPR:$src1),
3727 (v4f32 (NEONvduplane (v4f32 QPR:$src2), imm:$lane)))),
3728 (v4f32 (VMULslfq (v4f32 QPR:$src1),
3729 (v2f32 (EXTRACT_SUBREG QPR:$src2,
3730 (DSubReg_i32_reg imm:$lane))),
3731 (SubReg_i32_lane imm:$lane)))>;
3733 // VQDMULH : Vector Saturating Doubling Multiply Returning High Half
3734 defm VQDMULH : N3VInt_HS<0, 0, 0b1011, 0, N3RegFrm, IIC_VMULi16D, IIC_VMULi32D,
3735 IIC_VMULi16Q, IIC_VMULi32Q,
3736 "vqdmulh", "s", int_arm_neon_vqdmulh, 1>;
3737 defm VQDMULHsl: N3VIntSL_HS<0b1100, IIC_VMULi16D, IIC_VMULi32D,
3738 IIC_VMULi16Q, IIC_VMULi32Q,
3739 "vqdmulh", "s", int_arm_neon_vqdmulh>;
3740 def : Pat<(v8i16 (int_arm_neon_vqdmulh (v8i16 QPR:$src1),
3741 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
3743 (v8i16 (VQDMULHslv8i16 (v8i16 QPR:$src1),
3744 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3745 (DSubReg_i16_reg imm:$lane))),
3746 (SubReg_i16_lane imm:$lane)))>;
3747 def : Pat<(v4i32 (int_arm_neon_vqdmulh (v4i32 QPR:$src1),
3748 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
3750 (v4i32 (VQDMULHslv4i32 (v4i32 QPR:$src1),
3751 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3752 (DSubReg_i32_reg imm:$lane))),
3753 (SubReg_i32_lane imm:$lane)))>;
3755 // VQRDMULH : Vector Rounding Saturating Doubling Multiply Returning High Half
3756 defm VQRDMULH : N3VInt_HS<1, 0, 0b1011, 0, N3RegFrm,
3757 IIC_VMULi16D,IIC_VMULi32D,IIC_VMULi16Q,IIC_VMULi32Q,
3758 "vqrdmulh", "s", int_arm_neon_vqrdmulh, 1>;
3759 defm VQRDMULHsl : N3VIntSL_HS<0b1101, IIC_VMULi16D, IIC_VMULi32D,
3760 IIC_VMULi16Q, IIC_VMULi32Q,
3761 "vqrdmulh", "s", int_arm_neon_vqrdmulh>;
3762 def : Pat<(v8i16 (int_arm_neon_vqrdmulh (v8i16 QPR:$src1),
3763 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
3765 (v8i16 (VQRDMULHslv8i16 (v8i16 QPR:$src1),
3766 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3767 (DSubReg_i16_reg imm:$lane))),
3768 (SubReg_i16_lane imm:$lane)))>;
3769 def : Pat<(v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$src1),
3770 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
3772 (v4i32 (VQRDMULHslv4i32 (v4i32 QPR:$src1),
3773 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3774 (DSubReg_i32_reg imm:$lane))),
3775 (SubReg_i32_lane imm:$lane)))>;
3777 // VMULL : Vector Multiply Long (integer and polynomial) (Q = D * D)
3778 defm VMULLs : N3VL_QHS<0,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
3779 "vmull", "s", NEONvmulls, 1>;
3780 defm VMULLu : N3VL_QHS<1,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
3781 "vmull", "u", NEONvmullu, 1>;
3782 def VMULLp : N3VLInt<0, 1, 0b00, 0b1110, 0, IIC_VMULi16D, "vmull", "p8",
3783 v8i16, v8i8, int_arm_neon_vmullp, 1>;
3784 defm VMULLsls : N3VLSL_HS<0, 0b1010, IIC_VMULi16D, "vmull", "s", NEONvmulls>;
3785 defm VMULLslu : N3VLSL_HS<1, 0b1010, IIC_VMULi16D, "vmull", "u", NEONvmullu>;
3787 // VQDMULL : Vector Saturating Doubling Multiply Long (Q = D * D)
3788 defm VQDMULL : N3VLInt_HS<0,1,0b1101,0, IIC_VMULi16D, IIC_VMULi32D,
3789 "vqdmull", "s", int_arm_neon_vqdmull, 1>;
3790 defm VQDMULLsl: N3VLIntSL_HS<0, 0b1011, IIC_VMULi16D,
3791 "vqdmull", "s", int_arm_neon_vqdmull>;
3793 // Vector Multiply-Accumulate and Multiply-Subtract Operations.
3795 // VMLA : Vector Multiply Accumulate (integer and floating-point)
3796 defm VMLA : N3VMulOp_QHS<0, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3797 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
3798 def VMLAfd : N3VDMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACD, "vmla", "f32",
3799 v2f32, fmul_su, fadd_mlx>,
3800 Requires<[HasNEON, UseFPVMLx]>;
3801 def VMLAfq : N3VQMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACQ, "vmla", "f32",
3802 v4f32, fmul_su, fadd_mlx>,
3803 Requires<[HasNEON, UseFPVMLx]>;
3804 defm VMLAsl : N3VMulOpSL_HS<0b0000, IIC_VMACi16D, IIC_VMACi32D,
3805 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
3806 def VMLAslfd : N3VDMulOpSL<0b10, 0b0001, IIC_VMACD, "vmla", "f32",
3807 v2f32, fmul_su, fadd_mlx>,
3808 Requires<[HasNEON, UseFPVMLx]>;
3809 def VMLAslfq : N3VQMulOpSL<0b10, 0b0001, IIC_VMACQ, "vmla", "f32",
3810 v4f32, v2f32, fmul_su, fadd_mlx>,
3811 Requires<[HasNEON, UseFPVMLx]>;
3813 def : Pat<(v8i16 (add (v8i16 QPR:$src1),
3814 (mul (v8i16 QPR:$src2),
3815 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
3816 (v8i16 (VMLAslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
3817 (v4i16 (EXTRACT_SUBREG QPR:$src3,
3818 (DSubReg_i16_reg imm:$lane))),
3819 (SubReg_i16_lane imm:$lane)))>;
3821 def : Pat<(v4i32 (add (v4i32 QPR:$src1),
3822 (mul (v4i32 QPR:$src2),
3823 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
3824 (v4i32 (VMLAslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
3825 (v2i32 (EXTRACT_SUBREG QPR:$src3,
3826 (DSubReg_i32_reg imm:$lane))),
3827 (SubReg_i32_lane imm:$lane)))>;
3829 def : Pat<(v4f32 (fadd_mlx (v4f32 QPR:$src1),
3830 (fmul_su (v4f32 QPR:$src2),
3831 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
3832 (v4f32 (VMLAslfq (v4f32 QPR:$src1),
3834 (v2f32 (EXTRACT_SUBREG QPR:$src3,
3835 (DSubReg_i32_reg imm:$lane))),
3836 (SubReg_i32_lane imm:$lane)))>,
3837 Requires<[HasNEON, UseFPVMLx]>;
3839 // VMLAL : Vector Multiply Accumulate Long (Q += D * D)
3840 defm VMLALs : N3VLMulOp_QHS<0,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
3841 "vmlal", "s", NEONvmulls, add>;
3842 defm VMLALu : N3VLMulOp_QHS<1,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
3843 "vmlal", "u", NEONvmullu, add>;
3845 defm VMLALsls : N3VLMulOpSL_HS<0, 0b0010, "vmlal", "s", NEONvmulls, add>;
3846 defm VMLALslu : N3VLMulOpSL_HS<1, 0b0010, "vmlal", "u", NEONvmullu, add>;
3848 // VQDMLAL : Vector Saturating Doubling Multiply Accumulate Long (Q += D * D)
3849 defm VQDMLAL : N3VLInt3_HS<0, 1, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3850 "vqdmlal", "s", int_arm_neon_vqdmlal>;
3851 defm VQDMLALsl: N3VLInt3SL_HS<0, 0b0011, "vqdmlal", "s", int_arm_neon_vqdmlal>;
3853 // VMLS : Vector Multiply Subtract (integer and floating-point)
3854 defm VMLS : N3VMulOp_QHS<1, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3855 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
3856 def VMLSfd : N3VDMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACD, "vmls", "f32",
3857 v2f32, fmul_su, fsub_mlx>,
3858 Requires<[HasNEON, UseFPVMLx]>;
3859 def VMLSfq : N3VQMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACQ, "vmls", "f32",
3860 v4f32, fmul_su, fsub_mlx>,
3861 Requires<[HasNEON, UseFPVMLx]>;
3862 defm VMLSsl : N3VMulOpSL_HS<0b0100, IIC_VMACi16D, IIC_VMACi32D,
3863 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
3864 def VMLSslfd : N3VDMulOpSL<0b10, 0b0101, IIC_VMACD, "vmls", "f32",
3865 v2f32, fmul_su, fsub_mlx>,
3866 Requires<[HasNEON, UseFPVMLx]>;
3867 def VMLSslfq : N3VQMulOpSL<0b10, 0b0101, IIC_VMACQ, "vmls", "f32",
3868 v4f32, v2f32, fmul_su, fsub_mlx>,
3869 Requires<[HasNEON, UseFPVMLx]>;
3871 def : Pat<(v8i16 (sub (v8i16 QPR:$src1),
3872 (mul (v8i16 QPR:$src2),
3873 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
3874 (v8i16 (VMLSslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
3875 (v4i16 (EXTRACT_SUBREG QPR:$src3,
3876 (DSubReg_i16_reg imm:$lane))),
3877 (SubReg_i16_lane imm:$lane)))>;
3879 def : Pat<(v4i32 (sub (v4i32 QPR:$src1),
3880 (mul (v4i32 QPR:$src2),
3881 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
3882 (v4i32 (VMLSslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
3883 (v2i32 (EXTRACT_SUBREG QPR:$src3,
3884 (DSubReg_i32_reg imm:$lane))),
3885 (SubReg_i32_lane imm:$lane)))>;
3887 def : Pat<(v4f32 (fsub_mlx (v4f32 QPR:$src1),
3888 (fmul_su (v4f32 QPR:$src2),
3889 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
3890 (v4f32 (VMLSslfq (v4f32 QPR:$src1), (v4f32 QPR:$src2),
3891 (v2f32 (EXTRACT_SUBREG QPR:$src3,
3892 (DSubReg_i32_reg imm:$lane))),
3893 (SubReg_i32_lane imm:$lane)))>,
3894 Requires<[HasNEON, UseFPVMLx]>;
3896 // VMLSL : Vector Multiply Subtract Long (Q -= D * D)
3897 defm VMLSLs : N3VLMulOp_QHS<0,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
3898 "vmlsl", "s", NEONvmulls, sub>;
3899 defm VMLSLu : N3VLMulOp_QHS<1,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
3900 "vmlsl", "u", NEONvmullu, sub>;
3902 defm VMLSLsls : N3VLMulOpSL_HS<0, 0b0110, "vmlsl", "s", NEONvmulls, sub>;
3903 defm VMLSLslu : N3VLMulOpSL_HS<1, 0b0110, "vmlsl", "u", NEONvmullu, sub>;
3905 // VQDMLSL : Vector Saturating Doubling Multiply Subtract Long (Q -= D * D)
3906 defm VQDMLSL : N3VLInt3_HS<0, 1, 0b1011, 0, IIC_VMACi16D, IIC_VMACi32D,
3907 "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
3908 defm VQDMLSLsl: N3VLInt3SL_HS<0, 0b111, "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
3910 // Vector Subtract Operations.
3912 // VSUB : Vector Subtract (integer and floating-point)
3913 defm VSUB : N3V_QHSD<1, 0, 0b1000, 0, IIC_VSUBiD, IIC_VSUBiQ,
3914 "vsub", "i", sub, 0>;
3915 def VSUBfd : N3VD<0, 0, 0b10, 0b1101, 0, IIC_VBIND, "vsub", "f32",
3916 v2f32, v2f32, fsub, 0>;
3917 def VSUBfq : N3VQ<0, 0, 0b10, 0b1101, 0, IIC_VBINQ, "vsub", "f32",
3918 v4f32, v4f32, fsub, 0>;
3919 // VSUBL : Vector Subtract Long (Q = D - D)
3920 defm VSUBLs : N3VLExt_QHS<0,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
3921 "vsubl", "s", sub, sext, 0>;
3922 defm VSUBLu : N3VLExt_QHS<1,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
3923 "vsubl", "u", sub, zext, 0>;
3924 // VSUBW : Vector Subtract Wide (Q = Q - D)
3925 defm VSUBWs : N3VW_QHS<0,1,0b0011,0, "vsubw", "s", sub, sext, 0>;
3926 defm VSUBWu : N3VW_QHS<1,1,0b0011,0, "vsubw", "u", sub, zext, 0>;
3927 // VHSUB : Vector Halving Subtract
3928 defm VHSUBs : N3VInt_QHS<0, 0, 0b0010, 0, N3RegFrm,
3929 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3930 "vhsub", "s", int_arm_neon_vhsubs, 0>;
3931 defm VHSUBu : N3VInt_QHS<1, 0, 0b0010, 0, N3RegFrm,
3932 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3933 "vhsub", "u", int_arm_neon_vhsubu, 0>;
3934 // VQSUB : Vector Saturing Subtract
3935 defm VQSUBs : N3VInt_QHSD<0, 0, 0b0010, 1, N3RegFrm,
3936 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3937 "vqsub", "s", int_arm_neon_vqsubs, 0>;
3938 defm VQSUBu : N3VInt_QHSD<1, 0, 0b0010, 1, N3RegFrm,
3939 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3940 "vqsub", "u", int_arm_neon_vqsubu, 0>;
3941 // VSUBHN : Vector Subtract and Narrow Returning High Half (D = Q - Q)
3942 defm VSUBHN : N3VNInt_HSD<0,1,0b0110,0, "vsubhn", "i",
3943 int_arm_neon_vsubhn, 0>;
3944 // VRSUBHN : Vector Rounding Subtract and Narrow Returning High Half (D=Q-Q)
3945 defm VRSUBHN : N3VNInt_HSD<1,1,0b0110,0, "vrsubhn", "i",
3946 int_arm_neon_vrsubhn, 0>;
3948 // Vector Comparisons.
