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 nImmSplatI64AsmOperand : AsmOperandClass { let Name = "NEONi64splat"; }
43 def nImmSplatI64 : Operand<i32> {
44 let PrintMethod = "printNEONModImmOperand";
45 let ParserMatchClass = nImmSplatI64AsmOperand;
48 def VectorIndex8Operand : AsmOperandClass { let Name = "VectorIndex8"; }
49 def VectorIndex16Operand : AsmOperandClass { let Name = "VectorIndex16"; }
50 def VectorIndex32Operand : AsmOperandClass { let Name = "VectorIndex32"; }
51 def VectorIndex8 : Operand<i32>, ImmLeaf<i32, [{
52 return ((uint64_t)Imm) < 8;
54 let ParserMatchClass = VectorIndex8Operand;
55 let PrintMethod = "printVectorIndex";
56 let MIOperandInfo = (ops i32imm);
58 def VectorIndex16 : Operand<i32>, ImmLeaf<i32, [{
59 return ((uint64_t)Imm) < 4;
61 let ParserMatchClass = VectorIndex16Operand;
62 let PrintMethod = "printVectorIndex";
63 let MIOperandInfo = (ops i32imm);
65 def VectorIndex32 : Operand<i32>, ImmLeaf<i32, [{
66 return ((uint64_t)Imm) < 2;
68 let ParserMatchClass = VectorIndex32Operand;
69 let PrintMethod = "printVectorIndex";
70 let MIOperandInfo = (ops i32imm);
73 def VecListOneDAsmOperand : AsmOperandClass {
74 let Name = "VecListOneD";
75 let ParserMethod = "parseVectorList";
77 def VecListOneD : RegisterOperand<DPR, "printVectorListOne"> {
78 let ParserMatchClass = VecListOneDAsmOperand;
80 // Register list of two sequential D registers.
81 def VecListTwoDAsmOperand : AsmOperandClass {
82 let Name = "VecListTwoD";
83 let ParserMethod = "parseVectorList";
85 def VecListTwoD : RegisterOperand<DPR, "printVectorListTwo"> {
86 let ParserMatchClass = VecListTwoDAsmOperand;
88 // Register list of three sequential D registers.
89 def VecListThreeDAsmOperand : AsmOperandClass {
90 let Name = "VecListThreeD";
91 let ParserMethod = "parseVectorList";
93 def VecListThreeD : RegisterOperand<DPR, "printVectorListThree"> {
94 let ParserMatchClass = VecListThreeDAsmOperand;
96 // Register list of four sequential D registers.
97 def VecListFourDAsmOperand : AsmOperandClass {
98 let Name = "VecListFourD";
99 let ParserMethod = "parseVectorList";
101 def VecListFourD : RegisterOperand<DPR, "printVectorListFour"> {
102 let ParserMatchClass = VecListFourDAsmOperand;
104 // Register list of two D registers spaced by 2 (two sequential Q registers).
105 def VecListTwoQAsmOperand : AsmOperandClass {
106 let Name = "VecListTwoQ";
107 let ParserMethod = "parseVectorList";
109 def VecListTwoQ : RegisterOperand<DPR, "printVectorListTwo"> {
110 let ParserMatchClass = VecListTwoQAsmOperand;
113 //===----------------------------------------------------------------------===//
114 // NEON-specific DAG Nodes.
115 //===----------------------------------------------------------------------===//
117 def SDTARMVCMP : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisSameAs<1, 2>]>;
118 def SDTARMVCMPZ : SDTypeProfile<1, 1, []>;
120 def NEONvceq : SDNode<"ARMISD::VCEQ", SDTARMVCMP>;
121 def NEONvceqz : SDNode<"ARMISD::VCEQZ", SDTARMVCMPZ>;
122 def NEONvcge : SDNode<"ARMISD::VCGE", SDTARMVCMP>;
123 def NEONvcgez : SDNode<"ARMISD::VCGEZ", SDTARMVCMPZ>;
124 def NEONvclez : SDNode<"ARMISD::VCLEZ", SDTARMVCMPZ>;
125 def NEONvcgeu : SDNode<"ARMISD::VCGEU", SDTARMVCMP>;
126 def NEONvcgt : SDNode<"ARMISD::VCGT", SDTARMVCMP>;
127 def NEONvcgtz : SDNode<"ARMISD::VCGTZ", SDTARMVCMPZ>;
128 def NEONvcltz : SDNode<"ARMISD::VCLTZ", SDTARMVCMPZ>;
129 def NEONvcgtu : SDNode<"ARMISD::VCGTU", SDTARMVCMP>;
130 def NEONvtst : SDNode<"ARMISD::VTST", SDTARMVCMP>;
132 // Types for vector shift by immediates. The "SHX" version is for long and
133 // narrow operations where the source and destination vectors have different
134 // types. The "SHINS" version is for shift and insert operations.
135 def SDTARMVSH : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
137 def SDTARMVSHX : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
139 def SDTARMVSHINS : SDTypeProfile<1, 3, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
140 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
142 def NEONvshl : SDNode<"ARMISD::VSHL", SDTARMVSH>;
143 def NEONvshrs : SDNode<"ARMISD::VSHRs", SDTARMVSH>;
144 def NEONvshru : SDNode<"ARMISD::VSHRu", SDTARMVSH>;
145 def NEONvshlls : SDNode<"ARMISD::VSHLLs", SDTARMVSHX>;
146 def NEONvshllu : SDNode<"ARMISD::VSHLLu", SDTARMVSHX>;
147 def NEONvshlli : SDNode<"ARMISD::VSHLLi", SDTARMVSHX>;
148 def NEONvshrn : SDNode<"ARMISD::VSHRN", SDTARMVSHX>;
150 def NEONvrshrs : SDNode<"ARMISD::VRSHRs", SDTARMVSH>;
151 def NEONvrshru : SDNode<"ARMISD::VRSHRu", SDTARMVSH>;
152 def NEONvrshrn : SDNode<"ARMISD::VRSHRN", SDTARMVSHX>;
154 def NEONvqshls : SDNode<"ARMISD::VQSHLs", SDTARMVSH>;
155 def NEONvqshlu : SDNode<"ARMISD::VQSHLu", SDTARMVSH>;
156 def NEONvqshlsu : SDNode<"ARMISD::VQSHLsu", SDTARMVSH>;
157 def NEONvqshrns : SDNode<"ARMISD::VQSHRNs", SDTARMVSHX>;
158 def NEONvqshrnu : SDNode<"ARMISD::VQSHRNu", SDTARMVSHX>;
159 def NEONvqshrnsu : SDNode<"ARMISD::VQSHRNsu", SDTARMVSHX>;
161 def NEONvqrshrns : SDNode<"ARMISD::VQRSHRNs", SDTARMVSHX>;
162 def NEONvqrshrnu : SDNode<"ARMISD::VQRSHRNu", SDTARMVSHX>;
163 def NEONvqrshrnsu : SDNode<"ARMISD::VQRSHRNsu", SDTARMVSHX>;
165 def NEONvsli : SDNode<"ARMISD::VSLI", SDTARMVSHINS>;
166 def NEONvsri : SDNode<"ARMISD::VSRI", SDTARMVSHINS>;
168 def SDTARMVGETLN : SDTypeProfile<1, 2, [SDTCisVT<0, i32>, SDTCisInt<1>,
170 def NEONvgetlaneu : SDNode<"ARMISD::VGETLANEu", SDTARMVGETLN>;
171 def NEONvgetlanes : SDNode<"ARMISD::VGETLANEs", SDTARMVGETLN>;
173 def SDTARMVMOVIMM : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisVT<1, i32>]>;
174 def NEONvmovImm : SDNode<"ARMISD::VMOVIMM", SDTARMVMOVIMM>;
175 def NEONvmvnImm : SDNode<"ARMISD::VMVNIMM", SDTARMVMOVIMM>;
177 def SDTARMVORRIMM : SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
179 def NEONvorrImm : SDNode<"ARMISD::VORRIMM", SDTARMVORRIMM>;
180 def NEONvbicImm : SDNode<"ARMISD::VBICIMM", SDTARMVORRIMM>;
182 def NEONvbsl : SDNode<"ARMISD::VBSL",
183 SDTypeProfile<1, 3, [SDTCisVec<0>,
186 SDTCisSameAs<0, 3>]>>;
188 def NEONvdup : SDNode<"ARMISD::VDUP", SDTypeProfile<1, 1, [SDTCisVec<0>]>>;
190 // VDUPLANE can produce a quad-register result from a double-register source,
191 // so the result is not constrained to match the source.
192 def NEONvduplane : SDNode<"ARMISD::VDUPLANE",
193 SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisVec<1>,
196 def SDTARMVEXT : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
197 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
198 def NEONvext : SDNode<"ARMISD::VEXT", SDTARMVEXT>;
200 def SDTARMVSHUF : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisSameAs<0, 1>]>;
201 def NEONvrev64 : SDNode<"ARMISD::VREV64", SDTARMVSHUF>;
202 def NEONvrev32 : SDNode<"ARMISD::VREV32", SDTARMVSHUF>;
203 def NEONvrev16 : SDNode<"ARMISD::VREV16", SDTARMVSHUF>;
205 def SDTARMVSHUF2 : SDTypeProfile<2, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
207 SDTCisSameAs<0, 3>]>;
208 def NEONzip : SDNode<"ARMISD::VZIP", SDTARMVSHUF2>;
209 def NEONuzp : SDNode<"ARMISD::VUZP", SDTARMVSHUF2>;
210 def NEONtrn : SDNode<"ARMISD::VTRN", SDTARMVSHUF2>;
212 def SDTARMVMULL : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
213 SDTCisSameAs<1, 2>]>;
214 def NEONvmulls : SDNode<"ARMISD::VMULLs", SDTARMVMULL>;
215 def NEONvmullu : SDNode<"ARMISD::VMULLu", SDTARMVMULL>;
217 def SDTARMFMAX : SDTypeProfile<1, 2, [SDTCisVT<0, f32>, SDTCisSameAs<0, 1>,
218 SDTCisSameAs<0, 2>]>;
219 def NEONfmax : SDNode<"ARMISD::FMAX", SDTARMFMAX>;
220 def NEONfmin : SDNode<"ARMISD::FMIN", SDTARMFMAX>;
222 def NEONimmAllZerosV: PatLeaf<(NEONvmovImm (i32 timm)), [{
223 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
224 unsigned EltBits = 0;
225 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
226 return (EltBits == 32 && EltVal == 0);
229 def NEONimmAllOnesV: PatLeaf<(NEONvmovImm (i32 timm)), [{
230 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
231 unsigned EltBits = 0;
232 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
233 return (EltBits == 8 && EltVal == 0xff);
236 //===----------------------------------------------------------------------===//
237 // NEON load / store instructions
238 //===----------------------------------------------------------------------===//
240 // Use VLDM to load a Q register as a D register pair.
241 // This is a pseudo instruction that is expanded to VLDMD after reg alloc.
243 : PseudoVFPLdStM<(outs QPR:$dst), (ins GPR:$Rn),
245 [(set QPR:$dst, (v2f64 (load GPR:$Rn)))]>;
247 // Use VSTM to store a Q register as a D register pair.
248 // This is a pseudo instruction that is expanded to VSTMD after reg alloc.
250 : PseudoVFPLdStM<(outs), (ins QPR:$src, GPR:$Rn),
252 [(store (v2f64 QPR:$src), GPR:$Rn)]>;
254 // Classes for VLD* pseudo-instructions with multi-register operands.
255 // These are expanded to real instructions after register allocation.
256 class VLDQPseudo<InstrItinClass itin>
257 : PseudoNLdSt<(outs QPR:$dst), (ins addrmode6:$addr), itin, "">;
258 class VLDQWBPseudo<InstrItinClass itin>
259 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
260 (ins addrmode6:$addr, am6offset:$offset), itin,
262 class VLDQQPseudo<InstrItinClass itin>
263 : PseudoNLdSt<(outs QQPR:$dst), (ins addrmode6:$addr), itin, "">;
264 class VLDQQWBPseudo<InstrItinClass itin>
265 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
266 (ins addrmode6:$addr, am6offset:$offset), itin,
268 class VLDQQQQPseudo<InstrItinClass itin>
269 : PseudoNLdSt<(outs QQQQPR:$dst), (ins addrmode6:$addr, QQQQPR:$src),itin,
271 class VLDQQQQWBPseudo<InstrItinClass itin>
272 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
273 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
274 "$addr.addr = $wb, $src = $dst">;
276 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
278 // VLD1 : Vector Load (multiple single elements)
279 class VLD1D<bits<4> op7_4, string Dt>
280 : NLdSt<0,0b10,0b0111,op7_4, (outs VecListOneD:$Vd),
281 (ins addrmode6:$Rn), IIC_VLD1,
282 "vld1", Dt, "$Vd, $Rn", "", []> {
285 let DecoderMethod = "DecodeVLDInstruction";
287 class VLD1Q<bits<4> op7_4, string Dt>
288 : NLdSt<0,0b10,0b1010,op7_4, (outs VecListTwoD:$Vd),
289 (ins addrmode6:$Rn), IIC_VLD1x2,
290 "vld1", Dt, "$Vd, $Rn", "", []> {
292 let Inst{5-4} = Rn{5-4};
293 let DecoderMethod = "DecodeVLDInstruction";
296 def VLD1d8 : VLD1D<{0,0,0,?}, "8">;
297 def VLD1d16 : VLD1D<{0,1,0,?}, "16">;
298 def VLD1d32 : VLD1D<{1,0,0,?}, "32">;
299 def VLD1d64 : VLD1D<{1,1,0,?}, "64">;
301 def VLD1q8 : VLD1Q<{0,0,?,?}, "8">;
302 def VLD1q16 : VLD1Q<{0,1,?,?}, "16">;
303 def VLD1q32 : VLD1Q<{1,0,?,?}, "32">;
304 def VLD1q64 : VLD1Q<{1,1,?,?}, "64">;
306 def VLD1q8Pseudo : VLDQPseudo<IIC_VLD1x2>;
307 def VLD1q16Pseudo : VLDQPseudo<IIC_VLD1x2>;
308 def VLD1q32Pseudo : VLDQPseudo<IIC_VLD1x2>;
309 def VLD1q64Pseudo : VLDQPseudo<IIC_VLD1x2>;
311 // ...with address register writeback:
312 class VLD1DWB<bits<4> op7_4, string Dt>
313 : NLdSt<0,0b10,0b0111,op7_4, (outs DPR:$Vd, GPR:$wb),
314 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD1u,
315 "vld1", Dt, "\\{$Vd\\}, $Rn$Rm",
316 "$Rn.addr = $wb", []> {
318 let DecoderMethod = "DecodeVLDInstruction";
320 class VLD1QWB<bits<4> op7_4, string Dt>
321 : NLdSt<0,0b10,0b1010,op7_4, (outs VecListTwoD:$Vd, GPR:$wb),
322 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD1x2u,
323 "vld1", Dt, "$Vd, $Rn$Rm",
324 "$Rn.addr = $wb", []> {
325 let Inst{5-4} = Rn{5-4};
326 let DecoderMethod = "DecodeVLDInstruction";
329 def VLD1d8_UPD : VLD1DWB<{0,0,0,?}, "8">;
330 def VLD1d16_UPD : VLD1DWB<{0,1,0,?}, "16">;
331 def VLD1d32_UPD : VLD1DWB<{1,0,0,?}, "32">;
332 def VLD1d64_UPD : VLD1DWB<{1,1,0,?}, "64">;
334 def VLD1q8_UPD : VLD1QWB<{0,0,?,?}, "8">;
335 def VLD1q16_UPD : VLD1QWB<{0,1,?,?}, "16">;
336 def VLD1q32_UPD : VLD1QWB<{1,0,?,?}, "32">;
337 def VLD1q64_UPD : VLD1QWB<{1,1,?,?}, "64">;
339 def VLD1q8Pseudo_UPD : VLDQWBPseudo<IIC_VLD1x2u>;
340 def VLD1q16Pseudo_UPD : VLDQWBPseudo<IIC_VLD1x2u>;
341 def VLD1q32Pseudo_UPD : VLDQWBPseudo<IIC_VLD1x2u>;
342 def VLD1q64Pseudo_UPD : VLDQWBPseudo<IIC_VLD1x2u>;
344 // ...with 3 registers
345 class VLD1D3<bits<4> op7_4, string Dt>
346 : NLdSt<0,0b10,0b0110,op7_4, (outs VecListThreeD:$Vd),
347 (ins addrmode6:$Rn), IIC_VLD1x3, "vld1", Dt,
348 "$Vd, $Rn", "", []> {
351 let DecoderMethod = "DecodeVLDInstruction";
353 class VLD1D3WB<bits<4> op7_4, string Dt>
354 : NLdSt<0,0b10,0b0110,op7_4, (outs VecListThreeD:$Vd, GPR:$wb),
355 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD1x3u, "vld1", Dt,
356 "$Vd, $Rn$Rm", "$Rn.addr = $wb", []> {
358 let DecoderMethod = "DecodeVLDInstruction";
361 def VLD1d8T : VLD1D3<{0,0,0,?}, "8">;
362 def VLD1d16T : VLD1D3<{0,1,0,?}, "16">;
363 def VLD1d32T : VLD1D3<{1,0,0,?}, "32">;
364 def VLD1d64T : VLD1D3<{1,1,0,?}, "64">;
366 def VLD1d8T_UPD : VLD1D3WB<{0,0,0,?}, "8">;
367 def VLD1d16T_UPD : VLD1D3WB<{0,1,0,?}, "16">;
368 def VLD1d32T_UPD : VLD1D3WB<{1,0,0,?}, "32">;
369 def VLD1d64T_UPD : VLD1D3WB<{1,1,0,?}, "64">;
371 def VLD1d64TPseudo : VLDQQPseudo<IIC_VLD1x3>;
372 def VLD1d64TPseudo_UPD : VLDQQWBPseudo<IIC_VLD1x3u>;
374 // ...with 4 registers
375 class VLD1D4<bits<4> op7_4, string Dt>
376 : NLdSt<0, 0b10, 0b0010, op7_4, (outs VecListFourD:$Vd),
377 (ins addrmode6:$Rn), IIC_VLD1x4, "vld1", Dt,
378 "$Vd, $Rn", "", []> {
380 let Inst{5-4} = Rn{5-4};
381 let DecoderMethod = "DecodeVLDInstruction";
383 class VLD1D4WB<bits<4> op7_4, string Dt>
384 : NLdSt<0, 0b10, 0b0010, op7_4, (outs VecListFourD:$Vd, GPR:$wb),
385 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD1x4u, "vld1", Dt,
386 "$Vd, $Rn$Rm", "$Rn.addr = $wb",
388 let Inst{5-4} = Rn{5-4};
389 let DecoderMethod = "DecodeVLDInstruction";
392 def VLD1d8Q : VLD1D4<{0,0,?,?}, "8">;
393 def VLD1d16Q : VLD1D4<{0,1,?,?}, "16">;
394 def VLD1d32Q : VLD1D4<{1,0,?,?}, "32">;
395 def VLD1d64Q : VLD1D4<{1,1,?,?}, "64">;
397 def VLD1d8Q_UPD : VLD1D4WB<{0,0,?,?}, "8">;
398 def VLD1d16Q_UPD : VLD1D4WB<{0,1,?,?}, "16">;
399 def VLD1d32Q_UPD : VLD1D4WB<{1,0,?,?}, "32">;
400 def VLD1d64Q_UPD : VLD1D4WB<{1,1,?,?}, "64">;
402 def VLD1d64QPseudo : VLDQQPseudo<IIC_VLD1x4>;
403 def VLD1d64QPseudo_UPD : VLDQQWBPseudo<IIC_VLD1x4u>;
405 // VLD2 : Vector Load (multiple 2-element structures)
406 class VLD2D<bits<4> op11_8, bits<4> op7_4, string Dt, RegisterOperand VdTy>
407 : NLdSt<0, 0b10, op11_8, op7_4, (outs VdTy:$Vd),
408 (ins addrmode6:$Rn), IIC_VLD2,
409 "vld2", Dt, "$Vd, $Rn", "", []> {
411 let Inst{5-4} = Rn{5-4};
412 let DecoderMethod = "DecodeVLDInstruction";
414 class VLD2Q<bits<4> op7_4, string Dt>
415 : NLdSt<0, 0b10, 0b0011, op7_4,
416 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
417 (ins addrmode6:$Rn), IIC_VLD2x2,
418 "vld2", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn", "", []> {
420 let Inst{5-4} = Rn{5-4};
421 let DecoderMethod = "DecodeVLDInstruction";
424 def VLD2d8 : VLD2D<0b1000, {0,0,?,?}, "8", VecListTwoD>;
425 def VLD2d16 : VLD2D<0b1000, {0,1,?,?}, "16", VecListTwoD>;
426 def VLD2d32 : VLD2D<0b1000, {1,0,?,?}, "32", VecListTwoD>;
428 def VLD2q8 : VLD2Q<{0,0,?,?}, "8">;
429 def VLD2q16 : VLD2Q<{0,1,?,?}, "16">;
430 def VLD2q32 : VLD2Q<{1,0,?,?}, "32">;
432 def VLD2d8Pseudo : VLDQPseudo<IIC_VLD2>;
433 def VLD2d16Pseudo : VLDQPseudo<IIC_VLD2>;
434 def VLD2d32Pseudo : VLDQPseudo<IIC_VLD2>;
436 def VLD2q8Pseudo : VLDQQPseudo<IIC_VLD2x2>;
437 def VLD2q16Pseudo : VLDQQPseudo<IIC_VLD2x2>;
438 def VLD2q32Pseudo : VLDQQPseudo<IIC_VLD2x2>;
440 // ...with address register writeback:
441 class VLD2DWB<bits<4> op11_8, bits<4> op7_4, string Dt, RegisterOperand VdTy>
442 : NLdSt<0, 0b10, op11_8, op7_4, (outs VdTy:$Vd, GPR:$wb),
443 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD2u,
444 "vld2", Dt, "$Vd, $Rn$Rm",
445 "$Rn.addr = $wb", []> {
446 let Inst{5-4} = Rn{5-4};
447 let DecoderMethod = "DecodeVLDInstruction";
449 class VLD2QWB<bits<4> op7_4, string Dt>
450 : NLdSt<0, 0b10, 0b0011, op7_4,
451 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
452 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD2x2u,
453 "vld2", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn$Rm",
454 "$Rn.addr = $wb", []> {
455 let Inst{5-4} = Rn{5-4};
456 let DecoderMethod = "DecodeVLDInstruction";
459 def VLD2d8_UPD : VLD2DWB<0b1000, {0,0,?,?}, "8", VecListTwoD>;
460 def VLD2d16_UPD : VLD2DWB<0b1000, {0,1,?,?}, "16", VecListTwoD>;
461 def VLD2d32_UPD : VLD2DWB<0b1000, {1,0,?,?}, "32", VecListTwoD>;
463 def VLD2q8_UPD : VLD2QWB<{0,0,?,?}, "8">;
464 def VLD2q16_UPD : VLD2QWB<{0,1,?,?}, "16">;
465 def VLD2q32_UPD : VLD2QWB<{1,0,?,?}, "32">;
467 def VLD2d8Pseudo_UPD : VLDQWBPseudo<IIC_VLD2u>;
468 def VLD2d16Pseudo_UPD : VLDQWBPseudo<IIC_VLD2u>;
469 def VLD2d32Pseudo_UPD : VLDQWBPseudo<IIC_VLD2u>;
471 def VLD2q8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD2x2u>;
472 def VLD2q16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD2x2u>;
473 def VLD2q32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD2x2u>;
475 // ...with double-spaced registers
476 def VLD2b8 : VLD2D<0b1001, {0,0,?,?}, "8", VecListTwoQ>;
477 def VLD2b16 : VLD2D<0b1001, {0,1,?,?}, "16", VecListTwoQ>;
478 def VLD2b32 : VLD2D<0b1001, {1,0,?,?}, "32", VecListTwoQ>;
479 def VLD2b8_UPD : VLD2DWB<0b1001, {0,0,?,?}, "8", VecListTwoQ>;
480 def VLD2b16_UPD : VLD2DWB<0b1001, {0,1,?,?}, "16", VecListTwoQ>;
481 def VLD2b32_UPD : VLD2DWB<0b1001, {1,0,?,?}, "32", VecListTwoQ>;
483 // VLD3 : Vector Load (multiple 3-element structures)
484 class VLD3D<bits<4> op11_8, bits<4> op7_4, string Dt>
485 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
486 (ins addrmode6:$Rn), IIC_VLD3,
487 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn", "", []> {
490 let DecoderMethod = "DecodeVLDInstruction";
493 def VLD3d8 : VLD3D<0b0100, {0,0,0,?}, "8">;
494 def VLD3d16 : VLD3D<0b0100, {0,1,0,?}, "16">;
495 def VLD3d32 : VLD3D<0b0100, {1,0,0,?}, "32">;
497 def VLD3d8Pseudo : VLDQQPseudo<IIC_VLD3>;
498 def VLD3d16Pseudo : VLDQQPseudo<IIC_VLD3>;
499 def VLD3d32Pseudo : VLDQQPseudo<IIC_VLD3>;
501 // ...with address register writeback:
502 class VLD3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
503 : NLdSt<0, 0b10, op11_8, op7_4,
504 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
505 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD3u,
506 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn$Rm",
507 "$Rn.addr = $wb", []> {
509 let DecoderMethod = "DecodeVLDInstruction";
512 def VLD3d8_UPD : VLD3DWB<0b0100, {0,0,0,?}, "8">;
513 def VLD3d16_UPD : VLD3DWB<0b0100, {0,1,0,?}, "16">;
514 def VLD3d32_UPD : VLD3DWB<0b0100, {1,0,0,?}, "32">;
516 def VLD3d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
517 def VLD3d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
518 def VLD3d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
520 // ...with double-spaced registers:
521 def VLD3q8 : VLD3D<0b0101, {0,0,0,?}, "8">;
522 def VLD3q16 : VLD3D<0b0101, {0,1,0,?}, "16">;
523 def VLD3q32 : VLD3D<0b0101, {1,0,0,?}, "32">;
524 def VLD3q8_UPD : VLD3DWB<0b0101, {0,0,0,?}, "8">;
525 def VLD3q16_UPD : VLD3DWB<0b0101, {0,1,0,?}, "16">;
526 def VLD3q32_UPD : VLD3DWB<0b0101, {1,0,0,?}, "32">;
528 def VLD3q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
529 def VLD3q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
530 def VLD3q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
532 // ...alternate versions to be allocated odd register numbers:
533 def VLD3q8oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
534 def VLD3q16oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
535 def VLD3q32oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
537 def VLD3q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
538 def VLD3q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
539 def VLD3q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
541 // VLD4 : Vector Load (multiple 4-element structures)
542 class VLD4D<bits<4> op11_8, bits<4> op7_4, string Dt>
543 : NLdSt<0, 0b10, op11_8, op7_4,
544 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
545 (ins addrmode6:$Rn), IIC_VLD4,
546 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn", "", []> {
548 let Inst{5-4} = Rn{5-4};
549 let DecoderMethod = "DecodeVLDInstruction";
552 def VLD4d8 : VLD4D<0b0000, {0,0,?,?}, "8">;
553 def VLD4d16 : VLD4D<0b0000, {0,1,?,?}, "16">;
554 def VLD4d32 : VLD4D<0b0000, {1,0,?,?}, "32">;
556 def VLD4d8Pseudo : VLDQQPseudo<IIC_VLD4>;
557 def VLD4d16Pseudo : VLDQQPseudo<IIC_VLD4>;
558 def VLD4d32Pseudo : VLDQQPseudo<IIC_VLD4>;
560 // ...with address register writeback:
561 class VLD4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
562 : NLdSt<0, 0b10, op11_8, op7_4,
563 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
564 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD4u,
565 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn$Rm",
566 "$Rn.addr = $wb", []> {
567 let Inst{5-4} = Rn{5-4};
568 let DecoderMethod = "DecodeVLDInstruction";
571 def VLD4d8_UPD : VLD4DWB<0b0000, {0,0,?,?}, "8">;
572 def VLD4d16_UPD : VLD4DWB<0b0000, {0,1,?,?}, "16">;
573 def VLD4d32_UPD : VLD4DWB<0b0000, {1,0,?,?}, "32">;
575 def VLD4d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
576 def VLD4d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
577 def VLD4d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
579 // ...with double-spaced registers:
580 def VLD4q8 : VLD4D<0b0001, {0,0,?,?}, "8">;
581 def VLD4q16 : VLD4D<0b0001, {0,1,?,?}, "16">;
582 def VLD4q32 : VLD4D<0b0001, {1,0,?,?}, "32">;
583 def VLD4q8_UPD : VLD4DWB<0b0001, {0,0,?,?}, "8">;
584 def VLD4q16_UPD : VLD4DWB<0b0001, {0,1,?,?}, "16">;
585 def VLD4q32_UPD : VLD4DWB<0b0001, {1,0,?,?}, "32">;
587 def VLD4q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
588 def VLD4q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
589 def VLD4q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
591 // ...alternate versions to be allocated odd register numbers:
592 def VLD4q8oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
593 def VLD4q16oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
594 def VLD4q32oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
596 def VLD4q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
597 def VLD4q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
598 def VLD4q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
600 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
602 // Classes for VLD*LN pseudo-instructions with multi-register operands.
603 // These are expanded to real instructions after register allocation.
