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 def VecListOneDAsmOperand : AsmOperandClass {
78 let Name = "VecListOneD";
79 let ParserMethod = "parseVectorList";
81 def VecListOneD : RegisterOperand<DPR, "printVectorListOne"> {
82 let ParserMatchClass = VecListOneDAsmOperand;
84 // Register list of two sequential D registers.
85 def VecListTwoDAsmOperand : AsmOperandClass {
86 let Name = "VecListTwoD";
87 let ParserMethod = "parseVectorList";
89 def VecListTwoD : RegisterOperand<DPR, "printVectorListTwo"> {
90 let ParserMatchClass = VecListTwoDAsmOperand;
92 // Register list of three sequential D registers.
93 def VecListThreeDAsmOperand : AsmOperandClass {
94 let Name = "VecListThreeD";
95 let ParserMethod = "parseVectorList";
97 def VecListThreeD : RegisterOperand<DPR, "printVectorListThree"> {
98 let ParserMatchClass = VecListThreeDAsmOperand;
100 // Register list of four sequential D registers.
101 def VecListFourDAsmOperand : AsmOperandClass {
102 let Name = "VecListFourD";
103 let ParserMethod = "parseVectorList";
105 def VecListFourD : RegisterOperand<DPR, "printVectorListFour"> {
106 let ParserMatchClass = VecListFourDAsmOperand;
108 // Register list of two D registers spaced by 2 (two sequential Q registers).
109 def VecListTwoQAsmOperand : AsmOperandClass {
110 let Name = "VecListTwoQ";
111 let ParserMethod = "parseVectorList";
113 def VecListTwoQ : RegisterOperand<DPR, "printVectorListTwo"> {
114 let ParserMatchClass = VecListTwoQAsmOperand;
117 //===----------------------------------------------------------------------===//
118 // NEON-specific DAG Nodes.
119 //===----------------------------------------------------------------------===//
121 def SDTARMVCMP : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisSameAs<1, 2>]>;
122 def SDTARMVCMPZ : SDTypeProfile<1, 1, []>;
124 def NEONvceq : SDNode<"ARMISD::VCEQ", SDTARMVCMP>;
125 def NEONvceqz : SDNode<"ARMISD::VCEQZ", SDTARMVCMPZ>;
126 def NEONvcge : SDNode<"ARMISD::VCGE", SDTARMVCMP>;
127 def NEONvcgez : SDNode<"ARMISD::VCGEZ", SDTARMVCMPZ>;
128 def NEONvclez : SDNode<"ARMISD::VCLEZ", SDTARMVCMPZ>;
129 def NEONvcgeu : SDNode<"ARMISD::VCGEU", SDTARMVCMP>;
130 def NEONvcgt : SDNode<"ARMISD::VCGT", SDTARMVCMP>;
131 def NEONvcgtz : SDNode<"ARMISD::VCGTZ", SDTARMVCMPZ>;
132 def NEONvcltz : SDNode<"ARMISD::VCLTZ", SDTARMVCMPZ>;
133 def NEONvcgtu : SDNode<"ARMISD::VCGTU", SDTARMVCMP>;
134 def NEONvtst : SDNode<"ARMISD::VTST", SDTARMVCMP>;
136 // Types for vector shift by immediates. The "SHX" version is for long and
137 // narrow operations where the source and destination vectors have different
138 // types. The "SHINS" version is for shift and insert operations.
139 def SDTARMVSH : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
141 def SDTARMVSHX : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
143 def SDTARMVSHINS : SDTypeProfile<1, 3, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
144 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
146 def NEONvshl : SDNode<"ARMISD::VSHL", SDTARMVSH>;
147 def NEONvshrs : SDNode<"ARMISD::VSHRs", SDTARMVSH>;
148 def NEONvshru : SDNode<"ARMISD::VSHRu", SDTARMVSH>;
149 def NEONvshlls : SDNode<"ARMISD::VSHLLs", SDTARMVSHX>;
150 def NEONvshllu : SDNode<"ARMISD::VSHLLu", SDTARMVSHX>;
151 def NEONvshlli : SDNode<"ARMISD::VSHLLi", SDTARMVSHX>;
152 def NEONvshrn : SDNode<"ARMISD::VSHRN", SDTARMVSHX>;
154 def NEONvrshrs : SDNode<"ARMISD::VRSHRs", SDTARMVSH>;
155 def NEONvrshru : SDNode<"ARMISD::VRSHRu", SDTARMVSH>;
156 def NEONvrshrn : SDNode<"ARMISD::VRSHRN", SDTARMVSHX>;
158 def NEONvqshls : SDNode<"ARMISD::VQSHLs", SDTARMVSH>;
159 def NEONvqshlu : SDNode<"ARMISD::VQSHLu", SDTARMVSH>;
160 def NEONvqshlsu : SDNode<"ARMISD::VQSHLsu", SDTARMVSH>;
161 def NEONvqshrns : SDNode<"ARMISD::VQSHRNs", SDTARMVSHX>;
162 def NEONvqshrnu : SDNode<"ARMISD::VQSHRNu", SDTARMVSHX>;
163 def NEONvqshrnsu : SDNode<"ARMISD::VQSHRNsu", SDTARMVSHX>;
165 def NEONvqrshrns : SDNode<"ARMISD::VQRSHRNs", SDTARMVSHX>;
166 def NEONvqrshrnu : SDNode<"ARMISD::VQRSHRNu", SDTARMVSHX>;
167 def NEONvqrshrnsu : SDNode<"ARMISD::VQRSHRNsu", SDTARMVSHX>;
169 def NEONvsli : SDNode<"ARMISD::VSLI", SDTARMVSHINS>;
170 def NEONvsri : SDNode<"ARMISD::VSRI", SDTARMVSHINS>;
172 def SDTARMVGETLN : SDTypeProfile<1, 2, [SDTCisVT<0, i32>, SDTCisInt<1>,
174 def NEONvgetlaneu : SDNode<"ARMISD::VGETLANEu", SDTARMVGETLN>;
175 def NEONvgetlanes : SDNode<"ARMISD::VGETLANEs", SDTARMVGETLN>;
177 def SDTARMVMOVIMM : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisVT<1, i32>]>;
178 def NEONvmovImm : SDNode<"ARMISD::VMOVIMM", SDTARMVMOVIMM>;
179 def NEONvmvnImm : SDNode<"ARMISD::VMVNIMM", SDTARMVMOVIMM>;
180 def NEONvmovFPImm : SDNode<"ARMISD::VMOVFPIMM", SDTARMVMOVIMM>;
182 def SDTARMVORRIMM : SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
184 def NEONvorrImm : SDNode<"ARMISD::VORRIMM", SDTARMVORRIMM>;
185 def NEONvbicImm : SDNode<"ARMISD::VBICIMM", SDTARMVORRIMM>;
187 def NEONvbsl : SDNode<"ARMISD::VBSL",
188 SDTypeProfile<1, 3, [SDTCisVec<0>,
191 SDTCisSameAs<0, 3>]>>;
193 def NEONvdup : SDNode<"ARMISD::VDUP", SDTypeProfile<1, 1, [SDTCisVec<0>]>>;
195 // VDUPLANE can produce a quad-register result from a double-register source,
196 // so the result is not constrained to match the source.
197 def NEONvduplane : SDNode<"ARMISD::VDUPLANE",
198 SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisVec<1>,
201 def SDTARMVEXT : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
202 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
203 def NEONvext : SDNode<"ARMISD::VEXT", SDTARMVEXT>;
205 def SDTARMVSHUF : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisSameAs<0, 1>]>;
206 def NEONvrev64 : SDNode<"ARMISD::VREV64", SDTARMVSHUF>;
207 def NEONvrev32 : SDNode<"ARMISD::VREV32", SDTARMVSHUF>;
208 def NEONvrev16 : SDNode<"ARMISD::VREV16", SDTARMVSHUF>;
210 def SDTARMVSHUF2 : SDTypeProfile<2, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
212 SDTCisSameAs<0, 3>]>;
213 def NEONzip : SDNode<"ARMISD::VZIP", SDTARMVSHUF2>;
214 def NEONuzp : SDNode<"ARMISD::VUZP", SDTARMVSHUF2>;
215 def NEONtrn : SDNode<"ARMISD::VTRN", SDTARMVSHUF2>;
217 def SDTARMVMULL : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
218 SDTCisSameAs<1, 2>]>;
219 def NEONvmulls : SDNode<"ARMISD::VMULLs", SDTARMVMULL>;
220 def NEONvmullu : SDNode<"ARMISD::VMULLu", SDTARMVMULL>;
222 def SDTARMFMAX : SDTypeProfile<1, 2, [SDTCisVT<0, f32>, SDTCisSameAs<0, 1>,
223 SDTCisSameAs<0, 2>]>;
224 def NEONfmax : SDNode<"ARMISD::FMAX", SDTARMFMAX>;
225 def NEONfmin : SDNode<"ARMISD::FMIN", SDTARMFMAX>;
227 def NEONimmAllZerosV: PatLeaf<(NEONvmovImm (i32 timm)), [{
228 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
229 unsigned EltBits = 0;
230 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
231 return (EltBits == 32 && EltVal == 0);
234 def NEONimmAllOnesV: PatLeaf<(NEONvmovImm (i32 timm)), [{
235 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
236 unsigned EltBits = 0;
237 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
238 return (EltBits == 8 && EltVal == 0xff);
241 //===----------------------------------------------------------------------===//
242 // NEON load / store instructions
243 //===----------------------------------------------------------------------===//
245 // Use VLDM to load a Q register as a D register pair.
246 // This is a pseudo instruction that is expanded to VLDMD after reg alloc.
248 : PseudoVFPLdStM<(outs QPR:$dst), (ins GPR:$Rn),
250 [(set QPR:$dst, (v2f64 (load GPR:$Rn)))]>;
252 // Use VSTM to store a Q register as a D register pair.
253 // This is a pseudo instruction that is expanded to VSTMD after reg alloc.
255 : PseudoVFPLdStM<(outs), (ins QPR:$src, GPR:$Rn),
257 [(store (v2f64 QPR:$src), GPR:$Rn)]>;
259 // Classes for VLD* pseudo-instructions with multi-register operands.
260 // These are expanded to real instructions after register allocation.
261 class VLDQPseudo<InstrItinClass itin>
262 : PseudoNLdSt<(outs QPR:$dst), (ins addrmode6:$addr), itin, "">;
263 class VLDQWBPseudo<InstrItinClass itin>
264 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
265 (ins addrmode6:$addr, am6offset:$offset), itin,
267 class VLDQWBfixedPseudo<InstrItinClass itin>
268 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
269 (ins addrmode6:$addr), itin,
271 class VLDQWBregisterPseudo<InstrItinClass itin>
272 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
273 (ins addrmode6:$addr, rGPR:$offset), itin,
275 class VLDQQPseudo<InstrItinClass itin>
276 : PseudoNLdSt<(outs QQPR:$dst), (ins addrmode6:$addr), itin, "">;
277 class VLDQQWBPseudo<InstrItinClass itin>
278 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
279 (ins addrmode6:$addr, am6offset:$offset), itin,
281 class VLDQQQQPseudo<InstrItinClass itin>
282 : PseudoNLdSt<(outs QQQQPR:$dst), (ins addrmode6:$addr, QQQQPR:$src),itin,
284 class VLDQQQQWBPseudo<InstrItinClass itin>
285 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
286 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
287 "$addr.addr = $wb, $src = $dst">;
289 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
291 // VLD1 : Vector Load (multiple single elements)
292 class VLD1D<bits<4> op7_4, string Dt>
293 : NLdSt<0,0b10,0b0111,op7_4, (outs VecListOneD:$Vd),
294 (ins addrmode6:$Rn), IIC_VLD1,
295 "vld1", Dt, "$Vd, $Rn", "", []> {
298 let DecoderMethod = "DecodeVLDInstruction";
300 class VLD1Q<bits<4> op7_4, string Dt>
301 : NLdSt<0,0b10,0b1010,op7_4, (outs VecListTwoD:$Vd),
302 (ins addrmode6:$Rn), IIC_VLD1x2,
303 "vld1", Dt, "$Vd, $Rn", "", []> {
305 let Inst{5-4} = Rn{5-4};
306 let DecoderMethod = "DecodeVLDInstruction";
309 def VLD1d8 : VLD1D<{0,0,0,?}, "8">;
310 def VLD1d16 : VLD1D<{0,1,0,?}, "16">;
311 def VLD1d32 : VLD1D<{1,0,0,?}, "32">;
312 def VLD1d64 : VLD1D<{1,1,0,?}, "64">;
314 def VLD1q8 : VLD1Q<{0,0,?,?}, "8">;
315 def VLD1q16 : VLD1Q<{0,1,?,?}, "16">;
316 def VLD1q32 : VLD1Q<{1,0,?,?}, "32">;
317 def VLD1q64 : VLD1Q<{1,1,?,?}, "64">;
319 def VLD1q8Pseudo : VLDQPseudo<IIC_VLD1x2>;
320 def VLD1q16Pseudo : VLDQPseudo<IIC_VLD1x2>;
321 def VLD1q32Pseudo : VLDQPseudo<IIC_VLD1x2>;
322 def VLD1q64Pseudo : VLDQPseudo<IIC_VLD1x2>;
324 // ...with address register writeback:
325 multiclass VLD1DWB<bits<4> op7_4, string Dt> {
326 def _fixed : NLdSt<0,0b10, 0b0111,op7_4, (outs VecListOneD:$Vd, GPR:$wb),
327 (ins addrmode6:$Rn), IIC_VLD1u,
328 "vld1", Dt, "$Vd, $Rn!",
329 "$Rn.addr = $wb", []> {
330 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
332 let DecoderMethod = "DecodeVLDInstruction";
333 let AsmMatchConverter = "cvtVLDwbFixed";
335 def _register : NLdSt<0,0b10,0b0111,op7_4, (outs VecListOneD:$Vd, GPR:$wb),
336 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1u,
337 "vld1", Dt, "$Vd, $Rn, $Rm",
338 "$Rn.addr = $wb", []> {
340 let DecoderMethod = "DecodeVLDInstruction";
341 let AsmMatchConverter = "cvtVLDwbRegister";
344 multiclass VLD1QWB<bits<4> op7_4, string Dt> {
345 def _fixed : NLdSt<0,0b10,0b1010,op7_4, (outs VecListTwoD:$Vd, GPR:$wb),
346 (ins addrmode6:$Rn), IIC_VLD1x2u,
347 "vld1", Dt, "$Vd, $Rn!",
348 "$Rn.addr = $wb", []> {
349 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
350 let Inst{5-4} = Rn{5-4};
351 let DecoderMethod = "DecodeVLDInstruction";
352 let AsmMatchConverter = "cvtVLDwbFixed";
354 def _register : NLdSt<0,0b10,0b1010,op7_4, (outs VecListTwoD:$Vd, GPR:$wb),
355 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1x2u,
356 "vld1", Dt, "$Vd, $Rn, $Rm",
357 "$Rn.addr = $wb", []> {
358 let Inst{5-4} = Rn{5-4};
359 let DecoderMethod = "DecodeVLDInstruction";
360 let AsmMatchConverter = "cvtVLDwbRegister";
364 defm VLD1d8wb : VLD1DWB<{0,0,0,?}, "8">;
365 defm VLD1d16wb : VLD1DWB<{0,1,0,?}, "16">;
366 defm VLD1d32wb : VLD1DWB<{1,0,0,?}, "32">;
367 defm VLD1d64wb : VLD1DWB<{1,1,0,?}, "64">;
368 defm VLD1q8wb : VLD1QWB<{0,0,?,?}, "8">;
369 defm VLD1q16wb : VLD1QWB<{0,1,?,?}, "16">;
370 defm VLD1q32wb : VLD1QWB<{1,0,?,?}, "32">;
371 defm VLD1q64wb : VLD1QWB<{1,1,?,?}, "64">;
373 def VLD1q8PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD1x2u>;
374 def VLD1q16PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD1x2u>;
375 def VLD1q32PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD1x2u>;
376 def VLD1q64PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD1x2u>;
377 def VLD1q8PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD1x2u>;
378 def VLD1q16PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD1x2u>;
379 def VLD1q32PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD1x2u>;
380 def VLD1q64PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD1x2u>;
382 // ...with 3 registers
383 class VLD1D3<bits<4> op7_4, string Dt>
384 : NLdSt<0,0b10,0b0110,op7_4, (outs VecListThreeD:$Vd),
385 (ins addrmode6:$Rn), IIC_VLD1x3, "vld1", Dt,
386 "$Vd, $Rn", "", []> {
389 let DecoderMethod = "DecodeVLDInstruction";
391 multiclass VLD1D3WB<bits<4> op7_4, string Dt> {
392 def _fixed : NLdSt<0,0b10,0b0110, op7_4, (outs VecListThreeD:$Vd, GPR:$wb),
393 (ins addrmode6:$Rn), IIC_VLD1x2u,
394 "vld1", Dt, "$Vd, $Rn!",
395 "$Rn.addr = $wb", []> {
396 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
398 let DecoderMethod = "DecodeVLDInstruction";
399 let AsmMatchConverter = "cvtVLDwbFixed";
401 def _register : NLdSt<0,0b10,0b0110,op7_4, (outs VecListThreeD:$Vd, GPR:$wb),
402 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1x2u,
403 "vld1", Dt, "$Vd, $Rn, $Rm",
404 "$Rn.addr = $wb", []> {
406 let DecoderMethod = "DecodeVLDInstruction";
407 let AsmMatchConverter = "cvtVLDwbRegister";
411 def VLD1d8T : VLD1D3<{0,0,0,?}, "8">;
412 def VLD1d16T : VLD1D3<{0,1,0,?}, "16">;
413 def VLD1d32T : VLD1D3<{1,0,0,?}, "32">;
414 def VLD1d64T : VLD1D3<{1,1,0,?}, "64">;
416 defm VLD1d8Twb : VLD1D3WB<{0,0,0,?}, "8">;
417 defm VLD1d16Twb : VLD1D3WB<{0,1,0,?}, "16">;
418 defm VLD1d32Twb : VLD1D3WB<{1,0,0,?}, "32">;
419 defm VLD1d64Twb : VLD1D3WB<{1,1,0,?}, "64">;
421 def VLD1d64TPseudo : VLDQQPseudo<IIC_VLD1x3>;
423 // ...with 4 registers
424 class VLD1D4<bits<4> op7_4, string Dt>
425 : NLdSt<0, 0b10, 0b0010, op7_4, (outs VecListFourD:$Vd),
426 (ins addrmode6:$Rn), IIC_VLD1x4, "vld1", Dt,
427 "$Vd, $Rn", "", []> {
429 let Inst{5-4} = Rn{5-4};
430 let DecoderMethod = "DecodeVLDInstruction";
432 multiclass VLD1D4WB<bits<4> op7_4, string Dt> {
433 def _fixed : NLdSt<0,0b10,0b0010, op7_4, (outs VecListFourD:$Vd, GPR:$wb),
434 (ins addrmode6:$Rn), IIC_VLD1x2u,
435 "vld1", Dt, "$Vd, $Rn!",
436 "$Rn.addr = $wb", []> {
437 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
438 let Inst{5-4} = Rn{5-4};
439 let DecoderMethod = "DecodeVLDInstruction";
440 let AsmMatchConverter = "cvtVLDwbFixed";
442 def _register : NLdSt<0,0b10,0b0010,op7_4, (outs VecListFourD:$Vd, GPR:$wb),
443 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1x2u,
444 "vld1", Dt, "$Vd, $Rn, $Rm",
445 "$Rn.addr = $wb", []> {
446 let Inst{5-4} = Rn{5-4};
447 let DecoderMethod = "DecodeVLDInstruction";
448 let AsmMatchConverter = "cvtVLDwbRegister";
452 def VLD1d8Q : VLD1D4<{0,0,?,?}, "8">;
453 def VLD1d16Q : VLD1D4<{0,1,?,?}, "16">;
454 def VLD1d32Q : VLD1D4<{1,0,?,?}, "32">;
455 def VLD1d64Q : VLD1D4<{1,1,?,?}, "64">;
457 defm VLD1d8Qwb : VLD1D4WB<{0,0,?,?}, "8">;
458 defm VLD1d16Qwb : VLD1D4WB<{0,1,?,?}, "16">;
459 defm VLD1d32Qwb : VLD1D4WB<{1,0,?,?}, "32">;
460 defm VLD1d64Qwb : VLD1D4WB<{1,1,?,?}, "64">;
462 def VLD1d64QPseudo : VLDQQPseudo<IIC_VLD1x4>;
464 // VLD2 : Vector Load (multiple 2-element structures)
465 class VLD2D<bits<4> op11_8, bits<4> op7_4, string Dt, RegisterOperand VdTy>
466 : NLdSt<0, 0b10, op11_8, op7_4, (outs VdTy:$Vd),
467 (ins addrmode6:$Rn), IIC_VLD2,
468 "vld2", Dt, "$Vd, $Rn", "", []> {
470 let Inst{5-4} = Rn{5-4};
471 let DecoderMethod = "DecodeVLDInstruction";
473 class VLD2Q<bits<4> op7_4, string Dt, RegisterOperand VdTy>
474 : NLdSt<0, 0b10, 0b0011, op7_4,
476 (ins addrmode6:$Rn), IIC_VLD2x2,
477 "vld2", Dt, "$Vd, $Rn", "", []> {
479 let Inst{5-4} = Rn{5-4};
480 let DecoderMethod = "DecodeVLDInstruction";
483 def VLD2d8 : VLD2D<0b1000, {0,0,?,?}, "8", VecListTwoD>;
484 def VLD2d16 : VLD2D<0b1000, {0,1,?,?}, "16", VecListTwoD>;
485 def VLD2d32 : VLD2D<0b1000, {1,0,?,?}, "32", VecListTwoD>;
487 def VLD2q8 : VLD2Q<{0,0,?,?}, "8", VecListFourD>;
488 def VLD2q16 : VLD2Q<{0,1,?,?}, "16", VecListFourD>;
489 def VLD2q32 : VLD2Q<{1,0,?,?}, "32", VecListFourD>;
491 def VLD2d8Pseudo : VLDQPseudo<IIC_VLD2>;
492 def VLD2d16Pseudo : VLDQPseudo<IIC_VLD2>;
493 def VLD2d32Pseudo : VLDQPseudo<IIC_VLD2>;
495 def VLD2q8Pseudo : VLDQQPseudo<IIC_VLD2x2>;
496 def VLD2q16Pseudo : VLDQQPseudo<IIC_VLD2x2>;
497 def VLD2q32Pseudo : VLDQQPseudo<IIC_VLD2x2>;
499 // ...with address register writeback:
500 class VLD2DWB<bits<4> op11_8, bits<4> op7_4, string Dt, RegisterOperand VdTy>
501 : NLdSt<0, 0b10, op11_8, op7_4, (outs VdTy:$Vd, GPR:$wb),
502 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD2u,
503 "vld2", Dt, "$Vd, $Rn$Rm",
504 "$Rn.addr = $wb", []> {
505 let Inst{5-4} = Rn{5-4};
506 let DecoderMethod = "DecodeVLDInstruction";
508 class VLD2QWB<bits<4> op7_4, string Dt, RegisterOperand VdTy>
509 : NLdSt<0, 0b10, 0b0011, op7_4,
510 (outs VdTy:$Vd, GPR:$wb),
511 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD2x2u,
512 "vld2", Dt, "$Vd, $Rn$Rm",
513 "$Rn.addr = $wb", []> {
514 let Inst{5-4} = Rn{5-4};
515 let DecoderMethod = "DecodeVLDInstruction";
518 def VLD2d8_UPD : VLD2DWB<0b1000, {0,0,?,?}, "8", VecListTwoD>;
519 def VLD2d16_UPD : VLD2DWB<0b1000, {0,1,?,?}, "16", VecListTwoD>;
520 def VLD2d32_UPD : VLD2DWB<0b1000, {1,0,?,?}, "32", VecListTwoD>;
522 def VLD2q8_UPD : VLD2QWB<{0,0,?,?}, "8", VecListFourD>;
523 def VLD2q16_UPD : VLD2QWB<{0,1,?,?}, "16", VecListFourD>;
524 def VLD2q32_UPD : VLD2QWB<{1,0,?,?}, "32", VecListFourD>;
526 def VLD2d8Pseudo_UPD : VLDQWBPseudo<IIC_VLD2u>;
527 def VLD2d16Pseudo_UPD : VLDQWBPseudo<IIC_VLD2u>;
528 def VLD2d32Pseudo_UPD : VLDQWBPseudo<IIC_VLD2u>;
530 def VLD2q8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD2x2u>;
531 def VLD2q16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD2x2u>;
532 def VLD2q32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD2x2u>;
534 // ...with double-spaced registers
535 def VLD2b8 : VLD2D<0b1001, {0,0,?,?}, "8", VecListTwoQ>;
536 def VLD2b16 : VLD2D<0b1001, {0,1,?,?}, "16", VecListTwoQ>;
537 def VLD2b32 : VLD2D<0b1001, {1,0,?,?}, "32", VecListTwoQ>;
538 def VLD2b8_UPD : VLD2DWB<0b1001, {0,0,?,?}, "8", VecListTwoQ>;
539 def VLD2b16_UPD : VLD2DWB<0b1001, {0,1,?,?}, "16", VecListTwoQ>;
540 def VLD2b32_UPD : VLD2DWB<0b1001, {1,0,?,?}, "32", VecListTwoQ>;
542 // VLD3 : Vector Load (multiple 3-element structures)
543 class VLD3D<bits<4> op11_8, bits<4> op7_4, string Dt>
544 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
545 (ins addrmode6:$Rn), IIC_VLD3,
546 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn", "", []> {
549 let DecoderMethod = "DecodeVLDInstruction";
552 def VLD3d8 : VLD3D<0b0100, {0,0,0,?}, "8">;
553 def VLD3d16 : VLD3D<0b0100, {0,1,0,?}, "16">;
554 def VLD3d32 : VLD3D<0b0100, {1,0,0,?}, "32">;
556 def VLD3d8Pseudo : VLDQQPseudo<IIC_VLD3>;
557 def VLD3d16Pseudo : VLDQQPseudo<IIC_VLD3>;
558 def VLD3d32Pseudo : VLDQQPseudo<IIC_VLD3>;
560 // ...with address register writeback:
561 class VLD3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
562 : NLdSt<0, 0b10, op11_8, op7_4,
563 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
564 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD3u,
565 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn$Rm",
566 "$Rn.addr = $wb", []> {
568 let DecoderMethod = "DecodeVLDInstruction";
571 def VLD3d8_UPD : VLD3DWB<0b0100, {0,0,0,?}, "8">;
572 def VLD3d16_UPD : VLD3DWB<0b0100, {0,1,0,?}, "16">;
573 def VLD3d32_UPD : VLD3DWB<0b0100, {1,0,0,?}, "32">;
575 def VLD3d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
576 def VLD3d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
577 def VLD3d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
579 // ...with double-spaced registers:
580 def VLD3q8 : VLD3D<0b0101, {0,0,0,?}, "8">;
581 def VLD3q16 : VLD3D<0b0101, {0,1,0,?}, "16">;
582 def VLD3q32 : VLD3D<0b0101, {1,0,0,?}, "32">;
583 def VLD3q8_UPD : VLD3DWB<0b0101, {0,0,0,?}, "8">;
584 def VLD3q16_UPD : VLD3DWB<0b0101, {0,1,0,?}, "16">;
585 def VLD3q32_UPD : VLD3DWB<0b0101, {1,0,0,?}, "32">;
587 def VLD3q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
588 def VLD3q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
589 def VLD3q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
591 // ...alternate versions to be allocated odd register numbers:
592 def VLD3q8oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
593 def VLD3q16oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
594 def VLD3q32oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
596 def VLD3q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
597 def VLD3q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
598 def VLD3q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
600 // VLD4 : Vector Load (multiple 4-element structures)
601 class VLD4D<bits<4> op11_8, bits<4> op7_4, string Dt>
602 : NLdSt<0, 0b10, op11_8, op7_4,
603 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
604 (ins addrmode6:$Rn), IIC_VLD4,
605 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn", "", []> {
607 let Inst{5-4} = Rn{5-4};
608 let DecoderMethod = "DecodeVLDInstruction";
611 def VLD4d8 : VLD4D<0b0000, {0,0,?,?}, "8">;
612 def VLD4d16 : VLD4D<0b0000, {0,1,?,?}, "16">;
613 def VLD4d32 : VLD4D<0b0000, {1,0,?,?}, "32">;
615 def VLD4d8Pseudo : VLDQQPseudo<IIC_VLD4>;
616 def VLD4d16Pseudo : VLDQQPseudo<IIC_VLD4>;
617 def VLD4d32Pseudo : VLDQQPseudo<IIC_VLD4>;
619 // ...with address register writeback:
620 class VLD4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
621 : NLdSt<0, 0b10, op11_8, op7_4,
622 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
623 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD4u,
624 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn$Rm",
625 "$Rn.addr = $wb", []> {
626 let Inst{5-4} = Rn{5-4};
627 let DecoderMethod = "DecodeVLDInstruction";
630 def VLD4d8_UPD : VLD4DWB<0b0000, {0,0,?,?}, "8">;
631 def VLD4d16_UPD : VLD4DWB<0b0000, {0,1,?,?}, "16">;
632 def VLD4d32_UPD : VLD4DWB<0b0000, {1,0,?,?}, "32">;
634 def VLD4d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
635 def VLD4d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
636 def VLD4d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
638 // ...with double-spaced registers:
639 def VLD4q8 : VLD4D<0b0001, {0,0,?,?}, "8">;
640 def VLD4q16 : VLD4D<0b0001, {0,1,?,?}, "16">;
641 def VLD4q32 : VLD4D<0b0001, {1,0,?,?}, "32">;
642 def VLD4q8_UPD : VLD4DWB<0b0001, {0,0,?,?}, "8">;
643 def VLD4q16_UPD : VLD4DWB<0b0001, {0,1,?,?}, "16">;
644 def VLD4q32_UPD : VLD4DWB<0b0001, {1,0,?,?}, "32">;
646 def VLD4q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
647 def VLD4q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
648 def VLD4q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
650 // ...alternate versions to be allocated odd register numbers:
651 def VLD4q8oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
652 def VLD4q16oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
653 def VLD4q32oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
655 def VLD4q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
656 def VLD4q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
657 def VLD4q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
659 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
661 // Classes for VLD*LN pseudo-instructions with multi-register operands.
662 // These are expanded to real instructions after register allocation.
