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>]>;
20 def NEONvceq : SDNode<"ARMISD::VCEQ", SDTARMVCMP>;
21 def NEONvcge : SDNode<"ARMISD::VCGE", SDTARMVCMP>;
22 def NEONvcgeu : SDNode<"ARMISD::VCGEU", SDTARMVCMP>;
23 def NEONvcgt : SDNode<"ARMISD::VCGT", SDTARMVCMP>;
24 def NEONvcgtu : SDNode<"ARMISD::VCGTU", SDTARMVCMP>;
25 def NEONvtst : SDNode<"ARMISD::VTST", SDTARMVCMP>;
27 // Types for vector shift by immediates. The "SHX" version is for long and
28 // narrow operations where the source and destination vectors have different
29 // types. The "SHINS" version is for shift and insert operations.
30 def SDTARMVSH : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
32 def SDTARMVSHX : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
34 def SDTARMVSHINS : SDTypeProfile<1, 3, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
35 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
37 def NEONvshl : SDNode<"ARMISD::VSHL", SDTARMVSH>;
38 def NEONvshrs : SDNode<"ARMISD::VSHRs", SDTARMVSH>;
39 def NEONvshru : SDNode<"ARMISD::VSHRu", SDTARMVSH>;
40 def NEONvshlls : SDNode<"ARMISD::VSHLLs", SDTARMVSHX>;
41 def NEONvshllu : SDNode<"ARMISD::VSHLLu", SDTARMVSHX>;
42 def NEONvshlli : SDNode<"ARMISD::VSHLLi", SDTARMVSHX>;
43 def NEONvshrn : SDNode<"ARMISD::VSHRN", SDTARMVSHX>;
45 def NEONvrshrs : SDNode<"ARMISD::VRSHRs", SDTARMVSH>;
46 def NEONvrshru : SDNode<"ARMISD::VRSHRu", SDTARMVSH>;
47 def NEONvrshrn : SDNode<"ARMISD::VRSHRN", SDTARMVSHX>;
49 def NEONvqshls : SDNode<"ARMISD::VQSHLs", SDTARMVSH>;
50 def NEONvqshlu : SDNode<"ARMISD::VQSHLu", SDTARMVSH>;
51 def NEONvqshlsu : SDNode<"ARMISD::VQSHLsu", SDTARMVSH>;
52 def NEONvqshrns : SDNode<"ARMISD::VQSHRNs", SDTARMVSHX>;
53 def NEONvqshrnu : SDNode<"ARMISD::VQSHRNu", SDTARMVSHX>;
54 def NEONvqshrnsu : SDNode<"ARMISD::VQSHRNsu", SDTARMVSHX>;
56 def NEONvqrshrns : SDNode<"ARMISD::VQRSHRNs", SDTARMVSHX>;
57 def NEONvqrshrnu : SDNode<"ARMISD::VQRSHRNu", SDTARMVSHX>;
58 def NEONvqrshrnsu : SDNode<"ARMISD::VQRSHRNsu", SDTARMVSHX>;
60 def NEONvsli : SDNode<"ARMISD::VSLI", SDTARMVSHINS>;
61 def NEONvsri : SDNode<"ARMISD::VSRI", SDTARMVSHINS>;
63 def SDTARMVGETLN : SDTypeProfile<1, 2, [SDTCisVT<0, i32>, SDTCisInt<1>,
65 def NEONvgetlaneu : SDNode<"ARMISD::VGETLANEu", SDTARMVGETLN>;
66 def NEONvgetlanes : SDNode<"ARMISD::VGETLANEs", SDTARMVGETLN>;
68 def SDTARMVMOVIMM : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisVT<1, i32>]>;
69 def NEONvmovImm : SDNode<"ARMISD::VMOVIMM", SDTARMVMOVIMM>;
70 def NEONvmvnImm : SDNode<"ARMISD::VMVNIMM", SDTARMVMOVIMM>;
72 def SDTARMVORRIMM : SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
74 def NEONvorrImm : SDNode<"ARMISD::VORRIMM", SDTARMVORRIMM>;
76 def NEONvdup : SDNode<"ARMISD::VDUP", SDTypeProfile<1, 1, [SDTCisVec<0>]>>;
78 // VDUPLANE can produce a quad-register result from a double-register source,
79 // so the result is not constrained to match the source.
80 def NEONvduplane : SDNode<"ARMISD::VDUPLANE",
81 SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisVec<1>,
84 def SDTARMVEXT : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
85 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
86 def NEONvext : SDNode<"ARMISD::VEXT", SDTARMVEXT>;
88 def SDTARMVSHUF : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisSameAs<0, 1>]>;
89 def NEONvrev64 : SDNode<"ARMISD::VREV64", SDTARMVSHUF>;
90 def NEONvrev32 : SDNode<"ARMISD::VREV32", SDTARMVSHUF>;
91 def NEONvrev16 : SDNode<"ARMISD::VREV16", SDTARMVSHUF>;
93 def SDTARMVSHUF2 : SDTypeProfile<2, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
96 def NEONzip : SDNode<"ARMISD::VZIP", SDTARMVSHUF2>;
97 def NEONuzp : SDNode<"ARMISD::VUZP", SDTARMVSHUF2>;
98 def NEONtrn : SDNode<"ARMISD::VTRN", SDTARMVSHUF2>;
100 def SDTARMVMULL : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
101 SDTCisSameAs<1, 2>]>;
102 def NEONvmulls : SDNode<"ARMISD::VMULLs", SDTARMVMULL>;
103 def NEONvmullu : SDNode<"ARMISD::VMULLu", SDTARMVMULL>;
105 def SDTARMFMAX : SDTypeProfile<1, 2, [SDTCisVT<0, f32>, SDTCisSameAs<0, 1>,
106 SDTCisSameAs<0, 2>]>;
107 def NEONfmax : SDNode<"ARMISD::FMAX", SDTARMFMAX>;
108 def NEONfmin : SDNode<"ARMISD::FMIN", SDTARMFMAX>;
110 def NEONimmAllZerosV: PatLeaf<(NEONvmovImm (i32 timm)), [{
111 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
112 unsigned EltBits = 0;
113 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
114 return (EltBits == 32 && EltVal == 0);
117 def NEONimmAllOnesV: PatLeaf<(NEONvmovImm (i32 timm)), [{
118 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
119 unsigned EltBits = 0;
120 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
121 return (EltBits == 8 && EltVal == 0xff);
124 //===----------------------------------------------------------------------===//
125 // NEON operand definitions
126 //===----------------------------------------------------------------------===//
128 def nModImm : Operand<i32> {
129 let PrintMethod = "printNEONModImmOperand";
132 //===----------------------------------------------------------------------===//
133 // NEON load / store instructions
134 //===----------------------------------------------------------------------===//
136 // Use VLDM to load a Q register as a D register pair.
137 // This is a pseudo instruction that is expanded to VLDMD after reg alloc.
139 : PseudoVFPLdStM<(outs QPR:$dst), (ins GPR:$Rn, ldstm_mode:$mode),
141 [(set QPR:$dst, (v2f64 (load GPR:$Rn)))]>;
143 // Use VSTM to store a Q register as a D register pair.
144 // This is a pseudo instruction that is expanded to VSTMD after reg alloc.
146 : PseudoVFPLdStM<(outs), (ins QPR:$src, GPR:$Rn, ldstm_mode:$mode),
148 [(store (v2f64 QPR:$src), GPR:$Rn)]>;
150 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
152 // Classes for VLD* pseudo-instructions with multi-register operands.
153 // These are expanded to real instructions after register allocation.
154 class VLDQPseudo<InstrItinClass itin>
155 : PseudoNLdSt<(outs QPR:$dst), (ins addrmode6:$addr), itin, "">;
156 class VLDQWBPseudo<InstrItinClass itin>
157 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
158 (ins addrmode6:$addr, am6offset:$offset), itin,
160 class VLDQQPseudo<InstrItinClass itin>
161 : PseudoNLdSt<(outs QQPR:$dst), (ins addrmode6:$addr), itin, "">;
162 class VLDQQWBPseudo<InstrItinClass itin>
163 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
164 (ins addrmode6:$addr, am6offset:$offset), itin,
166 class VLDQQQQWBPseudo<InstrItinClass itin>
167 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
168 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
169 "$addr.addr = $wb, $src = $dst">;
171 // VLD1 : Vector Load (multiple single elements)
172 class VLD1D<bits<4> op7_4, string Dt>
173 : NLdSt<0,0b10,0b0111,op7_4, (outs DPR:$Vd),
174 (ins addrmode6:$Rn), IIC_VLD1,
175 "vld1", Dt, "\\{$Vd\\}, $Rn", "", []> {
179 class VLD1Q<bits<4> op7_4, string Dt>
180 : NLdSt<0,0b10,0b1010,op7_4, (outs DPR:$Vd, DPR:$dst2),
181 (ins addrmode6:$Rn), IIC_VLD1x2,
182 "vld1", Dt, "\\{$Vd, $dst2\\}, $Rn", "", []> {
184 let Inst{5-4} = Rn{5-4};
187 def VLD1d8 : VLD1D<{0,0,0,?}, "8">;
188 def VLD1d16 : VLD1D<{0,1,0,?}, "16">;
189 def VLD1d32 : VLD1D<{1,0,0,?}, "32">;
190 def VLD1d64 : VLD1D<{1,1,0,?}, "64">;
192 def VLD1q8 : VLD1Q<{0,0,?,?}, "8">;
193 def VLD1q16 : VLD1Q<{0,1,?,?}, "16">;
194 def VLD1q32 : VLD1Q<{1,0,?,?}, "32">;
195 def VLD1q64 : VLD1Q<{1,1,?,?}, "64">;
197 def VLD1q8Pseudo : VLDQPseudo<IIC_VLD1x2>;
198 def VLD1q16Pseudo : VLDQPseudo<IIC_VLD1x2>;
199 def VLD1q32Pseudo : VLDQPseudo<IIC_VLD1x2>;
200 def VLD1q64Pseudo : VLDQPseudo<IIC_VLD1x2>;
202 // ...with address register writeback:
203 class VLD1DWB<bits<4> op7_4, string Dt>
204 : NLdSt<0,0b10,0b0111,op7_4, (outs DPR:$Vd, GPR:$wb),
205 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD1u,
206 "vld1", Dt, "\\{$Vd\\}, $Rn$Rm",
207 "$Rn.addr = $wb", []> {
210 class VLD1QWB<bits<4> op7_4, string Dt>
211 : NLdSt<0,0b10,0b1010,op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
212 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD1x2u,
213 "vld1", Dt, "\\{$Vd, $dst2\\}, $Rn$Rm",
214 "$Rn.addr = $wb", []> {
215 let Inst{5-4} = Rn{5-4};
218 def VLD1d8_UPD : VLD1DWB<{0,0,0,?}, "8">;
219 def VLD1d16_UPD : VLD1DWB<{0,1,0,?}, "16">;
220 def VLD1d32_UPD : VLD1DWB<{1,0,0,?}, "32">;
221 def VLD1d64_UPD : VLD1DWB<{1,1,0,?}, "64">;
223 def VLD1q8_UPD : VLD1QWB<{0,0,?,?}, "8">;
224 def VLD1q16_UPD : VLD1QWB<{0,1,?,?}, "16">;
225 def VLD1q32_UPD : VLD1QWB<{1,0,?,?}, "32">;
226 def VLD1q64_UPD : VLD1QWB<{1,1,?,?}, "64">;
228 def VLD1q8Pseudo_UPD : VLDQWBPseudo<IIC_VLD1x2u>;
229 def VLD1q16Pseudo_UPD : VLDQWBPseudo<IIC_VLD1x2u>;
230 def VLD1q32Pseudo_UPD : VLDQWBPseudo<IIC_VLD1x2u>;
231 def VLD1q64Pseudo_UPD : VLDQWBPseudo<IIC_VLD1x2u>;
233 // ...with 3 registers (some of these are only for the disassembler):
234 class VLD1D3<bits<4> op7_4, string Dt>
235 : NLdSt<0,0b10,0b0110,op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
236 (ins addrmode6:$Rn), IIC_VLD1x3, "vld1", Dt,
237 "\\{$Vd, $dst2, $dst3\\}, $Rn", "", []> {
241 class VLD1D3WB<bits<4> op7_4, string Dt>
242 : NLdSt<0,0b10,0b0110,op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
243 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD1x3u, "vld1", Dt,
244 "\\{$Vd, $dst2, $dst3\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
248 def VLD1d8T : VLD1D3<{0,0,0,?}, "8">;
249 def VLD1d16T : VLD1D3<{0,1,0,?}, "16">;
250 def VLD1d32T : VLD1D3<{1,0,0,?}, "32">;
251 def VLD1d64T : VLD1D3<{1,1,0,?}, "64">;
253 def VLD1d8T_UPD : VLD1D3WB<{0,0,0,?}, "8">;
254 def VLD1d16T_UPD : VLD1D3WB<{0,1,0,?}, "16">;
255 def VLD1d32T_UPD : VLD1D3WB<{1,0,0,?}, "32">;
256 def VLD1d64T_UPD : VLD1D3WB<{1,1,0,?}, "64">;
258 def VLD1d64TPseudo : VLDQQPseudo<IIC_VLD1x3>;
259 def VLD1d64TPseudo_UPD : VLDQQWBPseudo<IIC_VLD1x3u>;
261 // ...with 4 registers (some of these are only for the disassembler):
262 class VLD1D4<bits<4> op7_4, string Dt>
263 : NLdSt<0,0b10,0b0010,op7_4,(outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
264 (ins addrmode6:$Rn), IIC_VLD1x4, "vld1", Dt,
265 "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn", "", []> {
267 let Inst{5-4} = Rn{5-4};
269 class VLD1D4WB<bits<4> op7_4, string Dt>
270 : NLdSt<0,0b10,0b0010,op7_4,
271 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
272 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD4, "vld1", Dt,
273 "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn$Rm", "$Rn.addr = $wb",
275 let Inst{5-4} = Rn{5-4};
278 def VLD1d8Q : VLD1D4<{0,0,?,?}, "8">;
279 def VLD1d16Q : VLD1D4<{0,1,?,?}, "16">;
280 def VLD1d32Q : VLD1D4<{1,0,?,?}, "32">;
281 def VLD1d64Q : VLD1D4<{1,1,?,?}, "64">;
283 def VLD1d8Q_UPD : VLD1D4WB<{0,0,?,?}, "8">;
284 def VLD1d16Q_UPD : VLD1D4WB<{0,1,?,?}, "16">;
285 def VLD1d32Q_UPD : VLD1D4WB<{1,0,?,?}, "32">;
286 def VLD1d64Q_UPD : VLD1D4WB<{1,1,?,?}, "64">;
288 def VLD1d64QPseudo : VLDQQPseudo<IIC_VLD1x4>;
289 def VLD1d64QPseudo_UPD : VLDQQWBPseudo<IIC_VLD1x4u>;
291 // VLD2 : Vector Load (multiple 2-element structures)
292 class VLD2D<bits<4> op11_8, bits<4> op7_4, string Dt>
293 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2),
294 (ins addrmode6:$Rn), IIC_VLD2,
295 "vld2", Dt, "\\{$Vd, $dst2\\}, $Rn", "", []> {
297 let Inst{5-4} = Rn{5-4};
299 class VLD2Q<bits<4> op7_4, string Dt>
300 : NLdSt<0, 0b10, 0b0011, op7_4,
301 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
302 (ins addrmode6:$Rn), IIC_VLD2x2,
303 "vld2", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn", "", []> {
305 let Inst{5-4} = Rn{5-4};
308 def VLD2d8 : VLD2D<0b1000, {0,0,?,?}, "8">;
309 def VLD2d16 : VLD2D<0b1000, {0,1,?,?}, "16">;
310 def VLD2d32 : VLD2D<0b1000, {1,0,?,?}, "32">;
312 def VLD2q8 : VLD2Q<{0,0,?,?}, "8">;
313 def VLD2q16 : VLD2Q<{0,1,?,?}, "16">;
314 def VLD2q32 : VLD2Q<{1,0,?,?}, "32">;
316 def VLD2d8Pseudo : VLDQPseudo<IIC_VLD2>;
317 def VLD2d16Pseudo : VLDQPseudo<IIC_VLD2>;
318 def VLD2d32Pseudo : VLDQPseudo<IIC_VLD2>;
320 def VLD2q8Pseudo : VLDQQPseudo<IIC_VLD2x2>;
321 def VLD2q16Pseudo : VLDQQPseudo<IIC_VLD2x2>;
322 def VLD2q32Pseudo : VLDQQPseudo<IIC_VLD2x2>;
324 // ...with address register writeback:
325 class VLD2DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
326 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
327 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD2u,
328 "vld2", Dt, "\\{$Vd, $dst2\\}, $Rn$Rm",
329 "$Rn.addr = $wb", []> {
330 let Inst{5-4} = Rn{5-4};
332 class VLD2QWB<bits<4> op7_4, string Dt>
333 : NLdSt<0, 0b10, 0b0011, op7_4,
334 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
335 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD2x2u,
336 "vld2", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn$Rm",
337 "$Rn.addr = $wb", []> {
338 let Inst{5-4} = Rn{5-4};
341 def VLD2d8_UPD : VLD2DWB<0b1000, {0,0,?,?}, "8">;
342 def VLD2d16_UPD : VLD2DWB<0b1000, {0,1,?,?}, "16">;
343 def VLD2d32_UPD : VLD2DWB<0b1000, {1,0,?,?}, "32">;
345 def VLD2q8_UPD : VLD2QWB<{0,0,?,?}, "8">;
346 def VLD2q16_UPD : VLD2QWB<{0,1,?,?}, "16">;
347 def VLD2q32_UPD : VLD2QWB<{1,0,?,?}, "32">;
349 def VLD2d8Pseudo_UPD : VLDQWBPseudo<IIC_VLD2u>;
350 def VLD2d16Pseudo_UPD : VLDQWBPseudo<IIC_VLD2u>;
351 def VLD2d32Pseudo_UPD : VLDQWBPseudo<IIC_VLD2u>;
353 def VLD2q8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD2x2u>;
354 def VLD2q16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD2x2u>;
355 def VLD2q32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD2x2u>;
357 // ...with double-spaced registers (for disassembly only):
358 def VLD2b8 : VLD2D<0b1001, {0,0,?,?}, "8">;
359 def VLD2b16 : VLD2D<0b1001, {0,1,?,?}, "16">;
360 def VLD2b32 : VLD2D<0b1001, {1,0,?,?}, "32">;
361 def VLD2b8_UPD : VLD2DWB<0b1001, {0,0,?,?}, "8">;
362 def VLD2b16_UPD : VLD2DWB<0b1001, {0,1,?,?}, "16">;
363 def VLD2b32_UPD : VLD2DWB<0b1001, {1,0,?,?}, "32">;
365 // VLD3 : Vector Load (multiple 3-element structures)
366 class VLD3D<bits<4> op11_8, bits<4> op7_4, string Dt>
367 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
368 (ins addrmode6:$Rn), IIC_VLD3,
369 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn", "", []> {
374 def VLD3d8 : VLD3D<0b0100, {0,0,0,?}, "8">;
375 def VLD3d16 : VLD3D<0b0100, {0,1,0,?}, "16">;
376 def VLD3d32 : VLD3D<0b0100, {1,0,0,?}, "32">;
378 def VLD3d8Pseudo : VLDQQPseudo<IIC_VLD3>;
379 def VLD3d16Pseudo : VLDQQPseudo<IIC_VLD3>;
380 def VLD3d32Pseudo : VLDQQPseudo<IIC_VLD3>;
382 // ...with address register writeback:
383 class VLD3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
384 : NLdSt<0, 0b10, op11_8, op7_4,
385 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
386 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD3u,
387 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn$Rm",
388 "$Rn.addr = $wb", []> {
392 def VLD3d8_UPD : VLD3DWB<0b0100, {0,0,0,?}, "8">;
393 def VLD3d16_UPD : VLD3DWB<0b0100, {0,1,0,?}, "16">;
394 def VLD3d32_UPD : VLD3DWB<0b0100, {1,0,0,?}, "32">;
396 def VLD3d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
397 def VLD3d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
398 def VLD3d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
400 // ...with double-spaced registers (non-updating versions for disassembly only):
401 def VLD3q8 : VLD3D<0b0101, {0,0,0,?}, "8">;
402 def VLD3q16 : VLD3D<0b0101, {0,1,0,?}, "16">;
403 def VLD3q32 : VLD3D<0b0101, {1,0,0,?}, "32">;
404 def VLD3q8_UPD : VLD3DWB<0b0101, {0,0,0,?}, "8">;
405 def VLD3q16_UPD : VLD3DWB<0b0101, {0,1,0,?}, "16">;
406 def VLD3q32_UPD : VLD3DWB<0b0101, {1,0,0,?}, "32">;
408 def VLD3q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
409 def VLD3q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
410 def VLD3q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
412 // ...alternate versions to be allocated odd register numbers:
413 def VLD3q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
414 def VLD3q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
415 def VLD3q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
417 // VLD4 : Vector Load (multiple 4-element structures)
418 class VLD4D<bits<4> op11_8, bits<4> op7_4, string Dt>
419 : NLdSt<0, 0b10, op11_8, op7_4,
420 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
421 (ins addrmode6:$Rn), IIC_VLD4,
422 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn", "", []> {
424 let Inst{5-4} = Rn{5-4};
427 def VLD4d8 : VLD4D<0b0000, {0,0,?,?}, "8">;
428 def VLD4d16 : VLD4D<0b0000, {0,1,?,?}, "16">;
429 def VLD4d32 : VLD4D<0b0000, {1,0,?,?}, "32">;
431 def VLD4d8Pseudo : VLDQQPseudo<IIC_VLD4>;
432 def VLD4d16Pseudo : VLDQQPseudo<IIC_VLD4>;
433 def VLD4d32Pseudo : VLDQQPseudo<IIC_VLD4>;
435 // ...with address register writeback:
436 class VLD4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
437 : NLdSt<0, 0b10, op11_8, op7_4,
438 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
439 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD4,
440 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn$Rm",
441 "$Rn.addr = $wb", []> {
442 let Inst{5-4} = Rn{5-4};
445 def VLD4d8_UPD : VLD4DWB<0b0000, {0,0,?,?}, "8">;
446 def VLD4d16_UPD : VLD4DWB<0b0000, {0,1,?,?}, "16">;
447 def VLD4d32_UPD : VLD4DWB<0b0000, {1,0,?,?}, "32">;
449 def VLD4d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4>;
450 def VLD4d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4>;
451 def VLD4d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4>;
453 // ...with double-spaced registers (non-updating versions for disassembly only):
454 def VLD4q8 : VLD4D<0b0001, {0,0,?,?}, "8">;
455 def VLD4q16 : VLD4D<0b0001, {0,1,?,?}, "16">;
456 def VLD4q32 : VLD4D<0b0001, {1,0,?,?}, "32">;
457 def VLD4q8_UPD : VLD4DWB<0b0001, {0,0,?,?}, "8">;
458 def VLD4q16_UPD : VLD4DWB<0b0001, {0,1,?,?}, "16">;
459 def VLD4q32_UPD : VLD4DWB<0b0001, {1,0,?,?}, "32">;
461 def VLD4q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4>;
462 def VLD4q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4>;
463 def VLD4q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4>;
465 // ...alternate versions to be allocated odd register numbers:
466 def VLD4q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4>;
467 def VLD4q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4>;
468 def VLD4q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4>;
470 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
472 // Classes for VLD*LN pseudo-instructions with multi-register operands.
473 // These are expanded to real instructions after register allocation.