3950 // VCEQ : Vector Compare Equal
3951 defm VCEQ : N3V_QHS<1, 0, 0b1000, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3952 IIC_VSUBi4Q, "vceq", "i", NEONvceq, 1>;
3953 def VCEQfd : N3VD<0,0,0b00,0b1110,0, IIC_VBIND, "vceq", "f32", v2i32, v2f32,
3955 def VCEQfq : N3VQ<0,0,0b00,0b1110,0, IIC_VBINQ, "vceq", "f32", v4i32, v4f32,
3958 defm VCEQz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00010, 0, "vceq", "i",
3959 "$Vd, $Vm, #0", NEONvceqz>;
3961 // VCGE : Vector Compare Greater Than or Equal
3962 defm VCGEs : N3V_QHS<0, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3963 IIC_VSUBi4Q, "vcge", "s", NEONvcge, 0>;
3964 defm VCGEu : N3V_QHS<1, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3965 IIC_VSUBi4Q, "vcge", "u", NEONvcgeu, 0>;
3966 def VCGEfd : N3VD<1,0,0b00,0b1110,0, IIC_VBIND, "vcge", "f32", v2i32, v2f32,
3968 def VCGEfq : N3VQ<1,0,0b00,0b1110,0, IIC_VBINQ, "vcge", "f32", v4i32, v4f32,
3971 defm VCGEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00001, 0, "vcge", "s",
3972 "$Vd, $Vm, #0", NEONvcgez>;
3973 defm VCLEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00011, 0, "vcle", "s",
3974 "$Vd, $Vm, #0", NEONvclez>;
3976 // VCGT : Vector Compare Greater Than
3977 defm VCGTs : N3V_QHS<0, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3978 IIC_VSUBi4Q, "vcgt", "s", NEONvcgt, 0>;
3979 defm VCGTu : N3V_QHS<1, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3980 IIC_VSUBi4Q, "vcgt", "u", NEONvcgtu, 0>;
3981 def VCGTfd : N3VD<1,0,0b10,0b1110,0, IIC_VBIND, "vcgt", "f32", v2i32, v2f32,
3983 def VCGTfq : N3VQ<1,0,0b10,0b1110,0, IIC_VBINQ, "vcgt", "f32", v4i32, v4f32,
3986 defm VCGTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00000, 0, "vcgt", "s",
3987 "$Vd, $Vm, #0", NEONvcgtz>;
3988 defm VCLTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00100, 0, "vclt", "s",
3989 "$Vd, $Vm, #0", NEONvcltz>;
3991 // VACGE : Vector Absolute Compare Greater Than or Equal (aka VCAGE)
3992 def VACGEd : N3VDInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacge",
3993 "f32", v2i32, v2f32, int_arm_neon_vacged, 0>;
3994 def VACGEq : N3VQInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacge",
3995 "f32", v4i32, v4f32, int_arm_neon_vacgeq, 0>;
3996 // VACGT : Vector Absolute Compare Greater Than (aka VCAGT)
3997 def VACGTd : N3VDInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacgt",
3998 "f32", v2i32, v2f32, int_arm_neon_vacgtd, 0>;
3999 def VACGTq : N3VQInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacgt",
4000 "f32", v4i32, v4f32, int_arm_neon_vacgtq, 0>;
4001 // VTST : Vector Test Bits
4002 defm VTST : N3V_QHS<0, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
4003 IIC_VBINi4Q, "vtst", "", NEONvtst, 1>;
4005 // Vector Bitwise Operations.
4007 def vnotd : PatFrag<(ops node:$in),
4008 (xor node:$in, (bitconvert (v8i8 NEONimmAllOnesV)))>;
4009 def vnotq : PatFrag<(ops node:$in),
4010 (xor node:$in, (bitconvert (v16i8 NEONimmAllOnesV)))>;
4013 // VAND : Vector Bitwise AND
4014 def VANDd : N3VDX<0, 0, 0b00, 0b0001, 1, IIC_VBINiD, "vand",
4015 v2i32, v2i32, and, 1>;
4016 def VANDq : N3VQX<0, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "vand",
4017 v4i32, v4i32, and, 1>;
4019 // VEOR : Vector Bitwise Exclusive OR
4020 def VEORd : N3VDX<1, 0, 0b00, 0b0001, 1, IIC_VBINiD, "veor",
4021 v2i32, v2i32, xor, 1>;
4022 def VEORq : N3VQX<1, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "veor",
4023 v4i32, v4i32, xor, 1>;
4025 // VORR : Vector Bitwise OR
4026 def VORRd : N3VDX<0, 0, 0b10, 0b0001, 1, IIC_VBINiD, "vorr",
4027 v2i32, v2i32, or, 1>;
4028 def VORRq : N3VQX<0, 0, 0b10, 0b0001, 1, IIC_VBINiQ, "vorr",
4029 v4i32, v4i32, or, 1>;
4031 def VORRiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 0, 1,
4032 (outs DPR:$Vd), (ins nImmSplatI16:$SIMM, DPR:$src),
4034 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
4036 (v4i16 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
4037 let Inst{9} = SIMM{9};
4040 def VORRiv2i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 0, 0, 1,
4041 (outs DPR:$Vd), (ins nImmSplatI32:$SIMM, DPR:$src),
4043 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
4045 (v2i32 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
4046 let Inst{10-9} = SIMM{10-9};
4049 def VORRiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 0, 1,
4050 (outs QPR:$Vd), (ins nImmSplatI16:$SIMM, QPR:$src),
4052 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
4054 (v8i16 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
4055 let Inst{9} = SIMM{9};
4058 def VORRiv4i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 1, 0, 1,
4059 (outs QPR:$Vd), (ins nImmSplatI32:$SIMM, QPR:$src),
4061 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
4063 (v4i32 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
4064 let Inst{10-9} = SIMM{10-9};
4068 // VBIC : Vector Bitwise Bit Clear (AND NOT)
4069 def VBICd : N3VX<0, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
4070 (ins DPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VBINiD,
4071 "vbic", "$Vd, $Vn, $Vm", "",
4072 [(set DPR:$Vd, (v2i32 (and DPR:$Vn,
4073 (vnotd DPR:$Vm))))]>;
4074 def VBICq : N3VX<0, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
4075 (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINiQ,
4076 "vbic", "$Vd, $Vn, $Vm", "",
4077 [(set QPR:$Vd, (v4i32 (and QPR:$Vn,
4078 (vnotq QPR:$Vm))))]>;
4080 def VBICiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 1, 1,
4081 (outs DPR:$Vd), (ins nImmSplatI16:$SIMM, DPR:$src),
4083 "vbic", "i16", "$Vd, $SIMM", "$src = $Vd",
4085 (v4i16 (NEONvbicImm DPR:$src, timm:$SIMM)))]> {
4086 let Inst{9} = SIMM{9};
4089 def VBICiv2i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 0, 1, 1,
4090 (outs DPR:$Vd), (ins nImmSplatI32:$SIMM, DPR:$src),
4092 "vbic", "i32", "$Vd, $SIMM", "$src = $Vd",
4094 (v2i32 (NEONvbicImm DPR:$src, timm:$SIMM)))]> {
4095 let Inst{10-9} = SIMM{10-9};
4098 def VBICiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 1, 1,
4099 (outs QPR:$Vd), (ins nImmSplatI16:$SIMM, QPR:$src),
4101 "vbic", "i16", "$Vd, $SIMM", "$src = $Vd",
4103 (v8i16 (NEONvbicImm QPR:$src, timm:$SIMM)))]> {
4104 let Inst{9} = SIMM{9};
4107 def VBICiv4i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 1, 1, 1,
4108 (outs QPR:$Vd), (ins nImmSplatI32:$SIMM, QPR:$src),
4110 "vbic", "i32", "$Vd, $SIMM", "$src = $Vd",
4112 (v4i32 (NEONvbicImm QPR:$src, timm:$SIMM)))]> {
4113 let Inst{10-9} = SIMM{10-9};
4116 // VORN : Vector Bitwise OR NOT
4117 def VORNd : N3VX<0, 0, 0b11, 0b0001, 0, 1, (outs DPR:$Vd),
4118 (ins DPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VBINiD,
4119 "vorn", "$Vd, $Vn, $Vm", "",
4120 [(set DPR:$Vd, (v2i32 (or DPR:$Vn,
4121 (vnotd DPR:$Vm))))]>;
4122 def VORNq : N3VX<0, 0, 0b11, 0b0001, 1, 1, (outs QPR:$Vd),
4123 (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINiQ,
4124 "vorn", "$Vd, $Vn, $Vm", "",
4125 [(set QPR:$Vd, (v4i32 (or QPR:$Vn,
4126 (vnotq QPR:$Vm))))]>;
4128 // VMVN : Vector Bitwise NOT (Immediate)
4130 let isReMaterializable = 1 in {
4132 def VMVNv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 1, 1, (outs DPR:$Vd),
4133 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4134 "vmvn", "i16", "$Vd, $SIMM", "",
4135 [(set DPR:$Vd, (v4i16 (NEONvmvnImm timm:$SIMM)))]> {
4136 let Inst{9} = SIMM{9};
4139 def VMVNv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 1, 1, (outs QPR:$Vd),
4140 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4141 "vmvn", "i16", "$Vd, $SIMM", "",
4142 [(set QPR:$Vd, (v8i16 (NEONvmvnImm timm:$SIMM)))]> {
4143 let Inst{9} = SIMM{9};
4146 def VMVNv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 1, 1, (outs DPR:$Vd),
4147 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4148 "vmvn", "i32", "$Vd, $SIMM", "",
4149 [(set DPR:$Vd, (v2i32 (NEONvmvnImm timm:$SIMM)))]> {
4150 let Inst{11-8} = SIMM{11-8};
4153 def VMVNv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 1, 1, (outs QPR:$Vd),
4154 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4155 "vmvn", "i32", "$Vd, $SIMM", "",
4156 [(set QPR:$Vd, (v4i32 (NEONvmvnImm timm:$SIMM)))]> {
4157 let Inst{11-8} = SIMM{11-8};
4161 // VMVN : Vector Bitwise NOT
4162 def VMVNd : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 0, 0,
4163 (outs DPR:$Vd), (ins DPR:$Vm), IIC_VSUBiD,
4164 "vmvn", "$Vd, $Vm", "",
4165 [(set DPR:$Vd, (v2i32 (vnotd DPR:$Vm)))]>;
4166 def VMVNq : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 1, 0,
4167 (outs QPR:$Vd), (ins QPR:$Vm), IIC_VSUBiD,
4168 "vmvn", "$Vd, $Vm", "",
4169 [(set QPR:$Vd, (v4i32 (vnotq QPR:$Vm)))]>;
4170 def : Pat<(v2i32 (vnotd DPR:$src)), (VMVNd DPR:$src)>;
4171 def : Pat<(v4i32 (vnotq QPR:$src)), (VMVNq QPR:$src)>;
4173 // VBSL : Vector Bitwise Select
4174 def VBSLd : N3VX<1, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
4175 (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
4176 N3RegFrm, IIC_VCNTiD,
4177 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4179 (v2i32 (NEONvbsl DPR:$src1, DPR:$Vn, DPR:$Vm)))]>;
4181 def : Pat<(v2i32 (or (and DPR:$Vn, DPR:$Vd),
4182 (and DPR:$Vm, (vnotd DPR:$Vd)))),
4183 (VBSLd DPR:$Vd, DPR:$Vn, DPR:$Vm)>;
4185 def VBSLq : N3VX<1, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
4186 (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
4187 N3RegFrm, IIC_VCNTiQ,
4188 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4190 (v4i32 (NEONvbsl QPR:$src1, QPR:$Vn, QPR:$Vm)))]>;
4192 def : Pat<(v4i32 (or (and QPR:$Vn, QPR:$Vd),
4193 (and QPR:$Vm, (vnotq QPR:$Vd)))),
4194 (VBSLq QPR:$Vd, QPR:$Vn, QPR:$Vm)>;
4196 // VBIF : Vector Bitwise Insert if False
4197 // like VBSL but with: "vbif $dst, $src3, $src1", "$src2 = $dst",
4198 // FIXME: This instruction's encoding MAY NOT BE correct.
4199 def VBIFd : N3VX<1, 0, 0b11, 0b0001, 0, 1,
4200 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
4201 N3RegFrm, IIC_VBINiD,
4202 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4204 def VBIFq : N3VX<1, 0, 0b11, 0b0001, 1, 1,
4205 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
4206 N3RegFrm, IIC_VBINiQ,
4207 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4210 // VBIT : Vector Bitwise Insert if True
4211 // like VBSL but with: "vbit $dst, $src2, $src1", "$src3 = $dst",
4212 // FIXME: This instruction's encoding MAY NOT BE correct.
4213 def VBITd : N3VX<1, 0, 0b10, 0b0001, 0, 1,
4214 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
4215 N3RegFrm, IIC_VBINiD,
4216 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4218 def VBITq : N3VX<1, 0, 0b10, 0b0001, 1, 1,
4219 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
4220 N3RegFrm, IIC_VBINiQ,
4221 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4224 // VBIT/VBIF are not yet implemented. The TwoAddress pass will not go looking
4225 // for equivalent operations with different register constraints; it just
4228 // Vector Absolute Differences.
4230 // VABD : Vector Absolute Difference
4231 defm VABDs : N3VInt_QHS<0, 0, 0b0111, 0, N3RegFrm,
4232 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4233 "vabd", "s", int_arm_neon_vabds, 1>;
4234 defm VABDu : N3VInt_QHS<1, 0, 0b0111, 0, N3RegFrm,
4235 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4236 "vabd", "u", int_arm_neon_vabdu, 1>;
4237 def VABDfd : N3VDInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBIND,
4238 "vabd", "f32", v2f32, v2f32, int_arm_neon_vabds, 1>;
4239 def VABDfq : N3VQInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBINQ,
4240 "vabd", "f32", v4f32, v4f32, int_arm_neon_vabds, 1>;
4242 // VABDL : Vector Absolute Difference Long (Q = | D - D |)
4243 defm VABDLs : N3VLIntExt_QHS<0,1,0b0111,0, IIC_VSUBi4Q,
4244 "vabdl", "s", int_arm_neon_vabds, zext, 1>;
4245 defm VABDLu : N3VLIntExt_QHS<1,1,0b0111,0, IIC_VSUBi4Q,
4246 "vabdl", "u", int_arm_neon_vabdu, zext, 1>;
4248 // VABA : Vector Absolute Difference and Accumulate
4249 defm VABAs : N3VIntOp_QHS<0,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
4250 "vaba", "s", int_arm_neon_vabds, add>;
4251 defm VABAu : N3VIntOp_QHS<1,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
4252 "vaba", "u", int_arm_neon_vabdu, add>;
4254 // VABAL : Vector Absolute Difference and Accumulate Long (Q += | D - D |)
4255 defm VABALs : N3VLIntExtOp_QHS<0,1,0b0101,0, IIC_VABAD,
4256 "vabal", "s", int_arm_neon_vabds, zext, add>;
4257 defm VABALu : N3VLIntExtOp_QHS<1,1,0b0101,0, IIC_VABAD,
4258 "vabal", "u", int_arm_neon_vabdu, zext, add>;
4260 // Vector Maximum and Minimum.
4262 // VMAX : Vector Maximum
4263 defm VMAXs : N3VInt_QHS<0, 0, 0b0110, 0, N3RegFrm,
4264 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4265 "vmax", "s", int_arm_neon_vmaxs, 1>;
4266 defm VMAXu : N3VInt_QHS<1, 0, 0b0110, 0, N3RegFrm,
4267 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4268 "vmax", "u", int_arm_neon_vmaxu, 1>;
4269 def VMAXfd : N3VDInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBIND,
4271 v2f32, v2f32, int_arm_neon_vmaxs, 1>;
4272 def VMAXfq : N3VQInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBINQ,
4274 v4f32, v4f32, int_arm_neon_vmaxs, 1>;
4276 // VMIN : Vector Minimum
4277 defm VMINs : N3VInt_QHS<0, 0, 0b0110, 1, N3RegFrm,
4278 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4279 "vmin", "s", int_arm_neon_vmins, 1>;
4280 defm VMINu : N3VInt_QHS<1, 0, 0b0110, 1, N3RegFrm,
4281 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4282 "vmin", "u", int_arm_neon_vminu, 1>;
4283 def VMINfd : N3VDInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBIND,
4285 v2f32, v2f32, int_arm_neon_vmins, 1>;
4286 def VMINfq : N3VQInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBINQ,
4288 v4f32, v4f32, int_arm_neon_vmins, 1>;
4290 // Vector Pairwise Operations.
