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 //===----------------------------------------------------------------------===//
14 //===----------------------------------------------------------------------===//
15 // NEON-specific DAG Nodes.
16 //===----------------------------------------------------------------------===//
18 def SDTARMVCMP : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisSameAs<1, 2>]>;
19 def SDTARMVCMPZ : SDTypeProfile<1, 1, []>;
21 def NEONvceq : SDNode<"ARMISD::VCEQ", SDTARMVCMP>;
22 def NEONvceqz : SDNode<"ARMISD::VCEQZ", SDTARMVCMPZ>;
23 def NEONvcge : SDNode<"ARMISD::VCGE", SDTARMVCMP>;
24 def NEONvcgez : SDNode<"ARMISD::VCGEZ", SDTARMVCMPZ>;
25 def NEONvclez : SDNode<"ARMISD::VCLEZ", SDTARMVCMPZ>;
26 def NEONvcgeu : SDNode<"ARMISD::VCGEU", SDTARMVCMP>;
27 def NEONvcgt : SDNode<"ARMISD::VCGT", SDTARMVCMP>;
28 def NEONvcgtz : SDNode<"ARMISD::VCGTZ", SDTARMVCMPZ>;
29 def NEONvcltz : SDNode<"ARMISD::VCLTZ", SDTARMVCMPZ>;
30 def NEONvcgtu : SDNode<"ARMISD::VCGTU", SDTARMVCMP>;
31 def NEONvtst : SDNode<"ARMISD::VTST", SDTARMVCMP>;
33 // Types for vector shift by immediates. The "SHX" version is for long and
34 // narrow operations where the source and destination vectors have different
35 // types. The "SHINS" version is for shift and insert operations.
36 def SDTARMVSH : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
38 def SDTARMVSHX : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
40 def SDTARMVSHINS : SDTypeProfile<1, 3, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
41 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
43 def NEONvshl : SDNode<"ARMISD::VSHL", SDTARMVSH>;
44 def NEONvshrs : SDNode<"ARMISD::VSHRs", SDTARMVSH>;
45 def NEONvshru : SDNode<"ARMISD::VSHRu", SDTARMVSH>;
46 def NEONvshlls : SDNode<"ARMISD::VSHLLs", SDTARMVSHX>;
47 def NEONvshllu : SDNode<"ARMISD::VSHLLu", SDTARMVSHX>;
48 def NEONvshlli : SDNode<"ARMISD::VSHLLi", SDTARMVSHX>;
49 def NEONvshrn : SDNode<"ARMISD::VSHRN", SDTARMVSHX>;
51 def NEONvrshrs : SDNode<"ARMISD::VRSHRs", SDTARMVSH>;
52 def NEONvrshru : SDNode<"ARMISD::VRSHRu", SDTARMVSH>;
53 def NEONvrshrn : SDNode<"ARMISD::VRSHRN", SDTARMVSHX>;
55 def NEONvqshls : SDNode<"ARMISD::VQSHLs", SDTARMVSH>;
56 def NEONvqshlu : SDNode<"ARMISD::VQSHLu", SDTARMVSH>;
57 def NEONvqshlsu : SDNode<"ARMISD::VQSHLsu", SDTARMVSH>;
58 def NEONvqshrns : SDNode<"ARMISD::VQSHRNs", SDTARMVSHX>;
59 def NEONvqshrnu : SDNode<"ARMISD::VQSHRNu", SDTARMVSHX>;
60 def NEONvqshrnsu : SDNode<"ARMISD::VQSHRNsu", SDTARMVSHX>;
62 def NEONvqrshrns : SDNode<"ARMISD::VQRSHRNs", SDTARMVSHX>;
63 def NEONvqrshrnu : SDNode<"ARMISD::VQRSHRNu", SDTARMVSHX>;
64 def NEONvqrshrnsu : SDNode<"ARMISD::VQRSHRNsu", SDTARMVSHX>;
66 def NEONvsli : SDNode<"ARMISD::VSLI", SDTARMVSHINS>;
67 def NEONvsri : SDNode<"ARMISD::VSRI", SDTARMVSHINS>;
69 def SDTARMVGETLN : SDTypeProfile<1, 2, [SDTCisVT<0, i32>, SDTCisInt<1>,
71 def NEONvgetlaneu : SDNode<"ARMISD::VGETLANEu", SDTARMVGETLN>;
72 def NEONvgetlanes : SDNode<"ARMISD::VGETLANEs", SDTARMVGETLN>;
74 def SDTARMVMOVIMM : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisVT<1, i32>]>;
75 def NEONvmovImm : SDNode<"ARMISD::VMOVIMM", SDTARMVMOVIMM>;
76 def NEONvmvnImm : SDNode<"ARMISD::VMVNIMM", SDTARMVMOVIMM>;
78 def SDTARMVORRIMM : SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
80 def NEONvorrImm : SDNode<"ARMISD::VORRIMM", SDTARMVORRIMM>;
81 def NEONvbicImm : SDNode<"ARMISD::VBICIMM", SDTARMVORRIMM>;
83 def NEONvbsl : SDNode<"ARMISD::VBSL",
84 SDTypeProfile<1, 3, [SDTCisVec<0>,
87 SDTCisSameAs<0, 3>]>>;
89 def NEONvdup : SDNode<"ARMISD::VDUP", SDTypeProfile<1, 1, [SDTCisVec<0>]>>;
91 // VDUPLANE can produce a quad-register result from a double-register source,
92 // so the result is not constrained to match the source.
93 def NEONvduplane : SDNode<"ARMISD::VDUPLANE",
94 SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisVec<1>,
97 def SDTARMVEXT : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
98 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
99 def NEONvext : SDNode<"ARMISD::VEXT", SDTARMVEXT>;
101 def SDTARMVSHUF : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisSameAs<0, 1>]>;
102 def NEONvrev64 : SDNode<"ARMISD::VREV64", SDTARMVSHUF>;
103 def NEONvrev32 : SDNode<"ARMISD::VREV32", SDTARMVSHUF>;
104 def NEONvrev16 : SDNode<"ARMISD::VREV16", SDTARMVSHUF>;
106 def SDTARMVSHUF2 : SDTypeProfile<2, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
108 SDTCisSameAs<0, 3>]>;
109 def NEONzip : SDNode<"ARMISD::VZIP", SDTARMVSHUF2>;
110 def NEONuzp : SDNode<"ARMISD::VUZP", SDTARMVSHUF2>;
111 def NEONtrn : SDNode<"ARMISD::VTRN", SDTARMVSHUF2>;
113 def SDTARMVMULL : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
114 SDTCisSameAs<1, 2>]>;
115 def NEONvmulls : SDNode<"ARMISD::VMULLs", SDTARMVMULL>;
116 def NEONvmullu : SDNode<"ARMISD::VMULLu", SDTARMVMULL>;
118 def SDTARMFMAX : SDTypeProfile<1, 2, [SDTCisVT<0, f32>, SDTCisSameAs<0, 1>,
119 SDTCisSameAs<0, 2>]>;
120 def NEONfmax : SDNode<"ARMISD::FMAX", SDTARMFMAX>;
121 def NEONfmin : SDNode<"ARMISD::FMIN", SDTARMFMAX>;
123 def NEONimmAllZerosV: PatLeaf<(NEONvmovImm (i32 timm)), [{
124 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
125 unsigned EltBits = 0;
126 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
127 return (EltBits == 32 && EltVal == 0);
130 def NEONimmAllOnesV: PatLeaf<(NEONvmovImm (i32 timm)), [{
131 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
132 unsigned EltBits = 0;
133 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
134 return (EltBits == 8 && EltVal == 0xff);
137 //===----------------------------------------------------------------------===//
138 // NEON operand definitions
139 //===----------------------------------------------------------------------===//
141 def nModImm : Operand<i32> {
142 let PrintMethod = "printNEONModImmOperand";
145 //===----------------------------------------------------------------------===//
146 // NEON load / store instructions
147 //===----------------------------------------------------------------------===//
149 // Use VLDM to load a Q register as a D register pair.
150 // This is a pseudo instruction that is expanded to VLDMD after reg alloc.
152 : PseudoVFPLdStM<(outs QPR:$dst), (ins GPR:$Rn),
154 [(set QPR:$dst, (v2f64 (load GPR:$Rn)))]>;
156 // Use VSTM to store a Q register as a D register pair.
157 // This is a pseudo instruction that is expanded to VSTMD after reg alloc.
159 : PseudoVFPLdStM<(outs), (ins QPR:$src, GPR:$Rn),
161 [(store (v2f64 QPR:$src), GPR:$Rn)]>;
163 // Classes for VLD* pseudo-instructions with multi-register operands.
164 // These are expanded to real instructions after register allocation.
165 class VLDQPseudo<InstrItinClass itin>
166 : PseudoNLdSt<(outs QPR:$dst), (ins addrmode6:$addr), itin, "">;
167 class VLDQWBPseudo<InstrItinClass itin>
168 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
169 (ins addrmode6:$addr, am6offset:$offset), itin,
171 class VLDQQPseudo<InstrItinClass itin>
172 : PseudoNLdSt<(outs QQPR:$dst), (ins addrmode6:$addr), itin, "">;
173 class VLDQQWBPseudo<InstrItinClass itin>
174 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
175 (ins addrmode6:$addr, am6offset:$offset), itin,
177 class VLDQQQQPseudo<InstrItinClass itin>
178 : PseudoNLdSt<(outs QQQQPR:$dst), (ins addrmode6:$addr, QQQQPR:$src),itin,
180 class VLDQQQQWBPseudo<InstrItinClass itin>
181 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
182 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
183 "$addr.addr = $wb, $src = $dst">;
185 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
187 // VLD1 : Vector Load (multiple single elements)
188 class VLD1D<bits<4> op7_4, string Dt>
189 : NLdSt<0,0b10,0b0111,op7_4, (outs DPR:$Vd),
190 (ins addrmode6:$Rn), IIC_VLD1,
191 "vld1", Dt, "\\{$Vd\\}, $Rn", "", []> {
194 let DecoderMethod = "DecodeVLDInstruction";
196 class VLD1Q<bits<4> op7_4, string Dt>
197 : NLdSt<0,0b10,0b1010,op7_4, (outs DPR:$Vd, DPR:$dst2),
198 (ins addrmode6:$Rn), IIC_VLD1x2,
199 "vld1", Dt, "\\{$Vd, $dst2\\}, $Rn", "", []> {
201 let Inst{5-4} = Rn{5-4};
202 let DecoderMethod = "DecodeVLDInstruction";
205 def VLD1d8 : VLD1D<{0,0,0,?}, "8">;
206 def VLD1d16 : VLD1D<{0,1,0,?}, "16">;
207 def VLD1d32 : VLD1D<{1,0,0,?}, "32">;
208 def VLD1d64 : VLD1D<{1,1,0,?}, "64">;
210 def VLD1q8 : VLD1Q<{0,0,?,?}, "8">;
211 def VLD1q16 : VLD1Q<{0,1,?,?}, "16">;
212 def VLD1q32 : VLD1Q<{1,0,?,?}, "32">;
213 def VLD1q64 : VLD1Q<{1,1,?,?}, "64">;
215 def VLD1q8Pseudo : VLDQPseudo<IIC_VLD1x2>;
216 def VLD1q16Pseudo : VLDQPseudo<IIC_VLD1x2>;
217 def VLD1q32Pseudo : VLDQPseudo<IIC_VLD1x2>;
218 def VLD1q64Pseudo : VLDQPseudo<IIC_VLD1x2>;
220 // ...with address register writeback:
221 class VLD1DWB<bits<4> op7_4, string Dt>
222 : NLdSt<0,0b10,0b0111,op7_4, (outs DPR:$Vd, GPR:$wb),
223 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD1u,
224 "vld1", Dt, "\\{$Vd\\}, $Rn$Rm",
225 "$Rn.addr = $wb", []> {
227 let DecoderMethod = "DecodeVLDInstruction";
229 class VLD1QWB<bits<4> op7_4, string Dt>
230 : NLdSt<0,0b10,0b1010,op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
231 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD1x2u,
232 "vld1", Dt, "\\{$Vd, $dst2\\}, $Rn$Rm",
233 "$Rn.addr = $wb", []> {
234 let Inst{5-4} = Rn{5-4};
235 let DecoderMethod = "DecodeVLDInstruction";
238 def VLD1d8_UPD : VLD1DWB<{0,0,0,?}, "8">;
239 def VLD1d16_UPD : VLD1DWB<{0,1,0,?}, "16">;
240 def VLD1d32_UPD : VLD1DWB<{1,0,0,?}, "32">;
241 def VLD1d64_UPD : VLD1DWB<{1,1,0,?}, "64">;
243 def VLD1q8_UPD : VLD1QWB<{0,0,?,?}, "8">;
244 def VLD1q16_UPD : VLD1QWB<{0,1,?,?}, "16">;
245 def VLD1q32_UPD : VLD1QWB<{1,0,?,?}, "32">;
246 def VLD1q64_UPD : VLD1QWB<{1,1,?,?}, "64">;
248 def VLD1q8Pseudo_UPD : VLDQWBPseudo<IIC_VLD1x2u>;
249 def VLD1q16Pseudo_UPD : VLDQWBPseudo<IIC_VLD1x2u>;
250 def VLD1q32Pseudo_UPD : VLDQWBPseudo<IIC_VLD1x2u>;
251 def VLD1q64Pseudo_UPD : VLDQWBPseudo<IIC_VLD1x2u>;
253 // ...with 3 registers (some of these are only for the disassembler):
254 class VLD1D3<bits<4> op7_4, string Dt>
255 : NLdSt<0,0b10,0b0110,op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
256 (ins addrmode6:$Rn), IIC_VLD1x3, "vld1", Dt,
257 "\\{$Vd, $dst2, $dst3\\}, $Rn", "", []> {
260 let DecoderMethod = "DecodeVLDInstruction";
262 class VLD1D3WB<bits<4> op7_4, string Dt>
263 : NLdSt<0,0b10,0b0110,op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
264 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD1x3u, "vld1", Dt,
265 "\\{$Vd, $dst2, $dst3\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
267 let DecoderMethod = "DecodeVLDInstruction";
270 def VLD1d8T : VLD1D3<{0,0,0,?}, "8">;
271 def VLD1d16T : VLD1D3<{0,1,0,?}, "16">;
272 def VLD1d32T : VLD1D3<{1,0,0,?}, "32">;
273 def VLD1d64T : VLD1D3<{1,1,0,?}, "64">;
275 def VLD1d8T_UPD : VLD1D3WB<{0,0,0,?}, "8">;
276 def VLD1d16T_UPD : VLD1D3WB<{0,1,0,?}, "16">;
277 def VLD1d32T_UPD : VLD1D3WB<{1,0,0,?}, "32">;
278 def VLD1d64T_UPD : VLD1D3WB<{1,1,0,?}, "64">;
280 def VLD1d64TPseudo : VLDQQPseudo<IIC_VLD1x3>;
281 def VLD1d64TPseudo_UPD : VLDQQWBPseudo<IIC_VLD1x3u>;
283 // ...with 4 registers (some of these are only for the disassembler):
284 class VLD1D4<bits<4> op7_4, string Dt>
285 : NLdSt<0,0b10,0b0010,op7_4,(outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
286 (ins addrmode6:$Rn), IIC_VLD1x4, "vld1", Dt,
287 "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn", "", []> {
289 let Inst{5-4} = Rn{5-4};
290 let DecoderMethod = "DecodeVLDInstruction";
292 class VLD1D4WB<bits<4> op7_4, string Dt>
293 : NLdSt<0,0b10,0b0010,op7_4,
294 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
295 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD1x4u, "vld1", Dt,
296 "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn$Rm", "$Rn.addr = $wb",
298 let Inst{5-4} = Rn{5-4};
299 let DecoderMethod = "DecodeVLDInstruction";
302 def VLD1d8Q : VLD1D4<{0,0,?,?}, "8">;
303 def VLD1d16Q : VLD1D4<{0,1,?,?}, "16">;
304 def VLD1d32Q : VLD1D4<{1,0,?,?}, "32">;
305 def VLD1d64Q : VLD1D4<{1,1,?,?}, "64">;
307 def VLD1d8Q_UPD : VLD1D4WB<{0,0,?,?}, "8">;
308 def VLD1d16Q_UPD : VLD1D4WB<{0,1,?,?}, "16">;
309 def VLD1d32Q_UPD : VLD1D4WB<{1,0,?,?}, "32">;
310 def VLD1d64Q_UPD : VLD1D4WB<{1,1,?,?}, "64">;
312 def VLD1d64QPseudo : VLDQQPseudo<IIC_VLD1x4>;
313 def VLD1d64QPseudo_UPD : VLDQQWBPseudo<IIC_VLD1x4u>;
315 // VLD2 : Vector Load (multiple 2-element structures)
316 class VLD2D<bits<4> op11_8, bits<4> op7_4, string Dt>
317 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2),
318 (ins addrmode6:$Rn), IIC_VLD2,
319 "vld2", Dt, "\\{$Vd, $dst2\\}, $Rn", "", []> {
321 let Inst{5-4} = Rn{5-4};
322 let DecoderMethod = "DecodeVLDInstruction";
324 class VLD2Q<bits<4> op7_4, string Dt>
325 : NLdSt<0, 0b10, 0b0011, op7_4,
326 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
327 (ins addrmode6:$Rn), IIC_VLD2x2,
328 "vld2", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn", "", []> {
330 let Inst{5-4} = Rn{5-4};
331 let DecoderMethod = "DecodeVLDInstruction";
334 def VLD2d8 : VLD2D<0b1000, {0,0,?,?}, "8">;
335 def VLD2d16 : VLD2D<0b1000, {0,1,?,?}, "16">;
336 def VLD2d32 : VLD2D<0b1000, {1,0,?,?}, "32">;
338 def VLD2q8 : VLD2Q<{0,0,?,?}, "8">;
339 def VLD2q16 : VLD2Q<{0,1,?,?}, "16">;
340 def VLD2q32 : VLD2Q<{1,0,?,?}, "32">;
342 def VLD2d8Pseudo : VLDQPseudo<IIC_VLD2>;
343 def VLD2d16Pseudo : VLDQPseudo<IIC_VLD2>;
344 def VLD2d32Pseudo : VLDQPseudo<IIC_VLD2>;
346 def VLD2q8Pseudo : VLDQQPseudo<IIC_VLD2x2>;
347 def VLD2q16Pseudo : VLDQQPseudo<IIC_VLD2x2>;
348 def VLD2q32Pseudo : VLDQQPseudo<IIC_VLD2x2>;
350 // ...with address register writeback:
351 class VLD2DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
352 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
353 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD2u,
354 "vld2", Dt, "\\{$Vd, $dst2\\}, $Rn$Rm",
355 "$Rn.addr = $wb", []> {
356 let Inst{5-4} = Rn{5-4};
357 let DecoderMethod = "DecodeVLDInstruction";
359 class VLD2QWB<bits<4> op7_4, string Dt>
360 : NLdSt<0, 0b10, 0b0011, op7_4,
361 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
362 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD2x2u,
363 "vld2", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn$Rm",
364 "$Rn.addr = $wb", []> {
365 let Inst{5-4} = Rn{5-4};
366 let DecoderMethod = "DecodeVLDInstruction";
369 def VLD2d8_UPD : VLD2DWB<0b1000, {0,0,?,?}, "8">;
370 def VLD2d16_UPD : VLD2DWB<0b1000, {0,1,?,?}, "16">;
371 def VLD2d32_UPD : VLD2DWB<0b1000, {1,0,?,?}, "32">;
373 def VLD2q8_UPD : VLD2QWB<{0,0,?,?}, "8">;
374 def VLD2q16_UPD : VLD2QWB<{0,1,?,?}, "16">;
375 def VLD2q32_UPD : VLD2QWB<{1,0,?,?}, "32">;
377 def VLD2d8Pseudo_UPD : VLDQWBPseudo<IIC_VLD2u>;
378 def VLD2d16Pseudo_UPD : VLDQWBPseudo<IIC_VLD2u>;
379 def VLD2d32Pseudo_UPD : VLDQWBPseudo<IIC_VLD2u>;
381 def VLD2q8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD2x2u>;
382 def VLD2q16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD2x2u>;
383 def VLD2q32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD2x2u>;
385 // ...with double-spaced registers (for disassembly only):
386 def VLD2b8 : VLD2D<0b1001, {0,0,?,?}, "8">;
387 def VLD2b16 : VLD2D<0b1001, {0,1,?,?}, "16">;
388 def VLD2b32 : VLD2D<0b1001, {1,0,?,?}, "32">;
389 def VLD2b8_UPD : VLD2DWB<0b1001, {0,0,?,?}, "8">;
390 def VLD2b16_UPD : VLD2DWB<0b1001, {0,1,?,?}, "16">;
391 def VLD2b32_UPD : VLD2DWB<0b1001, {1,0,?,?}, "32">;
393 // VLD3 : Vector Load (multiple 3-element structures)
394 class VLD3D<bits<4> op11_8, bits<4> op7_4, string Dt>
395 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
396 (ins addrmode6:$Rn), IIC_VLD3,
397 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn", "", []> {
400 let DecoderMethod = "DecodeVLDInstruction";
403 def VLD3d8 : VLD3D<0b0100, {0,0,0,?}, "8">;
404 def VLD3d16 : VLD3D<0b0100, {0,1,0,?}, "16">;
405 def VLD3d32 : VLD3D<0b0100, {1,0,0,?}, "32">;
407 def VLD3d8Pseudo : VLDQQPseudo<IIC_VLD3>;
408 def VLD3d16Pseudo : VLDQQPseudo<IIC_VLD3>;
409 def VLD3d32Pseudo : VLDQQPseudo<IIC_VLD3>;
411 // ...with address register writeback:
412 class VLD3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
413 : NLdSt<0, 0b10, op11_8, op7_4,
414 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
415 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD3u,
416 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn$Rm",
417 "$Rn.addr = $wb", []> {
419 let DecoderMethod = "DecodeVLDInstruction";
422 def VLD3d8_UPD : VLD3DWB<0b0100, {0,0,0,?}, "8">;
423 def VLD3d16_UPD : VLD3DWB<0b0100, {0,1,0,?}, "16">;
424 def VLD3d32_UPD : VLD3DWB<0b0100, {1,0,0,?}, "32">;
426 def VLD3d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
427 def VLD3d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
428 def VLD3d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
430 // ...with double-spaced registers:
431 def VLD3q8 : VLD3D<0b0101, {0,0,0,?}, "8">;
432 def VLD3q16 : VLD3D<0b0101, {0,1,0,?}, "16">;
433 def VLD3q32 : VLD3D<0b0101, {1,0,0,?}, "32">;
434 def VLD3q8_UPD : VLD3DWB<0b0101, {0,0,0,?}, "8">;
435 def VLD3q16_UPD : VLD3DWB<0b0101, {0,1,0,?}, "16">;
436 def VLD3q32_UPD : VLD3DWB<0b0101, {1,0,0,?}, "32">;
438 def VLD3q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
439 def VLD3q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
440 def VLD3q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
442 // ...alternate versions to be allocated odd register numbers:
443 def VLD3q8oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
444 def VLD3q16oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
445 def VLD3q32oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
447 def VLD3q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
448 def VLD3q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
449 def VLD3q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
451 // VLD4 : Vector Load (multiple 4-element structures)
452 class VLD4D<bits<4> op11_8, bits<4> op7_4, string Dt>
453 : NLdSt<0, 0b10, op11_8, op7_4,
454 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
455 (ins addrmode6:$Rn), IIC_VLD4,
456 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn", "", []> {
458 let Inst{5-4} = Rn{5-4};
459 let DecoderMethod = "DecodeVLDInstruction";
462 def VLD4d8 : VLD4D<0b0000, {0,0,?,?}, "8">;
463 def VLD4d16 : VLD4D<0b0000, {0,1,?,?}, "16">;
464 def VLD4d32 : VLD4D<0b0000, {1,0,?,?}, "32">;
466 def VLD4d8Pseudo : VLDQQPseudo<IIC_VLD4>;
467 def VLD4d16Pseudo : VLDQQPseudo<IIC_VLD4>;
468 def VLD4d32Pseudo : VLDQQPseudo<IIC_VLD4>;
470 // ...with address register writeback:
471 class VLD4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
472 : NLdSt<0, 0b10, op11_8, op7_4,
473 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
474 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD4u,
475 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn$Rm",
476 "$Rn.addr = $wb", []> {
477 let Inst{5-4} = Rn{5-4};
478 let DecoderMethod = "DecodeVLDInstruction";
481 def VLD4d8_UPD : VLD4DWB<0b0000, {0,0,?,?}, "8">;
482 def VLD4d16_UPD : VLD4DWB<0b0000, {0,1,?,?}, "16">;
483 def VLD4d32_UPD : VLD4DWB<0b0000, {1,0,?,?}, "32">;
485 def VLD4d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
486 def VLD4d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
487 def VLD4d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
489 // ...with double-spaced registers:
490 def VLD4q8 : VLD4D<0b0001, {0,0,?,?}, "8">;
491 def VLD4q16 : VLD4D<0b0001, {0,1,?,?}, "16">;
492 def VLD4q32 : VLD4D<0b0001, {1,0,?,?}, "32">;
493 def VLD4q8_UPD : VLD4DWB<0b0001, {0,0,?,?}, "8">;
494 def VLD4q16_UPD : VLD4DWB<0b0001, {0,1,?,?}, "16">;
495 def VLD4q32_UPD : VLD4DWB<0b0001, {1,0,?,?}, "32">;
497 def VLD4q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
498 def VLD4q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
499 def VLD4q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
501 // ...alternate versions to be allocated odd register numbers:
502 def VLD4q8oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
503 def VLD4q16oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
504 def VLD4q32oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
506 def VLD4q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
507 def VLD4q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
508 def VLD4q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
510 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
512 // Classes for VLD*LN pseudo-instructions with multi-register operands.
513 // These are expanded to real instructions after register allocation.