474 class VLDQLNPseudo<InstrItinClass itin>
475 : PseudoNLdSt<(outs QPR:$dst),
476 (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
477 itin, "$src = $dst">;
478 class VLDQLNWBPseudo<InstrItinClass itin>
479 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
480 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
481 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
482 class VLDQQLNPseudo<InstrItinClass itin>
483 : PseudoNLdSt<(outs QQPR:$dst),
484 (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
485 itin, "$src = $dst">;
486 class VLDQQLNWBPseudo<InstrItinClass itin>
487 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
488 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
489 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
490 class VLDQQQQLNPseudo<InstrItinClass itin>
491 : PseudoNLdSt<(outs QQQQPR:$dst),
492 (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
493 itin, "$src = $dst">;
494 class VLDQQQQLNWBPseudo<InstrItinClass itin>
495 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
496 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
497 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
499 // VLD1LN : Vector Load (single element to one lane)
500 class VLD1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
502 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd),
503 (ins addrmode6:$Rn, DPR:$src, nohash_imm:$lane),
504 IIC_VLD1ln, "vld1", Dt, "\\{$Vd[$lane]\\}, $Rn",
506 [(set DPR:$Vd, (vector_insert (Ty DPR:$src),
507 (i32 (LoadOp addrmode6:$Rn)),
511 class VLD1QLNPseudo<ValueType Ty, PatFrag LoadOp> : VLDQLNPseudo<IIC_VLD1ln> {
512 let Pattern = [(set QPR:$dst, (vector_insert (Ty QPR:$src),
513 (i32 (LoadOp addrmode6:$addr)),
517 def VLD1LNd8 : VLD1LN<0b0000, {?,?,?,0}, "8", v8i8, extloadi8> {
518 let Inst{7-5} = lane{2-0};
520 def VLD1LNd16 : VLD1LN<0b0100, {?,?,0,?}, "16", v4i16, extloadi16> {
521 let Inst{7-6} = lane{1-0};
524 def VLD1LNd32 : VLD1LN<0b1000, {?,0,?,?}, "32", v2i32, load> {
525 let Inst{7} = lane{0};
530 def VLD1LNq8Pseudo : VLD1QLNPseudo<v16i8, extloadi8>;
531 def VLD1LNq16Pseudo : VLD1QLNPseudo<v8i16, extloadi16>;
532 def VLD1LNq32Pseudo : VLD1QLNPseudo<v4i32, load>;
534 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
536 // ...with address register writeback:
537 class VLD1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
538 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, GPR:$wb),
539 (ins addrmode6:$Rn, am6offset:$Rm,
540 DPR:$src, nohash_imm:$lane), IIC_VLD1lnu, "vld1", Dt,
541 "\\{$Vd[$lane]\\}, $Rn$Rm",
542 "$src = $Vd, $Rn.addr = $wb", []>;
544 def VLD1LNd8_UPD : VLD1LNWB<0b0000, {?,?,?,0}, "8"> {
545 let Inst{7-5} = lane{2-0};
547 def VLD1LNd16_UPD : VLD1LNWB<0b0100, {?,?,0,?}, "16"> {
548 let Inst{7-6} = lane{1-0};
551 def VLD1LNd32_UPD : VLD1LNWB<0b1000, {?,0,?,?}, "32"> {
552 let Inst{7} = lane{0};
557 def VLD1LNq8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
558 def VLD1LNq16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
559 def VLD1LNq32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
561 // VLD2LN : Vector Load (single 2-element structure to one lane)
562 class VLD2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
563 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2),
564 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, nohash_imm:$lane),
565 IIC_VLD2ln, "vld2", Dt, "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn",
566 "$src1 = $Vd, $src2 = $dst2", []> {
571 def VLD2LNd8 : VLD2LN<0b0001, {?,?,?,?}, "8"> {
572 let Inst{7-5} = lane{2-0};
574 def VLD2LNd16 : VLD2LN<0b0101, {?,?,0,?}, "16"> {
575 let Inst{7-6} = lane{1-0};
577 def VLD2LNd32 : VLD2LN<0b1001, {?,0,0,?}, "32"> {
578 let Inst{7} = lane{0};
581 def VLD2LNd8Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
582 def VLD2LNd16Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
583 def VLD2LNd32Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
585 // ...with double-spaced registers:
586 def VLD2LNq16 : VLD2LN<0b0101, {?,?,1,?}, "16"> {
587 let Inst{7-6} = lane{1-0};
589 def VLD2LNq32 : VLD2LN<0b1001, {?,1,0,?}, "32"> {
590 let Inst{7} = lane{0};
593 def VLD2LNq16Pseudo : VLDQQLNPseudo<IIC_VLD2ln>;
594 def VLD2LNq32Pseudo : VLDQQLNPseudo<IIC_VLD2ln>;
596 // ...with address register writeback:
597 class VLD2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
598 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
599 (ins addrmode6:$Rn, am6offset:$Rm,
600 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VLD2lnu, "vld2", Dt,
601 "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn$Rm",
602 "$src1 = $Vd, $src2 = $dst2, $Rn.addr = $wb", []> {
606 def VLD2LNd8_UPD : VLD2LNWB<0b0001, {?,?,?,?}, "8"> {
607 let Inst{7-5} = lane{2-0};
609 def VLD2LNd16_UPD : VLD2LNWB<0b0101, {?,?,0,?}, "16"> {
610 let Inst{7-6} = lane{1-0};
612 def VLD2LNd32_UPD : VLD2LNWB<0b1001, {?,0,0,?}, "32"> {
613 let Inst{7} = lane{0};
616 def VLD2LNd8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
617 def VLD2LNd16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
618 def VLD2LNd32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
620 def VLD2LNq16_UPD : VLD2LNWB<0b0101, {?,?,1,?}, "16"> {
621 let Inst{7-6} = lane{1-0};
623 def VLD2LNq32_UPD : VLD2LNWB<0b1001, {?,1,0,?}, "32"> {
624 let Inst{7} = lane{0};
627 def VLD2LNq16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
628 def VLD2LNq32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
630 // VLD3LN : Vector Load (single 3-element structure to one lane)
631 class VLD3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
632 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
633 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3,
634 nohash_imm:$lane), IIC_VLD3ln, "vld3", Dt,
635 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn",
636 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3", []> {
640 def VLD3LNd8 : VLD3LN<0b0010, {?,?,?,0}, "8"> {
641 let Inst{7-5} = lane{2-0};
643 def VLD3LNd16 : VLD3LN<0b0110, {?,?,0,0}, "16"> {
644 let Inst{7-6} = lane{1-0};
646 def VLD3LNd32 : VLD3LN<0b1010, {?,0,0,0}, "32"> {
647 let Inst{7} = lane{0};
650 def VLD3LNd8Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
651 def VLD3LNd16Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
652 def VLD3LNd32Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
654 // ...with double-spaced registers:
655 def VLD3LNq16 : VLD3LN<0b0110, {?,?,1,0}, "16"> {
656 let Inst{7-6} = lane{1-0};
658 def VLD3LNq32 : VLD3LN<0b1010, {?,1,0,0}, "32"> {
659 let Inst{7} = lane{0};
662 def VLD3LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
663 def VLD3LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
665 // ...with address register writeback:
666 class VLD3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
667 : NLdStLn<1, 0b10, op11_8, op7_4,
668 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
669 (ins addrmode6:$Rn, am6offset:$Rm,
670 DPR:$src1, DPR:$src2, DPR:$src3, nohash_imm:$lane),
671 IIC_VLD3lnu, "vld3", Dt,
672 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn$Rm",
673 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $Rn.addr = $wb",
676 def VLD3LNd8_UPD : VLD3LNWB<0b0010, {?,?,?,0}, "8"> {
677 let Inst{7-5} = lane{2-0};
679 def VLD3LNd16_UPD : VLD3LNWB<0b0110, {?,?,0,0}, "16"> {
680 let Inst{7-6} = lane{1-0};
682 def VLD3LNd32_UPD : VLD3LNWB<0b1010, {?,0,0,0}, "32"> {
683 let Inst{7} = lane{0};
686 def VLD3LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
687 def VLD3LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
688 def VLD3LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
690 def VLD3LNq16_UPD : VLD3LNWB<0b0110, {?,?,1,0}, "16"> {
691 let Inst{7-6} = lane{1-0};
693 def VLD3LNq32_UPD : VLD3LNWB<0b1010, {?,1,0,0}, "32"> {
694 let Inst{7} = lane{0};
697 def VLD3LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
698 def VLD3LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
700 // VLD4LN : Vector Load (single 4-element structure to one lane)
701 class VLD4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
702 : NLdStLn<1, 0b10, op11_8, op7_4,
703 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
704 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
705 nohash_imm:$lane), IIC_VLD4ln, "vld4", Dt,
706 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn",
707 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4", []> {
712 def VLD4LNd8 : VLD4LN<0b0011, {?,?,?,?}, "8"> {
713 let Inst{7-5} = lane{2-0};
715 def VLD4LNd16 : VLD4LN<0b0111, {?,?,0,?}, "16"> {
716 let Inst{7-6} = lane{1-0};
718 def VLD4LNd32 : VLD4LN<0b1011, {?,0,?,?}, "32"> {
719 let Inst{7} = lane{0};
723 def VLD4LNd8Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
724 def VLD4LNd16Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
725 def VLD4LNd32Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
727 // ...with double-spaced registers:
728 def VLD4LNq16 : VLD4LN<0b0111, {?,?,1,?}, "16"> {
729 let Inst{7-6} = lane{1-0};
731 def VLD4LNq32 : VLD4LN<0b1011, {?,1,?,?}, "32"> {
732 let Inst{7} = lane{0};
736 def VLD4LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>;
737 def VLD4LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>;
739 // ...with address register writeback:
740 class VLD4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
741 : NLdStLn<1, 0b10, op11_8, op7_4,
742 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
743 (ins addrmode6:$Rn, am6offset:$Rm,
744 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
745 IIC_VLD4ln, "vld4", Dt,
746 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn$Rm",
747 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4, $Rn.addr = $wb",
752 def VLD4LNd8_UPD : VLD4LNWB<0b0011, {?,?,?,?}, "8"> {
753 let Inst{7-5} = lane{2-0};
755 def VLD4LNd16_UPD : VLD4LNWB<0b0111, {?,?,0,?}, "16"> {
756 let Inst{7-6} = lane{1-0};
758 def VLD4LNd32_UPD : VLD4LNWB<0b1011, {?,0,?,?}, "32"> {
759 let Inst{7} = lane{0};
763 def VLD4LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
764 def VLD4LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
765 def VLD4LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
767 def VLD4LNq16_UPD : VLD4LNWB<0b0111, {?,?,1,?}, "16"> {
768 let Inst{7-6} = lane{1-0};
770 def VLD4LNq32_UPD : VLD4LNWB<0b1011, {?,1,?,?}, "32"> {
771 let Inst{7} = lane{0};
775 def VLD4LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
776 def VLD4LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
778 // VLD1DUP : Vector Load (single element to all lanes)
779 // VLD2DUP : Vector Load (single 2-element structure to all lanes)
780 // VLD3DUP : Vector Load (single 3-element structure to all lanes)
781 // VLD4DUP : Vector Load (single 4-element structure to all lanes)
782 // FIXME: Not yet implemented.
783 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
785 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
787 // Classes for VST* pseudo-instructions with multi-register operands.
788 // These are expanded to real instructions after register allocation.
789 class VSTQPseudo<InstrItinClass itin>
790 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src), itin, "">;
791 class VSTQWBPseudo<InstrItinClass itin>
792 : PseudoNLdSt<(outs GPR:$wb),
793 (ins addrmode6:$addr, am6offset:$offset, QPR:$src), itin,
795 class VSTQQPseudo<InstrItinClass itin>
796 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src), itin, "">;
797 class VSTQQWBPseudo<InstrItinClass itin>
798 : PseudoNLdSt<(outs GPR:$wb),
799 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src), itin,
801 class VSTQQQQWBPseudo<InstrItinClass itin>
802 : PseudoNLdSt<(outs GPR:$wb),
803 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
806 // VST1 : Vector Store (multiple single elements)
807 class VST1D<bits<4> op7_4, string Dt>
808 : NLdSt<0,0b00,0b0111,op7_4, (outs), (ins addrmode6:$Rn, DPR:$Vd),
809 IIC_VST1, "vst1", Dt, "\\{$Vd\\}, $Rn", "", []> {
813 class VST1Q<bits<4> op7_4, string Dt>
814 : NLdSt<0,0b00,0b1010,op7_4, (outs),
815 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2), IIC_VST1x2,
816 "vst1", Dt, "\\{$Vd, $src2\\}, $Rn", "", []> {
818 let Inst{5-4} = Rn{5-4};
821 def VST1d8 : VST1D<{0,0,0,?}, "8">;
822 def VST1d16 : VST1D<{0,1,0,?}, "16">;
823 def VST1d32 : VST1D<{1,0,0,?}, "32">;
824 def VST1d64 : VST1D<{1,1,0,?}, "64">;
826 def VST1q8 : VST1Q<{0,0,?,?}, "8">;
827 def VST1q16 : VST1Q<{0,1,?,?}, "16">;
828 def VST1q32 : VST1Q<{1,0,?,?}, "32">;
829 def VST1q64 : VST1Q<{1,1,?,?}, "64">;
831 def VST1q8Pseudo : VSTQPseudo<IIC_VST1x2>;
832 def VST1q16Pseudo : VSTQPseudo<IIC_VST1x2>;
833 def VST1q32Pseudo : VSTQPseudo<IIC_VST1x2>;
834 def VST1q64Pseudo : VSTQPseudo<IIC_VST1x2>;
836 // ...with address register writeback:
837 class VST1DWB<bits<4> op7_4, string Dt>
838 : NLdSt<0, 0b00, 0b0111, op7_4, (outs GPR:$wb),
839 (ins addrmode6:$Rn, am6offset:$Rm, DPR:$Vd), IIC_VST1u,
840 "vst1", Dt, "\\{$Vd\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
843 class VST1QWB<bits<4> op7_4, string Dt>
844 : NLdSt<0, 0b00, 0b1010, op7_4, (outs GPR:$wb),
845 (ins addrmode6:$Rn, am6offset:$Rm, DPR:$Vd, DPR:$src2),
846 IIC_VST1x2u, "vst1", Dt, "\\{$Vd, $src2\\}, $Rn$Rm",
847 "$Rn.addr = $wb", []> {
848 let Inst{5-4} = Rn{5-4};
851 def VST1d8_UPD : VST1DWB<{0,0,0,?}, "8">;
852 def VST1d16_UPD : VST1DWB<{0,1,0,?}, "16">;
853 def VST1d32_UPD : VST1DWB<{1,0,0,?}, "32">;
854 def VST1d64_UPD : VST1DWB<{1,1,0,?}, "64">;
856 def VST1q8_UPD : VST1QWB<{0,0,?,?}, "8">;
857 def VST1q16_UPD : VST1QWB<{0,1,?,?}, "16">;
858 def VST1q32_UPD : VST1QWB<{1,0,?,?}, "32">;
859 def VST1q64_UPD : VST1QWB<{1,1,?,?}, "64">;
861 def VST1q8Pseudo_UPD : VSTQWBPseudo<IIC_VST1x2u>;
862 def VST1q16Pseudo_UPD : VSTQWBPseudo<IIC_VST1x2u>;
863 def VST1q32Pseudo_UPD : VSTQWBPseudo<IIC_VST1x2u>;
864 def VST1q64Pseudo_UPD : VSTQWBPseudo<IIC_VST1x2u>;
866 // ...with 3 registers (some of these are only for the disassembler):
867 class VST1D3<bits<4> op7_4, string Dt>
868 : NLdSt<0, 0b00, 0b0110, op7_4, (outs),
869 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3),
870 IIC_VST1x3, "vst1", Dt, "\\{$Vd, $src2, $src3\\}, $Rn", "", []> {
874 class VST1D3WB<bits<4> op7_4, string Dt>
875 : NLdSt<0, 0b00, 0b0110, op7_4, (outs GPR:$wb),
876 (ins addrmode6:$Rn, am6offset:$Rm,
877 DPR:$Vd, DPR:$src2, DPR:$src3),
878 IIC_VST1x3u, "vst1", Dt, "\\{$Vd, $src2, $src3\\}, $Rn$Rm",
879 "$Rn.addr = $wb", []> {
883 def VST1d8T : VST1D3<{0,0,0,?}, "8">;
884 def VST1d16T : VST1D3<{0,1,0,?}, "16">;
885 def VST1d32T : VST1D3<{1,0,0,?}, "32">;
886 def VST1d64T : VST1D3<{1,1,0,?}, "64">;
888 def VST1d8T_UPD : VST1D3WB<{0,0,0,?}, "8">;
889 def VST1d16T_UPD : VST1D3WB<{0,1,0,?}, "16">;
890 def VST1d32T_UPD : VST1D3WB<{1,0,0,?}, "32">;
891 def VST1d64T_UPD : VST1D3WB<{1,1,0,?}, "64">;
893 def VST1d64TPseudo : VSTQQPseudo<IIC_VST1x3>;
894 def VST1d64TPseudo_UPD : VSTQQWBPseudo<IIC_VST1x3u>;
896 // ...with 4 registers (some of these are only for the disassembler):
897 class VST1D4<bits<4> op7_4, string Dt>
898 : NLdSt<0, 0b00, 0b0010, op7_4, (outs),
899 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
900 IIC_VST1x4, "vst1", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn", "",
903 let Inst{5-4} = Rn{5-4};
905 class VST1D4WB<bits<4> op7_4, string Dt>
906 : NLdSt<0, 0b00, 0b0010, op7_4, (outs GPR:$wb),
907 (ins addrmode6:$Rn, am6offset:$Rm,
908 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST1x4u,
909 "vst1", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
910 "$Rn.addr = $wb", []> {
911 let Inst{5-4} = Rn{5-4};
914 def VST1d8Q : VST1D4<{0,0,?,?}, "8">;
915 def VST1d16Q : VST1D4<{0,1,?,?}, "16">;
916 def VST1d32Q : VST1D4<{1,0,?,?}, "32">;
917 def VST1d64Q : VST1D4<{1,1,?,?}, "64">;
919 def VST1d8Q_UPD : VST1D4WB<{0,0,?,?}, "8">;
920 def VST1d16Q_UPD : VST1D4WB<{0,1,?,?}, "16">;
921 def VST1d32Q_UPD : VST1D4WB<{1,0,?,?}, "32">;
922 def VST1d64Q_UPD : VST1D4WB<{1,1,?,?}, "64">;
924 def VST1d64QPseudo : VSTQQPseudo<IIC_VST1x4>;
925 def VST1d64QPseudo_UPD : VSTQQWBPseudo<IIC_VST1x4u>;
927 // VST2 : Vector Store (multiple 2-element structures)
928 class VST2D<bits<4> op11_8, bits<4> op7_4, string Dt>
929 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
930 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2),
931 IIC_VST2, "vst2", Dt, "\\{$Vd, $src2\\}, $Rn", "", []> {
933 let Inst{5-4} = Rn{5-4};
935 class VST2Q<bits<4> op7_4, string Dt>
936 : NLdSt<0, 0b00, 0b0011, op7_4, (outs),
937 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
938 IIC_VST2x2, "vst2", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn",
941 let Inst{5-4} = Rn{5-4};
944 def VST2d8 : VST2D<0b1000, {0,0,?,?}, "8">;
945 def VST2d16 : VST2D<0b1000, {0,1,?,?}, "16">;
946 def VST2d32 : VST2D<0b1000, {1,0,?,?}, "32">;
948 def VST2q8 : VST2Q<{0,0,?,?}, "8">;
949 def VST2q16 : VST2Q<{0,1,?,?}, "16">;
950 def VST2q32 : VST2Q<{1,0,?,?}, "32">;
952 def VST2d8Pseudo : VSTQPseudo<IIC_VST2>;
953 def VST2d16Pseudo : VSTQPseudo<IIC_VST2>;
954 def VST2d32Pseudo : VSTQPseudo<IIC_VST2>;
956 def VST2q8Pseudo : VSTQQPseudo<IIC_VST2x2>;
957 def VST2q16Pseudo : VSTQQPseudo<IIC_VST2x2>;
958 def VST2q32Pseudo : VSTQQPseudo<IIC_VST2x2>;
960 // ...with address register writeback:
961 class VST2DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
962 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
963 (ins addrmode6:$Rn, am6offset:$Rm, DPR:$Vd, DPR:$src2),
964 IIC_VST2u, "vst2", Dt, "\\{$Vd, $src2\\}, $Rn$Rm",
965 "$Rn.addr = $wb", []> {
966 let Inst{5-4} = Rn{5-4};
968 class VST2QWB<bits<4> op7_4, string Dt>
969 : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
970 (ins addrmode6:$Rn, am6offset:$Rm,
971 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST2x2u,
972 "vst2", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
973 "$Rn.addr = $wb", []> {
974 let Inst{5-4} = Rn{5-4};
977 def VST2d8_UPD : VST2DWB<0b1000, {0,0,?,?}, "8">;
978 def VST2d16_UPD : VST2DWB<0b1000, {0,1,?,?}, "16">;
979 def VST2d32_UPD : VST2DWB<0b1000, {1,0,?,?}, "32">;
981 def VST2q8_UPD : VST2QWB<{0,0,?,?}, "8">;
982 def VST2q16_UPD : VST2QWB<{0,1,?,?}, "16">;
983 def VST2q32_UPD : VST2QWB<{1,0,?,?}, "32">;
985 def VST2d8Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
986 def VST2d16Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
987 def VST2d32Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
989 def VST2q8Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
990 def VST2q16Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
991 def VST2q32Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
993 // ...with double-spaced registers (for disassembly only):
994 def VST2b8 : VST2D<0b1001, {0,0,?,?}, "8">;
995 def VST2b16 : VST2D<0b1001, {0,1,?,?}, "16">;
996 def VST2b32 : VST2D<0b1001, {1,0,?,?}, "32">;
997 def VST2b8_UPD : VST2DWB<0b1001, {0,0,?,?}, "8">;
998 def VST2b16_UPD : VST2DWB<0b1001, {0,1,?,?}, "16">;
999 def VST2b32_UPD : VST2DWB<0b1001, {1,0,?,?}, "32">;
1001 // VST3 : Vector Store (multiple 3-element structures)
1002 class VST3D<bits<4> op11_8, bits<4> op7_4, string Dt>
1003 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1004 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3,
1005 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn", "", []> {
1007 let Inst{4} = Rn{4};
1010 def VST3d8 : VST3D<0b0100, {0,0,0,?}, "8">;
1011 def VST3d16 : VST3D<0b0100, {0,1,0,?}, "16">;
1012 def VST3d32 : VST3D<0b0100, {1,0,0,?}, "32">;
1014 def VST3d8Pseudo : VSTQQPseudo<IIC_VST3>;
1015 def VST3d16Pseudo : VSTQQPseudo<IIC_VST3>;
1016 def VST3d32Pseudo : VSTQQPseudo<IIC_VST3>;
1018 // ...with address register writeback:
1019 class VST3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1020 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1021 (ins addrmode6:$Rn, am6offset:$Rm,
1022 DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3u,
1023 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn$Rm",
1024 "$Rn.addr = $wb", []> {
1025 let Inst{4} = Rn{4};
1028 def VST3d8_UPD : VST3DWB<0b0100, {0,0,0,?}, "8">;
1029 def VST3d16_UPD : VST3DWB<0b0100, {0,1,0,?}, "16">;
1030 def VST3d32_UPD : VST3DWB<0b0100, {1,0,0,?}, "32">;
1032 def VST3d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1033 def VST3d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1034 def VST3d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1036 // ...with double-spaced registers (non-updating versions for disassembly only):
1037 def VST3q8 : VST3D<0b0101, {0,0,0,?}, "8">;
1038 def VST3q16 : VST3D<0b0101, {0,1,0,?}, "16">;
1039 def VST3q32 : VST3D<0b0101, {1,0,0,?}, "32">;
1040 def VST3q8_UPD : VST3DWB<0b0101, {0,0,0,?}, "8">;
1041 def VST3q16_UPD : VST3DWB<0b0101, {0,1,0,?}, "16">;
1042 def VST3q32_UPD : VST3DWB<0b0101, {1,0,0,?}, "32">;
1044 def VST3q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1045 def VST3q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1046 def VST3q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1048 // ...alternate versions to be allocated odd register numbers:
1049 def VST3q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1050 def VST3q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1051 def VST3q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1053 // VST4 : Vector Store (multiple 4-element structures)
1054 class VST4D<bits<4> op11_8, bits<4> op7_4, string Dt>
1055 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1056 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
1057 IIC_VST4, "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn",
1060 let Inst{5-4} = Rn{5-4};
1063 def VST4d8 : VST4D<0b0000, {0,0,?,?}, "8">;
1064 def VST4d16 : VST4D<0b0000, {0,1,?,?}, "16">;
1065 def VST4d32 : VST4D<0b0000, {1,0,?,?}, "32">;
1067 def VST4d8Pseudo : VSTQQPseudo<IIC_VST4>;
1068 def VST4d16Pseudo : VSTQQPseudo<IIC_VST4>;
1069 def VST4d32Pseudo : VSTQQPseudo<IIC_VST4>;
1071 // ...with address register writeback:
1072 class VST4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1073 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1074 (ins addrmode6:$Rn, am6offset:$Rm,
1075 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST4u,
1076 "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
1077 "$Rn.addr = $wb", []> {
1078 let Inst{5-4} = Rn{5-4};
1081 def VST4d8_UPD : VST4DWB<0b0000, {0,0,?,?}, "8">;
1082 def VST4d16_UPD : VST4DWB<0b0000, {0,1,?,?}, "16">;
1083 def VST4d32_UPD : VST4DWB<0b0000, {1,0,?,?}, "32">;
1085 def VST4d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1086 def VST4d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1087 def VST4d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1089 // ...with double-spaced registers (non-updating versions for disassembly only):
1090 def VST4q8 : VST4D<0b0001, {0,0,?,?}, "8">;
1091 def VST4q16 : VST4D<0b0001, {0,1,?,?}, "16">;
1092 def VST4q32 : VST4D<0b0001, {1,0,?,?}, "32">;
1093 def VST4q8_UPD : VST4DWB<0b0001, {0,0,?,?}, "8">;
1094 def VST4q16_UPD : VST4DWB<0b0001, {0,1,?,?}, "16">;
1095 def VST4q32_UPD : VST4DWB<0b0001, {1,0,?,?}, "32">;
1097 def VST4q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1098 def VST4q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1099 def VST4q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1101 // ...alternate versions to be allocated odd register numbers:
1102 def VST4q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1103 def VST4q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1104 def VST4q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1106 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
1108 // Classes for VST*LN pseudo-instructions with multi-register operands.
1109 // These are expanded to real instructions after register allocation.