4292 // VPADD : Vector Pairwise Add
4293 def VPADDi8 : N3VDInt<0, 0, 0b00, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4295 v8i8, v8i8, int_arm_neon_vpadd, 0>;
4296 def VPADDi16 : N3VDInt<0, 0, 0b01, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4298 v4i16, v4i16, int_arm_neon_vpadd, 0>;
4299 def VPADDi32 : N3VDInt<0, 0, 0b10, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4301 v2i32, v2i32, int_arm_neon_vpadd, 0>;
4302 def VPADDf : N3VDInt<1, 0, 0b00, 0b1101, 0, N3RegFrm,
4303 IIC_VPBIND, "vpadd", "f32",
4304 v2f32, v2f32, int_arm_neon_vpadd, 0>;
4306 // VPADDL : Vector Pairwise Add Long
4307 defm VPADDLs : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00100, 0, "vpaddl", "s",
4308 int_arm_neon_vpaddls>;
4309 defm VPADDLu : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00101, 0, "vpaddl", "u",
4310 int_arm_neon_vpaddlu>;
4312 // VPADAL : Vector Pairwise Add and Accumulate Long
4313 defm VPADALs : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01100, 0, "vpadal", "s",
4314 int_arm_neon_vpadals>;
4315 defm VPADALu : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01101, 0, "vpadal", "u",
4316 int_arm_neon_vpadalu>;
4318 // VPMAX : Vector Pairwise Maximum
4319 def VPMAXs8 : N3VDInt<0, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4320 "s8", v8i8, v8i8, int_arm_neon_vpmaxs, 0>;
4321 def VPMAXs16 : N3VDInt<0, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4322 "s16", v4i16, v4i16, int_arm_neon_vpmaxs, 0>;
4323 def VPMAXs32 : N3VDInt<0, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4324 "s32", v2i32, v2i32, int_arm_neon_vpmaxs, 0>;
4325 def VPMAXu8 : N3VDInt<1, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4326 "u8", v8i8, v8i8, int_arm_neon_vpmaxu, 0>;
4327 def VPMAXu16 : N3VDInt<1, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4328 "u16", v4i16, v4i16, int_arm_neon_vpmaxu, 0>;
4329 def VPMAXu32 : N3VDInt<1, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4330 "u32", v2i32, v2i32, int_arm_neon_vpmaxu, 0>;
4331 def VPMAXf : N3VDInt<1, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmax",
4332 "f32", v2f32, v2f32, int_arm_neon_vpmaxs, 0>;
4334 // VPMIN : Vector Pairwise Minimum
4335 def VPMINs8 : N3VDInt<0, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4336 "s8", v8i8, v8i8, int_arm_neon_vpmins, 0>;
4337 def VPMINs16 : N3VDInt<0, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4338 "s16", v4i16, v4i16, int_arm_neon_vpmins, 0>;
4339 def VPMINs32 : N3VDInt<0, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4340 "s32", v2i32, v2i32, int_arm_neon_vpmins, 0>;
4341 def VPMINu8 : N3VDInt<1, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4342 "u8", v8i8, v8i8, int_arm_neon_vpminu, 0>;
4343 def VPMINu16 : N3VDInt<1, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4344 "u16", v4i16, v4i16, int_arm_neon_vpminu, 0>;
4345 def VPMINu32 : N3VDInt<1, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4346 "u32", v2i32, v2i32, int_arm_neon_vpminu, 0>;
4347 def VPMINf : N3VDInt<1, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmin",
4348 "f32", v2f32, v2f32, int_arm_neon_vpmins, 0>;
4350 // Vector Reciprocal and Reciprocal Square Root Estimate and Step.
4352 // VRECPE : Vector Reciprocal Estimate
4353 def VRECPEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
4354 IIC_VUNAD, "vrecpe", "u32",
4355 v2i32, v2i32, int_arm_neon_vrecpe>;
4356 def VRECPEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
4357 IIC_VUNAQ, "vrecpe", "u32",
4358 v4i32, v4i32, int_arm_neon_vrecpe>;
4359 def VRECPEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
4360 IIC_VUNAD, "vrecpe", "f32",
4361 v2f32, v2f32, int_arm_neon_vrecpe>;
4362 def VRECPEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
4363 IIC_VUNAQ, "vrecpe", "f32",
4364 v4f32, v4f32, int_arm_neon_vrecpe>;
4366 // VRECPS : Vector Reciprocal Step
4367 def VRECPSfd : N3VDInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
4368 IIC_VRECSD, "vrecps", "f32",
4369 v2f32, v2f32, int_arm_neon_vrecps, 1>;
4370 def VRECPSfq : N3VQInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
4371 IIC_VRECSQ, "vrecps", "f32",
4372 v4f32, v4f32, int_arm_neon_vrecps, 1>;
4374 // VRSQRTE : Vector Reciprocal Square Root Estimate
4375 def VRSQRTEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
4376 IIC_VUNAD, "vrsqrte", "u32",
4377 v2i32, v2i32, int_arm_neon_vrsqrte>;
4378 def VRSQRTEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
4379 IIC_VUNAQ, "vrsqrte", "u32",
4380 v4i32, v4i32, int_arm_neon_vrsqrte>;
4381 def VRSQRTEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
4382 IIC_VUNAD, "vrsqrte", "f32",
4383 v2f32, v2f32, int_arm_neon_vrsqrte>;
4384 def VRSQRTEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
4385 IIC_VUNAQ, "vrsqrte", "f32",
4386 v4f32, v4f32, int_arm_neon_vrsqrte>;
4388 // VRSQRTS : Vector Reciprocal Square Root Step
4389 def VRSQRTSfd : N3VDInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
4390 IIC_VRECSD, "vrsqrts", "f32",
4391 v2f32, v2f32, int_arm_neon_vrsqrts, 1>;
4392 def VRSQRTSfq : N3VQInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
4393 IIC_VRECSQ, "vrsqrts", "f32",
4394 v4f32, v4f32, int_arm_neon_vrsqrts, 1>;
4398 // VSHL : Vector Shift
4399 defm VSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 0, N3RegVShFrm,
4400 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
4401 "vshl", "s", int_arm_neon_vshifts>;
4402 defm VSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 0, N3RegVShFrm,
4403 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
4404 "vshl", "u", int_arm_neon_vshiftu>;
4406 // VSHL : Vector Shift Left (Immediate)
4407 defm VSHLi : N2VShL_QHSD<0, 1, 0b0101, 1, IIC_VSHLiD, "vshl", "i", NEONvshl>;
4409 // VSHR : Vector Shift Right (Immediate)
4410 defm VSHRs : N2VShR_QHSD<0, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "s",NEONvshrs>;
4411 defm VSHRu : N2VShR_QHSD<1, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "u",NEONvshru>;
4413 // VSHLL : Vector Shift Left Long
4414 defm VSHLLs : N2VLSh_QHS<0, 1, 0b1010, 0, 0, 1, "vshll", "s", NEONvshlls>;
4415 defm VSHLLu : N2VLSh_QHS<1, 1, 0b1010, 0, 0, 1, "vshll", "u", NEONvshllu>;
4417 // VSHLL : Vector Shift Left Long (with maximum shift count)
4418 class N2VLShMax<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
4419 bit op6, bit op4, string OpcodeStr, string Dt, ValueType ResTy,
4420 ValueType OpTy, Operand ImmTy, SDNode OpNode>
4421 : N2VLSh<op24, op23, op11_8, op7, op6, op4, OpcodeStr, Dt,
4422 ResTy, OpTy, ImmTy, OpNode> {
4423 let Inst{21-16} = op21_16;
4424 let DecoderMethod = "DecodeVSHLMaxInstruction";
4426 def VSHLLi8 : N2VLShMax<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll", "i8",
4427 v8i16, v8i8, imm8, NEONvshlli>;
4428 def VSHLLi16 : N2VLShMax<1, 1, 0b110110, 0b0011, 0, 0, 0, "vshll", "i16",
4429 v4i32, v4i16, imm16, NEONvshlli>;
4430 def VSHLLi32 : N2VLShMax<1, 1, 0b111010, 0b0011, 0, 0, 0, "vshll", "i32",
4431 v2i64, v2i32, imm32, NEONvshlli>;
4433 // VSHRN : Vector Shift Right and Narrow
4434 defm VSHRN : N2VNSh_HSD<0,1,0b1000,0,0,1, IIC_VSHLiD, "vshrn", "i",
4437 // VRSHL : Vector Rounding Shift
4438 defm VRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 0, N3RegVShFrm,
4439 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4440 "vrshl", "s", int_arm_neon_vrshifts>;
4441 defm VRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 0, N3RegVShFrm,
4442 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4443 "vrshl", "u", int_arm_neon_vrshiftu>;
4444 // VRSHR : Vector Rounding Shift Right
4445 defm VRSHRs : N2VShR_QHSD<0,1,0b0010,1, IIC_VSHLi4D, "vrshr", "s",NEONvrshrs>;
4446 defm VRSHRu : N2VShR_QHSD<1,1,0b0010,1, IIC_VSHLi4D, "vrshr", "u",NEONvrshru>;
4448 // VRSHRN : Vector Rounding Shift Right and Narrow
4449 defm VRSHRN : N2VNSh_HSD<0, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vrshrn", "i",
4452 // VQSHL : Vector Saturating Shift
4453 defm VQSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 1, N3RegVShFrm,
4454 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4455 "vqshl", "s", int_arm_neon_vqshifts>;
4456 defm VQSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 1, N3RegVShFrm,
4457 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4458 "vqshl", "u", int_arm_neon_vqshiftu>;
4459 // VQSHL : Vector Saturating Shift Left (Immediate)
4460 defm VQSHLsi : N2VShL_QHSD<0,1,0b0111,1, IIC_VSHLi4D, "vqshl", "s",NEONvqshls>;
4461 defm VQSHLui : N2VShL_QHSD<1,1,0b0111,1, IIC_VSHLi4D, "vqshl", "u",NEONvqshlu>;
4463 // VQSHLU : Vector Saturating Shift Left (Immediate, Unsigned)
4464 defm VQSHLsu : N2VShL_QHSD<1,1,0b0110,1, IIC_VSHLi4D,"vqshlu","s",NEONvqshlsu>;
4466 // VQSHRN : Vector Saturating Shift Right and Narrow
4467 defm VQSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "s",
4469 defm VQSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "u",
4472 // VQSHRUN : Vector Saturating Shift Right and Narrow (Unsigned)
4473 defm VQSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 0, 1, IIC_VSHLi4D, "vqshrun", "s",
4476 // VQRSHL : Vector Saturating Rounding Shift
4477 defm VQRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 1, N3RegVShFrm,
4478 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4479 "vqrshl", "s", int_arm_neon_vqrshifts>;
4480 defm VQRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 1, N3RegVShFrm,
4481 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4482 "vqrshl", "u", int_arm_neon_vqrshiftu>;
4484 // VQRSHRN : Vector Saturating Rounding Shift Right and Narrow
4485 defm VQRSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "s",
4487 defm VQRSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "u",
4490 // VQRSHRUN : Vector Saturating Rounding Shift Right and Narrow (Unsigned)
4491 defm VQRSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vqrshrun", "s",
4494 // VSRA : Vector Shift Right and Accumulate
4495 defm VSRAs : N2VShAdd_QHSD<0, 1, 0b0001, 1, "vsra", "s", NEONvshrs>;
4496 defm VSRAu : N2VShAdd_QHSD<1, 1, 0b0001, 1, "vsra", "u", NEONvshru>;
4497 // VRSRA : Vector Rounding Shift Right and Accumulate
4498 defm VRSRAs : N2VShAdd_QHSD<0, 1, 0b0011, 1, "vrsra", "s", NEONvrshrs>;
4499 defm VRSRAu : N2VShAdd_QHSD<1, 1, 0b0011, 1, "vrsra", "u", NEONvrshru>;
4501 // VSLI : Vector Shift Left and Insert
4502 defm VSLI : N2VShInsL_QHSD<1, 1, 0b0101, 1, "vsli">;
4504 // VSRI : Vector Shift Right and Insert
4505 defm VSRI : N2VShInsR_QHSD<1, 1, 0b0100, 1, "vsri">;
4507 // Vector Absolute and Saturating Absolute.
4509 // VABS : Vector Absolute Value
4510 defm VABS : N2VInt_QHS<0b11, 0b11, 0b01, 0b00110, 0,
4511 IIC_VUNAiD, IIC_VUNAiQ, "vabs", "s",
4513 def VABSfd : N2VDInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
4514 IIC_VUNAD, "vabs", "f32",
4515 v2f32, v2f32, int_arm_neon_vabs>;
4516 def VABSfq : N2VQInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
4517 IIC_VUNAQ, "vabs", "f32",
4518 v4f32, v4f32, int_arm_neon_vabs>;
4520 // VQABS : Vector Saturating Absolute Value
4521 defm VQABS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01110, 0,
4522 IIC_VQUNAiD, IIC_VQUNAiQ, "vqabs", "s",
4523 int_arm_neon_vqabs>;
4527 def vnegd : PatFrag<(ops node:$in),
4528 (sub (bitconvert (v2i32 NEONimmAllZerosV)), node:$in)>;
4529 def vnegq : PatFrag<(ops node:$in),
4530 (sub (bitconvert (v4i32 NEONimmAllZerosV)), node:$in)>;
4532 class VNEGD<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
4533 : N2V<0b11, 0b11, size, 0b01, 0b00111, 0, 0, (outs DPR:$Vd), (ins DPR:$Vm),
4534 IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
4535 [(set DPR:$Vd, (Ty (vnegd DPR:$Vm)))]>;
4536 class VNEGQ<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
4537 : N2V<0b11, 0b11, size, 0b01, 0b00111, 1, 0, (outs QPR:$Vd), (ins QPR:$Vm),
4538 IIC_VSHLiQ, OpcodeStr, Dt, "$Vd, $Vm", "",
4539 [(set QPR:$Vd, (Ty (vnegq QPR:$Vm)))]>;
4541 // VNEG : Vector Negate (integer)
4542 def VNEGs8d : VNEGD<0b00, "vneg", "s8", v8i8>;
4543 def VNEGs16d : VNEGD<0b01, "vneg", "s16", v4i16>;
4544 def VNEGs32d : VNEGD<0b10, "vneg", "s32", v2i32>;
4545 def VNEGs8q : VNEGQ<0b00, "vneg", "s8", v16i8>;
4546 def VNEGs16q : VNEGQ<0b01, "vneg", "s16", v8i16>;
4547 def VNEGs32q : VNEGQ<0b10, "vneg", "s32", v4i32>;
4549 // VNEG : Vector Negate (floating-point)
4550 def VNEGfd : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
4551 (outs DPR:$Vd), (ins DPR:$Vm), IIC_VUNAD,
4552 "vneg", "f32", "$Vd, $Vm", "",
4553 [(set DPR:$Vd, (v2f32 (fneg DPR:$Vm)))]>;
4554 def VNEGf32q : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 1, 0,
4555 (outs QPR:$Vd), (ins QPR:$Vm), IIC_VUNAQ,
4556 "vneg", "f32", "$Vd, $Vm", "",
4557 [(set QPR:$Vd, (v4f32 (fneg QPR:$Vm)))]>;
4559 def : Pat<(v8i8 (vnegd DPR:$src)), (VNEGs8d DPR:$src)>;
4560 def : Pat<(v4i16 (vnegd DPR:$src)), (VNEGs16d DPR:$src)>;
4561 def : Pat<(v2i32 (vnegd DPR:$src)), (VNEGs32d DPR:$src)>;
4562 def : Pat<(v16i8 (vnegq QPR:$src)), (VNEGs8q QPR:$src)>;
4563 def : Pat<(v8i16 (vnegq QPR:$src)), (VNEGs16q QPR:$src)>;
4564 def : Pat<(v4i32 (vnegq QPR:$src)), (VNEGs32q QPR:$src)>;
4566 // VQNEG : Vector Saturating Negate
4567 defm VQNEG : N2VInt_QHS<0b11, 0b11, 0b00, 0b01111, 0,
4568 IIC_VQUNAiD, IIC_VQUNAiQ, "vqneg", "s",
4569 int_arm_neon_vqneg>;
4571 // Vector Bit Counting Operations.