663 class VLDQLNPseudo<InstrItinClass itin>
664 : PseudoNLdSt<(outs QPR:$dst),
665 (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
666 itin, "$src = $dst">;
667 class VLDQLNWBPseudo<InstrItinClass itin>
668 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
669 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
670 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
671 class VLDQQLNPseudo<InstrItinClass itin>
672 : PseudoNLdSt<(outs QQPR:$dst),
673 (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
674 itin, "$src = $dst">;
675 class VLDQQLNWBPseudo<InstrItinClass itin>
676 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
677 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
678 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
679 class VLDQQQQLNPseudo<InstrItinClass itin>
680 : PseudoNLdSt<(outs QQQQPR:$dst),
681 (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
682 itin, "$src = $dst">;
683 class VLDQQQQLNWBPseudo<InstrItinClass itin>
684 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
685 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
686 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
688 // VLD1LN : Vector Load (single element to one lane)
689 class VLD1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
691 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd),
692 (ins addrmode6:$Rn, DPR:$src, nohash_imm:$lane),
693 IIC_VLD1ln, "vld1", Dt, "\\{$Vd[$lane]\\}, $Rn",
695 [(set DPR:$Vd, (vector_insert (Ty DPR:$src),
696 (i32 (LoadOp addrmode6:$Rn)),
699 let DecoderMethod = "DecodeVLD1LN";
701 class VLD1LN32<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
703 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd),
704 (ins addrmode6oneL32:$Rn, DPR:$src, nohash_imm:$lane),
705 IIC_VLD1ln, "vld1", Dt, "\\{$Vd[$lane]\\}, $Rn",
707 [(set DPR:$Vd, (vector_insert (Ty DPR:$src),
708 (i32 (LoadOp addrmode6oneL32:$Rn)),
711 let DecoderMethod = "DecodeVLD1LN";
713 class VLD1QLNPseudo<ValueType Ty, PatFrag LoadOp> : VLDQLNPseudo<IIC_VLD1ln> {
714 let Pattern = [(set QPR:$dst, (vector_insert (Ty QPR:$src),
715 (i32 (LoadOp addrmode6:$addr)),
719 def VLD1LNd8 : VLD1LN<0b0000, {?,?,?,0}, "8", v8i8, extloadi8> {
720 let Inst{7-5} = lane{2-0};
722 def VLD1LNd16 : VLD1LN<0b0100, {?,?,0,?}, "16", v4i16, extloadi16> {
723 let Inst{7-6} = lane{1-0};
726 def VLD1LNd32 : VLD1LN32<0b1000, {?,0,?,?}, "32", v2i32, load> {
727 let Inst{7} = lane{0};
732 def VLD1LNq8Pseudo : VLD1QLNPseudo<v16i8, extloadi8>;
733 def VLD1LNq16Pseudo : VLD1QLNPseudo<v8i16, extloadi16>;
734 def VLD1LNq32Pseudo : VLD1QLNPseudo<v4i32, load>;
736 def : Pat<(vector_insert (v2f32 DPR:$src),
737 (f32 (load addrmode6:$addr)), imm:$lane),
738 (VLD1LNd32 addrmode6:$addr, DPR:$src, imm:$lane)>;
739 def : Pat<(vector_insert (v4f32 QPR:$src),
740 (f32 (load addrmode6:$addr)), imm:$lane),
741 (VLD1LNq32Pseudo addrmode6:$addr, QPR:$src, imm:$lane)>;
743 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
745 // ...with address register writeback:
746 class VLD1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
747 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, GPR:$wb),
748 (ins addrmode6:$Rn, am6offset:$Rm,
749 DPR:$src, nohash_imm:$lane), IIC_VLD1lnu, "vld1", Dt,
750 "\\{$Vd[$lane]\\}, $Rn$Rm",
751 "$src = $Vd, $Rn.addr = $wb", []> {
752 let DecoderMethod = "DecodeVLD1LN";
755 def VLD1LNd8_UPD : VLD1LNWB<0b0000, {?,?,?,0}, "8"> {
756 let Inst{7-5} = lane{2-0};
758 def VLD1LNd16_UPD : VLD1LNWB<0b0100, {?,?,0,?}, "16"> {
759 let Inst{7-6} = lane{1-0};
762 def VLD1LNd32_UPD : VLD1LNWB<0b1000, {?,0,?,?}, "32"> {
763 let Inst{7} = lane{0};
768 def VLD1LNq8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
769 def VLD1LNq16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
770 def VLD1LNq32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
772 // VLD2LN : Vector Load (single 2-element structure to one lane)
773 class VLD2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
774 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2),
775 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, nohash_imm:$lane),
776 IIC_VLD2ln, "vld2", Dt, "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn",
777 "$src1 = $Vd, $src2 = $dst2", []> {
780 let DecoderMethod = "DecodeVLD2LN";
783 def VLD2LNd8 : VLD2LN<0b0001, {?,?,?,?}, "8"> {
784 let Inst{7-5} = lane{2-0};
786 def VLD2LNd16 : VLD2LN<0b0101, {?,?,0,?}, "16"> {
787 let Inst{7-6} = lane{1-0};
789 def VLD2LNd32 : VLD2LN<0b1001, {?,0,0,?}, "32"> {
790 let Inst{7} = lane{0};
793 def VLD2LNd8Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
794 def VLD2LNd16Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
795 def VLD2LNd32Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
797 // ...with double-spaced registers:
798 def VLD2LNq16 : VLD2LN<0b0101, {?,?,1,?}, "16"> {
799 let Inst{7-6} = lane{1-0};
801 def VLD2LNq32 : VLD2LN<0b1001, {?,1,0,?}, "32"> {
802 let Inst{7} = lane{0};
805 def VLD2LNq16Pseudo : VLDQQLNPseudo<IIC_VLD2ln>;
806 def VLD2LNq32Pseudo : VLDQQLNPseudo<IIC_VLD2ln>;
808 // ...with address register writeback:
809 class VLD2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
810 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
811 (ins addrmode6:$Rn, am6offset:$Rm,
812 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VLD2lnu, "vld2", Dt,
813 "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn$Rm",
814 "$src1 = $Vd, $src2 = $dst2, $Rn.addr = $wb", []> {
816 let DecoderMethod = "DecodeVLD2LN";
819 def VLD2LNd8_UPD : VLD2LNWB<0b0001, {?,?,?,?}, "8"> {
820 let Inst{7-5} = lane{2-0};
822 def VLD2LNd16_UPD : VLD2LNWB<0b0101, {?,?,0,?}, "16"> {
823 let Inst{7-6} = lane{1-0};
825 def VLD2LNd32_UPD : VLD2LNWB<0b1001, {?,0,0,?}, "32"> {
826 let Inst{7} = lane{0};
829 def VLD2LNd8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
830 def VLD2LNd16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
831 def VLD2LNd32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
833 def VLD2LNq16_UPD : VLD2LNWB<0b0101, {?,?,1,?}, "16"> {
834 let Inst{7-6} = lane{1-0};
836 def VLD2LNq32_UPD : VLD2LNWB<0b1001, {?,1,0,?}, "32"> {
837 let Inst{7} = lane{0};
840 def VLD2LNq16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
841 def VLD2LNq32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
843 // VLD3LN : Vector Load (single 3-element structure to one lane)
844 class VLD3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
845 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
846 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3,
847 nohash_imm:$lane), IIC_VLD3ln, "vld3", Dt,
848 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn",
849 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3", []> {
851 let DecoderMethod = "DecodeVLD3LN";
854 def VLD3LNd8 : VLD3LN<0b0010, {?,?,?,0}, "8"> {
855 let Inst{7-5} = lane{2-0};
857 def VLD3LNd16 : VLD3LN<0b0110, {?,?,0,0}, "16"> {
858 let Inst{7-6} = lane{1-0};
860 def VLD3LNd32 : VLD3LN<0b1010, {?,0,0,0}, "32"> {
861 let Inst{7} = lane{0};
864 def VLD3LNd8Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
865 def VLD3LNd16Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
866 def VLD3LNd32Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
868 // ...with double-spaced registers:
869 def VLD3LNq16 : VLD3LN<0b0110, {?,?,1,0}, "16"> {
870 let Inst{7-6} = lane{1-0};
872 def VLD3LNq32 : VLD3LN<0b1010, {?,1,0,0}, "32"> {
873 let Inst{7} = lane{0};
876 def VLD3LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
877 def VLD3LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
879 // ...with address register writeback:
880 class VLD3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
881 : NLdStLn<1, 0b10, op11_8, op7_4,
882 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
883 (ins addrmode6:$Rn, am6offset:$Rm,
884 DPR:$src1, DPR:$src2, DPR:$src3, nohash_imm:$lane),
885 IIC_VLD3lnu, "vld3", Dt,
886 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn$Rm",
887 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $Rn.addr = $wb",
889 let DecoderMethod = "DecodeVLD3LN";
892 def VLD3LNd8_UPD : VLD3LNWB<0b0010, {?,?,?,0}, "8"> {
893 let Inst{7-5} = lane{2-0};
895 def VLD3LNd16_UPD : VLD3LNWB<0b0110, {?,?,0,0}, "16"> {
896 let Inst{7-6} = lane{1-0};
898 def VLD3LNd32_UPD : VLD3LNWB<0b1010, {?,0,0,0}, "32"> {
899 let Inst{7} = lane{0};
902 def VLD3LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
903 def VLD3LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
904 def VLD3LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
906 def VLD3LNq16_UPD : VLD3LNWB<0b0110, {?,?,1,0}, "16"> {
907 let Inst{7-6} = lane{1-0};
909 def VLD3LNq32_UPD : VLD3LNWB<0b1010, {?,1,0,0}, "32"> {
910 let Inst{7} = lane{0};
913 def VLD3LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
914 def VLD3LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
916 // VLD4LN : Vector Load (single 4-element structure to one lane)
917 class VLD4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
918 : NLdStLn<1, 0b10, op11_8, op7_4,
919 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
920 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
921 nohash_imm:$lane), IIC_VLD4ln, "vld4", Dt,
922 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn",
923 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4", []> {
926 let DecoderMethod = "DecodeVLD4LN";
929 def VLD4LNd8 : VLD4LN<0b0011, {?,?,?,?}, "8"> {
930 let Inst{7-5} = lane{2-0};
932 def VLD4LNd16 : VLD4LN<0b0111, {?,?,0,?}, "16"> {
933 let Inst{7-6} = lane{1-0};
935 def VLD4LNd32 : VLD4LN<0b1011, {?,0,?,?}, "32"> {
936 let Inst{7} = lane{0};
940 def VLD4LNd8Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
941 def VLD4LNd16Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
942 def VLD4LNd32Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
944 // ...with double-spaced registers:
945 def VLD4LNq16 : VLD4LN<0b0111, {?,?,1,?}, "16"> {
946 let Inst{7-6} = lane{1-0};
948 def VLD4LNq32 : VLD4LN<0b1011, {?,1,?,?}, "32"> {
949 let Inst{7} = lane{0};
953 def VLD4LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>;
954 def VLD4LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>;
956 // ...with address register writeback:
957 class VLD4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
958 : NLdStLn<1, 0b10, op11_8, op7_4,
959 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
960 (ins addrmode6:$Rn, am6offset:$Rm,
961 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
962 IIC_VLD4lnu, "vld4", Dt,
963 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn$Rm",
964 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4, $Rn.addr = $wb",
967 let DecoderMethod = "DecodeVLD4LN" ;
970 def VLD4LNd8_UPD : VLD4LNWB<0b0011, {?,?,?,?}, "8"> {
971 let Inst{7-5} = lane{2-0};
973 def VLD4LNd16_UPD : VLD4LNWB<0b0111, {?,?,0,?}, "16"> {
974 let Inst{7-6} = lane{1-0};
976 def VLD4LNd32_UPD : VLD4LNWB<0b1011, {?,0,?,?}, "32"> {
977 let Inst{7} = lane{0};
981 def VLD4LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
982 def VLD4LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
983 def VLD4LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
985 def VLD4LNq16_UPD : VLD4LNWB<0b0111, {?,?,1,?}, "16"> {
986 let Inst{7-6} = lane{1-0};
988 def VLD4LNq32_UPD : VLD4LNWB<0b1011, {?,1,?,?}, "32"> {
989 let Inst{7} = lane{0};
993 def VLD4LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
994 def VLD4LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
996 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
998 // VLD1DUP : Vector Load (single element to all lanes)
999 class VLD1DUP<bits<4> op7_4, string Dt, ValueType Ty, PatFrag LoadOp>
1000 : NLdSt<1, 0b10, 0b1100, op7_4, (outs DPR:$Vd), (ins addrmode6dup:$Rn),
1001 IIC_VLD1dup, "vld1", Dt, "\\{$Vd[]\\}, $Rn", "",
1002 [(set DPR:$Vd, (Ty (NEONvdup (i32 (LoadOp addrmode6dup:$Rn)))))]> {
1004 let Inst{4} = Rn{4};
1005 let DecoderMethod = "DecodeVLD1DupInstruction";
1007 class VLD1QDUPPseudo<ValueType Ty, PatFrag LoadOp> : VLDQPseudo<IIC_VLD1dup> {
1008 let Pattern = [(set QPR:$dst,
1009 (Ty (NEONvdup (i32 (LoadOp addrmode6dup:$addr)))))];
1012 def VLD1DUPd8 : VLD1DUP<{0,0,0,?}, "8", v8i8, extloadi8>;
1013 def VLD1DUPd16 : VLD1DUP<{0,1,0,?}, "16", v4i16, extloadi16>;
1014 def VLD1DUPd32 : VLD1DUP<{1,0,0,?}, "32", v2i32, load>;
1016 def VLD1DUPq8Pseudo : VLD1QDUPPseudo<v16i8, extloadi8>;
1017 def VLD1DUPq16Pseudo : VLD1QDUPPseudo<v8i16, extloadi16>;
1018 def VLD1DUPq32Pseudo : VLD1QDUPPseudo<v4i32, load>;
1020 def : Pat<(v2f32 (NEONvdup (f32 (load addrmode6dup:$addr)))),
1021 (VLD1DUPd32 addrmode6:$addr)>;
1022 def : Pat<(v4f32 (NEONvdup (f32 (load addrmode6dup:$addr)))),
1023 (VLD1DUPq32Pseudo addrmode6:$addr)>;
1025 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
1027 class VLD1QDUP<bits<4> op7_4, string Dt>
1028 : NLdSt<1, 0b10, 0b1100, op7_4, (outs DPR:$Vd, DPR:$dst2),
1029 (ins addrmode6dup:$Rn), IIC_VLD1dup,
1030 "vld1", Dt, "\\{$Vd[], $dst2[]\\}, $Rn", "", []> {
1032 let Inst{4} = Rn{4};
1033 let DecoderMethod = "DecodeVLD1DupInstruction";
1036 def VLD1DUPq8 : VLD1QDUP<{0,0,1,0}, "8">;
1037 def VLD1DUPq16 : VLD1QDUP<{0,1,1,?}, "16">;
1038 def VLD1DUPq32 : VLD1QDUP<{1,0,1,?}, "32">;
1040 // ...with address register writeback:
1041 class VLD1DUPWB<bits<4> op7_4, string Dt>
1042 : NLdSt<1, 0b10, 0b1100, op7_4, (outs DPR:$Vd, GPR:$wb),
1043 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD1dupu,
1044 "vld1", Dt, "\\{$Vd[]\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
1045 let Inst{4} = Rn{4};
1046 let DecoderMethod = "DecodeVLD1DupInstruction";
1048 class VLD1QDUPWB<bits<4> op7_4, string Dt>
1049 : NLdSt<1, 0b10, 0b1100, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
1050 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD1dupu,
1051 "vld1", Dt, "\\{$Vd[], $dst2[]\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
1052 let Inst{4} = Rn{4};
1053 let DecoderMethod = "DecodeVLD1DupInstruction";
1056 def VLD1DUPd8_UPD : VLD1DUPWB<{0,0,0,0}, "8">;
1057 def VLD1DUPd16_UPD : VLD1DUPWB<{0,1,0,?}, "16">;
1058 def VLD1DUPd32_UPD : VLD1DUPWB<{1,0,0,?}, "32">;
1060 def VLD1DUPq8_UPD : VLD1QDUPWB<{0,0,1,0}, "8">;
1061 def VLD1DUPq16_UPD : VLD1QDUPWB<{0,1,1,?}, "16">;
1062 def VLD1DUPq32_UPD : VLD1QDUPWB<{1,0,1,?}, "32">;
1064 def VLD1DUPq8Pseudo_UPD : VLDQWBPseudo<IIC_VLD1dupu>;
1065 def VLD1DUPq16Pseudo_UPD : VLDQWBPseudo<IIC_VLD1dupu>;
1066 def VLD1DUPq32Pseudo_UPD : VLDQWBPseudo<IIC_VLD1dupu>;
1068 // VLD2DUP : Vector Load (single 2-element structure to all lanes)
1069 class VLD2DUP<bits<4> op7_4, string Dt>
1070 : NLdSt<1, 0b10, 0b1101, op7_4, (outs DPR:$Vd, DPR:$dst2),
1071 (ins addrmode6dup:$Rn), IIC_VLD2dup,
1072 "vld2", Dt, "\\{$Vd[], $dst2[]\\}, $Rn", "", []> {
1074 let Inst{4} = Rn{4};
1075 let DecoderMethod = "DecodeVLD2DupInstruction";
1078 def VLD2DUPd8 : VLD2DUP<{0,0,0,?}, "8">;
1079 def VLD2DUPd16 : VLD2DUP<{0,1,0,?}, "16">;
1080 def VLD2DUPd32 : VLD2DUP<{1,0,0,?}, "32">;
1082 def VLD2DUPd8Pseudo : VLDQPseudo<IIC_VLD2dup>;
1083 def VLD2DUPd16Pseudo : VLDQPseudo<IIC_VLD2dup>;
1084 def VLD2DUPd32Pseudo : VLDQPseudo<IIC_VLD2dup>;
1086 // ...with double-spaced registers (not used for codegen):
1087 def VLD2DUPd8x2 : VLD2DUP<{0,0,1,?}, "8">;
1088 def VLD2DUPd16x2 : VLD2DUP<{0,1,1,?}, "16">;
1089 def VLD2DUPd32x2 : VLD2DUP<{1,0,1,?}, "32">;
1091 // ...with address register writeback:
1092 class VLD2DUPWB<bits<4> op7_4, string Dt>
1093 : NLdSt<1, 0b10, 0b1101, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
1094 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD2dupu,
1095 "vld2", Dt, "\\{$Vd[], $dst2[]\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
1096 let Inst{4} = Rn{4};
1097 let DecoderMethod = "DecodeVLD2DupInstruction";
1100 def VLD2DUPd8_UPD : VLD2DUPWB<{0,0,0,0}, "8">;
1101 def VLD2DUPd16_UPD : VLD2DUPWB<{0,1,0,?}, "16">;
1102 def VLD2DUPd32_UPD : VLD2DUPWB<{1,0,0,?}, "32">;
1104 def VLD2DUPd8x2_UPD : VLD2DUPWB<{0,0,1,0}, "8">;
1105 def VLD2DUPd16x2_UPD : VLD2DUPWB<{0,1,1,?}, "16">;
1106 def VLD2DUPd32x2_UPD : VLD2DUPWB<{1,0,1,?}, "32">;
1108 def VLD2DUPd8Pseudo_UPD : VLDQWBPseudo<IIC_VLD2dupu>;
1109 def VLD2DUPd16Pseudo_UPD : VLDQWBPseudo<IIC_VLD2dupu>;
1110 def VLD2DUPd32Pseudo_UPD : VLDQWBPseudo<IIC_VLD2dupu>;
1112 // VLD3DUP : Vector Load (single 3-element structure to all lanes)
1113 class VLD3DUP<bits<4> op7_4, string Dt>
1114 : NLdSt<1, 0b10, 0b1110, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
1115 (ins addrmode6dup:$Rn), IIC_VLD3dup,
1116 "vld3", Dt, "\\{$Vd[], $dst2[], $dst3[]\\}, $Rn", "", []> {
1119 let DecoderMethod = "DecodeVLD3DupInstruction";
1122 def VLD3DUPd8 : VLD3DUP<{0,0,0,?}, "8">;
1123 def VLD3DUPd16 : VLD3DUP<{0,1,0,?}, "16">;
1124 def VLD3DUPd32 : VLD3DUP<{1,0,0,?}, "32">;
1126 def VLD3DUPd8Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1127 def VLD3DUPd16Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1128 def VLD3DUPd32Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1130 // ...with double-spaced registers (not used for codegen):
1131 def VLD3DUPd8x2 : VLD3DUP<{0,0,1,?}, "8">;
1132 def VLD3DUPd16x2 : VLD3DUP<{0,1,1,?}, "16">;
1133 def VLD3DUPd32x2 : VLD3DUP<{1,0,1,?}, "32">;
1135 // ...with address register writeback:
1136 class VLD3DUPWB<bits<4> op7_4, string Dt>
1137 : NLdSt<1, 0b10, 0b1110, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
1138 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD3dupu,
1139 "vld3", Dt, "\\{$Vd[], $dst2[], $dst3[]\\}, $Rn$Rm",
1140 "$Rn.addr = $wb", []> {
1142 let DecoderMethod = "DecodeVLD3DupInstruction";
1145 def VLD3DUPd8_UPD : VLD3DUPWB<{0,0,0,0}, "8">;
1146 def VLD3DUPd16_UPD : VLD3DUPWB<{0,1,0,?}, "16">;
1147 def VLD3DUPd32_UPD : VLD3DUPWB<{1,0,0,?}, "32">;
1149 def VLD3DUPd8x2_UPD : VLD3DUPWB<{0,0,1,0}, "8">;
1150 def VLD3DUPd16x2_UPD : VLD3DUPWB<{0,1,1,?}, "16">;
1151 def VLD3DUPd32x2_UPD : VLD3DUPWB<{1,0,1,?}, "32">;
1153 def VLD3DUPd8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1154 def VLD3DUPd16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1155 def VLD3DUPd32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1157 // VLD4DUP : Vector Load (single 4-element structure to all lanes)
1158 class VLD4DUP<bits<4> op7_4, string Dt>
1159 : NLdSt<1, 0b10, 0b1111, op7_4,
1160 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
1161 (ins addrmode6dup:$Rn), IIC_VLD4dup,
1162 "vld4", Dt, "\\{$Vd[], $dst2[], $dst3[], $dst4[]\\}, $Rn", "", []> {
1164 let Inst{4} = Rn{4};
1165 let DecoderMethod = "DecodeVLD4DupInstruction";
1168 def VLD4DUPd8 : VLD4DUP<{0,0,0,?}, "8">;
1169 def VLD4DUPd16 : VLD4DUP<{0,1,0,?}, "16">;
1170 def VLD4DUPd32 : VLD4DUP<{1,?,0,?}, "32"> { let Inst{6} = Rn{5}; }
1172 def VLD4DUPd8Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1173 def VLD4DUPd16Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1174 def VLD4DUPd32Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1176 // ...with double-spaced registers (not used for codegen):
1177 def VLD4DUPd8x2 : VLD4DUP<{0,0,1,?}, "8">;
1178 def VLD4DUPd16x2 : VLD4DUP<{0,1,1,?}, "16">;
1179 def VLD4DUPd32x2 : VLD4DUP<{1,?,1,?}, "32"> { let Inst{6} = Rn{5}; }
1181 // ...with address register writeback:
1182 class VLD4DUPWB<bits<4> op7_4, string Dt>
1183 : NLdSt<1, 0b10, 0b1111, op7_4,
1184 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
1185 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD4dupu,
1186 "vld4", Dt, "\\{$Vd[], $dst2[], $dst3[], $dst4[]\\}, $Rn$Rm",
1187 "$Rn.addr = $wb", []> {
1188 let Inst{4} = Rn{4};
1189 let DecoderMethod = "DecodeVLD4DupInstruction";
1192 def VLD4DUPd8_UPD : VLD4DUPWB<{0,0,0,0}, "8">;
1193 def VLD4DUPd16_UPD : VLD4DUPWB<{0,1,0,?}, "16">;
1194 def VLD4DUPd32_UPD : VLD4DUPWB<{1,?,0,?}, "32"> { let Inst{6} = Rn{5}; }
1196 def VLD4DUPd8x2_UPD : VLD4DUPWB<{0,0,1,0}, "8">;
1197 def VLD4DUPd16x2_UPD : VLD4DUPWB<{0,1,1,?}, "16">;
1198 def VLD4DUPd32x2_UPD : VLD4DUPWB<{1,?,1,?}, "32"> { let Inst{6} = Rn{5}; }
1200 def VLD4DUPd8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1201 def VLD4DUPd16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1202 def VLD4DUPd32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1204 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
1206 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
1208 // Classes for VST* pseudo-instructions with multi-register operands.
1209 // These are expanded to real instructions after register allocation.
1210 class VSTQPseudo<InstrItinClass itin>
1211 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src), itin, "">;
1212 class VSTQWBPseudo<InstrItinClass itin>
1213 : PseudoNLdSt<(outs GPR:$wb),
1214 (ins addrmode6:$addr, am6offset:$offset, QPR:$src), itin,
1215 "$addr.addr = $wb">;
1216 class VSTQWBfixedPseudo<InstrItinClass itin>
1217 : PseudoNLdSt<(outs GPR:$wb),
1218 (ins addrmode6:$addr, QPR:$src), itin,
1219 "$addr.addr = $wb">;
1220 class VSTQWBregisterPseudo<InstrItinClass itin>
1221 : PseudoNLdSt<(outs GPR:$wb),
1222 (ins addrmode6:$addr, rGPR:$offset, QPR:$src), itin,
1223 "$addr.addr = $wb">;
1224 class VSTQQPseudo<InstrItinClass itin>
1225 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src), itin, "">;
1226 class VSTQQWBPseudo<InstrItinClass itin>
1227 : PseudoNLdSt<(outs GPR:$wb),
1228 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src), itin,
1229 "$addr.addr = $wb">;
1230 class VSTQQQQPseudo<InstrItinClass itin>
1231 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src), itin, "">;
1232 class VSTQQQQWBPseudo<InstrItinClass itin>
1233 : PseudoNLdSt<(outs GPR:$wb),
1234 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
1235 "$addr.addr = $wb">;
1237 // VST1 : Vector Store (multiple single elements)
1238 class VST1D<bits<4> op7_4, string Dt>
1239 : NLdSt<0,0b00,0b0111,op7_4, (outs), (ins addrmode6:$Rn, VecListOneD:$Vd),
1240 IIC_VST1, "vst1", Dt, "$Vd, $Rn", "", []> {
1242 let Inst{4} = Rn{4};
1243 let DecoderMethod = "DecodeVSTInstruction";
1245 class VST1Q<bits<4> op7_4, string Dt>
1246 : NLdSt<0,0b00,0b1010,op7_4, (outs), (ins addrmode6:$Rn, VecListTwoD:$Vd),
1247 IIC_VST1x2, "vst1", Dt, "$Vd, $Rn", "", []> {
1249 let Inst{5-4} = Rn{5-4};
1250 let DecoderMethod = "DecodeVSTInstruction";
1253 def VST1d8 : VST1D<{0,0,0,?}, "8">;
1254 def VST1d16 : VST1D<{0,1,0,?}, "16">;
1255 def VST1d32 : VST1D<{1,0,0,?}, "32">;
1256 def VST1d64 : VST1D<{1,1,0,?}, "64">;
1258 def VST1q8 : VST1Q<{0,0,?,?}, "8">;
1259 def VST1q16 : VST1Q<{0,1,?,?}, "16">;
1260 def VST1q32 : VST1Q<{1,0,?,?}, "32">;
1261 def VST1q64 : VST1Q<{1,1,?,?}, "64">;
1263 def VST1q8Pseudo : VSTQPseudo<IIC_VST1x2>;
1264 def VST1q16Pseudo : VSTQPseudo<IIC_VST1x2>;
1265 def VST1q32Pseudo : VSTQPseudo<IIC_VST1x2>;
1266 def VST1q64Pseudo : VSTQPseudo<IIC_VST1x2>;
1268 // ...with address register writeback:
1269 multiclass VST1DWB<bits<4> op7_4, string Dt> {
1270 def _fixed : NLdSt<0,0b00, 0b0111,op7_4, (outs GPR:$wb),
1271 (ins addrmode6:$Rn, VecListOneD:$Vd), IIC_VLD1u,
1272 "vst1", Dt, "$Vd, $Rn!",
1273 "$Rn.addr = $wb", []> {
1274 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1275 let Inst{4} = Rn{4};
1276 let DecoderMethod = "DecodeVSTInstruction";
1277 let AsmMatchConverter = "cvtVSTwbFixed";
1279 def _register : NLdSt<0,0b00,0b0111,op7_4, (outs GPR:$wb),
1280 (ins addrmode6:$Rn, rGPR:$Rm, VecListOneD:$Vd),
1282 "vst1", Dt, "$Vd, $Rn, $Rm",
1283 "$Rn.addr = $wb", []> {
1284 let Inst{4} = Rn{4};
1285 let DecoderMethod = "DecodeVSTInstruction";
1286 let AsmMatchConverter = "cvtVSTwbRegister";
1289 multiclass VST1QWB<bits<4> op7_4, string Dt> {
1290 def _fixed : NLdSt<0,0b00,0b1010,op7_4, (outs GPR:$wb),
1291 (ins addrmode6:$Rn, VecListTwoD:$Vd), IIC_VLD1x2u,
1292 "vst1", Dt, "$Vd, $Rn!",
1293 "$Rn.addr = $wb", []> {
1294 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1295 let Inst{5-4} = Rn{5-4};
1296 let DecoderMethod = "DecodeVSTInstruction";
1297 let AsmMatchConverter = "cvtVSTwbFixed";
1299 def _register : NLdSt<0,0b00,0b1010,op7_4, (outs GPR:$wb),
1300 (ins addrmode6:$Rn, rGPR:$Rm, VecListTwoD:$Vd),
1302 "vst1", Dt, "$Vd, $Rn, $Rm",
1303 "$Rn.addr = $wb", []> {
1304 let Inst{5-4} = Rn{5-4};
1305 let DecoderMethod = "DecodeVSTInstruction";
1306 let AsmMatchConverter = "cvtVSTwbRegister";
1310 defm VST1d8wb : VST1DWB<{0,0,0,?}, "8">;
1311 defm VST1d16wb : VST1DWB<{0,1,0,?}, "16">;
1312 defm VST1d32wb : VST1DWB<{1,0,0,?}, "32">;
1313 defm VST1d64wb : VST1DWB<{1,1,0,?}, "64">;
1315 defm VST1q8wb : VST1QWB<{0,0,?,?}, "8">;
1316 defm VST1q16wb : VST1QWB<{0,1,?,?}, "16">;
1317 defm VST1q32wb : VST1QWB<{1,0,?,?}, "32">;
1318 defm VST1q64wb : VST1QWB<{1,1,?,?}, "64">;
1320 def VST1q8PseudoWB_fixed : VSTQWBfixedPseudo<IIC_VST1x2u>;
1321 def VST1q16PseudoWB_fixed : VSTQWBfixedPseudo<IIC_VST1x2u>;
1322 def VST1q32PseudoWB_fixed : VSTQWBfixedPseudo<IIC_VST1x2u>;
1323 def VST1q64PseudoWB_fixed : VSTQWBfixedPseudo<IIC_VST1x2u>;
1324 def VST1q8PseudoWB_register : VSTQWBregisterPseudo<IIC_VST1x2u>;
1325 def VST1q16PseudoWB_register : VSTQWBregisterPseudo<IIC_VST1x2u>;
1326 def VST1q32PseudoWB_register : VSTQWBregisterPseudo<IIC_VST1x2u>;
1327 def VST1q64PseudoWB_register : VSTQWBregisterPseudo<IIC_VST1x2u>;
1329 // ...with 3 registers
1330 class VST1D3<bits<4> op7_4, string Dt>
1331 : NLdSt<0, 0b00, 0b0110, op7_4, (outs),
1332 (ins addrmode6:$Rn, VecListThreeD:$Vd),
1333 IIC_VST1x3, "vst1", Dt, "$Vd, $Rn", "", []> {
1335 let Inst{4} = Rn{4};
1336 let DecoderMethod = "DecodeVSTInstruction";
1338 multiclass VST1D3WB<bits<4> op7_4, string Dt> {
1339 def _fixed : NLdSt<0,0b00,0b0110,op7_4, (outs GPR:$wb),
1340 (ins addrmode6:$Rn, VecListThreeD:$Vd), IIC_VLD1x3u,
1341 "vst1", Dt, "$Vd, $Rn!",
1342 "$Rn.addr = $wb", []> {
1343 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1344 let Inst{5-4} = Rn{5-4};
1345 let DecoderMethod = "DecodeVSTInstruction";
1346 let AsmMatchConverter = "cvtVSTwbFixed";
1348 def _register : NLdSt<0,0b00,0b0110,op7_4, (outs GPR:$wb),
1349 (ins addrmode6:$Rn, rGPR:$Rm, VecListThreeD:$Vd),
1351 "vst1", Dt, "$Vd, $Rn, $Rm",
1352 "$Rn.addr = $wb", []> {
1353 let Inst{5-4} = Rn{5-4};
1354 let DecoderMethod = "DecodeVSTInstruction";
1355 let AsmMatchConverter = "cvtVSTwbRegister";
1359 def VST1d8T : VST1D3<{0,0,0,?}, "8">;
1360 def VST1d16T : VST1D3<{0,1,0,?}, "16">;
1361 def VST1d32T : VST1D3<{1,0,0,?}, "32">;
1362 def VST1d64T : VST1D3<{1,1,0,?}, "64">;
1364 defm VST1d8Twb : VST1D3WB<{0,0,0,?}, "8">;
1365 defm VST1d16Twb : VST1D3WB<{0,1,0,?}, "16">;
1366 defm VST1d32Twb : VST1D3WB<{1,0,0,?}, "32">;
1367 defm VST1d64Twb : VST1D3WB<{1,1,0,?}, "64">;
1369 def VST1d64TPseudo : VSTQQPseudo<IIC_VST1x3>;
1370 def VST1d64TPseudoWB_fixed : VSTQQWBPseudo<IIC_VST1x3u>;
1371 def VST1d64TPseudoWB_register : VSTQQWBPseudo<IIC_VST1x3u>;
1373 // ...with 4 registers
1374 class VST1D4<bits<4> op7_4, string Dt>
1375 : NLdSt<0, 0b00, 0b0010, op7_4, (outs),
1376 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
1377 IIC_VST1x4, "vst1", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn", "",
1380 let Inst{5-4} = Rn{5-4};
1381 let DecoderMethod = "DecodeVSTInstruction";
1383 class VST1D4WB<bits<4> op7_4, string Dt>
1384 : NLdSt<0, 0b00, 0b0010, op7_4, (outs GPR:$wb),
1385 (ins addrmode6:$Rn, am6offset:$Rm,
1386 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST1x4u,
1387 "vst1", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
1388 "$Rn.addr = $wb", []> {
1389 let Inst{5-4} = Rn{5-4};
1390 let DecoderMethod = "DecodeVSTInstruction";
1393 def VST1d8Q : VST1D4<{0,0,?,?}, "8">;
1394 def VST1d16Q : VST1D4<{0,1,?,?}, "16">;
1395 def VST1d32Q : VST1D4<{1,0,?,?}, "32">;
1396 def VST1d64Q : VST1D4<{1,1,?,?}, "64">;
1398 def VST1d8Q_UPD : VST1D4WB<{0,0,?,?}, "8">;
1399 def VST1d16Q_UPD : VST1D4WB<{0,1,?,?}, "16">;
1400 def VST1d32Q_UPD : VST1D4WB<{1,0,?,?}, "32">;
1401 def VST1d64Q_UPD : VST1D4WB<{1,1,?,?}, "64">;
1403 def VST1d64QPseudo : VSTQQPseudo<IIC_VST1x4>;
1404 def VST1d64QPseudo_UPD : VSTQQWBPseudo<IIC_VST1x4u>;
1406 // VST2 : Vector Store (multiple 2-element structures)
1407 class VST2D<bits<4> op11_8, bits<4> op7_4, string Dt>
1408 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1409 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2),
1410 IIC_VST2, "vst2", Dt, "\\{$Vd, $src2\\}, $Rn", "", []> {
1412 let Inst{5-4} = Rn{5-4};
1413 let DecoderMethod = "DecodeVSTInstruction";
1415 class VST2Q<bits<4> op7_4, string Dt>
1416 : NLdSt<0, 0b00, 0b0011, op7_4, (outs),
1417 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
1418 IIC_VST2x2, "vst2", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn",
1421 let Inst{5-4} = Rn{5-4};
1422 let DecoderMethod = "DecodeVSTInstruction";
1425 def VST2d8 : VST2D<0b1000, {0,0,?,?}, "8">;
1426 def VST2d16 : VST2D<0b1000, {0,1,?,?}, "16">;
1427 def VST2d32 : VST2D<0b1000, {1,0,?,?}, "32">;
1429 def VST2q8 : VST2Q<{0,0,?,?}, "8">;
1430 def VST2q16 : VST2Q<{0,1,?,?}, "16">;
1431 def VST2q32 : VST2Q<{1,0,?,?}, "32">;
1433 def VST2d8Pseudo : VSTQPseudo<IIC_VST2>;
1434 def VST2d16Pseudo : VSTQPseudo<IIC_VST2>;
1435 def VST2d32Pseudo : VSTQPseudo<IIC_VST2>;
1437 def VST2q8Pseudo : VSTQQPseudo<IIC_VST2x2>;
1438 def VST2q16Pseudo : VSTQQPseudo<IIC_VST2x2>;
1439 def VST2q32Pseudo : VSTQQPseudo<IIC_VST2x2>;
1441 // ...with address register writeback:
1442 class VST2DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1443 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1444 (ins addrmode6:$Rn, am6offset:$Rm, DPR:$Vd, DPR:$src2),
1445 IIC_VST2u, "vst2", Dt, "\\{$Vd, $src2\\}, $Rn$Rm",
1446 "$Rn.addr = $wb", []> {
1447 let Inst{5-4} = Rn{5-4};
1448 let DecoderMethod = "DecodeVSTInstruction";
1450 class VST2QWB<bits<4> op7_4, string Dt>
1451 : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
1452 (ins addrmode6:$Rn, am6offset:$Rm,
1453 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST2x2u,
1454 "vst2", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
1455 "$Rn.addr = $wb", []> {
1456 let Inst{5-4} = Rn{5-4};
1457 let DecoderMethod = "DecodeVSTInstruction";
1460 def VST2d8_UPD : VST2DWB<0b1000, {0,0,?,?}, "8">;
1461 def VST2d16_UPD : VST2DWB<0b1000, {0,1,?,?}, "16">;
1462 def VST2d32_UPD : VST2DWB<0b1000, {1,0,?,?}, "32">;
1464 def VST2q8_UPD : VST2QWB<{0,0,?,?}, "8">;
1465 def VST2q16_UPD : VST2QWB<{0,1,?,?}, "16">;
1466 def VST2q32_UPD : VST2QWB<{1,0,?,?}, "32">;
1468 def VST2d8Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
1469 def VST2d16Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
1470 def VST2d32Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
1472 def VST2q8Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
1473 def VST2q16Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
1474 def VST2q32Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
1476 // ...with double-spaced registers
1477 def VST2b8 : VST2D<0b1001, {0,0,?,?}, "8">;
1478 def VST2b16 : VST2D<0b1001, {0,1,?,?}, "16">;
1479 def VST2b32 : VST2D<0b1001, {1,0,?,?}, "32">;
1480 def VST2b8_UPD : VST2DWB<0b1001, {0,0,?,?}, "8">;
1481 def VST2b16_UPD : VST2DWB<0b1001, {0,1,?,?}, "16">;
1482 def VST2b32_UPD : VST2DWB<0b1001, {1,0,?,?}, "32">;
1484 // VST3 : Vector Store (multiple 3-element structures)
1485 class VST3D<bits<4> op11_8, bits<4> op7_4, string Dt>
1486 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1487 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3,
1488 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn", "", []> {
1490 let Inst{4} = Rn{4};
1491 let DecoderMethod = "DecodeVSTInstruction";
1494 def VST3d8 : VST3D<0b0100, {0,0,0,?}, "8">;
1495 def VST3d16 : VST3D<0b0100, {0,1,0,?}, "16">;
1496 def VST3d32 : VST3D<0b0100, {1,0,0,?}, "32">;
1498 def VST3d8Pseudo : VSTQQPseudo<IIC_VST3>;
1499 def VST3d16Pseudo : VSTQQPseudo<IIC_VST3>;
1500 def VST3d32Pseudo : VSTQQPseudo<IIC_VST3>;
1502 // ...with address register writeback:
1503 class VST3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1504 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1505 (ins addrmode6:$Rn, am6offset:$Rm,
1506 DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3u,
1507 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn$Rm",
1508 "$Rn.addr = $wb", []> {
1509 let Inst{4} = Rn{4};
1510 let DecoderMethod = "DecodeVSTInstruction";
1513 def VST3d8_UPD : VST3DWB<0b0100, {0,0,0,?}, "8">;
1514 def VST3d16_UPD : VST3DWB<0b0100, {0,1,0,?}, "16">;
1515 def VST3d32_UPD : VST3DWB<0b0100, {1,0,0,?}, "32">;
1517 def VST3d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1518 def VST3d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1519 def VST3d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1521 // ...with double-spaced registers:
1522 def VST3q8 : VST3D<0b0101, {0,0,0,?}, "8">;
1523 def VST3q16 : VST3D<0b0101, {0,1,0,?}, "16">;
1524 def VST3q32 : VST3D<0b0101, {1,0,0,?}, "32">;
1525 def VST3q8_UPD : VST3DWB<0b0101, {0,0,0,?}, "8">;
1526 def VST3q16_UPD : VST3DWB<0b0101, {0,1,0,?}, "16">;
1527 def VST3q32_UPD : VST3DWB<0b0101, {1,0,0,?}, "32">;
1529 def VST3q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1530 def VST3q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1531 def VST3q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1533 // ...alternate versions to be allocated odd register numbers:
1534 def VST3q8oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1535 def VST3q16oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1536 def VST3q32oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1538 def VST3q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1539 def VST3q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1540 def VST3q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1542 // VST4 : Vector Store (multiple 4-element structures)
1543 class VST4D<bits<4> op11_8, bits<4> op7_4, string Dt>
1544 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1545 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
1546 IIC_VST4, "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn",
1549 let Inst{5-4} = Rn{5-4};
1550 let DecoderMethod = "DecodeVSTInstruction";
1553 def VST4d8 : VST4D<0b0000, {0,0,?,?}, "8">;
1554 def VST4d16 : VST4D<0b0000, {0,1,?,?}, "16">;
1555 def VST4d32 : VST4D<0b0000, {1,0,?,?}, "32">;
1557 def VST4d8Pseudo : VSTQQPseudo<IIC_VST4>;
1558 def VST4d16Pseudo : VSTQQPseudo<IIC_VST4>;
1559 def VST4d32Pseudo : VSTQQPseudo<IIC_VST4>;
1561 // ...with address register writeback:
1562 class VST4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1563 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1564 (ins addrmode6:$Rn, am6offset:$Rm,
1565 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST4u,
1566 "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
1567 "$Rn.addr = $wb", []> {
1568 let Inst{5-4} = Rn{5-4};
1569 let DecoderMethod = "DecodeVSTInstruction";
1572 def VST4d8_UPD : VST4DWB<0b0000, {0,0,?,?}, "8">;
1573 def VST4d16_UPD : VST4DWB<0b0000, {0,1,?,?}, "16">;
1574 def VST4d32_UPD : VST4DWB<0b0000, {1,0,?,?}, "32">;
1576 def VST4d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1577 def VST4d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1578 def VST4d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1580 // ...with double-spaced registers:
1581 def VST4q8 : VST4D<0b0001, {0,0,?,?}, "8">;
1582 def VST4q16 : VST4D<0b0001, {0,1,?,?}, "16">;
1583 def VST4q32 : VST4D<0b0001, {1,0,?,?}, "32">;
1584 def VST4q8_UPD : VST4DWB<0b0001, {0,0,?,?}, "8">;
1585 def VST4q16_UPD : VST4DWB<0b0001, {0,1,?,?}, "16">;
1586 def VST4q32_UPD : VST4DWB<0b0001, {1,0,?,?}, "32">;
1588 def VST4q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1589 def VST4q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1590 def VST4q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1592 // ...alternate versions to be allocated odd register numbers:
1593 def VST4q8oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1594 def VST4q16oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1595 def VST4q32oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1597 def VST4q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1598 def VST4q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1599 def VST4q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1601 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
1603 // Classes for VST*LN pseudo-instructions with multi-register operands.