604 class VLDQLNPseudo<InstrItinClass itin>
605 : PseudoNLdSt<(outs QPR:$dst),
606 (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
607 itin, "$src = $dst">;
608 class VLDQLNWBPseudo<InstrItinClass itin>
609 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
610 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
611 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
612 class VLDQQLNPseudo<InstrItinClass itin>
613 : PseudoNLdSt<(outs QQPR:$dst),
614 (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
615 itin, "$src = $dst">;
616 class VLDQQLNWBPseudo<InstrItinClass itin>
617 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
618 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
619 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
620 class VLDQQQQLNPseudo<InstrItinClass itin>
621 : PseudoNLdSt<(outs QQQQPR:$dst),
622 (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
623 itin, "$src = $dst">;
624 class VLDQQQQLNWBPseudo<InstrItinClass itin>
625 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
626 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
627 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
629 // VLD1LN : Vector Load (single element to one lane)
630 class VLD1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
632 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd),
633 (ins addrmode6:$Rn, DPR:$src, nohash_imm:$lane),
634 IIC_VLD1ln, "vld1", Dt, "\\{$Vd[$lane]\\}, $Rn",
636 [(set DPR:$Vd, (vector_insert (Ty DPR:$src),
637 (i32 (LoadOp addrmode6:$Rn)),
640 let DecoderMethod = "DecodeVLD1LN";
642 class VLD1LN32<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
644 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd),
645 (ins addrmode6oneL32:$Rn, DPR:$src, nohash_imm:$lane),
646 IIC_VLD1ln, "vld1", Dt, "\\{$Vd[$lane]\\}, $Rn",
648 [(set DPR:$Vd, (vector_insert (Ty DPR:$src),
649 (i32 (LoadOp addrmode6oneL32:$Rn)),
652 let DecoderMethod = "DecodeVLD1LN";
654 class VLD1QLNPseudo<ValueType Ty, PatFrag LoadOp> : VLDQLNPseudo<IIC_VLD1ln> {
655 let Pattern = [(set QPR:$dst, (vector_insert (Ty QPR:$src),
656 (i32 (LoadOp addrmode6:$addr)),
660 def VLD1LNd8 : VLD1LN<0b0000, {?,?,?,0}, "8", v8i8, extloadi8> {
661 let Inst{7-5} = lane{2-0};
663 def VLD1LNd16 : VLD1LN<0b0100, {?,?,0,?}, "16", v4i16, extloadi16> {
664 let Inst{7-6} = lane{1-0};
667 def VLD1LNd32 : VLD1LN32<0b1000, {?,0,?,?}, "32", v2i32, load> {
668 let Inst{7} = lane{0};
673 def VLD1LNq8Pseudo : VLD1QLNPseudo<v16i8, extloadi8>;
674 def VLD1LNq16Pseudo : VLD1QLNPseudo<v8i16, extloadi16>;
675 def VLD1LNq32Pseudo : VLD1QLNPseudo<v4i32, load>;
677 def : Pat<(vector_insert (v2f32 DPR:$src),
678 (f32 (load addrmode6:$addr)), imm:$lane),
679 (VLD1LNd32 addrmode6:$addr, DPR:$src, imm:$lane)>;
680 def : Pat<(vector_insert (v4f32 QPR:$src),
681 (f32 (load addrmode6:$addr)), imm:$lane),
682 (VLD1LNq32Pseudo addrmode6:$addr, QPR:$src, imm:$lane)>;
684 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
686 // ...with address register writeback:
687 class VLD1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
688 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, GPR:$wb),
689 (ins addrmode6:$Rn, am6offset:$Rm,
690 DPR:$src, nohash_imm:$lane), IIC_VLD1lnu, "vld1", Dt,
691 "\\{$Vd[$lane]\\}, $Rn$Rm",
692 "$src = $Vd, $Rn.addr = $wb", []> {
693 let DecoderMethod = "DecodeVLD1LN";
696 def VLD1LNd8_UPD : VLD1LNWB<0b0000, {?,?,?,0}, "8"> {
697 let Inst{7-5} = lane{2-0};
699 def VLD1LNd16_UPD : VLD1LNWB<0b0100, {?,?,0,?}, "16"> {
700 let Inst{7-6} = lane{1-0};
703 def VLD1LNd32_UPD : VLD1LNWB<0b1000, {?,0,?,?}, "32"> {
704 let Inst{7} = lane{0};
709 def VLD1LNq8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
710 def VLD1LNq16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
711 def VLD1LNq32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
713 // VLD2LN : Vector Load (single 2-element structure to one lane)
714 class VLD2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
715 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2),
716 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, nohash_imm:$lane),
717 IIC_VLD2ln, "vld2", Dt, "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn",
718 "$src1 = $Vd, $src2 = $dst2", []> {
721 let DecoderMethod = "DecodeVLD2LN";
724 def VLD2LNd8 : VLD2LN<0b0001, {?,?,?,?}, "8"> {
725 let Inst{7-5} = lane{2-0};
727 def VLD2LNd16 : VLD2LN<0b0101, {?,?,0,?}, "16"> {
728 let Inst{7-6} = lane{1-0};
730 def VLD2LNd32 : VLD2LN<0b1001, {?,0,0,?}, "32"> {
731 let Inst{7} = lane{0};
734 def VLD2LNd8Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
735 def VLD2LNd16Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
736 def VLD2LNd32Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
738 // ...with double-spaced registers:
739 def VLD2LNq16 : VLD2LN<0b0101, {?,?,1,?}, "16"> {
740 let Inst{7-6} = lane{1-0};
742 def VLD2LNq32 : VLD2LN<0b1001, {?,1,0,?}, "32"> {
743 let Inst{7} = lane{0};
746 def VLD2LNq16Pseudo : VLDQQLNPseudo<IIC_VLD2ln>;
747 def VLD2LNq32Pseudo : VLDQQLNPseudo<IIC_VLD2ln>;
749 // ...with address register writeback:
750 class VLD2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
751 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
752 (ins addrmode6:$Rn, am6offset:$Rm,
753 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VLD2lnu, "vld2", Dt,
754 "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn$Rm",
755 "$src1 = $Vd, $src2 = $dst2, $Rn.addr = $wb", []> {
757 let DecoderMethod = "DecodeVLD2LN";
760 def VLD2LNd8_UPD : VLD2LNWB<0b0001, {?,?,?,?}, "8"> {
761 let Inst{7-5} = lane{2-0};
763 def VLD2LNd16_UPD : VLD2LNWB<0b0101, {?,?,0,?}, "16"> {
764 let Inst{7-6} = lane{1-0};
766 def VLD2LNd32_UPD : VLD2LNWB<0b1001, {?,0,0,?}, "32"> {
767 let Inst{7} = lane{0};
770 def VLD2LNd8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
771 def VLD2LNd16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
772 def VLD2LNd32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
774 def VLD2LNq16_UPD : VLD2LNWB<0b0101, {?,?,1,?}, "16"> {
775 let Inst{7-6} = lane{1-0};
777 def VLD2LNq32_UPD : VLD2LNWB<0b1001, {?,1,0,?}, "32"> {
778 let Inst{7} = lane{0};
781 def VLD2LNq16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
782 def VLD2LNq32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
784 // VLD3LN : Vector Load (single 3-element structure to one lane)
785 class VLD3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
786 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
787 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3,
788 nohash_imm:$lane), IIC_VLD3ln, "vld3", Dt,
789 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn",
790 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3", []> {
792 let DecoderMethod = "DecodeVLD3LN";
795 def VLD3LNd8 : VLD3LN<0b0010, {?,?,?,0}, "8"> {
796 let Inst{7-5} = lane{2-0};
798 def VLD3LNd16 : VLD3LN<0b0110, {?,?,0,0}, "16"> {
799 let Inst{7-6} = lane{1-0};
801 def VLD3LNd32 : VLD3LN<0b1010, {?,0,0,0}, "32"> {
802 let Inst{7} = lane{0};
805 def VLD3LNd8Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
806 def VLD3LNd16Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
807 def VLD3LNd32Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
809 // ...with double-spaced registers:
810 def VLD3LNq16 : VLD3LN<0b0110, {?,?,1,0}, "16"> {
811 let Inst{7-6} = lane{1-0};
813 def VLD3LNq32 : VLD3LN<0b1010, {?,1,0,0}, "32"> {
814 let Inst{7} = lane{0};
817 def VLD3LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
818 def VLD3LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
820 // ...with address register writeback:
821 class VLD3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
822 : NLdStLn<1, 0b10, op11_8, op7_4,
823 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
824 (ins addrmode6:$Rn, am6offset:$Rm,
825 DPR:$src1, DPR:$src2, DPR:$src3, nohash_imm:$lane),
826 IIC_VLD3lnu, "vld3", Dt,
827 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn$Rm",
828 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $Rn.addr = $wb",
830 let DecoderMethod = "DecodeVLD3LN";
833 def VLD3LNd8_UPD : VLD3LNWB<0b0010, {?,?,?,0}, "8"> {
834 let Inst{7-5} = lane{2-0};
836 def VLD3LNd16_UPD : VLD3LNWB<0b0110, {?,?,0,0}, "16"> {
837 let Inst{7-6} = lane{1-0};
839 def VLD3LNd32_UPD : VLD3LNWB<0b1010, {?,0,0,0}, "32"> {
840 let Inst{7} = lane{0};
843 def VLD3LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
844 def VLD3LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
845 def VLD3LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
847 def VLD3LNq16_UPD : VLD3LNWB<0b0110, {?,?,1,0}, "16"> {
848 let Inst{7-6} = lane{1-0};
850 def VLD3LNq32_UPD : VLD3LNWB<0b1010, {?,1,0,0}, "32"> {
851 let Inst{7} = lane{0};
854 def VLD3LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
855 def VLD3LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
857 // VLD4LN : Vector Load (single 4-element structure to one lane)
858 class VLD4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
859 : NLdStLn<1, 0b10, op11_8, op7_4,
860 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
861 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
862 nohash_imm:$lane), IIC_VLD4ln, "vld4", Dt,
863 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn",
864 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4", []> {
867 let DecoderMethod = "DecodeVLD4LN";
870 def VLD4LNd8 : VLD4LN<0b0011, {?,?,?,?}, "8"> {
871 let Inst{7-5} = lane{2-0};
873 def VLD4LNd16 : VLD4LN<0b0111, {?,?,0,?}, "16"> {
874 let Inst{7-6} = lane{1-0};
876 def VLD4LNd32 : VLD4LN<0b1011, {?,0,?,?}, "32"> {
877 let Inst{7} = lane{0};
881 def VLD4LNd8Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
882 def VLD4LNd16Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
883 def VLD4LNd32Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
885 // ...with double-spaced registers:
886 def VLD4LNq16 : VLD4LN<0b0111, {?,?,1,?}, "16"> {
887 let Inst{7-6} = lane{1-0};
889 def VLD4LNq32 : VLD4LN<0b1011, {?,1,?,?}, "32"> {
890 let Inst{7} = lane{0};
894 def VLD4LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>;
895 def VLD4LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>;
897 // ...with address register writeback:
898 class VLD4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
899 : NLdStLn<1, 0b10, op11_8, op7_4,
900 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
901 (ins addrmode6:$Rn, am6offset:$Rm,
902 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
903 IIC_VLD4lnu, "vld4", Dt,
904 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn$Rm",
905 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4, $Rn.addr = $wb",
908 let DecoderMethod = "DecodeVLD4LN" ;
911 def VLD4LNd8_UPD : VLD4LNWB<0b0011, {?,?,?,?}, "8"> {
912 let Inst{7-5} = lane{2-0};
914 def VLD4LNd16_UPD : VLD4LNWB<0b0111, {?,?,0,?}, "16"> {
915 let Inst{7-6} = lane{1-0};
917 def VLD4LNd32_UPD : VLD4LNWB<0b1011, {?,0,?,?}, "32"> {
918 let Inst{7} = lane{0};
922 def VLD4LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
923 def VLD4LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
924 def VLD4LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
926 def VLD4LNq16_UPD : VLD4LNWB<0b0111, {?,?,1,?}, "16"> {
927 let Inst{7-6} = lane{1-0};
929 def VLD4LNq32_UPD : VLD4LNWB<0b1011, {?,1,?,?}, "32"> {
930 let Inst{7} = lane{0};
934 def VLD4LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
935 def VLD4LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
937 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
939 // VLD1DUP : Vector Load (single element to all lanes)
940 class VLD1DUP<bits<4> op7_4, string Dt, ValueType Ty, PatFrag LoadOp>
941 : NLdSt<1, 0b10, 0b1100, op7_4, (outs DPR:$Vd), (ins addrmode6dup:$Rn),
942 IIC_VLD1dup, "vld1", Dt, "\\{$Vd[]\\}, $Rn", "",
943 [(set DPR:$Vd, (Ty (NEONvdup (i32 (LoadOp addrmode6dup:$Rn)))))]> {
946 let DecoderMethod = "DecodeVLD1DupInstruction";
948 class VLD1QDUPPseudo<ValueType Ty, PatFrag LoadOp> : VLDQPseudo<IIC_VLD1dup> {
949 let Pattern = [(set QPR:$dst,
950 (Ty (NEONvdup (i32 (LoadOp addrmode6dup:$addr)))))];
953 def VLD1DUPd8 : VLD1DUP<{0,0,0,?}, "8", v8i8, extloadi8>;
954 def VLD1DUPd16 : VLD1DUP<{0,1,0,?}, "16", v4i16, extloadi16>;
955 def VLD1DUPd32 : VLD1DUP<{1,0,0,?}, "32", v2i32, load>;
957 def VLD1DUPq8Pseudo : VLD1QDUPPseudo<v16i8, extloadi8>;
958 def VLD1DUPq16Pseudo : VLD1QDUPPseudo<v8i16, extloadi16>;
959 def VLD1DUPq32Pseudo : VLD1QDUPPseudo<v4i32, load>;
961 def : Pat<(v2f32 (NEONvdup (f32 (load addrmode6dup:$addr)))),
962 (VLD1DUPd32 addrmode6:$addr)>;
963 def : Pat<(v4f32 (NEONvdup (f32 (load addrmode6dup:$addr)))),
964 (VLD1DUPq32Pseudo addrmode6:$addr)>;
966 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
968 class VLD1QDUP<bits<4> op7_4, string Dt>
969 : NLdSt<1, 0b10, 0b1100, op7_4, (outs DPR:$Vd, DPR:$dst2),
970 (ins addrmode6dup:$Rn), IIC_VLD1dup,
971 "vld1", Dt, "\\{$Vd[], $dst2[]\\}, $Rn", "", []> {
974 let DecoderMethod = "DecodeVLD1DupInstruction";
977 def VLD1DUPq8 : VLD1QDUP<{0,0,1,0}, "8">;
978 def VLD1DUPq16 : VLD1QDUP<{0,1,1,?}, "16">;
979 def VLD1DUPq32 : VLD1QDUP<{1,0,1,?}, "32">;
981 // ...with address register writeback:
982 class VLD1DUPWB<bits<4> op7_4, string Dt>
983 : NLdSt<1, 0b10, 0b1100, op7_4, (outs DPR:$Vd, GPR:$wb),
984 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD1dupu,
985 "vld1", Dt, "\\{$Vd[]\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
987 let DecoderMethod = "DecodeVLD1DupInstruction";
989 class VLD1QDUPWB<bits<4> op7_4, string Dt>
990 : NLdSt<1, 0b10, 0b1100, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
991 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD1dupu,
992 "vld1", Dt, "\\{$Vd[], $dst2[]\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
994 let DecoderMethod = "DecodeVLD1DupInstruction";
997 def VLD1DUPd8_UPD : VLD1DUPWB<{0,0,0,0}, "8">;
998 def VLD1DUPd16_UPD : VLD1DUPWB<{0,1,0,?}, "16">;
999 def VLD1DUPd32_UPD : VLD1DUPWB<{1,0,0,?}, "32">;
1001 def VLD1DUPq8_UPD : VLD1QDUPWB<{0,0,1,0}, "8">;
1002 def VLD1DUPq16_UPD : VLD1QDUPWB<{0,1,1,?}, "16">;
1003 def VLD1DUPq32_UPD : VLD1QDUPWB<{1,0,1,?}, "32">;
1005 def VLD1DUPq8Pseudo_UPD : VLDQWBPseudo<IIC_VLD1dupu>;
1006 def VLD1DUPq16Pseudo_UPD : VLDQWBPseudo<IIC_VLD1dupu>;
1007 def VLD1DUPq32Pseudo_UPD : VLDQWBPseudo<IIC_VLD1dupu>;
1009 // VLD2DUP : Vector Load (single 2-element structure to all lanes)
1010 class VLD2DUP<bits<4> op7_4, string Dt>
1011 : NLdSt<1, 0b10, 0b1101, op7_4, (outs DPR:$Vd, DPR:$dst2),
1012 (ins addrmode6dup:$Rn), IIC_VLD2dup,
1013 "vld2", Dt, "\\{$Vd[], $dst2[]\\}, $Rn", "", []> {
1015 let Inst{4} = Rn{4};
1016 let DecoderMethod = "DecodeVLD2DupInstruction";
1019 def VLD2DUPd8 : VLD2DUP<{0,0,0,?}, "8">;
1020 def VLD2DUPd16 : VLD2DUP<{0,1,0,?}, "16">;
1021 def VLD2DUPd32 : VLD2DUP<{1,0,0,?}, "32">;
1023 def VLD2DUPd8Pseudo : VLDQPseudo<IIC_VLD2dup>;
1024 def VLD2DUPd16Pseudo : VLDQPseudo<IIC_VLD2dup>;
1025 def VLD2DUPd32Pseudo : VLDQPseudo<IIC_VLD2dup>;
1027 // ...with double-spaced registers (not used for codegen):
1028 def VLD2DUPd8x2 : VLD2DUP<{0,0,1,?}, "8">;
1029 def VLD2DUPd16x2 : VLD2DUP<{0,1,1,?}, "16">;
1030 def VLD2DUPd32x2 : VLD2DUP<{1,0,1,?}, "32">;
1032 // ...with address register writeback:
1033 class VLD2DUPWB<bits<4> op7_4, string Dt>
1034 : NLdSt<1, 0b10, 0b1101, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
1035 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD2dupu,
1036 "vld2", Dt, "\\{$Vd[], $dst2[]\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
1037 let Inst{4} = Rn{4};
1038 let DecoderMethod = "DecodeVLD2DupInstruction";
1041 def VLD2DUPd8_UPD : VLD2DUPWB<{0,0,0,0}, "8">;
1042 def VLD2DUPd16_UPD : VLD2DUPWB<{0,1,0,?}, "16">;
1043 def VLD2DUPd32_UPD : VLD2DUPWB<{1,0,0,?}, "32">;
1045 def VLD2DUPd8x2_UPD : VLD2DUPWB<{0,0,1,0}, "8">;
1046 def VLD2DUPd16x2_UPD : VLD2DUPWB<{0,1,1,?}, "16">;
1047 def VLD2DUPd32x2_UPD : VLD2DUPWB<{1,0,1,?}, "32">;
1049 def VLD2DUPd8Pseudo_UPD : VLDQWBPseudo<IIC_VLD2dupu>;
1050 def VLD2DUPd16Pseudo_UPD : VLDQWBPseudo<IIC_VLD2dupu>;
1051 def VLD2DUPd32Pseudo_UPD : VLDQWBPseudo<IIC_VLD2dupu>;
1053 // VLD3DUP : Vector Load (single 3-element structure to all lanes)
1054 class VLD3DUP<bits<4> op7_4, string Dt>
1055 : NLdSt<1, 0b10, 0b1110, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
1056 (ins addrmode6dup:$Rn), IIC_VLD3dup,
1057 "vld3", Dt, "\\{$Vd[], $dst2[], $dst3[]\\}, $Rn", "", []> {
1060 let DecoderMethod = "DecodeVLD3DupInstruction";
1063 def VLD3DUPd8 : VLD3DUP<{0,0,0,?}, "8">;
1064 def VLD3DUPd16 : VLD3DUP<{0,1,0,?}, "16">;
1065 def VLD3DUPd32 : VLD3DUP<{1,0,0,?}, "32">;
1067 def VLD3DUPd8Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1068 def VLD3DUPd16Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1069 def VLD3DUPd32Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1071 // ...with double-spaced registers (not used for codegen):
1072 def VLD3DUPd8x2 : VLD3DUP<{0,0,1,?}, "8">;
1073 def VLD3DUPd16x2 : VLD3DUP<{0,1,1,?}, "16">;
1074 def VLD3DUPd32x2 : VLD3DUP<{1,0,1,?}, "32">;
1076 // ...with address register writeback:
1077 class VLD3DUPWB<bits<4> op7_4, string Dt>
1078 : NLdSt<1, 0b10, 0b1110, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
1079 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD3dupu,
1080 "vld3", Dt, "\\{$Vd[], $dst2[], $dst3[]\\}, $Rn$Rm",
1081 "$Rn.addr = $wb", []> {
1083 let DecoderMethod = "DecodeVLD3DupInstruction";
1086 def VLD3DUPd8_UPD : VLD3DUPWB<{0,0,0,0}, "8">;
1087 def VLD3DUPd16_UPD : VLD3DUPWB<{0,1,0,?}, "16">;
1088 def VLD3DUPd32_UPD : VLD3DUPWB<{1,0,0,?}, "32">;
1090 def VLD3DUPd8x2_UPD : VLD3DUPWB<{0,0,1,0}, "8">;
1091 def VLD3DUPd16x2_UPD : VLD3DUPWB<{0,1,1,?}, "16">;
1092 def VLD3DUPd32x2_UPD : VLD3DUPWB<{1,0,1,?}, "32">;
1094 def VLD3DUPd8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1095 def VLD3DUPd16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1096 def VLD3DUPd32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1098 // VLD4DUP : Vector Load (single 4-element structure to all lanes)
1099 class VLD4DUP<bits<4> op7_4, string Dt>
1100 : NLdSt<1, 0b10, 0b1111, op7_4,
1101 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
1102 (ins addrmode6dup:$Rn), IIC_VLD4dup,
1103 "vld4", Dt, "\\{$Vd[], $dst2[], $dst3[], $dst4[]\\}, $Rn", "", []> {
1105 let Inst{4} = Rn{4};
1106 let DecoderMethod = "DecodeVLD4DupInstruction";
1109 def VLD4DUPd8 : VLD4DUP<{0,0,0,?}, "8">;
1110 def VLD4DUPd16 : VLD4DUP<{0,1,0,?}, "16">;
1111 def VLD4DUPd32 : VLD4DUP<{1,?,0,?}, "32"> { let Inst{6} = Rn{5}; }
1113 def VLD4DUPd8Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1114 def VLD4DUPd16Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1115 def VLD4DUPd32Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1117 // ...with double-spaced registers (not used for codegen):
1118 def VLD4DUPd8x2 : VLD4DUP<{0,0,1,?}, "8">;
1119 def VLD4DUPd16x2 : VLD4DUP<{0,1,1,?}, "16">;
1120 def VLD4DUPd32x2 : VLD4DUP<{1,?,1,?}, "32"> { let Inst{6} = Rn{5}; }
1122 // ...with address register writeback:
1123 class VLD4DUPWB<bits<4> op7_4, string Dt>
1124 : NLdSt<1, 0b10, 0b1111, op7_4,
1125 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
1126 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD4dupu,
1127 "vld4", Dt, "\\{$Vd[], $dst2[], $dst3[], $dst4[]\\}, $Rn$Rm",
1128 "$Rn.addr = $wb", []> {
1129 let Inst{4} = Rn{4};
1130 let DecoderMethod = "DecodeVLD4DupInstruction";
1133 def VLD4DUPd8_UPD : VLD4DUPWB<{0,0,0,0}, "8">;
1134 def VLD4DUPd16_UPD : VLD4DUPWB<{0,1,0,?}, "16">;
1135 def VLD4DUPd32_UPD : VLD4DUPWB<{1,?,0,?}, "32"> { let Inst{6} = Rn{5}; }
1137 def VLD4DUPd8x2_UPD : VLD4DUPWB<{0,0,1,0}, "8">;
1138 def VLD4DUPd16x2_UPD : VLD4DUPWB<{0,1,1,?}, "16">;
1139 def VLD4DUPd32x2_UPD : VLD4DUPWB<{1,?,1,?}, "32"> { let Inst{6} = Rn{5}; }
1141 def VLD4DUPd8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1142 def VLD4DUPd16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1143 def VLD4DUPd32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1145 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
1147 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
1149 // Classes for VST* pseudo-instructions with multi-register operands.
1150 // These are expanded to real instructions after register allocation.
