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;
28 def VectorIndex8Operand : AsmOperandClass { let Name = "VectorIndex8"; }
29 def VectorIndex16Operand : AsmOperandClass { let Name = "VectorIndex16"; }
30 def VectorIndex32Operand : AsmOperandClass { let Name = "VectorIndex32"; }
31 def VectorIndex8 : Operand<i32>, ImmLeaf<i32, [{
32 return ((uint64_t)Imm) < 8;
34 let ParserMatchClass = VectorIndex8Operand;
35 let PrintMethod = "printVectorIndex";
36 let MIOperandInfo = (ops i32imm);
38 def VectorIndex16 : Operand<i32>, ImmLeaf<i32, [{
39 return ((uint64_t)Imm) < 4;
41 let ParserMatchClass = VectorIndex16Operand;
42 let PrintMethod = "printVectorIndex";
43 let MIOperandInfo = (ops i32imm);
45 def VectorIndex32 : Operand<i32>, ImmLeaf<i32, [{
46 return ((uint64_t)Imm) < 2;
48 let ParserMatchClass = VectorIndex32Operand;
49 let PrintMethod = "printVectorIndex";
50 let MIOperandInfo = (ops i32imm);
53 //===----------------------------------------------------------------------===//
54 // NEON-specific DAG Nodes.
55 //===----------------------------------------------------------------------===//
57 def SDTARMVCMP : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisSameAs<1, 2>]>;
58 def SDTARMVCMPZ : SDTypeProfile<1, 1, []>;
60 def NEONvceq : SDNode<"ARMISD::VCEQ", SDTARMVCMP>;
61 def NEONvceqz : SDNode<"ARMISD::VCEQZ", SDTARMVCMPZ>;
62 def NEONvcge : SDNode<"ARMISD::VCGE", SDTARMVCMP>;
63 def NEONvcgez : SDNode<"ARMISD::VCGEZ", SDTARMVCMPZ>;
64 def NEONvclez : SDNode<"ARMISD::VCLEZ", SDTARMVCMPZ>;
65 def NEONvcgeu : SDNode<"ARMISD::VCGEU", SDTARMVCMP>;
66 def NEONvcgt : SDNode<"ARMISD::VCGT", SDTARMVCMP>;
67 def NEONvcgtz : SDNode<"ARMISD::VCGTZ", SDTARMVCMPZ>;
68 def NEONvcltz : SDNode<"ARMISD::VCLTZ", SDTARMVCMPZ>;
69 def NEONvcgtu : SDNode<"ARMISD::VCGTU", SDTARMVCMP>;
70 def NEONvtst : SDNode<"ARMISD::VTST", SDTARMVCMP>;
72 // Types for vector shift by immediates. The "SHX" version is for long and
73 // narrow operations where the source and destination vectors have different
74 // types. The "SHINS" version is for shift and insert operations.
75 def SDTARMVSH : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
77 def SDTARMVSHX : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
79 def SDTARMVSHINS : SDTypeProfile<1, 3, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
80 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
82 def NEONvshl : SDNode<"ARMISD::VSHL", SDTARMVSH>;
83 def NEONvshrs : SDNode<"ARMISD::VSHRs", SDTARMVSH>;
84 def NEONvshru : SDNode<"ARMISD::VSHRu", SDTARMVSH>;
85 def NEONvshlls : SDNode<"ARMISD::VSHLLs", SDTARMVSHX>;
86 def NEONvshllu : SDNode<"ARMISD::VSHLLu", SDTARMVSHX>;
87 def NEONvshlli : SDNode<"ARMISD::VSHLLi", SDTARMVSHX>;
88 def NEONvshrn : SDNode<"ARMISD::VSHRN", SDTARMVSHX>;
90 def NEONvrshrs : SDNode<"ARMISD::VRSHRs", SDTARMVSH>;
91 def NEONvrshru : SDNode<"ARMISD::VRSHRu", SDTARMVSH>;
92 def NEONvrshrn : SDNode<"ARMISD::VRSHRN", SDTARMVSHX>;
94 def NEONvqshls : SDNode<"ARMISD::VQSHLs", SDTARMVSH>;
95 def NEONvqshlu : SDNode<"ARMISD::VQSHLu", SDTARMVSH>;
96 def NEONvqshlsu : SDNode<"ARMISD::VQSHLsu", SDTARMVSH>;
97 def NEONvqshrns : SDNode<"ARMISD::VQSHRNs", SDTARMVSHX>;
98 def NEONvqshrnu : SDNode<"ARMISD::VQSHRNu", SDTARMVSHX>;
99 def NEONvqshrnsu : SDNode<"ARMISD::VQSHRNsu", SDTARMVSHX>;
101 def NEONvqrshrns : SDNode<"ARMISD::VQRSHRNs", SDTARMVSHX>;
102 def NEONvqrshrnu : SDNode<"ARMISD::VQRSHRNu", SDTARMVSHX>;
103 def NEONvqrshrnsu : SDNode<"ARMISD::VQRSHRNsu", SDTARMVSHX>;
105 def NEONvsli : SDNode<"ARMISD::VSLI", SDTARMVSHINS>;
106 def NEONvsri : SDNode<"ARMISD::VSRI", SDTARMVSHINS>;
108 def SDTARMVGETLN : SDTypeProfile<1, 2, [SDTCisVT<0, i32>, SDTCisInt<1>,
110 def NEONvgetlaneu : SDNode<"ARMISD::VGETLANEu", SDTARMVGETLN>;
111 def NEONvgetlanes : SDNode<"ARMISD::VGETLANEs", SDTARMVGETLN>;
113 def SDTARMVMOVIMM : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisVT<1, i32>]>;
114 def NEONvmovImm : SDNode<"ARMISD::VMOVIMM", SDTARMVMOVIMM>;
115 def NEONvmvnImm : SDNode<"ARMISD::VMVNIMM", SDTARMVMOVIMM>;
117 def SDTARMVORRIMM : SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
119 def NEONvorrImm : SDNode<"ARMISD::VORRIMM", SDTARMVORRIMM>;
120 def NEONvbicImm : SDNode<"ARMISD::VBICIMM", SDTARMVORRIMM>;
122 def NEONvbsl : SDNode<"ARMISD::VBSL",
123 SDTypeProfile<1, 3, [SDTCisVec<0>,
126 SDTCisSameAs<0, 3>]>>;
128 def NEONvdup : SDNode<"ARMISD::VDUP", SDTypeProfile<1, 1, [SDTCisVec<0>]>>;
130 // VDUPLANE can produce a quad-register result from a double-register source,
131 // so the result is not constrained to match the source.
132 def NEONvduplane : SDNode<"ARMISD::VDUPLANE",
133 SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisVec<1>,
136 def SDTARMVEXT : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
137 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
138 def NEONvext : SDNode<"ARMISD::VEXT", SDTARMVEXT>;
140 def SDTARMVSHUF : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisSameAs<0, 1>]>;
141 def NEONvrev64 : SDNode<"ARMISD::VREV64", SDTARMVSHUF>;
142 def NEONvrev32 : SDNode<"ARMISD::VREV32", SDTARMVSHUF>;
143 def NEONvrev16 : SDNode<"ARMISD::VREV16", SDTARMVSHUF>;
145 def SDTARMVSHUF2 : SDTypeProfile<2, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
147 SDTCisSameAs<0, 3>]>;
148 def NEONzip : SDNode<"ARMISD::VZIP", SDTARMVSHUF2>;
149 def NEONuzp : SDNode<"ARMISD::VUZP", SDTARMVSHUF2>;
150 def NEONtrn : SDNode<"ARMISD::VTRN", SDTARMVSHUF2>;
152 def SDTARMVMULL : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
153 SDTCisSameAs<1, 2>]>;
154 def NEONvmulls : SDNode<"ARMISD::VMULLs", SDTARMVMULL>;
155 def NEONvmullu : SDNode<"ARMISD::VMULLu", SDTARMVMULL>;
157 def SDTARMFMAX : SDTypeProfile<1, 2, [SDTCisVT<0, f32>, SDTCisSameAs<0, 1>,
158 SDTCisSameAs<0, 2>]>;
159 def NEONfmax : SDNode<"ARMISD::FMAX", SDTARMFMAX>;
160 def NEONfmin : SDNode<"ARMISD::FMIN", SDTARMFMAX>;
162 def NEONimmAllZerosV: PatLeaf<(NEONvmovImm (i32 timm)), [{
163 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
164 unsigned EltBits = 0;
165 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
166 return (EltBits == 32 && EltVal == 0);
169 def NEONimmAllOnesV: PatLeaf<(NEONvmovImm (i32 timm)), [{
170 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
171 unsigned EltBits = 0;
172 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
173 return (EltBits == 8 && EltVal == 0xff);
176 //===----------------------------------------------------------------------===//
177 // NEON load / store instructions
178 //===----------------------------------------------------------------------===//
180 // Use VLDM to load a Q register as a D register pair.
181 // This is a pseudo instruction that is expanded to VLDMD after reg alloc.
183 : PseudoVFPLdStM<(outs QPR:$dst), (ins GPR:$Rn),
185 [(set QPR:$dst, (v2f64 (load GPR:$Rn)))]>;
187 // Use VSTM to store a Q register as a D register pair.
188 // This is a pseudo instruction that is expanded to VSTMD after reg alloc.
190 : PseudoVFPLdStM<(outs), (ins QPR:$src, GPR:$Rn),
192 [(store (v2f64 QPR:$src), GPR:$Rn)]>;
194 // Classes for VLD* pseudo-instructions with multi-register operands.
195 // These are expanded to real instructions after register allocation.
196 class VLDQPseudo<InstrItinClass itin>
197 : PseudoNLdSt<(outs QPR:$dst), (ins addrmode6:$addr), itin, "">;
198 class VLDQWBPseudo<InstrItinClass itin>
199 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
200 (ins addrmode6:$addr, am6offset:$offset), itin,
202 class VLDQQPseudo<InstrItinClass itin>
203 : PseudoNLdSt<(outs QQPR:$dst), (ins addrmode6:$addr), itin, "">;
204 class VLDQQWBPseudo<InstrItinClass itin>
205 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
206 (ins addrmode6:$addr, am6offset:$offset), itin,
208 class VLDQQQQPseudo<InstrItinClass itin>
209 : PseudoNLdSt<(outs QQQQPR:$dst), (ins addrmode6:$addr, QQQQPR:$src),itin,
211 class VLDQQQQWBPseudo<InstrItinClass itin>
212 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
213 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
214 "$addr.addr = $wb, $src = $dst">;
216 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
218 // VLD1 : Vector Load (multiple single elements)
219 class VLD1D<bits<4> op7_4, string Dt>
220 : NLdSt<0,0b10,0b0111,op7_4, (outs DPR:$Vd),
221 (ins addrmode6:$Rn), IIC_VLD1,
222 "vld1", Dt, "\\{$Vd\\}, $Rn", "", []> {
225 let DecoderMethod = "DecodeVLDInstruction";
227 class VLD1Q<bits<4> op7_4, string Dt>
228 : NLdSt<0,0b10,0b1010,op7_4, (outs DPR:$Vd, DPR:$dst2),
229 (ins addrmode6:$Rn), IIC_VLD1x2,
230 "vld1", Dt, "\\{$Vd, $dst2\\}, $Rn", "", []> {
232 let Inst{5-4} = Rn{5-4};
233 let DecoderMethod = "DecodeVLDInstruction";
236 def VLD1d8 : VLD1D<{0,0,0,?}, "8">;
237 def VLD1d16 : VLD1D<{0,1,0,?}, "16">;
238 def VLD1d32 : VLD1D<{1,0,0,?}, "32">;
239 def VLD1d64 : VLD1D<{1,1,0,?}, "64">;
241 def VLD1q8 : VLD1Q<{0,0,?,?}, "8">;
242 def VLD1q16 : VLD1Q<{0,1,?,?}, "16">;
243 def VLD1q32 : VLD1Q<{1,0,?,?}, "32">;
244 def VLD1q64 : VLD1Q<{1,1,?,?}, "64">;
246 def VLD1q8Pseudo : VLDQPseudo<IIC_VLD1x2>;
247 def VLD1q16Pseudo : VLDQPseudo<IIC_VLD1x2>;
248 def VLD1q32Pseudo : VLDQPseudo<IIC_VLD1x2>;
249 def VLD1q64Pseudo : VLDQPseudo<IIC_VLD1x2>;
251 // ...with address register writeback:
252 class VLD1DWB<bits<4> op7_4, string Dt>
253 : NLdSt<0,0b10,0b0111,op7_4, (outs DPR:$Vd, GPR:$wb),
254 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD1u,
255 "vld1", Dt, "\\{$Vd\\}, $Rn$Rm",
256 "$Rn.addr = $wb", []> {
258 let DecoderMethod = "DecodeVLDInstruction";
260 class VLD1QWB<bits<4> op7_4, string Dt>
261 : NLdSt<0,0b10,0b1010,op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
262 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD1x2u,
263 "vld1", Dt, "\\{$Vd, $dst2\\}, $Rn$Rm",
264 "$Rn.addr = $wb", []> {
265 let Inst{5-4} = Rn{5-4};
266 let DecoderMethod = "DecodeVLDInstruction";
269 def VLD1d8_UPD : VLD1DWB<{0,0,0,?}, "8">;
270 def VLD1d16_UPD : VLD1DWB<{0,1,0,?}, "16">;
271 def VLD1d32_UPD : VLD1DWB<{1,0,0,?}, "32">;
272 def VLD1d64_UPD : VLD1DWB<{1,1,0,?}, "64">;
274 def VLD1q8_UPD : VLD1QWB<{0,0,?,?}, "8">;
275 def VLD1q16_UPD : VLD1QWB<{0,1,?,?}, "16">;
276 def VLD1q32_UPD : VLD1QWB<{1,0,?,?}, "32">;
277 def VLD1q64_UPD : VLD1QWB<{1,1,?,?}, "64">;
279 def VLD1q8Pseudo_UPD : VLDQWBPseudo<IIC_VLD1x2u>;
280 def VLD1q16Pseudo_UPD : VLDQWBPseudo<IIC_VLD1x2u>;
281 def VLD1q32Pseudo_UPD : VLDQWBPseudo<IIC_VLD1x2u>;
282 def VLD1q64Pseudo_UPD : VLDQWBPseudo<IIC_VLD1x2u>;
284 // ...with 3 registers (some of these are only for the disassembler):
285 class VLD1D3<bits<4> op7_4, string Dt>
286 : NLdSt<0,0b10,0b0110,op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
287 (ins addrmode6:$Rn), IIC_VLD1x3, "vld1", Dt,
288 "\\{$Vd, $dst2, $dst3\\}, $Rn", "", []> {
291 let DecoderMethod = "DecodeVLDInstruction";
293 class VLD1D3WB<bits<4> op7_4, string Dt>
294 : NLdSt<0,0b10,0b0110,op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
295 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD1x3u, "vld1", Dt,
296 "\\{$Vd, $dst2, $dst3\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
298 let DecoderMethod = "DecodeVLDInstruction";
301 def VLD1d8T : VLD1D3<{0,0,0,?}, "8">;
302 def VLD1d16T : VLD1D3<{0,1,0,?}, "16">;
303 def VLD1d32T : VLD1D3<{1,0,0,?}, "32">;
304 def VLD1d64T : VLD1D3<{1,1,0,?}, "64">;
306 def VLD1d8T_UPD : VLD1D3WB<{0,0,0,?}, "8">;
307 def VLD1d16T_UPD : VLD1D3WB<{0,1,0,?}, "16">;
308 def VLD1d32T_UPD : VLD1D3WB<{1,0,0,?}, "32">;
309 def VLD1d64T_UPD : VLD1D3WB<{1,1,0,?}, "64">;
311 def VLD1d64TPseudo : VLDQQPseudo<IIC_VLD1x3>;
312 def VLD1d64TPseudo_UPD : VLDQQWBPseudo<IIC_VLD1x3u>;
314 // ...with 4 registers (some of these are only for the disassembler):
315 class VLD1D4<bits<4> op7_4, string Dt>
316 : NLdSt<0,0b10,0b0010,op7_4,(outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
317 (ins addrmode6:$Rn), IIC_VLD1x4, "vld1", Dt,
318 "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn", "", []> {
320 let Inst{5-4} = Rn{5-4};
321 let DecoderMethod = "DecodeVLDInstruction";
323 class VLD1D4WB<bits<4> op7_4, string Dt>
324 : NLdSt<0,0b10,0b0010,op7_4,
325 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
326 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD1x4u, "vld1", Dt,
327 "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn$Rm", "$Rn.addr = $wb",
329 let Inst{5-4} = Rn{5-4};
330 let DecoderMethod = "DecodeVLDInstruction";
333 def VLD1d8Q : VLD1D4<{0,0,?,?}, "8">;
334 def VLD1d16Q : VLD1D4<{0,1,?,?}, "16">;
335 def VLD1d32Q : VLD1D4<{1,0,?,?}, "32">;
336 def VLD1d64Q : VLD1D4<{1,1,?,?}, "64">;
338 def VLD1d8Q_UPD : VLD1D4WB<{0,0,?,?}, "8">;
339 def VLD1d16Q_UPD : VLD1D4WB<{0,1,?,?}, "16">;
340 def VLD1d32Q_UPD : VLD1D4WB<{1,0,?,?}, "32">;
341 def VLD1d64Q_UPD : VLD1D4WB<{1,1,?,?}, "64">;
343 def VLD1d64QPseudo : VLDQQPseudo<IIC_VLD1x4>;
344 def VLD1d64QPseudo_UPD : VLDQQWBPseudo<IIC_VLD1x4u>;
346 // VLD2 : Vector Load (multiple 2-element structures)
347 class VLD2D<bits<4> op11_8, bits<4> op7_4, string Dt>
348 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2),
349 (ins addrmode6:$Rn), IIC_VLD2,
350 "vld2", Dt, "\\{$Vd, $dst2\\}, $Rn", "", []> {
352 let Inst{5-4} = Rn{5-4};
353 let DecoderMethod = "DecodeVLDInstruction";
355 class VLD2Q<bits<4> op7_4, string Dt>
356 : NLdSt<0, 0b10, 0b0011, op7_4,
357 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
358 (ins addrmode6:$Rn), IIC_VLD2x2,
359 "vld2", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn", "", []> {
361 let Inst{5-4} = Rn{5-4};
362 let DecoderMethod = "DecodeVLDInstruction";
365 def VLD2d8 : VLD2D<0b1000, {0,0,?,?}, "8">;
366 def VLD2d16 : VLD2D<0b1000, {0,1,?,?}, "16">;
367 def VLD2d32 : VLD2D<0b1000, {1,0,?,?}, "32">;
369 def VLD2q8 : VLD2Q<{0,0,?,?}, "8">;
370 def VLD2q16 : VLD2Q<{0,1,?,?}, "16">;
371 def VLD2q32 : VLD2Q<{1,0,?,?}, "32">;
373 def VLD2d8Pseudo : VLDQPseudo<IIC_VLD2>;
374 def VLD2d16Pseudo : VLDQPseudo<IIC_VLD2>;
375 def VLD2d32Pseudo : VLDQPseudo<IIC_VLD2>;
377 def VLD2q8Pseudo : VLDQQPseudo<IIC_VLD2x2>;
378 def VLD2q16Pseudo : VLDQQPseudo<IIC_VLD2x2>;
379 def VLD2q32Pseudo : VLDQQPseudo<IIC_VLD2x2>;
381 // ...with address register writeback:
382 class VLD2DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
383 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
384 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD2u,
385 "vld2", Dt, "\\{$Vd, $dst2\\}, $Rn$Rm",
386 "$Rn.addr = $wb", []> {
387 let Inst{5-4} = Rn{5-4};
388 let DecoderMethod = "DecodeVLDInstruction";
390 class VLD2QWB<bits<4> op7_4, string Dt>
391 : NLdSt<0, 0b10, 0b0011, op7_4,
392 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
393 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD2x2u,
394 "vld2", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn$Rm",
395 "$Rn.addr = $wb", []> {
396 let Inst{5-4} = Rn{5-4};
397 let DecoderMethod = "DecodeVLDInstruction";
400 def VLD2d8_UPD : VLD2DWB<0b1000, {0,0,?,?}, "8">;
401 def VLD2d16_UPD : VLD2DWB<0b1000, {0,1,?,?}, "16">;
402 def VLD2d32_UPD : VLD2DWB<0b1000, {1,0,?,?}, "32">;
404 def VLD2q8_UPD : VLD2QWB<{0,0,?,?}, "8">;
405 def VLD2q16_UPD : VLD2QWB<{0,1,?,?}, "16">;
406 def VLD2q32_UPD : VLD2QWB<{1,0,?,?}, "32">;
408 def VLD2d8Pseudo_UPD : VLDQWBPseudo<IIC_VLD2u>;
409 def VLD2d16Pseudo_UPD : VLDQWBPseudo<IIC_VLD2u>;
410 def VLD2d32Pseudo_UPD : VLDQWBPseudo<IIC_VLD2u>;
412 def VLD2q8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD2x2u>;
413 def VLD2q16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD2x2u>;
414 def VLD2q32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD2x2u>;
416 // ...with double-spaced registers (for disassembly only):
417 def VLD2b8 : VLD2D<0b1001, {0,0,?,?}, "8">;
418 def VLD2b16 : VLD2D<0b1001, {0,1,?,?}, "16">;
419 def VLD2b32 : VLD2D<0b1001, {1,0,?,?}, "32">;
420 def VLD2b8_UPD : VLD2DWB<0b1001, {0,0,?,?}, "8">;
421 def VLD2b16_UPD : VLD2DWB<0b1001, {0,1,?,?}, "16">;
422 def VLD2b32_UPD : VLD2DWB<0b1001, {1,0,?,?}, "32">;
424 // VLD3 : Vector Load (multiple 3-element structures)
425 class VLD3D<bits<4> op11_8, bits<4> op7_4, string Dt>
426 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
427 (ins addrmode6:$Rn), IIC_VLD3,
428 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn", "", []> {
431 let DecoderMethod = "DecodeVLDInstruction";
434 def VLD3d8 : VLD3D<0b0100, {0,0,0,?}, "8">;
435 def VLD3d16 : VLD3D<0b0100, {0,1,0,?}, "16">;
436 def VLD3d32 : VLD3D<0b0100, {1,0,0,?}, "32">;
438 def VLD3d8Pseudo : VLDQQPseudo<IIC_VLD3>;
439 def VLD3d16Pseudo : VLDQQPseudo<IIC_VLD3>;
440 def VLD3d32Pseudo : VLDQQPseudo<IIC_VLD3>;
442 // ...with address register writeback:
443 class VLD3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
444 : NLdSt<0, 0b10, op11_8, op7_4,
445 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
446 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD3u,
447 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn$Rm",
448 "$Rn.addr = $wb", []> {
450 let DecoderMethod = "DecodeVLDInstruction";
453 def VLD3d8_UPD : VLD3DWB<0b0100, {0,0,0,?}, "8">;
454 def VLD3d16_UPD : VLD3DWB<0b0100, {0,1,0,?}, "16">;
455 def VLD3d32_UPD : VLD3DWB<0b0100, {1,0,0,?}, "32">;
457 def VLD3d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
458 def VLD3d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
459 def VLD3d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
461 // ...with double-spaced registers:
462 def VLD3q8 : VLD3D<0b0101, {0,0,0,?}, "8">;
463 def VLD3q16 : VLD3D<0b0101, {0,1,0,?}, "16">;
464 def VLD3q32 : VLD3D<0b0101, {1,0,0,?}, "32">;
465 def VLD3q8_UPD : VLD3DWB<0b0101, {0,0,0,?}, "8">;
466 def VLD3q16_UPD : VLD3DWB<0b0101, {0,1,0,?}, "16">;
467 def VLD3q32_UPD : VLD3DWB<0b0101, {1,0,0,?}, "32">;
469 def VLD3q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
470 def VLD3q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
471 def VLD3q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
473 // ...alternate versions to be allocated odd register numbers:
474 def VLD3q8oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
475 def VLD3q16oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
476 def VLD3q32oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
478 def VLD3q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
479 def VLD3q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
480 def VLD3q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
482 // VLD4 : Vector Load (multiple 4-element structures)
483 class VLD4D<bits<4> op11_8, bits<4> op7_4, string Dt>
484 : NLdSt<0, 0b10, op11_8, op7_4,
485 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
486 (ins addrmode6:$Rn), IIC_VLD4,
487 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn", "", []> {
489 let Inst{5-4} = Rn{5-4};
490 let DecoderMethod = "DecodeVLDInstruction";
493 def VLD4d8 : VLD4D<0b0000, {0,0,?,?}, "8">;
494 def VLD4d16 : VLD4D<0b0000, {0,1,?,?}, "16">;
495 def VLD4d32 : VLD4D<0b0000, {1,0,?,?}, "32">;
497 def VLD4d8Pseudo : VLDQQPseudo<IIC_VLD4>;
498 def VLD4d16Pseudo : VLDQQPseudo<IIC_VLD4>;
499 def VLD4d32Pseudo : VLDQQPseudo<IIC_VLD4>;
501 // ...with address register writeback:
502 class VLD4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
503 : NLdSt<0, 0b10, op11_8, op7_4,
504 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
505 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD4u,
506 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn$Rm",
507 "$Rn.addr = $wb", []> {
508 let Inst{5-4} = Rn{5-4};
509 let DecoderMethod = "DecodeVLDInstruction";
512 def VLD4d8_UPD : VLD4DWB<0b0000, {0,0,?,?}, "8">;
513 def VLD4d16_UPD : VLD4DWB<0b0000, {0,1,?,?}, "16">;
514 def VLD4d32_UPD : VLD4DWB<0b0000, {1,0,?,?}, "32">;
516 def VLD4d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
517 def VLD4d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
518 def VLD4d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
520 // ...with double-spaced registers:
521 def VLD4q8 : VLD4D<0b0001, {0,0,?,?}, "8">;
522 def VLD4q16 : VLD4D<0b0001, {0,1,?,?}, "16">;
523 def VLD4q32 : VLD4D<0b0001, {1,0,?,?}, "32">;
524 def VLD4q8_UPD : VLD4DWB<0b0001, {0,0,?,?}, "8">;
525 def VLD4q16_UPD : VLD4DWB<0b0001, {0,1,?,?}, "16">;
526 def VLD4q32_UPD : VLD4DWB<0b0001, {1,0,?,?}, "32">;
528 def VLD4q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
529 def VLD4q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
530 def VLD4q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
532 // ...alternate versions to be allocated odd register numbers:
533 def VLD4q8oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
534 def VLD4q16oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
535 def VLD4q32oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
537 def VLD4q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
538 def VLD4q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
539 def VLD4q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
541 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
543 // Classes for VLD*LN pseudo-instructions with multi-register operands.
544 // These are expanded to real instructions after register allocation.
