1 //===- ARMInstrNEON.td - NEON support for ARM -----------------------------===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file describes the ARM NEON instruction set.
12 //===----------------------------------------------------------------------===//
15 //===----------------------------------------------------------------------===//
16 // NEON-specific Operands.
17 //===----------------------------------------------------------------------===//
18 def nModImm : Operand<i32> {
19 let PrintMethod = "printNEONModImmOperand";
22 def nImmSplatI8AsmOperand : AsmOperandClass { let Name = "NEONi8splat"; }
23 def nImmSplatI8 : Operand<i32> {
24 let PrintMethod = "printNEONModImmOperand";
25 let ParserMatchClass = nImmSplatI8AsmOperand;
27 def nImmSplatI16AsmOperand : AsmOperandClass { let Name = "NEONi16splat"; }
28 def nImmSplatI16 : Operand<i32> {
29 let PrintMethod = "printNEONModImmOperand";
30 let ParserMatchClass = nImmSplatI16AsmOperand;
32 def nImmSplatI32AsmOperand : AsmOperandClass { let Name = "NEONi32splat"; }
33 def nImmSplatI32 : Operand<i32> {
34 let PrintMethod = "printNEONModImmOperand";
35 let ParserMatchClass = nImmSplatI32AsmOperand;
37 def nImmVMOVI32AsmOperand : AsmOperandClass { let Name = "NEONi32vmov"; }
38 def nImmVMOVI32 : Operand<i32> {
39 let PrintMethod = "printNEONModImmOperand";
40 let ParserMatchClass = nImmVMOVI32AsmOperand;
42 def nImmVMOVF32 : Operand<i32> {
43 let PrintMethod = "printFPImmOperand";
44 let ParserMatchClass = FPImmOperand;
46 def nImmSplatI64AsmOperand : AsmOperandClass { let Name = "NEONi64splat"; }
47 def nImmSplatI64 : Operand<i32> {
48 let PrintMethod = "printNEONModImmOperand";
49 let ParserMatchClass = nImmSplatI64AsmOperand;
52 def VectorIndex8Operand : AsmOperandClass { let Name = "VectorIndex8"; }
53 def VectorIndex16Operand : AsmOperandClass { let Name = "VectorIndex16"; }
54 def VectorIndex32Operand : AsmOperandClass { let Name = "VectorIndex32"; }
55 def VectorIndex8 : Operand<i32>, ImmLeaf<i32, [{
56 return ((uint64_t)Imm) < 8;
58 let ParserMatchClass = VectorIndex8Operand;
59 let PrintMethod = "printVectorIndex";
60 let MIOperandInfo = (ops i32imm);
62 def VectorIndex16 : Operand<i32>, ImmLeaf<i32, [{
63 return ((uint64_t)Imm) < 4;
65 let ParserMatchClass = VectorIndex16Operand;
66 let PrintMethod = "printVectorIndex";
67 let MIOperandInfo = (ops i32imm);
69 def VectorIndex32 : Operand<i32>, ImmLeaf<i32, [{
70 return ((uint64_t)Imm) < 2;
72 let ParserMatchClass = VectorIndex32Operand;
73 let PrintMethod = "printVectorIndex";
74 let MIOperandInfo = (ops i32imm);
77 // Register list of one D register.
78 def VecListOneDAsmOperand : AsmOperandClass {
79 let Name = "VecListOneD";
80 let ParserMethod = "parseVectorList";
81 let RenderMethod = "addVecListOperands";
83 def VecListOneD : RegisterOperand<DPR, "printVectorListOne"> {
84 let ParserMatchClass = VecListOneDAsmOperand;
86 // Register list of two sequential D registers.
87 def VecListTwoDAsmOperand : AsmOperandClass {
88 let Name = "VecListTwoD";
89 let ParserMethod = "parseVectorList";
90 let RenderMethod = "addVecListOperands";
92 def VecListTwoD : RegisterOperand<DPR, "printVectorListTwo"> {
93 let ParserMatchClass = VecListTwoDAsmOperand;
95 // Register list of three sequential D registers.
96 def VecListThreeDAsmOperand : AsmOperandClass {
97 let Name = "VecListThreeD";
98 let ParserMethod = "parseVectorList";
99 let RenderMethod = "addVecListOperands";
101 def VecListThreeD : RegisterOperand<DPR, "printVectorListThree"> {
102 let ParserMatchClass = VecListThreeDAsmOperand;
104 // Register list of four sequential D registers.
105 def VecListFourDAsmOperand : AsmOperandClass {
106 let Name = "VecListFourD";
107 let ParserMethod = "parseVectorList";
108 let RenderMethod = "addVecListOperands";
110 def VecListFourD : RegisterOperand<DPR, "printVectorListFour"> {
111 let ParserMatchClass = VecListFourDAsmOperand;
113 // Register list of two D registers spaced by 2 (two sequential Q registers).
114 def VecListTwoQAsmOperand : AsmOperandClass {
115 let Name = "VecListTwoQ";
116 let ParserMethod = "parseVectorList";
117 let RenderMethod = "addVecListOperands";
119 def VecListTwoQ : RegisterOperand<DPR, "printVectorListTwo"> {
120 let ParserMatchClass = VecListTwoQAsmOperand;
123 // Register list of one D register, with "all lanes" subscripting.
124 def VecListOneDAllLanesAsmOperand : AsmOperandClass {
125 let Name = "VecListOneDAllLanes";
126 let ParserMethod = "parseVectorList";
127 let RenderMethod = "addVecListOperands";
129 def VecListOneDAllLanes : RegisterOperand<DPR, "printVectorListOneAllLanes"> {
130 let ParserMatchClass = VecListOneDAllLanesAsmOperand;
132 // Register list of two D registers, with "all lanes" subscripting.
133 def VecListTwoDAllLanesAsmOperand : AsmOperandClass {
134 let Name = "VecListTwoDAllLanes";
135 let ParserMethod = "parseVectorList";
136 let RenderMethod = "addVecListOperands";
138 def VecListTwoDAllLanes : RegisterOperand<DPR, "printVectorListTwoAllLanes"> {
139 let ParserMatchClass = VecListTwoDAllLanesAsmOperand;
142 //===----------------------------------------------------------------------===//
143 // NEON-specific DAG Nodes.
144 //===----------------------------------------------------------------------===//
146 def SDTARMVCMP : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisSameAs<1, 2>]>;
147 def SDTARMVCMPZ : SDTypeProfile<1, 1, []>;
149 def NEONvceq : SDNode<"ARMISD::VCEQ", SDTARMVCMP>;
150 def NEONvceqz : SDNode<"ARMISD::VCEQZ", SDTARMVCMPZ>;
151 def NEONvcge : SDNode<"ARMISD::VCGE", SDTARMVCMP>;
152 def NEONvcgez : SDNode<"ARMISD::VCGEZ", SDTARMVCMPZ>;
153 def NEONvclez : SDNode<"ARMISD::VCLEZ", SDTARMVCMPZ>;
154 def NEONvcgeu : SDNode<"ARMISD::VCGEU", SDTARMVCMP>;
155 def NEONvcgt : SDNode<"ARMISD::VCGT", SDTARMVCMP>;
156 def NEONvcgtz : SDNode<"ARMISD::VCGTZ", SDTARMVCMPZ>;
157 def NEONvcltz : SDNode<"ARMISD::VCLTZ", SDTARMVCMPZ>;
158 def NEONvcgtu : SDNode<"ARMISD::VCGTU", SDTARMVCMP>;
159 def NEONvtst : SDNode<"ARMISD::VTST", SDTARMVCMP>;
161 // Types for vector shift by immediates. The "SHX" version is for long and
162 // narrow operations where the source and destination vectors have different
163 // types. The "SHINS" version is for shift and insert operations.
164 def SDTARMVSH : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
166 def SDTARMVSHX : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
168 def SDTARMVSHINS : SDTypeProfile<1, 3, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
169 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
171 def NEONvshl : SDNode<"ARMISD::VSHL", SDTARMVSH>;
172 def NEONvshrs : SDNode<"ARMISD::VSHRs", SDTARMVSH>;
173 def NEONvshru : SDNode<"ARMISD::VSHRu", SDTARMVSH>;
174 def NEONvshlls : SDNode<"ARMISD::VSHLLs", SDTARMVSHX>;
175 def NEONvshllu : SDNode<"ARMISD::VSHLLu", SDTARMVSHX>;
176 def NEONvshlli : SDNode<"ARMISD::VSHLLi", SDTARMVSHX>;
177 def NEONvshrn : SDNode<"ARMISD::VSHRN", SDTARMVSHX>;
179 def NEONvrshrs : SDNode<"ARMISD::VRSHRs", SDTARMVSH>;
180 def NEONvrshru : SDNode<"ARMISD::VRSHRu", SDTARMVSH>;
181 def NEONvrshrn : SDNode<"ARMISD::VRSHRN", SDTARMVSHX>;
183 def NEONvqshls : SDNode<"ARMISD::VQSHLs", SDTARMVSH>;
184 def NEONvqshlu : SDNode<"ARMISD::VQSHLu", SDTARMVSH>;
185 def NEONvqshlsu : SDNode<"ARMISD::VQSHLsu", SDTARMVSH>;
186 def NEONvqshrns : SDNode<"ARMISD::VQSHRNs", SDTARMVSHX>;
187 def NEONvqshrnu : SDNode<"ARMISD::VQSHRNu", SDTARMVSHX>;
188 def NEONvqshrnsu : SDNode<"ARMISD::VQSHRNsu", SDTARMVSHX>;
190 def NEONvqrshrns : SDNode<"ARMISD::VQRSHRNs", SDTARMVSHX>;
191 def NEONvqrshrnu : SDNode<"ARMISD::VQRSHRNu", SDTARMVSHX>;
192 def NEONvqrshrnsu : SDNode<"ARMISD::VQRSHRNsu", SDTARMVSHX>;
194 def NEONvsli : SDNode<"ARMISD::VSLI", SDTARMVSHINS>;
195 def NEONvsri : SDNode<"ARMISD::VSRI", SDTARMVSHINS>;
197 def SDTARMVGETLN : SDTypeProfile<1, 2, [SDTCisVT<0, i32>, SDTCisInt<1>,
199 def NEONvgetlaneu : SDNode<"ARMISD::VGETLANEu", SDTARMVGETLN>;
200 def NEONvgetlanes : SDNode<"ARMISD::VGETLANEs", SDTARMVGETLN>;
202 def SDTARMVMOVIMM : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisVT<1, i32>]>;
203 def NEONvmovImm : SDNode<"ARMISD::VMOVIMM", SDTARMVMOVIMM>;
204 def NEONvmvnImm : SDNode<"ARMISD::VMVNIMM", SDTARMVMOVIMM>;
205 def NEONvmovFPImm : SDNode<"ARMISD::VMOVFPIMM", SDTARMVMOVIMM>;
207 def SDTARMVORRIMM : SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
209 def NEONvorrImm : SDNode<"ARMISD::VORRIMM", SDTARMVORRIMM>;
210 def NEONvbicImm : SDNode<"ARMISD::VBICIMM", SDTARMVORRIMM>;
212 def NEONvbsl : SDNode<"ARMISD::VBSL",
213 SDTypeProfile<1, 3, [SDTCisVec<0>,
216 SDTCisSameAs<0, 3>]>>;
218 def NEONvdup : SDNode<"ARMISD::VDUP", SDTypeProfile<1, 1, [SDTCisVec<0>]>>;
220 // VDUPLANE can produce a quad-register result from a double-register source,
221 // so the result is not constrained to match the source.
222 def NEONvduplane : SDNode<"ARMISD::VDUPLANE",
223 SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisVec<1>,
226 def SDTARMVEXT : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
227 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
228 def NEONvext : SDNode<"ARMISD::VEXT", SDTARMVEXT>;
230 def SDTARMVSHUF : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisSameAs<0, 1>]>;
231 def NEONvrev64 : SDNode<"ARMISD::VREV64", SDTARMVSHUF>;
232 def NEONvrev32 : SDNode<"ARMISD::VREV32", SDTARMVSHUF>;
233 def NEONvrev16 : SDNode<"ARMISD::VREV16", SDTARMVSHUF>;
235 def SDTARMVSHUF2 : SDTypeProfile<2, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
237 SDTCisSameAs<0, 3>]>;
238 def NEONzip : SDNode<"ARMISD::VZIP", SDTARMVSHUF2>;
239 def NEONuzp : SDNode<"ARMISD::VUZP", SDTARMVSHUF2>;
240 def NEONtrn : SDNode<"ARMISD::VTRN", SDTARMVSHUF2>;
242 def SDTARMVMULL : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
243 SDTCisSameAs<1, 2>]>;
244 def NEONvmulls : SDNode<"ARMISD::VMULLs", SDTARMVMULL>;
245 def NEONvmullu : SDNode<"ARMISD::VMULLu", SDTARMVMULL>;
247 def SDTARMFMAX : SDTypeProfile<1, 2, [SDTCisVT<0, f32>, SDTCisSameAs<0, 1>,
248 SDTCisSameAs<0, 2>]>;
249 def NEONfmax : SDNode<"ARMISD::FMAX", SDTARMFMAX>;
250 def NEONfmin : SDNode<"ARMISD::FMIN", SDTARMFMAX>;
252 def NEONimmAllZerosV: PatLeaf<(NEONvmovImm (i32 timm)), [{
253 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
254 unsigned EltBits = 0;
255 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
256 return (EltBits == 32 && EltVal == 0);
259 def NEONimmAllOnesV: PatLeaf<(NEONvmovImm (i32 timm)), [{
260 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
261 unsigned EltBits = 0;
262 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
263 return (EltBits == 8 && EltVal == 0xff);
266 //===----------------------------------------------------------------------===//
267 // NEON load / store instructions
268 //===----------------------------------------------------------------------===//
270 // Use VLDM to load a Q register as a D register pair.
271 // This is a pseudo instruction that is expanded to VLDMD after reg alloc.
273 : PseudoVFPLdStM<(outs QPR:$dst), (ins GPR:$Rn),
275 [(set QPR:$dst, (v2f64 (load GPR:$Rn)))]>;
277 // Use VSTM to store a Q register as a D register pair.
278 // This is a pseudo instruction that is expanded to VSTMD after reg alloc.
280 : PseudoVFPLdStM<(outs), (ins QPR:$src, GPR:$Rn),
282 [(store (v2f64 QPR:$src), GPR:$Rn)]>;
284 // Classes for VLD* pseudo-instructions with multi-register operands.
285 // These are expanded to real instructions after register allocation.
286 class VLDQPseudo<InstrItinClass itin>
287 : PseudoNLdSt<(outs QPR:$dst), (ins addrmode6:$addr), itin, "">;
288 class VLDQWBPseudo<InstrItinClass itin>
289 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
290 (ins addrmode6:$addr, am6offset:$offset), itin,
292 class VLDQWBfixedPseudo<InstrItinClass itin>
293 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
294 (ins addrmode6:$addr), itin,
296 class VLDQWBregisterPseudo<InstrItinClass itin>
297 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
298 (ins addrmode6:$addr, rGPR:$offset), itin,
300 class VLDQQPseudo<InstrItinClass itin>
301 : PseudoNLdSt<(outs QQPR:$dst), (ins addrmode6:$addr), itin, "">;
302 class VLDQQWBPseudo<InstrItinClass itin>
303 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
304 (ins addrmode6:$addr, am6offset:$offset), itin,
306 class VLDQQQQPseudo<InstrItinClass itin>
307 : PseudoNLdSt<(outs QQQQPR:$dst), (ins addrmode6:$addr, QQQQPR:$src),itin,
309 class VLDQQQQWBPseudo<InstrItinClass itin>
310 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
311 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
312 "$addr.addr = $wb, $src = $dst">;
314 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
316 // VLD1 : Vector Load (multiple single elements)
317 class VLD1D<bits<4> op7_4, string Dt>
318 : NLdSt<0,0b10,0b0111,op7_4, (outs VecListOneD:$Vd),
319 (ins addrmode6:$Rn), IIC_VLD1,
320 "vld1", Dt, "$Vd, $Rn", "", []> {
323 let DecoderMethod = "DecodeVLDInstruction";
325 class VLD1Q<bits<4> op7_4, string Dt>
326 : NLdSt<0,0b10,0b1010,op7_4, (outs VecListTwoD:$Vd),
327 (ins addrmode6:$Rn), IIC_VLD1x2,
328 "vld1", Dt, "$Vd, $Rn", "", []> {
330 let Inst{5-4} = Rn{5-4};
331 let DecoderMethod = "DecodeVLDInstruction";
334 def VLD1d8 : VLD1D<{0,0,0,?}, "8">;
335 def VLD1d16 : VLD1D<{0,1,0,?}, "16">;
336 def VLD1d32 : VLD1D<{1,0,0,?}, "32">;
337 def VLD1d64 : VLD1D<{1,1,0,?}, "64">;
339 def VLD1q8 : VLD1Q<{0,0,?,?}, "8">;
340 def VLD1q16 : VLD1Q<{0,1,?,?}, "16">;
341 def VLD1q32 : VLD1Q<{1,0,?,?}, "32">;
342 def VLD1q64 : VLD1Q<{1,1,?,?}, "64">;
344 def VLD1q8Pseudo : VLDQPseudo<IIC_VLD1x2>;
345 def VLD1q16Pseudo : VLDQPseudo<IIC_VLD1x2>;
346 def VLD1q32Pseudo : VLDQPseudo<IIC_VLD1x2>;
347 def VLD1q64Pseudo : VLDQPseudo<IIC_VLD1x2>;
349 // ...with address register writeback:
350 multiclass VLD1DWB<bits<4> op7_4, string Dt> {
351 def _fixed : NLdSt<0,0b10, 0b0111,op7_4, (outs VecListOneD:$Vd, GPR:$wb),
352 (ins addrmode6:$Rn), IIC_VLD1u,
353 "vld1", Dt, "$Vd, $Rn!",
354 "$Rn.addr = $wb", []> {
355 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
357 let DecoderMethod = "DecodeVLDInstruction";
358 let AsmMatchConverter = "cvtVLDwbFixed";
360 def _register : NLdSt<0,0b10,0b0111,op7_4, (outs VecListOneD:$Vd, GPR:$wb),
361 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1u,
362 "vld1", Dt, "$Vd, $Rn, $Rm",
363 "$Rn.addr = $wb", []> {
365 let DecoderMethod = "DecodeVLDInstruction";
366 let AsmMatchConverter = "cvtVLDwbRegister";
369 multiclass VLD1QWB<bits<4> op7_4, string Dt> {
370 def _fixed : NLdSt<0,0b10,0b1010,op7_4, (outs VecListTwoD:$Vd, GPR:$wb),
371 (ins addrmode6:$Rn), IIC_VLD1x2u,
372 "vld1", Dt, "$Vd, $Rn!",
373 "$Rn.addr = $wb", []> {
374 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
375 let Inst{5-4} = Rn{5-4};
376 let DecoderMethod = "DecodeVLDInstruction";
377 let AsmMatchConverter = "cvtVLDwbFixed";
379 def _register : NLdSt<0,0b10,0b1010,op7_4, (outs VecListTwoD:$Vd, GPR:$wb),
380 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1x2u,
381 "vld1", Dt, "$Vd, $Rn, $Rm",
382 "$Rn.addr = $wb", []> {
383 let Inst{5-4} = Rn{5-4};
384 let DecoderMethod = "DecodeVLDInstruction";
385 let AsmMatchConverter = "cvtVLDwbRegister";
389 defm VLD1d8wb : VLD1DWB<{0,0,0,?}, "8">;
390 defm VLD1d16wb : VLD1DWB<{0,1,0,?}, "16">;
391 defm VLD1d32wb : VLD1DWB<{1,0,0,?}, "32">;
392 defm VLD1d64wb : VLD1DWB<{1,1,0,?}, "64">;
393 defm VLD1q8wb : VLD1QWB<{0,0,?,?}, "8">;
394 defm VLD1q16wb : VLD1QWB<{0,1,?,?}, "16">;
395 defm VLD1q32wb : VLD1QWB<{1,0,?,?}, "32">;
396 defm VLD1q64wb : VLD1QWB<{1,1,?,?}, "64">;
398 def VLD1q8PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD1x2u>;
399 def VLD1q16PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD1x2u>;
400 def VLD1q32PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD1x2u>;
401 def VLD1q64PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD1x2u>;
402 def VLD1q8PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD1x2u>;
403 def VLD1q16PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD1x2u>;
404 def VLD1q32PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD1x2u>;
405 def VLD1q64PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD1x2u>;
407 // ...with 3 registers
408 class VLD1D3<bits<4> op7_4, string Dt>
409 : NLdSt<0,0b10,0b0110,op7_4, (outs VecListThreeD:$Vd),
410 (ins addrmode6:$Rn), IIC_VLD1x3, "vld1", Dt,
411 "$Vd, $Rn", "", []> {
414 let DecoderMethod = "DecodeVLDInstruction";
416 multiclass VLD1D3WB<bits<4> op7_4, string Dt> {
417 def _fixed : NLdSt<0,0b10,0b0110, op7_4, (outs VecListThreeD:$Vd, GPR:$wb),
418 (ins addrmode6:$Rn), IIC_VLD1x2u,
419 "vld1", Dt, "$Vd, $Rn!",
420 "$Rn.addr = $wb", []> {
421 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
423 let DecoderMethod = "DecodeVLDInstruction";
424 let AsmMatchConverter = "cvtVLDwbFixed";
426 def _register : NLdSt<0,0b10,0b0110,op7_4, (outs VecListThreeD:$Vd, GPR:$wb),
427 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1x2u,
428 "vld1", Dt, "$Vd, $Rn, $Rm",
429 "$Rn.addr = $wb", []> {
431 let DecoderMethod = "DecodeVLDInstruction";
432 let AsmMatchConverter = "cvtVLDwbRegister";
436 def VLD1d8T : VLD1D3<{0,0,0,?}, "8">;
437 def VLD1d16T : VLD1D3<{0,1,0,?}, "16">;
438 def VLD1d32T : VLD1D3<{1,0,0,?}, "32">;
439 def VLD1d64T : VLD1D3<{1,1,0,?}, "64">;
441 defm VLD1d8Twb : VLD1D3WB<{0,0,0,?}, "8">;
442 defm VLD1d16Twb : VLD1D3WB<{0,1,0,?}, "16">;
443 defm VLD1d32Twb : VLD1D3WB<{1,0,0,?}, "32">;
444 defm VLD1d64Twb : VLD1D3WB<{1,1,0,?}, "64">;
446 def VLD1d64TPseudo : VLDQQPseudo<IIC_VLD1x3>;
448 // ...with 4 registers
449 class VLD1D4<bits<4> op7_4, string Dt>
450 : NLdSt<0, 0b10, 0b0010, op7_4, (outs VecListFourD:$Vd),
451 (ins addrmode6:$Rn), IIC_VLD1x4, "vld1", Dt,
452 "$Vd, $Rn", "", []> {
454 let Inst{5-4} = Rn{5-4};
455 let DecoderMethod = "DecodeVLDInstruction";
457 multiclass VLD1D4WB<bits<4> op7_4, string Dt> {
458 def _fixed : NLdSt<0,0b10,0b0010, op7_4, (outs VecListFourD:$Vd, GPR:$wb),
459 (ins addrmode6:$Rn), IIC_VLD1x2u,
460 "vld1", Dt, "$Vd, $Rn!",
461 "$Rn.addr = $wb", []> {
462 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
463 let Inst{5-4} = Rn{5-4};
464 let DecoderMethod = "DecodeVLDInstruction";
465 let AsmMatchConverter = "cvtVLDwbFixed";
467 def _register : NLdSt<0,0b10,0b0010,op7_4, (outs VecListFourD:$Vd, GPR:$wb),
468 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1x2u,
469 "vld1", Dt, "$Vd, $Rn, $Rm",
470 "$Rn.addr = $wb", []> {
471 let Inst{5-4} = Rn{5-4};
472 let DecoderMethod = "DecodeVLDInstruction";
473 let AsmMatchConverter = "cvtVLDwbRegister";
477 def VLD1d8Q : VLD1D4<{0,0,?,?}, "8">;
478 def VLD1d16Q : VLD1D4<{0,1,?,?}, "16">;
479 def VLD1d32Q : VLD1D4<{1,0,?,?}, "32">;
480 def VLD1d64Q : VLD1D4<{1,1,?,?}, "64">;
482 defm VLD1d8Qwb : VLD1D4WB<{0,0,?,?}, "8">;
483 defm VLD1d16Qwb : VLD1D4WB<{0,1,?,?}, "16">;
484 defm VLD1d32Qwb : VLD1D4WB<{1,0,?,?}, "32">;
485 defm VLD1d64Qwb : VLD1D4WB<{1,1,?,?}, "64">;
487 def VLD1d64QPseudo : VLDQQPseudo<IIC_VLD1x4>;
489 // VLD2 : Vector Load (multiple 2-element structures)
490 class VLD2D<bits<4> op11_8, bits<4> op7_4, string Dt, RegisterOperand VdTy>
491 : NLdSt<0, 0b10, op11_8, op7_4, (outs VdTy:$Vd),
492 (ins addrmode6:$Rn), IIC_VLD2,
493 "vld2", Dt, "$Vd, $Rn", "", []> {
495 let Inst{5-4} = Rn{5-4};
496 let DecoderMethod = "DecodeVLDInstruction";
498 class VLD2Q<bits<4> op7_4, string Dt, RegisterOperand VdTy>
499 : NLdSt<0, 0b10, 0b0011, op7_4,
501 (ins addrmode6:$Rn), IIC_VLD2x2,
502 "vld2", Dt, "$Vd, $Rn", "", []> {
504 let Inst{5-4} = Rn{5-4};
505 let DecoderMethod = "DecodeVLDInstruction";
508 def VLD2d8 : VLD2D<0b1000, {0,0,?,?}, "8", VecListTwoD>;
509 def VLD2d16 : VLD2D<0b1000, {0,1,?,?}, "16", VecListTwoD>;
510 def VLD2d32 : VLD2D<0b1000, {1,0,?,?}, "32", VecListTwoD>;
512 def VLD2q8 : VLD2Q<{0,0,?,?}, "8", VecListFourD>;
513 def VLD2q16 : VLD2Q<{0,1,?,?}, "16", VecListFourD>;
514 def VLD2q32 : VLD2Q<{1,0,?,?}, "32", VecListFourD>;
516 def VLD2d8Pseudo : VLDQPseudo<IIC_VLD2>;
517 def VLD2d16Pseudo : VLDQPseudo<IIC_VLD2>;
518 def VLD2d32Pseudo : VLDQPseudo<IIC_VLD2>;
520 def VLD2q8Pseudo : VLDQQPseudo<IIC_VLD2x2>;
521 def VLD2q16Pseudo : VLDQQPseudo<IIC_VLD2x2>;
522 def VLD2q32Pseudo : VLDQQPseudo<IIC_VLD2x2>;
524 // ...with address register writeback:
525 class VLD2DWB<bits<4> op11_8, bits<4> op7_4, string Dt, RegisterOperand VdTy>
526 : NLdSt<0, 0b10, op11_8, op7_4, (outs VdTy:$Vd, GPR:$wb),
527 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD2u,
528 "vld2", Dt, "$Vd, $Rn$Rm",
529 "$Rn.addr = $wb", []> {
530 let Inst{5-4} = Rn{5-4};
531 let DecoderMethod = "DecodeVLDInstruction";
533 class VLD2QWB<bits<4> op7_4, string Dt, RegisterOperand VdTy>
534 : NLdSt<0, 0b10, 0b0011, op7_4,
535 (outs VdTy:$Vd, GPR:$wb),
536 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD2x2u,
537 "vld2", Dt, "$Vd, $Rn$Rm",
538 "$Rn.addr = $wb", []> {
539 let Inst{5-4} = Rn{5-4};
540 let DecoderMethod = "DecodeVLDInstruction";
543 def VLD2d8_UPD : VLD2DWB<0b1000, {0,0,?,?}, "8", VecListTwoD>;
544 def VLD2d16_UPD : VLD2DWB<0b1000, {0,1,?,?}, "16", VecListTwoD>;
545 def VLD2d32_UPD : VLD2DWB<0b1000, {1,0,?,?}, "32", VecListTwoD>;
547 def VLD2q8_UPD : VLD2QWB<{0,0,?,?}, "8", VecListFourD>;
548 def VLD2q16_UPD : VLD2QWB<{0,1,?,?}, "16", VecListFourD>;
549 def VLD2q32_UPD : VLD2QWB<{1,0,?,?}, "32", VecListFourD>;
551 def VLD2d8Pseudo_UPD : VLDQWBPseudo<IIC_VLD2u>;
552 def VLD2d16Pseudo_UPD : VLDQWBPseudo<IIC_VLD2u>;
553 def VLD2d32Pseudo_UPD : VLDQWBPseudo<IIC_VLD2u>;
555 def VLD2q8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD2x2u>;
556 def VLD2q16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD2x2u>;
557 def VLD2q32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD2x2u>;
559 // ...with double-spaced registers
560 def VLD2b8 : VLD2D<0b1001, {0,0,?,?}, "8", VecListTwoQ>;
561 def VLD2b16 : VLD2D<0b1001, {0,1,?,?}, "16", VecListTwoQ>;
562 def VLD2b32 : VLD2D<0b1001, {1,0,?,?}, "32", VecListTwoQ>;
563 def VLD2b8_UPD : VLD2DWB<0b1001, {0,0,?,?}, "8", VecListTwoQ>;
564 def VLD2b16_UPD : VLD2DWB<0b1001, {0,1,?,?}, "16", VecListTwoQ>;
565 def VLD2b32_UPD : VLD2DWB<0b1001, {1,0,?,?}, "32", VecListTwoQ>;
567 // VLD3 : Vector Load (multiple 3-element structures)
568 class VLD3D<bits<4> op11_8, bits<4> op7_4, string Dt>
569 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
570 (ins addrmode6:$Rn), IIC_VLD3,
571 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn", "", []> {
574 let DecoderMethod = "DecodeVLDInstruction";
577 def VLD3d8 : VLD3D<0b0100, {0,0,0,?}, "8">;
578 def VLD3d16 : VLD3D<0b0100, {0,1,0,?}, "16">;
579 def VLD3d32 : VLD3D<0b0100, {1,0,0,?}, "32">;
581 def VLD3d8Pseudo : VLDQQPseudo<IIC_VLD3>;
582 def VLD3d16Pseudo : VLDQQPseudo<IIC_VLD3>;
583 def VLD3d32Pseudo : VLDQQPseudo<IIC_VLD3>;
585 // ...with address register writeback:
586 class VLD3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
587 : NLdSt<0, 0b10, op11_8, op7_4,
588 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
589 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD3u,
590 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn$Rm",
591 "$Rn.addr = $wb", []> {
593 let DecoderMethod = "DecodeVLDInstruction";
596 def VLD3d8_UPD : VLD3DWB<0b0100, {0,0,0,?}, "8">;
597 def VLD3d16_UPD : VLD3DWB<0b0100, {0,1,0,?}, "16">;
598 def VLD3d32_UPD : VLD3DWB<0b0100, {1,0,0,?}, "32">;
600 def VLD3d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
601 def VLD3d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
602 def VLD3d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
604 // ...with double-spaced registers:
605 def VLD3q8 : VLD3D<0b0101, {0,0,0,?}, "8">;
606 def VLD3q16 : VLD3D<0b0101, {0,1,0,?}, "16">;
607 def VLD3q32 : VLD3D<0b0101, {1,0,0,?}, "32">;
608 def VLD3q8_UPD : VLD3DWB<0b0101, {0,0,0,?}, "8">;
609 def VLD3q16_UPD : VLD3DWB<0b0101, {0,1,0,?}, "16">;
610 def VLD3q32_UPD : VLD3DWB<0b0101, {1,0,0,?}, "32">;
612 def VLD3q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
613 def VLD3q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
614 def VLD3q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
616 // ...alternate versions to be allocated odd register numbers:
617 def VLD3q8oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
618 def VLD3q16oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
619 def VLD3q32oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
621 def VLD3q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
622 def VLD3q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
623 def VLD3q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
625 // VLD4 : Vector Load (multiple 4-element structures)
626 class VLD4D<bits<4> op11_8, bits<4> op7_4, string Dt>
627 : NLdSt<0, 0b10, op11_8, op7_4,
628 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
629 (ins addrmode6:$Rn), IIC_VLD4,
630 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn", "", []> {
632 let Inst{5-4} = Rn{5-4};
633 let DecoderMethod = "DecodeVLDInstruction";
636 def VLD4d8 : VLD4D<0b0000, {0,0,?,?}, "8">;
637 def VLD4d16 : VLD4D<0b0000, {0,1,?,?}, "16">;
638 def VLD4d32 : VLD4D<0b0000, {1,0,?,?}, "32">;
640 def VLD4d8Pseudo : VLDQQPseudo<IIC_VLD4>;
641 def VLD4d16Pseudo : VLDQQPseudo<IIC_VLD4>;
642 def VLD4d32Pseudo : VLDQQPseudo<IIC_VLD4>;
644 // ...with address register writeback:
645 class VLD4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
646 : NLdSt<0, 0b10, op11_8, op7_4,
647 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
648 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD4u,
649 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn$Rm",
650 "$Rn.addr = $wb", []> {
651 let Inst{5-4} = Rn{5-4};
652 let DecoderMethod = "DecodeVLDInstruction";
655 def VLD4d8_UPD : VLD4DWB<0b0000, {0,0,?,?}, "8">;
656 def VLD4d16_UPD : VLD4DWB<0b0000, {0,1,?,?}, "16">;
657 def VLD4d32_UPD : VLD4DWB<0b0000, {1,0,?,?}, "32">;
659 def VLD4d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
660 def VLD4d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
661 def VLD4d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
663 // ...with double-spaced registers:
664 def VLD4q8 : VLD4D<0b0001, {0,0,?,?}, "8">;
665 def VLD4q16 : VLD4D<0b0001, {0,1,?,?}, "16">;
666 def VLD4q32 : VLD4D<0b0001, {1,0,?,?}, "32">;
667 def VLD4q8_UPD : VLD4DWB<0b0001, {0,0,?,?}, "8">;
668 def VLD4q16_UPD : VLD4DWB<0b0001, {0,1,?,?}, "16">;
669 def VLD4q32_UPD : VLD4DWB<0b0001, {1,0,?,?}, "32">;
671 def VLD4q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
672 def VLD4q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
673 def VLD4q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
675 // ...alternate versions to be allocated odd register numbers:
676 def VLD4q8oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
677 def VLD4q16oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
678 def VLD4q32oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
680 def VLD4q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
681 def VLD4q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
682 def VLD4q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
684 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
686 // Classes for VLD*LN pseudo-instructions with multi-register operands.