1604 // These are expanded to real instructions after register allocation.
1605 class VSTQLNPseudo<InstrItinClass itin>
1606 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
1608 class VSTQLNWBPseudo<InstrItinClass itin>
1609 : PseudoNLdSt<(outs GPR:$wb),
1610 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
1611 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1612 class VSTQQLNPseudo<InstrItinClass itin>
1613 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
1615 class VSTQQLNWBPseudo<InstrItinClass itin>
1616 : PseudoNLdSt<(outs GPR:$wb),
1617 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
1618 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1619 class VSTQQQQLNPseudo<InstrItinClass itin>
1620 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
1622 class VSTQQQQLNWBPseudo<InstrItinClass itin>
1623 : PseudoNLdSt<(outs GPR:$wb),
1624 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
1625 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1627 // VST1LN : Vector Store (single element from one lane)
1628 class VST1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1629 PatFrag StoreOp, SDNode ExtractOp>
1630 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1631 (ins addrmode6:$Rn, DPR:$Vd, nohash_imm:$lane),
1632 IIC_VST1ln, "vst1", Dt, "\\{$Vd[$lane]\\}, $Rn", "",
1633 [(StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane), addrmode6:$Rn)]> {
1635 let DecoderMethod = "DecodeVST1LN";
1637 class VST1LN32<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1638 PatFrag StoreOp, SDNode ExtractOp>
1639 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1640 (ins addrmode6oneL32:$Rn, DPR:$Vd, nohash_imm:$lane),
1641 IIC_VST1ln, "vst1", Dt, "\\{$Vd[$lane]\\}, $Rn", "",
1642 [(StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane), addrmode6oneL32:$Rn)]>{
1644 let DecoderMethod = "DecodeVST1LN";
1646 class VST1QLNPseudo<ValueType Ty, PatFrag StoreOp, SDNode ExtractOp>
1647 : VSTQLNPseudo<IIC_VST1ln> {
1648 let Pattern = [(StoreOp (ExtractOp (Ty QPR:$src), imm:$lane),
1652 def VST1LNd8 : VST1LN<0b0000, {?,?,?,0}, "8", v8i8, truncstorei8,
1654 let Inst{7-5} = lane{2-0};
1656 def VST1LNd16 : VST1LN<0b0100, {?,?,0,?}, "16", v4i16, truncstorei16,
1658 let Inst{7-6} = lane{1-0};
1659 let Inst{4} = Rn{5};
1662 def VST1LNd32 : VST1LN32<0b1000, {?,0,?,?}, "32", v2i32, store, extractelt> {
1663 let Inst{7} = lane{0};
1664 let Inst{5-4} = Rn{5-4};
1667 def VST1LNq8Pseudo : VST1QLNPseudo<v16i8, truncstorei8, NEONvgetlaneu>;
1668 def VST1LNq16Pseudo : VST1QLNPseudo<v8i16, truncstorei16, NEONvgetlaneu>;
1669 def VST1LNq32Pseudo : VST1QLNPseudo<v4i32, store, extractelt>;
1671 def : Pat<(store (extractelt (v2f32 DPR:$src), imm:$lane), addrmode6:$addr),
1672 (VST1LNd32 addrmode6:$addr, DPR:$src, imm:$lane)>;
1673 def : Pat<(store (extractelt (v4f32 QPR:$src), imm:$lane), addrmode6:$addr),
1674 (VST1LNq32Pseudo addrmode6:$addr, QPR:$src, imm:$lane)>;
1676 // ...with address register writeback:
1677 class VST1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1678 PatFrag StoreOp, SDNode ExtractOp>
1679 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1680 (ins addrmode6:$Rn, am6offset:$Rm,
1681 DPR:$Vd, nohash_imm:$lane), IIC_VST1lnu, "vst1", Dt,
1682 "\\{$Vd[$lane]\\}, $Rn$Rm",
1684 [(set GPR:$wb, (StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane),
1685 addrmode6:$Rn, am6offset:$Rm))]> {
1686 let DecoderMethod = "DecodeVST1LN";
1688 class VST1QLNWBPseudo<ValueType Ty, PatFrag StoreOp, SDNode ExtractOp>
1689 : VSTQLNWBPseudo<IIC_VST1lnu> {
1690 let Pattern = [(set GPR:$wb, (StoreOp (ExtractOp (Ty QPR:$src), imm:$lane),
1691 addrmode6:$addr, am6offset:$offset))];
1694 def VST1LNd8_UPD : VST1LNWB<0b0000, {?,?,?,0}, "8", v8i8, post_truncsti8,
1696 let Inst{7-5} = lane{2-0};
1698 def VST1LNd16_UPD : VST1LNWB<0b0100, {?,?,0,?}, "16", v4i16, post_truncsti16,
1700 let Inst{7-6} = lane{1-0};
1701 let Inst{4} = Rn{5};
1703 def VST1LNd32_UPD : VST1LNWB<0b1000, {?,0,?,?}, "32", v2i32, post_store,
1705 let Inst{7} = lane{0};
1706 let Inst{5-4} = Rn{5-4};
1709 def VST1LNq8Pseudo_UPD : VST1QLNWBPseudo<v16i8, post_truncsti8, NEONvgetlaneu>;
1710 def VST1LNq16Pseudo_UPD : VST1QLNWBPseudo<v8i16, post_truncsti16,NEONvgetlaneu>;
1711 def VST1LNq32Pseudo_UPD : VST1QLNWBPseudo<v4i32, post_store, extractelt>;
1713 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
1715 // VST2LN : Vector Store (single 2-element structure from one lane)
1716 class VST2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1717 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1718 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, nohash_imm:$lane),
1719 IIC_VST2ln, "vst2", Dt, "\\{$Vd[$lane], $src2[$lane]\\}, $Rn",
1722 let Inst{4} = Rn{4};
1723 let DecoderMethod = "DecodeVST2LN";
1726 def VST2LNd8 : VST2LN<0b0001, {?,?,?,?}, "8"> {
1727 let Inst{7-5} = lane{2-0};
1729 def VST2LNd16 : VST2LN<0b0101, {?,?,0,?}, "16"> {
1730 let Inst{7-6} = lane{1-0};
1732 def VST2LNd32 : VST2LN<0b1001, {?,0,0,?}, "32"> {
1733 let Inst{7} = lane{0};
1736 def VST2LNd8Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1737 def VST2LNd16Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1738 def VST2LNd32Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1740 // ...with double-spaced registers:
1741 def VST2LNq16 : VST2LN<0b0101, {?,?,1,?}, "16"> {
1742 let Inst{7-6} = lane{1-0};
1743 let Inst{4} = Rn{4};
1745 def VST2LNq32 : VST2LN<0b1001, {?,1,0,?}, "32"> {
1746 let Inst{7} = lane{0};
1747 let Inst{4} = Rn{4};
1750 def VST2LNq16Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
1751 def VST2LNq32Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
1753 // ...with address register writeback:
1754 class VST2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1755 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1756 (ins addrmode6:$addr, am6offset:$offset,
1757 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VST2lnu, "vst2", Dt,
1758 "\\{$src1[$lane], $src2[$lane]\\}, $addr$offset",
1759 "$addr.addr = $wb", []> {
1760 let Inst{4} = Rn{4};
1761 let DecoderMethod = "DecodeVST2LN";
1764 def VST2LNd8_UPD : VST2LNWB<0b0001, {?,?,?,?}, "8"> {
1765 let Inst{7-5} = lane{2-0};
1767 def VST2LNd16_UPD : VST2LNWB<0b0101, {?,?,0,?}, "16"> {
1768 let Inst{7-6} = lane{1-0};
1770 def VST2LNd32_UPD : VST2LNWB<0b1001, {?,0,0,?}, "32"> {
1771 let Inst{7} = lane{0};
1774 def VST2LNd8Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1775 def VST2LNd16Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1776 def VST2LNd32Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1778 def VST2LNq16_UPD : VST2LNWB<0b0101, {?,?,1,?}, "16"> {
1779 let Inst{7-6} = lane{1-0};
1781 def VST2LNq32_UPD : VST2LNWB<0b1001, {?,1,0,?}, "32"> {
1782 let Inst{7} = lane{0};
1785 def VST2LNq16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
1786 def VST2LNq32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
1788 // VST3LN : Vector Store (single 3-element structure from one lane)
1789 class VST3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1790 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1791 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3,
1792 nohash_imm:$lane), IIC_VST3ln, "vst3", Dt,
1793 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn", "", []> {
1795 let DecoderMethod = "DecodeVST3LN";
1798 def VST3LNd8 : VST3LN<0b0010, {?,?,?,0}, "8"> {
1799 let Inst{7-5} = lane{2-0};
1801 def VST3LNd16 : VST3LN<0b0110, {?,?,0,0}, "16"> {
1802 let Inst{7-6} = lane{1-0};
1804 def VST3LNd32 : VST3LN<0b1010, {?,0,0,0}, "32"> {
1805 let Inst{7} = lane{0};
1808 def VST3LNd8Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1809 def VST3LNd16Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1810 def VST3LNd32Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1812 // ...with double-spaced registers:
1813 def VST3LNq16 : VST3LN<0b0110, {?,?,1,0}, "16"> {
1814 let Inst{7-6} = lane{1-0};
1816 def VST3LNq32 : VST3LN<0b1010, {?,1,0,0}, "32"> {
1817 let Inst{7} = lane{0};
1820 def VST3LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
1821 def VST3LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
1823 // ...with address register writeback:
1824 class VST3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1825 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1826 (ins addrmode6:$Rn, am6offset:$Rm,
1827 DPR:$Vd, DPR:$src2, DPR:$src3, nohash_imm:$lane),
1828 IIC_VST3lnu, "vst3", Dt,
1829 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn$Rm",
1830 "$Rn.addr = $wb", []> {
1831 let DecoderMethod = "DecodeVST3LN";
1834 def VST3LNd8_UPD : VST3LNWB<0b0010, {?,?,?,0}, "8"> {
1835 let Inst{7-5} = lane{2-0};
1837 def VST3LNd16_UPD : VST3LNWB<0b0110, {?,?,0,0}, "16"> {
1838 let Inst{7-6} = lane{1-0};
1840 def VST3LNd32_UPD : VST3LNWB<0b1010, {?,0,0,0}, "32"> {
1841 let Inst{7} = lane{0};
1844 def VST3LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1845 def VST3LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1846 def VST3LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1848 def VST3LNq16_UPD : VST3LNWB<0b0110, {?,?,1,0}, "16"> {
1849 let Inst{7-6} = lane{1-0};
1851 def VST3LNq32_UPD : VST3LNWB<0b1010, {?,1,0,0}, "32"> {
1852 let Inst{7} = lane{0};
1855 def VST3LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
1856 def VST3LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
1858 // VST4LN : Vector Store (single 4-element structure from one lane)
1859 class VST4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1860 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1861 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4,
1862 nohash_imm:$lane), IIC_VST4ln, "vst4", Dt,
1863 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn",
1866 let Inst{4} = Rn{4};
1867 let DecoderMethod = "DecodeVST4LN";
1870 def VST4LNd8 : VST4LN<0b0011, {?,?,?,?}, "8"> {
1871 let Inst{7-5} = lane{2-0};
1873 def VST4LNd16 : VST4LN<0b0111, {?,?,0,?}, "16"> {
1874 let Inst{7-6} = lane{1-0};
1876 def VST4LNd32 : VST4LN<0b1011, {?,0,?,?}, "32"> {
1877 let Inst{7} = lane{0};
1878 let Inst{5} = Rn{5};
1881 def VST4LNd8Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1882 def VST4LNd16Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1883 def VST4LNd32Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1885 // ...with double-spaced registers:
1886 def VST4LNq16 : VST4LN<0b0111, {?,?,1,?}, "16"> {
1887 let Inst{7-6} = lane{1-0};
1889 def VST4LNq32 : VST4LN<0b1011, {?,1,?,?}, "32"> {
1890 let Inst{7} = lane{0};
1891 let Inst{5} = Rn{5};
1894 def VST4LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
1895 def VST4LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
1897 // ...with address register writeback:
1898 class VST4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1899 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1900 (ins addrmode6:$Rn, am6offset:$Rm,
1901 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
1902 IIC_VST4lnu, "vst4", Dt,
1903 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn$Rm",
1904 "$Rn.addr = $wb", []> {
1905 let Inst{4} = Rn{4};
1906 let DecoderMethod = "DecodeVST4LN";
1909 def VST4LNd8_UPD : VST4LNWB<0b0011, {?,?,?,?}, "8"> {
1910 let Inst{7-5} = lane{2-0};
1912 def VST4LNd16_UPD : VST4LNWB<0b0111, {?,?,0,?}, "16"> {
1913 let Inst{7-6} = lane{1-0};
1915 def VST4LNd32_UPD : VST4LNWB<0b1011, {?,0,?,?}, "32"> {
1916 let Inst{7} = lane{0};
1917 let Inst{5} = Rn{5};
1920 def VST4LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
1921 def VST4LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
1922 def VST4LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
1924 def VST4LNq16_UPD : VST4LNWB<0b0111, {?,?,1,?}, "16"> {
1925 let Inst{7-6} = lane{1-0};
1927 def VST4LNq32_UPD : VST4LNWB<0b1011, {?,1,?,?}, "32"> {
1928 let Inst{7} = lane{0};
1929 let Inst{5} = Rn{5};
1932 def VST4LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
1933 def VST4LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
1935 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
1938 //===----------------------------------------------------------------------===//
1939 // NEON pattern fragments
1940 //===----------------------------------------------------------------------===//
1942 // Extract D sub-registers of Q registers.
1943 def DSubReg_i8_reg : SDNodeXForm<imm, [{
1944 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1945 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/8, MVT::i32);
1947 def DSubReg_i16_reg : SDNodeXForm<imm, [{
1948 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1949 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/4, MVT::i32);
1951 def DSubReg_i32_reg : SDNodeXForm<imm, [{
1952 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1953 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/2, MVT::i32);
1955 def DSubReg_f64_reg : SDNodeXForm<imm, [{
1956 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1957 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue(), MVT::i32);
1960 // Extract S sub-registers of Q/D registers.
1961 def SSubReg_f32_reg : SDNodeXForm<imm, [{
1962 assert(ARM::ssub_3 == ARM::ssub_0+3 && "Unexpected subreg numbering");
1963 return CurDAG->getTargetConstant(ARM::ssub_0 + N->getZExtValue(), MVT::i32);
1966 // Translate lane numbers from Q registers to D subregs.
1967 def SubReg_i8_lane : SDNodeXForm<imm, [{
1968 return CurDAG->getTargetConstant(N->getZExtValue() & 7, MVT::i32);
1970 def SubReg_i16_lane : SDNodeXForm<imm, [{
1971 return CurDAG->getTargetConstant(N->getZExtValue() & 3, MVT::i32);
1973 def SubReg_i32_lane : SDNodeXForm<imm, [{
1974 return CurDAG->getTargetConstant(N->getZExtValue() & 1, MVT::i32);
1977 //===----------------------------------------------------------------------===//
1978 // Instruction Classes
1979 //===----------------------------------------------------------------------===//
1981 // Basic 2-register operations: double- and quad-register.
1982 class N2VD<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1983 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
1984 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
1985 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
1986 (ins DPR:$Vm), IIC_VUNAD, OpcodeStr, Dt,"$Vd, $Vm", "",
1987 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vm))))]>;
1988 class N2VQ<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1989 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
1990 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
1991 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
1992 (ins QPR:$Vm), IIC_VUNAQ, OpcodeStr, Dt,"$Vd, $Vm", "",
1993 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vm))))]>;
1995 // Basic 2-register intrinsics, both double- and quad-register.
1996 class N2VDInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1997 bits<2> op17_16, bits<5> op11_7, bit op4,
1998 InstrItinClass itin, string OpcodeStr, string Dt,
1999 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2000 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
2001 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2002 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm))))]>;
2003 class N2VQInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2004 bits<2> op17_16, bits<5> op11_7, bit op4,
2005 InstrItinClass itin, string OpcodeStr, string Dt,
2006 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2007 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
2008 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2009 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm))))]>;
2011 // Narrow 2-register operations.
2012 class N2VN<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2013 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2014 InstrItinClass itin, string OpcodeStr, string Dt,
2015 ValueType TyD, ValueType TyQ, SDNode OpNode>
2016 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$Vd),
2017 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2018 [(set DPR:$Vd, (TyD (OpNode (TyQ QPR:$Vm))))]>;
2020 // Narrow 2-register intrinsics.
2021 class N2VNInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2022 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2023 InstrItinClass itin, string OpcodeStr, string Dt,
2024 ValueType TyD, ValueType TyQ, Intrinsic IntOp>
2025 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$Vd),
2026 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2027 [(set DPR:$Vd, (TyD (IntOp (TyQ QPR:$Vm))))]>;
2029 // Long 2-register operations (currently only used for VMOVL).
2030 class N2VL<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2031 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2032 InstrItinClass itin, string OpcodeStr, string Dt,
2033 ValueType TyQ, ValueType TyD, SDNode OpNode>
2034 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$Vd),
2035 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2036 [(set QPR:$Vd, (TyQ (OpNode (TyD DPR:$Vm))))]>;
2038 // Long 2-register intrinsics.
2039 class N2VLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2040 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2041 InstrItinClass itin, string OpcodeStr, string Dt,
2042 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
2043 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$Vd),
2044 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2045 [(set QPR:$Vd, (TyQ (IntOp (TyD DPR:$Vm))))]>;
2047 // 2-register shuffles (VTRN/VZIP/VUZP), both double- and quad-register.
2048 class N2VDShuffle<bits<2> op19_18, bits<5> op11_7, string OpcodeStr, string Dt>
2049 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 0, 0, (outs DPR:$Vd, DPR:$Vm),
2050 (ins DPR:$src1, DPR:$src2), IIC_VPERMD,
2051 OpcodeStr, Dt, "$Vd, $Vm",
2052 "$src1 = $Vd, $src2 = $Vm", []>;
2053 class N2VQShuffle<bits<2> op19_18, bits<5> op11_7,
2054 InstrItinClass itin, string OpcodeStr, string Dt>
2055 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 1, 0, (outs QPR:$Vd, QPR:$Vm),
2056 (ins QPR:$src1, QPR:$src2), itin, OpcodeStr, Dt, "$Vd, $Vm",
2057 "$src1 = $Vd, $src2 = $Vm", []>;
2059 // Basic 3-register operations: double- and quad-register.
2060 class N3VD<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2061 InstrItinClass itin, string OpcodeStr, string Dt,
2062 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
2063 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2064 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2065 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2066 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
2067 let isCommutable = Commutable;
2069 // Same as N3VD but no data type.
2070 class N3VDX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2071 InstrItinClass itin, string OpcodeStr,
2072 ValueType ResTy, ValueType OpTy,
2073 SDNode OpNode, bit Commutable>
2074 : N3VX<op24, op23, op21_20, op11_8, 0, op4,
2075 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2076 OpcodeStr, "$Vd, $Vn, $Vm", "",
2077 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>{
2078 let isCommutable = Commutable;
2081 class N3VDSL<bits<2> op21_20, bits<4> op11_8,
2082 InstrItinClass itin, string OpcodeStr, string Dt,
2083 ValueType Ty, SDNode ShOp>
2084 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2085 (outs DPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2086 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2088 (Ty (ShOp (Ty DPR:$Vn),
2089 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),imm:$lane)))))]> {
2090 let isCommutable = 0;
2092 class N3VDSL16<bits<2> op21_20, bits<4> op11_8,
2093 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
2094 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2095 (outs DPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2096 NVMulSLFrm, IIC_VMULi16D, OpcodeStr, Dt,"$Vd, $Vn, $Vm$lane","",
2098 (Ty (ShOp (Ty DPR:$Vn),
2099 (Ty (NEONvduplane (Ty DPR_8:$Vm), imm:$lane)))))]> {
2100 let isCommutable = 0;
2103 class N3VQ<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2104 InstrItinClass itin, string OpcodeStr, string Dt,
2105 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
2106 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2107 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2108 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2109 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
2110 let isCommutable = Commutable;
2112 class N3VQX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2113 InstrItinClass itin, string OpcodeStr,
2114 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
2115 : N3VX<op24, op23, op21_20, op11_8, 1, op4,
2116 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2117 OpcodeStr, "$Vd, $Vn, $Vm", "",
2118 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]>{
2119 let isCommutable = Commutable;
2121 class N3VQSL<bits<2> op21_20, bits<4> op11_8,
2122 InstrItinClass itin, string OpcodeStr, string Dt,
2123 ValueType ResTy, ValueType OpTy, SDNode ShOp>
2124 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2125 (outs QPR:$Vd), (ins QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2126 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2127 [(set (ResTy QPR:$Vd),
2128 (ResTy (ShOp (ResTy QPR:$Vn),
2129 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2131 let isCommutable = 0;
2133 class N3VQSL16<bits<2> op21_20, bits<4> op11_8, string OpcodeStr, string Dt,
2134 ValueType ResTy, ValueType OpTy, SDNode ShOp>
2135 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2136 (outs QPR:$Vd), (ins QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2137 NVMulSLFrm, IIC_VMULi16Q, OpcodeStr, Dt,"$Vd, $Vn, $Vm$lane", "",
2138 [(set (ResTy QPR:$Vd),
2139 (ResTy (ShOp (ResTy QPR:$Vn),
2140 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2142 let isCommutable = 0;
2145 // Basic 3-register intrinsics, both double- and quad-register.
2146 class N3VDInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2147 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2148 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
2149 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2150 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), f, itin,
2151 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2152 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
2153 let isCommutable = Commutable;
2155 class N3VDIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2156 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
2157 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2158 (outs DPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2159 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2161 (Ty (IntOp (Ty DPR:$Vn),
2162 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),
2164 let isCommutable = 0;
2166 class N3VDIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2167 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
2168 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2169 (outs DPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2170 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2172 (Ty (IntOp (Ty DPR:$Vn),
2173 (Ty (NEONvduplane (Ty DPR_8:$Vm), imm:$lane)))))]> {
2174 let isCommutable = 0;
2176 class N3VDIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2177 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2178 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2179 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2180 (outs DPR:$Vd), (ins DPR:$Vm, DPR:$Vn), f, itin,
2181 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
2182 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (OpTy DPR:$Vn))))]> {
2183 let isCommutable = 0;
2186 class N3VQInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2187 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2188 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
2189 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2190 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), f, itin,
2191 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2192 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
2193 let isCommutable = Commutable;
2195 class N3VQIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2196 string OpcodeStr, string Dt,
2197 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2198 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2199 (outs QPR:$Vd), (ins QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2200 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2201 [(set (ResTy QPR:$Vd),
2202 (ResTy (IntOp (ResTy QPR:$Vn),
2203 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2205 let isCommutable = 0;
2207 class N3VQIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2208 string OpcodeStr, string Dt,
2209 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2210 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2211 (outs QPR:$Vd), (ins QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2212 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2213 [(set (ResTy QPR:$Vd),
2214 (ResTy (IntOp (ResTy QPR:$Vn),
2215 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2217 let isCommutable = 0;
2219 class N3VQIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2220 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2221 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2222 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2223 (outs QPR:$Vd), (ins QPR:$Vm, QPR:$Vn), f, itin,
2224 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
2225 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (OpTy QPR:$Vn))))]> {
2226 let isCommutable = 0;
2229 // Multiply-Add/Sub operations: double- and quad-register.
2230 class N3VDMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2231 InstrItinClass itin, string OpcodeStr, string Dt,
2232 ValueType Ty, SDPatternOperator MulOp, SDPatternOperator OpNode>
2233 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2234 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2235 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2236 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
2237 (Ty (MulOp DPR:$Vn, DPR:$Vm)))))]>;
2239 class N3VDMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2240 string OpcodeStr, string Dt,
2241 ValueType Ty, SDPatternOperator MulOp, SDPatternOperator ShOp>
2242 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2244 (ins DPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2246 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2248 (Ty (ShOp (Ty DPR:$src1),
2250 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),
2252 class N3VDMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2253 string OpcodeStr, string Dt,
2254 ValueType Ty, SDNode MulOp, SDNode ShOp>
2255 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2257 (ins DPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2259 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2261 (Ty (ShOp (Ty DPR:$src1),
2263 (Ty (NEONvduplane (Ty DPR_8:$Vm),
2266 class N3VQMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2267 InstrItinClass itin, string OpcodeStr, string Dt, ValueType Ty,
2268 SDPatternOperator MulOp, SDPatternOperator OpNode>
2269 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2270 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2271 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2272 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
2273 (Ty (MulOp QPR:$Vn, QPR:$Vm)))))]>;
2274 class N3VQMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2275 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2276 SDPatternOperator MulOp, SDPatternOperator ShOp>
2277 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2279 (ins QPR:$src1, QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2281 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2282 [(set (ResTy QPR:$Vd),
2283 (ResTy (ShOp (ResTy QPR:$src1),
2284 (ResTy (MulOp QPR:$Vn,
2285 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2287 class N3VQMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2288 string OpcodeStr, string Dt,
2289 ValueType ResTy, ValueType OpTy,
2290 SDNode MulOp, SDNode ShOp>
2291 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2293 (ins QPR:$src1, QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2295 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2296 [(set (ResTy QPR:$Vd),
2297 (ResTy (ShOp (ResTy QPR:$src1),
2298 (ResTy (MulOp QPR:$Vn,
2299 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2302 // Neon Intrinsic-Op instructions (VABA): double- and quad-register.