545 class VLDQLNPseudo<InstrItinClass itin>
546 : PseudoNLdSt<(outs QPR:$dst),
547 (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
548 itin, "$src = $dst">;
549 class VLDQLNWBPseudo<InstrItinClass itin>
550 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
551 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
552 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
553 class VLDQQLNPseudo<InstrItinClass itin>
554 : PseudoNLdSt<(outs QQPR:$dst),
555 (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
556 itin, "$src = $dst">;
557 class VLDQQLNWBPseudo<InstrItinClass itin>
558 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
559 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
560 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
561 class VLDQQQQLNPseudo<InstrItinClass itin>
562 : PseudoNLdSt<(outs QQQQPR:$dst),
563 (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
564 itin, "$src = $dst">;
565 class VLDQQQQLNWBPseudo<InstrItinClass itin>
566 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
567 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
568 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
570 // VLD1LN : Vector Load (single element to one lane)
571 class VLD1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
573 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd),
574 (ins addrmode6:$Rn, DPR:$src, nohash_imm:$lane),
575 IIC_VLD1ln, "vld1", Dt, "\\{$Vd[$lane]\\}, $Rn",
577 [(set DPR:$Vd, (vector_insert (Ty DPR:$src),
578 (i32 (LoadOp addrmode6:$Rn)),
581 let DecoderMethod = "DecodeVLD1LN";
583 class VLD1LN32<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
585 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd),
586 (ins addrmode6oneL32:$Rn, DPR:$src, nohash_imm:$lane),
587 IIC_VLD1ln, "vld1", Dt, "\\{$Vd[$lane]\\}, $Rn",
589 [(set DPR:$Vd, (vector_insert (Ty DPR:$src),
590 (i32 (LoadOp addrmode6oneL32:$Rn)),
593 let DecoderMethod = "DecodeVLD1LN";
595 class VLD1QLNPseudo<ValueType Ty, PatFrag LoadOp> : VLDQLNPseudo<IIC_VLD1ln> {
596 let Pattern = [(set QPR:$dst, (vector_insert (Ty QPR:$src),
597 (i32 (LoadOp addrmode6:$addr)),
601 def VLD1LNd8 : VLD1LN<0b0000, {?,?,?,0}, "8", v8i8, extloadi8> {
602 let Inst{7-5} = lane{2-0};
604 def VLD1LNd16 : VLD1LN<0b0100, {?,?,0,?}, "16", v4i16, extloadi16> {
605 let Inst{7-6} = lane{1-0};
608 def VLD1LNd32 : VLD1LN32<0b1000, {?,0,?,?}, "32", v2i32, load> {
609 let Inst{7} = lane{0};
614 def VLD1LNq8Pseudo : VLD1QLNPseudo<v16i8, extloadi8>;
615 def VLD1LNq16Pseudo : VLD1QLNPseudo<v8i16, extloadi16>;
616 def VLD1LNq32Pseudo : VLD1QLNPseudo<v4i32, load>;
618 def : Pat<(vector_insert (v2f32 DPR:$src),
619 (f32 (load addrmode6:$addr)), imm:$lane),
620 (VLD1LNd32 addrmode6:$addr, DPR:$src, imm:$lane)>;
621 def : Pat<(vector_insert (v4f32 QPR:$src),
622 (f32 (load addrmode6:$addr)), imm:$lane),
623 (VLD1LNq32Pseudo addrmode6:$addr, QPR:$src, imm:$lane)>;
625 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
627 // ...with address register writeback:
628 class VLD1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
629 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, GPR:$wb),
630 (ins addrmode6:$Rn, am6offset:$Rm,
631 DPR:$src, nohash_imm:$lane), IIC_VLD1lnu, "vld1", Dt,
632 "\\{$Vd[$lane]\\}, $Rn$Rm",
633 "$src = $Vd, $Rn.addr = $wb", []> {
634 let DecoderMethod = "DecodeVLD1LN";
637 def VLD1LNd8_UPD : VLD1LNWB<0b0000, {?,?,?,0}, "8"> {
638 let Inst{7-5} = lane{2-0};
640 def VLD1LNd16_UPD : VLD1LNWB<0b0100, {?,?,0,?}, "16"> {
641 let Inst{7-6} = lane{1-0};
644 def VLD1LNd32_UPD : VLD1LNWB<0b1000, {?,0,?,?}, "32"> {
645 let Inst{7} = lane{0};
650 def VLD1LNq8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
651 def VLD1LNq16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
652 def VLD1LNq32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
654 // VLD2LN : Vector Load (single 2-element structure to one lane)
655 class VLD2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
656 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2),
657 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, nohash_imm:$lane),
658 IIC_VLD2ln, "vld2", Dt, "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn",
659 "$src1 = $Vd, $src2 = $dst2", []> {
662 let DecoderMethod = "DecodeVLD2LN";
665 def VLD2LNd8 : VLD2LN<0b0001, {?,?,?,?}, "8"> {
666 let Inst{7-5} = lane{2-0};
668 def VLD2LNd16 : VLD2LN<0b0101, {?,?,0,?}, "16"> {
669 let Inst{7-6} = lane{1-0};
671 def VLD2LNd32 : VLD2LN<0b1001, {?,0,0,?}, "32"> {
672 let Inst{7} = lane{0};
675 def VLD2LNd8Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
676 def VLD2LNd16Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
677 def VLD2LNd32Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
679 // ...with double-spaced registers:
680 def VLD2LNq16 : VLD2LN<0b0101, {?,?,1,?}, "16"> {
681 let Inst{7-6} = lane{1-0};
683 def VLD2LNq32 : VLD2LN<0b1001, {?,1,0,?}, "32"> {
684 let Inst{7} = lane{0};
687 def VLD2LNq16Pseudo : VLDQQLNPseudo<IIC_VLD2ln>;
688 def VLD2LNq32Pseudo : VLDQQLNPseudo<IIC_VLD2ln>;
690 // ...with address register writeback:
691 class VLD2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
692 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
693 (ins addrmode6:$Rn, am6offset:$Rm,
694 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VLD2lnu, "vld2", Dt,
695 "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn$Rm",
696 "$src1 = $Vd, $src2 = $dst2, $Rn.addr = $wb", []> {
698 let DecoderMethod = "DecodeVLD2LN";
701 def VLD2LNd8_UPD : VLD2LNWB<0b0001, {?,?,?,?}, "8"> {
702 let Inst{7-5} = lane{2-0};
704 def VLD2LNd16_UPD : VLD2LNWB<0b0101, {?,?,0,?}, "16"> {
705 let Inst{7-6} = lane{1-0};
707 def VLD2LNd32_UPD : VLD2LNWB<0b1001, {?,0,0,?}, "32"> {
708 let Inst{7} = lane{0};
711 def VLD2LNd8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
712 def VLD2LNd16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
713 def VLD2LNd32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
715 def VLD2LNq16_UPD : VLD2LNWB<0b0101, {?,?,1,?}, "16"> {
716 let Inst{7-6} = lane{1-0};
718 def VLD2LNq32_UPD : VLD2LNWB<0b1001, {?,1,0,?}, "32"> {
719 let Inst{7} = lane{0};
722 def VLD2LNq16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
723 def VLD2LNq32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
725 // VLD3LN : Vector Load (single 3-element structure to one lane)
726 class VLD3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
727 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
728 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3,
729 nohash_imm:$lane), IIC_VLD3ln, "vld3", Dt,
730 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn",
731 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3", []> {
733 let DecoderMethod = "DecodeVLD3LN";
736 def VLD3LNd8 : VLD3LN<0b0010, {?,?,?,0}, "8"> {
737 let Inst{7-5} = lane{2-0};
739 def VLD3LNd16 : VLD3LN<0b0110, {?,?,0,0}, "16"> {
740 let Inst{7-6} = lane{1-0};
742 def VLD3LNd32 : VLD3LN<0b1010, {?,0,0,0}, "32"> {
743 let Inst{7} = lane{0};
746 def VLD3LNd8Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
747 def VLD3LNd16Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
748 def VLD3LNd32Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
750 // ...with double-spaced registers:
751 def VLD3LNq16 : VLD3LN<0b0110, {?,?,1,0}, "16"> {
752 let Inst{7-6} = lane{1-0};
754 def VLD3LNq32 : VLD3LN<0b1010, {?,1,0,0}, "32"> {
755 let Inst{7} = lane{0};
758 def VLD3LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
759 def VLD3LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
761 // ...with address register writeback:
762 class VLD3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
763 : NLdStLn<1, 0b10, op11_8, op7_4,
764 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
765 (ins addrmode6:$Rn, am6offset:$Rm,
766 DPR:$src1, DPR:$src2, DPR:$src3, nohash_imm:$lane),
767 IIC_VLD3lnu, "vld3", Dt,
768 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn$Rm",
769 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $Rn.addr = $wb",
771 let DecoderMethod = "DecodeVLD3LN";
774 def VLD3LNd8_UPD : VLD3LNWB<0b0010, {?,?,?,0}, "8"> {
775 let Inst{7-5} = lane{2-0};
777 def VLD3LNd16_UPD : VLD3LNWB<0b0110, {?,?,0,0}, "16"> {
778 let Inst{7-6} = lane{1-0};
780 def VLD3LNd32_UPD : VLD3LNWB<0b1010, {?,0,0,0}, "32"> {
781 let Inst{7} = lane{0};
784 def VLD3LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
785 def VLD3LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
786 def VLD3LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
788 def VLD3LNq16_UPD : VLD3LNWB<0b0110, {?,?,1,0}, "16"> {
789 let Inst{7-6} = lane{1-0};
791 def VLD3LNq32_UPD : VLD3LNWB<0b1010, {?,1,0,0}, "32"> {
792 let Inst{7} = lane{0};
795 def VLD3LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
796 def VLD3LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
798 // VLD4LN : Vector Load (single 4-element structure to one lane)
799 class VLD4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
800 : NLdStLn<1, 0b10, op11_8, op7_4,
801 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
802 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
803 nohash_imm:$lane), IIC_VLD4ln, "vld4", Dt,
804 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn",
805 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4", []> {
808 let DecoderMethod = "DecodeVLD4LN";
811 def VLD4LNd8 : VLD4LN<0b0011, {?,?,?,?}, "8"> {
812 let Inst{7-5} = lane{2-0};
814 def VLD4LNd16 : VLD4LN<0b0111, {?,?,0,?}, "16"> {
815 let Inst{7-6} = lane{1-0};
817 def VLD4LNd32 : VLD4LN<0b1011, {?,0,?,?}, "32"> {
818 let Inst{7} = lane{0};
822 def VLD4LNd8Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
823 def VLD4LNd16Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
824 def VLD4LNd32Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
826 // ...with double-spaced registers:
827 def VLD4LNq16 : VLD4LN<0b0111, {?,?,1,?}, "16"> {
828 let Inst{7-6} = lane{1-0};
830 def VLD4LNq32 : VLD4LN<0b1011, {?,1,?,?}, "32"> {
831 let Inst{7} = lane{0};
835 def VLD4LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>;
836 def VLD4LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>;
838 // ...with address register writeback:
839 class VLD4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
840 : NLdStLn<1, 0b10, op11_8, op7_4,
841 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
842 (ins addrmode6:$Rn, am6offset:$Rm,
843 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
844 IIC_VLD4lnu, "vld4", Dt,
845 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn$Rm",
846 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4, $Rn.addr = $wb",
849 let DecoderMethod = "DecodeVLD4LN" ;
852 def VLD4LNd8_UPD : VLD4LNWB<0b0011, {?,?,?,?}, "8"> {
853 let Inst{7-5} = lane{2-0};
855 def VLD4LNd16_UPD : VLD4LNWB<0b0111, {?,?,0,?}, "16"> {
856 let Inst{7-6} = lane{1-0};
858 def VLD4LNd32_UPD : VLD4LNWB<0b1011, {?,0,?,?}, "32"> {
859 let Inst{7} = lane{0};
863 def VLD4LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
864 def VLD4LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
865 def VLD4LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
867 def VLD4LNq16_UPD : VLD4LNWB<0b0111, {?,?,1,?}, "16"> {
868 let Inst{7-6} = lane{1-0};
870 def VLD4LNq32_UPD : VLD4LNWB<0b1011, {?,1,?,?}, "32"> {
871 let Inst{7} = lane{0};
875 def VLD4LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
876 def VLD4LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
878 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
880 // VLD1DUP : Vector Load (single element to all lanes)
881 class VLD1DUP<bits<4> op7_4, string Dt, ValueType Ty, PatFrag LoadOp>
882 : NLdSt<1, 0b10, 0b1100, op7_4, (outs DPR:$Vd), (ins addrmode6dup:$Rn),
883 IIC_VLD1dup, "vld1", Dt, "\\{$Vd[]\\}, $Rn", "",
884 [(set DPR:$Vd, (Ty (NEONvdup (i32 (LoadOp addrmode6dup:$Rn)))))]> {
887 let DecoderMethod = "DecodeVLD1DupInstruction";
889 class VLD1QDUPPseudo<ValueType Ty, PatFrag LoadOp> : VLDQPseudo<IIC_VLD1dup> {
890 let Pattern = [(set QPR:$dst,
891 (Ty (NEONvdup (i32 (LoadOp addrmode6dup:$addr)))))];
894 def VLD1DUPd8 : VLD1DUP<{0,0,0,?}, "8", v8i8, extloadi8>;
895 def VLD1DUPd16 : VLD1DUP<{0,1,0,?}, "16", v4i16, extloadi16>;
896 def VLD1DUPd32 : VLD1DUP<{1,0,0,?}, "32", v2i32, load>;
898 def VLD1DUPq8Pseudo : VLD1QDUPPseudo<v16i8, extloadi8>;
899 def VLD1DUPq16Pseudo : VLD1QDUPPseudo<v8i16, extloadi16>;
900 def VLD1DUPq32Pseudo : VLD1QDUPPseudo<v4i32, load>;
902 def : Pat<(v2f32 (NEONvdup (f32 (load addrmode6dup:$addr)))),
903 (VLD1DUPd32 addrmode6:$addr)>;
904 def : Pat<(v4f32 (NEONvdup (f32 (load addrmode6dup:$addr)))),
905 (VLD1DUPq32Pseudo addrmode6:$addr)>;
907 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
909 class VLD1QDUP<bits<4> op7_4, string Dt>
910 : NLdSt<1, 0b10, 0b1100, op7_4, (outs DPR:$Vd, DPR:$dst2),
911 (ins addrmode6dup:$Rn), IIC_VLD1dup,
912 "vld1", Dt, "\\{$Vd[], $dst2[]\\}, $Rn", "", []> {
915 let DecoderMethod = "DecodeVLD1DupInstruction";
918 def VLD1DUPq8 : VLD1QDUP<{0,0,1,0}, "8">;
919 def VLD1DUPq16 : VLD1QDUP<{0,1,1,?}, "16">;
920 def VLD1DUPq32 : VLD1QDUP<{1,0,1,?}, "32">;
922 // ...with address register writeback:
923 class VLD1DUPWB<bits<4> op7_4, string Dt>
924 : NLdSt<1, 0b10, 0b1100, op7_4, (outs DPR:$Vd, GPR:$wb),
925 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD1dupu,
926 "vld1", Dt, "\\{$Vd[]\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
928 let DecoderMethod = "DecodeVLD1DupInstruction";
930 class VLD1QDUPWB<bits<4> op7_4, string Dt>
931 : NLdSt<1, 0b10, 0b1100, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
932 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD1dupu,
933 "vld1", Dt, "\\{$Vd[], $dst2[]\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
935 let DecoderMethod = "DecodeVLD1DupInstruction";
938 def VLD1DUPd8_UPD : VLD1DUPWB<{0,0,0,0}, "8">;
939 def VLD1DUPd16_UPD : VLD1DUPWB<{0,1,0,?}, "16">;
940 def VLD1DUPd32_UPD : VLD1DUPWB<{1,0,0,?}, "32">;
942 def VLD1DUPq8_UPD : VLD1QDUPWB<{0,0,1,0}, "8">;
943 def VLD1DUPq16_UPD : VLD1QDUPWB<{0,1,1,?}, "16">;
944 def VLD1DUPq32_UPD : VLD1QDUPWB<{1,0,1,?}, "32">;
946 def VLD1DUPq8Pseudo_UPD : VLDQWBPseudo<IIC_VLD1dupu>;
947 def VLD1DUPq16Pseudo_UPD : VLDQWBPseudo<IIC_VLD1dupu>;
948 def VLD1DUPq32Pseudo_UPD : VLDQWBPseudo<IIC_VLD1dupu>;
950 // VLD2DUP : Vector Load (single 2-element structure to all lanes)
951 class VLD2DUP<bits<4> op7_4, string Dt>
952 : NLdSt<1, 0b10, 0b1101, op7_4, (outs DPR:$Vd, DPR:$dst2),
953 (ins addrmode6dup:$Rn), IIC_VLD2dup,
954 "vld2", Dt, "\\{$Vd[], $dst2[]\\}, $Rn", "", []> {
957 let DecoderMethod = "DecodeVLD2DupInstruction";
960 def VLD2DUPd8 : VLD2DUP<{0,0,0,?}, "8">;
961 def VLD2DUPd16 : VLD2DUP<{0,1,0,?}, "16">;
962 def VLD2DUPd32 : VLD2DUP<{1,0,0,?}, "32">;
964 def VLD2DUPd8Pseudo : VLDQPseudo<IIC_VLD2dup>;
965 def VLD2DUPd16Pseudo : VLDQPseudo<IIC_VLD2dup>;
966 def VLD2DUPd32Pseudo : VLDQPseudo<IIC_VLD2dup>;
968 // ...with double-spaced registers (not used for codegen):
969 def VLD2DUPd8x2 : VLD2DUP<{0,0,1,?}, "8">;
970 def VLD2DUPd16x2 : VLD2DUP<{0,1,1,?}, "16">;
971 def VLD2DUPd32x2 : VLD2DUP<{1,0,1,?}, "32">;
973 // ...with address register writeback:
974 class VLD2DUPWB<bits<4> op7_4, string Dt>
975 : NLdSt<1, 0b10, 0b1101, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
976 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD2dupu,
977 "vld2", Dt, "\\{$Vd[], $dst2[]\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
979 let DecoderMethod = "DecodeVLD2DupInstruction";
982 def VLD2DUPd8_UPD : VLD2DUPWB<{0,0,0,0}, "8">;
983 def VLD2DUPd16_UPD : VLD2DUPWB<{0,1,0,?}, "16">;
984 def VLD2DUPd32_UPD : VLD2DUPWB<{1,0,0,?}, "32">;
986 def VLD2DUPd8x2_UPD : VLD2DUPWB<{0,0,1,0}, "8">;
987 def VLD2DUPd16x2_UPD : VLD2DUPWB<{0,1,1,?}, "16">;
988 def VLD2DUPd32x2_UPD : VLD2DUPWB<{1,0,1,?}, "32">;
990 def VLD2DUPd8Pseudo_UPD : VLDQWBPseudo<IIC_VLD2dupu>;
991 def VLD2DUPd16Pseudo_UPD : VLDQWBPseudo<IIC_VLD2dupu>;
992 def VLD2DUPd32Pseudo_UPD : VLDQWBPseudo<IIC_VLD2dupu>;
994 // VLD3DUP : Vector Load (single 3-element structure to all lanes)
995 class VLD3DUP<bits<4> op7_4, string Dt>
996 : NLdSt<1, 0b10, 0b1110, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
997 (ins addrmode6dup:$Rn), IIC_VLD3dup,
998 "vld3", Dt, "\\{$Vd[], $dst2[], $dst3[]\\}, $Rn", "", []> {
1001 let DecoderMethod = "DecodeVLD3DupInstruction";
1004 def VLD3DUPd8 : VLD3DUP<{0,0,0,?}, "8">;
1005 def VLD3DUPd16 : VLD3DUP<{0,1,0,?}, "16">;
1006 def VLD3DUPd32 : VLD3DUP<{1,0,0,?}, "32">;
1008 def VLD3DUPd8Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1009 def VLD3DUPd16Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1010 def VLD3DUPd32Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1012 // ...with double-spaced registers (not used for codegen):
1013 def VLD3DUPd8x2 : VLD3DUP<{0,0,1,?}, "8">;
1014 def VLD3DUPd16x2 : VLD3DUP<{0,1,1,?}, "16">;
1015 def VLD3DUPd32x2 : VLD3DUP<{1,0,1,?}, "32">;
1017 // ...with address register writeback:
1018 class VLD3DUPWB<bits<4> op7_4, string Dt>
1019 : NLdSt<1, 0b10, 0b1110, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
1020 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD3dupu,
1021 "vld3", Dt, "\\{$Vd[], $dst2[], $dst3[]\\}, $Rn$Rm",
1022 "$Rn.addr = $wb", []> {
1024 let DecoderMethod = "DecodeVLD3DupInstruction";
1027 def VLD3DUPd8_UPD : VLD3DUPWB<{0,0,0,0}, "8">;
1028 def VLD3DUPd16_UPD : VLD3DUPWB<{0,1,0,?}, "16">;
1029 def VLD3DUPd32_UPD : VLD3DUPWB<{1,0,0,?}, "32">;
1031 def VLD3DUPd8x2_UPD : VLD3DUPWB<{0,0,1,0}, "8">;
1032 def VLD3DUPd16x2_UPD : VLD3DUPWB<{0,1,1,?}, "16">;
1033 def VLD3DUPd32x2_UPD : VLD3DUPWB<{1,0,1,?}, "32">;
1035 def VLD3DUPd8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1036 def VLD3DUPd16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1037 def VLD3DUPd32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1039 // VLD4DUP : Vector Load (single 4-element structure to all lanes)
1040 class VLD4DUP<bits<4> op7_4, string Dt>
1041 : NLdSt<1, 0b10, 0b1111, op7_4,
1042 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
1043 (ins addrmode6dup:$Rn), IIC_VLD4dup,
1044 "vld4", Dt, "\\{$Vd[], $dst2[], $dst3[], $dst4[]\\}, $Rn", "", []> {
1046 let Inst{4} = Rn{4};
1047 let DecoderMethod = "DecodeVLD4DupInstruction";
1050 def VLD4DUPd8 : VLD4DUP<{0,0,0,?}, "8">;
1051 def VLD4DUPd16 : VLD4DUP<{0,1,0,?}, "16">;
1052 def VLD4DUPd32 : VLD4DUP<{1,?,0,?}, "32"> { let Inst{6} = Rn{5}; }
1054 def VLD4DUPd8Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1055 def VLD4DUPd16Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1056 def VLD4DUPd32Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1058 // ...with double-spaced registers (not used for codegen):
1059 def VLD4DUPd8x2 : VLD4DUP<{0,0,1,?}, "8">;
1060 def VLD4DUPd16x2 : VLD4DUP<{0,1,1,?}, "16">;
1061 def VLD4DUPd32x2 : VLD4DUP<{1,?,1,?}, "32"> { let Inst{6} = Rn{5}; }
1063 // ...with address register writeback:
1064 class VLD4DUPWB<bits<4> op7_4, string Dt>
1065 : NLdSt<1, 0b10, 0b1111, op7_4,
1066 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
1067 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD4dupu,
1068 "vld4", Dt, "\\{$Vd[], $dst2[], $dst3[], $dst4[]\\}, $Rn$Rm",
1069 "$Rn.addr = $wb", []> {
1070 let Inst{4} = Rn{4};
1071 let DecoderMethod = "DecodeVLD4DupInstruction";
1074 def VLD4DUPd8_UPD : VLD4DUPWB<{0,0,0,0}, "8">;
1075 def VLD4DUPd16_UPD : VLD4DUPWB<{0,1,0,?}, "16">;
1076 def VLD4DUPd32_UPD : VLD4DUPWB<{1,?,0,?}, "32"> { let Inst{6} = Rn{5}; }
1078 def VLD4DUPd8x2_UPD : VLD4DUPWB<{0,0,1,0}, "8">;
1079 def VLD4DUPd16x2_UPD : VLD4DUPWB<{0,1,1,?}, "16">;
1080 def VLD4DUPd32x2_UPD : VLD4DUPWB<{1,?,1,?}, "32"> { let Inst{6} = Rn{5}; }
1082 def VLD4DUPd8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1083 def VLD4DUPd16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1084 def VLD4DUPd32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1086 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
1088 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
1090 // Classes for VST* pseudo-instructions with multi-register operands.
1091 // These are expanded to real instructions after register allocation.
1092 class VSTQPseudo<InstrItinClass itin>
1093 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src), itin, "">;
1094 class VSTQWBPseudo<InstrItinClass itin>
1095 : PseudoNLdSt<(outs GPR:$wb),
1096 (ins addrmode6:$addr, am6offset:$offset, QPR:$src), itin,
1097 "$addr.addr = $wb">;
1098 class VSTQQPseudo<InstrItinClass itin>
1099 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src), itin, "">;
1100 class VSTQQWBPseudo<InstrItinClass itin>
1101 : PseudoNLdSt<(outs GPR:$wb),
1102 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src), itin,
1103 "$addr.addr = $wb">;
1104 class VSTQQQQPseudo<InstrItinClass itin>
1105 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src), itin, "">;
1106 class VSTQQQQWBPseudo<InstrItinClass itin>
1107 : PseudoNLdSt<(outs GPR:$wb),
1108 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
1109 "$addr.addr = $wb">;
1111 // VST1 : Vector Store (multiple single elements)
1112 class VST1D<bits<4> op7_4, string Dt>
1113 : NLdSt<0,0b00,0b0111,op7_4, (outs), (ins addrmode6:$Rn, DPR:$Vd),
1114 IIC_VST1, "vst1", Dt, "\\{$Vd\\}, $Rn", "", []> {
1116 let Inst{4} = Rn{4};
1117 let DecoderMethod = "DecodeVSTInstruction";
1119 class VST1Q<bits<4> op7_4, string Dt>
1120 : NLdSt<0,0b00,0b1010,op7_4, (outs),
1121 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2), IIC_VST1x2,
1122 "vst1", Dt, "\\{$Vd, $src2\\}, $Rn", "", []> {
1124 let Inst{5-4} = Rn{5-4};
1125 let DecoderMethod = "DecodeVSTInstruction";
1128 def VST1d8 : VST1D<{0,0,0,?}, "8">;
1129 def VST1d16 : VST1D<{0,1,0,?}, "16">;
1130 def VST1d32 : VST1D<{1,0,0,?}, "32">;
1131 def VST1d64 : VST1D<{1,1,0,?}, "64">;
1133 def VST1q8 : VST1Q<{0,0,?,?}, "8">;
1134 def VST1q16 : VST1Q<{0,1,?,?}, "16">;
1135 def VST1q32 : VST1Q<{1,0,?,?}, "32">;
1136 def VST1q64 : VST1Q<{1,1,?,?}, "64">;
1138 def VST1q8Pseudo : VSTQPseudo<IIC_VST1x2>;
1139 def VST1q16Pseudo : VSTQPseudo<IIC_VST1x2>;
1140 def VST1q32Pseudo : VSTQPseudo<IIC_VST1x2>;
1141 def VST1q64Pseudo : VSTQPseudo<IIC_VST1x2>;
1143 // ...with address register writeback:
1144 class VST1DWB<bits<4> op7_4, string Dt>
1145 : NLdSt<0, 0b00, 0b0111, op7_4, (outs GPR:$wb),
1146 (ins addrmode6:$Rn, am6offset:$Rm, DPR:$Vd), IIC_VST1u,
1147 "vst1", Dt, "\\{$Vd\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
1148 let Inst{4} = Rn{4};
1149 let DecoderMethod = "DecodeVSTInstruction";
1151 class VST1QWB<bits<4> op7_4, string Dt>
1152 : NLdSt<0, 0b00, 0b1010, op7_4, (outs GPR:$wb),
1153 (ins addrmode6:$Rn, am6offset:$Rm, DPR:$Vd, DPR:$src2),
1154 IIC_VST1x2u, "vst1", Dt, "\\{$Vd, $src2\\}, $Rn$Rm",
1155 "$Rn.addr = $wb", []> {
1156 let Inst{5-4} = Rn{5-4};
1157 let DecoderMethod = "DecodeVSTInstruction";
1160 def VST1d8_UPD : VST1DWB<{0,0,0,?}, "8">;
1161 def VST1d16_UPD : VST1DWB<{0,1,0,?}, "16">;
1162 def VST1d32_UPD : VST1DWB<{1,0,0,?}, "32">;
1163 def VST1d64_UPD : VST1DWB<{1,1,0,?}, "64">;
1165 def VST1q8_UPD : VST1QWB<{0,0,?,?}, "8">;
1166 def VST1q16_UPD : VST1QWB<{0,1,?,?}, "16">;
1167 def VST1q32_UPD : VST1QWB<{1,0,?,?}, "32">;
1168 def VST1q64_UPD : VST1QWB<{1,1,?,?}, "64">;
1170 def VST1q8Pseudo_UPD : VSTQWBPseudo<IIC_VST1x2u>;
1171 def VST1q16Pseudo_UPD : VSTQWBPseudo<IIC_VST1x2u>;
1172 def VST1q32Pseudo_UPD : VSTQWBPseudo<IIC_VST1x2u>;
1173 def VST1q64Pseudo_UPD : VSTQWBPseudo<IIC_VST1x2u>;
1175 // ...with 3 registers (some of these are only for the disassembler):
1176 class VST1D3<bits<4> op7_4, string Dt>
1177 : NLdSt<0, 0b00, 0b0110, op7_4, (outs),
1178 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3),
1179 IIC_VST1x3, "vst1", Dt, "\\{$Vd, $src2, $src3\\}, $Rn", "", []> {
1181 let Inst{4} = Rn{4};
1182 let DecoderMethod = "DecodeVSTInstruction";
1184 class VST1D3WB<bits<4> op7_4, string Dt>
1185 : NLdSt<0, 0b00, 0b0110, op7_4, (outs GPR:$wb),
1186 (ins addrmode6:$Rn, am6offset:$Rm,
1187 DPR:$Vd, DPR:$src2, DPR:$src3),
1188 IIC_VST1x3u, "vst1", Dt, "\\{$Vd, $src2, $src3\\}, $Rn$Rm",
1189 "$Rn.addr = $wb", []> {
1190 let Inst{4} = Rn{4};
1191 let DecoderMethod = "DecodeVSTInstruction";
1194 def VST1d8T : VST1D3<{0,0,0,?}, "8">;
1195 def VST1d16T : VST1D3<{0,1,0,?}, "16">;
1196 def VST1d32T : VST1D3<{1,0,0,?}, "32">;
1197 def VST1d64T : VST1D3<{1,1,0,?}, "64">;
1199 def VST1d8T_UPD : VST1D3WB<{0,0,0,?}, "8">;
1200 def VST1d16T_UPD : VST1D3WB<{0,1,0,?}, "16">;
1201 def VST1d32T_UPD : VST1D3WB<{1,0,0,?}, "32">;
1202 def VST1d64T_UPD : VST1D3WB<{1,1,0,?}, "64">;
1204 def VST1d64TPseudo : VSTQQPseudo<IIC_VST1x3>;
1205 def VST1d64TPseudo_UPD : VSTQQWBPseudo<IIC_VST1x3u>;
1207 // ...with 4 registers (some of these are only for the disassembler):
1208 class VST1D4<bits<4> op7_4, string Dt>
1209 : NLdSt<0, 0b00, 0b0010, op7_4, (outs),
1210 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
1211 IIC_VST1x4, "vst1", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn", "",
1214 let Inst{5-4} = Rn{5-4};
1215 let DecoderMethod = "DecodeVSTInstruction";
1217 class VST1D4WB<bits<4> op7_4, string Dt>
1218 : NLdSt<0, 0b00, 0b0010, op7_4, (outs GPR:$wb),
1219 (ins addrmode6:$Rn, am6offset:$Rm,
1220 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST1x4u,
1221 "vst1", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
1222 "$Rn.addr = $wb", []> {
1223 let Inst{5-4} = Rn{5-4};
1224 let DecoderMethod = "DecodeVSTInstruction";
1227 def VST1d8Q : VST1D4<{0,0,?,?}, "8">;
1228 def VST1d16Q : VST1D4<{0,1,?,?}, "16">;
1229 def VST1d32Q : VST1D4<{1,0,?,?}, "32">;
1230 def VST1d64Q : VST1D4<{1,1,?,?}, "64">;
1232 def VST1d8Q_UPD : VST1D4WB<{0,0,?,?}, "8">;
1233 def VST1d16Q_UPD : VST1D4WB<{0,1,?,?}, "16">;
1234 def VST1d32Q_UPD : VST1D4WB<{1,0,?,?}, "32">;
1235 def VST1d64Q_UPD : VST1D4WB<{1,1,?,?}, "64">;
1237 def VST1d64QPseudo : VSTQQPseudo<IIC_VST1x4>;
1238 def VST1d64QPseudo_UPD : VSTQQWBPseudo<IIC_VST1x4u>;
1240 // VST2 : Vector Store (multiple 2-element structures)
1241 class VST2D<bits<4> op11_8, bits<4> op7_4, string Dt>
1242 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1243 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2),
1244 IIC_VST2, "vst2", Dt, "\\{$Vd, $src2\\}, $Rn", "", []> {
1246 let Inst{5-4} = Rn{5-4};
1247 let DecoderMethod = "DecodeVSTInstruction";
1249 class VST2Q<bits<4> op7_4, string Dt>
1250 : NLdSt<0, 0b00, 0b0011, op7_4, (outs),
1251 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
1252 IIC_VST2x2, "vst2", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn",
1255 let Inst{5-4} = Rn{5-4};
1256 let DecoderMethod = "DecodeVSTInstruction";
1259 def VST2d8 : VST2D<0b1000, {0,0,?,?}, "8">;
1260 def VST2d16 : VST2D<0b1000, {0,1,?,?}, "16">;
1261 def VST2d32 : VST2D<0b1000, {1,0,?,?}, "32">;
1263 def VST2q8 : VST2Q<{0,0,?,?}, "8">;
1264 def VST2q16 : VST2Q<{0,1,?,?}, "16">;
1265 def VST2q32 : VST2Q<{1,0,?,?}, "32">;
1267 def VST2d8Pseudo : VSTQPseudo<IIC_VST2>;
1268 def VST2d16Pseudo : VSTQPseudo<IIC_VST2>;
1269 def VST2d32Pseudo : VSTQPseudo<IIC_VST2>;
1271 def VST2q8Pseudo : VSTQQPseudo<IIC_VST2x2>;
1272 def VST2q16Pseudo : VSTQQPseudo<IIC_VST2x2>;
1273 def VST2q32Pseudo : VSTQQPseudo<IIC_VST2x2>;
1275 // ...with address register writeback:
1276 class VST2DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1277 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1278 (ins addrmode6:$Rn, am6offset:$Rm, DPR:$Vd, DPR:$src2),
1279 IIC_VST2u, "vst2", Dt, "\\{$Vd, $src2\\}, $Rn$Rm",
1280 "$Rn.addr = $wb", []> {
1281 let Inst{5-4} = Rn{5-4};
1282 let DecoderMethod = "DecodeVSTInstruction";
1284 class VST2QWB<bits<4> op7_4, string Dt>
1285 : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
1286 (ins addrmode6:$Rn, am6offset:$Rm,
1287 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST2x2u,
1288 "vst2", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
1289 "$Rn.addr = $wb", []> {
1290 let Inst{5-4} = Rn{5-4};
1291 let DecoderMethod = "DecodeVSTInstruction";
1294 def VST2d8_UPD : VST2DWB<0b1000, {0,0,?,?}, "8">;
1295 def VST2d16_UPD : VST2DWB<0b1000, {0,1,?,?}, "16">;
1296 def VST2d32_UPD : VST2DWB<0b1000, {1,0,?,?}, "32">;
1298 def VST2q8_UPD : VST2QWB<{0,0,?,?}, "8">;
1299 def VST2q16_UPD : VST2QWB<{0,1,?,?}, "16">;
1300 def VST2q32_UPD : VST2QWB<{1,0,?,?}, "32">;
1302 def VST2d8Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
1303 def VST2d16Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
1304 def VST2d32Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
1306 def VST2q8Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
1307 def VST2q16Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
1308 def VST2q32Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
1310 // ...with double-spaced registers (for disassembly only):
1311 def VST2b8 : VST2D<0b1001, {0,0,?,?}, "8">;
1312 def VST2b16 : VST2D<0b1001, {0,1,?,?}, "16">;
1313 def VST2b32 : VST2D<0b1001, {1,0,?,?}, "32">;
1314 def VST2b8_UPD : VST2DWB<0b1001, {0,0,?,?}, "8">;
1315 def VST2b16_UPD : VST2DWB<0b1001, {0,1,?,?}, "16">;
1316 def VST2b32_UPD : VST2DWB<0b1001, {1,0,?,?}, "32">;
1318 // VST3 : Vector Store (multiple 3-element structures)
1319 class VST3D<bits<4> op11_8, bits<4> op7_4, string Dt>
1320 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1321 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3,
1322 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn", "", []> {
1324 let Inst{4} = Rn{4};
1325 let DecoderMethod = "DecodeVSTInstruction";
1328 def VST3d8 : VST3D<0b0100, {0,0,0,?}, "8">;
1329 def VST3d16 : VST3D<0b0100, {0,1,0,?}, "16">;
1330 def VST3d32 : VST3D<0b0100, {1,0,0,?}, "32">;
1332 def VST3d8Pseudo : VSTQQPseudo<IIC_VST3>;
1333 def VST3d16Pseudo : VSTQQPseudo<IIC_VST3>;
1334 def VST3d32Pseudo : VSTQQPseudo<IIC_VST3>;
1336 // ...with address register writeback:
1337 class VST3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1338 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1339 (ins addrmode6:$Rn, am6offset:$Rm,
1340 DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3u,
1341 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn$Rm",
1342 "$Rn.addr = $wb", []> {
1343 let Inst{4} = Rn{4};
1344 let DecoderMethod = "DecodeVSTInstruction";
1347 def VST3d8_UPD : VST3DWB<0b0100, {0,0,0,?}, "8">;
1348 def VST3d16_UPD : VST3DWB<0b0100, {0,1,0,?}, "16">;
1349 def VST3d32_UPD : VST3DWB<0b0100, {1,0,0,?}, "32">;
1351 def VST3d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1352 def VST3d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1353 def VST3d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1355 // ...with double-spaced registers:
1356 def VST3q8 : VST3D<0b0101, {0,0,0,?}, "8">;
1357 def VST3q16 : VST3D<0b0101, {0,1,0,?}, "16">;
1358 def VST3q32 : VST3D<0b0101, {1,0,0,?}, "32">;
1359 def VST3q8_UPD : VST3DWB<0b0101, {0,0,0,?}, "8">;
1360 def VST3q16_UPD : VST3DWB<0b0101, {0,1,0,?}, "16">;
1361 def VST3q32_UPD : VST3DWB<0b0101, {1,0,0,?}, "32">;
1363 def VST3q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1364 def VST3q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1365 def VST3q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1367 // ...alternate versions to be allocated odd register numbers:
1368 def VST3q8oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1369 def VST3q16oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1370 def VST3q32oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1372 def VST3q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1373 def VST3q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1374 def VST3q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1376 // VST4 : Vector Store (multiple 4-element structures)
1377 class VST4D<bits<4> op11_8, bits<4> op7_4, string Dt>
1378 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1379 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
1380 IIC_VST4, "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn",
1383 let Inst{5-4} = Rn{5-4};
1384 let DecoderMethod = "DecodeVSTInstruction";
1387 def VST4d8 : VST4D<0b0000, {0,0,?,?}, "8">;
1388 def VST4d16 : VST4D<0b0000, {0,1,?,?}, "16">;
1389 def VST4d32 : VST4D<0b0000, {1,0,?,?}, "32">;
1391 def VST4d8Pseudo : VSTQQPseudo<IIC_VST4>;
1392 def VST4d16Pseudo : VSTQQPseudo<IIC_VST4>;
1393 def VST4d32Pseudo : VSTQQPseudo<IIC_VST4>;
1395 // ...with address register writeback:
1396 class VST4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1397 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1398 (ins addrmode6:$Rn, am6offset:$Rm,
1399 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST4u,
1400 "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
1401 "$Rn.addr = $wb", []> {
1402 let Inst{5-4} = Rn{5-4};
1403 let DecoderMethod = "DecodeVSTInstruction";
1406 def VST4d8_UPD : VST4DWB<0b0000, {0,0,?,?}, "8">;
1407 def VST4d16_UPD : VST4DWB<0b0000, {0,1,?,?}, "16">;
1408 def VST4d32_UPD : VST4DWB<0b0000, {1,0,?,?}, "32">;
1410 def VST4d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1411 def VST4d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1412 def VST4d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1414 // ...with double-spaced registers:
1415 def VST4q8 : VST4D<0b0001, {0,0,?,?}, "8">;
1416 def VST4q16 : VST4D<0b0001, {0,1,?,?}, "16">;
1417 def VST4q32 : VST4D<0b0001, {1,0,?,?}, "32">;
1418 def VST4q8_UPD : VST4DWB<0b0001, {0,0,?,?}, "8">;
1419 def VST4q16_UPD : VST4DWB<0b0001, {0,1,?,?}, "16">;
1420 def VST4q32_UPD : VST4DWB<0b0001, {1,0,?,?}, "32">;
1422 def VST4q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1423 def VST4q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1424 def VST4q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1426 // ...alternate versions to be allocated odd register numbers:
1427 def VST4q8oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1428 def VST4q16oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1429 def VST4q32oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1431 def VST4q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1432 def VST4q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1433 def VST4q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1435 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
1437 // Classes for VST*LN pseudo-instructions with multi-register operands.