687 // These are expanded to real instructions after register allocation.
688 class VLDQLNPseudo<InstrItinClass itin>
689 : PseudoNLdSt<(outs QPR:$dst),
690 (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
691 itin, "$src = $dst">;
692 class VLDQLNWBPseudo<InstrItinClass itin>
693 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
694 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
695 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
696 class VLDQQLNPseudo<InstrItinClass itin>
697 : PseudoNLdSt<(outs QQPR:$dst),
698 (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
699 itin, "$src = $dst">;
700 class VLDQQLNWBPseudo<InstrItinClass itin>
701 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
702 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
703 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
704 class VLDQQQQLNPseudo<InstrItinClass itin>
705 : PseudoNLdSt<(outs QQQQPR:$dst),
706 (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
707 itin, "$src = $dst">;
708 class VLDQQQQLNWBPseudo<InstrItinClass itin>
709 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
710 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
711 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
713 // VLD1LN : Vector Load (single element to one lane)
714 class VLD1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
716 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd),
717 (ins addrmode6:$Rn, DPR:$src, nohash_imm:$lane),
718 IIC_VLD1ln, "vld1", Dt, "\\{$Vd[$lane]\\}, $Rn",
720 [(set DPR:$Vd, (vector_insert (Ty DPR:$src),
721 (i32 (LoadOp addrmode6:$Rn)),
724 let DecoderMethod = "DecodeVLD1LN";
726 class VLD1LN32<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
728 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd),
729 (ins addrmode6oneL32:$Rn, DPR:$src, nohash_imm:$lane),
730 IIC_VLD1ln, "vld1", Dt, "\\{$Vd[$lane]\\}, $Rn",
732 [(set DPR:$Vd, (vector_insert (Ty DPR:$src),
733 (i32 (LoadOp addrmode6oneL32:$Rn)),
736 let DecoderMethod = "DecodeVLD1LN";
738 class VLD1QLNPseudo<ValueType Ty, PatFrag LoadOp> : VLDQLNPseudo<IIC_VLD1ln> {
739 let Pattern = [(set QPR:$dst, (vector_insert (Ty QPR:$src),
740 (i32 (LoadOp addrmode6:$addr)),
744 def VLD1LNd8 : VLD1LN<0b0000, {?,?,?,0}, "8", v8i8, extloadi8> {
745 let Inst{7-5} = lane{2-0};
747 def VLD1LNd16 : VLD1LN<0b0100, {?,?,0,?}, "16", v4i16, extloadi16> {
748 let Inst{7-6} = lane{1-0};
751 def VLD1LNd32 : VLD1LN32<0b1000, {?,0,?,?}, "32", v2i32, load> {
752 let Inst{7} = lane{0};
757 def VLD1LNq8Pseudo : VLD1QLNPseudo<v16i8, extloadi8>;
758 def VLD1LNq16Pseudo : VLD1QLNPseudo<v8i16, extloadi16>;
759 def VLD1LNq32Pseudo : VLD1QLNPseudo<v4i32, load>;
761 def : Pat<(vector_insert (v2f32 DPR:$src),
762 (f32 (load addrmode6:$addr)), imm:$lane),
763 (VLD1LNd32 addrmode6:$addr, DPR:$src, imm:$lane)>;
764 def : Pat<(vector_insert (v4f32 QPR:$src),
765 (f32 (load addrmode6:$addr)), imm:$lane),
766 (VLD1LNq32Pseudo addrmode6:$addr, QPR:$src, imm:$lane)>;
768 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
770 // ...with address register writeback:
771 class VLD1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
772 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, GPR:$wb),
773 (ins addrmode6:$Rn, am6offset:$Rm,
774 DPR:$src, nohash_imm:$lane), IIC_VLD1lnu, "vld1", Dt,
775 "\\{$Vd[$lane]\\}, $Rn$Rm",
776 "$src = $Vd, $Rn.addr = $wb", []> {
777 let DecoderMethod = "DecodeVLD1LN";
780 def VLD1LNd8_UPD : VLD1LNWB<0b0000, {?,?,?,0}, "8"> {
781 let Inst{7-5} = lane{2-0};
783 def VLD1LNd16_UPD : VLD1LNWB<0b0100, {?,?,0,?}, "16"> {
784 let Inst{7-6} = lane{1-0};
787 def VLD1LNd32_UPD : VLD1LNWB<0b1000, {?,0,?,?}, "32"> {
788 let Inst{7} = lane{0};
793 def VLD1LNq8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
794 def VLD1LNq16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
795 def VLD1LNq32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
797 // VLD2LN : Vector Load (single 2-element structure to one lane)
798 class VLD2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
799 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2),
800 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, nohash_imm:$lane),
801 IIC_VLD2ln, "vld2", Dt, "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn",
802 "$src1 = $Vd, $src2 = $dst2", []> {
805 let DecoderMethod = "DecodeVLD2LN";
808 def VLD2LNd8 : VLD2LN<0b0001, {?,?,?,?}, "8"> {
809 let Inst{7-5} = lane{2-0};
811 def VLD2LNd16 : VLD2LN<0b0101, {?,?,0,?}, "16"> {
812 let Inst{7-6} = lane{1-0};
814 def VLD2LNd32 : VLD2LN<0b1001, {?,0,0,?}, "32"> {
815 let Inst{7} = lane{0};
818 def VLD2LNd8Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
819 def VLD2LNd16Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
820 def VLD2LNd32Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
822 // ...with double-spaced registers:
823 def VLD2LNq16 : VLD2LN<0b0101, {?,?,1,?}, "16"> {
824 let Inst{7-6} = lane{1-0};
826 def VLD2LNq32 : VLD2LN<0b1001, {?,1,0,?}, "32"> {
827 let Inst{7} = lane{0};
830 def VLD2LNq16Pseudo : VLDQQLNPseudo<IIC_VLD2ln>;
831 def VLD2LNq32Pseudo : VLDQQLNPseudo<IIC_VLD2ln>;
833 // ...with address register writeback:
834 class VLD2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
835 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
836 (ins addrmode6:$Rn, am6offset:$Rm,
837 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VLD2lnu, "vld2", Dt,
838 "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn$Rm",
839 "$src1 = $Vd, $src2 = $dst2, $Rn.addr = $wb", []> {
841 let DecoderMethod = "DecodeVLD2LN";
844 def VLD2LNd8_UPD : VLD2LNWB<0b0001, {?,?,?,?}, "8"> {
845 let Inst{7-5} = lane{2-0};
847 def VLD2LNd16_UPD : VLD2LNWB<0b0101, {?,?,0,?}, "16"> {
848 let Inst{7-6} = lane{1-0};
850 def VLD2LNd32_UPD : VLD2LNWB<0b1001, {?,0,0,?}, "32"> {
851 let Inst{7} = lane{0};
854 def VLD2LNd8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
855 def VLD2LNd16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
856 def VLD2LNd32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
858 def VLD2LNq16_UPD : VLD2LNWB<0b0101, {?,?,1,?}, "16"> {
859 let Inst{7-6} = lane{1-0};
861 def VLD2LNq32_UPD : VLD2LNWB<0b1001, {?,1,0,?}, "32"> {
862 let Inst{7} = lane{0};
865 def VLD2LNq16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
866 def VLD2LNq32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
868 // VLD3LN : Vector Load (single 3-element structure to one lane)
869 class VLD3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
870 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
871 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3,
872 nohash_imm:$lane), IIC_VLD3ln, "vld3", Dt,
873 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn",
874 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3", []> {
876 let DecoderMethod = "DecodeVLD3LN";
879 def VLD3LNd8 : VLD3LN<0b0010, {?,?,?,0}, "8"> {
880 let Inst{7-5} = lane{2-0};
882 def VLD3LNd16 : VLD3LN<0b0110, {?,?,0,0}, "16"> {
883 let Inst{7-6} = lane{1-0};
885 def VLD3LNd32 : VLD3LN<0b1010, {?,0,0,0}, "32"> {
886 let Inst{7} = lane{0};
889 def VLD3LNd8Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
890 def VLD3LNd16Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
891 def VLD3LNd32Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
893 // ...with double-spaced registers:
894 def VLD3LNq16 : VLD3LN<0b0110, {?,?,1,0}, "16"> {
895 let Inst{7-6} = lane{1-0};
897 def VLD3LNq32 : VLD3LN<0b1010, {?,1,0,0}, "32"> {
898 let Inst{7} = lane{0};
901 def VLD3LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
902 def VLD3LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
904 // ...with address register writeback:
905 class VLD3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
906 : NLdStLn<1, 0b10, op11_8, op7_4,
907 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
908 (ins addrmode6:$Rn, am6offset:$Rm,
909 DPR:$src1, DPR:$src2, DPR:$src3, nohash_imm:$lane),
910 IIC_VLD3lnu, "vld3", Dt,
911 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn$Rm",
912 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $Rn.addr = $wb",
914 let DecoderMethod = "DecodeVLD3LN";
917 def VLD3LNd8_UPD : VLD3LNWB<0b0010, {?,?,?,0}, "8"> {
918 let Inst{7-5} = lane{2-0};
920 def VLD3LNd16_UPD : VLD3LNWB<0b0110, {?,?,0,0}, "16"> {
921 let Inst{7-6} = lane{1-0};
923 def VLD3LNd32_UPD : VLD3LNWB<0b1010, {?,0,0,0}, "32"> {
924 let Inst{7} = lane{0};
927 def VLD3LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
928 def VLD3LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
929 def VLD3LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
931 def VLD3LNq16_UPD : VLD3LNWB<0b0110, {?,?,1,0}, "16"> {
932 let Inst{7-6} = lane{1-0};
934 def VLD3LNq32_UPD : VLD3LNWB<0b1010, {?,1,0,0}, "32"> {
935 let Inst{7} = lane{0};
938 def VLD3LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
939 def VLD3LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
941 // VLD4LN : Vector Load (single 4-element structure to one lane)
942 class VLD4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
943 : NLdStLn<1, 0b10, op11_8, op7_4,
944 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
945 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
946 nohash_imm:$lane), IIC_VLD4ln, "vld4", Dt,
947 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn",
948 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4", []> {
951 let DecoderMethod = "DecodeVLD4LN";
954 def VLD4LNd8 : VLD4LN<0b0011, {?,?,?,?}, "8"> {
955 let Inst{7-5} = lane{2-0};
957 def VLD4LNd16 : VLD4LN<0b0111, {?,?,0,?}, "16"> {
958 let Inst{7-6} = lane{1-0};
960 def VLD4LNd32 : VLD4LN<0b1011, {?,0,?,?}, "32"> {
961 let Inst{7} = lane{0};
965 def VLD4LNd8Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
966 def VLD4LNd16Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
967 def VLD4LNd32Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
969 // ...with double-spaced registers:
970 def VLD4LNq16 : VLD4LN<0b0111, {?,?,1,?}, "16"> {
971 let Inst{7-6} = lane{1-0};
973 def VLD4LNq32 : VLD4LN<0b1011, {?,1,?,?}, "32"> {
974 let Inst{7} = lane{0};
978 def VLD4LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>;
979 def VLD4LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>;
981 // ...with address register writeback:
982 class VLD4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
983 : NLdStLn<1, 0b10, op11_8, op7_4,
984 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
985 (ins addrmode6:$Rn, am6offset:$Rm,
986 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
987 IIC_VLD4lnu, "vld4", Dt,
988 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn$Rm",
989 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4, $Rn.addr = $wb",
992 let DecoderMethod = "DecodeVLD4LN" ;
995 def VLD4LNd8_UPD : VLD4LNWB<0b0011, {?,?,?,?}, "8"> {
996 let Inst{7-5} = lane{2-0};
998 def VLD4LNd16_UPD : VLD4LNWB<0b0111, {?,?,0,?}, "16"> {
999 let Inst{7-6} = lane{1-0};
1001 def VLD4LNd32_UPD : VLD4LNWB<0b1011, {?,0,?,?}, "32"> {
1002 let Inst{7} = lane{0};
1003 let Inst{5} = Rn{5};
1006 def VLD4LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
1007 def VLD4LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
1008 def VLD4LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
1010 def VLD4LNq16_UPD : VLD4LNWB<0b0111, {?,?,1,?}, "16"> {
1011 let Inst{7-6} = lane{1-0};
1013 def VLD4LNq32_UPD : VLD4LNWB<0b1011, {?,1,?,?}, "32"> {
1014 let Inst{7} = lane{0};
1015 let Inst{5} = Rn{5};
1018 def VLD4LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
1019 def VLD4LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
1021 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
1023 // VLD1DUP : Vector Load (single element to all lanes)
1024 class VLD1DUP<bits<4> op7_4, string Dt, ValueType Ty, PatFrag LoadOp>
1025 : NLdSt<1, 0b10, 0b1100, op7_4, (outs VecListOneDAllLanes:$Vd),
1026 (ins addrmode6dup:$Rn),
1027 IIC_VLD1dup, "vld1", Dt, "$Vd, $Rn", "",
1028 [(set VecListOneDAllLanes:$Vd,
1029 (Ty (NEONvdup (i32 (LoadOp addrmode6dup:$Rn)))))]> {
1031 let Inst{4} = Rn{4};
1032 let DecoderMethod = "DecodeVLD1DupInstruction";
1034 class VLD1QDUPPseudo<ValueType Ty, PatFrag LoadOp> : VLDQPseudo<IIC_VLD1dup> {
1035 let Pattern = [(set QPR:$dst,
1036 (Ty (NEONvdup (i32 (LoadOp addrmode6dup:$addr)))))];
1039 def VLD1DUPd8 : VLD1DUP<{0,0,0,?}, "8", v8i8, extloadi8>;
1040 def VLD1DUPd16 : VLD1DUP<{0,1,0,?}, "16", v4i16, extloadi16>;
1041 def VLD1DUPd32 : VLD1DUP<{1,0,0,?}, "32", v2i32, load>;
1043 def VLD1DUPq8Pseudo : VLD1QDUPPseudo<v16i8, extloadi8>;
1044 def VLD1DUPq16Pseudo : VLD1QDUPPseudo<v8i16, extloadi16>;
1045 def VLD1DUPq32Pseudo : VLD1QDUPPseudo<v4i32, load>;
1047 def : Pat<(v2f32 (NEONvdup (f32 (load addrmode6dup:$addr)))),
1048 (VLD1DUPd32 addrmode6:$addr)>;
1049 def : Pat<(v4f32 (NEONvdup (f32 (load addrmode6dup:$addr)))),
1050 (VLD1DUPq32Pseudo addrmode6:$addr)>;
1052 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
1054 class VLD1QDUP<bits<4> op7_4, string Dt>
1055 : NLdSt<1, 0b10, 0b1100, op7_4, (outs VecListTwoDAllLanes:$Vd),
1056 (ins addrmode6dup:$Rn), IIC_VLD1dup,
1057 "vld1", Dt, "$Vd, $Rn", "", []> {
1059 let Inst{4} = Rn{4};
1060 let DecoderMethod = "DecodeVLD1DupInstruction";
1063 def VLD1DUPq8 : VLD1QDUP<{0,0,1,0}, "8">;
1064 def VLD1DUPq16 : VLD1QDUP<{0,1,1,?}, "16">;
1065 def VLD1DUPq32 : VLD1QDUP<{1,0,1,?}, "32">;
1067 // ...with address register writeback:
1068 class VLD1DUPWB<bits<4> op7_4, string Dt>
1069 : NLdSt<1, 0b10, 0b1100, op7_4, (outs DPR:$Vd, GPR:$wb),
1070 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD1dupu,
1071 "vld1", Dt, "\\{$Vd[]\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
1072 let Inst{4} = Rn{4};
1073 let DecoderMethod = "DecodeVLD1DupInstruction";
1075 class VLD1QDUPWB<bits<4> op7_4, string Dt>
1076 : NLdSt<1, 0b10, 0b1100, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
1077 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD1dupu,
1078 "vld1", Dt, "\\{$Vd[], $dst2[]\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
1079 let Inst{4} = Rn{4};
1080 let DecoderMethod = "DecodeVLD1DupInstruction";
1083 def VLD1DUPd8_UPD : VLD1DUPWB<{0,0,0,0}, "8">;
1084 def VLD1DUPd16_UPD : VLD1DUPWB<{0,1,0,?}, "16">;
1085 def VLD1DUPd32_UPD : VLD1DUPWB<{1,0,0,?}, "32">;
1087 def VLD1DUPq8_UPD : VLD1QDUPWB<{0,0,1,0}, "8">;
1088 def VLD1DUPq16_UPD : VLD1QDUPWB<{0,1,1,?}, "16">;
1089 def VLD1DUPq32_UPD : VLD1QDUPWB<{1,0,1,?}, "32">;
1091 def VLD1DUPq8Pseudo_UPD : VLDQWBPseudo<IIC_VLD1dupu>;
1092 def VLD1DUPq16Pseudo_UPD : VLDQWBPseudo<IIC_VLD1dupu>;
1093 def VLD1DUPq32Pseudo_UPD : VLDQWBPseudo<IIC_VLD1dupu>;
1095 // VLD2DUP : Vector Load (single 2-element structure to all lanes)
1096 class VLD2DUP<bits<4> op7_4, string Dt>
1097 : NLdSt<1, 0b10, 0b1101, op7_4, (outs DPR:$Vd, DPR:$dst2),
1098 (ins addrmode6dup:$Rn), IIC_VLD2dup,
1099 "vld2", Dt, "\\{$Vd[], $dst2[]\\}, $Rn", "", []> {
1101 let Inst{4} = Rn{4};
1102 let DecoderMethod = "DecodeVLD2DupInstruction";
1105 def VLD2DUPd8 : VLD2DUP<{0,0,0,?}, "8">;
1106 def VLD2DUPd16 : VLD2DUP<{0,1,0,?}, "16">;
1107 def VLD2DUPd32 : VLD2DUP<{1,0,0,?}, "32">;
1109 def VLD2DUPd8Pseudo : VLDQPseudo<IIC_VLD2dup>;
1110 def VLD2DUPd16Pseudo : VLDQPseudo<IIC_VLD2dup>;
1111 def VLD2DUPd32Pseudo : VLDQPseudo<IIC_VLD2dup>;
1113 // ...with double-spaced registers (not used for codegen):
1114 def VLD2DUPd8x2 : VLD2DUP<{0,0,1,?}, "8">;
1115 def VLD2DUPd16x2 : VLD2DUP<{0,1,1,?}, "16">;
1116 def VLD2DUPd32x2 : VLD2DUP<{1,0,1,?}, "32">;
1118 // ...with address register writeback:
1119 class VLD2DUPWB<bits<4> op7_4, string Dt>
1120 : NLdSt<1, 0b10, 0b1101, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
1121 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD2dupu,
1122 "vld2", Dt, "\\{$Vd[], $dst2[]\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
1123 let Inst{4} = Rn{4};
1124 let DecoderMethod = "DecodeVLD2DupInstruction";
1127 def VLD2DUPd8_UPD : VLD2DUPWB<{0,0,0,0}, "8">;
1128 def VLD2DUPd16_UPD : VLD2DUPWB<{0,1,0,?}, "16">;
1129 def VLD2DUPd32_UPD : VLD2DUPWB<{1,0,0,?}, "32">;
1131 def VLD2DUPd8x2_UPD : VLD2DUPWB<{0,0,1,0}, "8">;
1132 def VLD2DUPd16x2_UPD : VLD2DUPWB<{0,1,1,?}, "16">;
1133 def VLD2DUPd32x2_UPD : VLD2DUPWB<{1,0,1,?}, "32">;
1135 def VLD2DUPd8Pseudo_UPD : VLDQWBPseudo<IIC_VLD2dupu>;
1136 def VLD2DUPd16Pseudo_UPD : VLDQWBPseudo<IIC_VLD2dupu>;
1137 def VLD2DUPd32Pseudo_UPD : VLDQWBPseudo<IIC_VLD2dupu>;
1139 // VLD3DUP : Vector Load (single 3-element structure to all lanes)
1140 class VLD3DUP<bits<4> op7_4, string Dt>
1141 : NLdSt<1, 0b10, 0b1110, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
1142 (ins addrmode6dup:$Rn), IIC_VLD3dup,
1143 "vld3", Dt, "\\{$Vd[], $dst2[], $dst3[]\\}, $Rn", "", []> {
1146 let DecoderMethod = "DecodeVLD3DupInstruction";
1149 def VLD3DUPd8 : VLD3DUP<{0,0,0,?}, "8">;
1150 def VLD3DUPd16 : VLD3DUP<{0,1,0,?}, "16">;
1151 def VLD3DUPd32 : VLD3DUP<{1,0,0,?}, "32">;
1153 def VLD3DUPd8Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1154 def VLD3DUPd16Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1155 def VLD3DUPd32Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1157 // ...with double-spaced registers (not used for codegen):
1158 def VLD3DUPd8x2 : VLD3DUP<{0,0,1,?}, "8">;
1159 def VLD3DUPd16x2 : VLD3DUP<{0,1,1,?}, "16">;
1160 def VLD3DUPd32x2 : VLD3DUP<{1,0,1,?}, "32">;
1162 // ...with address register writeback:
1163 class VLD3DUPWB<bits<4> op7_4, string Dt>
1164 : NLdSt<1, 0b10, 0b1110, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
1165 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD3dupu,
1166 "vld3", Dt, "\\{$Vd[], $dst2[], $dst3[]\\}, $Rn$Rm",
1167 "$Rn.addr = $wb", []> {
1169 let DecoderMethod = "DecodeVLD3DupInstruction";
1172 def VLD3DUPd8_UPD : VLD3DUPWB<{0,0,0,0}, "8">;
1173 def VLD3DUPd16_UPD : VLD3DUPWB<{0,1,0,?}, "16">;
1174 def VLD3DUPd32_UPD : VLD3DUPWB<{1,0,0,?}, "32">;
1176 def VLD3DUPd8x2_UPD : VLD3DUPWB<{0,0,1,0}, "8">;
1177 def VLD3DUPd16x2_UPD : VLD3DUPWB<{0,1,1,?}, "16">;
1178 def VLD3DUPd32x2_UPD : VLD3DUPWB<{1,0,1,?}, "32">;
1180 def VLD3DUPd8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1181 def VLD3DUPd16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1182 def VLD3DUPd32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1184 // VLD4DUP : Vector Load (single 4-element structure to all lanes)
1185 class VLD4DUP<bits<4> op7_4, string Dt>
1186 : NLdSt<1, 0b10, 0b1111, op7_4,
1187 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
1188 (ins addrmode6dup:$Rn), IIC_VLD4dup,
1189 "vld4", Dt, "\\{$Vd[], $dst2[], $dst3[], $dst4[]\\}, $Rn", "", []> {
1191 let Inst{4} = Rn{4};
1192 let DecoderMethod = "DecodeVLD4DupInstruction";
1195 def VLD4DUPd8 : VLD4DUP<{0,0,0,?}, "8">;
1196 def VLD4DUPd16 : VLD4DUP<{0,1,0,?}, "16">;
1197 def VLD4DUPd32 : VLD4DUP<{1,?,0,?}, "32"> { let Inst{6} = Rn{5}; }
1199 def VLD4DUPd8Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1200 def VLD4DUPd16Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1201 def VLD4DUPd32Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1203 // ...with double-spaced registers (not used for codegen):
1204 def VLD4DUPd8x2 : VLD4DUP<{0,0,1,?}, "8">;
1205 def VLD4DUPd16x2 : VLD4DUP<{0,1,1,?}, "16">;
1206 def VLD4DUPd32x2 : VLD4DUP<{1,?,1,?}, "32"> { let Inst{6} = Rn{5}; }
1208 // ...with address register writeback:
1209 class VLD4DUPWB<bits<4> op7_4, string Dt>
1210 : NLdSt<1, 0b10, 0b1111, op7_4,
1211 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
1212 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD4dupu,
1213 "vld4", Dt, "\\{$Vd[], $dst2[], $dst3[], $dst4[]\\}, $Rn$Rm",
1214 "$Rn.addr = $wb", []> {
1215 let Inst{4} = Rn{4};
1216 let DecoderMethod = "DecodeVLD4DupInstruction";
1219 def VLD4DUPd8_UPD : VLD4DUPWB<{0,0,0,0}, "8">;
1220 def VLD4DUPd16_UPD : VLD4DUPWB<{0,1,0,?}, "16">;
1221 def VLD4DUPd32_UPD : VLD4DUPWB<{1,?,0,?}, "32"> { let Inst{6} = Rn{5}; }
1223 def VLD4DUPd8x2_UPD : VLD4DUPWB<{0,0,1,0}, "8">;
1224 def VLD4DUPd16x2_UPD : VLD4DUPWB<{0,1,1,?}, "16">;
1225 def VLD4DUPd32x2_UPD : VLD4DUPWB<{1,?,1,?}, "32"> { let Inst{6} = Rn{5}; }
1227 def VLD4DUPd8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1228 def VLD4DUPd16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1229 def VLD4DUPd32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1231 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
1233 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
1235 // Classes for VST* pseudo-instructions with multi-register operands.
1236 // These are expanded to real instructions after register allocation.