4573 // VCLS : Vector Count Leading Sign Bits
4574 defm VCLS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01000, 0,
4575 IIC_VCNTiD, IIC_VCNTiQ, "vcls", "s",
4577 // VCLZ : Vector Count Leading Zeros
4578 defm VCLZ : N2VInt_QHS<0b11, 0b11, 0b00, 0b01001, 0,
4579 IIC_VCNTiD, IIC_VCNTiQ, "vclz", "i",
4581 // VCNT : Vector Count One Bits
4582 def VCNTd : N2VDInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
4583 IIC_VCNTiD, "vcnt", "8",
4584 v8i8, v8i8, int_arm_neon_vcnt>;
4585 def VCNTq : N2VQInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
4586 IIC_VCNTiQ, "vcnt", "8",
4587 v16i8, v16i8, int_arm_neon_vcnt>;
4590 def VSWPd : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 0, 0,
4591 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
4592 "vswp", "$Vd, $Vm", "", []>;
4593 def VSWPq : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 1, 0,
4594 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
4595 "vswp", "$Vd, $Vm", "", []>;
4597 // Vector Move Operations.
4599 // VMOV : Vector Move (Register)
4600 def : InstAlias<"vmov${p} $Vd, $Vm",
4601 (VORRd DPR:$Vd, DPR:$Vm, DPR:$Vm, pred:$p)>;
4602 def : InstAlias<"vmov${p} $Vd, $Vm",
4603 (VORRq QPR:$Vd, QPR:$Vm, QPR:$Vm, pred:$p)>;
4605 // VMOV : Vector Move (Immediate)
4607 let isReMaterializable = 1 in {
4608 def VMOVv8i8 : N1ModImm<1, 0b000, 0b1110, 0, 0, 0, 1, (outs DPR:$Vd),
4609 (ins nImmSplatI8:$SIMM), IIC_VMOVImm,
4610 "vmov", "i8", "$Vd, $SIMM", "",
4611 [(set DPR:$Vd, (v8i8 (NEONvmovImm timm:$SIMM)))]>;
4612 def VMOVv16i8 : N1ModImm<1, 0b000, 0b1110, 0, 1, 0, 1, (outs QPR:$Vd),
4613 (ins nImmSplatI8:$SIMM), IIC_VMOVImm,
4614 "vmov", "i8", "$Vd, $SIMM", "",
4615 [(set QPR:$Vd, (v16i8 (NEONvmovImm timm:$SIMM)))]>;
4617 def VMOVv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 0, 1, (outs DPR:$Vd),
4618 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4619 "vmov", "i16", "$Vd, $SIMM", "",
4620 [(set DPR:$Vd, (v4i16 (NEONvmovImm timm:$SIMM)))]> {
4621 let Inst{9} = SIMM{9};
4624 def VMOVv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 0, 1, (outs QPR:$Vd),
4625 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4626 "vmov", "i16", "$Vd, $SIMM", "",
4627 [(set QPR:$Vd, (v8i16 (NEONvmovImm timm:$SIMM)))]> {
4628 let Inst{9} = SIMM{9};
4631 def VMOVv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 0, 1, (outs DPR:$Vd),
4632 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4633 "vmov", "i32", "$Vd, $SIMM", "",
4634 [(set DPR:$Vd, (v2i32 (NEONvmovImm timm:$SIMM)))]> {
4635 let Inst{11-8} = SIMM{11-8};
4638 def VMOVv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 0, 1, (outs QPR:$Vd),
4639 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4640 "vmov", "i32", "$Vd, $SIMM", "",
4641 [(set QPR:$Vd, (v4i32 (NEONvmovImm timm:$SIMM)))]> {
4642 let Inst{11-8} = SIMM{11-8};
4645 def VMOVv1i64 : N1ModImm<1, 0b000, 0b1110, 0, 0, 1, 1, (outs DPR:$Vd),
4646 (ins nImmSplatI64:$SIMM), IIC_VMOVImm,
4647 "vmov", "i64", "$Vd, $SIMM", "",
4648 [(set DPR:$Vd, (v1i64 (NEONvmovImm timm:$SIMM)))]>;
4649 def VMOVv2i64 : N1ModImm<1, 0b000, 0b1110, 0, 1, 1, 1, (outs QPR:$Vd),
4650 (ins nImmSplatI64:$SIMM), IIC_VMOVImm,
4651 "vmov", "i64", "$Vd, $SIMM", "",
4652 [(set QPR:$Vd, (v2i64 (NEONvmovImm timm:$SIMM)))]>;
4654 def VMOVv2f32 : N1ModImm<1, 0b000, 0b1111, 0, 0, 0, 1, (outs DPR:$Vd),
4655 (ins nImmVMOVF32:$SIMM), IIC_VMOVImm,
4656 "vmov", "f32", "$Vd, $SIMM", "",
4657 [(set DPR:$Vd, (v2f32 (NEONvmovFPImm timm:$SIMM)))]>;
4658 def VMOVv4f32 : N1ModImm<1, 0b000, 0b1111, 0, 1, 0, 1, (outs QPR:$Vd),
4659 (ins nImmVMOVF32:$SIMM), IIC_VMOVImm,
4660 "vmov", "f32", "$Vd, $SIMM", "",
4661 [(set QPR:$Vd, (v4f32 (NEONvmovFPImm timm:$SIMM)))]>;
4662 } // isReMaterializable
4664 // VMOV : Vector Get Lane (move scalar to ARM core register)
4666 def VGETLNs8 : NVGetLane<{1,1,1,0,0,1,?,1}, 0b1011, {?,?},
4667 (outs GPR:$R), (ins DPR:$V, VectorIndex8:$lane),
4668 IIC_VMOVSI, "vmov", "s8", "$R, $V$lane",
4669 [(set GPR:$R, (NEONvgetlanes (v8i8 DPR:$V),
4671 let Inst{21} = lane{2};
4672 let Inst{6-5} = lane{1-0};
4674 def VGETLNs16 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, {?,1},
4675 (outs GPR:$R), (ins DPR:$V, VectorIndex16:$lane),
4676 IIC_VMOVSI, "vmov", "s16", "$R, $V$lane",
4677 [(set GPR:$R, (NEONvgetlanes (v4i16 DPR:$V),
4679 let Inst{21} = lane{1};
4680 let Inst{6} = lane{0};
4682 def VGETLNu8 : NVGetLane<{1,1,1,0,1,1,?,1}, 0b1011, {?,?},
4683 (outs GPR:$R), (ins DPR:$V, VectorIndex8:$lane),
4684 IIC_VMOVSI, "vmov", "u8", "$R, $V$lane",
4685 [(set GPR:$R, (NEONvgetlaneu (v8i8 DPR:$V),
4687 let Inst{21} = lane{2};
4688 let Inst{6-5} = lane{1-0};
4690 def VGETLNu16 : NVGetLane<{1,1,1,0,1,0,?,1}, 0b1011, {?,1},
4691 (outs GPR:$R), (ins DPR:$V, VectorIndex16:$lane),
4692 IIC_VMOVSI, "vmov", "u16", "$R, $V$lane",
4693 [(set GPR:$R, (NEONvgetlaneu (v4i16 DPR:$V),
4695 let Inst{21} = lane{1};
4696 let Inst{6} = lane{0};
4698 def VGETLNi32 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, 0b00,
4699 (outs GPR:$R), (ins DPR:$V, VectorIndex32:$lane),
4700 IIC_VMOVSI, "vmov", "32", "$R, $V$lane",
4701 [(set GPR:$R, (extractelt (v2i32 DPR:$V),
4703 let Inst{21} = lane{0};
4705 // def VGETLNf32: see FMRDH and FMRDL in ARMInstrVFP.td
4706 def : Pat<(NEONvgetlanes (v16i8 QPR:$src), imm:$lane),
4707 (VGETLNs8 (v8i8 (EXTRACT_SUBREG QPR:$src,
4708 (DSubReg_i8_reg imm:$lane))),
4709 (SubReg_i8_lane imm:$lane))>;
4710 def : Pat<(NEONvgetlanes (v8i16 QPR:$src), imm:$lane),
4711 (VGETLNs16 (v4i16 (EXTRACT_SUBREG QPR:$src,
4712 (DSubReg_i16_reg imm:$lane))),
4713 (SubReg_i16_lane imm:$lane))>;
4714 def : Pat<(NEONvgetlaneu (v16i8 QPR:$src), imm:$lane),
4715 (VGETLNu8 (v8i8 (EXTRACT_SUBREG QPR:$src,
4716 (DSubReg_i8_reg imm:$lane))),
4717 (SubReg_i8_lane imm:$lane))>;
4718 def : Pat<(NEONvgetlaneu (v8i16 QPR:$src), imm:$lane),
4719 (VGETLNu16 (v4i16 (EXTRACT_SUBREG QPR:$src,
4720 (DSubReg_i16_reg imm:$lane))),
4721 (SubReg_i16_lane imm:$lane))>;
4722 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
4723 (VGETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src,
4724 (DSubReg_i32_reg imm:$lane))),
4725 (SubReg_i32_lane imm:$lane))>;
4726 def : Pat<(extractelt (v2f32 DPR:$src1), imm:$src2),
4727 (EXTRACT_SUBREG (v2f32 (COPY_TO_REGCLASS (v2f32 DPR:$src1),DPR_VFP2)),
4728 (SSubReg_f32_reg imm:$src2))>;
4729 def : Pat<(extractelt (v4f32 QPR:$src1), imm:$src2),
4730 (EXTRACT_SUBREG (v4f32 (COPY_TO_REGCLASS (v4f32 QPR:$src1),QPR_VFP2)),
4731 (SSubReg_f32_reg imm:$src2))>;
4732 //def : Pat<(extractelt (v2i64 QPR:$src1), imm:$src2),
4733 // (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
4734 def : Pat<(extractelt (v2f64 QPR:$src1), imm:$src2),
4735 (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
4738 // VMOV : Vector Set Lane (move ARM core register to scalar)
4740 let Constraints = "$src1 = $V" in {
4741 def VSETLNi8 : NVSetLane<{1,1,1,0,0,1,?,0}, 0b1011, {?,?}, (outs DPR:$V),
4742 (ins DPR:$src1, GPR:$R, VectorIndex8:$lane),
4743 IIC_VMOVISL, "vmov", "8", "$V$lane, $R",
4744 [(set DPR:$V, (vector_insert (v8i8 DPR:$src1),
4745 GPR:$R, imm:$lane))]> {
4746 let Inst{21} = lane{2};
4747 let Inst{6-5} = lane{1-0};
4749 def VSETLNi16 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, {?,1}, (outs DPR:$V),
4750 (ins DPR:$src1, GPR:$R, VectorIndex16:$lane),
4751 IIC_VMOVISL, "vmov", "16", "$V$lane, $R",
4752 [(set DPR:$V, (vector_insert (v4i16 DPR:$src1),
4753 GPR:$R, imm:$lane))]> {
4754 let Inst{21} = lane{1};
4755 let Inst{6} = lane{0};
4757 def VSETLNi32 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, 0b00, (outs DPR:$V),
4758 (ins DPR:$src1, GPR:$R, VectorIndex32:$lane),
4759 IIC_VMOVISL, "vmov", "32", "$V$lane, $R",
4760 [(set DPR:$V, (insertelt (v2i32 DPR:$src1),
4761 GPR:$R, imm:$lane))]> {
4762 let Inst{21} = lane{0};
4765 def : Pat<(vector_insert (v16i8 QPR:$src1), GPR:$src2, imm:$lane),
4766 (v16i8 (INSERT_SUBREG QPR:$src1,
4767 (v8i8 (VSETLNi8 (v8i8 (EXTRACT_SUBREG QPR:$src1,
4768 (DSubReg_i8_reg imm:$lane))),
4769 GPR:$src2, (SubReg_i8_lane imm:$lane))),
4770 (DSubReg_i8_reg imm:$lane)))>;
4771 def : Pat<(vector_insert (v8i16 QPR:$src1), GPR:$src2, imm:$lane),
4772 (v8i16 (INSERT_SUBREG QPR:$src1,
4773 (v4i16 (VSETLNi16 (v4i16 (EXTRACT_SUBREG QPR:$src1,
4774 (DSubReg_i16_reg imm:$lane))),
4775 GPR:$src2, (SubReg_i16_lane imm:$lane))),
4776 (DSubReg_i16_reg imm:$lane)))>;
4777 def : Pat<(insertelt (v4i32 QPR:$src1), GPR:$src2, imm:$lane),
4778 (v4i32 (INSERT_SUBREG QPR:$src1,
4779 (v2i32 (VSETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src1,
4780 (DSubReg_i32_reg imm:$lane))),
4781 GPR:$src2, (SubReg_i32_lane imm:$lane))),
4782 (DSubReg_i32_reg imm:$lane)))>;
4784 def : Pat<(v2f32 (insertelt DPR:$src1, SPR:$src2, imm:$src3)),
4785 (INSERT_SUBREG (v2f32 (COPY_TO_REGCLASS DPR:$src1, DPR_VFP2)),
4786 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
4787 def : Pat<(v4f32 (insertelt QPR:$src1, SPR:$src2, imm:$src3)),
4788 (INSERT_SUBREG (v4f32 (COPY_TO_REGCLASS QPR:$src1, QPR_VFP2)),
4789 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
4791 //def : Pat<(v2i64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
4792 // (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
4793 def : Pat<(v2f64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
4794 (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
4796 def : Pat<(v2f32 (scalar_to_vector SPR:$src)),
4797 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
4798 def : Pat<(v2f64 (scalar_to_vector (f64 DPR:$src))),
4799 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
4800 def : Pat<(v4f32 (scalar_to_vector SPR:$src)),
4801 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
4803 def : Pat<(v8i8 (scalar_to_vector GPR:$src)),
4804 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4805 def : Pat<(v4i16 (scalar_to_vector GPR:$src)),
4806 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4807 def : Pat<(v2i32 (scalar_to_vector GPR:$src)),
4808 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4810 def : Pat<(v16i8 (scalar_to_vector GPR:$src)),
4811 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4812 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4814 def : Pat<(v8i16 (scalar_to_vector GPR:$src)),
4815 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
4816 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4818 def : Pat<(v4i32 (scalar_to_vector GPR:$src)),
4819 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
4820 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4823 // VDUP : Vector Duplicate (from ARM core register to all elements)
4825 class VDUPD<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
4826 : NVDup<opcod1, 0b1011, opcod3, (outs DPR:$V), (ins GPR:$R),
4827 IIC_VMOVIS, "vdup", Dt, "$V, $R",
4828 [(set DPR:$V, (Ty (NEONvdup (i32 GPR:$R))))]>;
4829 class VDUPQ<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
4830 : NVDup<opcod1, 0b1011, opcod3, (outs QPR:$V), (ins GPR:$R),
4831 IIC_VMOVIS, "vdup", Dt, "$V, $R",
4832 [(set QPR:$V, (Ty (NEONvdup (i32 GPR:$R))))]>;
4834 def VDUP8d : VDUPD<0b11101100, 0b00, "8", v8i8>;
4835 def VDUP16d : VDUPD<0b11101000, 0b01, "16", v4i16>;
4836 def VDUP32d : VDUPD<0b11101000, 0b00, "32", v2i32>;
4837 def VDUP8q : VDUPQ<0b11101110, 0b00, "8", v16i8>;
4838 def VDUP16q : VDUPQ<0b11101010, 0b01, "16", v8i16>;
4839 def VDUP32q : VDUPQ<0b11101010, 0b00, "32", v4i32>;
4841 def : Pat<(v2f32 (NEONvdup (f32 (bitconvert GPR:$R)))), (VDUP32d GPR:$R)>;
4842 def : Pat<(v4f32 (NEONvdup (f32 (bitconvert GPR:$R)))), (VDUP32q GPR:$R)>;
4844 // VDUP : Vector Duplicate Lane (from scalar to all elements)
4846 class VDUPLND<bits<4> op19_16, string OpcodeStr, string Dt,
4847 ValueType Ty, Operand IdxTy>
4848 : NVDupLane<op19_16, 0, (outs DPR:$Vd), (ins DPR:$Vm, IdxTy:$lane),
4849 IIC_VMOVD, OpcodeStr, Dt, "$Vd, $Vm$lane",
4850 [(set DPR:$Vd, (Ty (NEONvduplane (Ty DPR:$Vm), imm:$lane)))]>;
4852 class VDUPLNQ<bits<4> op19_16, string OpcodeStr, string Dt,
4853 ValueType ResTy, ValueType OpTy, Operand IdxTy>
4854 : NVDupLane<op19_16, 1, (outs QPR:$Vd), (ins DPR:$Vm, IdxTy:$lane),
4855 IIC_VMOVQ, OpcodeStr, Dt, "$Vd, $Vm$lane",
4856 [(set QPR:$Vd, (ResTy (NEONvduplane (OpTy DPR:$Vm),
4857 VectorIndex32:$lane)))]>;
4859 // Inst{19-16} is partially specified depending on the element size.
