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 NEONvdup : SDNode<"ARMISD::VDUP", SDTypeProfile<1, 1, [SDTCisVec<0>]>>;
74 // VDUPLANE can produce a quad-register result from a double-register source,
75 // so the result is not constrained to match the source.
76 def NEONvduplane : SDNode<"ARMISD::VDUPLANE",
77 SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisVec<1>,
80 def SDTARMVEXT : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
81 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
82 def NEONvext : SDNode<"ARMISD::VEXT", SDTARMVEXT>;
84 def SDTARMVSHUF : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisSameAs<0, 1>]>;
85 def NEONvrev64 : SDNode<"ARMISD::VREV64", SDTARMVSHUF>;
86 def NEONvrev32 : SDNode<"ARMISD::VREV32", SDTARMVSHUF>;
87 def NEONvrev16 : SDNode<"ARMISD::VREV16", SDTARMVSHUF>;
89 def SDTARMVSHUF2 : SDTypeProfile<2, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
92 def NEONzip : SDNode<"ARMISD::VZIP", SDTARMVSHUF2>;
93 def NEONuzp : SDNode<"ARMISD::VUZP", SDTARMVSHUF2>;
94 def NEONtrn : SDNode<"ARMISD::VTRN", SDTARMVSHUF2>;
96 def SDTARMVMULL : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
98 def NEONvmulls : SDNode<"ARMISD::VMULLs", SDTARMVMULL>;
99 def NEONvmullu : SDNode<"ARMISD::VMULLu", SDTARMVMULL>;
101 def SDTARMFMAX : SDTypeProfile<1, 2, [SDTCisVT<0, f32>, SDTCisSameAs<0, 1>,
102 SDTCisSameAs<0, 2>]>;
103 def NEONfmax : SDNode<"ARMISD::FMAX", SDTARMFMAX>;
104 def NEONfmin : SDNode<"ARMISD::FMIN", SDTARMFMAX>;
106 def NEONimmAllZerosV: PatLeaf<(NEONvmovImm (i32 timm)), [{
107 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
108 unsigned EltBits = 0;
109 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
110 return (EltBits == 32 && EltVal == 0);
113 def NEONimmAllOnesV: PatLeaf<(NEONvmovImm (i32 timm)), [{
114 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
115 unsigned EltBits = 0;
116 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
117 return (EltBits == 8 && EltVal == 0xff);
120 //===----------------------------------------------------------------------===//
121 // NEON operand definitions
122 //===----------------------------------------------------------------------===//
124 def nModImm : Operand<i32> {
125 let PrintMethod = "printNEONModImmOperand";
128 //===----------------------------------------------------------------------===//
129 // NEON load / store instructions
130 //===----------------------------------------------------------------------===//
132 // Use VLDM to load a Q register as a D register pair.
133 // This is a pseudo instruction that is expanded to VLDMD after reg alloc.
135 : PseudoVFPLdStM<(outs QPR:$dst), (ins GPR:$Rn, ldstm_mode:$mode),
137 [(set QPR:$dst, (v2f64 (load GPR:$Rn)))]>;
139 // Use VSTM to store a Q register as a D register pair.
140 // This is a pseudo instruction that is expanded to VSTMD after reg alloc.
142 : PseudoVFPLdStM<(outs), (ins QPR:$src, GPR:$Rn, ldstm_mode:$mode),
144 [(store (v2f64 QPR:$src), GPR:$Rn)]>;
146 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
148 // Classes for VLD* pseudo-instructions with multi-register operands.
149 // These are expanded to real instructions after register allocation.