1237 class VSTQPseudo<InstrItinClass itin>
1238 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src), itin, "">;
1239 class VSTQWBPseudo<InstrItinClass itin>
1240 : PseudoNLdSt<(outs GPR:$wb),
1241 (ins addrmode6:$addr, am6offset:$offset, QPR:$src), itin,
1242 "$addr.addr = $wb">;
1243 class VSTQWBfixedPseudo<InstrItinClass itin>
1244 : PseudoNLdSt<(outs GPR:$wb),
1245 (ins addrmode6:$addr, QPR:$src), itin,
1246 "$addr.addr = $wb">;
1247 class VSTQWBregisterPseudo<InstrItinClass itin>
1248 : PseudoNLdSt<(outs GPR:$wb),
1249 (ins addrmode6:$addr, rGPR:$offset, QPR:$src), itin,
1250 "$addr.addr = $wb">;
1251 class VSTQQPseudo<InstrItinClass itin>
1252 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src), itin, "">;
1253 class VSTQQWBPseudo<InstrItinClass itin>
1254 : PseudoNLdSt<(outs GPR:$wb),
1255 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src), itin,
1256 "$addr.addr = $wb">;
1257 class VSTQQQQPseudo<InstrItinClass itin>
1258 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src), itin, "">;
1259 class VSTQQQQWBPseudo<InstrItinClass itin>
1260 : PseudoNLdSt<(outs GPR:$wb),
1261 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
1262 "$addr.addr = $wb">;
1264 // VST1 : Vector Store (multiple single elements)
1265 class VST1D<bits<4> op7_4, string Dt>
1266 : NLdSt<0,0b00,0b0111,op7_4, (outs), (ins addrmode6:$Rn, VecListOneD:$Vd),
1267 IIC_VST1, "vst1", Dt, "$Vd, $Rn", "", []> {
1269 let Inst{4} = Rn{4};
1270 let DecoderMethod = "DecodeVSTInstruction";
1272 class VST1Q<bits<4> op7_4, string Dt>
1273 : NLdSt<0,0b00,0b1010,op7_4, (outs), (ins addrmode6:$Rn, VecListTwoD:$Vd),
1274 IIC_VST1x2, "vst1", Dt, "$Vd, $Rn", "", []> {
1276 let Inst{5-4} = Rn{5-4};
1277 let DecoderMethod = "DecodeVSTInstruction";
1280 def VST1d8 : VST1D<{0,0,0,?}, "8">;
1281 def VST1d16 : VST1D<{0,1,0,?}, "16">;
1282 def VST1d32 : VST1D<{1,0,0,?}, "32">;
1283 def VST1d64 : VST1D<{1,1,0,?}, "64">;
1285 def VST1q8 : VST1Q<{0,0,?,?}, "8">;
1286 def VST1q16 : VST1Q<{0,1,?,?}, "16">;
1287 def VST1q32 : VST1Q<{1,0,?,?}, "32">;
1288 def VST1q64 : VST1Q<{1,1,?,?}, "64">;
1290 def VST1q8Pseudo : VSTQPseudo<IIC_VST1x2>;
1291 def VST1q16Pseudo : VSTQPseudo<IIC_VST1x2>;
1292 def VST1q32Pseudo : VSTQPseudo<IIC_VST1x2>;
1293 def VST1q64Pseudo : VSTQPseudo<IIC_VST1x2>;
1295 // ...with address register writeback:
1296 multiclass VST1DWB<bits<4> op7_4, string Dt> {
1297 def _fixed : NLdSt<0,0b00, 0b0111,op7_4, (outs GPR:$wb),
1298 (ins addrmode6:$Rn, VecListOneD:$Vd), IIC_VLD1u,
1299 "vst1", Dt, "$Vd, $Rn!",
1300 "$Rn.addr = $wb", []> {
1301 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1302 let Inst{4} = Rn{4};
1303 let DecoderMethod = "DecodeVSTInstruction";
1304 let AsmMatchConverter = "cvtVSTwbFixed";
1306 def _register : NLdSt<0,0b00,0b0111,op7_4, (outs GPR:$wb),
1307 (ins addrmode6:$Rn, rGPR:$Rm, VecListOneD:$Vd),
1309 "vst1", Dt, "$Vd, $Rn, $Rm",
1310 "$Rn.addr = $wb", []> {
1311 let Inst{4} = Rn{4};
1312 let DecoderMethod = "DecodeVSTInstruction";
1313 let AsmMatchConverter = "cvtVSTwbRegister";
1316 multiclass VST1QWB<bits<4> op7_4, string Dt> {
1317 def _fixed : NLdSt<0,0b00,0b1010,op7_4, (outs GPR:$wb),
1318 (ins addrmode6:$Rn, VecListTwoD:$Vd), IIC_VLD1x2u,
1319 "vst1", Dt, "$Vd, $Rn!",
1320 "$Rn.addr = $wb", []> {
1321 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1322 let Inst{5-4} = Rn{5-4};
1323 let DecoderMethod = "DecodeVSTInstruction";
1324 let AsmMatchConverter = "cvtVSTwbFixed";
1326 def _register : NLdSt<0,0b00,0b1010,op7_4, (outs GPR:$wb),
1327 (ins addrmode6:$Rn, rGPR:$Rm, VecListTwoD:$Vd),
1329 "vst1", Dt, "$Vd, $Rn, $Rm",
1330 "$Rn.addr = $wb", []> {
1331 let Inst{5-4} = Rn{5-4};
1332 let DecoderMethod = "DecodeVSTInstruction";
1333 let AsmMatchConverter = "cvtVSTwbRegister";
1337 defm VST1d8wb : VST1DWB<{0,0,0,?}, "8">;
1338 defm VST1d16wb : VST1DWB<{0,1,0,?}, "16">;
1339 defm VST1d32wb : VST1DWB<{1,0,0,?}, "32">;
1340 defm VST1d64wb : VST1DWB<{1,1,0,?}, "64">;
1342 defm VST1q8wb : VST1QWB<{0,0,?,?}, "8">;
1343 defm VST1q16wb : VST1QWB<{0,1,?,?}, "16">;
1344 defm VST1q32wb : VST1QWB<{1,0,?,?}, "32">;
1345 defm VST1q64wb : VST1QWB<{1,1,?,?}, "64">;
1347 def VST1q8PseudoWB_fixed : VSTQWBfixedPseudo<IIC_VST1x2u>;
1348 def VST1q16PseudoWB_fixed : VSTQWBfixedPseudo<IIC_VST1x2u>;
1349 def VST1q32PseudoWB_fixed : VSTQWBfixedPseudo<IIC_VST1x2u>;
1350 def VST1q64PseudoWB_fixed : VSTQWBfixedPseudo<IIC_VST1x2u>;
1351 def VST1q8PseudoWB_register : VSTQWBregisterPseudo<IIC_VST1x2u>;
1352 def VST1q16PseudoWB_register : VSTQWBregisterPseudo<IIC_VST1x2u>;
1353 def VST1q32PseudoWB_register : VSTQWBregisterPseudo<IIC_VST1x2u>;
1354 def VST1q64PseudoWB_register : VSTQWBregisterPseudo<IIC_VST1x2u>;
1356 // ...with 3 registers
1357 class VST1D3<bits<4> op7_4, string Dt>
1358 : NLdSt<0, 0b00, 0b0110, op7_4, (outs),
1359 (ins addrmode6:$Rn, VecListThreeD:$Vd),
1360 IIC_VST1x3, "vst1", Dt, "$Vd, $Rn", "", []> {
1362 let Inst{4} = Rn{4};
1363 let DecoderMethod = "DecodeVSTInstruction";
1365 multiclass VST1D3WB<bits<4> op7_4, string Dt> {
1366 def _fixed : NLdSt<0,0b00,0b0110,op7_4, (outs GPR:$wb),
1367 (ins addrmode6:$Rn, VecListThreeD:$Vd), IIC_VLD1x3u,
1368 "vst1", Dt, "$Vd, $Rn!",
1369 "$Rn.addr = $wb", []> {
1370 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1371 let Inst{5-4} = Rn{5-4};
1372 let DecoderMethod = "DecodeVSTInstruction";
1373 let AsmMatchConverter = "cvtVSTwbFixed";
1375 def _register : NLdSt<0,0b00,0b0110,op7_4, (outs GPR:$wb),
1376 (ins addrmode6:$Rn, rGPR:$Rm, VecListThreeD:$Vd),
1378 "vst1", Dt, "$Vd, $Rn, $Rm",
1379 "$Rn.addr = $wb", []> {
1380 let Inst{5-4} = Rn{5-4};
1381 let DecoderMethod = "DecodeVSTInstruction";
1382 let AsmMatchConverter = "cvtVSTwbRegister";
1386 def VST1d8T : VST1D3<{0,0,0,?}, "8">;
1387 def VST1d16T : VST1D3<{0,1,0,?}, "16">;
1388 def VST1d32T : VST1D3<{1,0,0,?}, "32">;
1389 def VST1d64T : VST1D3<{1,1,0,?}, "64">;
1391 defm VST1d8Twb : VST1D3WB<{0,0,0,?}, "8">;
1392 defm VST1d16Twb : VST1D3WB<{0,1,0,?}, "16">;
1393 defm VST1d32Twb : VST1D3WB<{1,0,0,?}, "32">;
1394 defm VST1d64Twb : VST1D3WB<{1,1,0,?}, "64">;
1396 def VST1d64TPseudo : VSTQQPseudo<IIC_VST1x3>;
1397 def VST1d64TPseudoWB_fixed : VSTQQWBPseudo<IIC_VST1x3u>;
1398 def VST1d64TPseudoWB_register : VSTQQWBPseudo<IIC_VST1x3u>;
1400 // ...with 4 registers
1401 class VST1D4<bits<4> op7_4, string Dt>
1402 : NLdSt<0, 0b00, 0b0010, op7_4, (outs),
1403 (ins addrmode6:$Rn, VecListFourD:$Vd),
1404 IIC_VST1x4, "vst1", Dt, "$Vd, $Rn", "",
1407 let Inst{5-4} = Rn{5-4};
1408 let DecoderMethod = "DecodeVSTInstruction";
1410 multiclass VST1D4WB<bits<4> op7_4, string Dt> {
1411 def _fixed : NLdSt<0,0b00,0b0010,op7_4, (outs GPR:$wb),
1412 (ins addrmode6:$Rn, VecListFourD:$Vd), IIC_VLD1x4u,
1413 "vst1", Dt, "$Vd, $Rn!",
1414 "$Rn.addr = $wb", []> {
1415 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1416 let Inst{5-4} = Rn{5-4};
1417 let DecoderMethod = "DecodeVSTInstruction";
1418 let AsmMatchConverter = "cvtVSTwbFixed";
1420 def _register : NLdSt<0,0b00,0b0010,op7_4, (outs GPR:$wb),
1421 (ins addrmode6:$Rn, rGPR:$Rm, VecListFourD:$Vd),
1423 "vst1", Dt, "$Vd, $Rn, $Rm",
1424 "$Rn.addr = $wb", []> {
1425 let Inst{5-4} = Rn{5-4};
1426 let DecoderMethod = "DecodeVSTInstruction";
1427 let AsmMatchConverter = "cvtVSTwbRegister";
1431 def VST1d8Q : VST1D4<{0,0,?,?}, "8">;
1432 def VST1d16Q : VST1D4<{0,1,?,?}, "16">;
1433 def VST1d32Q : VST1D4<{1,0,?,?}, "32">;
1434 def VST1d64Q : VST1D4<{1,1,?,?}, "64">;
1436 defm VST1d8Qwb : VST1D4WB<{0,0,?,?}, "8">;
1437 defm VST1d16Qwb : VST1D4WB<{0,1,?,?}, "16">;
1438 defm VST1d32Qwb : VST1D4WB<{1,0,?,?}, "32">;
1439 defm VST1d64Qwb : VST1D4WB<{1,1,?,?}, "64">;
1441 def VST1d64QPseudo : VSTQQPseudo<IIC_VST1x4>;
1442 def VST1d64QPseudoWB_fixed : VSTQQWBPseudo<IIC_VST1x4u>;
1443 def VST1d64QPseudoWB_register : VSTQQWBPseudo<IIC_VST1x4u>;
1445 // VST2 : Vector Store (multiple 2-element structures)
1446 class VST2D<bits<4> op11_8, bits<4> op7_4, string Dt>
1447 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1448 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2),
1449 IIC_VST2, "vst2", Dt, "\\{$Vd, $src2\\}, $Rn", "", []> {
1451 let Inst{5-4} = Rn{5-4};
1452 let DecoderMethod = "DecodeVSTInstruction";
1454 class VST2Q<bits<4> op7_4, string Dt>
1455 : NLdSt<0, 0b00, 0b0011, op7_4, (outs),
1456 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
1457 IIC_VST2x2, "vst2", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn",
1460 let Inst{5-4} = Rn{5-4};
1461 let DecoderMethod = "DecodeVSTInstruction";
1464 def VST2d8 : VST2D<0b1000, {0,0,?,?}, "8">;
1465 def VST2d16 : VST2D<0b1000, {0,1,?,?}, "16">;
1466 def VST2d32 : VST2D<0b1000, {1,0,?,?}, "32">;
1468 def VST2q8 : VST2Q<{0,0,?,?}, "8">;
1469 def VST2q16 : VST2Q<{0,1,?,?}, "16">;
1470 def VST2q32 : VST2Q<{1,0,?,?}, "32">;
1472 def VST2d8Pseudo : VSTQPseudo<IIC_VST2>;
1473 def VST2d16Pseudo : VSTQPseudo<IIC_VST2>;
1474 def VST2d32Pseudo : VSTQPseudo<IIC_VST2>;
1476 def VST2q8Pseudo : VSTQQPseudo<IIC_VST2x2>;
1477 def VST2q16Pseudo : VSTQQPseudo<IIC_VST2x2>;
1478 def VST2q32Pseudo : VSTQQPseudo<IIC_VST2x2>;
1480 // ...with address register writeback:
1481 class VST2DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1482 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1483 (ins addrmode6:$Rn, am6offset:$Rm, DPR:$Vd, DPR:$src2),
1484 IIC_VST2u, "vst2", Dt, "\\{$Vd, $src2\\}, $Rn$Rm",
1485 "$Rn.addr = $wb", []> {
1486 let Inst{5-4} = Rn{5-4};
1487 let DecoderMethod = "DecodeVSTInstruction";
1489 class VST2QWB<bits<4> op7_4, string Dt>
1490 : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
1491 (ins addrmode6:$Rn, am6offset:$Rm,
1492 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST2x2u,
1493 "vst2", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
1494 "$Rn.addr = $wb", []> {
1495 let Inst{5-4} = Rn{5-4};
1496 let DecoderMethod = "DecodeVSTInstruction";
1499 def VST2d8_UPD : VST2DWB<0b1000, {0,0,?,?}, "8">;
1500 def VST2d16_UPD : VST2DWB<0b1000, {0,1,?,?}, "16">;
1501 def VST2d32_UPD : VST2DWB<0b1000, {1,0,?,?}, "32">;
1503 def VST2q8_UPD : VST2QWB<{0,0,?,?}, "8">;
1504 def VST2q16_UPD : VST2QWB<{0,1,?,?}, "16">;
1505 def VST2q32_UPD : VST2QWB<{1,0,?,?}, "32">;
1507 def VST2d8Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
1508 def VST2d16Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
1509 def VST2d32Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
1511 def VST2q8Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
1512 def VST2q16Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
1513 def VST2q32Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
1515 // ...with double-spaced registers
1516 def VST2b8 : VST2D<0b1001, {0,0,?,?}, "8">;
1517 def VST2b16 : VST2D<0b1001, {0,1,?,?}, "16">;
1518 def VST2b32 : VST2D<0b1001, {1,0,?,?}, "32">;
1519 def VST2b8_UPD : VST2DWB<0b1001, {0,0,?,?}, "8">;
1520 def VST2b16_UPD : VST2DWB<0b1001, {0,1,?,?}, "16">;
1521 def VST2b32_UPD : VST2DWB<0b1001, {1,0,?,?}, "32">;
1523 // VST3 : Vector Store (multiple 3-element structures)
1524 class VST3D<bits<4> op11_8, bits<4> op7_4, string Dt>
1525 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1526 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3,
1527 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn", "", []> {
1529 let Inst{4} = Rn{4};
1530 let DecoderMethod = "DecodeVSTInstruction";
1533 def VST3d8 : VST3D<0b0100, {0,0,0,?}, "8">;
1534 def VST3d16 : VST3D<0b0100, {0,1,0,?}, "16">;
1535 def VST3d32 : VST3D<0b0100, {1,0,0,?}, "32">;
1537 def VST3d8Pseudo : VSTQQPseudo<IIC_VST3>;
1538 def VST3d16Pseudo : VSTQQPseudo<IIC_VST3>;
1539 def VST3d32Pseudo : VSTQQPseudo<IIC_VST3>;
1541 // ...with address register writeback:
1542 class VST3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1543 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1544 (ins addrmode6:$Rn, am6offset:$Rm,
1545 DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3u,
1546 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn$Rm",
1547 "$Rn.addr = $wb", []> {
1548 let Inst{4} = Rn{4};
1549 let DecoderMethod = "DecodeVSTInstruction";
1552 def VST3d8_UPD : VST3DWB<0b0100, {0,0,0,?}, "8">;
1553 def VST3d16_UPD : VST3DWB<0b0100, {0,1,0,?}, "16">;
1554 def VST3d32_UPD : VST3DWB<0b0100, {1,0,0,?}, "32">;
1556 def VST3d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1557 def VST3d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1558 def VST3d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1560 // ...with double-spaced registers:
1561 def VST3q8 : VST3D<0b0101, {0,0,0,?}, "8">;
1562 def VST3q16 : VST3D<0b0101, {0,1,0,?}, "16">;
1563 def VST3q32 : VST3D<0b0101, {1,0,0,?}, "32">;
1564 def VST3q8_UPD : VST3DWB<0b0101, {0,0,0,?}, "8">;
1565 def VST3q16_UPD : VST3DWB<0b0101, {0,1,0,?}, "16">;
1566 def VST3q32_UPD : VST3DWB<0b0101, {1,0,0,?}, "32">;
1568 def VST3q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1569 def VST3q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1570 def VST3q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1572 // ...alternate versions to be allocated odd register numbers:
1573 def VST3q8oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1574 def VST3q16oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1575 def VST3q32oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1577 def VST3q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1578 def VST3q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1579 def VST3q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1581 // VST4 : Vector Store (multiple 4-element structures)
1582 class VST4D<bits<4> op11_8, bits<4> op7_4, string Dt>
1583 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1584 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
1585 IIC_VST4, "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn",
1588 let Inst{5-4} = Rn{5-4};
1589 let DecoderMethod = "DecodeVSTInstruction";
1592 def VST4d8 : VST4D<0b0000, {0,0,?,?}, "8">;
1593 def VST4d16 : VST4D<0b0000, {0,1,?,?}, "16">;
1594 def VST4d32 : VST4D<0b0000, {1,0,?,?}, "32">;
1596 def VST4d8Pseudo : VSTQQPseudo<IIC_VST4>;
1597 def VST4d16Pseudo : VSTQQPseudo<IIC_VST4>;
1598 def VST4d32Pseudo : VSTQQPseudo<IIC_VST4>;
1600 // ...with address register writeback:
1601 class VST4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1602 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1603 (ins addrmode6:$Rn, am6offset:$Rm,
1604 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST4u,
1605 "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
1606 "$Rn.addr = $wb", []> {
1607 let Inst{5-4} = Rn{5-4};
1608 let DecoderMethod = "DecodeVSTInstruction";
1611 def VST4d8_UPD : VST4DWB<0b0000, {0,0,?,?}, "8">;
1612 def VST4d16_UPD : VST4DWB<0b0000, {0,1,?,?}, "16">;
1613 def VST4d32_UPD : VST4DWB<0b0000, {1,0,?,?}, "32">;
1615 def VST4d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1616 def VST4d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1617 def VST4d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1619 // ...with double-spaced registers:
1620 def VST4q8 : VST4D<0b0001, {0,0,?,?}, "8">;
1621 def VST4q16 : VST4D<0b0001, {0,1,?,?}, "16">;
1622 def VST4q32 : VST4D<0b0001, {1,0,?,?}, "32">;
1623 def VST4q8_UPD : VST4DWB<0b0001, {0,0,?,?}, "8">;
1624 def VST4q16_UPD : VST4DWB<0b0001, {0,1,?,?}, "16">;
1625 def VST4q32_UPD : VST4DWB<0b0001, {1,0,?,?}, "32">;
1627 def VST4q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1628 def VST4q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1629 def VST4q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1631 // ...alternate versions to be allocated odd register numbers:
1632 def VST4q8oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1633 def VST4q16oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1634 def VST4q32oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1636 def VST4q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1637 def VST4q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1638 def VST4q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1640 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
1642 // Classes for VST*LN pseudo-instructions with multi-register operands.
1643 // These are expanded to real instructions after register allocation.
1644 class VSTQLNPseudo<InstrItinClass itin>
1645 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
1647 class VSTQLNWBPseudo<InstrItinClass itin>
1648 : PseudoNLdSt<(outs GPR:$wb),
1649 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
1650 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1651 class VSTQQLNPseudo<InstrItinClass itin>
1652 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
1654 class VSTQQLNWBPseudo<InstrItinClass itin>
1655 : PseudoNLdSt<(outs GPR:$wb),
1656 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
1657 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1658 class VSTQQQQLNPseudo<InstrItinClass itin>
1659 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
1661 class VSTQQQQLNWBPseudo<InstrItinClass itin>
1662 : PseudoNLdSt<(outs GPR:$wb),
1663 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
1664 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1666 // VST1LN : Vector Store (single element from one lane)
1667 class VST1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1668 PatFrag StoreOp, SDNode ExtractOp>
1669 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1670 (ins addrmode6:$Rn, DPR:$Vd, nohash_imm:$lane),
1671 IIC_VST1ln, "vst1", Dt, "\\{$Vd[$lane]\\}, $Rn", "",
1672 [(StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane), addrmode6:$Rn)]> {
1674 let DecoderMethod = "DecodeVST1LN";
1676 class VST1LN32<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1677 PatFrag StoreOp, SDNode ExtractOp>
1678 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1679 (ins addrmode6oneL32:$Rn, DPR:$Vd, nohash_imm:$lane),
1680 IIC_VST1ln, "vst1", Dt, "\\{$Vd[$lane]\\}, $Rn", "",
1681 [(StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane), addrmode6oneL32:$Rn)]>{
1683 let DecoderMethod = "DecodeVST1LN";
1685 class VST1QLNPseudo<ValueType Ty, PatFrag StoreOp, SDNode ExtractOp>
1686 : VSTQLNPseudo<IIC_VST1ln> {
1687 let Pattern = [(StoreOp (ExtractOp (Ty QPR:$src), imm:$lane),
1691 def VST1LNd8 : VST1LN<0b0000, {?,?,?,0}, "8", v8i8, truncstorei8,
1693 let Inst{7-5} = lane{2-0};
1695 def VST1LNd16 : VST1LN<0b0100, {?,?,0,?}, "16", v4i16, truncstorei16,
1697 let Inst{7-6} = lane{1-0};
1698 let Inst{4} = Rn{5};
1701 def VST1LNd32 : VST1LN32<0b1000, {?,0,?,?}, "32", v2i32, store, extractelt> {
1702 let Inst{7} = lane{0};
1703 let Inst{5-4} = Rn{5-4};
1706 def VST1LNq8Pseudo : VST1QLNPseudo<v16i8, truncstorei8, NEONvgetlaneu>;
1707 def VST1LNq16Pseudo : VST1QLNPseudo<v8i16, truncstorei16, NEONvgetlaneu>;
1708 def VST1LNq32Pseudo : VST1QLNPseudo<v4i32, store, extractelt>;
1710 def : Pat<(store (extractelt (v2f32 DPR:$src), imm:$lane), addrmode6:$addr),
1711 (VST1LNd32 addrmode6:$addr, DPR:$src, imm:$lane)>;
1712 def : Pat<(store (extractelt (v4f32 QPR:$src), imm:$lane), addrmode6:$addr),
1713 (VST1LNq32Pseudo addrmode6:$addr, QPR:$src, imm:$lane)>;
1715 // ...with address register writeback:
1716 class VST1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1717 PatFrag StoreOp, SDNode ExtractOp>
1718 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1719 (ins addrmode6:$Rn, am6offset:$Rm,
1720 DPR:$Vd, nohash_imm:$lane), IIC_VST1lnu, "vst1", Dt,
1721 "\\{$Vd[$lane]\\}, $Rn$Rm",
1723 [(set GPR:$wb, (StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane),
1724 addrmode6:$Rn, am6offset:$Rm))]> {
1725 let DecoderMethod = "DecodeVST1LN";
1727 class VST1QLNWBPseudo<ValueType Ty, PatFrag StoreOp, SDNode ExtractOp>
1728 : VSTQLNWBPseudo<IIC_VST1lnu> {
1729 let Pattern = [(set GPR:$wb, (StoreOp (ExtractOp (Ty QPR:$src), imm:$lane),
1730 addrmode6:$addr, am6offset:$offset))];
1733 def VST1LNd8_UPD : VST1LNWB<0b0000, {?,?,?,0}, "8", v8i8, post_truncsti8,
1735 let Inst{7-5} = lane{2-0};
1737 def VST1LNd16_UPD : VST1LNWB<0b0100, {?,?,0,?}, "16", v4i16, post_truncsti16,
1739 let Inst{7-6} = lane{1-0};
1740 let Inst{4} = Rn{5};
1742 def VST1LNd32_UPD : VST1LNWB<0b1000, {?,0,?,?}, "32", v2i32, post_store,
1744 let Inst{7} = lane{0};
1745 let Inst{5-4} = Rn{5-4};
1748 def VST1LNq8Pseudo_UPD : VST1QLNWBPseudo<v16i8, post_truncsti8, NEONvgetlaneu>;
1749 def VST1LNq16Pseudo_UPD : VST1QLNWBPseudo<v8i16, post_truncsti16,NEONvgetlaneu>;
1750 def VST1LNq32Pseudo_UPD : VST1QLNWBPseudo<v4i32, post_store, extractelt>;
1752 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
1754 // VST2LN : Vector Store (single 2-element structure from one lane)
1755 class VST2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1756 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1757 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, nohash_imm:$lane),
1758 IIC_VST2ln, "vst2", Dt, "\\{$Vd[$lane], $src2[$lane]\\}, $Rn",
1761 let Inst{4} = Rn{4};
1762 let DecoderMethod = "DecodeVST2LN";
1765 def VST2LNd8 : VST2LN<0b0001, {?,?,?,?}, "8"> {
1766 let Inst{7-5} = lane{2-0};
1768 def VST2LNd16 : VST2LN<0b0101, {?,?,0,?}, "16"> {
1769 let Inst{7-6} = lane{1-0};
1771 def VST2LNd32 : VST2LN<0b1001, {?,0,0,?}, "32"> {
1772 let Inst{7} = lane{0};
1775 def VST2LNd8Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1776 def VST2LNd16Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1777 def VST2LNd32Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1779 // ...with double-spaced registers:
1780 def VST2LNq16 : VST2LN<0b0101, {?,?,1,?}, "16"> {
1781 let Inst{7-6} = lane{1-0};
1782 let Inst{4} = Rn{4};
1784 def VST2LNq32 : VST2LN<0b1001, {?,1,0,?}, "32"> {
1785 let Inst{7} = lane{0};
1786 let Inst{4} = Rn{4};
1789 def VST2LNq16Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
1790 def VST2LNq32Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
1792 // ...with address register writeback:
1793 class VST2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1794 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1795 (ins addrmode6:$addr, am6offset:$offset,
1796 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VST2lnu, "vst2", Dt,
1797 "\\{$src1[$lane], $src2[$lane]\\}, $addr$offset",
1798 "$addr.addr = $wb", []> {
1799 let Inst{4} = Rn{4};
1800 let DecoderMethod = "DecodeVST2LN";
1803 def VST2LNd8_UPD : VST2LNWB<0b0001, {?,?,?,?}, "8"> {
1804 let Inst{7-5} = lane{2-0};
1806 def VST2LNd16_UPD : VST2LNWB<0b0101, {?,?,0,?}, "16"> {
1807 let Inst{7-6} = lane{1-0};
1809 def VST2LNd32_UPD : VST2LNWB<0b1001, {?,0,0,?}, "32"> {
1810 let Inst{7} = lane{0};
1813 def VST2LNd8Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1814 def VST2LNd16Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1815 def VST2LNd32Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1817 def VST2LNq16_UPD : VST2LNWB<0b0101, {?,?,1,?}, "16"> {
1818 let Inst{7-6} = lane{1-0};
1820 def VST2LNq32_UPD : VST2LNWB<0b1001, {?,1,0,?}, "32"> {
1821 let Inst{7} = lane{0};
1824 def VST2LNq16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
1825 def VST2LNq32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
1827 // VST3LN : Vector Store (single 3-element structure from one lane)
1828 class VST3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1829 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1830 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3,
1831 nohash_imm:$lane), IIC_VST3ln, "vst3", Dt,
1832 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn", "", []> {
1834 let DecoderMethod = "DecodeVST3LN";
1837 def VST3LNd8 : VST3LN<0b0010, {?,?,?,0}, "8"> {
1838 let Inst{7-5} = lane{2-0};
1840 def VST3LNd16 : VST3LN<0b0110, {?,?,0,0}, "16"> {
1841 let Inst{7-6} = lane{1-0};
1843 def VST3LNd32 : VST3LN<0b1010, {?,0,0,0}, "32"> {
1844 let Inst{7} = lane{0};
1847 def VST3LNd8Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1848 def VST3LNd16Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1849 def VST3LNd32Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1851 // ...with double-spaced registers:
1852 def VST3LNq16 : VST3LN<0b0110, {?,?,1,0}, "16"> {
1853 let Inst{7-6} = lane{1-0};
1855 def VST3LNq32 : VST3LN<0b1010, {?,1,0,0}, "32"> {
1856 let Inst{7} = lane{0};
1859 def VST3LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
1860 def VST3LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
1862 // ...with address register writeback:
1863 class VST3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1864 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1865 (ins addrmode6:$Rn, am6offset:$Rm,
1866 DPR:$Vd, DPR:$src2, DPR:$src3, nohash_imm:$lane),
1867 IIC_VST3lnu, "vst3", Dt,
1868 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn$Rm",
1869 "$Rn.addr = $wb", []> {
1870 let DecoderMethod = "DecodeVST3LN";
1873 def VST3LNd8_UPD : VST3LNWB<0b0010, {?,?,?,0}, "8"> {
1874 let Inst{7-5} = lane{2-0};
1876 def VST3LNd16_UPD : VST3LNWB<0b0110, {?,?,0,0}, "16"> {
1877 let Inst{7-6} = lane{1-0};
1879 def VST3LNd32_UPD : VST3LNWB<0b1010, {?,0,0,0}, "32"> {
1880 let Inst{7} = lane{0};
1883 def VST3LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1884 def VST3LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1885 def VST3LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1887 def VST3LNq16_UPD : VST3LNWB<0b0110, {?,?,1,0}, "16"> {
1888 let Inst{7-6} = lane{1-0};
1890 def VST3LNq32_UPD : VST3LNWB<0b1010, {?,1,0,0}, "32"> {
1891 let Inst{7} = lane{0};
1894 def VST3LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
1895 def VST3LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
1897 // VST4LN : Vector Store (single 4-element structure from one lane)
1898 class VST4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1899 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1900 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4,
1901 nohash_imm:$lane), IIC_VST4ln, "vst4", Dt,
1902 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn",
1905 let Inst{4} = Rn{4};
1906 let DecoderMethod = "DecodeVST4LN";
1909 def VST4LNd8 : VST4LN<0b0011, {?,?,?,?}, "8"> {
1910 let Inst{7-5} = lane{2-0};
1912 def VST4LNd16 : VST4LN<0b0111, {?,?,0,?}, "16"> {
1913 let Inst{7-6} = lane{1-0};
1915 def VST4LNd32 : VST4LN<0b1011, {?,0,?,?}, "32"> {
1916 let Inst{7} = lane{0};
1917 let Inst{5} = Rn{5};
1920 def VST4LNd8Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1921 def VST4LNd16Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1922 def VST4LNd32Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1924 // ...with double-spaced registers:
1925 def VST4LNq16 : VST4LN<0b0111, {?,?,1,?}, "16"> {
1926 let Inst{7-6} = lane{1-0};
1928 def VST4LNq32 : VST4LN<0b1011, {?,1,?,?}, "32"> {
1929 let Inst{7} = lane{0};
1930 let Inst{5} = Rn{5};
1933 def VST4LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
1934 def VST4LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
1936 // ...with address register writeback:
1937 class VST4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1938 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1939 (ins addrmode6:$Rn, am6offset:$Rm,
1940 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
1941 IIC_VST4lnu, "vst4", Dt,
1942 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn$Rm",
1943 "$Rn.addr = $wb", []> {
1944 let Inst{4} = Rn{4};
1945 let DecoderMethod = "DecodeVST4LN";
1948 def VST4LNd8_UPD : VST4LNWB<0b0011, {?,?,?,?}, "8"> {
1949 let Inst{7-5} = lane{2-0};
1951 def VST4LNd16_UPD : VST4LNWB<0b0111, {?,?,0,?}, "16"> {
1952 let Inst{7-6} = lane{1-0};
1954 def VST4LNd32_UPD : VST4LNWB<0b1011, {?,0,?,?}, "32"> {
1955 let Inst{7} = lane{0};
1956 let Inst{5} = Rn{5};
1959 def VST4LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
1960 def VST4LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
1961 def VST4LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
1963 def VST4LNq16_UPD : VST4LNWB<0b0111, {?,?,1,?}, "16"> {
1964 let Inst{7-6} = lane{1-0};
1966 def VST4LNq32_UPD : VST4LNWB<0b1011, {?,1,?,?}, "32"> {
1967 let Inst{7} = lane{0};
1968 let Inst{5} = Rn{5};
1971 def VST4LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
1972 def VST4LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
1974 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
1977 //===----------------------------------------------------------------------===//
1978 // NEON pattern fragments
1979 //===----------------------------------------------------------------------===//
1981 // Extract D sub-registers of Q registers.
1982 def DSubReg_i8_reg : SDNodeXForm<imm, [{
1983 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1984 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/8, MVT::i32);
1986 def DSubReg_i16_reg : SDNodeXForm<imm, [{
1987 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1988 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/4, MVT::i32);
1990 def DSubReg_i32_reg : SDNodeXForm<imm, [{
1991 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1992 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/2, MVT::i32);
1994 def DSubReg_f64_reg : SDNodeXForm<imm, [{
1995 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1996 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue(), MVT::i32);
1999 // Extract S sub-registers of Q/D registers.
2000 def SSubReg_f32_reg : SDNodeXForm<imm, [{
2001 assert(ARM::ssub_3 == ARM::ssub_0+3 && "Unexpected subreg numbering");
2002 return CurDAG->getTargetConstant(ARM::ssub_0 + N->getZExtValue(), MVT::i32);
2005 // Translate lane numbers from Q registers to D subregs.
2006 def SubReg_i8_lane : SDNodeXForm<imm, [{
2007 return CurDAG->getTargetConstant(N->getZExtValue() & 7, MVT::i32);
2009 def SubReg_i16_lane : SDNodeXForm<imm, [{
2010 return CurDAG->getTargetConstant(N->getZExtValue() & 3, MVT::i32);
2012 def SubReg_i32_lane : SDNodeXForm<imm, [{
2013 return CurDAG->getTargetConstant(N->getZExtValue() & 1, MVT::i32);
2016 //===----------------------------------------------------------------------===//
2017 // Instruction Classes
2018 //===----------------------------------------------------------------------===//
2020 // Basic 2-register operations: double- and quad-register.
2021 class N2VD<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2022 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
2023 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
2024 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
2025 (ins DPR:$Vm), IIC_VUNAD, OpcodeStr, Dt,"$Vd, $Vm", "",
2026 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vm))))]>;
2027 class N2VQ<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2028 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
2029 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
2030 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
2031 (ins QPR:$Vm), IIC_VUNAQ, OpcodeStr, Dt,"$Vd, $Vm", "",
2032 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vm))))]>;
2034 // Basic 2-register intrinsics, both double- and quad-register.
2035 class N2VDInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2036 bits<2> op17_16, bits<5> op11_7, bit op4,
2037 InstrItinClass itin, string OpcodeStr, string Dt,
2038 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2039 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
2040 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2041 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm))))]>;
2042 class N2VQInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2043 bits<2> op17_16, bits<5> op11_7, bit op4,
2044 InstrItinClass itin, string OpcodeStr, string Dt,
2045 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2046 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
2047 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2048 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm))))]>;
2050 // Narrow 2-register operations.
2051 class N2VN<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2052 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2053 InstrItinClass itin, string OpcodeStr, string Dt,
2054 ValueType TyD, ValueType TyQ, SDNode OpNode>
2055 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$Vd),
2056 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2057 [(set DPR:$Vd, (TyD (OpNode (TyQ QPR:$Vm))))]>;
2059 // Narrow 2-register intrinsics.
2060 class N2VNInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2061 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2062 InstrItinClass itin, string OpcodeStr, string Dt,
2063 ValueType TyD, ValueType TyQ, Intrinsic IntOp>
2064 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$Vd),
2065 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2066 [(set DPR:$Vd, (TyD (IntOp (TyQ QPR:$Vm))))]>;
2068 // Long 2-register operations (currently only used for VMOVL).
2069 class N2VL<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2070 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2071 InstrItinClass itin, string OpcodeStr, string Dt,
2072 ValueType TyQ, ValueType TyD, SDNode OpNode>
2073 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$Vd),
2074 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2075 [(set QPR:$Vd, (TyQ (OpNode (TyD DPR:$Vm))))]>;
2077 // Long 2-register intrinsics.
2078 class N2VLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2079 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2080 InstrItinClass itin, string OpcodeStr, string Dt,
2081 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
2082 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$Vd),
2083 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2084 [(set QPR:$Vd, (TyQ (IntOp (TyD DPR:$Vm))))]>;
2086 // 2-register shuffles (VTRN/VZIP/VUZP), both double- and quad-register.
2087 class N2VDShuffle<bits<2> op19_18, bits<5> op11_7, string OpcodeStr, string Dt>
2088 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 0, 0, (outs DPR:$Vd, DPR:$Vm),
2089 (ins DPR:$src1, DPR:$src2), IIC_VPERMD,
2090 OpcodeStr, Dt, "$Vd, $Vm",
2091 "$src1 = $Vd, $src2 = $Vm", []>;
2092 class N2VQShuffle<bits<2> op19_18, bits<5> op11_7,
2093 InstrItinClass itin, string OpcodeStr, string Dt>
2094 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 1, 0, (outs QPR:$Vd, QPR:$Vm),
2095 (ins QPR:$src1, QPR:$src2), itin, OpcodeStr, Dt, "$Vd, $Vm",
2096 "$src1 = $Vd, $src2 = $Vm", []>;
2098 // Basic 3-register operations: double- and quad-register.
2099 class N3VD<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2100 InstrItinClass itin, string OpcodeStr, string Dt,
2101 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
2102 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2103 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2104 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2105 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
2106 let isCommutable = Commutable;
2108 // Same as N3VD but no data type.
2109 class N3VDX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2110 InstrItinClass itin, string OpcodeStr,
2111 ValueType ResTy, ValueType OpTy,
2112 SDNode OpNode, bit Commutable>
2113 : N3VX<op24, op23, op21_20, op11_8, 0, op4,
2114 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2115 OpcodeStr, "$Vd, $Vn, $Vm", "",
2116 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>{
2117 let isCommutable = Commutable;
2120 class N3VDSL<bits<2> op21_20, bits<4> op11_8,
2121 InstrItinClass itin, string OpcodeStr, string Dt,
2122 ValueType Ty, SDNode ShOp>
2123 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2124 (outs DPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2125 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2127 (Ty (ShOp (Ty DPR:$Vn),
2128 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),imm:$lane)))))]> {
2129 let isCommutable = 0;
2131 class N3VDSL16<bits<2> op21_20, bits<4> op11_8,
2132 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
2133 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2134 (outs DPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2135 NVMulSLFrm, IIC_VMULi16D, OpcodeStr, Dt,"$Vd, $Vn, $Vm$lane","",
2137 (Ty (ShOp (Ty DPR:$Vn),
2138 (Ty (NEONvduplane (Ty DPR_8:$Vm), imm:$lane)))))]> {
2139 let isCommutable = 0;
2142 class N3VQ<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2143 InstrItinClass itin, string OpcodeStr, string Dt,
2144 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
2145 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2146 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2147 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2148 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
2149 let isCommutable = Commutable;
2151 class N3VQX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2152 InstrItinClass itin, string OpcodeStr,
2153 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
2154 : N3VX<op24, op23, op21_20, op11_8, 1, op4,
2155 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2156 OpcodeStr, "$Vd, $Vn, $Vm", "",
2157 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]>{
2158 let isCommutable = Commutable;
2160 class N3VQSL<bits<2> op21_20, bits<4> op11_8,
2161 InstrItinClass itin, string OpcodeStr, string Dt,
2162 ValueType ResTy, ValueType OpTy, SDNode ShOp>
2163 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2164 (outs QPR:$Vd), (ins QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2165 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2166 [(set (ResTy QPR:$Vd),
2167 (ResTy (ShOp (ResTy QPR:$Vn),
2168 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2170 let isCommutable = 0;
2172 class N3VQSL16<bits<2> op21_20, bits<4> op11_8, string OpcodeStr, string Dt,
2173 ValueType ResTy, ValueType OpTy, SDNode ShOp>
2174 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2175 (outs QPR:$Vd), (ins QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2176 NVMulSLFrm, IIC_VMULi16Q, OpcodeStr, Dt,"$Vd, $Vn, $Vm$lane", "",
2177 [(set (ResTy QPR:$Vd),
2178 (ResTy (ShOp (ResTy QPR:$Vn),
2179 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2181 let isCommutable = 0;
2184 // Basic 3-register intrinsics, both double- and quad-register.
2185 class N3VDInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2186 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2187 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
2188 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2189 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), f, itin,
2190 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2191 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
2192 let isCommutable = Commutable;
2194 class N3VDIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2195 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
2196 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2197 (outs DPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2198 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2200 (Ty (IntOp (Ty DPR:$Vn),
2201 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),
2203 let isCommutable = 0;
2205 class N3VDIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2206 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
2207 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2208 (outs DPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2209 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2211 (Ty (IntOp (Ty DPR:$Vn),
2212 (Ty (NEONvduplane (Ty DPR_8:$Vm), imm:$lane)))))]> {
2213 let isCommutable = 0;
2215 class N3VDIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2216 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2217 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2218 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2219 (outs DPR:$Vd), (ins DPR:$Vm, DPR:$Vn), f, itin,
2220 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
2221 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (OpTy DPR:$Vn))))]> {
2222 let isCommutable = 0;
2225 class N3VQInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2226 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2227 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
2228 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2229 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), f, itin,
2230 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2231 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
2232 let isCommutable = Commutable;
2234 class N3VQIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2235 string OpcodeStr, string Dt,
2236 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2237 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2238 (outs QPR:$Vd), (ins QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2239 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2240 [(set (ResTy QPR:$Vd),
2241 (ResTy (IntOp (ResTy QPR:$Vn),
2242 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2244 let isCommutable = 0;
2246 class N3VQIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2247 string OpcodeStr, string Dt,
2248 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2249 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2250 (outs QPR:$Vd), (ins QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2251 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2252 [(set (ResTy QPR:$Vd),
2253 (ResTy (IntOp (ResTy QPR:$Vn),
2254 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2256 let isCommutable = 0;
2258 class N3VQIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2259 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2260 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2261 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2262 (outs QPR:$Vd), (ins QPR:$Vm, QPR:$Vn), f, itin,
2263 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
2264 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (OpTy QPR:$Vn))))]> {
2265 let isCommutable = 0;
2268 // Multiply-Add/Sub operations: double- and quad-register.