4861 def VDUPLN8d : VDUPLND<{?,?,?,1}, "vdup", "8", v8i8, VectorIndex8> {
4863 let Inst{19-17} = lane{2-0};
4865 def VDUPLN16d : VDUPLND<{?,?,1,0}, "vdup", "16", v4i16, VectorIndex16> {
4867 let Inst{19-18} = lane{1-0};
4869 def VDUPLN32d : VDUPLND<{?,1,0,0}, "vdup", "32", v2i32, VectorIndex32> {
4871 let Inst{19} = lane{0};
4873 def VDUPLN8q : VDUPLNQ<{?,?,?,1}, "vdup", "8", v16i8, v8i8, VectorIndex8> {
4875 let Inst{19-17} = lane{2-0};
4877 def VDUPLN16q : VDUPLNQ<{?,?,1,0}, "vdup", "16", v8i16, v4i16, VectorIndex16> {
4879 let Inst{19-18} = lane{1-0};
4881 def VDUPLN32q : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4i32, v2i32, VectorIndex32> {
4883 let Inst{19} = lane{0};
4886 def : Pat<(v2f32 (NEONvduplane (v2f32 DPR:$Vm), imm:$lane)),
4887 (VDUPLN32d DPR:$Vm, imm:$lane)>;
4889 def : Pat<(v4f32 (NEONvduplane (v2f32 DPR:$Vm), imm:$lane)),
4890 (VDUPLN32q DPR:$Vm, imm:$lane)>;
4892 def : Pat<(v16i8 (NEONvduplane (v16i8 QPR:$src), imm:$lane)),
4893 (v16i8 (VDUPLN8q (v8i8 (EXTRACT_SUBREG QPR:$src,
4894 (DSubReg_i8_reg imm:$lane))),
4895 (SubReg_i8_lane imm:$lane)))>;
4896 def : Pat<(v8i16 (NEONvduplane (v8i16 QPR:$src), imm:$lane)),
4897 (v8i16 (VDUPLN16q (v4i16 (EXTRACT_SUBREG QPR:$src,
4898 (DSubReg_i16_reg imm:$lane))),
4899 (SubReg_i16_lane imm:$lane)))>;
4900 def : Pat<(v4i32 (NEONvduplane (v4i32 QPR:$src), imm:$lane)),
4901 (v4i32 (VDUPLN32q (v2i32 (EXTRACT_SUBREG QPR:$src,
4902 (DSubReg_i32_reg imm:$lane))),
4903 (SubReg_i32_lane imm:$lane)))>;
4904 def : Pat<(v4f32 (NEONvduplane (v4f32 QPR:$src), imm:$lane)),
4905 (v4f32 (VDUPLN32q (v2f32 (EXTRACT_SUBREG QPR:$src,
4906 (DSubReg_i32_reg imm:$lane))),
4907 (SubReg_i32_lane imm:$lane)))>;
4909 def VDUPfdf : PseudoNeonI<(outs DPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
4910 [(set DPR:$dst, (v2f32 (NEONvdup (f32 SPR:$src))))]>;
4911 def VDUPfqf : PseudoNeonI<(outs QPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
4912 [(set QPR:$dst, (v4f32 (NEONvdup (f32 SPR:$src))))]>;
4914 // VMOVN : Vector Narrowing Move
4915 defm VMOVN : N2VN_HSD<0b11,0b11,0b10,0b00100,0,0, IIC_VMOVN,
4916 "vmovn", "i", trunc>;
4917 // VQMOVN : Vector Saturating Narrowing Move
4918 defm VQMOVNs : N2VNInt_HSD<0b11,0b11,0b10,0b00101,0,0, IIC_VQUNAiD,
4919 "vqmovn", "s", int_arm_neon_vqmovns>;
4920 defm VQMOVNu : N2VNInt_HSD<0b11,0b11,0b10,0b00101,1,0, IIC_VQUNAiD,
4921 "vqmovn", "u", int_arm_neon_vqmovnu>;
4922 defm VQMOVNsu : N2VNInt_HSD<0b11,0b11,0b10,0b00100,1,0, IIC_VQUNAiD,
4923 "vqmovun", "s", int_arm_neon_vqmovnsu>;
4924 // VMOVL : Vector Lengthening Move
4925 defm VMOVLs : N2VL_QHS<0b01,0b10100,0,1, "vmovl", "s", sext>;
4926 defm VMOVLu : N2VL_QHS<0b11,0b10100,0,1, "vmovl", "u", zext>;
4928 // Vector Conversions.
4930 // VCVT : Vector Convert Between Floating-Point and Integers
4931 def VCVTf2sd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
4932 v2i32, v2f32, fp_to_sint>;
4933 def VCVTf2ud : N2VD<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
4934 v2i32, v2f32, fp_to_uint>;
4935 def VCVTs2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
4936 v2f32, v2i32, sint_to_fp>;
4937 def VCVTu2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
4938 v2f32, v2i32, uint_to_fp>;
4940 def VCVTf2sq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
4941 v4i32, v4f32, fp_to_sint>;
4942 def VCVTf2uq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
4943 v4i32, v4f32, fp_to_uint>;
4944 def VCVTs2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
4945 v4f32, v4i32, sint_to_fp>;
4946 def VCVTu2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
4947 v4f32, v4i32, uint_to_fp>;
4949 // VCVT : Vector Convert Between Floating-Point and Fixed-Point.
4950 let DecoderMethod = "DecodeVCVTD" in {
4951 def VCVTf2xsd : N2VCvtD<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
4952 v2i32, v2f32, int_arm_neon_vcvtfp2fxs>;
4953 def VCVTf2xud : N2VCvtD<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
4954 v2i32, v2f32, int_arm_neon_vcvtfp2fxu>;
4955 def VCVTxs2fd : N2VCvtD<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
4956 v2f32, v2i32, int_arm_neon_vcvtfxs2fp>;
4957 def VCVTxu2fd : N2VCvtD<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
4958 v2f32, v2i32, int_arm_neon_vcvtfxu2fp>;
4961 let DecoderMethod = "DecodeVCVTQ" in {
4962 def VCVTf2xsq : N2VCvtQ<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
4963 v4i32, v4f32, int_arm_neon_vcvtfp2fxs>;
4964 def VCVTf2xuq : N2VCvtQ<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
4965 v4i32, v4f32, int_arm_neon_vcvtfp2fxu>;
4966 def VCVTxs2fq : N2VCvtQ<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
4967 v4f32, v4i32, int_arm_neon_vcvtfxs2fp>;
4968 def VCVTxu2fq : N2VCvtQ<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
4969 v4f32, v4i32, int_arm_neon_vcvtfxu2fp>;
4972 // VCVT : Vector Convert Between Half-Precision and Single-Precision.
4973 def VCVTf2h : N2VNInt<0b11, 0b11, 0b01, 0b10, 0b01100, 0, 0,
4974 IIC_VUNAQ, "vcvt", "f16.f32",
4975 v4i16, v4f32, int_arm_neon_vcvtfp2hf>,
4976 Requires<[HasNEON, HasFP16]>;
4977 def VCVTh2f : N2VLInt<0b11, 0b11, 0b01, 0b10, 0b01110, 0, 0,
4978 IIC_VUNAQ, "vcvt", "f32.f16",
4979 v4f32, v4i16, int_arm_neon_vcvthf2fp>,
4980 Requires<[HasNEON, HasFP16]>;
4984 // VREV64 : Vector Reverse elements within 64-bit doublewords
4986 class VREV64D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4987 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 0, 0, (outs DPR:$Vd),
4988 (ins DPR:$Vm), IIC_VMOVD,
4989 OpcodeStr, Dt, "$Vd, $Vm", "",
4990 [(set DPR:$Vd, (Ty (NEONvrev64 (Ty DPR:$Vm))))]>;
4991 class VREV64Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4992 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 1, 0, (outs QPR:$Vd),
4993 (ins QPR:$Vm), IIC_VMOVQ,
4994 OpcodeStr, Dt, "$Vd, $Vm", "",
4995 [(set QPR:$Vd, (Ty (NEONvrev64 (Ty QPR:$Vm))))]>;
4997 def VREV64d8 : VREV64D<0b00, "vrev64", "8", v8i8>;
4998 def VREV64d16 : VREV64D<0b01, "vrev64", "16", v4i16>;
4999 def VREV64d32 : VREV64D<0b10, "vrev64", "32", v2i32>;
5000 def : Pat<(v2f32 (NEONvrev64 (v2f32 DPR:$Vm))), (VREV64d32 DPR:$Vm)>;
5002 def VREV64q8 : VREV64Q<0b00, "vrev64", "8", v16i8>;
5003 def VREV64q16 : VREV64Q<0b01, "vrev64", "16", v8i16>;
5004 def VREV64q32 : VREV64Q<0b10, "vrev64", "32", v4i32>;
5005 def : Pat<(v4f32 (NEONvrev64 (v4f32 QPR:$Vm))), (VREV64q32 QPR:$Vm)>;
5007 // VREV32 : Vector Reverse elements within 32-bit words
5009 class VREV32D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
5010 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 0, 0, (outs DPR:$Vd),
5011 (ins DPR:$Vm), IIC_VMOVD,
5012 OpcodeStr, Dt, "$Vd, $Vm", "",
5013 [(set DPR:$Vd, (Ty (NEONvrev32 (Ty DPR:$Vm))))]>;
5014 class VREV32Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
5015 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 1, 0, (outs QPR:$Vd),
5016 (ins QPR:$Vm), IIC_VMOVQ,
5017 OpcodeStr, Dt, "$Vd, $Vm", "",
5018 [(set QPR:$Vd, (Ty (NEONvrev32 (Ty QPR:$Vm))))]>;
5020 def VREV32d8 : VREV32D<0b00, "vrev32", "8", v8i8>;
5021 def VREV32d16 : VREV32D<0b01, "vrev32", "16", v4i16>;
5023 def VREV32q8 : VREV32Q<0b00, "vrev32", "8", v16i8>;
5024 def VREV32q16 : VREV32Q<0b01, "vrev32", "16", v8i16>;
5026 // VREV16 : Vector Reverse elements within 16-bit halfwords
5028 class VREV16D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
5029 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 0, 0, (outs DPR:$Vd),
5030 (ins DPR:$Vm), IIC_VMOVD,
5031 OpcodeStr, Dt, "$Vd, $Vm", "",
5032 [(set DPR:$Vd, (Ty (NEONvrev16 (Ty DPR:$Vm))))]>;
5033 class VREV16Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
5034 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 1, 0, (outs QPR:$Vd),
5035 (ins QPR:$Vm), IIC_VMOVQ,
5036 OpcodeStr, Dt, "$Vd, $Vm", "",
5037 [(set QPR:$Vd, (Ty (NEONvrev16 (Ty QPR:$Vm))))]>;
5039 def VREV16d8 : VREV16D<0b00, "vrev16", "8", v8i8>;
5040 def VREV16q8 : VREV16Q<0b00, "vrev16", "8", v16i8>;
5042 // Other Vector Shuffles.
5044 // Aligned extractions: really just dropping registers
5046 class AlignedVEXTq<ValueType DestTy, ValueType SrcTy, SDNodeXForm LaneCVT>
5047 : Pat<(DestTy (vector_extract_subvec (SrcTy QPR:$src), (i32 imm:$start))),
5048 (EXTRACT_SUBREG (SrcTy QPR:$src), (LaneCVT imm:$start))>;
5050 def : AlignedVEXTq<v8i8, v16i8, DSubReg_i8_reg>;
5052 def : AlignedVEXTq<v4i16, v8i16, DSubReg_i16_reg>;
5054 def : AlignedVEXTq<v2i32, v4i32, DSubReg_i32_reg>;
5056 def : AlignedVEXTq<v1i64, v2i64, DSubReg_f64_reg>;
5058 def : AlignedVEXTq<v2f32, v4f32, DSubReg_i32_reg>;
5061 // VEXT : Vector Extract
5063 class VEXTd<string OpcodeStr, string Dt, ValueType Ty, Operand immTy>
5064 : N3V<0,1,0b11,{?,?,?,?},0,0, (outs DPR:$Vd),
5065 (ins DPR:$Vn, DPR:$Vm, immTy:$index), NVExtFrm,
5066 IIC_VEXTD, OpcodeStr, Dt, "$Vd, $Vn, $Vm, $index", "",
5067 [(set DPR:$Vd, (Ty (NEONvext (Ty DPR:$Vn),
5068 (Ty DPR:$Vm), imm:$index)))]> {
5070 let Inst{11-8} = index{3-0};
5073 class VEXTq<string OpcodeStr, string Dt, ValueType Ty, Operand immTy>
5074 : N3V<0,1,0b11,{?,?,?,?},1,0, (outs QPR:$Vd),
5075 (ins QPR:$Vn, QPR:$Vm, imm0_15:$index), NVExtFrm,
5076 IIC_VEXTQ, OpcodeStr, Dt, "$Vd, $Vn, $Vm, $index", "",
5077 [(set QPR:$Vd, (Ty (NEONvext (Ty QPR:$Vn),
5078 (Ty QPR:$Vm), imm:$index)))]> {
5080 let Inst{11-8} = index{3-0};
5083 def VEXTd8 : VEXTd<"vext", "8", v8i8, imm0_7> {
5084 let Inst{11-8} = index{3-0};
5086 def VEXTd16 : VEXTd<"vext", "16", v4i16, imm0_3> {
5087 let Inst{11-9} = index{2-0};
5090 def VEXTd32 : VEXTd<"vext", "32", v2i32, imm0_1> {
5091 let Inst{11-10} = index{1-0};
5092 let Inst{9-8} = 0b00;
5094 def : Pat<(v2f32 (NEONvext (v2f32 DPR:$Vn),
5097 (VEXTd32 DPR:$Vn, DPR:$Vm, imm:$index)>;
5099 def VEXTq8 : VEXTq<"vext", "8", v16i8, imm0_15> {
5100 let Inst{11-8} = index{3-0};
5102 def VEXTq16 : VEXTq<"vext", "16", v8i16, imm0_7> {
5103 let Inst{11-9} = index{2-0};
5106 def VEXTq32 : VEXTq<"vext", "32", v4i32, imm0_3> {
5107 let Inst{11-10} = index{1-0};
5108 let Inst{9-8} = 0b00;
5110 def VEXTq64 : VEXTq<"vext", "64", v2i64, imm0_1> {
5111 let Inst{11} = index{0};
5112 let Inst{10-8} = 0b000;
5114 def : Pat<(v4f32 (NEONvext (v4f32 QPR:$Vn),
5117 (VEXTq32 QPR:$Vn, QPR:$Vm, imm:$index)>;
5119 // VTRN : Vector Transpose
5121 def VTRNd8 : N2VDShuffle<0b00, 0b00001, "vtrn", "8">;
5122 def VTRNd16 : N2VDShuffle<0b01, 0b00001, "vtrn", "16">;
5123 def VTRNd32 : N2VDShuffle<0b10, 0b00001, "vtrn", "32">;
5125 def VTRNq8 : N2VQShuffle<0b00, 0b00001, IIC_VPERMQ, "vtrn", "8">;
5126 def VTRNq16 : N2VQShuffle<0b01, 0b00001, IIC_VPERMQ, "vtrn", "16">;
5127 def VTRNq32 : N2VQShuffle<0b10, 0b00001, IIC_VPERMQ, "vtrn", "32">;
5129 // VUZP : Vector Unzip (Deinterleave)
5131 def VUZPd8 : N2VDShuffle<0b00, 0b00010, "vuzp", "8">;
5132 def VUZPd16 : N2VDShuffle<0b01, 0b00010, "vuzp", "16">;
5133 def VUZPd32 : N2VDShuffle<0b10, 0b00010, "vuzp", "32">;
5135 def VUZPq8 : N2VQShuffle<0b00, 0b00010, IIC_VPERMQ3, "vuzp", "8">;
5136 def VUZPq16 : N2VQShuffle<0b01, 0b00010, IIC_VPERMQ3, "vuzp", "16">;
5137 def VUZPq32 : N2VQShuffle<0b10, 0b00010, IIC_VPERMQ3, "vuzp", "32">;
5139 // VZIP : Vector Zip (Interleave)
5141 def VZIPd8 : N2VDShuffle<0b00, 0b00011, "vzip", "8">;
5142 def VZIPd16 : N2VDShuffle<0b01, 0b00011, "vzip", "16">;
5143 def VZIPd32 : N2VDShuffle<0b10, 0b00011, "vzip", "32">;
5145 def VZIPq8 : N2VQShuffle<0b00, 0b00011, IIC_VPERMQ3, "vzip", "8">;
5146 def VZIPq16 : N2VQShuffle<0b01, 0b00011, IIC_VPERMQ3, "vzip", "16">;
5147 def VZIPq32 : N2VQShuffle<0b10, 0b00011, IIC_VPERMQ3, "vzip", "32">;
5149 // Vector Table Lookup and Table Extension.