2303 class N3VDIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2304 InstrItinClass itin, string OpcodeStr, string Dt,
2305 ValueType Ty, Intrinsic IntOp, SDNode OpNode>
2306 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2307 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2308 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2309 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
2310 (Ty (IntOp (Ty DPR:$Vn), (Ty DPR:$Vm))))))]>;
2311 class N3VQIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2312 InstrItinClass itin, string OpcodeStr, string Dt,
2313 ValueType Ty, Intrinsic IntOp, SDNode OpNode>
2314 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2315 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2316 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2317 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
2318 (Ty (IntOp (Ty QPR:$Vn), (Ty QPR:$Vm))))))]>;
2320 // Neon 3-argument intrinsics, both double- and quad-register.
2321 // The destination register is also used as the first source operand register.
2322 class N3VDInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2323 InstrItinClass itin, string OpcodeStr, string Dt,
2324 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2325 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2326 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2327 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2328 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$src1),
2329 (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>;
2330 class N3VQInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2331 InstrItinClass itin, string OpcodeStr, string Dt,
2332 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2333 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2334 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2335 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2336 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$src1),
2337 (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]>;
2339 // Long Multiply-Add/Sub operations.
2340 class N3VLMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2341 InstrItinClass itin, string OpcodeStr, string Dt,
2342 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2343 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2344 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2345 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2346 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
2347 (TyQ (MulOp (TyD DPR:$Vn),
2348 (TyD DPR:$Vm)))))]>;
2349 class N3VLMulOpSL<bit op24, bits<2> op21_20, bits<4> op11_8,
2350 InstrItinClass itin, string OpcodeStr, string Dt,
2351 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2352 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$Vd),
2353 (ins QPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2355 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2357 (OpNode (TyQ QPR:$src1),
2358 (TyQ (MulOp (TyD DPR:$Vn),
2359 (TyD (NEONvduplane (TyD DPR_VFP2:$Vm),
2361 class N3VLMulOpSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2362 InstrItinClass itin, string OpcodeStr, string Dt,
2363 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2364 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$Vd),
2365 (ins QPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2367 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2369 (OpNode (TyQ QPR:$src1),
2370 (TyQ (MulOp (TyD DPR:$Vn),
2371 (TyD (NEONvduplane (TyD DPR_8:$Vm),
2374 // Long Intrinsic-Op vector operations with explicit extend (VABAL).
2375 class N3VLIntExtOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2376 InstrItinClass itin, string OpcodeStr, string Dt,
2377 ValueType TyQ, ValueType TyD, Intrinsic IntOp, SDNode ExtOp,
2379 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2380 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2381 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2382 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
2383 (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
2384 (TyD DPR:$Vm)))))))]>;
2386 // Neon Long 3-argument intrinsic. The destination register is
2387 // a quad-register and is also used as the first source operand register.
2388 class N3VLInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2389 InstrItinClass itin, string OpcodeStr, string Dt,
2390 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
2391 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2392 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2393 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2395 (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$Vn), (TyD DPR:$Vm))))]>;
2396 class N3VLInt3SL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2397 string OpcodeStr, string Dt,
2398 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2399 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2401 (ins QPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2403 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2404 [(set (ResTy QPR:$Vd),
2405 (ResTy (IntOp (ResTy QPR:$src1),
2407 (OpTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2409 class N3VLInt3SL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2410 InstrItinClass itin, string OpcodeStr, string Dt,
2411 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2412 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2414 (ins QPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2416 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2417 [(set (ResTy QPR:$Vd),
2418 (ResTy (IntOp (ResTy QPR:$src1),
2420 (OpTy (NEONvduplane (OpTy DPR_8:$Vm),
2423 // Narrowing 3-register intrinsics.
2424 class N3VNInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2425 string OpcodeStr, string Dt, ValueType TyD, ValueType TyQ,
2426 Intrinsic IntOp, bit Commutable>
2427 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2428 (outs DPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINi4D,
2429 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2430 [(set DPR:$Vd, (TyD (IntOp (TyQ QPR:$Vn), (TyQ QPR:$Vm))))]> {
2431 let isCommutable = Commutable;
2434 // Long 3-register operations.
2435 class N3VL<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2436 InstrItinClass itin, string OpcodeStr, string Dt,
2437 ValueType TyQ, ValueType TyD, SDNode OpNode, bit Commutable>
2438 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2439 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2440 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2441 [(set QPR:$Vd, (TyQ (OpNode (TyD DPR:$Vn), (TyD DPR:$Vm))))]> {
2442 let isCommutable = Commutable;
2444 class N3VLSL<bit op24, bits<2> op21_20, bits<4> op11_8,
2445 InstrItinClass itin, string OpcodeStr, string Dt,
2446 ValueType TyQ, ValueType TyD, SDNode OpNode>
2447 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2448 (outs QPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2449 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2451 (TyQ (OpNode (TyD DPR:$Vn),
2452 (TyD (NEONvduplane (TyD DPR_VFP2:$Vm),imm:$lane)))))]>;
2453 class N3VLSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2454 InstrItinClass itin, string OpcodeStr, string Dt,
2455 ValueType TyQ, ValueType TyD, SDNode OpNode>
2456 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2457 (outs QPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2458 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2460 (TyQ (OpNode (TyD DPR:$Vn),
2461 (TyD (NEONvduplane (TyD DPR_8:$Vm), imm:$lane)))))]>;
2463 // Long 3-register operations with explicitly extended operands.
2464 class N3VLExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2465 InstrItinClass itin, string OpcodeStr, string Dt,
2466 ValueType TyQ, ValueType TyD, SDNode OpNode, SDNode ExtOp,
2468 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2469 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2470 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2471 [(set QPR:$Vd, (OpNode (TyQ (ExtOp (TyD DPR:$Vn))),
2472 (TyQ (ExtOp (TyD DPR:$Vm)))))]> {
2473 let isCommutable = Commutable;
2476 // Long 3-register intrinsics with explicit extend (VABDL).
2477 class N3VLIntExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2478 InstrItinClass itin, string OpcodeStr, string Dt,
2479 ValueType TyQ, ValueType TyD, Intrinsic IntOp, SDNode ExtOp,
2481 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2482 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2483 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2484 [(set QPR:$Vd, (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
2485 (TyD DPR:$Vm))))))]> {
2486 let isCommutable = Commutable;
2489 // Long 3-register intrinsics.
2490 class N3VLInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2491 InstrItinClass itin, string OpcodeStr, string Dt,
2492 ValueType TyQ, ValueType TyD, Intrinsic IntOp, bit Commutable>
2493 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2494 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2495 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2496 [(set QPR:$Vd, (TyQ (IntOp (TyD DPR:$Vn), (TyD DPR:$Vm))))]> {
2497 let isCommutable = Commutable;
2499 class N3VLIntSL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2500 string OpcodeStr, string Dt,
2501 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2502 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2503 (outs QPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2504 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2505 [(set (ResTy QPR:$Vd),
2506 (ResTy (IntOp (OpTy DPR:$Vn),
2507 (OpTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2509 class N3VLIntSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2510 InstrItinClass itin, string OpcodeStr, string Dt,
2511 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2512 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2513 (outs QPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2514 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2515 [(set (ResTy QPR:$Vd),
2516 (ResTy (IntOp (OpTy DPR:$Vn),
2517 (OpTy (NEONvduplane (OpTy DPR_8:$Vm),
2520 // Wide 3-register operations.
2521 class N3VW<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2522 string OpcodeStr, string Dt, ValueType TyQ, ValueType TyD,
2523 SDNode OpNode, SDNode ExtOp, bit Commutable>
2524 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2525 (outs QPR:$Vd), (ins QPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VSUBiD,
2526 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2527 [(set QPR:$Vd, (OpNode (TyQ QPR:$Vn),
2528 (TyQ (ExtOp (TyD DPR:$Vm)))))]> {
2529 let isCommutable = Commutable;
2532 // Pairwise long 2-register intrinsics, both double- and quad-register.
2533 class N2VDPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2534 bits<2> op17_16, bits<5> op11_7, bit op4,
2535 string OpcodeStr, string Dt,
2536 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2537 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
2538 (ins DPR:$Vm), IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
2539 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm))))]>;
2540 class N2VQPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2541 bits<2> op17_16, bits<5> op11_7, bit op4,
2542 string OpcodeStr, string Dt,
2543 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2544 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
2545 (ins QPR:$Vm), IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
2546 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm))))]>;
2548 // Pairwise long 2-register accumulate intrinsics,
2549 // both double- and quad-register.
2550 // The destination register is also used as the first source operand register.
2551 class N2VDPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2552 bits<2> op17_16, bits<5> op11_7, bit op4,
2553 string OpcodeStr, string Dt,
2554 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2555 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
2556 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vm), IIC_VPALiD,
2557 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
2558 [(set DPR:$Vd, (ResTy (IntOp (ResTy DPR:$src1), (OpTy DPR:$Vm))))]>;
2559 class N2VQPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2560 bits<2> op17_16, bits<5> op11_7, bit op4,
2561 string OpcodeStr, string Dt,
2562 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2563 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4,
2564 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vm), IIC_VPALiQ,
2565 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
2566 [(set QPR:$Vd, (ResTy (IntOp (ResTy QPR:$src1), (OpTy QPR:$Vm))))]>;
2568 // Shift by immediate,
2569 // both double- and quad-register.
2570 class N2VDSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2571 Format f, InstrItinClass itin, Operand ImmTy,
2572 string OpcodeStr, string Dt, ValueType Ty, SDNode OpNode>
2573 : N2VImm<op24, op23, op11_8, op7, 0, op4,
2574 (outs DPR:$Vd), (ins DPR:$Vm, ImmTy:$SIMM), f, itin,
2575 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2576 [(set DPR:$Vd, (Ty (OpNode (Ty DPR:$Vm), (i32 imm:$SIMM))))]>;
2577 class N2VQSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2578 Format f, InstrItinClass itin, Operand ImmTy,
2579 string OpcodeStr, string Dt, ValueType Ty, SDNode OpNode>
2580 : N2VImm<op24, op23, op11_8, op7, 1, op4,
2581 (outs QPR:$Vd), (ins QPR:$Vm, ImmTy:$SIMM), f, itin,
2582 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2583 [(set QPR:$Vd, (Ty (OpNode (Ty QPR:$Vm), (i32 imm:$SIMM))))]>;
2585 // Long shift by immediate.
2586 class N2VLSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
2587 string OpcodeStr, string Dt,
2588 ValueType ResTy, ValueType OpTy, SDNode OpNode>
2589 : N2VImm<op24, op23, op11_8, op7, op6, op4,
2590 (outs QPR:$Vd), (ins DPR:$Vm, i32imm:$SIMM), N2RegVShLFrm,
2591 IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2592 [(set QPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vm),
2593 (i32 imm:$SIMM))))]>;
2595 // Narrow shift by immediate.
2596 class N2VNSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
2597 InstrItinClass itin, string OpcodeStr, string Dt,
2598 ValueType ResTy, ValueType OpTy, Operand ImmTy, SDNode OpNode>
2599 : N2VImm<op24, op23, op11_8, op7, op6, op4,
2600 (outs DPR:$Vd), (ins QPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, itin,
2601 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2602 [(set DPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vm),
2603 (i32 imm:$SIMM))))]>;
2605 // Shift right by immediate and accumulate,
2606 // both double- and quad-register.
2607 class N2VDShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2608 Operand ImmTy, string OpcodeStr, string Dt,
2609 ValueType Ty, SDNode ShOp>
2610 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
2611 (ins DPR:$src1, DPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, IIC_VPALiD,
2612 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2613 [(set DPR:$Vd, (Ty (add DPR:$src1,
2614 (Ty (ShOp DPR:$Vm, (i32 imm:$SIMM))))))]>;
2615 class N2VQShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2616 Operand ImmTy, string OpcodeStr, string Dt,
2617 ValueType Ty, SDNode ShOp>
2618 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
2619 (ins QPR:$src1, QPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, IIC_VPALiD,
2620 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2621 [(set QPR:$Vd, (Ty (add QPR:$src1,
2622 (Ty (ShOp QPR:$Vm, (i32 imm:$SIMM))))))]>;
2624 // Shift by immediate and insert,
2625 // both double- and quad-register.
2626 class N2VDShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2627 Operand ImmTy, Format f, string OpcodeStr, string Dt,
2628 ValueType Ty,SDNode ShOp>
2629 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
2630 (ins DPR:$src1, DPR:$Vm, ImmTy:$SIMM), f, IIC_VSHLiD,
2631 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2632 [(set DPR:$Vd, (Ty (ShOp DPR:$src1, DPR:$Vm, (i32 imm:$SIMM))))]>;
2633 class N2VQShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2634 Operand ImmTy, Format f, string OpcodeStr, string Dt,
2635 ValueType Ty,SDNode ShOp>
2636 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
2637 (ins QPR:$src1, QPR:$Vm, ImmTy:$SIMM), f, IIC_VSHLiQ,
2638 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2639 [(set QPR:$Vd, (Ty (ShOp QPR:$src1, QPR:$Vm, (i32 imm:$SIMM))))]>;
2641 // Convert, with fractional bits immediate,
2642 // both double- and quad-register.
2643 class N2VCvtD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2644 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2646 : N2VImm<op24, op23, op11_8, op7, 0, op4,
2647 (outs DPR:$Vd), (ins DPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
2648 IIC_VUNAD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2649 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (i32 imm:$SIMM))))]>;
2650 class N2VCvtQ<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2651 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2653 : N2VImm<op24, op23, op11_8, op7, 1, op4,
2654 (outs QPR:$Vd), (ins QPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
2655 IIC_VUNAQ, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2656 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (i32 imm:$SIMM))))]>;
2658 //===----------------------------------------------------------------------===//
2660 //===----------------------------------------------------------------------===//
2662 // Abbreviations used in multiclass suffixes:
2663 // Q = quarter int (8 bit) elements
2664 // H = half int (16 bit) elements
2665 // S = single int (32 bit) elements
2666 // D = double int (64 bit) elements
2668 // Neon 2-register vector operations and intrinsics.
2670 // Neon 2-register comparisons.
2671 // source operand element sizes of 8, 16 and 32 bits:
2672 multiclass N2V_QHS_cmp<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2673 bits<5> op11_7, bit op4, string opc, string Dt,
2674 string asm, SDNode OpNode> {
2675 // 64-bit vector types.
2676 def v8i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 0, op4,
2677 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2678 opc, !strconcat(Dt, "8"), asm, "",
2679 [(set DPR:$Vd, (v8i8 (OpNode (v8i8 DPR:$Vm))))]>;
2680 def v4i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 0, op4,
2681 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2682 opc, !strconcat(Dt, "16"), asm, "",
2683 [(set DPR:$Vd, (v4i16 (OpNode (v4i16 DPR:$Vm))))]>;
2684 def v2i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
2685 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2686 opc, !strconcat(Dt, "32"), asm, "",
2687 [(set DPR:$Vd, (v2i32 (OpNode (v2i32 DPR:$Vm))))]>;
2688 def v2f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
2689 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2690 opc, "f32", asm, "",
2691 [(set DPR:$Vd, (v2i32 (OpNode (v2f32 DPR:$Vm))))]> {
2692 let Inst{10} = 1; // overwrite F = 1
2695 // 128-bit vector types.
2696 def v16i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 1, op4,
2697 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2698 opc, !strconcat(Dt, "8"), asm, "",
2699 [(set QPR:$Vd, (v16i8 (OpNode (v16i8 QPR:$Vm))))]>;
2700 def v8i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 1, op4,
2701 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2702 opc, !strconcat(Dt, "16"), asm, "",
2703 [(set QPR:$Vd, (v8i16 (OpNode (v8i16 QPR:$Vm))))]>;
2704 def v4i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
2705 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2706 opc, !strconcat(Dt, "32"), asm, "",
2707 [(set QPR:$Vd, (v4i32 (OpNode (v4i32 QPR:$Vm))))]>;
2708 def v4f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
2709 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2710 opc, "f32", asm, "",
2711 [(set QPR:$Vd, (v4i32 (OpNode (v4f32 QPR:$Vm))))]> {
2712 let Inst{10} = 1; // overwrite F = 1
2717 // Neon 2-register vector intrinsics,
2718 // element sizes of 8, 16 and 32 bits:
2719 multiclass N2VInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2720 bits<5> op11_7, bit op4,
2721 InstrItinClass itinD, InstrItinClass itinQ,
2722 string OpcodeStr, string Dt, Intrinsic IntOp> {
2723 // 64-bit vector types.
2724 def v8i8 : N2VDInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2725 itinD, OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
2726 def v4i16 : N2VDInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2727 itinD, OpcodeStr, !strconcat(Dt, "16"),v4i16,v4i16,IntOp>;
2728 def v2i32 : N2VDInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2729 itinD, OpcodeStr, !strconcat(Dt, "32"),v2i32,v2i32,IntOp>;
2731 // 128-bit vector types.
2732 def v16i8 : N2VQInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2733 itinQ, OpcodeStr, !strconcat(Dt, "8"), v16i8,v16i8,IntOp>;
2734 def v8i16 : N2VQInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2735 itinQ, OpcodeStr, !strconcat(Dt, "16"),v8i16,v8i16,IntOp>;
2736 def v4i32 : N2VQInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2737 itinQ, OpcodeStr, !strconcat(Dt, "32"),v4i32,v4i32,IntOp>;
2741 // Neon Narrowing 2-register vector operations,
2742 // source operand element sizes of 16, 32 and 64 bits:
2743 multiclass N2VN_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2744 bits<5> op11_7, bit op6, bit op4,
2745 InstrItinClass itin, string OpcodeStr, string Dt,
2747 def v8i8 : N2VN<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
2748 itin, OpcodeStr, !strconcat(Dt, "16"),
2749 v8i8, v8i16, OpNode>;
2750 def v4i16 : N2VN<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
2751 itin, OpcodeStr, !strconcat(Dt, "32"),
2752 v4i16, v4i32, OpNode>;
2753 def v2i32 : N2VN<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
2754 itin, OpcodeStr, !strconcat(Dt, "64"),
2755 v2i32, v2i64, OpNode>;
2758 // Neon Narrowing 2-register vector intrinsics,
2759 // source operand element sizes of 16, 32 and 64 bits:
2760 multiclass N2VNInt_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2761 bits<5> op11_7, bit op6, bit op4,
2762 InstrItinClass itin, string OpcodeStr, string Dt,
2764 def v8i8 : N2VNInt<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
2765 itin, OpcodeStr, !strconcat(Dt, "16"),
2766 v8i8, v8i16, IntOp>;
2767 def v4i16 : N2VNInt<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
2768 itin, OpcodeStr, !strconcat(Dt, "32"),
2769 v4i16, v4i32, IntOp>;
2770 def v2i32 : N2VNInt<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
2771 itin, OpcodeStr, !strconcat(Dt, "64"),
2772 v2i32, v2i64, IntOp>;
2776 // Neon Lengthening 2-register vector intrinsic (currently specific to VMOVL).
2777 // source operand element sizes of 16, 32 and 64 bits:
2778 multiclass N2VL_QHS<bits<2> op24_23, bits<5> op11_7, bit op6, bit op4,
2779 string OpcodeStr, string Dt, SDNode OpNode> {
2780 def v8i16 : N2VL<op24_23, 0b00, 0b10, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2781 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode>;
2782 def v4i32 : N2VL<op24_23, 0b01, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2783 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
2784 def v2i64 : N2VL<op24_23, 0b10, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2785 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
2789 // Neon 3-register vector operations.
2791 // First with only element sizes of 8, 16 and 32 bits:
2792 multiclass N3V_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2793 InstrItinClass itinD16, InstrItinClass itinD32,
2794 InstrItinClass itinQ16, InstrItinClass itinQ32,
2795 string OpcodeStr, string Dt,
2796 SDNode OpNode, bit Commutable = 0> {
2797 // 64-bit vector types.
2798 def v8i8 : N3VD<op24, op23, 0b00, op11_8, op4, itinD16,
2799 OpcodeStr, !strconcat(Dt, "8"),
2800 v8i8, v8i8, OpNode, Commutable>;
2801 def v4i16 : N3VD<op24, op23, 0b01, op11_8, op4, itinD16,
2802 OpcodeStr, !strconcat(Dt, "16"),
2803 v4i16, v4i16, OpNode, Commutable>;
2804 def v2i32 : N3VD<op24, op23, 0b10, op11_8, op4, itinD32,
2805 OpcodeStr, !strconcat(Dt, "32"),
2806 v2i32, v2i32, OpNode, Commutable>;
2808 // 128-bit vector types.
2809 def v16i8 : N3VQ<op24, op23, 0b00, op11_8, op4, itinQ16,
2810 OpcodeStr, !strconcat(Dt, "8"),
2811 v16i8, v16i8, OpNode, Commutable>;
2812 def v8i16 : N3VQ<op24, op23, 0b01, op11_8, op4, itinQ16,
2813 OpcodeStr, !strconcat(Dt, "16"),
2814 v8i16, v8i16, OpNode, Commutable>;
2815 def v4i32 : N3VQ<op24, op23, 0b10, op11_8, op4, itinQ32,
2816 OpcodeStr, !strconcat(Dt, "32"),
2817 v4i32, v4i32, OpNode, Commutable>;
2820 multiclass N3VSL_HS<bits<4> op11_8, string OpcodeStr, string Dt, SDNode ShOp> {
2821 def v4i16 : N3VDSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
2823 def v2i32 : N3VDSL<0b10, op11_8, IIC_VMULi32D, OpcodeStr, !strconcat(Dt,"32"),
2825 def v8i16 : N3VQSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
2826 v8i16, v4i16, ShOp>;
2827 def v4i32 : N3VQSL<0b10, op11_8, IIC_VMULi32Q, OpcodeStr, !strconcat(Dt,"32"),
2828 v4i32, v2i32, ShOp>;
2831 // ....then also with element size 64 bits:
2832 multiclass N3V_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2833 InstrItinClass itinD, InstrItinClass itinQ,
2834 string OpcodeStr, string Dt,
2835 SDNode OpNode, bit Commutable = 0>
2836 : N3V_QHS<op24, op23, op11_8, op4, itinD, itinD, itinQ, itinQ,
2837 OpcodeStr, Dt, OpNode, Commutable> {
2838 def v1i64 : N3VD<op24, op23, 0b11, op11_8, op4, itinD,
2839 OpcodeStr, !strconcat(Dt, "64"),
2840 v1i64, v1i64, OpNode, Commutable>;
2841 def v2i64 : N3VQ<op24, op23, 0b11, op11_8, op4, itinQ,
2842 OpcodeStr, !strconcat(Dt, "64"),
2843 v2i64, v2i64, OpNode, Commutable>;
2847 // Neon 3-register vector intrinsics.
2849 // First with only element sizes of 16 and 32 bits:
2850 multiclass N3VInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2851 InstrItinClass itinD16, InstrItinClass itinD32,
2852 InstrItinClass itinQ16, InstrItinClass itinQ32,
2853 string OpcodeStr, string Dt,
2854 Intrinsic IntOp, bit Commutable = 0> {
2855 // 64-bit vector types.
2856 def v4i16 : N3VDInt<op24, op23, 0b01, op11_8, op4, f, itinD16,
2857 OpcodeStr, !strconcat(Dt, "16"),
2858 v4i16, v4i16, IntOp, Commutable>;
2859 def v2i32 : N3VDInt<op24, op23, 0b10, op11_8, op4, f, itinD32,
2860 OpcodeStr, !strconcat(Dt, "32"),
2861 v2i32, v2i32, IntOp, Commutable>;
2863 // 128-bit vector types.
2864 def v8i16 : N3VQInt<op24, op23, 0b01, op11_8, op4, f, itinQ16,
2865 OpcodeStr, !strconcat(Dt, "16"),
2866 v8i16, v8i16, IntOp, Commutable>;
2867 def v4i32 : N3VQInt<op24, op23, 0b10, op11_8, op4, f, itinQ32,
2868 OpcodeStr, !strconcat(Dt, "32"),
2869 v4i32, v4i32, IntOp, Commutable>;
2871 multiclass N3VInt_HSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2872 InstrItinClass itinD16, InstrItinClass itinD32,
2873 InstrItinClass itinQ16, InstrItinClass itinQ32,
2874 string OpcodeStr, string Dt,
2876 // 64-bit vector types.
2877 def v4i16 : N3VDIntSh<op24, op23, 0b01, op11_8, op4, f, itinD16,
2878 OpcodeStr, !strconcat(Dt, "16"),
2879 v4i16, v4i16, IntOp>;
2880 def v2i32 : N3VDIntSh<op24, op23, 0b10, op11_8, op4, f, itinD32,
2881 OpcodeStr, !strconcat(Dt, "32"),
2882 v2i32, v2i32, IntOp>;
2884 // 128-bit vector types.
2885 def v8i16 : N3VQIntSh<op24, op23, 0b01, op11_8, op4, f, itinQ16,
2886 OpcodeStr, !strconcat(Dt, "16"),
2887 v8i16, v8i16, IntOp>;
2888 def v4i32 : N3VQIntSh<op24, op23, 0b10, op11_8, op4, f, itinQ32,
2889 OpcodeStr, !strconcat(Dt, "32"),
2890 v4i32, v4i32, IntOp>;
2893 multiclass N3VIntSL_HS<bits<4> op11_8,
2894 InstrItinClass itinD16, InstrItinClass itinD32,
2895 InstrItinClass itinQ16, InstrItinClass itinQ32,
2896 string OpcodeStr, string Dt, Intrinsic IntOp> {
2897 def v4i16 : N3VDIntSL16<0b01, op11_8, itinD16,
2898 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp>;
2899 def v2i32 : N3VDIntSL<0b10, op11_8, itinD32,
2900 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp>;
2901 def v8i16 : N3VQIntSL16<0b01, op11_8, itinQ16,
2902 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16, IntOp>;
2903 def v4i32 : N3VQIntSL<0b10, op11_8, itinQ32,
2904 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32, IntOp>;
2907 // ....then also with element size of 8 bits:
2908 multiclass N3VInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2909 InstrItinClass itinD16, InstrItinClass itinD32,
2910 InstrItinClass itinQ16, InstrItinClass itinQ32,
2911 string OpcodeStr, string Dt,
2912 Intrinsic IntOp, bit Commutable = 0>
2913 : N3VInt_HS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2914 OpcodeStr, Dt, IntOp, Commutable> {
2915 def v8i8 : N3VDInt<op24, op23, 0b00, op11_8, op4, f, itinD16,
2916 OpcodeStr, !strconcat(Dt, "8"),
2917 v8i8, v8i8, IntOp, Commutable>;
2918 def v16i8 : N3VQInt<op24, op23, 0b00, op11_8, op4, f, itinQ16,
2919 OpcodeStr, !strconcat(Dt, "8"),
2920 v16i8, v16i8, IntOp, Commutable>;
2922 multiclass N3VInt_QHSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2923 InstrItinClass itinD16, InstrItinClass itinD32,
2924 InstrItinClass itinQ16, InstrItinClass itinQ32,
2925 string OpcodeStr, string Dt,
2927 : N3VInt_HSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2928 OpcodeStr, Dt, IntOp> {
2929 def v8i8 : N3VDIntSh<op24, op23, 0b00, op11_8, op4, f, itinD16,
2930 OpcodeStr, !strconcat(Dt, "8"),
2932 def v16i8 : N3VQIntSh<op24, op23, 0b00, op11_8, op4, f, itinQ16,
2933 OpcodeStr, !strconcat(Dt, "8"),
2934 v16i8, v16i8, IntOp>;
2938 // ....then also with element size of 64 bits:
2939 multiclass N3VInt_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2940 InstrItinClass itinD16, InstrItinClass itinD32,
2941 InstrItinClass itinQ16, InstrItinClass itinQ32,
2942 string OpcodeStr, string Dt,
2943 Intrinsic IntOp, bit Commutable = 0>
2944 : N3VInt_QHS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2945 OpcodeStr, Dt, IntOp, Commutable> {
2946 def v1i64 : N3VDInt<op24, op23, 0b11, op11_8, op4, f, itinD32,
2947 OpcodeStr, !strconcat(Dt, "64"),
2948 v1i64, v1i64, IntOp, Commutable>;
2949 def v2i64 : N3VQInt<op24, op23, 0b11, op11_8, op4, f, itinQ32,
2950 OpcodeStr, !strconcat(Dt, "64"),
2951 v2i64, v2i64, IntOp, Commutable>;
2953 multiclass N3VInt_QHSDSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2954 InstrItinClass itinD16, InstrItinClass itinD32,
2955 InstrItinClass itinQ16, InstrItinClass itinQ32,
2956 string OpcodeStr, string Dt,
2958 : N3VInt_QHSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2959 OpcodeStr, Dt, IntOp> {
2960 def v1i64 : N3VDIntSh<op24, op23, 0b11, op11_8, op4, f, itinD32,
2961 OpcodeStr, !strconcat(Dt, "64"),
2962 v1i64, v1i64, IntOp>;
2963 def v2i64 : N3VQIntSh<op24, op23, 0b11, op11_8, op4, f, itinQ32,
2964 OpcodeStr, !strconcat(Dt, "64"),
2965 v2i64, v2i64, IntOp>;
2968 // Neon Narrowing 3-register vector intrinsics,
2969 // source operand element sizes of 16, 32 and 64 bits:
2970 multiclass N3VNInt_HSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2971 string OpcodeStr, string Dt,
2972 Intrinsic IntOp, bit Commutable = 0> {
2973 def v8i8 : N3VNInt<op24, op23, 0b00, op11_8, op4,
2974 OpcodeStr, !strconcat(Dt, "16"),
2975 v8i8, v8i16, IntOp, Commutable>;
2976 def v4i16 : N3VNInt<op24, op23, 0b01, op11_8, op4,
2977 OpcodeStr, !strconcat(Dt, "32"),
2978 v4i16, v4i32, IntOp, Commutable>;
2979 def v2i32 : N3VNInt<op24, op23, 0b10, op11_8, op4,
2980 OpcodeStr, !strconcat(Dt, "64"),
2981 v2i32, v2i64, IntOp, Commutable>;
2985 // Neon Long 3-register vector operations.
2987 multiclass N3VL_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2988 InstrItinClass itin16, InstrItinClass itin32,
2989 string OpcodeStr, string Dt,
2990 SDNode OpNode, bit Commutable = 0> {
2991 def v8i16 : N3VL<op24, op23, 0b00, op11_8, op4, itin16,
2992 OpcodeStr, !strconcat(Dt, "8"),
2993 v8i16, v8i8, OpNode, Commutable>;
2994 def v4i32 : N3VL<op24, op23, 0b01, op11_8, op4, itin16,
2995 OpcodeStr, !strconcat(Dt, "16"),
2996 v4i32, v4i16, OpNode, Commutable>;
2997 def v2i64 : N3VL<op24, op23, 0b10, op11_8, op4, itin32,
2998 OpcodeStr, !strconcat(Dt, "32"),
2999 v2i64, v2i32, OpNode, Commutable>;
3002 multiclass N3VLSL_HS<bit op24, bits<4> op11_8,
3003 InstrItinClass itin, string OpcodeStr, string Dt,
3005 def v4i16 : N3VLSL16<op24, 0b01, op11_8, itin, OpcodeStr,
3006 !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
3007 def v2i32 : N3VLSL<op24, 0b10, op11_8, itin, OpcodeStr,
3008 !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
3011 multiclass N3VLExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3012 InstrItinClass itin16, InstrItinClass itin32,
3013 string OpcodeStr, string Dt,
3014 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
3015 def v8i16 : N3VLExt<op24, op23, 0b00, op11_8, op4, itin16,
3016 OpcodeStr, !strconcat(Dt, "8"),
3017 v8i16, v8i8, OpNode, ExtOp, Commutable>;
3018 def v4i32 : N3VLExt<op24, op23, 0b01, op11_8, op4, itin16,
3019 OpcodeStr, !strconcat(Dt, "16"),
3020 v4i32, v4i16, OpNode, ExtOp, Commutable>;
3021 def v2i64 : N3VLExt<op24, op23, 0b10, op11_8, op4, itin32,
3022 OpcodeStr, !strconcat(Dt, "32"),
3023 v2i64, v2i32, OpNode, ExtOp, Commutable>;
3026 // Neon Long 3-register vector intrinsics.
3028 // First with only element sizes of 16 and 32 bits:
3029 multiclass N3VLInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
3030 InstrItinClass itin16, InstrItinClass itin32,
3031 string OpcodeStr, string Dt,
3032 Intrinsic IntOp, bit Commutable = 0> {
3033 def v4i32 : N3VLInt<op24, op23, 0b01, op11_8, op4, itin16,
3034 OpcodeStr, !strconcat(Dt, "16"),
3035 v4i32, v4i16, IntOp, Commutable>;
3036 def v2i64 : N3VLInt<op24, op23, 0b10, op11_8, op4, itin32,
3037 OpcodeStr, !strconcat(Dt, "32"),
3038 v2i64, v2i32, IntOp, Commutable>;
3041 multiclass N3VLIntSL_HS<bit op24, bits<4> op11_8,
3042 InstrItinClass itin, string OpcodeStr, string Dt,
3044 def v4i16 : N3VLIntSL16<op24, 0b01, op11_8, itin,
3045 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
3046 def v2i32 : N3VLIntSL<op24, 0b10, op11_8, itin,
3047 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3050 // ....then also with element size of 8 bits:
3051 multiclass N3VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3052 InstrItinClass itin16, InstrItinClass itin32,
3053 string OpcodeStr, string Dt,
3054 Intrinsic IntOp, bit Commutable = 0>
3055 : N3VLInt_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt,
3056 IntOp, Commutable> {
3057 def v8i16 : N3VLInt<op24, op23, 0b00, op11_8, op4, itin16,
3058 OpcodeStr, !strconcat(Dt, "8"),
3059 v8i16, v8i8, IntOp, Commutable>;
3062 // ....with explicit extend (VABDL).