2269 class N3VDMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2270 InstrItinClass itin, string OpcodeStr, string Dt,
2271 ValueType Ty, SDPatternOperator MulOp, SDPatternOperator OpNode>
2272 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2273 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2274 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2275 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
2276 (Ty (MulOp DPR:$Vn, DPR:$Vm)))))]>;
2278 class N3VDMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2279 string OpcodeStr, string Dt,
2280 ValueType Ty, SDPatternOperator MulOp, SDPatternOperator ShOp>
2281 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2283 (ins DPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2285 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2287 (Ty (ShOp (Ty DPR:$src1),
2289 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),
2291 class N3VDMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2292 string OpcodeStr, string Dt,
2293 ValueType Ty, SDNode MulOp, SDNode ShOp>
2294 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2296 (ins DPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2298 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2300 (Ty (ShOp (Ty DPR:$src1),
2302 (Ty (NEONvduplane (Ty DPR_8:$Vm),
2305 class N3VQMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2306 InstrItinClass itin, string OpcodeStr, string Dt, ValueType Ty,
2307 SDPatternOperator MulOp, SDPatternOperator OpNode>
2308 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2309 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2310 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2311 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
2312 (Ty (MulOp QPR:$Vn, QPR:$Vm)))))]>;
2313 class N3VQMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2314 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2315 SDPatternOperator MulOp, SDPatternOperator ShOp>
2316 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2318 (ins QPR:$src1, QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2320 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2321 [(set (ResTy QPR:$Vd),
2322 (ResTy (ShOp (ResTy QPR:$src1),
2323 (ResTy (MulOp QPR:$Vn,
2324 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2326 class N3VQMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2327 string OpcodeStr, string Dt,
2328 ValueType ResTy, ValueType OpTy,
2329 SDNode MulOp, SDNode ShOp>
2330 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2332 (ins QPR:$src1, QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2334 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2335 [(set (ResTy QPR:$Vd),
2336 (ResTy (ShOp (ResTy QPR:$src1),
2337 (ResTy (MulOp QPR:$Vn,
2338 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2341 // Neon Intrinsic-Op instructions (VABA): double- and quad-register.
2342 class N3VDIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2343 InstrItinClass itin, string OpcodeStr, string Dt,
2344 ValueType Ty, Intrinsic IntOp, SDNode OpNode>
2345 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2346 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2347 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2348 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
2349 (Ty (IntOp (Ty DPR:$Vn), (Ty DPR:$Vm))))))]>;
2350 class N3VQIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2351 InstrItinClass itin, string OpcodeStr, string Dt,
2352 ValueType Ty, Intrinsic IntOp, SDNode OpNode>
2353 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2354 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2355 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2356 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
2357 (Ty (IntOp (Ty QPR:$Vn), (Ty QPR:$Vm))))))]>;
2359 // Neon 3-argument intrinsics, both double- and quad-register.
2360 // The destination register is also used as the first source operand register.
2361 class N3VDInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2362 InstrItinClass itin, string OpcodeStr, string Dt,
2363 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2364 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2365 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2366 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2367 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$src1),
2368 (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>;
2369 class N3VQInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2370 InstrItinClass itin, string OpcodeStr, string Dt,
2371 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2372 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2373 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2374 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2375 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$src1),
2376 (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]>;
2378 // Long Multiply-Add/Sub operations.
2379 class N3VLMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2380 InstrItinClass itin, string OpcodeStr, string Dt,
2381 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2382 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2383 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2384 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2385 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
2386 (TyQ (MulOp (TyD DPR:$Vn),
2387 (TyD DPR:$Vm)))))]>;
2388 class N3VLMulOpSL<bit op24, bits<2> op21_20, bits<4> op11_8,
2389 InstrItinClass itin, string OpcodeStr, string Dt,
2390 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2391 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$Vd),
2392 (ins QPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2394 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2396 (OpNode (TyQ QPR:$src1),
2397 (TyQ (MulOp (TyD DPR:$Vn),
2398 (TyD (NEONvduplane (TyD DPR_VFP2:$Vm),
2400 class N3VLMulOpSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2401 InstrItinClass itin, string OpcodeStr, string Dt,
2402 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2403 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$Vd),
2404 (ins QPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2406 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2408 (OpNode (TyQ QPR:$src1),
2409 (TyQ (MulOp (TyD DPR:$Vn),
2410 (TyD (NEONvduplane (TyD DPR_8:$Vm),
2413 // Long Intrinsic-Op vector operations with explicit extend (VABAL).
2414 class N3VLIntExtOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2415 InstrItinClass itin, string OpcodeStr, string Dt,
2416 ValueType TyQ, ValueType TyD, Intrinsic IntOp, SDNode ExtOp,
2418 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2419 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2420 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2421 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
2422 (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
2423 (TyD DPR:$Vm)))))))]>;
2425 // Neon Long 3-argument intrinsic. The destination register is
2426 // a quad-register and is also used as the first source operand register.
2427 class N3VLInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2428 InstrItinClass itin, string OpcodeStr, string Dt,
2429 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
2430 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2431 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2432 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2434 (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$Vn), (TyD DPR:$Vm))))]>;
2435 class N3VLInt3SL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2436 string OpcodeStr, string Dt,
2437 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2438 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2440 (ins QPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2442 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2443 [(set (ResTy QPR:$Vd),
2444 (ResTy (IntOp (ResTy QPR:$src1),
2446 (OpTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2448 class N3VLInt3SL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2449 InstrItinClass itin, string OpcodeStr, string Dt,
2450 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2451 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2453 (ins QPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2455 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2456 [(set (ResTy QPR:$Vd),
2457 (ResTy (IntOp (ResTy QPR:$src1),
2459 (OpTy (NEONvduplane (OpTy DPR_8:$Vm),
2462 // Narrowing 3-register intrinsics.
2463 class N3VNInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2464 string OpcodeStr, string Dt, ValueType TyD, ValueType TyQ,
2465 Intrinsic IntOp, bit Commutable>
2466 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2467 (outs DPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINi4D,
2468 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2469 [(set DPR:$Vd, (TyD (IntOp (TyQ QPR:$Vn), (TyQ QPR:$Vm))))]> {
2470 let isCommutable = Commutable;
2473 // Long 3-register operations.
2474 class N3VL<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2475 InstrItinClass itin, string OpcodeStr, string Dt,
2476 ValueType TyQ, ValueType TyD, SDNode OpNode, bit Commutable>
2477 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2478 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2479 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2480 [(set QPR:$Vd, (TyQ (OpNode (TyD DPR:$Vn), (TyD DPR:$Vm))))]> {
2481 let isCommutable = Commutable;
2483 class N3VLSL<bit op24, bits<2> op21_20, bits<4> op11_8,
2484 InstrItinClass itin, string OpcodeStr, string Dt,
2485 ValueType TyQ, ValueType TyD, SDNode OpNode>
2486 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2487 (outs QPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2488 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2490 (TyQ (OpNode (TyD DPR:$Vn),
2491 (TyD (NEONvduplane (TyD DPR_VFP2:$Vm),imm:$lane)))))]>;
2492 class N3VLSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2493 InstrItinClass itin, string OpcodeStr, string Dt,
2494 ValueType TyQ, ValueType TyD, SDNode OpNode>
2495 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2496 (outs QPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2497 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2499 (TyQ (OpNode (TyD DPR:$Vn),
2500 (TyD (NEONvduplane (TyD DPR_8:$Vm), imm:$lane)))))]>;
2502 // Long 3-register operations with explicitly extended operands.
2503 class N3VLExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2504 InstrItinClass itin, string OpcodeStr, string Dt,
2505 ValueType TyQ, ValueType TyD, SDNode OpNode, SDNode ExtOp,
2507 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2508 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2509 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2510 [(set QPR:$Vd, (OpNode (TyQ (ExtOp (TyD DPR:$Vn))),
2511 (TyQ (ExtOp (TyD DPR:$Vm)))))]> {
2512 let isCommutable = Commutable;
2515 // Long 3-register intrinsics with explicit extend (VABDL).
2516 class N3VLIntExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2517 InstrItinClass itin, string OpcodeStr, string Dt,
2518 ValueType TyQ, ValueType TyD, Intrinsic IntOp, SDNode ExtOp,
2520 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2521 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2522 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2523 [(set QPR:$Vd, (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
2524 (TyD DPR:$Vm))))))]> {
2525 let isCommutable = Commutable;
2528 // Long 3-register intrinsics.
2529 class N3VLInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2530 InstrItinClass itin, string OpcodeStr, string Dt,
2531 ValueType TyQ, ValueType TyD, Intrinsic IntOp, bit Commutable>
2532 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2533 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2534 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2535 [(set QPR:$Vd, (TyQ (IntOp (TyD DPR:$Vn), (TyD DPR:$Vm))))]> {
2536 let isCommutable = Commutable;
2538 class N3VLIntSL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2539 string OpcodeStr, string Dt,
2540 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2541 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2542 (outs QPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2543 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2544 [(set (ResTy QPR:$Vd),
2545 (ResTy (IntOp (OpTy DPR:$Vn),
2546 (OpTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2548 class N3VLIntSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2549 InstrItinClass itin, string OpcodeStr, string Dt,
2550 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2551 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2552 (outs QPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2553 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2554 [(set (ResTy QPR:$Vd),
2555 (ResTy (IntOp (OpTy DPR:$Vn),
2556 (OpTy (NEONvduplane (OpTy DPR_8:$Vm),
2559 // Wide 3-register operations.
2560 class N3VW<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2561 string OpcodeStr, string Dt, ValueType TyQ, ValueType TyD,
2562 SDNode OpNode, SDNode ExtOp, bit Commutable>
2563 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2564 (outs QPR:$Vd), (ins QPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VSUBiD,
2565 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2566 [(set QPR:$Vd, (OpNode (TyQ QPR:$Vn),
2567 (TyQ (ExtOp (TyD DPR:$Vm)))))]> {
2568 let isCommutable = Commutable;
2571 // Pairwise long 2-register intrinsics, both double- and quad-register.
2572 class N2VDPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2573 bits<2> op17_16, bits<5> op11_7, bit op4,
2574 string OpcodeStr, string Dt,
2575 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2576 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
2577 (ins DPR:$Vm), IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
2578 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm))))]>;
2579 class N2VQPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2580 bits<2> op17_16, bits<5> op11_7, bit op4,
2581 string OpcodeStr, string Dt,
2582 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2583 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
2584 (ins QPR:$Vm), IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
2585 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm))))]>;
2587 // Pairwise long 2-register accumulate intrinsics,
2588 // both double- and quad-register.
2589 // The destination register is also used as the first source operand register.
2590 class N2VDPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2591 bits<2> op17_16, bits<5> op11_7, bit op4,
2592 string OpcodeStr, string Dt,
2593 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2594 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
2595 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vm), IIC_VPALiD,
2596 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
2597 [(set DPR:$Vd, (ResTy (IntOp (ResTy DPR:$src1), (OpTy DPR:$Vm))))]>;
2598 class N2VQPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2599 bits<2> op17_16, bits<5> op11_7, bit op4,
2600 string OpcodeStr, string Dt,
2601 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2602 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4,
2603 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vm), IIC_VPALiQ,
2604 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
2605 [(set QPR:$Vd, (ResTy (IntOp (ResTy QPR:$src1), (OpTy QPR:$Vm))))]>;
2607 // Shift by immediate,
2608 // both double- and quad-register.
2609 class N2VDSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2610 Format f, InstrItinClass itin, Operand ImmTy,
2611 string OpcodeStr, string Dt, ValueType Ty, SDNode OpNode>
2612 : N2VImm<op24, op23, op11_8, op7, 0, op4,
2613 (outs DPR:$Vd), (ins DPR:$Vm, ImmTy:$SIMM), f, itin,
2614 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2615 [(set DPR:$Vd, (Ty (OpNode (Ty DPR:$Vm), (i32 imm:$SIMM))))]>;
2616 class N2VQSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2617 Format f, InstrItinClass itin, Operand ImmTy,
2618 string OpcodeStr, string Dt, ValueType Ty, SDNode OpNode>
2619 : N2VImm<op24, op23, op11_8, op7, 1, op4,
2620 (outs QPR:$Vd), (ins QPR:$Vm, ImmTy:$SIMM), f, itin,
2621 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2622 [(set QPR:$Vd, (Ty (OpNode (Ty QPR:$Vm), (i32 imm:$SIMM))))]>;
2624 // Long shift by immediate.
2625 class N2VLSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
2626 string OpcodeStr, string Dt,
2627 ValueType ResTy, ValueType OpTy, SDNode OpNode>
2628 : N2VImm<op24, op23, op11_8, op7, op6, op4,
2629 (outs QPR:$Vd), (ins DPR:$Vm, i32imm:$SIMM), N2RegVShLFrm,
2630 IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2631 [(set QPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vm),
2632 (i32 imm:$SIMM))))]>;
2634 // Narrow shift by immediate.
2635 class N2VNSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
2636 InstrItinClass itin, string OpcodeStr, string Dt,
2637 ValueType ResTy, ValueType OpTy, Operand ImmTy, SDNode OpNode>
2638 : N2VImm<op24, op23, op11_8, op7, op6, op4,
2639 (outs DPR:$Vd), (ins QPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, itin,
2640 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2641 [(set DPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vm),
2642 (i32 imm:$SIMM))))]>;
2644 // Shift right by immediate and accumulate,
2645 // both double- and quad-register.
2646 class N2VDShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2647 Operand ImmTy, string OpcodeStr, string Dt,
2648 ValueType Ty, SDNode ShOp>
2649 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
2650 (ins DPR:$src1, DPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, IIC_VPALiD,
2651 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2652 [(set DPR:$Vd, (Ty (add DPR:$src1,
2653 (Ty (ShOp DPR:$Vm, (i32 imm:$SIMM))))))]>;
2654 class N2VQShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2655 Operand ImmTy, string OpcodeStr, string Dt,
2656 ValueType Ty, SDNode ShOp>
2657 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
2658 (ins QPR:$src1, QPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, IIC_VPALiD,
2659 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2660 [(set QPR:$Vd, (Ty (add QPR:$src1,
2661 (Ty (ShOp QPR:$Vm, (i32 imm:$SIMM))))))]>;
2663 // Shift by immediate and insert,
2664 // both double- and quad-register.
2665 class N2VDShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2666 Operand ImmTy, Format f, string OpcodeStr, string Dt,
2667 ValueType Ty,SDNode ShOp>
2668 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
2669 (ins DPR:$src1, DPR:$Vm, ImmTy:$SIMM), f, IIC_VSHLiD,
2670 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2671 [(set DPR:$Vd, (Ty (ShOp DPR:$src1, DPR:$Vm, (i32 imm:$SIMM))))]>;
2672 class N2VQShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2673 Operand ImmTy, Format f, string OpcodeStr, string Dt,
2674 ValueType Ty,SDNode ShOp>
2675 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
2676 (ins QPR:$src1, QPR:$Vm, ImmTy:$SIMM), f, IIC_VSHLiQ,
2677 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2678 [(set QPR:$Vd, (Ty (ShOp QPR:$src1, QPR:$Vm, (i32 imm:$SIMM))))]>;
2680 // Convert, with fractional bits immediate,
2681 // both double- and quad-register.
2682 class N2VCvtD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2683 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2685 : N2VImm<op24, op23, op11_8, op7, 0, op4,
2686 (outs DPR:$Vd), (ins DPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
2687 IIC_VUNAD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2688 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (i32 imm:$SIMM))))]>;
2689 class N2VCvtQ<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2690 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2692 : N2VImm<op24, op23, op11_8, op7, 1, op4,
2693 (outs QPR:$Vd), (ins QPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
2694 IIC_VUNAQ, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2695 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (i32 imm:$SIMM))))]>;
2697 //===----------------------------------------------------------------------===//
2699 //===----------------------------------------------------------------------===//
2701 // Abbreviations used in multiclass suffixes:
2702 // Q = quarter int (8 bit) elements
2703 // H = half int (16 bit) elements
2704 // S = single int (32 bit) elements
2705 // D = double int (64 bit) elements
2707 // Neon 2-register vector operations and intrinsics.
2709 // Neon 2-register comparisons.
2710 // source operand element sizes of 8, 16 and 32 bits:
2711 multiclass N2V_QHS_cmp<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2712 bits<5> op11_7, bit op4, string opc, string Dt,
2713 string asm, SDNode OpNode> {
2714 // 64-bit vector types.
2715 def v8i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 0, op4,
2716 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2717 opc, !strconcat(Dt, "8"), asm, "",
2718 [(set DPR:$Vd, (v8i8 (OpNode (v8i8 DPR:$Vm))))]>;
2719 def v4i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 0, op4,
2720 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2721 opc, !strconcat(Dt, "16"), asm, "",
2722 [(set DPR:$Vd, (v4i16 (OpNode (v4i16 DPR:$Vm))))]>;
2723 def v2i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
2724 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2725 opc, !strconcat(Dt, "32"), asm, "",
2726 [(set DPR:$Vd, (v2i32 (OpNode (v2i32 DPR:$Vm))))]>;
2727 def v2f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
2728 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2729 opc, "f32", asm, "",
2730 [(set DPR:$Vd, (v2i32 (OpNode (v2f32 DPR:$Vm))))]> {
2731 let Inst{10} = 1; // overwrite F = 1
2734 // 128-bit vector types.
2735 def v16i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 1, op4,
2736 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2737 opc, !strconcat(Dt, "8"), asm, "",
2738 [(set QPR:$Vd, (v16i8 (OpNode (v16i8 QPR:$Vm))))]>;
2739 def v8i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 1, op4,
2740 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2741 opc, !strconcat(Dt, "16"), asm, "",
2742 [(set QPR:$Vd, (v8i16 (OpNode (v8i16 QPR:$Vm))))]>;
2743 def v4i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
2744 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2745 opc, !strconcat(Dt, "32"), asm, "",
2746 [(set QPR:$Vd, (v4i32 (OpNode (v4i32 QPR:$Vm))))]>;
2747 def v4f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
2748 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2749 opc, "f32", asm, "",
2750 [(set QPR:$Vd, (v4i32 (OpNode (v4f32 QPR:$Vm))))]> {
2751 let Inst{10} = 1; // overwrite F = 1
2756 // Neon 2-register vector intrinsics,
2757 // element sizes of 8, 16 and 32 bits:
2758 multiclass N2VInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2759 bits<5> op11_7, bit op4,
2760 InstrItinClass itinD, InstrItinClass itinQ,
2761 string OpcodeStr, string Dt, Intrinsic IntOp> {
2762 // 64-bit vector types.
2763 def v8i8 : N2VDInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2764 itinD, OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
2765 def v4i16 : N2VDInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2766 itinD, OpcodeStr, !strconcat(Dt, "16"),v4i16,v4i16,IntOp>;
2767 def v2i32 : N2VDInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2768 itinD, OpcodeStr, !strconcat(Dt, "32"),v2i32,v2i32,IntOp>;
2770 // 128-bit vector types.
2771 def v16i8 : N2VQInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2772 itinQ, OpcodeStr, !strconcat(Dt, "8"), v16i8,v16i8,IntOp>;
2773 def v8i16 : N2VQInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2774 itinQ, OpcodeStr, !strconcat(Dt, "16"),v8i16,v8i16,IntOp>;
2775 def v4i32 : N2VQInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2776 itinQ, OpcodeStr, !strconcat(Dt, "32"),v4i32,v4i32,IntOp>;
2780 // Neon Narrowing 2-register vector operations,
2781 // source operand element sizes of 16, 32 and 64 bits:
2782 multiclass N2VN_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2783 bits<5> op11_7, bit op6, bit op4,
2784 InstrItinClass itin, string OpcodeStr, string Dt,
2786 def v8i8 : N2VN<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
2787 itin, OpcodeStr, !strconcat(Dt, "16"),
2788 v8i8, v8i16, OpNode>;
2789 def v4i16 : N2VN<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
2790 itin, OpcodeStr, !strconcat(Dt, "32"),
2791 v4i16, v4i32, OpNode>;
2792 def v2i32 : N2VN<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
2793 itin, OpcodeStr, !strconcat(Dt, "64"),
2794 v2i32, v2i64, OpNode>;
2797 // Neon Narrowing 2-register vector intrinsics,
2798 // source operand element sizes of 16, 32 and 64 bits:
2799 multiclass N2VNInt_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2800 bits<5> op11_7, bit op6, bit op4,
2801 InstrItinClass itin, string OpcodeStr, string Dt,
2803 def v8i8 : N2VNInt<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
2804 itin, OpcodeStr, !strconcat(Dt, "16"),
2805 v8i8, v8i16, IntOp>;
2806 def v4i16 : N2VNInt<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
2807 itin, OpcodeStr, !strconcat(Dt, "32"),
2808 v4i16, v4i32, IntOp>;
2809 def v2i32 : N2VNInt<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
2810 itin, OpcodeStr, !strconcat(Dt, "64"),
2811 v2i32, v2i64, IntOp>;
2815 // Neon Lengthening 2-register vector intrinsic (currently specific to VMOVL).
2816 // source operand element sizes of 16, 32 and 64 bits:
2817 multiclass N2VL_QHS<bits<2> op24_23, bits<5> op11_7, bit op6, bit op4,
2818 string OpcodeStr, string Dt, SDNode OpNode> {
2819 def v8i16 : N2VL<op24_23, 0b00, 0b10, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2820 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode>;
2821 def v4i32 : N2VL<op24_23, 0b01, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2822 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
2823 def v2i64 : N2VL<op24_23, 0b10, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2824 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
2828 // Neon 3-register vector operations.
2830 // First with only element sizes of 8, 16 and 32 bits:
2831 multiclass N3V_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2832 InstrItinClass itinD16, InstrItinClass itinD32,
2833 InstrItinClass itinQ16, InstrItinClass itinQ32,
2834 string OpcodeStr, string Dt,
2835 SDNode OpNode, bit Commutable = 0> {
2836 // 64-bit vector types.
2837 def v8i8 : N3VD<op24, op23, 0b00, op11_8, op4, itinD16,
2838 OpcodeStr, !strconcat(Dt, "8"),
2839 v8i8, v8i8, OpNode, Commutable>;
2840 def v4i16 : N3VD<op24, op23, 0b01, op11_8, op4, itinD16,
2841 OpcodeStr, !strconcat(Dt, "16"),
2842 v4i16, v4i16, OpNode, Commutable>;
2843 def v2i32 : N3VD<op24, op23, 0b10, op11_8, op4, itinD32,
2844 OpcodeStr, !strconcat(Dt, "32"),
2845 v2i32, v2i32, OpNode, Commutable>;
2847 // 128-bit vector types.
2848 def v16i8 : N3VQ<op24, op23, 0b00, op11_8, op4, itinQ16,
2849 OpcodeStr, !strconcat(Dt, "8"),
2850 v16i8, v16i8, OpNode, Commutable>;
2851 def v8i16 : N3VQ<op24, op23, 0b01, op11_8, op4, itinQ16,
2852 OpcodeStr, !strconcat(Dt, "16"),
2853 v8i16, v8i16, OpNode, Commutable>;
2854 def v4i32 : N3VQ<op24, op23, 0b10, op11_8, op4, itinQ32,
2855 OpcodeStr, !strconcat(Dt, "32"),
2856 v4i32, v4i32, OpNode, Commutable>;
2859 multiclass N3VSL_HS<bits<4> op11_8, string OpcodeStr, string Dt, SDNode ShOp> {
2860 def v4i16 : N3VDSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
2862 def v2i32 : N3VDSL<0b10, op11_8, IIC_VMULi32D, OpcodeStr, !strconcat(Dt,"32"),
2864 def v8i16 : N3VQSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
2865 v8i16, v4i16, ShOp>;
2866 def v4i32 : N3VQSL<0b10, op11_8, IIC_VMULi32Q, OpcodeStr, !strconcat(Dt,"32"),
2867 v4i32, v2i32, ShOp>;
2870 // ....then also with element size 64 bits:
2871 multiclass N3V_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2872 InstrItinClass itinD, InstrItinClass itinQ,
2873 string OpcodeStr, string Dt,
2874 SDNode OpNode, bit Commutable = 0>
2875 : N3V_QHS<op24, op23, op11_8, op4, itinD, itinD, itinQ, itinQ,
2876 OpcodeStr, Dt, OpNode, Commutable> {
2877 def v1i64 : N3VD<op24, op23, 0b11, op11_8, op4, itinD,
2878 OpcodeStr, !strconcat(Dt, "64"),
2879 v1i64, v1i64, OpNode, Commutable>;
2880 def v2i64 : N3VQ<op24, op23, 0b11, op11_8, op4, itinQ,
2881 OpcodeStr, !strconcat(Dt, "64"),
2882 v2i64, v2i64, OpNode, Commutable>;
2886 // Neon 3-register vector intrinsics.
2888 // First with only element sizes of 16 and 32 bits:
2889 multiclass N3VInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2890 InstrItinClass itinD16, InstrItinClass itinD32,
2891 InstrItinClass itinQ16, InstrItinClass itinQ32,
2892 string OpcodeStr, string Dt,
2893 Intrinsic IntOp, bit Commutable = 0> {
2894 // 64-bit vector types.
2895 def v4i16 : N3VDInt<op24, op23, 0b01, op11_8, op4, f, itinD16,
2896 OpcodeStr, !strconcat(Dt, "16"),
2897 v4i16, v4i16, IntOp, Commutable>;
2898 def v2i32 : N3VDInt<op24, op23, 0b10, op11_8, op4, f, itinD32,
2899 OpcodeStr, !strconcat(Dt, "32"),
2900 v2i32, v2i32, IntOp, Commutable>;
2902 // 128-bit vector types.
2903 def v8i16 : N3VQInt<op24, op23, 0b01, op11_8, op4, f, itinQ16,
2904 OpcodeStr, !strconcat(Dt, "16"),
2905 v8i16, v8i16, IntOp, Commutable>;
2906 def v4i32 : N3VQInt<op24, op23, 0b10, op11_8, op4, f, itinQ32,
2907 OpcodeStr, !strconcat(Dt, "32"),
2908 v4i32, v4i32, IntOp, Commutable>;
2910 multiclass N3VInt_HSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2911 InstrItinClass itinD16, InstrItinClass itinD32,
2912 InstrItinClass itinQ16, InstrItinClass itinQ32,
2913 string OpcodeStr, string Dt,
2915 // 64-bit vector types.
2916 def v4i16 : N3VDIntSh<op24, op23, 0b01, op11_8, op4, f, itinD16,
2917 OpcodeStr, !strconcat(Dt, "16"),
2918 v4i16, v4i16, IntOp>;
2919 def v2i32 : N3VDIntSh<op24, op23, 0b10, op11_8, op4, f, itinD32,
2920 OpcodeStr, !strconcat(Dt, "32"),
2921 v2i32, v2i32, IntOp>;
2923 // 128-bit vector types.
2924 def v8i16 : N3VQIntSh<op24, op23, 0b01, op11_8, op4, f, itinQ16,
2925 OpcodeStr, !strconcat(Dt, "16"),
2926 v8i16, v8i16, IntOp>;
2927 def v4i32 : N3VQIntSh<op24, op23, 0b10, op11_8, op4, f, itinQ32,
2928 OpcodeStr, !strconcat(Dt, "32"),
2929 v4i32, v4i32, IntOp>;
2932 multiclass N3VIntSL_HS<bits<4> op11_8,
2933 InstrItinClass itinD16, InstrItinClass itinD32,
2934 InstrItinClass itinQ16, InstrItinClass itinQ32,
2935 string OpcodeStr, string Dt, Intrinsic IntOp> {
2936 def v4i16 : N3VDIntSL16<0b01, op11_8, itinD16,
2937 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp>;
2938 def v2i32 : N3VDIntSL<0b10, op11_8, itinD32,
2939 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp>;
2940 def v8i16 : N3VQIntSL16<0b01, op11_8, itinQ16,
2941 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16, IntOp>;
2942 def v4i32 : N3VQIntSL<0b10, op11_8, itinQ32,
2943 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32, IntOp>;
2946 // ....then also with element size of 8 bits:
2947 multiclass N3VInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2948 InstrItinClass itinD16, InstrItinClass itinD32,
2949 InstrItinClass itinQ16, InstrItinClass itinQ32,
2950 string OpcodeStr, string Dt,
2951 Intrinsic IntOp, bit Commutable = 0>
2952 : N3VInt_HS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2953 OpcodeStr, Dt, IntOp, Commutable> {
2954 def v8i8 : N3VDInt<op24, op23, 0b00, op11_8, op4, f, itinD16,
2955 OpcodeStr, !strconcat(Dt, "8"),
2956 v8i8, v8i8, IntOp, Commutable>;
2957 def v16i8 : N3VQInt<op24, op23, 0b00, op11_8, op4, f, itinQ16,
2958 OpcodeStr, !strconcat(Dt, "8"),
2959 v16i8, v16i8, IntOp, Commutable>;
2961 multiclass N3VInt_QHSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2962 InstrItinClass itinD16, InstrItinClass itinD32,
2963 InstrItinClass itinQ16, InstrItinClass itinQ32,
2964 string OpcodeStr, string Dt,
2966 : N3VInt_HSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2967 OpcodeStr, Dt, IntOp> {
2968 def v8i8 : N3VDIntSh<op24, op23, 0b00, op11_8, op4, f, itinD16,
2969 OpcodeStr, !strconcat(Dt, "8"),
2971 def v16i8 : N3VQIntSh<op24, op23, 0b00, op11_8, op4, f, itinQ16,
2972 OpcodeStr, !strconcat(Dt, "8"),
2973 v16i8, v16i8, IntOp>;
2977 // ....then also with element size of 64 bits:
2978 multiclass N3VInt_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2979 InstrItinClass itinD16, InstrItinClass itinD32,
2980 InstrItinClass itinQ16, InstrItinClass itinQ32,
2981 string OpcodeStr, string Dt,
2982 Intrinsic IntOp, bit Commutable = 0>
2983 : N3VInt_QHS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2984 OpcodeStr, Dt, IntOp, Commutable> {
2985 def v1i64 : N3VDInt<op24, op23, 0b11, op11_8, op4, f, itinD32,
2986 OpcodeStr, !strconcat(Dt, "64"),
2987 v1i64, v1i64, IntOp, Commutable>;
2988 def v2i64 : N3VQInt<op24, op23, 0b11, op11_8, op4, f, itinQ32,
2989 OpcodeStr, !strconcat(Dt, "64"),
2990 v2i64, v2i64, IntOp, Commutable>;
2992 multiclass N3VInt_QHSDSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2993 InstrItinClass itinD16, InstrItinClass itinD32,
2994 InstrItinClass itinQ16, InstrItinClass itinQ32,
2995 string OpcodeStr, string Dt,
2997 : N3VInt_QHSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2998 OpcodeStr, Dt, IntOp> {
2999 def v1i64 : N3VDIntSh<op24, op23, 0b11, op11_8, op4, f, itinD32,
3000 OpcodeStr, !strconcat(Dt, "64"),
3001 v1i64, v1i64, IntOp>;
3002 def v2i64 : N3VQIntSh<op24, op23, 0b11, op11_8, op4, f, itinQ32,
3003 OpcodeStr, !strconcat(Dt, "64"),
3004 v2i64, v2i64, IntOp>;
3007 // Neon Narrowing 3-register vector intrinsics,
3008 // source operand element sizes of 16, 32 and 64 bits:
3009 multiclass N3VNInt_HSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3010 string OpcodeStr, string Dt,
3011 Intrinsic IntOp, bit Commutable = 0> {
3012 def v8i8 : N3VNInt<op24, op23, 0b00, op11_8, op4,
3013 OpcodeStr, !strconcat(Dt, "16"),
3014 v8i8, v8i16, IntOp, Commutable>;
3015 def v4i16 : N3VNInt<op24, op23, 0b01, op11_8, op4,
3016 OpcodeStr, !strconcat(Dt, "32"),
3017 v4i16, v4i32, IntOp, Commutable>;
3018 def v2i32 : N3VNInt<op24, op23, 0b10, op11_8, op4,
3019 OpcodeStr, !strconcat(Dt, "64"),
3020 v2i32, v2i64, IntOp, Commutable>;
3024 // Neon Long 3-register vector operations.
3026 multiclass N3VL_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3027 InstrItinClass itin16, InstrItinClass itin32,
3028 string OpcodeStr, string Dt,
3029 SDNode OpNode, bit Commutable = 0> {
3030 def v8i16 : N3VL<op24, op23, 0b00, op11_8, op4, itin16,
3031 OpcodeStr, !strconcat(Dt, "8"),
3032 v8i16, v8i8, OpNode, Commutable>;
3033 def v4i32 : N3VL<op24, op23, 0b01, op11_8, op4, itin16,
3034 OpcodeStr, !strconcat(Dt, "16"),
3035 v4i32, v4i16, OpNode, Commutable>;
3036 def v2i64 : N3VL<op24, op23, 0b10, op11_8, op4, itin32,
3037 OpcodeStr, !strconcat(Dt, "32"),
3038 v2i64, v2i32, OpNode, Commutable>;
3041 multiclass N3VLSL_HS<bit op24, bits<4> op11_8,
3042 InstrItinClass itin, string OpcodeStr, string Dt,
3044 def v4i16 : N3VLSL16<op24, 0b01, op11_8, itin, OpcodeStr,
3045 !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
3046 def v2i32 : N3VLSL<op24, 0b10, op11_8, itin, OpcodeStr,
3047 !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
3050 multiclass N3VLExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3051 InstrItinClass itin16, InstrItinClass itin32,
3052 string OpcodeStr, string Dt,
3053 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
3054 def v8i16 : N3VLExt<op24, op23, 0b00, op11_8, op4, itin16,
3055 OpcodeStr, !strconcat(Dt, "8"),
3056 v8i16, v8i8, OpNode, ExtOp, Commutable>;
3057 def v4i32 : N3VLExt<op24, op23, 0b01, op11_8, op4, itin16,
3058 OpcodeStr, !strconcat(Dt, "16"),
3059 v4i32, v4i16, OpNode, ExtOp, Commutable>;
3060 def v2i64 : N3VLExt<op24, op23, 0b10, op11_8, op4, itin32,
3061 OpcodeStr, !strconcat(Dt, "32"),
3062 v2i64, v2i32, OpNode, ExtOp, Commutable>;
3065 // Neon Long 3-register vector intrinsics.