5151 // VTBL : Vector Table Lookup
5152 let DecoderMethod = "DecodeTBLInstruction" in {
5154 : N3V<1,1,0b11,0b1000,0,0, (outs DPR:$Vd),
5155 (ins VecListOneD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTB1,
5156 "vtbl", "8", "$Vd, $Vn, $Vm", "",
5157 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbl1 VecListOneD:$Vn, DPR:$Vm)))]>;
5158 let hasExtraSrcRegAllocReq = 1 in {
5160 : N3V<1,1,0b11,0b1001,0,0, (outs DPR:$Vd),
5161 (ins DPR:$Vn, DPR:$tbl2, DPR:$Vm), NVTBLFrm, IIC_VTB2,
5162 "vtbl", "8", "$Vd, \\{$Vn, $tbl2\\}, $Vm", "", []>;
5164 : N3V<1,1,0b11,0b1010,0,0, (outs DPR:$Vd),
5165 (ins DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$Vm), NVTBLFrm, IIC_VTB3,
5166 "vtbl", "8", "$Vd, \\{$Vn, $tbl2, $tbl3\\}, $Vm", "", []>;
5168 : N3V<1,1,0b11,0b1011,0,0, (outs DPR:$Vd),
5169 (ins DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$Vm),
5171 "vtbl", "8", "$Vd, \\{$Vn, $tbl2, $tbl3, $tbl4\\}, $Vm", "", []>;
5172 } // hasExtraSrcRegAllocReq = 1
5175 : PseudoNeonI<(outs DPR:$dst), (ins QPR:$tbl, DPR:$src), IIC_VTB2, "", []>;
5177 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB3, "", []>;
5179 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB4, "", []>;
5181 // VTBX : Vector Table Extension
5183 : N3V<1,1,0b11,0b1000,1,0, (outs DPR:$Vd),
5184 (ins DPR:$orig, VecListOneD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTBX1,
5185 "vtbx", "8", "$Vd, $Vn, $Vm", "$orig = $Vd",
5186 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbx1
5187 DPR:$orig, VecListOneD:$Vn, DPR:$Vm)))]>;
5188 let hasExtraSrcRegAllocReq = 1 in {
5190 : N3V<1,1,0b11,0b1001,1,0, (outs DPR:$Vd),
5191 (ins DPR:$orig, DPR:$Vn, DPR:$tbl2, DPR:$Vm), NVTBLFrm, IIC_VTBX2,
5192 "vtbx", "8", "$Vd, \\{$Vn, $tbl2\\}, $Vm", "$orig = $Vd", []>;
5194 : N3V<1,1,0b11,0b1010,1,0, (outs DPR:$Vd),
5195 (ins DPR:$orig, DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$Vm),
5196 NVTBLFrm, IIC_VTBX3,
5197 "vtbx", "8", "$Vd, \\{$Vn, $tbl2, $tbl3\\}, $Vm",
5200 : N3V<1,1,0b11,0b1011,1,0, (outs DPR:$Vd), (ins DPR:$orig, DPR:$Vn,
5201 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$Vm), NVTBLFrm, IIC_VTBX4,
5202 "vtbx", "8", "$Vd, \\{$Vn, $tbl2, $tbl3, $tbl4\\}, $Vm",
5204 } // hasExtraSrcRegAllocReq = 1
5207 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QPR:$tbl, DPR:$src),
5208 IIC_VTBX2, "$orig = $dst", []>;
5210 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
5211 IIC_VTBX3, "$orig = $dst", []>;
5213 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
5214 IIC_VTBX4, "$orig = $dst", []>;
5215 } // DecoderMethod = "DecodeTBLInstruction"
5217 //===----------------------------------------------------------------------===//
5218 // NEON instructions for single-precision FP math
5219 //===----------------------------------------------------------------------===//
5221 class N2VSPat<SDNode OpNode, NeonI Inst>
5222 : NEONFPPat<(f32 (OpNode SPR:$a)),
5224 (v2f32 (COPY_TO_REGCLASS (Inst
5226 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5227 SPR:$a, ssub_0)), DPR_VFP2)), ssub_0)>;
5229 class N3VSPat<SDNode OpNode, NeonI Inst>
5230 : NEONFPPat<(f32 (OpNode SPR:$a, SPR:$b)),
5232 (v2f32 (COPY_TO_REGCLASS (Inst
5234 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5237 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5238 SPR:$b, ssub_0)), DPR_VFP2)), ssub_0)>;
5240 class N3VSMulOpPat<SDNode MulNode, SDNode OpNode, NeonI Inst>
5241 : NEONFPPat<(f32 (OpNode SPR:$acc, (f32 (MulNode SPR:$a, SPR:$b)))),
5243 (v2f32 (COPY_TO_REGCLASS (Inst
5245 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5248 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5251 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5252 SPR:$b, ssub_0)), DPR_VFP2)), ssub_0)>;
5254 def : N3VSPat<fadd, VADDfd>;
5255 def : N3VSPat<fsub, VSUBfd>;
5256 def : N3VSPat<fmul, VMULfd>;
5257 def : N3VSMulOpPat<fmul, fadd, VMLAfd>,
5258 Requires<[HasNEON, UseNEONForFP, UseFPVMLx]>;
5259 def : N3VSMulOpPat<fmul, fsub, VMLSfd>,
5260 Requires<[HasNEON, UseNEONForFP, UseFPVMLx]>;
5261 def : N2VSPat<fabs, VABSfd>;
5262 def : N2VSPat<fneg, VNEGfd>;
5263 def : N3VSPat<NEONfmax, VMAXfd>;
5264 def : N3VSPat<NEONfmin, VMINfd>;
5265 def : N2VSPat<arm_ftosi, VCVTf2sd>;
5266 def : N2VSPat<arm_ftoui, VCVTf2ud>;
5267 def : N2VSPat<arm_sitof, VCVTs2fd>;
5268 def : N2VSPat<arm_uitof, VCVTu2fd>;
5270 //===----------------------------------------------------------------------===//
5271 // Non-Instruction Patterns
5272 //===----------------------------------------------------------------------===//
5275 def : Pat<(v1i64 (bitconvert (v2i32 DPR:$src))), (v1i64 DPR:$src)>;
5276 def : Pat<(v1i64 (bitconvert (v4i16 DPR:$src))), (v1i64 DPR:$src)>;
5277 def : Pat<(v1i64 (bitconvert (v8i8 DPR:$src))), (v1i64 DPR:$src)>;
5278 def : Pat<(v1i64 (bitconvert (f64 DPR:$src))), (v1i64 DPR:$src)>;
5279 def : Pat<(v1i64 (bitconvert (v2f32 DPR:$src))), (v1i64 DPR:$src)>;
5280 def : Pat<(v2i32 (bitconvert (v1i64 DPR:$src))), (v2i32 DPR:$src)>;
5281 def : Pat<(v2i32 (bitconvert (v4i16 DPR:$src))), (v2i32 DPR:$src)>;
5282 def : Pat<(v2i32 (bitconvert (v8i8 DPR:$src))), (v2i32 DPR:$src)>;
5283 def : Pat<(v2i32 (bitconvert (f64 DPR:$src))), (v2i32 DPR:$src)>;
5284 def : Pat<(v2i32 (bitconvert (v2f32 DPR:$src))), (v2i32 DPR:$src)>;
5285 def : Pat<(v4i16 (bitconvert (v1i64 DPR:$src))), (v4i16 DPR:$src)>;
5286 def : Pat<(v4i16 (bitconvert (v2i32 DPR:$src))), (v4i16 DPR:$src)>;
5287 def : Pat<(v4i16 (bitconvert (v8i8 DPR:$src))), (v4i16 DPR:$src)>;
5288 def : Pat<(v4i16 (bitconvert (f64 DPR:$src))), (v4i16 DPR:$src)>;
5289 def : Pat<(v4i16 (bitconvert (v2f32 DPR:$src))), (v4i16 DPR:$src)>;
5290 def : Pat<(v8i8 (bitconvert (v1i64 DPR:$src))), (v8i8 DPR:$src)>;
5291 def : Pat<(v8i8 (bitconvert (v2i32 DPR:$src))), (v8i8 DPR:$src)>;
5292 def : Pat<(v8i8 (bitconvert (v4i16 DPR:$src))), (v8i8 DPR:$src)>;
5293 def : Pat<(v8i8 (bitconvert (f64 DPR:$src))), (v8i8 DPR:$src)>;
5294 def : Pat<(v8i8 (bitconvert (v2f32 DPR:$src))), (v8i8 DPR:$src)>;
5295 def : Pat<(f64 (bitconvert (v1i64 DPR:$src))), (f64 DPR:$src)>;
5296 def : Pat<(f64 (bitconvert (v2i32 DPR:$src))), (f64 DPR:$src)>;
5297 def : Pat<(f64 (bitconvert (v4i16 DPR:$src))), (f64 DPR:$src)>;
5298 def : Pat<(f64 (bitconvert (v8i8 DPR:$src))), (f64 DPR:$src)>;
5299 def : Pat<(f64 (bitconvert (v2f32 DPR:$src))), (f64 DPR:$src)>;
5300 def : Pat<(v2f32 (bitconvert (f64 DPR:$src))), (v2f32 DPR:$src)>;
5301 def : Pat<(v2f32 (bitconvert (v1i64 DPR:$src))), (v2f32 DPR:$src)>;
5302 def : Pat<(v2f32 (bitconvert (v2i32 DPR:$src))), (v2f32 DPR:$src)>;
5303 def : Pat<(v2f32 (bitconvert (v4i16 DPR:$src))), (v2f32 DPR:$src)>;
5304 def : Pat<(v2f32 (bitconvert (v8i8 DPR:$src))), (v2f32 DPR:$src)>;
5306 def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (v2i64 QPR:$src)>;
5307 def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (v2i64 QPR:$src)>;
5308 def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (v2i64 QPR:$src)>;
5309 def : Pat<(v2i64 (bitconvert (v2f64 QPR:$src))), (v2i64 QPR:$src)>;
5310 def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (v2i64 QPR:$src)>;
5311 def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (v4i32 QPR:$src)>;
5312 def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (v4i32 QPR:$src)>;
5313 def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (v4i32 QPR:$src)>;
5314 def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (v4i32 QPR:$src)>;
5315 def : Pat<(v4i32 (bitconvert (v4f32 QPR:$src))), (v4i32 QPR:$src)>;
5316 def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (v8i16 QPR:$src)>;
5317 def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (v8i16 QPR:$src)>;
5318 def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (v8i16 QPR:$src)>;
5319 def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (v8i16 QPR:$src)>;
5320 def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (v8i16 QPR:$src)>;
5321 def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (v16i8 QPR:$src)>;
5322 def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (v16i8 QPR:$src)>;
5323 def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (v16i8 QPR:$src)>;
5324 def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (v16i8 QPR:$src)>;
5325 def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (v16i8 QPR:$src)>;
5326 def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (v4f32 QPR:$src)>;
5327 def : Pat<(v4f32 (bitconvert (v4i32 QPR:$src))), (v4f32 QPR:$src)>;
5328 def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (v4f32 QPR:$src)>;
5329 def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (v4f32 QPR:$src)>;
5330 def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (v4f32 QPR:$src)>;
5331 def : Pat<(v2f64 (bitconvert (v2i64 QPR:$src))), (v2f64 QPR:$src)>;
5332 def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (v2f64 QPR:$src)>;
5333 def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 QPR:$src)>;
5334 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 QPR:$src)>;
5335 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 QPR:$src)>;
5338 //===----------------------------------------------------------------------===//
5339 // Assembler aliases
5342 def : VFP2InstAlias<"fmdhr${p} $Dd, $Rn",
5343 (VSETLNi32 DPR:$Dd, GPR:$Rn, 1, pred:$p)>;
5344 def : VFP2InstAlias<"fmdlr${p} $Dd, $Rn",
5345 (VSETLNi32 DPR:$Dd, GPR:$Rn, 0, pred:$p)>;
5348 // VADD two-operand aliases.
5349 def : NEONInstAlias<"vadd${p}.i8 $Vdn, $Vm",
5350 (VADDv16i8 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5351 def : NEONInstAlias<"vadd${p}.i16 $Vdn, $Vm",
5352 (VADDv8i16 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5353 def : NEONInstAlias<"vadd${p}.i32 $Vdn, $Vm",
5354 (VADDv4i32 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5355 def : NEONInstAlias<"vadd${p}.i64 $Vdn, $Vm",
5356 (VADDv2i64 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5358 def : NEONInstAlias<"vadd${p}.i8 $Vdn, $Vm",
5359 (VADDv8i8 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5360 def : NEONInstAlias<"vadd${p}.i16 $Vdn, $Vm",
5361 (VADDv4i16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5362 def : NEONInstAlias<"vadd${p}.i32 $Vdn, $Vm",
5363 (VADDv2i32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5364 def : NEONInstAlias<"vadd${p}.i64 $Vdn, $Vm",
5365 (VADDv1i64 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5367 def : NEONInstAlias<"vadd${p}.f32 $Vdn, $Vm",
5368 (VADDfd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5369 def : NEONInstAlias<"vadd${p}.f32 $Vdn, $Vm",
5370 (VADDfq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5372 // VSUB two-operand aliases.