3063 multiclass N3VLIntExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3064 InstrItinClass itin, string OpcodeStr, string Dt,
3065 Intrinsic IntOp, SDNode ExtOp, bit Commutable = 0> {
3066 def v8i16 : N3VLIntExt<op24, op23, 0b00, op11_8, op4, itin,
3067 OpcodeStr, !strconcat(Dt, "8"),
3068 v8i16, v8i8, IntOp, ExtOp, Commutable>;
3069 def v4i32 : N3VLIntExt<op24, op23, 0b01, op11_8, op4, itin,
3070 OpcodeStr, !strconcat(Dt, "16"),
3071 v4i32, v4i16, IntOp, ExtOp, Commutable>;
3072 def v2i64 : N3VLIntExt<op24, op23, 0b10, op11_8, op4, itin,
3073 OpcodeStr, !strconcat(Dt, "32"),
3074 v2i64, v2i32, IntOp, ExtOp, Commutable>;
3078 // Neon Wide 3-register vector intrinsics,
3079 // source operand element sizes of 8, 16 and 32 bits:
3080 multiclass N3VW_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3081 string OpcodeStr, string Dt,
3082 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
3083 def v8i16 : N3VW<op24, op23, 0b00, op11_8, op4,
3084 OpcodeStr, !strconcat(Dt, "8"),
3085 v8i16, v8i8, OpNode, ExtOp, Commutable>;
3086 def v4i32 : N3VW<op24, op23, 0b01, op11_8, op4,
3087 OpcodeStr, !strconcat(Dt, "16"),
3088 v4i32, v4i16, OpNode, ExtOp, Commutable>;
3089 def v2i64 : N3VW<op24, op23, 0b10, op11_8, op4,
3090 OpcodeStr, !strconcat(Dt, "32"),
3091 v2i64, v2i32, OpNode, ExtOp, Commutable>;
3095 // Neon Multiply-Op vector operations,
3096 // element sizes of 8, 16 and 32 bits:
3097 multiclass N3VMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3098 InstrItinClass itinD16, InstrItinClass itinD32,
3099 InstrItinClass itinQ16, InstrItinClass itinQ32,
3100 string OpcodeStr, string Dt, SDNode OpNode> {
3101 // 64-bit vector types.
3102 def v8i8 : N3VDMulOp<op24, op23, 0b00, op11_8, op4, itinD16,
3103 OpcodeStr, !strconcat(Dt, "8"), v8i8, mul, OpNode>;
3104 def v4i16 : N3VDMulOp<op24, op23, 0b01, op11_8, op4, itinD16,
3105 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, OpNode>;
3106 def v2i32 : N3VDMulOp<op24, op23, 0b10, op11_8, op4, itinD32,
3107 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, OpNode>;
3109 // 128-bit vector types.
3110 def v16i8 : N3VQMulOp<op24, op23, 0b00, op11_8, op4, itinQ16,
3111 OpcodeStr, !strconcat(Dt, "8"), v16i8, mul, OpNode>;
3112 def v8i16 : N3VQMulOp<op24, op23, 0b01, op11_8, op4, itinQ16,
3113 OpcodeStr, !strconcat(Dt, "16"), v8i16, mul, OpNode>;
3114 def v4i32 : N3VQMulOp<op24, op23, 0b10, op11_8, op4, itinQ32,
3115 OpcodeStr, !strconcat(Dt, "32"), v4i32, mul, OpNode>;
3118 multiclass N3VMulOpSL_HS<bits<4> op11_8,
3119 InstrItinClass itinD16, InstrItinClass itinD32,
3120 InstrItinClass itinQ16, InstrItinClass itinQ32,
3121 string OpcodeStr, string Dt, SDNode ShOp> {
3122 def v4i16 : N3VDMulOpSL16<0b01, op11_8, itinD16,
3123 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, ShOp>;
3124 def v2i32 : N3VDMulOpSL<0b10, op11_8, itinD32,
3125 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, ShOp>;
3126 def v8i16 : N3VQMulOpSL16<0b01, op11_8, itinQ16,
3127 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16,
3129 def v4i32 : N3VQMulOpSL<0b10, op11_8, itinQ32,
3130 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32,
3134 // Neon Intrinsic-Op vector operations,
3135 // element sizes of 8, 16 and 32 bits:
3136 multiclass N3VIntOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3137 InstrItinClass itinD, InstrItinClass itinQ,
3138 string OpcodeStr, string Dt, Intrinsic IntOp,
3140 // 64-bit vector types.
3141 def v8i8 : N3VDIntOp<op24, op23, 0b00, op11_8, op4, itinD,
3142 OpcodeStr, !strconcat(Dt, "8"), v8i8, IntOp, OpNode>;
3143 def v4i16 : N3VDIntOp<op24, op23, 0b01, op11_8, op4, itinD,
3144 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp, OpNode>;
3145 def v2i32 : N3VDIntOp<op24, op23, 0b10, op11_8, op4, itinD,
3146 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp, OpNode>;
3148 // 128-bit vector types.
3149 def v16i8 : N3VQIntOp<op24, op23, 0b00, op11_8, op4, itinQ,
3150 OpcodeStr, !strconcat(Dt, "8"), v16i8, IntOp, OpNode>;
3151 def v8i16 : N3VQIntOp<op24, op23, 0b01, op11_8, op4, itinQ,
3152 OpcodeStr, !strconcat(Dt, "16"), v8i16, IntOp, OpNode>;
3153 def v4i32 : N3VQIntOp<op24, op23, 0b10, op11_8, op4, itinQ,
3154 OpcodeStr, !strconcat(Dt, "32"), v4i32, IntOp, OpNode>;
3157 // Neon 3-argument intrinsics,
3158 // element sizes of 8, 16 and 32 bits:
3159 multiclass N3VInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3160 InstrItinClass itinD, InstrItinClass itinQ,
3161 string OpcodeStr, string Dt, Intrinsic IntOp> {
3162 // 64-bit vector types.
3163 def v8i8 : N3VDInt3<op24, op23, 0b00, op11_8, op4, itinD,
3164 OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
3165 def v4i16 : N3VDInt3<op24, op23, 0b01, op11_8, op4, itinD,
3166 OpcodeStr, !strconcat(Dt, "16"), v4i16, v4i16, IntOp>;
3167 def v2i32 : N3VDInt3<op24, op23, 0b10, op11_8, op4, itinD,
3168 OpcodeStr, !strconcat(Dt, "32"), v2i32, v2i32, IntOp>;
3170 // 128-bit vector types.
3171 def v16i8 : N3VQInt3<op24, op23, 0b00, op11_8, op4, itinQ,
3172 OpcodeStr, !strconcat(Dt, "8"), v16i8, v16i8, IntOp>;
3173 def v8i16 : N3VQInt3<op24, op23, 0b01, op11_8, op4, itinQ,
3174 OpcodeStr, !strconcat(Dt, "16"), v8i16, v8i16, IntOp>;
3175 def v4i32 : N3VQInt3<op24, op23, 0b10, op11_8, op4, itinQ,
3176 OpcodeStr, !strconcat(Dt, "32"), v4i32, v4i32, IntOp>;
3180 // Neon Long Multiply-Op vector operations,
3181 // element sizes of 8, 16 and 32 bits:
3182 multiclass N3VLMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3183 InstrItinClass itin16, InstrItinClass itin32,
3184 string OpcodeStr, string Dt, SDNode MulOp,
3186 def v8i16 : N3VLMulOp<op24, op23, 0b00, op11_8, op4, itin16, OpcodeStr,
3187 !strconcat(Dt, "8"), v8i16, v8i8, MulOp, OpNode>;
3188 def v4i32 : N3VLMulOp<op24, op23, 0b01, op11_8, op4, itin16, OpcodeStr,
3189 !strconcat(Dt, "16"), v4i32, v4i16, MulOp, OpNode>;
3190 def v2i64 : N3VLMulOp<op24, op23, 0b10, op11_8, op4, itin32, OpcodeStr,
3191 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
3194 multiclass N3VLMulOpSL_HS<bit op24, bits<4> op11_8, string OpcodeStr,
3195 string Dt, SDNode MulOp, SDNode OpNode> {
3196 def v4i16 : N3VLMulOpSL16<op24, 0b01, op11_8, IIC_VMACi16D, OpcodeStr,
3197 !strconcat(Dt,"16"), v4i32, v4i16, MulOp, OpNode>;
3198 def v2i32 : N3VLMulOpSL<op24, 0b10, op11_8, IIC_VMACi32D, OpcodeStr,
3199 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
3203 // Neon Long 3-argument intrinsics.
3205 // First with only element sizes of 16 and 32 bits:
3206 multiclass N3VLInt3_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
3207 InstrItinClass itin16, InstrItinClass itin32,
3208 string OpcodeStr, string Dt, Intrinsic IntOp> {
3209 def v4i32 : N3VLInt3<op24, op23, 0b01, op11_8, op4, itin16,
3210 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
3211 def v2i64 : N3VLInt3<op24, op23, 0b10, op11_8, op4, itin32,
3212 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3215 multiclass N3VLInt3SL_HS<bit op24, bits<4> op11_8,
3216 string OpcodeStr, string Dt, Intrinsic IntOp> {
3217 def v4i16 : N3VLInt3SL16<op24, 0b01, op11_8, IIC_VMACi16D,
3218 OpcodeStr, !strconcat(Dt,"16"), v4i32, v4i16, IntOp>;
3219 def v2i32 : N3VLInt3SL<op24, 0b10, op11_8, IIC_VMACi32D,
3220 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3223 // ....then also with element size of 8 bits:
3224 multiclass N3VLInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3225 InstrItinClass itin16, InstrItinClass itin32,
3226 string OpcodeStr, string Dt, Intrinsic IntOp>
3227 : N3VLInt3_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt, IntOp> {
3228 def v8i16 : N3VLInt3<op24, op23, 0b00, op11_8, op4, itin16,
3229 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
3232 // ....with explicit extend (VABAL).
3233 multiclass N3VLIntExtOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3234 InstrItinClass itin, string OpcodeStr, string Dt,
3235 Intrinsic IntOp, SDNode ExtOp, SDNode OpNode> {
3236 def v8i16 : N3VLIntExtOp<op24, op23, 0b00, op11_8, op4, itin,
3237 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8,
3238 IntOp, ExtOp, OpNode>;
3239 def v4i32 : N3VLIntExtOp<op24, op23, 0b01, op11_8, op4, itin,
3240 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16,
3241 IntOp, ExtOp, OpNode>;
3242 def v2i64 : N3VLIntExtOp<op24, op23, 0b10, op11_8, op4, itin,
3243 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32,
3244 IntOp, ExtOp, OpNode>;
3248 // Neon Pairwise long 2-register intrinsics,
3249 // element sizes of 8, 16 and 32 bits:
3250 multiclass N2VPLInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3251 bits<5> op11_7, bit op4,
3252 string OpcodeStr, string Dt, Intrinsic IntOp> {
3253 // 64-bit vector types.
3254 def v8i8 : N2VDPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3255 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
3256 def v4i16 : N2VDPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3257 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
3258 def v2i32 : N2VDPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3259 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
3261 // 128-bit vector types.
3262 def v16i8 : N2VQPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3263 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
3264 def v8i16 : N2VQPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3265 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
3266 def v4i32 : N2VQPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3267 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
3271 // Neon Pairwise long 2-register accumulate intrinsics,
3272 // element sizes of 8, 16 and 32 bits:
3273 multiclass N2VPLInt2_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3274 bits<5> op11_7, bit op4,
3275 string OpcodeStr, string Dt, Intrinsic IntOp> {
3276 // 64-bit vector types.
3277 def v8i8 : N2VDPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3278 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
3279 def v4i16 : N2VDPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3280 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
3281 def v2i32 : N2VDPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3282 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
3284 // 128-bit vector types.
3285 def v16i8 : N2VQPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3286 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
3287 def v8i16 : N2VQPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3288 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
3289 def v4i32 : N2VQPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3290 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
3294 // Neon 2-register vector shift by immediate,
3295 // with f of either N2RegVShLFrm or N2RegVShRFrm
3296 // element sizes of 8, 16, 32 and 64 bits:
3297 multiclass N2VShL_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3298 InstrItinClass itin, string OpcodeStr, string Dt,
3300 // 64-bit vector types.
3301 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3302 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
3303 let Inst{21-19} = 0b001; // imm6 = 001xxx
3305 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3306 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
3307 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3309 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3310 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
3311 let Inst{21} = 0b1; // imm6 = 1xxxxx
3313 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, N2RegVShLFrm, itin, i32imm,
3314 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
3317 // 128-bit vector types.
3318 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3319 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
3320 let Inst{21-19} = 0b001; // imm6 = 001xxx
3322 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3323 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
3324 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3326 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3327 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
3328 let Inst{21} = 0b1; // imm6 = 1xxxxx
3330 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, N2RegVShLFrm, itin, i32imm,
3331 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
3334 multiclass N2VShR_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3335 InstrItinClass itin, string OpcodeStr, string Dt,
3337 // 64-bit vector types.
3338 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm8,
3339 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
3340 let Inst{21-19} = 0b001; // imm6 = 001xxx
3342 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm16,
3343 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
3344 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3346 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm32,
3347 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
3348 let Inst{21} = 0b1; // imm6 = 1xxxxx
3350 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, N2RegVShRFrm, itin, shr_imm64,
3351 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
3354 // 128-bit vector types.
3355 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm8,
3356 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
3357 let Inst{21-19} = 0b001; // imm6 = 001xxx
3359 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm16,
3360 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
3361 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3363 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm32,
3364 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
3365 let Inst{21} = 0b1; // imm6 = 1xxxxx
3367 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, N2RegVShRFrm, itin, shr_imm64,
3368 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
3372 // Neon Shift-Accumulate vector operations,
3373 // element sizes of 8, 16, 32 and 64 bits:
3374 multiclass N2VShAdd_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3375 string OpcodeStr, string Dt, SDNode ShOp> {
3376 // 64-bit vector types.
3377 def v8i8 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm8,
3378 OpcodeStr, !strconcat(Dt, "8"), v8i8, ShOp> {
3379 let Inst{21-19} = 0b001; // imm6 = 001xxx
3381 def v4i16 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm16,
3382 OpcodeStr, !strconcat(Dt, "16"), v4i16, ShOp> {
3383 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3385 def v2i32 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm32,
3386 OpcodeStr, !strconcat(Dt, "32"), v2i32, ShOp> {
3387 let Inst{21} = 0b1; // imm6 = 1xxxxx
3389 def v1i64 : N2VDShAdd<op24, op23, op11_8, 1, op4, shr_imm64,
3390 OpcodeStr, !strconcat(Dt, "64"), v1i64, ShOp>;
3393 // 128-bit vector types.
3394 def v16i8 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm8,
3395 OpcodeStr, !strconcat(Dt, "8"), v16i8, ShOp> {
3396 let Inst{21-19} = 0b001; // imm6 = 001xxx
3398 def v8i16 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm16,
3399 OpcodeStr, !strconcat(Dt, "16"), v8i16, ShOp> {
3400 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3402 def v4i32 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm32,
3403 OpcodeStr, !strconcat(Dt, "32"), v4i32, ShOp> {
3404 let Inst{21} = 0b1; // imm6 = 1xxxxx
3406 def v2i64 : N2VQShAdd<op24, op23, op11_8, 1, op4, shr_imm64,
3407 OpcodeStr, !strconcat(Dt, "64"), v2i64, ShOp>;
3411 // Neon Shift-Insert vector operations,
3412 // with f of either N2RegVShLFrm or N2RegVShRFrm
3413 // element sizes of 8, 16, 32 and 64 bits:
3414 multiclass N2VShInsL_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3416 // 64-bit vector types.
3417 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3418 N2RegVShLFrm, OpcodeStr, "8", v8i8, NEONvsli> {
3419 let Inst{21-19} = 0b001; // imm6 = 001xxx
3421 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3422 N2RegVShLFrm, OpcodeStr, "16", v4i16, NEONvsli> {
3423 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3425 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3426 N2RegVShLFrm, OpcodeStr, "32", v2i32, NEONvsli> {
3427 let Inst{21} = 0b1; // imm6 = 1xxxxx
3429 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4, i32imm,
3430 N2RegVShLFrm, OpcodeStr, "64", v1i64, NEONvsli>;
3433 // 128-bit vector types.
3434 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3435 N2RegVShLFrm, OpcodeStr, "8", v16i8, NEONvsli> {
3436 let Inst{21-19} = 0b001; // imm6 = 001xxx
3438 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3439 N2RegVShLFrm, OpcodeStr, "16", v8i16, NEONvsli> {
3440 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3442 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3443 N2RegVShLFrm, OpcodeStr, "32", v4i32, NEONvsli> {
3444 let Inst{21} = 0b1; // imm6 = 1xxxxx
3446 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4, i32imm,
3447 N2RegVShLFrm, OpcodeStr, "64", v2i64, NEONvsli>;
3450 multiclass N2VShInsR_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3452 // 64-bit vector types.
3453 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm8,
3454 N2RegVShRFrm, OpcodeStr, "8", v8i8, NEONvsri> {
3455 let Inst{21-19} = 0b001; // imm6 = 001xxx
3457 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm16,
3458 N2RegVShRFrm, OpcodeStr, "16", v4i16, NEONvsri> {
3459 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3461 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm32,
3462 N2RegVShRFrm, OpcodeStr, "32", v2i32, NEONvsri> {
3463 let Inst{21} = 0b1; // imm6 = 1xxxxx
3465 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4, shr_imm64,
3466 N2RegVShRFrm, OpcodeStr, "64", v1i64, NEONvsri>;
3469 // 128-bit vector types.
3470 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm8,
3471 N2RegVShRFrm, OpcodeStr, "8", v16i8, NEONvsri> {
3472 let Inst{21-19} = 0b001; // imm6 = 001xxx
3474 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm16,
3475 N2RegVShRFrm, OpcodeStr, "16", v8i16, NEONvsri> {
3476 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3478 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm32,
3479 N2RegVShRFrm, OpcodeStr, "32", v4i32, NEONvsri> {
3480 let Inst{21} = 0b1; // imm6 = 1xxxxx
3482 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4, shr_imm64,
3483 N2RegVShRFrm, OpcodeStr, "64", v2i64, NEONvsri>;
3487 // Neon Shift Long operations,
3488 // element sizes of 8, 16, 32 bits:
3489 multiclass N2VLSh_QHS<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
3490 bit op4, string OpcodeStr, string Dt, SDNode OpNode> {
3491 def v8i16 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3492 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode> {
3493 let Inst{21-19} = 0b001; // imm6 = 001xxx
3495 def v4i32 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3496 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode> {
3497 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3499 def v2i64 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3500 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode> {
3501 let Inst{21} = 0b1; // imm6 = 1xxxxx
3505 // Neon Shift Narrow operations,
3506 // element sizes of 16, 32, 64 bits:
3507 multiclass N2VNSh_HSD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
3508 bit op4, InstrItinClass itin, string OpcodeStr, string Dt,
3510 def v8i8 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3511 OpcodeStr, !strconcat(Dt, "16"),
3512 v8i8, v8i16, shr_imm8, OpNode> {
3513 let Inst{21-19} = 0b001; // imm6 = 001xxx
3515 def v4i16 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3516 OpcodeStr, !strconcat(Dt, "32"),
3517 v4i16, v4i32, shr_imm16, OpNode> {
3518 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3520 def v2i32 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3521 OpcodeStr, !strconcat(Dt, "64"),
3522 v2i32, v2i64, shr_imm32, OpNode> {
3523 let Inst{21} = 0b1; // imm6 = 1xxxxx
3527 //===----------------------------------------------------------------------===//
3528 // Instruction Definitions.
3529 //===----------------------------------------------------------------------===//
3531 // Vector Add Operations.
3533 // VADD : Vector Add (integer and floating-point)
3534 defm VADD : N3V_QHSD<0, 0, 0b1000, 0, IIC_VBINiD, IIC_VBINiQ, "vadd", "i",
3536 def VADDfd : N3VD<0, 0, 0b00, 0b1101, 0, IIC_VBIND, "vadd", "f32",
3537 v2f32, v2f32, fadd, 1>;
3538 def VADDfq : N3VQ<0, 0, 0b00, 0b1101, 0, IIC_VBINQ, "vadd", "f32",
3539 v4f32, v4f32, fadd, 1>;
3540 // VADDL : Vector Add Long (Q = D + D)
3541 defm VADDLs : N3VLExt_QHS<0,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
3542 "vaddl", "s", add, sext, 1>;
3543 defm VADDLu : N3VLExt_QHS<1,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
3544 "vaddl", "u", add, zext, 1>;
3545 // VADDW : Vector Add Wide (Q = Q + D)
3546 defm VADDWs : N3VW_QHS<0,1,0b0001,0, "vaddw", "s", add, sext, 0>;
3547 defm VADDWu : N3VW_QHS<1,1,0b0001,0, "vaddw", "u", add, zext, 0>;
3548 // VHADD : Vector Halving Add
3549 defm VHADDs : N3VInt_QHS<0, 0, 0b0000, 0, N3RegFrm,
3550 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3551 "vhadd", "s", int_arm_neon_vhadds, 1>;
3552 defm VHADDu : N3VInt_QHS<1, 0, 0b0000, 0, N3RegFrm,
3553 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3554 "vhadd", "u", int_arm_neon_vhaddu, 1>;
3555 // VRHADD : Vector Rounding Halving Add
3556 defm VRHADDs : N3VInt_QHS<0, 0, 0b0001, 0, N3RegFrm,
3557 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3558 "vrhadd", "s", int_arm_neon_vrhadds, 1>;
3559 defm VRHADDu : N3VInt_QHS<1, 0, 0b0001, 0, N3RegFrm,
3560 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3561 "vrhadd", "u", int_arm_neon_vrhaddu, 1>;
3562 // VQADD : Vector Saturating Add
3563 defm VQADDs : N3VInt_QHSD<0, 0, 0b0000, 1, N3RegFrm,
3564 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3565 "vqadd", "s", int_arm_neon_vqadds, 1>;
3566 defm VQADDu : N3VInt_QHSD<1, 0, 0b0000, 1, N3RegFrm,
3567 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3568 "vqadd", "u", int_arm_neon_vqaddu, 1>;
3569 // VADDHN : Vector Add and Narrow Returning High Half (D = Q + Q)
3570 defm VADDHN : N3VNInt_HSD<0,1,0b0100,0, "vaddhn", "i",
3571 int_arm_neon_vaddhn, 1>;
3572 // VRADDHN : Vector Rounding Add and Narrow Returning High Half (D = Q + Q)
3573 defm VRADDHN : N3VNInt_HSD<1,1,0b0100,0, "vraddhn", "i",
3574 int_arm_neon_vraddhn, 1>;
3576 // Vector Multiply Operations.
3578 // VMUL : Vector Multiply (integer, polynomial and floating-point)
3579 defm VMUL : N3V_QHS<0, 0, 0b1001, 1, IIC_VMULi16D, IIC_VMULi32D,
3580 IIC_VMULi16Q, IIC_VMULi32Q, "vmul", "i", mul, 1>;
3581 def VMULpd : N3VDInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16D, "vmul",
3582 "p8", v8i8, v8i8, int_arm_neon_vmulp, 1>;
3583 def VMULpq : N3VQInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16Q, "vmul",
3584 "p8", v16i8, v16i8, int_arm_neon_vmulp, 1>;
3585 def VMULfd : N3VD<1, 0, 0b00, 0b1101, 1, IIC_VFMULD, "vmul", "f32",
3586 v2f32, v2f32, fmul, 1>;
3587 def VMULfq : N3VQ<1, 0, 0b00, 0b1101, 1, IIC_VFMULQ, "vmul", "f32",
3588 v4f32, v4f32, fmul, 1>;
3589 defm VMULsl : N3VSL_HS<0b1000, "vmul", "i", mul>;
3590 def VMULslfd : N3VDSL<0b10, 0b1001, IIC_VBIND, "vmul", "f32", v2f32, fmul>;
3591 def VMULslfq : N3VQSL<0b10, 0b1001, IIC_VBINQ, "vmul", "f32", v4f32,
3594 def : Pat<(v8i16 (mul (v8i16 QPR:$src1),
3595 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
3596 (v8i16 (VMULslv8i16 (v8i16 QPR:$src1),
3597 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3598 (DSubReg_i16_reg imm:$lane))),
3599 (SubReg_i16_lane imm:$lane)))>;
3600 def : Pat<(v4i32 (mul (v4i32 QPR:$src1),
3601 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
3602 (v4i32 (VMULslv4i32 (v4i32 QPR:$src1),
3603 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3604 (DSubReg_i32_reg imm:$lane))),
3605 (SubReg_i32_lane imm:$lane)))>;
3606 def : Pat<(v4f32 (fmul (v4f32 QPR:$src1),
3607 (v4f32 (NEONvduplane (v4f32 QPR:$src2), imm:$lane)))),
3608 (v4f32 (VMULslfq (v4f32 QPR:$src1),
3609 (v2f32 (EXTRACT_SUBREG QPR:$src2,
3610 (DSubReg_i32_reg imm:$lane))),
3611 (SubReg_i32_lane imm:$lane)))>;
3613 // VQDMULH : Vector Saturating Doubling Multiply Returning High Half
3614 defm VQDMULH : N3VInt_HS<0, 0, 0b1011, 0, N3RegFrm, IIC_VMULi16D, IIC_VMULi32D,
3615 IIC_VMULi16Q, IIC_VMULi32Q,
3616 "vqdmulh", "s", int_arm_neon_vqdmulh, 1>;
3617 defm VQDMULHsl: N3VIntSL_HS<0b1100, IIC_VMULi16D, IIC_VMULi32D,
3618 IIC_VMULi16Q, IIC_VMULi32Q,
3619 "vqdmulh", "s", int_arm_neon_vqdmulh>;
3620 def : Pat<(v8i16 (int_arm_neon_vqdmulh (v8i16 QPR:$src1),
3621 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
3623 (v8i16 (VQDMULHslv8i16 (v8i16 QPR:$src1),
3624 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3625 (DSubReg_i16_reg imm:$lane))),
3626 (SubReg_i16_lane imm:$lane)))>;
3627 def : Pat<(v4i32 (int_arm_neon_vqdmulh (v4i32 QPR:$src1),
3628 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
3630 (v4i32 (VQDMULHslv4i32 (v4i32 QPR:$src1),
3631 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3632 (DSubReg_i32_reg imm:$lane))),
3633 (SubReg_i32_lane imm:$lane)))>;
3635 // VQRDMULH : Vector Rounding Saturating Doubling Multiply Returning High Half
3636 defm VQRDMULH : N3VInt_HS<1, 0, 0b1011, 0, N3RegFrm,
3637 IIC_VMULi16D,IIC_VMULi32D,IIC_VMULi16Q,IIC_VMULi32Q,
3638 "vqrdmulh", "s", int_arm_neon_vqrdmulh, 1>;
3639 defm VQRDMULHsl : N3VIntSL_HS<0b1101, IIC_VMULi16D, IIC_VMULi32D,
3640 IIC_VMULi16Q, IIC_VMULi32Q,
3641 "vqrdmulh", "s", int_arm_neon_vqrdmulh>;
3642 def : Pat<(v8i16 (int_arm_neon_vqrdmulh (v8i16 QPR:$src1),
3643 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
3645 (v8i16 (VQRDMULHslv8i16 (v8i16 QPR:$src1),
3646 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3647 (DSubReg_i16_reg imm:$lane))),
3648 (SubReg_i16_lane imm:$lane)))>;
3649 def : Pat<(v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$src1),
3650 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
3652 (v4i32 (VQRDMULHslv4i32 (v4i32 QPR:$src1),
3653 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3654 (DSubReg_i32_reg imm:$lane))),
3655 (SubReg_i32_lane imm:$lane)))>;
3657 // VMULL : Vector Multiply Long (integer and polynomial) (Q = D * D)
3658 defm VMULLs : N3VL_QHS<0,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
3659 "vmull", "s", NEONvmulls, 1>;
3660 defm VMULLu : N3VL_QHS<1,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
3661 "vmull", "u", NEONvmullu, 1>;
3662 def VMULLp : N3VLInt<0, 1, 0b00, 0b1110, 0, IIC_VMULi16D, "vmull", "p8",
3663 v8i16, v8i8, int_arm_neon_vmullp, 1>;
3664 defm VMULLsls : N3VLSL_HS<0, 0b1010, IIC_VMULi16D, "vmull", "s", NEONvmulls>;
3665 defm VMULLslu : N3VLSL_HS<1, 0b1010, IIC_VMULi16D, "vmull", "u", NEONvmullu>;
3667 // VQDMULL : Vector Saturating Doubling Multiply Long (Q = D * D)
3668 defm VQDMULL : N3VLInt_HS<0,1,0b1101,0, IIC_VMULi16D, IIC_VMULi32D,
3669 "vqdmull", "s", int_arm_neon_vqdmull, 1>;
3670 defm VQDMULLsl: N3VLIntSL_HS<0, 0b1011, IIC_VMULi16D,
3671 "vqdmull", "s", int_arm_neon_vqdmull>;
3673 // Vector Multiply-Accumulate and Multiply-Subtract Operations.
3675 // VMLA : Vector Multiply Accumulate (integer and floating-point)
3676 defm VMLA : N3VMulOp_QHS<0, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3677 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
3678 def VMLAfd : N3VDMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACD, "vmla", "f32",
3679 v2f32, fmul_su, fadd_mlx>,
3680 Requires<[HasNEON, UseFPVMLx]>;
3681 def VMLAfq : N3VQMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACQ, "vmla", "f32",
3682 v4f32, fmul_su, fadd_mlx>,
3683 Requires<[HasNEON, UseFPVMLx]>;
3684 defm VMLAsl : N3VMulOpSL_HS<0b0000, IIC_VMACi16D, IIC_VMACi32D,
3685 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
3686 def VMLAslfd : N3VDMulOpSL<0b10, 0b0001, IIC_VMACD, "vmla", "f32",
3687 v2f32, fmul_su, fadd_mlx>,
3688 Requires<[HasNEON, UseFPVMLx]>;
3689 def VMLAslfq : N3VQMulOpSL<0b10, 0b0001, IIC_VMACQ, "vmla", "f32",
3690 v4f32, v2f32, fmul_su, fadd_mlx>,
3691 Requires<[HasNEON, UseFPVMLx]>;
3693 def : Pat<(v8i16 (add (v8i16 QPR:$src1),
3694 (mul (v8i16 QPR:$src2),
3695 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
3696 (v8i16 (VMLAslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
3697 (v4i16 (EXTRACT_SUBREG QPR:$src3,
3698 (DSubReg_i16_reg imm:$lane))),
3699 (SubReg_i16_lane imm:$lane)))>;
3701 def : Pat<(v4i32 (add (v4i32 QPR:$src1),
3702 (mul (v4i32 QPR:$src2),
3703 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
3704 (v4i32 (VMLAslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
3705 (v2i32 (EXTRACT_SUBREG QPR:$src3,
3706 (DSubReg_i32_reg imm:$lane))),
3707 (SubReg_i32_lane imm:$lane)))>;
3709 def : Pat<(v4f32 (fadd_mlx (v4f32 QPR:$src1),
3710 (fmul_su (v4f32 QPR:$src2),
3711 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
3712 (v4f32 (VMLAslfq (v4f32 QPR:$src1),
3714 (v2f32 (EXTRACT_SUBREG QPR:$src3,
3715 (DSubReg_i32_reg imm:$lane))),
3716 (SubReg_i32_lane imm:$lane)))>,
3717 Requires<[HasNEON, UseFPVMLx]>;
3719 // VMLAL : Vector Multiply Accumulate Long (Q += D * D)
3720 defm VMLALs : N3VLMulOp_QHS<0,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
3721 "vmlal", "s", NEONvmulls, add>;
3722 defm VMLALu : N3VLMulOp_QHS<1,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
3723 "vmlal", "u", NEONvmullu, add>;
3725 defm VMLALsls : N3VLMulOpSL_HS<0, 0b0010, "vmlal", "s", NEONvmulls, add>;
3726 defm VMLALslu : N3VLMulOpSL_HS<1, 0b0010, "vmlal", "u", NEONvmullu, add>;
3728 // VQDMLAL : Vector Saturating Doubling Multiply Accumulate Long (Q += D * D)
3729 defm VQDMLAL : N3VLInt3_HS<0, 1, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3730 "vqdmlal", "s", int_arm_neon_vqdmlal>;
3731 defm VQDMLALsl: N3VLInt3SL_HS<0, 0b0011, "vqdmlal", "s", int_arm_neon_vqdmlal>;
3733 // VMLS : Vector Multiply Subtract (integer and floating-point)
3734 defm VMLS : N3VMulOp_QHS<1, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3735 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
3736 def VMLSfd : N3VDMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACD, "vmls", "f32",
3737 v2f32, fmul_su, fsub_mlx>,
3738 Requires<[HasNEON, UseFPVMLx]>;
3739 def VMLSfq : N3VQMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACQ, "vmls", "f32",
3740 v4f32, fmul_su, fsub_mlx>,
3741 Requires<[HasNEON, UseFPVMLx]>;
3742 defm VMLSsl : N3VMulOpSL_HS<0b0100, IIC_VMACi16D, IIC_VMACi32D,
3743 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
3744 def VMLSslfd : N3VDMulOpSL<0b10, 0b0101, IIC_VMACD, "vmls", "f32",
3745 v2f32, fmul_su, fsub_mlx>,
3746 Requires<[HasNEON, UseFPVMLx]>;
3747 def VMLSslfq : N3VQMulOpSL<0b10, 0b0101, IIC_VMACQ, "vmls", "f32",
3748 v4f32, v2f32, fmul_su, fsub_mlx>,
3749 Requires<[HasNEON, UseFPVMLx]>;
3751 def : Pat<(v8i16 (sub (v8i16 QPR:$src1),
3752 (mul (v8i16 QPR:$src2),
3753 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
3754 (v8i16 (VMLSslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
3755 (v4i16 (EXTRACT_SUBREG QPR:$src3,
3756 (DSubReg_i16_reg imm:$lane))),
3757 (SubReg_i16_lane imm:$lane)))>;
3759 def : Pat<(v4i32 (sub (v4i32 QPR:$src1),
3760 (mul (v4i32 QPR:$src2),
3761 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
3762 (v4i32 (VMLSslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
3763 (v2i32 (EXTRACT_SUBREG QPR:$src3,
3764 (DSubReg_i32_reg imm:$lane))),
3765 (SubReg_i32_lane imm:$lane)))>;
3767 def : Pat<(v4f32 (fsub_mlx (v4f32 QPR:$src1),
3768 (fmul_su (v4f32 QPR:$src2),
3769 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
3770 (v4f32 (VMLSslfq (v4f32 QPR:$src1), (v4f32 QPR:$src2),
3771 (v2f32 (EXTRACT_SUBREG QPR:$src3,
3772 (DSubReg_i32_reg imm:$lane))),
3773 (SubReg_i32_lane imm:$lane)))>,
3774 Requires<[HasNEON, UseFPVMLx]>;
3776 // VMLSL : Vector Multiply Subtract Long (Q -= D * D)
3777 defm VMLSLs : N3VLMulOp_QHS<0,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
3778 "vmlsl", "s", NEONvmulls, sub>;
3779 defm VMLSLu : N3VLMulOp_QHS<1,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
3780 "vmlsl", "u", NEONvmullu, sub>;
3782 defm VMLSLsls : N3VLMulOpSL_HS<0, 0b0110, "vmlsl", "s", NEONvmulls, sub>;
3783 defm VMLSLslu : N3VLMulOpSL_HS<1, 0b0110, "vmlsl", "u", NEONvmullu, sub>;
3785 // VQDMLSL : Vector Saturating Doubling Multiply Subtract Long (Q -= D * D)
3786 defm VQDMLSL : N3VLInt3_HS<0, 1, 0b1011, 0, IIC_VMACi16D, IIC_VMACi32D,
3787 "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
3788 defm VQDMLSLsl: N3VLInt3SL_HS<0, 0b111, "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
3790 // Vector Subtract Operations.