1110 class VSTQLNPseudo<InstrItinClass itin>
1111 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
1113 class VSTQLNWBPseudo<InstrItinClass itin>
1114 : PseudoNLdSt<(outs GPR:$wb),
1115 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
1116 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1117 class VSTQQLNPseudo<InstrItinClass itin>
1118 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
1120 class VSTQQLNWBPseudo<InstrItinClass itin>
1121 : PseudoNLdSt<(outs GPR:$wb),
1122 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
1123 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1124 class VSTQQQQLNPseudo<InstrItinClass itin>
1125 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
1127 class VSTQQQQLNWBPseudo<InstrItinClass itin>
1128 : PseudoNLdSt<(outs GPR:$wb),
1129 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
1130 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1132 // VST1LN : Vector Store (single element from one lane)
1133 class VST1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1134 PatFrag StoreOp, SDNode ExtractOp>
1135 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1136 (ins addrmode6:$Rn, DPR:$Vd, nohash_imm:$lane),
1137 IIC_VST1ln, "vst1", Dt, "\\{$Vd[$lane]\\}, $Rn", "",
1138 [(StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane), addrmode6:$Rn)]> {
1141 class VST1QLNPseudo<ValueType Ty, PatFrag StoreOp, SDNode ExtractOp>
1142 : VSTQLNPseudo<IIC_VST1ln> {
1143 let Pattern = [(StoreOp (ExtractOp (Ty QPR:$src), imm:$lane),
1147 def VST1LNd8 : VST1LN<0b0000, {?,?,?,0}, "8", v8i8, truncstorei8,
1149 let Inst{7-5} = lane{2-0};
1151 def VST1LNd16 : VST1LN<0b0100, {?,?,0,?}, "16", v4i16, truncstorei16,
1153 let Inst{7-6} = lane{1-0};
1154 let Inst{4} = Rn{5};
1156 def VST1LNd32 : VST1LN<0b1000, {?,0,?,?}, "32", v2i32, store, extractelt> {
1157 let Inst{7} = lane{0};
1158 let Inst{5-4} = Rn{5-4};
1161 def VST1LNq8Pseudo : VST1QLNPseudo<v16i8, truncstorei8, NEONvgetlaneu>;
1162 def VST1LNq16Pseudo : VST1QLNPseudo<v8i16, truncstorei16, NEONvgetlaneu>;
1163 def VST1LNq32Pseudo : VST1QLNPseudo<v4i32, store, extractelt>;
1165 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
1167 // ...with address register writeback:
1168 class VST1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1169 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1170 (ins addrmode6:$Rn, am6offset:$Rm,
1171 DPR:$Vd, nohash_imm:$lane), IIC_VST1lnu, "vst1", Dt,
1172 "\\{$Vd[$lane]\\}, $Rn$Rm",
1173 "$Rn.addr = $wb", []>;
1175 def VST1LNd8_UPD : VST1LNWB<0b0000, {?,?,?,0}, "8"> {
1176 let Inst{7-5} = lane{2-0};
1178 def VST1LNd16_UPD : VST1LNWB<0b0100, {?,?,0,?}, "16"> {
1179 let Inst{7-6} = lane{1-0};
1180 let Inst{4} = Rn{5};
1182 def VST1LNd32_UPD : VST1LNWB<0b1000, {?,0,?,?}, "32"> {
1183 let Inst{7} = lane{0};
1184 let Inst{5-4} = Rn{5-4};
1187 def VST1LNq8Pseudo_UPD : VSTQLNWBPseudo<IIC_VST1lnu>;
1188 def VST1LNq16Pseudo_UPD : VSTQLNWBPseudo<IIC_VST1lnu>;
1189 def VST1LNq32Pseudo_UPD : VSTQLNWBPseudo<IIC_VST1lnu>;
1191 // VST2LN : Vector Store (single 2-element structure from one lane)
1192 class VST2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1193 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1194 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, nohash_imm:$lane),
1195 IIC_VST2ln, "vst2", Dt, "\\{$Vd[$lane], $src2[$lane]\\}, $Rn",
1198 let Inst{4} = Rn{4};
1201 def VST2LNd8 : VST2LN<0b0001, {?,?,?,?}, "8"> {
1202 let Inst{7-5} = lane{2-0};
1204 def VST2LNd16 : VST2LN<0b0101, {?,?,0,?}, "16"> {
1205 let Inst{7-6} = lane{1-0};
1207 def VST2LNd32 : VST2LN<0b1001, {?,0,0,?}, "32"> {
1208 let Inst{7} = lane{0};
1211 def VST2LNd8Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1212 def VST2LNd16Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1213 def VST2LNd32Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1215 // ...with double-spaced registers:
1216 def VST2LNq16 : VST2LN<0b0101, {?,?,1,?}, "16"> {
1217 let Inst{7-6} = lane{1-0};
1218 let Inst{4} = Rn{4};
1220 def VST2LNq32 : VST2LN<0b1001, {?,1,0,?}, "32"> {
1221 let Inst{7} = lane{0};
1222 let Inst{4} = Rn{4};
1225 def VST2LNq16Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
1226 def VST2LNq32Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
1228 // ...with address register writeback:
1229 class VST2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1230 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1231 (ins addrmode6:$addr, am6offset:$offset,
1232 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VST2lnu, "vst2", Dt,
1233 "\\{$src1[$lane], $src2[$lane]\\}, $addr$offset",
1234 "$addr.addr = $wb", []> {
1235 let Inst{4} = Rn{4};
1238 def VST2LNd8_UPD : VST2LNWB<0b0001, {?,?,?,?}, "8"> {
1239 let Inst{7-5} = lane{2-0};
1241 def VST2LNd16_UPD : VST2LNWB<0b0101, {?,?,0,?}, "16"> {
1242 let Inst{7-6} = lane{1-0};
1244 def VST2LNd32_UPD : VST2LNWB<0b1001, {?,0,0,?}, "32"> {
1245 let Inst{7} = lane{0};
1248 def VST2LNd8Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1249 def VST2LNd16Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1250 def VST2LNd32Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1252 def VST2LNq16_UPD : VST2LNWB<0b0101, {?,?,1,?}, "16"> {
1253 let Inst{7-6} = lane{1-0};
1255 def VST2LNq32_UPD : VST2LNWB<0b1001, {?,1,0,?}, "32"> {
1256 let Inst{7} = lane{0};
1259 def VST2LNq16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
1260 def VST2LNq32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
1262 // VST3LN : Vector Store (single 3-element structure from one lane)
1263 class VST3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1264 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1265 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3,
1266 nohash_imm:$lane), IIC_VST3ln, "vst3", Dt,
1267 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn", "", []> {
1271 def VST3LNd8 : VST3LN<0b0010, {?,?,?,0}, "8"> {
1272 let Inst{7-5} = lane{2-0};
1274 def VST3LNd16 : VST3LN<0b0110, {?,?,0,0}, "16"> {
1275 let Inst{7-6} = lane{1-0};
1277 def VST3LNd32 : VST3LN<0b1010, {?,0,0,0}, "32"> {
1278 let Inst{7} = lane{0};
1281 def VST3LNd8Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1282 def VST3LNd16Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1283 def VST3LNd32Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1285 // ...with double-spaced registers:
1286 def VST3LNq16 : VST3LN<0b0110, {?,?,1,0}, "16"> {
1287 let Inst{7-6} = lane{1-0};
1289 def VST3LNq32 : VST3LN<0b1010, {?,1,0,0}, "32"> {
1290 let Inst{7} = lane{0};
1293 def VST3LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
1294 def VST3LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
1296 // ...with address register writeback:
1297 class VST3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1298 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1299 (ins addrmode6:$Rn, am6offset:$Rm,
1300 DPR:$Vd, DPR:$src2, DPR:$src3, nohash_imm:$lane),
1301 IIC_VST3lnu, "vst3", Dt,
1302 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn$Rm",
1303 "$Rn.addr = $wb", []>;
1305 def VST3LNd8_UPD : VST3LNWB<0b0010, {?,?,?,0}, "8"> {
1306 let Inst{7-5} = lane{2-0};
1308 def VST3LNd16_UPD : VST3LNWB<0b0110, {?,?,0,0}, "16"> {
1309 let Inst{7-6} = lane{1-0};
1311 def VST3LNd32_UPD : VST3LNWB<0b1010, {?,0,0,0}, "32"> {
1312 let Inst{7} = lane{0};
1315 def VST3LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1316 def VST3LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1317 def VST3LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1319 def VST3LNq16_UPD : VST3LNWB<0b0110, {?,?,1,0}, "16"> {
1320 let Inst{7-6} = lane{1-0};
1322 def VST3LNq32_UPD : VST3LNWB<0b1010, {?,1,0,0}, "32"> {
1323 let Inst{7} = lane{0};
1326 def VST3LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
1327 def VST3LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
1329 // VST4LN : Vector Store (single 4-element structure from one lane)
1330 class VST4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1331 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1332 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4,
1333 nohash_imm:$lane), IIC_VST4ln, "vst4", Dt,
1334 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn",
1337 let Inst{4} = Rn{4};
1340 def VST4LNd8 : VST4LN<0b0011, {?,?,?,?}, "8"> {
1341 let Inst{7-5} = lane{2-0};
1343 def VST4LNd16 : VST4LN<0b0111, {?,?,0,?}, "16"> {
1344 let Inst{7-6} = lane{1-0};
1346 def VST4LNd32 : VST4LN<0b1011, {?,0,?,?}, "32"> {
1347 let Inst{7} = lane{0};
1348 let Inst{5} = Rn{5};
1351 def VST4LNd8Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1352 def VST4LNd16Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1353 def VST4LNd32Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1355 // ...with double-spaced registers:
1356 def VST4LNq16 : VST4LN<0b0111, {?,?,1,?}, "16"> {
1357 let Inst{7-6} = lane{1-0};
1359 def VST4LNq32 : VST4LN<0b1011, {?,1,?,?}, "32"> {
1360 let Inst{7} = lane{0};
1361 let Inst{5} = Rn{5};
1364 def VST4LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
1365 def VST4LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
1367 // ...with address register writeback:
1368 class VST4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1369 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1370 (ins addrmode6:$Rn, am6offset:$Rm,
1371 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
1372 IIC_VST4lnu, "vst4", Dt,
1373 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn$Rm",
1374 "$Rn.addr = $wb", []> {
1375 let Inst{4} = Rn{4};
1378 def VST4LNd8_UPD : VST4LNWB<0b0011, {?,?,?,?}, "8"> {
1379 let Inst{7-5} = lane{2-0};
1381 def VST4LNd16_UPD : VST4LNWB<0b0111, {?,?,0,?}, "16"> {
1382 let Inst{7-6} = lane{1-0};
1384 def VST4LNd32_UPD : VST4LNWB<0b1011, {?,0,?,?}, "32"> {
1385 let Inst{7} = lane{0};
1386 let Inst{5} = Rn{5};
1389 def VST4LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
1390 def VST4LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
1391 def VST4LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
1393 def VST4LNq16_UPD : VST4LNWB<0b0111, {?,?,1,?}, "16"> {
1394 let Inst{7-6} = lane{1-0};
1396 def VST4LNq32_UPD : VST4LNWB<0b1011, {?,1,?,?}, "32"> {
1397 let Inst{7} = lane{0};
1398 let Inst{5} = Rn{5};
1401 def VST4LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
1402 def VST4LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
1404 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
1407 //===----------------------------------------------------------------------===//
1408 // NEON pattern fragments
1409 //===----------------------------------------------------------------------===//
1411 // Extract D sub-registers of Q registers.
1412 def DSubReg_i8_reg : SDNodeXForm<imm, [{
1413 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1414 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/8, MVT::i32);
1416 def DSubReg_i16_reg : SDNodeXForm<imm, [{
1417 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1418 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/4, MVT::i32);
1420 def DSubReg_i32_reg : SDNodeXForm<imm, [{
1421 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1422 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/2, MVT::i32);
1424 def DSubReg_f64_reg : SDNodeXForm<imm, [{
1425 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1426 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue(), MVT::i32);
1429 // Extract S sub-registers of Q/D registers.
1430 def SSubReg_f32_reg : SDNodeXForm<imm, [{
1431 assert(ARM::ssub_3 == ARM::ssub_0+3 && "Unexpected subreg numbering");
1432 return CurDAG->getTargetConstant(ARM::ssub_0 + N->getZExtValue(), MVT::i32);
1435 // Translate lane numbers from Q registers to D subregs.
1436 def SubReg_i8_lane : SDNodeXForm<imm, [{
1437 return CurDAG->getTargetConstant(N->getZExtValue() & 7, MVT::i32);
1439 def SubReg_i16_lane : SDNodeXForm<imm, [{
1440 return CurDAG->getTargetConstant(N->getZExtValue() & 3, MVT::i32);
1442 def SubReg_i32_lane : SDNodeXForm<imm, [{
1443 return CurDAG->getTargetConstant(N->getZExtValue() & 1, MVT::i32);
1446 //===----------------------------------------------------------------------===//
1447 // Instruction Classes
1448 //===----------------------------------------------------------------------===//
1450 // Basic 2-register operations: single-, double- and quad-register.
1451 class N2VS<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1452 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
1453 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
1454 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
1455 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src),
1456 IIC_VUNAD, OpcodeStr, Dt, "$dst, $src", "", []>;
1457 class N2VD<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1458 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
1459 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
1460 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
1461 (ins DPR:$src), IIC_VUNAD, OpcodeStr, Dt,"$dst, $src", "",
1462 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src))))]>;
1463 class N2VQ<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1464 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
1465 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
1466 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
1467 (ins QPR:$src), IIC_VUNAQ, OpcodeStr, Dt,"$dst, $src", "",
1468 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src))))]>;
1470 // Basic 2-register intrinsics, both double- and quad-register.
1471 class N2VDInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1472 bits<2> op17_16, bits<5> op11_7, bit op4,
1473 InstrItinClass itin, string OpcodeStr, string Dt,
1474 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1475 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
1476 (ins DPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
1477 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
1478 class N2VQInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1479 bits<2> op17_16, bits<5> op11_7, bit op4,
1480 InstrItinClass itin, string OpcodeStr, string Dt,
1481 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1482 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
1483 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
1484 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
1486 // Narrow 2-register operations.
1487 class N2VN<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1488 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1489 InstrItinClass itin, string OpcodeStr, string Dt,
1490 ValueType TyD, ValueType TyQ, SDNode OpNode>
1491 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$dst),
1492 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
1493 [(set DPR:$dst, (TyD (OpNode (TyQ QPR:$src))))]>;
1495 // Narrow 2-register intrinsics.
1496 class N2VNInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1497 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1498 InstrItinClass itin, string OpcodeStr, string Dt,
1499 ValueType TyD, ValueType TyQ, Intrinsic IntOp>
1500 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$dst),
1501 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
1502 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src))))]>;
1504 // Long 2-register operations (currently only used for VMOVL).
1505 class N2VL<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1506 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1507 InstrItinClass itin, string OpcodeStr, string Dt,
1508 ValueType TyQ, ValueType TyD, SDNode OpNode>
1509 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$dst),
1510 (ins DPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
1511 [(set QPR:$dst, (TyQ (OpNode (TyD DPR:$src))))]>;
1513 // 2-register shuffles (VTRN/VZIP/VUZP), both double- and quad-register.
1514 class N2VDShuffle<bits<2> op19_18, bits<5> op11_7, string OpcodeStr, string Dt>
1515 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 0, 0, (outs DPR:$dst1, DPR:$dst2),
1516 (ins DPR:$src1, DPR:$src2), IIC_VPERMD,
1517 OpcodeStr, Dt, "$dst1, $dst2",
1518 "$src1 = $dst1, $src2 = $dst2", []>;
1519 class N2VQShuffle<bits<2> op19_18, bits<5> op11_7,
1520 InstrItinClass itin, string OpcodeStr, string Dt>
1521 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 1, 0, (outs QPR:$dst1, QPR:$dst2),
1522 (ins QPR:$src1, QPR:$src2), itin, OpcodeStr, Dt, "$dst1, $dst2",
1523 "$src1 = $dst1, $src2 = $dst2", []>;
1525 // Basic 3-register operations: single-, double- and quad-register.
1526 class N3VS<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1527 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1528 SDNode OpNode, bit Commutable>
1529 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1530 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src1, DPR_VFP2:$src2), N3RegFrm,
1531 IIC_VBIND, OpcodeStr, Dt, "$dst, $src1, $src2", "", []> {
1532 let isCommutable = Commutable;
1535 class N3VD<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1536 InstrItinClass itin, string OpcodeStr, string Dt,
1537 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
1538 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1539 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1540 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
1541 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
1542 let isCommutable = Commutable;
1544 // Same as N3VD but no data type.
1545 class N3VDX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1546 InstrItinClass itin, string OpcodeStr,
1547 ValueType ResTy, ValueType OpTy,
1548 SDNode OpNode, bit Commutable>
1549 : N3VX<op24, op23, op21_20, op11_8, 0, op4,
1550 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
1551 OpcodeStr, "$dst, $src1, $src2", "",
1552 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src1), (OpTy DPR:$src2))))]>{
1553 let isCommutable = Commutable;
1556 class N3VDSL<bits<2> op21_20, bits<4> op11_8,
1557 InstrItinClass itin, string OpcodeStr, string Dt,
1558 ValueType Ty, SDNode ShOp>
1559 : N3V<0, 1, op21_20, op11_8, 1, 0,
1560 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1561 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1562 [(set (Ty DPR:$dst),
1563 (Ty (ShOp (Ty DPR:$src1),
1564 (Ty (NEONvduplane (Ty DPR_VFP2:$src2),imm:$lane)))))]> {
1565 let isCommutable = 0;
1567 class N3VDSL16<bits<2> op21_20, bits<4> op11_8,
1568 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1569 : N3V<0, 1, op21_20, op11_8, 1, 0,
1570 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1571 NVMulSLFrm, IIC_VMULi16D, OpcodeStr, Dt,"$dst, $src1, $src2[$lane]","",
1572 [(set (Ty DPR:$dst),
1573 (Ty (ShOp (Ty DPR:$src1),
1574 (Ty (NEONvduplane (Ty DPR_8:$src2), imm:$lane)))))]> {
1575 let isCommutable = 0;
1578 class N3VQ<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1579 InstrItinClass itin, string OpcodeStr, string Dt,
1580 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
1581 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1582 (outs QPR:$Qd), (ins QPR:$Qn, QPR:$Qm), N3RegFrm, itin,
1583 OpcodeStr, Dt, "$Qd, $Qn, $Qm", "",
1584 [(set QPR:$Qd, (ResTy (OpNode (OpTy QPR:$Qn), (OpTy QPR:$Qm))))]> {
1585 let isCommutable = Commutable;
1587 class N3VQX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1588 InstrItinClass itin, string OpcodeStr,
1589 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
1590 : N3VX<op24, op23, op21_20, op11_8, 1, op4,
1591 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), N3RegFrm, itin,
1592 OpcodeStr, "$dst, $src1, $src2", "",
1593 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src1), (OpTy QPR:$src2))))]>{
1594 let isCommutable = Commutable;
1596 class N3VQSL<bits<2> op21_20, bits<4> op11_8,
1597 InstrItinClass itin, string OpcodeStr, string Dt,
1598 ValueType ResTy, ValueType OpTy, SDNode ShOp>
1599 : N3V<1, 1, op21_20, op11_8, 1, 0,
1600 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1601 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1602 [(set (ResTy QPR:$dst),
1603 (ResTy (ShOp (ResTy QPR:$src1),
1604 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1606 let isCommutable = 0;
1608 class N3VQSL16<bits<2> op21_20, bits<4> op11_8, string OpcodeStr, string Dt,
1609 ValueType ResTy, ValueType OpTy, SDNode ShOp>
1610 : N3V<1, 1, op21_20, op11_8, 1, 0,
1611 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1612 NVMulSLFrm, IIC_VMULi16Q, OpcodeStr, Dt,"$dst, $src1, $src2[$lane]","",
1613 [(set (ResTy QPR:$dst),
1614 (ResTy (ShOp (ResTy QPR:$src1),
1615 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
1617 let isCommutable = 0;
1620 // Basic 3-register intrinsics, both double- and quad-register.
1621 class N3VDInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1622 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1623 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
1624 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1625 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), f, itin,
1626 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
1627 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
1628 let isCommutable = Commutable;
1630 class N3VDIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1631 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
1632 : N3V<0, 1, op21_20, op11_8, 1, 0,
1633 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1634 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1635 [(set (Ty DPR:$dst),
1636 (Ty (IntOp (Ty DPR:$src1),
1637 (Ty (NEONvduplane (Ty DPR_VFP2:$src2),
1639 let isCommutable = 0;
1641 class N3VDIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1642 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
1643 : N3V<0, 1, op21_20, op11_8, 1, 0,
1644 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1645 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1646 [(set (Ty DPR:$dst),
1647 (Ty (IntOp (Ty DPR:$src1),
1648 (Ty (NEONvduplane (Ty DPR_8:$src2), imm:$lane)))))]> {
1649 let isCommutable = 0;
1651 class N3VDIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1652 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1653 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1654 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1655 (outs DPR:$Vd), (ins DPR:$Vm, DPR:$Vn), f, itin,
1656 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
1657 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (OpTy DPR:$Vn))))]> {
1658 let isCommutable = 0;
1661 class N3VQInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1662 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1663 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
1664 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1665 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), f, itin,
1666 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
1667 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
1668 let isCommutable = Commutable;
1670 class N3VQIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1671 string OpcodeStr, string Dt,
1672 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1673 : N3V<1, 1, op21_20, op11_8, 1, 0,
1674 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1675 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1676 [(set (ResTy QPR:$dst),
1677 (ResTy (IntOp (ResTy QPR:$src1),
1678 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1680 let isCommutable = 0;
1682 class N3VQIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1683 string OpcodeStr, string Dt,
1684 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1685 : N3V<1, 1, op21_20, op11_8, 1, 0,
1686 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1687 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1688 [(set (ResTy QPR:$dst),
1689 (ResTy (IntOp (ResTy QPR:$src1),
1690 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
1692 let isCommutable = 0;
1694 class N3VQIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1695 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1696 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1697 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1698 (outs QPR:$Vd), (ins QPR:$Vm, QPR:$Vn), f, itin,
1699 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
1700 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (OpTy QPR:$Vn))))]> {
1701 let isCommutable = 0;
1704 // Multiply-Add/Sub operations: single-, double- and quad-register.
1705 class N3VSMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1706 InstrItinClass itin, string OpcodeStr, string Dt,
1707 ValueType Ty, SDNode MulOp, SDNode OpNode>
1708 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1709 (outs DPR_VFP2:$dst),
1710 (ins DPR_VFP2:$src1, DPR_VFP2:$src2, DPR_VFP2:$src3), N3RegFrm, itin,
1711 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst", []>;
1713 class N3VDMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1714 InstrItinClass itin, string OpcodeStr, string Dt,
1715 ValueType Ty, SDNode MulOp, SDNode OpNode>
1716 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1717 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1718 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1719 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
1720 (Ty (MulOp DPR:$Vn, DPR:$Vm)))))]>;
1722 class N3VDMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1723 string OpcodeStr, string Dt,
1724 ValueType Ty, SDNode MulOp, SDNode ShOp>
1725 : N3V<0, 1, op21_20, op11_8, 1, 0,
1727 (ins DPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1729 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1730 [(set (Ty DPR:$dst),
1731 (Ty (ShOp (Ty DPR:$src1),
1732 (Ty (MulOp DPR:$src2,
1733 (Ty (NEONvduplane (Ty DPR_VFP2:$src3),
1735 class N3VDMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1736 string OpcodeStr, string Dt,
1737 ValueType Ty, SDNode MulOp, SDNode ShOp>
1738 : N3V<0, 1, op21_20, op11_8, 1, 0,
1740 (ins DPR:$src1, DPR:$Vn, DPR_8:$Vm, nohash_imm:$lane),
1742 OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "$src1 = $Vd",
1744 (Ty (ShOp (Ty DPR:$src1),
1746 (Ty (NEONvduplane (Ty DPR_8:$Vm),
1749 class N3VQMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1750 InstrItinClass itin, string OpcodeStr, string Dt, ValueType Ty,
1751 SDNode MulOp, SDNode OpNode>
1752 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1753 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
1754 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1755 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
1756 (Ty (MulOp QPR:$Vn, QPR:$Vm)))))]>;
1757 class N3VQMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1758 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1759 SDNode MulOp, SDNode ShOp>
1760 : N3V<1, 1, op21_20, op11_8, 1, 0,
1762 (ins QPR:$src1, QPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1764 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1765 [(set (ResTy QPR:$dst),
1766 (ResTy (ShOp (ResTy QPR:$src1),
1767 (ResTy (MulOp QPR:$src2,
1768 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src3),
1770 class N3VQMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1771 string OpcodeStr, string Dt,
1772 ValueType ResTy, ValueType OpTy,
1773 SDNode MulOp, SDNode ShOp>
1774 : N3V<1, 1, op21_20, op11_8, 1, 0,
1776 (ins QPR:$src1, QPR:$src2, DPR_8:$src3, nohash_imm:$lane),
1778 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1779 [(set (ResTy QPR:$dst),
1780 (ResTy (ShOp (ResTy QPR:$src1),
1781 (ResTy (MulOp QPR:$src2,
1782 (ResTy (NEONvduplane (OpTy DPR_8:$src3),
1785 // Neon Intrinsic-Op instructions (VABA): double- and quad-register.
1786 class N3VDIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1787 InstrItinClass itin, string OpcodeStr, string Dt,
1788 ValueType Ty, Intrinsic IntOp, SDNode OpNode>
1789 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1790 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1791 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1792 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
1793 (Ty (IntOp (Ty DPR:$Vn), (Ty DPR:$Vm))))))]>;
1794 class N3VQIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1795 InstrItinClass itin, string OpcodeStr, string Dt,
1796 ValueType Ty, Intrinsic IntOp, SDNode OpNode>
1797 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1798 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
1799 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1800 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
1801 (Ty (IntOp (Ty QPR:$Vn), (Ty QPR:$Vm))))))]>;
1803 // Neon 3-argument intrinsics, both double- and quad-register.
1804 // The destination register is also used as the first source operand register.
1805 class N3VDInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1806 InstrItinClass itin, string OpcodeStr, string Dt,
1807 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1808 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1809 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3), N3RegFrm, itin,
1810 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1811 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src1),
1812 (OpTy DPR:$src2), (OpTy DPR:$src3))))]>;
1813 class N3VQInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1814 InstrItinClass itin, string OpcodeStr, string Dt,
1815 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1816 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1817 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3), N3RegFrm, itin,
1818 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1819 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src1),
1820 (OpTy QPR:$src2), (OpTy QPR:$src3))))]>;
1822 // Long Multiply-Add/Sub operations.
1823 class N3VLMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1824 InstrItinClass itin, string OpcodeStr, string Dt,
1825 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
1826 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1827 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1828 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1829 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
1830 (TyQ (MulOp (TyD DPR:$Vn),
1831 (TyD DPR:$Vm)))))]>;
1832 class N3VLMulOpSL<bit op24, bits<2> op21_20, bits<4> op11_8,
1833 InstrItinClass itin, string OpcodeStr, string Dt,
1834 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
1835 : N3V<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$dst),
1836 (ins QPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1838 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1840 (OpNode (TyQ QPR:$src1),
1841 (TyQ (MulOp (TyD DPR:$src2),
1842 (TyD (NEONvduplane (TyD DPR_VFP2:$src3),
1844 class N3VLMulOpSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1845 InstrItinClass itin, string OpcodeStr, string Dt,
1846 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
1847 : N3V<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$dst),
1848 (ins QPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane),
1850 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1852 (OpNode (TyQ QPR:$src1),
1853 (TyQ (MulOp (TyD DPR:$src2),
1854 (TyD (NEONvduplane (TyD DPR_8:$src3),
1857 // Long Intrinsic-Op vector operations with explicit extend (VABAL).
1858 class N3VLIntExtOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1859 InstrItinClass itin, string OpcodeStr, string Dt,
1860 ValueType TyQ, ValueType TyD, Intrinsic IntOp, SDNode ExtOp,
1862 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1863 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1864 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1865 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
1866 (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
1867 (TyD DPR:$Vm)))))))]>;
1869 // Neon Long 3-argument intrinsic. The destination register is
1870 // a quad-register and is also used as the first source operand register.
1871 class N3VLInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1872 InstrItinClass itin, string OpcodeStr, string Dt,
1873 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
1874 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1875 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1876 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1878 (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$Vn), (TyD DPR:$Vm))))]>;
1879 class N3VLInt3SL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1880 string OpcodeStr, string Dt,
1881 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1882 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1884 (ins QPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1886 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1887 [(set (ResTy QPR:$dst),
1888 (ResTy (IntOp (ResTy QPR:$src1),
1890 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src3),
1892 class N3VLInt3SL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1893 InstrItinClass itin, string OpcodeStr, string Dt,
1894 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1895 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1897 (ins QPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane),
1899 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1900 [(set (ResTy QPR:$dst),
1901 (ResTy (IntOp (ResTy QPR:$src1),
1903 (OpTy (NEONvduplane (OpTy DPR_8:$src3),
1906 // Narrowing 3-register intrinsics.
1907 class N3VNInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1908 string OpcodeStr, string Dt, ValueType TyD, ValueType TyQ,
1909 Intrinsic IntOp, bit Commutable>
1910 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1911 (outs DPR:$dst), (ins QPR:$src1, QPR:$src2), N3RegFrm, IIC_VBINi4D,
1912 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1913 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src1), (TyQ QPR:$src2))))]> {
1914 let isCommutable = Commutable;
1917 // Long 3-register operations.
1918 class N3VL<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1919 InstrItinClass itin, string OpcodeStr, string Dt,
1920 ValueType TyQ, ValueType TyD, SDNode OpNode, bit Commutable>
1921 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1922 (outs QPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
1923 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1924 [(set QPR:$dst, (TyQ (OpNode (TyD DPR:$src1), (TyD DPR:$src2))))]> {
1925 let isCommutable = Commutable;
1927 class N3VLSL<bit op24, bits<2> op21_20, bits<4> op11_8,
1928 InstrItinClass itin, string OpcodeStr, string Dt,
1929 ValueType TyQ, ValueType TyD, SDNode OpNode>
1930 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1931 (outs QPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1932 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1934 (TyQ (OpNode (TyD DPR:$src1),
1935 (TyD (NEONvduplane (TyD DPR_VFP2:$src2),imm:$lane)))))]>;
1936 class N3VLSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1937 InstrItinClass itin, string OpcodeStr, string Dt,
1938 ValueType TyQ, ValueType TyD, SDNode OpNode>
1939 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1940 (outs QPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1941 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1943 (TyQ (OpNode (TyD DPR:$src1),
1944 (TyD (NEONvduplane (TyD DPR_8:$src2), imm:$lane)))))]>;
1946 // Long 3-register operations with explicitly extended operands.
1947 class N3VLExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1948 InstrItinClass itin, string OpcodeStr, string Dt,
1949 ValueType TyQ, ValueType TyD, SDNode OpNode, SDNode ExtOp,
1951 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1952 (outs QPR:$Qd), (ins DPR:$Dn, DPR:$Dm), N3RegFrm, itin,
1953 OpcodeStr, Dt, "$Qd, $Dn, $Dm", "",
1954 [(set QPR:$Qd, (OpNode (TyQ (ExtOp (TyD DPR:$Dn))),
1955 (TyQ (ExtOp (TyD DPR:$Dm)))))]> {
1956 let isCommutable = Commutable;
1959 // Long 3-register intrinsics with explicit extend (VABDL).
1960 class N3VLIntExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1961 InstrItinClass itin, string OpcodeStr, string Dt,
1962 ValueType TyQ, ValueType TyD, Intrinsic IntOp, SDNode ExtOp,
1964 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1965 (outs QPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
1966 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1967 [(set QPR:$dst, (TyQ (ExtOp (TyD (IntOp (TyD DPR:$src1),
1968 (TyD DPR:$src2))))))]> {
1969 let isCommutable = Commutable;
1972 // Long 3-register intrinsics.
1973 class N3VLInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1974 InstrItinClass itin, string OpcodeStr, string Dt,
1975 ValueType TyQ, ValueType TyD, Intrinsic IntOp, bit Commutable>
1976 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1977 (outs QPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
1978 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1979 [(set QPR:$dst, (TyQ (IntOp (TyD DPR:$src1), (TyD DPR:$src2))))]> {
1980 let isCommutable = Commutable;
1982 class N3VLIntSL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1983 string OpcodeStr, string Dt,
1984 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1985 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1986 (outs QPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1987 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1988 [(set (ResTy QPR:$dst),
1989 (ResTy (IntOp (OpTy DPR:$src1),
1990 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1992 class N3VLIntSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1993 InstrItinClass itin, string OpcodeStr, string Dt,
1994 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1995 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1996 (outs QPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1997 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1998 [(set (ResTy QPR:$dst),
1999 (ResTy (IntOp (OpTy DPR:$src1),
2000 (OpTy (NEONvduplane (OpTy DPR_8:$src2),
2003 // Wide 3-register operations.