5373 def : NEONInstAlias<"vsub${p}.i8 $Vdn, $Vm",
5374 (VSUBv16i8 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5375 def : NEONInstAlias<"vsub${p}.i16 $Vdn, $Vm",
5376 (VSUBv8i16 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5377 def : NEONInstAlias<"vsub${p}.i32 $Vdn, $Vm",
5378 (VSUBv4i32 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5379 def : NEONInstAlias<"vsub${p}.i64 $Vdn, $Vm",
5380 (VSUBv2i64 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5382 def : NEONInstAlias<"vsub${p}.i8 $Vdn, $Vm",
5383 (VSUBv8i8 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5384 def : NEONInstAlias<"vsub${p}.i16 $Vdn, $Vm",
5385 (VSUBv4i16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5386 def : NEONInstAlias<"vsub${p}.i32 $Vdn, $Vm",
5387 (VSUBv2i32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5388 def : NEONInstAlias<"vsub${p}.i64 $Vdn, $Vm",
5389 (VSUBv1i64 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5391 def : NEONInstAlias<"vsub${p}.f32 $Vdn, $Vm",
5392 (VSUBfd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5393 def : NEONInstAlias<"vsub${p}.f32 $Vdn, $Vm",
5394 (VSUBfq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5396 // VADDW two-operand aliases.
5397 def : NEONInstAlias<"vaddw${p}.s8 $Vdn, $Vm",
5398 (VADDWsv8i16 QPR:$Vdn, QPR:$Vdn, DPR:$Vm, pred:$p)>;
5399 def : NEONInstAlias<"vaddw${p}.s16 $Vdn, $Vm",
5400 (VADDWsv4i32 QPR:$Vdn, QPR:$Vdn, DPR:$Vm, pred:$p)>;
5401 def : NEONInstAlias<"vaddw${p}.s32 $Vdn, $Vm",
5402 (VADDWsv2i64 QPR:$Vdn, QPR:$Vdn, DPR:$Vm, pred:$p)>;
5403 def : NEONInstAlias<"vaddw${p}.u8 $Vdn, $Vm",
5404 (VADDWuv8i16 QPR:$Vdn, QPR:$Vdn, DPR:$Vm, pred:$p)>;
5405 def : NEONInstAlias<"vaddw${p}.u16 $Vdn, $Vm",
5406 (VADDWuv4i32 QPR:$Vdn, QPR:$Vdn, DPR:$Vm, pred:$p)>;
5407 def : NEONInstAlias<"vaddw${p}.u32 $Vdn, $Vm",
5408 (VADDWuv2i64 QPR:$Vdn, QPR:$Vdn, DPR:$Vm, pred:$p)>;
5410 // VAND/VBIC/VEOR/VORR accept but do not require a type suffix.
5411 defm : VFPDTAnyInstAlias<"vand${p}", "$Vd, $Vn, $Vm",
5412 (VANDd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
5413 defm : VFPDTAnyInstAlias<"vand${p}", "$Vd, $Vn, $Vm",
5414 (VANDq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
5415 defm : VFPDTAnyInstAlias<"vbic${p}", "$Vd, $Vn, $Vm",
5416 (VBICd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
5417 defm : VFPDTAnyInstAlias<"vbic${p}", "$Vd, $Vn, $Vm",
5418 (VBICq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
5419 defm : VFPDTAnyInstAlias<"veor${p}", "$Vd, $Vn, $Vm",
5420 (VEORd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
5421 defm : VFPDTAnyInstAlias<"veor${p}", "$Vd, $Vn, $Vm",
5422 (VEORq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
5423 defm : VFPDTAnyInstAlias<"vorr${p}", "$Vd, $Vn, $Vm",
5424 (VORRd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
5425 defm : VFPDTAnyInstAlias<"vorr${p}", "$Vd, $Vn, $Vm",
5426 (VORRq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
5427 // ... two-operand aliases
5428 def : NEONInstAlias<"vand${p} $Vdn, $Vm",
5429 (VANDd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5430 def : NEONInstAlias<"vand${p} $Vdn, $Vm",
5431 (VANDq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5432 def : NEONInstAlias<"vbic${p} $Vdn, $Vm",
5433 (VBICd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5434 def : NEONInstAlias<"vbic${p} $Vdn, $Vm",
5435 (VBICq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5436 def : NEONInstAlias<"veor${p} $Vdn, $Vm",
5437 (VEORd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5438 def : NEONInstAlias<"veor${p} $Vdn, $Vm",
5439 (VEORq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5440 def : NEONInstAlias<"vorr${p} $Vdn, $Vm",
5441 (VORRd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5442 def : NEONInstAlias<"vorr${p} $Vdn, $Vm",
5443 (VORRq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5445 defm : VFPDTAnyInstAlias<"vand${p}", "$Vdn, $Vm",
5446 (VANDd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5447 defm : VFPDTAnyInstAlias<"vand${p}", "$Vdn, $Vm",
5448 (VANDq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5449 defm : VFPDTAnyInstAlias<"veor${p}", "$Vdn, $Vm",
5450 (VEORd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5451 defm : VFPDTAnyInstAlias<"veor${p}", "$Vdn, $Vm",
5452 (VEORq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5453 defm : VFPDTAnyInstAlias<"vorr${p}", "$Vdn, $Vm",
5454 (VORRd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5455 defm : VFPDTAnyInstAlias<"vorr${p}", "$Vdn, $Vm",
5456 (VORRq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5458 // VMUL two-operand aliases.
5459 def : NEONInstAlias<"vmul${p}.p8 $Qdn, $Qm",
5460 (VMULpq QPR:$Qdn, QPR:$Qdn, QPR:$Qm, pred:$p)>;
5461 def : NEONInstAlias<"vmul${p}.i8 $Qdn, $Qm",
5462 (VMULv16i8 QPR:$Qdn, QPR:$Qdn, QPR:$Qm, pred:$p)>;
5463 def : NEONInstAlias<"vmul${p}.i16 $Qdn, $Qm",
5464 (VMULv8i16 QPR:$Qdn, QPR:$Qdn, QPR:$Qm, pred:$p)>;
5465 def : NEONInstAlias<"vmul${p}.i32 $Qdn, $Qm",
5466 (VMULv4i32 QPR:$Qdn, QPR:$Qdn, QPR:$Qm, pred:$p)>;
5468 def : NEONInstAlias<"vmul${p}.p8 $Ddn, $Dm",
5469 (VMULpd DPR:$Ddn, DPR:$Ddn, DPR:$Dm, pred:$p)>;
5470 def : NEONInstAlias<"vmul${p}.i8 $Ddn, $Dm",
5471 (VMULv8i8 DPR:$Ddn, DPR:$Ddn, DPR:$Dm, pred:$p)>;
5472 def : NEONInstAlias<"vmul${p}.i16 $Ddn, $Dm",
5473 (VMULv4i16 DPR:$Ddn, DPR:$Ddn, DPR:$Dm, pred:$p)>;
5474 def : NEONInstAlias<"vmul${p}.i32 $Ddn, $Dm",
5475 (VMULv2i32 DPR:$Ddn, DPR:$Ddn, DPR:$Dm, pred:$p)>;
5477 def : NEONInstAlias<"vmul${p}.f32 $Qdn, $Qm",
5478 (VMULfq QPR:$Qdn, QPR:$Qdn, QPR:$Qm, pred:$p)>;
5479 def : NEONInstAlias<"vmul${p}.f32 $Ddn, $Dm",
5480 (VMULfd DPR:$Ddn, DPR:$Ddn, DPR:$Dm, pred:$p)>;
5482 def : NEONInstAlias<"vmul${p}.i16 $Ddn, $Dm$lane",
5483 (VMULslv4i16 DPR:$Ddn, DPR:$Ddn, DPR_8:$Dm,
5484 VectorIndex16:$lane, pred:$p)>;
5485 def : NEONInstAlias<"vmul${p}.i16 $Qdn, $Dm$lane",
5486 (VMULslv8i16 QPR:$Qdn, QPR:$Qdn, DPR_8:$Dm,
5487 VectorIndex16:$lane, pred:$p)>;
5489 def : NEONInstAlias<"vmul${p}.i32 $Ddn, $Dm$lane",
5490 (VMULslv2i32 DPR:$Ddn, DPR:$Ddn, DPR_VFP2:$Dm,
5491 VectorIndex32:$lane, pred:$p)>;
5492 def : NEONInstAlias<"vmul${p}.i32 $Qdn, $Dm$lane",
5493 (VMULslv4i32 QPR:$Qdn, QPR:$Qdn, DPR_VFP2:$Dm,
5494 VectorIndex32:$lane, pred:$p)>;
5496 def : NEONInstAlias<"vmul${p}.f32 $Ddn, $Dm$lane",
5497 (VMULslfd DPR:$Ddn, DPR:$Ddn, DPR_VFP2:$Dm,
5498 VectorIndex32:$lane, pred:$p)>;
5499 def : NEONInstAlias<"vmul${p}.f32 $Qdn, $Dm$lane",
5500 (VMULslfq QPR:$Qdn, QPR:$Qdn, DPR_VFP2:$Dm,
5501 VectorIndex32:$lane, pred:$p)>;
5503 // VQADD (register) two-operand aliases.
5504 def : NEONInstAlias<"vqadd${p}.s8 $Vdn, $Vm",
5505 (VQADDsv8i8 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5506 def : NEONInstAlias<"vqadd${p}.s16 $Vdn, $Vm",
5507 (VQADDsv4i16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5508 def : NEONInstAlias<"vqadd${p}.s32 $Vdn, $Vm",
5509 (VQADDsv2i32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5510 def : NEONInstAlias<"vqadd${p}.s64 $Vdn, $Vm",
5511 (VQADDsv1i64 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5512 def : NEONInstAlias<"vqadd${p}.u8 $Vdn, $Vm",
5513 (VQADDuv8i8 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5514 def : NEONInstAlias<"vqadd${p}.u16 $Vdn, $Vm",
5515 (VQADDuv4i16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5516 def : NEONInstAlias<"vqadd${p}.u32 $Vdn, $Vm",
5517 (VQADDuv2i32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5518 def : NEONInstAlias<"vqadd${p}.u64 $Vdn, $Vm",
5519 (VQADDuv1i64 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5521 def : NEONInstAlias<"vqadd${p}.s8 $Vdn, $Vm",
5522 (VQADDsv16i8 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5523 def : NEONInstAlias<"vqadd${p}.s16 $Vdn, $Vm",
5524 (VQADDsv8i16 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5525 def : NEONInstAlias<"vqadd${p}.s32 $Vdn, $Vm",
5526 (VQADDsv4i32 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5527 def : NEONInstAlias<"vqadd${p}.s64 $Vdn, $Vm",
5528 (VQADDsv2i64 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5529 def : NEONInstAlias<"vqadd${p}.u8 $Vdn, $Vm",
5530 (VQADDuv16i8 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5531 def : NEONInstAlias<"vqadd${p}.u16 $Vdn, $Vm",
5532 (VQADDuv8i16 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5533 def : NEONInstAlias<"vqadd${p}.u32 $Vdn, $Vm",
5534 (VQADDuv4i32 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5535 def : NEONInstAlias<"vqadd${p}.u64 $Vdn, $Vm",
5536 (VQADDuv2i64 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5538 // VSHL (immediate) two-operand aliases.
5539 def : NEONInstAlias<"vshl${p}.i8 $Vdn, $imm",
5540 (VSHLiv8i8 DPR:$Vdn, DPR:$Vdn, imm0_7:$imm, pred:$p)>;
5541 def : NEONInstAlias<"vshl${p}.i16 $Vdn, $imm",
5542 (VSHLiv4i16 DPR:$Vdn, DPR:$Vdn, imm0_15:$imm, pred:$p)>;
5543 def : NEONInstAlias<"vshl${p}.i32 $Vdn, $imm",
5544 (VSHLiv2i32 DPR:$Vdn, DPR:$Vdn, imm0_31:$imm, pred:$p)>;
5545 def : NEONInstAlias<"vshl${p}.i64 $Vdn, $imm",
5546 (VSHLiv1i64 DPR:$Vdn, DPR:$Vdn, imm0_63:$imm, pred:$p)>;
5548 def : NEONInstAlias<"vshl${p}.i8 $Vdn, $imm",
5549 (VSHLiv16i8 QPR:$Vdn, QPR:$Vdn, imm0_7:$imm, pred:$p)>;
5550 def : NEONInstAlias<"vshl${p}.i16 $Vdn, $imm",
5551 (VSHLiv8i16 QPR:$Vdn, QPR:$Vdn, imm0_15:$imm, pred:$p)>;
5552 def : NEONInstAlias<"vshl${p}.i32 $Vdn, $imm",
5553 (VSHLiv4i32 QPR:$Vdn, QPR:$Vdn, imm0_31:$imm, pred:$p)>;
5554 def : NEONInstAlias<"vshl${p}.i64 $Vdn, $imm",
5555 (VSHLiv2i64 QPR:$Vdn, QPR:$Vdn, imm0_63:$imm, pred:$p)>;
5557 // VSHL (register) two-operand aliases.
5558 def : NEONInstAlias<"vshl${p}.s8 $Vdn, $Vm",
5559 (VSHLsv8i8 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5560 def : NEONInstAlias<"vshl${p}.s16 $Vdn, $Vm",
5561 (VSHLsv4i16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5562 def : NEONInstAlias<"vshl${p}.s32 $Vdn, $Vm",
5563 (VSHLsv2i32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5564 def : NEONInstAlias<"vshl${p}.s64 $Vdn, $Vm",
5565 (VSHLsv1i64 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5566 def : NEONInstAlias<"vshl${p}.u8 $Vdn, $Vm",
5567 (VSHLuv8i8 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5568 def : NEONInstAlias<"vshl${p}.u16 $Vdn, $Vm",
5569 (VSHLuv4i16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5570 def : NEONInstAlias<"vshl${p}.u32 $Vdn, $Vm",
5571 (VSHLuv2i32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5572 def : NEONInstAlias<"vshl${p}.u64 $Vdn, $Vm",
5573 (VSHLuv1i64 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5575 def : NEONInstAlias<"vshl${p}.s8 $Vdn, $Vm",
5576 (VSHLsv16i8 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5577 def : NEONInstAlias<"vshl${p}.s16 $Vdn, $Vm",
5578 (VSHLsv8i16 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5579 def : NEONInstAlias<"vshl${p}.s32 $Vdn, $Vm",
5580 (VSHLsv4i32 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5581 def : NEONInstAlias<"vshl${p}.s64 $Vdn, $Vm",
5582 (VSHLsv2i64 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5583 def : NEONInstAlias<"vshl${p}.u8 $Vdn, $Vm",
5584 (VSHLuv16i8 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5585 def : NEONInstAlias<"vshl${p}.u16 $Vdn, $Vm",
5586 (VSHLuv8i16 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5587 def : NEONInstAlias<"vshl${p}.u32 $Vdn, $Vm",
5588 (VSHLuv4i32 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5589 def : NEONInstAlias<"vshl${p}.u64 $Vdn, $Vm",
5590 (VSHLuv2i64 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5592 // VSHL (immediate) two-operand aliases.