1438 // These are expanded to real instructions after register allocation.
1439 class VSTQLNPseudo<InstrItinClass itin>
1440 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
1442 class VSTQLNWBPseudo<InstrItinClass itin>
1443 : PseudoNLdSt<(outs GPR:$wb),
1444 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
1445 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1446 class VSTQQLNPseudo<InstrItinClass itin>
1447 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
1449 class VSTQQLNWBPseudo<InstrItinClass itin>
1450 : PseudoNLdSt<(outs GPR:$wb),
1451 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
1452 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1453 class VSTQQQQLNPseudo<InstrItinClass itin>
1454 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
1456 class VSTQQQQLNWBPseudo<InstrItinClass itin>
1457 : PseudoNLdSt<(outs GPR:$wb),
1458 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
1459 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1461 // VST1LN : Vector Store (single element from one lane)
1462 class VST1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1463 PatFrag StoreOp, SDNode ExtractOp>
1464 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1465 (ins addrmode6:$Rn, DPR:$Vd, nohash_imm:$lane),
1466 IIC_VST1ln, "vst1", Dt, "\\{$Vd[$lane]\\}, $Rn", "",
1467 [(StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane), addrmode6:$Rn)]> {
1469 let DecoderMethod = "DecodeVST1LN";
1471 class VST1LN32<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1472 PatFrag StoreOp, SDNode ExtractOp>
1473 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1474 (ins addrmode6oneL32:$Rn, DPR:$Vd, nohash_imm:$lane),
1475 IIC_VST1ln, "vst1", Dt, "\\{$Vd[$lane]\\}, $Rn", "",
1476 [(StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane), addrmode6oneL32:$Rn)]>{
1478 let DecoderMethod = "DecodeVST1LN";
1480 class VST1QLNPseudo<ValueType Ty, PatFrag StoreOp, SDNode ExtractOp>
1481 : VSTQLNPseudo<IIC_VST1ln> {
1482 let Pattern = [(StoreOp (ExtractOp (Ty QPR:$src), imm:$lane),
1486 def VST1LNd8 : VST1LN<0b0000, {?,?,?,0}, "8", v8i8, truncstorei8,
1488 let Inst{7-5} = lane{2-0};
1490 def VST1LNd16 : VST1LN<0b0100, {?,?,0,?}, "16", v4i16, truncstorei16,
1492 let Inst{7-6} = lane{1-0};
1493 let Inst{4} = Rn{5};
1496 def VST1LNd32 : VST1LN32<0b1000, {?,0,?,?}, "32", v2i32, store, extractelt> {
1497 let Inst{7} = lane{0};
1498 let Inst{5-4} = Rn{5-4};
1501 def VST1LNq8Pseudo : VST1QLNPseudo<v16i8, truncstorei8, NEONvgetlaneu>;
1502 def VST1LNq16Pseudo : VST1QLNPseudo<v8i16, truncstorei16, NEONvgetlaneu>;
1503 def VST1LNq32Pseudo : VST1QLNPseudo<v4i32, store, extractelt>;
1505 def : Pat<(store (extractelt (v2f32 DPR:$src), imm:$lane), addrmode6:$addr),
1506 (VST1LNd32 addrmode6:$addr, DPR:$src, imm:$lane)>;
1507 def : Pat<(store (extractelt (v4f32 QPR:$src), imm:$lane), addrmode6:$addr),
1508 (VST1LNq32Pseudo addrmode6:$addr, QPR:$src, imm:$lane)>;
1510 // ...with address register writeback:
1511 class VST1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1512 PatFrag StoreOp, SDNode ExtractOp>
1513 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1514 (ins addrmode6:$Rn, am6offset:$Rm,
1515 DPR:$Vd, nohash_imm:$lane), IIC_VST1lnu, "vst1", Dt,
1516 "\\{$Vd[$lane]\\}, $Rn$Rm",
1518 [(set GPR:$wb, (StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane),
1519 addrmode6:$Rn, am6offset:$Rm))]> {
1520 let DecoderMethod = "DecodeVST1LN";
1522 class VST1QLNWBPseudo<ValueType Ty, PatFrag StoreOp, SDNode ExtractOp>
1523 : VSTQLNWBPseudo<IIC_VST1lnu> {
1524 let Pattern = [(set GPR:$wb, (StoreOp (ExtractOp (Ty QPR:$src), imm:$lane),
1525 addrmode6:$addr, am6offset:$offset))];
1528 def VST1LNd8_UPD : VST1LNWB<0b0000, {?,?,?,0}, "8", v8i8, post_truncsti8,
1530 let Inst{7-5} = lane{2-0};
1532 def VST1LNd16_UPD : VST1LNWB<0b0100, {?,?,0,?}, "16", v4i16, post_truncsti16,
1534 let Inst{7-6} = lane{1-0};
1535 let Inst{4} = Rn{5};
1537 def VST1LNd32_UPD : VST1LNWB<0b1000, {?,0,?,?}, "32", v2i32, post_store,
1539 let Inst{7} = lane{0};
1540 let Inst{5-4} = Rn{5-4};
1543 def VST1LNq8Pseudo_UPD : VST1QLNWBPseudo<v16i8, post_truncsti8, NEONvgetlaneu>;
1544 def VST1LNq16Pseudo_UPD : VST1QLNWBPseudo<v8i16, post_truncsti16,NEONvgetlaneu>;
1545 def VST1LNq32Pseudo_UPD : VST1QLNWBPseudo<v4i32, post_store, extractelt>;
1547 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
1549 // VST2LN : Vector Store (single 2-element structure from one lane)
1550 class VST2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1551 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1552 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, nohash_imm:$lane),
1553 IIC_VST2ln, "vst2", Dt, "\\{$Vd[$lane], $src2[$lane]\\}, $Rn",
1556 let Inst{4} = Rn{4};
1557 let DecoderMethod = "DecodeVST2LN";
1560 def VST2LNd8 : VST2LN<0b0001, {?,?,?,?}, "8"> {
1561 let Inst{7-5} = lane{2-0};
1563 def VST2LNd16 : VST2LN<0b0101, {?,?,0,?}, "16"> {
1564 let Inst{7-6} = lane{1-0};
1566 def VST2LNd32 : VST2LN<0b1001, {?,0,0,?}, "32"> {
1567 let Inst{7} = lane{0};
1570 def VST2LNd8Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1571 def VST2LNd16Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1572 def VST2LNd32Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1574 // ...with double-spaced registers:
1575 def VST2LNq16 : VST2LN<0b0101, {?,?,1,?}, "16"> {
1576 let Inst{7-6} = lane{1-0};
1577 let Inst{4} = Rn{4};
1579 def VST2LNq32 : VST2LN<0b1001, {?,1,0,?}, "32"> {
1580 let Inst{7} = lane{0};
1581 let Inst{4} = Rn{4};
1584 def VST2LNq16Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
1585 def VST2LNq32Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
1587 // ...with address register writeback:
1588 class VST2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1589 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1590 (ins addrmode6:$addr, am6offset:$offset,
1591 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VST2lnu, "vst2", Dt,
1592 "\\{$src1[$lane], $src2[$lane]\\}, $addr$offset",
1593 "$addr.addr = $wb", []> {
1594 let Inst{4} = Rn{4};
1595 let DecoderMethod = "DecodeVST2LN";
1598 def VST2LNd8_UPD : VST2LNWB<0b0001, {?,?,?,?}, "8"> {
1599 let Inst{7-5} = lane{2-0};
1601 def VST2LNd16_UPD : VST2LNWB<0b0101, {?,?,0,?}, "16"> {
1602 let Inst{7-6} = lane{1-0};
1604 def VST2LNd32_UPD : VST2LNWB<0b1001, {?,0,0,?}, "32"> {
1605 let Inst{7} = lane{0};
1608 def VST2LNd8Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1609 def VST2LNd16Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1610 def VST2LNd32Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1612 def VST2LNq16_UPD : VST2LNWB<0b0101, {?,?,1,?}, "16"> {
1613 let Inst{7-6} = lane{1-0};
1615 def VST2LNq32_UPD : VST2LNWB<0b1001, {?,1,0,?}, "32"> {
1616 let Inst{7} = lane{0};
1619 def VST2LNq16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
1620 def VST2LNq32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
1622 // VST3LN : Vector Store (single 3-element structure from one lane)
1623 class VST3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1624 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1625 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3,
1626 nohash_imm:$lane), IIC_VST3ln, "vst3", Dt,
1627 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn", "", []> {
1629 let DecoderMethod = "DecodeVST3LN";
1632 def VST3LNd8 : VST3LN<0b0010, {?,?,?,0}, "8"> {
1633 let Inst{7-5} = lane{2-0};
1635 def VST3LNd16 : VST3LN<0b0110, {?,?,0,0}, "16"> {
1636 let Inst{7-6} = lane{1-0};
1638 def VST3LNd32 : VST3LN<0b1010, {?,0,0,0}, "32"> {
1639 let Inst{7} = lane{0};
1642 def VST3LNd8Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1643 def VST3LNd16Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1644 def VST3LNd32Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1646 // ...with double-spaced registers:
1647 def VST3LNq16 : VST3LN<0b0110, {?,?,1,0}, "16"> {
1648 let Inst{7-6} = lane{1-0};
1650 def VST3LNq32 : VST3LN<0b1010, {?,1,0,0}, "32"> {
1651 let Inst{7} = lane{0};
1654 def VST3LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
1655 def VST3LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
1657 // ...with address register writeback:
1658 class VST3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1659 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1660 (ins addrmode6:$Rn, am6offset:$Rm,
1661 DPR:$Vd, DPR:$src2, DPR:$src3, nohash_imm:$lane),
1662 IIC_VST3lnu, "vst3", Dt,
1663 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn$Rm",
1664 "$Rn.addr = $wb", []> {
1665 let DecoderMethod = "DecodeVST3LN";
1668 def VST3LNd8_UPD : VST3LNWB<0b0010, {?,?,?,0}, "8"> {
1669 let Inst{7-5} = lane{2-0};
1671 def VST3LNd16_UPD : VST3LNWB<0b0110, {?,?,0,0}, "16"> {
1672 let Inst{7-6} = lane{1-0};
1674 def VST3LNd32_UPD : VST3LNWB<0b1010, {?,0,0,0}, "32"> {
1675 let Inst{7} = lane{0};
1678 def VST3LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1679 def VST3LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1680 def VST3LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1682 def VST3LNq16_UPD : VST3LNWB<0b0110, {?,?,1,0}, "16"> {
1683 let Inst{7-6} = lane{1-0};
1685 def VST3LNq32_UPD : VST3LNWB<0b1010, {?,1,0,0}, "32"> {
1686 let Inst{7} = lane{0};
1689 def VST3LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
1690 def VST3LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
1692 // VST4LN : Vector Store (single 4-element structure from one lane)
1693 class VST4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1694 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1695 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4,
1696 nohash_imm:$lane), IIC_VST4ln, "vst4", Dt,
1697 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn",
1700 let Inst{4} = Rn{4};
1701 let DecoderMethod = "DecodeVST4LN";
1704 def VST4LNd8 : VST4LN<0b0011, {?,?,?,?}, "8"> {
1705 let Inst{7-5} = lane{2-0};
1707 def VST4LNd16 : VST4LN<0b0111, {?,?,0,?}, "16"> {
1708 let Inst{7-6} = lane{1-0};
1710 def VST4LNd32 : VST4LN<0b1011, {?,0,?,?}, "32"> {
1711 let Inst{7} = lane{0};
1712 let Inst{5} = Rn{5};
1715 def VST4LNd8Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1716 def VST4LNd16Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1717 def VST4LNd32Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1719 // ...with double-spaced registers:
1720 def VST4LNq16 : VST4LN<0b0111, {?,?,1,?}, "16"> {
1721 let Inst{7-6} = lane{1-0};
1723 def VST4LNq32 : VST4LN<0b1011, {?,1,?,?}, "32"> {
1724 let Inst{7} = lane{0};
1725 let Inst{5} = Rn{5};
1728 def VST4LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
1729 def VST4LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
1731 // ...with address register writeback:
1732 class VST4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1733 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1734 (ins addrmode6:$Rn, am6offset:$Rm,
1735 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
1736 IIC_VST4lnu, "vst4", Dt,
1737 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn$Rm",
1738 "$Rn.addr = $wb", []> {
1739 let Inst{4} = Rn{4};
1740 let DecoderMethod = "DecodeVST4LN";
1743 def VST4LNd8_UPD : VST4LNWB<0b0011, {?,?,?,?}, "8"> {
1744 let Inst{7-5} = lane{2-0};
1746 def VST4LNd16_UPD : VST4LNWB<0b0111, {?,?,0,?}, "16"> {
1747 let Inst{7-6} = lane{1-0};
1749 def VST4LNd32_UPD : VST4LNWB<0b1011, {?,0,?,?}, "32"> {
1750 let Inst{7} = lane{0};
1751 let Inst{5} = Rn{5};
1754 def VST4LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
1755 def VST4LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
1756 def VST4LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
1758 def VST4LNq16_UPD : VST4LNWB<0b0111, {?,?,1,?}, "16"> {
1759 let Inst{7-6} = lane{1-0};
1761 def VST4LNq32_UPD : VST4LNWB<0b1011, {?,1,?,?}, "32"> {
1762 let Inst{7} = lane{0};
1763 let Inst{5} = Rn{5};
1766 def VST4LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
1767 def VST4LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
1769 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
1772 //===----------------------------------------------------------------------===//
1773 // NEON pattern fragments
1774 //===----------------------------------------------------------------------===//
1776 // Extract D sub-registers of Q registers.
1777 def DSubReg_i8_reg : SDNodeXForm<imm, [{
1778 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1779 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/8, MVT::i32);
1781 def DSubReg_i16_reg : SDNodeXForm<imm, [{
1782 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1783 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/4, MVT::i32);
1785 def DSubReg_i32_reg : SDNodeXForm<imm, [{
1786 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1787 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/2, MVT::i32);
1789 def DSubReg_f64_reg : SDNodeXForm<imm, [{
1790 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1791 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue(), MVT::i32);
1794 // Extract S sub-registers of Q/D registers.
1795 def SSubReg_f32_reg : SDNodeXForm<imm, [{
1796 assert(ARM::ssub_3 == ARM::ssub_0+3 && "Unexpected subreg numbering");
1797 return CurDAG->getTargetConstant(ARM::ssub_0 + N->getZExtValue(), MVT::i32);
1800 // Translate lane numbers from Q registers to D subregs.
1801 def SubReg_i8_lane : SDNodeXForm<imm, [{
1802 return CurDAG->getTargetConstant(N->getZExtValue() & 7, MVT::i32);
1804 def SubReg_i16_lane : SDNodeXForm<imm, [{
1805 return CurDAG->getTargetConstant(N->getZExtValue() & 3, MVT::i32);
1807 def SubReg_i32_lane : SDNodeXForm<imm, [{
1808 return CurDAG->getTargetConstant(N->getZExtValue() & 1, MVT::i32);
1811 //===----------------------------------------------------------------------===//
1812 // Instruction Classes
1813 //===----------------------------------------------------------------------===//
1815 // Basic 2-register operations: double- and quad-register.
1816 class N2VD<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1817 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
1818 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
1819 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
1820 (ins DPR:$Vm), IIC_VUNAD, OpcodeStr, Dt,"$Vd, $Vm", "",
1821 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vm))))]>;
1822 class N2VQ<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1823 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
1824 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
1825 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
1826 (ins QPR:$Vm), IIC_VUNAQ, OpcodeStr, Dt,"$Vd, $Vm", "",
1827 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vm))))]>;
1829 // Basic 2-register intrinsics, both double- and quad-register.
1830 class N2VDInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1831 bits<2> op17_16, bits<5> op11_7, bit op4,
1832 InstrItinClass itin, string OpcodeStr, string Dt,
1833 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1834 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
1835 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
1836 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm))))]>;
1837 class N2VQInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1838 bits<2> op17_16, bits<5> op11_7, bit op4,
1839 InstrItinClass itin, string OpcodeStr, string Dt,
1840 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1841 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
1842 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
1843 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm))))]>;
1845 // Narrow 2-register operations.
1846 class N2VN<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1847 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1848 InstrItinClass itin, string OpcodeStr, string Dt,
1849 ValueType TyD, ValueType TyQ, SDNode OpNode>
1850 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$Vd),
1851 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
1852 [(set DPR:$Vd, (TyD (OpNode (TyQ QPR:$Vm))))]>;
1854 // Narrow 2-register intrinsics.
1855 class N2VNInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1856 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1857 InstrItinClass itin, string OpcodeStr, string Dt,
1858 ValueType TyD, ValueType TyQ, Intrinsic IntOp>
1859 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$Vd),
1860 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
1861 [(set DPR:$Vd, (TyD (IntOp (TyQ QPR:$Vm))))]>;
1863 // Long 2-register operations (currently only used for VMOVL).
1864 class N2VL<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1865 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1866 InstrItinClass itin, string OpcodeStr, string Dt,
1867 ValueType TyQ, ValueType TyD, SDNode OpNode>
1868 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$Vd),
1869 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
1870 [(set QPR:$Vd, (TyQ (OpNode (TyD DPR:$Vm))))]>;
1872 // Long 2-register intrinsics.
1873 class N2VLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1874 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1875 InstrItinClass itin, string OpcodeStr, string Dt,
1876 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
1877 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$Vd),
1878 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
1879 [(set QPR:$Vd, (TyQ (IntOp (TyD DPR:$Vm))))]>;
1881 // 2-register shuffles (VTRN/VZIP/VUZP), both double- and quad-register.
1882 class N2VDShuffle<bits<2> op19_18, bits<5> op11_7, string OpcodeStr, string Dt>
1883 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 0, 0, (outs DPR:$Vd, DPR:$Vm),
1884 (ins DPR:$src1, DPR:$src2), IIC_VPERMD,
1885 OpcodeStr, Dt, "$Vd, $Vm",
1886 "$src1 = $Vd, $src2 = $Vm", []>;
1887 class N2VQShuffle<bits<2> op19_18, bits<5> op11_7,
1888 InstrItinClass itin, string OpcodeStr, string Dt>
1889 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 1, 0, (outs QPR:$Vd, QPR:$Vm),
1890 (ins QPR:$src1, QPR:$src2), itin, OpcodeStr, Dt, "$Vd, $Vm",
1891 "$src1 = $Vd, $src2 = $Vm", []>;
1893 // Basic 3-register operations: double- and quad-register.
1894 class N3VD<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1895 InstrItinClass itin, string OpcodeStr, string Dt,
1896 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
1897 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1898 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1899 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
1900 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
1901 let isCommutable = Commutable;
1903 // Same as N3VD but no data type.
1904 class N3VDX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1905 InstrItinClass itin, string OpcodeStr,
1906 ValueType ResTy, ValueType OpTy,
1907 SDNode OpNode, bit Commutable>
1908 : N3VX<op24, op23, op21_20, op11_8, 0, op4,
1909 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1910 OpcodeStr, "$Vd, $Vn, $Vm", "",
1911 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>{
1912 let isCommutable = Commutable;
1915 class N3VDSL<bits<2> op21_20, bits<4> op11_8,
1916 InstrItinClass itin, string OpcodeStr, string Dt,
1917 ValueType Ty, SDNode ShOp>
1918 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
1919 (outs DPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, nohash_imm:$lane),
1920 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "",
1922 (Ty (ShOp (Ty DPR:$Vn),
1923 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),imm:$lane)))))]> {
1924 let isCommutable = 0;
1926 class N3VDSL16<bits<2> op21_20, bits<4> op11_8,
1927 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1928 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
1929 (outs DPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, nohash_imm:$lane),
1930 NVMulSLFrm, IIC_VMULi16D, OpcodeStr, Dt,"$Vd, $Vn, $Vm[$lane]","",
1932 (Ty (ShOp (Ty DPR:$Vn),
1933 (Ty (NEONvduplane (Ty DPR_8:$Vm), imm:$lane)))))]> {
1934 let isCommutable = 0;
1937 class N3VQ<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1938 InstrItinClass itin, string OpcodeStr, string Dt,
1939 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
1940 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1941 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
1942 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
1943 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
1944 let isCommutable = Commutable;
1946 class N3VQX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1947 InstrItinClass itin, string OpcodeStr,
1948 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
1949 : N3VX<op24, op23, op21_20, op11_8, 1, op4,
1950 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
1951 OpcodeStr, "$Vd, $Vn, $Vm", "",
1952 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]>{
1953 let isCommutable = Commutable;
1955 class N3VQSL<bits<2> op21_20, bits<4> op11_8,
1956 InstrItinClass itin, string OpcodeStr, string Dt,
1957 ValueType ResTy, ValueType OpTy, SDNode ShOp>
1958 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
1959 (outs QPR:$Vd), (ins QPR:$Vn, DPR_VFP2:$Vm, nohash_imm:$lane),
1960 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "",
1961 [(set (ResTy QPR:$Vd),
1962 (ResTy (ShOp (ResTy QPR:$Vn),
1963 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
1965 let isCommutable = 0;
1967 class N3VQSL16<bits<2> op21_20, bits<4> op11_8, string OpcodeStr, string Dt,
1968 ValueType ResTy, ValueType OpTy, SDNode ShOp>
1969 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
1970 (outs QPR:$Vd), (ins QPR:$Vn, DPR_8:$Vm, nohash_imm:$lane),
1971 NVMulSLFrm, IIC_VMULi16Q, OpcodeStr, Dt,"$Vd, $Vn, $Vm[$lane]","",
1972 [(set (ResTy QPR:$Vd),
1973 (ResTy (ShOp (ResTy QPR:$Vn),
1974 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
1976 let isCommutable = 0;
1979 // Basic 3-register intrinsics, both double- and quad-register.
1980 class N3VDInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1981 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1982 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
1983 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1984 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), f, itin,
1985 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
1986 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
1987 let isCommutable = Commutable;
1989 class N3VDIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1990 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
1991 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
1992 (outs DPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, nohash_imm:$lane),
1993 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "",
1995 (Ty (IntOp (Ty DPR:$Vn),
1996 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),
1998 let isCommutable = 0;
2000 class N3VDIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2001 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
2002 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2003 (outs DPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, nohash_imm:$lane),
2004 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "",
2006 (Ty (IntOp (Ty DPR:$Vn),
2007 (Ty (NEONvduplane (Ty DPR_8:$Vm), imm:$lane)))))]> {
2008 let isCommutable = 0;
2010 class N3VDIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2011 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2012 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2013 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2014 (outs DPR:$Vd), (ins DPR:$Vm, DPR:$Vn), f, itin,
2015 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
2016 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (OpTy DPR:$Vn))))]> {
2017 let isCommutable = 0;
2020 class N3VQInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2021 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2022 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
2023 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2024 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), f, itin,
2025 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2026 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
2027 let isCommutable = Commutable;
2029 class N3VQIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2030 string OpcodeStr, string Dt,
2031 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2032 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2033 (outs QPR:$Vd), (ins QPR:$Vn, DPR_VFP2:$Vm, nohash_imm:$lane),
2034 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "",
2035 [(set (ResTy QPR:$Vd),
2036 (ResTy (IntOp (ResTy QPR:$Vn),
2037 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2039 let isCommutable = 0;
2041 class N3VQIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2042 string OpcodeStr, string Dt,
2043 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2044 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2045 (outs QPR:$Vd), (ins QPR:$Vn, DPR_8:$Vm, nohash_imm:$lane),
2046 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "",
2047 [(set (ResTy QPR:$Vd),
2048 (ResTy (IntOp (ResTy QPR:$Vn),
2049 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2051 let isCommutable = 0;
2053 class N3VQIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2054 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2055 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2056 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2057 (outs QPR:$Vd), (ins QPR:$Vm, QPR:$Vn), f, itin,
2058 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
2059 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (OpTy QPR:$Vn))))]> {
2060 let isCommutable = 0;
2063 // Multiply-Add/Sub operations: double- and quad-register.