2004 class N3VW<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2005 string OpcodeStr, string Dt, ValueType TyQ, ValueType TyD,
2006 SDNode OpNode, SDNode ExtOp, bit Commutable>
2007 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2008 (outs QPR:$Qd), (ins QPR:$Qn, DPR:$Dm), N3RegFrm, IIC_VSUBiD,
2009 OpcodeStr, Dt, "$Qd, $Qn, $Dm", "",
2010 [(set QPR:$Qd, (OpNode (TyQ QPR:$Qn),
2011 (TyQ (ExtOp (TyD DPR:$Dm)))))]> {
2012 let isCommutable = Commutable;
2015 // Pairwise long 2-register intrinsics, both double- and quad-register.
2016 class N2VDPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2017 bits<2> op17_16, bits<5> op11_7, bit op4,
2018 string OpcodeStr, string Dt,
2019 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2020 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
2021 (ins DPR:$src), IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
2022 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
2023 class N2VQPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2024 bits<2> op17_16, bits<5> op11_7, bit op4,
2025 string OpcodeStr, string Dt,
2026 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2027 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
2028 (ins QPR:$src), IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
2029 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
2031 // Pairwise long 2-register accumulate intrinsics,
2032 // both double- and quad-register.
2033 // The destination register is also used as the first source operand register.
2034 class N2VDPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2035 bits<2> op17_16, bits<5> op11_7, bit op4,
2036 string OpcodeStr, string Dt,
2037 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2038 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
2039 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vm), IIC_VPALiD,
2040 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
2041 [(set DPR:$Vd, (ResTy (IntOp (ResTy DPR:$src1), (OpTy DPR:$Vm))))]>;
2042 class N2VQPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2043 bits<2> op17_16, bits<5> op11_7, bit op4,
2044 string OpcodeStr, string Dt,
2045 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2046 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4,
2047 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vm), IIC_VPALiQ,
2048 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
2049 [(set QPR:$Vd, (ResTy (IntOp (ResTy QPR:$src1), (OpTy QPR:$Vm))))]>;
2051 // Shift by immediate,
2052 // both double- and quad-register.
2053 class N2VDSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2054 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2055 ValueType Ty, SDNode OpNode>
2056 : N2VImm<op24, op23, op11_8, op7, 0, op4,
2057 (outs DPR:$dst), (ins DPR:$src, i32imm:$SIMM), f, itin,
2058 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
2059 [(set DPR:$dst, (Ty (OpNode (Ty DPR:$src), (i32 imm:$SIMM))))]>;
2060 class N2VQSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2061 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2062 ValueType Ty, SDNode OpNode>
2063 : N2VImm<op24, op23, op11_8, op7, 1, op4,
2064 (outs QPR:$dst), (ins QPR:$src, i32imm:$SIMM), f, itin,
2065 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
2066 [(set QPR:$dst, (Ty (OpNode (Ty QPR:$src), (i32 imm:$SIMM))))]>;
2068 // Long shift by immediate.
2069 class N2VLSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
2070 string OpcodeStr, string Dt,
2071 ValueType ResTy, ValueType OpTy, SDNode OpNode>
2072 : N2VImm<op24, op23, op11_8, op7, op6, op4,
2073 (outs QPR:$dst), (ins DPR:$src, i32imm:$SIMM), N2RegVShLFrm,
2074 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src, $SIMM", "",
2075 [(set QPR:$dst, (ResTy (OpNode (OpTy DPR:$src),
2076 (i32 imm:$SIMM))))]>;
2078 // Narrow shift by immediate.
2079 class N2VNSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
2080 InstrItinClass itin, string OpcodeStr, string Dt,
2081 ValueType ResTy, ValueType OpTy, SDNode OpNode>
2082 : N2VImm<op24, op23, op11_8, op7, op6, op4,
2083 (outs DPR:$dst), (ins QPR:$src, i32imm:$SIMM), N2RegVShRFrm, itin,
2084 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
2085 [(set DPR:$dst, (ResTy (OpNode (OpTy QPR:$src),
2086 (i32 imm:$SIMM))))]>;
2088 // Shift right by immediate and accumulate,
2089 // both double- and quad-register.
2090 class N2VDShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2091 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
2092 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
2093 (ins DPR:$src1, DPR:$Vm, i32imm:$SIMM), N2RegVShRFrm, IIC_VPALiD,
2094 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2095 [(set DPR:$Vd, (Ty (add DPR:$src1,
2096 (Ty (ShOp DPR:$Vm, (i32 imm:$SIMM))))))]>;
2097 class N2VQShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2098 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
2099 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
2100 (ins QPR:$src1, QPR:$Vm, i32imm:$SIMM), N2RegVShRFrm, IIC_VPALiD,
2101 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2102 [(set QPR:$Vd, (Ty (add QPR:$src1,
2103 (Ty (ShOp QPR:$Vm, (i32 imm:$SIMM))))))]>;
2105 // Shift by immediate and insert,
2106 // both double- and quad-register.
2107 class N2VDShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2108 Format f, string OpcodeStr, string Dt, ValueType Ty,SDNode ShOp>
2109 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
2110 (ins DPR:$src1, DPR:$Vm, i32imm:$SIMM), f, IIC_VSHLiD,
2111 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2112 [(set DPR:$Vd, (Ty (ShOp DPR:$src1, DPR:$Vm, (i32 imm:$SIMM))))]>;
2113 class N2VQShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2114 Format f, string OpcodeStr, string Dt, ValueType Ty,SDNode ShOp>
2115 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
2116 (ins QPR:$src1, QPR:$Vm, i32imm:$SIMM), f, IIC_VSHLiQ,
2117 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2118 [(set QPR:$Vd, (Ty (ShOp QPR:$src1, QPR:$Vm, (i32 imm:$SIMM))))]>;
2120 // Convert, with fractional bits immediate,
2121 // both double- and quad-register.
2122 class N2VCvtD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2123 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2125 : N2VImm<op24, op23, op11_8, op7, 0, op4,
2126 (outs DPR:$Vd), (ins DPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
2127 IIC_VUNAD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2128 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (i32 imm:$SIMM))))]>;
2129 class N2VCvtQ<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2130 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2132 : N2VImm<op24, op23, op11_8, op7, 1, op4,
2133 (outs QPR:$Vd), (ins QPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
2134 IIC_VUNAQ, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2135 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (i32 imm:$SIMM))))]>;
2137 //===----------------------------------------------------------------------===//
2139 //===----------------------------------------------------------------------===//
2141 // Abbreviations used in multiclass suffixes:
2142 // Q = quarter int (8 bit) elements
2143 // H = half int (16 bit) elements
2144 // S = single int (32 bit) elements
2145 // D = double int (64 bit) elements
2147 // Neon 2-register vector operations -- for disassembly only.
2149 // First with only element sizes of 8, 16 and 32 bits:
2150 multiclass N2V_QHS_cmp<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2151 bits<5> op11_7, bit op4, string opc, string Dt,
2153 // 64-bit vector types.
2154 def v8i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 0, op4,
2155 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
2156 opc, !strconcat(Dt, "8"), asm, "", []>;
2157 def v4i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 0, op4,
2158 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
2159 opc, !strconcat(Dt, "16"), asm, "", []>;
2160 def v2i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
2161 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
2162 opc, !strconcat(Dt, "32"), asm, "", []>;
2163 def v2f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
2164 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
2165 opc, "f32", asm, "", []> {
2166 let Inst{10} = 1; // overwrite F = 1
2169 // 128-bit vector types.
2170 def v16i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 1, op4,
2171 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
2172 opc, !strconcat(Dt, "8"), asm, "", []>;
2173 def v8i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 1, op4,
2174 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
2175 opc, !strconcat(Dt, "16"), asm, "", []>;
2176 def v4i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
2177 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
2178 opc, !strconcat(Dt, "32"), asm, "", []>;
2179 def v4f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
2180 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
2181 opc, "f32", asm, "", []> {
2182 let Inst{10} = 1; // overwrite F = 1
2186 // Neon 3-register vector operations.
2188 // First with only element sizes of 8, 16 and 32 bits:
2189 multiclass N3V_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2190 InstrItinClass itinD16, InstrItinClass itinD32,
2191 InstrItinClass itinQ16, InstrItinClass itinQ32,
2192 string OpcodeStr, string Dt,
2193 SDNode OpNode, bit Commutable = 0> {
2194 // 64-bit vector types.
2195 def v8i8 : N3VD<op24, op23, 0b00, op11_8, op4, itinD16,
2196 OpcodeStr, !strconcat(Dt, "8"),
2197 v8i8, v8i8, OpNode, Commutable>;
2198 def v4i16 : N3VD<op24, op23, 0b01, op11_8, op4, itinD16,
2199 OpcodeStr, !strconcat(Dt, "16"),
2200 v4i16, v4i16, OpNode, Commutable>;
2201 def v2i32 : N3VD<op24, op23, 0b10, op11_8, op4, itinD32,
2202 OpcodeStr, !strconcat(Dt, "32"),
2203 v2i32, v2i32, OpNode, Commutable>;
2205 // 128-bit vector types.
2206 def v16i8 : N3VQ<op24, op23, 0b00, op11_8, op4, itinQ16,
2207 OpcodeStr, !strconcat(Dt, "8"),
2208 v16i8, v16i8, OpNode, Commutable>;
2209 def v8i16 : N3VQ<op24, op23, 0b01, op11_8, op4, itinQ16,
2210 OpcodeStr, !strconcat(Dt, "16"),
2211 v8i16, v8i16, OpNode, Commutable>;
2212 def v4i32 : N3VQ<op24, op23, 0b10, op11_8, op4, itinQ32,
2213 OpcodeStr, !strconcat(Dt, "32"),
2214 v4i32, v4i32, OpNode, Commutable>;
2217 multiclass N3VSL_HS<bits<4> op11_8, string OpcodeStr, string Dt, SDNode ShOp> {
2218 def v4i16 : N3VDSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
2220 def v2i32 : N3VDSL<0b10, op11_8, IIC_VMULi32D, OpcodeStr, !strconcat(Dt,"32"),
2222 def v8i16 : N3VQSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
2223 v8i16, v4i16, ShOp>;
2224 def v4i32 : N3VQSL<0b10, op11_8, IIC_VMULi32Q, OpcodeStr, !strconcat(Dt,"32"),
2225 v4i32, v2i32, ShOp>;
2228 // ....then also with element size 64 bits:
2229 multiclass N3V_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2230 InstrItinClass itinD, InstrItinClass itinQ,
2231 string OpcodeStr, string Dt,
2232 SDNode OpNode, bit Commutable = 0>
2233 : N3V_QHS<op24, op23, op11_8, op4, itinD, itinD, itinQ, itinQ,
2234 OpcodeStr, Dt, OpNode, Commutable> {
2235 def v1i64 : N3VD<op24, op23, 0b11, op11_8, op4, itinD,
2236 OpcodeStr, !strconcat(Dt, "64"),
2237 v1i64, v1i64, OpNode, Commutable>;
2238 def v2i64 : N3VQ<op24, op23, 0b11, op11_8, op4, itinQ,
2239 OpcodeStr, !strconcat(Dt, "64"),
2240 v2i64, v2i64, OpNode, Commutable>;
2244 // Neon Narrowing 2-register vector operations,
2245 // source operand element sizes of 16, 32 and 64 bits:
2246 multiclass N2VN_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2247 bits<5> op11_7, bit op6, bit op4,
2248 InstrItinClass itin, string OpcodeStr, string Dt,
2250 def v8i8 : N2VN<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
2251 itin, OpcodeStr, !strconcat(Dt, "16"),
2252 v8i8, v8i16, OpNode>;
2253 def v4i16 : N2VN<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
2254 itin, OpcodeStr, !strconcat(Dt, "32"),
2255 v4i16, v4i32, OpNode>;
2256 def v2i32 : N2VN<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
2257 itin, OpcodeStr, !strconcat(Dt, "64"),
2258 v2i32, v2i64, OpNode>;
2261 // Neon Narrowing 2-register vector intrinsics,
2262 // source operand element sizes of 16, 32 and 64 bits:
2263 multiclass N2VNInt_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2264 bits<5> op11_7, bit op6, bit op4,
2265 InstrItinClass itin, string OpcodeStr, string Dt,
2267 def v8i8 : N2VNInt<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
2268 itin, OpcodeStr, !strconcat(Dt, "16"),
2269 v8i8, v8i16, IntOp>;
2270 def v4i16 : N2VNInt<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
2271 itin, OpcodeStr, !strconcat(Dt, "32"),
2272 v4i16, v4i32, IntOp>;
2273 def v2i32 : N2VNInt<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
2274 itin, OpcodeStr, !strconcat(Dt, "64"),
2275 v2i32, v2i64, IntOp>;
2279 // Neon Lengthening 2-register vector intrinsic (currently specific to VMOVL).
2280 // source operand element sizes of 16, 32 and 64 bits:
2281 multiclass N2VL_QHS<bits<2> op24_23, bits<5> op11_7, bit op6, bit op4,
2282 string OpcodeStr, string Dt, SDNode OpNode> {
2283 def v8i16 : N2VL<op24_23, 0b00, 0b10, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2284 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode>;
2285 def v4i32 : N2VL<op24_23, 0b01, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2286 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
2287 def v2i64 : N2VL<op24_23, 0b10, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2288 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
2292 // Neon 3-register vector intrinsics.
2294 // First with only element sizes of 16 and 32 bits:
2295 multiclass N3VInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2296 InstrItinClass itinD16, InstrItinClass itinD32,
2297 InstrItinClass itinQ16, InstrItinClass itinQ32,
2298 string OpcodeStr, string Dt,
2299 Intrinsic IntOp, bit Commutable = 0> {
2300 // 64-bit vector types.
2301 def v4i16 : N3VDInt<op24, op23, 0b01, op11_8, op4, f, itinD16,
2302 OpcodeStr, !strconcat(Dt, "16"),
2303 v4i16, v4i16, IntOp, Commutable>;
2304 def v2i32 : N3VDInt<op24, op23, 0b10, op11_8, op4, f, itinD32,
2305 OpcodeStr, !strconcat(Dt, "32"),
2306 v2i32, v2i32, IntOp, Commutable>;
2308 // 128-bit vector types.
2309 def v8i16 : N3VQInt<op24, op23, 0b01, op11_8, op4, f, itinQ16,
2310 OpcodeStr, !strconcat(Dt, "16"),
2311 v8i16, v8i16, IntOp, Commutable>;
2312 def v4i32 : N3VQInt<op24, op23, 0b10, op11_8, op4, f, itinQ32,
2313 OpcodeStr, !strconcat(Dt, "32"),
2314 v4i32, v4i32, IntOp, Commutable>;
2316 multiclass N3VInt_HSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2317 InstrItinClass itinD16, InstrItinClass itinD32,
2318 InstrItinClass itinQ16, InstrItinClass itinQ32,
2319 string OpcodeStr, string Dt,
2321 // 64-bit vector types.
2322 def v4i16 : N3VDIntSh<op24, op23, 0b01, op11_8, op4, f, itinD16,
2323 OpcodeStr, !strconcat(Dt, "16"),
2324 v4i16, v4i16, IntOp>;
2325 def v2i32 : N3VDIntSh<op24, op23, 0b10, op11_8, op4, f, itinD32,
2326 OpcodeStr, !strconcat(Dt, "32"),
2327 v2i32, v2i32, IntOp>;
2329 // 128-bit vector types.
2330 def v8i16 : N3VQIntSh<op24, op23, 0b01, op11_8, op4, f, itinQ16,
2331 OpcodeStr, !strconcat(Dt, "16"),
2332 v8i16, v8i16, IntOp>;
2333 def v4i32 : N3VQIntSh<op24, op23, 0b10, op11_8, op4, f, itinQ32,
2334 OpcodeStr, !strconcat(Dt, "32"),
2335 v4i32, v4i32, IntOp>;
2338 multiclass N3VIntSL_HS<bits<4> op11_8,
2339 InstrItinClass itinD16, InstrItinClass itinD32,
2340 InstrItinClass itinQ16, InstrItinClass itinQ32,
2341 string OpcodeStr, string Dt, Intrinsic IntOp> {
2342 def v4i16 : N3VDIntSL16<0b01, op11_8, itinD16,
2343 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp>;
2344 def v2i32 : N3VDIntSL<0b10, op11_8, itinD32,
2345 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp>;
2346 def v8i16 : N3VQIntSL16<0b01, op11_8, itinQ16,
2347 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16, IntOp>;
2348 def v4i32 : N3VQIntSL<0b10, op11_8, itinQ32,
2349 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32, IntOp>;
2352 // ....then also with element size of 8 bits:
2353 multiclass N3VInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2354 InstrItinClass itinD16, InstrItinClass itinD32,
2355 InstrItinClass itinQ16, InstrItinClass itinQ32,
2356 string OpcodeStr, string Dt,
2357 Intrinsic IntOp, bit Commutable = 0>
2358 : N3VInt_HS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2359 OpcodeStr, Dt, IntOp, Commutable> {
2360 def v8i8 : N3VDInt<op24, op23, 0b00, op11_8, op4, f, itinD16,
2361 OpcodeStr, !strconcat(Dt, "8"),
2362 v8i8, v8i8, IntOp, Commutable>;
2363 def v16i8 : N3VQInt<op24, op23, 0b00, op11_8, op4, f, itinQ16,
2364 OpcodeStr, !strconcat(Dt, "8"),
2365 v16i8, v16i8, IntOp, Commutable>;
2367 multiclass N3VInt_QHSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2368 InstrItinClass itinD16, InstrItinClass itinD32,
2369 InstrItinClass itinQ16, InstrItinClass itinQ32,
2370 string OpcodeStr, string Dt,
2372 : N3VInt_HSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2373 OpcodeStr, Dt, IntOp> {
2374 def v8i8 : N3VDIntSh<op24, op23, 0b00, op11_8, op4, f, itinD16,
2375 OpcodeStr, !strconcat(Dt, "8"),
2377 def v16i8 : N3VQIntSh<op24, op23, 0b00, op11_8, op4, f, itinQ16,
2378 OpcodeStr, !strconcat(Dt, "8"),
2379 v16i8, v16i8, IntOp>;
2383 // ....then also with element size of 64 bits:
2384 multiclass N3VInt_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2385 InstrItinClass itinD16, InstrItinClass itinD32,
2386 InstrItinClass itinQ16, InstrItinClass itinQ32,
2387 string OpcodeStr, string Dt,
2388 Intrinsic IntOp, bit Commutable = 0>
2389 : N3VInt_QHS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2390 OpcodeStr, Dt, IntOp, Commutable> {
2391 def v1i64 : N3VDInt<op24, op23, 0b11, op11_8, op4, f, itinD32,
2392 OpcodeStr, !strconcat(Dt, "64"),
2393 v1i64, v1i64, IntOp, Commutable>;
2394 def v2i64 : N3VQInt<op24, op23, 0b11, op11_8, op4, f, itinQ32,
2395 OpcodeStr, !strconcat(Dt, "64"),
2396 v2i64, v2i64, IntOp, Commutable>;
2398 multiclass N3VInt_QHSDSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2399 InstrItinClass itinD16, InstrItinClass itinD32,
2400 InstrItinClass itinQ16, InstrItinClass itinQ32,
2401 string OpcodeStr, string Dt,
2403 : N3VInt_QHSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2404 OpcodeStr, Dt, IntOp> {
2405 def v1i64 : N3VDIntSh<op24, op23, 0b11, op11_8, op4, f, itinD32,
2406 OpcodeStr, !strconcat(Dt, "64"),
2407 v1i64, v1i64, IntOp>;
2408 def v2i64 : N3VQIntSh<op24, op23, 0b11, op11_8, op4, f, itinQ32,
2409 OpcodeStr, !strconcat(Dt, "64"),
2410 v2i64, v2i64, IntOp>;
2413 // Neon Narrowing 3-register vector intrinsics,
2414 // source operand element sizes of 16, 32 and 64 bits:
2415 multiclass N3VNInt_HSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2416 string OpcodeStr, string Dt,
2417 Intrinsic IntOp, bit Commutable = 0> {
2418 def v8i8 : N3VNInt<op24, op23, 0b00, op11_8, op4,
2419 OpcodeStr, !strconcat(Dt, "16"),
2420 v8i8, v8i16, IntOp, Commutable>;
2421 def v4i16 : N3VNInt<op24, op23, 0b01, op11_8, op4,
2422 OpcodeStr, !strconcat(Dt, "32"),
2423 v4i16, v4i32, IntOp, Commutable>;
2424 def v2i32 : N3VNInt<op24, op23, 0b10, op11_8, op4,
2425 OpcodeStr, !strconcat(Dt, "64"),
2426 v2i32, v2i64, IntOp, Commutable>;
2430 // Neon Long 3-register vector operations.
2432 multiclass N3VL_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2433 InstrItinClass itin16, InstrItinClass itin32,
2434 string OpcodeStr, string Dt,
2435 SDNode OpNode, bit Commutable = 0> {
2436 def v8i16 : N3VL<op24, op23, 0b00, op11_8, op4, itin16,
2437 OpcodeStr, !strconcat(Dt, "8"),
2438 v8i16, v8i8, OpNode, Commutable>;
2439 def v4i32 : N3VL<op24, op23, 0b01, op11_8, op4, itin16,
2440 OpcodeStr, !strconcat(Dt, "16"),
2441 v4i32, v4i16, OpNode, Commutable>;
2442 def v2i64 : N3VL<op24, op23, 0b10, op11_8, op4, itin32,
2443 OpcodeStr, !strconcat(Dt, "32"),
2444 v2i64, v2i32, OpNode, Commutable>;
2447 multiclass N3VLSL_HS<bit op24, bits<4> op11_8,
2448 InstrItinClass itin, string OpcodeStr, string Dt,
2450 def v4i16 : N3VLSL16<op24, 0b01, op11_8, itin, OpcodeStr,
2451 !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
2452 def v2i32 : N3VLSL<op24, 0b10, op11_8, itin, OpcodeStr,
2453 !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
2456 multiclass N3VLExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2457 InstrItinClass itin16, InstrItinClass itin32,
2458 string OpcodeStr, string Dt,
2459 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
2460 def v8i16 : N3VLExt<op24, op23, 0b00, op11_8, op4, itin16,
2461 OpcodeStr, !strconcat(Dt, "8"),
2462 v8i16, v8i8, OpNode, ExtOp, Commutable>;
2463 def v4i32 : N3VLExt<op24, op23, 0b01, op11_8, op4, itin16,
2464 OpcodeStr, !strconcat(Dt, "16"),
2465 v4i32, v4i16, OpNode, ExtOp, Commutable>;
2466 def v2i64 : N3VLExt<op24, op23, 0b10, op11_8, op4, itin32,
2467 OpcodeStr, !strconcat(Dt, "32"),
2468 v2i64, v2i32, OpNode, ExtOp, Commutable>;
2471 // Neon Long 3-register vector intrinsics.
2473 // First with only element sizes of 16 and 32 bits:
2474 multiclass N3VLInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
2475 InstrItinClass itin16, InstrItinClass itin32,
2476 string OpcodeStr, string Dt,
2477 Intrinsic IntOp, bit Commutable = 0> {
2478 def v4i32 : N3VLInt<op24, op23, 0b01, op11_8, op4, itin16,
2479 OpcodeStr, !strconcat(Dt, "16"),
2480 v4i32, v4i16, IntOp, Commutable>;
2481 def v2i64 : N3VLInt<op24, op23, 0b10, op11_8, op4, itin32,
2482 OpcodeStr, !strconcat(Dt, "32"),
2483 v2i64, v2i32, IntOp, Commutable>;
2486 multiclass N3VLIntSL_HS<bit op24, bits<4> op11_8,
2487 InstrItinClass itin, string OpcodeStr, string Dt,
2489 def v4i16 : N3VLIntSL16<op24, 0b01, op11_8, itin,
2490 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
2491 def v2i32 : N3VLIntSL<op24, 0b10, op11_8, itin,
2492 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
2495 // ....then also with element size of 8 bits:
2496 multiclass N3VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2497 InstrItinClass itin16, InstrItinClass itin32,
2498 string OpcodeStr, string Dt,
2499 Intrinsic IntOp, bit Commutable = 0>
2500 : N3VLInt_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt,
2501 IntOp, Commutable> {
2502 def v8i16 : N3VLInt<op24, op23, 0b00, op11_8, op4, itin16,
2503 OpcodeStr, !strconcat(Dt, "8"),
2504 v8i16, v8i8, IntOp, Commutable>;
2507 // ....with explicit extend (VABDL).
2508 multiclass N3VLIntExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2509 InstrItinClass itin, string OpcodeStr, string Dt,
2510 Intrinsic IntOp, SDNode ExtOp, bit Commutable = 0> {
2511 def v8i16 : N3VLIntExt<op24, op23, 0b00, op11_8, op4, itin,
2512 OpcodeStr, !strconcat(Dt, "8"),
2513 v8i16, v8i8, IntOp, ExtOp, Commutable>;
2514 def v4i32 : N3VLIntExt<op24, op23, 0b01, op11_8, op4, itin,
2515 OpcodeStr, !strconcat(Dt, "16"),
2516 v4i32, v4i16, IntOp, ExtOp, Commutable>;
2517 def v2i64 : N3VLIntExt<op24, op23, 0b10, op11_8, op4, itin,
2518 OpcodeStr, !strconcat(Dt, "32"),
2519 v2i64, v2i32, IntOp, ExtOp, Commutable>;
2523 // Neon Wide 3-register vector intrinsics,
2524 // source operand element sizes of 8, 16 and 32 bits:
2525 multiclass N3VW_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2526 string OpcodeStr, string Dt,
2527 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
2528 def v8i16 : N3VW<op24, op23, 0b00, op11_8, op4,
2529 OpcodeStr, !strconcat(Dt, "8"),
2530 v8i16, v8i8, OpNode, ExtOp, Commutable>;
2531 def v4i32 : N3VW<op24, op23, 0b01, op11_8, op4,
2532 OpcodeStr, !strconcat(Dt, "16"),
2533 v4i32, v4i16, OpNode, ExtOp, Commutable>;
2534 def v2i64 : N3VW<op24, op23, 0b10, op11_8, op4,
2535 OpcodeStr, !strconcat(Dt, "32"),
2536 v2i64, v2i32, OpNode, ExtOp, Commutable>;
2540 // Neon Multiply-Op vector operations,
2541 // element sizes of 8, 16 and 32 bits:
2542 multiclass N3VMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2543 InstrItinClass itinD16, InstrItinClass itinD32,
2544 InstrItinClass itinQ16, InstrItinClass itinQ32,
2545 string OpcodeStr, string Dt, SDNode OpNode> {
2546 // 64-bit vector types.
2547 def v8i8 : N3VDMulOp<op24, op23, 0b00, op11_8, op4, itinD16,
2548 OpcodeStr, !strconcat(Dt, "8"), v8i8, mul, OpNode>;
2549 def v4i16 : N3VDMulOp<op24, op23, 0b01, op11_8, op4, itinD16,
2550 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, OpNode>;
2551 def v2i32 : N3VDMulOp<op24, op23, 0b10, op11_8, op4, itinD32,
2552 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, OpNode>;
2554 // 128-bit vector types.