3792 // VSUB : Vector Subtract (integer and floating-point)
3793 defm VSUB : N3V_QHSD<1, 0, 0b1000, 0, IIC_VSUBiD, IIC_VSUBiQ,
3794 "vsub", "i", sub, 0>;
3795 def VSUBfd : N3VD<0, 0, 0b10, 0b1101, 0, IIC_VBIND, "vsub", "f32",
3796 v2f32, v2f32, fsub, 0>;
3797 def VSUBfq : N3VQ<0, 0, 0b10, 0b1101, 0, IIC_VBINQ, "vsub", "f32",
3798 v4f32, v4f32, fsub, 0>;
3799 // VSUBL : Vector Subtract Long (Q = D - D)
3800 defm VSUBLs : N3VLExt_QHS<0,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
3801 "vsubl", "s", sub, sext, 0>;
3802 defm VSUBLu : N3VLExt_QHS<1,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
3803 "vsubl", "u", sub, zext, 0>;
3804 // VSUBW : Vector Subtract Wide (Q = Q - D)
3805 defm VSUBWs : N3VW_QHS<0,1,0b0011,0, "vsubw", "s", sub, sext, 0>;
3806 defm VSUBWu : N3VW_QHS<1,1,0b0011,0, "vsubw", "u", sub, zext, 0>;
3807 // VHSUB : Vector Halving Subtract
3808 defm VHSUBs : N3VInt_QHS<0, 0, 0b0010, 0, N3RegFrm,
3809 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3810 "vhsub", "s", int_arm_neon_vhsubs, 0>;
3811 defm VHSUBu : N3VInt_QHS<1, 0, 0b0010, 0, N3RegFrm,
3812 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3813 "vhsub", "u", int_arm_neon_vhsubu, 0>;
3814 // VQSUB : Vector Saturing Subtract
3815 defm VQSUBs : N3VInt_QHSD<0, 0, 0b0010, 1, N3RegFrm,
3816 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3817 "vqsub", "s", int_arm_neon_vqsubs, 0>;
3818 defm VQSUBu : N3VInt_QHSD<1, 0, 0b0010, 1, N3RegFrm,
3819 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3820 "vqsub", "u", int_arm_neon_vqsubu, 0>;
3821 // VSUBHN : Vector Subtract and Narrow Returning High Half (D = Q - Q)
3822 defm VSUBHN : N3VNInt_HSD<0,1,0b0110,0, "vsubhn", "i",
3823 int_arm_neon_vsubhn, 0>;
3824 // VRSUBHN : Vector Rounding Subtract and Narrow Returning High Half (D=Q-Q)
3825 defm VRSUBHN : N3VNInt_HSD<1,1,0b0110,0, "vrsubhn", "i",
3826 int_arm_neon_vrsubhn, 0>;
3828 // Vector Comparisons.
3830 // VCEQ : Vector Compare Equal
3831 defm VCEQ : N3V_QHS<1, 0, 0b1000, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3832 IIC_VSUBi4Q, "vceq", "i", NEONvceq, 1>;
3833 def VCEQfd : N3VD<0,0,0b00,0b1110,0, IIC_VBIND, "vceq", "f32", v2i32, v2f32,
3835 def VCEQfq : N3VQ<0,0,0b00,0b1110,0, IIC_VBINQ, "vceq", "f32", v4i32, v4f32,
3838 defm VCEQz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00010, 0, "vceq", "i",
3839 "$Vd, $Vm, #0", NEONvceqz>;
3841 // VCGE : Vector Compare Greater Than or Equal
3842 defm VCGEs : N3V_QHS<0, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3843 IIC_VSUBi4Q, "vcge", "s", NEONvcge, 0>;
3844 defm VCGEu : N3V_QHS<1, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3845 IIC_VSUBi4Q, "vcge", "u", NEONvcgeu, 0>;
3846 def VCGEfd : N3VD<1,0,0b00,0b1110,0, IIC_VBIND, "vcge", "f32", v2i32, v2f32,
3848 def VCGEfq : N3VQ<1,0,0b00,0b1110,0, IIC_VBINQ, "vcge", "f32", v4i32, v4f32,
3851 defm VCGEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00001, 0, "vcge", "s",
3852 "$Vd, $Vm, #0", NEONvcgez>;
3853 defm VCLEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00011, 0, "vcle", "s",
3854 "$Vd, $Vm, #0", NEONvclez>;
3856 // VCGT : Vector Compare Greater Than
3857 defm VCGTs : N3V_QHS<0, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3858 IIC_VSUBi4Q, "vcgt", "s", NEONvcgt, 0>;
3859 defm VCGTu : N3V_QHS<1, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3860 IIC_VSUBi4Q, "vcgt", "u", NEONvcgtu, 0>;
3861 def VCGTfd : N3VD<1,0,0b10,0b1110,0, IIC_VBIND, "vcgt", "f32", v2i32, v2f32,
3863 def VCGTfq : N3VQ<1,0,0b10,0b1110,0, IIC_VBINQ, "vcgt", "f32", v4i32, v4f32,
3866 defm VCGTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00000, 0, "vcgt", "s",
3867 "$Vd, $Vm, #0", NEONvcgtz>;
3868 defm VCLTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00100, 0, "vclt", "s",
3869 "$Vd, $Vm, #0", NEONvcltz>;
3871 // VACGE : Vector Absolute Compare Greater Than or Equal (aka VCAGE)
3872 def VACGEd : N3VDInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacge",
3873 "f32", v2i32, v2f32, int_arm_neon_vacged, 0>;
3874 def VACGEq : N3VQInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacge",
3875 "f32", v4i32, v4f32, int_arm_neon_vacgeq, 0>;
3876 // VACGT : Vector Absolute Compare Greater Than (aka VCAGT)
3877 def VACGTd : N3VDInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacgt",
3878 "f32", v2i32, v2f32, int_arm_neon_vacgtd, 0>;
3879 def VACGTq : N3VQInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacgt",
3880 "f32", v4i32, v4f32, int_arm_neon_vacgtq, 0>;
3881 // VTST : Vector Test Bits
3882 defm VTST : N3V_QHS<0, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
3883 IIC_VBINi4Q, "vtst", "", NEONvtst, 1>;
3885 // Vector Bitwise Operations.
3887 def vnotd : PatFrag<(ops node:$in),
3888 (xor node:$in, (bitconvert (v8i8 NEONimmAllOnesV)))>;
3889 def vnotq : PatFrag<(ops node:$in),
3890 (xor node:$in, (bitconvert (v16i8 NEONimmAllOnesV)))>;
3893 // VAND : Vector Bitwise AND
3894 def VANDd : N3VDX<0, 0, 0b00, 0b0001, 1, IIC_VBINiD, "vand",
3895 v2i32, v2i32, and, 1>;
3896 def VANDq : N3VQX<0, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "vand",
3897 v4i32, v4i32, and, 1>;
3899 // VEOR : Vector Bitwise Exclusive OR
3900 def VEORd : N3VDX<1, 0, 0b00, 0b0001, 1, IIC_VBINiD, "veor",
3901 v2i32, v2i32, xor, 1>;
3902 def VEORq : N3VQX<1, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "veor",
3903 v4i32, v4i32, xor, 1>;
3905 // VORR : Vector Bitwise OR
3906 def VORRd : N3VDX<0, 0, 0b10, 0b0001, 1, IIC_VBINiD, "vorr",
3907 v2i32, v2i32, or, 1>;
3908 def VORRq : N3VQX<0, 0, 0b10, 0b0001, 1, IIC_VBINiQ, "vorr",
3909 v4i32, v4i32, or, 1>;
3911 def VORRiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 0, 1,
3912 (outs DPR:$Vd), (ins nImmSplatI16:$SIMM, DPR:$src),
3914 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
3916 (v4i16 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
3917 let Inst{9} = SIMM{9};
3920 def VORRiv2i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 0, 0, 1,
3921 (outs DPR:$Vd), (ins nImmSplatI32:$SIMM, DPR:$src),
3923 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
3925 (v2i32 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
3926 let Inst{10-9} = SIMM{10-9};
3929 def VORRiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 0, 1,
3930 (outs QPR:$Vd), (ins nImmSplatI16:$SIMM, QPR:$src),
3932 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
3934 (v8i16 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
3935 let Inst{9} = SIMM{9};
3938 def VORRiv4i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 1, 0, 1,
3939 (outs QPR:$Vd), (ins nImmSplatI32:$SIMM, QPR:$src),
3941 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
3943 (v4i32 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
3944 let Inst{10-9} = SIMM{10-9};
3948 // VBIC : Vector Bitwise Bit Clear (AND NOT)
3949 def VBICd : N3VX<0, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
3950 (ins DPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VBINiD,
3951 "vbic", "$Vd, $Vn, $Vm", "",
3952 [(set DPR:$Vd, (v2i32 (and DPR:$Vn,
3953 (vnotd DPR:$Vm))))]>;
3954 def VBICq : N3VX<0, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
3955 (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINiQ,
3956 "vbic", "$Vd, $Vn, $Vm", "",
3957 [(set QPR:$Vd, (v4i32 (and QPR:$Vn,
3958 (vnotq QPR:$Vm))))]>;
3960 def VBICiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 1, 1,
3961 (outs DPR:$Vd), (ins nImmSplatI16:$SIMM, DPR:$src),
3963 "vbic", "i16", "$Vd, $SIMM", "$src = $Vd",
3965 (v4i16 (NEONvbicImm DPR:$src, timm:$SIMM)))]> {
3966 let Inst{9} = SIMM{9};
3969 def VBICiv2i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 0, 1, 1,
3970 (outs DPR:$Vd), (ins nImmSplatI32:$SIMM, DPR:$src),
3972 "vbic", "i32", "$Vd, $SIMM", "$src = $Vd",
3974 (v2i32 (NEONvbicImm DPR:$src, timm:$SIMM)))]> {
3975 let Inst{10-9} = SIMM{10-9};
3978 def VBICiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 1, 1,
3979 (outs QPR:$Vd), (ins nImmSplatI16:$SIMM, QPR:$src),
3981 "vbic", "i16", "$Vd, $SIMM", "$src = $Vd",
3983 (v8i16 (NEONvbicImm QPR:$src, timm:$SIMM)))]> {
3984 let Inst{9} = SIMM{9};
3987 def VBICiv4i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 1, 1, 1,
3988 (outs QPR:$Vd), (ins nImmSplatI32:$SIMM, QPR:$src),
3990 "vbic", "i32", "$Vd, $SIMM", "$src = $Vd",
3992 (v4i32 (NEONvbicImm QPR:$src, timm:$SIMM)))]> {
3993 let Inst{10-9} = SIMM{10-9};
3996 // VORN : Vector Bitwise OR NOT
3997 def VORNd : N3VX<0, 0, 0b11, 0b0001, 0, 1, (outs DPR:$Vd),
3998 (ins DPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VBINiD,
3999 "vorn", "$Vd, $Vn, $Vm", "",
4000 [(set DPR:$Vd, (v2i32 (or DPR:$Vn,
4001 (vnotd DPR:$Vm))))]>;
4002 def VORNq : N3VX<0, 0, 0b11, 0b0001, 1, 1, (outs QPR:$Vd),
4003 (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINiQ,
4004 "vorn", "$Vd, $Vn, $Vm", "",
4005 [(set QPR:$Vd, (v4i32 (or QPR:$Vn,
4006 (vnotq QPR:$Vm))))]>;
4008 // VMVN : Vector Bitwise NOT (Immediate)
4010 let isReMaterializable = 1 in {
4012 def VMVNv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 1, 1, (outs DPR:$Vd),
4013 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4014 "vmvn", "i16", "$Vd, $SIMM", "",
4015 [(set DPR:$Vd, (v4i16 (NEONvmvnImm timm:$SIMM)))]> {
4016 let Inst{9} = SIMM{9};
4019 def VMVNv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 1, 1, (outs QPR:$Vd),
4020 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4021 "vmvn", "i16", "$Vd, $SIMM", "",
4022 [(set QPR:$Vd, (v8i16 (NEONvmvnImm timm:$SIMM)))]> {
4023 let Inst{9} = SIMM{9};
4026 def VMVNv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 1, 1, (outs DPR:$Vd),
4027 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4028 "vmvn", "i32", "$Vd, $SIMM", "",
4029 [(set DPR:$Vd, (v2i32 (NEONvmvnImm timm:$SIMM)))]> {
4030 let Inst{11-8} = SIMM{11-8};
4033 def VMVNv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 1, 1, (outs QPR:$Vd),
4034 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4035 "vmvn", "i32", "$Vd, $SIMM", "",
4036 [(set QPR:$Vd, (v4i32 (NEONvmvnImm timm:$SIMM)))]> {
4037 let Inst{11-8} = SIMM{11-8};
4041 // VMVN : Vector Bitwise NOT
4042 def VMVNd : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 0, 0,
4043 (outs DPR:$Vd), (ins DPR:$Vm), IIC_VSUBiD,
4044 "vmvn", "$Vd, $Vm", "",
4045 [(set DPR:$Vd, (v2i32 (vnotd DPR:$Vm)))]>;
4046 def VMVNq : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 1, 0,
4047 (outs QPR:$Vd), (ins QPR:$Vm), IIC_VSUBiD,
4048 "vmvn", "$Vd, $Vm", "",
4049 [(set QPR:$Vd, (v4i32 (vnotq QPR:$Vm)))]>;
4050 def : Pat<(v2i32 (vnotd DPR:$src)), (VMVNd DPR:$src)>;
4051 def : Pat<(v4i32 (vnotq QPR:$src)), (VMVNq QPR:$src)>;
4053 // VBSL : Vector Bitwise Select
4054 def VBSLd : N3VX<1, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
4055 (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
4056 N3RegFrm, IIC_VCNTiD,
4057 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4059 (v2i32 (NEONvbsl DPR:$src1, DPR:$Vn, DPR:$Vm)))]>;
4061 def : Pat<(v2i32 (or (and DPR:$Vn, DPR:$Vd),
4062 (and DPR:$Vm, (vnotd DPR:$Vd)))),
4063 (VBSLd DPR:$Vd, DPR:$Vn, DPR:$Vm)>;
4065 def VBSLq : N3VX<1, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
4066 (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
4067 N3RegFrm, IIC_VCNTiQ,
4068 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4070 (v4i32 (NEONvbsl QPR:$src1, QPR:$Vn, QPR:$Vm)))]>;
4072 def : Pat<(v4i32 (or (and QPR:$Vn, QPR:$Vd),
4073 (and QPR:$Vm, (vnotq QPR:$Vd)))),
4074 (VBSLq QPR:$Vd, QPR:$Vn, QPR:$Vm)>;
4076 // VBIF : Vector Bitwise Insert if False
4077 // like VBSL but with: "vbif $dst, $src3, $src1", "$src2 = $dst",
4078 // FIXME: This instruction's encoding MAY NOT BE correct.
4079 def VBIFd : N3VX<1, 0, 0b11, 0b0001, 0, 1,
4080 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
4081 N3RegFrm, IIC_VBINiD,
4082 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4084 def VBIFq : N3VX<1, 0, 0b11, 0b0001, 1, 1,
4085 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
4086 N3RegFrm, IIC_VBINiQ,
4087 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4090 // VBIT : Vector Bitwise Insert if True
4091 // like VBSL but with: "vbit $dst, $src2, $src1", "$src3 = $dst",
4092 // FIXME: This instruction's encoding MAY NOT BE correct.
4093 def VBITd : N3VX<1, 0, 0b10, 0b0001, 0, 1,
4094 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
4095 N3RegFrm, IIC_VBINiD,
4096 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4098 def VBITq : N3VX<1, 0, 0b10, 0b0001, 1, 1,
4099 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
4100 N3RegFrm, IIC_VBINiQ,
4101 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4104 // VBIT/VBIF are not yet implemented. The TwoAddress pass will not go looking
4105 // for equivalent operations with different register constraints; it just
4108 // Vector Absolute Differences.
4110 // VABD : Vector Absolute Difference
4111 defm VABDs : N3VInt_QHS<0, 0, 0b0111, 0, N3RegFrm,
4112 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4113 "vabd", "s", int_arm_neon_vabds, 1>;
4114 defm VABDu : N3VInt_QHS<1, 0, 0b0111, 0, N3RegFrm,
4115 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4116 "vabd", "u", int_arm_neon_vabdu, 1>;
4117 def VABDfd : N3VDInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBIND,
4118 "vabd", "f32", v2f32, v2f32, int_arm_neon_vabds, 1>;
4119 def VABDfq : N3VQInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBINQ,
4120 "vabd", "f32", v4f32, v4f32, int_arm_neon_vabds, 1>;
4122 // VABDL : Vector Absolute Difference Long (Q = | D - D |)
4123 defm VABDLs : N3VLIntExt_QHS<0,1,0b0111,0, IIC_VSUBi4Q,
4124 "vabdl", "s", int_arm_neon_vabds, zext, 1>;
4125 defm VABDLu : N3VLIntExt_QHS<1,1,0b0111,0, IIC_VSUBi4Q,
4126 "vabdl", "u", int_arm_neon_vabdu, zext, 1>;
4128 // VABA : Vector Absolute Difference and Accumulate
4129 defm VABAs : N3VIntOp_QHS<0,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
4130 "vaba", "s", int_arm_neon_vabds, add>;
4131 defm VABAu : N3VIntOp_QHS<1,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
4132 "vaba", "u", int_arm_neon_vabdu, add>;
4134 // VABAL : Vector Absolute Difference and Accumulate Long (Q += | D - D |)
4135 defm VABALs : N3VLIntExtOp_QHS<0,1,0b0101,0, IIC_VABAD,
4136 "vabal", "s", int_arm_neon_vabds, zext, add>;
4137 defm VABALu : N3VLIntExtOp_QHS<1,1,0b0101,0, IIC_VABAD,
4138 "vabal", "u", int_arm_neon_vabdu, zext, add>;
4140 // Vector Maximum and Minimum.
4142 // VMAX : Vector Maximum
4143 defm VMAXs : N3VInt_QHS<0, 0, 0b0110, 0, N3RegFrm,
4144 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4145 "vmax", "s", int_arm_neon_vmaxs, 1>;
4146 defm VMAXu : N3VInt_QHS<1, 0, 0b0110, 0, N3RegFrm,
4147 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4148 "vmax", "u", int_arm_neon_vmaxu, 1>;
4149 def VMAXfd : N3VDInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBIND,
4151 v2f32, v2f32, int_arm_neon_vmaxs, 1>;
4152 def VMAXfq : N3VQInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBINQ,
4154 v4f32, v4f32, int_arm_neon_vmaxs, 1>;
4156 // VMIN : Vector Minimum
4157 defm VMINs : N3VInt_QHS<0, 0, 0b0110, 1, N3RegFrm,
4158 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4159 "vmin", "s", int_arm_neon_vmins, 1>;
4160 defm VMINu : N3VInt_QHS<1, 0, 0b0110, 1, N3RegFrm,
4161 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4162 "vmin", "u", int_arm_neon_vminu, 1>;
4163 def VMINfd : N3VDInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBIND,
4165 v2f32, v2f32, int_arm_neon_vmins, 1>;
4166 def VMINfq : N3VQInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBINQ,
4168 v4f32, v4f32, int_arm_neon_vmins, 1>;
4170 // Vector Pairwise Operations.
4172 // VPADD : Vector Pairwise Add
4173 def VPADDi8 : N3VDInt<0, 0, 0b00, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4175 v8i8, v8i8, int_arm_neon_vpadd, 0>;
4176 def VPADDi16 : N3VDInt<0, 0, 0b01, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4178 v4i16, v4i16, int_arm_neon_vpadd, 0>;
4179 def VPADDi32 : N3VDInt<0, 0, 0b10, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4181 v2i32, v2i32, int_arm_neon_vpadd, 0>;
4182 def VPADDf : N3VDInt<1, 0, 0b00, 0b1101, 0, N3RegFrm,
4183 IIC_VPBIND, "vpadd", "f32",
4184 v2f32, v2f32, int_arm_neon_vpadd, 0>;
4186 // VPADDL : Vector Pairwise Add Long
4187 defm VPADDLs : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00100, 0, "vpaddl", "s",
4188 int_arm_neon_vpaddls>;
4189 defm VPADDLu : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00101, 0, "vpaddl", "u",
4190 int_arm_neon_vpaddlu>;
4192 // VPADAL : Vector Pairwise Add and Accumulate Long
4193 defm VPADALs : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01100, 0, "vpadal", "s",
4194 int_arm_neon_vpadals>;
4195 defm VPADALu : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01101, 0, "vpadal", "u",
4196 int_arm_neon_vpadalu>;
4198 // VPMAX : Vector Pairwise Maximum
4199 def VPMAXs8 : N3VDInt<0, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4200 "s8", v8i8, v8i8, int_arm_neon_vpmaxs, 0>;
4201 def VPMAXs16 : N3VDInt<0, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4202 "s16", v4i16, v4i16, int_arm_neon_vpmaxs, 0>;
4203 def VPMAXs32 : N3VDInt<0, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4204 "s32", v2i32, v2i32, int_arm_neon_vpmaxs, 0>;
4205 def VPMAXu8 : N3VDInt<1, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4206 "u8", v8i8, v8i8, int_arm_neon_vpmaxu, 0>;
4207 def VPMAXu16 : N3VDInt<1, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4208 "u16", v4i16, v4i16, int_arm_neon_vpmaxu, 0>;
4209 def VPMAXu32 : N3VDInt<1, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4210 "u32", v2i32, v2i32, int_arm_neon_vpmaxu, 0>;
4211 def VPMAXf : N3VDInt<1, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmax",
4212 "f32", v2f32, v2f32, int_arm_neon_vpmaxs, 0>;
4214 // VPMIN : Vector Pairwise Minimum
4215 def VPMINs8 : N3VDInt<0, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4216 "s8", v8i8, v8i8, int_arm_neon_vpmins, 0>;
4217 def VPMINs16 : N3VDInt<0, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4218 "s16", v4i16, v4i16, int_arm_neon_vpmins, 0>;
4219 def VPMINs32 : N3VDInt<0, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4220 "s32", v2i32, v2i32, int_arm_neon_vpmins, 0>;
4221 def VPMINu8 : N3VDInt<1, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4222 "u8", v8i8, v8i8, int_arm_neon_vpminu, 0>;
4223 def VPMINu16 : N3VDInt<1, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4224 "u16", v4i16, v4i16, int_arm_neon_vpminu, 0>;
4225 def VPMINu32 : N3VDInt<1, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4226 "u32", v2i32, v2i32, int_arm_neon_vpminu, 0>;
4227 def VPMINf : N3VDInt<1, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmin",
4228 "f32", v2f32, v2f32, int_arm_neon_vpmins, 0>;
4230 // Vector Reciprocal and Reciprocal Square Root Estimate and Step.
4232 // VRECPE : Vector Reciprocal Estimate
4233 def VRECPEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
4234 IIC_VUNAD, "vrecpe", "u32",
4235 v2i32, v2i32, int_arm_neon_vrecpe>;
4236 def VRECPEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
4237 IIC_VUNAQ, "vrecpe", "u32",
4238 v4i32, v4i32, int_arm_neon_vrecpe>;
4239 def VRECPEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
4240 IIC_VUNAD, "vrecpe", "f32",
4241 v2f32, v2f32, int_arm_neon_vrecpe>;
4242 def VRECPEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
4243 IIC_VUNAQ, "vrecpe", "f32",
4244 v4f32, v4f32, int_arm_neon_vrecpe>;
4246 // VRECPS : Vector Reciprocal Step
4247 def VRECPSfd : N3VDInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
4248 IIC_VRECSD, "vrecps", "f32",
4249 v2f32, v2f32, int_arm_neon_vrecps, 1>;
4250 def VRECPSfq : N3VQInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
4251 IIC_VRECSQ, "vrecps", "f32",
4252 v4f32, v4f32, int_arm_neon_vrecps, 1>;
4254 // VRSQRTE : Vector Reciprocal Square Root Estimate
4255 def VRSQRTEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
4256 IIC_VUNAD, "vrsqrte", "u32",
4257 v2i32, v2i32, int_arm_neon_vrsqrte>;
4258 def VRSQRTEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
4259 IIC_VUNAQ, "vrsqrte", "u32",
4260 v4i32, v4i32, int_arm_neon_vrsqrte>;
4261 def VRSQRTEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
4262 IIC_VUNAD, "vrsqrte", "f32",
4263 v2f32, v2f32, int_arm_neon_vrsqrte>;
4264 def VRSQRTEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
4265 IIC_VUNAQ, "vrsqrte", "f32",
4266 v4f32, v4f32, int_arm_neon_vrsqrte>;
4268 // VRSQRTS : Vector Reciprocal Square Root Step
4269 def VRSQRTSfd : N3VDInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
4270 IIC_VRECSD, "vrsqrts", "f32",
4271 v2f32, v2f32, int_arm_neon_vrsqrts, 1>;
4272 def VRSQRTSfq : N3VQInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
4273 IIC_VRECSQ, "vrsqrts", "f32",
4274 v4f32, v4f32, int_arm_neon_vrsqrts, 1>;
4278 // VSHL : Vector Shift
4279 defm VSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 0, N3RegVShFrm,
4280 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
4281 "vshl", "s", int_arm_neon_vshifts>;
4282 defm VSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 0, N3RegVShFrm,
4283 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
4284 "vshl", "u", int_arm_neon_vshiftu>;
4286 // VSHL : Vector Shift Left (Immediate)
4287 defm VSHLi : N2VShL_QHSD<0, 1, 0b0101, 1, IIC_VSHLiD, "vshl", "i", NEONvshl>;
4289 // VSHR : Vector Shift Right (Immediate)
4290 defm VSHRs : N2VShR_QHSD<0, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "s",NEONvshrs>;
4291 defm VSHRu : N2VShR_QHSD<1, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "u",NEONvshru>;
4293 // VSHLL : Vector Shift Left Long
4294 defm VSHLLs : N2VLSh_QHS<0, 1, 0b1010, 0, 0, 1, "vshll", "s", NEONvshlls>;
4295 defm VSHLLu : N2VLSh_QHS<1, 1, 0b1010, 0, 0, 1, "vshll", "u", NEONvshllu>;
4297 // VSHLL : Vector Shift Left Long (with maximum shift count)
4298 class N2VLShMax<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
4299 bit op6, bit op4, string OpcodeStr, string Dt, ValueType ResTy,
4300 ValueType OpTy, SDNode OpNode>
4301 : N2VLSh<op24, op23, op11_8, op7, op6, op4, OpcodeStr, Dt,
4302 ResTy, OpTy, OpNode> {
4303 let Inst{21-16} = op21_16;
4304 let DecoderMethod = "DecodeVSHLMaxInstruction";
4306 def VSHLLi8 : N2VLShMax<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll", "i8",
4307 v8i16, v8i8, NEONvshlli>;
4308 def VSHLLi16 : N2VLShMax<1, 1, 0b110110, 0b0011, 0, 0, 0, "vshll", "i16",
4309 v4i32, v4i16, NEONvshlli>;
4310 def VSHLLi32 : N2VLShMax<1, 1, 0b111010, 0b0011, 0, 0, 0, "vshll", "i32",
4311 v2i64, v2i32, NEONvshlli>;
4313 // VSHRN : Vector Shift Right and Narrow
4314 defm VSHRN : N2VNSh_HSD<0,1,0b1000,0,0,1, IIC_VSHLiD, "vshrn", "i",
4317 // VRSHL : Vector Rounding Shift
4318 defm VRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 0, N3RegVShFrm,
4319 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4320 "vrshl", "s", int_arm_neon_vrshifts>;
4321 defm VRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 0, N3RegVShFrm,
4322 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4323 "vrshl", "u", int_arm_neon_vrshiftu>;
4324 // VRSHR : Vector Rounding Shift Right
4325 defm VRSHRs : N2VShR_QHSD<0,1,0b0010,1, IIC_VSHLi4D, "vrshr", "s",NEONvrshrs>;
4326 defm VRSHRu : N2VShR_QHSD<1,1,0b0010,1, IIC_VSHLi4D, "vrshr", "u",NEONvrshru>;
4328 // VRSHRN : Vector Rounding Shift Right and Narrow
4329 defm VRSHRN : N2VNSh_HSD<0, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vrshrn", "i",
4332 // VQSHL : Vector Saturating Shift
4333 defm VQSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 1, N3RegVShFrm,
4334 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4335 "vqshl", "s", int_arm_neon_vqshifts>;
4336 defm VQSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 1, N3RegVShFrm,
4337 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4338 "vqshl", "u", int_arm_neon_vqshiftu>;
4339 // VQSHL : Vector Saturating Shift Left (Immediate)
4340 defm VQSHLsi : N2VShL_QHSD<0,1,0b0111,1, IIC_VSHLi4D, "vqshl", "s",NEONvqshls>;
4341 defm VQSHLui : N2VShL_QHSD<1,1,0b0111,1, IIC_VSHLi4D, "vqshl", "u",NEONvqshlu>;
4343 // VQSHLU : Vector Saturating Shift Left (Immediate, Unsigned)
4344 defm VQSHLsu : N2VShL_QHSD<1,1,0b0110,1, IIC_VSHLi4D,"vqshlu","s",NEONvqshlsu>;
4346 // VQSHRN : Vector Saturating Shift Right and Narrow
4347 defm VQSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "s",
4349 defm VQSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "u",
4352 // VQSHRUN : Vector Saturating Shift Right and Narrow (Unsigned)
4353 defm VQSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 0, 1, IIC_VSHLi4D, "vqshrun", "s",
4356 // VQRSHL : Vector Saturating Rounding Shift
4357 defm VQRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 1, N3RegVShFrm,
4358 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4359 "vqrshl", "s", int_arm_neon_vqrshifts>;
4360 defm VQRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 1, N3RegVShFrm,
4361 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4362 "vqrshl", "u", int_arm_neon_vqrshiftu>;
4364 // VQRSHRN : Vector Saturating Rounding Shift Right and Narrow
4365 defm VQRSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "s",
4367 defm VQRSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "u",
4370 // VQRSHRUN : Vector Saturating Rounding Shift Right and Narrow (Unsigned)
4371 defm VQRSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vqrshrun", "s",
4374 // VSRA : Vector Shift Right and Accumulate
4375 defm VSRAs : N2VShAdd_QHSD<0, 1, 0b0001, 1, "vsra", "s", NEONvshrs>;
4376 defm VSRAu : N2VShAdd_QHSD<1, 1, 0b0001, 1, "vsra", "u", NEONvshru>;
4377 // VRSRA : Vector Rounding Shift Right and Accumulate
4378 defm VRSRAs : N2VShAdd_QHSD<0, 1, 0b0011, 1, "vrsra", "s", NEONvrshrs>;
4379 defm VRSRAu : N2VShAdd_QHSD<1, 1, 0b0011, 1, "vrsra", "u", NEONvrshru>;
4381 // VSLI : Vector Shift Left and Insert
4382 defm VSLI : N2VShInsL_QHSD<1, 1, 0b0101, 1, "vsli">;
4384 // VSRI : Vector Shift Right and Insert
4385 defm VSRI : N2VShInsR_QHSD<1, 1, 0b0100, 1, "vsri">;
4387 // Vector Absolute and Saturating Absolute.