1151 class VSTQPseudo<InstrItinClass itin>
1152 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src), itin, "">;
1153 class VSTQWBPseudo<InstrItinClass itin>
1154 : PseudoNLdSt<(outs GPR:$wb),
1155 (ins addrmode6:$addr, am6offset:$offset, QPR:$src), itin,
1156 "$addr.addr = $wb">;
1157 class VSTQQPseudo<InstrItinClass itin>
1158 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src), itin, "">;
1159 class VSTQQWBPseudo<InstrItinClass itin>
1160 : PseudoNLdSt<(outs GPR:$wb),
1161 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src), itin,
1162 "$addr.addr = $wb">;
1163 class VSTQQQQPseudo<InstrItinClass itin>
1164 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src), itin, "">;
1165 class VSTQQQQWBPseudo<InstrItinClass itin>
1166 : PseudoNLdSt<(outs GPR:$wb),
1167 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
1168 "$addr.addr = $wb">;
1170 // VST1 : Vector Store (multiple single elements)
1171 class VST1D<bits<4> op7_4, string Dt>
1172 : NLdSt<0,0b00,0b0111,op7_4, (outs), (ins addrmode6:$Rn, VecListOneD:$Vd),
1173 IIC_VST1, "vst1", Dt, "$Vd, $Rn", "", []> {
1175 let Inst{4} = Rn{4};
1176 let DecoderMethod = "DecodeVSTInstruction";
1178 class VST1Q<bits<4> op7_4, string Dt>
1179 : NLdSt<0,0b00,0b1010,op7_4, (outs),
1180 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2), IIC_VST1x2,
1181 "vst1", Dt, "\\{$Vd, $src2\\}, $Rn", "", []> {
1183 let Inst{5-4} = Rn{5-4};
1184 let DecoderMethod = "DecodeVSTInstruction";
1187 def VST1d8 : VST1D<{0,0,0,?}, "8">;
1188 def VST1d16 : VST1D<{0,1,0,?}, "16">;
1189 def VST1d32 : VST1D<{1,0,0,?}, "32">;
1190 def VST1d64 : VST1D<{1,1,0,?}, "64">;
1192 def VST1q8 : VST1Q<{0,0,?,?}, "8">;
1193 def VST1q16 : VST1Q<{0,1,?,?}, "16">;
1194 def VST1q32 : VST1Q<{1,0,?,?}, "32">;
1195 def VST1q64 : VST1Q<{1,1,?,?}, "64">;
1197 def VST1q8Pseudo : VSTQPseudo<IIC_VST1x2>;
1198 def VST1q16Pseudo : VSTQPseudo<IIC_VST1x2>;
1199 def VST1q32Pseudo : VSTQPseudo<IIC_VST1x2>;
1200 def VST1q64Pseudo : VSTQPseudo<IIC_VST1x2>;
1202 // ...with address register writeback:
1203 class VST1DWB<bits<4> op7_4, string Dt>
1204 : NLdSt<0, 0b00, 0b0111, op7_4, (outs GPR:$wb),
1205 (ins addrmode6:$Rn, am6offset:$Rm, DPR:$Vd), IIC_VST1u,
1206 "vst1", Dt, "\\{$Vd\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
1207 let Inst{4} = Rn{4};
1208 let DecoderMethod = "DecodeVSTInstruction";
1210 class VST1QWB<bits<4> op7_4, string Dt>
1211 : NLdSt<0, 0b00, 0b1010, op7_4, (outs GPR:$wb),
1212 (ins addrmode6:$Rn, am6offset:$Rm, DPR:$Vd, DPR:$src2),
1213 IIC_VST1x2u, "vst1", Dt, "\\{$Vd, $src2\\}, $Rn$Rm",
1214 "$Rn.addr = $wb", []> {
1215 let Inst{5-4} = Rn{5-4};
1216 let DecoderMethod = "DecodeVSTInstruction";
1219 def VST1d8_UPD : VST1DWB<{0,0,0,?}, "8">;
1220 def VST1d16_UPD : VST1DWB<{0,1,0,?}, "16">;
1221 def VST1d32_UPD : VST1DWB<{1,0,0,?}, "32">;
1222 def VST1d64_UPD : VST1DWB<{1,1,0,?}, "64">;
1224 def VST1q8_UPD : VST1QWB<{0,0,?,?}, "8">;
1225 def VST1q16_UPD : VST1QWB<{0,1,?,?}, "16">;
1226 def VST1q32_UPD : VST1QWB<{1,0,?,?}, "32">;
1227 def VST1q64_UPD : VST1QWB<{1,1,?,?}, "64">;
1229 def VST1q8Pseudo_UPD : VSTQWBPseudo<IIC_VST1x2u>;
1230 def VST1q16Pseudo_UPD : VSTQWBPseudo<IIC_VST1x2u>;
1231 def VST1q32Pseudo_UPD : VSTQWBPseudo<IIC_VST1x2u>;
1232 def VST1q64Pseudo_UPD : VSTQWBPseudo<IIC_VST1x2u>;
1234 // ...with 3 registers
1235 class VST1D3<bits<4> op7_4, string Dt>
1236 : NLdSt<0, 0b00, 0b0110, op7_4, (outs),
1237 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3),
1238 IIC_VST1x3, "vst1", Dt, "\\{$Vd, $src2, $src3\\}, $Rn", "", []> {
1240 let Inst{4} = Rn{4};
1241 let DecoderMethod = "DecodeVSTInstruction";
1243 class VST1D3WB<bits<4> op7_4, string Dt>
1244 : NLdSt<0, 0b00, 0b0110, op7_4, (outs GPR:$wb),
1245 (ins addrmode6:$Rn, am6offset:$Rm,
1246 DPR:$Vd, DPR:$src2, DPR:$src3),
1247 IIC_VST1x3u, "vst1", Dt, "\\{$Vd, $src2, $src3\\}, $Rn$Rm",
1248 "$Rn.addr = $wb", []> {
1249 let Inst{4} = Rn{4};
1250 let DecoderMethod = "DecodeVSTInstruction";
1253 def VST1d8T : VST1D3<{0,0,0,?}, "8">;
1254 def VST1d16T : VST1D3<{0,1,0,?}, "16">;
1255 def VST1d32T : VST1D3<{1,0,0,?}, "32">;
1256 def VST1d64T : VST1D3<{1,1,0,?}, "64">;
1258 def VST1d8T_UPD : VST1D3WB<{0,0,0,?}, "8">;
1259 def VST1d16T_UPD : VST1D3WB<{0,1,0,?}, "16">;
1260 def VST1d32T_UPD : VST1D3WB<{1,0,0,?}, "32">;
1261 def VST1d64T_UPD : VST1D3WB<{1,1,0,?}, "64">;
1263 def VST1d64TPseudo : VSTQQPseudo<IIC_VST1x3>;
1264 def VST1d64TPseudo_UPD : VSTQQWBPseudo<IIC_VST1x3u>;
1266 // ...with 4 registers
1267 class VST1D4<bits<4> op7_4, string Dt>
1268 : NLdSt<0, 0b00, 0b0010, op7_4, (outs),
1269 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
1270 IIC_VST1x4, "vst1", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn", "",
1273 let Inst{5-4} = Rn{5-4};
1274 let DecoderMethod = "DecodeVSTInstruction";
1276 class VST1D4WB<bits<4> op7_4, string Dt>
1277 : NLdSt<0, 0b00, 0b0010, op7_4, (outs GPR:$wb),
1278 (ins addrmode6:$Rn, am6offset:$Rm,
1279 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST1x4u,
1280 "vst1", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
1281 "$Rn.addr = $wb", []> {
1282 let Inst{5-4} = Rn{5-4};
1283 let DecoderMethod = "DecodeVSTInstruction";
1286 def VST1d8Q : VST1D4<{0,0,?,?}, "8">;
1287 def VST1d16Q : VST1D4<{0,1,?,?}, "16">;
1288 def VST1d32Q : VST1D4<{1,0,?,?}, "32">;
1289 def VST1d64Q : VST1D4<{1,1,?,?}, "64">;
1291 def VST1d8Q_UPD : VST1D4WB<{0,0,?,?}, "8">;
1292 def VST1d16Q_UPD : VST1D4WB<{0,1,?,?}, "16">;
1293 def VST1d32Q_UPD : VST1D4WB<{1,0,?,?}, "32">;
1294 def VST1d64Q_UPD : VST1D4WB<{1,1,?,?}, "64">;
1296 def VST1d64QPseudo : VSTQQPseudo<IIC_VST1x4>;
1297 def VST1d64QPseudo_UPD : VSTQQWBPseudo<IIC_VST1x4u>;
1299 // VST2 : Vector Store (multiple 2-element structures)
1300 class VST2D<bits<4> op11_8, bits<4> op7_4, string Dt>
1301 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1302 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2),
1303 IIC_VST2, "vst2", Dt, "\\{$Vd, $src2\\}, $Rn", "", []> {
1305 let Inst{5-4} = Rn{5-4};
1306 let DecoderMethod = "DecodeVSTInstruction";
1308 class VST2Q<bits<4> op7_4, string Dt>
1309 : NLdSt<0, 0b00, 0b0011, op7_4, (outs),
1310 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
1311 IIC_VST2x2, "vst2", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn",
1314 let Inst{5-4} = Rn{5-4};
1315 let DecoderMethod = "DecodeVSTInstruction";
1318 def VST2d8 : VST2D<0b1000, {0,0,?,?}, "8">;
1319 def VST2d16 : VST2D<0b1000, {0,1,?,?}, "16">;
1320 def VST2d32 : VST2D<0b1000, {1,0,?,?}, "32">;
1322 def VST2q8 : VST2Q<{0,0,?,?}, "8">;
1323 def VST2q16 : VST2Q<{0,1,?,?}, "16">;
1324 def VST2q32 : VST2Q<{1,0,?,?}, "32">;
1326 def VST2d8Pseudo : VSTQPseudo<IIC_VST2>;
1327 def VST2d16Pseudo : VSTQPseudo<IIC_VST2>;
1328 def VST2d32Pseudo : VSTQPseudo<IIC_VST2>;
1330 def VST2q8Pseudo : VSTQQPseudo<IIC_VST2x2>;
1331 def VST2q16Pseudo : VSTQQPseudo<IIC_VST2x2>;
1332 def VST2q32Pseudo : VSTQQPseudo<IIC_VST2x2>;
1334 // ...with address register writeback:
1335 class VST2DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1336 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1337 (ins addrmode6:$Rn, am6offset:$Rm, DPR:$Vd, DPR:$src2),
1338 IIC_VST2u, "vst2", Dt, "\\{$Vd, $src2\\}, $Rn$Rm",
1339 "$Rn.addr = $wb", []> {
1340 let Inst{5-4} = Rn{5-4};
1341 let DecoderMethod = "DecodeVSTInstruction";
1343 class VST2QWB<bits<4> op7_4, string Dt>
1344 : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
1345 (ins addrmode6:$Rn, am6offset:$Rm,
1346 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST2x2u,
1347 "vst2", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
1348 "$Rn.addr = $wb", []> {
1349 let Inst{5-4} = Rn{5-4};
1350 let DecoderMethod = "DecodeVSTInstruction";
1353 def VST2d8_UPD : VST2DWB<0b1000, {0,0,?,?}, "8">;
1354 def VST2d16_UPD : VST2DWB<0b1000, {0,1,?,?}, "16">;
1355 def VST2d32_UPD : VST2DWB<0b1000, {1,0,?,?}, "32">;
1357 def VST2q8_UPD : VST2QWB<{0,0,?,?}, "8">;
1358 def VST2q16_UPD : VST2QWB<{0,1,?,?}, "16">;
1359 def VST2q32_UPD : VST2QWB<{1,0,?,?}, "32">;
1361 def VST2d8Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
1362 def VST2d16Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
1363 def VST2d32Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
1365 def VST2q8Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
1366 def VST2q16Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
1367 def VST2q32Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
1369 // ...with double-spaced registers
1370 def VST2b8 : VST2D<0b1001, {0,0,?,?}, "8">;
1371 def VST2b16 : VST2D<0b1001, {0,1,?,?}, "16">;
1372 def VST2b32 : VST2D<0b1001, {1,0,?,?}, "32">;
1373 def VST2b8_UPD : VST2DWB<0b1001, {0,0,?,?}, "8">;
1374 def VST2b16_UPD : VST2DWB<0b1001, {0,1,?,?}, "16">;
1375 def VST2b32_UPD : VST2DWB<0b1001, {1,0,?,?}, "32">;
1377 // VST3 : Vector Store (multiple 3-element structures)
1378 class VST3D<bits<4> op11_8, bits<4> op7_4, string Dt>
1379 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1380 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3,
1381 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn", "", []> {
1383 let Inst{4} = Rn{4};
1384 let DecoderMethod = "DecodeVSTInstruction";
1387 def VST3d8 : VST3D<0b0100, {0,0,0,?}, "8">;
1388 def VST3d16 : VST3D<0b0100, {0,1,0,?}, "16">;
1389 def VST3d32 : VST3D<0b0100, {1,0,0,?}, "32">;
1391 def VST3d8Pseudo : VSTQQPseudo<IIC_VST3>;
1392 def VST3d16Pseudo : VSTQQPseudo<IIC_VST3>;
1393 def VST3d32Pseudo : VSTQQPseudo<IIC_VST3>;
1395 // ...with address register writeback:
1396 class VST3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1397 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1398 (ins addrmode6:$Rn, am6offset:$Rm,
1399 DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3u,
1400 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn$Rm",
1401 "$Rn.addr = $wb", []> {
1402 let Inst{4} = Rn{4};
1403 let DecoderMethod = "DecodeVSTInstruction";
1406 def VST3d8_UPD : VST3DWB<0b0100, {0,0,0,?}, "8">;
1407 def VST3d16_UPD : VST3DWB<0b0100, {0,1,0,?}, "16">;
1408 def VST3d32_UPD : VST3DWB<0b0100, {1,0,0,?}, "32">;
1410 def VST3d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1411 def VST3d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1412 def VST3d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1414 // ...with double-spaced registers:
1415 def VST3q8 : VST3D<0b0101, {0,0,0,?}, "8">;
1416 def VST3q16 : VST3D<0b0101, {0,1,0,?}, "16">;
1417 def VST3q32 : VST3D<0b0101, {1,0,0,?}, "32">;
1418 def VST3q8_UPD : VST3DWB<0b0101, {0,0,0,?}, "8">;
1419 def VST3q16_UPD : VST3DWB<0b0101, {0,1,0,?}, "16">;
1420 def VST3q32_UPD : VST3DWB<0b0101, {1,0,0,?}, "32">;
1422 def VST3q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1423 def VST3q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1424 def VST3q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1426 // ...alternate versions to be allocated odd register numbers:
1427 def VST3q8oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1428 def VST3q16oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1429 def VST3q32oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1431 def VST3q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1432 def VST3q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1433 def VST3q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1435 // VST4 : Vector Store (multiple 4-element structures)
1436 class VST4D<bits<4> op11_8, bits<4> op7_4, string Dt>
1437 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1438 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
1439 IIC_VST4, "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn",
1442 let Inst{5-4} = Rn{5-4};
1443 let DecoderMethod = "DecodeVSTInstruction";
1446 def VST4d8 : VST4D<0b0000, {0,0,?,?}, "8">;
1447 def VST4d16 : VST4D<0b0000, {0,1,?,?}, "16">;
1448 def VST4d32 : VST4D<0b0000, {1,0,?,?}, "32">;
1450 def VST4d8Pseudo : VSTQQPseudo<IIC_VST4>;
1451 def VST4d16Pseudo : VSTQQPseudo<IIC_VST4>;
1452 def VST4d32Pseudo : VSTQQPseudo<IIC_VST4>;
1454 // ...with address register writeback:
1455 class VST4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1456 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1457 (ins addrmode6:$Rn, am6offset:$Rm,
1458 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST4u,
1459 "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
1460 "$Rn.addr = $wb", []> {
1461 let Inst{5-4} = Rn{5-4};
1462 let DecoderMethod = "DecodeVSTInstruction";
1465 def VST4d8_UPD : VST4DWB<0b0000, {0,0,?,?}, "8">;
1466 def VST4d16_UPD : VST4DWB<0b0000, {0,1,?,?}, "16">;
1467 def VST4d32_UPD : VST4DWB<0b0000, {1,0,?,?}, "32">;
1469 def VST4d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1470 def VST4d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1471 def VST4d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1473 // ...with double-spaced registers:
1474 def VST4q8 : VST4D<0b0001, {0,0,?,?}, "8">;
1475 def VST4q16 : VST4D<0b0001, {0,1,?,?}, "16">;
1476 def VST4q32 : VST4D<0b0001, {1,0,?,?}, "32">;
1477 def VST4q8_UPD : VST4DWB<0b0001, {0,0,?,?}, "8">;
1478 def VST4q16_UPD : VST4DWB<0b0001, {0,1,?,?}, "16">;
1479 def VST4q32_UPD : VST4DWB<0b0001, {1,0,?,?}, "32">;
1481 def VST4q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1482 def VST4q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1483 def VST4q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1485 // ...alternate versions to be allocated odd register numbers:
1486 def VST4q8oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1487 def VST4q16oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1488 def VST4q32oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1490 def VST4q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1491 def VST4q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1492 def VST4q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1494 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
1496 // Classes for VST*LN pseudo-instructions with multi-register operands.
1497 // These are expanded to real instructions after register allocation.
1498 class VSTQLNPseudo<InstrItinClass itin>
1499 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
1501 class VSTQLNWBPseudo<InstrItinClass itin>
1502 : PseudoNLdSt<(outs GPR:$wb),
1503 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
1504 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1505 class VSTQQLNPseudo<InstrItinClass itin>
1506 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
1508 class VSTQQLNWBPseudo<InstrItinClass itin>
1509 : PseudoNLdSt<(outs GPR:$wb),
1510 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
1511 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1512 class VSTQQQQLNPseudo<InstrItinClass itin>
1513 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
1515 class VSTQQQQLNWBPseudo<InstrItinClass itin>
1516 : PseudoNLdSt<(outs GPR:$wb),
1517 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
1518 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1520 // VST1LN : Vector Store (single element from one lane)
1521 class VST1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1522 PatFrag StoreOp, SDNode ExtractOp>
1523 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1524 (ins addrmode6:$Rn, DPR:$Vd, nohash_imm:$lane),
1525 IIC_VST1ln, "vst1", Dt, "\\{$Vd[$lane]\\}, $Rn", "",
1526 [(StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane), addrmode6:$Rn)]> {
1528 let DecoderMethod = "DecodeVST1LN";
1530 class VST1LN32<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1531 PatFrag StoreOp, SDNode ExtractOp>
1532 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1533 (ins addrmode6oneL32:$Rn, DPR:$Vd, nohash_imm:$lane),
1534 IIC_VST1ln, "vst1", Dt, "\\{$Vd[$lane]\\}, $Rn", "",
1535 [(StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane), addrmode6oneL32:$Rn)]>{
1537 let DecoderMethod = "DecodeVST1LN";
1539 class VST1QLNPseudo<ValueType Ty, PatFrag StoreOp, SDNode ExtractOp>
1540 : VSTQLNPseudo<IIC_VST1ln> {
1541 let Pattern = [(StoreOp (ExtractOp (Ty QPR:$src), imm:$lane),
1545 def VST1LNd8 : VST1LN<0b0000, {?,?,?,0}, "8", v8i8, truncstorei8,
1547 let Inst{7-5} = lane{2-0};
1549 def VST1LNd16 : VST1LN<0b0100, {?,?,0,?}, "16", v4i16, truncstorei16,
1551 let Inst{7-6} = lane{1-0};
1552 let Inst{4} = Rn{5};
1555 def VST1LNd32 : VST1LN32<0b1000, {?,0,?,?}, "32", v2i32, store, extractelt> {
1556 let Inst{7} = lane{0};
1557 let Inst{5-4} = Rn{5-4};
1560 def VST1LNq8Pseudo : VST1QLNPseudo<v16i8, truncstorei8, NEONvgetlaneu>;
1561 def VST1LNq16Pseudo : VST1QLNPseudo<v8i16, truncstorei16, NEONvgetlaneu>;
1562 def VST1LNq32Pseudo : VST1QLNPseudo<v4i32, store, extractelt>;
1564 def : Pat<(store (extractelt (v2f32 DPR:$src), imm:$lane), addrmode6:$addr),
1565 (VST1LNd32 addrmode6:$addr, DPR:$src, imm:$lane)>;
1566 def : Pat<(store (extractelt (v4f32 QPR:$src), imm:$lane), addrmode6:$addr),
1567 (VST1LNq32Pseudo addrmode6:$addr, QPR:$src, imm:$lane)>;
1569 // ...with address register writeback:
1570 class VST1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1571 PatFrag StoreOp, SDNode ExtractOp>
1572 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1573 (ins addrmode6:$Rn, am6offset:$Rm,
1574 DPR:$Vd, nohash_imm:$lane), IIC_VST1lnu, "vst1", Dt,
1575 "\\{$Vd[$lane]\\}, $Rn$Rm",
1577 [(set GPR:$wb, (StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane),
1578 addrmode6:$Rn, am6offset:$Rm))]> {
1579 let DecoderMethod = "DecodeVST1LN";
1581 class VST1QLNWBPseudo<ValueType Ty, PatFrag StoreOp, SDNode ExtractOp>
1582 : VSTQLNWBPseudo<IIC_VST1lnu> {
1583 let Pattern = [(set GPR:$wb, (StoreOp (ExtractOp (Ty QPR:$src), imm:$lane),
1584 addrmode6:$addr, am6offset:$offset))];
1587 def VST1LNd8_UPD : VST1LNWB<0b0000, {?,?,?,0}, "8", v8i8, post_truncsti8,
1589 let Inst{7-5} = lane{2-0};
1591 def VST1LNd16_UPD : VST1LNWB<0b0100, {?,?,0,?}, "16", v4i16, post_truncsti16,
1593 let Inst{7-6} = lane{1-0};
1594 let Inst{4} = Rn{5};
1596 def VST1LNd32_UPD : VST1LNWB<0b1000, {?,0,?,?}, "32", v2i32, post_store,
1598 let Inst{7} = lane{0};
1599 let Inst{5-4} = Rn{5-4};
1602 def VST1LNq8Pseudo_UPD : VST1QLNWBPseudo<v16i8, post_truncsti8, NEONvgetlaneu>;
1603 def VST1LNq16Pseudo_UPD : VST1QLNWBPseudo<v8i16, post_truncsti16,NEONvgetlaneu>;
1604 def VST1LNq32Pseudo_UPD : VST1QLNWBPseudo<v4i32, post_store, extractelt>;
1606 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
1608 // VST2LN : Vector Store (single 2-element structure from one lane)
1609 class VST2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1610 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1611 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, nohash_imm:$lane),
1612 IIC_VST2ln, "vst2", Dt, "\\{$Vd[$lane], $src2[$lane]\\}, $Rn",
1615 let Inst{4} = Rn{4};
1616 let DecoderMethod = "DecodeVST2LN";
1619 def VST2LNd8 : VST2LN<0b0001, {?,?,?,?}, "8"> {
1620 let Inst{7-5} = lane{2-0};
1622 def VST2LNd16 : VST2LN<0b0101, {?,?,0,?}, "16"> {
1623 let Inst{7-6} = lane{1-0};
1625 def VST2LNd32 : VST2LN<0b1001, {?,0,0,?}, "32"> {
1626 let Inst{7} = lane{0};
1629 def VST2LNd8Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1630 def VST2LNd16Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1631 def VST2LNd32Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1633 // ...with double-spaced registers:
1634 def VST2LNq16 : VST2LN<0b0101, {?,?,1,?}, "16"> {
1635 let Inst{7-6} = lane{1-0};
1636 let Inst{4} = Rn{4};
1638 def VST2LNq32 : VST2LN<0b1001, {?,1,0,?}, "32"> {
1639 let Inst{7} = lane{0};
1640 let Inst{4} = Rn{4};
1643 def VST2LNq16Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
1644 def VST2LNq32Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
1646 // ...with address register writeback:
1647 class VST2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1648 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1649 (ins addrmode6:$addr, am6offset:$offset,
1650 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VST2lnu, "vst2", Dt,
1651 "\\{$src1[$lane], $src2[$lane]\\}, $addr$offset",
1652 "$addr.addr = $wb", []> {
1653 let Inst{4} = Rn{4};
1654 let DecoderMethod = "DecodeVST2LN";
1657 def VST2LNd8_UPD : VST2LNWB<0b0001, {?,?,?,?}, "8"> {
1658 let Inst{7-5} = lane{2-0};
1660 def VST2LNd16_UPD : VST2LNWB<0b0101, {?,?,0,?}, "16"> {
1661 let Inst{7-6} = lane{1-0};
1663 def VST2LNd32_UPD : VST2LNWB<0b1001, {?,0,0,?}, "32"> {
1664 let Inst{7} = lane{0};
1667 def VST2LNd8Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1668 def VST2LNd16Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1669 def VST2LNd32Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1671 def VST2LNq16_UPD : VST2LNWB<0b0101, {?,?,1,?}, "16"> {
1672 let Inst{7-6} = lane{1-0};
1674 def VST2LNq32_UPD : VST2LNWB<0b1001, {?,1,0,?}, "32"> {
1675 let Inst{7} = lane{0};
1678 def VST2LNq16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
1679 def VST2LNq32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
1681 // VST3LN : Vector Store (single 3-element structure from one lane)
1682 class VST3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1683 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1684 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3,
1685 nohash_imm:$lane), IIC_VST3ln, "vst3", Dt,
1686 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn", "", []> {
1688 let DecoderMethod = "DecodeVST3LN";
1691 def VST3LNd8 : VST3LN<0b0010, {?,?,?,0}, "8"> {
1692 let Inst{7-5} = lane{2-0};
1694 def VST3LNd16 : VST3LN<0b0110, {?,?,0,0}, "16"> {
1695 let Inst{7-6} = lane{1-0};
1697 def VST3LNd32 : VST3LN<0b1010, {?,0,0,0}, "32"> {
1698 let Inst{7} = lane{0};
1701 def VST3LNd8Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1702 def VST3LNd16Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1703 def VST3LNd32Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1705 // ...with double-spaced registers:
1706 def VST3LNq16 : VST3LN<0b0110, {?,?,1,0}, "16"> {
1707 let Inst{7-6} = lane{1-0};
1709 def VST3LNq32 : VST3LN<0b1010, {?,1,0,0}, "32"> {
1710 let Inst{7} = lane{0};
1713 def VST3LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
1714 def VST3LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
1716 // ...with address register writeback:
1717 class VST3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1718 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1719 (ins addrmode6:$Rn, am6offset:$Rm,
1720 DPR:$Vd, DPR:$src2, DPR:$src3, nohash_imm:$lane),
1721 IIC_VST3lnu, "vst3", Dt,
1722 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn$Rm",
1723 "$Rn.addr = $wb", []> {
1724 let DecoderMethod = "DecodeVST3LN";
1727 def VST3LNd8_UPD : VST3LNWB<0b0010, {?,?,?,0}, "8"> {
1728 let Inst{7-5} = lane{2-0};
1730 def VST3LNd16_UPD : VST3LNWB<0b0110, {?,?,0,0}, "16"> {
1731 let Inst{7-6} = lane{1-0};
1733 def VST3LNd32_UPD : VST3LNWB<0b1010, {?,0,0,0}, "32"> {
1734 let Inst{7} = lane{0};
1737 def VST3LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1738 def VST3LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1739 def VST3LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1741 def VST3LNq16_UPD : VST3LNWB<0b0110, {?,?,1,0}, "16"> {
1742 let Inst{7-6} = lane{1-0};
1744 def VST3LNq32_UPD : VST3LNWB<0b1010, {?,1,0,0}, "32"> {
1745 let Inst{7} = lane{0};
1748 def VST3LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
1749 def VST3LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
1751 // VST4LN : Vector Store (single 4-element structure from one lane)
1752 class VST4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1753 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1754 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4,
1755 nohash_imm:$lane), IIC_VST4ln, "vst4", Dt,
1756 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn",
1759 let Inst{4} = Rn{4};
1760 let DecoderMethod = "DecodeVST4LN";
1763 def VST4LNd8 : VST4LN<0b0011, {?,?,?,?}, "8"> {
1764 let Inst{7-5} = lane{2-0};
1766 def VST4LNd16 : VST4LN<0b0111, {?,?,0,?}, "16"> {
1767 let Inst{7-6} = lane{1-0};
1769 def VST4LNd32 : VST4LN<0b1011, {?,0,?,?}, "32"> {
1770 let Inst{7} = lane{0};
1771 let Inst{5} = Rn{5};
1774 def VST4LNd8Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1775 def VST4LNd16Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1776 def VST4LNd32Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1778 // ...with double-spaced registers:
1779 def VST4LNq16 : VST4LN<0b0111, {?,?,1,?}, "16"> {
1780 let Inst{7-6} = lane{1-0};
1782 def VST4LNq32 : VST4LN<0b1011, {?,1,?,?}, "32"> {
1783 let Inst{7} = lane{0};
1784 let Inst{5} = Rn{5};
1787 def VST4LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
1788 def VST4LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
1790 // ...with address register writeback:
1791 class VST4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1792 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1793 (ins addrmode6:$Rn, am6offset:$Rm,
1794 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
1795 IIC_VST4lnu, "vst4", Dt,
1796 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn$Rm",
1797 "$Rn.addr = $wb", []> {
1798 let Inst{4} = Rn{4};
1799 let DecoderMethod = "DecodeVST4LN";
1802 def VST4LNd8_UPD : VST4LNWB<0b0011, {?,?,?,?}, "8"> {
1803 let Inst{7-5} = lane{2-0};
1805 def VST4LNd16_UPD : VST4LNWB<0b0111, {?,?,0,?}, "16"> {
1806 let Inst{7-6} = lane{1-0};
1808 def VST4LNd32_UPD : VST4LNWB<0b1011, {?,0,?,?}, "32"> {
1809 let Inst{7} = lane{0};
1810 let Inst{5} = Rn{5};
1813 def VST4LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
1814 def VST4LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
1815 def VST4LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
1817 def VST4LNq16_UPD : VST4LNWB<0b0111, {?,?,1,?}, "16"> {
1818 let Inst{7-6} = lane{1-0};
1820 def VST4LNq32_UPD : VST4LNWB<0b1011, {?,1,?,?}, "32"> {
1821 let Inst{7} = lane{0};
1822 let Inst{5} = Rn{5};
1825 def VST4LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
1826 def VST4LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
1828 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
1831 //===----------------------------------------------------------------------===//
1832 // NEON pattern fragments
1833 //===----------------------------------------------------------------------===//
1835 // Extract D sub-registers of Q registers.
1836 def DSubReg_i8_reg : SDNodeXForm<imm, [{
1837 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1838 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/8, MVT::i32);
1840 def DSubReg_i16_reg : SDNodeXForm<imm, [{
1841 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1842 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/4, MVT::i32);
1844 def DSubReg_i32_reg : SDNodeXForm<imm, [{
1845 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1846 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/2, MVT::i32);
1848 def DSubReg_f64_reg : SDNodeXForm<imm, [{
1849 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1850 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue(), MVT::i32);
1853 // Extract S sub-registers of Q/D registers.
1854 def SSubReg_f32_reg : SDNodeXForm<imm, [{
1855 assert(ARM::ssub_3 == ARM::ssub_0+3 && "Unexpected subreg numbering");
1856 return CurDAG->getTargetConstant(ARM::ssub_0 + N->getZExtValue(), MVT::i32);
1859 // Translate lane numbers from Q registers to D subregs.
1860 def SubReg_i8_lane : SDNodeXForm<imm, [{
1861 return CurDAG->getTargetConstant(N->getZExtValue() & 7, MVT::i32);
1863 def SubReg_i16_lane : SDNodeXForm<imm, [{
1864 return CurDAG->getTargetConstant(N->getZExtValue() & 3, MVT::i32);
1866 def SubReg_i32_lane : SDNodeXForm<imm, [{
1867 return CurDAG->getTargetConstant(N->getZExtValue() & 1, MVT::i32);
1870 //===----------------------------------------------------------------------===//
1871 // Instruction Classes
1872 //===----------------------------------------------------------------------===//
1874 // Basic 2-register operations: double- and quad-register.
1875 class N2VD<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1876 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
1877 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
1878 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
1879 (ins DPR:$Vm), IIC_VUNAD, OpcodeStr, Dt,"$Vd, $Vm", "",
1880 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vm))))]>;
1881 class N2VQ<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1882 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
1883 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
1884 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
1885 (ins QPR:$Vm), IIC_VUNAQ, OpcodeStr, Dt,"$Vd, $Vm", "",
1886 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vm))))]>;
1888 // Basic 2-register intrinsics, both double- and quad-register.
1889 class N2VDInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1890 bits<2> op17_16, bits<5> op11_7, bit op4,
1891 InstrItinClass itin, string OpcodeStr, string Dt,
1892 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1893 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
1894 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
1895 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm))))]>;
1896 class N2VQInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1897 bits<2> op17_16, bits<5> op11_7, bit op4,
1898 InstrItinClass itin, string OpcodeStr, string Dt,
1899 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1900 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
1901 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
1902 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm))))]>;
1904 // Narrow 2-register operations.
1905 class N2VN<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1906 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1907 InstrItinClass itin, string OpcodeStr, string Dt,
1908 ValueType TyD, ValueType TyQ, SDNode OpNode>
1909 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$Vd),
1910 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
1911 [(set DPR:$Vd, (TyD (OpNode (TyQ QPR:$Vm))))]>;
1913 // Narrow 2-register intrinsics.
1914 class N2VNInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1915 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1916 InstrItinClass itin, string OpcodeStr, string Dt,
1917 ValueType TyD, ValueType TyQ, Intrinsic IntOp>
1918 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$Vd),
1919 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
1920 [(set DPR:$Vd, (TyD (IntOp (TyQ QPR:$Vm))))]>;
1922 // Long 2-register operations (currently only used for VMOVL).
1923 class N2VL<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1924 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1925 InstrItinClass itin, string OpcodeStr, string Dt,
1926 ValueType TyQ, ValueType TyD, SDNode OpNode>
1927 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$Vd),
1928 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
1929 [(set QPR:$Vd, (TyQ (OpNode (TyD DPR:$Vm))))]>;
1931 // Long 2-register intrinsics.
1932 class N2VLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1933 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1934 InstrItinClass itin, string OpcodeStr, string Dt,
1935 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
1936 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$Vd),
1937 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
1938 [(set QPR:$Vd, (TyQ (IntOp (TyD DPR:$Vm))))]>;
1940 // 2-register shuffles (VTRN/VZIP/VUZP), both double- and quad-register.
1941 class N2VDShuffle<bits<2> op19_18, bits<5> op11_7, string OpcodeStr, string Dt>
1942 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 0, 0, (outs DPR:$Vd, DPR:$Vm),
1943 (ins DPR:$src1, DPR:$src2), IIC_VPERMD,
1944 OpcodeStr, Dt, "$Vd, $Vm",
1945 "$src1 = $Vd, $src2 = $Vm", []>;
1946 class N2VQShuffle<bits<2> op19_18, bits<5> op11_7,
1947 InstrItinClass itin, string OpcodeStr, string Dt>
1948 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 1, 0, (outs QPR:$Vd, QPR:$Vm),
1949 (ins QPR:$src1, QPR:$src2), itin, OpcodeStr, Dt, "$Vd, $Vm",
1950 "$src1 = $Vd, $src2 = $Vm", []>;
1952 // Basic 3-register operations: double- and quad-register.
1953 class N3VD<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1954 InstrItinClass itin, string OpcodeStr, string Dt,
1955 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
1956 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1957 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1958 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
1959 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
1960 let isCommutable = Commutable;
1962 // Same as N3VD but no data type.
1963 class N3VDX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1964 InstrItinClass itin, string OpcodeStr,
1965 ValueType ResTy, ValueType OpTy,
1966 SDNode OpNode, bit Commutable>
1967 : N3VX<op24, op23, op21_20, op11_8, 0, op4,
1968 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1969 OpcodeStr, "$Vd, $Vn, $Vm", "",
1970 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>{
1971 let isCommutable = Commutable;
1974 class N3VDSL<bits<2> op21_20, bits<4> op11_8,
1975 InstrItinClass itin, string OpcodeStr, string Dt,
1976 ValueType Ty, SDNode ShOp>
1977 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
1978 (outs DPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
1979 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
1981 (Ty (ShOp (Ty DPR:$Vn),
1982 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),imm:$lane)))))]> {
1983 let isCommutable = 0;
1985 class N3VDSL16<bits<2> op21_20, bits<4> op11_8,
1986 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1987 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
1988 (outs DPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
1989 NVMulSLFrm, IIC_VMULi16D, OpcodeStr, Dt,"$Vd, $Vn, $Vm$lane","",
1991 (Ty (ShOp (Ty DPR:$Vn),
1992 (Ty (NEONvduplane (Ty DPR_8:$Vm), imm:$lane)))))]> {
1993 let isCommutable = 0;
1996 class N3VQ<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1997 InstrItinClass itin, string OpcodeStr, string Dt,
1998 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
1999 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2000 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2001 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2002 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
2003 let isCommutable = Commutable;
2005 class N3VQX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2006 InstrItinClass itin, string OpcodeStr,
2007 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
2008 : N3VX<op24, op23, op21_20, op11_8, 1, op4,
2009 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2010 OpcodeStr, "$Vd, $Vn, $Vm", "",
2011 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]>{
2012 let isCommutable = Commutable;
2014 class N3VQSL<bits<2> op21_20, bits<4> op11_8,
2015 InstrItinClass itin, string OpcodeStr, string Dt,
2016 ValueType ResTy, ValueType OpTy, SDNode ShOp>
2017 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2018 (outs QPR:$Vd), (ins QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2019 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2020 [(set (ResTy QPR:$Vd),
2021 (ResTy (ShOp (ResTy QPR:$Vn),
2022 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2024 let isCommutable = 0;
2026 class N3VQSL16<bits<2> op21_20, bits<4> op11_8, string OpcodeStr, string Dt,
2027 ValueType ResTy, ValueType OpTy, SDNode ShOp>
2028 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2029 (outs QPR:$Vd), (ins QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2030 NVMulSLFrm, IIC_VMULi16Q, OpcodeStr, Dt,"$Vd, $Vn, $Vm$lane", "",
2031 [(set (ResTy QPR:$Vd),
2032 (ResTy (ShOp (ResTy QPR:$Vn),
2033 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2035 let isCommutable = 0;
2038 // Basic 3-register intrinsics, both double- and quad-register.