2555 def v16i8 : N3VQMulOp<op24, op23, 0b00, op11_8, op4, itinQ16,
2556 OpcodeStr, !strconcat(Dt, "8"), v16i8, mul, OpNode>;
2557 def v8i16 : N3VQMulOp<op24, op23, 0b01, op11_8, op4, itinQ16,
2558 OpcodeStr, !strconcat(Dt, "16"), v8i16, mul, OpNode>;
2559 def v4i32 : N3VQMulOp<op24, op23, 0b10, op11_8, op4, itinQ32,
2560 OpcodeStr, !strconcat(Dt, "32"), v4i32, mul, OpNode>;
2563 multiclass N3VMulOpSL_HS<bits<4> op11_8,
2564 InstrItinClass itinD16, InstrItinClass itinD32,
2565 InstrItinClass itinQ16, InstrItinClass itinQ32,
2566 string OpcodeStr, string Dt, SDNode ShOp> {
2567 def v4i16 : N3VDMulOpSL16<0b01, op11_8, itinD16,
2568 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, ShOp>;
2569 def v2i32 : N3VDMulOpSL<0b10, op11_8, itinD32,
2570 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, ShOp>;
2571 def v8i16 : N3VQMulOpSL16<0b01, op11_8, itinQ16,
2572 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16,
2574 def v4i32 : N3VQMulOpSL<0b10, op11_8, itinQ32,
2575 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32,
2579 // Neon Intrinsic-Op vector operations,
2580 // element sizes of 8, 16 and 32 bits:
2581 multiclass N3VIntOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2582 InstrItinClass itinD, InstrItinClass itinQ,
2583 string OpcodeStr, string Dt, Intrinsic IntOp,
2585 // 64-bit vector types.
2586 def v8i8 : N3VDIntOp<op24, op23, 0b00, op11_8, op4, itinD,
2587 OpcodeStr, !strconcat(Dt, "8"), v8i8, IntOp, OpNode>;
2588 def v4i16 : N3VDIntOp<op24, op23, 0b01, op11_8, op4, itinD,
2589 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp, OpNode>;
2590 def v2i32 : N3VDIntOp<op24, op23, 0b10, op11_8, op4, itinD,
2591 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp, OpNode>;
2593 // 128-bit vector types.
2594 def v16i8 : N3VQIntOp<op24, op23, 0b00, op11_8, op4, itinQ,
2595 OpcodeStr, !strconcat(Dt, "8"), v16i8, IntOp, OpNode>;
2596 def v8i16 : N3VQIntOp<op24, op23, 0b01, op11_8, op4, itinQ,
2597 OpcodeStr, !strconcat(Dt, "16"), v8i16, IntOp, OpNode>;
2598 def v4i32 : N3VQIntOp<op24, op23, 0b10, op11_8, op4, itinQ,
2599 OpcodeStr, !strconcat(Dt, "32"), v4i32, IntOp, OpNode>;
2602 // Neon 3-argument intrinsics,
2603 // element sizes of 8, 16 and 32 bits:
2604 multiclass N3VInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2605 InstrItinClass itinD, InstrItinClass itinQ,
2606 string OpcodeStr, string Dt, Intrinsic IntOp> {
2607 // 64-bit vector types.
2608 def v8i8 : N3VDInt3<op24, op23, 0b00, op11_8, op4, itinD,
2609 OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
2610 def v4i16 : N3VDInt3<op24, op23, 0b01, op11_8, op4, itinD,
2611 OpcodeStr, !strconcat(Dt, "16"), v4i16, v4i16, IntOp>;
2612 def v2i32 : N3VDInt3<op24, op23, 0b10, op11_8, op4, itinD,
2613 OpcodeStr, !strconcat(Dt, "32"), v2i32, v2i32, IntOp>;
2615 // 128-bit vector types.
2616 def v16i8 : N3VQInt3<op24, op23, 0b00, op11_8, op4, itinQ,
2617 OpcodeStr, !strconcat(Dt, "8"), v16i8, v16i8, IntOp>;
2618 def v8i16 : N3VQInt3<op24, op23, 0b01, op11_8, op4, itinQ,
2619 OpcodeStr, !strconcat(Dt, "16"), v8i16, v8i16, IntOp>;
2620 def v4i32 : N3VQInt3<op24, op23, 0b10, op11_8, op4, itinQ,
2621 OpcodeStr, !strconcat(Dt, "32"), v4i32, v4i32, IntOp>;
2625 // Neon Long Multiply-Op vector operations,
2626 // element sizes of 8, 16 and 32 bits:
2627 multiclass N3VLMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2628 InstrItinClass itin16, InstrItinClass itin32,
2629 string OpcodeStr, string Dt, SDNode MulOp,
2631 def v8i16 : N3VLMulOp<op24, op23, 0b00, op11_8, op4, itin16, OpcodeStr,
2632 !strconcat(Dt, "8"), v8i16, v8i8, MulOp, OpNode>;
2633 def v4i32 : N3VLMulOp<op24, op23, 0b01, op11_8, op4, itin16, OpcodeStr,
2634 !strconcat(Dt, "16"), v4i32, v4i16, MulOp, OpNode>;
2635 def v2i64 : N3VLMulOp<op24, op23, 0b10, op11_8, op4, itin32, OpcodeStr,
2636 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
2639 multiclass N3VLMulOpSL_HS<bit op24, bits<4> op11_8, string OpcodeStr,
2640 string Dt, SDNode MulOp, SDNode OpNode> {
2641 def v4i16 : N3VLMulOpSL16<op24, 0b01, op11_8, IIC_VMACi16D, OpcodeStr,
2642 !strconcat(Dt,"16"), v4i32, v4i16, MulOp, OpNode>;
2643 def v2i32 : N3VLMulOpSL<op24, 0b10, op11_8, IIC_VMACi32D, OpcodeStr,
2644 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
2648 // Neon Long 3-argument intrinsics.
2650 // First with only element sizes of 16 and 32 bits:
2651 multiclass N3VLInt3_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
2652 InstrItinClass itin16, InstrItinClass itin32,
2653 string OpcodeStr, string Dt, Intrinsic IntOp> {
2654 def v4i32 : N3VLInt3<op24, op23, 0b01, op11_8, op4, itin16,
2655 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
2656 def v2i64 : N3VLInt3<op24, op23, 0b10, op11_8, op4, itin32,
2657 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
2660 multiclass N3VLInt3SL_HS<bit op24, bits<4> op11_8,
2661 string OpcodeStr, string Dt, Intrinsic IntOp> {
2662 def v4i16 : N3VLInt3SL16<op24, 0b01, op11_8, IIC_VMACi16D,
2663 OpcodeStr, !strconcat(Dt,"16"), v4i32, v4i16, IntOp>;
2664 def v2i32 : N3VLInt3SL<op24, 0b10, op11_8, IIC_VMACi32D,
2665 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
2668 // ....then also with element size of 8 bits:
2669 multiclass N3VLInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2670 InstrItinClass itin16, InstrItinClass itin32,
2671 string OpcodeStr, string Dt, Intrinsic IntOp>
2672 : N3VLInt3_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt, IntOp> {
2673 def v8i16 : N3VLInt3<op24, op23, 0b00, op11_8, op4, itin16,
2674 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
2677 // ....with explicit extend (VABAL).
2678 multiclass N3VLIntExtOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2679 InstrItinClass itin, string OpcodeStr, string Dt,
2680 Intrinsic IntOp, SDNode ExtOp, SDNode OpNode> {
2681 def v8i16 : N3VLIntExtOp<op24, op23, 0b00, op11_8, op4, itin,
2682 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8,
2683 IntOp, ExtOp, OpNode>;
2684 def v4i32 : N3VLIntExtOp<op24, op23, 0b01, op11_8, op4, itin,
2685 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16,
2686 IntOp, ExtOp, OpNode>;
2687 def v2i64 : N3VLIntExtOp<op24, op23, 0b10, op11_8, op4, itin,
2688 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32,
2689 IntOp, ExtOp, OpNode>;
2693 // Neon 2-register vector intrinsics,
2694 // element sizes of 8, 16 and 32 bits:
2695 multiclass N2VInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2696 bits<5> op11_7, bit op4,
2697 InstrItinClass itinD, InstrItinClass itinQ,
2698 string OpcodeStr, string Dt, Intrinsic IntOp> {
2699 // 64-bit vector types.
2700 def v8i8 : N2VDInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2701 itinD, OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
2702 def v4i16 : N2VDInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2703 itinD, OpcodeStr, !strconcat(Dt, "16"),v4i16,v4i16,IntOp>;
2704 def v2i32 : N2VDInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2705 itinD, OpcodeStr, !strconcat(Dt, "32"),v2i32,v2i32,IntOp>;
2707 // 128-bit vector types.
2708 def v16i8 : N2VQInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2709 itinQ, OpcodeStr, !strconcat(Dt, "8"), v16i8,v16i8,IntOp>;
2710 def v8i16 : N2VQInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2711 itinQ, OpcodeStr, !strconcat(Dt, "16"),v8i16,v8i16,IntOp>;
2712 def v4i32 : N2VQInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2713 itinQ, OpcodeStr, !strconcat(Dt, "32"),v4i32,v4i32,IntOp>;
2717 // Neon Pairwise long 2-register intrinsics,
2718 // element sizes of 8, 16 and 32 bits:
2719 multiclass N2VPLInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2720 bits<5> op11_7, bit op4,
2721 string OpcodeStr, string Dt, Intrinsic IntOp> {
2722 // 64-bit vector types.
2723 def v8i8 : N2VDPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2724 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
2725 def v4i16 : N2VDPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2726 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
2727 def v2i32 : N2VDPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2728 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
2730 // 128-bit vector types.
2731 def v16i8 : N2VQPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2732 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
2733 def v8i16 : N2VQPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2734 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
2735 def v4i32 : N2VQPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2736 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
2740 // Neon Pairwise long 2-register accumulate intrinsics,
2741 // element sizes of 8, 16 and 32 bits:
2742 multiclass N2VPLInt2_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2743 bits<5> op11_7, bit op4,
2744 string OpcodeStr, string Dt, Intrinsic IntOp> {
2745 // 64-bit vector types.
2746 def v8i8 : N2VDPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2747 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
2748 def v4i16 : N2VDPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2749 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
2750 def v2i32 : N2VDPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2751 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
2753 // 128-bit vector types.
2754 def v16i8 : N2VQPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2755 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
2756 def v8i16 : N2VQPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2757 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
2758 def v4i32 : N2VQPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2759 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
2763 // Neon 2-register vector shift by immediate,
2764 // with f of either N2RegVShLFrm or N2RegVShRFrm
2765 // element sizes of 8, 16, 32 and 64 bits:
2766 multiclass N2VSh_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2767 InstrItinClass itin, string OpcodeStr, string Dt,
2768 SDNode OpNode, Format f> {
2769 // 64-bit vector types.
2770 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
2771 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
2772 let Inst{21-19} = 0b001; // imm6 = 001xxx
2774 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
2775 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
2776 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2778 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
2779 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
2780 let Inst{21} = 0b1; // imm6 = 1xxxxx
2782 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, f, itin,
2783 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
2786 // 128-bit vector types.
2787 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
2788 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
2789 let Inst{21-19} = 0b001; // imm6 = 001xxx
2791 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
2792 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
2793 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2795 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
2796 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
2797 let Inst{21} = 0b1; // imm6 = 1xxxxx
2799 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, f, itin,
2800 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
2804 // Neon Shift-Accumulate vector operations,
2805 // element sizes of 8, 16, 32 and 64 bits:
2806 multiclass N2VShAdd_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2807 string OpcodeStr, string Dt, SDNode ShOp> {
2808 // 64-bit vector types.
2809 def v8i8 : N2VDShAdd<op24, op23, op11_8, 0, op4,
2810 OpcodeStr, !strconcat(Dt, "8"), v8i8, ShOp> {
2811 let Inst{21-19} = 0b001; // imm6 = 001xxx
2813 def v4i16 : N2VDShAdd<op24, op23, op11_8, 0, op4,
2814 OpcodeStr, !strconcat(Dt, "16"), v4i16, ShOp> {
2815 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2817 def v2i32 : N2VDShAdd<op24, op23, op11_8, 0, op4,
2818 OpcodeStr, !strconcat(Dt, "32"), v2i32, ShOp> {
2819 let Inst{21} = 0b1; // imm6 = 1xxxxx
2821 def v1i64 : N2VDShAdd<op24, op23, op11_8, 1, op4,
2822 OpcodeStr, !strconcat(Dt, "64"), v1i64, ShOp>;
2825 // 128-bit vector types.
2826 def v16i8 : N2VQShAdd<op24, op23, op11_8, 0, op4,
2827 OpcodeStr, !strconcat(Dt, "8"), v16i8, ShOp> {
2828 let Inst{21-19} = 0b001; // imm6 = 001xxx
2830 def v8i16 : N2VQShAdd<op24, op23, op11_8, 0, op4,
2831 OpcodeStr, !strconcat(Dt, "16"), v8i16, ShOp> {
2832 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2834 def v4i32 : N2VQShAdd<op24, op23, op11_8, 0, op4,
2835 OpcodeStr, !strconcat(Dt, "32"), v4i32, ShOp> {
2836 let Inst{21} = 0b1; // imm6 = 1xxxxx
2838 def v2i64 : N2VQShAdd<op24, op23, op11_8, 1, op4,
2839 OpcodeStr, !strconcat(Dt, "64"), v2i64, ShOp>;
2844 // Neon Shift-Insert vector operations,
2845 // with f of either N2RegVShLFrm or N2RegVShRFrm
2846 // element sizes of 8, 16, 32 and 64 bits:
2847 multiclass N2VShIns_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2848 string OpcodeStr, SDNode ShOp,
2850 // 64-bit vector types.
2851 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4,
2852 f, OpcodeStr, "8", v8i8, ShOp> {
2853 let Inst{21-19} = 0b001; // imm6 = 001xxx
2855 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4,
2856 f, OpcodeStr, "16", v4i16, ShOp> {
2857 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2859 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4,
2860 f, OpcodeStr, "32", v2i32, ShOp> {
2861 let Inst{21} = 0b1; // imm6 = 1xxxxx
2863 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4,
2864 f, OpcodeStr, "64", v1i64, ShOp>;
2867 // 128-bit vector types.
2868 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4,
2869 f, OpcodeStr, "8", v16i8, ShOp> {
2870 let Inst{21-19} = 0b001; // imm6 = 001xxx
2872 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4,
2873 f, OpcodeStr, "16", v8i16, ShOp> {
2874 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2876 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4,
2877 f, OpcodeStr, "32", v4i32, ShOp> {
2878 let Inst{21} = 0b1; // imm6 = 1xxxxx
2880 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4,
2881 f, OpcodeStr, "64", v2i64, ShOp>;
2885 // Neon Shift Long operations,
2886 // element sizes of 8, 16, 32 bits:
2887 multiclass N2VLSh_QHS<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
2888 bit op4, string OpcodeStr, string Dt, SDNode OpNode> {
2889 def v8i16 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
2890 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode> {
2891 let Inst{21-19} = 0b001; // imm6 = 001xxx
2893 def v4i32 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
2894 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode> {
2895 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2897 def v2i64 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
2898 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode> {
2899 let Inst{21} = 0b1; // imm6 = 1xxxxx
2903 // Neon Shift Narrow operations,
2904 // element sizes of 16, 32, 64 bits:
2905 multiclass N2VNSh_HSD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
2906 bit op4, InstrItinClass itin, string OpcodeStr, string Dt,
2908 def v8i8 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
2909 OpcodeStr, !strconcat(Dt, "16"), v8i8, v8i16, OpNode> {
2910 let Inst{21-19} = 0b001; // imm6 = 001xxx
2912 def v4i16 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
2913 OpcodeStr, !strconcat(Dt, "32"), v4i16, v4i32, OpNode> {
2914 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2916 def v2i32 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
2917 OpcodeStr, !strconcat(Dt, "64"), v2i32, v2i64, OpNode> {
2918 let Inst{21} = 0b1; // imm6 = 1xxxxx
2922 //===----------------------------------------------------------------------===//
2923 // Instruction Definitions.
2924 //===----------------------------------------------------------------------===//
2926 // Vector Add Operations.
2928 // VADD : Vector Add (integer and floating-point)
2929 defm VADD : N3V_QHSD<0, 0, 0b1000, 0, IIC_VBINiD, IIC_VBINiQ, "vadd", "i",
2931 def VADDfd : N3VD<0, 0, 0b00, 0b1101, 0, IIC_VBIND, "vadd", "f32",
2932 v2f32, v2f32, fadd, 1>;
2933 def VADDfq : N3VQ<0, 0, 0b00, 0b1101, 0, IIC_VBINQ, "vadd", "f32",
2934 v4f32, v4f32, fadd, 1>;
2935 // VADDL : Vector Add Long (Q = D + D)
2936 defm VADDLs : N3VLExt_QHS<0,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
2937 "vaddl", "s", add, sext, 1>;
2938 defm VADDLu : N3VLExt_QHS<1,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
2939 "vaddl", "u", add, zext, 1>;
2940 // VADDW : Vector Add Wide (Q = Q + D)
2941 defm VADDWs : N3VW_QHS<0,1,0b0001,0, "vaddw", "s", add, sext, 0>;
2942 defm VADDWu : N3VW_QHS<1,1,0b0001,0, "vaddw", "u", add, zext, 0>;
2943 // VHADD : Vector Halving Add
2944 defm VHADDs : N3VInt_QHS<0, 0, 0b0000, 0, N3RegFrm,
2945 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2946 "vhadd", "s", int_arm_neon_vhadds, 1>;
2947 defm VHADDu : N3VInt_QHS<1, 0, 0b0000, 0, N3RegFrm,
2948 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2949 "vhadd", "u", int_arm_neon_vhaddu, 1>;
2950 // VRHADD : Vector Rounding Halving Add
2951 defm VRHADDs : N3VInt_QHS<0, 0, 0b0001, 0, N3RegFrm,
2952 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2953 "vrhadd", "s", int_arm_neon_vrhadds, 1>;
2954 defm VRHADDu : N3VInt_QHS<1, 0, 0b0001, 0, N3RegFrm,
2955 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2956 "vrhadd", "u", int_arm_neon_vrhaddu, 1>;
2957 // VQADD : Vector Saturating Add
2958 defm VQADDs : N3VInt_QHSD<0, 0, 0b0000, 1, N3RegFrm,
2959 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2960 "vqadd", "s", int_arm_neon_vqadds, 1>;
2961 defm VQADDu : N3VInt_QHSD<1, 0, 0b0000, 1, N3RegFrm,
2962 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2963 "vqadd", "u", int_arm_neon_vqaddu, 1>;
2964 // VADDHN : Vector Add and Narrow Returning High Half (D = Q + Q)
2965 defm VADDHN : N3VNInt_HSD<0,1,0b0100,0, "vaddhn", "i",
2966 int_arm_neon_vaddhn, 1>;
2967 // VRADDHN : Vector Rounding Add and Narrow Returning High Half (D = Q + Q)
2968 defm VRADDHN : N3VNInt_HSD<1,1,0b0100,0, "vraddhn", "i",
2969 int_arm_neon_vraddhn, 1>;
2971 // Vector Multiply Operations.
2973 // VMUL : Vector Multiply (integer, polynomial and floating-point)
2974 defm VMUL : N3V_QHS<0, 0, 0b1001, 1, IIC_VMULi16D, IIC_VMULi32D,
2975 IIC_VMULi16Q, IIC_VMULi32Q, "vmul", "i", mul, 1>;
2976 def VMULpd : N3VDInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16D, "vmul",
2977 "p8", v8i8, v8i8, int_arm_neon_vmulp, 1>;
2978 def VMULpq : N3VQInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16Q, "vmul",
2979 "p8", v16i8, v16i8, int_arm_neon_vmulp, 1>;
2980 def VMULfd : N3VD<1, 0, 0b00, 0b1101, 1, IIC_VFMULD, "vmul", "f32",
2981 v2f32, v2f32, fmul, 1>;
2982 def VMULfq : N3VQ<1, 0, 0b00, 0b1101, 1, IIC_VFMULQ, "vmul", "f32",
2983 v4f32, v4f32, fmul, 1>;
2984 defm VMULsl : N3VSL_HS<0b1000, "vmul", "i", mul>;
2985 def VMULslfd : N3VDSL<0b10, 0b1001, IIC_VBIND, "vmul", "f32", v2f32, fmul>;
2986 def VMULslfq : N3VQSL<0b10, 0b1001, IIC_VBINQ, "vmul", "f32", v4f32,
2989 def : Pat<(v8i16 (mul (v8i16 QPR:$src1),
2990 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
2991 (v8i16 (VMULslv8i16 (v8i16 QPR:$src1),
2992 (v4i16 (EXTRACT_SUBREG QPR:$src2,
2993 (DSubReg_i16_reg imm:$lane))),
2994 (SubReg_i16_lane imm:$lane)))>;
2995 def : Pat<(v4i32 (mul (v4i32 QPR:$src1),
2996 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
2997 (v4i32 (VMULslv4i32 (v4i32 QPR:$src1),
2998 (v2i32 (EXTRACT_SUBREG QPR:$src2,
2999 (DSubReg_i32_reg imm:$lane))),
3000 (SubReg_i32_lane imm:$lane)))>;
3001 def : Pat<(v4f32 (fmul (v4f32 QPR:$src1),
3002 (v4f32 (NEONvduplane (v4f32 QPR:$src2), imm:$lane)))),
3003 (v4f32 (VMULslfq (v4f32 QPR:$src1),
3004 (v2f32 (EXTRACT_SUBREG QPR:$src2,
3005 (DSubReg_i32_reg imm:$lane))),
3006 (SubReg_i32_lane imm:$lane)))>;
3008 // VQDMULH : Vector Saturating Doubling Multiply Returning High Half
3009 defm VQDMULH : N3VInt_HS<0, 0, 0b1011, 0, N3RegFrm, IIC_VMULi16D, IIC_VMULi32D,
3010 IIC_VMULi16Q, IIC_VMULi32Q,
3011 "vqdmulh", "s", int_arm_neon_vqdmulh, 1>;
3012 defm VQDMULHsl: N3VIntSL_HS<0b1100, IIC_VMULi16D, IIC_VMULi32D,
3013 IIC_VMULi16Q, IIC_VMULi32Q,
3014 "vqdmulh", "s", int_arm_neon_vqdmulh>;
3015 def : Pat<(v8i16 (int_arm_neon_vqdmulh (v8i16 QPR:$src1),
3016 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
3018 (v8i16 (VQDMULHslv8i16 (v8i16 QPR:$src1),
3019 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3020 (DSubReg_i16_reg imm:$lane))),
3021 (SubReg_i16_lane imm:$lane)))>;
3022 def : Pat<(v4i32 (int_arm_neon_vqdmulh (v4i32 QPR:$src1),
3023 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
3025 (v4i32 (VQDMULHslv4i32 (v4i32 QPR:$src1),
3026 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3027 (DSubReg_i32_reg imm:$lane))),
3028 (SubReg_i32_lane imm:$lane)))>;
3030 // VQRDMULH : Vector Rounding Saturating Doubling Multiply Returning High Half
3031 defm VQRDMULH : N3VInt_HS<1, 0, 0b1011, 0, N3RegFrm,
3032 IIC_VMULi16D,IIC_VMULi32D,IIC_VMULi16Q,IIC_VMULi32Q,
3033 "vqrdmulh", "s", int_arm_neon_vqrdmulh, 1>;
3034 defm VQRDMULHsl : N3VIntSL_HS<0b1101, IIC_VMULi16D, IIC_VMULi32D,
3035 IIC_VMULi16Q, IIC_VMULi32Q,
3036 "vqrdmulh", "s", int_arm_neon_vqrdmulh>;
3037 def : Pat<(v8i16 (int_arm_neon_vqrdmulh (v8i16 QPR:$src1),
3038 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
3040 (v8i16 (VQRDMULHslv8i16 (v8i16 QPR:$src1),
3041 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3042 (DSubReg_i16_reg imm:$lane))),
3043 (SubReg_i16_lane imm:$lane)))>;
3044 def : Pat<(v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$src1),
3045 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
3047 (v4i32 (VQRDMULHslv4i32 (v4i32 QPR:$src1),
3048 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3049 (DSubReg_i32_reg imm:$lane))),
3050 (SubReg_i32_lane imm:$lane)))>;
3052 // VMULL : Vector Multiply Long (integer and polynomial) (Q = D * D)
3053 defm VMULLs : N3VL_QHS<0,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
3054 "vmull", "s", NEONvmulls, 1>;
3055 defm VMULLu : N3VL_QHS<1,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
3056 "vmull", "u", NEONvmullu, 1>;
3057 def VMULLp : N3VLInt<0, 1, 0b00, 0b1110, 0, IIC_VMULi16D, "vmull", "p8",
3058 v8i16, v8i8, int_arm_neon_vmullp, 1>;
3059 defm VMULLsls : N3VLSL_HS<0, 0b1010, IIC_VMULi16D, "vmull", "s", NEONvmulls>;
3060 defm VMULLslu : N3VLSL_HS<1, 0b1010, IIC_VMULi16D, "vmull", "u", NEONvmullu>;
3062 // VQDMULL : Vector Saturating Doubling Multiply Long (Q = D * D)
3063 defm VQDMULL : N3VLInt_HS<0,1,0b1101,0, IIC_VMULi16D, IIC_VMULi32D,
3064 "vqdmull", "s", int_arm_neon_vqdmull, 1>;
3065 defm VQDMULLsl: N3VLIntSL_HS<0, 0b1011, IIC_VMULi16D,
3066 "vqdmull", "s", int_arm_neon_vqdmull>;
3068 // Vector Multiply-Accumulate and Multiply-Subtract Operations.