3067 // First with only element sizes of 16 and 32 bits:
3068 multiclass N3VLInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
3069 InstrItinClass itin16, InstrItinClass itin32,
3070 string OpcodeStr, string Dt,
3071 Intrinsic IntOp, bit Commutable = 0> {
3072 def v4i32 : N3VLInt<op24, op23, 0b01, op11_8, op4, itin16,
3073 OpcodeStr, !strconcat(Dt, "16"),
3074 v4i32, v4i16, IntOp, Commutable>;
3075 def v2i64 : N3VLInt<op24, op23, 0b10, op11_8, op4, itin32,
3076 OpcodeStr, !strconcat(Dt, "32"),
3077 v2i64, v2i32, IntOp, Commutable>;
3080 multiclass N3VLIntSL_HS<bit op24, bits<4> op11_8,
3081 InstrItinClass itin, string OpcodeStr, string Dt,
3083 def v4i16 : N3VLIntSL16<op24, 0b01, op11_8, itin,
3084 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
3085 def v2i32 : N3VLIntSL<op24, 0b10, op11_8, itin,
3086 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3089 // ....then also with element size of 8 bits:
3090 multiclass N3VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3091 InstrItinClass itin16, InstrItinClass itin32,
3092 string OpcodeStr, string Dt,
3093 Intrinsic IntOp, bit Commutable = 0>
3094 : N3VLInt_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt,
3095 IntOp, Commutable> {
3096 def v8i16 : N3VLInt<op24, op23, 0b00, op11_8, op4, itin16,
3097 OpcodeStr, !strconcat(Dt, "8"),
3098 v8i16, v8i8, IntOp, Commutable>;
3101 // ....with explicit extend (VABDL).
3102 multiclass N3VLIntExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3103 InstrItinClass itin, string OpcodeStr, string Dt,
3104 Intrinsic IntOp, SDNode ExtOp, bit Commutable = 0> {
3105 def v8i16 : N3VLIntExt<op24, op23, 0b00, op11_8, op4, itin,
3106 OpcodeStr, !strconcat(Dt, "8"),
3107 v8i16, v8i8, IntOp, ExtOp, Commutable>;
3108 def v4i32 : N3VLIntExt<op24, op23, 0b01, op11_8, op4, itin,
3109 OpcodeStr, !strconcat(Dt, "16"),
3110 v4i32, v4i16, IntOp, ExtOp, Commutable>;
3111 def v2i64 : N3VLIntExt<op24, op23, 0b10, op11_8, op4, itin,
3112 OpcodeStr, !strconcat(Dt, "32"),
3113 v2i64, v2i32, IntOp, ExtOp, Commutable>;
3117 // Neon Wide 3-register vector intrinsics,
3118 // source operand element sizes of 8, 16 and 32 bits:
3119 multiclass N3VW_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3120 string OpcodeStr, string Dt,
3121 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
3122 def v8i16 : N3VW<op24, op23, 0b00, op11_8, op4,
3123 OpcodeStr, !strconcat(Dt, "8"),
3124 v8i16, v8i8, OpNode, ExtOp, Commutable>;
3125 def v4i32 : N3VW<op24, op23, 0b01, op11_8, op4,
3126 OpcodeStr, !strconcat(Dt, "16"),
3127 v4i32, v4i16, OpNode, ExtOp, Commutable>;
3128 def v2i64 : N3VW<op24, op23, 0b10, op11_8, op4,
3129 OpcodeStr, !strconcat(Dt, "32"),
3130 v2i64, v2i32, OpNode, ExtOp, Commutable>;
3134 // Neon Multiply-Op vector operations,
3135 // element sizes of 8, 16 and 32 bits:
3136 multiclass N3VMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3137 InstrItinClass itinD16, InstrItinClass itinD32,
3138 InstrItinClass itinQ16, InstrItinClass itinQ32,
3139 string OpcodeStr, string Dt, SDNode OpNode> {
3140 // 64-bit vector types.
3141 def v8i8 : N3VDMulOp<op24, op23, 0b00, op11_8, op4, itinD16,
3142 OpcodeStr, !strconcat(Dt, "8"), v8i8, mul, OpNode>;
3143 def v4i16 : N3VDMulOp<op24, op23, 0b01, op11_8, op4, itinD16,
3144 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, OpNode>;
3145 def v2i32 : N3VDMulOp<op24, op23, 0b10, op11_8, op4, itinD32,
3146 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, OpNode>;
3148 // 128-bit vector types.
3149 def v16i8 : N3VQMulOp<op24, op23, 0b00, op11_8, op4, itinQ16,
3150 OpcodeStr, !strconcat(Dt, "8"), v16i8, mul, OpNode>;
3151 def v8i16 : N3VQMulOp<op24, op23, 0b01, op11_8, op4, itinQ16,
3152 OpcodeStr, !strconcat(Dt, "16"), v8i16, mul, OpNode>;
3153 def v4i32 : N3VQMulOp<op24, op23, 0b10, op11_8, op4, itinQ32,
3154 OpcodeStr, !strconcat(Dt, "32"), v4i32, mul, OpNode>;
3157 multiclass N3VMulOpSL_HS<bits<4> op11_8,
3158 InstrItinClass itinD16, InstrItinClass itinD32,
3159 InstrItinClass itinQ16, InstrItinClass itinQ32,
3160 string OpcodeStr, string Dt, SDNode ShOp> {
3161 def v4i16 : N3VDMulOpSL16<0b01, op11_8, itinD16,
3162 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, ShOp>;
3163 def v2i32 : N3VDMulOpSL<0b10, op11_8, itinD32,
3164 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, ShOp>;
3165 def v8i16 : N3VQMulOpSL16<0b01, op11_8, itinQ16,
3166 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16,
3168 def v4i32 : N3VQMulOpSL<0b10, op11_8, itinQ32,
3169 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32,
3173 // Neon Intrinsic-Op vector operations,
3174 // element sizes of 8, 16 and 32 bits:
3175 multiclass N3VIntOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3176 InstrItinClass itinD, InstrItinClass itinQ,
3177 string OpcodeStr, string Dt, Intrinsic IntOp,
3179 // 64-bit vector types.
3180 def v8i8 : N3VDIntOp<op24, op23, 0b00, op11_8, op4, itinD,
3181 OpcodeStr, !strconcat(Dt, "8"), v8i8, IntOp, OpNode>;
3182 def v4i16 : N3VDIntOp<op24, op23, 0b01, op11_8, op4, itinD,
3183 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp, OpNode>;
3184 def v2i32 : N3VDIntOp<op24, op23, 0b10, op11_8, op4, itinD,
3185 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp, OpNode>;
3187 // 128-bit vector types.
3188 def v16i8 : N3VQIntOp<op24, op23, 0b00, op11_8, op4, itinQ,
3189 OpcodeStr, !strconcat(Dt, "8"), v16i8, IntOp, OpNode>;
3190 def v8i16 : N3VQIntOp<op24, op23, 0b01, op11_8, op4, itinQ,
3191 OpcodeStr, !strconcat(Dt, "16"), v8i16, IntOp, OpNode>;
3192 def v4i32 : N3VQIntOp<op24, op23, 0b10, op11_8, op4, itinQ,
3193 OpcodeStr, !strconcat(Dt, "32"), v4i32, IntOp, OpNode>;
3196 // Neon 3-argument intrinsics,
3197 // element sizes of 8, 16 and 32 bits:
3198 multiclass N3VInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3199 InstrItinClass itinD, InstrItinClass itinQ,
3200 string OpcodeStr, string Dt, Intrinsic IntOp> {
3201 // 64-bit vector types.
3202 def v8i8 : N3VDInt3<op24, op23, 0b00, op11_8, op4, itinD,
3203 OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
3204 def v4i16 : N3VDInt3<op24, op23, 0b01, op11_8, op4, itinD,
3205 OpcodeStr, !strconcat(Dt, "16"), v4i16, v4i16, IntOp>;
3206 def v2i32 : N3VDInt3<op24, op23, 0b10, op11_8, op4, itinD,
3207 OpcodeStr, !strconcat(Dt, "32"), v2i32, v2i32, IntOp>;
3209 // 128-bit vector types.
3210 def v16i8 : N3VQInt3<op24, op23, 0b00, op11_8, op4, itinQ,
3211 OpcodeStr, !strconcat(Dt, "8"), v16i8, v16i8, IntOp>;
3212 def v8i16 : N3VQInt3<op24, op23, 0b01, op11_8, op4, itinQ,
3213 OpcodeStr, !strconcat(Dt, "16"), v8i16, v8i16, IntOp>;
3214 def v4i32 : N3VQInt3<op24, op23, 0b10, op11_8, op4, itinQ,
3215 OpcodeStr, !strconcat(Dt, "32"), v4i32, v4i32, IntOp>;
3219 // Neon Long Multiply-Op vector operations,
3220 // element sizes of 8, 16 and 32 bits:
3221 multiclass N3VLMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3222 InstrItinClass itin16, InstrItinClass itin32,
3223 string OpcodeStr, string Dt, SDNode MulOp,
3225 def v8i16 : N3VLMulOp<op24, op23, 0b00, op11_8, op4, itin16, OpcodeStr,
3226 !strconcat(Dt, "8"), v8i16, v8i8, MulOp, OpNode>;
3227 def v4i32 : N3VLMulOp<op24, op23, 0b01, op11_8, op4, itin16, OpcodeStr,
3228 !strconcat(Dt, "16"), v4i32, v4i16, MulOp, OpNode>;
3229 def v2i64 : N3VLMulOp<op24, op23, 0b10, op11_8, op4, itin32, OpcodeStr,
3230 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
3233 multiclass N3VLMulOpSL_HS<bit op24, bits<4> op11_8, string OpcodeStr,
3234 string Dt, SDNode MulOp, SDNode OpNode> {
3235 def v4i16 : N3VLMulOpSL16<op24, 0b01, op11_8, IIC_VMACi16D, OpcodeStr,
3236 !strconcat(Dt,"16"), v4i32, v4i16, MulOp, OpNode>;
3237 def v2i32 : N3VLMulOpSL<op24, 0b10, op11_8, IIC_VMACi32D, OpcodeStr,
3238 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
3242 // Neon Long 3-argument intrinsics.
3244 // First with only element sizes of 16 and 32 bits:
3245 multiclass N3VLInt3_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
3246 InstrItinClass itin16, InstrItinClass itin32,
3247 string OpcodeStr, string Dt, Intrinsic IntOp> {
3248 def v4i32 : N3VLInt3<op24, op23, 0b01, op11_8, op4, itin16,
3249 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
3250 def v2i64 : N3VLInt3<op24, op23, 0b10, op11_8, op4, itin32,
3251 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3254 multiclass N3VLInt3SL_HS<bit op24, bits<4> op11_8,
3255 string OpcodeStr, string Dt, Intrinsic IntOp> {
3256 def v4i16 : N3VLInt3SL16<op24, 0b01, op11_8, IIC_VMACi16D,
3257 OpcodeStr, !strconcat(Dt,"16"), v4i32, v4i16, IntOp>;
3258 def v2i32 : N3VLInt3SL<op24, 0b10, op11_8, IIC_VMACi32D,
3259 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3262 // ....then also with element size of 8 bits:
3263 multiclass N3VLInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3264 InstrItinClass itin16, InstrItinClass itin32,
3265 string OpcodeStr, string Dt, Intrinsic IntOp>
3266 : N3VLInt3_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt, IntOp> {
3267 def v8i16 : N3VLInt3<op24, op23, 0b00, op11_8, op4, itin16,
3268 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
3271 // ....with explicit extend (VABAL).
3272 multiclass N3VLIntExtOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3273 InstrItinClass itin, string OpcodeStr, string Dt,
3274 Intrinsic IntOp, SDNode ExtOp, SDNode OpNode> {
3275 def v8i16 : N3VLIntExtOp<op24, op23, 0b00, op11_8, op4, itin,
3276 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8,
3277 IntOp, ExtOp, OpNode>;
3278 def v4i32 : N3VLIntExtOp<op24, op23, 0b01, op11_8, op4, itin,
3279 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16,
3280 IntOp, ExtOp, OpNode>;
3281 def v2i64 : N3VLIntExtOp<op24, op23, 0b10, op11_8, op4, itin,
3282 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32,
3283 IntOp, ExtOp, OpNode>;
3287 // Neon Pairwise long 2-register intrinsics,
3288 // element sizes of 8, 16 and 32 bits:
3289 multiclass N2VPLInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3290 bits<5> op11_7, bit op4,
3291 string OpcodeStr, string Dt, Intrinsic IntOp> {
3292 // 64-bit vector types.
3293 def v8i8 : N2VDPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3294 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
3295 def v4i16 : N2VDPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3296 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
3297 def v2i32 : N2VDPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3298 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
3300 // 128-bit vector types.
3301 def v16i8 : N2VQPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3302 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
3303 def v8i16 : N2VQPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3304 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
3305 def v4i32 : N2VQPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3306 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
3310 // Neon Pairwise long 2-register accumulate intrinsics,
3311 // element sizes of 8, 16 and 32 bits:
3312 multiclass N2VPLInt2_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3313 bits<5> op11_7, bit op4,
3314 string OpcodeStr, string Dt, Intrinsic IntOp> {
3315 // 64-bit vector types.
3316 def v8i8 : N2VDPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3317 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
3318 def v4i16 : N2VDPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3319 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
3320 def v2i32 : N2VDPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3321 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
3323 // 128-bit vector types.
3324 def v16i8 : N2VQPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3325 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
3326 def v8i16 : N2VQPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3327 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
3328 def v4i32 : N2VQPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3329 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
3333 // Neon 2-register vector shift by immediate,
3334 // with f of either N2RegVShLFrm or N2RegVShRFrm
3335 // element sizes of 8, 16, 32 and 64 bits:
3336 multiclass N2VShL_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3337 InstrItinClass itin, string OpcodeStr, string Dt,
3339 // 64-bit vector types.
3340 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3341 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
3342 let Inst{21-19} = 0b001; // imm6 = 001xxx
3344 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3345 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
3346 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3348 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3349 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
3350 let Inst{21} = 0b1; // imm6 = 1xxxxx
3352 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, N2RegVShLFrm, itin, i32imm,
3353 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
3356 // 128-bit vector types.
3357 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3358 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
3359 let Inst{21-19} = 0b001; // imm6 = 001xxx
3361 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3362 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
3363 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3365 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3366 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
3367 let Inst{21} = 0b1; // imm6 = 1xxxxx
3369 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, N2RegVShLFrm, itin, i32imm,
3370 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
3373 multiclass N2VShR_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3374 InstrItinClass itin, string OpcodeStr, string Dt,
3376 // 64-bit vector types.
3377 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm8,
3378 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
3379 let Inst{21-19} = 0b001; // imm6 = 001xxx
3381 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm16,
3382 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
3383 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3385 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm32,
3386 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
3387 let Inst{21} = 0b1; // imm6 = 1xxxxx
3389 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, N2RegVShRFrm, itin, shr_imm64,
3390 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
3393 // 128-bit vector types.
3394 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm8,
3395 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
3396 let Inst{21-19} = 0b001; // imm6 = 001xxx
3398 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm16,
3399 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
3400 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3402 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm32,
3403 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
3404 let Inst{21} = 0b1; // imm6 = 1xxxxx
3406 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, N2RegVShRFrm, itin, shr_imm64,
3407 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
3411 // Neon Shift-Accumulate vector operations,
3412 // element sizes of 8, 16, 32 and 64 bits:
3413 multiclass N2VShAdd_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3414 string OpcodeStr, string Dt, SDNode ShOp> {
3415 // 64-bit vector types.
3416 def v8i8 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm8,
3417 OpcodeStr, !strconcat(Dt, "8"), v8i8, ShOp> {
3418 let Inst{21-19} = 0b001; // imm6 = 001xxx
3420 def v4i16 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm16,
3421 OpcodeStr, !strconcat(Dt, "16"), v4i16, ShOp> {
3422 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3424 def v2i32 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm32,
3425 OpcodeStr, !strconcat(Dt, "32"), v2i32, ShOp> {
3426 let Inst{21} = 0b1; // imm6 = 1xxxxx
3428 def v1i64 : N2VDShAdd<op24, op23, op11_8, 1, op4, shr_imm64,
3429 OpcodeStr, !strconcat(Dt, "64"), v1i64, ShOp>;
3432 // 128-bit vector types.
3433 def v16i8 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm8,
3434 OpcodeStr, !strconcat(Dt, "8"), v16i8, ShOp> {
3435 let Inst{21-19} = 0b001; // imm6 = 001xxx
3437 def v8i16 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm16,
3438 OpcodeStr, !strconcat(Dt, "16"), v8i16, ShOp> {
3439 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3441 def v4i32 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm32,
3442 OpcodeStr, !strconcat(Dt, "32"), v4i32, ShOp> {
3443 let Inst{21} = 0b1; // imm6 = 1xxxxx
3445 def v2i64 : N2VQShAdd<op24, op23, op11_8, 1, op4, shr_imm64,
3446 OpcodeStr, !strconcat(Dt, "64"), v2i64, ShOp>;
3450 // Neon Shift-Insert vector operations,
3451 // with f of either N2RegVShLFrm or N2RegVShRFrm
3452 // element sizes of 8, 16, 32 and 64 bits:
3453 multiclass N2VShInsL_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3455 // 64-bit vector types.
3456 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3457 N2RegVShLFrm, OpcodeStr, "8", v8i8, NEONvsli> {
3458 let Inst{21-19} = 0b001; // imm6 = 001xxx
3460 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3461 N2RegVShLFrm, OpcodeStr, "16", v4i16, NEONvsli> {
3462 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3464 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3465 N2RegVShLFrm, OpcodeStr, "32", v2i32, NEONvsli> {
3466 let Inst{21} = 0b1; // imm6 = 1xxxxx
3468 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4, i32imm,
3469 N2RegVShLFrm, OpcodeStr, "64", v1i64, NEONvsli>;
3472 // 128-bit vector types.
3473 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3474 N2RegVShLFrm, OpcodeStr, "8", v16i8, NEONvsli> {
3475 let Inst{21-19} = 0b001; // imm6 = 001xxx
3477 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3478 N2RegVShLFrm, OpcodeStr, "16", v8i16, NEONvsli> {
3479 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3481 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3482 N2RegVShLFrm, OpcodeStr, "32", v4i32, NEONvsli> {
3483 let Inst{21} = 0b1; // imm6 = 1xxxxx
3485 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4, i32imm,
3486 N2RegVShLFrm, OpcodeStr, "64", v2i64, NEONvsli>;
3489 multiclass N2VShInsR_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3491 // 64-bit vector types.
3492 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm8,
3493 N2RegVShRFrm, OpcodeStr, "8", v8i8, NEONvsri> {
3494 let Inst{21-19} = 0b001; // imm6 = 001xxx
3496 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm16,
3497 N2RegVShRFrm, OpcodeStr, "16", v4i16, NEONvsri> {
3498 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3500 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm32,
3501 N2RegVShRFrm, OpcodeStr, "32", v2i32, NEONvsri> {
3502 let Inst{21} = 0b1; // imm6 = 1xxxxx
3504 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4, shr_imm64,
3505 N2RegVShRFrm, OpcodeStr, "64", v1i64, NEONvsri>;
3508 // 128-bit vector types.
3509 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm8,
3510 N2RegVShRFrm, OpcodeStr, "8", v16i8, NEONvsri> {
3511 let Inst{21-19} = 0b001; // imm6 = 001xxx
3513 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm16,
3514 N2RegVShRFrm, OpcodeStr, "16", v8i16, NEONvsri> {
3515 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3517 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm32,
3518 N2RegVShRFrm, OpcodeStr, "32", v4i32, NEONvsri> {
3519 let Inst{21} = 0b1; // imm6 = 1xxxxx
3521 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4, shr_imm64,
3522 N2RegVShRFrm, OpcodeStr, "64", v2i64, NEONvsri>;
3526 // Neon Shift Long operations,
3527 // element sizes of 8, 16, 32 bits:
3528 multiclass N2VLSh_QHS<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
3529 bit op4, string OpcodeStr, string Dt, SDNode OpNode> {
3530 def v8i16 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3531 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode> {
3532 let Inst{21-19} = 0b001; // imm6 = 001xxx
3534 def v4i32 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3535 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode> {
3536 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3538 def v2i64 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3539 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode> {
3540 let Inst{21} = 0b1; // imm6 = 1xxxxx
3544 // Neon Shift Narrow operations,
3545 // element sizes of 16, 32, 64 bits:
3546 multiclass N2VNSh_HSD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
3547 bit op4, InstrItinClass itin, string OpcodeStr, string Dt,
3549 def v8i8 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3550 OpcodeStr, !strconcat(Dt, "16"),
3551 v8i8, v8i16, shr_imm8, OpNode> {
3552 let Inst{21-19} = 0b001; // imm6 = 001xxx
3554 def v4i16 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3555 OpcodeStr, !strconcat(Dt, "32"),
3556 v4i16, v4i32, shr_imm16, OpNode> {
3557 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3559 def v2i32 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3560 OpcodeStr, !strconcat(Dt, "64"),
3561 v2i32, v2i64, shr_imm32, OpNode> {
3562 let Inst{21} = 0b1; // imm6 = 1xxxxx
3566 //===----------------------------------------------------------------------===//
3567 // Instruction Definitions.
3568 //===----------------------------------------------------------------------===//
3570 // Vector Add Operations.
3572 // VADD : Vector Add (integer and floating-point)
3573 defm VADD : N3V_QHSD<0, 0, 0b1000, 0, IIC_VBINiD, IIC_VBINiQ, "vadd", "i",
3575 def VADDfd : N3VD<0, 0, 0b00, 0b1101, 0, IIC_VBIND, "vadd", "f32",
3576 v2f32, v2f32, fadd, 1>;
3577 def VADDfq : N3VQ<0, 0, 0b00, 0b1101, 0, IIC_VBINQ, "vadd", "f32",
3578 v4f32, v4f32, fadd, 1>;
3579 // VADDL : Vector Add Long (Q = D + D)
3580 defm VADDLs : N3VLExt_QHS<0,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
3581 "vaddl", "s", add, sext, 1>;
3582 defm VADDLu : N3VLExt_QHS<1,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
3583 "vaddl", "u", add, zext, 1>;
3584 // VADDW : Vector Add Wide (Q = Q + D)
3585 defm VADDWs : N3VW_QHS<0,1,0b0001,0, "vaddw", "s", add, sext, 0>;
3586 defm VADDWu : N3VW_QHS<1,1,0b0001,0, "vaddw", "u", add, zext, 0>;
3587 // VHADD : Vector Halving Add
3588 defm VHADDs : N3VInt_QHS<0, 0, 0b0000, 0, N3RegFrm,
3589 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3590 "vhadd", "s", int_arm_neon_vhadds, 1>;
3591 defm VHADDu : N3VInt_QHS<1, 0, 0b0000, 0, N3RegFrm,
3592 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3593 "vhadd", "u", int_arm_neon_vhaddu, 1>;
3594 // VRHADD : Vector Rounding Halving Add
3595 defm VRHADDs : N3VInt_QHS<0, 0, 0b0001, 0, N3RegFrm,
3596 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3597 "vrhadd", "s", int_arm_neon_vrhadds, 1>;
3598 defm VRHADDu : N3VInt_QHS<1, 0, 0b0001, 0, N3RegFrm,
3599 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3600 "vrhadd", "u", int_arm_neon_vrhaddu, 1>;
3601 // VQADD : Vector Saturating Add
3602 defm VQADDs : N3VInt_QHSD<0, 0, 0b0000, 1, N3RegFrm,
3603 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3604 "vqadd", "s", int_arm_neon_vqadds, 1>;
3605 defm VQADDu : N3VInt_QHSD<1, 0, 0b0000, 1, N3RegFrm,
3606 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3607 "vqadd", "u", int_arm_neon_vqaddu, 1>;
3608 // VADDHN : Vector Add and Narrow Returning High Half (D = Q + Q)
3609 defm VADDHN : N3VNInt_HSD<0,1,0b0100,0, "vaddhn", "i",
3610 int_arm_neon_vaddhn, 1>;
3611 // VRADDHN : Vector Rounding Add and Narrow Returning High Half (D = Q + Q)
3612 defm VRADDHN : N3VNInt_HSD<1,1,0b0100,0, "vraddhn", "i",
3613 int_arm_neon_vraddhn, 1>;
3615 // Vector Multiply Operations.
3617 // VMUL : Vector Multiply (integer, polynomial and floating-point)
3618 defm VMUL : N3V_QHS<0, 0, 0b1001, 1, IIC_VMULi16D, IIC_VMULi32D,
3619 IIC_VMULi16Q, IIC_VMULi32Q, "vmul", "i", mul, 1>;
3620 def VMULpd : N3VDInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16D, "vmul",
3621 "p8", v8i8, v8i8, int_arm_neon_vmulp, 1>;
3622 def VMULpq : N3VQInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16Q, "vmul",
3623 "p8", v16i8, v16i8, int_arm_neon_vmulp, 1>;
3624 def VMULfd : N3VD<1, 0, 0b00, 0b1101, 1, IIC_VFMULD, "vmul", "f32",
3625 v2f32, v2f32, fmul, 1>;
3626 def VMULfq : N3VQ<1, 0, 0b00, 0b1101, 1, IIC_VFMULQ, "vmul", "f32",
3627 v4f32, v4f32, fmul, 1>;
3628 defm VMULsl : N3VSL_HS<0b1000, "vmul", "i", mul>;
3629 def VMULslfd : N3VDSL<0b10, 0b1001, IIC_VBIND, "vmul", "f32", v2f32, fmul>;
3630 def VMULslfq : N3VQSL<0b10, 0b1001, IIC_VBINQ, "vmul", "f32", v4f32,
3633 def : Pat<(v8i16 (mul (v8i16 QPR:$src1),
3634 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
3635 (v8i16 (VMULslv8i16 (v8i16 QPR:$src1),
3636 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3637 (DSubReg_i16_reg imm:$lane))),
3638 (SubReg_i16_lane imm:$lane)))>;
3639 def : Pat<(v4i32 (mul (v4i32 QPR:$src1),
3640 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
3641 (v4i32 (VMULslv4i32 (v4i32 QPR:$src1),
3642 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3643 (DSubReg_i32_reg imm:$lane))),
3644 (SubReg_i32_lane imm:$lane)))>;
3645 def : Pat<(v4f32 (fmul (v4f32 QPR:$src1),
3646 (v4f32 (NEONvduplane (v4f32 QPR:$src2), imm:$lane)))),
3647 (v4f32 (VMULslfq (v4f32 QPR:$src1),
3648 (v2f32 (EXTRACT_SUBREG QPR:$src2,
3649 (DSubReg_i32_reg imm:$lane))),
3650 (SubReg_i32_lane imm:$lane)))>;
3652 // VQDMULH : Vector Saturating Doubling Multiply Returning High Half
3653 defm VQDMULH : N3VInt_HS<0, 0, 0b1011, 0, N3RegFrm, IIC_VMULi16D, IIC_VMULi32D,
3654 IIC_VMULi16Q, IIC_VMULi32Q,
3655 "vqdmulh", "s", int_arm_neon_vqdmulh, 1>;
3656 defm VQDMULHsl: N3VIntSL_HS<0b1100, IIC_VMULi16D, IIC_VMULi32D,
3657 IIC_VMULi16Q, IIC_VMULi32Q,
3658 "vqdmulh", "s", int_arm_neon_vqdmulh>;
3659 def : Pat<(v8i16 (int_arm_neon_vqdmulh (v8i16 QPR:$src1),
3660 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
3662 (v8i16 (VQDMULHslv8i16 (v8i16 QPR:$src1),
3663 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3664 (DSubReg_i16_reg imm:$lane))),
3665 (SubReg_i16_lane imm:$lane)))>;
3666 def : Pat<(v4i32 (int_arm_neon_vqdmulh (v4i32 QPR:$src1),
3667 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
3669 (v4i32 (VQDMULHslv4i32 (v4i32 QPR:$src1),
3670 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3671 (DSubReg_i32_reg imm:$lane))),
3672 (SubReg_i32_lane imm:$lane)))>;
3674 // VQRDMULH : Vector Rounding Saturating Doubling Multiply Returning High Half
3675 defm VQRDMULH : N3VInt_HS<1, 0, 0b1011, 0, N3RegFrm,
3676 IIC_VMULi16D,IIC_VMULi32D,IIC_VMULi16Q,IIC_VMULi32Q,
3677 "vqrdmulh", "s", int_arm_neon_vqrdmulh, 1>;
3678 defm VQRDMULHsl : N3VIntSL_HS<0b1101, IIC_VMULi16D, IIC_VMULi32D,
3679 IIC_VMULi16Q, IIC_VMULi32Q,
3680 "vqrdmulh", "s", int_arm_neon_vqrdmulh>;
3681 def : Pat<(v8i16 (int_arm_neon_vqrdmulh (v8i16 QPR:$src1),
3682 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
3684 (v8i16 (VQRDMULHslv8i16 (v8i16 QPR:$src1),
3685 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3686 (DSubReg_i16_reg imm:$lane))),
3687 (SubReg_i16_lane imm:$lane)))>;
3688 def : Pat<(v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$src1),
3689 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
3691 (v4i32 (VQRDMULHslv4i32 (v4i32 QPR:$src1),
3692 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3693 (DSubReg_i32_reg imm:$lane))),
3694 (SubReg_i32_lane imm:$lane)))>;
3696 // VMULL : Vector Multiply Long (integer and polynomial) (Q = D * D)
3697 defm VMULLs : N3VL_QHS<0,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
3698 "vmull", "s", NEONvmulls, 1>;
3699 defm VMULLu : N3VL_QHS<1,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
3700 "vmull", "u", NEONvmullu, 1>;
3701 def VMULLp : N3VLInt<0, 1, 0b00, 0b1110, 0, IIC_VMULi16D, "vmull", "p8",
3702 v8i16, v8i8, int_arm_neon_vmullp, 1>;
3703 defm VMULLsls : N3VLSL_HS<0, 0b1010, IIC_VMULi16D, "vmull", "s", NEONvmulls>;
3704 defm VMULLslu : N3VLSL_HS<1, 0b1010, IIC_VMULi16D, "vmull", "u", NEONvmullu>;
3706 // VQDMULL : Vector Saturating Doubling Multiply Long (Q = D * D)
3707 defm VQDMULL : N3VLInt_HS<0,1,0b1101,0, IIC_VMULi16D, IIC_VMULi32D,
3708 "vqdmull", "s", int_arm_neon_vqdmull, 1>;
3709 defm VQDMULLsl: N3VLIntSL_HS<0, 0b1011, IIC_VMULi16D,
3710 "vqdmull", "s", int_arm_neon_vqdmull>;
3712 // Vector Multiply-Accumulate and Multiply-Subtract Operations.
3714 // VMLA : Vector Multiply Accumulate (integer and floating-point)
3715 defm VMLA : N3VMulOp_QHS<0, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3716 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
3717 def VMLAfd : N3VDMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACD, "vmla", "f32",
3718 v2f32, fmul_su, fadd_mlx>,
3719 Requires<[HasNEON, UseFPVMLx]>;
3720 def VMLAfq : N3VQMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACQ, "vmla", "f32",
3721 v4f32, fmul_su, fadd_mlx>,
3722 Requires<[HasNEON, UseFPVMLx]>;
3723 defm VMLAsl : N3VMulOpSL_HS<0b0000, IIC_VMACi16D, IIC_VMACi32D,
3724 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
3725 def VMLAslfd : N3VDMulOpSL<0b10, 0b0001, IIC_VMACD, "vmla", "f32",
3726 v2f32, fmul_su, fadd_mlx>,
3727 Requires<[HasNEON, UseFPVMLx]>;
3728 def VMLAslfq : N3VQMulOpSL<0b10, 0b0001, IIC_VMACQ, "vmla", "f32",
3729 v4f32, v2f32, fmul_su, fadd_mlx>,
3730 Requires<[HasNEON, UseFPVMLx]>;
3732 def : Pat<(v8i16 (add (v8i16 QPR:$src1),
3733 (mul (v8i16 QPR:$src2),
3734 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
3735 (v8i16 (VMLAslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
3736 (v4i16 (EXTRACT_SUBREG QPR:$src3,
3737 (DSubReg_i16_reg imm:$lane))),
3738 (SubReg_i16_lane imm:$lane)))>;
3740 def : Pat<(v4i32 (add (v4i32 QPR:$src1),
3741 (mul (v4i32 QPR:$src2),
3742 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
3743 (v4i32 (VMLAslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
3744 (v2i32 (EXTRACT_SUBREG QPR:$src3,
3745 (DSubReg_i32_reg imm:$lane))),
3746 (SubReg_i32_lane imm:$lane)))>;
3748 def : Pat<(v4f32 (fadd_mlx (v4f32 QPR:$src1),
3749 (fmul_su (v4f32 QPR:$src2),
3750 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
3751 (v4f32 (VMLAslfq (v4f32 QPR:$src1),
3753 (v2f32 (EXTRACT_SUBREG QPR:$src3,
3754 (DSubReg_i32_reg imm:$lane))),
3755 (SubReg_i32_lane imm:$lane)))>,
3756 Requires<[HasNEON, UseFPVMLx]>;
3758 // VMLAL : Vector Multiply Accumulate Long (Q += D * D)
3759 defm VMLALs : N3VLMulOp_QHS<0,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
3760 "vmlal", "s", NEONvmulls, add>;
3761 defm VMLALu : N3VLMulOp_QHS<1,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
3762 "vmlal", "u", NEONvmullu, add>;
3764 defm VMLALsls : N3VLMulOpSL_HS<0, 0b0010, "vmlal", "s", NEONvmulls, add>;
3765 defm VMLALslu : N3VLMulOpSL_HS<1, 0b0010, "vmlal", "u", NEONvmullu, add>;
3767 // VQDMLAL : Vector Saturating Doubling Multiply Accumulate Long (Q += D * D)
3768 defm VQDMLAL : N3VLInt3_HS<0, 1, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3769 "vqdmlal", "s", int_arm_neon_vqdmlal>;
3770 defm VQDMLALsl: N3VLInt3SL_HS<0, 0b0011, "vqdmlal", "s", int_arm_neon_vqdmlal>;
3772 // VMLS : Vector Multiply Subtract (integer and floating-point)
3773 defm VMLS : N3VMulOp_QHS<1, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3774 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
3775 def VMLSfd : N3VDMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACD, "vmls", "f32",
3776 v2f32, fmul_su, fsub_mlx>,
3777 Requires<[HasNEON, UseFPVMLx]>;
3778 def VMLSfq : N3VQMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACQ, "vmls", "f32",
3779 v4f32, fmul_su, fsub_mlx>,
3780 Requires<[HasNEON, UseFPVMLx]>;
3781 defm VMLSsl : N3VMulOpSL_HS<0b0100, IIC_VMACi16D, IIC_VMACi32D,
3782 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
3783 def VMLSslfd : N3VDMulOpSL<0b10, 0b0101, IIC_VMACD, "vmls", "f32",
3784 v2f32, fmul_su, fsub_mlx>,
3785 Requires<[HasNEON, UseFPVMLx]>;
3786 def VMLSslfq : N3VQMulOpSL<0b10, 0b0101, IIC_VMACQ, "vmls", "f32",
3787 v4f32, v2f32, fmul_su, fsub_mlx>,
3788 Requires<[HasNEON, UseFPVMLx]>;
3790 def : Pat<(v8i16 (sub (v8i16 QPR:$src1),
3791 (mul (v8i16 QPR:$src2),
3792 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
3793 (v8i16 (VMLSslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
3794 (v4i16 (EXTRACT_SUBREG QPR:$src3,
3795 (DSubReg_i16_reg imm:$lane))),
3796 (SubReg_i16_lane imm:$lane)))>;
3798 def : Pat<(v4i32 (sub (v4i32 QPR:$src1),
3799 (mul (v4i32 QPR:$src2),
3800 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
3801 (v4i32 (VMLSslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
3802 (v2i32 (EXTRACT_SUBREG QPR:$src3,
3803 (DSubReg_i32_reg imm:$lane))),
3804 (SubReg_i32_lane imm:$lane)))>;
3806 def : Pat<(v4f32 (fsub_mlx (v4f32 QPR:$src1),
3807 (fmul_su (v4f32 QPR:$src2),
3808 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
3809 (v4f32 (VMLSslfq (v4f32 QPR:$src1), (v4f32 QPR:$src2),
3810 (v2f32 (EXTRACT_SUBREG QPR:$src3,
3811 (DSubReg_i32_reg imm:$lane))),
3812 (SubReg_i32_lane imm:$lane)))>,
3813 Requires<[HasNEON, UseFPVMLx]>;
3815 // VMLSL : Vector Multiply Subtract Long (Q -= D * D)
3816 defm VMLSLs : N3VLMulOp_QHS<0,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
3817 "vmlsl", "s", NEONvmulls, sub>;
3818 defm VMLSLu : N3VLMulOp_QHS<1,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
3819 "vmlsl", "u", NEONvmullu, sub>;
3821 defm VMLSLsls : N3VLMulOpSL_HS<0, 0b0110, "vmlsl", "s", NEONvmulls, sub>;
3822 defm VMLSLslu : N3VLMulOpSL_HS<1, 0b0110, "vmlsl", "u", NEONvmullu, sub>;
3824 // VQDMLSL : Vector Saturating Doubling Multiply Subtract Long (Q -= D * D)
3825 defm VQDMLSL : N3VLInt3_HS<0, 1, 0b1011, 0, IIC_VMACi16D, IIC_VMACi32D,
3826 "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
3827 defm VQDMLSLsl: N3VLInt3SL_HS<0, 0b111, "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
3829 // Vector Subtract Operations.