2039 class N3VDInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2040 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2041 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
2042 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2043 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), f, itin,
2044 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2045 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
2046 let isCommutable = Commutable;
2048 class N3VDIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2049 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
2050 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2051 (outs DPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2052 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2054 (Ty (IntOp (Ty DPR:$Vn),
2055 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),
2057 let isCommutable = 0;
2059 class N3VDIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2060 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
2061 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2062 (outs DPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2063 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2065 (Ty (IntOp (Ty DPR:$Vn),
2066 (Ty (NEONvduplane (Ty DPR_8:$Vm), imm:$lane)))))]> {
2067 let isCommutable = 0;
2069 class N3VDIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2070 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2071 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2072 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2073 (outs DPR:$Vd), (ins DPR:$Vm, DPR:$Vn), f, itin,
2074 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
2075 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (OpTy DPR:$Vn))))]> {
2076 let isCommutable = 0;
2079 class N3VQInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2080 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2081 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
2082 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2083 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), f, itin,
2084 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2085 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
2086 let isCommutable = Commutable;
2088 class N3VQIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2089 string OpcodeStr, string Dt,
2090 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2091 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2092 (outs QPR:$Vd), (ins QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2093 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2094 [(set (ResTy QPR:$Vd),
2095 (ResTy (IntOp (ResTy QPR:$Vn),
2096 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2098 let isCommutable = 0;
2100 class N3VQIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2101 string OpcodeStr, string Dt,
2102 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2103 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2104 (outs QPR:$Vd), (ins QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2105 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2106 [(set (ResTy QPR:$Vd),
2107 (ResTy (IntOp (ResTy QPR:$Vn),
2108 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2110 let isCommutable = 0;
2112 class N3VQIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2113 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2114 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2115 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2116 (outs QPR:$Vd), (ins QPR:$Vm, QPR:$Vn), f, itin,
2117 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
2118 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (OpTy QPR:$Vn))))]> {
2119 let isCommutable = 0;
2122 // Multiply-Add/Sub operations: double- and quad-register.
2123 class N3VDMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2124 InstrItinClass itin, string OpcodeStr, string Dt,
2125 ValueType Ty, SDPatternOperator MulOp, SDPatternOperator OpNode>
2126 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2127 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2128 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2129 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
2130 (Ty (MulOp DPR:$Vn, DPR:$Vm)))))]>;
2132 class N3VDMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2133 string OpcodeStr, string Dt,
2134 ValueType Ty, SDPatternOperator MulOp, SDPatternOperator ShOp>
2135 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2137 (ins DPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2139 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2141 (Ty (ShOp (Ty DPR:$src1),
2143 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),
2145 class N3VDMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2146 string OpcodeStr, string Dt,
2147 ValueType Ty, SDNode MulOp, SDNode ShOp>
2148 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2150 (ins DPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2152 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2154 (Ty (ShOp (Ty DPR:$src1),
2156 (Ty (NEONvduplane (Ty DPR_8:$Vm),
2159 class N3VQMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2160 InstrItinClass itin, string OpcodeStr, string Dt, ValueType Ty,
2161 SDPatternOperator MulOp, SDPatternOperator OpNode>
2162 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2163 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2164 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2165 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
2166 (Ty (MulOp QPR:$Vn, QPR:$Vm)))))]>;
2167 class N3VQMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2168 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2169 SDPatternOperator MulOp, SDPatternOperator ShOp>
2170 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2172 (ins QPR:$src1, QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2174 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2175 [(set (ResTy QPR:$Vd),
2176 (ResTy (ShOp (ResTy QPR:$src1),
2177 (ResTy (MulOp QPR:$Vn,
2178 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2180 class N3VQMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2181 string OpcodeStr, string Dt,
2182 ValueType ResTy, ValueType OpTy,
2183 SDNode MulOp, SDNode ShOp>
2184 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2186 (ins QPR:$src1, QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2188 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2189 [(set (ResTy QPR:$Vd),
2190 (ResTy (ShOp (ResTy QPR:$src1),
2191 (ResTy (MulOp QPR:$Vn,
2192 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2195 // Neon Intrinsic-Op instructions (VABA): double- and quad-register.
2196 class N3VDIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2197 InstrItinClass itin, string OpcodeStr, string Dt,
2198 ValueType Ty, Intrinsic IntOp, SDNode OpNode>
2199 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2200 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2201 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2202 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
2203 (Ty (IntOp (Ty DPR:$Vn), (Ty DPR:$Vm))))))]>;
2204 class N3VQIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2205 InstrItinClass itin, string OpcodeStr, string Dt,
2206 ValueType Ty, Intrinsic IntOp, SDNode OpNode>
2207 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2208 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2209 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2210 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
2211 (Ty (IntOp (Ty QPR:$Vn), (Ty QPR:$Vm))))))]>;
2213 // Neon 3-argument intrinsics, both double- and quad-register.
2214 // The destination register is also used as the first source operand register.
2215 class N3VDInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2216 InstrItinClass itin, string OpcodeStr, string Dt,
2217 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2218 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2219 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2220 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2221 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$src1),
2222 (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>;
2223 class N3VQInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2224 InstrItinClass itin, string OpcodeStr, string Dt,
2225 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2226 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2227 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2228 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2229 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$src1),
2230 (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]>;
2232 // Long Multiply-Add/Sub operations.
2233 class N3VLMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2234 InstrItinClass itin, string OpcodeStr, string Dt,
2235 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2236 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2237 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2238 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2239 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
2240 (TyQ (MulOp (TyD DPR:$Vn),
2241 (TyD DPR:$Vm)))))]>;
2242 class N3VLMulOpSL<bit op24, bits<2> op21_20, bits<4> op11_8,
2243 InstrItinClass itin, string OpcodeStr, string Dt,
2244 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2245 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$Vd),
2246 (ins QPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2248 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2250 (OpNode (TyQ QPR:$src1),
2251 (TyQ (MulOp (TyD DPR:$Vn),
2252 (TyD (NEONvduplane (TyD DPR_VFP2:$Vm),
2254 class N3VLMulOpSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2255 InstrItinClass itin, string OpcodeStr, string Dt,
2256 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2257 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$Vd),
2258 (ins QPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2260 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2262 (OpNode (TyQ QPR:$src1),
2263 (TyQ (MulOp (TyD DPR:$Vn),
2264 (TyD (NEONvduplane (TyD DPR_8:$Vm),
2267 // Long Intrinsic-Op vector operations with explicit extend (VABAL).
2268 class N3VLIntExtOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2269 InstrItinClass itin, string OpcodeStr, string Dt,
2270 ValueType TyQ, ValueType TyD, Intrinsic IntOp, SDNode ExtOp,
2272 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2273 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2274 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2275 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
2276 (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
2277 (TyD DPR:$Vm)))))))]>;
2279 // Neon Long 3-argument intrinsic. The destination register is
2280 // a quad-register and is also used as the first source operand register.
2281 class N3VLInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2282 InstrItinClass itin, string OpcodeStr, string Dt,
2283 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
2284 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2285 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2286 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2288 (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$Vn), (TyD DPR:$Vm))))]>;
2289 class N3VLInt3SL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2290 string OpcodeStr, string Dt,
2291 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2292 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2294 (ins QPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2296 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2297 [(set (ResTy QPR:$Vd),
2298 (ResTy (IntOp (ResTy QPR:$src1),
2300 (OpTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2302 class N3VLInt3SL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2303 InstrItinClass itin, string OpcodeStr, string Dt,
2304 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2305 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2307 (ins QPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2309 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2310 [(set (ResTy QPR:$Vd),
2311 (ResTy (IntOp (ResTy QPR:$src1),
2313 (OpTy (NEONvduplane (OpTy DPR_8:$Vm),
2316 // Narrowing 3-register intrinsics.
2317 class N3VNInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2318 string OpcodeStr, string Dt, ValueType TyD, ValueType TyQ,
2319 Intrinsic IntOp, bit Commutable>
2320 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2321 (outs DPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINi4D,
2322 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2323 [(set DPR:$Vd, (TyD (IntOp (TyQ QPR:$Vn), (TyQ QPR:$Vm))))]> {
2324 let isCommutable = Commutable;
2327 // Long 3-register operations.
2328 class N3VL<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2329 InstrItinClass itin, string OpcodeStr, string Dt,
2330 ValueType TyQ, ValueType TyD, SDNode OpNode, bit Commutable>
2331 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2332 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2333 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2334 [(set QPR:$Vd, (TyQ (OpNode (TyD DPR:$Vn), (TyD DPR:$Vm))))]> {
2335 let isCommutable = Commutable;
2337 class N3VLSL<bit op24, bits<2> op21_20, bits<4> op11_8,
2338 InstrItinClass itin, string OpcodeStr, string Dt,
2339 ValueType TyQ, ValueType TyD, SDNode OpNode>
2340 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2341 (outs QPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2342 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2344 (TyQ (OpNode (TyD DPR:$Vn),
2345 (TyD (NEONvduplane (TyD DPR_VFP2:$Vm),imm:$lane)))))]>;
2346 class N3VLSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2347 InstrItinClass itin, string OpcodeStr, string Dt,
2348 ValueType TyQ, ValueType TyD, SDNode OpNode>
2349 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2350 (outs QPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2351 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2353 (TyQ (OpNode (TyD DPR:$Vn),
2354 (TyD (NEONvduplane (TyD DPR_8:$Vm), imm:$lane)))))]>;
2356 // Long 3-register operations with explicitly extended operands.
2357 class N3VLExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2358 InstrItinClass itin, string OpcodeStr, string Dt,
2359 ValueType TyQ, ValueType TyD, SDNode OpNode, SDNode ExtOp,
2361 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2362 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2363 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2364 [(set QPR:$Vd, (OpNode (TyQ (ExtOp (TyD DPR:$Vn))),
2365 (TyQ (ExtOp (TyD DPR:$Vm)))))]> {
2366 let isCommutable = Commutable;
2369 // Long 3-register intrinsics with explicit extend (VABDL).
2370 class N3VLIntExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2371 InstrItinClass itin, string OpcodeStr, string Dt,
2372 ValueType TyQ, ValueType TyD, Intrinsic IntOp, SDNode ExtOp,
2374 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2375 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2376 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2377 [(set QPR:$Vd, (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
2378 (TyD DPR:$Vm))))))]> {
2379 let isCommutable = Commutable;
2382 // Long 3-register intrinsics.
2383 class N3VLInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2384 InstrItinClass itin, string OpcodeStr, string Dt,
2385 ValueType TyQ, ValueType TyD, Intrinsic IntOp, bit Commutable>
2386 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2387 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2388 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2389 [(set QPR:$Vd, (TyQ (IntOp (TyD DPR:$Vn), (TyD DPR:$Vm))))]> {
2390 let isCommutable = Commutable;
2392 class N3VLIntSL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2393 string OpcodeStr, string Dt,
2394 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2395 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2396 (outs QPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2397 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2398 [(set (ResTy QPR:$Vd),
2399 (ResTy (IntOp (OpTy DPR:$Vn),
2400 (OpTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2402 class N3VLIntSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2403 InstrItinClass itin, string OpcodeStr, string Dt,
2404 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2405 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2406 (outs QPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2407 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2408 [(set (ResTy QPR:$Vd),
2409 (ResTy (IntOp (OpTy DPR:$Vn),
2410 (OpTy (NEONvduplane (OpTy DPR_8:$Vm),
2413 // Wide 3-register operations.
2414 class N3VW<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2415 string OpcodeStr, string Dt, ValueType TyQ, ValueType TyD,
2416 SDNode OpNode, SDNode ExtOp, bit Commutable>
2417 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2418 (outs QPR:$Vd), (ins QPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VSUBiD,
2419 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2420 [(set QPR:$Vd, (OpNode (TyQ QPR:$Vn),
2421 (TyQ (ExtOp (TyD DPR:$Vm)))))]> {
2422 let isCommutable = Commutable;
2425 // Pairwise long 2-register intrinsics, both double- and quad-register.
2426 class N2VDPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2427 bits<2> op17_16, bits<5> op11_7, bit op4,
2428 string OpcodeStr, string Dt,
2429 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2430 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
2431 (ins DPR:$Vm), IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
2432 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm))))]>;
2433 class N2VQPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2434 bits<2> op17_16, bits<5> op11_7, bit op4,
2435 string OpcodeStr, string Dt,
2436 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2437 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
2438 (ins QPR:$Vm), IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
2439 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm))))]>;
2441 // Pairwise long 2-register accumulate intrinsics,
2442 // both double- and quad-register.
2443 // The destination register is also used as the first source operand register.
2444 class N2VDPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2445 bits<2> op17_16, bits<5> op11_7, bit op4,
2446 string OpcodeStr, string Dt,
2447 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2448 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
2449 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vm), IIC_VPALiD,
2450 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
2451 [(set DPR:$Vd, (ResTy (IntOp (ResTy DPR:$src1), (OpTy DPR:$Vm))))]>;
2452 class N2VQPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2453 bits<2> op17_16, bits<5> op11_7, bit op4,
2454 string OpcodeStr, string Dt,
2455 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2456 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4,
2457 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vm), IIC_VPALiQ,
2458 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
2459 [(set QPR:$Vd, (ResTy (IntOp (ResTy QPR:$src1), (OpTy QPR:$Vm))))]>;
2461 // Shift by immediate,
2462 // both double- and quad-register.
2463 class N2VDSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2464 Format f, InstrItinClass itin, Operand ImmTy,
2465 string OpcodeStr, string Dt, ValueType Ty, SDNode OpNode>
2466 : N2VImm<op24, op23, op11_8, op7, 0, op4,
2467 (outs DPR:$Vd), (ins DPR:$Vm, ImmTy:$SIMM), f, itin,
2468 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2469 [(set DPR:$Vd, (Ty (OpNode (Ty DPR:$Vm), (i32 imm:$SIMM))))]>;
2470 class N2VQSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2471 Format f, InstrItinClass itin, Operand ImmTy,
2472 string OpcodeStr, string Dt, ValueType Ty, SDNode OpNode>
2473 : N2VImm<op24, op23, op11_8, op7, 1, op4,
2474 (outs QPR:$Vd), (ins QPR:$Vm, ImmTy:$SIMM), f, itin,
2475 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2476 [(set QPR:$Vd, (Ty (OpNode (Ty QPR:$Vm), (i32 imm:$SIMM))))]>;
2478 // Long shift by immediate.
2479 class N2VLSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
2480 string OpcodeStr, string Dt,
2481 ValueType ResTy, ValueType OpTy, SDNode OpNode>
2482 : N2VImm<op24, op23, op11_8, op7, op6, op4,
2483 (outs QPR:$Vd), (ins DPR:$Vm, i32imm:$SIMM), N2RegVShLFrm,
2484 IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2485 [(set QPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vm),
2486 (i32 imm:$SIMM))))]>;
2488 // Narrow shift by immediate.
2489 class N2VNSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
2490 InstrItinClass itin, string OpcodeStr, string Dt,
2491 ValueType ResTy, ValueType OpTy, Operand ImmTy, SDNode OpNode>
2492 : N2VImm<op24, op23, op11_8, op7, op6, op4,
2493 (outs DPR:$Vd), (ins QPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, itin,
2494 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2495 [(set DPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vm),
2496 (i32 imm:$SIMM))))]>;
2498 // Shift right by immediate and accumulate,
2499 // both double- and quad-register.
2500 class N2VDShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2501 Operand ImmTy, string OpcodeStr, string Dt,
2502 ValueType Ty, SDNode ShOp>
2503 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
2504 (ins DPR:$src1, DPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, IIC_VPALiD,
2505 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2506 [(set DPR:$Vd, (Ty (add DPR:$src1,
2507 (Ty (ShOp DPR:$Vm, (i32 imm:$SIMM))))))]>;
2508 class N2VQShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2509 Operand ImmTy, string OpcodeStr, string Dt,
2510 ValueType Ty, SDNode ShOp>
2511 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
2512 (ins QPR:$src1, QPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, IIC_VPALiD,
2513 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2514 [(set QPR:$Vd, (Ty (add QPR:$src1,
2515 (Ty (ShOp QPR:$Vm, (i32 imm:$SIMM))))))]>;
2517 // Shift by immediate and insert,
2518 // both double- and quad-register.
2519 class N2VDShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2520 Operand ImmTy, Format f, string OpcodeStr, string Dt,
2521 ValueType Ty,SDNode ShOp>
2522 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
2523 (ins DPR:$src1, DPR:$Vm, ImmTy:$SIMM), f, IIC_VSHLiD,
2524 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2525 [(set DPR:$Vd, (Ty (ShOp DPR:$src1, DPR:$Vm, (i32 imm:$SIMM))))]>;
2526 class N2VQShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2527 Operand ImmTy, Format f, string OpcodeStr, string Dt,
2528 ValueType Ty,SDNode ShOp>
2529 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
2530 (ins QPR:$src1, QPR:$Vm, ImmTy:$SIMM), f, IIC_VSHLiQ,
2531 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2532 [(set QPR:$Vd, (Ty (ShOp QPR:$src1, QPR:$Vm, (i32 imm:$SIMM))))]>;
2534 // Convert, with fractional bits immediate,
2535 // both double- and quad-register.
2536 class N2VCvtD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2537 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2539 : N2VImm<op24, op23, op11_8, op7, 0, op4,
2540 (outs DPR:$Vd), (ins DPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
2541 IIC_VUNAD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2542 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (i32 imm:$SIMM))))]>;
2543 class N2VCvtQ<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2544 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2546 : N2VImm<op24, op23, op11_8, op7, 1, op4,
2547 (outs QPR:$Vd), (ins QPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
2548 IIC_VUNAQ, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2549 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (i32 imm:$SIMM))))]>;
2551 //===----------------------------------------------------------------------===//
2553 //===----------------------------------------------------------------------===//
2555 // Abbreviations used in multiclass suffixes:
2556 // Q = quarter int (8 bit) elements
2557 // H = half int (16 bit) elements
2558 // S = single int (32 bit) elements
2559 // D = double int (64 bit) elements
2561 // Neon 2-register vector operations and intrinsics.
2563 // Neon 2-register comparisons.
2564 // source operand element sizes of 8, 16 and 32 bits:
2565 multiclass N2V_QHS_cmp<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2566 bits<5> op11_7, bit op4, string opc, string Dt,
2567 string asm, SDNode OpNode> {
2568 // 64-bit vector types.
2569 def v8i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 0, op4,
2570 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2571 opc, !strconcat(Dt, "8"), asm, "",
2572 [(set DPR:$Vd, (v8i8 (OpNode (v8i8 DPR:$Vm))))]>;
2573 def v4i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 0, op4,
2574 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2575 opc, !strconcat(Dt, "16"), asm, "",
2576 [(set DPR:$Vd, (v4i16 (OpNode (v4i16 DPR:$Vm))))]>;
2577 def v2i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
2578 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2579 opc, !strconcat(Dt, "32"), asm, "",
2580 [(set DPR:$Vd, (v2i32 (OpNode (v2i32 DPR:$Vm))))]>;
2581 def v2f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
2582 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2583 opc, "f32", asm, "",
2584 [(set DPR:$Vd, (v2i32 (OpNode (v2f32 DPR:$Vm))))]> {
2585 let Inst{10} = 1; // overwrite F = 1
2588 // 128-bit vector types.
2589 def v16i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 1, op4,
2590 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2591 opc, !strconcat(Dt, "8"), asm, "",
2592 [(set QPR:$Vd, (v16i8 (OpNode (v16i8 QPR:$Vm))))]>;
2593 def v8i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 1, op4,
2594 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2595 opc, !strconcat(Dt, "16"), asm, "",
2596 [(set QPR:$Vd, (v8i16 (OpNode (v8i16 QPR:$Vm))))]>;
2597 def v4i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
2598 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2599 opc, !strconcat(Dt, "32"), asm, "",
2600 [(set QPR:$Vd, (v4i32 (OpNode (v4i32 QPR:$Vm))))]>;
2601 def v4f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
2602 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2603 opc, "f32", asm, "",
2604 [(set QPR:$Vd, (v4i32 (OpNode (v4f32 QPR:$Vm))))]> {
2605 let Inst{10} = 1; // overwrite F = 1
2610 // Neon 2-register vector intrinsics,
2611 // element sizes of 8, 16 and 32 bits:
2612 multiclass N2VInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2613 bits<5> op11_7, bit op4,
2614 InstrItinClass itinD, InstrItinClass itinQ,
2615 string OpcodeStr, string Dt, Intrinsic IntOp> {
2616 // 64-bit vector types.
2617 def v8i8 : N2VDInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2618 itinD, OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
2619 def v4i16 : N2VDInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2620 itinD, OpcodeStr, !strconcat(Dt, "16"),v4i16,v4i16,IntOp>;
2621 def v2i32 : N2VDInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2622 itinD, OpcodeStr, !strconcat(Dt, "32"),v2i32,v2i32,IntOp>;
2624 // 128-bit vector types.
2625 def v16i8 : N2VQInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2626 itinQ, OpcodeStr, !strconcat(Dt, "8"), v16i8,v16i8,IntOp>;
2627 def v8i16 : N2VQInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2628 itinQ, OpcodeStr, !strconcat(Dt, "16"),v8i16,v8i16,IntOp>;
2629 def v4i32 : N2VQInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2630 itinQ, OpcodeStr, !strconcat(Dt, "32"),v4i32,v4i32,IntOp>;
2634 // Neon Narrowing 2-register vector operations,
2635 // source operand element sizes of 16, 32 and 64 bits:
2636 multiclass N2VN_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2637 bits<5> op11_7, bit op6, bit op4,
2638 InstrItinClass itin, string OpcodeStr, string Dt,
2640 def v8i8 : N2VN<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
2641 itin, OpcodeStr, !strconcat(Dt, "16"),
2642 v8i8, v8i16, OpNode>;
2643 def v4i16 : N2VN<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
2644 itin, OpcodeStr, !strconcat(Dt, "32"),
2645 v4i16, v4i32, OpNode>;
2646 def v2i32 : N2VN<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
2647 itin, OpcodeStr, !strconcat(Dt, "64"),
2648 v2i32, v2i64, OpNode>;
2651 // Neon Narrowing 2-register vector intrinsics,
2652 // source operand element sizes of 16, 32 and 64 bits:
2653 multiclass N2VNInt_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2654 bits<5> op11_7, bit op6, bit op4,
2655 InstrItinClass itin, string OpcodeStr, string Dt,
2657 def v8i8 : N2VNInt<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
2658 itin, OpcodeStr, !strconcat(Dt, "16"),
2659 v8i8, v8i16, IntOp>;
2660 def v4i16 : N2VNInt<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
2661 itin, OpcodeStr, !strconcat(Dt, "32"),
2662 v4i16, v4i32, IntOp>;
2663 def v2i32 : N2VNInt<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
2664 itin, OpcodeStr, !strconcat(Dt, "64"),
2665 v2i32, v2i64, IntOp>;
2669 // Neon Lengthening 2-register vector intrinsic (currently specific to VMOVL).
2670 // source operand element sizes of 16, 32 and 64 bits:
2671 multiclass N2VL_QHS<bits<2> op24_23, bits<5> op11_7, bit op6, bit op4,
2672 string OpcodeStr, string Dt, SDNode OpNode> {
2673 def v8i16 : N2VL<op24_23, 0b00, 0b10, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2674 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode>;
2675 def v4i32 : N2VL<op24_23, 0b01, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2676 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
2677 def v2i64 : N2VL<op24_23, 0b10, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2678 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
2682 // Neon 3-register vector operations.
2684 // First with only element sizes of 8, 16 and 32 bits:
2685 multiclass N3V_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2686 InstrItinClass itinD16, InstrItinClass itinD32,
2687 InstrItinClass itinQ16, InstrItinClass itinQ32,
2688 string OpcodeStr, string Dt,
2689 SDNode OpNode, bit Commutable = 0> {
2690 // 64-bit vector types.
2691 def v8i8 : N3VD<op24, op23, 0b00, op11_8, op4, itinD16,
2692 OpcodeStr, !strconcat(Dt, "8"),
2693 v8i8, v8i8, OpNode, Commutable>;
2694 def v4i16 : N3VD<op24, op23, 0b01, op11_8, op4, itinD16,
2695 OpcodeStr, !strconcat(Dt, "16"),
2696 v4i16, v4i16, OpNode, Commutable>;
2697 def v2i32 : N3VD<op24, op23, 0b10, op11_8, op4, itinD32,
2698 OpcodeStr, !strconcat(Dt, "32"),
2699 v2i32, v2i32, OpNode, Commutable>;
2701 // 128-bit vector types.
2702 def v16i8 : N3VQ<op24, op23, 0b00, op11_8, op4, itinQ16,
2703 OpcodeStr, !strconcat(Dt, "8"),
2704 v16i8, v16i8, OpNode, Commutable>;
2705 def v8i16 : N3VQ<op24, op23, 0b01, op11_8, op4, itinQ16,
2706 OpcodeStr, !strconcat(Dt, "16"),
2707 v8i16, v8i16, OpNode, Commutable>;
2708 def v4i32 : N3VQ<op24, op23, 0b10, op11_8, op4, itinQ32,
2709 OpcodeStr, !strconcat(Dt, "32"),
2710 v4i32, v4i32, OpNode, Commutable>;
2713 multiclass N3VSL_HS<bits<4> op11_8, string OpcodeStr, string Dt, SDNode ShOp> {
2714 def v4i16 : N3VDSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
2716 def v2i32 : N3VDSL<0b10, op11_8, IIC_VMULi32D, OpcodeStr, !strconcat(Dt,"32"),
2718 def v8i16 : N3VQSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
2719 v8i16, v4i16, ShOp>;
2720 def v4i32 : N3VQSL<0b10, op11_8, IIC_VMULi32Q, OpcodeStr, !strconcat(Dt,"32"),
2721 v4i32, v2i32, ShOp>;
2724 // ....then also with element size 64 bits:
2725 multiclass N3V_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2726 InstrItinClass itinD, InstrItinClass itinQ,
2727 string OpcodeStr, string Dt,
2728 SDNode OpNode, bit Commutable = 0>
2729 : N3V_QHS<op24, op23, op11_8, op4, itinD, itinD, itinQ, itinQ,
2730 OpcodeStr, Dt, OpNode, Commutable> {
2731 def v1i64 : N3VD<op24, op23, 0b11, op11_8, op4, itinD,
2732 OpcodeStr, !strconcat(Dt, "64"),
2733 v1i64, v1i64, OpNode, Commutable>;
2734 def v2i64 : N3VQ<op24, op23, 0b11, op11_8, op4, itinQ,
2735 OpcodeStr, !strconcat(Dt, "64"),
2736 v2i64, v2i64, OpNode, Commutable>;
2740 // Neon 3-register vector intrinsics.
2742 // First with only element sizes of 16 and 32 bits:
2743 multiclass N3VInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2744 InstrItinClass itinD16, InstrItinClass itinD32,
2745 InstrItinClass itinQ16, InstrItinClass itinQ32,
2746 string OpcodeStr, string Dt,
2747 Intrinsic IntOp, bit Commutable = 0> {
2748 // 64-bit vector types.
2749 def v4i16 : N3VDInt<op24, op23, 0b01, op11_8, op4, f, itinD16,
2750 OpcodeStr, !strconcat(Dt, "16"),
2751 v4i16, v4i16, IntOp, Commutable>;
2752 def v2i32 : N3VDInt<op24, op23, 0b10, op11_8, op4, f, itinD32,
2753 OpcodeStr, !strconcat(Dt, "32"),
2754 v2i32, v2i32, IntOp, Commutable>;
2756 // 128-bit vector types.
2757 def v8i16 : N3VQInt<op24, op23, 0b01, op11_8, op4, f, itinQ16,
2758 OpcodeStr, !strconcat(Dt, "16"),
2759 v8i16, v8i16, IntOp, Commutable>;
2760 def v4i32 : N3VQInt<op24, op23, 0b10, op11_8, op4, f, itinQ32,
2761 OpcodeStr, !strconcat(Dt, "32"),
2762 v4i32, v4i32, IntOp, Commutable>;
2764 multiclass N3VInt_HSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2765 InstrItinClass itinD16, InstrItinClass itinD32,
2766 InstrItinClass itinQ16, InstrItinClass itinQ32,
2767 string OpcodeStr, string Dt,
2769 // 64-bit vector types.
2770 def v4i16 : N3VDIntSh<op24, op23, 0b01, op11_8, op4, f, itinD16,
2771 OpcodeStr, !strconcat(Dt, "16"),
2772 v4i16, v4i16, IntOp>;
2773 def v2i32 : N3VDIntSh<op24, op23, 0b10, op11_8, op4, f, itinD32,
2774 OpcodeStr, !strconcat(Dt, "32"),
2775 v2i32, v2i32, IntOp>;
2777 // 128-bit vector types.