4389 // VABS : Vector Absolute Value
4390 defm VABS : N2VInt_QHS<0b11, 0b11, 0b01, 0b00110, 0,
4391 IIC_VUNAiD, IIC_VUNAiQ, "vabs", "s",
4393 def VABSfd : N2VDInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
4394 IIC_VUNAD, "vabs", "f32",
4395 v2f32, v2f32, int_arm_neon_vabs>;
4396 def VABSfq : N2VQInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
4397 IIC_VUNAQ, "vabs", "f32",
4398 v4f32, v4f32, int_arm_neon_vabs>;
4400 // VQABS : Vector Saturating Absolute Value
4401 defm VQABS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01110, 0,
4402 IIC_VQUNAiD, IIC_VQUNAiQ, "vqabs", "s",
4403 int_arm_neon_vqabs>;
4407 def vnegd : PatFrag<(ops node:$in),
4408 (sub (bitconvert (v2i32 NEONimmAllZerosV)), node:$in)>;
4409 def vnegq : PatFrag<(ops node:$in),
4410 (sub (bitconvert (v4i32 NEONimmAllZerosV)), node:$in)>;
4412 class VNEGD<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
4413 : N2V<0b11, 0b11, size, 0b01, 0b00111, 0, 0, (outs DPR:$Vd), (ins DPR:$Vm),
4414 IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
4415 [(set DPR:$Vd, (Ty (vnegd DPR:$Vm)))]>;
4416 class VNEGQ<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
4417 : N2V<0b11, 0b11, size, 0b01, 0b00111, 1, 0, (outs QPR:$Vd), (ins QPR:$Vm),
4418 IIC_VSHLiQ, OpcodeStr, Dt, "$Vd, $Vm", "",
4419 [(set QPR:$Vd, (Ty (vnegq QPR:$Vm)))]>;
4421 // VNEG : Vector Negate (integer)
4422 def VNEGs8d : VNEGD<0b00, "vneg", "s8", v8i8>;
4423 def VNEGs16d : VNEGD<0b01, "vneg", "s16", v4i16>;
4424 def VNEGs32d : VNEGD<0b10, "vneg", "s32", v2i32>;
4425 def VNEGs8q : VNEGQ<0b00, "vneg", "s8", v16i8>;
4426 def VNEGs16q : VNEGQ<0b01, "vneg", "s16", v8i16>;
4427 def VNEGs32q : VNEGQ<0b10, "vneg", "s32", v4i32>;
4429 // VNEG : Vector Negate (floating-point)
4430 def VNEGfd : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
4431 (outs DPR:$Vd), (ins DPR:$Vm), IIC_VUNAD,
4432 "vneg", "f32", "$Vd, $Vm", "",
4433 [(set DPR:$Vd, (v2f32 (fneg DPR:$Vm)))]>;
4434 def VNEGf32q : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 1, 0,
4435 (outs QPR:$Vd), (ins QPR:$Vm), IIC_VUNAQ,
4436 "vneg", "f32", "$Vd, $Vm", "",
4437 [(set QPR:$Vd, (v4f32 (fneg QPR:$Vm)))]>;
4439 def : Pat<(v8i8 (vnegd DPR:$src)), (VNEGs8d DPR:$src)>;
4440 def : Pat<(v4i16 (vnegd DPR:$src)), (VNEGs16d DPR:$src)>;
4441 def : Pat<(v2i32 (vnegd DPR:$src)), (VNEGs32d DPR:$src)>;
4442 def : Pat<(v16i8 (vnegq QPR:$src)), (VNEGs8q QPR:$src)>;
4443 def : Pat<(v8i16 (vnegq QPR:$src)), (VNEGs16q QPR:$src)>;
4444 def : Pat<(v4i32 (vnegq QPR:$src)), (VNEGs32q QPR:$src)>;
4446 // VQNEG : Vector Saturating Negate
4447 defm VQNEG : N2VInt_QHS<0b11, 0b11, 0b00, 0b01111, 0,
4448 IIC_VQUNAiD, IIC_VQUNAiQ, "vqneg", "s",
4449 int_arm_neon_vqneg>;
4451 // Vector Bit Counting Operations.
4453 // VCLS : Vector Count Leading Sign Bits
4454 defm VCLS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01000, 0,
4455 IIC_VCNTiD, IIC_VCNTiQ, "vcls", "s",
4457 // VCLZ : Vector Count Leading Zeros
4458 defm VCLZ : N2VInt_QHS<0b11, 0b11, 0b00, 0b01001, 0,
4459 IIC_VCNTiD, IIC_VCNTiQ, "vclz", "i",
4461 // VCNT : Vector Count One Bits
4462 def VCNTd : N2VDInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
4463 IIC_VCNTiD, "vcnt", "8",
4464 v8i8, v8i8, int_arm_neon_vcnt>;
4465 def VCNTq : N2VQInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
4466 IIC_VCNTiQ, "vcnt", "8",
4467 v16i8, v16i8, int_arm_neon_vcnt>;
4470 def VSWPd : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 0, 0,
4471 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
4472 "vswp", "$Vd, $Vm", "", []>;
4473 def VSWPq : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 1, 0,
4474 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
4475 "vswp", "$Vd, $Vm", "", []>;
4477 // Vector Move Operations.
4479 // VMOV : Vector Move (Register)
4480 def : InstAlias<"vmov${p} $Vd, $Vm",
4481 (VORRd DPR:$Vd, DPR:$Vm, DPR:$Vm, pred:$p)>;
4482 def : InstAlias<"vmov${p} $Vd, $Vm",
4483 (VORRq QPR:$Vd, QPR:$Vm, QPR:$Vm, pred:$p)>;
4484 defm : VFPDTAnyNoF64InstAlias<"vmov${p}", "$Vd, $Vm",
4485 (VORRd DPR:$Vd, DPR:$Vm, DPR:$Vm, pred:$p)>;
4486 defm : VFPDTAnyNoF64InstAlias<"vmov${p}", "$Vd, $Vm",
4487 (VORRq QPR:$Vd, QPR:$Vm, QPR:$Vm, pred:$p)>;
4489 // VMOV : Vector Move (Immediate)
4491 let isReMaterializable = 1 in {
4492 def VMOVv8i8 : N1ModImm<1, 0b000, 0b1110, 0, 0, 0, 1, (outs DPR:$Vd),
4493 (ins nImmSplatI8:$SIMM), IIC_VMOVImm,
4494 "vmov", "i8", "$Vd, $SIMM", "",
4495 [(set DPR:$Vd, (v8i8 (NEONvmovImm timm:$SIMM)))]>;
4496 def VMOVv16i8 : N1ModImm<1, 0b000, 0b1110, 0, 1, 0, 1, (outs QPR:$Vd),
4497 (ins nImmSplatI8:$SIMM), IIC_VMOVImm,
4498 "vmov", "i8", "$Vd, $SIMM", "",
4499 [(set QPR:$Vd, (v16i8 (NEONvmovImm timm:$SIMM)))]>;
4501 def VMOVv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 0, 1, (outs DPR:$Vd),
4502 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4503 "vmov", "i16", "$Vd, $SIMM", "",
4504 [(set DPR:$Vd, (v4i16 (NEONvmovImm timm:$SIMM)))]> {
4505 let Inst{9} = SIMM{9};
4508 def VMOVv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 0, 1, (outs QPR:$Vd),
4509 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4510 "vmov", "i16", "$Vd, $SIMM", "",
4511 [(set QPR:$Vd, (v8i16 (NEONvmovImm timm:$SIMM)))]> {
4512 let Inst{9} = SIMM{9};
4515 def VMOVv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 0, 1, (outs DPR:$Vd),
4516 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4517 "vmov", "i32", "$Vd, $SIMM", "",
4518 [(set DPR:$Vd, (v2i32 (NEONvmovImm timm:$SIMM)))]> {
4519 let Inst{11-8} = SIMM{11-8};
4522 def VMOVv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 0, 1, (outs QPR:$Vd),
4523 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4524 "vmov", "i32", "$Vd, $SIMM", "",
4525 [(set QPR:$Vd, (v4i32 (NEONvmovImm timm:$SIMM)))]> {
4526 let Inst{11-8} = SIMM{11-8};
4529 def VMOVv1i64 : N1ModImm<1, 0b000, 0b1110, 0, 0, 1, 1, (outs DPR:$Vd),
4530 (ins nImmSplatI64:$SIMM), IIC_VMOVImm,
4531 "vmov", "i64", "$Vd, $SIMM", "",
4532 [(set DPR:$Vd, (v1i64 (NEONvmovImm timm:$SIMM)))]>;
4533 def VMOVv2i64 : N1ModImm<1, 0b000, 0b1110, 0, 1, 1, 1, (outs QPR:$Vd),
4534 (ins nImmSplatI64:$SIMM), IIC_VMOVImm,
4535 "vmov", "i64", "$Vd, $SIMM", "",
4536 [(set QPR:$Vd, (v2i64 (NEONvmovImm timm:$SIMM)))]>;
4538 def VMOVv2f32 : N1ModImm<1, 0b000, 0b1111, 0, 0, 0, 1, (outs DPR:$Vd),
4539 (ins nImmVMOVF32:$SIMM), IIC_VMOVImm,
4540 "vmov", "f32", "$Vd, $SIMM", "",
4541 [(set DPR:$Vd, (v2f32 (NEONvmovFPImm timm:$SIMM)))]>;
4542 def VMOVv4f32 : N1ModImm<1, 0b000, 0b1111, 0, 1, 0, 1, (outs QPR:$Vd),
4543 (ins nImmVMOVF32:$SIMM), IIC_VMOVImm,
4544 "vmov", "f32", "$Vd, $SIMM", "",
4545 [(set QPR:$Vd, (v4f32 (NEONvmovFPImm timm:$SIMM)))]>;
4546 } // isReMaterializable
4548 // VMOV : Vector Get Lane (move scalar to ARM core register)
4550 def VGETLNs8 : NVGetLane<{1,1,1,0,0,1,?,1}, 0b1011, {?,?},
4551 (outs GPR:$R), (ins DPR:$V, VectorIndex8:$lane),
4552 IIC_VMOVSI, "vmov", "s8", "$R, $V$lane",
4553 [(set GPR:$R, (NEONvgetlanes (v8i8 DPR:$V),
4555 let Inst{21} = lane{2};
4556 let Inst{6-5} = lane{1-0};
4558 def VGETLNs16 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, {?,1},
4559 (outs GPR:$R), (ins DPR:$V, VectorIndex16:$lane),
4560 IIC_VMOVSI, "vmov", "s16", "$R, $V$lane",
4561 [(set GPR:$R, (NEONvgetlanes (v4i16 DPR:$V),
4563 let Inst{21} = lane{1};
4564 let Inst{6} = lane{0};
4566 def VGETLNu8 : NVGetLane<{1,1,1,0,1,1,?,1}, 0b1011, {?,?},
4567 (outs GPR:$R), (ins DPR:$V, VectorIndex8:$lane),
4568 IIC_VMOVSI, "vmov", "u8", "$R, $V$lane",
4569 [(set GPR:$R, (NEONvgetlaneu (v8i8 DPR:$V),
4571 let Inst{21} = lane{2};
4572 let Inst{6-5} = lane{1-0};
4574 def VGETLNu16 : NVGetLane<{1,1,1,0,1,0,?,1}, 0b1011, {?,1},
4575 (outs GPR:$R), (ins DPR:$V, VectorIndex16:$lane),
4576 IIC_VMOVSI, "vmov", "u16", "$R, $V$lane",
4577 [(set GPR:$R, (NEONvgetlaneu (v4i16 DPR:$V),
4579 let Inst{21} = lane{1};
4580 let Inst{6} = lane{0};
4582 def VGETLNi32 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, 0b00,
4583 (outs GPR:$R), (ins DPR:$V, VectorIndex32:$lane),
4584 IIC_VMOVSI, "vmov", "32", "$R, $V$lane",
4585 [(set GPR:$R, (extractelt (v2i32 DPR:$V),
4587 let Inst{21} = lane{0};
4589 // def VGETLNf32: see FMRDH and FMRDL in ARMInstrVFP.td
4590 def : Pat<(NEONvgetlanes (v16i8 QPR:$src), imm:$lane),
4591 (VGETLNs8 (v8i8 (EXTRACT_SUBREG QPR:$src,
4592 (DSubReg_i8_reg imm:$lane))),
4593 (SubReg_i8_lane imm:$lane))>;
4594 def : Pat<(NEONvgetlanes (v8i16 QPR:$src), imm:$lane),
4595 (VGETLNs16 (v4i16 (EXTRACT_SUBREG QPR:$src,
4596 (DSubReg_i16_reg imm:$lane))),
4597 (SubReg_i16_lane imm:$lane))>;
4598 def : Pat<(NEONvgetlaneu (v16i8 QPR:$src), imm:$lane),
4599 (VGETLNu8 (v8i8 (EXTRACT_SUBREG QPR:$src,
4600 (DSubReg_i8_reg imm:$lane))),
4601 (SubReg_i8_lane imm:$lane))>;
4602 def : Pat<(NEONvgetlaneu (v8i16 QPR:$src), imm:$lane),
4603 (VGETLNu16 (v4i16 (EXTRACT_SUBREG QPR:$src,
4604 (DSubReg_i16_reg imm:$lane))),
4605 (SubReg_i16_lane imm:$lane))>;
4606 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
4607 (VGETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src,
4608 (DSubReg_i32_reg imm:$lane))),
4609 (SubReg_i32_lane imm:$lane))>;
4610 def : Pat<(extractelt (v2f32 DPR:$src1), imm:$src2),
4611 (EXTRACT_SUBREG (v2f32 (COPY_TO_REGCLASS (v2f32 DPR:$src1),DPR_VFP2)),
4612 (SSubReg_f32_reg imm:$src2))>;
4613 def : Pat<(extractelt (v4f32 QPR:$src1), imm:$src2),
4614 (EXTRACT_SUBREG (v4f32 (COPY_TO_REGCLASS (v4f32 QPR:$src1),QPR_VFP2)),
4615 (SSubReg_f32_reg imm:$src2))>;
4616 //def : Pat<(extractelt (v2i64 QPR:$src1), imm:$src2),
4617 // (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
4618 def : Pat<(extractelt (v2f64 QPR:$src1), imm:$src2),
4619 (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
4622 // VMOV : Vector Set Lane (move ARM core register to scalar)
4624 let Constraints = "$src1 = $V" in {
4625 def VSETLNi8 : NVSetLane<{1,1,1,0,0,1,?,0}, 0b1011, {?,?}, (outs DPR:$V),
4626 (ins DPR:$src1, GPR:$R, VectorIndex8:$lane),
4627 IIC_VMOVISL, "vmov", "8", "$V$lane, $R",
4628 [(set DPR:$V, (vector_insert (v8i8 DPR:$src1),
4629 GPR:$R, imm:$lane))]> {
4630 let Inst{21} = lane{2};
4631 let Inst{6-5} = lane{1-0};
4633 def VSETLNi16 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, {?,1}, (outs DPR:$V),
4634 (ins DPR:$src1, GPR:$R, VectorIndex16:$lane),
4635 IIC_VMOVISL, "vmov", "16", "$V$lane, $R",
4636 [(set DPR:$V, (vector_insert (v4i16 DPR:$src1),
4637 GPR:$R, imm:$lane))]> {
4638 let Inst{21} = lane{1};
4639 let Inst{6} = lane{0};
4641 def VSETLNi32 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, 0b00, (outs DPR:$V),
4642 (ins DPR:$src1, GPR:$R, VectorIndex32:$lane),
4643 IIC_VMOVISL, "vmov", "32", "$V$lane, $R",
4644 [(set DPR:$V, (insertelt (v2i32 DPR:$src1),
4645 GPR:$R, imm:$lane))]> {
4646 let Inst{21} = lane{0};
4649 def : Pat<(vector_insert (v16i8 QPR:$src1), GPR:$src2, imm:$lane),
4650 (v16i8 (INSERT_SUBREG QPR:$src1,
4651 (v8i8 (VSETLNi8 (v8i8 (EXTRACT_SUBREG QPR:$src1,
4652 (DSubReg_i8_reg imm:$lane))),
4653 GPR:$src2, (SubReg_i8_lane imm:$lane))),
4654 (DSubReg_i8_reg imm:$lane)))>;
4655 def : Pat<(vector_insert (v8i16 QPR:$src1), GPR:$src2, imm:$lane),
4656 (v8i16 (INSERT_SUBREG QPR:$src1,
4657 (v4i16 (VSETLNi16 (v4i16 (EXTRACT_SUBREG QPR:$src1,
4658 (DSubReg_i16_reg imm:$lane))),
4659 GPR:$src2, (SubReg_i16_lane imm:$lane))),
4660 (DSubReg_i16_reg imm:$lane)))>;
4661 def : Pat<(insertelt (v4i32 QPR:$src1), GPR:$src2, imm:$lane),
4662 (v4i32 (INSERT_SUBREG QPR:$src1,
4663 (v2i32 (VSETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src1,
4664 (DSubReg_i32_reg imm:$lane))),
4665 GPR:$src2, (SubReg_i32_lane imm:$lane))),
4666 (DSubReg_i32_reg imm:$lane)))>;
4668 def : Pat<(v2f32 (insertelt DPR:$src1, SPR:$src2, imm:$src3)),
4669 (INSERT_SUBREG (v2f32 (COPY_TO_REGCLASS DPR:$src1, DPR_VFP2)),
4670 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
4671 def : Pat<(v4f32 (insertelt QPR:$src1, SPR:$src2, imm:$src3)),
4672 (INSERT_SUBREG (v4f32 (COPY_TO_REGCLASS QPR:$src1, QPR_VFP2)),
4673 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
4675 //def : Pat<(v2i64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
4676 // (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
4677 def : Pat<(v2f64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
4678 (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
4680 def : Pat<(v2f32 (scalar_to_vector SPR:$src)),
4681 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
4682 def : Pat<(v2f64 (scalar_to_vector (f64 DPR:$src))),
4683 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
4684 def : Pat<(v4f32 (scalar_to_vector SPR:$src)),
4685 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
4687 def : Pat<(v8i8 (scalar_to_vector GPR:$src)),
4688 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4689 def : Pat<(v4i16 (scalar_to_vector GPR:$src)),
4690 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4691 def : Pat<(v2i32 (scalar_to_vector GPR:$src)),
4692 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4694 def : Pat<(v16i8 (scalar_to_vector GPR:$src)),
4695 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4696 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4698 def : Pat<(v8i16 (scalar_to_vector GPR:$src)),
4699 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
4700 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4702 def : Pat<(v4i32 (scalar_to_vector GPR:$src)),
4703 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
4704 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4707 // VDUP : Vector Duplicate (from ARM core register to all elements)
4709 class VDUPD<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
4710 : NVDup<opcod1, 0b1011, opcod3, (outs DPR:$V), (ins GPR:$R),
4711 IIC_VMOVIS, "vdup", Dt, "$V, $R",
4712 [(set DPR:$V, (Ty (NEONvdup (i32 GPR:$R))))]>;
4713 class VDUPQ<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
4714 : NVDup<opcod1, 0b1011, opcod3, (outs QPR:$V), (ins GPR:$R),
4715 IIC_VMOVIS, "vdup", Dt, "$V, $R",
4716 [(set QPR:$V, (Ty (NEONvdup (i32 GPR:$R))))]>;
4718 def VDUP8d : VDUPD<0b11101100, 0b00, "8", v8i8>;
4719 def VDUP16d : VDUPD<0b11101000, 0b01, "16", v4i16>;
4720 def VDUP32d : VDUPD<0b11101000, 0b00, "32", v2i32>;
4721 def VDUP8q : VDUPQ<0b11101110, 0b00, "8", v16i8>;
4722 def VDUP16q : VDUPQ<0b11101010, 0b01, "16", v8i16>;
4723 def VDUP32q : VDUPQ<0b11101010, 0b00, "32", v4i32>;
4725 def : Pat<(v2f32 (NEONvdup (f32 (bitconvert GPR:$R)))), (VDUP32d GPR:$R)>;
4726 def : Pat<(v4f32 (NEONvdup (f32 (bitconvert GPR:$R)))), (VDUP32q GPR:$R)>;
4728 // VDUP : Vector Duplicate Lane (from scalar to all elements)
4730 class VDUPLND<bits<4> op19_16, string OpcodeStr, string Dt,
4731 ValueType Ty, Operand IdxTy>
4732 : NVDupLane<op19_16, 0, (outs DPR:$Vd), (ins DPR:$Vm, IdxTy:$lane),
4733 IIC_VMOVD, OpcodeStr, Dt, "$Vd, $Vm$lane",
4734 [(set DPR:$Vd, (Ty (NEONvduplane (Ty DPR:$Vm), imm:$lane)))]>;
4736 class VDUPLNQ<bits<4> op19_16, string OpcodeStr, string Dt,
4737 ValueType ResTy, ValueType OpTy, Operand IdxTy>
4738 : NVDupLane<op19_16, 1, (outs QPR:$Vd), (ins DPR:$Vm, IdxTy:$lane),
4739 IIC_VMOVQ, OpcodeStr, Dt, "$Vd, $Vm$lane",
4740 [(set QPR:$Vd, (ResTy (NEONvduplane (OpTy DPR:$Vm),
4741 VectorIndex32:$lane)))]>;
4743 // Inst{19-16} is partially specified depending on the element size.
4745 def VDUPLN8d : VDUPLND<{?,?,?,1}, "vdup", "8", v8i8, VectorIndex8> {
4747 let Inst{19-17} = lane{2-0};
4749 def VDUPLN16d : VDUPLND<{?,?,1,0}, "vdup", "16", v4i16, VectorIndex16> {
4751 let Inst{19-18} = lane{1-0};
4753 def VDUPLN32d : VDUPLND<{?,1,0,0}, "vdup", "32", v2i32, VectorIndex32> {
4755 let Inst{19} = lane{0};
4757 def VDUPLN8q : VDUPLNQ<{?,?,?,1}, "vdup", "8", v16i8, v8i8, VectorIndex8> {
4759 let Inst{19-17} = lane{2-0};
4761 def VDUPLN16q : VDUPLNQ<{?,?,1,0}, "vdup", "16", v8i16, v4i16, VectorIndex16> {
4763 let Inst{19-18} = lane{1-0};
4765 def VDUPLN32q : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4i32, v2i32, VectorIndex32> {
4767 let Inst{19} = lane{0};
4770 def : Pat<(v2f32 (NEONvduplane (v2f32 DPR:$Vm), imm:$lane)),
4771 (VDUPLN32d DPR:$Vm, imm:$lane)>;
4773 def : Pat<(v4f32 (NEONvduplane (v2f32 DPR:$Vm), imm:$lane)),
4774 (VDUPLN32q DPR:$Vm, imm:$lane)>;
4776 def : Pat<(v16i8 (NEONvduplane (v16i8 QPR:$src), imm:$lane)),
4777 (v16i8 (VDUPLN8q (v8i8 (EXTRACT_SUBREG QPR:$src,
4778 (DSubReg_i8_reg imm:$lane))),
4779 (SubReg_i8_lane imm:$lane)))>;
4780 def : Pat<(v8i16 (NEONvduplane (v8i16 QPR:$src), imm:$lane)),
4781 (v8i16 (VDUPLN16q (v4i16 (EXTRACT_SUBREG QPR:$src,
4782 (DSubReg_i16_reg imm:$lane))),
4783 (SubReg_i16_lane imm:$lane)))>;
4784 def : Pat<(v4i32 (NEONvduplane (v4i32 QPR:$src), imm:$lane)),
4785 (v4i32 (VDUPLN32q (v2i32 (EXTRACT_SUBREG QPR:$src,
4786 (DSubReg_i32_reg imm:$lane))),
4787 (SubReg_i32_lane imm:$lane)))>;
4788 def : Pat<(v4f32 (NEONvduplane (v4f32 QPR:$src), imm:$lane)),
4789 (v4f32 (VDUPLN32q (v2f32 (EXTRACT_SUBREG QPR:$src,
4790 (DSubReg_i32_reg imm:$lane))),
4791 (SubReg_i32_lane imm:$lane)))>;
4793 def VDUPfdf : PseudoNeonI<(outs DPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
4794 [(set DPR:$dst, (v2f32 (NEONvdup (f32 SPR:$src))))]>;
4795 def VDUPfqf : PseudoNeonI<(outs QPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
4796 [(set QPR:$dst, (v4f32 (NEONvdup (f32 SPR:$src))))]>;
4798 // VMOVN : Vector Narrowing Move
4799 defm VMOVN : N2VN_HSD<0b11,0b11,0b10,0b00100,0,0, IIC_VMOVN,
4800 "vmovn", "i", trunc>;
4801 // VQMOVN : Vector Saturating Narrowing Move
4802 defm VQMOVNs : N2VNInt_HSD<0b11,0b11,0b10,0b00101,0,0, IIC_VQUNAiD,
4803 "vqmovn", "s", int_arm_neon_vqmovns>;
4804 defm VQMOVNu : N2VNInt_HSD<0b11,0b11,0b10,0b00101,1,0, IIC_VQUNAiD,
4805 "vqmovn", "u", int_arm_neon_vqmovnu>;
4806 defm VQMOVNsu : N2VNInt_HSD<0b11,0b11,0b10,0b00100,1,0, IIC_VQUNAiD,
4807 "vqmovun", "s", int_arm_neon_vqmovnsu>;
4808 // VMOVL : Vector Lengthening Move
4809 defm VMOVLs : N2VL_QHS<0b01,0b10100,0,1, "vmovl", "s", sext>;
4810 defm VMOVLu : N2VL_QHS<0b11,0b10100,0,1, "vmovl", "u", zext>;
4812 // Vector Conversions.
4814 // VCVT : Vector Convert Between Floating-Point and Integers
4815 def VCVTf2sd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
4816 v2i32, v2f32, fp_to_sint>;
4817 def VCVTf2ud : N2VD<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
4818 v2i32, v2f32, fp_to_uint>;
4819 def VCVTs2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
4820 v2f32, v2i32, sint_to_fp>;
4821 def VCVTu2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
4822 v2f32, v2i32, uint_to_fp>;
4824 def VCVTf2sq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
4825 v4i32, v4f32, fp_to_sint>;
4826 def VCVTf2uq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
4827 v4i32, v4f32, fp_to_uint>;
4828 def VCVTs2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
4829 v4f32, v4i32, sint_to_fp>;
4830 def VCVTu2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
4831 v4f32, v4i32, uint_to_fp>;
4833 // VCVT : Vector Convert Between Floating-Point and Fixed-Point.
4834 let DecoderMethod = "DecodeVCVTD" in {
4835 def VCVTf2xsd : N2VCvtD<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
4836 v2i32, v2f32, int_arm_neon_vcvtfp2fxs>;
4837 def VCVTf2xud : N2VCvtD<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
4838 v2i32, v2f32, int_arm_neon_vcvtfp2fxu>;
4839 def VCVTxs2fd : N2VCvtD<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
4840 v2f32, v2i32, int_arm_neon_vcvtfxs2fp>;
4841 def VCVTxu2fd : N2VCvtD<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
4842 v2f32, v2i32, int_arm_neon_vcvtfxu2fp>;
4845 let DecoderMethod = "DecodeVCVTQ" in {
4846 def VCVTf2xsq : N2VCvtQ<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
4847 v4i32, v4f32, int_arm_neon_vcvtfp2fxs>;
4848 def VCVTf2xuq : N2VCvtQ<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
4849 v4i32, v4f32, int_arm_neon_vcvtfp2fxu>;
4850 def VCVTxs2fq : N2VCvtQ<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
4851 v4f32, v4i32, int_arm_neon_vcvtfxs2fp>;
4852 def VCVTxu2fq : N2VCvtQ<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
4853 v4f32, v4i32, int_arm_neon_vcvtfxu2fp>;
4856 // VCVT : Vector Convert Between Half-Precision and Single-Precision.
4857 def VCVTf2h : N2VNInt<0b11, 0b11, 0b01, 0b10, 0b01100, 0, 0,
4858 IIC_VUNAQ, "vcvt", "f16.f32",
4859 v4i16, v4f32, int_arm_neon_vcvtfp2hf>,
4860 Requires<[HasNEON, HasFP16]>;
4861 def VCVTh2f : N2VLInt<0b11, 0b11, 0b01, 0b10, 0b01110, 0, 0,
4862 IIC_VUNAQ, "vcvt", "f32.f16",
4863 v4f32, v4i16, int_arm_neon_vcvthf2fp>,
4864 Requires<[HasNEON, HasFP16]>;
4868 // VREV64 : Vector Reverse elements within 64-bit doublewords
4870 class VREV64D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4871 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 0, 0, (outs DPR:$Vd),
4872 (ins DPR:$Vm), IIC_VMOVD,
4873 OpcodeStr, Dt, "$Vd, $Vm", "",
4874 [(set DPR:$Vd, (Ty (NEONvrev64 (Ty DPR:$Vm))))]>;
4875 class VREV64Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4876 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 1, 0, (outs QPR:$Vd),
4877 (ins QPR:$Vm), IIC_VMOVQ,
4878 OpcodeStr, Dt, "$Vd, $Vm", "",
4879 [(set QPR:$Vd, (Ty (NEONvrev64 (Ty QPR:$Vm))))]>;
4881 def VREV64d8 : VREV64D<0b00, "vrev64", "8", v8i8>;
4882 def VREV64d16 : VREV64D<0b01, "vrev64", "16", v4i16>;
4883 def VREV64d32 : VREV64D<0b10, "vrev64", "32", v2i32>;
4884 def : Pat<(v2f32 (NEONvrev64 (v2f32 DPR:$Vm))), (VREV64d32 DPR:$Vm)>;
4886 def VREV64q8 : VREV64Q<0b00, "vrev64", "8", v16i8>;
4887 def VREV64q16 : VREV64Q<0b01, "vrev64", "16", v8i16>;
4888 def VREV64q32 : VREV64Q<0b10, "vrev64", "32", v4i32>;
4889 def : Pat<(v4f32 (NEONvrev64 (v4f32 QPR:$Vm))), (VREV64q32 QPR:$Vm)>;
4891 // VREV32 : Vector Reverse elements within 32-bit words
4893 class VREV32D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4894 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 0, 0, (outs DPR:$Vd),
4895 (ins DPR:$Vm), IIC_VMOVD,
4896 OpcodeStr, Dt, "$Vd, $Vm", "",
4897 [(set DPR:$Vd, (Ty (NEONvrev32 (Ty DPR:$Vm))))]>;
4898 class VREV32Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4899 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 1, 0, (outs QPR:$Vd),
4900 (ins QPR:$Vm), IIC_VMOVQ,
4901 OpcodeStr, Dt, "$Vd, $Vm", "",
4902 [(set QPR:$Vd, (Ty (NEONvrev32 (Ty QPR:$Vm))))]>;
4904 def VREV32d8 : VREV32D<0b00, "vrev32", "8", v8i8>;
4905 def VREV32d16 : VREV32D<0b01, "vrev32", "16", v4i16>;
4907 def VREV32q8 : VREV32Q<0b00, "vrev32", "8", v16i8>;
4908 def VREV32q16 : VREV32Q<0b01, "vrev32", "16", v8i16>;
4910 // VREV16 : Vector Reverse elements within 16-bit halfwords
4912 class VREV16D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4913 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 0, 0, (outs DPR:$Vd),
4914 (ins DPR:$Vm), IIC_VMOVD,
4915 OpcodeStr, Dt, "$Vd, $Vm", "",
4916 [(set DPR:$Vd, (Ty (NEONvrev16 (Ty DPR:$Vm))))]>;
4917 class VREV16Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4918 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 1, 0, (outs QPR:$Vd),
4919 (ins QPR:$Vm), IIC_VMOVQ,
4920 OpcodeStr, Dt, "$Vd, $Vm", "",
4921 [(set QPR:$Vd, (Ty (NEONvrev16 (Ty QPR:$Vm))))]>;
4923 def VREV16d8 : VREV16D<0b00, "vrev16", "8", v8i8>;
4924 def VREV16q8 : VREV16Q<0b00, "vrev16", "8", v16i8>;
4926 // Other Vector Shuffles.