3070 // VMLA : Vector Multiply Accumulate (integer and floating-point)
3071 defm VMLA : N3VMulOp_QHS<0, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3072 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
3073 def VMLAfd : N3VDMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACD, "vmla", "f32",
3075 def VMLAfq : N3VQMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACQ, "vmla", "f32",
3077 defm VMLAsl : N3VMulOpSL_HS<0b0000, IIC_VMACi16D, IIC_VMACi32D,
3078 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
3079 def VMLAslfd : N3VDMulOpSL<0b10, 0b0001, IIC_VMACD, "vmla", "f32",
3081 def VMLAslfq : N3VQMulOpSL<0b10, 0b0001, IIC_VMACQ, "vmla", "f32",
3082 v4f32, v2f32, fmul, fadd>;
3084 def : Pat<(v8i16 (add (v8i16 QPR:$src1),
3085 (mul (v8i16 QPR:$src2),
3086 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
3087 (v8i16 (VMLAslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
3088 (v4i16 (EXTRACT_SUBREG QPR:$src3,
3089 (DSubReg_i16_reg imm:$lane))),
3090 (SubReg_i16_lane imm:$lane)))>;
3092 def : Pat<(v4i32 (add (v4i32 QPR:$src1),
3093 (mul (v4i32 QPR:$src2),
3094 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
3095 (v4i32 (VMLAslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
3096 (v2i32 (EXTRACT_SUBREG QPR:$src3,
3097 (DSubReg_i32_reg imm:$lane))),
3098 (SubReg_i32_lane imm:$lane)))>;
3100 def : Pat<(v4f32 (fadd (v4f32 QPR:$src1),
3101 (fmul (v4f32 QPR:$src2),
3102 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
3103 (v4f32 (VMLAslfq (v4f32 QPR:$src1),
3105 (v2f32 (EXTRACT_SUBREG QPR:$src3,
3106 (DSubReg_i32_reg imm:$lane))),
3107 (SubReg_i32_lane imm:$lane)))>;
3109 // VMLAL : Vector Multiply Accumulate Long (Q += D * D)
3110 defm VMLALs : N3VLMulOp_QHS<0,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
3111 "vmlal", "s", NEONvmulls, add>;
3112 defm VMLALu : N3VLMulOp_QHS<1,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
3113 "vmlal", "u", NEONvmullu, add>;
3115 defm VMLALsls : N3VLMulOpSL_HS<0, 0b0010, "vmlal", "s", NEONvmulls, add>;
3116 defm VMLALslu : N3VLMulOpSL_HS<1, 0b0010, "vmlal", "u", NEONvmullu, add>;
3118 // VQDMLAL : Vector Saturating Doubling Multiply Accumulate Long (Q += D * D)
3119 defm VQDMLAL : N3VLInt3_HS<0, 1, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3120 "vqdmlal", "s", int_arm_neon_vqdmlal>;
3121 defm VQDMLALsl: N3VLInt3SL_HS<0, 0b0011, "vqdmlal", "s", int_arm_neon_vqdmlal>;
3123 // VMLS : Vector Multiply Subtract (integer and floating-point)
3124 defm VMLS : N3VMulOp_QHS<1, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3125 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
3126 def VMLSfd : N3VDMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACD, "vmls", "f32",
3128 def VMLSfq : N3VQMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACQ, "vmls", "f32",
3130 defm VMLSsl : N3VMulOpSL_HS<0b0100, IIC_VMACi16D, IIC_VMACi32D,
3131 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
3132 def VMLSslfd : N3VDMulOpSL<0b10, 0b0101, IIC_VMACD, "vmls", "f32",
3134 def VMLSslfq : N3VQMulOpSL<0b10, 0b0101, IIC_VMACQ, "vmls", "f32",
3135 v4f32, v2f32, fmul, fsub>;
3137 def : Pat<(v8i16 (sub (v8i16 QPR:$src1),
3138 (mul (v8i16 QPR:$src2),
3139 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
3140 (v8i16 (VMLSslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
3141 (v4i16 (EXTRACT_SUBREG QPR:$src3,
3142 (DSubReg_i16_reg imm:$lane))),
3143 (SubReg_i16_lane imm:$lane)))>;
3145 def : Pat<(v4i32 (sub (v4i32 QPR:$src1),
3146 (mul (v4i32 QPR:$src2),
3147 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
3148 (v4i32 (VMLSslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
3149 (v2i32 (EXTRACT_SUBREG QPR:$src3,
3150 (DSubReg_i32_reg imm:$lane))),
3151 (SubReg_i32_lane imm:$lane)))>;
3153 def : Pat<(v4f32 (fsub (v4f32 QPR:$src1),
3154 (fmul (v4f32 QPR:$src2),
3155 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
3156 (v4f32 (VMLSslfq (v4f32 QPR:$src1), (v4f32 QPR:$src2),
3157 (v2f32 (EXTRACT_SUBREG QPR:$src3,
3158 (DSubReg_i32_reg imm:$lane))),
3159 (SubReg_i32_lane imm:$lane)))>;
3161 // VMLSL : Vector Multiply Subtract Long (Q -= D * D)
3162 defm VMLSLs : N3VLMulOp_QHS<0,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
3163 "vmlsl", "s", NEONvmulls, sub>;
3164 defm VMLSLu : N3VLMulOp_QHS<1,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
3165 "vmlsl", "u", NEONvmullu, sub>;
3167 defm VMLSLsls : N3VLMulOpSL_HS<0, 0b0110, "vmlsl", "s", NEONvmulls, sub>;
3168 defm VMLSLslu : N3VLMulOpSL_HS<1, 0b0110, "vmlsl", "u", NEONvmullu, sub>;
3170 // VQDMLSL : Vector Saturating Doubling Multiply Subtract Long (Q -= D * D)
3171 defm VQDMLSL : N3VLInt3_HS<0, 1, 0b1011, 0, IIC_VMACi16D, IIC_VMACi32D,
3172 "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
3173 defm VQDMLSLsl: N3VLInt3SL_HS<0, 0b111, "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
3175 // Vector Subtract Operations.
3177 // VSUB : Vector Subtract (integer and floating-point)
3178 defm VSUB : N3V_QHSD<1, 0, 0b1000, 0, IIC_VSUBiD, IIC_VSUBiQ,
3179 "vsub", "i", sub, 0>;
3180 def VSUBfd : N3VD<0, 0, 0b10, 0b1101, 0, IIC_VBIND, "vsub", "f32",
3181 v2f32, v2f32, fsub, 0>;
3182 def VSUBfq : N3VQ<0, 0, 0b10, 0b1101, 0, IIC_VBINQ, "vsub", "f32",
3183 v4f32, v4f32, fsub, 0>;
3184 // VSUBL : Vector Subtract Long (Q = D - D)
3185 defm VSUBLs : N3VLExt_QHS<0,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
3186 "vsubl", "s", sub, sext, 0>;
3187 defm VSUBLu : N3VLExt_QHS<1,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
3188 "vsubl", "u", sub, zext, 0>;
3189 // VSUBW : Vector Subtract Wide (Q = Q - D)
3190 defm VSUBWs : N3VW_QHS<0,1,0b0011,0, "vsubw", "s", sub, sext, 0>;
3191 defm VSUBWu : N3VW_QHS<1,1,0b0011,0, "vsubw", "u", sub, zext, 0>;
3192 // VHSUB : Vector Halving Subtract
3193 defm VHSUBs : N3VInt_QHS<0, 0, 0b0010, 0, N3RegFrm,
3194 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3195 "vhsub", "s", int_arm_neon_vhsubs, 0>;
3196 defm VHSUBu : N3VInt_QHS<1, 0, 0b0010, 0, N3RegFrm,
3197 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3198 "vhsub", "u", int_arm_neon_vhsubu, 0>;
3199 // VQSUB : Vector Saturing Subtract
3200 defm VQSUBs : N3VInt_QHSD<0, 0, 0b0010, 1, N3RegFrm,
3201 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3202 "vqsub", "s", int_arm_neon_vqsubs, 0>;
3203 defm VQSUBu : N3VInt_QHSD<1, 0, 0b0010, 1, N3RegFrm,
3204 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3205 "vqsub", "u", int_arm_neon_vqsubu, 0>;
3206 // VSUBHN : Vector Subtract and Narrow Returning High Half (D = Q - Q)
3207 defm VSUBHN : N3VNInt_HSD<0,1,0b0110,0, "vsubhn", "i",
3208 int_arm_neon_vsubhn, 0>;
3209 // VRSUBHN : Vector Rounding Subtract and Narrow Returning High Half (D=Q-Q)
3210 defm VRSUBHN : N3VNInt_HSD<1,1,0b0110,0, "vrsubhn", "i",
3211 int_arm_neon_vrsubhn, 0>;
3213 // Vector Comparisons.
3215 // VCEQ : Vector Compare Equal
3216 defm VCEQ : N3V_QHS<1, 0, 0b1000, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3217 IIC_VSUBi4Q, "vceq", "i", NEONvceq, 1>;
3218 def VCEQfd : N3VD<0,0,0b00,0b1110,0, IIC_VBIND, "vceq", "f32", v2i32, v2f32,
3220 def VCEQfq : N3VQ<0,0,0b00,0b1110,0, IIC_VBINQ, "vceq", "f32", v4i32, v4f32,
3222 // For disassembly only.
3223 defm VCEQz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00010, 0, "vceq", "i",
3226 // VCGE : Vector Compare Greater Than or Equal
3227 defm VCGEs : N3V_QHS<0, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3228 IIC_VSUBi4Q, "vcge", "s", NEONvcge, 0>;
3229 defm VCGEu : N3V_QHS<1, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3230 IIC_VSUBi4Q, "vcge", "u", NEONvcgeu, 0>;
3231 def VCGEfd : N3VD<1,0,0b00,0b1110,0, IIC_VBIND, "vcge", "f32", v2i32, v2f32,
3233 def VCGEfq : N3VQ<1,0,0b00,0b1110,0, IIC_VBINQ, "vcge", "f32", v4i32, v4f32,
3235 // For disassembly only.
3236 // FIXME: This instruction's encoding MAY NOT BE correct.
3237 defm VCGEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00001, 0, "vcge", "s",
3239 // For disassembly only.
3240 // FIXME: This instruction's encoding MAY NOT BE correct.
3241 defm VCLEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00011, 0, "vcle", "s",
3244 // VCGT : Vector Compare Greater Than
3245 defm VCGTs : N3V_QHS<0, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3246 IIC_VSUBi4Q, "vcgt", "s", NEONvcgt, 0>;
3247 defm VCGTu : N3V_QHS<1, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3248 IIC_VSUBi4Q, "vcgt", "u", NEONvcgtu, 0>;
3249 def VCGTfd : N3VD<1,0,0b10,0b1110,0, IIC_VBIND, "vcgt", "f32", v2i32, v2f32,
3251 def VCGTfq : N3VQ<1,0,0b10,0b1110,0, IIC_VBINQ, "vcgt", "f32", v4i32, v4f32,
3253 // For disassembly only.
3254 // FIXME: This instruction's encoding MAY NOT BE correct.
3255 defm VCGTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00000, 0, "vcgt", "s",
3257 // For disassembly only.
3258 // FIXME: This instruction's encoding MAY NOT BE correct.
3259 defm VCLTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00100, 0, "vclt", "s",
3262 // VACGE : Vector Absolute Compare Greater Than or Equal (aka VCAGE)
3263 def VACGEd : N3VDInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacge",
3264 "f32", v2i32, v2f32, int_arm_neon_vacged, 0>;
3265 def VACGEq : N3VQInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacge",
3266 "f32", v4i32, v4f32, int_arm_neon_vacgeq, 0>;
3267 // VACGT : Vector Absolute Compare Greater Than (aka VCAGT)
3268 def VACGTd : N3VDInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacgt",
3269 "f32", v2i32, v2f32, int_arm_neon_vacgtd, 0>;
3270 def VACGTq : N3VQInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacgt",
3271 "f32", v4i32, v4f32, int_arm_neon_vacgtq, 0>;
3272 // VTST : Vector Test Bits
3273 defm VTST : N3V_QHS<0, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
3274 IIC_VBINi4Q, "vtst", "", NEONvtst, 1>;
3276 // Vector Bitwise Operations.
3278 def vnotd : PatFrag<(ops node:$in),
3279 (xor node:$in, (bitconvert (v8i8 NEONimmAllOnesV)))>;
3280 def vnotq : PatFrag<(ops node:$in),
3281 (xor node:$in, (bitconvert (v16i8 NEONimmAllOnesV)))>;
3284 // VAND : Vector Bitwise AND
3285 def VANDd : N3VDX<0, 0, 0b00, 0b0001, 1, IIC_VBINiD, "vand",
3286 v2i32, v2i32, and, 1>;
3287 def VANDq : N3VQX<0, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "vand",
3288 v4i32, v4i32, and, 1>;
3290 // VEOR : Vector Bitwise Exclusive OR
3291 def VEORd : N3VDX<1, 0, 0b00, 0b0001, 1, IIC_VBINiD, "veor",
3292 v2i32, v2i32, xor, 1>;
3293 def VEORq : N3VQX<1, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "veor",
3294 v4i32, v4i32, xor, 1>;
3296 // VORR : Vector Bitwise OR
3297 def VORRd : N3VDX<0, 0, 0b10, 0b0001, 1, IIC_VBINiD, "vorr",
3298 v2i32, v2i32, or, 1>;
3299 def VORRq : N3VQX<0, 0, 0b10, 0b0001, 1, IIC_VBINiQ, "vorr",
3300 v4i32, v4i32, or, 1>;
3302 def VORRiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 0, 1,
3303 (outs DPR:$Vd), (ins nModImm:$SIMM, DPR:$src),
3305 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
3307 (v4i16 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
3308 let Inst{9} = SIMM{9};
3311 def VORRiv2i32 : N1ModImm<1, 0b000, {?,?,?,1}, 0, 0, 0, 1,
3312 (outs DPR:$Vd), (ins nModImm:$SIMM, DPR:$src),
3314 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
3316 (v2i32 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
3317 let Inst{11-9} = SIMM{11-9};
3320 def VORRiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 0, 1,
3321 (outs QPR:$Vd), (ins nModImm:$SIMM, QPR:$src),
3323 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
3325 (v8i16 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
3326 let Inst{9} = SIMM{9};
3329 def VORRiv4i32 : N1ModImm<1, 0b000, {?,?,?,1}, 0, 1, 0, 1,
3330 (outs QPR:$Vd), (ins nModImm:$SIMM, QPR:$src),
3332 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
3334 (v4i32 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
3335 let Inst{11-9} = SIMM{11-9};
3339 // VBIC : Vector Bitwise Bit Clear (AND NOT)
3340 def VBICd : N3VX<0, 0, 0b01, 0b0001, 0, 1, (outs DPR:$dst),
3341 (ins DPR:$src1, DPR:$src2), N3RegFrm, IIC_VBINiD,
3342 "vbic", "$dst, $src1, $src2", "",
3343 [(set DPR:$dst, (v2i32 (and DPR:$src1,
3344 (vnotd DPR:$src2))))]>;
3345 def VBICq : N3VX<0, 0, 0b01, 0b0001, 1, 1, (outs QPR:$dst),
3346 (ins QPR:$src1, QPR:$src2), N3RegFrm, IIC_VBINiQ,
3347 "vbic", "$dst, $src1, $src2", "",
3348 [(set QPR:$dst, (v4i32 (and QPR:$src1,
3349 (vnotq QPR:$src2))))]>;
3351 // VORN : Vector Bitwise OR NOT
3352 def VORNd : N3VX<0, 0, 0b11, 0b0001, 0, 1, (outs DPR:$dst),
3353 (ins DPR:$src1, DPR:$src2), N3RegFrm, IIC_VBINiD,
3354 "vorn", "$dst, $src1, $src2", "",
3355 [(set DPR:$dst, (v2i32 (or DPR:$src1,
3356 (vnotd DPR:$src2))))]>;
3357 def VORNq : N3VX<0, 0, 0b11, 0b0001, 1, 1, (outs QPR:$dst),
3358 (ins QPR:$src1, QPR:$src2), N3RegFrm, IIC_VBINiQ,
3359 "vorn", "$dst, $src1, $src2", "",
3360 [(set QPR:$dst, (v4i32 (or QPR:$src1,
3361 (vnotq QPR:$src2))))]>;
3363 // VMVN : Vector Bitwise NOT (Immediate)
3365 let isReMaterializable = 1 in {
3367 def VMVNv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 1, 1, (outs DPR:$dst),
3368 (ins nModImm:$SIMM), IIC_VMOVImm,
3369 "vmvn", "i16", "$dst, $SIMM", "",
3370 [(set DPR:$dst, (v4i16 (NEONvmvnImm timm:$SIMM)))]> {
3371 let Inst{9} = SIMM{9};
3374 def VMVNv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 1, 1, (outs QPR:$dst),
3375 (ins nModImm:$SIMM), IIC_VMOVImm,
3376 "vmvn", "i16", "$dst, $SIMM", "",
3377 [(set QPR:$dst, (v8i16 (NEONvmvnImm timm:$SIMM)))]> {
3378 let Inst{9} = SIMM{9};
3381 def VMVNv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 1, 1, (outs DPR:$dst),
3382 (ins nModImm:$SIMM), IIC_VMOVImm,
3383 "vmvn", "i32", "$dst, $SIMM", "",
3384 [(set DPR:$dst, (v2i32 (NEONvmvnImm timm:$SIMM)))]> {
3385 let Inst{11-8} = SIMM{11-8};
3388 def VMVNv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 1, 1, (outs QPR:$dst),
3389 (ins nModImm:$SIMM), IIC_VMOVImm,
3390 "vmvn", "i32", "$dst, $SIMM", "",
3391 [(set QPR:$dst, (v4i32 (NEONvmvnImm timm:$SIMM)))]> {
3392 let Inst{11-8} = SIMM{11-8};
3396 // VMVN : Vector Bitwise NOT
3397 def VMVNd : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 0, 0,
3398 (outs DPR:$dst), (ins DPR:$src), IIC_VSUBiD,
3399 "vmvn", "$dst, $src", "",
3400 [(set DPR:$dst, (v2i32 (vnotd DPR:$src)))]>;
3401 def VMVNq : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 1, 0,
3402 (outs QPR:$dst), (ins QPR:$src), IIC_VSUBiD,
3403 "vmvn", "$dst, $src", "",
3404 [(set QPR:$dst, (v4i32 (vnotq QPR:$src)))]>;
3405 def : Pat<(v2i32 (vnotd DPR:$src)), (VMVNd DPR:$src)>;
3406 def : Pat<(v4i32 (vnotq QPR:$src)), (VMVNq QPR:$src)>;
3408 // VBSL : Vector Bitwise Select
3409 def VBSLd : N3VX<1, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
3410 (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
3411 N3RegFrm, IIC_VCNTiD,
3412 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3414 (v2i32 (or (and DPR:$Vn, DPR:$src1),
3415 (and DPR:$Vm, (vnotd DPR:$src1)))))]>;
3416 def VBSLq : N3VX<1, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
3417 (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
3418 N3RegFrm, IIC_VCNTiQ,
3419 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3421 (v4i32 (or (and QPR:$Vn, QPR:$src1),
3422 (and QPR:$Vm, (vnotq QPR:$src1)))))]>;
3424 // VBIF : Vector Bitwise Insert if False
3425 // like VBSL but with: "vbif $dst, $src3, $src1", "$src2 = $dst",
3426 // FIXME: This instruction's encoding MAY NOT BE correct.
3427 def VBIFd : N3VX<1, 0, 0b11, 0b0001, 0, 1,
3428 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
3429 N3RegFrm, IIC_VBINiD,
3430 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3431 [/* For disassembly only; pattern left blank */]>;
3432 def VBIFq : N3VX<1, 0, 0b11, 0b0001, 1, 1,
3433 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
3434 N3RegFrm, IIC_VBINiQ,
3435 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3436 [/* For disassembly only; pattern left blank */]>;
3438 // VBIT : Vector Bitwise Insert if True
3439 // like VBSL but with: "vbit $dst, $src2, $src1", "$src3 = $dst",
3440 // FIXME: This instruction's encoding MAY NOT BE correct.
3441 def VBITd : N3VX<1, 0, 0b10, 0b0001, 0, 1,
3442 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
3443 N3RegFrm, IIC_VBINiD,
3444 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3445 [/* For disassembly only; pattern left blank */]>;
3446 def VBITq : N3VX<1, 0, 0b10, 0b0001, 1, 1,
3447 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
3448 N3RegFrm, IIC_VBINiQ,
3449 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3450 [/* For disassembly only; pattern left blank */]>;
3452 // VBIT/VBIF are not yet implemented. The TwoAddress pass will not go looking
3453 // for equivalent operations with different register constraints; it just
3456 // Vector Absolute Differences.
3458 // VABD : Vector Absolute Difference
3459 defm VABDs : N3VInt_QHS<0, 0, 0b0111, 0, N3RegFrm,
3460 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3461 "vabd", "s", int_arm_neon_vabds, 1>;
3462 defm VABDu : N3VInt_QHS<1, 0, 0b0111, 0, N3RegFrm,
3463 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3464 "vabd", "u", int_arm_neon_vabdu, 1>;
3465 def VABDfd : N3VDInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBIND,
3466 "vabd", "f32", v2f32, v2f32, int_arm_neon_vabds, 1>;
3467 def VABDfq : N3VQInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBINQ,
3468 "vabd", "f32", v4f32, v4f32, int_arm_neon_vabds, 1>;
3470 // VABDL : Vector Absolute Difference Long (Q = | D - D |)
3471 defm VABDLs : N3VLIntExt_QHS<0,1,0b0111,0, IIC_VSUBi4Q,
3472 "vabdl", "s", int_arm_neon_vabds, zext, 1>;
3473 defm VABDLu : N3VLIntExt_QHS<1,1,0b0111,0, IIC_VSUBi4Q,
3474 "vabdl", "u", int_arm_neon_vabdu, zext, 1>;
3476 // VABA : Vector Absolute Difference and Accumulate
3477 defm VABAs : N3VIntOp_QHS<0,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
3478 "vaba", "s", int_arm_neon_vabds, add>;
3479 defm VABAu : N3VIntOp_QHS<1,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
3480 "vaba", "u", int_arm_neon_vabdu, add>;
3482 // VABAL : Vector Absolute Difference and Accumulate Long (Q += | D - D |)
3483 defm VABALs : N3VLIntExtOp_QHS<0,1,0b0101,0, IIC_VABAD,
3484 "vabal", "s", int_arm_neon_vabds, zext, add>;
3485 defm VABALu : N3VLIntExtOp_QHS<1,1,0b0101,0, IIC_VABAD,
3486 "vabal", "u", int_arm_neon_vabdu, zext, add>;
3488 // Vector Maximum and Minimum.
3490 // VMAX : Vector Maximum
3491 defm VMAXs : N3VInt_QHS<0, 0, 0b0110, 0, N3RegFrm,
3492 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3493 "vmax", "s", int_arm_neon_vmaxs, 1>;
3494 defm VMAXu : N3VInt_QHS<1, 0, 0b0110, 0, N3RegFrm,
3495 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3496 "vmax", "u", int_arm_neon_vmaxu, 1>;
3497 def VMAXfd : N3VDInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBIND,
3499 v2f32, v2f32, int_arm_neon_vmaxs, 1>;
3500 def VMAXfq : N3VQInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBINQ,
3502 v4f32, v4f32, int_arm_neon_vmaxs, 1>;
3504 // VMIN : Vector Minimum
3505 defm VMINs : N3VInt_QHS<0, 0, 0b0110, 1, N3RegFrm,
3506 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3507 "vmin", "s", int_arm_neon_vmins, 1>;
3508 defm VMINu : N3VInt_QHS<1, 0, 0b0110, 1, N3RegFrm,
3509 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3510 "vmin", "u", int_arm_neon_vminu, 1>;
3511 def VMINfd : N3VDInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBIND,
3513 v2f32, v2f32, int_arm_neon_vmins, 1>;
3514 def VMINfq : N3VQInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBINQ,
3516 v4f32, v4f32, int_arm_neon_vmins, 1>;
3518 // Vector Pairwise Operations.
3520 // VPADD : Vector Pairwise Add
3521 def VPADDi8 : N3VDInt<0, 0, 0b00, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
3523 v8i8, v8i8, int_arm_neon_vpadd, 0>;
3524 def VPADDi16 : N3VDInt<0, 0, 0b01, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
3526 v4i16, v4i16, int_arm_neon_vpadd, 0>;
3527 def VPADDi32 : N3VDInt<0, 0, 0b10, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
3529 v2i32, v2i32, int_arm_neon_vpadd, 0>;
3530 def VPADDf : N3VDInt<1, 0, 0b00, 0b1101, 0, N3RegFrm,
3531 IIC_VPBIND, "vpadd", "f32",
3532 v2f32, v2f32, int_arm_neon_vpadd, 0>;
3534 // VPADDL : Vector Pairwise Add Long
3535 defm VPADDLs : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00100, 0, "vpaddl", "s",
3536 int_arm_neon_vpaddls>;
3537 defm VPADDLu : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00101, 0, "vpaddl", "u",
3538 int_arm_neon_vpaddlu>;
3540 // VPADAL : Vector Pairwise Add and Accumulate Long
3541 defm VPADALs : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01100, 0, "vpadal", "s",
3542 int_arm_neon_vpadals>;
3543 defm VPADALu : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01101, 0, "vpadal", "u",
3544 int_arm_neon_vpadalu>;
3546 // VPMAX : Vector Pairwise Maximum
3547 def VPMAXs8 : N3VDInt<0, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3548 "s8", v8i8, v8i8, int_arm_neon_vpmaxs, 0>;
3549 def VPMAXs16 : N3VDInt<0, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3550 "s16", v4i16, v4i16, int_arm_neon_vpmaxs, 0>;
3551 def VPMAXs32 : N3VDInt<0, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3552 "s32", v2i32, v2i32, int_arm_neon_vpmaxs, 0>;
3553 def VPMAXu8 : N3VDInt<1, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3554 "u8", v8i8, v8i8, int_arm_neon_vpmaxu, 0>;
3555 def VPMAXu16 : N3VDInt<1, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3556 "u16", v4i16, v4i16, int_arm_neon_vpmaxu, 0>;
3557 def VPMAXu32 : N3VDInt<1, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3558 "u32", v2i32, v2i32, int_arm_neon_vpmaxu, 0>;
3559 def VPMAXf : N3VDInt<1, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmax",
3560 "f32", v2f32, v2f32, int_arm_neon_vpmaxs, 0>;
3562 // VPMIN : Vector Pairwise Minimum
3563 def VPMINs8 : N3VDInt<0, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3564 "s8", v8i8, v8i8, int_arm_neon_vpmins, 0>;
3565 def VPMINs16 : N3VDInt<0, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3566 "s16", v4i16, v4i16, int_arm_neon_vpmins, 0>;
3567 def VPMINs32 : N3VDInt<0, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3568 "s32", v2i32, v2i32, int_arm_neon_vpmins, 0>;
3569 def VPMINu8 : N3VDInt<1, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3570 "u8", v8i8, v8i8, int_arm_neon_vpminu, 0>;
3571 def VPMINu16 : N3VDInt<1, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3572 "u16", v4i16, v4i16, int_arm_neon_vpminu, 0>;
3573 def VPMINu32 : N3VDInt<1, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3574 "u32", v2i32, v2i32, int_arm_neon_vpminu, 0>;
3575 def VPMINf : N3VDInt<1, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmin",
3576 "f32", v2f32, v2f32, int_arm_neon_vpmins, 0>;
3578 // Vector Reciprocal and Reciprocal Square Root Estimate and Step.