514 class VLDQLNPseudo<InstrItinClass itin>
515 : PseudoNLdSt<(outs QPR:$dst),
516 (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
517 itin, "$src = $dst">;
518 class VLDQLNWBPseudo<InstrItinClass itin>
519 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
520 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
521 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
522 class VLDQQLNPseudo<InstrItinClass itin>
523 : PseudoNLdSt<(outs QQPR:$dst),
524 (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
525 itin, "$src = $dst">;
526 class VLDQQLNWBPseudo<InstrItinClass itin>
527 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
528 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
529 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
530 class VLDQQQQLNPseudo<InstrItinClass itin>
531 : PseudoNLdSt<(outs QQQQPR:$dst),
532 (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
533 itin, "$src = $dst">;
534 class VLDQQQQLNWBPseudo<InstrItinClass itin>
535 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
536 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
537 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
539 // VLD1LN : Vector Load (single element to one lane)
540 class VLD1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
542 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd),
543 (ins addrmode6:$Rn, DPR:$src, nohash_imm:$lane),
544 IIC_VLD1ln, "vld1", Dt, "\\{$Vd[$lane]\\}, $Rn",
546 [(set DPR:$Vd, (vector_insert (Ty DPR:$src),
547 (i32 (LoadOp addrmode6:$Rn)),
551 class VLD1LN32<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
553 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd),
554 (ins addrmode6oneL32:$Rn, DPR:$src, nohash_imm:$lane),
555 IIC_VLD1ln, "vld1", Dt, "\\{$Vd[$lane]\\}, $Rn",
557 [(set DPR:$Vd, (vector_insert (Ty DPR:$src),
558 (i32 (LoadOp addrmode6oneL32:$Rn)),
562 class VLD1QLNPseudo<ValueType Ty, PatFrag LoadOp> : VLDQLNPseudo<IIC_VLD1ln> {
563 let Pattern = [(set QPR:$dst, (vector_insert (Ty QPR:$src),
564 (i32 (LoadOp addrmode6:$addr)),
568 def VLD1LNd8 : VLD1LN<0b0000, {?,?,?,0}, "8", v8i8, extloadi8> {
569 let Inst{7-5} = lane{2-0};
571 def VLD1LNd16 : VLD1LN<0b0100, {?,?,0,?}, "16", v4i16, extloadi16> {
572 let Inst{7-6} = lane{1-0};
575 def VLD1LNd32 : VLD1LN32<0b1000, {?,0,?,?}, "32", v2i32, load> {
576 let Inst{7} = lane{0};
581 def VLD1LNq8Pseudo : VLD1QLNPseudo<v16i8, extloadi8>;
582 def VLD1LNq16Pseudo : VLD1QLNPseudo<v8i16, extloadi16>;
583 def VLD1LNq32Pseudo : VLD1QLNPseudo<v4i32, load>;
585 def : Pat<(vector_insert (v2f32 DPR:$src),
586 (f32 (load addrmode6:$addr)), imm:$lane),
587 (VLD1LNd32 addrmode6:$addr, DPR:$src, imm:$lane)>;
588 def : Pat<(vector_insert (v4f32 QPR:$src),
589 (f32 (load addrmode6:$addr)), imm:$lane),
590 (VLD1LNq32Pseudo addrmode6:$addr, QPR:$src, imm:$lane)>;
592 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
594 // ...with address register writeback:
595 class VLD1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
596 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, GPR:$wb),
597 (ins addrmode6:$Rn, am6offset:$Rm,
598 DPR:$src, nohash_imm:$lane), IIC_VLD1lnu, "vld1", Dt,
599 "\\{$Vd[$lane]\\}, $Rn$Rm",
600 "$src = $Vd, $Rn.addr = $wb", []>;
602 def VLD1LNd8_UPD : VLD1LNWB<0b0000, {?,?,?,0}, "8"> {
603 let Inst{7-5} = lane{2-0};
605 def VLD1LNd16_UPD : VLD1LNWB<0b0100, {?,?,0,?}, "16"> {
606 let Inst{7-6} = lane{1-0};
609 def VLD1LNd32_UPD : VLD1LNWB<0b1000, {?,0,?,?}, "32"> {
610 let Inst{7} = lane{0};
615 def VLD1LNq8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
616 def VLD1LNq16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
617 def VLD1LNq32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
619 // VLD2LN : Vector Load (single 2-element structure to one lane)
620 class VLD2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
621 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2),
622 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, nohash_imm:$lane),
623 IIC_VLD2ln, "vld2", Dt, "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn",
624 "$src1 = $Vd, $src2 = $dst2", []> {
629 def VLD2LNd8 : VLD2LN<0b0001, {?,?,?,?}, "8"> {
630 let Inst{7-5} = lane{2-0};
632 def VLD2LNd16 : VLD2LN<0b0101, {?,?,0,?}, "16"> {
633 let Inst{7-6} = lane{1-0};
635 def VLD2LNd32 : VLD2LN<0b1001, {?,0,0,?}, "32"> {
636 let Inst{7} = lane{0};
639 def VLD2LNd8Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
640 def VLD2LNd16Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
641 def VLD2LNd32Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
643 // ...with double-spaced registers:
644 def VLD2LNq16 : VLD2LN<0b0101, {?,?,1,?}, "16"> {
645 let Inst{7-6} = lane{1-0};
647 def VLD2LNq32 : VLD2LN<0b1001, {?,1,0,?}, "32"> {
648 let Inst{7} = lane{0};
651 def VLD2LNq16Pseudo : VLDQQLNPseudo<IIC_VLD2ln>;
652 def VLD2LNq32Pseudo : VLDQQLNPseudo<IIC_VLD2ln>;
654 // ...with address register writeback:
655 class VLD2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
656 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
657 (ins addrmode6:$Rn, am6offset:$Rm,
658 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VLD2lnu, "vld2", Dt,
659 "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn$Rm",
660 "$src1 = $Vd, $src2 = $dst2, $Rn.addr = $wb", []> {
664 def VLD2LNd8_UPD : VLD2LNWB<0b0001, {?,?,?,?}, "8"> {
665 let Inst{7-5} = lane{2-0};
667 def VLD2LNd16_UPD : VLD2LNWB<0b0101, {?,?,0,?}, "16"> {
668 let Inst{7-6} = lane{1-0};
670 def VLD2LNd32_UPD : VLD2LNWB<0b1001, {?,0,0,?}, "32"> {
671 let Inst{7} = lane{0};
674 def VLD2LNd8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
675 def VLD2LNd16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
676 def VLD2LNd32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
678 def VLD2LNq16_UPD : VLD2LNWB<0b0101, {?,?,1,?}, "16"> {
679 let Inst{7-6} = lane{1-0};
681 def VLD2LNq32_UPD : VLD2LNWB<0b1001, {?,1,0,?}, "32"> {
682 let Inst{7} = lane{0};
685 def VLD2LNq16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
686 def VLD2LNq32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
688 // VLD3LN : Vector Load (single 3-element structure to one lane)
689 class VLD3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
690 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
691 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3,
692 nohash_imm:$lane), IIC_VLD3ln, "vld3", Dt,
693 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn",
694 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3", []> {
698 def VLD3LNd8 : VLD3LN<0b0010, {?,?,?,0}, "8"> {
699 let Inst{7-5} = lane{2-0};
701 def VLD3LNd16 : VLD3LN<0b0110, {?,?,0,0}, "16"> {
702 let Inst{7-6} = lane{1-0};
704 def VLD3LNd32 : VLD3LN<0b1010, {?,0,0,0}, "32"> {
705 let Inst{7} = lane{0};
708 def VLD3LNd8Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
709 def VLD3LNd16Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
710 def VLD3LNd32Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
712 // ...with double-spaced registers:
713 def VLD3LNq16 : VLD3LN<0b0110, {?,?,1,0}, "16"> {
714 let Inst{7-6} = lane{1-0};
716 def VLD3LNq32 : VLD3LN<0b1010, {?,1,0,0}, "32"> {
717 let Inst{7} = lane{0};
720 def VLD3LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
721 def VLD3LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
723 // ...with address register writeback:
724 class VLD3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
725 : NLdStLn<1, 0b10, op11_8, op7_4,
726 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
727 (ins addrmode6:$Rn, am6offset:$Rm,
728 DPR:$src1, DPR:$src2, DPR:$src3, nohash_imm:$lane),
729 IIC_VLD3lnu, "vld3", Dt,
730 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn$Rm",
731 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $Rn.addr = $wb",
734 def VLD3LNd8_UPD : VLD3LNWB<0b0010, {?,?,?,0}, "8"> {
735 let Inst{7-5} = lane{2-0};
737 def VLD3LNd16_UPD : VLD3LNWB<0b0110, {?,?,0,0}, "16"> {
738 let Inst{7-6} = lane{1-0};
740 def VLD3LNd32_UPD : VLD3LNWB<0b1010, {?,0,0,0}, "32"> {
741 let Inst{7} = lane{0};
744 def VLD3LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
745 def VLD3LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
746 def VLD3LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
748 def VLD3LNq16_UPD : VLD3LNWB<0b0110, {?,?,1,0}, "16"> {
749 let Inst{7-6} = lane{1-0};
751 def VLD3LNq32_UPD : VLD3LNWB<0b1010, {?,1,0,0}, "32"> {
752 let Inst{7} = lane{0};
755 def VLD3LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
756 def VLD3LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
758 // VLD4LN : Vector Load (single 4-element structure to one lane)
759 class VLD4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
760 : NLdStLn<1, 0b10, op11_8, op7_4,
761 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
762 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
763 nohash_imm:$lane), IIC_VLD4ln, "vld4", Dt,
764 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn",
765 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4", []> {
770 def VLD4LNd8 : VLD4LN<0b0011, {?,?,?,?}, "8"> {
771 let Inst{7-5} = lane{2-0};
773 def VLD4LNd16 : VLD4LN<0b0111, {?,?,0,?}, "16"> {
774 let Inst{7-6} = lane{1-0};
776 def VLD4LNd32 : VLD4LN<0b1011, {?,0,?,?}, "32"> {
777 let Inst{7} = lane{0};
781 def VLD4LNd8Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
782 def VLD4LNd16Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
783 def VLD4LNd32Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
785 // ...with double-spaced registers:
786 def VLD4LNq16 : VLD4LN<0b0111, {?,?,1,?}, "16"> {
787 let Inst{7-6} = lane{1-0};
789 def VLD4LNq32 : VLD4LN<0b1011, {?,1,?,?}, "32"> {
790 let Inst{7} = lane{0};
794 def VLD4LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>;
795 def VLD4LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>;
797 // ...with address register writeback:
798 class VLD4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
799 : NLdStLn<1, 0b10, op11_8, op7_4,
800 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
801 (ins addrmode6:$Rn, am6offset:$Rm,
802 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
803 IIC_VLD4lnu, "vld4", Dt,
804 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn$Rm",
805 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4, $Rn.addr = $wb",
810 def VLD4LNd8_UPD : VLD4LNWB<0b0011, {?,?,?,?}, "8"> {
811 let Inst{7-5} = lane{2-0};
813 def VLD4LNd16_UPD : VLD4LNWB<0b0111, {?,?,0,?}, "16"> {
814 let Inst{7-6} = lane{1-0};
816 def VLD4LNd32_UPD : VLD4LNWB<0b1011, {?,0,?,?}, "32"> {
817 let Inst{7} = lane{0};
821 def VLD4LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
822 def VLD4LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
823 def VLD4LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
825 def VLD4LNq16_UPD : VLD4LNWB<0b0111, {?,?,1,?}, "16"> {
826 let Inst{7-6} = lane{1-0};
828 def VLD4LNq32_UPD : VLD4LNWB<0b1011, {?,1,?,?}, "32"> {
829 let Inst{7} = lane{0};
833 def VLD4LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
834 def VLD4LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
836 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
838 // VLD1DUP : Vector Load (single element to all lanes)
839 class VLD1DUP<bits<4> op7_4, string Dt, ValueType Ty, PatFrag LoadOp>
840 : NLdSt<1, 0b10, 0b1100, op7_4, (outs DPR:$Vd), (ins addrmode6dup:$Rn),
841 IIC_VLD1dup, "vld1", Dt, "\\{$Vd[]\\}, $Rn", "",
842 [(set DPR:$Vd, (Ty (NEONvdup (i32 (LoadOp addrmode6dup:$Rn)))))]> {
845 let DecoderMethod = "DecodeVLD1DupInstruction";
847 class VLD1QDUPPseudo<ValueType Ty, PatFrag LoadOp> : VLDQPseudo<IIC_VLD1dup> {
848 let Pattern = [(set QPR:$dst,
849 (Ty (NEONvdup (i32 (LoadOp addrmode6dup:$addr)))))];
852 def VLD1DUPd8 : VLD1DUP<{0,0,0,?}, "8", v8i8, extloadi8>;
853 def VLD1DUPd16 : VLD1DUP<{0,1,0,?}, "16", v4i16, extloadi16>;
854 def VLD1DUPd32 : VLD1DUP<{1,0,0,?}, "32", v2i32, load>;
856 def VLD1DUPq8Pseudo : VLD1QDUPPseudo<v16i8, extloadi8>;
857 def VLD1DUPq16Pseudo : VLD1QDUPPseudo<v8i16, extloadi16>;
858 def VLD1DUPq32Pseudo : VLD1QDUPPseudo<v4i32, load>;
860 def : Pat<(v2f32 (NEONvdup (f32 (load addrmode6dup:$addr)))),
861 (VLD1DUPd32 addrmode6:$addr)>;
862 def : Pat<(v4f32 (NEONvdup (f32 (load addrmode6dup:$addr)))),
863 (VLD1DUPq32Pseudo addrmode6:$addr)>;
865 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
867 class VLD1QDUP<bits<4> op7_4, string Dt>
868 : NLdSt<1, 0b10, 0b1100, op7_4, (outs DPR:$Vd, DPR:$dst2),
869 (ins addrmode6dup:$Rn), IIC_VLD1dup,
870 "vld1", Dt, "\\{$Vd[], $dst2[]\\}, $Rn", "", []> {
873 let DecoderMethod = "DecodeVLD1DupInstruction";
876 def VLD1DUPq8 : VLD1QDUP<{0,0,1,0}, "8">;
877 def VLD1DUPq16 : VLD1QDUP<{0,1,1,?}, "16">;
878 def VLD1DUPq32 : VLD1QDUP<{1,0,1,?}, "32">;
880 // ...with address register writeback:
881 class VLD1DUPWB<bits<4> op7_4, string Dt>
882 : NLdSt<1, 0b10, 0b1100, op7_4, (outs DPR:$Vd, GPR:$wb),
883 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD1dupu,
884 "vld1", Dt, "\\{$Vd[]\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
886 let DecoderMethod = "DecodeVLD1DupInstruction";
888 class VLD1QDUPWB<bits<4> op7_4, string Dt>
889 : NLdSt<1, 0b10, 0b1100, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
890 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD1dupu,
891 "vld1", Dt, "\\{$Vd[], $dst2[]\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
893 let DecoderMethod = "DecodeVLD1DupInstruction";
896 def VLD1DUPd8_UPD : VLD1DUPWB<{0,0,0,0}, "8">;
897 def VLD1DUPd16_UPD : VLD1DUPWB<{0,1,0,?}, "16">;
898 def VLD1DUPd32_UPD : VLD1DUPWB<{1,0,0,?}, "32">;
900 def VLD1DUPq8_UPD : VLD1QDUPWB<{0,0,1,0}, "8">;
901 def VLD1DUPq16_UPD : VLD1QDUPWB<{0,1,1,?}, "16">;
902 def VLD1DUPq32_UPD : VLD1QDUPWB<{1,0,1,?}, "32">;
904 def VLD1DUPq8Pseudo_UPD : VLDQWBPseudo<IIC_VLD1dupu>;
905 def VLD1DUPq16Pseudo_UPD : VLDQWBPseudo<IIC_VLD1dupu>;
906 def VLD1DUPq32Pseudo_UPD : VLDQWBPseudo<IIC_VLD1dupu>;
908 // VLD2DUP : Vector Load (single 2-element structure to all lanes)
909 class VLD2DUP<bits<4> op7_4, string Dt>
910 : NLdSt<1, 0b10, 0b1101, op7_4, (outs DPR:$Vd, DPR:$dst2),
911 (ins addrmode6dup:$Rn), IIC_VLD2dup,
912 "vld2", Dt, "\\{$Vd[], $dst2[]\\}, $Rn", "", []> {
915 let DecoderMethod = "DecodeVLD2DupInstruction";
918 def VLD2DUPd8 : VLD2DUP<{0,0,0,?}, "8">;
919 def VLD2DUPd16 : VLD2DUP<{0,1,0,?}, "16">;
920 def VLD2DUPd32 : VLD2DUP<{1,0,0,?}, "32">;
922 def VLD2DUPd8Pseudo : VLDQPseudo<IIC_VLD2dup>;
923 def VLD2DUPd16Pseudo : VLDQPseudo<IIC_VLD2dup>;
924 def VLD2DUPd32Pseudo : VLDQPseudo<IIC_VLD2dup>;
926 // ...with double-spaced registers (not used for codegen):
927 def VLD2DUPd8x2 : VLD2DUP<{0,0,1,?}, "8">;
928 def VLD2DUPd16x2 : VLD2DUP<{0,1,1,?}, "16">;
929 def VLD2DUPd32x2 : VLD2DUP<{1,0,1,?}, "32">;
931 // ...with address register writeback:
932 class VLD2DUPWB<bits<4> op7_4, string Dt>
933 : NLdSt<1, 0b10, 0b1101, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
934 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD2dupu,
935 "vld2", Dt, "\\{$Vd[], $dst2[]\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
937 let DecoderMethod = "DecodeVLD2DupInstruction";
940 def VLD2DUPd8_UPD : VLD2DUPWB<{0,0,0,0}, "8">;
941 def VLD2DUPd16_UPD : VLD2DUPWB<{0,1,0,?}, "16">;
942 def VLD2DUPd32_UPD : VLD2DUPWB<{1,0,0,?}, "32">;
944 def VLD2DUPd8x2_UPD : VLD2DUPWB<{0,0,1,0}, "8">;
945 def VLD2DUPd16x2_UPD : VLD2DUPWB<{0,1,1,?}, "16">;
946 def VLD2DUPd32x2_UPD : VLD2DUPWB<{1,0,1,?}, "32">;
948 def VLD2DUPd8Pseudo_UPD : VLDQWBPseudo<IIC_VLD2dupu>;
949 def VLD2DUPd16Pseudo_UPD : VLDQWBPseudo<IIC_VLD2dupu>;
950 def VLD2DUPd32Pseudo_UPD : VLDQWBPseudo<IIC_VLD2dupu>;
952 // VLD3DUP : Vector Load (single 3-element structure to all lanes)
953 class VLD3DUP<bits<4> op7_4, string Dt>
954 : NLdSt<1, 0b10, 0b1110, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
955 (ins addrmode6dup:$Rn), IIC_VLD3dup,
956 "vld3", Dt, "\\{$Vd[], $dst2[], $dst3[]\\}, $Rn", "", []> {
959 let DecoderMethod = "DecodeVLD3DupInstruction";
962 def VLD3DUPd8 : VLD3DUP<{0,0,0,?}, "8">;
963 def VLD3DUPd16 : VLD3DUP<{0,1,0,?}, "16">;
964 def VLD3DUPd32 : VLD3DUP<{1,0,0,?}, "32">;
966 def VLD3DUPd8Pseudo : VLDQQPseudo<IIC_VLD3dup>;
967 def VLD3DUPd16Pseudo : VLDQQPseudo<IIC_VLD3dup>;
968 def VLD3DUPd32Pseudo : VLDQQPseudo<IIC_VLD3dup>;
970 // ...with double-spaced registers (not used for codegen):
971 def VLD3DUPd8x2 : VLD3DUP<{0,0,1,?}, "8">;
972 def VLD3DUPd16x2 : VLD3DUP<{0,1,1,?}, "16">;
973 def VLD3DUPd32x2 : VLD3DUP<{1,0,1,?}, "32">;
975 // ...with address register writeback:
976 class VLD3DUPWB<bits<4> op7_4, string Dt>
977 : NLdSt<1, 0b10, 0b1110, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
978 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD3dupu,
979 "vld3", Dt, "\\{$Vd[], $dst2[], $dst3[]\\}, $Rn$Rm",
980 "$Rn.addr = $wb", []> {
982 let DecoderMethod = "DecodeVLD3DupInstruction";
985 def VLD3DUPd8_UPD : VLD3DUPWB<{0,0,0,0}, "8">;
986 def VLD3DUPd16_UPD : VLD3DUPWB<{0,1,0,?}, "16">;
987 def VLD3DUPd32_UPD : VLD3DUPWB<{1,0,0,?}, "32">;
989 def VLD3DUPd8x2_UPD : VLD3DUPWB<{0,0,1,0}, "8">;
990 def VLD3DUPd16x2_UPD : VLD3DUPWB<{0,1,1,?}, "16">;
991 def VLD3DUPd32x2_UPD : VLD3DUPWB<{1,0,1,?}, "32">;
993 def VLD3DUPd8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
994 def VLD3DUPd16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
995 def VLD3DUPd32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
997 // VLD4DUP : Vector Load (single 4-element structure to all lanes)
998 class VLD4DUP<bits<4> op7_4, string Dt>
999 : NLdSt<1, 0b10, 0b1111, op7_4,
1000 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
1001 (ins addrmode6dup:$Rn), IIC_VLD4dup,
1002 "vld4", Dt, "\\{$Vd[], $dst2[], $dst3[], $dst4[]\\}, $Rn", "", []> {
1004 let Inst{4} = Rn{4};
1005 let DecoderMethod = "DecodeVLD4DupInstruction";
1008 def VLD4DUPd8 : VLD4DUP<{0,0,0,?}, "8">;
1009 def VLD4DUPd16 : VLD4DUP<{0,1,0,?}, "16">;
1010 def VLD4DUPd32 : VLD4DUP<{1,?,0,?}, "32"> { let Inst{6} = Rn{5}; }
1012 def VLD4DUPd8Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1013 def VLD4DUPd16Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1014 def VLD4DUPd32Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1016 // ...with double-spaced registers (not used for codegen):
1017 def VLD4DUPd8x2 : VLD4DUP<{0,0,1,?}, "8">;
1018 def VLD4DUPd16x2 : VLD4DUP<{0,1,1,?}, "16">;
1019 def VLD4DUPd32x2 : VLD4DUP<{1,?,1,?}, "32"> { let Inst{6} = Rn{5}; }
1021 // ...with address register writeback:
1022 class VLD4DUPWB<bits<4> op7_4, string Dt>
1023 : NLdSt<1, 0b10, 0b1111, op7_4,
1024 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
1025 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD4dupu,
1026 "vld4", Dt, "\\{$Vd[], $dst2[], $dst3[], $dst4[]\\}, $Rn$Rm",
1027 "$Rn.addr = $wb", []> {
1028 let Inst{4} = Rn{4};
1029 let DecoderMethod = "DecodeVLD4DupInstruction";
1032 def VLD4DUPd8_UPD : VLD4DUPWB<{0,0,0,0}, "8">;
1033 def VLD4DUPd16_UPD : VLD4DUPWB<{0,1,0,?}, "16">;
1034 def VLD4DUPd32_UPD : VLD4DUPWB<{1,?,0,?}, "32"> { let Inst{6} = Rn{5}; }
1036 def VLD4DUPd8x2_UPD : VLD4DUPWB<{0,0,1,0}, "8">;
1037 def VLD4DUPd16x2_UPD : VLD4DUPWB<{0,1,1,?}, "16">;
1038 def VLD4DUPd32x2_UPD : VLD4DUPWB<{1,?,1,?}, "32"> { let Inst{6} = Rn{5}; }
1040 def VLD4DUPd8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1041 def VLD4DUPd16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1042 def VLD4DUPd32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1044 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
1046 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
1048 // Classes for VST* pseudo-instructions with multi-register operands.
1049 // These are expanded to real instructions after register allocation.
1050 class VSTQPseudo<InstrItinClass itin>
1051 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src), itin, "">;
1052 class VSTQWBPseudo<InstrItinClass itin>
1053 : PseudoNLdSt<(outs GPR:$wb),
1054 (ins addrmode6:$addr, am6offset:$offset, QPR:$src), itin,
1055 "$addr.addr = $wb">;
1056 class VSTQQPseudo<InstrItinClass itin>
1057 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src), itin, "">;
1058 class VSTQQWBPseudo<InstrItinClass itin>
1059 : PseudoNLdSt<(outs GPR:$wb),
1060 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src), itin,
1061 "$addr.addr = $wb">;
1062 class VSTQQQQPseudo<InstrItinClass itin>
1063 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src), itin, "">;
1064 class VSTQQQQWBPseudo<InstrItinClass itin>
1065 : PseudoNLdSt<(outs GPR:$wb),
1066 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
1067 "$addr.addr = $wb">;
1069 // VST1 : Vector Store (multiple single elements)
1070 class VST1D<bits<4> op7_4, string Dt>
1071 : NLdSt<0,0b00,0b0111,op7_4, (outs), (ins addrmode6:$Rn, DPR:$Vd),
1072 IIC_VST1, "vst1", Dt, "\\{$Vd\\}, $Rn", "", []> {
1074 let Inst{4} = Rn{4};
1075 let DecoderMethod = "DecodeVSTInstruction";
1077 class VST1Q<bits<4> op7_4, string Dt>
1078 : NLdSt<0,0b00,0b1010,op7_4, (outs),
1079 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2), IIC_VST1x2,
1080 "vst1", Dt, "\\{$Vd, $src2\\}, $Rn", "", []> {
1082 let Inst{5-4} = Rn{5-4};
1083 let DecoderMethod = "DecodeVSTInstruction";
1086 def VST1d8 : VST1D<{0,0,0,?}, "8">;
1087 def VST1d16 : VST1D<{0,1,0,?}, "16">;
1088 def VST1d32 : VST1D<{1,0,0,?}, "32">;
1089 def VST1d64 : VST1D<{1,1,0,?}, "64">;
1091 def VST1q8 : VST1Q<{0,0,?,?}, "8">;
1092 def VST1q16 : VST1Q<{0,1,?,?}, "16">;
1093 def VST1q32 : VST1Q<{1,0,?,?}, "32">;
1094 def VST1q64 : VST1Q<{1,1,?,?}, "64">;
1096 def VST1q8Pseudo : VSTQPseudo<IIC_VST1x2>;
1097 def VST1q16Pseudo : VSTQPseudo<IIC_VST1x2>;
1098 def VST1q32Pseudo : VSTQPseudo<IIC_VST1x2>;
1099 def VST1q64Pseudo : VSTQPseudo<IIC_VST1x2>;
1101 // ...with address register writeback:
1102 class VST1DWB<bits<4> op7_4, string Dt>
1103 : NLdSt<0, 0b00, 0b0111, op7_4, (outs GPR:$wb),
1104 (ins addrmode6:$Rn, am6offset:$Rm, DPR:$Vd), IIC_VST1u,
1105 "vst1", Dt, "\\{$Vd\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
1106 let Inst{4} = Rn{4};
1107 let DecoderMethod = "DecodeVSTInstruction";
1109 class VST1QWB<bits<4> op7_4, string Dt>
1110 : NLdSt<0, 0b00, 0b1010, op7_4, (outs GPR:$wb),
1111 (ins addrmode6:$Rn, am6offset:$Rm, DPR:$Vd, DPR:$src2),
1112 IIC_VST1x2u, "vst1", Dt, "\\{$Vd, $src2\\}, $Rn$Rm",
1113 "$Rn.addr = $wb", []> {
1114 let Inst{5-4} = Rn{5-4};
1115 let DecoderMethod = "DecodeVSTInstruction";
1118 def VST1d8_UPD : VST1DWB<{0,0,0,?}, "8">;
1119 def VST1d16_UPD : VST1DWB<{0,1,0,?}, "16">;
1120 def VST1d32_UPD : VST1DWB<{1,0,0,?}, "32">;
1121 def VST1d64_UPD : VST1DWB<{1,1,0,?}, "64">;
1123 def VST1q8_UPD : VST1QWB<{0,0,?,?}, "8">;
1124 def VST1q16_UPD : VST1QWB<{0,1,?,?}, "16">;
1125 def VST1q32_UPD : VST1QWB<{1,0,?,?}, "32">;
1126 def VST1q64_UPD : VST1QWB<{1,1,?,?}, "64">;
1128 def VST1q8Pseudo_UPD : VSTQWBPseudo<IIC_VST1x2u>;
1129 def VST1q16Pseudo_UPD : VSTQWBPseudo<IIC_VST1x2u>;
1130 def VST1q32Pseudo_UPD : VSTQWBPseudo<IIC_VST1x2u>;
1131 def VST1q64Pseudo_UPD : VSTQWBPseudo<IIC_VST1x2u>;
1133 // ...with 3 registers (some of these are only for the disassembler):
1134 class VST1D3<bits<4> op7_4, string Dt>
1135 : NLdSt<0, 0b00, 0b0110, op7_4, (outs),
1136 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3),
1137 IIC_VST1x3, "vst1", Dt, "\\{$Vd, $src2, $src3\\}, $Rn", "", []> {
1139 let Inst{4} = Rn{4};
1140 let DecoderMethod = "DecodeVSTInstruction";
1142 class VST1D3WB<bits<4> op7_4, string Dt>
1143 : NLdSt<0, 0b00, 0b0110, op7_4, (outs GPR:$wb),
1144 (ins addrmode6:$Rn, am6offset:$Rm,
1145 DPR:$Vd, DPR:$src2, DPR:$src3),
1146 IIC_VST1x3u, "vst1", Dt, "\\{$Vd, $src2, $src3\\}, $Rn$Rm",
1147 "$Rn.addr = $wb", []> {
1148 let Inst{4} = Rn{4};
1149 let DecoderMethod = "DecodeVSTInstruction";
1152 def VST1d8T : VST1D3<{0,0,0,?}, "8">;
1153 def VST1d16T : VST1D3<{0,1,0,?}, "16">;
1154 def VST1d32T : VST1D3<{1,0,0,?}, "32">;
1155 def VST1d64T : VST1D3<{1,1,0,?}, "64">;
1157 def VST1d8T_UPD : VST1D3WB<{0,0,0,?}, "8">;
1158 def VST1d16T_UPD : VST1D3WB<{0,1,0,?}, "16">;
1159 def VST1d32T_UPD : VST1D3WB<{1,0,0,?}, "32">;
1160 def VST1d64T_UPD : VST1D3WB<{1,1,0,?}, "64">;
1162 def VST1d64TPseudo : VSTQQPseudo<IIC_VST1x3>;
1163 def VST1d64TPseudo_UPD : VSTQQWBPseudo<IIC_VST1x3u>;
1165 // ...with 4 registers (some of these are only for the disassembler):
1166 class VST1D4<bits<4> op7_4, string Dt>
1167 : NLdSt<0, 0b00, 0b0010, op7_4, (outs),
1168 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
1169 IIC_VST1x4, "vst1", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn", "",
1172 let Inst{5-4} = Rn{5-4};
1173 let DecoderMethod = "DecodeVSTInstruction";
1175 class VST1D4WB<bits<4> op7_4, string Dt>
1176 : NLdSt<0, 0b00, 0b0010, op7_4, (outs GPR:$wb),
1177 (ins addrmode6:$Rn, am6offset:$Rm,
1178 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST1x4u,
1179 "vst1", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
1180 "$Rn.addr = $wb", []> {
1181 let Inst{5-4} = Rn{5-4};
1182 let DecoderMethod = "DecodeVSTInstruction";
1185 def VST1d8Q : VST1D4<{0,0,?,?}, "8">;
1186 def VST1d16Q : VST1D4<{0,1,?,?}, "16">;
1187 def VST1d32Q : VST1D4<{1,0,?,?}, "32">;
1188 def VST1d64Q : VST1D4<{1,1,?,?}, "64">;
1190 def VST1d8Q_UPD : VST1D4WB<{0,0,?,?}, "8">;
1191 def VST1d16Q_UPD : VST1D4WB<{0,1,?,?}, "16">;
1192 def VST1d32Q_UPD : VST1D4WB<{1,0,?,?}, "32">;
1193 def VST1d64Q_UPD : VST1D4WB<{1,1,?,?}, "64">;
1195 def VST1d64QPseudo : VSTQQPseudo<IIC_VST1x4>;
1196 def VST1d64QPseudo_UPD : VSTQQWBPseudo<IIC_VST1x4u>;
1198 // VST2 : Vector Store (multiple 2-element structures)
1199 class VST2D<bits<4> op11_8, bits<4> op7_4, string Dt>
1200 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1201 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2),
1202 IIC_VST2, "vst2", Dt, "\\{$Vd, $src2\\}, $Rn", "", []> {
1204 let Inst{5-4} = Rn{5-4};
1205 let DecoderMethod = "DecodeVSTInstruction";
1207 class VST2Q<bits<4> op7_4, string Dt>
1208 : NLdSt<0, 0b00, 0b0011, op7_4, (outs),
1209 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
1210 IIC_VST2x2, "vst2", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn",
1213 let Inst{5-4} = Rn{5-4};
1214 let DecoderMethod = "DecodeVSTInstruction";
1217 def VST2d8 : VST2D<0b1000, {0,0,?,?}, "8">;
1218 def VST2d16 : VST2D<0b1000, {0,1,?,?}, "16">;
1219 def VST2d32 : VST2D<0b1000, {1,0,?,?}, "32">;
1221 def VST2q8 : VST2Q<{0,0,?,?}, "8">;
1222 def VST2q16 : VST2Q<{0,1,?,?}, "16">;
1223 def VST2q32 : VST2Q<{1,0,?,?}, "32">;
1225 def VST2d8Pseudo : VSTQPseudo<IIC_VST2>;
1226 def VST2d16Pseudo : VSTQPseudo<IIC_VST2>;
1227 def VST2d32Pseudo : VSTQPseudo<IIC_VST2>;
1229 def VST2q8Pseudo : VSTQQPseudo<IIC_VST2x2>;
1230 def VST2q16Pseudo : VSTQQPseudo<IIC_VST2x2>;
1231 def VST2q32Pseudo : VSTQQPseudo<IIC_VST2x2>;
1233 // ...with address register writeback:
1234 class VST2DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1235 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1236 (ins addrmode6:$Rn, am6offset:$Rm, DPR:$Vd, DPR:$src2),
1237 IIC_VST2u, "vst2", Dt, "\\{$Vd, $src2\\}, $Rn$Rm",
1238 "$Rn.addr = $wb", []> {
1239 let Inst{5-4} = Rn{5-4};
1240 let DecoderMethod = "DecodeVSTInstruction";
1242 class VST2QWB<bits<4> op7_4, string Dt>
1243 : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
1244 (ins addrmode6:$Rn, am6offset:$Rm,
1245 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST2x2u,
1246 "vst2", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
1247 "$Rn.addr = $wb", []> {
1248 let Inst{5-4} = Rn{5-4};
1249 let DecoderMethod = "DecodeVSTInstruction";
1252 def VST2d8_UPD : VST2DWB<0b1000, {0,0,?,?}, "8">;
1253 def VST2d16_UPD : VST2DWB<0b1000, {0,1,?,?}, "16">;
1254 def VST2d32_UPD : VST2DWB<0b1000, {1,0,?,?}, "32">;
1256 def VST2q8_UPD : VST2QWB<{0,0,?,?}, "8">;
1257 def VST2q16_UPD : VST2QWB<{0,1,?,?}, "16">;
1258 def VST2q32_UPD : VST2QWB<{1,0,?,?}, "32">;
1260 def VST2d8Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
1261 def VST2d16Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
1262 def VST2d32Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
1264 def VST2q8Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
1265 def VST2q16Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
1266 def VST2q32Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
1268 // ...with double-spaced registers (for disassembly only):
1269 def VST2b8 : VST2D<0b1001, {0,0,?,?}, "8">;
1270 def VST2b16 : VST2D<0b1001, {0,1,?,?}, "16">;
1271 def VST2b32 : VST2D<0b1001, {1,0,?,?}, "32">;
1272 def VST2b8_UPD : VST2DWB<0b1001, {0,0,?,?}, "8">;
1273 def VST2b16_UPD : VST2DWB<0b1001, {0,1,?,?}, "16">;
1274 def VST2b32_UPD : VST2DWB<0b1001, {1,0,?,?}, "32">;
1276 // VST3 : Vector Store (multiple 3-element structures)
1277 class VST3D<bits<4> op11_8, bits<4> op7_4, string Dt>
1278 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1279 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3,
1280 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn", "", []> {
1282 let Inst{4} = Rn{4};
1283 let DecoderMethod = "DecodeVSTInstruction";
1286 def VST3d8 : VST3D<0b0100, {0,0,0,?}, "8">;
1287 def VST3d16 : VST3D<0b0100, {0,1,0,?}, "16">;
1288 def VST3d32 : VST3D<0b0100, {1,0,0,?}, "32">;
1290 def VST3d8Pseudo : VSTQQPseudo<IIC_VST3>;
1291 def VST3d16Pseudo : VSTQQPseudo<IIC_VST3>;
1292 def VST3d32Pseudo : VSTQQPseudo<IIC_VST3>;
1294 // ...with address register writeback:
1295 class VST3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1296 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1297 (ins addrmode6:$Rn, am6offset:$Rm,
1298 DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3u,
1299 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn$Rm",
1300 "$Rn.addr = $wb", []> {
1301 let Inst{4} = Rn{4};
1302 let DecoderMethod = "DecodeVSTInstruction";
1305 def VST3d8_UPD : VST3DWB<0b0100, {0,0,0,?}, "8">;
1306 def VST3d16_UPD : VST3DWB<0b0100, {0,1,0,?}, "16">;
1307 def VST3d32_UPD : VST3DWB<0b0100, {1,0,0,?}, "32">;
1309 def VST3d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1310 def VST3d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1311 def VST3d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1313 // ...with double-spaced registers:
1314 def VST3q8 : VST3D<0b0101, {0,0,0,?}, "8">;
1315 def VST3q16 : VST3D<0b0101, {0,1,0,?}, "16">;
1316 def VST3q32 : VST3D<0b0101, {1,0,0,?}, "32">;
1317 def VST3q8_UPD : VST3DWB<0b0101, {0,0,0,?}, "8">;
1318 def VST3q16_UPD : VST3DWB<0b0101, {0,1,0,?}, "16">;
1319 def VST3q32_UPD : VST3DWB<0b0101, {1,0,0,?}, "32">;
1321 def VST3q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1322 def VST3q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1323 def VST3q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1325 // ...alternate versions to be allocated odd register numbers:
1326 def VST3q8oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1327 def VST3q16oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1328 def VST3q32oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1330 def VST3q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1331 def VST3q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1332 def VST3q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1334 // VST4 : Vector Store (multiple 4-element structures)
1335 class VST4D<bits<4> op11_8, bits<4> op7_4, string Dt>
1336 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1337 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
1338 IIC_VST4, "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn",
1341 let Inst{5-4} = Rn{5-4};
1342 let DecoderMethod = "DecodeVSTInstruction";
1345 def VST4d8 : VST4D<0b0000, {0,0,?,?}, "8">;
1346 def VST4d16 : VST4D<0b0000, {0,1,?,?}, "16">;
1347 def VST4d32 : VST4D<0b0000, {1,0,?,?}, "32">;
1349 def VST4d8Pseudo : VSTQQPseudo<IIC_VST4>;
1350 def VST4d16Pseudo : VSTQQPseudo<IIC_VST4>;
1351 def VST4d32Pseudo : VSTQQPseudo<IIC_VST4>;
1353 // ...with address register writeback:
1354 class VST4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1355 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1356 (ins addrmode6:$Rn, am6offset:$Rm,
1357 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST4u,
1358 "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
1359 "$Rn.addr = $wb", []> {
1360 let Inst{5-4} = Rn{5-4};
1361 let DecoderMethod = "DecodeVSTInstruction";
1364 def VST4d8_UPD : VST4DWB<0b0000, {0,0,?,?}, "8">;
1365 def VST4d16_UPD : VST4DWB<0b0000, {0,1,?,?}, "16">;
1366 def VST4d32_UPD : VST4DWB<0b0000, {1,0,?,?}, "32">;
1368 def VST4d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1369 def VST4d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1370 def VST4d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1372 // ...with double-spaced registers:
1373 def VST4q8 : VST4D<0b0001, {0,0,?,?}, "8">;
1374 def VST4q16 : VST4D<0b0001, {0,1,?,?}, "16">;
1375 def VST4q32 : VST4D<0b0001, {1,0,?,?}, "32">;
1376 def VST4q8_UPD : VST4DWB<0b0001, {0,0,?,?}, "8">;
1377 def VST4q16_UPD : VST4DWB<0b0001, {0,1,?,?}, "16">;
1378 def VST4q32_UPD : VST4DWB<0b0001, {1,0,?,?}, "32">;
1380 def VST4q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1381 def VST4q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1382 def VST4q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1384 // ...alternate versions to be allocated odd register numbers:
1385 def VST4q8oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1386 def VST4q16oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1387 def VST4q32oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1389 def VST4q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1390 def VST4q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1391 def VST4q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1393 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
1395 // Classes for VST*LN pseudo-instructions with multi-register operands.