3831 // VSUB : Vector Subtract (integer and floating-point)
3832 defm VSUB : N3V_QHSD<1, 0, 0b1000, 0, IIC_VSUBiD, IIC_VSUBiQ,
3833 "vsub", "i", sub, 0>;
3834 def VSUBfd : N3VD<0, 0, 0b10, 0b1101, 0, IIC_VBIND, "vsub", "f32",
3835 v2f32, v2f32, fsub, 0>;
3836 def VSUBfq : N3VQ<0, 0, 0b10, 0b1101, 0, IIC_VBINQ, "vsub", "f32",
3837 v4f32, v4f32, fsub, 0>;
3838 // VSUBL : Vector Subtract Long (Q = D - D)
3839 defm VSUBLs : N3VLExt_QHS<0,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
3840 "vsubl", "s", sub, sext, 0>;
3841 defm VSUBLu : N3VLExt_QHS<1,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
3842 "vsubl", "u", sub, zext, 0>;
3843 // VSUBW : Vector Subtract Wide (Q = Q - D)
3844 defm VSUBWs : N3VW_QHS<0,1,0b0011,0, "vsubw", "s", sub, sext, 0>;
3845 defm VSUBWu : N3VW_QHS<1,1,0b0011,0, "vsubw", "u", sub, zext, 0>;
3846 // VHSUB : Vector Halving Subtract
3847 defm VHSUBs : N3VInt_QHS<0, 0, 0b0010, 0, N3RegFrm,
3848 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3849 "vhsub", "s", int_arm_neon_vhsubs, 0>;
3850 defm VHSUBu : N3VInt_QHS<1, 0, 0b0010, 0, N3RegFrm,
3851 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3852 "vhsub", "u", int_arm_neon_vhsubu, 0>;
3853 // VQSUB : Vector Saturing Subtract
3854 defm VQSUBs : N3VInt_QHSD<0, 0, 0b0010, 1, N3RegFrm,
3855 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3856 "vqsub", "s", int_arm_neon_vqsubs, 0>;
3857 defm VQSUBu : N3VInt_QHSD<1, 0, 0b0010, 1, N3RegFrm,
3858 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3859 "vqsub", "u", int_arm_neon_vqsubu, 0>;
3860 // VSUBHN : Vector Subtract and Narrow Returning High Half (D = Q - Q)
3861 defm VSUBHN : N3VNInt_HSD<0,1,0b0110,0, "vsubhn", "i",
3862 int_arm_neon_vsubhn, 0>;
3863 // VRSUBHN : Vector Rounding Subtract and Narrow Returning High Half (D=Q-Q)
3864 defm VRSUBHN : N3VNInt_HSD<1,1,0b0110,0, "vrsubhn", "i",
3865 int_arm_neon_vrsubhn, 0>;
3867 // Vector Comparisons.
3869 // VCEQ : Vector Compare Equal
3870 defm VCEQ : N3V_QHS<1, 0, 0b1000, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3871 IIC_VSUBi4Q, "vceq", "i", NEONvceq, 1>;
3872 def VCEQfd : N3VD<0,0,0b00,0b1110,0, IIC_VBIND, "vceq", "f32", v2i32, v2f32,
3874 def VCEQfq : N3VQ<0,0,0b00,0b1110,0, IIC_VBINQ, "vceq", "f32", v4i32, v4f32,
3877 defm VCEQz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00010, 0, "vceq", "i",
3878 "$Vd, $Vm, #0", NEONvceqz>;
3880 // VCGE : Vector Compare Greater Than or Equal
3881 defm VCGEs : N3V_QHS<0, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3882 IIC_VSUBi4Q, "vcge", "s", NEONvcge, 0>;
3883 defm VCGEu : N3V_QHS<1, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3884 IIC_VSUBi4Q, "vcge", "u", NEONvcgeu, 0>;
3885 def VCGEfd : N3VD<1,0,0b00,0b1110,0, IIC_VBIND, "vcge", "f32", v2i32, v2f32,
3887 def VCGEfq : N3VQ<1,0,0b00,0b1110,0, IIC_VBINQ, "vcge", "f32", v4i32, v4f32,
3890 defm VCGEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00001, 0, "vcge", "s",
3891 "$Vd, $Vm, #0", NEONvcgez>;
3892 defm VCLEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00011, 0, "vcle", "s",
3893 "$Vd, $Vm, #0", NEONvclez>;
3895 // VCGT : Vector Compare Greater Than
3896 defm VCGTs : N3V_QHS<0, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3897 IIC_VSUBi4Q, "vcgt", "s", NEONvcgt, 0>;
3898 defm VCGTu : N3V_QHS<1, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3899 IIC_VSUBi4Q, "vcgt", "u", NEONvcgtu, 0>;
3900 def VCGTfd : N3VD<1,0,0b10,0b1110,0, IIC_VBIND, "vcgt", "f32", v2i32, v2f32,
3902 def VCGTfq : N3VQ<1,0,0b10,0b1110,0, IIC_VBINQ, "vcgt", "f32", v4i32, v4f32,
3905 defm VCGTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00000, 0, "vcgt", "s",
3906 "$Vd, $Vm, #0", NEONvcgtz>;
3907 defm VCLTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00100, 0, "vclt", "s",
3908 "$Vd, $Vm, #0", NEONvcltz>;
3910 // VACGE : Vector Absolute Compare Greater Than or Equal (aka VCAGE)
3911 def VACGEd : N3VDInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacge",
3912 "f32", v2i32, v2f32, int_arm_neon_vacged, 0>;
3913 def VACGEq : N3VQInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacge",
3914 "f32", v4i32, v4f32, int_arm_neon_vacgeq, 0>;
3915 // VACGT : Vector Absolute Compare Greater Than (aka VCAGT)
3916 def VACGTd : N3VDInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacgt",
3917 "f32", v2i32, v2f32, int_arm_neon_vacgtd, 0>;
3918 def VACGTq : N3VQInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacgt",
3919 "f32", v4i32, v4f32, int_arm_neon_vacgtq, 0>;
3920 // VTST : Vector Test Bits
3921 defm VTST : N3V_QHS<0, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
3922 IIC_VBINi4Q, "vtst", "", NEONvtst, 1>;
3924 // Vector Bitwise Operations.
3926 def vnotd : PatFrag<(ops node:$in),
3927 (xor node:$in, (bitconvert (v8i8 NEONimmAllOnesV)))>;
3928 def vnotq : PatFrag<(ops node:$in),
3929 (xor node:$in, (bitconvert (v16i8 NEONimmAllOnesV)))>;
3932 // VAND : Vector Bitwise AND
3933 def VANDd : N3VDX<0, 0, 0b00, 0b0001, 1, IIC_VBINiD, "vand",
3934 v2i32, v2i32, and, 1>;
3935 def VANDq : N3VQX<0, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "vand",
3936 v4i32, v4i32, and, 1>;
3938 // VEOR : Vector Bitwise Exclusive OR
3939 def VEORd : N3VDX<1, 0, 0b00, 0b0001, 1, IIC_VBINiD, "veor",
3940 v2i32, v2i32, xor, 1>;
3941 def VEORq : N3VQX<1, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "veor",
3942 v4i32, v4i32, xor, 1>;
3944 // VORR : Vector Bitwise OR
3945 def VORRd : N3VDX<0, 0, 0b10, 0b0001, 1, IIC_VBINiD, "vorr",
3946 v2i32, v2i32, or, 1>;
3947 def VORRq : N3VQX<0, 0, 0b10, 0b0001, 1, IIC_VBINiQ, "vorr",
3948 v4i32, v4i32, or, 1>;
3950 def VORRiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 0, 1,
3951 (outs DPR:$Vd), (ins nImmSplatI16:$SIMM, DPR:$src),
3953 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
3955 (v4i16 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
3956 let Inst{9} = SIMM{9};
3959 def VORRiv2i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 0, 0, 1,
3960 (outs DPR:$Vd), (ins nImmSplatI32:$SIMM, DPR:$src),
3962 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
3964 (v2i32 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
3965 let Inst{10-9} = SIMM{10-9};
3968 def VORRiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 0, 1,
3969 (outs QPR:$Vd), (ins nImmSplatI16:$SIMM, QPR:$src),
3971 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
3973 (v8i16 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
3974 let Inst{9} = SIMM{9};
3977 def VORRiv4i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 1, 0, 1,
3978 (outs QPR:$Vd), (ins nImmSplatI32:$SIMM, QPR:$src),
3980 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
3982 (v4i32 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
3983 let Inst{10-9} = SIMM{10-9};
3987 // VBIC : Vector Bitwise Bit Clear (AND NOT)
3988 def VBICd : N3VX<0, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
3989 (ins DPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VBINiD,
3990 "vbic", "$Vd, $Vn, $Vm", "",
3991 [(set DPR:$Vd, (v2i32 (and DPR:$Vn,
3992 (vnotd DPR:$Vm))))]>;
3993 def VBICq : N3VX<0, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
3994 (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINiQ,
3995 "vbic", "$Vd, $Vn, $Vm", "",
3996 [(set QPR:$Vd, (v4i32 (and QPR:$Vn,
3997 (vnotq QPR:$Vm))))]>;
3999 def VBICiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 1, 1,
4000 (outs DPR:$Vd), (ins nImmSplatI16:$SIMM, DPR:$src),
4002 "vbic", "i16", "$Vd, $SIMM", "$src = $Vd",
4004 (v4i16 (NEONvbicImm DPR:$src, timm:$SIMM)))]> {
4005 let Inst{9} = SIMM{9};
4008 def VBICiv2i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 0, 1, 1,
4009 (outs DPR:$Vd), (ins nImmSplatI32:$SIMM, DPR:$src),
4011 "vbic", "i32", "$Vd, $SIMM", "$src = $Vd",
4013 (v2i32 (NEONvbicImm DPR:$src, timm:$SIMM)))]> {
4014 let Inst{10-9} = SIMM{10-9};
4017 def VBICiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 1, 1,
4018 (outs QPR:$Vd), (ins nImmSplatI16:$SIMM, QPR:$src),
4020 "vbic", "i16", "$Vd, $SIMM", "$src = $Vd",
4022 (v8i16 (NEONvbicImm QPR:$src, timm:$SIMM)))]> {
4023 let Inst{9} = SIMM{9};
4026 def VBICiv4i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 1, 1, 1,
4027 (outs QPR:$Vd), (ins nImmSplatI32:$SIMM, QPR:$src),
4029 "vbic", "i32", "$Vd, $SIMM", "$src = $Vd",
4031 (v4i32 (NEONvbicImm QPR:$src, timm:$SIMM)))]> {
4032 let Inst{10-9} = SIMM{10-9};
4035 // VORN : Vector Bitwise OR NOT
4036 def VORNd : N3VX<0, 0, 0b11, 0b0001, 0, 1, (outs DPR:$Vd),
4037 (ins DPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VBINiD,
4038 "vorn", "$Vd, $Vn, $Vm", "",
4039 [(set DPR:$Vd, (v2i32 (or DPR:$Vn,
4040 (vnotd DPR:$Vm))))]>;
4041 def VORNq : N3VX<0, 0, 0b11, 0b0001, 1, 1, (outs QPR:$Vd),
4042 (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINiQ,
4043 "vorn", "$Vd, $Vn, $Vm", "",
4044 [(set QPR:$Vd, (v4i32 (or QPR:$Vn,
4045 (vnotq QPR:$Vm))))]>;
4047 // VMVN : Vector Bitwise NOT (Immediate)
4049 let isReMaterializable = 1 in {
4051 def VMVNv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 1, 1, (outs DPR:$Vd),
4052 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4053 "vmvn", "i16", "$Vd, $SIMM", "",
4054 [(set DPR:$Vd, (v4i16 (NEONvmvnImm timm:$SIMM)))]> {
4055 let Inst{9} = SIMM{9};
4058 def VMVNv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 1, 1, (outs QPR:$Vd),
4059 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4060 "vmvn", "i16", "$Vd, $SIMM", "",
4061 [(set QPR:$Vd, (v8i16 (NEONvmvnImm timm:$SIMM)))]> {
4062 let Inst{9} = SIMM{9};
4065 def VMVNv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 1, 1, (outs DPR:$Vd),
4066 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4067 "vmvn", "i32", "$Vd, $SIMM", "",
4068 [(set DPR:$Vd, (v2i32 (NEONvmvnImm timm:$SIMM)))]> {
4069 let Inst{11-8} = SIMM{11-8};
4072 def VMVNv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 1, 1, (outs QPR:$Vd),
4073 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4074 "vmvn", "i32", "$Vd, $SIMM", "",
4075 [(set QPR:$Vd, (v4i32 (NEONvmvnImm timm:$SIMM)))]> {
4076 let Inst{11-8} = SIMM{11-8};
4080 // VMVN : Vector Bitwise NOT
4081 def VMVNd : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 0, 0,
4082 (outs DPR:$Vd), (ins DPR:$Vm), IIC_VSUBiD,
4083 "vmvn", "$Vd, $Vm", "",
4084 [(set DPR:$Vd, (v2i32 (vnotd DPR:$Vm)))]>;
4085 def VMVNq : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 1, 0,
4086 (outs QPR:$Vd), (ins QPR:$Vm), IIC_VSUBiD,
4087 "vmvn", "$Vd, $Vm", "",
4088 [(set QPR:$Vd, (v4i32 (vnotq QPR:$Vm)))]>;
4089 def : Pat<(v2i32 (vnotd DPR:$src)), (VMVNd DPR:$src)>;
4090 def : Pat<(v4i32 (vnotq QPR:$src)), (VMVNq QPR:$src)>;
4092 // VBSL : Vector Bitwise Select
4093 def VBSLd : N3VX<1, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
4094 (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
4095 N3RegFrm, IIC_VCNTiD,
4096 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4098 (v2i32 (NEONvbsl DPR:$src1, DPR:$Vn, DPR:$Vm)))]>;
4100 def : Pat<(v2i32 (or (and DPR:$Vn, DPR:$Vd),
4101 (and DPR:$Vm, (vnotd DPR:$Vd)))),
4102 (VBSLd DPR:$Vd, DPR:$Vn, DPR:$Vm)>;
4104 def VBSLq : N3VX<1, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
4105 (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
4106 N3RegFrm, IIC_VCNTiQ,
4107 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4109 (v4i32 (NEONvbsl QPR:$src1, QPR:$Vn, QPR:$Vm)))]>;
4111 def : Pat<(v4i32 (or (and QPR:$Vn, QPR:$Vd),
4112 (and QPR:$Vm, (vnotq QPR:$Vd)))),
4113 (VBSLq QPR:$Vd, QPR:$Vn, QPR:$Vm)>;
4115 // VBIF : Vector Bitwise Insert if False
4116 // like VBSL but with: "vbif $dst, $src3, $src1", "$src2 = $dst",
4117 // FIXME: This instruction's encoding MAY NOT BE correct.
4118 def VBIFd : N3VX<1, 0, 0b11, 0b0001, 0, 1,
4119 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
4120 N3RegFrm, IIC_VBINiD,
4121 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4123 def VBIFq : N3VX<1, 0, 0b11, 0b0001, 1, 1,
4124 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
4125 N3RegFrm, IIC_VBINiQ,
4126 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4129 // VBIT : Vector Bitwise Insert if True
4130 // like VBSL but with: "vbit $dst, $src2, $src1", "$src3 = $dst",
4131 // FIXME: This instruction's encoding MAY NOT BE correct.
4132 def VBITd : N3VX<1, 0, 0b10, 0b0001, 0, 1,
4133 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
4134 N3RegFrm, IIC_VBINiD,
4135 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4137 def VBITq : N3VX<1, 0, 0b10, 0b0001, 1, 1,
4138 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
4139 N3RegFrm, IIC_VBINiQ,
4140 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4143 // VBIT/VBIF are not yet implemented. The TwoAddress pass will not go looking
4144 // for equivalent operations with different register constraints; it just
4147 // Vector Absolute Differences.
4149 // VABD : Vector Absolute Difference
4150 defm VABDs : N3VInt_QHS<0, 0, 0b0111, 0, N3RegFrm,
4151 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4152 "vabd", "s", int_arm_neon_vabds, 1>;
4153 defm VABDu : N3VInt_QHS<1, 0, 0b0111, 0, N3RegFrm,
4154 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4155 "vabd", "u", int_arm_neon_vabdu, 1>;
4156 def VABDfd : N3VDInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBIND,
4157 "vabd", "f32", v2f32, v2f32, int_arm_neon_vabds, 1>;
4158 def VABDfq : N3VQInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBINQ,
4159 "vabd", "f32", v4f32, v4f32, int_arm_neon_vabds, 1>;
4161 // VABDL : Vector Absolute Difference Long (Q = | D - D |)
4162 defm VABDLs : N3VLIntExt_QHS<0,1,0b0111,0, IIC_VSUBi4Q,
4163 "vabdl", "s", int_arm_neon_vabds, zext, 1>;
4164 defm VABDLu : N3VLIntExt_QHS<1,1,0b0111,0, IIC_VSUBi4Q,
4165 "vabdl", "u", int_arm_neon_vabdu, zext, 1>;
4167 // VABA : Vector Absolute Difference and Accumulate
4168 defm VABAs : N3VIntOp_QHS<0,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
4169 "vaba", "s", int_arm_neon_vabds, add>;
4170 defm VABAu : N3VIntOp_QHS<1,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
4171 "vaba", "u", int_arm_neon_vabdu, add>;
4173 // VABAL : Vector Absolute Difference and Accumulate Long (Q += | D - D |)
4174 defm VABALs : N3VLIntExtOp_QHS<0,1,0b0101,0, IIC_VABAD,
4175 "vabal", "s", int_arm_neon_vabds, zext, add>;
4176 defm VABALu : N3VLIntExtOp_QHS<1,1,0b0101,0, IIC_VABAD,
4177 "vabal", "u", int_arm_neon_vabdu, zext, add>;
4179 // Vector Maximum and Minimum.
4181 // VMAX : Vector Maximum
4182 defm VMAXs : N3VInt_QHS<0, 0, 0b0110, 0, N3RegFrm,
4183 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4184 "vmax", "s", int_arm_neon_vmaxs, 1>;
4185 defm VMAXu : N3VInt_QHS<1, 0, 0b0110, 0, N3RegFrm,
4186 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4187 "vmax", "u", int_arm_neon_vmaxu, 1>;
4188 def VMAXfd : N3VDInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBIND,
4190 v2f32, v2f32, int_arm_neon_vmaxs, 1>;
4191 def VMAXfq : N3VQInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBINQ,
4193 v4f32, v4f32, int_arm_neon_vmaxs, 1>;
4195 // VMIN : Vector Minimum
4196 defm VMINs : N3VInt_QHS<0, 0, 0b0110, 1, N3RegFrm,
4197 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4198 "vmin", "s", int_arm_neon_vmins, 1>;
4199 defm VMINu : N3VInt_QHS<1, 0, 0b0110, 1, N3RegFrm,
4200 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4201 "vmin", "u", int_arm_neon_vminu, 1>;
4202 def VMINfd : N3VDInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBIND,
4204 v2f32, v2f32, int_arm_neon_vmins, 1>;
4205 def VMINfq : N3VQInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBINQ,
4207 v4f32, v4f32, int_arm_neon_vmins, 1>;
4209 // Vector Pairwise Operations.
4211 // VPADD : Vector Pairwise Add
4212 def VPADDi8 : N3VDInt<0, 0, 0b00, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4214 v8i8, v8i8, int_arm_neon_vpadd, 0>;
4215 def VPADDi16 : N3VDInt<0, 0, 0b01, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4217 v4i16, v4i16, int_arm_neon_vpadd, 0>;
4218 def VPADDi32 : N3VDInt<0, 0, 0b10, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4220 v2i32, v2i32, int_arm_neon_vpadd, 0>;
4221 def VPADDf : N3VDInt<1, 0, 0b00, 0b1101, 0, N3RegFrm,
4222 IIC_VPBIND, "vpadd", "f32",
4223 v2f32, v2f32, int_arm_neon_vpadd, 0>;
4225 // VPADDL : Vector Pairwise Add Long
4226 defm VPADDLs : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00100, 0, "vpaddl", "s",
4227 int_arm_neon_vpaddls>;
4228 defm VPADDLu : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00101, 0, "vpaddl", "u",
4229 int_arm_neon_vpaddlu>;
4231 // VPADAL : Vector Pairwise Add and Accumulate Long
4232 defm VPADALs : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01100, 0, "vpadal", "s",
4233 int_arm_neon_vpadals>;
4234 defm VPADALu : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01101, 0, "vpadal", "u",
4235 int_arm_neon_vpadalu>;
4237 // VPMAX : Vector Pairwise Maximum
4238 def VPMAXs8 : N3VDInt<0, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4239 "s8", v8i8, v8i8, int_arm_neon_vpmaxs, 0>;
4240 def VPMAXs16 : N3VDInt<0, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4241 "s16", v4i16, v4i16, int_arm_neon_vpmaxs, 0>;
4242 def VPMAXs32 : N3VDInt<0, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4243 "s32", v2i32, v2i32, int_arm_neon_vpmaxs, 0>;
4244 def VPMAXu8 : N3VDInt<1, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4245 "u8", v8i8, v8i8, int_arm_neon_vpmaxu, 0>;
4246 def VPMAXu16 : N3VDInt<1, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4247 "u16", v4i16, v4i16, int_arm_neon_vpmaxu, 0>;
4248 def VPMAXu32 : N3VDInt<1, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4249 "u32", v2i32, v2i32, int_arm_neon_vpmaxu, 0>;
4250 def VPMAXf : N3VDInt<1, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmax",
4251 "f32", v2f32, v2f32, int_arm_neon_vpmaxs, 0>;
4253 // VPMIN : Vector Pairwise Minimum
4254 def VPMINs8 : N3VDInt<0, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4255 "s8", v8i8, v8i8, int_arm_neon_vpmins, 0>;
4256 def VPMINs16 : N3VDInt<0, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4257 "s16", v4i16, v4i16, int_arm_neon_vpmins, 0>;
4258 def VPMINs32 : N3VDInt<0, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4259 "s32", v2i32, v2i32, int_arm_neon_vpmins, 0>;
4260 def VPMINu8 : N3VDInt<1, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4261 "u8", v8i8, v8i8, int_arm_neon_vpminu, 0>;
4262 def VPMINu16 : N3VDInt<1, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4263 "u16", v4i16, v4i16, int_arm_neon_vpminu, 0>;
4264 def VPMINu32 : N3VDInt<1, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4265 "u32", v2i32, v2i32, int_arm_neon_vpminu, 0>;
4266 def VPMINf : N3VDInt<1, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmin",
4267 "f32", v2f32, v2f32, int_arm_neon_vpmins, 0>;
4269 // Vector Reciprocal and Reciprocal Square Root Estimate and Step.
4271 // VRECPE : Vector Reciprocal Estimate
4272 def VRECPEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
4273 IIC_VUNAD, "vrecpe", "u32",
4274 v2i32, v2i32, int_arm_neon_vrecpe>;
4275 def VRECPEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
4276 IIC_VUNAQ, "vrecpe", "u32",
4277 v4i32, v4i32, int_arm_neon_vrecpe>;
4278 def VRECPEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
4279 IIC_VUNAD, "vrecpe", "f32",
4280 v2f32, v2f32, int_arm_neon_vrecpe>;
4281 def VRECPEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
4282 IIC_VUNAQ, "vrecpe", "f32",
4283 v4f32, v4f32, int_arm_neon_vrecpe>;
4285 // VRECPS : Vector Reciprocal Step
4286 def VRECPSfd : N3VDInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
4287 IIC_VRECSD, "vrecps", "f32",
4288 v2f32, v2f32, int_arm_neon_vrecps, 1>;
4289 def VRECPSfq : N3VQInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
4290 IIC_VRECSQ, "vrecps", "f32",
4291 v4f32, v4f32, int_arm_neon_vrecps, 1>;
4293 // VRSQRTE : Vector Reciprocal Square Root Estimate
4294 def VRSQRTEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
4295 IIC_VUNAD, "vrsqrte", "u32",
4296 v2i32, v2i32, int_arm_neon_vrsqrte>;
4297 def VRSQRTEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
4298 IIC_VUNAQ, "vrsqrte", "u32",
4299 v4i32, v4i32, int_arm_neon_vrsqrte>;
4300 def VRSQRTEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
4301 IIC_VUNAD, "vrsqrte", "f32",
4302 v2f32, v2f32, int_arm_neon_vrsqrte>;
4303 def VRSQRTEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
4304 IIC_VUNAQ, "vrsqrte", "f32",
4305 v4f32, v4f32, int_arm_neon_vrsqrte>;
4307 // VRSQRTS : Vector Reciprocal Square Root Step
4308 def VRSQRTSfd : N3VDInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
4309 IIC_VRECSD, "vrsqrts", "f32",
4310 v2f32, v2f32, int_arm_neon_vrsqrts, 1>;
4311 def VRSQRTSfq : N3VQInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
4312 IIC_VRECSQ, "vrsqrts", "f32",
4313 v4f32, v4f32, int_arm_neon_vrsqrts, 1>;
4317 // VSHL : Vector Shift
4318 defm VSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 0, N3RegVShFrm,
4319 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
4320 "vshl", "s", int_arm_neon_vshifts>;
4321 defm VSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 0, N3RegVShFrm,
4322 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
4323 "vshl", "u", int_arm_neon_vshiftu>;
4325 // VSHL : Vector Shift Left (Immediate)
4326 defm VSHLi : N2VShL_QHSD<0, 1, 0b0101, 1, IIC_VSHLiD, "vshl", "i", NEONvshl>;
4328 // VSHR : Vector Shift Right (Immediate)
4329 defm VSHRs : N2VShR_QHSD<0, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "s",NEONvshrs>;
4330 defm VSHRu : N2VShR_QHSD<1, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "u",NEONvshru>;
4332 // VSHLL : Vector Shift Left Long
4333 defm VSHLLs : N2VLSh_QHS<0, 1, 0b1010, 0, 0, 1, "vshll", "s", NEONvshlls>;
4334 defm VSHLLu : N2VLSh_QHS<1, 1, 0b1010, 0, 0, 1, "vshll", "u", NEONvshllu>;
4336 // VSHLL : Vector Shift Left Long (with maximum shift count)
4337 class N2VLShMax<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
4338 bit op6, bit op4, string OpcodeStr, string Dt, ValueType ResTy,
4339 ValueType OpTy, SDNode OpNode>
4340 : N2VLSh<op24, op23, op11_8, op7, op6, op4, OpcodeStr, Dt,
4341 ResTy, OpTy, OpNode> {
4342 let Inst{21-16} = op21_16;
4343 let DecoderMethod = "DecodeVSHLMaxInstruction";
4345 def VSHLLi8 : N2VLShMax<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll", "i8",
4346 v8i16, v8i8, NEONvshlli>;
4347 def VSHLLi16 : N2VLShMax<1, 1, 0b110110, 0b0011, 0, 0, 0, "vshll", "i16",
4348 v4i32, v4i16, NEONvshlli>;
4349 def VSHLLi32 : N2VLShMax<1, 1, 0b111010, 0b0011, 0, 0, 0, "vshll", "i32",
4350 v2i64, v2i32, NEONvshlli>;
4352 // VSHRN : Vector Shift Right and Narrow
4353 defm VSHRN : N2VNSh_HSD<0,1,0b1000,0,0,1, IIC_VSHLiD, "vshrn", "i",
4356 // VRSHL : Vector Rounding Shift
4357 defm VRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 0, N3RegVShFrm,
4358 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4359 "vrshl", "s", int_arm_neon_vrshifts>;
4360 defm VRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 0, N3RegVShFrm,
4361 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4362 "vrshl", "u", int_arm_neon_vrshiftu>;
4363 // VRSHR : Vector Rounding Shift Right
4364 defm VRSHRs : N2VShR_QHSD<0,1,0b0010,1, IIC_VSHLi4D, "vrshr", "s",NEONvrshrs>;
4365 defm VRSHRu : N2VShR_QHSD<1,1,0b0010,1, IIC_VSHLi4D, "vrshr", "u",NEONvrshru>;
4367 // VRSHRN : Vector Rounding Shift Right and Narrow
4368 defm VRSHRN : N2VNSh_HSD<0, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vrshrn", "i",
4371 // VQSHL : Vector Saturating Shift
4372 defm VQSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 1, N3RegVShFrm,
4373 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4374 "vqshl", "s", int_arm_neon_vqshifts>;
4375 defm VQSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 1, N3RegVShFrm,
4376 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4377 "vqshl", "u", int_arm_neon_vqshiftu>;
4378 // VQSHL : Vector Saturating Shift Left (Immediate)
4379 defm VQSHLsi : N2VShL_QHSD<0,1,0b0111,1, IIC_VSHLi4D, "vqshl", "s",NEONvqshls>;
4380 defm VQSHLui : N2VShL_QHSD<1,1,0b0111,1, IIC_VSHLi4D, "vqshl", "u",NEONvqshlu>;
4382 // VQSHLU : Vector Saturating Shift Left (Immediate, Unsigned)
4383 defm VQSHLsu : N2VShL_QHSD<1,1,0b0110,1, IIC_VSHLi4D,"vqshlu","s",NEONvqshlsu>;
4385 // VQSHRN : Vector Saturating Shift Right and Narrow
4386 defm VQSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "s",
4388 defm VQSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "u",
4391 // VQSHRUN : Vector Saturating Shift Right and Narrow (Unsigned)
4392 defm VQSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 0, 1, IIC_VSHLi4D, "vqshrun", "s",
4395 // VQRSHL : Vector Saturating Rounding Shift
4396 defm VQRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 1, N3RegVShFrm,
4397 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4398 "vqrshl", "s", int_arm_neon_vqrshifts>;
4399 defm VQRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 1, N3RegVShFrm,
4400 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4401 "vqrshl", "u", int_arm_neon_vqrshiftu>;
4403 // VQRSHRN : Vector Saturating Rounding Shift Right and Narrow
4404 defm VQRSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "s",
4406 defm VQRSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "u",
4409 // VQRSHRUN : Vector Saturating Rounding Shift Right and Narrow (Unsigned)
4410 defm VQRSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vqrshrun", "s",
4413 // VSRA : Vector Shift Right and Accumulate
4414 defm VSRAs : N2VShAdd_QHSD<0, 1, 0b0001, 1, "vsra", "s", NEONvshrs>;
4415 defm VSRAu : N2VShAdd_QHSD<1, 1, 0b0001, 1, "vsra", "u", NEONvshru>;
4416 // VRSRA : Vector Rounding Shift Right and Accumulate
4417 defm VRSRAs : N2VShAdd_QHSD<0, 1, 0b0011, 1, "vrsra", "s", NEONvrshrs>;
4418 defm VRSRAu : N2VShAdd_QHSD<1, 1, 0b0011, 1, "vrsra", "u", NEONvrshru>;
4420 // VSLI : Vector Shift Left and Insert
4421 defm VSLI : N2VShInsL_QHSD<1, 1, 0b0101, 1, "vsli">;
4423 // VSRI : Vector Shift Right and Insert
4424 defm VSRI : N2VShInsR_QHSD<1, 1, 0b0100, 1, "vsri">;
4426 // Vector Absolute and Saturating Absolute.