3580 // VRECPE : Vector Reciprocal Estimate
3581 def VRECPEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
3582 IIC_VUNAD, "vrecpe", "u32",
3583 v2i32, v2i32, int_arm_neon_vrecpe>;
3584 def VRECPEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
3585 IIC_VUNAQ, "vrecpe", "u32",
3586 v4i32, v4i32, int_arm_neon_vrecpe>;
3587 def VRECPEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
3588 IIC_VUNAD, "vrecpe", "f32",
3589 v2f32, v2f32, int_arm_neon_vrecpe>;
3590 def VRECPEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
3591 IIC_VUNAQ, "vrecpe", "f32",
3592 v4f32, v4f32, int_arm_neon_vrecpe>;
3594 // VRECPS : Vector Reciprocal Step
3595 def VRECPSfd : N3VDInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
3596 IIC_VRECSD, "vrecps", "f32",
3597 v2f32, v2f32, int_arm_neon_vrecps, 1>;
3598 def VRECPSfq : N3VQInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
3599 IIC_VRECSQ, "vrecps", "f32",
3600 v4f32, v4f32, int_arm_neon_vrecps, 1>;
3602 // VRSQRTE : Vector Reciprocal Square Root Estimate
3603 def VRSQRTEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
3604 IIC_VUNAD, "vrsqrte", "u32",
3605 v2i32, v2i32, int_arm_neon_vrsqrte>;
3606 def VRSQRTEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
3607 IIC_VUNAQ, "vrsqrte", "u32",
3608 v4i32, v4i32, int_arm_neon_vrsqrte>;
3609 def VRSQRTEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
3610 IIC_VUNAD, "vrsqrte", "f32",
3611 v2f32, v2f32, int_arm_neon_vrsqrte>;
3612 def VRSQRTEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
3613 IIC_VUNAQ, "vrsqrte", "f32",
3614 v4f32, v4f32, int_arm_neon_vrsqrte>;
3616 // VRSQRTS : Vector Reciprocal Square Root Step
3617 def VRSQRTSfd : N3VDInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
3618 IIC_VRECSD, "vrsqrts", "f32",
3619 v2f32, v2f32, int_arm_neon_vrsqrts, 1>;
3620 def VRSQRTSfq : N3VQInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
3621 IIC_VRECSQ, "vrsqrts", "f32",
3622 v4f32, v4f32, int_arm_neon_vrsqrts, 1>;
3626 // VSHL : Vector Shift
3627 defm VSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 0, N3RegVShFrm,
3628 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
3629 "vshl", "s", int_arm_neon_vshifts>;
3630 defm VSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 0, N3RegVShFrm,
3631 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
3632 "vshl", "u", int_arm_neon_vshiftu>;
3633 // VSHL : Vector Shift Left (Immediate)
3634 defm VSHLi : N2VSh_QHSD<0, 1, 0b0101, 1, IIC_VSHLiD, "vshl", "i", NEONvshl,
3636 // VSHR : Vector Shift Right (Immediate)
3637 defm VSHRs : N2VSh_QHSD<0, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "s", NEONvshrs,
3639 defm VSHRu : N2VSh_QHSD<1, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "u", NEONvshru,
3642 // VSHLL : Vector Shift Left Long
3643 defm VSHLLs : N2VLSh_QHS<0, 1, 0b1010, 0, 0, 1, "vshll", "s", NEONvshlls>;
3644 defm VSHLLu : N2VLSh_QHS<1, 1, 0b1010, 0, 0, 1, "vshll", "u", NEONvshllu>;
3646 // VSHLL : Vector Shift Left Long (with maximum shift count)
3647 class N2VLShMax<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
3648 bit op6, bit op4, string OpcodeStr, string Dt, ValueType ResTy,
3649 ValueType OpTy, SDNode OpNode>
3650 : N2VLSh<op24, op23, op11_8, op7, op6, op4, OpcodeStr, Dt,
3651 ResTy, OpTy, OpNode> {
3652 let Inst{21-16} = op21_16;
3654 def VSHLLi8 : N2VLShMax<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll", "i8",
3655 v8i16, v8i8, NEONvshlli>;
3656 def VSHLLi16 : N2VLShMax<1, 1, 0b110110, 0b0011, 0, 0, 0, "vshll", "i16",
3657 v4i32, v4i16, NEONvshlli>;
3658 def VSHLLi32 : N2VLShMax<1, 1, 0b111010, 0b0011, 0, 0, 0, "vshll", "i32",
3659 v2i64, v2i32, NEONvshlli>;
3661 // VSHRN : Vector Shift Right and Narrow
3662 defm VSHRN : N2VNSh_HSD<0,1,0b1000,0,0,1, IIC_VSHLiD, "vshrn", "i",
3665 // VRSHL : Vector Rounding Shift
3666 defm VRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 0, N3RegVShFrm,
3667 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3668 "vrshl", "s", int_arm_neon_vrshifts>;
3669 defm VRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 0, N3RegVShFrm,
3670 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3671 "vrshl", "u", int_arm_neon_vrshiftu>;
3672 // VRSHR : Vector Rounding Shift Right
3673 defm VRSHRs : N2VSh_QHSD<0,1,0b0010,1, IIC_VSHLi4D, "vrshr", "s", NEONvrshrs,
3675 defm VRSHRu : N2VSh_QHSD<1,1,0b0010,1, IIC_VSHLi4D, "vrshr", "u", NEONvrshru,
3678 // VRSHRN : Vector Rounding Shift Right and Narrow
3679 defm VRSHRN : N2VNSh_HSD<0, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vrshrn", "i",
3682 // VQSHL : Vector Saturating Shift
3683 defm VQSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 1, N3RegVShFrm,
3684 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3685 "vqshl", "s", int_arm_neon_vqshifts>;
3686 defm VQSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 1, N3RegVShFrm,
3687 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3688 "vqshl", "u", int_arm_neon_vqshiftu>;
3689 // VQSHL : Vector Saturating Shift Left (Immediate)
3690 defm VQSHLsi : N2VSh_QHSD<0,1,0b0111,1, IIC_VSHLi4D, "vqshl", "s",NEONvqshls,
3692 defm VQSHLui : N2VSh_QHSD<1,1,0b0111,1, IIC_VSHLi4D, "vqshl", "u",NEONvqshlu,
3694 // VQSHLU : Vector Saturating Shift Left (Immediate, Unsigned)
3695 defm VQSHLsu : N2VSh_QHSD<1,1,0b0110,1, IIC_VSHLi4D,"vqshlu","s",NEONvqshlsu,
3698 // VQSHRN : Vector Saturating Shift Right and Narrow
3699 defm VQSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "s",
3701 defm VQSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "u",
3704 // VQSHRUN : Vector Saturating Shift Right and Narrow (Unsigned)
3705 defm VQSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 0, 1, IIC_VSHLi4D, "vqshrun", "s",
3708 // VQRSHL : Vector Saturating Rounding Shift
3709 defm VQRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 1, N3RegVShFrm,
3710 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3711 "vqrshl", "s", int_arm_neon_vqrshifts>;
3712 defm VQRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 1, N3RegVShFrm,
3713 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3714 "vqrshl", "u", int_arm_neon_vqrshiftu>;
3716 // VQRSHRN : Vector Saturating Rounding Shift Right and Narrow
3717 defm VQRSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "s",
3719 defm VQRSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "u",
3722 // VQRSHRUN : Vector Saturating Rounding Shift Right and Narrow (Unsigned)
3723 defm VQRSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vqrshrun", "s",
3726 // VSRA : Vector Shift Right and Accumulate
3727 defm VSRAs : N2VShAdd_QHSD<0, 1, 0b0001, 1, "vsra", "s", NEONvshrs>;
3728 defm VSRAu : N2VShAdd_QHSD<1, 1, 0b0001, 1, "vsra", "u", NEONvshru>;
3729 // VRSRA : Vector Rounding Shift Right and Accumulate
3730 defm VRSRAs : N2VShAdd_QHSD<0, 1, 0b0011, 1, "vrsra", "s", NEONvrshrs>;
3731 defm VRSRAu : N2VShAdd_QHSD<1, 1, 0b0011, 1, "vrsra", "u", NEONvrshru>;
3733 // VSLI : Vector Shift Left and Insert
3734 defm VSLI : N2VShIns_QHSD<1, 1, 0b0101, 1, "vsli", NEONvsli, N2RegVShLFrm>;
3735 // VSRI : Vector Shift Right and Insert
3736 defm VSRI : N2VShIns_QHSD<1, 1, 0b0100, 1, "vsri", NEONvsri, N2RegVShRFrm>;
3738 // Vector Absolute and Saturating Absolute.
3740 // VABS : Vector Absolute Value
3741 defm VABS : N2VInt_QHS<0b11, 0b11, 0b01, 0b00110, 0,
3742 IIC_VUNAiD, IIC_VUNAiQ, "vabs", "s",
3744 def VABSfd : N2VDInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
3745 IIC_VUNAD, "vabs", "f32",
3746 v2f32, v2f32, int_arm_neon_vabs>;
3747 def VABSfq : N2VQInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
3748 IIC_VUNAQ, "vabs", "f32",
3749 v4f32, v4f32, int_arm_neon_vabs>;
3751 // VQABS : Vector Saturating Absolute Value
3752 defm VQABS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01110, 0,
3753 IIC_VQUNAiD, IIC_VQUNAiQ, "vqabs", "s",
3754 int_arm_neon_vqabs>;
3758 def vnegd : PatFrag<(ops node:$in),
3759 (sub (bitconvert (v2i32 NEONimmAllZerosV)), node:$in)>;
3760 def vnegq : PatFrag<(ops node:$in),
3761 (sub (bitconvert (v4i32 NEONimmAllZerosV)), node:$in)>;
3763 class VNEGD<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
3764 : N2V<0b11, 0b11, size, 0b01, 0b00111, 0, 0, (outs DPR:$dst), (ins DPR:$src),
3765 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
3766 [(set DPR:$dst, (Ty (vnegd DPR:$src)))]>;
3767 class VNEGQ<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
3768 : N2V<0b11, 0b11, size, 0b01, 0b00111, 1, 0, (outs QPR:$dst), (ins QPR:$src),
3769 IIC_VSHLiQ, OpcodeStr, Dt, "$dst, $src", "",
3770 [(set QPR:$dst, (Ty (vnegq QPR:$src)))]>;
3772 // VNEG : Vector Negate (integer)
3773 def VNEGs8d : VNEGD<0b00, "vneg", "s8", v8i8>;
3774 def VNEGs16d : VNEGD<0b01, "vneg", "s16", v4i16>;
3775 def VNEGs32d : VNEGD<0b10, "vneg", "s32", v2i32>;
3776 def VNEGs8q : VNEGQ<0b00, "vneg", "s8", v16i8>;
3777 def VNEGs16q : VNEGQ<0b01, "vneg", "s16", v8i16>;
3778 def VNEGs32q : VNEGQ<0b10, "vneg", "s32", v4i32>;
3780 // VNEG : Vector Negate (floating-point)
3781 def VNEGfd : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
3782 (outs DPR:$dst), (ins DPR:$src), IIC_VUNAD,
3783 "vneg", "f32", "$dst, $src", "",
3784 [(set DPR:$dst, (v2f32 (fneg DPR:$src)))]>;
3785 def VNEGf32q : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 1, 0,
3786 (outs QPR:$dst), (ins QPR:$src), IIC_VUNAQ,
3787 "vneg", "f32", "$dst, $src", "",
3788 [(set QPR:$dst, (v4f32 (fneg QPR:$src)))]>;
3790 def : Pat<(v8i8 (vnegd DPR:$src)), (VNEGs8d DPR:$src)>;
3791 def : Pat<(v4i16 (vnegd DPR:$src)), (VNEGs16d DPR:$src)>;
3792 def : Pat<(v2i32 (vnegd DPR:$src)), (VNEGs32d DPR:$src)>;
3793 def : Pat<(v16i8 (vnegq QPR:$src)), (VNEGs8q QPR:$src)>;
3794 def : Pat<(v8i16 (vnegq QPR:$src)), (VNEGs16q QPR:$src)>;
3795 def : Pat<(v4i32 (vnegq QPR:$src)), (VNEGs32q QPR:$src)>;
3797 // VQNEG : Vector Saturating Negate
3798 defm VQNEG : N2VInt_QHS<0b11, 0b11, 0b00, 0b01111, 0,
3799 IIC_VQUNAiD, IIC_VQUNAiQ, "vqneg", "s",
3800 int_arm_neon_vqneg>;
3802 // Vector Bit Counting Operations.
3804 // VCLS : Vector Count Leading Sign Bits
3805 defm VCLS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01000, 0,
3806 IIC_VCNTiD, IIC_VCNTiQ, "vcls", "s",
3808 // VCLZ : Vector Count Leading Zeros
3809 defm VCLZ : N2VInt_QHS<0b11, 0b11, 0b00, 0b01001, 0,
3810 IIC_VCNTiD, IIC_VCNTiQ, "vclz", "i",
3812 // VCNT : Vector Count One Bits
3813 def VCNTd : N2VDInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
3814 IIC_VCNTiD, "vcnt", "8",
3815 v8i8, v8i8, int_arm_neon_vcnt>;
3816 def VCNTq : N2VQInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
3817 IIC_VCNTiQ, "vcnt", "8",
3818 v16i8, v16i8, int_arm_neon_vcnt>;
3820 // Vector Swap -- for disassembly only.
3821 def VSWPd : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 0, 0,
3822 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
3823 "vswp", "$dst, $src", "", []>;
3824 def VSWPq : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 1, 0,
3825 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
3826 "vswp", "$dst, $src", "", []>;
3828 // Vector Move Operations.
3830 // VMOV : Vector Move (Register)
3832 let neverHasSideEffects = 1 in {
3833 def VMOVDneon: N3VX<0, 0, 0b10, 0b0001, 0, 1, (outs DPR:$dst), (ins DPR:$src),
3834 N3RegFrm, IIC_VMOV, "vmov", "$dst, $src", "", []>;
3835 def VMOVQ : N3VX<0, 0, 0b10, 0b0001, 1, 1, (outs QPR:$dst), (ins QPR:$src),
3836 N3RegFrm, IIC_VMOV, "vmov", "$dst, $src", "", []>;
3838 // Pseudo vector move instructions for QQ and QQQQ registers. This should
3839 // be expanded after register allocation is completed.
3840 def VMOVQQ : PseudoInst<(outs QQPR:$dst), (ins QQPR:$src),
3841 NoItinerary, "", []>;
3843 def VMOVQQQQ : PseudoInst<(outs QQQQPR:$dst), (ins QQQQPR:$src),
3844 NoItinerary, "", []>;
3845 } // neverHasSideEffects
3847 // VMOV : Vector Move (Immediate)
3849 let isReMaterializable = 1 in {
3850 def VMOVv8i8 : N1ModImm<1, 0b000, 0b1110, 0, 0, 0, 1, (outs DPR:$dst),
3851 (ins nModImm:$SIMM), IIC_VMOVImm,
3852 "vmov", "i8", "$dst, $SIMM", "",
3853 [(set DPR:$dst, (v8i8 (NEONvmovImm timm:$SIMM)))]>;
3854 def VMOVv16i8 : N1ModImm<1, 0b000, 0b1110, 0, 1, 0, 1, (outs QPR:$dst),
3855 (ins nModImm:$SIMM), IIC_VMOVImm,
3856 "vmov", "i8", "$dst, $SIMM", "",
3857 [(set QPR:$dst, (v16i8 (NEONvmovImm timm:$SIMM)))]>;
3859 def VMOVv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 0, 1, (outs DPR:$dst),
3860 (ins nModImm:$SIMM), IIC_VMOVImm,
3861 "vmov", "i16", "$dst, $SIMM", "",
3862 [(set DPR:$dst, (v4i16 (NEONvmovImm timm:$SIMM)))]> {
3863 let Inst{9} = SIMM{9};
3866 def VMOVv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 0, 1, (outs QPR:$dst),
3867 (ins nModImm:$SIMM), IIC_VMOVImm,
3868 "vmov", "i16", "$dst, $SIMM", "",
3869 [(set QPR:$dst, (v8i16 (NEONvmovImm timm:$SIMM)))]> {
3870 let Inst{9} = SIMM{9};
3873 def VMOVv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 0, 1, (outs DPR:$dst),
3874 (ins nModImm:$SIMM), IIC_VMOVImm,
3875 "vmov", "i32", "$dst, $SIMM", "",
3876 [(set DPR:$dst, (v2i32 (NEONvmovImm timm:$SIMM)))]> {
3877 let Inst{11-8} = SIMM{11-8};
3880 def VMOVv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 0, 1, (outs QPR:$dst),
3881 (ins nModImm:$SIMM), IIC_VMOVImm,
3882 "vmov", "i32", "$dst, $SIMM", "",
3883 [(set QPR:$dst, (v4i32 (NEONvmovImm timm:$SIMM)))]> {
3884 let Inst{11-8} = SIMM{11-8};
3887 def VMOVv1i64 : N1ModImm<1, 0b000, 0b1110, 0, 0, 1, 1, (outs DPR:$dst),
3888 (ins nModImm:$SIMM), IIC_VMOVImm,
3889 "vmov", "i64", "$dst, $SIMM", "",
3890 [(set DPR:$dst, (v1i64 (NEONvmovImm timm:$SIMM)))]>;
3891 def VMOVv2i64 : N1ModImm<1, 0b000, 0b1110, 0, 1, 1, 1, (outs QPR:$dst),
3892 (ins nModImm:$SIMM), IIC_VMOVImm,
3893 "vmov", "i64", "$dst, $SIMM", "",
3894 [(set QPR:$dst, (v2i64 (NEONvmovImm timm:$SIMM)))]>;
3895 } // isReMaterializable
3897 // VMOV : Vector Get Lane (move scalar to ARM core register)
3899 def VGETLNs8 : NVGetLane<{1,1,1,0,0,1,?,1}, 0b1011, {?,?},
3900 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
3901 IIC_VMOVSI, "vmov", "s8", "$R, $V[$lane]",
3902 [(set GPR:$R, (NEONvgetlanes (v8i8 DPR:$V),
3904 let Inst{21} = lane{2};
3905 let Inst{6-5} = lane{1-0};
3907 def VGETLNs16 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, {?,1},
3908 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
3909 IIC_VMOVSI, "vmov", "s16", "$R, $V[$lane]",
3910 [(set GPR:$R, (NEONvgetlanes (v4i16 DPR:$V),
3912 let Inst{21} = lane{1};
3913 let Inst{6} = lane{0};
3915 def VGETLNu8 : NVGetLane<{1,1,1,0,1,1,?,1}, 0b1011, {?,?},
3916 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
3917 IIC_VMOVSI, "vmov", "u8", "$R, $V[$lane]",
3918 [(set GPR:$R, (NEONvgetlaneu (v8i8 DPR:$V),
3920 let Inst{21} = lane{2};
3921 let Inst{6-5} = lane{1-0};
3923 def VGETLNu16 : NVGetLane<{1,1,1,0,1,0,?,1}, 0b1011, {?,1},
3924 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
3925 IIC_VMOVSI, "vmov", "u16", "$R, $V[$lane]",
3926 [(set GPR:$R, (NEONvgetlaneu (v4i16 DPR:$V),
3928 let Inst{21} = lane{1};
3929 let Inst{6} = lane{0};
3931 def VGETLNi32 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, 0b00,
3932 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
3933 IIC_VMOVSI, "vmov", "32", "$R, $V[$lane]",
3934 [(set GPR:$R, (extractelt (v2i32 DPR:$V),
3936 let Inst{21} = lane{0};
3938 // def VGETLNf32: see FMRDH and FMRDL in ARMInstrVFP.td
3939 def : Pat<(NEONvgetlanes (v16i8 QPR:$src), imm:$lane),
3940 (VGETLNs8 (v8i8 (EXTRACT_SUBREG QPR:$src,
3941 (DSubReg_i8_reg imm:$lane))),
3942 (SubReg_i8_lane imm:$lane))>;
3943 def : Pat<(NEONvgetlanes (v8i16 QPR:$src), imm:$lane),
3944 (VGETLNs16 (v4i16 (EXTRACT_SUBREG QPR:$src,
3945 (DSubReg_i16_reg imm:$lane))),
3946 (SubReg_i16_lane imm:$lane))>;
3947 def : Pat<(NEONvgetlaneu (v16i8 QPR:$src), imm:$lane),
3948 (VGETLNu8 (v8i8 (EXTRACT_SUBREG QPR:$src,
3949 (DSubReg_i8_reg imm:$lane))),
3950 (SubReg_i8_lane imm:$lane))>;
3951 def : Pat<(NEONvgetlaneu (v8i16 QPR:$src), imm:$lane),
3952 (VGETLNu16 (v4i16 (EXTRACT_SUBREG QPR:$src,
3953 (DSubReg_i16_reg imm:$lane))),
3954 (SubReg_i16_lane imm:$lane))>;
3955 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
3956 (VGETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src,
3957 (DSubReg_i32_reg imm:$lane))),
3958 (SubReg_i32_lane imm:$lane))>;
3959 def : Pat<(extractelt (v2f32 DPR:$src1), imm:$src2),
3960 (EXTRACT_SUBREG (v2f32 (COPY_TO_REGCLASS (v2f32 DPR:$src1),DPR_VFP2)),
3961 (SSubReg_f32_reg imm:$src2))>;
3962 def : Pat<(extractelt (v4f32 QPR:$src1), imm:$src2),
3963 (EXTRACT_SUBREG (v4f32 (COPY_TO_REGCLASS (v4f32 QPR:$src1),QPR_VFP2)),
3964 (SSubReg_f32_reg imm:$src2))>;
3965 //def : Pat<(extractelt (v2i64 QPR:$src1), imm:$src2),
3966 // (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
3967 def : Pat<(extractelt (v2f64 QPR:$src1), imm:$src2),
3968 (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
3971 // VMOV : Vector Set Lane (move ARM core register to scalar)
3973 let Constraints = "$src1 = $V" in {
3974 def VSETLNi8 : NVSetLane<{1,1,1,0,0,1,?,0}, 0b1011, {?,?}, (outs DPR:$V),
3975 (ins DPR:$src1, GPR:$R, nohash_imm:$lane),
3976 IIC_VMOVISL, "vmov", "8", "$V[$lane], $R",
3977 [(set DPR:$V, (vector_insert (v8i8 DPR:$src1),
3978 GPR:$R, imm:$lane))]> {
3979 let Inst{21} = lane{2};
3980 let Inst{6-5} = lane{1-0};
3982 def VSETLNi16 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, {?,1}, (outs DPR:$V),
3983 (ins DPR:$src1, GPR:$R, nohash_imm:$lane),
3984 IIC_VMOVISL, "vmov", "16", "$V[$lane], $R",
3985 [(set DPR:$V, (vector_insert (v4i16 DPR:$src1),
3986 GPR:$R, imm:$lane))]> {
3987 let Inst{21} = lane{1};
3988 let Inst{6} = lane{0};
3990 def VSETLNi32 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, 0b00, (outs DPR:$V),
3991 (ins DPR:$src1, GPR:$R, nohash_imm:$lane),
3992 IIC_VMOVISL, "vmov", "32", "$V[$lane], $R",
3993 [(set DPR:$V, (insertelt (v2i32 DPR:$src1),
3994 GPR:$R, imm:$lane))]> {
3995 let Inst{21} = lane{0};
3998 def : Pat<(vector_insert (v16i8 QPR:$src1), GPR:$src2, imm:$lane),
3999 (v16i8 (INSERT_SUBREG QPR:$src1,
4000 (v8i8 (VSETLNi8 (v8i8 (EXTRACT_SUBREG QPR:$src1,
4001 (DSubReg_i8_reg imm:$lane))),
4002 GPR:$src2, (SubReg_i8_lane imm:$lane))),
4003 (DSubReg_i8_reg imm:$lane)))>;
4004 def : Pat<(vector_insert (v8i16 QPR:$src1), GPR:$src2, imm:$lane),
4005 (v8i16 (INSERT_SUBREG QPR:$src1,
4006 (v4i16 (VSETLNi16 (v4i16 (EXTRACT_SUBREG QPR:$src1,
4007 (DSubReg_i16_reg imm:$lane))),
4008 GPR:$src2, (SubReg_i16_lane imm:$lane))),
4009 (DSubReg_i16_reg imm:$lane)))>;
4010 def : Pat<(insertelt (v4i32 QPR:$src1), GPR:$src2, imm:$lane),
4011 (v4i32 (INSERT_SUBREG QPR:$src1,
4012 (v2i32 (VSETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src1,
4013 (DSubReg_i32_reg imm:$lane))),
4014 GPR:$src2, (SubReg_i32_lane imm:$lane))),
4015 (DSubReg_i32_reg imm:$lane)))>;
4017 def : Pat<(v2f32 (insertelt DPR:$src1, SPR:$src2, imm:$src3)),
4018 (INSERT_SUBREG (v2f32 (COPY_TO_REGCLASS DPR:$src1, DPR_VFP2)),
4019 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
4020 def : Pat<(v4f32 (insertelt QPR:$src1, SPR:$src2, imm:$src3)),
4021 (INSERT_SUBREG (v4f32 (COPY_TO_REGCLASS QPR:$src1, QPR_VFP2)),
4022 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
4024 //def : Pat<(v2i64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
4025 // (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
4026 def : Pat<(v2f64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
4027 (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
4029 def : Pat<(v2f32 (scalar_to_vector SPR:$src)),
4030 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
4031 def : Pat<(v2f64 (scalar_to_vector (f64 DPR:$src))),
4032 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
4033 def : Pat<(v4f32 (scalar_to_vector SPR:$src)),
4034 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
4036 def : Pat<(v8i8 (scalar_to_vector GPR:$src)),
4037 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4038 def : Pat<(v4i16 (scalar_to_vector GPR:$src)),
4039 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4040 def : Pat<(v2i32 (scalar_to_vector GPR:$src)),
4041 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4043 def : Pat<(v16i8 (scalar_to_vector GPR:$src)),
4044 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4045 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4047 def : Pat<(v8i16 (scalar_to_vector GPR:$src)),
4048 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
4049 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4051 def : Pat<(v4i32 (scalar_to_vector GPR:$src)),
4052 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
4053 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4056 // VDUP : Vector Duplicate (from ARM core register to all elements)
4058 class VDUPD<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
4059 : NVDup<opcod1, 0b1011, opcod3, (outs DPR:$dst), (ins GPR:$src),
4060 IIC_VMOVIS, "vdup", Dt, "$dst, $src",
4061 [(set DPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
4062 class VDUPQ<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
4063 : NVDup<opcod1, 0b1011, opcod3, (outs QPR:$dst), (ins GPR:$src),
4064 IIC_VMOVIS, "vdup", Dt, "$dst, $src",
4065 [(set QPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
4067 def VDUP8d : VDUPD<0b11101100, 0b00, "8", v8i8>;
4068 def VDUP16d : VDUPD<0b11101000, 0b01, "16", v4i16>;
4069 def VDUP32d : VDUPD<0b11101000, 0b00, "32", v2i32>;
4070 def VDUP8q : VDUPQ<0b11101110, 0b00, "8", v16i8>;
4071 def VDUP16q : VDUPQ<0b11101010, 0b01, "16", v8i16>;
4072 def VDUP32q : VDUPQ<0b11101010, 0b00, "32", v4i32>;
4074 def VDUPfd : NVDup<0b11101000, 0b1011, 0b00, (outs DPR:$dst), (ins GPR:$src),
4075 IIC_VMOVIS, "vdup", "32", "$dst, $src",
4076 [(set DPR:$dst, (v2f32 (NEONvdup
4077 (f32 (bitconvert GPR:$src)))))]>;
4078 def VDUPfq : NVDup<0b11101010, 0b1011, 0b00, (outs QPR:$dst), (ins GPR:$src),
4079 IIC_VMOVIS, "vdup", "32", "$dst, $src",
4080 [(set QPR:$dst, (v4f32 (NEONvdup
4081 (f32 (bitconvert GPR:$src)))))]>;
4083 // VDUP : Vector Duplicate Lane (from scalar to all elements)
4085 class VDUPLND<bits<4> op19_16, string OpcodeStr, string Dt,
4087 : NVDupLane<op19_16, 0, (outs DPR:$dst), (ins DPR:$src, nohash_imm:$lane),
4088 IIC_VMOVD, OpcodeStr, Dt, "$dst, $src[$lane]",
4089 [(set DPR:$dst, (Ty (NEONvduplane (Ty DPR:$src), imm:$lane)))]>;
4091 class VDUPLNQ<bits<4> op19_16, string OpcodeStr, string Dt,
4092 ValueType ResTy, ValueType OpTy>
4093 : NVDupLane<op19_16, 1, (outs QPR:$dst), (ins DPR:$src, nohash_imm:$lane),
4094 IIC_VMOVQ, OpcodeStr, Dt, "$dst, $src[$lane]",
4095 [(set QPR:$dst, (ResTy (NEONvduplane (OpTy DPR:$src),
4098 // Inst{19-16} is partially specified depending on the element size.