2064 class N3VDMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2065 InstrItinClass itin, string OpcodeStr, string Dt,
2066 ValueType Ty, SDPatternOperator MulOp, SDPatternOperator OpNode>
2067 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2068 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2069 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2070 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
2071 (Ty (MulOp DPR:$Vn, DPR:$Vm)))))]>;
2073 class N3VDMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2074 string OpcodeStr, string Dt,
2075 ValueType Ty, SDPatternOperator MulOp, SDPatternOperator ShOp>
2076 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2078 (ins DPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, nohash_imm:$lane),
2080 OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "$src1 = $Vd",
2082 (Ty (ShOp (Ty DPR:$src1),
2084 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),
2086 class N3VDMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2087 string OpcodeStr, string Dt,
2088 ValueType Ty, SDNode MulOp, SDNode ShOp>
2089 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2091 (ins DPR:$src1, DPR:$Vn, DPR_8:$Vm, nohash_imm:$lane),
2093 OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "$src1 = $Vd",
2095 (Ty (ShOp (Ty DPR:$src1),
2097 (Ty (NEONvduplane (Ty DPR_8:$Vm),
2100 class N3VQMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2101 InstrItinClass itin, string OpcodeStr, string Dt, ValueType Ty,
2102 SDPatternOperator MulOp, SDPatternOperator OpNode>
2103 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2104 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2105 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2106 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
2107 (Ty (MulOp QPR:$Vn, QPR:$Vm)))))]>;
2108 class N3VQMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2109 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2110 SDPatternOperator MulOp, SDPatternOperator ShOp>
2111 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2113 (ins QPR:$src1, QPR:$Vn, DPR_VFP2:$Vm, nohash_imm:$lane),
2115 OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "$src1 = $Vd",
2116 [(set (ResTy QPR:$Vd),
2117 (ResTy (ShOp (ResTy QPR:$src1),
2118 (ResTy (MulOp QPR:$Vn,
2119 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2121 class N3VQMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2122 string OpcodeStr, string Dt,
2123 ValueType ResTy, ValueType OpTy,
2124 SDNode MulOp, SDNode ShOp>
2125 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2127 (ins QPR:$src1, QPR:$Vn, DPR_8:$Vm, nohash_imm:$lane),
2129 OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "$src1 = $Vd",
2130 [(set (ResTy QPR:$Vd),
2131 (ResTy (ShOp (ResTy QPR:$src1),
2132 (ResTy (MulOp QPR:$Vn,
2133 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2136 // Neon Intrinsic-Op instructions (VABA): double- and quad-register.
2137 class N3VDIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2138 InstrItinClass itin, string OpcodeStr, string Dt,
2139 ValueType Ty, Intrinsic IntOp, SDNode OpNode>
2140 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2141 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2142 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2143 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
2144 (Ty (IntOp (Ty DPR:$Vn), (Ty DPR:$Vm))))))]>;
2145 class N3VQIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2146 InstrItinClass itin, string OpcodeStr, string Dt,
2147 ValueType Ty, Intrinsic IntOp, SDNode OpNode>
2148 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2149 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2150 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2151 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
2152 (Ty (IntOp (Ty QPR:$Vn), (Ty QPR:$Vm))))))]>;
2154 // Neon 3-argument intrinsics, both double- and quad-register.
2155 // The destination register is also used as the first source operand register.
2156 class N3VDInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2157 InstrItinClass itin, string OpcodeStr, string Dt,
2158 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2159 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2160 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2161 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2162 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$src1),
2163 (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>;
2164 class N3VQInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2165 InstrItinClass itin, string OpcodeStr, string Dt,
2166 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2167 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2168 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2169 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2170 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$src1),
2171 (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]>;
2173 // Long Multiply-Add/Sub operations.
2174 class N3VLMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2175 InstrItinClass itin, string OpcodeStr, string Dt,
2176 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2177 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2178 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2179 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2180 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
2181 (TyQ (MulOp (TyD DPR:$Vn),
2182 (TyD DPR:$Vm)))))]>;
2183 class N3VLMulOpSL<bit op24, bits<2> op21_20, bits<4> op11_8,
2184 InstrItinClass itin, string OpcodeStr, string Dt,
2185 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2186 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$Vd),
2187 (ins QPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, nohash_imm:$lane),
2189 OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "$src1 = $Vd",
2191 (OpNode (TyQ QPR:$src1),
2192 (TyQ (MulOp (TyD DPR:$Vn),
2193 (TyD (NEONvduplane (TyD DPR_VFP2:$Vm),
2195 class N3VLMulOpSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2196 InstrItinClass itin, string OpcodeStr, string Dt,
2197 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2198 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$Vd),
2199 (ins QPR:$src1, DPR:$Vn, DPR_8:$Vm, nohash_imm:$lane),
2201 OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "$src1 = $Vd",
2203 (OpNode (TyQ QPR:$src1),
2204 (TyQ (MulOp (TyD DPR:$Vn),
2205 (TyD (NEONvduplane (TyD DPR_8:$Vm),
2208 // Long Intrinsic-Op vector operations with explicit extend (VABAL).
2209 class N3VLIntExtOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2210 InstrItinClass itin, string OpcodeStr, string Dt,
2211 ValueType TyQ, ValueType TyD, Intrinsic IntOp, SDNode ExtOp,
2213 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2214 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2215 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2216 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
2217 (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
2218 (TyD DPR:$Vm)))))))]>;
2220 // Neon Long 3-argument intrinsic. The destination register is
2221 // a quad-register and is also used as the first source operand register.
2222 class N3VLInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2223 InstrItinClass itin, string OpcodeStr, string Dt,
2224 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
2225 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2226 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2227 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2229 (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$Vn), (TyD DPR:$Vm))))]>;
2230 class N3VLInt3SL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2231 string OpcodeStr, string Dt,
2232 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2233 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2235 (ins QPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, nohash_imm:$lane),
2237 OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "$src1 = $Vd",
2238 [(set (ResTy QPR:$Vd),
2239 (ResTy (IntOp (ResTy QPR:$src1),
2241 (OpTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2243 class N3VLInt3SL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2244 InstrItinClass itin, string OpcodeStr, string Dt,
2245 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2246 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2248 (ins QPR:$src1, DPR:$Vn, DPR_8:$Vm, nohash_imm:$lane),
2250 OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "$src1 = $Vd",
2251 [(set (ResTy QPR:$Vd),
2252 (ResTy (IntOp (ResTy QPR:$src1),
2254 (OpTy (NEONvduplane (OpTy DPR_8:$Vm),
2257 // Narrowing 3-register intrinsics.
2258 class N3VNInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2259 string OpcodeStr, string Dt, ValueType TyD, ValueType TyQ,
2260 Intrinsic IntOp, bit Commutable>
2261 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2262 (outs DPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINi4D,
2263 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2264 [(set DPR:$Vd, (TyD (IntOp (TyQ QPR:$Vn), (TyQ QPR:$Vm))))]> {
2265 let isCommutable = Commutable;
2268 // Long 3-register operations.
2269 class N3VL<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2270 InstrItinClass itin, string OpcodeStr, string Dt,
2271 ValueType TyQ, ValueType TyD, SDNode OpNode, bit Commutable>
2272 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2273 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2274 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2275 [(set QPR:$Vd, (TyQ (OpNode (TyD DPR:$Vn), (TyD DPR:$Vm))))]> {
2276 let isCommutable = Commutable;
2278 class N3VLSL<bit op24, bits<2> op21_20, bits<4> op11_8,
2279 InstrItinClass itin, string OpcodeStr, string Dt,
2280 ValueType TyQ, ValueType TyD, SDNode OpNode>
2281 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2282 (outs QPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, nohash_imm:$lane),
2283 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "",
2285 (TyQ (OpNode (TyD DPR:$Vn),
2286 (TyD (NEONvduplane (TyD DPR_VFP2:$Vm),imm:$lane)))))]>;
2287 class N3VLSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2288 InstrItinClass itin, string OpcodeStr, string Dt,
2289 ValueType TyQ, ValueType TyD, SDNode OpNode>
2290 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2291 (outs QPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, nohash_imm:$lane),
2292 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "",
2294 (TyQ (OpNode (TyD DPR:$Vn),
2295 (TyD (NEONvduplane (TyD DPR_8:$Vm), imm:$lane)))))]>;
2297 // Long 3-register operations with explicitly extended operands.
2298 class N3VLExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2299 InstrItinClass itin, string OpcodeStr, string Dt,
2300 ValueType TyQ, ValueType TyD, SDNode OpNode, SDNode ExtOp,
2302 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2303 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2304 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2305 [(set QPR:$Vd, (OpNode (TyQ (ExtOp (TyD DPR:$Vn))),
2306 (TyQ (ExtOp (TyD DPR:$Vm)))))]> {
2307 let isCommutable = Commutable;
2310 // Long 3-register intrinsics with explicit extend (VABDL).
2311 class N3VLIntExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2312 InstrItinClass itin, string OpcodeStr, string Dt,
2313 ValueType TyQ, ValueType TyD, Intrinsic IntOp, SDNode ExtOp,
2315 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2316 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2317 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2318 [(set QPR:$Vd, (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
2319 (TyD DPR:$Vm))))))]> {
2320 let isCommutable = Commutable;
2323 // Long 3-register intrinsics.
2324 class N3VLInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2325 InstrItinClass itin, string OpcodeStr, string Dt,
2326 ValueType TyQ, ValueType TyD, Intrinsic IntOp, bit Commutable>
2327 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2328 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2329 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2330 [(set QPR:$Vd, (TyQ (IntOp (TyD DPR:$Vn), (TyD DPR:$Vm))))]> {
2331 let isCommutable = Commutable;
2333 class N3VLIntSL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2334 string OpcodeStr, string Dt,
2335 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2336 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2337 (outs QPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, nohash_imm:$lane),
2338 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "",
2339 [(set (ResTy QPR:$Vd),
2340 (ResTy (IntOp (OpTy DPR:$Vn),
2341 (OpTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2343 class N3VLIntSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2344 InstrItinClass itin, string OpcodeStr, string Dt,
2345 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2346 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2347 (outs QPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, nohash_imm:$lane),
2348 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "",
2349 [(set (ResTy QPR:$Vd),
2350 (ResTy (IntOp (OpTy DPR:$Vn),
2351 (OpTy (NEONvduplane (OpTy DPR_8:$Vm),
2354 // Wide 3-register operations.
2355 class N3VW<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2356 string OpcodeStr, string Dt, ValueType TyQ, ValueType TyD,
2357 SDNode OpNode, SDNode ExtOp, bit Commutable>
2358 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2359 (outs QPR:$Vd), (ins QPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VSUBiD,
2360 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2361 [(set QPR:$Vd, (OpNode (TyQ QPR:$Vn),
2362 (TyQ (ExtOp (TyD DPR:$Vm)))))]> {
2363 let isCommutable = Commutable;
2366 // Pairwise long 2-register intrinsics, both double- and quad-register.
2367 class N2VDPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2368 bits<2> op17_16, bits<5> op11_7, bit op4,
2369 string OpcodeStr, string Dt,
2370 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2371 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
2372 (ins DPR:$Vm), IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
2373 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm))))]>;
2374 class N2VQPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2375 bits<2> op17_16, bits<5> op11_7, bit op4,
2376 string OpcodeStr, string Dt,
2377 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2378 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
2379 (ins QPR:$Vm), IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
2380 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm))))]>;
2382 // Pairwise long 2-register accumulate intrinsics,
2383 // both double- and quad-register.
2384 // The destination register is also used as the first source operand register.
2385 class N2VDPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2386 bits<2> op17_16, bits<5> op11_7, bit op4,
2387 string OpcodeStr, string Dt,
2388 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2389 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
2390 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vm), IIC_VPALiD,
2391 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
2392 [(set DPR:$Vd, (ResTy (IntOp (ResTy DPR:$src1), (OpTy DPR:$Vm))))]>;
2393 class N2VQPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2394 bits<2> op17_16, bits<5> op11_7, bit op4,
2395 string OpcodeStr, string Dt,
2396 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2397 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4,
2398 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vm), IIC_VPALiQ,
2399 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
2400 [(set QPR:$Vd, (ResTy (IntOp (ResTy QPR:$src1), (OpTy QPR:$Vm))))]>;
2402 // Shift by immediate,
2403 // both double- and quad-register.
2404 class N2VDSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2405 Format f, InstrItinClass itin, Operand ImmTy,
2406 string OpcodeStr, string Dt, ValueType Ty, SDNode OpNode>
2407 : N2VImm<op24, op23, op11_8, op7, 0, op4,
2408 (outs DPR:$Vd), (ins DPR:$Vm, ImmTy:$SIMM), f, itin,
2409 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2410 [(set DPR:$Vd, (Ty (OpNode (Ty DPR:$Vm), (i32 imm:$SIMM))))]>;
2411 class N2VQSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2412 Format f, InstrItinClass itin, Operand ImmTy,
2413 string OpcodeStr, string Dt, ValueType Ty, SDNode OpNode>
2414 : N2VImm<op24, op23, op11_8, op7, 1, op4,
2415 (outs QPR:$Vd), (ins QPR:$Vm, ImmTy:$SIMM), f, itin,
2416 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2417 [(set QPR:$Vd, (Ty (OpNode (Ty QPR:$Vm), (i32 imm:$SIMM))))]>;
2419 // Long shift by immediate.
2420 class N2VLSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
2421 string OpcodeStr, string Dt,
2422 ValueType ResTy, ValueType OpTy, SDNode OpNode>
2423 : N2VImm<op24, op23, op11_8, op7, op6, op4,
2424 (outs QPR:$Vd), (ins DPR:$Vm, i32imm:$SIMM), N2RegVShLFrm,
2425 IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2426 [(set QPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vm),
2427 (i32 imm:$SIMM))))]>;
2429 // Narrow shift by immediate.
2430 class N2VNSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
2431 InstrItinClass itin, string OpcodeStr, string Dt,
2432 ValueType ResTy, ValueType OpTy, Operand ImmTy, SDNode OpNode>
2433 : N2VImm<op24, op23, op11_8, op7, op6, op4,
2434 (outs DPR:$Vd), (ins QPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, itin,
2435 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2436 [(set DPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vm),
2437 (i32 imm:$SIMM))))]>;
2439 // Shift right by immediate and accumulate,
2440 // both double- and quad-register.
2441 class N2VDShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2442 Operand ImmTy, string OpcodeStr, string Dt,
2443 ValueType Ty, SDNode ShOp>
2444 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
2445 (ins DPR:$src1, DPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, IIC_VPALiD,
2446 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2447 [(set DPR:$Vd, (Ty (add DPR:$src1,
2448 (Ty (ShOp DPR:$Vm, (i32 imm:$SIMM))))))]>;
2449 class N2VQShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2450 Operand ImmTy, string OpcodeStr, string Dt,
2451 ValueType Ty, SDNode ShOp>
2452 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
2453 (ins QPR:$src1, QPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, IIC_VPALiD,
2454 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2455 [(set QPR:$Vd, (Ty (add QPR:$src1,
2456 (Ty (ShOp QPR:$Vm, (i32 imm:$SIMM))))))]>;
2458 // Shift by immediate and insert,
2459 // both double- and quad-register.
2460 class N2VDShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2461 Operand ImmTy, Format f, string OpcodeStr, string Dt,
2462 ValueType Ty,SDNode ShOp>
2463 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
2464 (ins DPR:$src1, DPR:$Vm, ImmTy:$SIMM), f, IIC_VSHLiD,
2465 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2466 [(set DPR:$Vd, (Ty (ShOp DPR:$src1, DPR:$Vm, (i32 imm:$SIMM))))]>;
2467 class N2VQShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2468 Operand ImmTy, Format f, string OpcodeStr, string Dt,
2469 ValueType Ty,SDNode ShOp>
2470 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
2471 (ins QPR:$src1, QPR:$Vm, ImmTy:$SIMM), f, IIC_VSHLiQ,
2472 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2473 [(set QPR:$Vd, (Ty (ShOp QPR:$src1, QPR:$Vm, (i32 imm:$SIMM))))]>;
2475 // Convert, with fractional bits immediate,
2476 // both double- and quad-register.
2477 class N2VCvtD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2478 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2480 : N2VImm<op24, op23, op11_8, op7, 0, op4,
2481 (outs DPR:$Vd), (ins DPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
2482 IIC_VUNAD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2483 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (i32 imm:$SIMM))))]>;
2484 class N2VCvtQ<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2485 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2487 : N2VImm<op24, op23, op11_8, op7, 1, op4,
2488 (outs QPR:$Vd), (ins QPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
2489 IIC_VUNAQ, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2490 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (i32 imm:$SIMM))))]>;
2492 //===----------------------------------------------------------------------===//
2494 //===----------------------------------------------------------------------===//
2496 // Abbreviations used in multiclass suffixes:
2497 // Q = quarter int (8 bit) elements
2498 // H = half int (16 bit) elements
2499 // S = single int (32 bit) elements
2500 // D = double int (64 bit) elements
2502 // Neon 2-register vector operations and intrinsics.
2504 // Neon 2-register comparisons.
2505 // source operand element sizes of 8, 16 and 32 bits:
2506 multiclass N2V_QHS_cmp<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2507 bits<5> op11_7, bit op4, string opc, string Dt,
2508 string asm, SDNode OpNode> {
2509 // 64-bit vector types.
2510 def v8i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 0, op4,
2511 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2512 opc, !strconcat(Dt, "8"), asm, "",
2513 [(set DPR:$Vd, (v8i8 (OpNode (v8i8 DPR:$Vm))))]>;
2514 def v4i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 0, op4,
2515 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2516 opc, !strconcat(Dt, "16"), asm, "",
2517 [(set DPR:$Vd, (v4i16 (OpNode (v4i16 DPR:$Vm))))]>;
2518 def v2i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
2519 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2520 opc, !strconcat(Dt, "32"), asm, "",
2521 [(set DPR:$Vd, (v2i32 (OpNode (v2i32 DPR:$Vm))))]>;
2522 def v2f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
2523 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2524 opc, "f32", asm, "",
2525 [(set DPR:$Vd, (v2i32 (OpNode (v2f32 DPR:$Vm))))]> {
2526 let Inst{10} = 1; // overwrite F = 1
2529 // 128-bit vector types.
2530 def v16i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 1, op4,
2531 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2532 opc, !strconcat(Dt, "8"), asm, "",
2533 [(set QPR:$Vd, (v16i8 (OpNode (v16i8 QPR:$Vm))))]>;
2534 def v8i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 1, op4,
2535 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2536 opc, !strconcat(Dt, "16"), asm, "",
2537 [(set QPR:$Vd, (v8i16 (OpNode (v8i16 QPR:$Vm))))]>;
2538 def v4i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
2539 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2540 opc, !strconcat(Dt, "32"), asm, "",
2541 [(set QPR:$Vd, (v4i32 (OpNode (v4i32 QPR:$Vm))))]>;
2542 def v4f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
2543 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2544 opc, "f32", asm, "",
2545 [(set QPR:$Vd, (v4i32 (OpNode (v4f32 QPR:$Vm))))]> {
2546 let Inst{10} = 1; // overwrite F = 1
2551 // Neon 2-register vector intrinsics,
2552 // element sizes of 8, 16 and 32 bits:
2553 multiclass N2VInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2554 bits<5> op11_7, bit op4,
2555 InstrItinClass itinD, InstrItinClass itinQ,
2556 string OpcodeStr, string Dt, Intrinsic IntOp> {
2557 // 64-bit vector types.
2558 def v8i8 : N2VDInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2559 itinD, OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
2560 def v4i16 : N2VDInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2561 itinD, OpcodeStr, !strconcat(Dt, "16"),v4i16,v4i16,IntOp>;
2562 def v2i32 : N2VDInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2563 itinD, OpcodeStr, !strconcat(Dt, "32"),v2i32,v2i32,IntOp>;
2565 // 128-bit vector types.
2566 def v16i8 : N2VQInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2567 itinQ, OpcodeStr, !strconcat(Dt, "8"), v16i8,v16i8,IntOp>;
2568 def v8i16 : N2VQInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2569 itinQ, OpcodeStr, !strconcat(Dt, "16"),v8i16,v8i16,IntOp>;
2570 def v4i32 : N2VQInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2571 itinQ, OpcodeStr, !strconcat(Dt, "32"),v4i32,v4i32,IntOp>;
2575 // Neon Narrowing 2-register vector operations,
2576 // source operand element sizes of 16, 32 and 64 bits:
2577 multiclass N2VN_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2578 bits<5> op11_7, bit op6, bit op4,
2579 InstrItinClass itin, string OpcodeStr, string Dt,
2581 def v8i8 : N2VN<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
2582 itin, OpcodeStr, !strconcat(Dt, "16"),
2583 v8i8, v8i16, OpNode>;
2584 def v4i16 : N2VN<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
2585 itin, OpcodeStr, !strconcat(Dt, "32"),
2586 v4i16, v4i32, OpNode>;
2587 def v2i32 : N2VN<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
2588 itin, OpcodeStr, !strconcat(Dt, "64"),
2589 v2i32, v2i64, OpNode>;
2592 // Neon Narrowing 2-register vector intrinsics,
2593 // source operand element sizes of 16, 32 and 64 bits:
2594 multiclass N2VNInt_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2595 bits<5> op11_7, bit op6, bit op4,
2596 InstrItinClass itin, string OpcodeStr, string Dt,
2598 def v8i8 : N2VNInt<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
2599 itin, OpcodeStr, !strconcat(Dt, "16"),
2600 v8i8, v8i16, IntOp>;
2601 def v4i16 : N2VNInt<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
2602 itin, OpcodeStr, !strconcat(Dt, "32"),
2603 v4i16, v4i32, IntOp>;
2604 def v2i32 : N2VNInt<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
2605 itin, OpcodeStr, !strconcat(Dt, "64"),
2606 v2i32, v2i64, IntOp>;
2610 // Neon Lengthening 2-register vector intrinsic (currently specific to VMOVL).
2611 // source operand element sizes of 16, 32 and 64 bits:
2612 multiclass N2VL_QHS<bits<2> op24_23, bits<5> op11_7, bit op6, bit op4,
2613 string OpcodeStr, string Dt, SDNode OpNode> {
2614 def v8i16 : N2VL<op24_23, 0b00, 0b10, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2615 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode>;
2616 def v4i32 : N2VL<op24_23, 0b01, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2617 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
2618 def v2i64 : N2VL<op24_23, 0b10, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2619 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
2623 // Neon 3-register vector operations.
2625 // First with only element sizes of 8, 16 and 32 bits:
2626 multiclass N3V_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2627 InstrItinClass itinD16, InstrItinClass itinD32,
2628 InstrItinClass itinQ16, InstrItinClass itinQ32,
2629 string OpcodeStr, string Dt,
2630 SDNode OpNode, bit Commutable = 0> {
2631 // 64-bit vector types.
2632 def v8i8 : N3VD<op24, op23, 0b00, op11_8, op4, itinD16,
2633 OpcodeStr, !strconcat(Dt, "8"),
2634 v8i8, v8i8, OpNode, Commutable>;
2635 def v4i16 : N3VD<op24, op23, 0b01, op11_8, op4, itinD16,
2636 OpcodeStr, !strconcat(Dt, "16"),
2637 v4i16, v4i16, OpNode, Commutable>;
2638 def v2i32 : N3VD<op24, op23, 0b10, op11_8, op4, itinD32,
2639 OpcodeStr, !strconcat(Dt, "32"),
2640 v2i32, v2i32, OpNode, Commutable>;
2642 // 128-bit vector types.
2643 def v16i8 : N3VQ<op24, op23, 0b00, op11_8, op4, itinQ16,
2644 OpcodeStr, !strconcat(Dt, "8"),
2645 v16i8, v16i8, OpNode, Commutable>;
2646 def v8i16 : N3VQ<op24, op23, 0b01, op11_8, op4, itinQ16,
2647 OpcodeStr, !strconcat(Dt, "16"),
2648 v8i16, v8i16, OpNode, Commutable>;
2649 def v4i32 : N3VQ<op24, op23, 0b10, op11_8, op4, itinQ32,
2650 OpcodeStr, !strconcat(Dt, "32"),
2651 v4i32, v4i32, OpNode, Commutable>;
2654 multiclass N3VSL_HS<bits<4> op11_8, string OpcodeStr, string Dt, SDNode ShOp> {
2655 def v4i16 : N3VDSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
2657 def v2i32 : N3VDSL<0b10, op11_8, IIC_VMULi32D, OpcodeStr, !strconcat(Dt,"32"),
2659 def v8i16 : N3VQSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
2660 v8i16, v4i16, ShOp>;
2661 def v4i32 : N3VQSL<0b10, op11_8, IIC_VMULi32Q, OpcodeStr, !strconcat(Dt,"32"),
2662 v4i32, v2i32, ShOp>;
2665 // ....then also with element size 64 bits:
2666 multiclass N3V_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2667 InstrItinClass itinD, InstrItinClass itinQ,
2668 string OpcodeStr, string Dt,
2669 SDNode OpNode, bit Commutable = 0>
2670 : N3V_QHS<op24, op23, op11_8, op4, itinD, itinD, itinQ, itinQ,
2671 OpcodeStr, Dt, OpNode, Commutable> {
2672 def v1i64 : N3VD<op24, op23, 0b11, op11_8, op4, itinD,
2673 OpcodeStr, !strconcat(Dt, "64"),
2674 v1i64, v1i64, OpNode, Commutable>;
2675 def v2i64 : N3VQ<op24, op23, 0b11, op11_8, op4, itinQ,
2676 OpcodeStr, !strconcat(Dt, "64"),
2677 v2i64, v2i64, OpNode, Commutable>;
2681 // Neon 3-register vector intrinsics.
2683 // First with only element sizes of 16 and 32 bits:
2684 multiclass N3VInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2685 InstrItinClass itinD16, InstrItinClass itinD32,
2686 InstrItinClass itinQ16, InstrItinClass itinQ32,
2687 string OpcodeStr, string Dt,
2688 Intrinsic IntOp, bit Commutable = 0> {
2689 // 64-bit vector types.
2690 def v4i16 : N3VDInt<op24, op23, 0b01, op11_8, op4, f, itinD16,
2691 OpcodeStr, !strconcat(Dt, "16"),
2692 v4i16, v4i16, IntOp, Commutable>;
2693 def v2i32 : N3VDInt<op24, op23, 0b10, op11_8, op4, f, itinD32,
2694 OpcodeStr, !strconcat(Dt, "32"),
2695 v2i32, v2i32, IntOp, Commutable>;
2697 // 128-bit vector types.
2698 def v8i16 : N3VQInt<op24, op23, 0b01, op11_8, op4, f, itinQ16,
2699 OpcodeStr, !strconcat(Dt, "16"),
2700 v8i16, v8i16, IntOp, Commutable>;
2701 def v4i32 : N3VQInt<op24, op23, 0b10, op11_8, op4, f, itinQ32,
2702 OpcodeStr, !strconcat(Dt, "32"),
2703 v4i32, v4i32, IntOp, Commutable>;
2705 multiclass N3VInt_HSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2706 InstrItinClass itinD16, InstrItinClass itinD32,
2707 InstrItinClass itinQ16, InstrItinClass itinQ32,
2708 string OpcodeStr, string Dt,
2710 // 64-bit vector types.
2711 def v4i16 : N3VDIntSh<op24, op23, 0b01, op11_8, op4, f, itinD16,
2712 OpcodeStr, !strconcat(Dt, "16"),
2713 v4i16, v4i16, IntOp>;
2714 def v2i32 : N3VDIntSh<op24, op23, 0b10, op11_8, op4, f, itinD32,
2715 OpcodeStr, !strconcat(Dt, "32"),
2716 v2i32, v2i32, IntOp>;
2718 // 128-bit vector types.