4428 // VABS : Vector Absolute Value
4429 defm VABS : N2VInt_QHS<0b11, 0b11, 0b01, 0b00110, 0,
4430 IIC_VUNAiD, IIC_VUNAiQ, "vabs", "s",
4432 def VABSfd : N2VDInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
4433 IIC_VUNAD, "vabs", "f32",
4434 v2f32, v2f32, int_arm_neon_vabs>;
4435 def VABSfq : N2VQInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
4436 IIC_VUNAQ, "vabs", "f32",
4437 v4f32, v4f32, int_arm_neon_vabs>;
4439 // VQABS : Vector Saturating Absolute Value
4440 defm VQABS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01110, 0,
4441 IIC_VQUNAiD, IIC_VQUNAiQ, "vqabs", "s",
4442 int_arm_neon_vqabs>;
4446 def vnegd : PatFrag<(ops node:$in),
4447 (sub (bitconvert (v2i32 NEONimmAllZerosV)), node:$in)>;
4448 def vnegq : PatFrag<(ops node:$in),
4449 (sub (bitconvert (v4i32 NEONimmAllZerosV)), node:$in)>;
4451 class VNEGD<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
4452 : N2V<0b11, 0b11, size, 0b01, 0b00111, 0, 0, (outs DPR:$Vd), (ins DPR:$Vm),
4453 IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
4454 [(set DPR:$Vd, (Ty (vnegd DPR:$Vm)))]>;
4455 class VNEGQ<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
4456 : N2V<0b11, 0b11, size, 0b01, 0b00111, 1, 0, (outs QPR:$Vd), (ins QPR:$Vm),
4457 IIC_VSHLiQ, OpcodeStr, Dt, "$Vd, $Vm", "",
4458 [(set QPR:$Vd, (Ty (vnegq QPR:$Vm)))]>;
4460 // VNEG : Vector Negate (integer)
4461 def VNEGs8d : VNEGD<0b00, "vneg", "s8", v8i8>;
4462 def VNEGs16d : VNEGD<0b01, "vneg", "s16", v4i16>;
4463 def VNEGs32d : VNEGD<0b10, "vneg", "s32", v2i32>;
4464 def VNEGs8q : VNEGQ<0b00, "vneg", "s8", v16i8>;
4465 def VNEGs16q : VNEGQ<0b01, "vneg", "s16", v8i16>;
4466 def VNEGs32q : VNEGQ<0b10, "vneg", "s32", v4i32>;
4468 // VNEG : Vector Negate (floating-point)
4469 def VNEGfd : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
4470 (outs DPR:$Vd), (ins DPR:$Vm), IIC_VUNAD,
4471 "vneg", "f32", "$Vd, $Vm", "",
4472 [(set DPR:$Vd, (v2f32 (fneg DPR:$Vm)))]>;
4473 def VNEGf32q : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 1, 0,
4474 (outs QPR:$Vd), (ins QPR:$Vm), IIC_VUNAQ,
4475 "vneg", "f32", "$Vd, $Vm", "",
4476 [(set QPR:$Vd, (v4f32 (fneg QPR:$Vm)))]>;
4478 def : Pat<(v8i8 (vnegd DPR:$src)), (VNEGs8d DPR:$src)>;
4479 def : Pat<(v4i16 (vnegd DPR:$src)), (VNEGs16d DPR:$src)>;
4480 def : Pat<(v2i32 (vnegd DPR:$src)), (VNEGs32d DPR:$src)>;
4481 def : Pat<(v16i8 (vnegq QPR:$src)), (VNEGs8q QPR:$src)>;
4482 def : Pat<(v8i16 (vnegq QPR:$src)), (VNEGs16q QPR:$src)>;
4483 def : Pat<(v4i32 (vnegq QPR:$src)), (VNEGs32q QPR:$src)>;
4485 // VQNEG : Vector Saturating Negate
4486 defm VQNEG : N2VInt_QHS<0b11, 0b11, 0b00, 0b01111, 0,
4487 IIC_VQUNAiD, IIC_VQUNAiQ, "vqneg", "s",
4488 int_arm_neon_vqneg>;
4490 // Vector Bit Counting Operations.
4492 // VCLS : Vector Count Leading Sign Bits
4493 defm VCLS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01000, 0,
4494 IIC_VCNTiD, IIC_VCNTiQ, "vcls", "s",
4496 // VCLZ : Vector Count Leading Zeros
4497 defm VCLZ : N2VInt_QHS<0b11, 0b11, 0b00, 0b01001, 0,
4498 IIC_VCNTiD, IIC_VCNTiQ, "vclz", "i",
4500 // VCNT : Vector Count One Bits
4501 def VCNTd : N2VDInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
4502 IIC_VCNTiD, "vcnt", "8",
4503 v8i8, v8i8, int_arm_neon_vcnt>;
4504 def VCNTq : N2VQInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
4505 IIC_VCNTiQ, "vcnt", "8",
4506 v16i8, v16i8, int_arm_neon_vcnt>;
4509 def VSWPd : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 0, 0,
4510 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
4511 "vswp", "$Vd, $Vm", "", []>;
4512 def VSWPq : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 1, 0,
4513 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
4514 "vswp", "$Vd, $Vm", "", []>;
4516 // Vector Move Operations.
4518 // VMOV : Vector Move (Register)
4519 def : InstAlias<"vmov${p} $Vd, $Vm",
4520 (VORRd DPR:$Vd, DPR:$Vm, DPR:$Vm, pred:$p)>;
4521 def : InstAlias<"vmov${p} $Vd, $Vm",
4522 (VORRq QPR:$Vd, QPR:$Vm, QPR:$Vm, pred:$p)>;
4523 defm : VFPDTAnyNoF64InstAlias<"vmov${p}", "$Vd, $Vm",
4524 (VORRd DPR:$Vd, DPR:$Vm, DPR:$Vm, pred:$p)>;
4525 defm : VFPDTAnyNoF64InstAlias<"vmov${p}", "$Vd, $Vm",
4526 (VORRq QPR:$Vd, QPR:$Vm, QPR:$Vm, pred:$p)>;
4528 // VMOV : Vector Move (Immediate)
4530 let isReMaterializable = 1 in {
4531 def VMOVv8i8 : N1ModImm<1, 0b000, 0b1110, 0, 0, 0, 1, (outs DPR:$Vd),
4532 (ins nImmSplatI8:$SIMM), IIC_VMOVImm,
4533 "vmov", "i8", "$Vd, $SIMM", "",
4534 [(set DPR:$Vd, (v8i8 (NEONvmovImm timm:$SIMM)))]>;
4535 def VMOVv16i8 : N1ModImm<1, 0b000, 0b1110, 0, 1, 0, 1, (outs QPR:$Vd),
4536 (ins nImmSplatI8:$SIMM), IIC_VMOVImm,
4537 "vmov", "i8", "$Vd, $SIMM", "",
4538 [(set QPR:$Vd, (v16i8 (NEONvmovImm timm:$SIMM)))]>;
4540 def VMOVv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 0, 1, (outs DPR:$Vd),
4541 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4542 "vmov", "i16", "$Vd, $SIMM", "",
4543 [(set DPR:$Vd, (v4i16 (NEONvmovImm timm:$SIMM)))]> {
4544 let Inst{9} = SIMM{9};
4547 def VMOVv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 0, 1, (outs QPR:$Vd),
4548 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4549 "vmov", "i16", "$Vd, $SIMM", "",
4550 [(set QPR:$Vd, (v8i16 (NEONvmovImm timm:$SIMM)))]> {
4551 let Inst{9} = SIMM{9};
4554 def VMOVv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 0, 1, (outs DPR:$Vd),
4555 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4556 "vmov", "i32", "$Vd, $SIMM", "",
4557 [(set DPR:$Vd, (v2i32 (NEONvmovImm timm:$SIMM)))]> {
4558 let Inst{11-8} = SIMM{11-8};
4561 def VMOVv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 0, 1, (outs QPR:$Vd),
4562 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4563 "vmov", "i32", "$Vd, $SIMM", "",
4564 [(set QPR:$Vd, (v4i32 (NEONvmovImm timm:$SIMM)))]> {
4565 let Inst{11-8} = SIMM{11-8};
4568 def VMOVv1i64 : N1ModImm<1, 0b000, 0b1110, 0, 0, 1, 1, (outs DPR:$Vd),
4569 (ins nImmSplatI64:$SIMM), IIC_VMOVImm,
4570 "vmov", "i64", "$Vd, $SIMM", "",
4571 [(set DPR:$Vd, (v1i64 (NEONvmovImm timm:$SIMM)))]>;
4572 def VMOVv2i64 : N1ModImm<1, 0b000, 0b1110, 0, 1, 1, 1, (outs QPR:$Vd),
4573 (ins nImmSplatI64:$SIMM), IIC_VMOVImm,
4574 "vmov", "i64", "$Vd, $SIMM", "",
4575 [(set QPR:$Vd, (v2i64 (NEONvmovImm timm:$SIMM)))]>;
4577 def VMOVv2f32 : N1ModImm<1, 0b000, 0b1111, 0, 0, 0, 1, (outs DPR:$Vd),
4578 (ins nImmVMOVF32:$SIMM), IIC_VMOVImm,
4579 "vmov", "f32", "$Vd, $SIMM", "",
4580 [(set DPR:$Vd, (v2f32 (NEONvmovFPImm timm:$SIMM)))]>;
4581 def VMOVv4f32 : N1ModImm<1, 0b000, 0b1111, 0, 1, 0, 1, (outs QPR:$Vd),
4582 (ins nImmVMOVF32:$SIMM), IIC_VMOVImm,
4583 "vmov", "f32", "$Vd, $SIMM", "",
4584 [(set QPR:$Vd, (v4f32 (NEONvmovFPImm timm:$SIMM)))]>;
4585 } // isReMaterializable
4587 // VMOV : Vector Get Lane (move scalar to ARM core register)
4589 def VGETLNs8 : NVGetLane<{1,1,1,0,0,1,?,1}, 0b1011, {?,?},
4590 (outs GPR:$R), (ins DPR:$V, VectorIndex8:$lane),
4591 IIC_VMOVSI, "vmov", "s8", "$R, $V$lane",
4592 [(set GPR:$R, (NEONvgetlanes (v8i8 DPR:$V),
4594 let Inst{21} = lane{2};
4595 let Inst{6-5} = lane{1-0};
4597 def VGETLNs16 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, {?,1},
4598 (outs GPR:$R), (ins DPR:$V, VectorIndex16:$lane),
4599 IIC_VMOVSI, "vmov", "s16", "$R, $V$lane",
4600 [(set GPR:$R, (NEONvgetlanes (v4i16 DPR:$V),
4602 let Inst{21} = lane{1};
4603 let Inst{6} = lane{0};
4605 def VGETLNu8 : NVGetLane<{1,1,1,0,1,1,?,1}, 0b1011, {?,?},
4606 (outs GPR:$R), (ins DPR:$V, VectorIndex8:$lane),
4607 IIC_VMOVSI, "vmov", "u8", "$R, $V$lane",
4608 [(set GPR:$R, (NEONvgetlaneu (v8i8 DPR:$V),
4610 let Inst{21} = lane{2};
4611 let Inst{6-5} = lane{1-0};
4613 def VGETLNu16 : NVGetLane<{1,1,1,0,1,0,?,1}, 0b1011, {?,1},
4614 (outs GPR:$R), (ins DPR:$V, VectorIndex16:$lane),
4615 IIC_VMOVSI, "vmov", "u16", "$R, $V$lane",
4616 [(set GPR:$R, (NEONvgetlaneu (v4i16 DPR:$V),
4618 let Inst{21} = lane{1};
4619 let Inst{6} = lane{0};
4621 def VGETLNi32 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, 0b00,
4622 (outs GPR:$R), (ins DPR:$V, VectorIndex32:$lane),
4623 IIC_VMOVSI, "vmov", "32", "$R, $V$lane",
4624 [(set GPR:$R, (extractelt (v2i32 DPR:$V),
4626 let Inst{21} = lane{0};
4628 // def VGETLNf32: see FMRDH and FMRDL in ARMInstrVFP.td
4629 def : Pat<(NEONvgetlanes (v16i8 QPR:$src), imm:$lane),
4630 (VGETLNs8 (v8i8 (EXTRACT_SUBREG QPR:$src,
4631 (DSubReg_i8_reg imm:$lane))),
4632 (SubReg_i8_lane imm:$lane))>;
4633 def : Pat<(NEONvgetlanes (v8i16 QPR:$src), imm:$lane),
4634 (VGETLNs16 (v4i16 (EXTRACT_SUBREG QPR:$src,
4635 (DSubReg_i16_reg imm:$lane))),
4636 (SubReg_i16_lane imm:$lane))>;
4637 def : Pat<(NEONvgetlaneu (v16i8 QPR:$src), imm:$lane),
4638 (VGETLNu8 (v8i8 (EXTRACT_SUBREG QPR:$src,
4639 (DSubReg_i8_reg imm:$lane))),
4640 (SubReg_i8_lane imm:$lane))>;
4641 def : Pat<(NEONvgetlaneu (v8i16 QPR:$src), imm:$lane),
4642 (VGETLNu16 (v4i16 (EXTRACT_SUBREG QPR:$src,
4643 (DSubReg_i16_reg imm:$lane))),
4644 (SubReg_i16_lane imm:$lane))>;
4645 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
4646 (VGETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src,
4647 (DSubReg_i32_reg imm:$lane))),
4648 (SubReg_i32_lane imm:$lane))>;
4649 def : Pat<(extractelt (v2f32 DPR:$src1), imm:$src2),
4650 (EXTRACT_SUBREG (v2f32 (COPY_TO_REGCLASS (v2f32 DPR:$src1),DPR_VFP2)),
4651 (SSubReg_f32_reg imm:$src2))>;
4652 def : Pat<(extractelt (v4f32 QPR:$src1), imm:$src2),
4653 (EXTRACT_SUBREG (v4f32 (COPY_TO_REGCLASS (v4f32 QPR:$src1),QPR_VFP2)),
4654 (SSubReg_f32_reg imm:$src2))>;
4655 //def : Pat<(extractelt (v2i64 QPR:$src1), imm:$src2),
4656 // (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
4657 def : Pat<(extractelt (v2f64 QPR:$src1), imm:$src2),
4658 (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
4661 // VMOV : Vector Set Lane (move ARM core register to scalar)
4663 let Constraints = "$src1 = $V" in {
4664 def VSETLNi8 : NVSetLane<{1,1,1,0,0,1,?,0}, 0b1011, {?,?}, (outs DPR:$V),
4665 (ins DPR:$src1, GPR:$R, VectorIndex8:$lane),
4666 IIC_VMOVISL, "vmov", "8", "$V$lane, $R",
4667 [(set DPR:$V, (vector_insert (v8i8 DPR:$src1),
4668 GPR:$R, imm:$lane))]> {
4669 let Inst{21} = lane{2};
4670 let Inst{6-5} = lane{1-0};
4672 def VSETLNi16 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, {?,1}, (outs DPR:$V),
4673 (ins DPR:$src1, GPR:$R, VectorIndex16:$lane),
4674 IIC_VMOVISL, "vmov", "16", "$V$lane, $R",
4675 [(set DPR:$V, (vector_insert (v4i16 DPR:$src1),
4676 GPR:$R, imm:$lane))]> {
4677 let Inst{21} = lane{1};
4678 let Inst{6} = lane{0};
4680 def VSETLNi32 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, 0b00, (outs DPR:$V),
4681 (ins DPR:$src1, GPR:$R, VectorIndex32:$lane),
4682 IIC_VMOVISL, "vmov", "32", "$V$lane, $R",
4683 [(set DPR:$V, (insertelt (v2i32 DPR:$src1),
4684 GPR:$R, imm:$lane))]> {
4685 let Inst{21} = lane{0};
4688 def : Pat<(vector_insert (v16i8 QPR:$src1), GPR:$src2, imm:$lane),
4689 (v16i8 (INSERT_SUBREG QPR:$src1,
4690 (v8i8 (VSETLNi8 (v8i8 (EXTRACT_SUBREG QPR:$src1,
4691 (DSubReg_i8_reg imm:$lane))),
4692 GPR:$src2, (SubReg_i8_lane imm:$lane))),
4693 (DSubReg_i8_reg imm:$lane)))>;
4694 def : Pat<(vector_insert (v8i16 QPR:$src1), GPR:$src2, imm:$lane),
4695 (v8i16 (INSERT_SUBREG QPR:$src1,
4696 (v4i16 (VSETLNi16 (v4i16 (EXTRACT_SUBREG QPR:$src1,
4697 (DSubReg_i16_reg imm:$lane))),
4698 GPR:$src2, (SubReg_i16_lane imm:$lane))),
4699 (DSubReg_i16_reg imm:$lane)))>;
4700 def : Pat<(insertelt (v4i32 QPR:$src1), GPR:$src2, imm:$lane),
4701 (v4i32 (INSERT_SUBREG QPR:$src1,
4702 (v2i32 (VSETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src1,
4703 (DSubReg_i32_reg imm:$lane))),
4704 GPR:$src2, (SubReg_i32_lane imm:$lane))),
4705 (DSubReg_i32_reg imm:$lane)))>;
4707 def : Pat<(v2f32 (insertelt DPR:$src1, SPR:$src2, imm:$src3)),
4708 (INSERT_SUBREG (v2f32 (COPY_TO_REGCLASS DPR:$src1, DPR_VFP2)),
4709 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
4710 def : Pat<(v4f32 (insertelt QPR:$src1, SPR:$src2, imm:$src3)),
4711 (INSERT_SUBREG (v4f32 (COPY_TO_REGCLASS QPR:$src1, QPR_VFP2)),
4712 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
4714 //def : Pat<(v2i64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
4715 // (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
4716 def : Pat<(v2f64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
4717 (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
4719 def : Pat<(v2f32 (scalar_to_vector SPR:$src)),
4720 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
4721 def : Pat<(v2f64 (scalar_to_vector (f64 DPR:$src))),
4722 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
4723 def : Pat<(v4f32 (scalar_to_vector SPR:$src)),
4724 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
4726 def : Pat<(v8i8 (scalar_to_vector GPR:$src)),
4727 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4728 def : Pat<(v4i16 (scalar_to_vector GPR:$src)),
4729 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4730 def : Pat<(v2i32 (scalar_to_vector GPR:$src)),
4731 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4733 def : Pat<(v16i8 (scalar_to_vector GPR:$src)),
4734 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4735 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4737 def : Pat<(v8i16 (scalar_to_vector GPR:$src)),
4738 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
4739 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4741 def : Pat<(v4i32 (scalar_to_vector GPR:$src)),
4742 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
4743 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4746 // VDUP : Vector Duplicate (from ARM core register to all elements)
4748 class VDUPD<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
4749 : NVDup<opcod1, 0b1011, opcod3, (outs DPR:$V), (ins GPR:$R),
4750 IIC_VMOVIS, "vdup", Dt, "$V, $R",
4751 [(set DPR:$V, (Ty (NEONvdup (i32 GPR:$R))))]>;
4752 class VDUPQ<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
4753 : NVDup<opcod1, 0b1011, opcod3, (outs QPR:$V), (ins GPR:$R),
4754 IIC_VMOVIS, "vdup", Dt, "$V, $R",
4755 [(set QPR:$V, (Ty (NEONvdup (i32 GPR:$R))))]>;
4757 def VDUP8d : VDUPD<0b11101100, 0b00, "8", v8i8>;
4758 def VDUP16d : VDUPD<0b11101000, 0b01, "16", v4i16>;
4759 def VDUP32d : VDUPD<0b11101000, 0b00, "32", v2i32>;
4760 def VDUP8q : VDUPQ<0b11101110, 0b00, "8", v16i8>;
4761 def VDUP16q : VDUPQ<0b11101010, 0b01, "16", v8i16>;
4762 def VDUP32q : VDUPQ<0b11101010, 0b00, "32", v4i32>;
4764 def : Pat<(v2f32 (NEONvdup (f32 (bitconvert GPR:$R)))), (VDUP32d GPR:$R)>;
4765 def : Pat<(v4f32 (NEONvdup (f32 (bitconvert GPR:$R)))), (VDUP32q GPR:$R)>;
4767 // VDUP : Vector Duplicate Lane (from scalar to all elements)
4769 class VDUPLND<bits<4> op19_16, string OpcodeStr, string Dt,
4770 ValueType Ty, Operand IdxTy>
4771 : NVDupLane<op19_16, 0, (outs DPR:$Vd), (ins DPR:$Vm, IdxTy:$lane),
4772 IIC_VMOVD, OpcodeStr, Dt, "$Vd, $Vm$lane",
4773 [(set DPR:$Vd, (Ty (NEONvduplane (Ty DPR:$Vm), imm:$lane)))]>;
4775 class VDUPLNQ<bits<4> op19_16, string OpcodeStr, string Dt,
4776 ValueType ResTy, ValueType OpTy, Operand IdxTy>
4777 : NVDupLane<op19_16, 1, (outs QPR:$Vd), (ins DPR:$Vm, IdxTy:$lane),
4778 IIC_VMOVQ, OpcodeStr, Dt, "$Vd, $Vm$lane",
4779 [(set QPR:$Vd, (ResTy (NEONvduplane (OpTy DPR:$Vm),
4780 VectorIndex32:$lane)))]>;
4782 // Inst{19-16} is partially specified depending on the element size.
4784 def VDUPLN8d : VDUPLND<{?,?,?,1}, "vdup", "8", v8i8, VectorIndex8> {
4786 let Inst{19-17} = lane{2-0};
4788 def VDUPLN16d : VDUPLND<{?,?,1,0}, "vdup", "16", v4i16, VectorIndex16> {
4790 let Inst{19-18} = lane{1-0};
4792 def VDUPLN32d : VDUPLND<{?,1,0,0}, "vdup", "32", v2i32, VectorIndex32> {
4794 let Inst{19} = lane{0};
4796 def VDUPLN8q : VDUPLNQ<{?,?,?,1}, "vdup", "8", v16i8, v8i8, VectorIndex8> {
4798 let Inst{19-17} = lane{2-0};
4800 def VDUPLN16q : VDUPLNQ<{?,?,1,0}, "vdup", "16", v8i16, v4i16, VectorIndex16> {
4802 let Inst{19-18} = lane{1-0};
4804 def VDUPLN32q : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4i32, v2i32, VectorIndex32> {
4806 let Inst{19} = lane{0};
4809 def : Pat<(v2f32 (NEONvduplane (v2f32 DPR:$Vm), imm:$lane)),
4810 (VDUPLN32d DPR:$Vm, imm:$lane)>;
4812 def : Pat<(v4f32 (NEONvduplane (v2f32 DPR:$Vm), imm:$lane)),
4813 (VDUPLN32q DPR:$Vm, imm:$lane)>;
4815 def : Pat<(v16i8 (NEONvduplane (v16i8 QPR:$src), imm:$lane)),
4816 (v16i8 (VDUPLN8q (v8i8 (EXTRACT_SUBREG QPR:$src,
4817 (DSubReg_i8_reg imm:$lane))),
4818 (SubReg_i8_lane imm:$lane)))>;
4819 def : Pat<(v8i16 (NEONvduplane (v8i16 QPR:$src), imm:$lane)),
4820 (v8i16 (VDUPLN16q (v4i16 (EXTRACT_SUBREG QPR:$src,
4821 (DSubReg_i16_reg imm:$lane))),
4822 (SubReg_i16_lane imm:$lane)))>;
4823 def : Pat<(v4i32 (NEONvduplane (v4i32 QPR:$src), imm:$lane)),
4824 (v4i32 (VDUPLN32q (v2i32 (EXTRACT_SUBREG QPR:$src,
4825 (DSubReg_i32_reg imm:$lane))),
4826 (SubReg_i32_lane imm:$lane)))>;
4827 def : Pat<(v4f32 (NEONvduplane (v4f32 QPR:$src), imm:$lane)),
4828 (v4f32 (VDUPLN32q (v2f32 (EXTRACT_SUBREG QPR:$src,
4829 (DSubReg_i32_reg imm:$lane))),
4830 (SubReg_i32_lane imm:$lane)))>;
4832 def VDUPfdf : PseudoNeonI<(outs DPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
4833 [(set DPR:$dst, (v2f32 (NEONvdup (f32 SPR:$src))))]>;
4834 def VDUPfqf : PseudoNeonI<(outs QPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
4835 [(set QPR:$dst, (v4f32 (NEONvdup (f32 SPR:$src))))]>;
4837 // VMOVN : Vector Narrowing Move
4838 defm VMOVN : N2VN_HSD<0b11,0b11,0b10,0b00100,0,0, IIC_VMOVN,
4839 "vmovn", "i", trunc>;
4840 // VQMOVN : Vector Saturating Narrowing Move
4841 defm VQMOVNs : N2VNInt_HSD<0b11,0b11,0b10,0b00101,0,0, IIC_VQUNAiD,
4842 "vqmovn", "s", int_arm_neon_vqmovns>;
4843 defm VQMOVNu : N2VNInt_HSD<0b11,0b11,0b10,0b00101,1,0, IIC_VQUNAiD,
4844 "vqmovn", "u", int_arm_neon_vqmovnu>;
4845 defm VQMOVNsu : N2VNInt_HSD<0b11,0b11,0b10,0b00100,1,0, IIC_VQUNAiD,
4846 "vqmovun", "s", int_arm_neon_vqmovnsu>;
4847 // VMOVL : Vector Lengthening Move
4848 defm VMOVLs : N2VL_QHS<0b01,0b10100,0,1, "vmovl", "s", sext>;
4849 defm VMOVLu : N2VL_QHS<0b11,0b10100,0,1, "vmovl", "u", zext>;
4851 // Vector Conversions.
4853 // VCVT : Vector Convert Between Floating-Point and Integers
4854 def VCVTf2sd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
4855 v2i32, v2f32, fp_to_sint>;
4856 def VCVTf2ud : N2VD<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
4857 v2i32, v2f32, fp_to_uint>;
4858 def VCVTs2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
4859 v2f32, v2i32, sint_to_fp>;
4860 def VCVTu2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
4861 v2f32, v2i32, uint_to_fp>;
4863 def VCVTf2sq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
4864 v4i32, v4f32, fp_to_sint>;
4865 def VCVTf2uq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
4866 v4i32, v4f32, fp_to_uint>;
4867 def VCVTs2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
4868 v4f32, v4i32, sint_to_fp>;
4869 def VCVTu2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
4870 v4f32, v4i32, uint_to_fp>;
4872 // VCVT : Vector Convert Between Floating-Point and Fixed-Point.
4873 let DecoderMethod = "DecodeVCVTD" in {
4874 def VCVTf2xsd : N2VCvtD<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
4875 v2i32, v2f32, int_arm_neon_vcvtfp2fxs>;
4876 def VCVTf2xud : N2VCvtD<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
4877 v2i32, v2f32, int_arm_neon_vcvtfp2fxu>;
4878 def VCVTxs2fd : N2VCvtD<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
4879 v2f32, v2i32, int_arm_neon_vcvtfxs2fp>;
4880 def VCVTxu2fd : N2VCvtD<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
4881 v2f32, v2i32, int_arm_neon_vcvtfxu2fp>;
4884 let DecoderMethod = "DecodeVCVTQ" in {
4885 def VCVTf2xsq : N2VCvtQ<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
4886 v4i32, v4f32, int_arm_neon_vcvtfp2fxs>;
4887 def VCVTf2xuq : N2VCvtQ<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
4888 v4i32, v4f32, int_arm_neon_vcvtfp2fxu>;
4889 def VCVTxs2fq : N2VCvtQ<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
4890 v4f32, v4i32, int_arm_neon_vcvtfxs2fp>;
4891 def VCVTxu2fq : N2VCvtQ<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
4892 v4f32, v4i32, int_arm_neon_vcvtfxu2fp>;
4895 // VCVT : Vector Convert Between Half-Precision and Single-Precision.
4896 def VCVTf2h : N2VNInt<0b11, 0b11, 0b01, 0b10, 0b01100, 0, 0,
4897 IIC_VUNAQ, "vcvt", "f16.f32",
4898 v4i16, v4f32, int_arm_neon_vcvtfp2hf>,
4899 Requires<[HasNEON, HasFP16]>;
4900 def VCVTh2f : N2VLInt<0b11, 0b11, 0b01, 0b10, 0b01110, 0, 0,
4901 IIC_VUNAQ, "vcvt", "f32.f16",
4902 v4f32, v4i16, int_arm_neon_vcvthf2fp>,
4903 Requires<[HasNEON, HasFP16]>;
4907 // VREV64 : Vector Reverse elements within 64-bit doublewords
4909 class VREV64D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4910 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 0, 0, (outs DPR:$Vd),
4911 (ins DPR:$Vm), IIC_VMOVD,
4912 OpcodeStr, Dt, "$Vd, $Vm", "",
4913 [(set DPR:$Vd, (Ty (NEONvrev64 (Ty DPR:$Vm))))]>;
4914 class VREV64Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4915 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 1, 0, (outs QPR:$Vd),
4916 (ins QPR:$Vm), IIC_VMOVQ,
4917 OpcodeStr, Dt, "$Vd, $Vm", "",
4918 [(set QPR:$Vd, (Ty (NEONvrev64 (Ty QPR:$Vm))))]>;
4920 def VREV64d8 : VREV64D<0b00, "vrev64", "8", v8i8>;
4921 def VREV64d16 : VREV64D<0b01, "vrev64", "16", v4i16>;
4922 def VREV64d32 : VREV64D<0b10, "vrev64", "32", v2i32>;
4923 def : Pat<(v2f32 (NEONvrev64 (v2f32 DPR:$Vm))), (VREV64d32 DPR:$Vm)>;
4925 def VREV64q8 : VREV64Q<0b00, "vrev64", "8", v16i8>;
4926 def VREV64q16 : VREV64Q<0b01, "vrev64", "16", v8i16>;
4927 def VREV64q32 : VREV64Q<0b10, "vrev64", "32", v4i32>;
4928 def : Pat<(v4f32 (NEONvrev64 (v4f32 QPR:$Vm))), (VREV64q32 QPR:$Vm)>;
4930 // VREV32 : Vector Reverse elements within 32-bit words
4932 class VREV32D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4933 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 0, 0, (outs DPR:$Vd),
4934 (ins DPR:$Vm), IIC_VMOVD,
4935 OpcodeStr, Dt, "$Vd, $Vm", "",
4936 [(set DPR:$Vd, (Ty (NEONvrev32 (Ty DPR:$Vm))))]>;
4937 class VREV32Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4938 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 1, 0, (outs QPR:$Vd),
4939 (ins QPR:$Vm), IIC_VMOVQ,
4940 OpcodeStr, Dt, "$Vd, $Vm", "",
4941 [(set QPR:$Vd, (Ty (NEONvrev32 (Ty QPR:$Vm))))]>;
4943 def VREV32d8 : VREV32D<0b00, "vrev32", "8", v8i8>;
4944 def VREV32d16 : VREV32D<0b01, "vrev32", "16", v4i16>;
4946 def VREV32q8 : VREV32Q<0b00, "vrev32", "8", v16i8>;
4947 def VREV32q16 : VREV32Q<0b01, "vrev32", "16", v8i16>;
4949 // VREV16 : Vector Reverse elements within 16-bit halfwords
4951 class VREV16D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4952 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 0, 0, (outs DPR:$Vd),
4953 (ins DPR:$Vm), IIC_VMOVD,
4954 OpcodeStr, Dt, "$Vd, $Vm", "",
4955 [(set DPR:$Vd, (Ty (NEONvrev16 (Ty DPR:$Vm))))]>;
4956 class VREV16Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4957 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 1, 0, (outs QPR:$Vd),
4958 (ins QPR:$Vm), IIC_VMOVQ,
4959 OpcodeStr, Dt, "$Vd, $Vm", "",
4960 [(set QPR:$Vd, (Ty (NEONvrev16 (Ty QPR:$Vm))))]>;
4962 def VREV16d8 : VREV16D<0b00, "vrev16", "8", v8i8>;
4963 def VREV16q8 : VREV16Q<0b00, "vrev16", "8", v16i8>;
4965 // Other Vector Shuffles.
4967 // Aligned extractions: really just dropping registers
4969 class AlignedVEXTq<ValueType DestTy, ValueType SrcTy, SDNodeXForm LaneCVT>
4970 : Pat<(DestTy (vector_extract_subvec (SrcTy QPR:$src), (i32 imm:$start))),
4971 (EXTRACT_SUBREG (SrcTy QPR:$src), (LaneCVT imm:$start))>;
4973 def : AlignedVEXTq<v8i8, v16i8, DSubReg_i8_reg>;
4975 def : AlignedVEXTq<v4i16, v8i16, DSubReg_i16_reg>;
4977 def : AlignedVEXTq<v2i32, v4i32, DSubReg_i32_reg>;
4979 def : AlignedVEXTq<v1i64, v2i64, DSubReg_f64_reg>;
4981 def : AlignedVEXTq<v2f32, v4f32, DSubReg_i32_reg>;
4984 // VEXT : Vector Extract
4986 class VEXTd<string OpcodeStr, string Dt, ValueType Ty>
4987 : N3V<0,1,0b11,{?,?,?,?},0,0, (outs DPR:$Vd),
4988 (ins DPR:$Vn, DPR:$Vm, i32imm:$index), NVExtFrm,
4989 IIC_VEXTD, OpcodeStr, Dt, "$Vd, $Vn, $Vm, $index", "",
4990 [(set DPR:$Vd, (Ty (NEONvext (Ty DPR:$Vn),
4991 (Ty DPR:$Vm), imm:$index)))]> {
4993 let Inst{11-8} = index{3-0};
4996 class VEXTq<string OpcodeStr, string Dt, ValueType Ty>
4997 : N3V<0,1,0b11,{?,?,?,?},1,0, (outs QPR:$Vd),
4998 (ins QPR:$Vn, QPR:$Vm, i32imm:$index), NVExtFrm,
4999 IIC_VEXTQ, OpcodeStr, Dt, "$Vd, $Vn, $Vm, $index", "",
5000 [(set QPR:$Vd, (Ty (NEONvext (Ty QPR:$Vn),
5001 (Ty QPR:$Vm), imm:$index)))]> {
5003 let Inst{11-8} = index{3-0};
5006 def VEXTd8 : VEXTd<"vext", "8", v8i8> {
5007 let Inst{11-8} = index{3-0};
5009 def VEXTd16 : VEXTd<"vext", "16", v4i16> {
5010 let Inst{11-9} = index{2-0};
5013 def VEXTd32 : VEXTd<"vext", "32", v2i32> {
5014 let Inst{11-10} = index{1-0};
5015 let Inst{9-8} = 0b00;
5017 def : Pat<(v2f32 (NEONvext (v2f32 DPR:$Vn),
5020 (VEXTd32 DPR:$Vn, DPR:$Vm, imm:$index)>;
5022 def VEXTq8 : VEXTq<"vext", "8", v16i8> {
5023 let Inst{11-8} = index{3-0};
5025 def VEXTq16 : VEXTq<"vext", "16", v8i16> {
5026 let Inst{11-9} = index{2-0};
5029 def VEXTq32 : VEXTq<"vext", "32", v4i32> {
5030 let Inst{11-10} = index{1-0};
5031 let Inst{9-8} = 0b00;
5033 def : Pat<(v4f32 (NEONvext (v4f32 QPR:$Vn),
5036 (VEXTq32 QPR:$Vn, QPR:$Vm, imm:$index)>;
5038 // VTRN : Vector Transpose
5040 def VTRNd8 : N2VDShuffle<0b00, 0b00001, "vtrn", "8">;
5041 def VTRNd16 : N2VDShuffle<0b01, 0b00001, "vtrn", "16">;
5042 def VTRNd32 : N2VDShuffle<0b10, 0b00001, "vtrn", "32">;
5044 def VTRNq8 : N2VQShuffle<0b00, 0b00001, IIC_VPERMQ, "vtrn", "8">;
5045 def VTRNq16 : N2VQShuffle<0b01, 0b00001, IIC_VPERMQ, "vtrn", "16">;
5046 def VTRNq32 : N2VQShuffle<0b10, 0b00001, IIC_VPERMQ, "vtrn", "32">;
5048 // VUZP : Vector Unzip (Deinterleave)
5050 def VUZPd8 : N2VDShuffle<0b00, 0b00010, "vuzp", "8">;
5051 def VUZPd16 : N2VDShuffle<0b01, 0b00010, "vuzp", "16">;
5052 def VUZPd32 : N2VDShuffle<0b10, 0b00010, "vuzp", "32">;
5054 def VUZPq8 : N2VQShuffle<0b00, 0b00010, IIC_VPERMQ3, "vuzp", "8">;
5055 def VUZPq16 : N2VQShuffle<0b01, 0b00010, IIC_VPERMQ3, "vuzp", "16">;
5056 def VUZPq32 : N2VQShuffle<0b10, 0b00010, IIC_VPERMQ3, "vuzp", "32">;
5058 // VZIP : Vector Zip (Interleave)
5060 def VZIPd8 : N2VDShuffle<0b00, 0b00011, "vzip", "8">;
5061 def VZIPd16 : N2VDShuffle<0b01, 0b00011, "vzip", "16">;
5062 def VZIPd32 : N2VDShuffle<0b10, 0b00011, "vzip", "32">;
5064 def VZIPq8 : N2VQShuffle<0b00, 0b00011, IIC_VPERMQ3, "vzip", "8">;
5065 def VZIPq16 : N2VQShuffle<0b01, 0b00011, IIC_VPERMQ3, "vzip", "16">;
5066 def VZIPq32 : N2VQShuffle<0b10, 0b00011, IIC_VPERMQ3, "vzip", "32">;
5068 // Vector Table Lookup and Table Extension.