4928 // Aligned extractions: really just dropping registers
4930 class AlignedVEXTq<ValueType DestTy, ValueType SrcTy, SDNodeXForm LaneCVT>
4931 : Pat<(DestTy (vector_extract_subvec (SrcTy QPR:$src), (i32 imm:$start))),
4932 (EXTRACT_SUBREG (SrcTy QPR:$src), (LaneCVT imm:$start))>;
4934 def : AlignedVEXTq<v8i8, v16i8, DSubReg_i8_reg>;
4936 def : AlignedVEXTq<v4i16, v8i16, DSubReg_i16_reg>;
4938 def : AlignedVEXTq<v2i32, v4i32, DSubReg_i32_reg>;
4940 def : AlignedVEXTq<v1i64, v2i64, DSubReg_f64_reg>;
4942 def : AlignedVEXTq<v2f32, v4f32, DSubReg_i32_reg>;
4945 // VEXT : Vector Extract
4947 class VEXTd<string OpcodeStr, string Dt, ValueType Ty>
4948 : N3V<0,1,0b11,{?,?,?,?},0,0, (outs DPR:$Vd),
4949 (ins DPR:$Vn, DPR:$Vm, i32imm:$index), NVExtFrm,
4950 IIC_VEXTD, OpcodeStr, Dt, "$Vd, $Vn, $Vm, $index", "",
4951 [(set DPR:$Vd, (Ty (NEONvext (Ty DPR:$Vn),
4952 (Ty DPR:$Vm), imm:$index)))]> {
4954 let Inst{11-8} = index{3-0};
4957 class VEXTq<string OpcodeStr, string Dt, ValueType Ty>
4958 : N3V<0,1,0b11,{?,?,?,?},1,0, (outs QPR:$Vd),
4959 (ins QPR:$Vn, QPR:$Vm, i32imm:$index), NVExtFrm,
4960 IIC_VEXTQ, OpcodeStr, Dt, "$Vd, $Vn, $Vm, $index", "",
4961 [(set QPR:$Vd, (Ty (NEONvext (Ty QPR:$Vn),
4962 (Ty QPR:$Vm), imm:$index)))]> {
4964 let Inst{11-8} = index{3-0};
4967 def VEXTd8 : VEXTd<"vext", "8", v8i8> {
4968 let Inst{11-8} = index{3-0};
4970 def VEXTd16 : VEXTd<"vext", "16", v4i16> {
4971 let Inst{11-9} = index{2-0};
4974 def VEXTd32 : VEXTd<"vext", "32", v2i32> {
4975 let Inst{11-10} = index{1-0};
4976 let Inst{9-8} = 0b00;
4978 def : Pat<(v2f32 (NEONvext (v2f32 DPR:$Vn),
4981 (VEXTd32 DPR:$Vn, DPR:$Vm, imm:$index)>;
4983 def VEXTq8 : VEXTq<"vext", "8", v16i8> {
4984 let Inst{11-8} = index{3-0};
4986 def VEXTq16 : VEXTq<"vext", "16", v8i16> {
4987 let Inst{11-9} = index{2-0};
4990 def VEXTq32 : VEXTq<"vext", "32", v4i32> {
4991 let Inst{11-10} = index{1-0};
4992 let Inst{9-8} = 0b00;
4994 def : Pat<(v4f32 (NEONvext (v4f32 QPR:$Vn),
4997 (VEXTq32 QPR:$Vn, QPR:$Vm, imm:$index)>;
4999 // VTRN : Vector Transpose
5001 def VTRNd8 : N2VDShuffle<0b00, 0b00001, "vtrn", "8">;
5002 def VTRNd16 : N2VDShuffle<0b01, 0b00001, "vtrn", "16">;
5003 def VTRNd32 : N2VDShuffle<0b10, 0b00001, "vtrn", "32">;
5005 def VTRNq8 : N2VQShuffle<0b00, 0b00001, IIC_VPERMQ, "vtrn", "8">;
5006 def VTRNq16 : N2VQShuffle<0b01, 0b00001, IIC_VPERMQ, "vtrn", "16">;
5007 def VTRNq32 : N2VQShuffle<0b10, 0b00001, IIC_VPERMQ, "vtrn", "32">;
5009 // VUZP : Vector Unzip (Deinterleave)
5011 def VUZPd8 : N2VDShuffle<0b00, 0b00010, "vuzp", "8">;
5012 def VUZPd16 : N2VDShuffle<0b01, 0b00010, "vuzp", "16">;
5013 def VUZPd32 : N2VDShuffle<0b10, 0b00010, "vuzp", "32">;
5015 def VUZPq8 : N2VQShuffle<0b00, 0b00010, IIC_VPERMQ3, "vuzp", "8">;
5016 def VUZPq16 : N2VQShuffle<0b01, 0b00010, IIC_VPERMQ3, "vuzp", "16">;
5017 def VUZPq32 : N2VQShuffle<0b10, 0b00010, IIC_VPERMQ3, "vuzp", "32">;
5019 // VZIP : Vector Zip (Interleave)
5021 def VZIPd8 : N2VDShuffle<0b00, 0b00011, "vzip", "8">;
5022 def VZIPd16 : N2VDShuffle<0b01, 0b00011, "vzip", "16">;
5023 def VZIPd32 : N2VDShuffle<0b10, 0b00011, "vzip", "32">;
5025 def VZIPq8 : N2VQShuffle<0b00, 0b00011, IIC_VPERMQ3, "vzip", "8">;
5026 def VZIPq16 : N2VQShuffle<0b01, 0b00011, IIC_VPERMQ3, "vzip", "16">;
5027 def VZIPq32 : N2VQShuffle<0b10, 0b00011, IIC_VPERMQ3, "vzip", "32">;
5029 // Vector Table Lookup and Table Extension.
5031 // VTBL : Vector Table Lookup
5032 let DecoderMethod = "DecodeTBLInstruction" in {
5034 : N3V<1,1,0b11,0b1000,0,0, (outs DPR:$Vd),
5035 (ins VecListOneD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTB1,
5036 "vtbl", "8", "$Vd, $Vn, $Vm", "",
5037 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbl1 VecListOneD:$Vn, DPR:$Vm)))]>;
5038 let hasExtraSrcRegAllocReq = 1 in {
5040 : N3V<1,1,0b11,0b1001,0,0, (outs DPR:$Vd),
5041 (ins DPR:$Vn, DPR:$tbl2, DPR:$Vm), NVTBLFrm, IIC_VTB2,
5042 "vtbl", "8", "$Vd, \\{$Vn, $tbl2\\}, $Vm", "", []>;
5044 : N3V<1,1,0b11,0b1010,0,0, (outs DPR:$Vd),
5045 (ins DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$Vm), NVTBLFrm, IIC_VTB3,
5046 "vtbl", "8", "$Vd, \\{$Vn, $tbl2, $tbl3\\}, $Vm", "", []>;
5048 : N3V<1,1,0b11,0b1011,0,0, (outs DPR:$Vd),
5049 (ins DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$Vm),
5051 "vtbl", "8", "$Vd, \\{$Vn, $tbl2, $tbl3, $tbl4\\}, $Vm", "", []>;
5052 } // hasExtraSrcRegAllocReq = 1
5055 : PseudoNeonI<(outs DPR:$dst), (ins QPR:$tbl, DPR:$src), IIC_VTB2, "", []>;
5057 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB3, "", []>;
5059 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB4, "", []>;
5061 // VTBX : Vector Table Extension
5063 : N3V<1,1,0b11,0b1000,1,0, (outs DPR:$Vd),
5064 (ins DPR:$orig, VecListOneD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTBX1,
5065 "vtbx", "8", "$Vd, $Vn, $Vm", "$orig = $Vd",
5066 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbx1
5067 DPR:$orig, VecListOneD:$Vn, DPR:$Vm)))]>;
5068 let hasExtraSrcRegAllocReq = 1 in {
5070 : N3V<1,1,0b11,0b1001,1,0, (outs DPR:$Vd),
5071 (ins DPR:$orig, DPR:$Vn, DPR:$tbl2, DPR:$Vm), NVTBLFrm, IIC_VTBX2,
5072 "vtbx", "8", "$Vd, \\{$Vn, $tbl2\\}, $Vm", "$orig = $Vd", []>;
5074 : N3V<1,1,0b11,0b1010,1,0, (outs DPR:$Vd),
5075 (ins DPR:$orig, DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$Vm),
5076 NVTBLFrm, IIC_VTBX3,
5077 "vtbx", "8", "$Vd, \\{$Vn, $tbl2, $tbl3\\}, $Vm",
5080 : N3V<1,1,0b11,0b1011,1,0, (outs DPR:$Vd), (ins DPR:$orig, DPR:$Vn,
5081 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$Vm), NVTBLFrm, IIC_VTBX4,
5082 "vtbx", "8", "$Vd, \\{$Vn, $tbl2, $tbl3, $tbl4\\}, $Vm",
5084 } // hasExtraSrcRegAllocReq = 1
5087 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QPR:$tbl, DPR:$src),
5088 IIC_VTBX2, "$orig = $dst", []>;
5090 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
5091 IIC_VTBX3, "$orig = $dst", []>;
5093 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
5094 IIC_VTBX4, "$orig = $dst", []>;
5095 } // DecoderMethod = "DecodeTBLInstruction"
5097 //===----------------------------------------------------------------------===//
5098 // NEON instructions for single-precision FP math
5099 //===----------------------------------------------------------------------===//
5101 class N2VSPat<SDNode OpNode, NeonI Inst>
5102 : NEONFPPat<(f32 (OpNode SPR:$a)),
5104 (v2f32 (COPY_TO_REGCLASS (Inst
5106 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5107 SPR:$a, ssub_0)), DPR_VFP2)), ssub_0)>;
5109 class N3VSPat<SDNode OpNode, NeonI Inst>
5110 : NEONFPPat<(f32 (OpNode SPR:$a, SPR:$b)),
5112 (v2f32 (COPY_TO_REGCLASS (Inst
5114 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5117 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5118 SPR:$b, ssub_0)), DPR_VFP2)), ssub_0)>;
5120 class N3VSMulOpPat<SDNode MulNode, SDNode OpNode, NeonI Inst>
5121 : NEONFPPat<(f32 (OpNode SPR:$acc, (f32 (MulNode SPR:$a, SPR:$b)))),
5123 (v2f32 (COPY_TO_REGCLASS (Inst
5125 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5128 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5131 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5132 SPR:$b, ssub_0)), DPR_VFP2)), ssub_0)>;
5134 def : N3VSPat<fadd, VADDfd>;
5135 def : N3VSPat<fsub, VSUBfd>;
5136 def : N3VSPat<fmul, VMULfd>;
5137 def : N3VSMulOpPat<fmul, fadd, VMLAfd>,
5138 Requires<[HasNEON, UseNEONForFP, UseFPVMLx]>;
5139 def : N3VSMulOpPat<fmul, fsub, VMLSfd>,
5140 Requires<[HasNEON, UseNEONForFP, UseFPVMLx]>;
5141 def : N2VSPat<fabs, VABSfd>;
5142 def : N2VSPat<fneg, VNEGfd>;
5143 def : N3VSPat<NEONfmax, VMAXfd>;
5144 def : N3VSPat<NEONfmin, VMINfd>;
5145 def : N2VSPat<arm_ftosi, VCVTf2sd>;
5146 def : N2VSPat<arm_ftoui, VCVTf2ud>;
5147 def : N2VSPat<arm_sitof, VCVTs2fd>;
5148 def : N2VSPat<arm_uitof, VCVTu2fd>;
5150 //===----------------------------------------------------------------------===//
5151 // Non-Instruction Patterns
5152 //===----------------------------------------------------------------------===//
5155 def : Pat<(v1i64 (bitconvert (v2i32 DPR:$src))), (v1i64 DPR:$src)>;
5156 def : Pat<(v1i64 (bitconvert (v4i16 DPR:$src))), (v1i64 DPR:$src)>;
5157 def : Pat<(v1i64 (bitconvert (v8i8 DPR:$src))), (v1i64 DPR:$src)>;
5158 def : Pat<(v1i64 (bitconvert (f64 DPR:$src))), (v1i64 DPR:$src)>;
5159 def : Pat<(v1i64 (bitconvert (v2f32 DPR:$src))), (v1i64 DPR:$src)>;
5160 def : Pat<(v2i32 (bitconvert (v1i64 DPR:$src))), (v2i32 DPR:$src)>;
5161 def : Pat<(v2i32 (bitconvert (v4i16 DPR:$src))), (v2i32 DPR:$src)>;
5162 def : Pat<(v2i32 (bitconvert (v8i8 DPR:$src))), (v2i32 DPR:$src)>;
5163 def : Pat<(v2i32 (bitconvert (f64 DPR:$src))), (v2i32 DPR:$src)>;
5164 def : Pat<(v2i32 (bitconvert (v2f32 DPR:$src))), (v2i32 DPR:$src)>;
5165 def : Pat<(v4i16 (bitconvert (v1i64 DPR:$src))), (v4i16 DPR:$src)>;
5166 def : Pat<(v4i16 (bitconvert (v2i32 DPR:$src))), (v4i16 DPR:$src)>;
5167 def : Pat<(v4i16 (bitconvert (v8i8 DPR:$src))), (v4i16 DPR:$src)>;
5168 def : Pat<(v4i16 (bitconvert (f64 DPR:$src))), (v4i16 DPR:$src)>;
5169 def : Pat<(v4i16 (bitconvert (v2f32 DPR:$src))), (v4i16 DPR:$src)>;
5170 def : Pat<(v8i8 (bitconvert (v1i64 DPR:$src))), (v8i8 DPR:$src)>;
5171 def : Pat<(v8i8 (bitconvert (v2i32 DPR:$src))), (v8i8 DPR:$src)>;
5172 def : Pat<(v8i8 (bitconvert (v4i16 DPR:$src))), (v8i8 DPR:$src)>;
5173 def : Pat<(v8i8 (bitconvert (f64 DPR:$src))), (v8i8 DPR:$src)>;
5174 def : Pat<(v8i8 (bitconvert (v2f32 DPR:$src))), (v8i8 DPR:$src)>;
5175 def : Pat<(f64 (bitconvert (v1i64 DPR:$src))), (f64 DPR:$src)>;
5176 def : Pat<(f64 (bitconvert (v2i32 DPR:$src))), (f64 DPR:$src)>;
5177 def : Pat<(f64 (bitconvert (v4i16 DPR:$src))), (f64 DPR:$src)>;
5178 def : Pat<(f64 (bitconvert (v8i8 DPR:$src))), (f64 DPR:$src)>;
5179 def : Pat<(f64 (bitconvert (v2f32 DPR:$src))), (f64 DPR:$src)>;
5180 def : Pat<(v2f32 (bitconvert (f64 DPR:$src))), (v2f32 DPR:$src)>;
5181 def : Pat<(v2f32 (bitconvert (v1i64 DPR:$src))), (v2f32 DPR:$src)>;
5182 def : Pat<(v2f32 (bitconvert (v2i32 DPR:$src))), (v2f32 DPR:$src)>;
5183 def : Pat<(v2f32 (bitconvert (v4i16 DPR:$src))), (v2f32 DPR:$src)>;
5184 def : Pat<(v2f32 (bitconvert (v8i8 DPR:$src))), (v2f32 DPR:$src)>;
5186 def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (v2i64 QPR:$src)>;
5187 def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (v2i64 QPR:$src)>;
5188 def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (v2i64 QPR:$src)>;
5189 def : Pat<(v2i64 (bitconvert (v2f64 QPR:$src))), (v2i64 QPR:$src)>;
5190 def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (v2i64 QPR:$src)>;
5191 def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (v4i32 QPR:$src)>;
5192 def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (v4i32 QPR:$src)>;
5193 def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (v4i32 QPR:$src)>;
5194 def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (v4i32 QPR:$src)>;
5195 def : Pat<(v4i32 (bitconvert (v4f32 QPR:$src))), (v4i32 QPR:$src)>;
5196 def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (v8i16 QPR:$src)>;
5197 def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (v8i16 QPR:$src)>;
5198 def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (v8i16 QPR:$src)>;
5199 def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (v8i16 QPR:$src)>;
5200 def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (v8i16 QPR:$src)>;
5201 def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (v16i8 QPR:$src)>;
5202 def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (v16i8 QPR:$src)>;
5203 def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (v16i8 QPR:$src)>;
5204 def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (v16i8 QPR:$src)>;
5205 def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (v16i8 QPR:$src)>;
5206 def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (v4f32 QPR:$src)>;
5207 def : Pat<(v4f32 (bitconvert (v4i32 QPR:$src))), (v4f32 QPR:$src)>;
5208 def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (v4f32 QPR:$src)>;
5209 def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (v4f32 QPR:$src)>;
5210 def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (v4f32 QPR:$src)>;
5211 def : Pat<(v2f64 (bitconvert (v2i64 QPR:$src))), (v2f64 QPR:$src)>;
5212 def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (v2f64 QPR:$src)>;
5213 def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 QPR:$src)>;
5214 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 QPR:$src)>;
5215 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 QPR:$src)>;
5218 //===----------------------------------------------------------------------===//
5219 // Assembler aliases
5222 // VAND/VEOR/VORR accept but do not require a type suffix.
5223 defm : VFPDTAnyInstAlias<"vand${p}", "$Vd, $Vn, $Vm",
5224 (VANDd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
5225 defm : VFPDTAnyInstAlias<"vand${p}", "$Vd, $Vn, $Vm",
5226 (VANDq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
5227 defm : VFPDTAnyInstAlias<"veor${p}", "$Vd, $Vn, $Vm",
5228 (VEORd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
5229 defm : VFPDTAnyInstAlias<"veor${p}", "$Vd, $Vn, $Vm",
5230 (VEORq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
5231 defm : VFPDTAnyInstAlias<"vorr${p}", "$Vd, $Vn, $Vm",
5232 (VORRd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
5233 defm : VFPDTAnyInstAlias<"vorr${p}", "$Vd, $Vn, $Vm",
5234 (VORRq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
5236 // VLD1 requires a size suffix, but also accepts type specific variants.
5237 // Load one D register.
5238 defm : VFPDT8ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5239 (VLD1d8 VecListOneD:$Vd, addrmode6:$Rn, pred:$p)>;
5240 defm : VFPDT16ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5241 (VLD1d16 VecListOneD:$Vd, addrmode6:$Rn, pred:$p)>;
5242 defm : VFPDT32ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5243 (VLD1d32 VecListOneD:$Vd, addrmode6:$Rn, pred:$p)>;
5244 defm : VFPDT64ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5245 (VLD1d64 VecListOneD:$Vd, addrmode6:$Rn, pred:$p)>;
5246 // with writeback, fixed stride
5247 defm : VFPDT8ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5248 (VLD1d8wb_fixed VecListOneD:$Vd, zero_reg, addrmode6:$Rn, pred:$p)>;
5249 defm : VFPDT16ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5250 (VLD1d16wb_fixed VecListOneD:$Vd, zero_reg, addrmode6:$Rn, pred:$p)>;
5251 defm : VFPDT32ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5252 (VLD1d32wb_fixed VecListOneD:$Vd, zero_reg, addrmode6:$Rn, pred:$p)>;
5253 defm : VFPDT64ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5254 (VLD1d64wb_fixed VecListOneD:$Vd, zero_reg, addrmode6:$Rn, pred:$p)>;
5255 // with writeback, register stride
5256 defm : VFPDT8ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5257 (VLD1d8wb_register VecListOneD:$Vd, zero_reg, addrmode6:$Rn,
5258 rGPR:$Rm, pred:$p)>;
5259 defm : VFPDT16ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5260 (VLD1d16wb_register VecListOneD:$Vd, zero_reg, addrmode6:$Rn,
5261 rGPR:$Rm, pred:$p)>;
5262 defm : VFPDT32ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5263 (VLD1d32wb_register VecListOneD:$Vd, zero_reg, addrmode6:$Rn,
5264 rGPR:$Rm, pred:$p)>;
5265 defm : VFPDT64ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5266 (VLD1d64wb_register VecListOneD:$Vd, zero_reg, addrmode6:$Rn,
5267 rGPR:$Rm, pred:$p)>;
5269 // Load two D registers.
5270 defm : VFPDT8ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5271 (VLD1q8 VecListTwoD:$Vd, addrmode6:$Rn, pred:$p)>;
5272 defm : VFPDT16ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5273 (VLD1q16 VecListTwoD:$Vd, addrmode6:$Rn, pred:$p)>;
5274 defm : VFPDT32ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5275 (VLD1q32 VecListTwoD:$Vd, addrmode6:$Rn, pred:$p)>;
5276 defm : VFPDT64ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5277 (VLD1q64 VecListTwoD:$Vd, addrmode6:$Rn, pred:$p)>;
5278 // with writeback, fixed stride
5279 defm : VFPDT8ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5280 (VLD1q8wb_fixed VecListTwoD:$Vd, zero_reg, addrmode6:$Rn, pred:$p)>;
5281 defm : VFPDT16ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5282 (VLD1q16wb_fixed VecListTwoD:$Vd, zero_reg, addrmode6:$Rn, pred:$p)>;
5283 defm : VFPDT32ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5284 (VLD1q32wb_fixed VecListTwoD:$Vd, zero_reg, addrmode6:$Rn, pred:$p)>;
5285 defm : VFPDT64ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5286 (VLD1q64wb_fixed VecListTwoD:$Vd, zero_reg, addrmode6:$Rn, pred:$p)>;
5287 // with writeback, register stride
5288 defm : VFPDT8ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5289 (VLD1q8wb_register VecListTwoD:$Vd, zero_reg, addrmode6:$Rn,
5290 rGPR:$Rm, pred:$p)>;
5291 defm : VFPDT16ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5292 (VLD1q16wb_register VecListTwoD:$Vd, zero_reg, addrmode6:$Rn,
5293 rGPR:$Rm, pred:$p)>;
5294 defm : VFPDT32ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5295 (VLD1q32wb_register VecListTwoD:$Vd, zero_reg, addrmode6:$Rn,
5296 rGPR:$Rm, pred:$p)>;
5297 defm : VFPDT64ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5298 (VLD1q64wb_register VecListTwoD:$Vd, zero_reg, addrmode6:$Rn,
5299 rGPR:$Rm, pred:$p)>;
5301 // Load three D registers.
5302 defm : VFPDT8ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5303 (VLD1d8T VecListThreeD:$Vd, addrmode6:$Rn, pred:$p)>;
5304 defm : VFPDT16ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5305 (VLD1d16T VecListThreeD:$Vd, addrmode6:$Rn, pred:$p)>;
5306 defm : VFPDT32ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5307 (VLD1d32T VecListThreeD:$Vd, addrmode6:$Rn, pred:$p)>;
5308 defm : VFPDT64ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5309 (VLD1d64T VecListThreeD:$Vd, addrmode6:$Rn, pred:$p)>;
5310 // with writeback, fixed stride
5311 defm : VFPDT8ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5312 (VLD1d8Twb_fixed VecListThreeD:$Vd, zero_reg,
5313 addrmode6:$Rn, pred:$p)>;
5314 defm : VFPDT16ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5315 (VLD1d16Twb_fixed VecListThreeD:$Vd, zero_reg,
5316 addrmode6:$Rn, pred:$p)>;
5317 defm : VFPDT32ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5318 (VLD1d32Twb_fixed VecListThreeD:$Vd, zero_reg,
5319 addrmode6:$Rn, pred:$p)>;
5320 defm : VFPDT64ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5321 (VLD1d64Twb_fixed VecListThreeD:$Vd, zero_reg,
5322 addrmode6:$Rn, pred:$p)>;
5323 // with writeback, register stride
5324 defm : VFPDT8ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5325 (VLD1d8Twb_register VecListThreeD:$Vd, zero_reg,
5326 addrmode6:$Rn, rGPR:$Rm, pred:$p)>;
5327 defm : VFPDT16ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5328 (VLD1d16Twb_register VecListThreeD:$Vd, zero_reg,
5329 addrmode6:$Rn, rGPR:$Rm, pred:$p)>;
5330 defm : VFPDT32ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5331 (VLD1d32Twb_register VecListThreeD:$Vd, zero_reg,
5332 addrmode6:$Rn, rGPR:$Rm, pred:$p)>;
5333 defm : VFPDT64ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5334 (VLD1d64Twb_register VecListThreeD:$Vd, zero_reg,
5335 addrmode6:$Rn, rGPR:$Rm, pred:$p)>;
5338 // Load four D registers.
5339 defm : VFPDT8ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5340 (VLD1d8Q VecListFourD:$Vd, addrmode6:$Rn, pred:$p)>;
5341 defm : VFPDT16ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5342 (VLD1d16Q VecListFourD:$Vd, addrmode6:$Rn, pred:$p)>;
5343 defm : VFPDT32ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5344 (VLD1d32Q VecListFourD:$Vd, addrmode6:$Rn, pred:$p)>;
5345 defm : VFPDT64ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5346 (VLD1d64Q VecListFourD:$Vd, addrmode6:$Rn, pred:$p)>;
5347 // with writeback, fixed stride
5348 defm : VFPDT8ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5349 (VLD1d8Qwb_fixed VecListFourD:$Vd, zero_reg,
5350 addrmode6:$Rn, pred:$p)>;
5351 defm : VFPDT16ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5352 (VLD1d16Qwb_fixed VecListFourD:$Vd, zero_reg,
5353 addrmode6:$Rn, pred:$p)>;
5354 defm : VFPDT32ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5355 (VLD1d32Qwb_fixed VecListFourD:$Vd, zero_reg,
5356 addrmode6:$Rn, pred:$p)>;
5357 defm : VFPDT64ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5358 (VLD1d64Qwb_fixed VecListFourD:$Vd, zero_reg,
5359 addrmode6:$Rn, pred:$p)>;
5360 // with writeback, register stride
5361 defm : VFPDT8ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5362 (VLD1d8Qwb_register VecListFourD:$Vd, zero_reg,
5363 addrmode6:$Rn, rGPR:$Rm, pred:$p)>;
5364 defm : VFPDT16ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5365 (VLD1d16Qwb_register VecListFourD:$Vd, zero_reg,
5366 addrmode6:$Rn, rGPR:$Rm, pred:$p)>;
5367 defm : VFPDT32ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5368 (VLD1d32Qwb_register VecListFourD:$Vd, zero_reg,
5369 addrmode6:$Rn, rGPR:$Rm, pred:$p)>;
5370 defm : VFPDT64ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5371 (VLD1d64Qwb_register VecListFourD:$Vd, zero_reg,
5372 addrmode6:$Rn, rGPR:$Rm, pred:$p)>;
5374 // VST1 requires a size suffix, but also accepts type specific variants.
5375 // Store one D register.
5376 defm : VFPDT8ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5377 (VST1d8 addrmode6:$Rn, VecListOneD:$Vd, pred:$p)>;
5378 defm : VFPDT16ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5379 (VST1d16 addrmode6:$Rn, VecListOneD:$Vd, pred:$p)>;
5380 defm : VFPDT32ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5381 (VST1d32 addrmode6:$Rn, VecListOneD:$Vd, pred:$p)>;
5382 defm : VFPDT64ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5383 (VST1d64 addrmode6:$Rn, VecListOneD:$Vd, pred:$p)>;
5384 // with writeback, fixed stride
5385 defm : VFPDT8ReqInstAlias<"vst1${p}", "$Vd, $Rn!",
5386 (VST1d8wb_fixed zero_reg, addrmode6:$Rn, VecListOneD:$Vd, pred:$p)>;
5387 defm : VFPDT16ReqInstAlias<"vst1${p}", "$Vd, $Rn!",
5388 (VST1d16wb_fixed zero_reg, addrmode6:$Rn, VecListOneD:$Vd, pred:$p)>;
5389 defm : VFPDT32ReqInstAlias<"vst1${p}", "$Vd, $Rn!",
5390 (VST1d32wb_fixed zero_reg, addrmode6:$Rn, VecListOneD:$Vd, pred:$p)>;
5391 defm : VFPDT64ReqInstAlias<"vst1${p}", "$Vd, $Rn!",
5392 (VST1d64wb_fixed zero_reg, addrmode6:$Rn, VecListOneD:$Vd, pred:$p)>;
5393 // with writeback, register stride
5394 defm : VFPDT8ReqInstAlias<"vst1${p}", "$Vd, $Rn, $Rm",
5395 (VST1d8wb_register zero_reg, addrmode6:$Rn, rGPR:$Rm,
5396 VecListOneD:$Vd, pred:$p)>;
5397 defm : VFPDT16ReqInstAlias<"vst1${p}", "$Vd, $Rn, $Rm",
5398 (VST1d16wb_register zero_reg, addrmode6:$Rn, rGPR:$Rm,
5399 VecListOneD:$Vd, pred:$p)>;
5400 defm : VFPDT32ReqInstAlias<"vst1${p}", "$Vd, $Rn, $Rm",
5401 (VST1d32wb_register zero_reg, addrmode6:$Rn, rGPR:$Rm,
5402 VecListOneD:$Vd, pred:$p)>;
5403 defm : VFPDT64ReqInstAlias<"vst1${p}", "$Vd, $Rn, $Rm",
5404 (VST1d64wb_register zero_reg, addrmode6:$Rn, rGPR:$Rm,
5405 VecListOneD:$Vd, pred:$p)>;
5407 // Store two D registers.
5408 defm : VFPDT8ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5409 (VST1q8 addrmode6:$Rn, VecListTwoD:$Vd, pred:$p)>;
5410 defm : VFPDT16ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5411 (VST1q16 addrmode6:$Rn, VecListTwoD:$Vd, pred:$p)>;
5412 defm : VFPDT32ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5413 (VST1q32 addrmode6:$Rn, VecListTwoD:$Vd, pred:$p)>;
5414 defm : VFPDT64ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5415 (VST1q64 addrmode6:$Rn, VecListTwoD:$Vd, pred:$p)>;
5416 // with writeback, fixed stride
5417 defm : VFPDT8ReqInstAlias<"vst1${p}", "$Vd, $Rn!",
5418 (VST1q8wb_fixed zero_reg, addrmode6:$Rn, VecListTwoD:$Vd, pred:$p)>;
5419 defm : VFPDT16ReqInstAlias<"vst1${p}", "$Vd, $Rn!",
5420 (VST1q16wb_fixed zero_reg, addrmode6:$Rn, VecListTwoD:$Vd, pred:$p)>;
5421 defm : VFPDT32ReqInstAlias<"vst1${p}", "$Vd, $Rn!",
5422 (VST1q32wb_fixed zero_reg, addrmode6:$Rn, VecListTwoD:$Vd, pred:$p)>;
5423 defm : VFPDT64ReqInstAlias<"vst1${p}", "$Vd, $Rn!",
5424 (VST1q64wb_fixed zero_reg, addrmode6:$Rn, VecListTwoD:$Vd, pred:$p)>;
5425 // with writeback, register stride
5426 defm : VFPDT8ReqInstAlias<"vst1${p}", "$Vd, $Rn, $Rm",
5427 (VST1q8wb_register zero_reg, addrmode6:$Rn,
5428 rGPR:$Rm, VecListTwoD:$Vd, pred:$p)>;
5429 defm : VFPDT16ReqInstAlias<"vst1${p}", "$Vd, $Rn, $Rm",
5430 (VST1q16wb_register zero_reg, addrmode6:$Rn,
5431 rGPR:$Rm, VecListTwoD:$Vd, pred:$p)>;
5432 defm : VFPDT32ReqInstAlias<"vst1${p}", "$Vd, $Rn, $Rm",
5433 (VST1q32wb_register zero_reg, addrmode6:$Rn,
5434 rGPR:$Rm, VecListTwoD:$Vd, pred:$p)>;
5435 defm : VFPDT64ReqInstAlias<"vst1${p}", "$Vd, $Rn, $Rm",
5436 (VST1q64wb_register zero_reg, addrmode6:$Rn,
5437 rGPR:$Rm, VecListTwoD:$Vd, pred:$p)>;
5439 // FIXME: The three and four register VST1 instructions haven't been moved
5440 // to the VecList* encoding yet, so we can't do assembly parsing support
5441 // for them. Uncomment these when that happens.
5442 // Load three D registers.
5443 //defm : VFPDT8ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5444 // (VST1d8T addrmode6:$Rn, VecListThreeD:$Vd, pred:$p)>;
5445 //defm : VFPDT16ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5446 // (VST1d16T addrmode6:$Rn, VecListThreeD:$Vd, pred:$p)>;
5447 //defm : VFPDT32ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5448 // (VST1d32T addrmode6:$Rn, VecListThreeD:$Vd, pred:$p)>;
5449 //defm : VFPDT64ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5450 // (VST1d64T addrmode6:$Rn, VecListThreeD:$Vd, pred:$p)>;
5452 // Load four D registers.
5453 //defm : VFPDT8ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5454 // (VST1d8Q addrmode6:$Rn, VecListFourD:$Vd, pred:$p)>;
5455 //defm : VFPDT16ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5456 // (VST1d16Q addrmode6:$Rn, VecListFourD:$Vd, pred:$p)>;
5457 //defm : VFPDT32ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5458 // (VST1d32Q addrmode6:$Rn, VecListFourD:$Vd, pred:$p)>;
5459 //defm : VFPDT64ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5460 // (VST1d64Q addrmode6:$Rn, VecListFourD:$Vd, pred:$p)>;
5463 // VTRN instructions data type suffix aliases for more-specific types.
5464 defm : VFPDT8ReqInstAlias <"vtrn${p}", "$Dd, $Dm",
5465 (VTRNd8 DPR:$Dd, DPR:$Dm, pred:$p)>;
5466 defm : VFPDT16ReqInstAlias<"vtrn${p}", "$Dd, $Dm",
5467 (VTRNd16 DPR:$Dd, DPR:$Dm, pred:$p)>;
5468 defm : VFPDT32ReqInstAlias<"vtrn${p}", "$Dd, $Dm",
5469 (VTRNd32 DPR:$Dd, DPR:$Dm, pred:$p)>;
5471 defm : VFPDT8ReqInstAlias <"vtrn${p}", "$Qd, $Qm",
5472 (VTRNq8 QPR:$Qd, QPR:$Qm, pred:$p)>;
5473 defm : VFPDT16ReqInstAlias<"vtrn${p}", "$Qd, $Qm",
5474 (VTRNq16 QPR:$Qd, QPR:$Qm, pred:$p)>;
5475 defm : VFPDT32ReqInstAlias<"vtrn${p}", "$Qd, $Qm",
5476 (VTRNq32 QPR:$Qd, QPR:$Qm, pred:$p)>;