1396 // These are expanded to real instructions after register allocation.
1397 class VSTQLNPseudo<InstrItinClass itin>
1398 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
1400 class VSTQLNWBPseudo<InstrItinClass itin>
1401 : PseudoNLdSt<(outs GPR:$wb),
1402 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
1403 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1404 class VSTQQLNPseudo<InstrItinClass itin>
1405 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
1407 class VSTQQLNWBPseudo<InstrItinClass itin>
1408 : PseudoNLdSt<(outs GPR:$wb),
1409 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
1410 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1411 class VSTQQQQLNPseudo<InstrItinClass itin>
1412 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
1414 class VSTQQQQLNWBPseudo<InstrItinClass itin>
1415 : PseudoNLdSt<(outs GPR:$wb),
1416 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
1417 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1419 // VST1LN : Vector Store (single element from one lane)
1420 class VST1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1421 PatFrag StoreOp, SDNode ExtractOp>
1422 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1423 (ins addrmode6:$Rn, DPR:$Vd, nohash_imm:$lane),
1424 IIC_VST1ln, "vst1", Dt, "\\{$Vd[$lane]\\}, $Rn", "",
1425 [(StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane), addrmode6:$Rn)]> {
1428 class VST1LN32<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1429 PatFrag StoreOp, SDNode ExtractOp>
1430 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1431 (ins addrmode6oneL32:$Rn, DPR:$Vd, nohash_imm:$lane),
1432 IIC_VST1ln, "vst1", Dt, "\\{$Vd[$lane]\\}, $Rn", "",
1433 [(StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane), addrmode6oneL32:$Rn)]>{
1436 class VST1QLNPseudo<ValueType Ty, PatFrag StoreOp, SDNode ExtractOp>
1437 : VSTQLNPseudo<IIC_VST1ln> {
1438 let Pattern = [(StoreOp (ExtractOp (Ty QPR:$src), imm:$lane),
1442 def VST1LNd8 : VST1LN<0b0000, {?,?,?,0}, "8", v8i8, truncstorei8,
1444 let Inst{7-5} = lane{2-0};
1446 def VST1LNd16 : VST1LN<0b0100, {?,?,0,?}, "16", v4i16, truncstorei16,
1448 let Inst{7-6} = lane{1-0};
1449 let Inst{4} = Rn{5};
1452 def VST1LNd32 : VST1LN32<0b1000, {?,0,?,?}, "32", v2i32, store, extractelt> {
1453 let Inst{7} = lane{0};
1454 let Inst{5-4} = Rn{5-4};
1457 def VST1LNq8Pseudo : VST1QLNPseudo<v16i8, truncstorei8, NEONvgetlaneu>;
1458 def VST1LNq16Pseudo : VST1QLNPseudo<v8i16, truncstorei16, NEONvgetlaneu>;
1459 def VST1LNq32Pseudo : VST1QLNPseudo<v4i32, store, extractelt>;
1461 def : Pat<(store (extractelt (v2f32 DPR:$src), imm:$lane), addrmode6:$addr),
1462 (VST1LNd32 addrmode6:$addr, DPR:$src, imm:$lane)>;
1463 def : Pat<(store (extractelt (v4f32 QPR:$src), imm:$lane), addrmode6:$addr),
1464 (VST1LNq32Pseudo addrmode6:$addr, QPR:$src, imm:$lane)>;
1466 // ...with address register writeback:
1467 class VST1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1468 PatFrag StoreOp, SDNode ExtractOp>
1469 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1470 (ins addrmode6:$Rn, am6offset:$Rm,
1471 DPR:$Vd, nohash_imm:$lane), IIC_VST1lnu, "vst1", Dt,
1472 "\\{$Vd[$lane]\\}, $Rn$Rm",
1474 [(set GPR:$wb, (StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane),
1475 addrmode6:$Rn, am6offset:$Rm))]>;
1476 class VST1QLNWBPseudo<ValueType Ty, PatFrag StoreOp, SDNode ExtractOp>
1477 : VSTQLNWBPseudo<IIC_VST1lnu> {
1478 let Pattern = [(set GPR:$wb, (StoreOp (ExtractOp (Ty QPR:$src), imm:$lane),
1479 addrmode6:$addr, am6offset:$offset))];
1482 def VST1LNd8_UPD : VST1LNWB<0b0000, {?,?,?,0}, "8", v8i8, post_truncsti8,
1484 let Inst{7-5} = lane{2-0};
1486 def VST1LNd16_UPD : VST1LNWB<0b0100, {?,?,0,?}, "16", v4i16, post_truncsti16,
1488 let Inst{7-6} = lane{1-0};
1489 let Inst{4} = Rn{5};
1491 def VST1LNd32_UPD : VST1LNWB<0b1000, {?,0,?,?}, "32", v2i32, post_store,
1493 let Inst{7} = lane{0};
1494 let Inst{5-4} = Rn{5-4};
1497 def VST1LNq8Pseudo_UPD : VST1QLNWBPseudo<v16i8, post_truncsti8, NEONvgetlaneu>;
1498 def VST1LNq16Pseudo_UPD : VST1QLNWBPseudo<v8i16, post_truncsti16,NEONvgetlaneu>;
1499 def VST1LNq32Pseudo_UPD : VST1QLNWBPseudo<v4i32, post_store, extractelt>;
1501 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
1503 // VST2LN : Vector Store (single 2-element structure from one lane)
1504 class VST2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1505 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1506 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, nohash_imm:$lane),
1507 IIC_VST2ln, "vst2", Dt, "\\{$Vd[$lane], $src2[$lane]\\}, $Rn",
1510 let Inst{4} = Rn{4};
1513 def VST2LNd8 : VST2LN<0b0001, {?,?,?,?}, "8"> {
1514 let Inst{7-5} = lane{2-0};
1516 def VST2LNd16 : VST2LN<0b0101, {?,?,0,?}, "16"> {
1517 let Inst{7-6} = lane{1-0};
1519 def VST2LNd32 : VST2LN<0b1001, {?,0,0,?}, "32"> {
1520 let Inst{7} = lane{0};
1523 def VST2LNd8Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1524 def VST2LNd16Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1525 def VST2LNd32Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1527 // ...with double-spaced registers:
1528 def VST2LNq16 : VST2LN<0b0101, {?,?,1,?}, "16"> {
1529 let Inst{7-6} = lane{1-0};
1530 let Inst{4} = Rn{4};
1532 def VST2LNq32 : VST2LN<0b1001, {?,1,0,?}, "32"> {
1533 let Inst{7} = lane{0};
1534 let Inst{4} = Rn{4};
1537 def VST2LNq16Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
1538 def VST2LNq32Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
1540 // ...with address register writeback:
1541 class VST2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1542 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1543 (ins addrmode6:$addr, am6offset:$offset,
1544 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VST2lnu, "vst2", Dt,
1545 "\\{$src1[$lane], $src2[$lane]\\}, $addr$offset",
1546 "$addr.addr = $wb", []> {
1547 let Inst{4} = Rn{4};
1550 def VST2LNd8_UPD : VST2LNWB<0b0001, {?,?,?,?}, "8"> {
1551 let Inst{7-5} = lane{2-0};
1553 def VST2LNd16_UPD : VST2LNWB<0b0101, {?,?,0,?}, "16"> {
1554 let Inst{7-6} = lane{1-0};
1556 def VST2LNd32_UPD : VST2LNWB<0b1001, {?,0,0,?}, "32"> {
1557 let Inst{7} = lane{0};
1560 def VST2LNd8Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1561 def VST2LNd16Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1562 def VST2LNd32Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1564 def VST2LNq16_UPD : VST2LNWB<0b0101, {?,?,1,?}, "16"> {
1565 let Inst{7-6} = lane{1-0};
1567 def VST2LNq32_UPD : VST2LNWB<0b1001, {?,1,0,?}, "32"> {
1568 let Inst{7} = lane{0};
1571 def VST2LNq16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
1572 def VST2LNq32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
1574 // VST3LN : Vector Store (single 3-element structure from one lane)
1575 class VST3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1576 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1577 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3,
1578 nohash_imm:$lane), IIC_VST3ln, "vst3", Dt,
1579 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn", "", []> {
1583 def VST3LNd8 : VST3LN<0b0010, {?,?,?,0}, "8"> {
1584 let Inst{7-5} = lane{2-0};
1586 def VST3LNd16 : VST3LN<0b0110, {?,?,0,0}, "16"> {
1587 let Inst{7-6} = lane{1-0};
1589 def VST3LNd32 : VST3LN<0b1010, {?,0,0,0}, "32"> {
1590 let Inst{7} = lane{0};
1593 def VST3LNd8Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1594 def VST3LNd16Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1595 def VST3LNd32Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1597 // ...with double-spaced registers:
1598 def VST3LNq16 : VST3LN<0b0110, {?,?,1,0}, "16"> {
1599 let Inst{7-6} = lane{1-0};
1601 def VST3LNq32 : VST3LN<0b1010, {?,1,0,0}, "32"> {
1602 let Inst{7} = lane{0};
1605 def VST3LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
1606 def VST3LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
1608 // ...with address register writeback:
1609 class VST3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1610 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1611 (ins addrmode6:$Rn, am6offset:$Rm,
1612 DPR:$Vd, DPR:$src2, DPR:$src3, nohash_imm:$lane),
1613 IIC_VST3lnu, "vst3", Dt,
1614 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn$Rm",
1615 "$Rn.addr = $wb", []>;
1617 def VST3LNd8_UPD : VST3LNWB<0b0010, {?,?,?,0}, "8"> {
1618 let Inst{7-5} = lane{2-0};
1620 def VST3LNd16_UPD : VST3LNWB<0b0110, {?,?,0,0}, "16"> {
1621 let Inst{7-6} = lane{1-0};
1623 def VST3LNd32_UPD : VST3LNWB<0b1010, {?,0,0,0}, "32"> {
1624 let Inst{7} = lane{0};
1627 def VST3LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1628 def VST3LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1629 def VST3LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1631 def VST3LNq16_UPD : VST3LNWB<0b0110, {?,?,1,0}, "16"> {
1632 let Inst{7-6} = lane{1-0};
1634 def VST3LNq32_UPD : VST3LNWB<0b1010, {?,1,0,0}, "32"> {
1635 let Inst{7} = lane{0};
1638 def VST3LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
1639 def VST3LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
1641 // VST4LN : Vector Store (single 4-element structure from one lane)
1642 class VST4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1643 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1644 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4,
1645 nohash_imm:$lane), IIC_VST4ln, "vst4", Dt,
1646 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn",
1649 let Inst{4} = Rn{4};
1652 def VST4LNd8 : VST4LN<0b0011, {?,?,?,?}, "8"> {
1653 let Inst{7-5} = lane{2-0};
1655 def VST4LNd16 : VST4LN<0b0111, {?,?,0,?}, "16"> {
1656 let Inst{7-6} = lane{1-0};
1658 def VST4LNd32 : VST4LN<0b1011, {?,0,?,?}, "32"> {
1659 let Inst{7} = lane{0};
1660 let Inst{5} = Rn{5};
1663 def VST4LNd8Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1664 def VST4LNd16Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1665 def VST4LNd32Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1667 // ...with double-spaced registers:
1668 def VST4LNq16 : VST4LN<0b0111, {?,?,1,?}, "16"> {
1669 let Inst{7-6} = lane{1-0};
1671 def VST4LNq32 : VST4LN<0b1011, {?,1,?,?}, "32"> {
1672 let Inst{7} = lane{0};
1673 let Inst{5} = Rn{5};
1676 def VST4LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
1677 def VST4LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
1679 // ...with address register writeback:
1680 class VST4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1681 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1682 (ins addrmode6:$Rn, am6offset:$Rm,
1683 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
1684 IIC_VST4lnu, "vst4", Dt,
1685 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn$Rm",
1686 "$Rn.addr = $wb", []> {
1687 let Inst{4} = Rn{4};
1690 def VST4LNd8_UPD : VST4LNWB<0b0011, {?,?,?,?}, "8"> {
1691 let Inst{7-5} = lane{2-0};
1693 def VST4LNd16_UPD : VST4LNWB<0b0111, {?,?,0,?}, "16"> {
1694 let Inst{7-6} = lane{1-0};
1696 def VST4LNd32_UPD : VST4LNWB<0b1011, {?,0,?,?}, "32"> {
1697 let Inst{7} = lane{0};
1698 let Inst{5} = Rn{5};
1701 def VST4LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
1702 def VST4LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
1703 def VST4LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
1705 def VST4LNq16_UPD : VST4LNWB<0b0111, {?,?,1,?}, "16"> {
1706 let Inst{7-6} = lane{1-0};
1708 def VST4LNq32_UPD : VST4LNWB<0b1011, {?,1,?,?}, "32"> {
1709 let Inst{7} = lane{0};
1710 let Inst{5} = Rn{5};
1713 def VST4LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
1714 def VST4LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
1716 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
1719 //===----------------------------------------------------------------------===//
1720 // NEON pattern fragments
1721 //===----------------------------------------------------------------------===//
1723 // Extract D sub-registers of Q registers.
1724 def DSubReg_i8_reg : SDNodeXForm<imm, [{
1725 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1726 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/8, MVT::i32);
1728 def DSubReg_i16_reg : SDNodeXForm<imm, [{
1729 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1730 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/4, MVT::i32);
1732 def DSubReg_i32_reg : SDNodeXForm<imm, [{
1733 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1734 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/2, MVT::i32);
1736 def DSubReg_f64_reg : SDNodeXForm<imm, [{
1737 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1738 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue(), MVT::i32);
1741 // Extract S sub-registers of Q/D registers.
1742 def SSubReg_f32_reg : SDNodeXForm<imm, [{
1743 assert(ARM::ssub_3 == ARM::ssub_0+3 && "Unexpected subreg numbering");
1744 return CurDAG->getTargetConstant(ARM::ssub_0 + N->getZExtValue(), MVT::i32);
1747 // Translate lane numbers from Q registers to D subregs.
1748 def SubReg_i8_lane : SDNodeXForm<imm, [{
1749 return CurDAG->getTargetConstant(N->getZExtValue() & 7, MVT::i32);
1751 def SubReg_i16_lane : SDNodeXForm<imm, [{
1752 return CurDAG->getTargetConstant(N->getZExtValue() & 3, MVT::i32);
1754 def SubReg_i32_lane : SDNodeXForm<imm, [{
1755 return CurDAG->getTargetConstant(N->getZExtValue() & 1, MVT::i32);
1758 //===----------------------------------------------------------------------===//
1759 // Instruction Classes
1760 //===----------------------------------------------------------------------===//
1762 // Basic 2-register operations: double- and quad-register.
1763 class N2VD<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1764 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
1765 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
1766 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
1767 (ins DPR:$Vm), IIC_VUNAD, OpcodeStr, Dt,"$Vd, $Vm", "",
1768 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vm))))]>;
1769 class N2VQ<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1770 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
1771 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
1772 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
1773 (ins QPR:$Vm), IIC_VUNAQ, OpcodeStr, Dt,"$Vd, $Vm", "",
1774 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vm))))]>;
1776 // Basic 2-register intrinsics, both double- and quad-register.
1777 class N2VDInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1778 bits<2> op17_16, bits<5> op11_7, bit op4,
1779 InstrItinClass itin, string OpcodeStr, string Dt,
1780 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1781 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
1782 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
1783 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm))))]>;
1784 class N2VQInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1785 bits<2> op17_16, bits<5> op11_7, bit op4,
1786 InstrItinClass itin, string OpcodeStr, string Dt,
1787 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1788 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
1789 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
1790 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm))))]>;
1792 // Narrow 2-register operations.
1793 class N2VN<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1794 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1795 InstrItinClass itin, string OpcodeStr, string Dt,
1796 ValueType TyD, ValueType TyQ, SDNode OpNode>
1797 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$Vd),
1798 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
1799 [(set DPR:$Vd, (TyD (OpNode (TyQ QPR:$Vm))))]>;
1801 // Narrow 2-register intrinsics.
1802 class N2VNInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1803 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1804 InstrItinClass itin, string OpcodeStr, string Dt,
1805 ValueType TyD, ValueType TyQ, Intrinsic IntOp>
1806 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$Vd),
1807 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
1808 [(set DPR:$Vd, (TyD (IntOp (TyQ QPR:$Vm))))]>;
1810 // Long 2-register operations (currently only used for VMOVL).
1811 class N2VL<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1812 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1813 InstrItinClass itin, string OpcodeStr, string Dt,
1814 ValueType TyQ, ValueType TyD, SDNode OpNode>
1815 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$Vd),
1816 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
1817 [(set QPR:$Vd, (TyQ (OpNode (TyD DPR:$Vm))))]>;
1819 // Long 2-register intrinsics.
1820 class N2VLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1821 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1822 InstrItinClass itin, string OpcodeStr, string Dt,
1823 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
1824 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$Vd),
1825 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
1826 [(set QPR:$Vd, (TyQ (IntOp (TyD DPR:$Vm))))]>;
1828 // 2-register shuffles (VTRN/VZIP/VUZP), both double- and quad-register.
1829 class N2VDShuffle<bits<2> op19_18, bits<5> op11_7, string OpcodeStr, string Dt>
1830 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 0, 0, (outs DPR:$Vd, DPR:$Vm),
1831 (ins DPR:$src1, DPR:$src2), IIC_VPERMD,
1832 OpcodeStr, Dt, "$Vd, $Vm",
1833 "$src1 = $Vd, $src2 = $Vm", []>;
1834 class N2VQShuffle<bits<2> op19_18, bits<5> op11_7,
1835 InstrItinClass itin, string OpcodeStr, string Dt>
1836 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 1, 0, (outs QPR:$Vd, QPR:$Vm),
1837 (ins QPR:$src1, QPR:$src2), itin, OpcodeStr, Dt, "$Vd, $Vm",
1838 "$src1 = $Vd, $src2 = $Vm", []>;
1840 // Basic 3-register operations: double- and quad-register.
1841 class N3VD<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1842 InstrItinClass itin, string OpcodeStr, string Dt,
1843 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
1844 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1845 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1846 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
1847 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
1848 let isCommutable = Commutable;
1850 // Same as N3VD but no data type.
1851 class N3VDX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1852 InstrItinClass itin, string OpcodeStr,
1853 ValueType ResTy, ValueType OpTy,
1854 SDNode OpNode, bit Commutable>
1855 : N3VX<op24, op23, op21_20, op11_8, 0, op4,
1856 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1857 OpcodeStr, "$Vd, $Vn, $Vm", "",
1858 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>{
1859 let isCommutable = Commutable;
1862 class N3VDSL<bits<2> op21_20, bits<4> op11_8,
1863 InstrItinClass itin, string OpcodeStr, string Dt,
1864 ValueType Ty, SDNode ShOp>
1865 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
1866 (outs DPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, nohash_imm:$lane),
1867 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "",
1869 (Ty (ShOp (Ty DPR:$Vn),
1870 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),imm:$lane)))))]> {
1871 let isCommutable = 0;
1873 class N3VDSL16<bits<2> op21_20, bits<4> op11_8,
1874 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1875 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
1876 (outs DPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, nohash_imm:$lane),
1877 NVMulSLFrm, IIC_VMULi16D, OpcodeStr, Dt,"$Vd, $Vn, $Vm[$lane]","",
1879 (Ty (ShOp (Ty DPR:$Vn),
1880 (Ty (NEONvduplane (Ty DPR_8:$Vm), imm:$lane)))))]> {
1881 let isCommutable = 0;
1884 class N3VQ<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1885 InstrItinClass itin, string OpcodeStr, string Dt,
1886 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
1887 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1888 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
1889 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
1890 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
1891 let isCommutable = Commutable;
1893 class N3VQX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1894 InstrItinClass itin, string OpcodeStr,
1895 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
1896 : N3VX<op24, op23, op21_20, op11_8, 1, op4,
1897 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
1898 OpcodeStr, "$Vd, $Vn, $Vm", "",
1899 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]>{
1900 let isCommutable = Commutable;
1902 class N3VQSL<bits<2> op21_20, bits<4> op11_8,
1903 InstrItinClass itin, string OpcodeStr, string Dt,
1904 ValueType ResTy, ValueType OpTy, SDNode ShOp>
1905 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
1906 (outs QPR:$Vd), (ins QPR:$Vn, DPR_VFP2:$Vm, nohash_imm:$lane),
1907 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "",
1908 [(set (ResTy QPR:$Vd),
1909 (ResTy (ShOp (ResTy QPR:$Vn),
1910 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
1912 let isCommutable = 0;
1914 class N3VQSL16<bits<2> op21_20, bits<4> op11_8, string OpcodeStr, string Dt,
1915 ValueType ResTy, ValueType OpTy, SDNode ShOp>
1916 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
1917 (outs QPR:$Vd), (ins QPR:$Vn, DPR_8:$Vm, nohash_imm:$lane),
1918 NVMulSLFrm, IIC_VMULi16Q, OpcodeStr, Dt,"$Vd, $Vn, $Vm[$lane]","",
1919 [(set (ResTy QPR:$Vd),
1920 (ResTy (ShOp (ResTy QPR:$Vn),
1921 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
1923 let isCommutable = 0;
1926 // Basic 3-register intrinsics, both double- and quad-register.
1927 class N3VDInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1928 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1929 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
1930 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1931 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), f, itin,
1932 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
1933 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
1934 let isCommutable = Commutable;
1936 class N3VDIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1937 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
1938 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
1939 (outs DPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, nohash_imm:$lane),
1940 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "",
1942 (Ty (IntOp (Ty DPR:$Vn),
1943 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),
1945 let isCommutable = 0;
1947 class N3VDIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1948 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
1949 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
1950 (outs DPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, nohash_imm:$lane),
1951 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "",
1953 (Ty (IntOp (Ty DPR:$Vn),
1954 (Ty (NEONvduplane (Ty DPR_8:$Vm), imm:$lane)))))]> {
1955 let isCommutable = 0;
1957 class N3VDIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1958 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1959 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1960 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1961 (outs DPR:$Vd), (ins DPR:$Vm, DPR:$Vn), f, itin,
1962 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
1963 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (OpTy DPR:$Vn))))]> {
1964 let isCommutable = 0;
1967 class N3VQInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1968 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1969 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
1970 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1971 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), f, itin,
1972 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
1973 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
1974 let isCommutable = Commutable;
1976 class N3VQIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1977 string OpcodeStr, string Dt,
1978 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1979 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
1980 (outs QPR:$Vd), (ins QPR:$Vn, DPR_VFP2:$Vm, nohash_imm:$lane),
1981 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "",
1982 [(set (ResTy QPR:$Vd),
1983 (ResTy (IntOp (ResTy QPR:$Vn),
1984 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
1986 let isCommutable = 0;
1988 class N3VQIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1989 string OpcodeStr, string Dt,
1990 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1991 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
1992 (outs QPR:$Vd), (ins QPR:$Vn, DPR_8:$Vm, nohash_imm:$lane),
1993 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "",
1994 [(set (ResTy QPR:$Vd),
1995 (ResTy (IntOp (ResTy QPR:$Vn),
1996 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
1998 let isCommutable = 0;
2000 class N3VQIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2001 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2002 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2003 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2004 (outs QPR:$Vd), (ins QPR:$Vm, QPR:$Vn), f, itin,
2005 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
2006 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (OpTy QPR:$Vn))))]> {
2007 let isCommutable = 0;
2010 // Multiply-Add/Sub operations: double- and quad-register.