2778 def v8i16 : N3VQIntSh<op24, op23, 0b01, op11_8, op4, f, itinQ16,
2779 OpcodeStr, !strconcat(Dt, "16"),
2780 v8i16, v8i16, IntOp>;
2781 def v4i32 : N3VQIntSh<op24, op23, 0b10, op11_8, op4, f, itinQ32,
2782 OpcodeStr, !strconcat(Dt, "32"),
2783 v4i32, v4i32, IntOp>;
2786 multiclass N3VIntSL_HS<bits<4> op11_8,
2787 InstrItinClass itinD16, InstrItinClass itinD32,
2788 InstrItinClass itinQ16, InstrItinClass itinQ32,
2789 string OpcodeStr, string Dt, Intrinsic IntOp> {
2790 def v4i16 : N3VDIntSL16<0b01, op11_8, itinD16,
2791 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp>;
2792 def v2i32 : N3VDIntSL<0b10, op11_8, itinD32,
2793 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp>;
2794 def v8i16 : N3VQIntSL16<0b01, op11_8, itinQ16,
2795 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16, IntOp>;
2796 def v4i32 : N3VQIntSL<0b10, op11_8, itinQ32,
2797 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32, IntOp>;
2800 // ....then also with element size of 8 bits:
2801 multiclass N3VInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2802 InstrItinClass itinD16, InstrItinClass itinD32,
2803 InstrItinClass itinQ16, InstrItinClass itinQ32,
2804 string OpcodeStr, string Dt,
2805 Intrinsic IntOp, bit Commutable = 0>
2806 : N3VInt_HS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2807 OpcodeStr, Dt, IntOp, Commutable> {
2808 def v8i8 : N3VDInt<op24, op23, 0b00, op11_8, op4, f, itinD16,
2809 OpcodeStr, !strconcat(Dt, "8"),
2810 v8i8, v8i8, IntOp, Commutable>;
2811 def v16i8 : N3VQInt<op24, op23, 0b00, op11_8, op4, f, itinQ16,
2812 OpcodeStr, !strconcat(Dt, "8"),
2813 v16i8, v16i8, IntOp, Commutable>;
2815 multiclass N3VInt_QHSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2816 InstrItinClass itinD16, InstrItinClass itinD32,
2817 InstrItinClass itinQ16, InstrItinClass itinQ32,
2818 string OpcodeStr, string Dt,
2820 : N3VInt_HSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2821 OpcodeStr, Dt, IntOp> {
2822 def v8i8 : N3VDIntSh<op24, op23, 0b00, op11_8, op4, f, itinD16,
2823 OpcodeStr, !strconcat(Dt, "8"),
2825 def v16i8 : N3VQIntSh<op24, op23, 0b00, op11_8, op4, f, itinQ16,
2826 OpcodeStr, !strconcat(Dt, "8"),
2827 v16i8, v16i8, IntOp>;
2831 // ....then also with element size of 64 bits:
2832 multiclass N3VInt_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2833 InstrItinClass itinD16, InstrItinClass itinD32,
2834 InstrItinClass itinQ16, InstrItinClass itinQ32,
2835 string OpcodeStr, string Dt,
2836 Intrinsic IntOp, bit Commutable = 0>
2837 : N3VInt_QHS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2838 OpcodeStr, Dt, IntOp, Commutable> {
2839 def v1i64 : N3VDInt<op24, op23, 0b11, op11_8, op4, f, itinD32,
2840 OpcodeStr, !strconcat(Dt, "64"),
2841 v1i64, v1i64, IntOp, Commutable>;
2842 def v2i64 : N3VQInt<op24, op23, 0b11, op11_8, op4, f, itinQ32,
2843 OpcodeStr, !strconcat(Dt, "64"),
2844 v2i64, v2i64, IntOp, Commutable>;
2846 multiclass N3VInt_QHSDSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2847 InstrItinClass itinD16, InstrItinClass itinD32,
2848 InstrItinClass itinQ16, InstrItinClass itinQ32,
2849 string OpcodeStr, string Dt,
2851 : N3VInt_QHSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2852 OpcodeStr, Dt, IntOp> {
2853 def v1i64 : N3VDIntSh<op24, op23, 0b11, op11_8, op4, f, itinD32,
2854 OpcodeStr, !strconcat(Dt, "64"),
2855 v1i64, v1i64, IntOp>;
2856 def v2i64 : N3VQIntSh<op24, op23, 0b11, op11_8, op4, f, itinQ32,
2857 OpcodeStr, !strconcat(Dt, "64"),
2858 v2i64, v2i64, IntOp>;
2861 // Neon Narrowing 3-register vector intrinsics,
2862 // source operand element sizes of 16, 32 and 64 bits:
2863 multiclass N3VNInt_HSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2864 string OpcodeStr, string Dt,
2865 Intrinsic IntOp, bit Commutable = 0> {
2866 def v8i8 : N3VNInt<op24, op23, 0b00, op11_8, op4,
2867 OpcodeStr, !strconcat(Dt, "16"),
2868 v8i8, v8i16, IntOp, Commutable>;
2869 def v4i16 : N3VNInt<op24, op23, 0b01, op11_8, op4,
2870 OpcodeStr, !strconcat(Dt, "32"),
2871 v4i16, v4i32, IntOp, Commutable>;
2872 def v2i32 : N3VNInt<op24, op23, 0b10, op11_8, op4,
2873 OpcodeStr, !strconcat(Dt, "64"),
2874 v2i32, v2i64, IntOp, Commutable>;
2878 // Neon Long 3-register vector operations.
2880 multiclass N3VL_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2881 InstrItinClass itin16, InstrItinClass itin32,
2882 string OpcodeStr, string Dt,
2883 SDNode OpNode, bit Commutable = 0> {
2884 def v8i16 : N3VL<op24, op23, 0b00, op11_8, op4, itin16,
2885 OpcodeStr, !strconcat(Dt, "8"),
2886 v8i16, v8i8, OpNode, Commutable>;
2887 def v4i32 : N3VL<op24, op23, 0b01, op11_8, op4, itin16,
2888 OpcodeStr, !strconcat(Dt, "16"),
2889 v4i32, v4i16, OpNode, Commutable>;
2890 def v2i64 : N3VL<op24, op23, 0b10, op11_8, op4, itin32,
2891 OpcodeStr, !strconcat(Dt, "32"),
2892 v2i64, v2i32, OpNode, Commutable>;
2895 multiclass N3VLSL_HS<bit op24, bits<4> op11_8,
2896 InstrItinClass itin, string OpcodeStr, string Dt,
2898 def v4i16 : N3VLSL16<op24, 0b01, op11_8, itin, OpcodeStr,
2899 !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
2900 def v2i32 : N3VLSL<op24, 0b10, op11_8, itin, OpcodeStr,
2901 !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
2904 multiclass N3VLExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2905 InstrItinClass itin16, InstrItinClass itin32,
2906 string OpcodeStr, string Dt,
2907 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
2908 def v8i16 : N3VLExt<op24, op23, 0b00, op11_8, op4, itin16,
2909 OpcodeStr, !strconcat(Dt, "8"),
2910 v8i16, v8i8, OpNode, ExtOp, Commutable>;
2911 def v4i32 : N3VLExt<op24, op23, 0b01, op11_8, op4, itin16,
2912 OpcodeStr, !strconcat(Dt, "16"),
2913 v4i32, v4i16, OpNode, ExtOp, Commutable>;
2914 def v2i64 : N3VLExt<op24, op23, 0b10, op11_8, op4, itin32,
2915 OpcodeStr, !strconcat(Dt, "32"),
2916 v2i64, v2i32, OpNode, ExtOp, Commutable>;
2919 // Neon Long 3-register vector intrinsics.
2921 // First with only element sizes of 16 and 32 bits:
2922 multiclass N3VLInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
2923 InstrItinClass itin16, InstrItinClass itin32,
2924 string OpcodeStr, string Dt,
2925 Intrinsic IntOp, bit Commutable = 0> {
2926 def v4i32 : N3VLInt<op24, op23, 0b01, op11_8, op4, itin16,
2927 OpcodeStr, !strconcat(Dt, "16"),
2928 v4i32, v4i16, IntOp, Commutable>;
2929 def v2i64 : N3VLInt<op24, op23, 0b10, op11_8, op4, itin32,
2930 OpcodeStr, !strconcat(Dt, "32"),
2931 v2i64, v2i32, IntOp, Commutable>;
2934 multiclass N3VLIntSL_HS<bit op24, bits<4> op11_8,
2935 InstrItinClass itin, string OpcodeStr, string Dt,
2937 def v4i16 : N3VLIntSL16<op24, 0b01, op11_8, itin,
2938 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
2939 def v2i32 : N3VLIntSL<op24, 0b10, op11_8, itin,
2940 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
2943 // ....then also with element size of 8 bits:
2944 multiclass N3VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2945 InstrItinClass itin16, InstrItinClass itin32,
2946 string OpcodeStr, string Dt,
2947 Intrinsic IntOp, bit Commutable = 0>
2948 : N3VLInt_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt,
2949 IntOp, Commutable> {
2950 def v8i16 : N3VLInt<op24, op23, 0b00, op11_8, op4, itin16,
2951 OpcodeStr, !strconcat(Dt, "8"),
2952 v8i16, v8i8, IntOp, Commutable>;
2955 // ....with explicit extend (VABDL).
2956 multiclass N3VLIntExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2957 InstrItinClass itin, string OpcodeStr, string Dt,
2958 Intrinsic IntOp, SDNode ExtOp, bit Commutable = 0> {
2959 def v8i16 : N3VLIntExt<op24, op23, 0b00, op11_8, op4, itin,
2960 OpcodeStr, !strconcat(Dt, "8"),
2961 v8i16, v8i8, IntOp, ExtOp, Commutable>;
2962 def v4i32 : N3VLIntExt<op24, op23, 0b01, op11_8, op4, itin,
2963 OpcodeStr, !strconcat(Dt, "16"),
2964 v4i32, v4i16, IntOp, ExtOp, Commutable>;
2965 def v2i64 : N3VLIntExt<op24, op23, 0b10, op11_8, op4, itin,
2966 OpcodeStr, !strconcat(Dt, "32"),
2967 v2i64, v2i32, IntOp, ExtOp, Commutable>;
2971 // Neon Wide 3-register vector intrinsics,
2972 // source operand element sizes of 8, 16 and 32 bits:
2973 multiclass N3VW_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2974 string OpcodeStr, string Dt,
2975 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
2976 def v8i16 : N3VW<op24, op23, 0b00, op11_8, op4,
2977 OpcodeStr, !strconcat(Dt, "8"),
2978 v8i16, v8i8, OpNode, ExtOp, Commutable>;
2979 def v4i32 : N3VW<op24, op23, 0b01, op11_8, op4,
2980 OpcodeStr, !strconcat(Dt, "16"),
2981 v4i32, v4i16, OpNode, ExtOp, Commutable>;
2982 def v2i64 : N3VW<op24, op23, 0b10, op11_8, op4,
2983 OpcodeStr, !strconcat(Dt, "32"),
2984 v2i64, v2i32, OpNode, ExtOp, Commutable>;
2988 // Neon Multiply-Op vector operations,
2989 // element sizes of 8, 16 and 32 bits:
2990 multiclass N3VMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2991 InstrItinClass itinD16, InstrItinClass itinD32,
2992 InstrItinClass itinQ16, InstrItinClass itinQ32,
2993 string OpcodeStr, string Dt, SDNode OpNode> {
2994 // 64-bit vector types.
2995 def v8i8 : N3VDMulOp<op24, op23, 0b00, op11_8, op4, itinD16,
2996 OpcodeStr, !strconcat(Dt, "8"), v8i8, mul, OpNode>;
2997 def v4i16 : N3VDMulOp<op24, op23, 0b01, op11_8, op4, itinD16,
2998 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, OpNode>;
2999 def v2i32 : N3VDMulOp<op24, op23, 0b10, op11_8, op4, itinD32,
3000 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, OpNode>;
3002 // 128-bit vector types.
3003 def v16i8 : N3VQMulOp<op24, op23, 0b00, op11_8, op4, itinQ16,
3004 OpcodeStr, !strconcat(Dt, "8"), v16i8, mul, OpNode>;
3005 def v8i16 : N3VQMulOp<op24, op23, 0b01, op11_8, op4, itinQ16,
3006 OpcodeStr, !strconcat(Dt, "16"), v8i16, mul, OpNode>;
3007 def v4i32 : N3VQMulOp<op24, op23, 0b10, op11_8, op4, itinQ32,
3008 OpcodeStr, !strconcat(Dt, "32"), v4i32, mul, OpNode>;
3011 multiclass N3VMulOpSL_HS<bits<4> op11_8,
3012 InstrItinClass itinD16, InstrItinClass itinD32,
3013 InstrItinClass itinQ16, InstrItinClass itinQ32,
3014 string OpcodeStr, string Dt, SDNode ShOp> {
3015 def v4i16 : N3VDMulOpSL16<0b01, op11_8, itinD16,
3016 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, ShOp>;
3017 def v2i32 : N3VDMulOpSL<0b10, op11_8, itinD32,
3018 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, ShOp>;
3019 def v8i16 : N3VQMulOpSL16<0b01, op11_8, itinQ16,
3020 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16,
3022 def v4i32 : N3VQMulOpSL<0b10, op11_8, itinQ32,
3023 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32,
3027 // Neon Intrinsic-Op vector operations,
3028 // element sizes of 8, 16 and 32 bits:
3029 multiclass N3VIntOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3030 InstrItinClass itinD, InstrItinClass itinQ,
3031 string OpcodeStr, string Dt, Intrinsic IntOp,
3033 // 64-bit vector types.
3034 def v8i8 : N3VDIntOp<op24, op23, 0b00, op11_8, op4, itinD,
3035 OpcodeStr, !strconcat(Dt, "8"), v8i8, IntOp, OpNode>;
3036 def v4i16 : N3VDIntOp<op24, op23, 0b01, op11_8, op4, itinD,
3037 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp, OpNode>;
3038 def v2i32 : N3VDIntOp<op24, op23, 0b10, op11_8, op4, itinD,
3039 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp, OpNode>;
3041 // 128-bit vector types.
3042 def v16i8 : N3VQIntOp<op24, op23, 0b00, op11_8, op4, itinQ,
3043 OpcodeStr, !strconcat(Dt, "8"), v16i8, IntOp, OpNode>;
3044 def v8i16 : N3VQIntOp<op24, op23, 0b01, op11_8, op4, itinQ,
3045 OpcodeStr, !strconcat(Dt, "16"), v8i16, IntOp, OpNode>;
3046 def v4i32 : N3VQIntOp<op24, op23, 0b10, op11_8, op4, itinQ,
3047 OpcodeStr, !strconcat(Dt, "32"), v4i32, IntOp, OpNode>;
3050 // Neon 3-argument intrinsics,
3051 // element sizes of 8, 16 and 32 bits:
3052 multiclass N3VInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3053 InstrItinClass itinD, InstrItinClass itinQ,
3054 string OpcodeStr, string Dt, Intrinsic IntOp> {
3055 // 64-bit vector types.
3056 def v8i8 : N3VDInt3<op24, op23, 0b00, op11_8, op4, itinD,
3057 OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
3058 def v4i16 : N3VDInt3<op24, op23, 0b01, op11_8, op4, itinD,
3059 OpcodeStr, !strconcat(Dt, "16"), v4i16, v4i16, IntOp>;
3060 def v2i32 : N3VDInt3<op24, op23, 0b10, op11_8, op4, itinD,
3061 OpcodeStr, !strconcat(Dt, "32"), v2i32, v2i32, IntOp>;
3063 // 128-bit vector types.
3064 def v16i8 : N3VQInt3<op24, op23, 0b00, op11_8, op4, itinQ,
3065 OpcodeStr, !strconcat(Dt, "8"), v16i8, v16i8, IntOp>;
3066 def v8i16 : N3VQInt3<op24, op23, 0b01, op11_8, op4, itinQ,
3067 OpcodeStr, !strconcat(Dt, "16"), v8i16, v8i16, IntOp>;
3068 def v4i32 : N3VQInt3<op24, op23, 0b10, op11_8, op4, itinQ,
3069 OpcodeStr, !strconcat(Dt, "32"), v4i32, v4i32, IntOp>;
3073 // Neon Long Multiply-Op vector operations,
3074 // element sizes of 8, 16 and 32 bits:
3075 multiclass N3VLMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3076 InstrItinClass itin16, InstrItinClass itin32,
3077 string OpcodeStr, string Dt, SDNode MulOp,
3079 def v8i16 : N3VLMulOp<op24, op23, 0b00, op11_8, op4, itin16, OpcodeStr,
3080 !strconcat(Dt, "8"), v8i16, v8i8, MulOp, OpNode>;
3081 def v4i32 : N3VLMulOp<op24, op23, 0b01, op11_8, op4, itin16, OpcodeStr,
3082 !strconcat(Dt, "16"), v4i32, v4i16, MulOp, OpNode>;
3083 def v2i64 : N3VLMulOp<op24, op23, 0b10, op11_8, op4, itin32, OpcodeStr,
3084 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
3087 multiclass N3VLMulOpSL_HS<bit op24, bits<4> op11_8, string OpcodeStr,
3088 string Dt, SDNode MulOp, SDNode OpNode> {
3089 def v4i16 : N3VLMulOpSL16<op24, 0b01, op11_8, IIC_VMACi16D, OpcodeStr,
3090 !strconcat(Dt,"16"), v4i32, v4i16, MulOp, OpNode>;
3091 def v2i32 : N3VLMulOpSL<op24, 0b10, op11_8, IIC_VMACi32D, OpcodeStr,
3092 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
3096 // Neon Long 3-argument intrinsics.
3098 // First with only element sizes of 16 and 32 bits:
3099 multiclass N3VLInt3_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
3100 InstrItinClass itin16, InstrItinClass itin32,
3101 string OpcodeStr, string Dt, Intrinsic IntOp> {
3102 def v4i32 : N3VLInt3<op24, op23, 0b01, op11_8, op4, itin16,
3103 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
3104 def v2i64 : N3VLInt3<op24, op23, 0b10, op11_8, op4, itin32,
3105 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3108 multiclass N3VLInt3SL_HS<bit op24, bits<4> op11_8,
3109 string OpcodeStr, string Dt, Intrinsic IntOp> {
3110 def v4i16 : N3VLInt3SL16<op24, 0b01, op11_8, IIC_VMACi16D,
3111 OpcodeStr, !strconcat(Dt,"16"), v4i32, v4i16, IntOp>;
3112 def v2i32 : N3VLInt3SL<op24, 0b10, op11_8, IIC_VMACi32D,
3113 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3116 // ....then also with element size of 8 bits:
3117 multiclass N3VLInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3118 InstrItinClass itin16, InstrItinClass itin32,
3119 string OpcodeStr, string Dt, Intrinsic IntOp>
3120 : N3VLInt3_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt, IntOp> {
3121 def v8i16 : N3VLInt3<op24, op23, 0b00, op11_8, op4, itin16,
3122 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
3125 // ....with explicit extend (VABAL).
3126 multiclass N3VLIntExtOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3127 InstrItinClass itin, string OpcodeStr, string Dt,
3128 Intrinsic IntOp, SDNode ExtOp, SDNode OpNode> {
3129 def v8i16 : N3VLIntExtOp<op24, op23, 0b00, op11_8, op4, itin,
3130 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8,
3131 IntOp, ExtOp, OpNode>;
3132 def v4i32 : N3VLIntExtOp<op24, op23, 0b01, op11_8, op4, itin,
3133 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16,
3134 IntOp, ExtOp, OpNode>;
3135 def v2i64 : N3VLIntExtOp<op24, op23, 0b10, op11_8, op4, itin,
3136 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32,
3137 IntOp, ExtOp, OpNode>;
3141 // Neon Pairwise long 2-register intrinsics,
3142 // element sizes of 8, 16 and 32 bits:
3143 multiclass N2VPLInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3144 bits<5> op11_7, bit op4,
3145 string OpcodeStr, string Dt, Intrinsic IntOp> {
3146 // 64-bit vector types.
3147 def v8i8 : N2VDPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3148 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
3149 def v4i16 : N2VDPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3150 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
3151 def v2i32 : N2VDPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3152 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
3154 // 128-bit vector types.
3155 def v16i8 : N2VQPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3156 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
3157 def v8i16 : N2VQPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3158 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
3159 def v4i32 : N2VQPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3160 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
3164 // Neon Pairwise long 2-register accumulate intrinsics,
3165 // element sizes of 8, 16 and 32 bits:
3166 multiclass N2VPLInt2_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3167 bits<5> op11_7, bit op4,
3168 string OpcodeStr, string Dt, Intrinsic IntOp> {
3169 // 64-bit vector types.
3170 def v8i8 : N2VDPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3171 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
3172 def v4i16 : N2VDPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3173 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
3174 def v2i32 : N2VDPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3175 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
3177 // 128-bit vector types.
3178 def v16i8 : N2VQPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3179 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
3180 def v8i16 : N2VQPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3181 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
3182 def v4i32 : N2VQPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3183 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
3187 // Neon 2-register vector shift by immediate,
3188 // with f of either N2RegVShLFrm or N2RegVShRFrm
3189 // element sizes of 8, 16, 32 and 64 bits:
3190 multiclass N2VShL_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3191 InstrItinClass itin, string OpcodeStr, string Dt,
3193 // 64-bit vector types.
3194 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3195 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
3196 let Inst{21-19} = 0b001; // imm6 = 001xxx
3198 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3199 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
3200 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3202 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3203 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
3204 let Inst{21} = 0b1; // imm6 = 1xxxxx
3206 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, N2RegVShLFrm, itin, i32imm,
3207 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
3210 // 128-bit vector types.
3211 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3212 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
3213 let Inst{21-19} = 0b001; // imm6 = 001xxx
3215 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3216 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
3217 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3219 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3220 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
3221 let Inst{21} = 0b1; // imm6 = 1xxxxx
3223 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, N2RegVShLFrm, itin, i32imm,
3224 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
3227 multiclass N2VShR_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3228 InstrItinClass itin, string OpcodeStr, string Dt,
3230 // 64-bit vector types.
3231 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm8,
3232 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
3233 let Inst{21-19} = 0b001; // imm6 = 001xxx
3235 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm16,
3236 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
3237 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3239 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm32,
3240 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
3241 let Inst{21} = 0b1; // imm6 = 1xxxxx
3243 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, N2RegVShRFrm, itin, shr_imm64,
3244 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
3247 // 128-bit vector types.
3248 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm8,
3249 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
3250 let Inst{21-19} = 0b001; // imm6 = 001xxx
3252 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm16,
3253 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
3254 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3256 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm32,
3257 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
3258 let Inst{21} = 0b1; // imm6 = 1xxxxx
3260 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, N2RegVShRFrm, itin, shr_imm64,
3261 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
3265 // Neon Shift-Accumulate vector operations,
3266 // element sizes of 8, 16, 32 and 64 bits:
3267 multiclass N2VShAdd_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3268 string OpcodeStr, string Dt, SDNode ShOp> {
3269 // 64-bit vector types.
3270 def v8i8 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm8,
3271 OpcodeStr, !strconcat(Dt, "8"), v8i8, ShOp> {
3272 let Inst{21-19} = 0b001; // imm6 = 001xxx
3274 def v4i16 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm16,
3275 OpcodeStr, !strconcat(Dt, "16"), v4i16, ShOp> {
3276 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3278 def v2i32 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm32,
3279 OpcodeStr, !strconcat(Dt, "32"), v2i32, ShOp> {
3280 let Inst{21} = 0b1; // imm6 = 1xxxxx
3282 def v1i64 : N2VDShAdd<op24, op23, op11_8, 1, op4, shr_imm64,
3283 OpcodeStr, !strconcat(Dt, "64"), v1i64, ShOp>;
3286 // 128-bit vector types.
3287 def v16i8 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm8,
3288 OpcodeStr, !strconcat(Dt, "8"), v16i8, ShOp> {
3289 let Inst{21-19} = 0b001; // imm6 = 001xxx
3291 def v8i16 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm16,
3292 OpcodeStr, !strconcat(Dt, "16"), v8i16, ShOp> {
3293 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3295 def v4i32 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm32,
3296 OpcodeStr, !strconcat(Dt, "32"), v4i32, ShOp> {
3297 let Inst{21} = 0b1; // imm6 = 1xxxxx
3299 def v2i64 : N2VQShAdd<op24, op23, op11_8, 1, op4, shr_imm64,
3300 OpcodeStr, !strconcat(Dt, "64"), v2i64, ShOp>;
3304 // Neon Shift-Insert vector operations,
3305 // with f of either N2RegVShLFrm or N2RegVShRFrm
3306 // element sizes of 8, 16, 32 and 64 bits:
3307 multiclass N2VShInsL_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3309 // 64-bit vector types.
3310 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3311 N2RegVShLFrm, OpcodeStr, "8", v8i8, NEONvsli> {
3312 let Inst{21-19} = 0b001; // imm6 = 001xxx
3314 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3315 N2RegVShLFrm, OpcodeStr, "16", v4i16, NEONvsli> {
3316 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3318 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3319 N2RegVShLFrm, OpcodeStr, "32", v2i32, NEONvsli> {
3320 let Inst{21} = 0b1; // imm6 = 1xxxxx
3322 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4, i32imm,
3323 N2RegVShLFrm, OpcodeStr, "64", v1i64, NEONvsli>;
3326 // 128-bit vector types.
3327 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3328 N2RegVShLFrm, OpcodeStr, "8", v16i8, NEONvsli> {
3329 let Inst{21-19} = 0b001; // imm6 = 001xxx
3331 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3332 N2RegVShLFrm, OpcodeStr, "16", v8i16, NEONvsli> {
3333 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3335 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3336 N2RegVShLFrm, OpcodeStr, "32", v4i32, NEONvsli> {
3337 let Inst{21} = 0b1; // imm6 = 1xxxxx
3339 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4, i32imm,
3340 N2RegVShLFrm, OpcodeStr, "64", v2i64, NEONvsli>;
3343 multiclass N2VShInsR_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3345 // 64-bit vector types.
3346 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm8,
3347 N2RegVShRFrm, OpcodeStr, "8", v8i8, NEONvsri> {
3348 let Inst{21-19} = 0b001; // imm6 = 001xxx
3350 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm16,
3351 N2RegVShRFrm, OpcodeStr, "16", v4i16, NEONvsri> {
3352 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3354 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm32,
3355 N2RegVShRFrm, OpcodeStr, "32", v2i32, NEONvsri> {
3356 let Inst{21} = 0b1; // imm6 = 1xxxxx
3358 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4, shr_imm64,
3359 N2RegVShRFrm, OpcodeStr, "64", v1i64, NEONvsri>;
3362 // 128-bit vector types.
3363 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm8,
3364 N2RegVShRFrm, OpcodeStr, "8", v16i8, NEONvsri> {
3365 let Inst{21-19} = 0b001; // imm6 = 001xxx
3367 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm16,
3368 N2RegVShRFrm, OpcodeStr, "16", v8i16, NEONvsri> {
3369 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3371 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm32,
3372 N2RegVShRFrm, OpcodeStr, "32", v4i32, NEONvsri> {
3373 let Inst{21} = 0b1; // imm6 = 1xxxxx
3375 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4, shr_imm64,
3376 N2RegVShRFrm, OpcodeStr, "64", v2i64, NEONvsri>;
3380 // Neon Shift Long operations,
3381 // element sizes of 8, 16, 32 bits:
3382 multiclass N2VLSh_QHS<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
3383 bit op4, string OpcodeStr, string Dt, SDNode OpNode> {
3384 def v8i16 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3385 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode> {
3386 let Inst{21-19} = 0b001; // imm6 = 001xxx
3388 def v4i32 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3389 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode> {
3390 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3392 def v2i64 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3393 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode> {
3394 let Inst{21} = 0b1; // imm6 = 1xxxxx
3398 // Neon Shift Narrow operations,
3399 // element sizes of 16, 32, 64 bits:
3400 multiclass N2VNSh_HSD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
3401 bit op4, InstrItinClass itin, string OpcodeStr, string Dt,
3403 def v8i8 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3404 OpcodeStr, !strconcat(Dt, "16"),
3405 v8i8, v8i16, shr_imm8, OpNode> {
3406 let Inst{21-19} = 0b001; // imm6 = 001xxx
3408 def v4i16 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3409 OpcodeStr, !strconcat(Dt, "32"),
3410 v4i16, v4i32, shr_imm16, OpNode> {
3411 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3413 def v2i32 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3414 OpcodeStr, !strconcat(Dt, "64"),
3415 v2i32, v2i64, shr_imm32, OpNode> {
3416 let Inst{21} = 0b1; // imm6 = 1xxxxx
3420 //===----------------------------------------------------------------------===//
3421 // Instruction Definitions.
3422 //===----------------------------------------------------------------------===//
3424 // Vector Add Operations.
3426 // VADD : Vector Add (integer and floating-point)
3427 defm VADD : N3V_QHSD<0, 0, 0b1000, 0, IIC_VBINiD, IIC_VBINiQ, "vadd", "i",
3429 def VADDfd : N3VD<0, 0, 0b00, 0b1101, 0, IIC_VBIND, "vadd", "f32",
3430 v2f32, v2f32, fadd, 1>;
3431 def VADDfq : N3VQ<0, 0, 0b00, 0b1101, 0, IIC_VBINQ, "vadd", "f32",
3432 v4f32, v4f32, fadd, 1>;
3433 // VADDL : Vector Add Long (Q = D + D)
3434 defm VADDLs : N3VLExt_QHS<0,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
3435 "vaddl", "s", add, sext, 1>;
3436 defm VADDLu : N3VLExt_QHS<1,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
3437 "vaddl", "u", add, zext, 1>;
3438 // VADDW : Vector Add Wide (Q = Q + D)
3439 defm VADDWs : N3VW_QHS<0,1,0b0001,0, "vaddw", "s", add, sext, 0>;
3440 defm VADDWu : N3VW_QHS<1,1,0b0001,0, "vaddw", "u", add, zext, 0>;
3441 // VHADD : Vector Halving Add
3442 defm VHADDs : N3VInt_QHS<0, 0, 0b0000, 0, N3RegFrm,
3443 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3444 "vhadd", "s", int_arm_neon_vhadds, 1>;
3445 defm VHADDu : N3VInt_QHS<1, 0, 0b0000, 0, N3RegFrm,
3446 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3447 "vhadd", "u", int_arm_neon_vhaddu, 1>;
3448 // VRHADD : Vector Rounding Halving Add
3449 defm VRHADDs : N3VInt_QHS<0, 0, 0b0001, 0, N3RegFrm,
3450 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3451 "vrhadd", "s", int_arm_neon_vrhadds, 1>;
3452 defm VRHADDu : N3VInt_QHS<1, 0, 0b0001, 0, N3RegFrm,
3453 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3454 "vrhadd", "u", int_arm_neon_vrhaddu, 1>;
3455 // VQADD : Vector Saturating Add
3456 defm VQADDs : N3VInt_QHSD<0, 0, 0b0000, 1, N3RegFrm,
3457 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3458 "vqadd", "s", int_arm_neon_vqadds, 1>;
3459 defm VQADDu : N3VInt_QHSD<1, 0, 0b0000, 1, N3RegFrm,
3460 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3461 "vqadd", "u", int_arm_neon_vqaddu, 1>;
3462 // VADDHN : Vector Add and Narrow Returning High Half (D = Q + Q)
3463 defm VADDHN : N3VNInt_HSD<0,1,0b0100,0, "vaddhn", "i",
3464 int_arm_neon_vaddhn, 1>;
3465 // VRADDHN : Vector Rounding Add and Narrow Returning High Half (D = Q + Q)
3466 defm VRADDHN : N3VNInt_HSD<1,1,0b0100,0, "vraddhn", "i",
3467 int_arm_neon_vraddhn, 1>;
3469 // Vector Multiply Operations.