2719 def v8i16 : N3VQIntSh<op24, op23, 0b01, op11_8, op4, f, itinQ16,
2720 OpcodeStr, !strconcat(Dt, "16"),
2721 v8i16, v8i16, IntOp>;
2722 def v4i32 : N3VQIntSh<op24, op23, 0b10, op11_8, op4, f, itinQ32,
2723 OpcodeStr, !strconcat(Dt, "32"),
2724 v4i32, v4i32, IntOp>;
2727 multiclass N3VIntSL_HS<bits<4> op11_8,
2728 InstrItinClass itinD16, InstrItinClass itinD32,
2729 InstrItinClass itinQ16, InstrItinClass itinQ32,
2730 string OpcodeStr, string Dt, Intrinsic IntOp> {
2731 def v4i16 : N3VDIntSL16<0b01, op11_8, itinD16,
2732 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp>;
2733 def v2i32 : N3VDIntSL<0b10, op11_8, itinD32,
2734 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp>;
2735 def v8i16 : N3VQIntSL16<0b01, op11_8, itinQ16,
2736 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16, IntOp>;
2737 def v4i32 : N3VQIntSL<0b10, op11_8, itinQ32,
2738 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32, IntOp>;
2741 // ....then also with element size of 8 bits:
2742 multiclass N3VInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2743 InstrItinClass itinD16, InstrItinClass itinD32,
2744 InstrItinClass itinQ16, InstrItinClass itinQ32,
2745 string OpcodeStr, string Dt,
2746 Intrinsic IntOp, bit Commutable = 0>
2747 : N3VInt_HS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2748 OpcodeStr, Dt, IntOp, Commutable> {
2749 def v8i8 : N3VDInt<op24, op23, 0b00, op11_8, op4, f, itinD16,
2750 OpcodeStr, !strconcat(Dt, "8"),
2751 v8i8, v8i8, IntOp, Commutable>;
2752 def v16i8 : N3VQInt<op24, op23, 0b00, op11_8, op4, f, itinQ16,
2753 OpcodeStr, !strconcat(Dt, "8"),
2754 v16i8, v16i8, IntOp, Commutable>;
2756 multiclass N3VInt_QHSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2757 InstrItinClass itinD16, InstrItinClass itinD32,
2758 InstrItinClass itinQ16, InstrItinClass itinQ32,
2759 string OpcodeStr, string Dt,
2761 : N3VInt_HSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2762 OpcodeStr, Dt, IntOp> {
2763 def v8i8 : N3VDIntSh<op24, op23, 0b00, op11_8, op4, f, itinD16,
2764 OpcodeStr, !strconcat(Dt, "8"),
2766 def v16i8 : N3VQIntSh<op24, op23, 0b00, op11_8, op4, f, itinQ16,
2767 OpcodeStr, !strconcat(Dt, "8"),
2768 v16i8, v16i8, IntOp>;
2772 // ....then also with element size of 64 bits:
2773 multiclass N3VInt_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2774 InstrItinClass itinD16, InstrItinClass itinD32,
2775 InstrItinClass itinQ16, InstrItinClass itinQ32,
2776 string OpcodeStr, string Dt,
2777 Intrinsic IntOp, bit Commutable = 0>
2778 : N3VInt_QHS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2779 OpcodeStr, Dt, IntOp, Commutable> {
2780 def v1i64 : N3VDInt<op24, op23, 0b11, op11_8, op4, f, itinD32,
2781 OpcodeStr, !strconcat(Dt, "64"),
2782 v1i64, v1i64, IntOp, Commutable>;
2783 def v2i64 : N3VQInt<op24, op23, 0b11, op11_8, op4, f, itinQ32,
2784 OpcodeStr, !strconcat(Dt, "64"),
2785 v2i64, v2i64, IntOp, Commutable>;
2787 multiclass N3VInt_QHSDSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2788 InstrItinClass itinD16, InstrItinClass itinD32,
2789 InstrItinClass itinQ16, InstrItinClass itinQ32,
2790 string OpcodeStr, string Dt,
2792 : N3VInt_QHSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2793 OpcodeStr, Dt, IntOp> {
2794 def v1i64 : N3VDIntSh<op24, op23, 0b11, op11_8, op4, f, itinD32,
2795 OpcodeStr, !strconcat(Dt, "64"),
2796 v1i64, v1i64, IntOp>;
2797 def v2i64 : N3VQIntSh<op24, op23, 0b11, op11_8, op4, f, itinQ32,
2798 OpcodeStr, !strconcat(Dt, "64"),
2799 v2i64, v2i64, IntOp>;
2802 // Neon Narrowing 3-register vector intrinsics,
2803 // source operand element sizes of 16, 32 and 64 bits:
2804 multiclass N3VNInt_HSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2805 string OpcodeStr, string Dt,
2806 Intrinsic IntOp, bit Commutable = 0> {
2807 def v8i8 : N3VNInt<op24, op23, 0b00, op11_8, op4,
2808 OpcodeStr, !strconcat(Dt, "16"),
2809 v8i8, v8i16, IntOp, Commutable>;
2810 def v4i16 : N3VNInt<op24, op23, 0b01, op11_8, op4,
2811 OpcodeStr, !strconcat(Dt, "32"),
2812 v4i16, v4i32, IntOp, Commutable>;
2813 def v2i32 : N3VNInt<op24, op23, 0b10, op11_8, op4,
2814 OpcodeStr, !strconcat(Dt, "64"),
2815 v2i32, v2i64, IntOp, Commutable>;
2819 // Neon Long 3-register vector operations.
2821 multiclass N3VL_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2822 InstrItinClass itin16, InstrItinClass itin32,
2823 string OpcodeStr, string Dt,
2824 SDNode OpNode, bit Commutable = 0> {
2825 def v8i16 : N3VL<op24, op23, 0b00, op11_8, op4, itin16,
2826 OpcodeStr, !strconcat(Dt, "8"),
2827 v8i16, v8i8, OpNode, Commutable>;
2828 def v4i32 : N3VL<op24, op23, 0b01, op11_8, op4, itin16,
2829 OpcodeStr, !strconcat(Dt, "16"),
2830 v4i32, v4i16, OpNode, Commutable>;
2831 def v2i64 : N3VL<op24, op23, 0b10, op11_8, op4, itin32,
2832 OpcodeStr, !strconcat(Dt, "32"),
2833 v2i64, v2i32, OpNode, Commutable>;
2836 multiclass N3VLSL_HS<bit op24, bits<4> op11_8,
2837 InstrItinClass itin, string OpcodeStr, string Dt,
2839 def v4i16 : N3VLSL16<op24, 0b01, op11_8, itin, OpcodeStr,
2840 !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
2841 def v2i32 : N3VLSL<op24, 0b10, op11_8, itin, OpcodeStr,
2842 !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
2845 multiclass N3VLExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2846 InstrItinClass itin16, InstrItinClass itin32,
2847 string OpcodeStr, string Dt,
2848 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
2849 def v8i16 : N3VLExt<op24, op23, 0b00, op11_8, op4, itin16,
2850 OpcodeStr, !strconcat(Dt, "8"),
2851 v8i16, v8i8, OpNode, ExtOp, Commutable>;
2852 def v4i32 : N3VLExt<op24, op23, 0b01, op11_8, op4, itin16,
2853 OpcodeStr, !strconcat(Dt, "16"),
2854 v4i32, v4i16, OpNode, ExtOp, Commutable>;
2855 def v2i64 : N3VLExt<op24, op23, 0b10, op11_8, op4, itin32,
2856 OpcodeStr, !strconcat(Dt, "32"),
2857 v2i64, v2i32, OpNode, ExtOp, Commutable>;
2860 // Neon Long 3-register vector intrinsics.
2862 // First with only element sizes of 16 and 32 bits:
2863 multiclass N3VLInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
2864 InstrItinClass itin16, InstrItinClass itin32,
2865 string OpcodeStr, string Dt,
2866 Intrinsic IntOp, bit Commutable = 0> {
2867 def v4i32 : N3VLInt<op24, op23, 0b01, op11_8, op4, itin16,
2868 OpcodeStr, !strconcat(Dt, "16"),
2869 v4i32, v4i16, IntOp, Commutable>;
2870 def v2i64 : N3VLInt<op24, op23, 0b10, op11_8, op4, itin32,
2871 OpcodeStr, !strconcat(Dt, "32"),
2872 v2i64, v2i32, IntOp, Commutable>;
2875 multiclass N3VLIntSL_HS<bit op24, bits<4> op11_8,
2876 InstrItinClass itin, string OpcodeStr, string Dt,
2878 def v4i16 : N3VLIntSL16<op24, 0b01, op11_8, itin,
2879 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
2880 def v2i32 : N3VLIntSL<op24, 0b10, op11_8, itin,
2881 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
2884 // ....then also with element size of 8 bits:
2885 multiclass N3VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2886 InstrItinClass itin16, InstrItinClass itin32,
2887 string OpcodeStr, string Dt,
2888 Intrinsic IntOp, bit Commutable = 0>
2889 : N3VLInt_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt,
2890 IntOp, Commutable> {
2891 def v8i16 : N3VLInt<op24, op23, 0b00, op11_8, op4, itin16,
2892 OpcodeStr, !strconcat(Dt, "8"),
2893 v8i16, v8i8, IntOp, Commutable>;
2896 // ....with explicit extend (VABDL).
2897 multiclass N3VLIntExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2898 InstrItinClass itin, string OpcodeStr, string Dt,
2899 Intrinsic IntOp, SDNode ExtOp, bit Commutable = 0> {
2900 def v8i16 : N3VLIntExt<op24, op23, 0b00, op11_8, op4, itin,
2901 OpcodeStr, !strconcat(Dt, "8"),
2902 v8i16, v8i8, IntOp, ExtOp, Commutable>;
2903 def v4i32 : N3VLIntExt<op24, op23, 0b01, op11_8, op4, itin,
2904 OpcodeStr, !strconcat(Dt, "16"),
2905 v4i32, v4i16, IntOp, ExtOp, Commutable>;
2906 def v2i64 : N3VLIntExt<op24, op23, 0b10, op11_8, op4, itin,
2907 OpcodeStr, !strconcat(Dt, "32"),
2908 v2i64, v2i32, IntOp, ExtOp, Commutable>;
2912 // Neon Wide 3-register vector intrinsics,
2913 // source operand element sizes of 8, 16 and 32 bits:
2914 multiclass N3VW_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2915 string OpcodeStr, string Dt,
2916 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
2917 def v8i16 : N3VW<op24, op23, 0b00, op11_8, op4,
2918 OpcodeStr, !strconcat(Dt, "8"),
2919 v8i16, v8i8, OpNode, ExtOp, Commutable>;
2920 def v4i32 : N3VW<op24, op23, 0b01, op11_8, op4,
2921 OpcodeStr, !strconcat(Dt, "16"),
2922 v4i32, v4i16, OpNode, ExtOp, Commutable>;
2923 def v2i64 : N3VW<op24, op23, 0b10, op11_8, op4,
2924 OpcodeStr, !strconcat(Dt, "32"),
2925 v2i64, v2i32, OpNode, ExtOp, Commutable>;
2929 // Neon Multiply-Op vector operations,
2930 // element sizes of 8, 16 and 32 bits:
2931 multiclass N3VMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2932 InstrItinClass itinD16, InstrItinClass itinD32,
2933 InstrItinClass itinQ16, InstrItinClass itinQ32,
2934 string OpcodeStr, string Dt, SDNode OpNode> {
2935 // 64-bit vector types.
2936 def v8i8 : N3VDMulOp<op24, op23, 0b00, op11_8, op4, itinD16,
2937 OpcodeStr, !strconcat(Dt, "8"), v8i8, mul, OpNode>;
2938 def v4i16 : N3VDMulOp<op24, op23, 0b01, op11_8, op4, itinD16,
2939 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, OpNode>;
2940 def v2i32 : N3VDMulOp<op24, op23, 0b10, op11_8, op4, itinD32,
2941 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, OpNode>;
2943 // 128-bit vector types.
2944 def v16i8 : N3VQMulOp<op24, op23, 0b00, op11_8, op4, itinQ16,
2945 OpcodeStr, !strconcat(Dt, "8"), v16i8, mul, OpNode>;
2946 def v8i16 : N3VQMulOp<op24, op23, 0b01, op11_8, op4, itinQ16,
2947 OpcodeStr, !strconcat(Dt, "16"), v8i16, mul, OpNode>;
2948 def v4i32 : N3VQMulOp<op24, op23, 0b10, op11_8, op4, itinQ32,
2949 OpcodeStr, !strconcat(Dt, "32"), v4i32, mul, OpNode>;
2952 multiclass N3VMulOpSL_HS<bits<4> op11_8,
2953 InstrItinClass itinD16, InstrItinClass itinD32,
2954 InstrItinClass itinQ16, InstrItinClass itinQ32,
2955 string OpcodeStr, string Dt, SDNode ShOp> {
2956 def v4i16 : N3VDMulOpSL16<0b01, op11_8, itinD16,
2957 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, ShOp>;
2958 def v2i32 : N3VDMulOpSL<0b10, op11_8, itinD32,
2959 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, ShOp>;
2960 def v8i16 : N3VQMulOpSL16<0b01, op11_8, itinQ16,
2961 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16,
2963 def v4i32 : N3VQMulOpSL<0b10, op11_8, itinQ32,
2964 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32,
2968 // Neon Intrinsic-Op vector operations,
2969 // element sizes of 8, 16 and 32 bits:
2970 multiclass N3VIntOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2971 InstrItinClass itinD, InstrItinClass itinQ,
2972 string OpcodeStr, string Dt, Intrinsic IntOp,
2974 // 64-bit vector types.
2975 def v8i8 : N3VDIntOp<op24, op23, 0b00, op11_8, op4, itinD,
2976 OpcodeStr, !strconcat(Dt, "8"), v8i8, IntOp, OpNode>;
2977 def v4i16 : N3VDIntOp<op24, op23, 0b01, op11_8, op4, itinD,
2978 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp, OpNode>;
2979 def v2i32 : N3VDIntOp<op24, op23, 0b10, op11_8, op4, itinD,
2980 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp, OpNode>;
2982 // 128-bit vector types.
2983 def v16i8 : N3VQIntOp<op24, op23, 0b00, op11_8, op4, itinQ,
2984 OpcodeStr, !strconcat(Dt, "8"), v16i8, IntOp, OpNode>;
2985 def v8i16 : N3VQIntOp<op24, op23, 0b01, op11_8, op4, itinQ,
2986 OpcodeStr, !strconcat(Dt, "16"), v8i16, IntOp, OpNode>;
2987 def v4i32 : N3VQIntOp<op24, op23, 0b10, op11_8, op4, itinQ,
2988 OpcodeStr, !strconcat(Dt, "32"), v4i32, IntOp, OpNode>;
2991 // Neon 3-argument intrinsics,
2992 // element sizes of 8, 16 and 32 bits:
2993 multiclass N3VInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2994 InstrItinClass itinD, InstrItinClass itinQ,
2995 string OpcodeStr, string Dt, Intrinsic IntOp> {
2996 // 64-bit vector types.
2997 def v8i8 : N3VDInt3<op24, op23, 0b00, op11_8, op4, itinD,
2998 OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
2999 def v4i16 : N3VDInt3<op24, op23, 0b01, op11_8, op4, itinD,
3000 OpcodeStr, !strconcat(Dt, "16"), v4i16, v4i16, IntOp>;
3001 def v2i32 : N3VDInt3<op24, op23, 0b10, op11_8, op4, itinD,
3002 OpcodeStr, !strconcat(Dt, "32"), v2i32, v2i32, IntOp>;
3004 // 128-bit vector types.
3005 def v16i8 : N3VQInt3<op24, op23, 0b00, op11_8, op4, itinQ,
3006 OpcodeStr, !strconcat(Dt, "8"), v16i8, v16i8, IntOp>;
3007 def v8i16 : N3VQInt3<op24, op23, 0b01, op11_8, op4, itinQ,
3008 OpcodeStr, !strconcat(Dt, "16"), v8i16, v8i16, IntOp>;
3009 def v4i32 : N3VQInt3<op24, op23, 0b10, op11_8, op4, itinQ,
3010 OpcodeStr, !strconcat(Dt, "32"), v4i32, v4i32, IntOp>;
3014 // Neon Long Multiply-Op vector operations,
3015 // element sizes of 8, 16 and 32 bits:
3016 multiclass N3VLMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3017 InstrItinClass itin16, InstrItinClass itin32,
3018 string OpcodeStr, string Dt, SDNode MulOp,
3020 def v8i16 : N3VLMulOp<op24, op23, 0b00, op11_8, op4, itin16, OpcodeStr,
3021 !strconcat(Dt, "8"), v8i16, v8i8, MulOp, OpNode>;
3022 def v4i32 : N3VLMulOp<op24, op23, 0b01, op11_8, op4, itin16, OpcodeStr,
3023 !strconcat(Dt, "16"), v4i32, v4i16, MulOp, OpNode>;
3024 def v2i64 : N3VLMulOp<op24, op23, 0b10, op11_8, op4, itin32, OpcodeStr,
3025 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
3028 multiclass N3VLMulOpSL_HS<bit op24, bits<4> op11_8, string OpcodeStr,
3029 string Dt, SDNode MulOp, SDNode OpNode> {
3030 def v4i16 : N3VLMulOpSL16<op24, 0b01, op11_8, IIC_VMACi16D, OpcodeStr,
3031 !strconcat(Dt,"16"), v4i32, v4i16, MulOp, OpNode>;
3032 def v2i32 : N3VLMulOpSL<op24, 0b10, op11_8, IIC_VMACi32D, OpcodeStr,
3033 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
3037 // Neon Long 3-argument intrinsics.
3039 // First with only element sizes of 16 and 32 bits:
3040 multiclass N3VLInt3_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
3041 InstrItinClass itin16, InstrItinClass itin32,
3042 string OpcodeStr, string Dt, Intrinsic IntOp> {
3043 def v4i32 : N3VLInt3<op24, op23, 0b01, op11_8, op4, itin16,
3044 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
3045 def v2i64 : N3VLInt3<op24, op23, 0b10, op11_8, op4, itin32,
3046 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3049 multiclass N3VLInt3SL_HS<bit op24, bits<4> op11_8,
3050 string OpcodeStr, string Dt, Intrinsic IntOp> {
3051 def v4i16 : N3VLInt3SL16<op24, 0b01, op11_8, IIC_VMACi16D,
3052 OpcodeStr, !strconcat(Dt,"16"), v4i32, v4i16, IntOp>;
3053 def v2i32 : N3VLInt3SL<op24, 0b10, op11_8, IIC_VMACi32D,
3054 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3057 // ....then also with element size of 8 bits:
3058 multiclass N3VLInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3059 InstrItinClass itin16, InstrItinClass itin32,
3060 string OpcodeStr, string Dt, Intrinsic IntOp>
3061 : N3VLInt3_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt, IntOp> {
3062 def v8i16 : N3VLInt3<op24, op23, 0b00, op11_8, op4, itin16,
3063 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
3066 // ....with explicit extend (VABAL).
3067 multiclass N3VLIntExtOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3068 InstrItinClass itin, string OpcodeStr, string Dt,
3069 Intrinsic IntOp, SDNode ExtOp, SDNode OpNode> {
3070 def v8i16 : N3VLIntExtOp<op24, op23, 0b00, op11_8, op4, itin,
3071 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8,
3072 IntOp, ExtOp, OpNode>;
3073 def v4i32 : N3VLIntExtOp<op24, op23, 0b01, op11_8, op4, itin,
3074 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16,
3075 IntOp, ExtOp, OpNode>;
3076 def v2i64 : N3VLIntExtOp<op24, op23, 0b10, op11_8, op4, itin,
3077 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32,
3078 IntOp, ExtOp, OpNode>;
3082 // Neon Pairwise long 2-register intrinsics,
3083 // element sizes of 8, 16 and 32 bits:
3084 multiclass N2VPLInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3085 bits<5> op11_7, bit op4,
3086 string OpcodeStr, string Dt, Intrinsic IntOp> {
3087 // 64-bit vector types.
3088 def v8i8 : N2VDPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3089 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
3090 def v4i16 : N2VDPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3091 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
3092 def v2i32 : N2VDPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3093 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
3095 // 128-bit vector types.
3096 def v16i8 : N2VQPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3097 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
3098 def v8i16 : N2VQPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3099 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
3100 def v4i32 : N2VQPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3101 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
3105 // Neon Pairwise long 2-register accumulate intrinsics,
3106 // element sizes of 8, 16 and 32 bits:
3107 multiclass N2VPLInt2_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3108 bits<5> op11_7, bit op4,
3109 string OpcodeStr, string Dt, Intrinsic IntOp> {
3110 // 64-bit vector types.
3111 def v8i8 : N2VDPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3112 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
3113 def v4i16 : N2VDPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3114 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
3115 def v2i32 : N2VDPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3116 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
3118 // 128-bit vector types.
3119 def v16i8 : N2VQPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3120 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
3121 def v8i16 : N2VQPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3122 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
3123 def v4i32 : N2VQPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3124 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
3128 // Neon 2-register vector shift by immediate,
3129 // with f of either N2RegVShLFrm or N2RegVShRFrm
3130 // element sizes of 8, 16, 32 and 64 bits:
3131 multiclass N2VShL_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3132 InstrItinClass itin, string OpcodeStr, string Dt,
3134 // 64-bit vector types.
3135 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3136 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
3137 let Inst{21-19} = 0b001; // imm6 = 001xxx
3139 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3140 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
3141 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3143 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3144 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
3145 let Inst{21} = 0b1; // imm6 = 1xxxxx
3147 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, N2RegVShLFrm, itin, i32imm,
3148 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
3151 // 128-bit vector types.
3152 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3153 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
3154 let Inst{21-19} = 0b001; // imm6 = 001xxx
3156 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3157 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
3158 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3160 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3161 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
3162 let Inst{21} = 0b1; // imm6 = 1xxxxx
3164 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, N2RegVShLFrm, itin, i32imm,
3165 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
3168 multiclass N2VShR_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3169 InstrItinClass itin, string OpcodeStr, string Dt,
3171 // 64-bit vector types.
3172 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm8,
3173 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
3174 let Inst{21-19} = 0b001; // imm6 = 001xxx
3176 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm16,
3177 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
3178 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3180 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm32,
3181 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
3182 let Inst{21} = 0b1; // imm6 = 1xxxxx
3184 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, N2RegVShRFrm, itin, shr_imm64,
3185 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
3188 // 128-bit vector types.
3189 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm8,
3190 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
3191 let Inst{21-19} = 0b001; // imm6 = 001xxx
3193 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm16,
3194 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
3195 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3197 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm32,
3198 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
3199 let Inst{21} = 0b1; // imm6 = 1xxxxx
3201 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, N2RegVShRFrm, itin, shr_imm64,
3202 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
3206 // Neon Shift-Accumulate vector operations,
3207 // element sizes of 8, 16, 32 and 64 bits:
3208 multiclass N2VShAdd_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3209 string OpcodeStr, string Dt, SDNode ShOp> {
3210 // 64-bit vector types.
3211 def v8i8 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm8,
3212 OpcodeStr, !strconcat(Dt, "8"), v8i8, ShOp> {
3213 let Inst{21-19} = 0b001; // imm6 = 001xxx
3215 def v4i16 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm16,
3216 OpcodeStr, !strconcat(Dt, "16"), v4i16, ShOp> {
3217 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3219 def v2i32 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm32,
3220 OpcodeStr, !strconcat(Dt, "32"), v2i32, ShOp> {
3221 let Inst{21} = 0b1; // imm6 = 1xxxxx
3223 def v1i64 : N2VDShAdd<op24, op23, op11_8, 1, op4, shr_imm64,
3224 OpcodeStr, !strconcat(Dt, "64"), v1i64, ShOp>;
3227 // 128-bit vector types.
3228 def v16i8 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm8,
3229 OpcodeStr, !strconcat(Dt, "8"), v16i8, ShOp> {
3230 let Inst{21-19} = 0b001; // imm6 = 001xxx
3232 def v8i16 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm16,
3233 OpcodeStr, !strconcat(Dt, "16"), v8i16, ShOp> {
3234 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3236 def v4i32 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm32,
3237 OpcodeStr, !strconcat(Dt, "32"), v4i32, ShOp> {
3238 let Inst{21} = 0b1; // imm6 = 1xxxxx
3240 def v2i64 : N2VQShAdd<op24, op23, op11_8, 1, op4, shr_imm64,
3241 OpcodeStr, !strconcat(Dt, "64"), v2i64, ShOp>;
3245 // Neon Shift-Insert vector operations,
3246 // with f of either N2RegVShLFrm or N2RegVShRFrm
3247 // element sizes of 8, 16, 32 and 64 bits:
3248 multiclass N2VShInsL_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3250 // 64-bit vector types.
3251 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3252 N2RegVShLFrm, OpcodeStr, "8", v8i8, NEONvsli> {
3253 let Inst{21-19} = 0b001; // imm6 = 001xxx
3255 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3256 N2RegVShLFrm, OpcodeStr, "16", v4i16, NEONvsli> {
3257 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3259 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3260 N2RegVShLFrm, OpcodeStr, "32", v2i32, NEONvsli> {
3261 let Inst{21} = 0b1; // imm6 = 1xxxxx
3263 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4, i32imm,
3264 N2RegVShLFrm, OpcodeStr, "64", v1i64, NEONvsli>;
3267 // 128-bit vector types.
3268 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3269 N2RegVShLFrm, OpcodeStr, "8", v16i8, NEONvsli> {
3270 let Inst{21-19} = 0b001; // imm6 = 001xxx
3272 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3273 N2RegVShLFrm, OpcodeStr, "16", v8i16, NEONvsli> {
3274 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3276 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3277 N2RegVShLFrm, OpcodeStr, "32", v4i32, NEONvsli> {
3278 let Inst{21} = 0b1; // imm6 = 1xxxxx
3280 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4, i32imm,
3281 N2RegVShLFrm, OpcodeStr, "64", v2i64, NEONvsli>;
3284 multiclass N2VShInsR_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3286 // 64-bit vector types.
3287 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm8,
3288 N2RegVShRFrm, OpcodeStr, "8", v8i8, NEONvsri> {
3289 let Inst{21-19} = 0b001; // imm6 = 001xxx
3291 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm16,
3292 N2RegVShRFrm, OpcodeStr, "16", v4i16, NEONvsri> {
3293 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3295 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm32,
3296 N2RegVShRFrm, OpcodeStr, "32", v2i32, NEONvsri> {
3297 let Inst{21} = 0b1; // imm6 = 1xxxxx
3299 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4, shr_imm64,
3300 N2RegVShRFrm, OpcodeStr, "64", v1i64, NEONvsri>;
3303 // 128-bit vector types.
3304 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm8,
3305 N2RegVShRFrm, OpcodeStr, "8", v16i8, NEONvsri> {
3306 let Inst{21-19} = 0b001; // imm6 = 001xxx
3308 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm16,
3309 N2RegVShRFrm, OpcodeStr, "16", v8i16, NEONvsri> {
3310 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3312 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm32,
3313 N2RegVShRFrm, OpcodeStr, "32", v4i32, NEONvsri> {
3314 let Inst{21} = 0b1; // imm6 = 1xxxxx
3316 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4, shr_imm64,
3317 N2RegVShRFrm, OpcodeStr, "64", v2i64, NEONvsri>;
3321 // Neon Shift Long operations,
3322 // element sizes of 8, 16, 32 bits:
3323 multiclass N2VLSh_QHS<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
3324 bit op4, string OpcodeStr, string Dt, SDNode OpNode> {
3325 def v8i16 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3326 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode> {
3327 let Inst{21-19} = 0b001; // imm6 = 001xxx
3329 def v4i32 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3330 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode> {
3331 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3333 def v2i64 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3334 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode> {
3335 let Inst{21} = 0b1; // imm6 = 1xxxxx
3339 // Neon Shift Narrow operations,
3340 // element sizes of 16, 32, 64 bits:
3341 multiclass N2VNSh_HSD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
3342 bit op4, InstrItinClass itin, string OpcodeStr, string Dt,
3344 def v8i8 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3345 OpcodeStr, !strconcat(Dt, "16"),
3346 v8i8, v8i16, shr_imm8, OpNode> {
3347 let Inst{21-19} = 0b001; // imm6 = 001xxx
3349 def v4i16 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3350 OpcodeStr, !strconcat(Dt, "32"),
3351 v4i16, v4i32, shr_imm16, OpNode> {
3352 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3354 def v2i32 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3355 OpcodeStr, !strconcat(Dt, "64"),
3356 v2i32, v2i64, shr_imm32, OpNode> {
3357 let Inst{21} = 0b1; // imm6 = 1xxxxx
3361 //===----------------------------------------------------------------------===//
3362 // Instruction Definitions.
3363 //===----------------------------------------------------------------------===//
3365 // Vector Add Operations.
3367 // VADD : Vector Add (integer and floating-point)
3368 defm VADD : N3V_QHSD<0, 0, 0b1000, 0, IIC_VBINiD, IIC_VBINiQ, "vadd", "i",
3370 def VADDfd : N3VD<0, 0, 0b00, 0b1101, 0, IIC_VBIND, "vadd", "f32",
3371 v2f32, v2f32, fadd, 1>;
3372 def VADDfq : N3VQ<0, 0, 0b00, 0b1101, 0, IIC_VBINQ, "vadd", "f32",
3373 v4f32, v4f32, fadd, 1>;
3374 // VADDL : Vector Add Long (Q = D + D)
3375 defm VADDLs : N3VLExt_QHS<0,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
3376 "vaddl", "s", add, sext, 1>;
3377 defm VADDLu : N3VLExt_QHS<1,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
3378 "vaddl", "u", add, zext, 1>;
3379 // VADDW : Vector Add Wide (Q = Q + D)
3380 defm VADDWs : N3VW_QHS<0,1,0b0001,0, "vaddw", "s", add, sext, 0>;
3381 defm VADDWu : N3VW_QHS<1,1,0b0001,0, "vaddw", "u", add, zext, 0>;
3382 // VHADD : Vector Halving Add
3383 defm VHADDs : N3VInt_QHS<0, 0, 0b0000, 0, N3RegFrm,
3384 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3385 "vhadd", "s", int_arm_neon_vhadds, 1>;
3386 defm VHADDu : N3VInt_QHS<1, 0, 0b0000, 0, N3RegFrm,
3387 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3388 "vhadd", "u", int_arm_neon_vhaddu, 1>;
3389 // VRHADD : Vector Rounding Halving Add
3390 defm VRHADDs : N3VInt_QHS<0, 0, 0b0001, 0, N3RegFrm,
3391 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3392 "vrhadd", "s", int_arm_neon_vrhadds, 1>;
3393 defm VRHADDu : N3VInt_QHS<1, 0, 0b0001, 0, N3RegFrm,
3394 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3395 "vrhadd", "u", int_arm_neon_vrhaddu, 1>;
3396 // VQADD : Vector Saturating Add
3397 defm VQADDs : N3VInt_QHSD<0, 0, 0b0000, 1, N3RegFrm,
3398 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3399 "vqadd", "s", int_arm_neon_vqadds, 1>;
3400 defm VQADDu : N3VInt_QHSD<1, 0, 0b0000, 1, N3RegFrm,
3401 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3402 "vqadd", "u", int_arm_neon_vqaddu, 1>;
3403 // VADDHN : Vector Add and Narrow Returning High Half (D = Q + Q)
3404 defm VADDHN : N3VNInt_HSD<0,1,0b0100,0, "vaddhn", "i",
3405 int_arm_neon_vaddhn, 1>;
3406 // VRADDHN : Vector Rounding Add and Narrow Returning High Half (D = Q + Q)
3407 defm VRADDHN : N3VNInt_HSD<1,1,0b0100,0, "vraddhn", "i",
3408 int_arm_neon_vraddhn, 1>;
3410 // Vector Multiply Operations.