2011 class N3VDMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2012 InstrItinClass itin, string OpcodeStr, string Dt,
2013 ValueType Ty, SDPatternOperator MulOp, SDPatternOperator OpNode>
2014 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2015 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2016 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2017 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
2018 (Ty (MulOp DPR:$Vn, DPR:$Vm)))))]>;
2020 class N3VDMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2021 string OpcodeStr, string Dt,
2022 ValueType Ty, SDPatternOperator MulOp, SDPatternOperator ShOp>
2023 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2025 (ins DPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, nohash_imm:$lane),
2027 OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "$src1 = $Vd",
2029 (Ty (ShOp (Ty DPR:$src1),
2031 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),
2033 class N3VDMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2034 string OpcodeStr, string Dt,
2035 ValueType Ty, SDNode MulOp, SDNode ShOp>
2036 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2038 (ins DPR:$src1, DPR:$Vn, DPR_8:$Vm, nohash_imm:$lane),
2040 OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "$src1 = $Vd",
2042 (Ty (ShOp (Ty DPR:$src1),
2044 (Ty (NEONvduplane (Ty DPR_8:$Vm),
2047 class N3VQMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2048 InstrItinClass itin, string OpcodeStr, string Dt, ValueType Ty,
2049 SDPatternOperator MulOp, SDPatternOperator OpNode>
2050 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2051 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2052 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2053 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
2054 (Ty (MulOp QPR:$Vn, QPR:$Vm)))))]>;
2055 class N3VQMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2056 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2057 SDPatternOperator MulOp, SDPatternOperator ShOp>
2058 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2060 (ins QPR:$src1, QPR:$Vn, DPR_VFP2:$Vm, nohash_imm:$lane),
2062 OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "$src1 = $Vd",
2063 [(set (ResTy QPR:$Vd),
2064 (ResTy (ShOp (ResTy QPR:$src1),
2065 (ResTy (MulOp QPR:$Vn,
2066 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2068 class N3VQMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2069 string OpcodeStr, string Dt,
2070 ValueType ResTy, ValueType OpTy,
2071 SDNode MulOp, SDNode ShOp>
2072 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2074 (ins QPR:$src1, QPR:$Vn, DPR_8:$Vm, nohash_imm:$lane),
2076 OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "$src1 = $Vd",
2077 [(set (ResTy QPR:$Vd),
2078 (ResTy (ShOp (ResTy QPR:$src1),
2079 (ResTy (MulOp QPR:$Vn,
2080 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2083 // Neon Intrinsic-Op instructions (VABA): double- and quad-register.
2084 class N3VDIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2085 InstrItinClass itin, string OpcodeStr, string Dt,
2086 ValueType Ty, Intrinsic IntOp, SDNode OpNode>
2087 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2088 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2089 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2090 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
2091 (Ty (IntOp (Ty DPR:$Vn), (Ty DPR:$Vm))))))]>;
2092 class N3VQIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2093 InstrItinClass itin, string OpcodeStr, string Dt,
2094 ValueType Ty, Intrinsic IntOp, SDNode OpNode>
2095 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2096 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2097 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2098 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
2099 (Ty (IntOp (Ty QPR:$Vn), (Ty QPR:$Vm))))))]>;
2101 // Neon 3-argument intrinsics, both double- and quad-register.
2102 // The destination register is also used as the first source operand register.
2103 class N3VDInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2104 InstrItinClass itin, string OpcodeStr, string Dt,
2105 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2106 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2107 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2108 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2109 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$src1),
2110 (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>;
2111 class N3VQInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2112 InstrItinClass itin, string OpcodeStr, string Dt,
2113 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2114 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2115 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2116 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2117 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$src1),
2118 (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]>;
2120 // Long Multiply-Add/Sub operations.
2121 class N3VLMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2122 InstrItinClass itin, string OpcodeStr, string Dt,
2123 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2124 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2125 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2126 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2127 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
2128 (TyQ (MulOp (TyD DPR:$Vn),
2129 (TyD DPR:$Vm)))))]>;
2130 class N3VLMulOpSL<bit op24, bits<2> op21_20, bits<4> op11_8,
2131 InstrItinClass itin, string OpcodeStr, string Dt,
2132 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2133 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$Vd),
2134 (ins QPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, nohash_imm:$lane),
2136 OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "$src1 = $Vd",
2138 (OpNode (TyQ QPR:$src1),
2139 (TyQ (MulOp (TyD DPR:$Vn),
2140 (TyD (NEONvduplane (TyD DPR_VFP2:$Vm),
2142 class N3VLMulOpSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2143 InstrItinClass itin, string OpcodeStr, string Dt,
2144 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2145 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$Vd),
2146 (ins QPR:$src1, DPR:$Vn, DPR_8:$Vm, nohash_imm:$lane),
2148 OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "$src1 = $Vd",
2150 (OpNode (TyQ QPR:$src1),
2151 (TyQ (MulOp (TyD DPR:$Vn),
2152 (TyD (NEONvduplane (TyD DPR_8:$Vm),
2155 // Long Intrinsic-Op vector operations with explicit extend (VABAL).
2156 class N3VLIntExtOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2157 InstrItinClass itin, string OpcodeStr, string Dt,
2158 ValueType TyQ, ValueType TyD, Intrinsic IntOp, SDNode ExtOp,
2160 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2161 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2162 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2163 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
2164 (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
2165 (TyD DPR:$Vm)))))))]>;
2167 // Neon Long 3-argument intrinsic. The destination register is
2168 // a quad-register and is also used as the first source operand register.
2169 class N3VLInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2170 InstrItinClass itin, string OpcodeStr, string Dt,
2171 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
2172 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2173 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2174 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2176 (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$Vn), (TyD DPR:$Vm))))]>;
2177 class N3VLInt3SL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2178 string OpcodeStr, string Dt,
2179 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2180 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2182 (ins QPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, nohash_imm:$lane),
2184 OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "$src1 = $Vd",
2185 [(set (ResTy QPR:$Vd),
2186 (ResTy (IntOp (ResTy QPR:$src1),
2188 (OpTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2190 class N3VLInt3SL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2191 InstrItinClass itin, string OpcodeStr, string Dt,
2192 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2193 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2195 (ins QPR:$src1, DPR:$Vn, DPR_8:$Vm, nohash_imm:$lane),
2197 OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "$src1 = $Vd",
2198 [(set (ResTy QPR:$Vd),
2199 (ResTy (IntOp (ResTy QPR:$src1),
2201 (OpTy (NEONvduplane (OpTy DPR_8:$Vm),
2204 // Narrowing 3-register intrinsics.
2205 class N3VNInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2206 string OpcodeStr, string Dt, ValueType TyD, ValueType TyQ,
2207 Intrinsic IntOp, bit Commutable>
2208 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2209 (outs DPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINi4D,
2210 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2211 [(set DPR:$Vd, (TyD (IntOp (TyQ QPR:$Vn), (TyQ QPR:$Vm))))]> {
2212 let isCommutable = Commutable;
2215 // Long 3-register operations.
2216 class N3VL<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2217 InstrItinClass itin, string OpcodeStr, string Dt,
2218 ValueType TyQ, ValueType TyD, SDNode OpNode, bit Commutable>
2219 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2220 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2221 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2222 [(set QPR:$Vd, (TyQ (OpNode (TyD DPR:$Vn), (TyD DPR:$Vm))))]> {
2223 let isCommutable = Commutable;
2225 class N3VLSL<bit op24, bits<2> op21_20, bits<4> op11_8,
2226 InstrItinClass itin, string OpcodeStr, string Dt,
2227 ValueType TyQ, ValueType TyD, SDNode OpNode>
2228 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2229 (outs QPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, nohash_imm:$lane),
2230 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "",
2232 (TyQ (OpNode (TyD DPR:$Vn),
2233 (TyD (NEONvduplane (TyD DPR_VFP2:$Vm),imm:$lane)))))]>;
2234 class N3VLSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2235 InstrItinClass itin, string OpcodeStr, string Dt,
2236 ValueType TyQ, ValueType TyD, SDNode OpNode>
2237 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2238 (outs QPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, nohash_imm:$lane),
2239 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "",
2241 (TyQ (OpNode (TyD DPR:$Vn),
2242 (TyD (NEONvduplane (TyD DPR_8:$Vm), imm:$lane)))))]>;
2244 // Long 3-register operations with explicitly extended operands.
2245 class N3VLExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2246 InstrItinClass itin, string OpcodeStr, string Dt,
2247 ValueType TyQ, ValueType TyD, SDNode OpNode, SDNode ExtOp,
2249 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2250 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2251 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2252 [(set QPR:$Vd, (OpNode (TyQ (ExtOp (TyD DPR:$Vn))),
2253 (TyQ (ExtOp (TyD DPR:$Vm)))))]> {
2254 let isCommutable = Commutable;
2257 // Long 3-register intrinsics with explicit extend (VABDL).
2258 class N3VLIntExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2259 InstrItinClass itin, string OpcodeStr, string Dt,
2260 ValueType TyQ, ValueType TyD, Intrinsic IntOp, SDNode ExtOp,
2262 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2263 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2264 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2265 [(set QPR:$Vd, (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
2266 (TyD DPR:$Vm))))))]> {
2267 let isCommutable = Commutable;
2270 // Long 3-register intrinsics.
2271 class N3VLInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2272 InstrItinClass itin, string OpcodeStr, string Dt,
2273 ValueType TyQ, ValueType TyD, Intrinsic IntOp, bit Commutable>
2274 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2275 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2276 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2277 [(set QPR:$Vd, (TyQ (IntOp (TyD DPR:$Vn), (TyD DPR:$Vm))))]> {
2278 let isCommutable = Commutable;
2280 class N3VLIntSL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2281 string OpcodeStr, string Dt,
2282 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2283 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2284 (outs QPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, nohash_imm:$lane),
2285 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "",
2286 [(set (ResTy QPR:$Vd),
2287 (ResTy (IntOp (OpTy DPR:$Vn),
2288 (OpTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2290 class N3VLIntSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2291 InstrItinClass itin, string OpcodeStr, string Dt,
2292 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2293 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2294 (outs QPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, nohash_imm:$lane),
2295 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "",
2296 [(set (ResTy QPR:$Vd),
2297 (ResTy (IntOp (OpTy DPR:$Vn),
2298 (OpTy (NEONvduplane (OpTy DPR_8:$Vm),
2301 // Wide 3-register operations.
2302 class N3VW<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2303 string OpcodeStr, string Dt, ValueType TyQ, ValueType TyD,
2304 SDNode OpNode, SDNode ExtOp, bit Commutable>
2305 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2306 (outs QPR:$Vd), (ins QPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VSUBiD,
2307 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2308 [(set QPR:$Vd, (OpNode (TyQ QPR:$Vn),
2309 (TyQ (ExtOp (TyD DPR:$Vm)))))]> {
2310 let isCommutable = Commutable;
2313 // Pairwise long 2-register intrinsics, both double- and quad-register.
2314 class N2VDPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2315 bits<2> op17_16, bits<5> op11_7, bit op4,
2316 string OpcodeStr, string Dt,
2317 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2318 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
2319 (ins DPR:$Vm), IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
2320 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm))))]>;
2321 class N2VQPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2322 bits<2> op17_16, bits<5> op11_7, bit op4,
2323 string OpcodeStr, string Dt,
2324 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2325 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
2326 (ins QPR:$Vm), IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
2327 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm))))]>;
2329 // Pairwise long 2-register accumulate intrinsics,
2330 // both double- and quad-register.
2331 // The destination register is also used as the first source operand register.
2332 class N2VDPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2333 bits<2> op17_16, bits<5> op11_7, bit op4,
2334 string OpcodeStr, string Dt,
2335 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2336 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
2337 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vm), IIC_VPALiD,
2338 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
2339 [(set DPR:$Vd, (ResTy (IntOp (ResTy DPR:$src1), (OpTy DPR:$Vm))))]>;
2340 class N2VQPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2341 bits<2> op17_16, bits<5> op11_7, bit op4,
2342 string OpcodeStr, string Dt,
2343 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2344 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4,
2345 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vm), IIC_VPALiQ,
2346 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
2347 [(set QPR:$Vd, (ResTy (IntOp (ResTy QPR:$src1), (OpTy QPR:$Vm))))]>;
2349 // Shift by immediate,
2350 // both double- and quad-register.
2351 class N2VDSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2352 Format f, InstrItinClass itin, Operand ImmTy,
2353 string OpcodeStr, string Dt, ValueType Ty, SDNode OpNode>
2354 : N2VImm<op24, op23, op11_8, op7, 0, op4,
2355 (outs DPR:$Vd), (ins DPR:$Vm, ImmTy:$SIMM), f, itin,
2356 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2357 [(set DPR:$Vd, (Ty (OpNode (Ty DPR:$Vm), (i32 imm:$SIMM))))]>;
2358 class N2VQSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2359 Format f, InstrItinClass itin, Operand ImmTy,
2360 string OpcodeStr, string Dt, ValueType Ty, SDNode OpNode>
2361 : N2VImm<op24, op23, op11_8, op7, 1, op4,
2362 (outs QPR:$Vd), (ins QPR:$Vm, ImmTy:$SIMM), f, itin,
2363 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2364 [(set QPR:$Vd, (Ty (OpNode (Ty QPR:$Vm), (i32 imm:$SIMM))))]>;
2366 // Long shift by immediate.
2367 class N2VLSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
2368 string OpcodeStr, string Dt,
2369 ValueType ResTy, ValueType OpTy, SDNode OpNode>
2370 : N2VImm<op24, op23, op11_8, op7, op6, op4,
2371 (outs QPR:$Vd), (ins DPR:$Vm, i32imm:$SIMM), N2RegVShLFrm,
2372 IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2373 [(set QPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vm),
2374 (i32 imm:$SIMM))))]>;
2376 // Narrow shift by immediate.
2377 class N2VNSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
2378 InstrItinClass itin, string OpcodeStr, string Dt,
2379 ValueType ResTy, ValueType OpTy, Operand ImmTy, SDNode OpNode>
2380 : N2VImm<op24, op23, op11_8, op7, op6, op4,
2381 (outs DPR:$Vd), (ins QPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, itin,
2382 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2383 [(set DPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vm),
2384 (i32 imm:$SIMM))))]>;
2386 // Shift right by immediate and accumulate,
2387 // both double- and quad-register.
2388 class N2VDShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2389 Operand ImmTy, string OpcodeStr, string Dt,
2390 ValueType Ty, SDNode ShOp>
2391 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
2392 (ins DPR:$src1, DPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, IIC_VPALiD,
2393 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2394 [(set DPR:$Vd, (Ty (add DPR:$src1,
2395 (Ty (ShOp DPR:$Vm, (i32 imm:$SIMM))))))]>;
2396 class N2VQShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2397 Operand ImmTy, string OpcodeStr, string Dt,
2398 ValueType Ty, SDNode ShOp>
2399 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
2400 (ins QPR:$src1, QPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, IIC_VPALiD,
2401 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2402 [(set QPR:$Vd, (Ty (add QPR:$src1,
2403 (Ty (ShOp QPR:$Vm, (i32 imm:$SIMM))))))]>;
2405 // Shift by immediate and insert,
2406 // both double- and quad-register.
2407 class N2VDShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2408 Operand ImmTy, Format f, string OpcodeStr, string Dt,
2409 ValueType Ty,SDNode ShOp>
2410 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
2411 (ins DPR:$src1, DPR:$Vm, ImmTy:$SIMM), f, IIC_VSHLiD,
2412 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2413 [(set DPR:$Vd, (Ty (ShOp DPR:$src1, DPR:$Vm, (i32 imm:$SIMM))))]>;
2414 class N2VQShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2415 Operand ImmTy, Format f, string OpcodeStr, string Dt,
2416 ValueType Ty,SDNode ShOp>
2417 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
2418 (ins QPR:$src1, QPR:$Vm, ImmTy:$SIMM), f, IIC_VSHLiQ,
2419 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2420 [(set QPR:$Vd, (Ty (ShOp QPR:$src1, QPR:$Vm, (i32 imm:$SIMM))))]>;
2422 // Convert, with fractional bits immediate,
2423 // both double- and quad-register.
2424 class N2VCvtD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2425 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2427 : N2VImm<op24, op23, op11_8, op7, 0, op4,
2428 (outs DPR:$Vd), (ins DPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
2429 IIC_VUNAD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2430 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (i32 imm:$SIMM))))]>;
2431 class N2VCvtQ<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2432 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2434 : N2VImm<op24, op23, op11_8, op7, 1, op4,
2435 (outs QPR:$Vd), (ins QPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
2436 IIC_VUNAQ, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2437 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (i32 imm:$SIMM))))]>;
2439 //===----------------------------------------------------------------------===//
2441 //===----------------------------------------------------------------------===//
2443 // Abbreviations used in multiclass suffixes:
2444 // Q = quarter int (8 bit) elements
2445 // H = half int (16 bit) elements
2446 // S = single int (32 bit) elements
2447 // D = double int (64 bit) elements
2449 // Neon 2-register vector operations and intrinsics.
2451 // Neon 2-register comparisons.
2452 // source operand element sizes of 8, 16 and 32 bits:
2453 multiclass N2V_QHS_cmp<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2454 bits<5> op11_7, bit op4, string opc, string Dt,
2455 string asm, SDNode OpNode> {
2456 // 64-bit vector types.
2457 def v8i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 0, op4,
2458 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2459 opc, !strconcat(Dt, "8"), asm, "",
2460 [(set DPR:$Vd, (v8i8 (OpNode (v8i8 DPR:$Vm))))]>;
2461 def v4i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 0, op4,
2462 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2463 opc, !strconcat(Dt, "16"), asm, "",
2464 [(set DPR:$Vd, (v4i16 (OpNode (v4i16 DPR:$Vm))))]>;
2465 def v2i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
2466 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2467 opc, !strconcat(Dt, "32"), asm, "",
2468 [(set DPR:$Vd, (v2i32 (OpNode (v2i32 DPR:$Vm))))]>;
2469 def v2f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
2470 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2471 opc, "f32", asm, "",
2472 [(set DPR:$Vd, (v2i32 (OpNode (v2f32 DPR:$Vm))))]> {
2473 let Inst{10} = 1; // overwrite F = 1
2476 // 128-bit vector types.
2477 def v16i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 1, op4,
2478 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2479 opc, !strconcat(Dt, "8"), asm, "",
2480 [(set QPR:$Vd, (v16i8 (OpNode (v16i8 QPR:$Vm))))]>;
2481 def v8i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 1, op4,
2482 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2483 opc, !strconcat(Dt, "16"), asm, "",
2484 [(set QPR:$Vd, (v8i16 (OpNode (v8i16 QPR:$Vm))))]>;
2485 def v4i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
2486 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2487 opc, !strconcat(Dt, "32"), asm, "",
2488 [(set QPR:$Vd, (v4i32 (OpNode (v4i32 QPR:$Vm))))]>;
2489 def v4f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
2490 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2491 opc, "f32", asm, "",
2492 [(set QPR:$Vd, (v4i32 (OpNode (v4f32 QPR:$Vm))))]> {
2493 let Inst{10} = 1; // overwrite F = 1
2498 // Neon 2-register vector intrinsics,
2499 // element sizes of 8, 16 and 32 bits:
2500 multiclass N2VInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2501 bits<5> op11_7, bit op4,
2502 InstrItinClass itinD, InstrItinClass itinQ,
2503 string OpcodeStr, string Dt, Intrinsic IntOp> {
2504 // 64-bit vector types.
2505 def v8i8 : N2VDInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2506 itinD, OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
2507 def v4i16 : N2VDInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2508 itinD, OpcodeStr, !strconcat(Dt, "16"),v4i16,v4i16,IntOp>;
2509 def v2i32 : N2VDInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2510 itinD, OpcodeStr, !strconcat(Dt, "32"),v2i32,v2i32,IntOp>;
2512 // 128-bit vector types.
2513 def v16i8 : N2VQInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2514 itinQ, OpcodeStr, !strconcat(Dt, "8"), v16i8,v16i8,IntOp>;
2515 def v8i16 : N2VQInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2516 itinQ, OpcodeStr, !strconcat(Dt, "16"),v8i16,v8i16,IntOp>;
2517 def v4i32 : N2VQInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2518 itinQ, OpcodeStr, !strconcat(Dt, "32"),v4i32,v4i32,IntOp>;
2522 // Neon Narrowing 2-register vector operations,
2523 // source operand element sizes of 16, 32 and 64 bits:
2524 multiclass N2VN_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2525 bits<5> op11_7, bit op6, bit op4,
2526 InstrItinClass itin, string OpcodeStr, string Dt,
2528 def v8i8 : N2VN<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
2529 itin, OpcodeStr, !strconcat(Dt, "16"),
2530 v8i8, v8i16, OpNode>;
2531 def v4i16 : N2VN<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
2532 itin, OpcodeStr, !strconcat(Dt, "32"),
2533 v4i16, v4i32, OpNode>;
2534 def v2i32 : N2VN<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
2535 itin, OpcodeStr, !strconcat(Dt, "64"),
2536 v2i32, v2i64, OpNode>;
2539 // Neon Narrowing 2-register vector intrinsics,
2540 // source operand element sizes of 16, 32 and 64 bits:
2541 multiclass N2VNInt_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2542 bits<5> op11_7, bit op6, bit op4,
2543 InstrItinClass itin, string OpcodeStr, string Dt,
2545 def v8i8 : N2VNInt<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
2546 itin, OpcodeStr, !strconcat(Dt, "16"),
2547 v8i8, v8i16, IntOp>;
2548 def v4i16 : N2VNInt<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
2549 itin, OpcodeStr, !strconcat(Dt, "32"),
2550 v4i16, v4i32, IntOp>;
2551 def v2i32 : N2VNInt<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
2552 itin, OpcodeStr, !strconcat(Dt, "64"),
2553 v2i32, v2i64, IntOp>;
2557 // Neon Lengthening 2-register vector intrinsic (currently specific to VMOVL).
2558 // source operand element sizes of 16, 32 and 64 bits:
2559 multiclass N2VL_QHS<bits<2> op24_23, bits<5> op11_7, bit op6, bit op4,
2560 string OpcodeStr, string Dt, SDNode OpNode> {
2561 def v8i16 : N2VL<op24_23, 0b00, 0b10, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2562 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode>;
2563 def v4i32 : N2VL<op24_23, 0b01, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2564 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
2565 def v2i64 : N2VL<op24_23, 0b10, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2566 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
2570 // Neon 3-register vector operations.
2572 // First with only element sizes of 8, 16 and 32 bits:
2573 multiclass N3V_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2574 InstrItinClass itinD16, InstrItinClass itinD32,
2575 InstrItinClass itinQ16, InstrItinClass itinQ32,
2576 string OpcodeStr, string Dt,
2577 SDNode OpNode, bit Commutable = 0> {
2578 // 64-bit vector types.
2579 def v8i8 : N3VD<op24, op23, 0b00, op11_8, op4, itinD16,
2580 OpcodeStr, !strconcat(Dt, "8"),
2581 v8i8, v8i8, OpNode, Commutable>;
2582 def v4i16 : N3VD<op24, op23, 0b01, op11_8, op4, itinD16,
2583 OpcodeStr, !strconcat(Dt, "16"),
2584 v4i16, v4i16, OpNode, Commutable>;
2585 def v2i32 : N3VD<op24, op23, 0b10, op11_8, op4, itinD32,
2586 OpcodeStr, !strconcat(Dt, "32"),
2587 v2i32, v2i32, OpNode, Commutable>;
2589 // 128-bit vector types.
2590 def v16i8 : N3VQ<op24, op23, 0b00, op11_8, op4, itinQ16,
2591 OpcodeStr, !strconcat(Dt, "8"),
2592 v16i8, v16i8, OpNode, Commutable>;
2593 def v8i16 : N3VQ<op24, op23, 0b01, op11_8, op4, itinQ16,
2594 OpcodeStr, !strconcat(Dt, "16"),
2595 v8i16, v8i16, OpNode, Commutable>;
2596 def v4i32 : N3VQ<op24, op23, 0b10, op11_8, op4, itinQ32,
2597 OpcodeStr, !strconcat(Dt, "32"),
2598 v4i32, v4i32, OpNode, Commutable>;
2601 multiclass N3VSL_HS<bits<4> op11_8, string OpcodeStr, string Dt, SDNode ShOp> {
2602 def v4i16 : N3VDSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
2604 def v2i32 : N3VDSL<0b10, op11_8, IIC_VMULi32D, OpcodeStr, !strconcat(Dt,"32"),
2606 def v8i16 : N3VQSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
2607 v8i16, v4i16, ShOp>;
2608 def v4i32 : N3VQSL<0b10, op11_8, IIC_VMULi32Q, OpcodeStr, !strconcat(Dt,"32"),
2609 v4i32, v2i32, ShOp>;
2612 // ....then also with element size 64 bits:
2613 multiclass N3V_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2614 InstrItinClass itinD, InstrItinClass itinQ,
2615 string OpcodeStr, string Dt,
2616 SDNode OpNode, bit Commutable = 0>
2617 : N3V_QHS<op24, op23, op11_8, op4, itinD, itinD, itinQ, itinQ,
2618 OpcodeStr, Dt, OpNode, Commutable> {
2619 def v1i64 : N3VD<op24, op23, 0b11, op11_8, op4, itinD,
2620 OpcodeStr, !strconcat(Dt, "64"),
2621 v1i64, v1i64, OpNode, Commutable>;
2622 def v2i64 : N3VQ<op24, op23, 0b11, op11_8, op4, itinQ,
2623 OpcodeStr, !strconcat(Dt, "64"),
2624 v2i64, v2i64, OpNode, Commutable>;
2628 // Neon 3-register vector intrinsics.
2630 // First with only element sizes of 16 and 32 bits:
2631 multiclass N3VInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2632 InstrItinClass itinD16, InstrItinClass itinD32,
2633 InstrItinClass itinQ16, InstrItinClass itinQ32,
2634 string OpcodeStr, string Dt,
2635 Intrinsic IntOp, bit Commutable = 0> {
2636 // 64-bit vector types.
2637 def v4i16 : N3VDInt<op24, op23, 0b01, op11_8, op4, f, itinD16,
2638 OpcodeStr, !strconcat(Dt, "16"),
2639 v4i16, v4i16, IntOp, Commutable>;
2640 def v2i32 : N3VDInt<op24, op23, 0b10, op11_8, op4, f, itinD32,
2641 OpcodeStr, !strconcat(Dt, "32"),
2642 v2i32, v2i32, IntOp, Commutable>;
2644 // 128-bit vector types.
2645 def v8i16 : N3VQInt<op24, op23, 0b01, op11_8, op4, f, itinQ16,
2646 OpcodeStr, !strconcat(Dt, "16"),
2647 v8i16, v8i16, IntOp, Commutable>;
2648 def v4i32 : N3VQInt<op24, op23, 0b10, op11_8, op4, f, itinQ32,
2649 OpcodeStr, !strconcat(Dt, "32"),
2650 v4i32, v4i32, IntOp, Commutable>;
2652 multiclass N3VInt_HSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2653 InstrItinClass itinD16, InstrItinClass itinD32,
2654 InstrItinClass itinQ16, InstrItinClass itinQ32,
2655 string OpcodeStr, string Dt,
2657 // 64-bit vector types.
2658 def v4i16 : N3VDIntSh<op24, op23, 0b01, op11_8, op4, f, itinD16,
2659 OpcodeStr, !strconcat(Dt, "16"),
2660 v4i16, v4i16, IntOp>;
2661 def v2i32 : N3VDIntSh<op24, op23, 0b10, op11_8, op4, f, itinD32,
2662 OpcodeStr, !strconcat(Dt, "32"),
2663 v2i32, v2i32, IntOp>;
2665 // 128-bit vector types.
2666 def v8i16 : N3VQIntSh<op24, op23, 0b01, op11_8, op4, f, itinQ16,
2667 OpcodeStr, !strconcat(Dt, "16"),
2668 v8i16, v8i16, IntOp>;
2669 def v4i32 : N3VQIntSh<op24, op23, 0b10, op11_8, op4, f, itinQ32,
2670 OpcodeStr, !strconcat(Dt, "32"),
2671 v4i32, v4i32, IntOp>;
2674 multiclass N3VIntSL_HS<bits<4> op11_8,
2675 InstrItinClass itinD16, InstrItinClass itinD32,
2676 InstrItinClass itinQ16, InstrItinClass itinQ32,
2677 string OpcodeStr, string Dt, Intrinsic IntOp> {
2678 def v4i16 : N3VDIntSL16<0b01, op11_8, itinD16,
2679 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp>;
2680 def v2i32 : N3VDIntSL<0b10, op11_8, itinD32,
2681 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp>;
2682 def v8i16 : N3VQIntSL16<0b01, op11_8, itinQ16,
2683 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16, IntOp>;
2684 def v4i32 : N3VQIntSL<0b10, op11_8, itinQ32,
2685 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32, IntOp>;
2688 // ....then also with element size of 8 bits:
2689 multiclass N3VInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2690 InstrItinClass itinD16, InstrItinClass itinD32,
2691 InstrItinClass itinQ16, InstrItinClass itinQ32,
2692 string OpcodeStr, string Dt,
2693 Intrinsic IntOp, bit Commutable = 0>
2694 : N3VInt_HS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2695 OpcodeStr, Dt, IntOp, Commutable> {
2696 def v8i8 : N3VDInt<op24, op23, 0b00, op11_8, op4, f, itinD16,
2697 OpcodeStr, !strconcat(Dt, "8"),
2698 v8i8, v8i8, IntOp, Commutable>;
2699 def v16i8 : N3VQInt<op24, op23, 0b00, op11_8, op4, f, itinQ16,
2700 OpcodeStr, !strconcat(Dt, "8"),
2701 v16i8, v16i8, IntOp, Commutable>;
2703 multiclass N3VInt_QHSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2704 InstrItinClass itinD16, InstrItinClass itinD32,
2705 InstrItinClass itinQ16, InstrItinClass itinQ32,
2706 string OpcodeStr, string Dt,
2708 : N3VInt_HSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2709 OpcodeStr, Dt, IntOp> {
2710 def v8i8 : N3VDIntSh<op24, op23, 0b00, op11_8, op4, f, itinD16,
2711 OpcodeStr, !strconcat(Dt, "8"),
2713 def v16i8 : N3VQIntSh<op24, op23, 0b00, op11_8, op4, f, itinQ16,
2714 OpcodeStr, !strconcat(Dt, "8"),
2715 v16i8, v16i8, IntOp>;
2719 // ....then also with element size of 64 bits:
2720 multiclass N3VInt_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2721 InstrItinClass itinD16, InstrItinClass itinD32,
2722 InstrItinClass itinQ16, InstrItinClass itinQ32,
2723 string OpcodeStr, string Dt,
2724 Intrinsic IntOp, bit Commutable = 0>
2725 : N3VInt_QHS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2726 OpcodeStr, Dt, IntOp, Commutable> {
2727 def v1i64 : N3VDInt<op24, op23, 0b11, op11_8, op4, f, itinD32,
2728 OpcodeStr, !strconcat(Dt, "64"),
2729 v1i64, v1i64, IntOp, Commutable>;
2730 def v2i64 : N3VQInt<op24, op23, 0b11, op11_8, op4, f, itinQ32,
2731 OpcodeStr, !strconcat(Dt, "64"),
2732 v2i64, v2i64, IntOp, Commutable>;
2734 multiclass N3VInt_QHSDSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2735 InstrItinClass itinD16, InstrItinClass itinD32,
2736 InstrItinClass itinQ16, InstrItinClass itinQ32,
2737 string OpcodeStr, string Dt,
2739 : N3VInt_QHSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2740 OpcodeStr, Dt, IntOp> {
2741 def v1i64 : N3VDIntSh<op24, op23, 0b11, op11_8, op4, f, itinD32,
2742 OpcodeStr, !strconcat(Dt, "64"),
2743 v1i64, v1i64, IntOp>;
2744 def v2i64 : N3VQIntSh<op24, op23, 0b11, op11_8, op4, f, itinQ32,
2745 OpcodeStr, !strconcat(Dt, "64"),
2746 v2i64, v2i64, IntOp>;
2749 // Neon Narrowing 3-register vector intrinsics,
2750 // source operand element sizes of 16, 32 and 64 bits:
2751 multiclass N3VNInt_HSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2752 string OpcodeStr, string Dt,
2753 Intrinsic IntOp, bit Commutable = 0> {
2754 def v8i8 : N3VNInt<op24, op23, 0b00, op11_8, op4,
2755 OpcodeStr, !strconcat(Dt, "16"),
2756 v8i8, v8i16, IntOp, Commutable>;
2757 def v4i16 : N3VNInt<op24, op23, 0b01, op11_8, op4,
2758 OpcodeStr, !strconcat(Dt, "32"),
2759 v4i16, v4i32, IntOp, Commutable>;
2760 def v2i32 : N3VNInt<op24, op23, 0b10, op11_8, op4,
2761 OpcodeStr, !strconcat(Dt, "64"),
2762 v2i32, v2i64, IntOp, Commutable>;
2766 // Neon Long 3-register vector operations.