3471 // VMUL : Vector Multiply (integer, polynomial and floating-point)
3472 defm VMUL : N3V_QHS<0, 0, 0b1001, 1, IIC_VMULi16D, IIC_VMULi32D,
3473 IIC_VMULi16Q, IIC_VMULi32Q, "vmul", "i", mul, 1>;
3474 def VMULpd : N3VDInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16D, "vmul",
3475 "p8", v8i8, v8i8, int_arm_neon_vmulp, 1>;
3476 def VMULpq : N3VQInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16Q, "vmul",
3477 "p8", v16i8, v16i8, int_arm_neon_vmulp, 1>;
3478 def VMULfd : N3VD<1, 0, 0b00, 0b1101, 1, IIC_VFMULD, "vmul", "f32",
3479 v2f32, v2f32, fmul, 1>;
3480 def VMULfq : N3VQ<1, 0, 0b00, 0b1101, 1, IIC_VFMULQ, "vmul", "f32",
3481 v4f32, v4f32, fmul, 1>;
3482 defm VMULsl : N3VSL_HS<0b1000, "vmul", "i", mul>;
3483 def VMULslfd : N3VDSL<0b10, 0b1001, IIC_VBIND, "vmul", "f32", v2f32, fmul>;
3484 def VMULslfq : N3VQSL<0b10, 0b1001, IIC_VBINQ, "vmul", "f32", v4f32,
3487 def : Pat<(v8i16 (mul (v8i16 QPR:$src1),
3488 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
3489 (v8i16 (VMULslv8i16 (v8i16 QPR:$src1),
3490 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3491 (DSubReg_i16_reg imm:$lane))),
3492 (SubReg_i16_lane imm:$lane)))>;
3493 def : Pat<(v4i32 (mul (v4i32 QPR:$src1),
3494 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
3495 (v4i32 (VMULslv4i32 (v4i32 QPR:$src1),
3496 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3497 (DSubReg_i32_reg imm:$lane))),
3498 (SubReg_i32_lane imm:$lane)))>;
3499 def : Pat<(v4f32 (fmul (v4f32 QPR:$src1),
3500 (v4f32 (NEONvduplane (v4f32 QPR:$src2), imm:$lane)))),
3501 (v4f32 (VMULslfq (v4f32 QPR:$src1),
3502 (v2f32 (EXTRACT_SUBREG QPR:$src2,
3503 (DSubReg_i32_reg imm:$lane))),
3504 (SubReg_i32_lane imm:$lane)))>;
3506 // VQDMULH : Vector Saturating Doubling Multiply Returning High Half
3507 defm VQDMULH : N3VInt_HS<0, 0, 0b1011, 0, N3RegFrm, IIC_VMULi16D, IIC_VMULi32D,
3508 IIC_VMULi16Q, IIC_VMULi32Q,
3509 "vqdmulh", "s", int_arm_neon_vqdmulh, 1>;
3510 defm VQDMULHsl: N3VIntSL_HS<0b1100, IIC_VMULi16D, IIC_VMULi32D,
3511 IIC_VMULi16Q, IIC_VMULi32Q,
3512 "vqdmulh", "s", int_arm_neon_vqdmulh>;
3513 def : Pat<(v8i16 (int_arm_neon_vqdmulh (v8i16 QPR:$src1),
3514 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
3516 (v8i16 (VQDMULHslv8i16 (v8i16 QPR:$src1),
3517 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3518 (DSubReg_i16_reg imm:$lane))),
3519 (SubReg_i16_lane imm:$lane)))>;
3520 def : Pat<(v4i32 (int_arm_neon_vqdmulh (v4i32 QPR:$src1),
3521 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
3523 (v4i32 (VQDMULHslv4i32 (v4i32 QPR:$src1),
3524 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3525 (DSubReg_i32_reg imm:$lane))),
3526 (SubReg_i32_lane imm:$lane)))>;
3528 // VQRDMULH : Vector Rounding Saturating Doubling Multiply Returning High Half
3529 defm VQRDMULH : N3VInt_HS<1, 0, 0b1011, 0, N3RegFrm,
3530 IIC_VMULi16D,IIC_VMULi32D,IIC_VMULi16Q,IIC_VMULi32Q,
3531 "vqrdmulh", "s", int_arm_neon_vqrdmulh, 1>;
3532 defm VQRDMULHsl : N3VIntSL_HS<0b1101, IIC_VMULi16D, IIC_VMULi32D,
3533 IIC_VMULi16Q, IIC_VMULi32Q,
3534 "vqrdmulh", "s", int_arm_neon_vqrdmulh>;
3535 def : Pat<(v8i16 (int_arm_neon_vqrdmulh (v8i16 QPR:$src1),
3536 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
3538 (v8i16 (VQRDMULHslv8i16 (v8i16 QPR:$src1),
3539 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3540 (DSubReg_i16_reg imm:$lane))),
3541 (SubReg_i16_lane imm:$lane)))>;
3542 def : Pat<(v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$src1),
3543 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
3545 (v4i32 (VQRDMULHslv4i32 (v4i32 QPR:$src1),
3546 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3547 (DSubReg_i32_reg imm:$lane))),
3548 (SubReg_i32_lane imm:$lane)))>;
3550 // VMULL : Vector Multiply Long (integer and polynomial) (Q = D * D)
3551 defm VMULLs : N3VL_QHS<0,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
3552 "vmull", "s", NEONvmulls, 1>;
3553 defm VMULLu : N3VL_QHS<1,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
3554 "vmull", "u", NEONvmullu, 1>;
3555 def VMULLp : N3VLInt<0, 1, 0b00, 0b1110, 0, IIC_VMULi16D, "vmull", "p8",
3556 v8i16, v8i8, int_arm_neon_vmullp, 1>;
3557 defm VMULLsls : N3VLSL_HS<0, 0b1010, IIC_VMULi16D, "vmull", "s", NEONvmulls>;
3558 defm VMULLslu : N3VLSL_HS<1, 0b1010, IIC_VMULi16D, "vmull", "u", NEONvmullu>;
3560 // VQDMULL : Vector Saturating Doubling Multiply Long (Q = D * D)
3561 defm VQDMULL : N3VLInt_HS<0,1,0b1101,0, IIC_VMULi16D, IIC_VMULi32D,
3562 "vqdmull", "s", int_arm_neon_vqdmull, 1>;
3563 defm VQDMULLsl: N3VLIntSL_HS<0, 0b1011, IIC_VMULi16D,
3564 "vqdmull", "s", int_arm_neon_vqdmull>;
3566 // Vector Multiply-Accumulate and Multiply-Subtract Operations.
3568 // VMLA : Vector Multiply Accumulate (integer and floating-point)
3569 defm VMLA : N3VMulOp_QHS<0, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3570 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
3571 def VMLAfd : N3VDMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACD, "vmla", "f32",
3572 v2f32, fmul_su, fadd_mlx>,
3573 Requires<[HasNEON, UseFPVMLx]>;
3574 def VMLAfq : N3VQMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACQ, "vmla", "f32",
3575 v4f32, fmul_su, fadd_mlx>,
3576 Requires<[HasNEON, UseFPVMLx]>;
3577 defm VMLAsl : N3VMulOpSL_HS<0b0000, IIC_VMACi16D, IIC_VMACi32D,
3578 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
3579 def VMLAslfd : N3VDMulOpSL<0b10, 0b0001, IIC_VMACD, "vmla", "f32",
3580 v2f32, fmul_su, fadd_mlx>,
3581 Requires<[HasNEON, UseFPVMLx]>;
3582 def VMLAslfq : N3VQMulOpSL<0b10, 0b0001, IIC_VMACQ, "vmla", "f32",
3583 v4f32, v2f32, fmul_su, fadd_mlx>,
3584 Requires<[HasNEON, UseFPVMLx]>;
3586 def : Pat<(v8i16 (add (v8i16 QPR:$src1),
3587 (mul (v8i16 QPR:$src2),
3588 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
3589 (v8i16 (VMLAslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
3590 (v4i16 (EXTRACT_SUBREG QPR:$src3,
3591 (DSubReg_i16_reg imm:$lane))),
3592 (SubReg_i16_lane imm:$lane)))>;
3594 def : Pat<(v4i32 (add (v4i32 QPR:$src1),
3595 (mul (v4i32 QPR:$src2),
3596 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
3597 (v4i32 (VMLAslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
3598 (v2i32 (EXTRACT_SUBREG QPR:$src3,
3599 (DSubReg_i32_reg imm:$lane))),
3600 (SubReg_i32_lane imm:$lane)))>;
3602 def : Pat<(v4f32 (fadd_mlx (v4f32 QPR:$src1),
3603 (fmul_su (v4f32 QPR:$src2),
3604 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
3605 (v4f32 (VMLAslfq (v4f32 QPR:$src1),
3607 (v2f32 (EXTRACT_SUBREG QPR:$src3,
3608 (DSubReg_i32_reg imm:$lane))),
3609 (SubReg_i32_lane imm:$lane)))>,
3610 Requires<[HasNEON, UseFPVMLx]>;
3612 // VMLAL : Vector Multiply Accumulate Long (Q += D * D)
3613 defm VMLALs : N3VLMulOp_QHS<0,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
3614 "vmlal", "s", NEONvmulls, add>;
3615 defm VMLALu : N3VLMulOp_QHS<1,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
3616 "vmlal", "u", NEONvmullu, add>;
3618 defm VMLALsls : N3VLMulOpSL_HS<0, 0b0010, "vmlal", "s", NEONvmulls, add>;
3619 defm VMLALslu : N3VLMulOpSL_HS<1, 0b0010, "vmlal", "u", NEONvmullu, add>;
3621 // VQDMLAL : Vector Saturating Doubling Multiply Accumulate Long (Q += D * D)
3622 defm VQDMLAL : N3VLInt3_HS<0, 1, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3623 "vqdmlal", "s", int_arm_neon_vqdmlal>;
3624 defm VQDMLALsl: N3VLInt3SL_HS<0, 0b0011, "vqdmlal", "s", int_arm_neon_vqdmlal>;
3626 // VMLS : Vector Multiply Subtract (integer and floating-point)
3627 defm VMLS : N3VMulOp_QHS<1, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3628 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
3629 def VMLSfd : N3VDMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACD, "vmls", "f32",
3630 v2f32, fmul_su, fsub_mlx>,
3631 Requires<[HasNEON, UseFPVMLx]>;
3632 def VMLSfq : N3VQMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACQ, "vmls", "f32",
3633 v4f32, fmul_su, fsub_mlx>,
3634 Requires<[HasNEON, UseFPVMLx]>;
3635 defm VMLSsl : N3VMulOpSL_HS<0b0100, IIC_VMACi16D, IIC_VMACi32D,
3636 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
3637 def VMLSslfd : N3VDMulOpSL<0b10, 0b0101, IIC_VMACD, "vmls", "f32",
3638 v2f32, fmul_su, fsub_mlx>,
3639 Requires<[HasNEON, UseFPVMLx]>;
3640 def VMLSslfq : N3VQMulOpSL<0b10, 0b0101, IIC_VMACQ, "vmls", "f32",
3641 v4f32, v2f32, fmul_su, fsub_mlx>,
3642 Requires<[HasNEON, UseFPVMLx]>;
3644 def : Pat<(v8i16 (sub (v8i16 QPR:$src1),
3645 (mul (v8i16 QPR:$src2),
3646 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
3647 (v8i16 (VMLSslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
3648 (v4i16 (EXTRACT_SUBREG QPR:$src3,
3649 (DSubReg_i16_reg imm:$lane))),
3650 (SubReg_i16_lane imm:$lane)))>;
3652 def : Pat<(v4i32 (sub (v4i32 QPR:$src1),
3653 (mul (v4i32 QPR:$src2),
3654 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
3655 (v4i32 (VMLSslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
3656 (v2i32 (EXTRACT_SUBREG QPR:$src3,
3657 (DSubReg_i32_reg imm:$lane))),
3658 (SubReg_i32_lane imm:$lane)))>;
3660 def : Pat<(v4f32 (fsub_mlx (v4f32 QPR:$src1),
3661 (fmul_su (v4f32 QPR:$src2),
3662 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
3663 (v4f32 (VMLSslfq (v4f32 QPR:$src1), (v4f32 QPR:$src2),
3664 (v2f32 (EXTRACT_SUBREG QPR:$src3,
3665 (DSubReg_i32_reg imm:$lane))),
3666 (SubReg_i32_lane imm:$lane)))>,
3667 Requires<[HasNEON, UseFPVMLx]>;
3669 // VMLSL : Vector Multiply Subtract Long (Q -= D * D)
3670 defm VMLSLs : N3VLMulOp_QHS<0,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
3671 "vmlsl", "s", NEONvmulls, sub>;
3672 defm VMLSLu : N3VLMulOp_QHS<1,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
3673 "vmlsl", "u", NEONvmullu, sub>;
3675 defm VMLSLsls : N3VLMulOpSL_HS<0, 0b0110, "vmlsl", "s", NEONvmulls, sub>;
3676 defm VMLSLslu : N3VLMulOpSL_HS<1, 0b0110, "vmlsl", "u", NEONvmullu, sub>;
3678 // VQDMLSL : Vector Saturating Doubling Multiply Subtract Long (Q -= D * D)
3679 defm VQDMLSL : N3VLInt3_HS<0, 1, 0b1011, 0, IIC_VMACi16D, IIC_VMACi32D,
3680 "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
3681 defm VQDMLSLsl: N3VLInt3SL_HS<0, 0b111, "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
3683 // Vector Subtract Operations.
3685 // VSUB : Vector Subtract (integer and floating-point)
3686 defm VSUB : N3V_QHSD<1, 0, 0b1000, 0, IIC_VSUBiD, IIC_VSUBiQ,
3687 "vsub", "i", sub, 0>;
3688 def VSUBfd : N3VD<0, 0, 0b10, 0b1101, 0, IIC_VBIND, "vsub", "f32",
3689 v2f32, v2f32, fsub, 0>;
3690 def VSUBfq : N3VQ<0, 0, 0b10, 0b1101, 0, IIC_VBINQ, "vsub", "f32",
3691 v4f32, v4f32, fsub, 0>;
3692 // VSUBL : Vector Subtract Long (Q = D - D)
3693 defm VSUBLs : N3VLExt_QHS<0,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
3694 "vsubl", "s", sub, sext, 0>;
3695 defm VSUBLu : N3VLExt_QHS<1,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
3696 "vsubl", "u", sub, zext, 0>;
3697 // VSUBW : Vector Subtract Wide (Q = Q - D)
3698 defm VSUBWs : N3VW_QHS<0,1,0b0011,0, "vsubw", "s", sub, sext, 0>;
3699 defm VSUBWu : N3VW_QHS<1,1,0b0011,0, "vsubw", "u", sub, zext, 0>;
3700 // VHSUB : Vector Halving Subtract
3701 defm VHSUBs : N3VInt_QHS<0, 0, 0b0010, 0, N3RegFrm,
3702 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3703 "vhsub", "s", int_arm_neon_vhsubs, 0>;
3704 defm VHSUBu : N3VInt_QHS<1, 0, 0b0010, 0, N3RegFrm,
3705 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3706 "vhsub", "u", int_arm_neon_vhsubu, 0>;
3707 // VQSUB : Vector Saturing Subtract
3708 defm VQSUBs : N3VInt_QHSD<0, 0, 0b0010, 1, N3RegFrm,
3709 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3710 "vqsub", "s", int_arm_neon_vqsubs, 0>;
3711 defm VQSUBu : N3VInt_QHSD<1, 0, 0b0010, 1, N3RegFrm,
3712 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3713 "vqsub", "u", int_arm_neon_vqsubu, 0>;
3714 // VSUBHN : Vector Subtract and Narrow Returning High Half (D = Q - Q)
3715 defm VSUBHN : N3VNInt_HSD<0,1,0b0110,0, "vsubhn", "i",
3716 int_arm_neon_vsubhn, 0>;
3717 // VRSUBHN : Vector Rounding Subtract and Narrow Returning High Half (D=Q-Q)
3718 defm VRSUBHN : N3VNInt_HSD<1,1,0b0110,0, "vrsubhn", "i",
3719 int_arm_neon_vrsubhn, 0>;
3721 // Vector Comparisons.
3723 // VCEQ : Vector Compare Equal
3724 defm VCEQ : N3V_QHS<1, 0, 0b1000, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3725 IIC_VSUBi4Q, "vceq", "i", NEONvceq, 1>;
3726 def VCEQfd : N3VD<0,0,0b00,0b1110,0, IIC_VBIND, "vceq", "f32", v2i32, v2f32,
3728 def VCEQfq : N3VQ<0,0,0b00,0b1110,0, IIC_VBINQ, "vceq", "f32", v4i32, v4f32,
3731 defm VCEQz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00010, 0, "vceq", "i",
3732 "$Vd, $Vm, #0", NEONvceqz>;
3734 // VCGE : Vector Compare Greater Than or Equal
3735 defm VCGEs : N3V_QHS<0, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3736 IIC_VSUBi4Q, "vcge", "s", NEONvcge, 0>;
3737 defm VCGEu : N3V_QHS<1, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3738 IIC_VSUBi4Q, "vcge", "u", NEONvcgeu, 0>;
3739 def VCGEfd : N3VD<1,0,0b00,0b1110,0, IIC_VBIND, "vcge", "f32", v2i32, v2f32,
3741 def VCGEfq : N3VQ<1,0,0b00,0b1110,0, IIC_VBINQ, "vcge", "f32", v4i32, v4f32,
3744 defm VCGEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00001, 0, "vcge", "s",
3745 "$Vd, $Vm, #0", NEONvcgez>;
3746 defm VCLEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00011, 0, "vcle", "s",
3747 "$Vd, $Vm, #0", NEONvclez>;
3749 // VCGT : Vector Compare Greater Than
3750 defm VCGTs : N3V_QHS<0, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3751 IIC_VSUBi4Q, "vcgt", "s", NEONvcgt, 0>;
3752 defm VCGTu : N3V_QHS<1, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3753 IIC_VSUBi4Q, "vcgt", "u", NEONvcgtu, 0>;
3754 def VCGTfd : N3VD<1,0,0b10,0b1110,0, IIC_VBIND, "vcgt", "f32", v2i32, v2f32,
3756 def VCGTfq : N3VQ<1,0,0b10,0b1110,0, IIC_VBINQ, "vcgt", "f32", v4i32, v4f32,
3759 defm VCGTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00000, 0, "vcgt", "s",
3760 "$Vd, $Vm, #0", NEONvcgtz>;
3761 defm VCLTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00100, 0, "vclt", "s",
3762 "$Vd, $Vm, #0", NEONvcltz>;
3764 // VACGE : Vector Absolute Compare Greater Than or Equal (aka VCAGE)
3765 def VACGEd : N3VDInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacge",
3766 "f32", v2i32, v2f32, int_arm_neon_vacged, 0>;
3767 def VACGEq : N3VQInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacge",
3768 "f32", v4i32, v4f32, int_arm_neon_vacgeq, 0>;
3769 // VACGT : Vector Absolute Compare Greater Than (aka VCAGT)
3770 def VACGTd : N3VDInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacgt",
3771 "f32", v2i32, v2f32, int_arm_neon_vacgtd, 0>;
3772 def VACGTq : N3VQInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacgt",
3773 "f32", v4i32, v4f32, int_arm_neon_vacgtq, 0>;
3774 // VTST : Vector Test Bits
3775 defm VTST : N3V_QHS<0, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
3776 IIC_VBINi4Q, "vtst", "", NEONvtst, 1>;
3778 // Vector Bitwise Operations.
3780 def vnotd : PatFrag<(ops node:$in),
3781 (xor node:$in, (bitconvert (v8i8 NEONimmAllOnesV)))>;
3782 def vnotq : PatFrag<(ops node:$in),
3783 (xor node:$in, (bitconvert (v16i8 NEONimmAllOnesV)))>;
3786 // VAND : Vector Bitwise AND
3787 def VANDd : N3VDX<0, 0, 0b00, 0b0001, 1, IIC_VBINiD, "vand",
3788 v2i32, v2i32, and, 1>;
3789 def VANDq : N3VQX<0, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "vand",
3790 v4i32, v4i32, and, 1>;
3792 // VEOR : Vector Bitwise Exclusive OR
3793 def VEORd : N3VDX<1, 0, 0b00, 0b0001, 1, IIC_VBINiD, "veor",
3794 v2i32, v2i32, xor, 1>;
3795 def VEORq : N3VQX<1, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "veor",
3796 v4i32, v4i32, xor, 1>;
3798 // VORR : Vector Bitwise OR
3799 def VORRd : N3VDX<0, 0, 0b10, 0b0001, 1, IIC_VBINiD, "vorr",
3800 v2i32, v2i32, or, 1>;
3801 def VORRq : N3VQX<0, 0, 0b10, 0b0001, 1, IIC_VBINiQ, "vorr",
3802 v4i32, v4i32, or, 1>;
3804 def VORRiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 0, 1,
3805 (outs DPR:$Vd), (ins nImmSplatI16:$SIMM, DPR:$src),
3807 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
3809 (v4i16 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
3810 let Inst{9} = SIMM{9};
3813 def VORRiv2i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 0, 0, 1,
3814 (outs DPR:$Vd), (ins nImmSplatI32:$SIMM, DPR:$src),
3816 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
3818 (v2i32 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
3819 let Inst{10-9} = SIMM{10-9};
3822 def VORRiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 0, 1,
3823 (outs QPR:$Vd), (ins nImmSplatI16:$SIMM, QPR:$src),
3825 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
3827 (v8i16 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
3828 let Inst{9} = SIMM{9};
3831 def VORRiv4i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 1, 0, 1,
3832 (outs QPR:$Vd), (ins nImmSplatI32:$SIMM, QPR:$src),
3834 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
3836 (v4i32 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
3837 let Inst{10-9} = SIMM{10-9};
3841 // VBIC : Vector Bitwise Bit Clear (AND NOT)
3842 def VBICd : N3VX<0, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
3843 (ins DPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VBINiD,
3844 "vbic", "$Vd, $Vn, $Vm", "",
3845 [(set DPR:$Vd, (v2i32 (and DPR:$Vn,
3846 (vnotd DPR:$Vm))))]>;
3847 def VBICq : N3VX<0, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
3848 (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINiQ,
3849 "vbic", "$Vd, $Vn, $Vm", "",
3850 [(set QPR:$Vd, (v4i32 (and QPR:$Vn,
3851 (vnotq QPR:$Vm))))]>;
3853 def VBICiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 1, 1,
3854 (outs DPR:$Vd), (ins nImmSplatI16:$SIMM, DPR:$src),
3856 "vbic", "i16", "$Vd, $SIMM", "$src = $Vd",
3858 (v4i16 (NEONvbicImm DPR:$src, timm:$SIMM)))]> {
3859 let Inst{9} = SIMM{9};
3862 def VBICiv2i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 0, 1, 1,
3863 (outs DPR:$Vd), (ins nImmSplatI32:$SIMM, DPR:$src),
3865 "vbic", "i32", "$Vd, $SIMM", "$src = $Vd",
3867 (v2i32 (NEONvbicImm DPR:$src, timm:$SIMM)))]> {
3868 let Inst{10-9} = SIMM{10-9};
3871 def VBICiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 1, 1,
3872 (outs QPR:$Vd), (ins nImmSplatI16:$SIMM, QPR:$src),
3874 "vbic", "i16", "$Vd, $SIMM", "$src = $Vd",
3876 (v8i16 (NEONvbicImm QPR:$src, timm:$SIMM)))]> {
3877 let Inst{9} = SIMM{9};
3880 def VBICiv4i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 1, 1, 1,
3881 (outs QPR:$Vd), (ins nImmSplatI32:$SIMM, QPR:$src),
3883 "vbic", "i32", "$Vd, $SIMM", "$src = $Vd",
3885 (v4i32 (NEONvbicImm QPR:$src, timm:$SIMM)))]> {
3886 let Inst{10-9} = SIMM{10-9};
3889 // VORN : Vector Bitwise OR NOT
3890 def VORNd : N3VX<0, 0, 0b11, 0b0001, 0, 1, (outs DPR:$Vd),
3891 (ins DPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VBINiD,
3892 "vorn", "$Vd, $Vn, $Vm", "",
3893 [(set DPR:$Vd, (v2i32 (or DPR:$Vn,
3894 (vnotd DPR:$Vm))))]>;
3895 def VORNq : N3VX<0, 0, 0b11, 0b0001, 1, 1, (outs QPR:$Vd),
3896 (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINiQ,
3897 "vorn", "$Vd, $Vn, $Vm", "",
3898 [(set QPR:$Vd, (v4i32 (or QPR:$Vn,
3899 (vnotq QPR:$Vm))))]>;
3901 // VMVN : Vector Bitwise NOT (Immediate)
3903 let isReMaterializable = 1 in {
3905 def VMVNv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 1, 1, (outs DPR:$Vd),
3906 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
3907 "vmvn", "i16", "$Vd, $SIMM", "",
3908 [(set DPR:$Vd, (v4i16 (NEONvmvnImm timm:$SIMM)))]> {
3909 let Inst{9} = SIMM{9};
3912 def VMVNv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 1, 1, (outs QPR:$Vd),
3913 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
3914 "vmvn", "i16", "$Vd, $SIMM", "",
3915 [(set QPR:$Vd, (v8i16 (NEONvmvnImm timm:$SIMM)))]> {
3916 let Inst{9} = SIMM{9};
3919 def VMVNv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 1, 1, (outs DPR:$Vd),
3920 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
3921 "vmvn", "i32", "$Vd, $SIMM", "",
3922 [(set DPR:$Vd, (v2i32 (NEONvmvnImm timm:$SIMM)))]> {
3923 let Inst{11-8} = SIMM{11-8};
3926 def VMVNv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 1, 1, (outs QPR:$Vd),
3927 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
3928 "vmvn", "i32", "$Vd, $SIMM", "",
3929 [(set QPR:$Vd, (v4i32 (NEONvmvnImm timm:$SIMM)))]> {
3930 let Inst{11-8} = SIMM{11-8};
3934 // VMVN : Vector Bitwise NOT
3935 def VMVNd : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 0, 0,
3936 (outs DPR:$Vd), (ins DPR:$Vm), IIC_VSUBiD,
3937 "vmvn", "$Vd, $Vm", "",
3938 [(set DPR:$Vd, (v2i32 (vnotd DPR:$Vm)))]>;
3939 def VMVNq : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 1, 0,
3940 (outs QPR:$Vd), (ins QPR:$Vm), IIC_VSUBiD,
3941 "vmvn", "$Vd, $Vm", "",
3942 [(set QPR:$Vd, (v4i32 (vnotq QPR:$Vm)))]>;
3943 def : Pat<(v2i32 (vnotd DPR:$src)), (VMVNd DPR:$src)>;
3944 def : Pat<(v4i32 (vnotq QPR:$src)), (VMVNq QPR:$src)>;
3946 // VBSL : Vector Bitwise Select
3947 def VBSLd : N3VX<1, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
3948 (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
3949 N3RegFrm, IIC_VCNTiD,
3950 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3952 (v2i32 (NEONvbsl DPR:$src1, DPR:$Vn, DPR:$Vm)))]>;
3954 def : Pat<(v2i32 (or (and DPR:$Vn, DPR:$Vd),
3955 (and DPR:$Vm, (vnotd DPR:$Vd)))),
3956 (VBSLd DPR:$Vd, DPR:$Vn, DPR:$Vm)>;
3958 def VBSLq : N3VX<1, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
3959 (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
3960 N3RegFrm, IIC_VCNTiQ,
3961 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3963 (v4i32 (NEONvbsl QPR:$src1, QPR:$Vn, QPR:$Vm)))]>;
3965 def : Pat<(v4i32 (or (and QPR:$Vn, QPR:$Vd),
3966 (and QPR:$Vm, (vnotq QPR:$Vd)))),
3967 (VBSLq QPR:$Vd, QPR:$Vn, QPR:$Vm)>;
3969 // VBIF : Vector Bitwise Insert if False
3970 // like VBSL but with: "vbif $dst, $src3, $src1", "$src2 = $dst",
3971 // FIXME: This instruction's encoding MAY NOT BE correct.
3972 def VBIFd : N3VX<1, 0, 0b11, 0b0001, 0, 1,
3973 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
3974 N3RegFrm, IIC_VBINiD,
3975 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3977 def VBIFq : N3VX<1, 0, 0b11, 0b0001, 1, 1,
3978 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
3979 N3RegFrm, IIC_VBINiQ,
3980 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3983 // VBIT : Vector Bitwise Insert if True
3984 // like VBSL but with: "vbit $dst, $src2, $src1", "$src3 = $dst",
3985 // FIXME: This instruction's encoding MAY NOT BE correct.
3986 def VBITd : N3VX<1, 0, 0b10, 0b0001, 0, 1,
3987 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
3988 N3RegFrm, IIC_VBINiD,
3989 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3991 def VBITq : N3VX<1, 0, 0b10, 0b0001, 1, 1,
3992 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
3993 N3RegFrm, IIC_VBINiQ,
3994 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3997 // VBIT/VBIF are not yet implemented. The TwoAddress pass will not go looking
3998 // for equivalent operations with different register constraints; it just
4001 // Vector Absolute Differences.
4003 // VABD : Vector Absolute Difference
4004 defm VABDs : N3VInt_QHS<0, 0, 0b0111, 0, N3RegFrm,
4005 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4006 "vabd", "s", int_arm_neon_vabds, 1>;
4007 defm VABDu : N3VInt_QHS<1, 0, 0b0111, 0, N3RegFrm,
4008 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4009 "vabd", "u", int_arm_neon_vabdu, 1>;
4010 def VABDfd : N3VDInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBIND,
4011 "vabd", "f32", v2f32, v2f32, int_arm_neon_vabds, 1>;
4012 def VABDfq : N3VQInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBINQ,
4013 "vabd", "f32", v4f32, v4f32, int_arm_neon_vabds, 1>;
4015 // VABDL : Vector Absolute Difference Long (Q = | D - D |)
4016 defm VABDLs : N3VLIntExt_QHS<0,1,0b0111,0, IIC_VSUBi4Q,
4017 "vabdl", "s", int_arm_neon_vabds, zext, 1>;
4018 defm VABDLu : N3VLIntExt_QHS<1,1,0b0111,0, IIC_VSUBi4Q,
4019 "vabdl", "u", int_arm_neon_vabdu, zext, 1>;
4021 // VABA : Vector Absolute Difference and Accumulate
4022 defm VABAs : N3VIntOp_QHS<0,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
4023 "vaba", "s", int_arm_neon_vabds, add>;
4024 defm VABAu : N3VIntOp_QHS<1,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
4025 "vaba", "u", int_arm_neon_vabdu, add>;
4027 // VABAL : Vector Absolute Difference and Accumulate Long (Q += | D - D |)
4028 defm VABALs : N3VLIntExtOp_QHS<0,1,0b0101,0, IIC_VABAD,
4029 "vabal", "s", int_arm_neon_vabds, zext, add>;
4030 defm VABALu : N3VLIntExtOp_QHS<1,1,0b0101,0, IIC_VABAD,
4031 "vabal", "u", int_arm_neon_vabdu, zext, add>;
4033 // Vector Maximum and Minimum.