3412 // VMUL : Vector Multiply (integer, polynomial and floating-point)
3413 defm VMUL : N3V_QHS<0, 0, 0b1001, 1, IIC_VMULi16D, IIC_VMULi32D,
3414 IIC_VMULi16Q, IIC_VMULi32Q, "vmul", "i", mul, 1>;
3415 def VMULpd : N3VDInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16D, "vmul",
3416 "p8", v8i8, v8i8, int_arm_neon_vmulp, 1>;
3417 def VMULpq : N3VQInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16Q, "vmul",
3418 "p8", v16i8, v16i8, int_arm_neon_vmulp, 1>;
3419 def VMULfd : N3VD<1, 0, 0b00, 0b1101, 1, IIC_VFMULD, "vmul", "f32",
3420 v2f32, v2f32, fmul, 1>;
3421 def VMULfq : N3VQ<1, 0, 0b00, 0b1101, 1, IIC_VFMULQ, "vmul", "f32",
3422 v4f32, v4f32, fmul, 1>;
3423 defm VMULsl : N3VSL_HS<0b1000, "vmul", "i", mul>;
3424 def VMULslfd : N3VDSL<0b10, 0b1001, IIC_VBIND, "vmul", "f32", v2f32, fmul>;
3425 def VMULslfq : N3VQSL<0b10, 0b1001, IIC_VBINQ, "vmul", "f32", v4f32,
3428 def : Pat<(v8i16 (mul (v8i16 QPR:$src1),
3429 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
3430 (v8i16 (VMULslv8i16 (v8i16 QPR:$src1),
3431 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3432 (DSubReg_i16_reg imm:$lane))),
3433 (SubReg_i16_lane imm:$lane)))>;
3434 def : Pat<(v4i32 (mul (v4i32 QPR:$src1),
3435 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
3436 (v4i32 (VMULslv4i32 (v4i32 QPR:$src1),
3437 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3438 (DSubReg_i32_reg imm:$lane))),
3439 (SubReg_i32_lane imm:$lane)))>;
3440 def : Pat<(v4f32 (fmul (v4f32 QPR:$src1),
3441 (v4f32 (NEONvduplane (v4f32 QPR:$src2), imm:$lane)))),
3442 (v4f32 (VMULslfq (v4f32 QPR:$src1),
3443 (v2f32 (EXTRACT_SUBREG QPR:$src2,
3444 (DSubReg_i32_reg imm:$lane))),
3445 (SubReg_i32_lane imm:$lane)))>;
3447 // VQDMULH : Vector Saturating Doubling Multiply Returning High Half
3448 defm VQDMULH : N3VInt_HS<0, 0, 0b1011, 0, N3RegFrm, IIC_VMULi16D, IIC_VMULi32D,
3449 IIC_VMULi16Q, IIC_VMULi32Q,
3450 "vqdmulh", "s", int_arm_neon_vqdmulh, 1>;
3451 defm VQDMULHsl: N3VIntSL_HS<0b1100, IIC_VMULi16D, IIC_VMULi32D,
3452 IIC_VMULi16Q, IIC_VMULi32Q,
3453 "vqdmulh", "s", int_arm_neon_vqdmulh>;
3454 def : Pat<(v8i16 (int_arm_neon_vqdmulh (v8i16 QPR:$src1),
3455 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
3457 (v8i16 (VQDMULHslv8i16 (v8i16 QPR:$src1),
3458 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3459 (DSubReg_i16_reg imm:$lane))),
3460 (SubReg_i16_lane imm:$lane)))>;
3461 def : Pat<(v4i32 (int_arm_neon_vqdmulh (v4i32 QPR:$src1),
3462 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
3464 (v4i32 (VQDMULHslv4i32 (v4i32 QPR:$src1),
3465 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3466 (DSubReg_i32_reg imm:$lane))),
3467 (SubReg_i32_lane imm:$lane)))>;
3469 // VQRDMULH : Vector Rounding Saturating Doubling Multiply Returning High Half
3470 defm VQRDMULH : N3VInt_HS<1, 0, 0b1011, 0, N3RegFrm,
3471 IIC_VMULi16D,IIC_VMULi32D,IIC_VMULi16Q,IIC_VMULi32Q,
3472 "vqrdmulh", "s", int_arm_neon_vqrdmulh, 1>;
3473 defm VQRDMULHsl : N3VIntSL_HS<0b1101, IIC_VMULi16D, IIC_VMULi32D,
3474 IIC_VMULi16Q, IIC_VMULi32Q,
3475 "vqrdmulh", "s", int_arm_neon_vqrdmulh>;
3476 def : Pat<(v8i16 (int_arm_neon_vqrdmulh (v8i16 QPR:$src1),
3477 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
3479 (v8i16 (VQRDMULHslv8i16 (v8i16 QPR:$src1),
3480 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3481 (DSubReg_i16_reg imm:$lane))),
3482 (SubReg_i16_lane imm:$lane)))>;
3483 def : Pat<(v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$src1),
3484 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
3486 (v4i32 (VQRDMULHslv4i32 (v4i32 QPR:$src1),
3487 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3488 (DSubReg_i32_reg imm:$lane))),
3489 (SubReg_i32_lane imm:$lane)))>;
3491 // VMULL : Vector Multiply Long (integer and polynomial) (Q = D * D)
3492 defm VMULLs : N3VL_QHS<0,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
3493 "vmull", "s", NEONvmulls, 1>;
3494 defm VMULLu : N3VL_QHS<1,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
3495 "vmull", "u", NEONvmullu, 1>;
3496 def VMULLp : N3VLInt<0, 1, 0b00, 0b1110, 0, IIC_VMULi16D, "vmull", "p8",
3497 v8i16, v8i8, int_arm_neon_vmullp, 1>;
3498 defm VMULLsls : N3VLSL_HS<0, 0b1010, IIC_VMULi16D, "vmull", "s", NEONvmulls>;
3499 defm VMULLslu : N3VLSL_HS<1, 0b1010, IIC_VMULi16D, "vmull", "u", NEONvmullu>;
3501 // VQDMULL : Vector Saturating Doubling Multiply Long (Q = D * D)
3502 defm VQDMULL : N3VLInt_HS<0,1,0b1101,0, IIC_VMULi16D, IIC_VMULi32D,
3503 "vqdmull", "s", int_arm_neon_vqdmull, 1>;
3504 defm VQDMULLsl: N3VLIntSL_HS<0, 0b1011, IIC_VMULi16D,
3505 "vqdmull", "s", int_arm_neon_vqdmull>;
3507 // Vector Multiply-Accumulate and Multiply-Subtract Operations.
3509 // VMLA : Vector Multiply Accumulate (integer and floating-point)
3510 defm VMLA : N3VMulOp_QHS<0, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3511 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
3512 def VMLAfd : N3VDMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACD, "vmla", "f32",
3513 v2f32, fmul_su, fadd_mlx>,
3514 Requires<[HasNEON, UseFPVMLx]>;
3515 def VMLAfq : N3VQMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACQ, "vmla", "f32",
3516 v4f32, fmul_su, fadd_mlx>,
3517 Requires<[HasNEON, UseFPVMLx]>;
3518 defm VMLAsl : N3VMulOpSL_HS<0b0000, IIC_VMACi16D, IIC_VMACi32D,
3519 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
3520 def VMLAslfd : N3VDMulOpSL<0b10, 0b0001, IIC_VMACD, "vmla", "f32",
3521 v2f32, fmul_su, fadd_mlx>,
3522 Requires<[HasNEON, UseFPVMLx]>;
3523 def VMLAslfq : N3VQMulOpSL<0b10, 0b0001, IIC_VMACQ, "vmla", "f32",
3524 v4f32, v2f32, fmul_su, fadd_mlx>,
3525 Requires<[HasNEON, UseFPVMLx]>;
3527 def : Pat<(v8i16 (add (v8i16 QPR:$src1),
3528 (mul (v8i16 QPR:$src2),
3529 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
3530 (v8i16 (VMLAslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
3531 (v4i16 (EXTRACT_SUBREG QPR:$src3,
3532 (DSubReg_i16_reg imm:$lane))),
3533 (SubReg_i16_lane imm:$lane)))>;
3535 def : Pat<(v4i32 (add (v4i32 QPR:$src1),
3536 (mul (v4i32 QPR:$src2),
3537 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
3538 (v4i32 (VMLAslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
3539 (v2i32 (EXTRACT_SUBREG QPR:$src3,
3540 (DSubReg_i32_reg imm:$lane))),
3541 (SubReg_i32_lane imm:$lane)))>;
3543 def : Pat<(v4f32 (fadd_mlx (v4f32 QPR:$src1),
3544 (fmul_su (v4f32 QPR:$src2),
3545 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
3546 (v4f32 (VMLAslfq (v4f32 QPR:$src1),
3548 (v2f32 (EXTRACT_SUBREG QPR:$src3,
3549 (DSubReg_i32_reg imm:$lane))),
3550 (SubReg_i32_lane imm:$lane)))>,
3551 Requires<[HasNEON, UseFPVMLx]>;
3553 // VMLAL : Vector Multiply Accumulate Long (Q += D * D)
3554 defm VMLALs : N3VLMulOp_QHS<0,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
3555 "vmlal", "s", NEONvmulls, add>;
3556 defm VMLALu : N3VLMulOp_QHS<1,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
3557 "vmlal", "u", NEONvmullu, add>;
3559 defm VMLALsls : N3VLMulOpSL_HS<0, 0b0010, "vmlal", "s", NEONvmulls, add>;
3560 defm VMLALslu : N3VLMulOpSL_HS<1, 0b0010, "vmlal", "u", NEONvmullu, add>;
3562 // VQDMLAL : Vector Saturating Doubling Multiply Accumulate Long (Q += D * D)
3563 defm VQDMLAL : N3VLInt3_HS<0, 1, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3564 "vqdmlal", "s", int_arm_neon_vqdmlal>;
3565 defm VQDMLALsl: N3VLInt3SL_HS<0, 0b0011, "vqdmlal", "s", int_arm_neon_vqdmlal>;
3567 // VMLS : Vector Multiply Subtract (integer and floating-point)
3568 defm VMLS : N3VMulOp_QHS<1, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3569 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
3570 def VMLSfd : N3VDMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACD, "vmls", "f32",
3571 v2f32, fmul_su, fsub_mlx>,
3572 Requires<[HasNEON, UseFPVMLx]>;
3573 def VMLSfq : N3VQMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACQ, "vmls", "f32",
3574 v4f32, fmul_su, fsub_mlx>,
3575 Requires<[HasNEON, UseFPVMLx]>;
3576 defm VMLSsl : N3VMulOpSL_HS<0b0100, IIC_VMACi16D, IIC_VMACi32D,
3577 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
3578 def VMLSslfd : N3VDMulOpSL<0b10, 0b0101, IIC_VMACD, "vmls", "f32",
3579 v2f32, fmul_su, fsub_mlx>,
3580 Requires<[HasNEON, UseFPVMLx]>;
3581 def VMLSslfq : N3VQMulOpSL<0b10, 0b0101, IIC_VMACQ, "vmls", "f32",
3582 v4f32, v2f32, fmul_su, fsub_mlx>,
3583 Requires<[HasNEON, UseFPVMLx]>;
3585 def : Pat<(v8i16 (sub (v8i16 QPR:$src1),
3586 (mul (v8i16 QPR:$src2),
3587 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
3588 (v8i16 (VMLSslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
3589 (v4i16 (EXTRACT_SUBREG QPR:$src3,
3590 (DSubReg_i16_reg imm:$lane))),
3591 (SubReg_i16_lane imm:$lane)))>;
3593 def : Pat<(v4i32 (sub (v4i32 QPR:$src1),
3594 (mul (v4i32 QPR:$src2),
3595 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
3596 (v4i32 (VMLSslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
3597 (v2i32 (EXTRACT_SUBREG QPR:$src3,
3598 (DSubReg_i32_reg imm:$lane))),
3599 (SubReg_i32_lane imm:$lane)))>;
3601 def : Pat<(v4f32 (fsub_mlx (v4f32 QPR:$src1),
3602 (fmul_su (v4f32 QPR:$src2),
3603 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
3604 (v4f32 (VMLSslfq (v4f32 QPR:$src1), (v4f32 QPR:$src2),
3605 (v2f32 (EXTRACT_SUBREG QPR:$src3,
3606 (DSubReg_i32_reg imm:$lane))),
3607 (SubReg_i32_lane imm:$lane)))>,
3608 Requires<[HasNEON, UseFPVMLx]>;
3610 // VMLSL : Vector Multiply Subtract Long (Q -= D * D)
3611 defm VMLSLs : N3VLMulOp_QHS<0,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
3612 "vmlsl", "s", NEONvmulls, sub>;
3613 defm VMLSLu : N3VLMulOp_QHS<1,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
3614 "vmlsl", "u", NEONvmullu, sub>;
3616 defm VMLSLsls : N3VLMulOpSL_HS<0, 0b0110, "vmlsl", "s", NEONvmulls, sub>;
3617 defm VMLSLslu : N3VLMulOpSL_HS<1, 0b0110, "vmlsl", "u", NEONvmullu, sub>;
3619 // VQDMLSL : Vector Saturating Doubling Multiply Subtract Long (Q -= D * D)
3620 defm VQDMLSL : N3VLInt3_HS<0, 1, 0b1011, 0, IIC_VMACi16D, IIC_VMACi32D,
3621 "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
3622 defm VQDMLSLsl: N3VLInt3SL_HS<0, 0b111, "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
3624 // Vector Subtract Operations.
3626 // VSUB : Vector Subtract (integer and floating-point)
3627 defm VSUB : N3V_QHSD<1, 0, 0b1000, 0, IIC_VSUBiD, IIC_VSUBiQ,
3628 "vsub", "i", sub, 0>;
3629 def VSUBfd : N3VD<0, 0, 0b10, 0b1101, 0, IIC_VBIND, "vsub", "f32",
3630 v2f32, v2f32, fsub, 0>;
3631 def VSUBfq : N3VQ<0, 0, 0b10, 0b1101, 0, IIC_VBINQ, "vsub", "f32",
3632 v4f32, v4f32, fsub, 0>;
3633 // VSUBL : Vector Subtract Long (Q = D - D)
3634 defm VSUBLs : N3VLExt_QHS<0,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
3635 "vsubl", "s", sub, sext, 0>;
3636 defm VSUBLu : N3VLExt_QHS<1,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
3637 "vsubl", "u", sub, zext, 0>;
3638 // VSUBW : Vector Subtract Wide (Q = Q - D)
3639 defm VSUBWs : N3VW_QHS<0,1,0b0011,0, "vsubw", "s", sub, sext, 0>;
3640 defm VSUBWu : N3VW_QHS<1,1,0b0011,0, "vsubw", "u", sub, zext, 0>;
3641 // VHSUB : Vector Halving Subtract
3642 defm VHSUBs : N3VInt_QHS<0, 0, 0b0010, 0, N3RegFrm,
3643 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3644 "vhsub", "s", int_arm_neon_vhsubs, 0>;
3645 defm VHSUBu : N3VInt_QHS<1, 0, 0b0010, 0, N3RegFrm,
3646 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3647 "vhsub", "u", int_arm_neon_vhsubu, 0>;
3648 // VQSUB : Vector Saturing Subtract
3649 defm VQSUBs : N3VInt_QHSD<0, 0, 0b0010, 1, N3RegFrm,
3650 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3651 "vqsub", "s", int_arm_neon_vqsubs, 0>;
3652 defm VQSUBu : N3VInt_QHSD<1, 0, 0b0010, 1, N3RegFrm,
3653 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3654 "vqsub", "u", int_arm_neon_vqsubu, 0>;
3655 // VSUBHN : Vector Subtract and Narrow Returning High Half (D = Q - Q)
3656 defm VSUBHN : N3VNInt_HSD<0,1,0b0110,0, "vsubhn", "i",
3657 int_arm_neon_vsubhn, 0>;
3658 // VRSUBHN : Vector Rounding Subtract and Narrow Returning High Half (D=Q-Q)
3659 defm VRSUBHN : N3VNInt_HSD<1,1,0b0110,0, "vrsubhn", "i",
3660 int_arm_neon_vrsubhn, 0>;
3662 // Vector Comparisons.
3664 // VCEQ : Vector Compare Equal
3665 defm VCEQ : N3V_QHS<1, 0, 0b1000, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3666 IIC_VSUBi4Q, "vceq", "i", NEONvceq, 1>;
3667 def VCEQfd : N3VD<0,0,0b00,0b1110,0, IIC_VBIND, "vceq", "f32", v2i32, v2f32,
3669 def VCEQfq : N3VQ<0,0,0b00,0b1110,0, IIC_VBINQ, "vceq", "f32", v4i32, v4f32,
3672 defm VCEQz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00010, 0, "vceq", "i",
3673 "$Vd, $Vm, #0", NEONvceqz>;
3675 // VCGE : Vector Compare Greater Than or Equal
3676 defm VCGEs : N3V_QHS<0, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3677 IIC_VSUBi4Q, "vcge", "s", NEONvcge, 0>;
3678 defm VCGEu : N3V_QHS<1, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3679 IIC_VSUBi4Q, "vcge", "u", NEONvcgeu, 0>;
3680 def VCGEfd : N3VD<1,0,0b00,0b1110,0, IIC_VBIND, "vcge", "f32", v2i32, v2f32,
3682 def VCGEfq : N3VQ<1,0,0b00,0b1110,0, IIC_VBINQ, "vcge", "f32", v4i32, v4f32,
3685 defm VCGEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00001, 0, "vcge", "s",
3686 "$Vd, $Vm, #0", NEONvcgez>;
3687 defm VCLEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00011, 0, "vcle", "s",
3688 "$Vd, $Vm, #0", NEONvclez>;
3690 // VCGT : Vector Compare Greater Than
3691 defm VCGTs : N3V_QHS<0, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3692 IIC_VSUBi4Q, "vcgt", "s", NEONvcgt, 0>;
3693 defm VCGTu : N3V_QHS<1, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3694 IIC_VSUBi4Q, "vcgt", "u", NEONvcgtu, 0>;
3695 def VCGTfd : N3VD<1,0,0b10,0b1110,0, IIC_VBIND, "vcgt", "f32", v2i32, v2f32,
3697 def VCGTfq : N3VQ<1,0,0b10,0b1110,0, IIC_VBINQ, "vcgt", "f32", v4i32, v4f32,
3700 defm VCGTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00000, 0, "vcgt", "s",
3701 "$Vd, $Vm, #0", NEONvcgtz>;
3702 defm VCLTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00100, 0, "vclt", "s",
3703 "$Vd, $Vm, #0", NEONvcltz>;
3705 // VACGE : Vector Absolute Compare Greater Than or Equal (aka VCAGE)
3706 def VACGEd : N3VDInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacge",
3707 "f32", v2i32, v2f32, int_arm_neon_vacged, 0>;
3708 def VACGEq : N3VQInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacge",
3709 "f32", v4i32, v4f32, int_arm_neon_vacgeq, 0>;
3710 // VACGT : Vector Absolute Compare Greater Than (aka VCAGT)
3711 def VACGTd : N3VDInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacgt",
3712 "f32", v2i32, v2f32, int_arm_neon_vacgtd, 0>;
3713 def VACGTq : N3VQInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacgt",
3714 "f32", v4i32, v4f32, int_arm_neon_vacgtq, 0>;
3715 // VTST : Vector Test Bits
3716 defm VTST : N3V_QHS<0, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
3717 IIC_VBINi4Q, "vtst", "", NEONvtst, 1>;
3719 // Vector Bitwise Operations.
3721 def vnotd : PatFrag<(ops node:$in),
3722 (xor node:$in, (bitconvert (v8i8 NEONimmAllOnesV)))>;
3723 def vnotq : PatFrag<(ops node:$in),
3724 (xor node:$in, (bitconvert (v16i8 NEONimmAllOnesV)))>;
3727 // VAND : Vector Bitwise AND
3728 def VANDd : N3VDX<0, 0, 0b00, 0b0001, 1, IIC_VBINiD, "vand",
3729 v2i32, v2i32, and, 1>;
3730 def VANDq : N3VQX<0, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "vand",
3731 v4i32, v4i32, and, 1>;
3733 // VEOR : Vector Bitwise Exclusive OR
3734 def VEORd : N3VDX<1, 0, 0b00, 0b0001, 1, IIC_VBINiD, "veor",
3735 v2i32, v2i32, xor, 1>;
3736 def VEORq : N3VQX<1, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "veor",
3737 v4i32, v4i32, xor, 1>;
3739 // VORR : Vector Bitwise OR
3740 def VORRd : N3VDX<0, 0, 0b10, 0b0001, 1, IIC_VBINiD, "vorr",
3741 v2i32, v2i32, or, 1>;
3742 def VORRq : N3VQX<0, 0, 0b10, 0b0001, 1, IIC_VBINiQ, "vorr",
3743 v4i32, v4i32, or, 1>;
3745 def VORRiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 0, 1,
3746 (outs DPR:$Vd), (ins nModImm:$SIMM, DPR:$src),
3748 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
3750 (v4i16 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
3751 let Inst{9} = SIMM{9};
3754 def VORRiv2i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 0, 0, 1,
3755 (outs DPR:$Vd), (ins nModImm:$SIMM, DPR:$src),
3757 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
3759 (v2i32 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
3760 let Inst{10-9} = SIMM{10-9};
3763 def VORRiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 0, 1,
3764 (outs QPR:$Vd), (ins nModImm:$SIMM, QPR:$src),
3766 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
3768 (v8i16 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
3769 let Inst{9} = SIMM{9};
3772 def VORRiv4i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 1, 0, 1,
3773 (outs QPR:$Vd), (ins nModImm:$SIMM, QPR:$src),
3775 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
3777 (v4i32 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
3778 let Inst{10-9} = SIMM{10-9};
3782 // VBIC : Vector Bitwise Bit Clear (AND NOT)
3783 def VBICd : N3VX<0, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
3784 (ins DPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VBINiD,
3785 "vbic", "$Vd, $Vn, $Vm", "",
3786 [(set DPR:$Vd, (v2i32 (and DPR:$Vn,
3787 (vnotd DPR:$Vm))))]>;
3788 def VBICq : N3VX<0, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
3789 (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINiQ,
3790 "vbic", "$Vd, $Vn, $Vm", "",
3791 [(set QPR:$Vd, (v4i32 (and QPR:$Vn,
3792 (vnotq QPR:$Vm))))]>;
3794 def VBICiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 1, 1,
3795 (outs DPR:$Vd), (ins nModImm:$SIMM, DPR:$src),
3797 "vbic", "i16", "$Vd, $SIMM", "$src = $Vd",
3799 (v4i16 (NEONvbicImm DPR:$src, timm:$SIMM)))]> {
3800 let Inst{9} = SIMM{9};
3803 def VBICiv2i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 0, 1, 1,
3804 (outs DPR:$Vd), (ins nModImm:$SIMM, DPR:$src),
3806 "vbic", "i32", "$Vd, $SIMM", "$src = $Vd",
3808 (v2i32 (NEONvbicImm DPR:$src, timm:$SIMM)))]> {
3809 let Inst{10-9} = SIMM{10-9};
3812 def VBICiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 1, 1,
3813 (outs QPR:$Vd), (ins nModImm:$SIMM, QPR:$src),
3815 "vbic", "i16", "$Vd, $SIMM", "$src = $Vd",
3817 (v8i16 (NEONvbicImm QPR:$src, timm:$SIMM)))]> {
3818 let Inst{9} = SIMM{9};
3821 def VBICiv4i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 1, 1, 1,
3822 (outs QPR:$Vd), (ins nModImm:$SIMM, QPR:$src),
3824 "vbic", "i32", "$Vd, $SIMM", "$src = $Vd",
3826 (v4i32 (NEONvbicImm QPR:$src, timm:$SIMM)))]> {
3827 let Inst{10-9} = SIMM{10-9};
3830 // VORN : Vector Bitwise OR NOT
3831 def VORNd : N3VX<0, 0, 0b11, 0b0001, 0, 1, (outs DPR:$Vd),
3832 (ins DPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VBINiD,
3833 "vorn", "$Vd, $Vn, $Vm", "",
3834 [(set DPR:$Vd, (v2i32 (or DPR:$Vn,
3835 (vnotd DPR:$Vm))))]>;
3836 def VORNq : N3VX<0, 0, 0b11, 0b0001, 1, 1, (outs QPR:$Vd),
3837 (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINiQ,
3838 "vorn", "$Vd, $Vn, $Vm", "",
3839 [(set QPR:$Vd, (v4i32 (or QPR:$Vn,
3840 (vnotq QPR:$Vm))))]>;
3842 // VMVN : Vector Bitwise NOT (Immediate)
3844 let isReMaterializable = 1 in {
3846 def VMVNv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 1, 1, (outs DPR:$Vd),
3847 (ins nModImm:$SIMM), IIC_VMOVImm,
3848 "vmvn", "i16", "$Vd, $SIMM", "",
3849 [(set DPR:$Vd, (v4i16 (NEONvmvnImm timm:$SIMM)))]> {
3850 let Inst{9} = SIMM{9};
3853 def VMVNv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 1, 1, (outs QPR:$Vd),
3854 (ins nModImm:$SIMM), IIC_VMOVImm,
3855 "vmvn", "i16", "$Vd, $SIMM", "",
3856 [(set QPR:$Vd, (v8i16 (NEONvmvnImm timm:$SIMM)))]> {
3857 let Inst{9} = SIMM{9};
3860 def VMVNv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 1, 1, (outs DPR:$Vd),
3861 (ins nModImm:$SIMM), IIC_VMOVImm,
3862 "vmvn", "i32", "$Vd, $SIMM", "",
3863 [(set DPR:$Vd, (v2i32 (NEONvmvnImm timm:$SIMM)))]> {
3864 let Inst{11-8} = SIMM{11-8};
3867 def VMVNv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 1, 1, (outs QPR:$Vd),
3868 (ins nModImm:$SIMM), IIC_VMOVImm,
3869 "vmvn", "i32", "$Vd, $SIMM", "",
3870 [(set QPR:$Vd, (v4i32 (NEONvmvnImm timm:$SIMM)))]> {
3871 let Inst{11-8} = SIMM{11-8};
3875 // VMVN : Vector Bitwise NOT
3876 def VMVNd : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 0, 0,
3877 (outs DPR:$Vd), (ins DPR:$Vm), IIC_VSUBiD,
3878 "vmvn", "$Vd, $Vm", "",
3879 [(set DPR:$Vd, (v2i32 (vnotd DPR:$Vm)))]>;
3880 def VMVNq : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 1, 0,
3881 (outs QPR:$Vd), (ins QPR:$Vm), IIC_VSUBiD,
3882 "vmvn", "$Vd, $Vm", "",
3883 [(set QPR:$Vd, (v4i32 (vnotq QPR:$Vm)))]>;
3884 def : Pat<(v2i32 (vnotd DPR:$src)), (VMVNd DPR:$src)>;
3885 def : Pat<(v4i32 (vnotq QPR:$src)), (VMVNq QPR:$src)>;
3887 // VBSL : Vector Bitwise Select
3888 def VBSLd : N3VX<1, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
3889 (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
3890 N3RegFrm, IIC_VCNTiD,
3891 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3893 (v2i32 (NEONvbsl DPR:$src1, DPR:$Vn, DPR:$Vm)))]>;
3895 def : Pat<(v2i32 (or (and DPR:$Vn, DPR:$Vd),
3896 (and DPR:$Vm, (vnotd DPR:$Vd)))),
3897 (VBSLd DPR:$Vd, DPR:$Vn, DPR:$Vm)>;
3899 def VBSLq : N3VX<1, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
3900 (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
3901 N3RegFrm, IIC_VCNTiQ,
3902 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3904 (v4i32 (NEONvbsl QPR:$src1, QPR:$Vn, QPR:$Vm)))]>;
3906 def : Pat<(v4i32 (or (and QPR:$Vn, QPR:$Vd),
3907 (and QPR:$Vm, (vnotq QPR:$Vd)))),
3908 (VBSLq QPR:$Vd, QPR:$Vn, QPR:$Vm)>;
3910 // VBIF : Vector Bitwise Insert if False
3911 // like VBSL but with: "vbif $dst, $src3, $src1", "$src2 = $dst",
3912 // FIXME: This instruction's encoding MAY NOT BE correct.
3913 def VBIFd : N3VX<1, 0, 0b11, 0b0001, 0, 1,
3914 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
3915 N3RegFrm, IIC_VBINiD,
3916 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3917 [/* For disassembly only; pattern left blank */]>;
3918 def VBIFq : N3VX<1, 0, 0b11, 0b0001, 1, 1,
3919 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
3920 N3RegFrm, IIC_VBINiQ,
3921 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3922 [/* For disassembly only; pattern left blank */]>;
3924 // VBIT : Vector Bitwise Insert if True
3925 // like VBSL but with: "vbit $dst, $src2, $src1", "$src3 = $dst",
3926 // FIXME: This instruction's encoding MAY NOT BE correct.
3927 def VBITd : N3VX<1, 0, 0b10, 0b0001, 0, 1,
3928 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
3929 N3RegFrm, IIC_VBINiD,
3930 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3931 [/* For disassembly only; pattern left blank */]>;
3932 def VBITq : N3VX<1, 0, 0b10, 0b0001, 1, 1,
3933 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
3934 N3RegFrm, IIC_VBINiQ,
3935 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3936 [/* For disassembly only; pattern left blank */]>;
3938 // VBIT/VBIF are not yet implemented. The TwoAddress pass will not go looking
3939 // for equivalent operations with different register constraints; it just
3942 // Vector Absolute Differences.
3944 // VABD : Vector Absolute Difference
3945 defm VABDs : N3VInt_QHS<0, 0, 0b0111, 0, N3RegFrm,
3946 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3947 "vabd", "s", int_arm_neon_vabds, 1>;
3948 defm VABDu : N3VInt_QHS<1, 0, 0b0111, 0, N3RegFrm,
3949 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3950 "vabd", "u", int_arm_neon_vabdu, 1>;
3951 def VABDfd : N3VDInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBIND,
3952 "vabd", "f32", v2f32, v2f32, int_arm_neon_vabds, 1>;
3953 def VABDfq : N3VQInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBINQ,
3954 "vabd", "f32", v4f32, v4f32, int_arm_neon_vabds, 1>;
3956 // VABDL : Vector Absolute Difference Long (Q = | D - D |)
3957 defm VABDLs : N3VLIntExt_QHS<0,1,0b0111,0, IIC_VSUBi4Q,
3958 "vabdl", "s", int_arm_neon_vabds, zext, 1>;
3959 defm VABDLu : N3VLIntExt_QHS<1,1,0b0111,0, IIC_VSUBi4Q,
3960 "vabdl", "u", int_arm_neon_vabdu, zext, 1>;
3962 // VABA : Vector Absolute Difference and Accumulate
3963 defm VABAs : N3VIntOp_QHS<0,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
3964 "vaba", "s", int_arm_neon_vabds, add>;
3965 defm VABAu : N3VIntOp_QHS<1,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
3966 "vaba", "u", int_arm_neon_vabdu, add>;
3968 // VABAL : Vector Absolute Difference and Accumulate Long (Q += | D - D |)
3969 defm VABALs : N3VLIntExtOp_QHS<0,1,0b0101,0, IIC_VABAD,
3970 "vabal", "s", int_arm_neon_vabds, zext, add>;
3971 defm VABALu : N3VLIntExtOp_QHS<1,1,0b0101,0, IIC_VABAD,
3972 "vabal", "u", int_arm_neon_vabdu, zext, add>;
3974 // Vector Maximum and Minimum.