2768 multiclass N3VL_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2769 InstrItinClass itin16, InstrItinClass itin32,
2770 string OpcodeStr, string Dt,
2771 SDNode OpNode, bit Commutable = 0> {
2772 def v8i16 : N3VL<op24, op23, 0b00, op11_8, op4, itin16,
2773 OpcodeStr, !strconcat(Dt, "8"),
2774 v8i16, v8i8, OpNode, Commutable>;
2775 def v4i32 : N3VL<op24, op23, 0b01, op11_8, op4, itin16,
2776 OpcodeStr, !strconcat(Dt, "16"),
2777 v4i32, v4i16, OpNode, Commutable>;
2778 def v2i64 : N3VL<op24, op23, 0b10, op11_8, op4, itin32,
2779 OpcodeStr, !strconcat(Dt, "32"),
2780 v2i64, v2i32, OpNode, Commutable>;
2783 multiclass N3VLSL_HS<bit op24, bits<4> op11_8,
2784 InstrItinClass itin, string OpcodeStr, string Dt,
2786 def v4i16 : N3VLSL16<op24, 0b01, op11_8, itin, OpcodeStr,
2787 !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
2788 def v2i32 : N3VLSL<op24, 0b10, op11_8, itin, OpcodeStr,
2789 !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
2792 multiclass N3VLExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2793 InstrItinClass itin16, InstrItinClass itin32,
2794 string OpcodeStr, string Dt,
2795 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
2796 def v8i16 : N3VLExt<op24, op23, 0b00, op11_8, op4, itin16,
2797 OpcodeStr, !strconcat(Dt, "8"),
2798 v8i16, v8i8, OpNode, ExtOp, Commutable>;
2799 def v4i32 : N3VLExt<op24, op23, 0b01, op11_8, op4, itin16,
2800 OpcodeStr, !strconcat(Dt, "16"),
2801 v4i32, v4i16, OpNode, ExtOp, Commutable>;
2802 def v2i64 : N3VLExt<op24, op23, 0b10, op11_8, op4, itin32,
2803 OpcodeStr, !strconcat(Dt, "32"),
2804 v2i64, v2i32, OpNode, ExtOp, Commutable>;
2807 // Neon Long 3-register vector intrinsics.
2809 // First with only element sizes of 16 and 32 bits:
2810 multiclass N3VLInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
2811 InstrItinClass itin16, InstrItinClass itin32,
2812 string OpcodeStr, string Dt,
2813 Intrinsic IntOp, bit Commutable = 0> {
2814 def v4i32 : N3VLInt<op24, op23, 0b01, op11_8, op4, itin16,
2815 OpcodeStr, !strconcat(Dt, "16"),
2816 v4i32, v4i16, IntOp, Commutable>;
2817 def v2i64 : N3VLInt<op24, op23, 0b10, op11_8, op4, itin32,
2818 OpcodeStr, !strconcat(Dt, "32"),
2819 v2i64, v2i32, IntOp, Commutable>;
2822 multiclass N3VLIntSL_HS<bit op24, bits<4> op11_8,
2823 InstrItinClass itin, string OpcodeStr, string Dt,
2825 def v4i16 : N3VLIntSL16<op24, 0b01, op11_8, itin,
2826 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
2827 def v2i32 : N3VLIntSL<op24, 0b10, op11_8, itin,
2828 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
2831 // ....then also with element size of 8 bits:
2832 multiclass N3VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2833 InstrItinClass itin16, InstrItinClass itin32,
2834 string OpcodeStr, string Dt,
2835 Intrinsic IntOp, bit Commutable = 0>
2836 : N3VLInt_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt,
2837 IntOp, Commutable> {
2838 def v8i16 : N3VLInt<op24, op23, 0b00, op11_8, op4, itin16,
2839 OpcodeStr, !strconcat(Dt, "8"),
2840 v8i16, v8i8, IntOp, Commutable>;
2843 // ....with explicit extend (VABDL).
2844 multiclass N3VLIntExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2845 InstrItinClass itin, string OpcodeStr, string Dt,
2846 Intrinsic IntOp, SDNode ExtOp, bit Commutable = 0> {
2847 def v8i16 : N3VLIntExt<op24, op23, 0b00, op11_8, op4, itin,
2848 OpcodeStr, !strconcat(Dt, "8"),
2849 v8i16, v8i8, IntOp, ExtOp, Commutable>;
2850 def v4i32 : N3VLIntExt<op24, op23, 0b01, op11_8, op4, itin,
2851 OpcodeStr, !strconcat(Dt, "16"),
2852 v4i32, v4i16, IntOp, ExtOp, Commutable>;
2853 def v2i64 : N3VLIntExt<op24, op23, 0b10, op11_8, op4, itin,
2854 OpcodeStr, !strconcat(Dt, "32"),
2855 v2i64, v2i32, IntOp, ExtOp, Commutable>;
2859 // Neon Wide 3-register vector intrinsics,
2860 // source operand element sizes of 8, 16 and 32 bits:
2861 multiclass N3VW_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2862 string OpcodeStr, string Dt,
2863 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
2864 def v8i16 : N3VW<op24, op23, 0b00, op11_8, op4,
2865 OpcodeStr, !strconcat(Dt, "8"),
2866 v8i16, v8i8, OpNode, ExtOp, Commutable>;
2867 def v4i32 : N3VW<op24, op23, 0b01, op11_8, op4,
2868 OpcodeStr, !strconcat(Dt, "16"),
2869 v4i32, v4i16, OpNode, ExtOp, Commutable>;
2870 def v2i64 : N3VW<op24, op23, 0b10, op11_8, op4,
2871 OpcodeStr, !strconcat(Dt, "32"),
2872 v2i64, v2i32, OpNode, ExtOp, Commutable>;
2876 // Neon Multiply-Op vector operations,
2877 // element sizes of 8, 16 and 32 bits:
2878 multiclass N3VMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2879 InstrItinClass itinD16, InstrItinClass itinD32,
2880 InstrItinClass itinQ16, InstrItinClass itinQ32,
2881 string OpcodeStr, string Dt, SDNode OpNode> {
2882 // 64-bit vector types.
2883 def v8i8 : N3VDMulOp<op24, op23, 0b00, op11_8, op4, itinD16,
2884 OpcodeStr, !strconcat(Dt, "8"), v8i8, mul, OpNode>;
2885 def v4i16 : N3VDMulOp<op24, op23, 0b01, op11_8, op4, itinD16,
2886 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, OpNode>;
2887 def v2i32 : N3VDMulOp<op24, op23, 0b10, op11_8, op4, itinD32,
2888 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, OpNode>;
2890 // 128-bit vector types.
2891 def v16i8 : N3VQMulOp<op24, op23, 0b00, op11_8, op4, itinQ16,
2892 OpcodeStr, !strconcat(Dt, "8"), v16i8, mul, OpNode>;
2893 def v8i16 : N3VQMulOp<op24, op23, 0b01, op11_8, op4, itinQ16,
2894 OpcodeStr, !strconcat(Dt, "16"), v8i16, mul, OpNode>;
2895 def v4i32 : N3VQMulOp<op24, op23, 0b10, op11_8, op4, itinQ32,
2896 OpcodeStr, !strconcat(Dt, "32"), v4i32, mul, OpNode>;
2899 multiclass N3VMulOpSL_HS<bits<4> op11_8,
2900 InstrItinClass itinD16, InstrItinClass itinD32,
2901 InstrItinClass itinQ16, InstrItinClass itinQ32,
2902 string OpcodeStr, string Dt, SDNode ShOp> {
2903 def v4i16 : N3VDMulOpSL16<0b01, op11_8, itinD16,
2904 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, ShOp>;
2905 def v2i32 : N3VDMulOpSL<0b10, op11_8, itinD32,
2906 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, ShOp>;
2907 def v8i16 : N3VQMulOpSL16<0b01, op11_8, itinQ16,
2908 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16,
2910 def v4i32 : N3VQMulOpSL<0b10, op11_8, itinQ32,
2911 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32,
2915 // Neon Intrinsic-Op vector operations,
2916 // element sizes of 8, 16 and 32 bits:
2917 multiclass N3VIntOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2918 InstrItinClass itinD, InstrItinClass itinQ,
2919 string OpcodeStr, string Dt, Intrinsic IntOp,
2921 // 64-bit vector types.
2922 def v8i8 : N3VDIntOp<op24, op23, 0b00, op11_8, op4, itinD,
2923 OpcodeStr, !strconcat(Dt, "8"), v8i8, IntOp, OpNode>;
2924 def v4i16 : N3VDIntOp<op24, op23, 0b01, op11_8, op4, itinD,
2925 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp, OpNode>;
2926 def v2i32 : N3VDIntOp<op24, op23, 0b10, op11_8, op4, itinD,
2927 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp, OpNode>;
2929 // 128-bit vector types.
2930 def v16i8 : N3VQIntOp<op24, op23, 0b00, op11_8, op4, itinQ,
2931 OpcodeStr, !strconcat(Dt, "8"), v16i8, IntOp, OpNode>;
2932 def v8i16 : N3VQIntOp<op24, op23, 0b01, op11_8, op4, itinQ,
2933 OpcodeStr, !strconcat(Dt, "16"), v8i16, IntOp, OpNode>;
2934 def v4i32 : N3VQIntOp<op24, op23, 0b10, op11_8, op4, itinQ,
2935 OpcodeStr, !strconcat(Dt, "32"), v4i32, IntOp, OpNode>;
2938 // Neon 3-argument intrinsics,
2939 // element sizes of 8, 16 and 32 bits:
2940 multiclass N3VInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2941 InstrItinClass itinD, InstrItinClass itinQ,
2942 string OpcodeStr, string Dt, Intrinsic IntOp> {
2943 // 64-bit vector types.
2944 def v8i8 : N3VDInt3<op24, op23, 0b00, op11_8, op4, itinD,
2945 OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
2946 def v4i16 : N3VDInt3<op24, op23, 0b01, op11_8, op4, itinD,
2947 OpcodeStr, !strconcat(Dt, "16"), v4i16, v4i16, IntOp>;
2948 def v2i32 : N3VDInt3<op24, op23, 0b10, op11_8, op4, itinD,
2949 OpcodeStr, !strconcat(Dt, "32"), v2i32, v2i32, IntOp>;
2951 // 128-bit vector types.
2952 def v16i8 : N3VQInt3<op24, op23, 0b00, op11_8, op4, itinQ,
2953 OpcodeStr, !strconcat(Dt, "8"), v16i8, v16i8, IntOp>;
2954 def v8i16 : N3VQInt3<op24, op23, 0b01, op11_8, op4, itinQ,
2955 OpcodeStr, !strconcat(Dt, "16"), v8i16, v8i16, IntOp>;
2956 def v4i32 : N3VQInt3<op24, op23, 0b10, op11_8, op4, itinQ,
2957 OpcodeStr, !strconcat(Dt, "32"), v4i32, v4i32, IntOp>;
2961 // Neon Long Multiply-Op vector operations,
2962 // element sizes of 8, 16 and 32 bits:
2963 multiclass N3VLMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2964 InstrItinClass itin16, InstrItinClass itin32,
2965 string OpcodeStr, string Dt, SDNode MulOp,
2967 def v8i16 : N3VLMulOp<op24, op23, 0b00, op11_8, op4, itin16, OpcodeStr,
2968 !strconcat(Dt, "8"), v8i16, v8i8, MulOp, OpNode>;
2969 def v4i32 : N3VLMulOp<op24, op23, 0b01, op11_8, op4, itin16, OpcodeStr,
2970 !strconcat(Dt, "16"), v4i32, v4i16, MulOp, OpNode>;
2971 def v2i64 : N3VLMulOp<op24, op23, 0b10, op11_8, op4, itin32, OpcodeStr,
2972 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
2975 multiclass N3VLMulOpSL_HS<bit op24, bits<4> op11_8, string OpcodeStr,
2976 string Dt, SDNode MulOp, SDNode OpNode> {
2977 def v4i16 : N3VLMulOpSL16<op24, 0b01, op11_8, IIC_VMACi16D, OpcodeStr,
2978 !strconcat(Dt,"16"), v4i32, v4i16, MulOp, OpNode>;
2979 def v2i32 : N3VLMulOpSL<op24, 0b10, op11_8, IIC_VMACi32D, OpcodeStr,
2980 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
2984 // Neon Long 3-argument intrinsics.
2986 // First with only element sizes of 16 and 32 bits:
2987 multiclass N3VLInt3_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
2988 InstrItinClass itin16, InstrItinClass itin32,
2989 string OpcodeStr, string Dt, Intrinsic IntOp> {
2990 def v4i32 : N3VLInt3<op24, op23, 0b01, op11_8, op4, itin16,
2991 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
2992 def v2i64 : N3VLInt3<op24, op23, 0b10, op11_8, op4, itin32,
2993 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
2996 multiclass N3VLInt3SL_HS<bit op24, bits<4> op11_8,
2997 string OpcodeStr, string Dt, Intrinsic IntOp> {
2998 def v4i16 : N3VLInt3SL16<op24, 0b01, op11_8, IIC_VMACi16D,
2999 OpcodeStr, !strconcat(Dt,"16"), v4i32, v4i16, IntOp>;
3000 def v2i32 : N3VLInt3SL<op24, 0b10, op11_8, IIC_VMACi32D,
3001 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3004 // ....then also with element size of 8 bits:
3005 multiclass N3VLInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3006 InstrItinClass itin16, InstrItinClass itin32,
3007 string OpcodeStr, string Dt, Intrinsic IntOp>
3008 : N3VLInt3_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt, IntOp> {
3009 def v8i16 : N3VLInt3<op24, op23, 0b00, op11_8, op4, itin16,
3010 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
3013 // ....with explicit extend (VABAL).
3014 multiclass N3VLIntExtOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3015 InstrItinClass itin, string OpcodeStr, string Dt,
3016 Intrinsic IntOp, SDNode ExtOp, SDNode OpNode> {
3017 def v8i16 : N3VLIntExtOp<op24, op23, 0b00, op11_8, op4, itin,
3018 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8,
3019 IntOp, ExtOp, OpNode>;
3020 def v4i32 : N3VLIntExtOp<op24, op23, 0b01, op11_8, op4, itin,
3021 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16,
3022 IntOp, ExtOp, OpNode>;
3023 def v2i64 : N3VLIntExtOp<op24, op23, 0b10, op11_8, op4, itin,
3024 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32,
3025 IntOp, ExtOp, OpNode>;
3029 // Neon Pairwise long 2-register intrinsics,
3030 // element sizes of 8, 16 and 32 bits:
3031 multiclass N2VPLInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3032 bits<5> op11_7, bit op4,
3033 string OpcodeStr, string Dt, Intrinsic IntOp> {
3034 // 64-bit vector types.
3035 def v8i8 : N2VDPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3036 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
3037 def v4i16 : N2VDPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3038 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
3039 def v2i32 : N2VDPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3040 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
3042 // 128-bit vector types.
3043 def v16i8 : N2VQPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3044 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
3045 def v8i16 : N2VQPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3046 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
3047 def v4i32 : N2VQPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3048 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
3052 // Neon Pairwise long 2-register accumulate intrinsics,
3053 // element sizes of 8, 16 and 32 bits:
3054 multiclass N2VPLInt2_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3055 bits<5> op11_7, bit op4,
3056 string OpcodeStr, string Dt, Intrinsic IntOp> {
3057 // 64-bit vector types.
3058 def v8i8 : N2VDPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3059 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
3060 def v4i16 : N2VDPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3061 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
3062 def v2i32 : N2VDPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3063 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
3065 // 128-bit vector types.
3066 def v16i8 : N2VQPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3067 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
3068 def v8i16 : N2VQPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3069 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
3070 def v4i32 : N2VQPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3071 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
3075 // Neon 2-register vector shift by immediate,
3076 // with f of either N2RegVShLFrm or N2RegVShRFrm
3077 // element sizes of 8, 16, 32 and 64 bits:
3078 multiclass N2VShL_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3079 InstrItinClass itin, string OpcodeStr, string Dt,
3081 // 64-bit vector types.
3082 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3083 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
3084 let Inst{21-19} = 0b001; // imm6 = 001xxx
3086 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3087 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
3088 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3090 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3091 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
3092 let Inst{21} = 0b1; // imm6 = 1xxxxx
3094 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, N2RegVShLFrm, itin, i32imm,
3095 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
3098 // 128-bit vector types.
3099 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3100 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
3101 let Inst{21-19} = 0b001; // imm6 = 001xxx
3103 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3104 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
3105 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3107 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3108 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
3109 let Inst{21} = 0b1; // imm6 = 1xxxxx
3111 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, N2RegVShLFrm, itin, i32imm,
3112 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
3115 multiclass N2VShR_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3116 InstrItinClass itin, string OpcodeStr, string Dt,
3118 // 64-bit vector types.
3119 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm8,
3120 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
3121 let Inst{21-19} = 0b001; // imm6 = 001xxx
3123 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm16,
3124 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
3125 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3127 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm32,
3128 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
3129 let Inst{21} = 0b1; // imm6 = 1xxxxx
3131 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, N2RegVShRFrm, itin, shr_imm64,
3132 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
3135 // 128-bit vector types.
3136 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm8,
3137 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
3138 let Inst{21-19} = 0b001; // imm6 = 001xxx
3140 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm16,
3141 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
3142 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3144 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm32,
3145 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
3146 let Inst{21} = 0b1; // imm6 = 1xxxxx
3148 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, N2RegVShRFrm, itin, shr_imm64,
3149 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
3153 // Neon Shift-Accumulate vector operations,
3154 // element sizes of 8, 16, 32 and 64 bits:
3155 multiclass N2VShAdd_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3156 string OpcodeStr, string Dt, SDNode ShOp> {
3157 // 64-bit vector types.
3158 def v8i8 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm8,
3159 OpcodeStr, !strconcat(Dt, "8"), v8i8, ShOp> {
3160 let Inst{21-19} = 0b001; // imm6 = 001xxx
3162 def v4i16 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm16,
3163 OpcodeStr, !strconcat(Dt, "16"), v4i16, ShOp> {
3164 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3166 def v2i32 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm32,
3167 OpcodeStr, !strconcat(Dt, "32"), v2i32, ShOp> {
3168 let Inst{21} = 0b1; // imm6 = 1xxxxx
3170 def v1i64 : N2VDShAdd<op24, op23, op11_8, 1, op4, shr_imm64,
3171 OpcodeStr, !strconcat(Dt, "64"), v1i64, ShOp>;
3174 // 128-bit vector types.
3175 def v16i8 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm8,
3176 OpcodeStr, !strconcat(Dt, "8"), v16i8, ShOp> {
3177 let Inst{21-19} = 0b001; // imm6 = 001xxx
3179 def v8i16 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm16,
3180 OpcodeStr, !strconcat(Dt, "16"), v8i16, ShOp> {
3181 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3183 def v4i32 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm32,
3184 OpcodeStr, !strconcat(Dt, "32"), v4i32, ShOp> {
3185 let Inst{21} = 0b1; // imm6 = 1xxxxx
3187 def v2i64 : N2VQShAdd<op24, op23, op11_8, 1, op4, shr_imm64,
3188 OpcodeStr, !strconcat(Dt, "64"), v2i64, ShOp>;
3192 // Neon Shift-Insert vector operations,
3193 // with f of either N2RegVShLFrm or N2RegVShRFrm
3194 // element sizes of 8, 16, 32 and 64 bits:
3195 multiclass N2VShInsL_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3197 // 64-bit vector types.
3198 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3199 N2RegVShLFrm, OpcodeStr, "8", v8i8, NEONvsli> {
3200 let Inst{21-19} = 0b001; // imm6 = 001xxx
3202 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3203 N2RegVShLFrm, OpcodeStr, "16", v4i16, NEONvsli> {
3204 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3206 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3207 N2RegVShLFrm, OpcodeStr, "32", v2i32, NEONvsli> {
3208 let Inst{21} = 0b1; // imm6 = 1xxxxx
3210 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4, i32imm,
3211 N2RegVShLFrm, OpcodeStr, "64", v1i64, NEONvsli>;
3214 // 128-bit vector types.
3215 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3216 N2RegVShLFrm, OpcodeStr, "8", v16i8, NEONvsli> {
3217 let Inst{21-19} = 0b001; // imm6 = 001xxx
3219 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3220 N2RegVShLFrm, OpcodeStr, "16", v8i16, NEONvsli> {
3221 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3223 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3224 N2RegVShLFrm, OpcodeStr, "32", v4i32, NEONvsli> {
3225 let Inst{21} = 0b1; // imm6 = 1xxxxx
3227 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4, i32imm,
3228 N2RegVShLFrm, OpcodeStr, "64", v2i64, NEONvsli>;
3231 multiclass N2VShInsR_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3233 // 64-bit vector types.
3234 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm8,
3235 N2RegVShRFrm, OpcodeStr, "8", v8i8, NEONvsri> {
3236 let Inst{21-19} = 0b001; // imm6 = 001xxx
3238 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm16,
3239 N2RegVShRFrm, OpcodeStr, "16", v4i16, NEONvsri> {
3240 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3242 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm32,
3243 N2RegVShRFrm, OpcodeStr, "32", v2i32, NEONvsri> {
3244 let Inst{21} = 0b1; // imm6 = 1xxxxx
3246 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4, shr_imm64,
3247 N2RegVShRFrm, OpcodeStr, "64", v1i64, NEONvsri>;
3250 // 128-bit vector types.
3251 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm8,
3252 N2RegVShRFrm, OpcodeStr, "8", v16i8, NEONvsri> {
3253 let Inst{21-19} = 0b001; // imm6 = 001xxx
3255 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm16,
3256 N2RegVShRFrm, OpcodeStr, "16", v8i16, NEONvsri> {
3257 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3259 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm32,
3260 N2RegVShRFrm, OpcodeStr, "32", v4i32, NEONvsri> {
3261 let Inst{21} = 0b1; // imm6 = 1xxxxx
3263 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4, shr_imm64,
3264 N2RegVShRFrm, OpcodeStr, "64", v2i64, NEONvsri>;
3268 // Neon Shift Long operations,
3269 // element sizes of 8, 16, 32 bits:
3270 multiclass N2VLSh_QHS<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
3271 bit op4, string OpcodeStr, string Dt, SDNode OpNode> {
3272 def v8i16 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3273 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode> {
3274 let Inst{21-19} = 0b001; // imm6 = 001xxx
3276 def v4i32 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3277 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode> {
3278 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3280 def v2i64 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3281 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode> {
3282 let Inst{21} = 0b1; // imm6 = 1xxxxx
3286 // Neon Shift Narrow operations,
3287 // element sizes of 16, 32, 64 bits:
3288 multiclass N2VNSh_HSD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
3289 bit op4, InstrItinClass itin, string OpcodeStr, string Dt,
3291 def v8i8 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3292 OpcodeStr, !strconcat(Dt, "16"),
3293 v8i8, v8i16, shr_imm8, OpNode> {
3294 let Inst{21-19} = 0b001; // imm6 = 001xxx
3296 def v4i16 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3297 OpcodeStr, !strconcat(Dt, "32"),
3298 v4i16, v4i32, shr_imm16, OpNode> {
3299 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3301 def v2i32 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3302 OpcodeStr, !strconcat(Dt, "64"),
3303 v2i32, v2i64, shr_imm32, OpNode> {
3304 let Inst{21} = 0b1; // imm6 = 1xxxxx
3308 //===----------------------------------------------------------------------===//
3309 // Instruction Definitions.
3310 //===----------------------------------------------------------------------===//
3312 // Vector Add Operations.
3314 // VADD : Vector Add (integer and floating-point)
3315 defm VADD : N3V_QHSD<0, 0, 0b1000, 0, IIC_VBINiD, IIC_VBINiQ, "vadd", "i",
3317 def VADDfd : N3VD<0, 0, 0b00, 0b1101, 0, IIC_VBIND, "vadd", "f32",
3318 v2f32, v2f32, fadd, 1>;
3319 def VADDfq : N3VQ<0, 0, 0b00, 0b1101, 0, IIC_VBINQ, "vadd", "f32",
3320 v4f32, v4f32, fadd, 1>;
3321 // VADDL : Vector Add Long (Q = D + D)
3322 defm VADDLs : N3VLExt_QHS<0,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
3323 "vaddl", "s", add, sext, 1>;
3324 defm VADDLu : N3VLExt_QHS<1,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
3325 "vaddl", "u", add, zext, 1>;
3326 // VADDW : Vector Add Wide (Q = Q + D)
3327 defm VADDWs : N3VW_QHS<0,1,0b0001,0, "vaddw", "s", add, sext, 0>;
3328 defm VADDWu : N3VW_QHS<1,1,0b0001,0, "vaddw", "u", add, zext, 0>;
3329 // VHADD : Vector Halving Add
3330 defm VHADDs : N3VInt_QHS<0, 0, 0b0000, 0, N3RegFrm,
3331 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3332 "vhadd", "s", int_arm_neon_vhadds, 1>;
3333 defm VHADDu : N3VInt_QHS<1, 0, 0b0000, 0, N3RegFrm,
3334 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3335 "vhadd", "u", int_arm_neon_vhaddu, 1>;
3336 // VRHADD : Vector Rounding Halving Add
3337 defm VRHADDs : N3VInt_QHS<0, 0, 0b0001, 0, N3RegFrm,
3338 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3339 "vrhadd", "s", int_arm_neon_vrhadds, 1>;
3340 defm VRHADDu : N3VInt_QHS<1, 0, 0b0001, 0, N3RegFrm,
3341 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3342 "vrhadd", "u", int_arm_neon_vrhaddu, 1>;
3343 // VQADD : Vector Saturating Add
3344 defm VQADDs : N3VInt_QHSD<0, 0, 0b0000, 1, N3RegFrm,
3345 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3346 "vqadd", "s", int_arm_neon_vqadds, 1>;
3347 defm VQADDu : N3VInt_QHSD<1, 0, 0b0000, 1, N3RegFrm,
3348 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3349 "vqadd", "u", int_arm_neon_vqaddu, 1>;
3350 // VADDHN : Vector Add and Narrow Returning High Half (D = Q + Q)
3351 defm VADDHN : N3VNInt_HSD<0,1,0b0100,0, "vaddhn", "i",
3352 int_arm_neon_vaddhn, 1>;
3353 // VRADDHN : Vector Rounding Add and Narrow Returning High Half (D = Q + Q)
3354 defm VRADDHN : N3VNInt_HSD<1,1,0b0100,0, "vraddhn", "i",
3355 int_arm_neon_vraddhn, 1>;
3357 // Vector Multiply Operations.