150 class VLDQPseudo<InstrItinClass itin>
151 : PseudoNLdSt<(outs QPR:$dst), (ins addrmode6:$addr), itin, "">;
152 class VLDQWBPseudo<InstrItinClass itin>
153 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
154 (ins addrmode6:$addr, am6offset:$offset), itin,
156 class VLDQQPseudo<InstrItinClass itin>
157 : PseudoNLdSt<(outs QQPR:$dst), (ins addrmode6:$addr), itin, "">;
158 class VLDQQWBPseudo<InstrItinClass itin>
159 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
160 (ins addrmode6:$addr, am6offset:$offset), itin,
162 class VLDQQQQWBPseudo<InstrItinClass itin>
163 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
164 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
165 "$addr.addr = $wb, $src = $dst">;
167 // VLD1 : Vector Load (multiple single elements)
168 class VLD1D<bits<4> op7_4, string Dt>
169 : NLdSt<0,0b10,0b0111,op7_4, (outs DPR:$Vd),
170 (ins addrmode6:$Rn), IIC_VLD1,
171 "vld1", Dt, "\\{$Vd\\}, $Rn", "", []> {
175 class VLD1Q<bits<4> op7_4, string Dt>
176 : NLdSt<0,0b10,0b1010,op7_4, (outs DPR:$Vd, DPR:$dst2),
177 (ins addrmode6:$Rn), IIC_VLD1x2,
178 "vld1", Dt, "\\{$Vd, $dst2\\}, $Rn", "", []> {
180 let Inst{5-4} = Rn{5-4};
183 def VLD1d8 : VLD1D<{0,0,0,?}, "8">;
184 def VLD1d16 : VLD1D<{0,1,0,?}, "16">;
185 def VLD1d32 : VLD1D<{1,0,0,?}, "32">;
186 def VLD1d64 : VLD1D<{1,1,0,?}, "64">;
188 def VLD1q8 : VLD1Q<{0,0,?,?}, "8">;
189 def VLD1q16 : VLD1Q<{0,1,?,?}, "16">;
190 def VLD1q32 : VLD1Q<{1,0,?,?}, "32">;
191 def VLD1q64 : VLD1Q<{1,1,?,?}, "64">;
193 def VLD1q8Pseudo : VLDQPseudo<IIC_VLD1x2>;
194 def VLD1q16Pseudo : VLDQPseudo<IIC_VLD1x2>;
195 def VLD1q32Pseudo : VLDQPseudo<IIC_VLD1x2>;
196 def VLD1q64Pseudo : VLDQPseudo<IIC_VLD1x2>;
198 // ...with address register writeback:
199 class VLD1DWB<bits<4> op7_4, string Dt>
200 : NLdSt<0,0b10,0b0111,op7_4, (outs DPR:$Vd, GPR:$wb),
201 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD1u,
202 "vld1", Dt, "\\{$Vd\\}, $Rn$Rm",
203 "$Rn.addr = $wb", []> {
206 class VLD1QWB<bits<4> op7_4, string Dt>
207 : NLdSt<0,0b10,0b1010,op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
208 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD1x2u,
209 "vld1", Dt, "\\{$Vd, $dst2\\}, $Rn$Rm",
210 "$Rn.addr = $wb", []> {
211 let Inst{5-4} = Rn{5-4};
214 def VLD1d8_UPD : VLD1DWB<{0,0,0,?}, "8">;
215 def VLD1d16_UPD : VLD1DWB<{0,1,0,?}, "16">;
216 def VLD1d32_UPD : VLD1DWB<{1,0,0,?}, "32">;
217 def VLD1d64_UPD : VLD1DWB<{1,1,0,?}, "64">;
219 def VLD1q8_UPD : VLD1QWB<{0,0,?,?}, "8">;
220 def VLD1q16_UPD : VLD1QWB<{0,1,?,?}, "16">;
221 def VLD1q32_UPD : VLD1QWB<{1,0,?,?}, "32">;
222 def VLD1q64_UPD : VLD1QWB<{1,1,?,?}, "64">;
224 def VLD1q8Pseudo_UPD : VLDQWBPseudo<IIC_VLD1x2u>;
225 def VLD1q16Pseudo_UPD : VLDQWBPseudo<IIC_VLD1x2u>;
226 def VLD1q32Pseudo_UPD : VLDQWBPseudo<IIC_VLD1x2u>;
227 def VLD1q64Pseudo_UPD : VLDQWBPseudo<IIC_VLD1x2u>;
229 // ...with 3 registers (some of these are only for the disassembler):
230 class VLD1D3<bits<4> op7_4, string Dt>
231 : NLdSt<0,0b10,0b0110,op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
232 (ins addrmode6:$Rn), IIC_VLD1x3, "vld1", Dt,
233 "\\{$Vd, $dst2, $dst3\\}, $Rn", "", []> {
237 class VLD1D3WB<bits<4> op7_4, string Dt>
238 : NLdSt<0,0b10,0b0110,op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
239 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD1x3u, "vld1", Dt,
240 "\\{$Vd, $dst2, $dst3\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
244 def VLD1d8T : VLD1D3<{0,0,0,?}, "8">;
245 def VLD1d16T : VLD1D3<{0,1,0,?}, "16">;