3976 // VMAX : Vector Maximum
3977 defm VMAXs : N3VInt_QHS<0, 0, 0b0110, 0, N3RegFrm,
3978 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3979 "vmax", "s", int_arm_neon_vmaxs, 1>;
3980 defm VMAXu : N3VInt_QHS<1, 0, 0b0110, 0, N3RegFrm,
3981 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3982 "vmax", "u", int_arm_neon_vmaxu, 1>;
3983 def VMAXfd : N3VDInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBIND,
3985 v2f32, v2f32, int_arm_neon_vmaxs, 1>;
3986 def VMAXfq : N3VQInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBINQ,
3988 v4f32, v4f32, int_arm_neon_vmaxs, 1>;
3990 // VMIN : Vector Minimum
3991 defm VMINs : N3VInt_QHS<0, 0, 0b0110, 1, N3RegFrm,
3992 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3993 "vmin", "s", int_arm_neon_vmins, 1>;
3994 defm VMINu : N3VInt_QHS<1, 0, 0b0110, 1, N3RegFrm,
3995 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3996 "vmin", "u", int_arm_neon_vminu, 1>;
3997 def VMINfd : N3VDInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBIND,
3999 v2f32, v2f32, int_arm_neon_vmins, 1>;
4000 def VMINfq : N3VQInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBINQ,
4002 v4f32, v4f32, int_arm_neon_vmins, 1>;
4004 // Vector Pairwise Operations.
4006 // VPADD : Vector Pairwise Add
4007 def VPADDi8 : N3VDInt<0, 0, 0b00, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4009 v8i8, v8i8, int_arm_neon_vpadd, 0>;
4010 def VPADDi16 : N3VDInt<0, 0, 0b01, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4012 v4i16, v4i16, int_arm_neon_vpadd, 0>;
4013 def VPADDi32 : N3VDInt<0, 0, 0b10, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4015 v2i32, v2i32, int_arm_neon_vpadd, 0>;
4016 def VPADDf : N3VDInt<1, 0, 0b00, 0b1101, 0, N3RegFrm,
4017 IIC_VPBIND, "vpadd", "f32",
4018 v2f32, v2f32, int_arm_neon_vpadd, 0>;
4020 // VPADDL : Vector Pairwise Add Long
4021 defm VPADDLs : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00100, 0, "vpaddl", "s",
4022 int_arm_neon_vpaddls>;
4023 defm VPADDLu : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00101, 0, "vpaddl", "u",
4024 int_arm_neon_vpaddlu>;
4026 // VPADAL : Vector Pairwise Add and Accumulate Long
4027 defm VPADALs : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01100, 0, "vpadal", "s",
4028 int_arm_neon_vpadals>;
4029 defm VPADALu : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01101, 0, "vpadal", "u",
4030 int_arm_neon_vpadalu>;
4032 // VPMAX : Vector Pairwise Maximum
4033 def VPMAXs8 : N3VDInt<0, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4034 "s8", v8i8, v8i8, int_arm_neon_vpmaxs, 0>;
4035 def VPMAXs16 : N3VDInt<0, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4036 "s16", v4i16, v4i16, int_arm_neon_vpmaxs, 0>;
4037 def VPMAXs32 : N3VDInt<0, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4038 "s32", v2i32, v2i32, int_arm_neon_vpmaxs, 0>;
4039 def VPMAXu8 : N3VDInt<1, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4040 "u8", v8i8, v8i8, int_arm_neon_vpmaxu, 0>;
4041 def VPMAXu16 : N3VDInt<1, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4042 "u16", v4i16, v4i16, int_arm_neon_vpmaxu, 0>;
4043 def VPMAXu32 : N3VDInt<1, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4044 "u32", v2i32, v2i32, int_arm_neon_vpmaxu, 0>;
4045 def VPMAXf : N3VDInt<1, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmax",
4046 "f32", v2f32, v2f32, int_arm_neon_vpmaxs, 0>;
4048 // VPMIN : Vector Pairwise Minimum
4049 def VPMINs8 : N3VDInt<0, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4050 "s8", v8i8, v8i8, int_arm_neon_vpmins, 0>;
4051 def VPMINs16 : N3VDInt<0, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4052 "s16", v4i16, v4i16, int_arm_neon_vpmins, 0>;
4053 def VPMINs32 : N3VDInt<0, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4054 "s32", v2i32, v2i32, int_arm_neon_vpmins, 0>;
4055 def VPMINu8 : N3VDInt<1, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4056 "u8", v8i8, v8i8, int_arm_neon_vpminu, 0>;
4057 def VPMINu16 : N3VDInt<1, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4058 "u16", v4i16, v4i16, int_arm_neon_vpminu, 0>;
4059 def VPMINu32 : N3VDInt<1, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4060 "u32", v2i32, v2i32, int_arm_neon_vpminu, 0>;
4061 def VPMINf : N3VDInt<1, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmin",
4062 "f32", v2f32, v2f32, int_arm_neon_vpmins, 0>;
4064 // Vector Reciprocal and Reciprocal Square Root Estimate and Step.
4066 // VRECPE : Vector Reciprocal Estimate
4067 def VRECPEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
4068 IIC_VUNAD, "vrecpe", "u32",
4069 v2i32, v2i32, int_arm_neon_vrecpe>;
4070 def VRECPEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
4071 IIC_VUNAQ, "vrecpe", "u32",
4072 v4i32, v4i32, int_arm_neon_vrecpe>;
4073 def VRECPEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
4074 IIC_VUNAD, "vrecpe", "f32",
4075 v2f32, v2f32, int_arm_neon_vrecpe>;
4076 def VRECPEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
4077 IIC_VUNAQ, "vrecpe", "f32",
4078 v4f32, v4f32, int_arm_neon_vrecpe>;
4080 // VRECPS : Vector Reciprocal Step
4081 def VRECPSfd : N3VDInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
4082 IIC_VRECSD, "vrecps", "f32",
4083 v2f32, v2f32, int_arm_neon_vrecps, 1>;
4084 def VRECPSfq : N3VQInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
4085 IIC_VRECSQ, "vrecps", "f32",
4086 v4f32, v4f32, int_arm_neon_vrecps, 1>;
4088 // VRSQRTE : Vector Reciprocal Square Root Estimate
4089 def VRSQRTEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
4090 IIC_VUNAD, "vrsqrte", "u32",
4091 v2i32, v2i32, int_arm_neon_vrsqrte>;
4092 def VRSQRTEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
4093 IIC_VUNAQ, "vrsqrte", "u32",
4094 v4i32, v4i32, int_arm_neon_vrsqrte>;
4095 def VRSQRTEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
4096 IIC_VUNAD, "vrsqrte", "f32",
4097 v2f32, v2f32, int_arm_neon_vrsqrte>;
4098 def VRSQRTEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
4099 IIC_VUNAQ, "vrsqrte", "f32",
4100 v4f32, v4f32, int_arm_neon_vrsqrte>;
4102 // VRSQRTS : Vector Reciprocal Square Root Step
4103 def VRSQRTSfd : N3VDInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
4104 IIC_VRECSD, "vrsqrts", "f32",
4105 v2f32, v2f32, int_arm_neon_vrsqrts, 1>;
4106 def VRSQRTSfq : N3VQInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
4107 IIC_VRECSQ, "vrsqrts", "f32",
4108 v4f32, v4f32, int_arm_neon_vrsqrts, 1>;
4112 // VSHL : Vector Shift
4113 defm VSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 0, N3RegVShFrm,
4114 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
4115 "vshl", "s", int_arm_neon_vshifts>;
4116 defm VSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 0, N3RegVShFrm,
4117 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
4118 "vshl", "u", int_arm_neon_vshiftu>;
4120 // VSHL : Vector Shift Left (Immediate)
4121 defm VSHLi : N2VShL_QHSD<0, 1, 0b0101, 1, IIC_VSHLiD, "vshl", "i", NEONvshl>;
4123 // VSHR : Vector Shift Right (Immediate)
4124 defm VSHRs : N2VShR_QHSD<0, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "s",NEONvshrs>;
4125 defm VSHRu : N2VShR_QHSD<1, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "u",NEONvshru>;
4127 // VSHLL : Vector Shift Left Long
4128 defm VSHLLs : N2VLSh_QHS<0, 1, 0b1010, 0, 0, 1, "vshll", "s", NEONvshlls>;
4129 defm VSHLLu : N2VLSh_QHS<1, 1, 0b1010, 0, 0, 1, "vshll", "u", NEONvshllu>;
4131 // VSHLL : Vector Shift Left Long (with maximum shift count)
4132 class N2VLShMax<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
4133 bit op6, bit op4, string OpcodeStr, string Dt, ValueType ResTy,
4134 ValueType OpTy, SDNode OpNode>
4135 : N2VLSh<op24, op23, op11_8, op7, op6, op4, OpcodeStr, Dt,
4136 ResTy, OpTy, OpNode> {
4137 let Inst{21-16} = op21_16;
4138 let DecoderMethod = "DecodeVSHLMaxInstruction";
4140 def VSHLLi8 : N2VLShMax<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll", "i8",
4141 v8i16, v8i8, NEONvshlli>;
4142 def VSHLLi16 : N2VLShMax<1, 1, 0b110110, 0b0011, 0, 0, 0, "vshll", "i16",
4143 v4i32, v4i16, NEONvshlli>;
4144 def VSHLLi32 : N2VLShMax<1, 1, 0b111010, 0b0011, 0, 0, 0, "vshll", "i32",
4145 v2i64, v2i32, NEONvshlli>;
4147 // VSHRN : Vector Shift Right and Narrow
4148 defm VSHRN : N2VNSh_HSD<0,1,0b1000,0,0,1, IIC_VSHLiD, "vshrn", "i",
4151 // VRSHL : Vector Rounding Shift
4152 defm VRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 0, N3RegVShFrm,
4153 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4154 "vrshl", "s", int_arm_neon_vrshifts>;
4155 defm VRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 0, N3RegVShFrm,
4156 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4157 "vrshl", "u", int_arm_neon_vrshiftu>;
4158 // VRSHR : Vector Rounding Shift Right
4159 defm VRSHRs : N2VShR_QHSD<0,1,0b0010,1, IIC_VSHLi4D, "vrshr", "s",NEONvrshrs>;
4160 defm VRSHRu : N2VShR_QHSD<1,1,0b0010,1, IIC_VSHLi4D, "vrshr", "u",NEONvrshru>;
4162 // VRSHRN : Vector Rounding Shift Right and Narrow
4163 defm VRSHRN : N2VNSh_HSD<0, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vrshrn", "i",
4166 // VQSHL : Vector Saturating Shift
4167 defm VQSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 1, N3RegVShFrm,
4168 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4169 "vqshl", "s", int_arm_neon_vqshifts>;
4170 defm VQSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 1, N3RegVShFrm,
4171 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4172 "vqshl", "u", int_arm_neon_vqshiftu>;
4173 // VQSHL : Vector Saturating Shift Left (Immediate)
4174 defm VQSHLsi : N2VShL_QHSD<0,1,0b0111,1, IIC_VSHLi4D, "vqshl", "s",NEONvqshls>;
4175 defm VQSHLui : N2VShL_QHSD<1,1,0b0111,1, IIC_VSHLi4D, "vqshl", "u",NEONvqshlu>;
4177 // VQSHLU : Vector Saturating Shift Left (Immediate, Unsigned)
4178 defm VQSHLsu : N2VShL_QHSD<1,1,0b0110,1, IIC_VSHLi4D,"vqshlu","s",NEONvqshlsu>;
4180 // VQSHRN : Vector Saturating Shift Right and Narrow
4181 defm VQSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "s",
4183 defm VQSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "u",
4186 // VQSHRUN : Vector Saturating Shift Right and Narrow (Unsigned)
4187 defm VQSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 0, 1, IIC_VSHLi4D, "vqshrun", "s",
4190 // VQRSHL : Vector Saturating Rounding Shift
4191 defm VQRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 1, N3RegVShFrm,
4192 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4193 "vqrshl", "s", int_arm_neon_vqrshifts>;
4194 defm VQRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 1, N3RegVShFrm,
4195 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4196 "vqrshl", "u", int_arm_neon_vqrshiftu>;
4198 // VQRSHRN : Vector Saturating Rounding Shift Right and Narrow
4199 defm VQRSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "s",
4201 defm VQRSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "u",
4204 // VQRSHRUN : Vector Saturating Rounding Shift Right and Narrow (Unsigned)
4205 defm VQRSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vqrshrun", "s",
4208 // VSRA : Vector Shift Right and Accumulate
4209 defm VSRAs : N2VShAdd_QHSD<0, 1, 0b0001, 1, "vsra", "s", NEONvshrs>;
4210 defm VSRAu : N2VShAdd_QHSD<1, 1, 0b0001, 1, "vsra", "u", NEONvshru>;
4211 // VRSRA : Vector Rounding Shift Right and Accumulate
4212 defm VRSRAs : N2VShAdd_QHSD<0, 1, 0b0011, 1, "vrsra", "s", NEONvrshrs>;
4213 defm VRSRAu : N2VShAdd_QHSD<1, 1, 0b0011, 1, "vrsra", "u", NEONvrshru>;
4215 // VSLI : Vector Shift Left and Insert
4216 defm VSLI : N2VShInsL_QHSD<1, 1, 0b0101, 1, "vsli">;
4218 // VSRI : Vector Shift Right and Insert
4219 defm VSRI : N2VShInsR_QHSD<1, 1, 0b0100, 1, "vsri">;
4221 // Vector Absolute and Saturating Absolute.
4223 // VABS : Vector Absolute Value
4224 defm VABS : N2VInt_QHS<0b11, 0b11, 0b01, 0b00110, 0,
4225 IIC_VUNAiD, IIC_VUNAiQ, "vabs", "s",
4227 def VABSfd : N2VDInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
4228 IIC_VUNAD, "vabs", "f32",
4229 v2f32, v2f32, int_arm_neon_vabs>;
4230 def VABSfq : N2VQInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
4231 IIC_VUNAQ, "vabs", "f32",
4232 v4f32, v4f32, int_arm_neon_vabs>;
4234 // VQABS : Vector Saturating Absolute Value
4235 defm VQABS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01110, 0,
4236 IIC_VQUNAiD, IIC_VQUNAiQ, "vqabs", "s",
4237 int_arm_neon_vqabs>;
4241 def vnegd : PatFrag<(ops node:$in),
4242 (sub (bitconvert (v2i32 NEONimmAllZerosV)), node:$in)>;
4243 def vnegq : PatFrag<(ops node:$in),
4244 (sub (bitconvert (v4i32 NEONimmAllZerosV)), node:$in)>;
4246 class VNEGD<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
4247 : N2V<0b11, 0b11, size, 0b01, 0b00111, 0, 0, (outs DPR:$Vd), (ins DPR:$Vm),
4248 IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
4249 [(set DPR:$Vd, (Ty (vnegd DPR:$Vm)))]>;
4250 class VNEGQ<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
4251 : N2V<0b11, 0b11, size, 0b01, 0b00111, 1, 0, (outs QPR:$Vd), (ins QPR:$Vm),
4252 IIC_VSHLiQ, OpcodeStr, Dt, "$Vd, $Vm", "",
4253 [(set QPR:$Vd, (Ty (vnegq QPR:$Vm)))]>;
4255 // VNEG : Vector Negate (integer)
4256 def VNEGs8d : VNEGD<0b00, "vneg", "s8", v8i8>;
4257 def VNEGs16d : VNEGD<0b01, "vneg", "s16", v4i16>;
4258 def VNEGs32d : VNEGD<0b10, "vneg", "s32", v2i32>;
4259 def VNEGs8q : VNEGQ<0b00, "vneg", "s8", v16i8>;
4260 def VNEGs16q : VNEGQ<0b01, "vneg", "s16", v8i16>;
4261 def VNEGs32q : VNEGQ<0b10, "vneg", "s32", v4i32>;
4263 // VNEG : Vector Negate (floating-point)
4264 def VNEGfd : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
4265 (outs DPR:$Vd), (ins DPR:$Vm), IIC_VUNAD,
4266 "vneg", "f32", "$Vd, $Vm", "",
4267 [(set DPR:$Vd, (v2f32 (fneg DPR:$Vm)))]>;
4268 def VNEGf32q : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 1, 0,
4269 (outs QPR:$Vd), (ins QPR:$Vm), IIC_VUNAQ,
4270 "vneg", "f32", "$Vd, $Vm", "",
4271 [(set QPR:$Vd, (v4f32 (fneg QPR:$Vm)))]>;
4273 def : Pat<(v8i8 (vnegd DPR:$src)), (VNEGs8d DPR:$src)>;
4274 def : Pat<(v4i16 (vnegd DPR:$src)), (VNEGs16d DPR:$src)>;
4275 def : Pat<(v2i32 (vnegd DPR:$src)), (VNEGs32d DPR:$src)>;
4276 def : Pat<(v16i8 (vnegq QPR:$src)), (VNEGs8q QPR:$src)>;
4277 def : Pat<(v8i16 (vnegq QPR:$src)), (VNEGs16q QPR:$src)>;
4278 def : Pat<(v4i32 (vnegq QPR:$src)), (VNEGs32q QPR:$src)>;
4280 // VQNEG : Vector Saturating Negate
4281 defm VQNEG : N2VInt_QHS<0b11, 0b11, 0b00, 0b01111, 0,
4282 IIC_VQUNAiD, IIC_VQUNAiQ, "vqneg", "s",
4283 int_arm_neon_vqneg>;
4285 // Vector Bit Counting Operations.
4287 // VCLS : Vector Count Leading Sign Bits
4288 defm VCLS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01000, 0,
4289 IIC_VCNTiD, IIC_VCNTiQ, "vcls", "s",
4291 // VCLZ : Vector Count Leading Zeros
4292 defm VCLZ : N2VInt_QHS<0b11, 0b11, 0b00, 0b01001, 0,
4293 IIC_VCNTiD, IIC_VCNTiQ, "vclz", "i",
4295 // VCNT : Vector Count One Bits
4296 def VCNTd : N2VDInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
4297 IIC_VCNTiD, "vcnt", "8",
4298 v8i8, v8i8, int_arm_neon_vcnt>;
4299 def VCNTq : N2VQInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
4300 IIC_VCNTiQ, "vcnt", "8",
4301 v16i8, v16i8, int_arm_neon_vcnt>;
4303 // Vector Swap -- for disassembly only.
4304 def VSWPd : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 0, 0,
4305 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
4306 "vswp", "$Vd, $Vm", "", []>;
4307 def VSWPq : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 1, 0,
4308 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
4309 "vswp", "$Vd, $Vm", "", []>;
4311 // Vector Move Operations.
4313 // VMOV : Vector Move (Register)
4314 def : InstAlias<"vmov${p} $Vd, $Vm",
4315 (VORRd DPR:$Vd, DPR:$Vm, DPR:$Vm, pred:$p)>;
4316 def : InstAlias<"vmov${p} $Vd, $Vm",
4317 (VORRq QPR:$Vd, QPR:$Vm, QPR:$Vm, pred:$p)>;
4319 // VMOV : Vector Move (Immediate)
4321 let isReMaterializable = 1 in {
4322 def VMOVv8i8 : N1ModImm<1, 0b000, 0b1110, 0, 0, 0, 1, (outs DPR:$Vd),
4323 (ins nImmSplatI8:$SIMM), IIC_VMOVImm,
4324 "vmov", "i8", "$Vd, $SIMM", "",
4325 [(set DPR:$Vd, (v8i8 (NEONvmovImm timm:$SIMM)))]>;
4326 def VMOVv16i8 : N1ModImm<1, 0b000, 0b1110, 0, 1, 0, 1, (outs QPR:$Vd),
4327 (ins nImmSplatI8:$SIMM), IIC_VMOVImm,
4328 "vmov", "i8", "$Vd, $SIMM", "",
4329 [(set QPR:$Vd, (v16i8 (NEONvmovImm timm:$SIMM)))]>;
4331 def VMOVv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 0, 1, (outs DPR:$Vd),
4332 (ins nModImm:$SIMM), IIC_VMOVImm,
4333 "vmov", "i16", "$Vd, $SIMM", "",
4334 [(set DPR:$Vd, (v4i16 (NEONvmovImm timm:$SIMM)))]> {
4335 let Inst{9} = SIMM{9};
4338 def VMOVv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 0, 1, (outs QPR:$Vd),
4339 (ins nModImm:$SIMM), IIC_VMOVImm,
4340 "vmov", "i16", "$Vd, $SIMM", "",
4341 [(set QPR:$Vd, (v8i16 (NEONvmovImm timm:$SIMM)))]> {
4342 let Inst{9} = SIMM{9};
4345 def VMOVv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 0, 1, (outs DPR:$Vd),
4346 (ins nModImm:$SIMM), IIC_VMOVImm,
4347 "vmov", "i32", "$Vd, $SIMM", "",
4348 [(set DPR:$Vd, (v2i32 (NEONvmovImm timm:$SIMM)))]> {
4349 let Inst{11-8} = SIMM{11-8};
4352 def VMOVv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 0, 1, (outs QPR:$Vd),
4353 (ins nModImm:$SIMM), IIC_VMOVImm,
4354 "vmov", "i32", "$Vd, $SIMM", "",
4355 [(set QPR:$Vd, (v4i32 (NEONvmovImm timm:$SIMM)))]> {
4356 let Inst{11-8} = SIMM{11-8};
4359 def VMOVv1i64 : N1ModImm<1, 0b000, 0b1110, 0, 0, 1, 1, (outs DPR:$Vd),
4360 (ins nModImm:$SIMM), IIC_VMOVImm,
4361 "vmov", "i64", "$Vd, $SIMM", "",
4362 [(set DPR:$Vd, (v1i64 (NEONvmovImm timm:$SIMM)))]>;
4363 def VMOVv2i64 : N1ModImm<1, 0b000, 0b1110, 0, 1, 1, 1, (outs QPR:$Vd),
4364 (ins nModImm:$SIMM), IIC_VMOVImm,
4365 "vmov", "i64", "$Vd, $SIMM", "",
4366 [(set QPR:$Vd, (v2i64 (NEONvmovImm timm:$SIMM)))]>;
4367 } // isReMaterializable
4369 // VMOV : Vector Get Lane (move scalar to ARM core register)
4371 def VGETLNs8 : NVGetLane<{1,1,1,0,0,1,?,1}, 0b1011, {?,?},
4372 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
4373 IIC_VMOVSI, "vmov", "s8", "$R, $V[$lane]",
4374 [(set GPR:$R, (NEONvgetlanes (v8i8 DPR:$V),
4376 let Inst{21} = lane{2};
4377 let Inst{6-5} = lane{1-0};
4379 def VGETLNs16 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, {?,1},
4380 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
4381 IIC_VMOVSI, "vmov", "s16", "$R, $V[$lane]",
4382 [(set GPR:$R, (NEONvgetlanes (v4i16 DPR:$V),
4384 let Inst{21} = lane{1};
4385 let Inst{6} = lane{0};
4387 def VGETLNu8 : NVGetLane<{1,1,1,0,1,1,?,1}, 0b1011, {?,?},
4388 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
4389 IIC_VMOVSI, "vmov", "u8", "$R, $V[$lane]",
4390 [(set GPR:$R, (NEONvgetlaneu (v8i8 DPR:$V),
4392 let Inst{21} = lane{2};
4393 let Inst{6-5} = lane{1-0};
4395 def VGETLNu16 : NVGetLane<{1,1,1,0,1,0,?,1}, 0b1011, {?,1},
4396 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
4397 IIC_VMOVSI, "vmov", "u16", "$R, $V[$lane]",
4398 [(set GPR:$R, (NEONvgetlaneu (v4i16 DPR:$V),
4400 let Inst{21} = lane{1};
4401 let Inst{6} = lane{0};
4403 def VGETLNi32 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, 0b00,
4404 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
4405 IIC_VMOVSI, "vmov", "32", "$R, $V[$lane]",
4406 [(set GPR:$R, (extractelt (v2i32 DPR:$V),
4408 let Inst{21} = lane{0};
4410 // def VGETLNf32: see FMRDH and FMRDL in ARMInstrVFP.td
4411 def : Pat<(NEONvgetlanes (v16i8 QPR:$src), imm:$lane),
4412 (VGETLNs8 (v8i8 (EXTRACT_SUBREG QPR:$src,
4413 (DSubReg_i8_reg imm:$lane))),
4414 (SubReg_i8_lane imm:$lane))>;
4415 def : Pat<(NEONvgetlanes (v8i16 QPR:$src), imm:$lane),
4416 (VGETLNs16 (v4i16 (EXTRACT_SUBREG QPR:$src,
4417 (DSubReg_i16_reg imm:$lane))),
4418 (SubReg_i16_lane imm:$lane))>;
4419 def : Pat<(NEONvgetlaneu (v16i8 QPR:$src), imm:$lane),
4420 (VGETLNu8 (v8i8 (EXTRACT_SUBREG QPR:$src,
4421 (DSubReg_i8_reg imm:$lane))),
4422 (SubReg_i8_lane imm:$lane))>;
4423 def : Pat<(NEONvgetlaneu (v8i16 QPR:$src), imm:$lane),
4424 (VGETLNu16 (v4i16 (EXTRACT_SUBREG QPR:$src,
4425 (DSubReg_i16_reg imm:$lane))),
4426 (SubReg_i16_lane imm:$lane))>;
4427 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
4428 (VGETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src,
4429 (DSubReg_i32_reg imm:$lane))),
4430 (SubReg_i32_lane imm:$lane))>;
4431 def : Pat<(extractelt (v2f32 DPR:$src1), imm:$src2),
4432 (EXTRACT_SUBREG (v2f32 (COPY_TO_REGCLASS (v2f32 DPR:$src1),DPR_VFP2)),
4433 (SSubReg_f32_reg imm:$src2))>;
4434 def : Pat<(extractelt (v4f32 QPR:$src1), imm:$src2),
4435 (EXTRACT_SUBREG (v4f32 (COPY_TO_REGCLASS (v4f32 QPR:$src1),QPR_VFP2)),
4436 (SSubReg_f32_reg imm:$src2))>;
4437 //def : Pat<(extractelt (v2i64 QPR:$src1), imm:$src2),
4438 // (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
4439 def : Pat<(extractelt (v2f64 QPR:$src1), imm:$src2),
4440 (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
4443 // VMOV : Vector Set Lane (move ARM core register to scalar)
4445 let Constraints = "$src1 = $V" in {
4446 def VSETLNi8 : NVSetLane<{1,1,1,0,0,1,?,0}, 0b1011, {?,?}, (outs DPR:$V),
4447 (ins DPR:$src1, GPR:$R, nohash_imm:$lane),
4448 IIC_VMOVISL, "vmov", "8", "$V[$lane], $R",
4449 [(set DPR:$V, (vector_insert (v8i8 DPR:$src1),
4450 GPR:$R, imm:$lane))]> {
4451 let Inst{21} = lane{2};
4452 let Inst{6-5} = lane{1-0};
4454 def VSETLNi16 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, {?,1}, (outs DPR:$V),
4455 (ins DPR:$src1, GPR:$R, nohash_imm:$lane),
4456 IIC_VMOVISL, "vmov", "16", "$V[$lane], $R",
4457 [(set DPR:$V, (vector_insert (v4i16 DPR:$src1),
4458 GPR:$R, imm:$lane))]> {
4459 let Inst{21} = lane{1};
4460 let Inst{6} = lane{0};
4462 def VSETLNi32 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, 0b00, (outs DPR:$V),
4463 (ins DPR:$src1, GPR:$R, nohash_imm:$lane),
4464 IIC_VMOVISL, "vmov", "32", "$V[$lane], $R",
4465 [(set DPR:$V, (insertelt (v2i32 DPR:$src1),
4466 GPR:$R, imm:$lane))]> {
4467 let Inst{21} = lane{0};
4470 def : Pat<(vector_insert (v16i8 QPR:$src1), GPR:$src2, imm:$lane),
4471 (v16i8 (INSERT_SUBREG QPR:$src1,
4472 (v8i8 (VSETLNi8 (v8i8 (EXTRACT_SUBREG QPR:$src1,
4473 (DSubReg_i8_reg imm:$lane))),
4474 GPR:$src2, (SubReg_i8_lane imm:$lane))),
4475 (DSubReg_i8_reg imm:$lane)))>;
4476 def : Pat<(vector_insert (v8i16 QPR:$src1), GPR:$src2, imm:$lane),
4477 (v8i16 (INSERT_SUBREG QPR:$src1,
4478 (v4i16 (VSETLNi16 (v4i16 (EXTRACT_SUBREG QPR:$src1,
4479 (DSubReg_i16_reg imm:$lane))),
4480 GPR:$src2, (SubReg_i16_lane imm:$lane))),
4481 (DSubReg_i16_reg imm:$lane)))>;
4482 def : Pat<(insertelt (v4i32 QPR:$src1), GPR:$src2, imm:$lane),
4483 (v4i32 (INSERT_SUBREG QPR:$src1,
4484 (v2i32 (VSETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src1,
4485 (DSubReg_i32_reg imm:$lane))),
4486 GPR:$src2, (SubReg_i32_lane imm:$lane))),
4487 (DSubReg_i32_reg imm:$lane)))>;
4489 def : Pat<(v2f32 (insertelt DPR:$src1, SPR:$src2, imm:$src3)),
4490 (INSERT_SUBREG (v2f32 (COPY_TO_REGCLASS DPR:$src1, DPR_VFP2)),
4491 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
4492 def : Pat<(v4f32 (insertelt QPR:$src1, SPR:$src2, imm:$src3)),
4493 (INSERT_SUBREG (v4f32 (COPY_TO_REGCLASS QPR:$src1, QPR_VFP2)),
4494 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
4496 //def : Pat<(v2i64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
4497 // (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
4498 def : Pat<(v2f64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
4499 (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
4501 def : Pat<(v2f32 (scalar_to_vector SPR:$src)),
4502 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
4503 def : Pat<(v2f64 (scalar_to_vector (f64 DPR:$src))),
4504 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
4505 def : Pat<(v4f32 (scalar_to_vector SPR:$src)),
4506 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
4508 def : Pat<(v8i8 (scalar_to_vector GPR:$src)),
4509 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4510 def : Pat<(v4i16 (scalar_to_vector GPR:$src)),
4511 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4512 def : Pat<(v2i32 (scalar_to_vector GPR:$src)),
4513 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4515 def : Pat<(v16i8 (scalar_to_vector GPR:$src)),
4516 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4517 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4519 def : Pat<(v8i16 (scalar_to_vector GPR:$src)),
4520 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
4521 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4523 def : Pat<(v4i32 (scalar_to_vector GPR:$src)),
4524 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
4525 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4528 // VDUP : Vector Duplicate (from ARM core register to all elements)
4530 class VDUPD<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
4531 : NVDup<opcod1, 0b1011, opcod3, (outs DPR:$V), (ins GPR:$R),
4532 IIC_VMOVIS, "vdup", Dt, "$V, $R",
4533 [(set DPR:$V, (Ty (NEONvdup (i32 GPR:$R))))]>;
4534 class VDUPQ<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
4535 : NVDup<opcod1, 0b1011, opcod3, (outs QPR:$V), (ins GPR:$R),
4536 IIC_VMOVIS, "vdup", Dt, "$V, $R",
4537 [(set QPR:$V, (Ty (NEONvdup (i32 GPR:$R))))]>;
4539 def VDUP8d : VDUPD<0b11101100, 0b00, "8", v8i8>;
4540 def VDUP16d : VDUPD<0b11101000, 0b01, "16", v4i16>;
4541 def VDUP32d : VDUPD<0b11101000, 0b00, "32", v2i32>;
4542 def VDUP8q : VDUPQ<0b11101110, 0b00, "8", v16i8>;
4543 def VDUP16q : VDUPQ<0b11101010, 0b01, "16", v8i16>;
4544 def VDUP32q : VDUPQ<0b11101010, 0b00, "32", v4i32>;
4546 def : Pat<(v2f32 (NEONvdup (f32 (bitconvert GPR:$R)))), (VDUP32d GPR:$R)>;
4547 def : Pat<(v4f32 (NEONvdup (f32 (bitconvert GPR:$R)))), (VDUP32q GPR:$R)>;
4549 // VDUP : Vector Duplicate Lane (from scalar to all elements)
4551 class VDUPLND<bits<4> op19_16, string OpcodeStr, string Dt,
4552 ValueType Ty, Operand IdxTy>
4553 : NVDupLane<op19_16, 0, (outs DPR:$Vd), (ins DPR:$Vm, IdxTy:$lane),
4554 IIC_VMOVD, OpcodeStr, Dt, "$Vd, $Vm$lane",
4555 [(set DPR:$Vd, (Ty (NEONvduplane (Ty DPR:$Vm), imm:$lane)))]>;
4557 class VDUPLNQ<bits<4> op19_16, string OpcodeStr, string Dt,
4558 ValueType ResTy, ValueType OpTy, Operand IdxTy>
4559 : NVDupLane<op19_16, 1, (outs QPR:$Vd), (ins DPR:$Vm, IdxTy:$lane),
4560 IIC_VMOVQ, OpcodeStr, Dt, "$Vd, $Vm$lane",
4561 [(set QPR:$Vd, (ResTy (NEONvduplane (OpTy DPR:$Vm),
4562 VectorIndex32:$lane)))]>;
4564 // Inst{19-16} is partially specified depending on the element size.