3359 // VMUL : Vector Multiply (integer, polynomial and floating-point)
3360 defm VMUL : N3V_QHS<0, 0, 0b1001, 1, IIC_VMULi16D, IIC_VMULi32D,
3361 IIC_VMULi16Q, IIC_VMULi32Q, "vmul", "i", mul, 1>;
3362 def VMULpd : N3VDInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16D, "vmul",
3363 "p8", v8i8, v8i8, int_arm_neon_vmulp, 1>;
3364 def VMULpq : N3VQInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16Q, "vmul",
3365 "p8", v16i8, v16i8, int_arm_neon_vmulp, 1>;
3366 def VMULfd : N3VD<1, 0, 0b00, 0b1101, 1, IIC_VFMULD, "vmul", "f32",
3367 v2f32, v2f32, fmul, 1>;
3368 def VMULfq : N3VQ<1, 0, 0b00, 0b1101, 1, IIC_VFMULQ, "vmul", "f32",
3369 v4f32, v4f32, fmul, 1>;
3370 defm VMULsl : N3VSL_HS<0b1000, "vmul", "i", mul>;
3371 def VMULslfd : N3VDSL<0b10, 0b1001, IIC_VBIND, "vmul", "f32", v2f32, fmul>;
3372 def VMULslfq : N3VQSL<0b10, 0b1001, IIC_VBINQ, "vmul", "f32", v4f32,
3375 def : Pat<(v8i16 (mul (v8i16 QPR:$src1),
3376 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
3377 (v8i16 (VMULslv8i16 (v8i16 QPR:$src1),
3378 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3379 (DSubReg_i16_reg imm:$lane))),
3380 (SubReg_i16_lane imm:$lane)))>;
3381 def : Pat<(v4i32 (mul (v4i32 QPR:$src1),
3382 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
3383 (v4i32 (VMULslv4i32 (v4i32 QPR:$src1),
3384 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3385 (DSubReg_i32_reg imm:$lane))),
3386 (SubReg_i32_lane imm:$lane)))>;
3387 def : Pat<(v4f32 (fmul (v4f32 QPR:$src1),
3388 (v4f32 (NEONvduplane (v4f32 QPR:$src2), imm:$lane)))),
3389 (v4f32 (VMULslfq (v4f32 QPR:$src1),
3390 (v2f32 (EXTRACT_SUBREG QPR:$src2,
3391 (DSubReg_i32_reg imm:$lane))),
3392 (SubReg_i32_lane imm:$lane)))>;
3394 // VQDMULH : Vector Saturating Doubling Multiply Returning High Half
3395 defm VQDMULH : N3VInt_HS<0, 0, 0b1011, 0, N3RegFrm, IIC_VMULi16D, IIC_VMULi32D,
3396 IIC_VMULi16Q, IIC_VMULi32Q,
3397 "vqdmulh", "s", int_arm_neon_vqdmulh, 1>;
3398 defm VQDMULHsl: N3VIntSL_HS<0b1100, IIC_VMULi16D, IIC_VMULi32D,
3399 IIC_VMULi16Q, IIC_VMULi32Q,
3400 "vqdmulh", "s", int_arm_neon_vqdmulh>;
3401 def : Pat<(v8i16 (int_arm_neon_vqdmulh (v8i16 QPR:$src1),
3402 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
3404 (v8i16 (VQDMULHslv8i16 (v8i16 QPR:$src1),
3405 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3406 (DSubReg_i16_reg imm:$lane))),
3407 (SubReg_i16_lane imm:$lane)))>;
3408 def : Pat<(v4i32 (int_arm_neon_vqdmulh (v4i32 QPR:$src1),
3409 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
3411 (v4i32 (VQDMULHslv4i32 (v4i32 QPR:$src1),
3412 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3413 (DSubReg_i32_reg imm:$lane))),
3414 (SubReg_i32_lane imm:$lane)))>;
3416 // VQRDMULH : Vector Rounding Saturating Doubling Multiply Returning High Half
3417 defm VQRDMULH : N3VInt_HS<1, 0, 0b1011, 0, N3RegFrm,
3418 IIC_VMULi16D,IIC_VMULi32D,IIC_VMULi16Q,IIC_VMULi32Q,
3419 "vqrdmulh", "s", int_arm_neon_vqrdmulh, 1>;
3420 defm VQRDMULHsl : N3VIntSL_HS<0b1101, IIC_VMULi16D, IIC_VMULi32D,
3421 IIC_VMULi16Q, IIC_VMULi32Q,
3422 "vqrdmulh", "s", int_arm_neon_vqrdmulh>;
3423 def : Pat<(v8i16 (int_arm_neon_vqrdmulh (v8i16 QPR:$src1),
3424 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
3426 (v8i16 (VQRDMULHslv8i16 (v8i16 QPR:$src1),
3427 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3428 (DSubReg_i16_reg imm:$lane))),
3429 (SubReg_i16_lane imm:$lane)))>;
3430 def : Pat<(v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$src1),
3431 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
3433 (v4i32 (VQRDMULHslv4i32 (v4i32 QPR:$src1),
3434 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3435 (DSubReg_i32_reg imm:$lane))),
3436 (SubReg_i32_lane imm:$lane)))>;
3438 // VMULL : Vector Multiply Long (integer and polynomial) (Q = D * D)
3439 defm VMULLs : N3VL_QHS<0,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
3440 "vmull", "s", NEONvmulls, 1>;
3441 defm VMULLu : N3VL_QHS<1,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
3442 "vmull", "u", NEONvmullu, 1>;
3443 def VMULLp : N3VLInt<0, 1, 0b00, 0b1110, 0, IIC_VMULi16D, "vmull", "p8",
3444 v8i16, v8i8, int_arm_neon_vmullp, 1>;
3445 defm VMULLsls : N3VLSL_HS<0, 0b1010, IIC_VMULi16D, "vmull", "s", NEONvmulls>;
3446 defm VMULLslu : N3VLSL_HS<1, 0b1010, IIC_VMULi16D, "vmull", "u", NEONvmullu>;
3448 // VQDMULL : Vector Saturating Doubling Multiply Long (Q = D * D)
3449 defm VQDMULL : N3VLInt_HS<0,1,0b1101,0, IIC_VMULi16D, IIC_VMULi32D,
3450 "vqdmull", "s", int_arm_neon_vqdmull, 1>;
3451 defm VQDMULLsl: N3VLIntSL_HS<0, 0b1011, IIC_VMULi16D,
3452 "vqdmull", "s", int_arm_neon_vqdmull>;
3454 // Vector Multiply-Accumulate and Multiply-Subtract Operations.
3456 // VMLA : Vector Multiply Accumulate (integer and floating-point)
3457 defm VMLA : N3VMulOp_QHS<0, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3458 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
3459 def VMLAfd : N3VDMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACD, "vmla", "f32",
3460 v2f32, fmul_su, fadd_mlx>,
3461 Requires<[HasNEON, UseFPVMLx]>;
3462 def VMLAfq : N3VQMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACQ, "vmla", "f32",
3463 v4f32, fmul_su, fadd_mlx>,
3464 Requires<[HasNEON, UseFPVMLx]>;
3465 defm VMLAsl : N3VMulOpSL_HS<0b0000, IIC_VMACi16D, IIC_VMACi32D,
3466 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
3467 def VMLAslfd : N3VDMulOpSL<0b10, 0b0001, IIC_VMACD, "vmla", "f32",
3468 v2f32, fmul_su, fadd_mlx>,
3469 Requires<[HasNEON, UseFPVMLx]>;
3470 def VMLAslfq : N3VQMulOpSL<0b10, 0b0001, IIC_VMACQ, "vmla", "f32",
3471 v4f32, v2f32, fmul_su, fadd_mlx>,
3472 Requires<[HasNEON, UseFPVMLx]>;
3474 def : Pat<(v8i16 (add (v8i16 QPR:$src1),
3475 (mul (v8i16 QPR:$src2),
3476 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
3477 (v8i16 (VMLAslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
3478 (v4i16 (EXTRACT_SUBREG QPR:$src3,
3479 (DSubReg_i16_reg imm:$lane))),
3480 (SubReg_i16_lane imm:$lane)))>;
3482 def : Pat<(v4i32 (add (v4i32 QPR:$src1),
3483 (mul (v4i32 QPR:$src2),
3484 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
3485 (v4i32 (VMLAslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
3486 (v2i32 (EXTRACT_SUBREG QPR:$src3,
3487 (DSubReg_i32_reg imm:$lane))),
3488 (SubReg_i32_lane imm:$lane)))>;
3490 def : Pat<(v4f32 (fadd_mlx (v4f32 QPR:$src1),
3491 (fmul_su (v4f32 QPR:$src2),
3492 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
3493 (v4f32 (VMLAslfq (v4f32 QPR:$src1),
3495 (v2f32 (EXTRACT_SUBREG QPR:$src3,
3496 (DSubReg_i32_reg imm:$lane))),
3497 (SubReg_i32_lane imm:$lane)))>,
3498 Requires<[HasNEON, UseFPVMLx]>;
3500 // VMLAL : Vector Multiply Accumulate Long (Q += D * D)
3501 defm VMLALs : N3VLMulOp_QHS<0,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
3502 "vmlal", "s", NEONvmulls, add>;
3503 defm VMLALu : N3VLMulOp_QHS<1,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
3504 "vmlal", "u", NEONvmullu, add>;
3506 defm VMLALsls : N3VLMulOpSL_HS<0, 0b0010, "vmlal", "s", NEONvmulls, add>;
3507 defm VMLALslu : N3VLMulOpSL_HS<1, 0b0010, "vmlal", "u", NEONvmullu, add>;
3509 // VQDMLAL : Vector Saturating Doubling Multiply Accumulate Long (Q += D * D)
3510 defm VQDMLAL : N3VLInt3_HS<0, 1, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3511 "vqdmlal", "s", int_arm_neon_vqdmlal>;
3512 defm VQDMLALsl: N3VLInt3SL_HS<0, 0b0011, "vqdmlal", "s", int_arm_neon_vqdmlal>;
3514 // VMLS : Vector Multiply Subtract (integer and floating-point)
3515 defm VMLS : N3VMulOp_QHS<1, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3516 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
3517 def VMLSfd : N3VDMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACD, "vmls", "f32",
3518 v2f32, fmul_su, fsub_mlx>,
3519 Requires<[HasNEON, UseFPVMLx]>;
3520 def VMLSfq : N3VQMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACQ, "vmls", "f32",
3521 v4f32, fmul_su, fsub_mlx>,
3522 Requires<[HasNEON, UseFPVMLx]>;
3523 defm VMLSsl : N3VMulOpSL_HS<0b0100, IIC_VMACi16D, IIC_VMACi32D,
3524 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
3525 def VMLSslfd : N3VDMulOpSL<0b10, 0b0101, IIC_VMACD, "vmls", "f32",
3526 v2f32, fmul_su, fsub_mlx>,
3527 Requires<[HasNEON, UseFPVMLx]>;
3528 def VMLSslfq : N3VQMulOpSL<0b10, 0b0101, IIC_VMACQ, "vmls", "f32",
3529 v4f32, v2f32, fmul_su, fsub_mlx>,
3530 Requires<[HasNEON, UseFPVMLx]>;
3532 def : Pat<(v8i16 (sub (v8i16 QPR:$src1),
3533 (mul (v8i16 QPR:$src2),
3534 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
3535 (v8i16 (VMLSslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
3536 (v4i16 (EXTRACT_SUBREG QPR:$src3,
3537 (DSubReg_i16_reg imm:$lane))),
3538 (SubReg_i16_lane imm:$lane)))>;
3540 def : Pat<(v4i32 (sub (v4i32 QPR:$src1),
3541 (mul (v4i32 QPR:$src2),
3542 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
3543 (v4i32 (VMLSslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
3544 (v2i32 (EXTRACT_SUBREG QPR:$src3,
3545 (DSubReg_i32_reg imm:$lane))),
3546 (SubReg_i32_lane imm:$lane)))>;
3548 def : Pat<(v4f32 (fsub_mlx (v4f32 QPR:$src1),
3549 (fmul_su (v4f32 QPR:$src2),
3550 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
3551 (v4f32 (VMLSslfq (v4f32 QPR:$src1), (v4f32 QPR:$src2),
3552 (v2f32 (EXTRACT_SUBREG QPR:$src3,
3553 (DSubReg_i32_reg imm:$lane))),
3554 (SubReg_i32_lane imm:$lane)))>,
3555 Requires<[HasNEON, UseFPVMLx]>;
3557 // VMLSL : Vector Multiply Subtract Long (Q -= D * D)
3558 defm VMLSLs : N3VLMulOp_QHS<0,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
3559 "vmlsl", "s", NEONvmulls, sub>;
3560 defm VMLSLu : N3VLMulOp_QHS<1,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
3561 "vmlsl", "u", NEONvmullu, sub>;
3563 defm VMLSLsls : N3VLMulOpSL_HS<0, 0b0110, "vmlsl", "s", NEONvmulls, sub>;
3564 defm VMLSLslu : N3VLMulOpSL_HS<1, 0b0110, "vmlsl", "u", NEONvmullu, sub>;
3566 // VQDMLSL : Vector Saturating Doubling Multiply Subtract Long (Q -= D * D)
3567 defm VQDMLSL : N3VLInt3_HS<0, 1, 0b1011, 0, IIC_VMACi16D, IIC_VMACi32D,
3568 "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
3569 defm VQDMLSLsl: N3VLInt3SL_HS<0, 0b111, "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
3571 // Vector Subtract Operations.
3573 // VSUB : Vector Subtract (integer and floating-point)
3574 defm VSUB : N3V_QHSD<1, 0, 0b1000, 0, IIC_VSUBiD, IIC_VSUBiQ,
3575 "vsub", "i", sub, 0>;
3576 def VSUBfd : N3VD<0, 0, 0b10, 0b1101, 0, IIC_VBIND, "vsub", "f32",
3577 v2f32, v2f32, fsub, 0>;
3578 def VSUBfq : N3VQ<0, 0, 0b10, 0b1101, 0, IIC_VBINQ, "vsub", "f32",
3579 v4f32, v4f32, fsub, 0>;
3580 // VSUBL : Vector Subtract Long (Q = D - D)
3581 defm VSUBLs : N3VLExt_QHS<0,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
3582 "vsubl", "s", sub, sext, 0>;
3583 defm VSUBLu : N3VLExt_QHS<1,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
3584 "vsubl", "u", sub, zext, 0>;
3585 // VSUBW : Vector Subtract Wide (Q = Q - D)
3586 defm VSUBWs : N3VW_QHS<0,1,0b0011,0, "vsubw", "s", sub, sext, 0>;
3587 defm VSUBWu : N3VW_QHS<1,1,0b0011,0, "vsubw", "u", sub, zext, 0>;
3588 // VHSUB : Vector Halving Subtract
3589 defm VHSUBs : N3VInt_QHS<0, 0, 0b0010, 0, N3RegFrm,
3590 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3591 "vhsub", "s", int_arm_neon_vhsubs, 0>;
3592 defm VHSUBu : N3VInt_QHS<1, 0, 0b0010, 0, N3RegFrm,
3593 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3594 "vhsub", "u", int_arm_neon_vhsubu, 0>;
3595 // VQSUB : Vector Saturing Subtract
3596 defm VQSUBs : N3VInt_QHSD<0, 0, 0b0010, 1, N3RegFrm,
3597 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3598 "vqsub", "s", int_arm_neon_vqsubs, 0>;
3599 defm VQSUBu : N3VInt_QHSD<1, 0, 0b0010, 1, N3RegFrm,
3600 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3601 "vqsub", "u", int_arm_neon_vqsubu, 0>;
3602 // VSUBHN : Vector Subtract and Narrow Returning High Half (D = Q - Q)
3603 defm VSUBHN : N3VNInt_HSD<0,1,0b0110,0, "vsubhn", "i",
3604 int_arm_neon_vsubhn, 0>;
3605 // VRSUBHN : Vector Rounding Subtract and Narrow Returning High Half (D=Q-Q)
3606 defm VRSUBHN : N3VNInt_HSD<1,1,0b0110,0, "vrsubhn", "i",
3607 int_arm_neon_vrsubhn, 0>;
3609 // Vector Comparisons.
3611 // VCEQ : Vector Compare Equal
3612 defm VCEQ : N3V_QHS<1, 0, 0b1000, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3613 IIC_VSUBi4Q, "vceq", "i", NEONvceq, 1>;
3614 def VCEQfd : N3VD<0,0,0b00,0b1110,0, IIC_VBIND, "vceq", "f32", v2i32, v2f32,
3616 def VCEQfq : N3VQ<0,0,0b00,0b1110,0, IIC_VBINQ, "vceq", "f32", v4i32, v4f32,
3619 defm VCEQz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00010, 0, "vceq", "i",
3620 "$Vd, $Vm, #0", NEONvceqz>;
3622 // VCGE : Vector Compare Greater Than or Equal
3623 defm VCGEs : N3V_QHS<0, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3624 IIC_VSUBi4Q, "vcge", "s", NEONvcge, 0>;
3625 defm VCGEu : N3V_QHS<1, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3626 IIC_VSUBi4Q, "vcge", "u", NEONvcgeu, 0>;
3627 def VCGEfd : N3VD<1,0,0b00,0b1110,0, IIC_VBIND, "vcge", "f32", v2i32, v2f32,
3629 def VCGEfq : N3VQ<1,0,0b00,0b1110,0, IIC_VBINQ, "vcge", "f32", v4i32, v4f32,
3632 defm VCGEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00001, 0, "vcge", "s",
3633 "$Vd, $Vm, #0", NEONvcgez>;
3634 defm VCLEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00011, 0, "vcle", "s",
3635 "$Vd, $Vm, #0", NEONvclez>;
3637 // VCGT : Vector Compare Greater Than
3638 defm VCGTs : N3V_QHS<0, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3639 IIC_VSUBi4Q, "vcgt", "s", NEONvcgt, 0>;
3640 defm VCGTu : N3V_QHS<1, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3641 IIC_VSUBi4Q, "vcgt", "u", NEONvcgtu, 0>;
3642 def VCGTfd : N3VD<1,0,0b10,0b1110,0, IIC_VBIND, "vcgt", "f32", v2i32, v2f32,
3644 def VCGTfq : N3VQ<1,0,0b10,0b1110,0, IIC_VBINQ, "vcgt", "f32", v4i32, v4f32,
3647 defm VCGTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00000, 0, "vcgt", "s",
3648 "$Vd, $Vm, #0", NEONvcgtz>;
3649 defm VCLTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00100, 0, "vclt", "s",
3650 "$Vd, $Vm, #0", NEONvcltz>;
3652 // VACGE : Vector Absolute Compare Greater Than or Equal (aka VCAGE)
3653 def VACGEd : N3VDInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacge",
3654 "f32", v2i32, v2f32, int_arm_neon_vacged, 0>;
3655 def VACGEq : N3VQInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacge",
3656 "f32", v4i32, v4f32, int_arm_neon_vacgeq, 0>;
3657 // VACGT : Vector Absolute Compare Greater Than (aka VCAGT)
3658 def VACGTd : N3VDInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacgt",
3659 "f32", v2i32, v2f32, int_arm_neon_vacgtd, 0>;
3660 def VACGTq : N3VQInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacgt",
3661 "f32", v4i32, v4f32, int_arm_neon_vacgtq, 0>;
3662 // VTST : Vector Test Bits
3663 defm VTST : N3V_QHS<0, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
3664 IIC_VBINi4Q, "vtst", "", NEONvtst, 1>;
3666 // Vector Bitwise Operations.
3668 def vnotd : PatFrag<(ops node:$in),
3669 (xor node:$in, (bitconvert (v8i8 NEONimmAllOnesV)))>;
3670 def vnotq : PatFrag<(ops node:$in),
3671 (xor node:$in, (bitconvert (v16i8 NEONimmAllOnesV)))>;
3674 // VAND : Vector Bitwise AND
3675 def VANDd : N3VDX<0, 0, 0b00, 0b0001, 1, IIC_VBINiD, "vand",
3676 v2i32, v2i32, and, 1>;
3677 def VANDq : N3VQX<0, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "vand",
3678 v4i32, v4i32, and, 1>;
3680 // VEOR : Vector Bitwise Exclusive OR
3681 def VEORd : N3VDX<1, 0, 0b00, 0b0001, 1, IIC_VBINiD, "veor",
3682 v2i32, v2i32, xor, 1>;
3683 def VEORq : N3VQX<1, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "veor",
3684 v4i32, v4i32, xor, 1>;
3686 // VORR : Vector Bitwise OR
3687 def VORRd : N3VDX<0, 0, 0b10, 0b0001, 1, IIC_VBINiD, "vorr",
3688 v2i32, v2i32, or, 1>;
3689 def VORRq : N3VQX<0, 0, 0b10, 0b0001, 1, IIC_VBINiQ, "vorr",
3690 v4i32, v4i32, or, 1>;
3692 def VORRiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 0, 1,
3693 (outs DPR:$Vd), (ins nModImm:$SIMM, DPR:$src),
3695 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
3697 (v4i16 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
3698 let Inst{9} = SIMM{9};
3701 def VORRiv2i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 0, 0, 1,
3702 (outs DPR:$Vd), (ins nModImm:$SIMM, DPR:$src),
3704 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
3706 (v2i32 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
3707 let Inst{10-9} = SIMM{10-9};
3710 def VORRiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 0, 1,
3711 (outs QPR:$Vd), (ins nModImm:$SIMM, QPR:$src),
3713 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
3715 (v8i16 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
3716 let Inst{9} = SIMM{9};
3719 def VORRiv4i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 1, 0, 1,
3720 (outs QPR:$Vd), (ins nModImm:$SIMM, QPR:$src),
3722 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
3724 (v4i32 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
3725 let Inst{10-9} = SIMM{10-9};
3729 // VBIC : Vector Bitwise Bit Clear (AND NOT)
3730 def VBICd : N3VX<0, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
3731 (ins DPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VBINiD,
3732 "vbic", "$Vd, $Vn, $Vm", "",
3733 [(set DPR:$Vd, (v2i32 (and DPR:$Vn,
3734 (vnotd DPR:$Vm))))]>;
3735 def VBICq : N3VX<0, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
3736 (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINiQ,
3737 "vbic", "$Vd, $Vn, $Vm", "",
3738 [(set QPR:$Vd, (v4i32 (and QPR:$Vn,
3739 (vnotq QPR:$Vm))))]>;
3741 def VBICiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 1, 1,
3742 (outs DPR:$Vd), (ins nModImm:$SIMM, DPR:$src),
3744 "vbic", "i16", "$Vd, $SIMM", "$src = $Vd",
3746 (v4i16 (NEONvbicImm DPR:$src, timm:$SIMM)))]> {
3747 let Inst{9} = SIMM{9};
3750 def VBICiv2i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 0, 1, 1,
3751 (outs DPR:$Vd), (ins nModImm:$SIMM, DPR:$src),
3753 "vbic", "i32", "$Vd, $SIMM", "$src = $Vd",
3755 (v2i32 (NEONvbicImm DPR:$src, timm:$SIMM)))]> {
3756 let Inst{10-9} = SIMM{10-9};
3759 def VBICiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 1, 1,
3760 (outs QPR:$Vd), (ins nModImm:$SIMM, QPR:$src),
3762 "vbic", "i16", "$Vd, $SIMM", "$src = $Vd",
3764 (v8i16 (NEONvbicImm QPR:$src, timm:$SIMM)))]> {
3765 let Inst{9} = SIMM{9};
3768 def VBICiv4i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 1, 1, 1,
3769 (outs QPR:$Vd), (ins nModImm:$SIMM, QPR:$src),
3771 "vbic", "i32", "$Vd, $SIMM", "$src = $Vd",
3773 (v4i32 (NEONvbicImm QPR:$src, timm:$SIMM)))]> {
3774 let Inst{10-9} = SIMM{10-9};
3777 // VORN : Vector Bitwise OR NOT
3778 def VORNd : N3VX<0, 0, 0b11, 0b0001, 0, 1, (outs DPR:$Vd),
3779 (ins DPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VBINiD,
3780 "vorn", "$Vd, $Vn, $Vm", "",
3781 [(set DPR:$Vd, (v2i32 (or DPR:$Vn,
3782 (vnotd DPR:$Vm))))]>;
3783 def VORNq : N3VX<0, 0, 0b11, 0b0001, 1, 1, (outs QPR:$Vd),
3784 (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINiQ,
3785 "vorn", "$Vd, $Vn, $Vm", "",
3786 [(set QPR:$Vd, (v4i32 (or QPR:$Vn,
3787 (vnotq QPR:$Vm))))]>;
3789 // VMVN : Vector Bitwise NOT (Immediate)
3791 let isReMaterializable = 1 in {
3793 def VMVNv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 1, 1, (outs DPR:$Vd),
3794 (ins nModImm:$SIMM), IIC_VMOVImm,
3795 "vmvn", "i16", "$Vd, $SIMM", "",
3796 [(set DPR:$Vd, (v4i16 (NEONvmvnImm timm:$SIMM)))]> {
3797 let Inst{9} = SIMM{9};
3800 def VMVNv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 1, 1, (outs QPR:$Vd),
3801 (ins nModImm:$SIMM), IIC_VMOVImm,
3802 "vmvn", "i16", "$Vd, $SIMM", "",
3803 [(set QPR:$Vd, (v8i16 (NEONvmvnImm timm:$SIMM)))]> {
3804 let Inst{9} = SIMM{9};
3807 def VMVNv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 1, 1, (outs DPR:$Vd),
3808 (ins nModImm:$SIMM), IIC_VMOVImm,
3809 "vmvn", "i32", "$Vd, $SIMM", "",
3810 [(set DPR:$Vd, (v2i32 (NEONvmvnImm timm:$SIMM)))]> {
3811 let Inst{11-8} = SIMM{11-8};
3814 def VMVNv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 1, 1, (outs QPR:$Vd),
3815 (ins nModImm:$SIMM), IIC_VMOVImm,
3816 "vmvn", "i32", "$Vd, $SIMM", "",
3817 [(set QPR:$Vd, (v4i32 (NEONvmvnImm timm:$SIMM)))]> {
3818 let Inst{11-8} = SIMM{11-8};
3822 // VMVN : Vector Bitwise NOT
3823 def VMVNd : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 0, 0,
3824 (outs DPR:$Vd), (ins DPR:$Vm), IIC_VSUBiD,
3825 "vmvn", "$Vd, $Vm", "",
3826 [(set DPR:$Vd, (v2i32 (vnotd DPR:$Vm)))]>;
3827 def VMVNq : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 1, 0,
3828 (outs QPR:$Vd), (ins QPR:$Vm), IIC_VSUBiD,
3829 "vmvn", "$Vd, $Vm", "",
3830 [(set QPR:$Vd, (v4i32 (vnotq QPR:$Vm)))]>;
3831 def : Pat<(v2i32 (vnotd DPR:$src)), (VMVNd DPR:$src)>;
3832 def : Pat<(v4i32 (vnotq QPR:$src)), (VMVNq QPR:$src)>;
3834 // VBSL : Vector Bitwise Select
3835 def VBSLd : N3VX<1, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
3836 (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
3837 N3RegFrm, IIC_VCNTiD,
3838 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3840 (v2i32 (NEONvbsl DPR:$src1, DPR:$Vn, DPR:$Vm)))]>;
3842 def : Pat<(v2i32 (or (and DPR:$Vn, DPR:$Vd),
3843 (and DPR:$Vm, (vnotd DPR:$Vd)))),
3844 (VBSLd DPR:$Vd, DPR:$Vn, DPR:$Vm)>;
3846 def VBSLq : N3VX<1, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
3847 (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
3848 N3RegFrm, IIC_VCNTiQ,
3849 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3851 (v4i32 (NEONvbsl QPR:$src1, QPR:$Vn, QPR:$Vm)))]>;
3853 def : Pat<(v4i32 (or (and QPR:$Vn, QPR:$Vd),
3854 (and QPR:$Vm, (vnotq QPR:$Vd)))),
3855 (VBSLq QPR:$Vd, QPR:$Vn, QPR:$Vm)>;
3857 // VBIF : Vector Bitwise Insert if False
3858 // like VBSL but with: "vbif $dst, $src3, $src1", "$src2 = $dst",
3859 // FIXME: This instruction's encoding MAY NOT BE correct.
3860 def VBIFd : N3VX<1, 0, 0b11, 0b0001, 0, 1,
3861 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
3862 N3RegFrm, IIC_VBINiD,
3863 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3864 [/* For disassembly only; pattern left blank */]>;
3865 def VBIFq : N3VX<1, 0, 0b11, 0b0001, 1, 1,
3866 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
3867 N3RegFrm, IIC_VBINiQ,
3868 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3869 [/* For disassembly only; pattern left blank */]>;
3871 // VBIT : Vector Bitwise Insert if True
3872 // like VBSL but with: "vbit $dst, $src2, $src1", "$src3 = $dst",
3873 // FIXME: This instruction's encoding MAY NOT BE correct.
3874 def VBITd : N3VX<1, 0, 0b10, 0b0001, 0, 1,
3875 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
3876 N3RegFrm, IIC_VBINiD,
3877 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3878 [/* For disassembly only; pattern left blank */]>;
3879 def VBITq : N3VX<1, 0, 0b10, 0b0001, 1, 1,
3880 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
3881 N3RegFrm, IIC_VBINiQ,
3882 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3883 [/* For disassembly only; pattern left blank */]>;
3885 // VBIT/VBIF are not yet implemented. The TwoAddress pass will not go looking
3886 // for equivalent operations with different register constraints; it just
3889 // Vector Absolute Differences.
3891 // VABD : Vector Absolute Difference
3892 defm VABDs : N3VInt_QHS<0, 0, 0b0111, 0, N3RegFrm,
3893 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3894 "vabd", "s", int_arm_neon_vabds, 1>;
3895 defm VABDu : N3VInt_QHS<1, 0, 0b0111, 0, N3RegFrm,
3896 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3897 "vabd", "u", int_arm_neon_vabdu, 1>;
3898 def VABDfd : N3VDInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBIND,
3899 "vabd", "f32", v2f32, v2f32, int_arm_neon_vabds, 1>;
3900 def VABDfq : N3VQInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBINQ,
3901 "vabd", "f32", v4f32, v4f32, int_arm_neon_vabds, 1>;
3903 // VABDL : Vector Absolute Difference Long (Q = | D - D |)
3904 defm VABDLs : N3VLIntExt_QHS<0,1,0b0111,0, IIC_VSUBi4Q,
3905 "vabdl", "s", int_arm_neon_vabds, zext, 1>;
3906 defm VABDLu : N3VLIntExt_QHS<1,1,0b0111,0, IIC_VSUBi4Q,
3907 "vabdl", "u", int_arm_neon_vabdu, zext, 1>;
3909 // VABA : Vector Absolute Difference and Accumulate
3910 defm VABAs : N3VIntOp_QHS<0,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
3911 "vaba", "s", int_arm_neon_vabds, add>;
3912 defm VABAu : N3VIntOp_QHS<1,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
3913 "vaba", "u", int_arm_neon_vabdu, add>;
3915 // VABAL : Vector Absolute Difference and Accumulate Long (Q += | D - D |)
3916 defm VABALs : N3VLIntExtOp_QHS<0,1,0b0101,0, IIC_VABAD,
3917 "vabal", "s", int_arm_neon_vabds, zext, add>;
3918 defm VABALu : N3VLIntExtOp_QHS<1,1,0b0101,0, IIC_VABAD,
3919 "vabal", "u", int_arm_neon_vabdu, zext, add>;
3921 // Vector Maximum and Minimum.