4035 // VMAX : Vector Maximum
4036 defm VMAXs : N3VInt_QHS<0, 0, 0b0110, 0, N3RegFrm,
4037 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4038 "vmax", "s", int_arm_neon_vmaxs, 1>;
4039 defm VMAXu : N3VInt_QHS<1, 0, 0b0110, 0, N3RegFrm,
4040 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4041 "vmax", "u", int_arm_neon_vmaxu, 1>;
4042 def VMAXfd : N3VDInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBIND,
4044 v2f32, v2f32, int_arm_neon_vmaxs, 1>;
4045 def VMAXfq : N3VQInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBINQ,
4047 v4f32, v4f32, int_arm_neon_vmaxs, 1>;
4049 // VMIN : Vector Minimum
4050 defm VMINs : N3VInt_QHS<0, 0, 0b0110, 1, N3RegFrm,
4051 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4052 "vmin", "s", int_arm_neon_vmins, 1>;
4053 defm VMINu : N3VInt_QHS<1, 0, 0b0110, 1, N3RegFrm,
4054 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4055 "vmin", "u", int_arm_neon_vminu, 1>;
4056 def VMINfd : N3VDInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBIND,
4058 v2f32, v2f32, int_arm_neon_vmins, 1>;
4059 def VMINfq : N3VQInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBINQ,
4061 v4f32, v4f32, int_arm_neon_vmins, 1>;
4063 // Vector Pairwise Operations.
4065 // VPADD : Vector Pairwise Add
4066 def VPADDi8 : N3VDInt<0, 0, 0b00, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4068 v8i8, v8i8, int_arm_neon_vpadd, 0>;
4069 def VPADDi16 : N3VDInt<0, 0, 0b01, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4071 v4i16, v4i16, int_arm_neon_vpadd, 0>;
4072 def VPADDi32 : N3VDInt<0, 0, 0b10, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4074 v2i32, v2i32, int_arm_neon_vpadd, 0>;
4075 def VPADDf : N3VDInt<1, 0, 0b00, 0b1101, 0, N3RegFrm,
4076 IIC_VPBIND, "vpadd", "f32",
4077 v2f32, v2f32, int_arm_neon_vpadd, 0>;
4079 // VPADDL : Vector Pairwise Add Long
4080 defm VPADDLs : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00100, 0, "vpaddl", "s",
4081 int_arm_neon_vpaddls>;
4082 defm VPADDLu : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00101, 0, "vpaddl", "u",
4083 int_arm_neon_vpaddlu>;
4085 // VPADAL : Vector Pairwise Add and Accumulate Long
4086 defm VPADALs : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01100, 0, "vpadal", "s",
4087 int_arm_neon_vpadals>;
4088 defm VPADALu : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01101, 0, "vpadal", "u",
4089 int_arm_neon_vpadalu>;
4091 // VPMAX : Vector Pairwise Maximum
4092 def VPMAXs8 : N3VDInt<0, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4093 "s8", v8i8, v8i8, int_arm_neon_vpmaxs, 0>;
4094 def VPMAXs16 : N3VDInt<0, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4095 "s16", v4i16, v4i16, int_arm_neon_vpmaxs, 0>;
4096 def VPMAXs32 : N3VDInt<0, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4097 "s32", v2i32, v2i32, int_arm_neon_vpmaxs, 0>;
4098 def VPMAXu8 : N3VDInt<1, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4099 "u8", v8i8, v8i8, int_arm_neon_vpmaxu, 0>;
4100 def VPMAXu16 : N3VDInt<1, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4101 "u16", v4i16, v4i16, int_arm_neon_vpmaxu, 0>;
4102 def VPMAXu32 : N3VDInt<1, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4103 "u32", v2i32, v2i32, int_arm_neon_vpmaxu, 0>;
4104 def VPMAXf : N3VDInt<1, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmax",
4105 "f32", v2f32, v2f32, int_arm_neon_vpmaxs, 0>;
4107 // VPMIN : Vector Pairwise Minimum
4108 def VPMINs8 : N3VDInt<0, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4109 "s8", v8i8, v8i8, int_arm_neon_vpmins, 0>;
4110 def VPMINs16 : N3VDInt<0, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4111 "s16", v4i16, v4i16, int_arm_neon_vpmins, 0>;
4112 def VPMINs32 : N3VDInt<0, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4113 "s32", v2i32, v2i32, int_arm_neon_vpmins, 0>;
4114 def VPMINu8 : N3VDInt<1, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4115 "u8", v8i8, v8i8, int_arm_neon_vpminu, 0>;
4116 def VPMINu16 : N3VDInt<1, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4117 "u16", v4i16, v4i16, int_arm_neon_vpminu, 0>;
4118 def VPMINu32 : N3VDInt<1, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4119 "u32", v2i32, v2i32, int_arm_neon_vpminu, 0>;
4120 def VPMINf : N3VDInt<1, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmin",
4121 "f32", v2f32, v2f32, int_arm_neon_vpmins, 0>;
4123 // Vector Reciprocal and Reciprocal Square Root Estimate and Step.
4125 // VRECPE : Vector Reciprocal Estimate
4126 def VRECPEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
4127 IIC_VUNAD, "vrecpe", "u32",
4128 v2i32, v2i32, int_arm_neon_vrecpe>;
4129 def VRECPEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
4130 IIC_VUNAQ, "vrecpe", "u32",
4131 v4i32, v4i32, int_arm_neon_vrecpe>;
4132 def VRECPEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
4133 IIC_VUNAD, "vrecpe", "f32",
4134 v2f32, v2f32, int_arm_neon_vrecpe>;
4135 def VRECPEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
4136 IIC_VUNAQ, "vrecpe", "f32",
4137 v4f32, v4f32, int_arm_neon_vrecpe>;
4139 // VRECPS : Vector Reciprocal Step
4140 def VRECPSfd : N3VDInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
4141 IIC_VRECSD, "vrecps", "f32",
4142 v2f32, v2f32, int_arm_neon_vrecps, 1>;
4143 def VRECPSfq : N3VQInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
4144 IIC_VRECSQ, "vrecps", "f32",
4145 v4f32, v4f32, int_arm_neon_vrecps, 1>;
4147 // VRSQRTE : Vector Reciprocal Square Root Estimate
4148 def VRSQRTEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
4149 IIC_VUNAD, "vrsqrte", "u32",
4150 v2i32, v2i32, int_arm_neon_vrsqrte>;
4151 def VRSQRTEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
4152 IIC_VUNAQ, "vrsqrte", "u32",
4153 v4i32, v4i32, int_arm_neon_vrsqrte>;
4154 def VRSQRTEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
4155 IIC_VUNAD, "vrsqrte", "f32",
4156 v2f32, v2f32, int_arm_neon_vrsqrte>;
4157 def VRSQRTEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
4158 IIC_VUNAQ, "vrsqrte", "f32",
4159 v4f32, v4f32, int_arm_neon_vrsqrte>;
4161 // VRSQRTS : Vector Reciprocal Square Root Step
4162 def VRSQRTSfd : N3VDInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
4163 IIC_VRECSD, "vrsqrts", "f32",
4164 v2f32, v2f32, int_arm_neon_vrsqrts, 1>;
4165 def VRSQRTSfq : N3VQInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
4166 IIC_VRECSQ, "vrsqrts", "f32",
4167 v4f32, v4f32, int_arm_neon_vrsqrts, 1>;
4171 // VSHL : Vector Shift
4172 defm VSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 0, N3RegVShFrm,
4173 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
4174 "vshl", "s", int_arm_neon_vshifts>;
4175 defm VSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 0, N3RegVShFrm,
4176 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
4177 "vshl", "u", int_arm_neon_vshiftu>;
4179 // VSHL : Vector Shift Left (Immediate)
4180 defm VSHLi : N2VShL_QHSD<0, 1, 0b0101, 1, IIC_VSHLiD, "vshl", "i", NEONvshl>;
4182 // VSHR : Vector Shift Right (Immediate)
4183 defm VSHRs : N2VShR_QHSD<0, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "s",NEONvshrs>;
4184 defm VSHRu : N2VShR_QHSD<1, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "u",NEONvshru>;
4186 // VSHLL : Vector Shift Left Long
4187 defm VSHLLs : N2VLSh_QHS<0, 1, 0b1010, 0, 0, 1, "vshll", "s", NEONvshlls>;
4188 defm VSHLLu : N2VLSh_QHS<1, 1, 0b1010, 0, 0, 1, "vshll", "u", NEONvshllu>;
4190 // VSHLL : Vector Shift Left Long (with maximum shift count)
4191 class N2VLShMax<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
4192 bit op6, bit op4, string OpcodeStr, string Dt, ValueType ResTy,
4193 ValueType OpTy, SDNode OpNode>
4194 : N2VLSh<op24, op23, op11_8, op7, op6, op4, OpcodeStr, Dt,
4195 ResTy, OpTy, OpNode> {
4196 let Inst{21-16} = op21_16;
4197 let DecoderMethod = "DecodeVSHLMaxInstruction";
4199 def VSHLLi8 : N2VLShMax<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll", "i8",
4200 v8i16, v8i8, NEONvshlli>;
4201 def VSHLLi16 : N2VLShMax<1, 1, 0b110110, 0b0011, 0, 0, 0, "vshll", "i16",
4202 v4i32, v4i16, NEONvshlli>;
4203 def VSHLLi32 : N2VLShMax<1, 1, 0b111010, 0b0011, 0, 0, 0, "vshll", "i32",
4204 v2i64, v2i32, NEONvshlli>;
4206 // VSHRN : Vector Shift Right and Narrow
4207 defm VSHRN : N2VNSh_HSD<0,1,0b1000,0,0,1, IIC_VSHLiD, "vshrn", "i",
4210 // VRSHL : Vector Rounding Shift
4211 defm VRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 0, N3RegVShFrm,
4212 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4213 "vrshl", "s", int_arm_neon_vrshifts>;
4214 defm VRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 0, N3RegVShFrm,
4215 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4216 "vrshl", "u", int_arm_neon_vrshiftu>;
4217 // VRSHR : Vector Rounding Shift Right
4218 defm VRSHRs : N2VShR_QHSD<0,1,0b0010,1, IIC_VSHLi4D, "vrshr", "s",NEONvrshrs>;
4219 defm VRSHRu : N2VShR_QHSD<1,1,0b0010,1, IIC_VSHLi4D, "vrshr", "u",NEONvrshru>;
4221 // VRSHRN : Vector Rounding Shift Right and Narrow
4222 defm VRSHRN : N2VNSh_HSD<0, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vrshrn", "i",
4225 // VQSHL : Vector Saturating Shift
4226 defm VQSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 1, N3RegVShFrm,
4227 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4228 "vqshl", "s", int_arm_neon_vqshifts>;
4229 defm VQSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 1, N3RegVShFrm,
4230 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4231 "vqshl", "u", int_arm_neon_vqshiftu>;
4232 // VQSHL : Vector Saturating Shift Left (Immediate)
4233 defm VQSHLsi : N2VShL_QHSD<0,1,0b0111,1, IIC_VSHLi4D, "vqshl", "s",NEONvqshls>;
4234 defm VQSHLui : N2VShL_QHSD<1,1,0b0111,1, IIC_VSHLi4D, "vqshl", "u",NEONvqshlu>;
4236 // VQSHLU : Vector Saturating Shift Left (Immediate, Unsigned)
4237 defm VQSHLsu : N2VShL_QHSD<1,1,0b0110,1, IIC_VSHLi4D,"vqshlu","s",NEONvqshlsu>;
4239 // VQSHRN : Vector Saturating Shift Right and Narrow
4240 defm VQSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "s",
4242 defm VQSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "u",
4245 // VQSHRUN : Vector Saturating Shift Right and Narrow (Unsigned)
4246 defm VQSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 0, 1, IIC_VSHLi4D, "vqshrun", "s",
4249 // VQRSHL : Vector Saturating Rounding Shift
4250 defm VQRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 1, N3RegVShFrm,
4251 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4252 "vqrshl", "s", int_arm_neon_vqrshifts>;
4253 defm VQRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 1, N3RegVShFrm,
4254 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4255 "vqrshl", "u", int_arm_neon_vqrshiftu>;
4257 // VQRSHRN : Vector Saturating Rounding Shift Right and Narrow
4258 defm VQRSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "s",
4260 defm VQRSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "u",
4263 // VQRSHRUN : Vector Saturating Rounding Shift Right and Narrow (Unsigned)
4264 defm VQRSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vqrshrun", "s",
4267 // VSRA : Vector Shift Right and Accumulate
4268 defm VSRAs : N2VShAdd_QHSD<0, 1, 0b0001, 1, "vsra", "s", NEONvshrs>;
4269 defm VSRAu : N2VShAdd_QHSD<1, 1, 0b0001, 1, "vsra", "u", NEONvshru>;
4270 // VRSRA : Vector Rounding Shift Right and Accumulate
4271 defm VRSRAs : N2VShAdd_QHSD<0, 1, 0b0011, 1, "vrsra", "s", NEONvrshrs>;
4272 defm VRSRAu : N2VShAdd_QHSD<1, 1, 0b0011, 1, "vrsra", "u", NEONvrshru>;
4274 // VSLI : Vector Shift Left and Insert
4275 defm VSLI : N2VShInsL_QHSD<1, 1, 0b0101, 1, "vsli">;
4277 // VSRI : Vector Shift Right and Insert
4278 defm VSRI : N2VShInsR_QHSD<1, 1, 0b0100, 1, "vsri">;
4280 // Vector Absolute and Saturating Absolute.
4282 // VABS : Vector Absolute Value
4283 defm VABS : N2VInt_QHS<0b11, 0b11, 0b01, 0b00110, 0,
4284 IIC_VUNAiD, IIC_VUNAiQ, "vabs", "s",
4286 def VABSfd : N2VDInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
4287 IIC_VUNAD, "vabs", "f32",
4288 v2f32, v2f32, int_arm_neon_vabs>;
4289 def VABSfq : N2VQInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
4290 IIC_VUNAQ, "vabs", "f32",
4291 v4f32, v4f32, int_arm_neon_vabs>;
4293 // VQABS : Vector Saturating Absolute Value
4294 defm VQABS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01110, 0,
4295 IIC_VQUNAiD, IIC_VQUNAiQ, "vqabs", "s",
4296 int_arm_neon_vqabs>;
4300 def vnegd : PatFrag<(ops node:$in),
4301 (sub (bitconvert (v2i32 NEONimmAllZerosV)), node:$in)>;
4302 def vnegq : PatFrag<(ops node:$in),
4303 (sub (bitconvert (v4i32 NEONimmAllZerosV)), node:$in)>;
4305 class VNEGD<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
4306 : N2V<0b11, 0b11, size, 0b01, 0b00111, 0, 0, (outs DPR:$Vd), (ins DPR:$Vm),
4307 IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
4308 [(set DPR:$Vd, (Ty (vnegd DPR:$Vm)))]>;
4309 class VNEGQ<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
4310 : N2V<0b11, 0b11, size, 0b01, 0b00111, 1, 0, (outs QPR:$Vd), (ins QPR:$Vm),
4311 IIC_VSHLiQ, OpcodeStr, Dt, "$Vd, $Vm", "",
4312 [(set QPR:$Vd, (Ty (vnegq QPR:$Vm)))]>;
4314 // VNEG : Vector Negate (integer)
4315 def VNEGs8d : VNEGD<0b00, "vneg", "s8", v8i8>;
4316 def VNEGs16d : VNEGD<0b01, "vneg", "s16", v4i16>;
4317 def VNEGs32d : VNEGD<0b10, "vneg", "s32", v2i32>;
4318 def VNEGs8q : VNEGQ<0b00, "vneg", "s8", v16i8>;
4319 def VNEGs16q : VNEGQ<0b01, "vneg", "s16", v8i16>;
4320 def VNEGs32q : VNEGQ<0b10, "vneg", "s32", v4i32>;
4322 // VNEG : Vector Negate (floating-point)
4323 def VNEGfd : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
4324 (outs DPR:$Vd), (ins DPR:$Vm), IIC_VUNAD,
4325 "vneg", "f32", "$Vd, $Vm", "",
4326 [(set DPR:$Vd, (v2f32 (fneg DPR:$Vm)))]>;
4327 def VNEGf32q : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 1, 0,
4328 (outs QPR:$Vd), (ins QPR:$Vm), IIC_VUNAQ,
4329 "vneg", "f32", "$Vd, $Vm", "",
4330 [(set QPR:$Vd, (v4f32 (fneg QPR:$Vm)))]>;
4332 def : Pat<(v8i8 (vnegd DPR:$src)), (VNEGs8d DPR:$src)>;
4333 def : Pat<(v4i16 (vnegd DPR:$src)), (VNEGs16d DPR:$src)>;
4334 def : Pat<(v2i32 (vnegd DPR:$src)), (VNEGs32d DPR:$src)>;
4335 def : Pat<(v16i8 (vnegq QPR:$src)), (VNEGs8q QPR:$src)>;
4336 def : Pat<(v8i16 (vnegq QPR:$src)), (VNEGs16q QPR:$src)>;
4337 def : Pat<(v4i32 (vnegq QPR:$src)), (VNEGs32q QPR:$src)>;
4339 // VQNEG : Vector Saturating Negate
4340 defm VQNEG : N2VInt_QHS<0b11, 0b11, 0b00, 0b01111, 0,
4341 IIC_VQUNAiD, IIC_VQUNAiQ, "vqneg", "s",
4342 int_arm_neon_vqneg>;
4344 // Vector Bit Counting Operations.
4346 // VCLS : Vector Count Leading Sign Bits
4347 defm VCLS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01000, 0,
4348 IIC_VCNTiD, IIC_VCNTiQ, "vcls", "s",
4350 // VCLZ : Vector Count Leading Zeros
4351 defm VCLZ : N2VInt_QHS<0b11, 0b11, 0b00, 0b01001, 0,
4352 IIC_VCNTiD, IIC_VCNTiQ, "vclz", "i",
4354 // VCNT : Vector Count One Bits
4355 def VCNTd : N2VDInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
4356 IIC_VCNTiD, "vcnt", "8",
4357 v8i8, v8i8, int_arm_neon_vcnt>;
4358 def VCNTq : N2VQInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
4359 IIC_VCNTiQ, "vcnt", "8",
4360 v16i8, v16i8, int_arm_neon_vcnt>;
4363 def VSWPd : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 0, 0,
4364 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
4365 "vswp", "$Vd, $Vm", "", []>;
4366 def VSWPq : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 1, 0,
4367 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
4368 "vswp", "$Vd, $Vm", "", []>;
4370 // Vector Move Operations.
4372 // VMOV : Vector Move (Register)
4373 def : InstAlias<"vmov${p} $Vd, $Vm",
4374 (VORRd DPR:$Vd, DPR:$Vm, DPR:$Vm, pred:$p)>;
4375 def : InstAlias<"vmov${p} $Vd, $Vm",
4376 (VORRq QPR:$Vd, QPR:$Vm, QPR:$Vm, pred:$p)>;
4378 // VMOV : Vector Move (Immediate)
4380 let isReMaterializable = 1 in {
4381 def VMOVv8i8 : N1ModImm<1, 0b000, 0b1110, 0, 0, 0, 1, (outs DPR:$Vd),
4382 (ins nImmSplatI8:$SIMM), IIC_VMOVImm,
4383 "vmov", "i8", "$Vd, $SIMM", "",
4384 [(set DPR:$Vd, (v8i8 (NEONvmovImm timm:$SIMM)))]>;
4385 def VMOVv16i8 : N1ModImm<1, 0b000, 0b1110, 0, 1, 0, 1, (outs QPR:$Vd),
4386 (ins nImmSplatI8:$SIMM), IIC_VMOVImm,
4387 "vmov", "i8", "$Vd, $SIMM", "",
4388 [(set QPR:$Vd, (v16i8 (NEONvmovImm timm:$SIMM)))]>;
4390 def VMOVv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 0, 1, (outs DPR:$Vd),
4391 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4392 "vmov", "i16", "$Vd, $SIMM", "",
4393 [(set DPR:$Vd, (v4i16 (NEONvmovImm timm:$SIMM)))]> {
4394 let Inst{9} = SIMM{9};
4397 def VMOVv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 0, 1, (outs QPR:$Vd),
4398 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4399 "vmov", "i16", "$Vd, $SIMM", "",
4400 [(set QPR:$Vd, (v8i16 (NEONvmovImm timm:$SIMM)))]> {
4401 let Inst{9} = SIMM{9};
4404 def VMOVv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 0, 1, (outs DPR:$Vd),
4405 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4406 "vmov", "i32", "$Vd, $SIMM", "",
4407 [(set DPR:$Vd, (v2i32 (NEONvmovImm timm:$SIMM)))]> {
4408 let Inst{11-8} = SIMM{11-8};
4411 def VMOVv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 0, 1, (outs QPR:$Vd),
4412 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4413 "vmov", "i32", "$Vd, $SIMM", "",
4414 [(set QPR:$Vd, (v4i32 (NEONvmovImm timm:$SIMM)))]> {
4415 let Inst{11-8} = SIMM{11-8};
4418 def VMOVv1i64 : N1ModImm<1, 0b000, 0b1110, 0, 0, 1, 1, (outs DPR:$Vd),
4419 (ins nImmSplatI64:$SIMM), IIC_VMOVImm,
4420 "vmov", "i64", "$Vd, $SIMM", "",
4421 [(set DPR:$Vd, (v1i64 (NEONvmovImm timm:$SIMM)))]>;
4422 def VMOVv2i64 : N1ModImm<1, 0b000, 0b1110, 0, 1, 1, 1, (outs QPR:$Vd),
4423 (ins nImmSplatI64:$SIMM), IIC_VMOVImm,
4424 "vmov", "i64", "$Vd, $SIMM", "",
4425 [(set QPR:$Vd, (v2i64 (NEONvmovImm timm:$SIMM)))]>;
4426 } // isReMaterializable
4428 // VMOV : Vector Get Lane (move scalar to ARM core register)
4430 def VGETLNs8 : NVGetLane<{1,1,1,0,0,1,?,1}, 0b1011, {?,?},
4431 (outs GPR:$R), (ins DPR:$V, VectorIndex8:$lane),
4432 IIC_VMOVSI, "vmov", "s8", "$R, $V$lane",
4433 [(set GPR:$R, (NEONvgetlanes (v8i8 DPR:$V),
4435 let Inst{21} = lane{2};
4436 let Inst{6-5} = lane{1-0};
4438 def VGETLNs16 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, {?,1},
4439 (outs GPR:$R), (ins DPR:$V, VectorIndex16:$lane),
4440 IIC_VMOVSI, "vmov", "s16", "$R, $V$lane",
4441 [(set GPR:$R, (NEONvgetlanes (v4i16 DPR:$V),
4443 let Inst{21} = lane{1};
4444 let Inst{6} = lane{0};
4446 def VGETLNu8 : NVGetLane<{1,1,1,0,1,1,?,1}, 0b1011, {?,?},
4447 (outs GPR:$R), (ins DPR:$V, VectorIndex8:$lane),
4448 IIC_VMOVSI, "vmov", "u8", "$R, $V$lane",
4449 [(set GPR:$R, (NEONvgetlaneu (v8i8 DPR:$V),
4451 let Inst{21} = lane{2};
4452 let Inst{6-5} = lane{1-0};
4454 def VGETLNu16 : NVGetLane<{1,1,1,0,1,0,?,1}, 0b1011, {?,1},
4455 (outs GPR:$R), (ins DPR:$V, VectorIndex16:$lane),
4456 IIC_VMOVSI, "vmov", "u16", "$R, $V$lane",
4457 [(set GPR:$R, (NEONvgetlaneu (v4i16 DPR:$V),
4459 let Inst{21} = lane{1};
4460 let Inst{6} = lane{0};
4462 def VGETLNi32 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, 0b00,
4463 (outs GPR:$R), (ins DPR:$V, VectorIndex32:$lane),
4464 IIC_VMOVSI, "vmov", "32", "$R, $V$lane",
4465 [(set GPR:$R, (extractelt (v2i32 DPR:$V),
4467 let Inst{21} = lane{0};
4469 // def VGETLNf32: see FMRDH and FMRDL in ARMInstrVFP.td
4470 def : Pat<(NEONvgetlanes (v16i8 QPR:$src), imm:$lane),
4471 (VGETLNs8 (v8i8 (EXTRACT_SUBREG QPR:$src,
4472 (DSubReg_i8_reg imm:$lane))),
4473 (SubReg_i8_lane imm:$lane))>;
4474 def : Pat<(NEONvgetlanes (v8i16 QPR:$src), imm:$lane),
4475 (VGETLNs16 (v4i16 (EXTRACT_SUBREG QPR:$src,
4476 (DSubReg_i16_reg imm:$lane))),
4477 (SubReg_i16_lane imm:$lane))>;
4478 def : Pat<(NEONvgetlaneu (v16i8 QPR:$src), imm:$lane),
4479 (VGETLNu8 (v8i8 (EXTRACT_SUBREG QPR:$src,
4480 (DSubReg_i8_reg imm:$lane))),
4481 (SubReg_i8_lane imm:$lane))>;
4482 def : Pat<(NEONvgetlaneu (v8i16 QPR:$src), imm:$lane),
4483 (VGETLNu16 (v4i16 (EXTRACT_SUBREG QPR:$src,
4484 (DSubReg_i16_reg imm:$lane))),
4485 (SubReg_i16_lane imm:$lane))>;
4486 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
4487 (VGETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src,
4488 (DSubReg_i32_reg imm:$lane))),
4489 (SubReg_i32_lane imm:$lane))>;
4490 def : Pat<(extractelt (v2f32 DPR:$src1), imm:$src2),
4491 (EXTRACT_SUBREG (v2f32 (COPY_TO_REGCLASS (v2f32 DPR:$src1),DPR_VFP2)),
4492 (SSubReg_f32_reg imm:$src2))>;
4493 def : Pat<(extractelt (v4f32 QPR:$src1), imm:$src2),
4494 (EXTRACT_SUBREG (v4f32 (COPY_TO_REGCLASS (v4f32 QPR:$src1),QPR_VFP2)),
4495 (SSubReg_f32_reg imm:$src2))>;
4496 //def : Pat<(extractelt (v2i64 QPR:$src1), imm:$src2),
4497 // (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
4498 def : Pat<(extractelt (v2f64 QPR:$src1), imm:$src2),
4499 (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
4502 // VMOV : Vector Set Lane (move ARM core register to scalar)
4504 let Constraints = "$src1 = $V" in {
4505 def VSETLNi8 : NVSetLane<{1,1,1,0,0,1,?,0}, 0b1011, {?,?}, (outs DPR:$V),
4506 (ins DPR:$src1, GPR:$R, VectorIndex8:$lane),
4507 IIC_VMOVISL, "vmov", "8", "$V$lane, $R",
4508 [(set DPR:$V, (vector_insert (v8i8 DPR:$src1),
4509 GPR:$R, imm:$lane))]> {
4510 let Inst{21} = lane{2};
4511 let Inst{6-5} = lane{1-0};
4513 def VSETLNi16 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, {?,1}, (outs DPR:$V),
4514 (ins DPR:$src1, GPR:$R, VectorIndex16:$lane),
4515 IIC_VMOVISL, "vmov", "16", "$V$lane, $R",
4516 [(set DPR:$V, (vector_insert (v4i16 DPR:$src1),
4517 GPR:$R, imm:$lane))]> {
4518 let Inst{21} = lane{1};
4519 let Inst{6} = lane{0};
4521 def VSETLNi32 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, 0b00, (outs DPR:$V),
4522 (ins DPR:$src1, GPR:$R, VectorIndex32:$lane),
4523 IIC_VMOVISL, "vmov", "32", "$V$lane, $R",
4524 [(set DPR:$V, (insertelt (v2i32 DPR:$src1),
4525 GPR:$R, imm:$lane))]> {
4526 let Inst{21} = lane{0};
4529 def : Pat<(vector_insert (v16i8 QPR:$src1), GPR:$src2, imm:$lane),
4530 (v16i8 (INSERT_SUBREG QPR:$src1,
4531 (v8i8 (VSETLNi8 (v8i8 (EXTRACT_SUBREG QPR:$src1,
4532 (DSubReg_i8_reg imm:$lane))),
4533 GPR:$src2, (SubReg_i8_lane imm:$lane))),
4534 (DSubReg_i8_reg imm:$lane)))>;
4535 def : Pat<(vector_insert (v8i16 QPR:$src1), GPR:$src2, imm:$lane),
4536 (v8i16 (INSERT_SUBREG QPR:$src1,
4537 (v4i16 (VSETLNi16 (v4i16 (EXTRACT_SUBREG QPR:$src1,
4538 (DSubReg_i16_reg imm:$lane))),
4539 GPR:$src2, (SubReg_i16_lane imm:$lane))),
4540 (DSubReg_i16_reg imm:$lane)))>;
4541 def : Pat<(insertelt (v4i32 QPR:$src1), GPR:$src2, imm:$lane),
4542 (v4i32 (INSERT_SUBREG QPR:$src1,
4543 (v2i32 (VSETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src1,
4544 (DSubReg_i32_reg imm:$lane))),
4545 GPR:$src2, (SubReg_i32_lane imm:$lane))),
4546 (DSubReg_i32_reg imm:$lane)))>;
4548 def : Pat<(v2f32 (insertelt DPR:$src1, SPR:$src2, imm:$src3)),
4549 (INSERT_SUBREG (v2f32 (COPY_TO_REGCLASS DPR:$src1, DPR_VFP2)),
4550 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
4551 def : Pat<(v4f32 (insertelt QPR:$src1, SPR:$src2, imm:$src3)),
4552 (INSERT_SUBREG (v4f32 (COPY_TO_REGCLASS QPR:$src1, QPR_VFP2)),
4553 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
4555 //def : Pat<(v2i64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
4556 // (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
4557 def : Pat<(v2f64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
4558 (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
4560 def : Pat<(v2f32 (scalar_to_vector SPR:$src)),
4561 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
4562 def : Pat<(v2f64 (scalar_to_vector (f64 DPR:$src))),
4563 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
4564 def : Pat<(v4f32 (scalar_to_vector SPR:$src)),
4565 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
4567 def : Pat<(v8i8 (scalar_to_vector GPR:$src)),
4568 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4569 def : Pat<(v4i16 (scalar_to_vector GPR:$src)),
4570 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4571 def : Pat<(v2i32 (scalar_to_vector GPR:$src)),
4572 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4574 def : Pat<(v16i8 (scalar_to_vector GPR:$src)),
4575 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4576 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4578 def : Pat<(v8i16 (scalar_to_vector GPR:$src)),
4579 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
4580 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4582 def : Pat<(v4i32 (scalar_to_vector GPR:$src)),
4583 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
4584 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4587 // VDUP : Vector Duplicate (from ARM core register to all elements)
4589 class VDUPD<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
4590 : NVDup<opcod1, 0b1011, opcod3, (outs DPR:$V), (ins GPR:$R),
4591 IIC_VMOVIS, "vdup", Dt, "$V, $R",
4592 [(set DPR:$V, (Ty (NEONvdup (i32 GPR:$R))))]>;
4593 class VDUPQ<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
4594 : NVDup<opcod1, 0b1011, opcod3, (outs QPR:$V), (ins GPR:$R),
4595 IIC_VMOVIS, "vdup", Dt, "$V, $R",
4596 [(set QPR:$V, (Ty (NEONvdup (i32 GPR:$R))))]>;
4598 def VDUP8d : VDUPD<0b11101100, 0b00, "8", v8i8>;
4599 def VDUP16d : VDUPD<0b11101000, 0b01, "16", v4i16>;
4600 def VDUP32d : VDUPD<0b11101000, 0b00, "32", v2i32>;
4601 def VDUP8q : VDUPQ<0b11101110, 0b00, "8", v16i8>;
4602 def VDUP16q : VDUPQ<0b11101010, 0b01, "16", v8i16>;
4603 def VDUP32q : VDUPQ<0b11101010, 0b00, "32", v4i32>;
4605 def : Pat<(v2f32 (NEONvdup (f32 (bitconvert GPR:$R)))), (VDUP32d GPR:$R)>;
4606 def : Pat<(v4f32 (NEONvdup (f32 (bitconvert GPR:$R)))), (VDUP32q GPR:$R)>;
4608 // VDUP : Vector Duplicate Lane (from scalar to all elements)
4610 class VDUPLND<bits<4> op19_16, string OpcodeStr, string Dt,
4611 ValueType Ty, Operand IdxTy>
4612 : NVDupLane<op19_16, 0, (outs DPR:$Vd), (ins DPR:$Vm, IdxTy:$lane),
4613 IIC_VMOVD, OpcodeStr, Dt, "$Vd, $Vm$lane",
4614 [(set DPR:$Vd, (Ty (NEONvduplane (Ty DPR:$Vm), imm:$lane)))]>;
4616 class VDUPLNQ<bits<4> op19_16, string OpcodeStr, string Dt,
4617 ValueType ResTy, ValueType OpTy, Operand IdxTy>
4618 : NVDupLane<op19_16, 1, (outs QPR:$Vd), (ins DPR:$Vm, IdxTy:$lane),
4619 IIC_VMOVQ, OpcodeStr, Dt, "$Vd, $Vm$lane",
4620 [(set QPR:$Vd, (ResTy (NEONvduplane (OpTy DPR:$Vm),
4621 VectorIndex32:$lane)))]>;
4623 // Inst{19-16} is partially specified depending on the element size.