4566 def VDUPLN8d : VDUPLND<{?,?,?,1}, "vdup", "8", v8i8, VectorIndex8> {
4568 let Inst{19-17} = lane{2-0};
4570 def VDUPLN16d : VDUPLND<{?,?,1,0}, "vdup", "16", v4i16, VectorIndex16> {
4572 let Inst{19-18} = lane{1-0};
4574 def VDUPLN32d : VDUPLND<{?,1,0,0}, "vdup", "32", v2i32, VectorIndex32> {
4576 let Inst{19} = lane{0};
4578 def VDUPLN8q : VDUPLNQ<{?,?,?,1}, "vdup", "8", v16i8, v8i8, VectorIndex8> {
4580 let Inst{19-17} = lane{2-0};
4582 def VDUPLN16q : VDUPLNQ<{?,?,1,0}, "vdup", "16", v8i16, v4i16, VectorIndex16> {
4584 let Inst{19-18} = lane{1-0};
4586 def VDUPLN32q : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4i32, v2i32, VectorIndex32> {
4588 let Inst{19} = lane{0};
4591 def : Pat<(v2f32 (NEONvduplane (v2f32 DPR:$Vm), imm:$lane)),
4592 (VDUPLN32d DPR:$Vm, imm:$lane)>;
4594 def : Pat<(v4f32 (NEONvduplane (v2f32 DPR:$Vm), imm:$lane)),
4595 (VDUPLN32q DPR:$Vm, imm:$lane)>;
4597 def : Pat<(v16i8 (NEONvduplane (v16i8 QPR:$src), imm:$lane)),
4598 (v16i8 (VDUPLN8q (v8i8 (EXTRACT_SUBREG QPR:$src,
4599 (DSubReg_i8_reg imm:$lane))),
4600 (SubReg_i8_lane imm:$lane)))>;
4601 def : Pat<(v8i16 (NEONvduplane (v8i16 QPR:$src), imm:$lane)),
4602 (v8i16 (VDUPLN16q (v4i16 (EXTRACT_SUBREG QPR:$src,
4603 (DSubReg_i16_reg imm:$lane))),
4604 (SubReg_i16_lane imm:$lane)))>;
4605 def : Pat<(v4i32 (NEONvduplane (v4i32 QPR:$src), imm:$lane)),
4606 (v4i32 (VDUPLN32q (v2i32 (EXTRACT_SUBREG QPR:$src,
4607 (DSubReg_i32_reg imm:$lane))),
4608 (SubReg_i32_lane imm:$lane)))>;
4609 def : Pat<(v4f32 (NEONvduplane (v4f32 QPR:$src), imm:$lane)),
4610 (v4f32 (VDUPLN32q (v2f32 (EXTRACT_SUBREG QPR:$src,
4611 (DSubReg_i32_reg imm:$lane))),
4612 (SubReg_i32_lane imm:$lane)))>;
4614 def VDUPfdf : PseudoNeonI<(outs DPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
4615 [(set DPR:$dst, (v2f32 (NEONvdup (f32 SPR:$src))))]>;
4616 def VDUPfqf : PseudoNeonI<(outs QPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
4617 [(set QPR:$dst, (v4f32 (NEONvdup (f32 SPR:$src))))]>;
4619 // VMOVN : Vector Narrowing Move
4620 defm VMOVN : N2VN_HSD<0b11,0b11,0b10,0b00100,0,0, IIC_VMOVN,
4621 "vmovn", "i", trunc>;
4622 // VQMOVN : Vector Saturating Narrowing Move
4623 defm VQMOVNs : N2VNInt_HSD<0b11,0b11,0b10,0b00101,0,0, IIC_VQUNAiD,
4624 "vqmovn", "s", int_arm_neon_vqmovns>;
4625 defm VQMOVNu : N2VNInt_HSD<0b11,0b11,0b10,0b00101,1,0, IIC_VQUNAiD,
4626 "vqmovn", "u", int_arm_neon_vqmovnu>;
4627 defm VQMOVNsu : N2VNInt_HSD<0b11,0b11,0b10,0b00100,1,0, IIC_VQUNAiD,
4628 "vqmovun", "s", int_arm_neon_vqmovnsu>;
4629 // VMOVL : Vector Lengthening Move
4630 defm VMOVLs : N2VL_QHS<0b01,0b10100,0,1, "vmovl", "s", sext>;
4631 defm VMOVLu : N2VL_QHS<0b11,0b10100,0,1, "vmovl", "u", zext>;
4633 // Vector Conversions.
4635 // VCVT : Vector Convert Between Floating-Point and Integers
4636 def VCVTf2sd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
4637 v2i32, v2f32, fp_to_sint>;
4638 def VCVTf2ud : N2VD<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
4639 v2i32, v2f32, fp_to_uint>;
4640 def VCVTs2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
4641 v2f32, v2i32, sint_to_fp>;
4642 def VCVTu2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
4643 v2f32, v2i32, uint_to_fp>;
4645 def VCVTf2sq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
4646 v4i32, v4f32, fp_to_sint>;
4647 def VCVTf2uq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
4648 v4i32, v4f32, fp_to_uint>;
4649 def VCVTs2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
4650 v4f32, v4i32, sint_to_fp>;
4651 def VCVTu2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
4652 v4f32, v4i32, uint_to_fp>;
4654 // VCVT : Vector Convert Between Floating-Point and Fixed-Point.
4655 def VCVTf2xsd : N2VCvtD<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
4656 v2i32, v2f32, int_arm_neon_vcvtfp2fxs>;
4657 def VCVTf2xud : N2VCvtD<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
4658 v2i32, v2f32, int_arm_neon_vcvtfp2fxu>;
4659 def VCVTxs2fd : N2VCvtD<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
4660 v2f32, v2i32, int_arm_neon_vcvtfxs2fp>;
4661 def VCVTxu2fd : N2VCvtD<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
4662 v2f32, v2i32, int_arm_neon_vcvtfxu2fp>;
4664 def VCVTf2xsq : N2VCvtQ<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
4665 v4i32, v4f32, int_arm_neon_vcvtfp2fxs>;
4666 def VCVTf2xuq : N2VCvtQ<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
4667 v4i32, v4f32, int_arm_neon_vcvtfp2fxu>;
4668 def VCVTxs2fq : N2VCvtQ<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
4669 v4f32, v4i32, int_arm_neon_vcvtfxs2fp>;
4670 def VCVTxu2fq : N2VCvtQ<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
4671 v4f32, v4i32, int_arm_neon_vcvtfxu2fp>;
4673 // VCVT : Vector Convert Between Half-Precision and Single-Precision.
4674 def VCVTf2h : N2VNInt<0b11, 0b11, 0b01, 0b10, 0b01100, 0, 0,
4675 IIC_VUNAQ, "vcvt", "f16.f32",
4676 v4i16, v4f32, int_arm_neon_vcvtfp2hf>,
4677 Requires<[HasNEON, HasFP16]>;
4678 def VCVTh2f : N2VLInt<0b11, 0b11, 0b01, 0b10, 0b01110, 0, 0,
4679 IIC_VUNAQ, "vcvt", "f32.f16",
4680 v4f32, v4i16, int_arm_neon_vcvthf2fp>,
4681 Requires<[HasNEON, HasFP16]>;
4685 // VREV64 : Vector Reverse elements within 64-bit doublewords
4687 class VREV64D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4688 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 0, 0, (outs DPR:$Vd),
4689 (ins DPR:$Vm), IIC_VMOVD,
4690 OpcodeStr, Dt, "$Vd, $Vm", "",
4691 [(set DPR:$Vd, (Ty (NEONvrev64 (Ty DPR:$Vm))))]>;
4692 class VREV64Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4693 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 1, 0, (outs QPR:$Vd),
4694 (ins QPR:$Vm), IIC_VMOVQ,
4695 OpcodeStr, Dt, "$Vd, $Vm", "",
4696 [(set QPR:$Vd, (Ty (NEONvrev64 (Ty QPR:$Vm))))]>;
4698 def VREV64d8 : VREV64D<0b00, "vrev64", "8", v8i8>;
4699 def VREV64d16 : VREV64D<0b01, "vrev64", "16", v4i16>;
4700 def VREV64d32 : VREV64D<0b10, "vrev64", "32", v2i32>;
4701 def : Pat<(v2f32 (NEONvrev64 (v2f32 DPR:$Vm))), (VREV64d32 DPR:$Vm)>;
4703 def VREV64q8 : VREV64Q<0b00, "vrev64", "8", v16i8>;
4704 def VREV64q16 : VREV64Q<0b01, "vrev64", "16", v8i16>;
4705 def VREV64q32 : VREV64Q<0b10, "vrev64", "32", v4i32>;
4706 def : Pat<(v4f32 (NEONvrev64 (v4f32 QPR:$Vm))), (VREV64q32 QPR:$Vm)>;
4708 // VREV32 : Vector Reverse elements within 32-bit words
4710 class VREV32D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4711 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 0, 0, (outs DPR:$Vd),
4712 (ins DPR:$Vm), IIC_VMOVD,
4713 OpcodeStr, Dt, "$Vd, $Vm", "",
4714 [(set DPR:$Vd, (Ty (NEONvrev32 (Ty DPR:$Vm))))]>;
4715 class VREV32Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4716 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 1, 0, (outs QPR:$Vd),
4717 (ins QPR:$Vm), IIC_VMOVQ,
4718 OpcodeStr, Dt, "$Vd, $Vm", "",
4719 [(set QPR:$Vd, (Ty (NEONvrev32 (Ty QPR:$Vm))))]>;
4721 def VREV32d8 : VREV32D<0b00, "vrev32", "8", v8i8>;
4722 def VREV32d16 : VREV32D<0b01, "vrev32", "16", v4i16>;
4724 def VREV32q8 : VREV32Q<0b00, "vrev32", "8", v16i8>;
4725 def VREV32q16 : VREV32Q<0b01, "vrev32", "16", v8i16>;
4727 // VREV16 : Vector Reverse elements within 16-bit halfwords
4729 class VREV16D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4730 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 0, 0, (outs DPR:$Vd),
4731 (ins DPR:$Vm), IIC_VMOVD,
4732 OpcodeStr, Dt, "$Vd, $Vm", "",
4733 [(set DPR:$Vd, (Ty (NEONvrev16 (Ty DPR:$Vm))))]>;
4734 class VREV16Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4735 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 1, 0, (outs QPR:$Vd),
4736 (ins QPR:$Vm), IIC_VMOVQ,
4737 OpcodeStr, Dt, "$Vd, $Vm", "",
4738 [(set QPR:$Vd, (Ty (NEONvrev16 (Ty QPR:$Vm))))]>;
4740 def VREV16d8 : VREV16D<0b00, "vrev16", "8", v8i8>;
4741 def VREV16q8 : VREV16Q<0b00, "vrev16", "8", v16i8>;
4743 // Other Vector Shuffles.
4745 // Aligned extractions: really just dropping registers
4747 class AlignedVEXTq<ValueType DestTy, ValueType SrcTy, SDNodeXForm LaneCVT>
4748 : Pat<(DestTy (vector_extract_subvec (SrcTy QPR:$src), (i32 imm:$start))),
4749 (EXTRACT_SUBREG (SrcTy QPR:$src), (LaneCVT imm:$start))>;
4751 def : AlignedVEXTq<v8i8, v16i8, DSubReg_i8_reg>;
4753 def : AlignedVEXTq<v4i16, v8i16, DSubReg_i16_reg>;
4755 def : AlignedVEXTq<v2i32, v4i32, DSubReg_i32_reg>;
4757 def : AlignedVEXTq<v1i64, v2i64, DSubReg_f64_reg>;
4759 def : AlignedVEXTq<v2f32, v4f32, DSubReg_i32_reg>;
4762 // VEXT : Vector Extract
4764 class VEXTd<string OpcodeStr, string Dt, ValueType Ty>
4765 : N3V<0,1,0b11,{?,?,?,?},0,0, (outs DPR:$Vd),
4766 (ins DPR:$Vn, DPR:$Vm, i32imm:$index), NVExtFrm,
4767 IIC_VEXTD, OpcodeStr, Dt, "$Vd, $Vn, $Vm, $index", "",
4768 [(set DPR:$Vd, (Ty (NEONvext (Ty DPR:$Vn),
4769 (Ty DPR:$Vm), imm:$index)))]> {
4771 let Inst{11-8} = index{3-0};
4774 class VEXTq<string OpcodeStr, string Dt, ValueType Ty>
4775 : N3V<0,1,0b11,{?,?,?,?},1,0, (outs QPR:$Vd),
4776 (ins QPR:$Vn, QPR:$Vm, i32imm:$index), NVExtFrm,
4777 IIC_VEXTQ, OpcodeStr, Dt, "$Vd, $Vn, $Vm, $index", "",
4778 [(set QPR:$Vd, (Ty (NEONvext (Ty QPR:$Vn),
4779 (Ty QPR:$Vm), imm:$index)))]> {
4781 let Inst{11-8} = index{3-0};
4784 def VEXTd8 : VEXTd<"vext", "8", v8i8> {
4785 let Inst{11-8} = index{3-0};
4787 def VEXTd16 : VEXTd<"vext", "16", v4i16> {
4788 let Inst{11-9} = index{2-0};
4791 def VEXTd32 : VEXTd<"vext", "32", v2i32> {
4792 let Inst{11-10} = index{1-0};
4793 let Inst{9-8} = 0b00;
4795 def : Pat<(v2f32 (NEONvext (v2f32 DPR:$Vn),
4798 (VEXTd32 DPR:$Vn, DPR:$Vm, imm:$index)>;
4800 def VEXTq8 : VEXTq<"vext", "8", v16i8> {
4801 let Inst{11-8} = index{3-0};
4803 def VEXTq16 : VEXTq<"vext", "16", v8i16> {
4804 let Inst{11-9} = index{2-0};
4807 def VEXTq32 : VEXTq<"vext", "32", v4i32> {
4808 let Inst{11-10} = index{1-0};
4809 let Inst{9-8} = 0b00;
4811 def : Pat<(v4f32 (NEONvext (v4f32 QPR:$Vn),
4814 (VEXTq32 QPR:$Vn, QPR:$Vm, imm:$index)>;
4816 // VTRN : Vector Transpose
4818 def VTRNd8 : N2VDShuffle<0b00, 0b00001, "vtrn", "8">;
4819 def VTRNd16 : N2VDShuffle<0b01, 0b00001, "vtrn", "16">;
4820 def VTRNd32 : N2VDShuffle<0b10, 0b00001, "vtrn", "32">;
4822 def VTRNq8 : N2VQShuffle<0b00, 0b00001, IIC_VPERMQ, "vtrn", "8">;
4823 def VTRNq16 : N2VQShuffle<0b01, 0b00001, IIC_VPERMQ, "vtrn", "16">;
4824 def VTRNq32 : N2VQShuffle<0b10, 0b00001, IIC_VPERMQ, "vtrn", "32">;
4826 // VUZP : Vector Unzip (Deinterleave)
4828 def VUZPd8 : N2VDShuffle<0b00, 0b00010, "vuzp", "8">;
4829 def VUZPd16 : N2VDShuffle<0b01, 0b00010, "vuzp", "16">;
4830 def VUZPd32 : N2VDShuffle<0b10, 0b00010, "vuzp", "32">;
4832 def VUZPq8 : N2VQShuffle<0b00, 0b00010, IIC_VPERMQ3, "vuzp", "8">;
4833 def VUZPq16 : N2VQShuffle<0b01, 0b00010, IIC_VPERMQ3, "vuzp", "16">;
4834 def VUZPq32 : N2VQShuffle<0b10, 0b00010, IIC_VPERMQ3, "vuzp", "32">;
4836 // VZIP : Vector Zip (Interleave)
4838 def VZIPd8 : N2VDShuffle<0b00, 0b00011, "vzip", "8">;
4839 def VZIPd16 : N2VDShuffle<0b01, 0b00011, "vzip", "16">;
4840 def VZIPd32 : N2VDShuffle<0b10, 0b00011, "vzip", "32">;
4842 def VZIPq8 : N2VQShuffle<0b00, 0b00011, IIC_VPERMQ3, "vzip", "8">;
4843 def VZIPq16 : N2VQShuffle<0b01, 0b00011, IIC_VPERMQ3, "vzip", "16">;
4844 def VZIPq32 : N2VQShuffle<0b10, 0b00011, IIC_VPERMQ3, "vzip", "32">;
4846 // Vector Table Lookup and Table Extension.
4848 // VTBL : Vector Table Lookup
4849 let DecoderMethod = "DecodeTBLInstruction" in {
4851 : N3V<1,1,0b11,0b1000,0,0, (outs DPR:$Vd),
4852 (ins DPR:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTB1,
4853 "vtbl", "8", "$Vd, \\{$Vn\\}, $Vm", "",
4854 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbl1 DPR:$Vn, DPR:$Vm)))]>;
4855 let hasExtraSrcRegAllocReq = 1 in {
4857 : N3V<1,1,0b11,0b1001,0,0, (outs DPR:$Vd),
4858 (ins DPR:$Vn, DPR:$tbl2, DPR:$Vm), NVTBLFrm, IIC_VTB2,
4859 "vtbl", "8", "$Vd, \\{$Vn, $tbl2\\}, $Vm", "", []>;
4861 : N3V<1,1,0b11,0b1010,0,0, (outs DPR:$Vd),
4862 (ins DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$Vm), NVTBLFrm, IIC_VTB3,
4863 "vtbl", "8", "$Vd, \\{$Vn, $tbl2, $tbl3\\}, $Vm", "", []>;
4865 : N3V<1,1,0b11,0b1011,0,0, (outs DPR:$Vd),
4866 (ins DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$Vm),
4868 "vtbl", "8", "$Vd, \\{$Vn, $tbl2, $tbl3, $tbl4\\}, $Vm", "", []>;
4869 } // hasExtraSrcRegAllocReq = 1
4872 : PseudoNeonI<(outs DPR:$dst), (ins QPR:$tbl, DPR:$src), IIC_VTB2, "", []>;
4874 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB3, "", []>;
4876 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB4, "", []>;
4878 // VTBX : Vector Table Extension
4880 : N3V<1,1,0b11,0b1000,1,0, (outs DPR:$Vd),
4881 (ins DPR:$orig, DPR:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTBX1,
4882 "vtbx", "8", "$Vd, \\{$Vn\\}, $Vm", "$orig = $Vd",
4883 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbx1
4884 DPR:$orig, DPR:$Vn, DPR:$Vm)))]>;
4885 let hasExtraSrcRegAllocReq = 1 in {
4887 : N3V<1,1,0b11,0b1001,1,0, (outs DPR:$Vd),
4888 (ins DPR:$orig, DPR:$Vn, DPR:$tbl2, DPR:$Vm), NVTBLFrm, IIC_VTBX2,
4889 "vtbx", "8", "$Vd, \\{$Vn, $tbl2\\}, $Vm", "$orig = $Vd", []>;
4891 : N3V<1,1,0b11,0b1010,1,0, (outs DPR:$Vd),
4892 (ins DPR:$orig, DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$Vm),
4893 NVTBLFrm, IIC_VTBX3,
4894 "vtbx", "8", "$Vd, \\{$Vn, $tbl2, $tbl3\\}, $Vm",
4897 : N3V<1,1,0b11,0b1011,1,0, (outs DPR:$Vd), (ins DPR:$orig, DPR:$Vn,
4898 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$Vm), NVTBLFrm, IIC_VTBX4,
4899 "vtbx", "8", "$Vd, \\{$Vn, $tbl2, $tbl3, $tbl4\\}, $Vm",
4901 } // hasExtraSrcRegAllocReq = 1
4904 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QPR:$tbl, DPR:$src),
4905 IIC_VTBX2, "$orig = $dst", []>;
4907 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
4908 IIC_VTBX3, "$orig = $dst", []>;
4910 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
4911 IIC_VTBX4, "$orig = $dst", []>;
4912 } // DecoderMethod = "DecodeTBLInstruction"
4914 //===----------------------------------------------------------------------===//
4915 // NEON instructions for single-precision FP math
4916 //===----------------------------------------------------------------------===//
4918 class N2VSPat<SDNode OpNode, NeonI Inst>
4919 : NEONFPPat<(f32 (OpNode SPR:$a)),
4921 (v2f32 (COPY_TO_REGCLASS (Inst
4923 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
4924 SPR:$a, ssub_0)), DPR_VFP2)), ssub_0)>;
4926 class N3VSPat<SDNode OpNode, NeonI Inst>
4927 : NEONFPPat<(f32 (OpNode SPR:$a, SPR:$b)),
4929 (v2f32 (COPY_TO_REGCLASS (Inst
4931 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
4934 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
4935 SPR:$b, ssub_0)), DPR_VFP2)), ssub_0)>;
4937 class N3VSMulOpPat<SDNode MulNode, SDNode OpNode, NeonI Inst>
4938 : NEONFPPat<(f32 (OpNode SPR:$acc, (f32 (MulNode SPR:$a, SPR:$b)))),
4940 (v2f32 (COPY_TO_REGCLASS (Inst
4942 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
4945 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
4948 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
4949 SPR:$b, ssub_0)), DPR_VFP2)), ssub_0)>;
4951 def : N3VSPat<fadd, VADDfd>;
4952 def : N3VSPat<fsub, VSUBfd>;
4953 def : N3VSPat<fmul, VMULfd>;
4954 def : N3VSMulOpPat<fmul, fadd, VMLAfd>,
4955 Requires<[HasNEON, UseNEONForFP, UseFPVMLx]>;
4956 def : N3VSMulOpPat<fmul, fsub, VMLSfd>,
4957 Requires<[HasNEON, UseNEONForFP, UseFPVMLx]>;
4958 def : N2VSPat<fabs, VABSfd>;
4959 def : N2VSPat<fneg, VNEGfd>;
4960 def : N3VSPat<NEONfmax, VMAXfd>;
4961 def : N3VSPat<NEONfmin, VMINfd>;
4962 def : N2VSPat<arm_ftosi, VCVTf2sd>;
4963 def : N2VSPat<arm_ftoui, VCVTf2ud>;
4964 def : N2VSPat<arm_sitof, VCVTs2fd>;
4965 def : N2VSPat<arm_uitof, VCVTu2fd>;
4967 //===----------------------------------------------------------------------===//
4968 // Non-Instruction Patterns
4969 //===----------------------------------------------------------------------===//
4972 def : Pat<(v1i64 (bitconvert (v2i32 DPR:$src))), (v1i64 DPR:$src)>;
4973 def : Pat<(v1i64 (bitconvert (v4i16 DPR:$src))), (v1i64 DPR:$src)>;
4974 def : Pat<(v1i64 (bitconvert (v8i8 DPR:$src))), (v1i64 DPR:$src)>;
4975 def : Pat<(v1i64 (bitconvert (f64 DPR:$src))), (v1i64 DPR:$src)>;
4976 def : Pat<(v1i64 (bitconvert (v2f32 DPR:$src))), (v1i64 DPR:$src)>;
4977 def : Pat<(v2i32 (bitconvert (v1i64 DPR:$src))), (v2i32 DPR:$src)>;
4978 def : Pat<(v2i32 (bitconvert (v4i16 DPR:$src))), (v2i32 DPR:$src)>;
4979 def : Pat<(v2i32 (bitconvert (v8i8 DPR:$src))), (v2i32 DPR:$src)>;
4980 def : Pat<(v2i32 (bitconvert (f64 DPR:$src))), (v2i32 DPR:$src)>;
4981 def : Pat<(v2i32 (bitconvert (v2f32 DPR:$src))), (v2i32 DPR:$src)>;
4982 def : Pat<(v4i16 (bitconvert (v1i64 DPR:$src))), (v4i16 DPR:$src)>;
4983 def : Pat<(v4i16 (bitconvert (v2i32 DPR:$src))), (v4i16 DPR:$src)>;
4984 def : Pat<(v4i16 (bitconvert (v8i8 DPR:$src))), (v4i16 DPR:$src)>;
4985 def : Pat<(v4i16 (bitconvert (f64 DPR:$src))), (v4i16 DPR:$src)>;
4986 def : Pat<(v4i16 (bitconvert (v2f32 DPR:$src))), (v4i16 DPR:$src)>;
4987 def : Pat<(v8i8 (bitconvert (v1i64 DPR:$src))), (v8i8 DPR:$src)>;
4988 def : Pat<(v8i8 (bitconvert (v2i32 DPR:$src))), (v8i8 DPR:$src)>;
4989 def : Pat<(v8i8 (bitconvert (v4i16 DPR:$src))), (v8i8 DPR:$src)>;
4990 def : Pat<(v8i8 (bitconvert (f64 DPR:$src))), (v8i8 DPR:$src)>;
4991 def : Pat<(v8i8 (bitconvert (v2f32 DPR:$src))), (v8i8 DPR:$src)>;
4992 def : Pat<(f64 (bitconvert (v1i64 DPR:$src))), (f64 DPR:$src)>;
4993 def : Pat<(f64 (bitconvert (v2i32 DPR:$src))), (f64 DPR:$src)>;
4994 def : Pat<(f64 (bitconvert (v4i16 DPR:$src))), (f64 DPR:$src)>;
4995 def : Pat<(f64 (bitconvert (v8i8 DPR:$src))), (f64 DPR:$src)>;
4996 def : Pat<(f64 (bitconvert (v2f32 DPR:$src))), (f64 DPR:$src)>;
4997 def : Pat<(v2f32 (bitconvert (f64 DPR:$src))), (v2f32 DPR:$src)>;
4998 def : Pat<(v2f32 (bitconvert (v1i64 DPR:$src))), (v2f32 DPR:$src)>;
4999 def : Pat<(v2f32 (bitconvert (v2i32 DPR:$src))), (v2f32 DPR:$src)>;
5000 def : Pat<(v2f32 (bitconvert (v4i16 DPR:$src))), (v2f32 DPR:$src)>;
5001 def : Pat<(v2f32 (bitconvert (v8i8 DPR:$src))), (v2f32 DPR:$src)>;
5003 def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (v2i64 QPR:$src)>;
5004 def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (v2i64 QPR:$src)>;
5005 def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (v2i64 QPR:$src)>;
5006 def : Pat<(v2i64 (bitconvert (v2f64 QPR:$src))), (v2i64 QPR:$src)>;
5007 def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (v2i64 QPR:$src)>;
5008 def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (v4i32 QPR:$src)>;
5009 def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (v4i32 QPR:$src)>;
5010 def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (v4i32 QPR:$src)>;
5011 def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (v4i32 QPR:$src)>;
5012 def : Pat<(v4i32 (bitconvert (v4f32 QPR:$src))), (v4i32 QPR:$src)>;
5013 def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (v8i16 QPR:$src)>;
5014 def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (v8i16 QPR:$src)>;
5015 def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (v8i16 QPR:$src)>;
5016 def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (v8i16 QPR:$src)>;
5017 def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (v8i16 QPR:$src)>;
5018 def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (v16i8 QPR:$src)>;
5019 def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (v16i8 QPR:$src)>;
5020 def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (v16i8 QPR:$src)>;
5021 def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (v16i8 QPR:$src)>;
5022 def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (v16i8 QPR:$src)>;
5023 def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (v4f32 QPR:$src)>;
5024 def : Pat<(v4f32 (bitconvert (v4i32 QPR:$src))), (v4f32 QPR:$src)>;
5025 def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (v4f32 QPR:$src)>;
5026 def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (v4f32 QPR:$src)>;
5027 def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (v4f32 QPR:$src)>;
5028 def : Pat<(v2f64 (bitconvert (v2i64 QPR:$src))), (v2f64 QPR:$src)>;
5029 def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (v2f64 QPR:$src)>;
5030 def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 QPR:$src)>;
5031 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 QPR:$src)>;
5032 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 QPR:$src)>;