3923 // VMAX : Vector Maximum
3924 defm VMAXs : N3VInt_QHS<0, 0, 0b0110, 0, N3RegFrm,
3925 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3926 "vmax", "s", int_arm_neon_vmaxs, 1>;
3927 defm VMAXu : N3VInt_QHS<1, 0, 0b0110, 0, N3RegFrm,
3928 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3929 "vmax", "u", int_arm_neon_vmaxu, 1>;
3930 def VMAXfd : N3VDInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBIND,
3932 v2f32, v2f32, int_arm_neon_vmaxs, 1>;
3933 def VMAXfq : N3VQInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBINQ,
3935 v4f32, v4f32, int_arm_neon_vmaxs, 1>;
3937 // VMIN : Vector Minimum
3938 defm VMINs : N3VInt_QHS<0, 0, 0b0110, 1, N3RegFrm,
3939 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3940 "vmin", "s", int_arm_neon_vmins, 1>;
3941 defm VMINu : N3VInt_QHS<1, 0, 0b0110, 1, N3RegFrm,
3942 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3943 "vmin", "u", int_arm_neon_vminu, 1>;
3944 def VMINfd : N3VDInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBIND,
3946 v2f32, v2f32, int_arm_neon_vmins, 1>;
3947 def VMINfq : N3VQInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBINQ,
3949 v4f32, v4f32, int_arm_neon_vmins, 1>;
3951 // Vector Pairwise Operations.
3953 // VPADD : Vector Pairwise Add
3954 def VPADDi8 : N3VDInt<0, 0, 0b00, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
3956 v8i8, v8i8, int_arm_neon_vpadd, 0>;
3957 def VPADDi16 : N3VDInt<0, 0, 0b01, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
3959 v4i16, v4i16, int_arm_neon_vpadd, 0>;
3960 def VPADDi32 : N3VDInt<0, 0, 0b10, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
3962 v2i32, v2i32, int_arm_neon_vpadd, 0>;
3963 def VPADDf : N3VDInt<1, 0, 0b00, 0b1101, 0, N3RegFrm,
3964 IIC_VPBIND, "vpadd", "f32",
3965 v2f32, v2f32, int_arm_neon_vpadd, 0>;
3967 // VPADDL : Vector Pairwise Add Long
3968 defm VPADDLs : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00100, 0, "vpaddl", "s",
3969 int_arm_neon_vpaddls>;
3970 defm VPADDLu : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00101, 0, "vpaddl", "u",
3971 int_arm_neon_vpaddlu>;
3973 // VPADAL : Vector Pairwise Add and Accumulate Long
3974 defm VPADALs : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01100, 0, "vpadal", "s",
3975 int_arm_neon_vpadals>;
3976 defm VPADALu : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01101, 0, "vpadal", "u",
3977 int_arm_neon_vpadalu>;
3979 // VPMAX : Vector Pairwise Maximum
3980 def VPMAXs8 : N3VDInt<0, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3981 "s8", v8i8, v8i8, int_arm_neon_vpmaxs, 0>;
3982 def VPMAXs16 : N3VDInt<0, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3983 "s16", v4i16, v4i16, int_arm_neon_vpmaxs, 0>;
3984 def VPMAXs32 : N3VDInt<0, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3985 "s32", v2i32, v2i32, int_arm_neon_vpmaxs, 0>;
3986 def VPMAXu8 : N3VDInt<1, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3987 "u8", v8i8, v8i8, int_arm_neon_vpmaxu, 0>;
3988 def VPMAXu16 : N3VDInt<1, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3989 "u16", v4i16, v4i16, int_arm_neon_vpmaxu, 0>;
3990 def VPMAXu32 : N3VDInt<1, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3991 "u32", v2i32, v2i32, int_arm_neon_vpmaxu, 0>;
3992 def VPMAXf : N3VDInt<1, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmax",
3993 "f32", v2f32, v2f32, int_arm_neon_vpmaxs, 0>;
3995 // VPMIN : Vector Pairwise Minimum
3996 def VPMINs8 : N3VDInt<0, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3997 "s8", v8i8, v8i8, int_arm_neon_vpmins, 0>;
3998 def VPMINs16 : N3VDInt<0, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3999 "s16", v4i16, v4i16, int_arm_neon_vpmins, 0>;
4000 def VPMINs32 : N3VDInt<0, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4001 "s32", v2i32, v2i32, int_arm_neon_vpmins, 0>;
4002 def VPMINu8 : N3VDInt<1, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4003 "u8", v8i8, v8i8, int_arm_neon_vpminu, 0>;
4004 def VPMINu16 : N3VDInt<1, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4005 "u16", v4i16, v4i16, int_arm_neon_vpminu, 0>;
4006 def VPMINu32 : N3VDInt<1, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4007 "u32", v2i32, v2i32, int_arm_neon_vpminu, 0>;
4008 def VPMINf : N3VDInt<1, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmin",
4009 "f32", v2f32, v2f32, int_arm_neon_vpmins, 0>;
4011 // Vector Reciprocal and Reciprocal Square Root Estimate and Step.
4013 // VRECPE : Vector Reciprocal Estimate
4014 def VRECPEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
4015 IIC_VUNAD, "vrecpe", "u32",
4016 v2i32, v2i32, int_arm_neon_vrecpe>;
4017 def VRECPEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
4018 IIC_VUNAQ, "vrecpe", "u32",
4019 v4i32, v4i32, int_arm_neon_vrecpe>;
4020 def VRECPEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
4021 IIC_VUNAD, "vrecpe", "f32",
4022 v2f32, v2f32, int_arm_neon_vrecpe>;
4023 def VRECPEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
4024 IIC_VUNAQ, "vrecpe", "f32",
4025 v4f32, v4f32, int_arm_neon_vrecpe>;
4027 // VRECPS : Vector Reciprocal Step
4028 def VRECPSfd : N3VDInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
4029 IIC_VRECSD, "vrecps", "f32",
4030 v2f32, v2f32, int_arm_neon_vrecps, 1>;
4031 def VRECPSfq : N3VQInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
4032 IIC_VRECSQ, "vrecps", "f32",
4033 v4f32, v4f32, int_arm_neon_vrecps, 1>;
4035 // VRSQRTE : Vector Reciprocal Square Root Estimate
4036 def VRSQRTEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
4037 IIC_VUNAD, "vrsqrte", "u32",
4038 v2i32, v2i32, int_arm_neon_vrsqrte>;
4039 def VRSQRTEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
4040 IIC_VUNAQ, "vrsqrte", "u32",
4041 v4i32, v4i32, int_arm_neon_vrsqrte>;
4042 def VRSQRTEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
4043 IIC_VUNAD, "vrsqrte", "f32",
4044 v2f32, v2f32, int_arm_neon_vrsqrte>;
4045 def VRSQRTEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
4046 IIC_VUNAQ, "vrsqrte", "f32",
4047 v4f32, v4f32, int_arm_neon_vrsqrte>;
4049 // VRSQRTS : Vector Reciprocal Square Root Step
4050 def VRSQRTSfd : N3VDInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
4051 IIC_VRECSD, "vrsqrts", "f32",
4052 v2f32, v2f32, int_arm_neon_vrsqrts, 1>;
4053 def VRSQRTSfq : N3VQInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
4054 IIC_VRECSQ, "vrsqrts", "f32",
4055 v4f32, v4f32, int_arm_neon_vrsqrts, 1>;
4059 // VSHL : Vector Shift
4060 defm VSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 0, N3RegVShFrm,
4061 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
4062 "vshl", "s", int_arm_neon_vshifts>;
4063 defm VSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 0, N3RegVShFrm,
4064 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
4065 "vshl", "u", int_arm_neon_vshiftu>;
4067 // VSHL : Vector Shift Left (Immediate)
4068 defm VSHLi : N2VShL_QHSD<0, 1, 0b0101, 1, IIC_VSHLiD, "vshl", "i", NEONvshl>;
4070 // VSHR : Vector Shift Right (Immediate)
4071 defm VSHRs : N2VShR_QHSD<0, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "s",NEONvshrs>;
4072 defm VSHRu : N2VShR_QHSD<1, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "u",NEONvshru>;
4074 // VSHLL : Vector Shift Left Long
4075 defm VSHLLs : N2VLSh_QHS<0, 1, 0b1010, 0, 0, 1, "vshll", "s", NEONvshlls>;
4076 defm VSHLLu : N2VLSh_QHS<1, 1, 0b1010, 0, 0, 1, "vshll", "u", NEONvshllu>;
4078 // VSHLL : Vector Shift Left Long (with maximum shift count)
4079 class N2VLShMax<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
4080 bit op6, bit op4, string OpcodeStr, string Dt, ValueType ResTy,
4081 ValueType OpTy, SDNode OpNode>
4082 : N2VLSh<op24, op23, op11_8, op7, op6, op4, OpcodeStr, Dt,
4083 ResTy, OpTy, OpNode> {
4084 let Inst{21-16} = op21_16;
4085 let DecoderMethod = "DecodeVSHLMaxInstruction";
4087 def VSHLLi8 : N2VLShMax<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll", "i8",
4088 v8i16, v8i8, NEONvshlli>;
4089 def VSHLLi16 : N2VLShMax<1, 1, 0b110110, 0b0011, 0, 0, 0, "vshll", "i16",
4090 v4i32, v4i16, NEONvshlli>;
4091 def VSHLLi32 : N2VLShMax<1, 1, 0b111010, 0b0011, 0, 0, 0, "vshll", "i32",
4092 v2i64, v2i32, NEONvshlli>;
4094 // VSHRN : Vector Shift Right and Narrow
4095 defm VSHRN : N2VNSh_HSD<0,1,0b1000,0,0,1, IIC_VSHLiD, "vshrn", "i",
4098 // VRSHL : Vector Rounding Shift
4099 defm VRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 0, N3RegVShFrm,
4100 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4101 "vrshl", "s", int_arm_neon_vrshifts>;
4102 defm VRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 0, N3RegVShFrm,
4103 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4104 "vrshl", "u", int_arm_neon_vrshiftu>;
4105 // VRSHR : Vector Rounding Shift Right
4106 defm VRSHRs : N2VShR_QHSD<0,1,0b0010,1, IIC_VSHLi4D, "vrshr", "s",NEONvrshrs>;
4107 defm VRSHRu : N2VShR_QHSD<1,1,0b0010,1, IIC_VSHLi4D, "vrshr", "u",NEONvrshru>;
4109 // VRSHRN : Vector Rounding Shift Right and Narrow
4110 defm VRSHRN : N2VNSh_HSD<0, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vrshrn", "i",
4113 // VQSHL : Vector Saturating Shift
4114 defm VQSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 1, N3RegVShFrm,
4115 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4116 "vqshl", "s", int_arm_neon_vqshifts>;
4117 defm VQSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 1, N3RegVShFrm,
4118 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4119 "vqshl", "u", int_arm_neon_vqshiftu>;
4120 // VQSHL : Vector Saturating Shift Left (Immediate)
4121 defm VQSHLsi : N2VShL_QHSD<0,1,0b0111,1, IIC_VSHLi4D, "vqshl", "s",NEONvqshls>;
4122 defm VQSHLui : N2VShL_QHSD<1,1,0b0111,1, IIC_VSHLi4D, "vqshl", "u",NEONvqshlu>;
4124 // VQSHLU : Vector Saturating Shift Left (Immediate, Unsigned)
4125 defm VQSHLsu : N2VShL_QHSD<1,1,0b0110,1, IIC_VSHLi4D,"vqshlu","s",NEONvqshlsu>;
4127 // VQSHRN : Vector Saturating Shift Right and Narrow
4128 defm VQSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "s",
4130 defm VQSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "u",
4133 // VQSHRUN : Vector Saturating Shift Right and Narrow (Unsigned)
4134 defm VQSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 0, 1, IIC_VSHLi4D, "vqshrun", "s",
4137 // VQRSHL : Vector Saturating Rounding Shift
4138 defm VQRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 1, N3RegVShFrm,
4139 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4140 "vqrshl", "s", int_arm_neon_vqrshifts>;
4141 defm VQRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 1, N3RegVShFrm,
4142 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4143 "vqrshl", "u", int_arm_neon_vqrshiftu>;
4145 // VQRSHRN : Vector Saturating Rounding Shift Right and Narrow
4146 defm VQRSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "s",
4148 defm VQRSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "u",
4151 // VQRSHRUN : Vector Saturating Rounding Shift Right and Narrow (Unsigned)
4152 defm VQRSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vqrshrun", "s",
4155 // VSRA : Vector Shift Right and Accumulate
4156 defm VSRAs : N2VShAdd_QHSD<0, 1, 0b0001, 1, "vsra", "s", NEONvshrs>;
4157 defm VSRAu : N2VShAdd_QHSD<1, 1, 0b0001, 1, "vsra", "u", NEONvshru>;
4158 // VRSRA : Vector Rounding Shift Right and Accumulate
4159 defm VRSRAs : N2VShAdd_QHSD<0, 1, 0b0011, 1, "vrsra", "s", NEONvrshrs>;
4160 defm VRSRAu : N2VShAdd_QHSD<1, 1, 0b0011, 1, "vrsra", "u", NEONvrshru>;
4162 // VSLI : Vector Shift Left and Insert
4163 defm VSLI : N2VShInsL_QHSD<1, 1, 0b0101, 1, "vsli">;
4165 // VSRI : Vector Shift Right and Insert
4166 defm VSRI : N2VShInsR_QHSD<1, 1, 0b0100, 1, "vsri">;
4168 // Vector Absolute and Saturating Absolute.
4170 // VABS : Vector Absolute Value
4171 defm VABS : N2VInt_QHS<0b11, 0b11, 0b01, 0b00110, 0,
4172 IIC_VUNAiD, IIC_VUNAiQ, "vabs", "s",
4174 def VABSfd : N2VDInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
4175 IIC_VUNAD, "vabs", "f32",
4176 v2f32, v2f32, int_arm_neon_vabs>;
4177 def VABSfq : N2VQInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
4178 IIC_VUNAQ, "vabs", "f32",
4179 v4f32, v4f32, int_arm_neon_vabs>;
4181 // VQABS : Vector Saturating Absolute Value
4182 defm VQABS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01110, 0,
4183 IIC_VQUNAiD, IIC_VQUNAiQ, "vqabs", "s",
4184 int_arm_neon_vqabs>;
4188 def vnegd : PatFrag<(ops node:$in),
4189 (sub (bitconvert (v2i32 NEONimmAllZerosV)), node:$in)>;
4190 def vnegq : PatFrag<(ops node:$in),
4191 (sub (bitconvert (v4i32 NEONimmAllZerosV)), node:$in)>;
4193 class VNEGD<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
4194 : N2V<0b11, 0b11, size, 0b01, 0b00111, 0, 0, (outs DPR:$Vd), (ins DPR:$Vm),
4195 IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
4196 [(set DPR:$Vd, (Ty (vnegd DPR:$Vm)))]>;
4197 class VNEGQ<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
4198 : N2V<0b11, 0b11, size, 0b01, 0b00111, 1, 0, (outs QPR:$Vd), (ins QPR:$Vm),
4199 IIC_VSHLiQ, OpcodeStr, Dt, "$Vd, $Vm", "",
4200 [(set QPR:$Vd, (Ty (vnegq QPR:$Vm)))]>;
4202 // VNEG : Vector Negate (integer)
4203 def VNEGs8d : VNEGD<0b00, "vneg", "s8", v8i8>;
4204 def VNEGs16d : VNEGD<0b01, "vneg", "s16", v4i16>;
4205 def VNEGs32d : VNEGD<0b10, "vneg", "s32", v2i32>;
4206 def VNEGs8q : VNEGQ<0b00, "vneg", "s8", v16i8>;
4207 def VNEGs16q : VNEGQ<0b01, "vneg", "s16", v8i16>;
4208 def VNEGs32q : VNEGQ<0b10, "vneg", "s32", v4i32>;
4210 // VNEG : Vector Negate (floating-point)
4211 def VNEGfd : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
4212 (outs DPR:$Vd), (ins DPR:$Vm), IIC_VUNAD,
4213 "vneg", "f32", "$Vd, $Vm", "",
4214 [(set DPR:$Vd, (v2f32 (fneg DPR:$Vm)))]>;
4215 def VNEGf32q : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 1, 0,
4216 (outs QPR:$Vd), (ins QPR:$Vm), IIC_VUNAQ,
4217 "vneg", "f32", "$Vd, $Vm", "",
4218 [(set QPR:$Vd, (v4f32 (fneg QPR:$Vm)))]>;
4220 def : Pat<(v8i8 (vnegd DPR:$src)), (VNEGs8d DPR:$src)>;
4221 def : Pat<(v4i16 (vnegd DPR:$src)), (VNEGs16d DPR:$src)>;
4222 def : Pat<(v2i32 (vnegd DPR:$src)), (VNEGs32d DPR:$src)>;
4223 def : Pat<(v16i8 (vnegq QPR:$src)), (VNEGs8q QPR:$src)>;
4224 def : Pat<(v8i16 (vnegq QPR:$src)), (VNEGs16q QPR:$src)>;
4225 def : Pat<(v4i32 (vnegq QPR:$src)), (VNEGs32q QPR:$src)>;
4227 // VQNEG : Vector Saturating Negate
4228 defm VQNEG : N2VInt_QHS<0b11, 0b11, 0b00, 0b01111, 0,
4229 IIC_VQUNAiD, IIC_VQUNAiQ, "vqneg", "s",
4230 int_arm_neon_vqneg>;
4232 // Vector Bit Counting Operations.
4234 // VCLS : Vector Count Leading Sign Bits
4235 defm VCLS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01000, 0,
4236 IIC_VCNTiD, IIC_VCNTiQ, "vcls", "s",
4238 // VCLZ : Vector Count Leading Zeros
4239 defm VCLZ : N2VInt_QHS<0b11, 0b11, 0b00, 0b01001, 0,
4240 IIC_VCNTiD, IIC_VCNTiQ, "vclz", "i",
4242 // VCNT : Vector Count One Bits
4243 def VCNTd : N2VDInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
4244 IIC_VCNTiD, "vcnt", "8",
4245 v8i8, v8i8, int_arm_neon_vcnt>;
4246 def VCNTq : N2VQInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
4247 IIC_VCNTiQ, "vcnt", "8",
4248 v16i8, v16i8, int_arm_neon_vcnt>;
4250 // Vector Swap -- for disassembly only.
4251 def VSWPd : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 0, 0,
4252 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
4253 "vswp", "$Vd, $Vm", "", []>;
4254 def VSWPq : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 1, 0,
4255 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
4256 "vswp", "$Vd, $Vm", "", []>;
4258 // Vector Move Operations.
4260 // VMOV : Vector Move (Register)
4261 def : InstAlias<"vmov${p} $Vd, $Vm",
4262 (VORRd DPR:$Vd, DPR:$Vm, DPR:$Vm, pred:$p)>;
4263 def : InstAlias<"vmov${p} $Vd, $Vm",
4264 (VORRq QPR:$Vd, QPR:$Vm, QPR:$Vm, pred:$p)>;
4266 let neverHasSideEffects = 1 in {
4267 // Pseudo vector move instructions for QQ and QQQQ registers. This should
4268 // be expanded after register allocation is completed.
4269 def VMOVQQ : PseudoInst<(outs QQPR:$dst), (ins QQPR:$src),
4272 def VMOVQQQQ : PseudoInst<(outs QQQQPR:$dst), (ins QQQQPR:$src),
4274 } // neverHasSideEffects
4276 // VMOV : Vector Move (Immediate)
4278 let isReMaterializable = 1 in {
4279 def VMOVv8i8 : N1ModImm<1, 0b000, 0b1110, 0, 0, 0, 1, (outs DPR:$Vd),
4280 (ins nModImm:$SIMM), IIC_VMOVImm,
4281 "vmov", "i8", "$Vd, $SIMM", "",
4282 [(set DPR:$Vd, (v8i8 (NEONvmovImm timm:$SIMM)))]>;
4283 def VMOVv16i8 : N1ModImm<1, 0b000, 0b1110, 0, 1, 0, 1, (outs QPR:$Vd),
4284 (ins nModImm:$SIMM), IIC_VMOVImm,
4285 "vmov", "i8", "$Vd, $SIMM", "",
4286 [(set QPR:$Vd, (v16i8 (NEONvmovImm timm:$SIMM)))]>;
4288 def VMOVv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 0, 1, (outs DPR:$Vd),
4289 (ins nModImm:$SIMM), IIC_VMOVImm,
4290 "vmov", "i16", "$Vd, $SIMM", "",
4291 [(set DPR:$Vd, (v4i16 (NEONvmovImm timm:$SIMM)))]> {
4292 let Inst{9} = SIMM{9};
4295 def VMOVv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 0, 1, (outs QPR:$Vd),
4296 (ins nModImm:$SIMM), IIC_VMOVImm,
4297 "vmov", "i16", "$Vd, $SIMM", "",
4298 [(set QPR:$Vd, (v8i16 (NEONvmovImm timm:$SIMM)))]> {
4299 let Inst{9} = SIMM{9};
4302 def VMOVv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 0, 1, (outs DPR:$Vd),
4303 (ins nModImm:$SIMM), IIC_VMOVImm,
4304 "vmov", "i32", "$Vd, $SIMM", "",
4305 [(set DPR:$Vd, (v2i32 (NEONvmovImm timm:$SIMM)))]> {
4306 let Inst{11-8} = SIMM{11-8};
4309 def VMOVv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 0, 1, (outs QPR:$Vd),
4310 (ins nModImm:$SIMM), IIC_VMOVImm,
4311 "vmov", "i32", "$Vd, $SIMM", "",
4312 [(set QPR:$Vd, (v4i32 (NEONvmovImm timm:$SIMM)))]> {
4313 let Inst{11-8} = SIMM{11-8};
4316 def VMOVv1i64 : N1ModImm<1, 0b000, 0b1110, 0, 0, 1, 1, (outs DPR:$Vd),
4317 (ins nModImm:$SIMM), IIC_VMOVImm,
4318 "vmov", "i64", "$Vd, $SIMM", "",
4319 [(set DPR:$Vd, (v1i64 (NEONvmovImm timm:$SIMM)))]>;
4320 def VMOVv2i64 : N1ModImm<1, 0b000, 0b1110, 0, 1, 1, 1, (outs QPR:$Vd),
4321 (ins nModImm:$SIMM), IIC_VMOVImm,
4322 "vmov", "i64", "$Vd, $SIMM", "",
4323 [(set QPR:$Vd, (v2i64 (NEONvmovImm timm:$SIMM)))]>;
4324 } // isReMaterializable
4326 // VMOV : Vector Get Lane (move scalar to ARM core register)
4328 def VGETLNs8 : NVGetLane<{1,1,1,0,0,1,?,1}, 0b1011, {?,?},
4329 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
4330 IIC_VMOVSI, "vmov", "s8", "$R, $V[$lane]",
4331 [(set GPR:$R, (NEONvgetlanes (v8i8 DPR:$V),
4333 let Inst{21} = lane{2};
4334 let Inst{6-5} = lane{1-0};
4336 def VGETLNs16 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, {?,1},
4337 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
4338 IIC_VMOVSI, "vmov", "s16", "$R, $V[$lane]",
4339 [(set GPR:$R, (NEONvgetlanes (v4i16 DPR:$V),
4341 let Inst{21} = lane{1};
4342 let Inst{6} = lane{0};
4344 def VGETLNu8 : NVGetLane<{1,1,1,0,1,1,?,1}, 0b1011, {?,?},
4345 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
4346 IIC_VMOVSI, "vmov", "u8", "$R, $V[$lane]",
4347 [(set GPR:$R, (NEONvgetlaneu (v8i8 DPR:$V),
4349 let Inst{21} = lane{2};
4350 let Inst{6-5} = lane{1-0};
4352 def VGETLNu16 : NVGetLane<{1,1,1,0,1,0,?,1}, 0b1011, {?,1},
4353 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
4354 IIC_VMOVSI, "vmov", "u16", "$R, $V[$lane]",
4355 [(set GPR:$R, (NEONvgetlaneu (v4i16 DPR:$V),
4357 let Inst{21} = lane{1};
4358 let Inst{6} = lane{0};
4360 def VGETLNi32 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, 0b00,
4361 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
4362 IIC_VMOVSI, "vmov", "32", "$R, $V[$lane]",
4363 [(set GPR:$R, (extractelt (v2i32 DPR:$V),
4365 let Inst{21} = lane{0};
4367 // def VGETLNf32: see FMRDH and FMRDL in ARMInstrVFP.td
4368 def : Pat<(NEONvgetlanes (v16i8 QPR:$src), imm:$lane),
4369 (VGETLNs8 (v8i8 (EXTRACT_SUBREG QPR:$src,
4370 (DSubReg_i8_reg imm:$lane))),
4371 (SubReg_i8_lane imm:$lane))>;
4372 def : Pat<(NEONvgetlanes (v8i16 QPR:$src), imm:$lane),
4373 (VGETLNs16 (v4i16 (EXTRACT_SUBREG QPR:$src,
4374 (DSubReg_i16_reg imm:$lane))),
4375 (SubReg_i16_lane imm:$lane))>;
4376 def : Pat<(NEONvgetlaneu (v16i8 QPR:$src), imm:$lane),
4377 (VGETLNu8 (v8i8 (EXTRACT_SUBREG QPR:$src,
4378 (DSubReg_i8_reg imm:$lane))),
4379 (SubReg_i8_lane imm:$lane))>;
4380 def : Pat<(NEONvgetlaneu (v8i16 QPR:$src), imm:$lane),
4381 (VGETLNu16 (v4i16 (EXTRACT_SUBREG QPR:$src,
4382 (DSubReg_i16_reg imm:$lane))),
4383 (SubReg_i16_lane imm:$lane))>;
4384 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
4385 (VGETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src,
4386 (DSubReg_i32_reg imm:$lane))),
4387 (SubReg_i32_lane imm:$lane))>;
4388 def : Pat<(extractelt (v2f32 DPR:$src1), imm:$src2),
4389 (EXTRACT_SUBREG (v2f32 (COPY_TO_REGCLASS (v2f32 DPR:$src1),DPR_VFP2)),
4390 (SSubReg_f32_reg imm:$src2))>;
4391 def : Pat<(extractelt (v4f32 QPR:$src1), imm:$src2),
4392 (EXTRACT_SUBREG (v4f32 (COPY_TO_REGCLASS (v4f32 QPR:$src1),QPR_VFP2)),
4393 (SSubReg_f32_reg imm:$src2))>;
4394 //def : Pat<(extractelt (v2i64 QPR:$src1), imm:$src2),
4395 // (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
4396 def : Pat<(extractelt (v2f64 QPR:$src1), imm:$src2),
4397 (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
4400 // VMOV : Vector Set Lane (move ARM core register to scalar)
4402 let Constraints = "$src1 = $V" in {
4403 def VSETLNi8 : NVSetLane<{1,1,1,0,0,1,?,0}, 0b1011, {?,?}, (outs DPR:$V),
4404 (ins DPR:$src1, GPR:$R, nohash_imm:$lane),
4405 IIC_VMOVISL, "vmov", "8", "$V[$lane], $R",
4406 [(set DPR:$V, (vector_insert (v8i8 DPR:$src1),
4407 GPR:$R, imm:$lane))]> {
4408 let Inst{21} = lane{2};
4409 let Inst{6-5} = lane{1-0};
4411 def VSETLNi16 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, {?,1}, (outs DPR:$V),
4412 (ins DPR:$src1, GPR:$R, nohash_imm:$lane),
4413 IIC_VMOVISL, "vmov", "16", "$V[$lane], $R",
4414 [(set DPR:$V, (vector_insert (v4i16 DPR:$src1),
4415 GPR:$R, imm:$lane))]> {
4416 let Inst{21} = lane{1};
4417 let Inst{6} = lane{0};
4419 def VSETLNi32 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, 0b00, (outs DPR:$V),
4420 (ins DPR:$src1, GPR:$R, nohash_imm:$lane),
4421 IIC_VMOVISL, "vmov", "32", "$V[$lane], $R",
4422 [(set DPR:$V, (insertelt (v2i32 DPR:$src1),
4423 GPR:$R, imm:$lane))]> {
4424 let Inst{21} = lane{0};
4427 def : Pat<(vector_insert (v16i8 QPR:$src1), GPR:$src2, imm:$lane),
4428 (v16i8 (INSERT_SUBREG QPR:$src1,
4429 (v8i8 (VSETLNi8 (v8i8 (EXTRACT_SUBREG QPR:$src1,
4430 (DSubReg_i8_reg imm:$lane))),
4431 GPR:$src2, (SubReg_i8_lane imm:$lane))),
4432 (DSubReg_i8_reg imm:$lane)))>;
4433 def : Pat<(vector_insert (v8i16 QPR:$src1), GPR:$src2, imm:$lane),
4434 (v8i16 (INSERT_SUBREG QPR:$src1,
4435 (v4i16 (VSETLNi16 (v4i16 (EXTRACT_SUBREG QPR:$src1,
4436 (DSubReg_i16_reg imm:$lane))),
4437 GPR:$src2, (SubReg_i16_lane imm:$lane))),
4438 (DSubReg_i16_reg imm:$lane)))>;
4439 def : Pat<(insertelt (v4i32 QPR:$src1), GPR:$src2, imm:$lane),
4440 (v4i32 (INSERT_SUBREG QPR:$src1,
4441 (v2i32 (VSETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src1,
4442 (DSubReg_i32_reg imm:$lane))),
4443 GPR:$src2, (SubReg_i32_lane imm:$lane))),
4444 (DSubReg_i32_reg imm:$lane)))>;
4446 def : Pat<(v2f32 (insertelt DPR:$src1, SPR:$src2, imm:$src3)),
4447 (INSERT_SUBREG (v2f32 (COPY_TO_REGCLASS DPR:$src1, DPR_VFP2)),
4448 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
4449 def : Pat<(v4f32 (insertelt QPR:$src1, SPR:$src2, imm:$src3)),
4450 (INSERT_SUBREG (v4f32 (COPY_TO_REGCLASS QPR:$src1, QPR_VFP2)),
4451 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
4453 //def : Pat<(v2i64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
4454 // (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
4455 def : Pat<(v2f64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
4456 (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
4458 def : Pat<(v2f32 (scalar_to_vector SPR:$src)),
4459 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
4460 def : Pat<(v2f64 (scalar_to_vector (f64 DPR:$src))),
4461 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
4462 def : Pat<(v4f32 (scalar_to_vector SPR:$src)),
4463 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
4465 def : Pat<(v8i8 (scalar_to_vector GPR:$src)),
4466 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4467 def : Pat<(v4i16 (scalar_to_vector GPR:$src)),
4468 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4469 def : Pat<(v2i32 (scalar_to_vector GPR:$src)),
4470 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4472 def : Pat<(v16i8 (scalar_to_vector GPR:$src)),
4473 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4474 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4476 def : Pat<(v8i16 (scalar_to_vector GPR:$src)),
4477 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
4478 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4480 def : Pat<(v4i32 (scalar_to_vector GPR:$src)),
4481 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
4482 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4485 // VDUP : Vector Duplicate (from ARM core register to all elements)
4487 class VDUPD<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
4488 : NVDup<opcod1, 0b1011, opcod3, (outs DPR:$V), (ins GPR:$R),
4489 IIC_VMOVIS, "vdup", Dt, "$V, $R",
4490 [(set DPR:$V, (Ty (NEONvdup (i32 GPR:$R))))]>;
4491 class VDUPQ<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
4492 : NVDup<opcod1, 0b1011, opcod3, (outs QPR:$V), (ins GPR:$R),
4493 IIC_VMOVIS, "vdup", Dt, "$V, $R",
4494 [(set QPR:$V, (Ty (NEONvdup (i32 GPR:$R))))]>;
4496 def VDUP8d : VDUPD<0b11101100, 0b00, "8", v8i8>;
4497 def VDUP16d : VDUPD<0b11101000, 0b01, "16", v4i16>;
4498 def VDUP32d : VDUPD<0b11101000, 0b00, "32", v2i32>;
4499 def VDUP8q : VDUPQ<0b11101110, 0b00, "8", v16i8>;
4500 def VDUP16q : VDUPQ<0b11101010, 0b01, "16", v8i16>;
4501 def VDUP32q : VDUPQ<0b11101010, 0b00, "32", v4i32>;
4503 def : Pat<(v2f32 (NEONvdup (f32 (bitconvert GPR:$R)))), (VDUP32d GPR:$R)>;
4504 def : Pat<(v4f32 (NEONvdup (f32 (bitconvert GPR:$R)))), (VDUP32q GPR:$R)>;
4506 // VDUP : Vector Duplicate Lane (from scalar to all elements)
4508 class VDUPLND<bits<4> op19_16, string OpcodeStr, string Dt,
4510 : NVDupLane<op19_16, 0, (outs DPR:$Vd), (ins DPR:$Vm, nohash_imm:$lane),
4511 IIC_VMOVD, OpcodeStr, Dt, "$Vd, $Vm[$lane]",
4512 [(set DPR:$Vd, (Ty (NEONvduplane (Ty DPR:$Vm), imm:$lane)))]>;
4514 class VDUPLNQ<bits<4> op19_16, string OpcodeStr, string Dt,
4515 ValueType ResTy, ValueType OpTy>
4516 : NVDupLane<op19_16, 1, (outs QPR:$Vd), (ins DPR:$Vm, nohash_imm:$lane),
4517 IIC_VMOVQ, OpcodeStr, Dt, "$Vd, $Vm[$lane]",
4518 [(set QPR:$Vd, (ResTy (NEONvduplane (OpTy DPR:$Vm),
4521 // Inst{19-16} is partially specified depending on the element size.