4100 def VDUPLN8d : VDUPLND<{?,?,?,1}, "vdup", "8", v8i8> {
4101 let Inst{19-17} = lane{2-0};
4103 def VDUPLN16d : VDUPLND<{?,?,1,0}, "vdup", "16", v4i16> {
4104 let Inst{19-18} = lane{1-0};
4106 def VDUPLN32d : VDUPLND<{?,1,0,0}, "vdup", "32", v2i32> {
4107 let Inst{19} = lane{0};
4109 def VDUPLNfd : VDUPLND<{?,1,0,0}, "vdup", "32", v2f32> {
4110 let Inst{19} = lane{0};
4112 def VDUPLN8q : VDUPLNQ<{?,?,?,1}, "vdup", "8", v16i8, v8i8> {
4113 let Inst{19-17} = lane{2-0};
4115 def VDUPLN16q : VDUPLNQ<{?,?,1,0}, "vdup", "16", v8i16, v4i16> {
4116 let Inst{19-18} = lane{1-0};
4118 def VDUPLN32q : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4i32, v2i32> {
4119 let Inst{19} = lane{0};
4121 def VDUPLNfq : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4f32, v2f32> {
4122 let Inst{19} = lane{0};
4125 def : Pat<(v16i8 (NEONvduplane (v16i8 QPR:$src), imm:$lane)),
4126 (v16i8 (VDUPLN8q (v8i8 (EXTRACT_SUBREG QPR:$src,
4127 (DSubReg_i8_reg imm:$lane))),
4128 (SubReg_i8_lane imm:$lane)))>;
4129 def : Pat<(v8i16 (NEONvduplane (v8i16 QPR:$src), imm:$lane)),
4130 (v8i16 (VDUPLN16q (v4i16 (EXTRACT_SUBREG QPR:$src,
4131 (DSubReg_i16_reg imm:$lane))),
4132 (SubReg_i16_lane imm:$lane)))>;
4133 def : Pat<(v4i32 (NEONvduplane (v4i32 QPR:$src), imm:$lane)),
4134 (v4i32 (VDUPLN32q (v2i32 (EXTRACT_SUBREG QPR:$src,
4135 (DSubReg_i32_reg imm:$lane))),
4136 (SubReg_i32_lane imm:$lane)))>;
4137 def : Pat<(v4f32 (NEONvduplane (v4f32 QPR:$src), imm:$lane)),
4138 (v4f32 (VDUPLNfq (v2f32 (EXTRACT_SUBREG QPR:$src,
4139 (DSubReg_i32_reg imm:$lane))),
4140 (SubReg_i32_lane imm:$lane)))>;
4142 def VDUPfdf : PseudoNeonI<(outs DPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
4143 [(set DPR:$dst, (v2f32 (NEONvdup (f32 SPR:$src))))]>;
4144 def VDUPfqf : PseudoNeonI<(outs QPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
4145 [(set QPR:$dst, (v4f32 (NEONvdup (f32 SPR:$src))))]>;
4147 // VMOVN : Vector Narrowing Move
4148 defm VMOVN : N2VN_HSD<0b11,0b11,0b10,0b00100,0,0, IIC_VMOVN,
4149 "vmovn", "i", trunc>;
4150 // VQMOVN : Vector Saturating Narrowing Move
4151 defm VQMOVNs : N2VNInt_HSD<0b11,0b11,0b10,0b00101,0,0, IIC_VQUNAiD,
4152 "vqmovn", "s", int_arm_neon_vqmovns>;
4153 defm VQMOVNu : N2VNInt_HSD<0b11,0b11,0b10,0b00101,1,0, IIC_VQUNAiD,
4154 "vqmovn", "u", int_arm_neon_vqmovnu>;
4155 defm VQMOVNsu : N2VNInt_HSD<0b11,0b11,0b10,0b00100,1,0, IIC_VQUNAiD,
4156 "vqmovun", "s", int_arm_neon_vqmovnsu>;
4157 // VMOVL : Vector Lengthening Move
4158 defm VMOVLs : N2VL_QHS<0b01,0b10100,0,1, "vmovl", "s", sext>;
4159 defm VMOVLu : N2VL_QHS<0b11,0b10100,0,1, "vmovl", "u", zext>;
4161 // Vector Conversions.
4163 // VCVT : Vector Convert Between Floating-Point and Integers
4164 def VCVTf2sd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
4165 v2i32, v2f32, fp_to_sint>;
4166 def VCVTf2ud : N2VD<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
4167 v2i32, v2f32, fp_to_uint>;
4168 def VCVTs2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
4169 v2f32, v2i32, sint_to_fp>;
4170 def VCVTu2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
4171 v2f32, v2i32, uint_to_fp>;
4173 def VCVTf2sq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
4174 v4i32, v4f32, fp_to_sint>;
4175 def VCVTf2uq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
4176 v4i32, v4f32, fp_to_uint>;
4177 def VCVTs2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
4178 v4f32, v4i32, sint_to_fp>;
4179 def VCVTu2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
4180 v4f32, v4i32, uint_to_fp>;
4182 // VCVT : Vector Convert Between Floating-Point and Fixed-Point.
4183 def VCVTf2xsd : N2VCvtD<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
4184 v2i32, v2f32, int_arm_neon_vcvtfp2fxs>;
4185 def VCVTf2xud : N2VCvtD<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
4186 v2i32, v2f32, int_arm_neon_vcvtfp2fxu>;
4187 def VCVTxs2fd : N2VCvtD<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
4188 v2f32, v2i32, int_arm_neon_vcvtfxs2fp>;
4189 def VCVTxu2fd : N2VCvtD<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
4190 v2f32, v2i32, int_arm_neon_vcvtfxu2fp>;
4192 def VCVTf2xsq : N2VCvtQ<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
4193 v4i32, v4f32, int_arm_neon_vcvtfp2fxs>;
4194 def VCVTf2xuq : N2VCvtQ<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
4195 v4i32, v4f32, int_arm_neon_vcvtfp2fxu>;
4196 def VCVTxs2fq : N2VCvtQ<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
4197 v4f32, v4i32, int_arm_neon_vcvtfxs2fp>;
4198 def VCVTxu2fq : N2VCvtQ<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
4199 v4f32, v4i32, int_arm_neon_vcvtfxu2fp>;
4203 // VREV64 : Vector Reverse elements within 64-bit doublewords
4205 class VREV64D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4206 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 0, 0, (outs DPR:$dst),
4207 (ins DPR:$src), IIC_VMOVD,
4208 OpcodeStr, Dt, "$dst, $src", "",
4209 [(set DPR:$dst, (Ty (NEONvrev64 (Ty DPR:$src))))]>;
4210 class VREV64Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4211 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 1, 0, (outs QPR:$dst),
4212 (ins QPR:$src), IIC_VMOVQ,
4213 OpcodeStr, Dt, "$dst, $src", "",
4214 [(set QPR:$dst, (Ty (NEONvrev64 (Ty QPR:$src))))]>;
4216 def VREV64d8 : VREV64D<0b00, "vrev64", "8", v8i8>;
4217 def VREV64d16 : VREV64D<0b01, "vrev64", "16", v4i16>;
4218 def VREV64d32 : VREV64D<0b10, "vrev64", "32", v2i32>;
4219 def VREV64df : VREV64D<0b10, "vrev64", "32", v2f32>;
4221 def VREV64q8 : VREV64Q<0b00, "vrev64", "8", v16i8>;
4222 def VREV64q16 : VREV64Q<0b01, "vrev64", "16", v8i16>;
4223 def VREV64q32 : VREV64Q<0b10, "vrev64", "32", v4i32>;
4224 def VREV64qf : VREV64Q<0b10, "vrev64", "32", v4f32>;
4226 // VREV32 : Vector Reverse elements within 32-bit words
4228 class VREV32D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4229 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 0, 0, (outs DPR:$dst),
4230 (ins DPR:$src), IIC_VMOVD,
4231 OpcodeStr, Dt, "$dst, $src", "",
4232 [(set DPR:$dst, (Ty (NEONvrev32 (Ty DPR:$src))))]>;
4233 class VREV32Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4234 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 1, 0, (outs QPR:$dst),
4235 (ins QPR:$src), IIC_VMOVQ,
4236 OpcodeStr, Dt, "$dst, $src", "",
4237 [(set QPR:$dst, (Ty (NEONvrev32 (Ty QPR:$src))))]>;
4239 def VREV32d8 : VREV32D<0b00, "vrev32", "8", v8i8>;
4240 def VREV32d16 : VREV32D<0b01, "vrev32", "16", v4i16>;
4242 def VREV32q8 : VREV32Q<0b00, "vrev32", "8", v16i8>;
4243 def VREV32q16 : VREV32Q<0b01, "vrev32", "16", v8i16>;
4245 // VREV16 : Vector Reverse elements within 16-bit halfwords
4247 class VREV16D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4248 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 0, 0, (outs DPR:$dst),
4249 (ins DPR:$src), IIC_VMOVD,
4250 OpcodeStr, Dt, "$dst, $src", "",
4251 [(set DPR:$dst, (Ty (NEONvrev16 (Ty DPR:$src))))]>;
4252 class VREV16Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4253 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 1, 0, (outs QPR:$dst),
4254 (ins QPR:$src), IIC_VMOVQ,
4255 OpcodeStr, Dt, "$dst, $src", "",
4256 [(set QPR:$dst, (Ty (NEONvrev16 (Ty QPR:$src))))]>;
4258 def VREV16d8 : VREV16D<0b00, "vrev16", "8", v8i8>;
4259 def VREV16q8 : VREV16Q<0b00, "vrev16", "8", v16i8>;
4261 // Other Vector Shuffles.
4263 // VEXT : Vector Extract
4265 class VEXTd<string OpcodeStr, string Dt, ValueType Ty>
4266 : N3V<0,1,0b11,{?,?,?,?},0,0, (outs DPR:$dst),
4267 (ins DPR:$lhs, DPR:$rhs, i32imm:$index), NVExtFrm,
4268 IIC_VEXTD, OpcodeStr, Dt, "$dst, $lhs, $rhs, $index", "",
4269 [(set DPR:$dst, (Ty (NEONvext (Ty DPR:$lhs),
4270 (Ty DPR:$rhs), imm:$index)))]> {
4272 let Inst{11-8} = index{3-0};
4275 class VEXTq<string OpcodeStr, string Dt, ValueType Ty>
4276 : N3V<0,1,0b11,{?,?,?,?},1,0, (outs QPR:$dst),
4277 (ins QPR:$lhs, QPR:$rhs, i32imm:$index), NVExtFrm,
4278 IIC_VEXTQ, OpcodeStr, Dt, "$dst, $lhs, $rhs, $index", "",
4279 [(set QPR:$dst, (Ty (NEONvext (Ty QPR:$lhs),
4280 (Ty QPR:$rhs), imm:$index)))]> {
4282 let Inst{11-8} = index{3-0};
4285 def VEXTd8 : VEXTd<"vext", "8", v8i8> {
4286 let Inst{11-8} = index{3-0};
4288 def VEXTd16 : VEXTd<"vext", "16", v4i16> {
4289 let Inst{11-9} = index{2-0};
4292 def VEXTd32 : VEXTd<"vext", "32", v2i32> {
4293 let Inst{11-10} = index{1-0};
4294 let Inst{9-8} = 0b00;
4296 def VEXTdf : VEXTd<"vext", "32", v2f32> {
4297 let Inst{11} = index{0};
4298 let Inst{10-8} = 0b000;
4301 def VEXTq8 : VEXTq<"vext", "8", v16i8> {
4302 let Inst{11-8} = index{3-0};
4304 def VEXTq16 : VEXTq<"vext", "16", v8i16> {
4305 let Inst{11-9} = index{2-0};
4308 def VEXTq32 : VEXTq<"vext", "32", v4i32> {
4309 let Inst{11-10} = index{1-0};
4310 let Inst{9-8} = 0b00;
4312 def VEXTqf : VEXTq<"vext", "32", v4f32> {
4313 let Inst{11} = index{0};
4314 let Inst{10-8} = 0b000;
4317 // VTRN : Vector Transpose
4319 def VTRNd8 : N2VDShuffle<0b00, 0b00001, "vtrn", "8">;
4320 def VTRNd16 : N2VDShuffle<0b01, 0b00001, "vtrn", "16">;
4321 def VTRNd32 : N2VDShuffle<0b10, 0b00001, "vtrn", "32">;
4323 def VTRNq8 : N2VQShuffle<0b00, 0b00001, IIC_VPERMQ, "vtrn", "8">;
4324 def VTRNq16 : N2VQShuffle<0b01, 0b00001, IIC_VPERMQ, "vtrn", "16">;
4325 def VTRNq32 : N2VQShuffle<0b10, 0b00001, IIC_VPERMQ, "vtrn", "32">;
4327 // VUZP : Vector Unzip (Deinterleave)
4329 def VUZPd8 : N2VDShuffle<0b00, 0b00010, "vuzp", "8">;
4330 def VUZPd16 : N2VDShuffle<0b01, 0b00010, "vuzp", "16">;
4331 def VUZPd32 : N2VDShuffle<0b10, 0b00010, "vuzp", "32">;
4333 def VUZPq8 : N2VQShuffle<0b00, 0b00010, IIC_VPERMQ3, "vuzp", "8">;
4334 def VUZPq16 : N2VQShuffle<0b01, 0b00010, IIC_VPERMQ3, "vuzp", "16">;
4335 def VUZPq32 : N2VQShuffle<0b10, 0b00010, IIC_VPERMQ3, "vuzp", "32">;
4337 // VZIP : Vector Zip (Interleave)
4339 def VZIPd8 : N2VDShuffle<0b00, 0b00011, "vzip", "8">;
4340 def VZIPd16 : N2VDShuffle<0b01, 0b00011, "vzip", "16">;
4341 def VZIPd32 : N2VDShuffle<0b10, 0b00011, "vzip", "32">;
4343 def VZIPq8 : N2VQShuffle<0b00, 0b00011, IIC_VPERMQ3, "vzip", "8">;
4344 def VZIPq16 : N2VQShuffle<0b01, 0b00011, IIC_VPERMQ3, "vzip", "16">;
4345 def VZIPq32 : N2VQShuffle<0b10, 0b00011, IIC_VPERMQ3, "vzip", "32">;
4347 // Vector Table Lookup and Table Extension.
4349 // VTBL : Vector Table Lookup
4351 : N3V<1,1,0b11,0b1000,0,0, (outs DPR:$Vd),
4352 (ins DPR:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTB1,
4353 "vtbl", "8", "$Vd, \\{$Vn\\}, $Vm", "",
4354 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbl1 DPR:$Vn, DPR:$Vm)))]>;
4355 let hasExtraSrcRegAllocReq = 1 in {
4357 : N3V<1,1,0b11,0b1001,0,0, (outs DPR:$Vd),
4358 (ins DPR:$Vn, DPR:$tbl2, DPR:$Vm), NVTBLFrm, IIC_VTB2,
4359 "vtbl", "8", "$Vd, \\{$Vn, $tbl2\\}, $Vm", "", []>;
4361 : N3V<1,1,0b11,0b1010,0,0, (outs DPR:$Vd),
4362 (ins DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$Vm), NVTBLFrm, IIC_VTB3,
4363 "vtbl", "8", "$Vd, \\{$Vn, $tbl2, $tbl3\\}, $Vm", "", []>;
4365 : N3V<1,1,0b11,0b1011,0,0, (outs DPR:$Vd),
4366 (ins DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$Vm),
4368 "vtbl", "8", "$Vd, \\{$Vn, $tbl2, $tbl3, $tbl4\\}, $Vm", "", []>;
4369 } // hasExtraSrcRegAllocReq = 1
4372 : PseudoNeonI<(outs DPR:$dst), (ins QPR:$tbl, DPR:$src), IIC_VTB2, "", []>;
4374 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB3, "", []>;
4376 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB4, "", []>;
4378 // VTBX : Vector Table Extension
4380 : N3V<1,1,0b11,0b1000,1,0, (outs DPR:$Vd),
4381 (ins DPR:$orig, DPR:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTBX1,
4382 "vtbx", "8", "$Vd, \\{$Vn\\}, $Vm", "$orig = $Vd",
4383 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbx1
4384 DPR:$orig, DPR:$Vn, DPR:$Vm)))]>;
4385 let hasExtraSrcRegAllocReq = 1 in {
4387 : N3V<1,1,0b11,0b1001,1,0, (outs DPR:$Vd),
4388 (ins DPR:$orig, DPR:$Vn, DPR:$tbl2, DPR:$Vm), NVTBLFrm, IIC_VTBX2,
4389 "vtbx", "8", "$Vd, \\{$Vn, $tbl2\\}, $Vm", "$orig = $Vd", []>;
4391 : N3V<1,1,0b11,0b1010,1,0, (outs DPR:$Vd),
4392 (ins DPR:$orig, DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$Vm),
4393 NVTBLFrm, IIC_VTBX3,
4394 "vtbx", "8", "$Vd, \\{$Vn, $tbl2, $tbl3\\}, $Vm",
4397 : N3V<1,1,0b11,0b1011,1,0, (outs DPR:$Vd), (ins DPR:$orig, DPR:$Vn,
4398 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$Vm), NVTBLFrm, IIC_VTBX4,
4399 "vtbx", "8", "$Vd, \\{$Vn, $tbl2, $tbl3, $tbl4\\}, $Vm",
4401 } // hasExtraSrcRegAllocReq = 1
4404 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QPR:$tbl, DPR:$src),
4405 IIC_VTBX2, "$orig = $dst", []>;
4407 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
4408 IIC_VTBX3, "$orig = $dst", []>;
4410 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
4411 IIC_VTBX4, "$orig = $dst", []>;
4413 //===----------------------------------------------------------------------===//
4414 // NEON instructions for single-precision FP math
4415 //===----------------------------------------------------------------------===//
4417 class N2VSPat<SDNode OpNode, ValueType ResTy, ValueType OpTy, NeonI Inst>
4418 : NEONFPPat<(ResTy (OpNode SPR:$a)),
4419 (EXTRACT_SUBREG (OpTy (Inst (INSERT_SUBREG (OpTy (IMPLICIT_DEF)),
4423 class N3VSPat<SDNode OpNode, NeonI Inst>
4424 : NEONFPPat<(f32 (OpNode SPR:$a, SPR:$b)),
4425 (EXTRACT_SUBREG (v2f32
4426 (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
4428 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
4432 class N3VSMulOpPat<SDNode MulNode, SDNode OpNode, NeonI Inst>
4433 : NEONFPPat<(f32 (OpNode SPR:$acc, (f32 (MulNode SPR:$a, SPR:$b)))),
4434 (EXTRACT_SUBREG (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
4436 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
4438 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
4442 // These need separate instructions because they must use DPR_VFP2 register
4443 // class which have SPR sub-registers.
4445 // Vector Add Operations used for single-precision FP
4446 let neverHasSideEffects = 1 in
4447 def VADDfd_sfp : N3VS<0,0,0b00,0b1101,0, "vadd", "f32", v2f32, v2f32, fadd, 1>;
4448 def : N3VSPat<fadd, VADDfd_sfp>;
4450 // Vector Sub Operations used for single-precision FP
4451 let neverHasSideEffects = 1 in
4452 def VSUBfd_sfp : N3VS<0,0,0b10,0b1101,0, "vsub", "f32", v2f32, v2f32, fsub, 0>;
4453 def : N3VSPat<fsub, VSUBfd_sfp>;
4455 // Vector Multiply Operations used for single-precision FP
4456 let neverHasSideEffects = 1 in
4457 def VMULfd_sfp : N3VS<1,0,0b00,0b1101,1, "vmul", "f32", v2f32, v2f32, fmul, 1>;
4458 def : N3VSPat<fmul, VMULfd_sfp>;
4460 // Vector Multiply-Accumulate/Subtract used for single-precision FP
4461 // vml[as].f32 can cause 4-8 cycle stalls in following ASIMD instructions, so
4462 // we want to avoid them for now. e.g., alternating vmla/vadd instructions.
4464 //let neverHasSideEffects = 1 in
4465 //def VMLAfd_sfp : N3VSMulOp<0,0,0b00,0b1101,1, IIC_VMACD, "vmla", "f32",
4466 // v2f32, fmul, fadd>;
4467 //def : N3VSMulOpPat<fmul, fadd, VMLAfd_sfp>;
4469 //let neverHasSideEffects = 1 in
4470 //def VMLSfd_sfp : N3VSMulOp<0,0,0b10,0b1101,1, IIC_VMACD, "vmls", "f32",
4471 // v2f32, fmul, fsub>;
4472 //def : N3VSMulOpPat<fmul, fsub, VMLSfd_sfp>;
4474 // Vector Absolute used for single-precision FP
4475 let neverHasSideEffects = 1 in
4476 def VABSfd_sfp : N2V<0b11, 0b11, 0b10, 0b01, 0b01110, 0, 0,
4477 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), IIC_VUNAD,
4478 "vabs", "f32", "$dst, $src", "", []>;
4479 def : N2VSPat<fabs, f32, v2f32, VABSfd_sfp>;
4481 // Vector Negate used for single-precision FP
4482 let neverHasSideEffects = 1 in
4483 def VNEGfd_sfp : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
4484 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), IIC_VUNAD,
4485 "vneg", "f32", "$dst, $src", "", []>;
4486 def : N2VSPat<fneg, f32, v2f32, VNEGfd_sfp>;
4488 // Vector Maximum used for single-precision FP
4489 let neverHasSideEffects = 1 in
4490 def VMAXfd_sfp : N3V<0, 0, 0b00, 0b1111, 0, 0, (outs DPR_VFP2:$dst),
4491 (ins DPR_VFP2:$src1, DPR_VFP2:$src2), N3RegFrm, IIC_VBIND,
4492 "vmax", "f32", "$dst, $src1, $src2", "", []>;
4493 def : N3VSPat<NEONfmax, VMAXfd_sfp>;
4495 // Vector Minimum used for single-precision FP
4496 let neverHasSideEffects = 1 in
4497 def VMINfd_sfp : N3V<0, 0, 0b00, 0b1111, 0, 0, (outs DPR_VFP2:$dst),
4498 (ins DPR_VFP2:$src1, DPR_VFP2:$src2), N3RegFrm, IIC_VBIND,
4499 "vmin", "f32", "$dst, $src1, $src2", "", []>;
4500 def : N3VSPat<NEONfmin, VMINfd_sfp>;
4502 // Vector Convert between single-precision FP and integer
4503 let neverHasSideEffects = 1 in
4504 def VCVTf2sd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
4505 v2i32, v2f32, fp_to_sint>;
4506 def : N2VSPat<arm_ftosi, f32, v2f32, VCVTf2sd_sfp>;
4508 let neverHasSideEffects = 1 in
4509 def VCVTf2ud_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
4510 v2i32, v2f32, fp_to_uint>;
4511 def : N2VSPat<arm_ftoui, f32, v2f32, VCVTf2ud_sfp>;
4513 let neverHasSideEffects = 1 in
4514 def VCVTs2fd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
4515 v2f32, v2i32, sint_to_fp>;
4516 def : N2VSPat<arm_sitof, f32, v2i32, VCVTs2fd_sfp>;
4518 let neverHasSideEffects = 1 in
4519 def VCVTu2fd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
4520 v2f32, v2i32, uint_to_fp>;
4521 def : N2VSPat<arm_uitof, f32, v2i32, VCVTu2fd_sfp>;
4523 //===----------------------------------------------------------------------===//
4524 // Non-Instruction Patterns
4525 //===----------------------------------------------------------------------===//
4528 def : Pat<(v1i64 (bitconvert (v2i32 DPR:$src))), (v1i64 DPR:$src)>;
4529 def : Pat<(v1i64 (bitconvert (v4i16 DPR:$src))), (v1i64 DPR:$src)>;
4530 def : Pat<(v1i64 (bitconvert (v8i8 DPR:$src))), (v1i64 DPR:$src)>;
4531 def : Pat<(v1i64 (bitconvert (f64 DPR:$src))), (v1i64 DPR:$src)>;
4532 def : Pat<(v1i64 (bitconvert (v2f32 DPR:$src))), (v1i64 DPR:$src)>;
4533 def : Pat<(v2i32 (bitconvert (v1i64 DPR:$src))), (v2i32 DPR:$src)>;
4534 def : Pat<(v2i32 (bitconvert (v4i16 DPR:$src))), (v2i32 DPR:$src)>;
4535 def : Pat<(v2i32 (bitconvert (v8i8 DPR:$src))), (v2i32 DPR:$src)>;
4536 def : Pat<(v2i32 (bitconvert (f64 DPR:$src))), (v2i32 DPR:$src)>;
4537 def : Pat<(v2i32 (bitconvert (v2f32 DPR:$src))), (v2i32 DPR:$src)>;
4538 def : Pat<(v4i16 (bitconvert (v1i64 DPR:$src))), (v4i16 DPR:$src)>;
4539 def : Pat<(v4i16 (bitconvert (v2i32 DPR:$src))), (v4i16 DPR:$src)>;
4540 def : Pat<(v4i16 (bitconvert (v8i8 DPR:$src))), (v4i16 DPR:$src)>;
4541 def : Pat<(v4i16 (bitconvert (f64 DPR:$src))), (v4i16 DPR:$src)>;
4542 def : Pat<(v4i16 (bitconvert (v2f32 DPR:$src))), (v4i16 DPR:$src)>;
4543 def : Pat<(v8i8 (bitconvert (v1i64 DPR:$src))), (v8i8 DPR:$src)>;
4544 def : Pat<(v8i8 (bitconvert (v2i32 DPR:$src))), (v8i8 DPR:$src)>;
4545 def : Pat<(v8i8 (bitconvert (v4i16 DPR:$src))), (v8i8 DPR:$src)>;
4546 def : Pat<(v8i8 (bitconvert (f64 DPR:$src))), (v8i8 DPR:$src)>;
4547 def : Pat<(v8i8 (bitconvert (v2f32 DPR:$src))), (v8i8 DPR:$src)>;
4548 def : Pat<(f64 (bitconvert (v1i64 DPR:$src))), (f64 DPR:$src)>;
4549 def : Pat<(f64 (bitconvert (v2i32 DPR:$src))), (f64 DPR:$src)>;
4550 def : Pat<(f64 (bitconvert (v4i16 DPR:$src))), (f64 DPR:$src)>;
4551 def : Pat<(f64 (bitconvert (v8i8 DPR:$src))), (f64 DPR:$src)>;
4552 def : Pat<(f64 (bitconvert (v2f32 DPR:$src))), (f64 DPR:$src)>;
4553 def : Pat<(v2f32 (bitconvert (f64 DPR:$src))), (v2f32 DPR:$src)>;
4554 def : Pat<(v2f32 (bitconvert (v1i64 DPR:$src))), (v2f32 DPR:$src)>;
4555 def : Pat<(v2f32 (bitconvert (v2i32 DPR:$src))), (v2f32 DPR:$src)>;
4556 def : Pat<(v2f32 (bitconvert (v4i16 DPR:$src))), (v2f32 DPR:$src)>;
4557 def : Pat<(v2f32 (bitconvert (v8i8 DPR:$src))), (v2f32 DPR:$src)>;
4559 def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (v2i64 QPR:$src)>;
4560 def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (v2i64 QPR:$src)>;
4561 def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (v2i64 QPR:$src)>;
4562 def : Pat<(v2i64 (bitconvert (v2f64 QPR:$src))), (v2i64 QPR:$src)>;
4563 def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (v2i64 QPR:$src)>;
4564 def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (v4i32 QPR:$src)>;
4565 def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (v4i32 QPR:$src)>;
4566 def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (v4i32 QPR:$src)>;
4567 def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (v4i32 QPR:$src)>;
4568 def : Pat<(v4i32 (bitconvert (v4f32 QPR:$src))), (v4i32 QPR:$src)>;
4569 def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (v8i16 QPR:$src)>;
4570 def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (v8i16 QPR:$src)>;
4571 def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (v8i16 QPR:$src)>;
4572 def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (v8i16 QPR:$src)>;
4573 def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (v8i16 QPR:$src)>;
4574 def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (v16i8 QPR:$src)>;
4575 def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (v16i8 QPR:$src)>;
4576 def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (v16i8 QPR:$src)>;
4577 def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (v16i8 QPR:$src)>;
4578 def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (v16i8 QPR:$src)>;
4579 def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (v4f32 QPR:$src)>;
4580 def : Pat<(v4f32 (bitconvert (v4i32 QPR:$src))), (v4f32 QPR:$src)>;
4581 def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (v4f32 QPR:$src)>;
4582 def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (v4f32 QPR:$src)>;
4583 def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (v4f32 QPR:$src)>;
4584 def : Pat<(v2f64 (bitconvert (v2i64 QPR:$src))), (v2f64 QPR:$src)>;
4585 def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (v2f64 QPR:$src)>;
4586 def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 QPR:$src)>;
4587 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 QPR:$src)>;
4588 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 QPR:$src)>;