4625 def VDUPLN8d : VDUPLND<{?,?,?,1}, "vdup", "8", v8i8, VectorIndex8> {
4627 let Inst{19-17} = lane{2-0};
4629 def VDUPLN16d : VDUPLND<{?,?,1,0}, "vdup", "16", v4i16, VectorIndex16> {
4631 let Inst{19-18} = lane{1-0};
4633 def VDUPLN32d : VDUPLND<{?,1,0,0}, "vdup", "32", v2i32, VectorIndex32> {
4635 let Inst{19} = lane{0};
4637 def VDUPLN8q : VDUPLNQ<{?,?,?,1}, "vdup", "8", v16i8, v8i8, VectorIndex8> {
4639 let Inst{19-17} = lane{2-0};
4641 def VDUPLN16q : VDUPLNQ<{?,?,1,0}, "vdup", "16", v8i16, v4i16, VectorIndex16> {
4643 let Inst{19-18} = lane{1-0};
4645 def VDUPLN32q : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4i32, v2i32, VectorIndex32> {
4647 let Inst{19} = lane{0};
4650 def : Pat<(v2f32 (NEONvduplane (v2f32 DPR:$Vm), imm:$lane)),
4651 (VDUPLN32d DPR:$Vm, imm:$lane)>;
4653 def : Pat<(v4f32 (NEONvduplane (v2f32 DPR:$Vm), imm:$lane)),
4654 (VDUPLN32q DPR:$Vm, imm:$lane)>;
4656 def : Pat<(v16i8 (NEONvduplane (v16i8 QPR:$src), imm:$lane)),
4657 (v16i8 (VDUPLN8q (v8i8 (EXTRACT_SUBREG QPR:$src,
4658 (DSubReg_i8_reg imm:$lane))),
4659 (SubReg_i8_lane imm:$lane)))>;
4660 def : Pat<(v8i16 (NEONvduplane (v8i16 QPR:$src), imm:$lane)),
4661 (v8i16 (VDUPLN16q (v4i16 (EXTRACT_SUBREG QPR:$src,
4662 (DSubReg_i16_reg imm:$lane))),
4663 (SubReg_i16_lane imm:$lane)))>;
4664 def : Pat<(v4i32 (NEONvduplane (v4i32 QPR:$src), imm:$lane)),
4665 (v4i32 (VDUPLN32q (v2i32 (EXTRACT_SUBREG QPR:$src,
4666 (DSubReg_i32_reg imm:$lane))),
4667 (SubReg_i32_lane imm:$lane)))>;
4668 def : Pat<(v4f32 (NEONvduplane (v4f32 QPR:$src), imm:$lane)),
4669 (v4f32 (VDUPLN32q (v2f32 (EXTRACT_SUBREG QPR:$src,
4670 (DSubReg_i32_reg imm:$lane))),
4671 (SubReg_i32_lane imm:$lane)))>;
4673 def VDUPfdf : PseudoNeonI<(outs DPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
4674 [(set DPR:$dst, (v2f32 (NEONvdup (f32 SPR:$src))))]>;
4675 def VDUPfqf : PseudoNeonI<(outs QPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
4676 [(set QPR:$dst, (v4f32 (NEONvdup (f32 SPR:$src))))]>;
4678 // VMOVN : Vector Narrowing Move
4679 defm VMOVN : N2VN_HSD<0b11,0b11,0b10,0b00100,0,0, IIC_VMOVN,
4680 "vmovn", "i", trunc>;
4681 // VQMOVN : Vector Saturating Narrowing Move
4682 defm VQMOVNs : N2VNInt_HSD<0b11,0b11,0b10,0b00101,0,0, IIC_VQUNAiD,
4683 "vqmovn", "s", int_arm_neon_vqmovns>;
4684 defm VQMOVNu : N2VNInt_HSD<0b11,0b11,0b10,0b00101,1,0, IIC_VQUNAiD,
4685 "vqmovn", "u", int_arm_neon_vqmovnu>;
4686 defm VQMOVNsu : N2VNInt_HSD<0b11,0b11,0b10,0b00100,1,0, IIC_VQUNAiD,
4687 "vqmovun", "s", int_arm_neon_vqmovnsu>;
4688 // VMOVL : Vector Lengthening Move
4689 defm VMOVLs : N2VL_QHS<0b01,0b10100,0,1, "vmovl", "s", sext>;
4690 defm VMOVLu : N2VL_QHS<0b11,0b10100,0,1, "vmovl", "u", zext>;
4692 // Vector Conversions.
4694 // VCVT : Vector Convert Between Floating-Point and Integers
4695 def VCVTf2sd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
4696 v2i32, v2f32, fp_to_sint>;
4697 def VCVTf2ud : N2VD<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
4698 v2i32, v2f32, fp_to_uint>;
4699 def VCVTs2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
4700 v2f32, v2i32, sint_to_fp>;
4701 def VCVTu2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
4702 v2f32, v2i32, uint_to_fp>;
4704 def VCVTf2sq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
4705 v4i32, v4f32, fp_to_sint>;
4706 def VCVTf2uq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
4707 v4i32, v4f32, fp_to_uint>;
4708 def VCVTs2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
4709 v4f32, v4i32, sint_to_fp>;
4710 def VCVTu2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
4711 v4f32, v4i32, uint_to_fp>;
4713 // VCVT : Vector Convert Between Floating-Point and Fixed-Point.
4714 def VCVTf2xsd : N2VCvtD<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
4715 v2i32, v2f32, int_arm_neon_vcvtfp2fxs>;
4716 def VCVTf2xud : N2VCvtD<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
4717 v2i32, v2f32, int_arm_neon_vcvtfp2fxu>;
4718 def VCVTxs2fd : N2VCvtD<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
4719 v2f32, v2i32, int_arm_neon_vcvtfxs2fp>;
4720 def VCVTxu2fd : N2VCvtD<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
4721 v2f32, v2i32, int_arm_neon_vcvtfxu2fp>;
4723 def VCVTf2xsq : N2VCvtQ<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
4724 v4i32, v4f32, int_arm_neon_vcvtfp2fxs>;
4725 def VCVTf2xuq : N2VCvtQ<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
4726 v4i32, v4f32, int_arm_neon_vcvtfp2fxu>;
4727 def VCVTxs2fq : N2VCvtQ<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
4728 v4f32, v4i32, int_arm_neon_vcvtfxs2fp>;
4729 def VCVTxu2fq : N2VCvtQ<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
4730 v4f32, v4i32, int_arm_neon_vcvtfxu2fp>;
4732 // VCVT : Vector Convert Between Half-Precision and Single-Precision.
4733 def VCVTf2h : N2VNInt<0b11, 0b11, 0b01, 0b10, 0b01100, 0, 0,
4734 IIC_VUNAQ, "vcvt", "f16.f32",
4735 v4i16, v4f32, int_arm_neon_vcvtfp2hf>,
4736 Requires<[HasNEON, HasFP16]>;
4737 def VCVTh2f : N2VLInt<0b11, 0b11, 0b01, 0b10, 0b01110, 0, 0,
4738 IIC_VUNAQ, "vcvt", "f32.f16",
4739 v4f32, v4i16, int_arm_neon_vcvthf2fp>,
4740 Requires<[HasNEON, HasFP16]>;
4744 // VREV64 : Vector Reverse elements within 64-bit doublewords
4746 class VREV64D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4747 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 0, 0, (outs DPR:$Vd),
4748 (ins DPR:$Vm), IIC_VMOVD,
4749 OpcodeStr, Dt, "$Vd, $Vm", "",
4750 [(set DPR:$Vd, (Ty (NEONvrev64 (Ty DPR:$Vm))))]>;
4751 class VREV64Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4752 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 1, 0, (outs QPR:$Vd),
4753 (ins QPR:$Vm), IIC_VMOVQ,
4754 OpcodeStr, Dt, "$Vd, $Vm", "",
4755 [(set QPR:$Vd, (Ty (NEONvrev64 (Ty QPR:$Vm))))]>;
4757 def VREV64d8 : VREV64D<0b00, "vrev64", "8", v8i8>;
4758 def VREV64d16 : VREV64D<0b01, "vrev64", "16", v4i16>;
4759 def VREV64d32 : VREV64D<0b10, "vrev64", "32", v2i32>;
4760 def : Pat<(v2f32 (NEONvrev64 (v2f32 DPR:$Vm))), (VREV64d32 DPR:$Vm)>;
4762 def VREV64q8 : VREV64Q<0b00, "vrev64", "8", v16i8>;
4763 def VREV64q16 : VREV64Q<0b01, "vrev64", "16", v8i16>;
4764 def VREV64q32 : VREV64Q<0b10, "vrev64", "32", v4i32>;
4765 def : Pat<(v4f32 (NEONvrev64 (v4f32 QPR:$Vm))), (VREV64q32 QPR:$Vm)>;
4767 // VREV32 : Vector Reverse elements within 32-bit words
4769 class VREV32D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4770 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 0, 0, (outs DPR:$Vd),
4771 (ins DPR:$Vm), IIC_VMOVD,
4772 OpcodeStr, Dt, "$Vd, $Vm", "",
4773 [(set DPR:$Vd, (Ty (NEONvrev32 (Ty DPR:$Vm))))]>;
4774 class VREV32Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4775 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 1, 0, (outs QPR:$Vd),
4776 (ins QPR:$Vm), IIC_VMOVQ,
4777 OpcodeStr, Dt, "$Vd, $Vm", "",
4778 [(set QPR:$Vd, (Ty (NEONvrev32 (Ty QPR:$Vm))))]>;
4780 def VREV32d8 : VREV32D<0b00, "vrev32", "8", v8i8>;
4781 def VREV32d16 : VREV32D<0b01, "vrev32", "16", v4i16>;
4783 def VREV32q8 : VREV32Q<0b00, "vrev32", "8", v16i8>;
4784 def VREV32q16 : VREV32Q<0b01, "vrev32", "16", v8i16>;
4786 // VREV16 : Vector Reverse elements within 16-bit halfwords
4788 class VREV16D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4789 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 0, 0, (outs DPR:$Vd),
4790 (ins DPR:$Vm), IIC_VMOVD,
4791 OpcodeStr, Dt, "$Vd, $Vm", "",
4792 [(set DPR:$Vd, (Ty (NEONvrev16 (Ty DPR:$Vm))))]>;
4793 class VREV16Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4794 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 1, 0, (outs QPR:$Vd),
4795 (ins QPR:$Vm), IIC_VMOVQ,
4796 OpcodeStr, Dt, "$Vd, $Vm", "",
4797 [(set QPR:$Vd, (Ty (NEONvrev16 (Ty QPR:$Vm))))]>;
4799 def VREV16d8 : VREV16D<0b00, "vrev16", "8", v8i8>;
4800 def VREV16q8 : VREV16Q<0b00, "vrev16", "8", v16i8>;
4802 // Other Vector Shuffles.
4804 // Aligned extractions: really just dropping registers
4806 class AlignedVEXTq<ValueType DestTy, ValueType SrcTy, SDNodeXForm LaneCVT>
4807 : Pat<(DestTy (vector_extract_subvec (SrcTy QPR:$src), (i32 imm:$start))),
4808 (EXTRACT_SUBREG (SrcTy QPR:$src), (LaneCVT imm:$start))>;
4810 def : AlignedVEXTq<v8i8, v16i8, DSubReg_i8_reg>;
4812 def : AlignedVEXTq<v4i16, v8i16, DSubReg_i16_reg>;
4814 def : AlignedVEXTq<v2i32, v4i32, DSubReg_i32_reg>;
4816 def : AlignedVEXTq<v1i64, v2i64, DSubReg_f64_reg>;
4818 def : AlignedVEXTq<v2f32, v4f32, DSubReg_i32_reg>;
4821 // VEXT : Vector Extract
4823 class VEXTd<string OpcodeStr, string Dt, ValueType Ty>
4824 : N3V<0,1,0b11,{?,?,?,?},0,0, (outs DPR:$Vd),
4825 (ins DPR:$Vn, DPR:$Vm, i32imm:$index), NVExtFrm,
4826 IIC_VEXTD, OpcodeStr, Dt, "$Vd, $Vn, $Vm, $index", "",
4827 [(set DPR:$Vd, (Ty (NEONvext (Ty DPR:$Vn),
4828 (Ty DPR:$Vm), imm:$index)))]> {
4830 let Inst{11-8} = index{3-0};
4833 class VEXTq<string OpcodeStr, string Dt, ValueType Ty>
4834 : N3V<0,1,0b11,{?,?,?,?},1,0, (outs QPR:$Vd),
4835 (ins QPR:$Vn, QPR:$Vm, i32imm:$index), NVExtFrm,
4836 IIC_VEXTQ, OpcodeStr, Dt, "$Vd, $Vn, $Vm, $index", "",
4837 [(set QPR:$Vd, (Ty (NEONvext (Ty QPR:$Vn),
4838 (Ty QPR:$Vm), imm:$index)))]> {
4840 let Inst{11-8} = index{3-0};
4843 def VEXTd8 : VEXTd<"vext", "8", v8i8> {
4844 let Inst{11-8} = index{3-0};
4846 def VEXTd16 : VEXTd<"vext", "16", v4i16> {
4847 let Inst{11-9} = index{2-0};
4850 def VEXTd32 : VEXTd<"vext", "32", v2i32> {
4851 let Inst{11-10} = index{1-0};
4852 let Inst{9-8} = 0b00;
4854 def : Pat<(v2f32 (NEONvext (v2f32 DPR:$Vn),
4857 (VEXTd32 DPR:$Vn, DPR:$Vm, imm:$index)>;
4859 def VEXTq8 : VEXTq<"vext", "8", v16i8> {
4860 let Inst{11-8} = index{3-0};
4862 def VEXTq16 : VEXTq<"vext", "16", v8i16> {
4863 let Inst{11-9} = index{2-0};
4866 def VEXTq32 : VEXTq<"vext", "32", v4i32> {
4867 let Inst{11-10} = index{1-0};
4868 let Inst{9-8} = 0b00;
4870 def : Pat<(v4f32 (NEONvext (v4f32 QPR:$Vn),
4873 (VEXTq32 QPR:$Vn, QPR:$Vm, imm:$index)>;
4875 // VTRN : Vector Transpose
4877 def VTRNd8 : N2VDShuffle<0b00, 0b00001, "vtrn", "8">;
4878 def VTRNd16 : N2VDShuffle<0b01, 0b00001, "vtrn", "16">;
4879 def VTRNd32 : N2VDShuffle<0b10, 0b00001, "vtrn", "32">;
4881 def VTRNq8 : N2VQShuffle<0b00, 0b00001, IIC_VPERMQ, "vtrn", "8">;
4882 def VTRNq16 : N2VQShuffle<0b01, 0b00001, IIC_VPERMQ, "vtrn", "16">;
4883 def VTRNq32 : N2VQShuffle<0b10, 0b00001, IIC_VPERMQ, "vtrn", "32">;
4885 // VUZP : Vector Unzip (Deinterleave)
4887 def VUZPd8 : N2VDShuffle<0b00, 0b00010, "vuzp", "8">;
4888 def VUZPd16 : N2VDShuffle<0b01, 0b00010, "vuzp", "16">;
4889 def VUZPd32 : N2VDShuffle<0b10, 0b00010, "vuzp", "32">;
4891 def VUZPq8 : N2VQShuffle<0b00, 0b00010, IIC_VPERMQ3, "vuzp", "8">;
4892 def VUZPq16 : N2VQShuffle<0b01, 0b00010, IIC_VPERMQ3, "vuzp", "16">;
4893 def VUZPq32 : N2VQShuffle<0b10, 0b00010, IIC_VPERMQ3, "vuzp", "32">;
4895 // VZIP : Vector Zip (Interleave)
4897 def VZIPd8 : N2VDShuffle<0b00, 0b00011, "vzip", "8">;
4898 def VZIPd16 : N2VDShuffle<0b01, 0b00011, "vzip", "16">;
4899 def VZIPd32 : N2VDShuffle<0b10, 0b00011, "vzip", "32">;
4901 def VZIPq8 : N2VQShuffle<0b00, 0b00011, IIC_VPERMQ3, "vzip", "8">;
4902 def VZIPq16 : N2VQShuffle<0b01, 0b00011, IIC_VPERMQ3, "vzip", "16">;
4903 def VZIPq32 : N2VQShuffle<0b10, 0b00011, IIC_VPERMQ3, "vzip", "32">;
4905 // Vector Table Lookup and Table Extension.
4907 // VTBL : Vector Table Lookup
4908 let DecoderMethod = "DecodeTBLInstruction" in {
4910 : N3V<1,1,0b11,0b1000,0,0, (outs DPR:$Vd),
4911 (ins VecListOneD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTB1,
4912 "vtbl", "8", "$Vd, $Vn, $Vm", "",
4913 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbl1 VecListOneD:$Vn, DPR:$Vm)))]>;
4914 let hasExtraSrcRegAllocReq = 1 in {
4916 : N3V<1,1,0b11,0b1001,0,0, (outs DPR:$Vd),
4917 (ins DPR:$Vn, DPR:$tbl2, DPR:$Vm), NVTBLFrm, IIC_VTB2,
4918 "vtbl", "8", "$Vd, \\{$Vn, $tbl2\\}, $Vm", "", []>;
4920 : N3V<1,1,0b11,0b1010,0,0, (outs DPR:$Vd),
4921 (ins DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$Vm), NVTBLFrm, IIC_VTB3,
4922 "vtbl", "8", "$Vd, \\{$Vn, $tbl2, $tbl3\\}, $Vm", "", []>;
4924 : N3V<1,1,0b11,0b1011,0,0, (outs DPR:$Vd),
4925 (ins DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$Vm),
4927 "vtbl", "8", "$Vd, \\{$Vn, $tbl2, $tbl3, $tbl4\\}, $Vm", "", []>;
4928 } // hasExtraSrcRegAllocReq = 1
4931 : PseudoNeonI<(outs DPR:$dst), (ins QPR:$tbl, DPR:$src), IIC_VTB2, "", []>;
4933 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB3, "", []>;
4935 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB4, "", []>;
4937 // VTBX : Vector Table Extension
4939 : N3V<1,1,0b11,0b1000,1,0, (outs DPR:$Vd),
4940 (ins DPR:$orig, VecListOneD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTBX1,
4941 "vtbx", "8", "$Vd, $Vn, $Vm", "$orig = $Vd",
4942 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbx1
4943 DPR:$orig, VecListOneD:$Vn, DPR:$Vm)))]>;
4944 let hasExtraSrcRegAllocReq = 1 in {
4946 : N3V<1,1,0b11,0b1001,1,0, (outs DPR:$Vd),
4947 (ins DPR:$orig, DPR:$Vn, DPR:$tbl2, DPR:$Vm), NVTBLFrm, IIC_VTBX2,
4948 "vtbx", "8", "$Vd, \\{$Vn, $tbl2\\}, $Vm", "$orig = $Vd", []>;
4950 : N3V<1,1,0b11,0b1010,1,0, (outs DPR:$Vd),
4951 (ins DPR:$orig, DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$Vm),
4952 NVTBLFrm, IIC_VTBX3,
4953 "vtbx", "8", "$Vd, \\{$Vn, $tbl2, $tbl3\\}, $Vm",
4956 : N3V<1,1,0b11,0b1011,1,0, (outs DPR:$Vd), (ins DPR:$orig, DPR:$Vn,
4957 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$Vm), NVTBLFrm, IIC_VTBX4,
4958 "vtbx", "8", "$Vd, \\{$Vn, $tbl2, $tbl3, $tbl4\\}, $Vm",
4960 } // hasExtraSrcRegAllocReq = 1
4963 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QPR:$tbl, DPR:$src),
4964 IIC_VTBX2, "$orig = $dst", []>;
4966 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
4967 IIC_VTBX3, "$orig = $dst", []>;
4969 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
4970 IIC_VTBX4, "$orig = $dst", []>;
4971 } // DecoderMethod = "DecodeTBLInstruction"
4973 //===----------------------------------------------------------------------===//
4974 // NEON instructions for single-precision FP math
4975 //===----------------------------------------------------------------------===//
4977 class N2VSPat<SDNode OpNode, NeonI Inst>
4978 : NEONFPPat<(f32 (OpNode SPR:$a)),
4980 (v2f32 (COPY_TO_REGCLASS (Inst
4982 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
4983 SPR:$a, ssub_0)), DPR_VFP2)), ssub_0)>;
4985 class N3VSPat<SDNode OpNode, NeonI Inst>
4986 : NEONFPPat<(f32 (OpNode SPR:$a, SPR:$b)),
4988 (v2f32 (COPY_TO_REGCLASS (Inst
4990 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
4993 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
4994 SPR:$b, ssub_0)), DPR_VFP2)), ssub_0)>;
4996 class N3VSMulOpPat<SDNode MulNode, SDNode OpNode, NeonI Inst>
4997 : NEONFPPat<(f32 (OpNode SPR:$acc, (f32 (MulNode SPR:$a, SPR:$b)))),
4999 (v2f32 (COPY_TO_REGCLASS (Inst
5001 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5004 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5007 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5008 SPR:$b, ssub_0)), DPR_VFP2)), ssub_0)>;
5010 def : N3VSPat<fadd, VADDfd>;
5011 def : N3VSPat<fsub, VSUBfd>;
5012 def : N3VSPat<fmul, VMULfd>;
5013 def : N3VSMulOpPat<fmul, fadd, VMLAfd>,
5014 Requires<[HasNEON, UseNEONForFP, UseFPVMLx]>;
5015 def : N3VSMulOpPat<fmul, fsub, VMLSfd>,
5016 Requires<[HasNEON, UseNEONForFP, UseFPVMLx]>;
5017 def : N2VSPat<fabs, VABSfd>;
5018 def : N2VSPat<fneg, VNEGfd>;
5019 def : N3VSPat<NEONfmax, VMAXfd>;
5020 def : N3VSPat<NEONfmin, VMINfd>;
5021 def : N2VSPat<arm_ftosi, VCVTf2sd>;
5022 def : N2VSPat<arm_ftoui, VCVTf2ud>;
5023 def : N2VSPat<arm_sitof, VCVTs2fd>;
5024 def : N2VSPat<arm_uitof, VCVTu2fd>;
5026 //===----------------------------------------------------------------------===//
5027 // Non-Instruction Patterns
5028 //===----------------------------------------------------------------------===//
5031 def : Pat<(v1i64 (bitconvert (v2i32 DPR:$src))), (v1i64 DPR:$src)>;
5032 def : Pat<(v1i64 (bitconvert (v4i16 DPR:$src))), (v1i64 DPR:$src)>;
5033 def : Pat<(v1i64 (bitconvert (v8i8 DPR:$src))), (v1i64 DPR:$src)>;
5034 def : Pat<(v1i64 (bitconvert (f64 DPR:$src))), (v1i64 DPR:$src)>;
5035 def : Pat<(v1i64 (bitconvert (v2f32 DPR:$src))), (v1i64 DPR:$src)>;
5036 def : Pat<(v2i32 (bitconvert (v1i64 DPR:$src))), (v2i32 DPR:$src)>;
5037 def : Pat<(v2i32 (bitconvert (v4i16 DPR:$src))), (v2i32 DPR:$src)>;
5038 def : Pat<(v2i32 (bitconvert (v8i8 DPR:$src))), (v2i32 DPR:$src)>;
5039 def : Pat<(v2i32 (bitconvert (f64 DPR:$src))), (v2i32 DPR:$src)>;
5040 def : Pat<(v2i32 (bitconvert (v2f32 DPR:$src))), (v2i32 DPR:$src)>;
5041 def : Pat<(v4i16 (bitconvert (v1i64 DPR:$src))), (v4i16 DPR:$src)>;
5042 def : Pat<(v4i16 (bitconvert (v2i32 DPR:$src))), (v4i16 DPR:$src)>;
5043 def : Pat<(v4i16 (bitconvert (v8i8 DPR:$src))), (v4i16 DPR:$src)>;
5044 def : Pat<(v4i16 (bitconvert (f64 DPR:$src))), (v4i16 DPR:$src)>;
5045 def : Pat<(v4i16 (bitconvert (v2f32 DPR:$src))), (v4i16 DPR:$src)>;
5046 def : Pat<(v8i8 (bitconvert (v1i64 DPR:$src))), (v8i8 DPR:$src)>;
5047 def : Pat<(v8i8 (bitconvert (v2i32 DPR:$src))), (v8i8 DPR:$src)>;
5048 def : Pat<(v8i8 (bitconvert (v4i16 DPR:$src))), (v8i8 DPR:$src)>;
5049 def : Pat<(v8i8 (bitconvert (f64 DPR:$src))), (v8i8 DPR:$src)>;
5050 def : Pat<(v8i8 (bitconvert (v2f32 DPR:$src))), (v8i8 DPR:$src)>;
5051 def : Pat<(f64 (bitconvert (v1i64 DPR:$src))), (f64 DPR:$src)>;
5052 def : Pat<(f64 (bitconvert (v2i32 DPR:$src))), (f64 DPR:$src)>;
5053 def : Pat<(f64 (bitconvert (v4i16 DPR:$src))), (f64 DPR:$src)>;
5054 def : Pat<(f64 (bitconvert (v8i8 DPR:$src))), (f64 DPR:$src)>;
5055 def : Pat<(f64 (bitconvert (v2f32 DPR:$src))), (f64 DPR:$src)>;
5056 def : Pat<(v2f32 (bitconvert (f64 DPR:$src))), (v2f32 DPR:$src)>;
5057 def : Pat<(v2f32 (bitconvert (v1i64 DPR:$src))), (v2f32 DPR:$src)>;
5058 def : Pat<(v2f32 (bitconvert (v2i32 DPR:$src))), (v2f32 DPR:$src)>;
5059 def : Pat<(v2f32 (bitconvert (v4i16 DPR:$src))), (v2f32 DPR:$src)>;
5060 def : Pat<(v2f32 (bitconvert (v8i8 DPR:$src))), (v2f32 DPR:$src)>;
5062 def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (v2i64 QPR:$src)>;
5063 def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (v2i64 QPR:$src)>;
5064 def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (v2i64 QPR:$src)>;
5065 def : Pat<(v2i64 (bitconvert (v2f64 QPR:$src))), (v2i64 QPR:$src)>;
5066 def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (v2i64 QPR:$src)>;
5067 def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (v4i32 QPR:$src)>;
5068 def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (v4i32 QPR:$src)>;
5069 def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (v4i32 QPR:$src)>;
5070 def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (v4i32 QPR:$src)>;
5071 def : Pat<(v4i32 (bitconvert (v4f32 QPR:$src))), (v4i32 QPR:$src)>;
5072 def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (v8i16 QPR:$src)>;
5073 def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (v8i16 QPR:$src)>;
5074 def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (v8i16 QPR:$src)>;
5075 def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (v8i16 QPR:$src)>;
5076 def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (v8i16 QPR:$src)>;
5077 def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (v16i8 QPR:$src)>;
5078 def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (v16i8 QPR:$src)>;
5079 def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (v16i8 QPR:$src)>;
5080 def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (v16i8 QPR:$src)>;
5081 def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (v16i8 QPR:$src)>;
5082 def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (v4f32 QPR:$src)>;
5083 def : Pat<(v4f32 (bitconvert (v4i32 QPR:$src))), (v4f32 QPR:$src)>;
5084 def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (v4f32 QPR:$src)>;
5085 def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (v4f32 QPR:$src)>;
5086 def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (v4f32 QPR:$src)>;
5087 def : Pat<(v2f64 (bitconvert (v2i64 QPR:$src))), (v2f64 QPR:$src)>;
5088 def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (v2f64 QPR:$src)>;
5089 def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 QPR:$src)>;
5090 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 QPR:$src)>;
5091 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 QPR:$src)>;