4523 def VDUPLN8d : VDUPLND<{?,?,?,1}, "vdup", "8", v8i8> {
4524 let Inst{19-17} = lane{2-0};
4526 def VDUPLN16d : VDUPLND<{?,?,1,0}, "vdup", "16", v4i16> {
4527 let Inst{19-18} = lane{1-0};
4529 def VDUPLN32d : VDUPLND<{?,1,0,0}, "vdup", "32", v2i32> {
4530 let Inst{19} = lane{0};
4532 def VDUPLN8q : VDUPLNQ<{?,?,?,1}, "vdup", "8", v16i8, v8i8> {
4533 let Inst{19-17} = lane{2-0};
4535 def VDUPLN16q : VDUPLNQ<{?,?,1,0}, "vdup", "16", v8i16, v4i16> {
4536 let Inst{19-18} = lane{1-0};
4538 def VDUPLN32q : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4i32, v2i32> {
4539 let Inst{19} = lane{0};
4542 def : Pat<(v2f32 (NEONvduplane (v2f32 DPR:$Vm), imm:$lane)),
4543 (VDUPLN32d DPR:$Vm, imm:$lane)>;
4545 def : Pat<(v4f32 (NEONvduplane (v2f32 DPR:$Vm), imm:$lane)),
4546 (VDUPLN32q DPR:$Vm, imm:$lane)>;
4548 def : Pat<(v16i8 (NEONvduplane (v16i8 QPR:$src), imm:$lane)),
4549 (v16i8 (VDUPLN8q (v8i8 (EXTRACT_SUBREG QPR:$src,
4550 (DSubReg_i8_reg imm:$lane))),
4551 (SubReg_i8_lane imm:$lane)))>;
4552 def : Pat<(v8i16 (NEONvduplane (v8i16 QPR:$src), imm:$lane)),
4553 (v8i16 (VDUPLN16q (v4i16 (EXTRACT_SUBREG QPR:$src,
4554 (DSubReg_i16_reg imm:$lane))),
4555 (SubReg_i16_lane imm:$lane)))>;
4556 def : Pat<(v4i32 (NEONvduplane (v4i32 QPR:$src), imm:$lane)),
4557 (v4i32 (VDUPLN32q (v2i32 (EXTRACT_SUBREG QPR:$src,
4558 (DSubReg_i32_reg imm:$lane))),
4559 (SubReg_i32_lane imm:$lane)))>;
4560 def : Pat<(v4f32 (NEONvduplane (v4f32 QPR:$src), imm:$lane)),
4561 (v4f32 (VDUPLN32q (v2f32 (EXTRACT_SUBREG QPR:$src,
4562 (DSubReg_i32_reg imm:$lane))),
4563 (SubReg_i32_lane imm:$lane)))>;
4565 def VDUPfdf : PseudoNeonI<(outs DPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
4566 [(set DPR:$dst, (v2f32 (NEONvdup (f32 SPR:$src))))]>;
4567 def VDUPfqf : PseudoNeonI<(outs QPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
4568 [(set QPR:$dst, (v4f32 (NEONvdup (f32 SPR:$src))))]>;
4570 // VMOVN : Vector Narrowing Move
4571 defm VMOVN : N2VN_HSD<0b11,0b11,0b10,0b00100,0,0, IIC_VMOVN,
4572 "vmovn", "i", trunc>;
4573 // VQMOVN : Vector Saturating Narrowing Move
4574 defm VQMOVNs : N2VNInt_HSD<0b11,0b11,0b10,0b00101,0,0, IIC_VQUNAiD,
4575 "vqmovn", "s", int_arm_neon_vqmovns>;
4576 defm VQMOVNu : N2VNInt_HSD<0b11,0b11,0b10,0b00101,1,0, IIC_VQUNAiD,
4577 "vqmovn", "u", int_arm_neon_vqmovnu>;
4578 defm VQMOVNsu : N2VNInt_HSD<0b11,0b11,0b10,0b00100,1,0, IIC_VQUNAiD,
4579 "vqmovun", "s", int_arm_neon_vqmovnsu>;
4580 // VMOVL : Vector Lengthening Move
4581 defm VMOVLs : N2VL_QHS<0b01,0b10100,0,1, "vmovl", "s", sext>;
4582 defm VMOVLu : N2VL_QHS<0b11,0b10100,0,1, "vmovl", "u", zext>;
4584 // Vector Conversions.
4586 // VCVT : Vector Convert Between Floating-Point and Integers
4587 def VCVTf2sd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
4588 v2i32, v2f32, fp_to_sint>;
4589 def VCVTf2ud : N2VD<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
4590 v2i32, v2f32, fp_to_uint>;
4591 def VCVTs2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
4592 v2f32, v2i32, sint_to_fp>;
4593 def VCVTu2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
4594 v2f32, v2i32, uint_to_fp>;
4596 def VCVTf2sq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
4597 v4i32, v4f32, fp_to_sint>;
4598 def VCVTf2uq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
4599 v4i32, v4f32, fp_to_uint>;
4600 def VCVTs2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
4601 v4f32, v4i32, sint_to_fp>;
4602 def VCVTu2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
4603 v4f32, v4i32, uint_to_fp>;
4605 // VCVT : Vector Convert Between Floating-Point and Fixed-Point.
4606 def VCVTf2xsd : N2VCvtD<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
4607 v2i32, v2f32, int_arm_neon_vcvtfp2fxs>;
4608 def VCVTf2xud : N2VCvtD<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
4609 v2i32, v2f32, int_arm_neon_vcvtfp2fxu>;
4610 def VCVTxs2fd : N2VCvtD<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
4611 v2f32, v2i32, int_arm_neon_vcvtfxs2fp>;
4612 def VCVTxu2fd : N2VCvtD<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
4613 v2f32, v2i32, int_arm_neon_vcvtfxu2fp>;
4615 def VCVTf2xsq : N2VCvtQ<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
4616 v4i32, v4f32, int_arm_neon_vcvtfp2fxs>;
4617 def VCVTf2xuq : N2VCvtQ<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
4618 v4i32, v4f32, int_arm_neon_vcvtfp2fxu>;
4619 def VCVTxs2fq : N2VCvtQ<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
4620 v4f32, v4i32, int_arm_neon_vcvtfxs2fp>;
4621 def VCVTxu2fq : N2VCvtQ<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
4622 v4f32, v4i32, int_arm_neon_vcvtfxu2fp>;
4624 // VCVT : Vector Convert Between Half-Precision and Single-Precision.
4625 def VCVTf2h : N2VNInt<0b11, 0b11, 0b01, 0b10, 0b01100, 0, 0,
4626 IIC_VUNAQ, "vcvt", "f16.f32",
4627 v4i16, v4f32, int_arm_neon_vcvtfp2hf>,
4628 Requires<[HasNEON, HasFP16]>;
4629 def VCVTh2f : N2VLInt<0b11, 0b11, 0b01, 0b10, 0b01110, 0, 0,
4630 IIC_VUNAQ, "vcvt", "f32.f16",
4631 v4f32, v4i16, int_arm_neon_vcvthf2fp>,
4632 Requires<[HasNEON, HasFP16]>;
4636 // VREV64 : Vector Reverse elements within 64-bit doublewords
4638 class VREV64D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4639 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 0, 0, (outs DPR:$Vd),
4640 (ins DPR:$Vm), IIC_VMOVD,
4641 OpcodeStr, Dt, "$Vd, $Vm", "",
4642 [(set DPR:$Vd, (Ty (NEONvrev64 (Ty DPR:$Vm))))]>;
4643 class VREV64Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4644 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 1, 0, (outs QPR:$Vd),
4645 (ins QPR:$Vm), IIC_VMOVQ,
4646 OpcodeStr, Dt, "$Vd, $Vm", "",
4647 [(set QPR:$Vd, (Ty (NEONvrev64 (Ty QPR:$Vm))))]>;
4649 def VREV64d8 : VREV64D<0b00, "vrev64", "8", v8i8>;
4650 def VREV64d16 : VREV64D<0b01, "vrev64", "16", v4i16>;
4651 def VREV64d32 : VREV64D<0b10, "vrev64", "32", v2i32>;
4652 def : Pat<(v2f32 (NEONvrev64 (v2f32 DPR:$Vm))), (VREV64d32 DPR:$Vm)>;
4654 def VREV64q8 : VREV64Q<0b00, "vrev64", "8", v16i8>;
4655 def VREV64q16 : VREV64Q<0b01, "vrev64", "16", v8i16>;
4656 def VREV64q32 : VREV64Q<0b10, "vrev64", "32", v4i32>;
4657 def : Pat<(v4f32 (NEONvrev64 (v4f32 QPR:$Vm))), (VREV64q32 QPR:$Vm)>;
4659 // VREV32 : Vector Reverse elements within 32-bit words
4661 class VREV32D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4662 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 0, 0, (outs DPR:$Vd),
4663 (ins DPR:$Vm), IIC_VMOVD,
4664 OpcodeStr, Dt, "$Vd, $Vm", "",
4665 [(set DPR:$Vd, (Ty (NEONvrev32 (Ty DPR:$Vm))))]>;
4666 class VREV32Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4667 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 1, 0, (outs QPR:$Vd),
4668 (ins QPR:$Vm), IIC_VMOVQ,
4669 OpcodeStr, Dt, "$Vd, $Vm", "",
4670 [(set QPR:$Vd, (Ty (NEONvrev32 (Ty QPR:$Vm))))]>;
4672 def VREV32d8 : VREV32D<0b00, "vrev32", "8", v8i8>;
4673 def VREV32d16 : VREV32D<0b01, "vrev32", "16", v4i16>;
4675 def VREV32q8 : VREV32Q<0b00, "vrev32", "8", v16i8>;
4676 def VREV32q16 : VREV32Q<0b01, "vrev32", "16", v8i16>;
4678 // VREV16 : Vector Reverse elements within 16-bit halfwords
4680 class VREV16D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4681 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 0, 0, (outs DPR:$Vd),
4682 (ins DPR:$Vm), IIC_VMOVD,
4683 OpcodeStr, Dt, "$Vd, $Vm", "",
4684 [(set DPR:$Vd, (Ty (NEONvrev16 (Ty DPR:$Vm))))]>;
4685 class VREV16Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4686 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 1, 0, (outs QPR:$Vd),
4687 (ins QPR:$Vm), IIC_VMOVQ,
4688 OpcodeStr, Dt, "$Vd, $Vm", "",
4689 [(set QPR:$Vd, (Ty (NEONvrev16 (Ty QPR:$Vm))))]>;
4691 def VREV16d8 : VREV16D<0b00, "vrev16", "8", v8i8>;
4692 def VREV16q8 : VREV16Q<0b00, "vrev16", "8", v16i8>;
4694 // Other Vector Shuffles.
4696 // Aligned extractions: really just dropping registers
4698 class AlignedVEXTq<ValueType DestTy, ValueType SrcTy, SDNodeXForm LaneCVT>
4699 : Pat<(DestTy (vector_extract_subvec (SrcTy QPR:$src), (i32 imm:$start))),
4700 (EXTRACT_SUBREG (SrcTy QPR:$src), (LaneCVT imm:$start))>;
4702 def : AlignedVEXTq<v8i8, v16i8, DSubReg_i8_reg>;
4704 def : AlignedVEXTq<v4i16, v8i16, DSubReg_i16_reg>;
4706 def : AlignedVEXTq<v2i32, v4i32, DSubReg_i32_reg>;
4708 def : AlignedVEXTq<v1i64, v2i64, DSubReg_f64_reg>;
4710 def : AlignedVEXTq<v2f32, v4f32, DSubReg_i32_reg>;
4713 // VEXT : Vector Extract
4715 class VEXTd<string OpcodeStr, string Dt, ValueType Ty>
4716 : N3V<0,1,0b11,{?,?,?,?},0,0, (outs DPR:$Vd),
4717 (ins DPR:$Vn, DPR:$Vm, i32imm:$index), NVExtFrm,
4718 IIC_VEXTD, OpcodeStr, Dt, "$Vd, $Vn, $Vm, $index", "",
4719 [(set DPR:$Vd, (Ty (NEONvext (Ty DPR:$Vn),
4720 (Ty DPR:$Vm), imm:$index)))]> {
4722 let Inst{11-8} = index{3-0};
4725 class VEXTq<string OpcodeStr, string Dt, ValueType Ty>
4726 : N3V<0,1,0b11,{?,?,?,?},1,0, (outs QPR:$Vd),
4727 (ins QPR:$Vn, QPR:$Vm, i32imm:$index), NVExtFrm,
4728 IIC_VEXTQ, OpcodeStr, Dt, "$Vd, $Vn, $Vm, $index", "",
4729 [(set QPR:$Vd, (Ty (NEONvext (Ty QPR:$Vn),
4730 (Ty QPR:$Vm), imm:$index)))]> {
4732 let Inst{11-8} = index{3-0};
4735 def VEXTd8 : VEXTd<"vext", "8", v8i8> {
4736 let Inst{11-8} = index{3-0};
4738 def VEXTd16 : VEXTd<"vext", "16", v4i16> {
4739 let Inst{11-9} = index{2-0};
4742 def VEXTd32 : VEXTd<"vext", "32", v2i32> {
4743 let Inst{11-10} = index{1-0};
4744 let Inst{9-8} = 0b00;
4746 def : Pat<(v2f32 (NEONvext (v2f32 DPR:$Vn),
4749 (VEXTd32 DPR:$Vn, DPR:$Vm, imm:$index)>;
4751 def VEXTq8 : VEXTq<"vext", "8", v16i8> {
4752 let Inst{11-8} = index{3-0};
4754 def VEXTq16 : VEXTq<"vext", "16", v8i16> {
4755 let Inst{11-9} = index{2-0};
4758 def VEXTq32 : VEXTq<"vext", "32", v4i32> {
4759 let Inst{11-10} = index{1-0};
4760 let Inst{9-8} = 0b00;
4762 def : Pat<(v4f32 (NEONvext (v4f32 QPR:$Vn),
4765 (VEXTq32 QPR:$Vn, QPR:$Vm, imm:$index)>;
4767 // VTRN : Vector Transpose
4769 def VTRNd8 : N2VDShuffle<0b00, 0b00001, "vtrn", "8">;
4770 def VTRNd16 : N2VDShuffle<0b01, 0b00001, "vtrn", "16">;
4771 def VTRNd32 : N2VDShuffle<0b10, 0b00001, "vtrn", "32">;
4773 def VTRNq8 : N2VQShuffle<0b00, 0b00001, IIC_VPERMQ, "vtrn", "8">;
4774 def VTRNq16 : N2VQShuffle<0b01, 0b00001, IIC_VPERMQ, "vtrn", "16">;
4775 def VTRNq32 : N2VQShuffle<0b10, 0b00001, IIC_VPERMQ, "vtrn", "32">;
4777 // VUZP : Vector Unzip (Deinterleave)
4779 def VUZPd8 : N2VDShuffle<0b00, 0b00010, "vuzp", "8">;
4780 def VUZPd16 : N2VDShuffle<0b01, 0b00010, "vuzp", "16">;
4781 def VUZPd32 : N2VDShuffle<0b10, 0b00010, "vuzp", "32">;
4783 def VUZPq8 : N2VQShuffle<0b00, 0b00010, IIC_VPERMQ3, "vuzp", "8">;
4784 def VUZPq16 : N2VQShuffle<0b01, 0b00010, IIC_VPERMQ3, "vuzp", "16">;
4785 def VUZPq32 : N2VQShuffle<0b10, 0b00010, IIC_VPERMQ3, "vuzp", "32">;
4787 // VZIP : Vector Zip (Interleave)
4789 def VZIPd8 : N2VDShuffle<0b00, 0b00011, "vzip", "8">;
4790 def VZIPd16 : N2VDShuffle<0b01, 0b00011, "vzip", "16">;
4791 def VZIPd32 : N2VDShuffle<0b10, 0b00011, "vzip", "32">;
4793 def VZIPq8 : N2VQShuffle<0b00, 0b00011, IIC_VPERMQ3, "vzip", "8">;
4794 def VZIPq16 : N2VQShuffle<0b01, 0b00011, IIC_VPERMQ3, "vzip", "16">;
4795 def VZIPq32 : N2VQShuffle<0b10, 0b00011, IIC_VPERMQ3, "vzip", "32">;
4797 // Vector Table Lookup and Table Extension.
4799 // VTBL : Vector Table Lookup
4800 let DecoderMethod = "DecodeTBLInstruction" in {
4802 : N3V<1,1,0b11,0b1000,0,0, (outs DPR:$Vd),
4803 (ins DPR:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTB1,
4804 "vtbl", "8", "$Vd, \\{$Vn\\}, $Vm", "",
4805 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbl1 DPR:$Vn, DPR:$Vm)))]>;
4806 let hasExtraSrcRegAllocReq = 1 in {
4808 : N3V<1,1,0b11,0b1001,0,0, (outs DPR:$Vd),
4809 (ins DPR:$Vn, DPR:$tbl2, DPR:$Vm), NVTBLFrm, IIC_VTB2,
4810 "vtbl", "8", "$Vd, \\{$Vn, $tbl2\\}, $Vm", "", []>;
4812 : N3V<1,1,0b11,0b1010,0,0, (outs DPR:$Vd),
4813 (ins DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$Vm), NVTBLFrm, IIC_VTB3,
4814 "vtbl", "8", "$Vd, \\{$Vn, $tbl2, $tbl3\\}, $Vm", "", []>;
4816 : N3V<1,1,0b11,0b1011,0,0, (outs DPR:$Vd),
4817 (ins DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$Vm),
4819 "vtbl", "8", "$Vd, \\{$Vn, $tbl2, $tbl3, $tbl4\\}, $Vm", "", []>;
4820 } // hasExtraSrcRegAllocReq = 1
4823 : PseudoNeonI<(outs DPR:$dst), (ins QPR:$tbl, DPR:$src), IIC_VTB2, "", []>;
4825 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB3, "", []>;
4827 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB4, "", []>;
4829 // VTBX : Vector Table Extension
4831 : N3V<1,1,0b11,0b1000,1,0, (outs DPR:$Vd),
4832 (ins DPR:$orig, DPR:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTBX1,
4833 "vtbx", "8", "$Vd, \\{$Vn\\}, $Vm", "$orig = $Vd",
4834 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbx1
4835 DPR:$orig, DPR:$Vn, DPR:$Vm)))]>;
4836 let hasExtraSrcRegAllocReq = 1 in {
4838 : N3V<1,1,0b11,0b1001,1,0, (outs DPR:$Vd),
4839 (ins DPR:$orig, DPR:$Vn, DPR:$tbl2, DPR:$Vm), NVTBLFrm, IIC_VTBX2,
4840 "vtbx", "8", "$Vd, \\{$Vn, $tbl2\\}, $Vm", "$orig = $Vd", []>;
4842 : N3V<1,1,0b11,0b1010,1,0, (outs DPR:$Vd),
4843 (ins DPR:$orig, DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$Vm),
4844 NVTBLFrm, IIC_VTBX3,
4845 "vtbx", "8", "$Vd, \\{$Vn, $tbl2, $tbl3\\}, $Vm",
4848 : N3V<1,1,0b11,0b1011,1,0, (outs DPR:$Vd), (ins DPR:$orig, DPR:$Vn,
4849 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$Vm), NVTBLFrm, IIC_VTBX4,
4850 "vtbx", "8", "$Vd, \\{$Vn, $tbl2, $tbl3, $tbl4\\}, $Vm",
4852 } // hasExtraSrcRegAllocReq = 1
4855 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QPR:$tbl, DPR:$src),
4856 IIC_VTBX2, "$orig = $dst", []>;
4858 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
4859 IIC_VTBX3, "$orig = $dst", []>;
4861 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
4862 IIC_VTBX4, "$orig = $dst", []>;
4863 } // DecoderMethod = "DecodeTBLInstruction"
4865 //===----------------------------------------------------------------------===//
4866 // NEON instructions for single-precision FP math
4867 //===----------------------------------------------------------------------===//
4869 class N2VSPat<SDNode OpNode, NeonI Inst>
4870 : NEONFPPat<(f32 (OpNode SPR:$a)),
4872 (v2f32 (COPY_TO_REGCLASS (Inst
4874 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
4875 SPR:$a, ssub_0)), DPR_VFP2)), ssub_0)>;
4877 class N3VSPat<SDNode OpNode, NeonI Inst>
4878 : NEONFPPat<(f32 (OpNode SPR:$a, SPR:$b)),
4880 (v2f32 (COPY_TO_REGCLASS (Inst
4882 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
4885 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
4886 SPR:$b, ssub_0)), DPR_VFP2)), ssub_0)>;
4888 class N3VSMulOpPat<SDNode MulNode, SDNode OpNode, NeonI Inst>
4889 : NEONFPPat<(f32 (OpNode SPR:$acc, (f32 (MulNode SPR:$a, SPR:$b)))),
4891 (v2f32 (COPY_TO_REGCLASS (Inst
4893 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
4896 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
4899 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
4900 SPR:$b, ssub_0)), DPR_VFP2)), ssub_0)>;
4902 def : N3VSPat<fadd, VADDfd>;
4903 def : N3VSPat<fsub, VSUBfd>;
4904 def : N3VSPat<fmul, VMULfd>;
4905 def : N3VSMulOpPat<fmul, fadd, VMLAfd>,
4906 Requires<[HasNEON, UseNEONForFP, UseFPVMLx]>;
4907 def : N3VSMulOpPat<fmul, fsub, VMLSfd>,
4908 Requires<[HasNEON, UseNEONForFP, UseFPVMLx]>;
4909 def : N2VSPat<fabs, VABSfd>;
4910 def : N2VSPat<fneg, VNEGfd>;
4911 def : N3VSPat<NEONfmax, VMAXfd>;
4912 def : N3VSPat<NEONfmin, VMINfd>;
4913 def : N2VSPat<arm_ftosi, VCVTf2sd>;
4914 def : N2VSPat<arm_ftoui, VCVTf2ud>;
4915 def : N2VSPat<arm_sitof, VCVTs2fd>;
4916 def : N2VSPat<arm_uitof, VCVTu2fd>;
4918 //===----------------------------------------------------------------------===//
4919 // Non-Instruction Patterns
4920 //===----------------------------------------------------------------------===//
4923 def : Pat<(v1i64 (bitconvert (v2i32 DPR:$src))), (v1i64 DPR:$src)>;
4924 def : Pat<(v1i64 (bitconvert (v4i16 DPR:$src))), (v1i64 DPR:$src)>;
4925 def : Pat<(v1i64 (bitconvert (v8i8 DPR:$src))), (v1i64 DPR:$src)>;
4926 def : Pat<(v1i64 (bitconvert (f64 DPR:$src))), (v1i64 DPR:$src)>;
4927 def : Pat<(v1i64 (bitconvert (v2f32 DPR:$src))), (v1i64 DPR:$src)>;
4928 def : Pat<(v2i32 (bitconvert (v1i64 DPR:$src))), (v2i32 DPR:$src)>;
4929 def : Pat<(v2i32 (bitconvert (v4i16 DPR:$src))), (v2i32 DPR:$src)>;
4930 def : Pat<(v2i32 (bitconvert (v8i8 DPR:$src))), (v2i32 DPR:$src)>;
4931 def : Pat<(v2i32 (bitconvert (f64 DPR:$src))), (v2i32 DPR:$src)>;
4932 def : Pat<(v2i32 (bitconvert (v2f32 DPR:$src))), (v2i32 DPR:$src)>;
4933 def : Pat<(v4i16 (bitconvert (v1i64 DPR:$src))), (v4i16 DPR:$src)>;
4934 def : Pat<(v4i16 (bitconvert (v2i32 DPR:$src))), (v4i16 DPR:$src)>;
4935 def : Pat<(v4i16 (bitconvert (v8i8 DPR:$src))), (v4i16 DPR:$src)>;
4936 def : Pat<(v4i16 (bitconvert (f64 DPR:$src))), (v4i16 DPR:$src)>;
4937 def : Pat<(v4i16 (bitconvert (v2f32 DPR:$src))), (v4i16 DPR:$src)>;
4938 def : Pat<(v8i8 (bitconvert (v1i64 DPR:$src))), (v8i8 DPR:$src)>;
4939 def : Pat<(v8i8 (bitconvert (v2i32 DPR:$src))), (v8i8 DPR:$src)>;
4940 def : Pat<(v8i8 (bitconvert (v4i16 DPR:$src))), (v8i8 DPR:$src)>;
4941 def : Pat<(v8i8 (bitconvert (f64 DPR:$src))), (v8i8 DPR:$src)>;
4942 def : Pat<(v8i8 (bitconvert (v2f32 DPR:$src))), (v8i8 DPR:$src)>;
4943 def : Pat<(f64 (bitconvert (v1i64 DPR:$src))), (f64 DPR:$src)>;
4944 def : Pat<(f64 (bitconvert (v2i32 DPR:$src))), (f64 DPR:$src)>;
4945 def : Pat<(f64 (bitconvert (v4i16 DPR:$src))), (f64 DPR:$src)>;
4946 def : Pat<(f64 (bitconvert (v8i8 DPR:$src))), (f64 DPR:$src)>;
4947 def : Pat<(f64 (bitconvert (v2f32 DPR:$src))), (f64 DPR:$src)>;
4948 def : Pat<(v2f32 (bitconvert (f64 DPR:$src))), (v2f32 DPR:$src)>;
4949 def : Pat<(v2f32 (bitconvert (v1i64 DPR:$src))), (v2f32 DPR:$src)>;
4950 def : Pat<(v2f32 (bitconvert (v2i32 DPR:$src))), (v2f32 DPR:$src)>;
4951 def : Pat<(v2f32 (bitconvert (v4i16 DPR:$src))), (v2f32 DPR:$src)>;
4952 def : Pat<(v2f32 (bitconvert (v8i8 DPR:$src))), (v2f32 DPR:$src)>;
4954 def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (v2i64 QPR:$src)>;
4955 def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (v2i64 QPR:$src)>;
4956 def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (v2i64 QPR:$src)>;
4957 def : Pat<(v2i64 (bitconvert (v2f64 QPR:$src))), (v2i64 QPR:$src)>;
4958 def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (v2i64 QPR:$src)>;
4959 def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (v4i32 QPR:$src)>;
4960 def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (v4i32 QPR:$src)>;
4961 def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (v4i32 QPR:$src)>;
4962 def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (v4i32 QPR:$src)>;
4963 def : Pat<(v4i32 (bitconvert (v4f32 QPR:$src))), (v4i32 QPR:$src)>;
4964 def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (v8i16 QPR:$src)>;
4965 def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (v8i16 QPR:$src)>;
4966 def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (v8i16 QPR:$src)>;
4967 def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (v8i16 QPR:$src)>;
4968 def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (v8i16 QPR:$src)>;
4969 def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (v16i8 QPR:$src)>;
4970 def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (v16i8 QPR:$src)>;
4971 def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (v16i8 QPR:$src)>;
4972 def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (v16i8 QPR:$src)>;
4973 def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (v16i8 QPR:$src)>;
4974 def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (v4f32 QPR:$src)>;
4975 def : Pat<(v4f32 (bitconvert (v4i32 QPR:$src))), (v4f32 QPR:$src)>;
4976 def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (v4f32 QPR:$src)>;
4977 def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (v4f32 QPR:$src)>;
4978 def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (v4f32 QPR:$src)>;
4979 def : Pat<(v2f64 (bitconvert (v2i64 QPR:$src))), (v2f64 QPR:$src)>;
4980 def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (v2f64 QPR:$src)>;
4981 def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 QPR:$src)>;
4982 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 QPR:$src)>;
4983 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 QPR:$src)>;