5593 def : NEONInstAlias<"vshr${p}.s8 $Vdn, $imm",
5594 (VSHRsv8i8 DPR:$Vdn, DPR:$Vdn, shr_imm8:$imm, pred:$p)>;
5595 def : NEONInstAlias<"vshr${p}.s16 $Vdn, $imm",
5596 (VSHRsv4i16 DPR:$Vdn, DPR:$Vdn, shr_imm16:$imm, pred:$p)>;
5597 def : NEONInstAlias<"vshr${p}.s32 $Vdn, $imm",
5598 (VSHRsv2i32 DPR:$Vdn, DPR:$Vdn, shr_imm32:$imm, pred:$p)>;
5599 def : NEONInstAlias<"vshr${p}.s64 $Vdn, $imm",
5600 (VSHRsv1i64 DPR:$Vdn, DPR:$Vdn, shr_imm64:$imm, pred:$p)>;
5602 def : NEONInstAlias<"vshr${p}.s8 $Vdn, $imm",
5603 (VSHRsv16i8 QPR:$Vdn, QPR:$Vdn, shr_imm8:$imm, pred:$p)>;
5604 def : NEONInstAlias<"vshr${p}.s16 $Vdn, $imm",
5605 (VSHRsv8i16 QPR:$Vdn, QPR:$Vdn, shr_imm16:$imm, pred:$p)>;
5606 def : NEONInstAlias<"vshr${p}.s32 $Vdn, $imm",
5607 (VSHRsv4i32 QPR:$Vdn, QPR:$Vdn, shr_imm32:$imm, pred:$p)>;
5608 def : NEONInstAlias<"vshr${p}.s64 $Vdn, $imm",
5609 (VSHRsv2i64 QPR:$Vdn, QPR:$Vdn, shr_imm64:$imm, pred:$p)>;
5611 def : NEONInstAlias<"vshr${p}.u8 $Vdn, $imm",
5612 (VSHRuv8i8 DPR:$Vdn, DPR:$Vdn, shr_imm8:$imm, pred:$p)>;
5613 def : NEONInstAlias<"vshr${p}.u16 $Vdn, $imm",
5614 (VSHRuv4i16 DPR:$Vdn, DPR:$Vdn, shr_imm16:$imm, pred:$p)>;
5615 def : NEONInstAlias<"vshr${p}.u32 $Vdn, $imm",
5616 (VSHRuv2i32 DPR:$Vdn, DPR:$Vdn, shr_imm32:$imm, pred:$p)>;
5617 def : NEONInstAlias<"vshr${p}.u64 $Vdn, $imm",
5618 (VSHRuv1i64 DPR:$Vdn, DPR:$Vdn, shr_imm64:$imm, pred:$p)>;
5620 def : NEONInstAlias<"vshr${p}.u8 $Vdn, $imm",
5621 (VSHRuv16i8 QPR:$Vdn, QPR:$Vdn, shr_imm8:$imm, pred:$p)>;
5622 def : NEONInstAlias<"vshr${p}.u16 $Vdn, $imm",
5623 (VSHRuv8i16 QPR:$Vdn, QPR:$Vdn, shr_imm16:$imm, pred:$p)>;
5624 def : NEONInstAlias<"vshr${p}.u32 $Vdn, $imm",
5625 (VSHRuv4i32 QPR:$Vdn, QPR:$Vdn, shr_imm32:$imm, pred:$p)>;
5626 def : NEONInstAlias<"vshr${p}.u64 $Vdn, $imm",
5627 (VSHRuv2i64 QPR:$Vdn, QPR:$Vdn, shr_imm64:$imm, pred:$p)>;
5629 // VLD1 single-lane pseudo-instructions. These need special handling for
5630 // the lane index that an InstAlias can't handle, so we use these instead.
5631 defm VLD1LNdAsm : NEONDT8AsmPseudoInst<"vld1${p}", "$list, $addr",
5632 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5633 defm VLD1LNdAsm : NEONDT16AsmPseudoInst<"vld1${p}", "$list, $addr",
5634 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5635 defm VLD1LNdAsm : NEONDT32AsmPseudoInst<"vld1${p}", "$list, $addr",
5636 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5638 defm VLD1LNdWB_fixed_Asm : NEONDT8AsmPseudoInst<"vld1${p}", "$list, $addr!",
5639 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5640 defm VLD1LNdWB_fixed_Asm : NEONDT16AsmPseudoInst<"vld1${p}", "$list, $addr!",
5641 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5642 defm VLD1LNdWB_fixed_Asm : NEONDT32AsmPseudoInst<"vld1${p}", "$list, $addr!",
5643 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5644 defm VLD1LNdWB_register_Asm :
5645 NEONDT8AsmPseudoInst<"vld1${p}", "$list, $addr, $Rm",
5646 (ins VecListOneDByteIndexed:$list, addrmode6:$addr,
5647 rGPR:$Rm, pred:$p)>;
5648 defm VLD1LNdWB_register_Asm :
5649 NEONDT16AsmPseudoInst<"vld1${p}", "$list, $addr, $Rm",
5650 (ins VecListOneDByteIndexed:$list, addrmode6:$addr,
5651 rGPR:$Rm, pred:$p)>;
5652 defm VLD1LNdWB_register_Asm :
5653 NEONDT32AsmPseudoInst<"vld1${p}", "$list, $addr, $Rm",
5654 (ins VecListOneDByteIndexed:$list, addrmode6:$addr,
5655 rGPR:$Rm, pred:$p)>;
5658 // VST1 single-lane pseudo-instructions. These need special handling for
5659 // the lane index that an InstAlias can't handle, so we use these instead.
5660 defm VST1LNdAsm : NEONDT8AsmPseudoInst<"vst1${p}", "$list, $addr",
5661 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5662 defm VST1LNdAsm : NEONDT16AsmPseudoInst<"vst1${p}", "$list, $addr",
5663 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5664 defm VST1LNdAsm : NEONDT32AsmPseudoInst<"vst1${p}", "$list, $addr",
5665 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5667 defm VST1LNdWB_fixed_Asm : NEONDT8AsmPseudoInst<"vst1${p}", "$list, $addr!",
5668 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5669 defm VST1LNdWB_fixed_Asm : NEONDT16AsmPseudoInst<"vst1${p}", "$list, $addr!",
5670 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5671 defm VST1LNdWB_fixed_Asm : NEONDT32AsmPseudoInst<"vst1${p}", "$list, $addr!",
5672 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5673 defm VST1LNdWB_register_Asm :
5674 NEONDT8AsmPseudoInst<"vst1${p}", "$list, $addr, $Rm",
5675 (ins VecListOneDByteIndexed:$list, addrmode6:$addr,
5676 rGPR:$Rm, pred:$p)>;
5677 defm VST1LNdWB_register_Asm :
5678 NEONDT16AsmPseudoInst<"vst1${p}", "$list, $addr, $Rm",
5679 (ins VecListOneDByteIndexed:$list, addrmode6:$addr,
5680 rGPR:$Rm, pred:$p)>;
5681 defm VST1LNdWB_register_Asm :
5682 NEONDT32AsmPseudoInst<"vst1${p}", "$list, $addr, $Rm",
5683 (ins VecListOneDByteIndexed:$list, addrmode6:$addr,
5684 rGPR:$Rm, pred:$p)>;
5686 // VLD2 single-lane pseudo-instructions. These need special handling for
5687 // the lane index that an InstAlias can't handle, so we use these instead.
5688 defm VLD2LNdAsm : NEONDT8AsmPseudoInst<"vld2${p}", "$list, $addr",
5689 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5690 defm VLD2LNdAsm : NEONDT16AsmPseudoInst<"vld2${p}", "$list, $addr",
5691 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5692 defm VLD2LNdAsm : NEONDT32AsmPseudoInst<"vld2${p}", "$list, $addr",
5693 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5695 defm VLD2LNdWB_fixed_Asm : NEONDT8AsmPseudoInst<"vld2${p}", "$list, $addr!",
5696 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5697 defm VLD2LNdWB_fixed_Asm : NEONDT16AsmPseudoInst<"vld2${p}", "$list, $addr!",
5698 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5699 defm VLD2LNdWB_fixed_Asm : NEONDT32AsmPseudoInst<"vld2${p}", "$list, $addr!",
5700 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5701 defm VLD2LNdWB_register_Asm :
5702 NEONDT8AsmPseudoInst<"vld2${p}", "$list, $addr, $Rm",
5703 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr,
5704 rGPR:$Rm, pred:$p)>;
5705 defm VLD2LNdWB_register_Asm :
5706 NEONDT16AsmPseudoInst<"vld2${p}", "$list, $addr, $Rm",
5707 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr,
5708 rGPR:$Rm, pred:$p)>;
5709 defm VLD2LNdWB_register_Asm :
5710 NEONDT32AsmPseudoInst<"vld2${p}", "$list, $addr, $Rm",
5711 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr,
5712 rGPR:$Rm, pred:$p)>;
5715 // VST2 single-lane pseudo-instructions. These need special handling for
5716 // the lane index that an InstAlias can't handle, so we use these instead.
5717 defm VST2LNdAsm : NEONDT8AsmPseudoInst<"vst2${p}", "$list, $addr",
5718 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5719 defm VST2LNdAsm : NEONDT16AsmPseudoInst<"vst2${p}", "$list, $addr",
5720 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5721 defm VST2LNdAsm : NEONDT32AsmPseudoInst<"vst2${p}", "$list, $addr",
5722 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5724 defm VST2LNdWB_fixed_Asm : NEONDT8AsmPseudoInst<"vst2${p}", "$list, $addr!",
5725 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5726 defm VST2LNdWB_fixed_Asm : NEONDT16AsmPseudoInst<"vst2${p}", "$list, $addr!",
5727 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5728 defm VST2LNdWB_fixed_Asm : NEONDT32AsmPseudoInst<"vst2${p}", "$list, $addr!",
5729 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5730 defm VST2LNdWB_register_Asm :
5731 NEONDT8AsmPseudoInst<"vst2${p}", "$list, $addr, $Rm",
5732 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr,
5733 rGPR:$Rm, pred:$p)>;
5734 defm VST2LNdWB_register_Asm :
5735 NEONDT16AsmPseudoInst<"vst2${p}", "$list, $addr, $Rm",
5736 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr,
5737 rGPR:$Rm, pred:$p)>;
5738 defm VST2LNdWB_register_Asm :
5739 NEONDT32AsmPseudoInst<"vst2${p}", "$list, $addr, $Rm",
5740 (ins VecListTwoDByteIndexed:$list, addrmode6:$addr,
5741 rGPR:$Rm, pred:$p)>;
5743 // VMOV takes an optional datatype suffix
5744 defm : VFPDTAnyInstAlias<"vmov${p}", "$Vd, $Vm",
5745 (VORRd DPR:$Vd, DPR:$Vm, DPR:$Vm, pred:$p)>;
5746 defm : VFPDTAnyInstAlias<"vmov${p}", "$Vd, $Vm",
5747 (VORRq QPR:$Vd, QPR:$Vm, QPR:$Vm, pred:$p)>;
5749 // VCLT (register) is an assembler alias for VCGT w/ the operands reversed.
5750 // D-register versions.
5751 def : NEONInstAlias<"vclt${p}.s8 $Dd, $Dn, $Dm",
5752 (VCGTsv8i8 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
5753 def : NEONInstAlias<"vclt${p}.s16 $Dd, $Dn, $Dm",
5754 (VCGTsv4i16 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
5755 def : NEONInstAlias<"vclt${p}.s32 $Dd, $Dn, $Dm",
5756 (VCGTsv2i32 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
5757 def : NEONInstAlias<"vclt${p}.u8 $Dd, $Dn, $Dm",
5758 (VCGTuv8i8 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
5759 def : NEONInstAlias<"vclt${p}.u16 $Dd, $Dn, $Dm",
5760 (VCGTuv4i16 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
5761 def : NEONInstAlias<"vclt${p}.u32 $Dd, $Dn, $Dm",
5762 (VCGTuv2i32 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
5763 def : NEONInstAlias<"vclt${p}.f32 $Dd, $Dn, $Dm",
5764 (VCGTfd DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
5765 // Q-register versions.
5766 def : NEONInstAlias<"vclt${p}.s8 $Qd, $Qn, $Qm",
5767 (VCGTsv16i8 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
5768 def : NEONInstAlias<"vclt${p}.s16 $Qd, $Qn, $Qm",
5769 (VCGTsv8i16 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
5770 def : NEONInstAlias<"vclt${p}.s32 $Qd, $Qn, $Qm",
5771 (VCGTsv4i32 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
5772 def : NEONInstAlias<"vclt${p}.u8 $Qd, $Qn, $Qm",
5773 (VCGTuv16i8 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
5774 def : NEONInstAlias<"vclt${p}.u16 $Qd, $Qn, $Qm",
5775 (VCGTuv8i16 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
5776 def : NEONInstAlias<"vclt${p}.u32 $Qd, $Qn, $Qm",
5777 (VCGTuv4i32 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
5778 def : NEONInstAlias<"vclt${p}.f32 $Qd, $Qn, $Qm",
5779 (VCGTfq QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
5781 // Two-operand variants for VEXT
5782 def : NEONInstAlias<"vext${p}.8 $Vdn, $Vm, $imm",
5783 (VEXTd8 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, imm0_7:$imm, pred:$p)>;
5784 def : NEONInstAlias<"vext${p}.16 $Vdn, $Vm, $imm",
5785 (VEXTd16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, imm0_3:$imm, pred:$p)>;
5786 def : NEONInstAlias<"vext${p}.32 $Vdn, $Vm, $imm",
5787 (VEXTd32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, imm0_1:$imm, pred:$p)>;
5789 def : NEONInstAlias<"vext${p}.8 $Vdn, $Vm, $imm",
5790 (VEXTq8 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, imm0_15:$imm, pred:$p)>;
5791 def : NEONInstAlias<"vext${p}.16 $Vdn, $Vm, $imm",
5792 (VEXTq16 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, imm0_7:$imm, pred:$p)>;
5793 def : NEONInstAlias<"vext${p}.32 $Vdn, $Vm, $imm",
5794 (VEXTq32 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, imm0_3:$imm, pred:$p)>;
5795 def : NEONInstAlias<"vext${p}.64 $Vdn, $Vm, $imm",
5796 (VEXTq64 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, imm0_1:$imm, pred:$p)>;
5798 // Two-operand variants for VQDMULH
5799 def : NEONInstAlias<"vqdmulh${p}.s16 $Vdn, $Vm",
5800 (VQDMULHv4i16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5801 def : NEONInstAlias<"vqdmulh${p}.s32 $Vdn, $Vm",
5802 (VQDMULHv2i32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5804 def : NEONInstAlias<"vqdmulh${p}.s16 $Vdn, $Vm",
5805 (VQDMULHv8i16 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5806 def : NEONInstAlias<"vqdmulh${p}.s32 $Vdn, $Vm",
5807 (VQDMULHv4i32 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5809 // 'gas' compatibility aliases for quad-word instructions. Strictly speaking,
5810 // these should restrict to just the Q register variants, but the register
5811 // classes are enough to match correctly regardless, so we keep it simple
5812 // and just use MnemonicAlias.
5813 def : NEONMnemonicAlias<"vbicq", "vbic">;
5814 def : NEONMnemonicAlias<"vandq", "vand">;
5815 def : NEONMnemonicAlias<"veorq", "veor">;
5816 def : NEONMnemonicAlias<"vorrq", "vorr">;
5818 def : NEONMnemonicAlias<"vmovq", "vmov">;
5819 def : NEONMnemonicAlias<"vmvnq", "vmvn">;
5821 def : NEONMnemonicAlias<"vaddq", "vadd">;
5822 def : NEONMnemonicAlias<"vsubq", "vsub">;
5824 def : NEONMnemonicAlias<"vminq", "vmin">;
5825 def : NEONMnemonicAlias<"vmaxq", "vmax">;
5827 def : NEONMnemonicAlias<"vmulq", "vmul">;
5829 def : NEONMnemonicAlias<"vabsq", "vabs">;
5831 def : NEONMnemonicAlias<"vshlq", "vshl">;
5832 def : NEONMnemonicAlias<"vshrq", "vshr">;
5834 def : NEONMnemonicAlias<"vcvtq", "vcvt">;
5836 def : NEONMnemonicAlias<"vcleq", "vcle">;
5837 def : NEONMnemonicAlias<"vceqq", "vceq">;