5070 // VTBL : Vector Table Lookup
5071 let DecoderMethod = "DecodeTBLInstruction" in {
5073 : N3V<1,1,0b11,0b1000,0,0, (outs DPR:$Vd),
5074 (ins VecListOneD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTB1,
5075 "vtbl", "8", "$Vd, $Vn, $Vm", "",
5076 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbl1 VecListOneD:$Vn, DPR:$Vm)))]>;
5077 let hasExtraSrcRegAllocReq = 1 in {
5079 : N3V<1,1,0b11,0b1001,0,0, (outs DPR:$Vd),
5080 (ins DPR:$Vn, DPR:$tbl2, DPR:$Vm), NVTBLFrm, IIC_VTB2,
5081 "vtbl", "8", "$Vd, \\{$Vn, $tbl2\\}, $Vm", "", []>;
5083 : N3V<1,1,0b11,0b1010,0,0, (outs DPR:$Vd),
5084 (ins DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$Vm), NVTBLFrm, IIC_VTB3,
5085 "vtbl", "8", "$Vd, \\{$Vn, $tbl2, $tbl3\\}, $Vm", "", []>;
5087 : N3V<1,1,0b11,0b1011,0,0, (outs DPR:$Vd),
5088 (ins DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$Vm),
5090 "vtbl", "8", "$Vd, \\{$Vn, $tbl2, $tbl3, $tbl4\\}, $Vm", "", []>;
5091 } // hasExtraSrcRegAllocReq = 1
5094 : PseudoNeonI<(outs DPR:$dst), (ins QPR:$tbl, DPR:$src), IIC_VTB2, "", []>;
5096 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB3, "", []>;
5098 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB4, "", []>;
5100 // VTBX : Vector Table Extension
5102 : N3V<1,1,0b11,0b1000,1,0, (outs DPR:$Vd),
5103 (ins DPR:$orig, VecListOneD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTBX1,
5104 "vtbx", "8", "$Vd, $Vn, $Vm", "$orig = $Vd",
5105 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbx1
5106 DPR:$orig, VecListOneD:$Vn, DPR:$Vm)))]>;
5107 let hasExtraSrcRegAllocReq = 1 in {
5109 : N3V<1,1,0b11,0b1001,1,0, (outs DPR:$Vd),
5110 (ins DPR:$orig, DPR:$Vn, DPR:$tbl2, DPR:$Vm), NVTBLFrm, IIC_VTBX2,
5111 "vtbx", "8", "$Vd, \\{$Vn, $tbl2\\}, $Vm", "$orig = $Vd", []>;
5113 : N3V<1,1,0b11,0b1010,1,0, (outs DPR:$Vd),
5114 (ins DPR:$orig, DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$Vm),
5115 NVTBLFrm, IIC_VTBX3,
5116 "vtbx", "8", "$Vd, \\{$Vn, $tbl2, $tbl3\\}, $Vm",
5119 : N3V<1,1,0b11,0b1011,1,0, (outs DPR:$Vd), (ins DPR:$orig, DPR:$Vn,
5120 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$Vm), NVTBLFrm, IIC_VTBX4,
5121 "vtbx", "8", "$Vd, \\{$Vn, $tbl2, $tbl3, $tbl4\\}, $Vm",
5123 } // hasExtraSrcRegAllocReq = 1
5126 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QPR:$tbl, DPR:$src),
5127 IIC_VTBX2, "$orig = $dst", []>;
5129 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
5130 IIC_VTBX3, "$orig = $dst", []>;
5132 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
5133 IIC_VTBX4, "$orig = $dst", []>;
5134 } // DecoderMethod = "DecodeTBLInstruction"
5136 //===----------------------------------------------------------------------===//
5137 // NEON instructions for single-precision FP math
5138 //===----------------------------------------------------------------------===//
5140 class N2VSPat<SDNode OpNode, NeonI Inst>
5141 : NEONFPPat<(f32 (OpNode SPR:$a)),
5143 (v2f32 (COPY_TO_REGCLASS (Inst
5145 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5146 SPR:$a, ssub_0)), DPR_VFP2)), ssub_0)>;
5148 class N3VSPat<SDNode OpNode, NeonI Inst>
5149 : NEONFPPat<(f32 (OpNode SPR:$a, SPR:$b)),
5151 (v2f32 (COPY_TO_REGCLASS (Inst
5153 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5156 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5157 SPR:$b, ssub_0)), DPR_VFP2)), ssub_0)>;
5159 class N3VSMulOpPat<SDNode MulNode, SDNode OpNode, NeonI Inst>
5160 : NEONFPPat<(f32 (OpNode SPR:$acc, (f32 (MulNode SPR:$a, SPR:$b)))),
5162 (v2f32 (COPY_TO_REGCLASS (Inst
5164 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5167 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5170 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5171 SPR:$b, ssub_0)), DPR_VFP2)), ssub_0)>;
5173 def : N3VSPat<fadd, VADDfd>;
5174 def : N3VSPat<fsub, VSUBfd>;
5175 def : N3VSPat<fmul, VMULfd>;
5176 def : N3VSMulOpPat<fmul, fadd, VMLAfd>,
5177 Requires<[HasNEON, UseNEONForFP, UseFPVMLx]>;
5178 def : N3VSMulOpPat<fmul, fsub, VMLSfd>,
5179 Requires<[HasNEON, UseNEONForFP, UseFPVMLx]>;
5180 def : N2VSPat<fabs, VABSfd>;
5181 def : N2VSPat<fneg, VNEGfd>;
5182 def : N3VSPat<NEONfmax, VMAXfd>;
5183 def : N3VSPat<NEONfmin, VMINfd>;
5184 def : N2VSPat<arm_ftosi, VCVTf2sd>;
5185 def : N2VSPat<arm_ftoui, VCVTf2ud>;
5186 def : N2VSPat<arm_sitof, VCVTs2fd>;
5187 def : N2VSPat<arm_uitof, VCVTu2fd>;
5189 //===----------------------------------------------------------------------===//
5190 // Non-Instruction Patterns
5191 //===----------------------------------------------------------------------===//
5194 def : Pat<(v1i64 (bitconvert (v2i32 DPR:$src))), (v1i64 DPR:$src)>;
5195 def : Pat<(v1i64 (bitconvert (v4i16 DPR:$src))), (v1i64 DPR:$src)>;
5196 def : Pat<(v1i64 (bitconvert (v8i8 DPR:$src))), (v1i64 DPR:$src)>;
5197 def : Pat<(v1i64 (bitconvert (f64 DPR:$src))), (v1i64 DPR:$src)>;
5198 def : Pat<(v1i64 (bitconvert (v2f32 DPR:$src))), (v1i64 DPR:$src)>;
5199 def : Pat<(v2i32 (bitconvert (v1i64 DPR:$src))), (v2i32 DPR:$src)>;
5200 def : Pat<(v2i32 (bitconvert (v4i16 DPR:$src))), (v2i32 DPR:$src)>;
5201 def : Pat<(v2i32 (bitconvert (v8i8 DPR:$src))), (v2i32 DPR:$src)>;
5202 def : Pat<(v2i32 (bitconvert (f64 DPR:$src))), (v2i32 DPR:$src)>;
5203 def : Pat<(v2i32 (bitconvert (v2f32 DPR:$src))), (v2i32 DPR:$src)>;
5204 def : Pat<(v4i16 (bitconvert (v1i64 DPR:$src))), (v4i16 DPR:$src)>;
5205 def : Pat<(v4i16 (bitconvert (v2i32 DPR:$src))), (v4i16 DPR:$src)>;
5206 def : Pat<(v4i16 (bitconvert (v8i8 DPR:$src))), (v4i16 DPR:$src)>;
5207 def : Pat<(v4i16 (bitconvert (f64 DPR:$src))), (v4i16 DPR:$src)>;
5208 def : Pat<(v4i16 (bitconvert (v2f32 DPR:$src))), (v4i16 DPR:$src)>;
5209 def : Pat<(v8i8 (bitconvert (v1i64 DPR:$src))), (v8i8 DPR:$src)>;
5210 def : Pat<(v8i8 (bitconvert (v2i32 DPR:$src))), (v8i8 DPR:$src)>;
5211 def : Pat<(v8i8 (bitconvert (v4i16 DPR:$src))), (v8i8 DPR:$src)>;
5212 def : Pat<(v8i8 (bitconvert (f64 DPR:$src))), (v8i8 DPR:$src)>;
5213 def : Pat<(v8i8 (bitconvert (v2f32 DPR:$src))), (v8i8 DPR:$src)>;
5214 def : Pat<(f64 (bitconvert (v1i64 DPR:$src))), (f64 DPR:$src)>;
5215 def : Pat<(f64 (bitconvert (v2i32 DPR:$src))), (f64 DPR:$src)>;
5216 def : Pat<(f64 (bitconvert (v4i16 DPR:$src))), (f64 DPR:$src)>;
5217 def : Pat<(f64 (bitconvert (v8i8 DPR:$src))), (f64 DPR:$src)>;
5218 def : Pat<(f64 (bitconvert (v2f32 DPR:$src))), (f64 DPR:$src)>;
5219 def : Pat<(v2f32 (bitconvert (f64 DPR:$src))), (v2f32 DPR:$src)>;
5220 def : Pat<(v2f32 (bitconvert (v1i64 DPR:$src))), (v2f32 DPR:$src)>;
5221 def : Pat<(v2f32 (bitconvert (v2i32 DPR:$src))), (v2f32 DPR:$src)>;
5222 def : Pat<(v2f32 (bitconvert (v4i16 DPR:$src))), (v2f32 DPR:$src)>;
5223 def : Pat<(v2f32 (bitconvert (v8i8 DPR:$src))), (v2f32 DPR:$src)>;
5225 def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (v2i64 QPR:$src)>;
5226 def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (v2i64 QPR:$src)>;
5227 def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (v2i64 QPR:$src)>;
5228 def : Pat<(v2i64 (bitconvert (v2f64 QPR:$src))), (v2i64 QPR:$src)>;
5229 def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (v2i64 QPR:$src)>;
5230 def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (v4i32 QPR:$src)>;
5231 def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (v4i32 QPR:$src)>;
5232 def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (v4i32 QPR:$src)>;
5233 def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (v4i32 QPR:$src)>;
5234 def : Pat<(v4i32 (bitconvert (v4f32 QPR:$src))), (v4i32 QPR:$src)>;
5235 def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (v8i16 QPR:$src)>;
5236 def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (v8i16 QPR:$src)>;
5237 def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (v8i16 QPR:$src)>;
5238 def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (v8i16 QPR:$src)>;
5239 def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (v8i16 QPR:$src)>;
5240 def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (v16i8 QPR:$src)>;
5241 def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (v16i8 QPR:$src)>;
5242 def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (v16i8 QPR:$src)>;
5243 def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (v16i8 QPR:$src)>;
5244 def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (v16i8 QPR:$src)>;
5245 def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (v4f32 QPR:$src)>;
5246 def : Pat<(v4f32 (bitconvert (v4i32 QPR:$src))), (v4f32 QPR:$src)>;
5247 def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (v4f32 QPR:$src)>;
5248 def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (v4f32 QPR:$src)>;
5249 def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (v4f32 QPR:$src)>;
5250 def : Pat<(v2f64 (bitconvert (v2i64 QPR:$src))), (v2f64 QPR:$src)>;
5251 def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (v2f64 QPR:$src)>;
5252 def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 QPR:$src)>;
5253 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 QPR:$src)>;
5254 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 QPR:$src)>;
5257 //===----------------------------------------------------------------------===//
5258 // Assembler aliases
5261 // VAND/VEOR/VORR accept but do not require a type suffix.
5262 defm : VFPDTAnyInstAlias<"vand${p}", "$Vd, $Vn, $Vm",
5263 (VANDd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
5264 defm : VFPDTAnyInstAlias<"vand${p}", "$Vd, $Vn, $Vm",
5265 (VANDq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
5266 defm : VFPDTAnyInstAlias<"veor${p}", "$Vd, $Vn, $Vm",
5267 (VEORd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
5268 defm : VFPDTAnyInstAlias<"veor${p}", "$Vd, $Vn, $Vm",
5269 (VEORq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
5270 defm : VFPDTAnyInstAlias<"vorr${p}", "$Vd, $Vn, $Vm",
5271 (VORRd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
5272 defm : VFPDTAnyInstAlias<"vorr${p}", "$Vd, $Vn, $Vm",
5273 (VORRq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
5275 // VLD1 requires a size suffix, but also accepts type specific variants.
5276 // Load one D register.
5277 defm : VFPDT8ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5278 (VLD1d8 VecListOneD:$Vd, addrmode6:$Rn, pred:$p)>;
5279 defm : VFPDT16ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5280 (VLD1d16 VecListOneD:$Vd, addrmode6:$Rn, pred:$p)>;
5281 defm : VFPDT32ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5282 (VLD1d32 VecListOneD:$Vd, addrmode6:$Rn, pred:$p)>;
5283 defm : VFPDT64ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5284 (VLD1d64 VecListOneD:$Vd, addrmode6:$Rn, pred:$p)>;
5285 // with writeback, fixed stride
5286 defm : VFPDT8ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5287 (VLD1d8wb_fixed VecListOneD:$Vd, zero_reg, addrmode6:$Rn, pred:$p)>;
5288 defm : VFPDT16ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5289 (VLD1d16wb_fixed VecListOneD:$Vd, zero_reg, addrmode6:$Rn, pred:$p)>;
5290 defm : VFPDT32ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5291 (VLD1d32wb_fixed VecListOneD:$Vd, zero_reg, addrmode6:$Rn, pred:$p)>;
5292 defm : VFPDT64ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5293 (VLD1d64wb_fixed VecListOneD:$Vd, zero_reg, addrmode6:$Rn, pred:$p)>;
5294 // with writeback, register stride
5295 defm : VFPDT8ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5296 (VLD1d8wb_register VecListOneD:$Vd, zero_reg, addrmode6:$Rn,
5297 rGPR:$Rm, pred:$p)>;
5298 defm : VFPDT16ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5299 (VLD1d16wb_register VecListOneD:$Vd, zero_reg, addrmode6:$Rn,
5300 rGPR:$Rm, pred:$p)>;
5301 defm : VFPDT32ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5302 (VLD1d32wb_register VecListOneD:$Vd, zero_reg, addrmode6:$Rn,
5303 rGPR:$Rm, pred:$p)>;
5304 defm : VFPDT64ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5305 (VLD1d64wb_register VecListOneD:$Vd, zero_reg, addrmode6:$Rn,
5306 rGPR:$Rm, pred:$p)>;
5308 // Load two D registers.
5309 defm : VFPDT8ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5310 (VLD1q8 VecListTwoD:$Vd, addrmode6:$Rn, pred:$p)>;
5311 defm : VFPDT16ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5312 (VLD1q16 VecListTwoD:$Vd, addrmode6:$Rn, pred:$p)>;
5313 defm : VFPDT32ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5314 (VLD1q32 VecListTwoD:$Vd, addrmode6:$Rn, pred:$p)>;
5315 defm : VFPDT64ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5316 (VLD1q64 VecListTwoD:$Vd, addrmode6:$Rn, pred:$p)>;
5317 // with writeback, fixed stride
5318 defm : VFPDT8ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5319 (VLD1q8wb_fixed VecListTwoD:$Vd, zero_reg, addrmode6:$Rn, pred:$p)>;
5320 defm : VFPDT16ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5321 (VLD1q16wb_fixed VecListTwoD:$Vd, zero_reg, addrmode6:$Rn, pred:$p)>;
5322 defm : VFPDT32ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5323 (VLD1q32wb_fixed VecListTwoD:$Vd, zero_reg, addrmode6:$Rn, pred:$p)>;
5324 defm : VFPDT64ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5325 (VLD1q64wb_fixed VecListTwoD:$Vd, zero_reg, addrmode6:$Rn, pred:$p)>;
5326 // with writeback, register stride
5327 defm : VFPDT8ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5328 (VLD1q8wb_register VecListTwoD:$Vd, zero_reg, addrmode6:$Rn,
5329 rGPR:$Rm, pred:$p)>;
5330 defm : VFPDT16ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5331 (VLD1q16wb_register VecListTwoD:$Vd, zero_reg, addrmode6:$Rn,
5332 rGPR:$Rm, pred:$p)>;
5333 defm : VFPDT32ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5334 (VLD1q32wb_register VecListTwoD:$Vd, zero_reg, addrmode6:$Rn,
5335 rGPR:$Rm, pred:$p)>;
5336 defm : VFPDT64ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5337 (VLD1q64wb_register VecListTwoD:$Vd, zero_reg, addrmode6:$Rn,
5338 rGPR:$Rm, pred:$p)>;
5340 // Load three D registers.
5341 defm : VFPDT8ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5342 (VLD1d8T VecListThreeD:$Vd, addrmode6:$Rn, pred:$p)>;
5343 defm : VFPDT16ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5344 (VLD1d16T VecListThreeD:$Vd, addrmode6:$Rn, pred:$p)>;
5345 defm : VFPDT32ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5346 (VLD1d32T VecListThreeD:$Vd, addrmode6:$Rn, pred:$p)>;
5347 defm : VFPDT64ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5348 (VLD1d64T VecListThreeD:$Vd, addrmode6:$Rn, pred:$p)>;
5349 // with writeback, fixed stride
5350 defm : VFPDT8ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5351 (VLD1d8Twb_fixed VecListThreeD:$Vd, zero_reg,
5352 addrmode6:$Rn, pred:$p)>;
5353 defm : VFPDT16ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5354 (VLD1d16Twb_fixed VecListThreeD:$Vd, zero_reg,
5355 addrmode6:$Rn, pred:$p)>;
5356 defm : VFPDT32ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5357 (VLD1d32Twb_fixed VecListThreeD:$Vd, zero_reg,
5358 addrmode6:$Rn, pred:$p)>;
5359 defm : VFPDT64ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5360 (VLD1d64Twb_fixed VecListThreeD:$Vd, zero_reg,
5361 addrmode6:$Rn, pred:$p)>;
5362 // with writeback, register stride
5363 defm : VFPDT8ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5364 (VLD1d8Twb_register VecListThreeD:$Vd, zero_reg,
5365 addrmode6:$Rn, rGPR:$Rm, pred:$p)>;
5366 defm : VFPDT16ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5367 (VLD1d16Twb_register VecListThreeD:$Vd, zero_reg,
5368 addrmode6:$Rn, rGPR:$Rm, pred:$p)>;
5369 defm : VFPDT32ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5370 (VLD1d32Twb_register VecListThreeD:$Vd, zero_reg,
5371 addrmode6:$Rn, rGPR:$Rm, pred:$p)>;
5372 defm : VFPDT64ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5373 (VLD1d64Twb_register VecListThreeD:$Vd, zero_reg,
5374 addrmode6:$Rn, rGPR:$Rm, pred:$p)>;
5377 // Load four D registers.
5378 defm : VFPDT8ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5379 (VLD1d8Q VecListFourD:$Vd, addrmode6:$Rn, pred:$p)>;
5380 defm : VFPDT16ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5381 (VLD1d16Q VecListFourD:$Vd, addrmode6:$Rn, pred:$p)>;
5382 defm : VFPDT32ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5383 (VLD1d32Q VecListFourD:$Vd, addrmode6:$Rn, pred:$p)>;
5384 defm : VFPDT64ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5385 (VLD1d64Q VecListFourD:$Vd, addrmode6:$Rn, pred:$p)>;
5386 // with writeback, fixed stride
5387 defm : VFPDT8ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5388 (VLD1d8Qwb_fixed VecListFourD:$Vd, zero_reg,
5389 addrmode6:$Rn, pred:$p)>;
5390 defm : VFPDT16ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5391 (VLD1d16Qwb_fixed VecListFourD:$Vd, zero_reg,
5392 addrmode6:$Rn, pred:$p)>;
5393 defm : VFPDT32ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5394 (VLD1d32Qwb_fixed VecListFourD:$Vd, zero_reg,
5395 addrmode6:$Rn, pred:$p)>;
5396 defm : VFPDT64ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5397 (VLD1d64Qwb_fixed VecListFourD:$Vd, zero_reg,
5398 addrmode6:$Rn, pred:$p)>;
5399 // with writeback, register stride
5400 defm : VFPDT8ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5401 (VLD1d8Qwb_register VecListFourD:$Vd, zero_reg,
5402 addrmode6:$Rn, rGPR:$Rm, pred:$p)>;
5403 defm : VFPDT16ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5404 (VLD1d16Qwb_register VecListFourD:$Vd, zero_reg,
5405 addrmode6:$Rn, rGPR:$Rm, pred:$p)>;
5406 defm : VFPDT32ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5407 (VLD1d32Qwb_register VecListFourD:$Vd, zero_reg,
5408 addrmode6:$Rn, rGPR:$Rm, pred:$p)>;
5409 defm : VFPDT64ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5410 (VLD1d64Qwb_register VecListFourD:$Vd, zero_reg,
5411 addrmode6:$Rn, rGPR:$Rm, pred:$p)>;
5413 // VST1 requires a size suffix, but also accepts type specific variants.
5414 // Store one D register.
5415 defm : VFPDT8ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5416 (VST1d8 addrmode6:$Rn, VecListOneD:$Vd, pred:$p)>;
5417 defm : VFPDT16ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5418 (VST1d16 addrmode6:$Rn, VecListOneD:$Vd, pred:$p)>;
5419 defm : VFPDT32ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5420 (VST1d32 addrmode6:$Rn, VecListOneD:$Vd, pred:$p)>;
5421 defm : VFPDT64ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5422 (VST1d64 addrmode6:$Rn, VecListOneD:$Vd, pred:$p)>;
5423 // with writeback, fixed stride
5424 defm : VFPDT8ReqInstAlias<"vst1${p}", "$Vd, $Rn!",
5425 (VST1d8wb_fixed zero_reg, addrmode6:$Rn, VecListOneD:$Vd, pred:$p)>;
5426 defm : VFPDT16ReqInstAlias<"vst1${p}", "$Vd, $Rn!",
5427 (VST1d16wb_fixed zero_reg, addrmode6:$Rn, VecListOneD:$Vd, pred:$p)>;
5428 defm : VFPDT32ReqInstAlias<"vst1${p}", "$Vd, $Rn!",
5429 (VST1d32wb_fixed zero_reg, addrmode6:$Rn, VecListOneD:$Vd, pred:$p)>;
5430 defm : VFPDT64ReqInstAlias<"vst1${p}", "$Vd, $Rn!",
5431 (VST1d64wb_fixed zero_reg, addrmode6:$Rn, VecListOneD:$Vd, pred:$p)>;
5432 // with writeback, register stride
5433 defm : VFPDT8ReqInstAlias<"vst1${p}", "$Vd, $Rn, $Rm",
5434 (VST1d8wb_register zero_reg, addrmode6:$Rn, rGPR:$Rm,
5435 VecListOneD:$Vd, pred:$p)>;
5436 defm : VFPDT16ReqInstAlias<"vst1${p}", "$Vd, $Rn, $Rm",
5437 (VST1d16wb_register zero_reg, addrmode6:$Rn, rGPR:$Rm,
5438 VecListOneD:$Vd, pred:$p)>;
5439 defm : VFPDT32ReqInstAlias<"vst1${p}", "$Vd, $Rn, $Rm",
5440 (VST1d32wb_register zero_reg, addrmode6:$Rn, rGPR:$Rm,
5441 VecListOneD:$Vd, pred:$p)>;
5442 defm : VFPDT64ReqInstAlias<"vst1${p}", "$Vd, $Rn, $Rm",
5443 (VST1d64wb_register zero_reg, addrmode6:$Rn, rGPR:$Rm,
5444 VecListOneD:$Vd, pred:$p)>;
5446 // Store two D registers.
5447 defm : VFPDT8ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5448 (VST1q8 addrmode6:$Rn, VecListTwoD:$Vd, pred:$p)>;
5449 defm : VFPDT16ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5450 (VST1q16 addrmode6:$Rn, VecListTwoD:$Vd, pred:$p)>;
5451 defm : VFPDT32ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5452 (VST1q32 addrmode6:$Rn, VecListTwoD:$Vd, pred:$p)>;
5453 defm : VFPDT64ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5454 (VST1q64 addrmode6:$Rn, VecListTwoD:$Vd, pred:$p)>;
5455 // with writeback, fixed stride
5456 defm : VFPDT8ReqInstAlias<"vst1${p}", "$Vd, $Rn!",
5457 (VST1q8wb_fixed zero_reg, addrmode6:$Rn, VecListTwoD:$Vd, pred:$p)>;
5458 defm : VFPDT16ReqInstAlias<"vst1${p}", "$Vd, $Rn!",
5459 (VST1q16wb_fixed zero_reg, addrmode6:$Rn, VecListTwoD:$Vd, pred:$p)>;
5460 defm : VFPDT32ReqInstAlias<"vst1${p}", "$Vd, $Rn!",
5461 (VST1q32wb_fixed zero_reg, addrmode6:$Rn, VecListTwoD:$Vd, pred:$p)>;
5462 defm : VFPDT64ReqInstAlias<"vst1${p}", "$Vd, $Rn!",
5463 (VST1q64wb_fixed zero_reg, addrmode6:$Rn, VecListTwoD:$Vd, pred:$p)>;
5464 // with writeback, register stride
5465 defm : VFPDT8ReqInstAlias<"vst1${p}", "$Vd, $Rn, $Rm",
5466 (VST1q8wb_register zero_reg, addrmode6:$Rn,
5467 rGPR:$Rm, VecListTwoD:$Vd, pred:$p)>;
5468 defm : VFPDT16ReqInstAlias<"vst1${p}", "$Vd, $Rn, $Rm",
5469 (VST1q16wb_register zero_reg, addrmode6:$Rn,
5470 rGPR:$Rm, VecListTwoD:$Vd, pred:$p)>;
5471 defm : VFPDT32ReqInstAlias<"vst1${p}", "$Vd, $Rn, $Rm",
5472 (VST1q32wb_register zero_reg, addrmode6:$Rn,
5473 rGPR:$Rm, VecListTwoD:$Vd, pred:$p)>;
5474 defm : VFPDT64ReqInstAlias<"vst1${p}", "$Vd, $Rn, $Rm",
5475 (VST1q64wb_register zero_reg, addrmode6:$Rn,
5476 rGPR:$Rm, VecListTwoD:$Vd, pred:$p)>;
5478 // Load three D registers.
5479 defm : VFPDT8ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5480 (VST1d8T addrmode6:$Rn, VecListThreeD:$Vd, pred:$p)>;
5481 defm : VFPDT16ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5482 (VST1d16T addrmode6:$Rn, VecListThreeD:$Vd, pred:$p)>;
5483 defm : VFPDT32ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5484 (VST1d32T addrmode6:$Rn, VecListThreeD:$Vd, pred:$p)>;
5485 defm : VFPDT64ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5486 (VST1d64T addrmode6:$Rn, VecListThreeD:$Vd, pred:$p)>;
5487 defm : VFPDT8ReqInstAlias<"vst1${p}", "$Vd, $Rn!",
5488 (VST1d8Twb_fixed zero_reg, addrmode6:$Rn, VecListThreeD:$Vd, pred:$p)>;
5489 defm : VFPDT16ReqInstAlias<"vst1${p}", "$Vd, $Rn!",
5490 (VST1d16Twb_fixed zero_reg, addrmode6:$Rn, VecListThreeD:$Vd, pred:$p)>;
5491 defm : VFPDT32ReqInstAlias<"vst1${p}", "$Vd, $Rn!",
5492 (VST1d32Twb_fixed zero_reg, addrmode6:$Rn, VecListThreeD:$Vd, pred:$p)>;
5493 defm : VFPDT64ReqInstAlias<"vst1${p}", "$Vd, $Rn!",
5494 (VST1d64Twb_fixed zero_reg, addrmode6:$Rn, VecListThreeD:$Vd, pred:$p)>;
5495 defm : VFPDT8ReqInstAlias<"vst1${p}", "$Vd, $Rn, $Rm",
5496 (VST1d8Twb_register zero_reg, addrmode6:$Rn, rGPR:$Rm,
5497 VecListThreeD:$Vd, pred:$p)>;
5498 defm : VFPDT16ReqInstAlias<"vst1${p}", "$Vd, $Rn, $Rm",
5499 (VST1d16Twb_register zero_reg, addrmode6:$Rn, rGPR:$Rm,
5500 VecListThreeD:$Vd, pred:$p)>;
5501 defm : VFPDT32ReqInstAlias<"vst1${p}", "$Vd, $Rn, $Rm",
5502 (VST1d32Twb_register zero_reg, addrmode6:$Rn, rGPR:$Rm,
5503 VecListThreeD:$Vd, pred:$p)>;
5504 defm : VFPDT64ReqInstAlias<"vst1${p}", "$Vd, $Rn, $Rm",
5505 (VST1d64Twb_register zero_reg, addrmode6:$Rn, rGPR:$Rm,
5506 VecListThreeD:$Vd, pred:$p)>;
5508 // Load four D registers.
5509 defm : VFPDT8ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5510 (VST1d8Q addrmode6:$Rn, VecListFourD:$Vd, pred:$p)>;
5511 defm : VFPDT16ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5512 (VST1d16Q addrmode6:$Rn, VecListFourD:$Vd, pred:$p)>;
5513 defm : VFPDT32ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5514 (VST1d32Q addrmode6:$Rn, VecListFourD:$Vd, pred:$p)>;
5515 defm : VFPDT64ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5516 (VST1d64Q addrmode6:$Rn, VecListFourD:$Vd, pred:$p)>;
5517 defm : VFPDT8ReqInstAlias<"vst1${p}", "$Vd, $Rn!",
5518 (VST1d8Qwb_fixed zero_reg, addrmode6:$Rn, VecListFourD:$Vd, pred:$p)>;
5519 defm : VFPDT16ReqInstAlias<"vst1${p}", "$Vd, $Rn!",
5520 (VST1d16Qwb_fixed zero_reg, addrmode6:$Rn, VecListFourD:$Vd, pred:$p)>;
5521 defm : VFPDT32ReqInstAlias<"vst1${p}", "$Vd, $Rn!",
5522 (VST1d32Qwb_fixed zero_reg, addrmode6:$Rn, VecListFourD:$Vd, pred:$p)>;
5523 defm : VFPDT64ReqInstAlias<"vst1${p}", "$Vd, $Rn!",
5524 (VST1d64Qwb_fixed zero_reg, addrmode6:$Rn, VecListFourD:$Vd, pred:$p)>;
5525 defm : VFPDT8ReqInstAlias<"vst1${p}", "$Vd, $Rn, $Rm",
5526 (VST1d8Qwb_register zero_reg, addrmode6:$Rn, rGPR:$Rm,
5527 VecListFourD:$Vd, pred:$p)>;
5528 defm : VFPDT16ReqInstAlias<"vst1${p}", "$Vd, $Rn, $Rm",
5529 (VST1d16Qwb_register zero_reg, addrmode6:$Rn, rGPR:$Rm,
5530 VecListFourD:$Vd, pred:$p)>;
5531 defm : VFPDT32ReqInstAlias<"vst1${p}", "$Vd, $Rn, $Rm",
5532 (VST1d32Qwb_register zero_reg, addrmode6:$Rn, rGPR:$Rm,
5533 VecListFourD:$Vd, pred:$p)>;
5534 defm : VFPDT64ReqInstAlias<"vst1${p}", "$Vd, $Rn, $Rm",
5535 (VST1d64Qwb_register zero_reg, addrmode6:$Rn, rGPR:$Rm,
5536 VecListFourD:$Vd, pred:$p)>;
5539 // VTRN instructions data type suffix aliases for more-specific types.
5540 defm : VFPDT8ReqInstAlias <"vtrn${p}", "$Dd, $Dm",
5541 (VTRNd8 DPR:$Dd, DPR:$Dm, pred:$p)>;
5542 defm : VFPDT16ReqInstAlias<"vtrn${p}", "$Dd, $Dm",
5543 (VTRNd16 DPR:$Dd, DPR:$Dm, pred:$p)>;
5544 defm : VFPDT32ReqInstAlias<"vtrn${p}", "$Dd, $Dm",
5545 (VTRNd32 DPR:$Dd, DPR:$Dm, pred:$p)>;
5547 defm : VFPDT8ReqInstAlias <"vtrn${p}", "$Qd, $Qm",
5548 (VTRNq8 QPR:$Qd, QPR:$Qm, pred:$p)>;
5549 defm : VFPDT16ReqInstAlias<"vtrn${p}", "$Qd, $Qm",
5550 (VTRNq16 QPR:$Qd, QPR:$Qm, pred:$p)>;
5551 defm : VFPDT32ReqInstAlias<"vtrn${p}", "$Qd, $Qm",
5552 (VTRNq32 QPR:$Qd, QPR:$Qm, pred:$p)>;