246 def VLD1d32T : VLD1D3<{1,0,0,?}, "32">;
247 def VLD1d64T : VLD1D3<{1,1,0,?}, "64">;
249 def VLD1d8T_UPD : VLD1D3WB<{0,0,0,?}, "8">;
250 def VLD1d16T_UPD : VLD1D3WB<{0,1,0,?}, "16">;
251 def VLD1d32T_UPD : VLD1D3WB<{1,0,0,?}, "32">;
252 def VLD1d64T_UPD : VLD1D3WB<{1,1,0,?}, "64">;
254 def VLD1d64TPseudo : VLDQQPseudo<IIC_VLD1x3>;
255 def VLD1d64TPseudo_UPD : VLDQQWBPseudo<IIC_VLD1x3u>;
257 // ...with 4 registers (some of these are only for the disassembler):
258 class VLD1D4<bits<4> op7_4, string Dt>
259 : NLdSt<0,0b10,0b0010,op7_4,(outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
260 (ins addrmode6:$Rn), IIC_VLD1x4, "vld1", Dt,
261 "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn", "", []> {
263 let Inst{5-4} = Rn{5-4};
265 class VLD1D4WB<bits<4> op7_4, string Dt>
266 : NLdSt<0,0b10,0b0010,op7_4,
267 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
268 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD4, "vld1", Dt,
269 "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn$Rm", "$Rn.addr = $wb",
271 let Inst{5-4} = Rn{5-4};
274 def VLD1d8Q : VLD1D4<{0,0,?,?}, "8">;
275 def VLD1d16Q : VLD1D4<{0,1,?,?}, "16">;
276 def VLD1d32Q : VLD1D4<{1,0,?,?}, "32">;
277 def VLD1d64Q : VLD1D4<{1,1,?,?}, "64">;
279 def VLD1d8Q_UPD : VLD1D4WB<{0,0,?,?}, "8">;
280 def VLD1d16Q_UPD : VLD1D4WB<{0,1,?,?}, "16">;
281 def VLD1d32Q_UPD : VLD1D4WB<{1,0,?,?}, "32">;
282 def VLD1d64Q_UPD : VLD1D4WB<{1,1,?,?}, "64">;
284 def VLD1d64QPseudo : VLDQQPseudo<IIC_VLD1x4>;
285 def VLD1d64QPseudo_UPD : VLDQQWBPseudo<IIC_VLD1x4u>;
287 // VLD2 : Vector Load (multiple 2-element structures)
288 class VLD2D<bits<4> op11_8, bits<4> op7_4, string Dt>
289 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2),
290 (ins addrmode6:$Rn), IIC_VLD2,
291 "vld2", Dt, "\\{$Vd, $dst2\\}, $Rn", "", []> {
293 let Inst{5-4} = Rn{5-4};
295 class VLD2Q<bits<4> op7_4, string Dt>
296 : NLdSt<0, 0b10, 0b0011, op7_4,
297 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
298 (ins addrmode6:$Rn), IIC_VLD2x2,
299 "vld2", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn", "", []> {
301 let Inst{5-4} = Rn{5-4};
304 def VLD2d8 : VLD2D<0b1000, {0,0,?,?}, "8">;
305 def VLD2d16 : VLD2D<0b1000, {0,1,?,?}, "16">;
306 def VLD2d32 : VLD2D<0b1000, {1,0,?,?}, "32">;
308 def VLD2q8 : VLD2Q<{0,0,?,?}, "8">;
309 def VLD2q16 : VLD2Q<{0,1,?,?}, "16">;
310 def VLD2q32 : VLD2Q<{1,0,?,?}, "32">;
312 def VLD2d8Pseudo : VLDQPseudo<IIC_VLD2>;
313 def VLD2d16Pseudo : VLDQPseudo<IIC_VLD2>;
314 def VLD2d32Pseudo : VLDQPseudo<IIC_VLD2>;
316 def VLD2q8Pseudo : VLDQQPseudo<IIC_VLD2x2>;
317 def VLD2q16Pseudo : VLDQQPseudo<IIC_VLD2x2>;
318 def VLD2q32Pseudo : VLDQQPseudo<IIC_VLD2x2>;
320 // ...with address register writeback:
321 class VLD2DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
322 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
323 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD2u,
324 "vld2", Dt, "\\{$Vd, $dst2\\}, $Rn$Rm",
325 "$Rn.addr = $wb", []> {
326 let Inst{5-4} = Rn{5-4};
328 class VLD2QWB<bits<4> op7_4, string Dt>
329 : NLdSt<0, 0b10, 0b0011, op7_4,
330 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
331 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD2x2u,
332 "vld2", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn$Rm",
333 "$Rn.addr = $wb", []> {
334 let Inst{5-4} = Rn{5-4};
337 def VLD2d8_UPD : VLD2DWB<0b1000, {0,0,?,?}, "8">;
338 def VLD2d16_UPD : VLD2DWB<0b1000, {0,1,?,?}, "16">;
339 def VLD2d32_UPD : VLD2DWB<0b1000, {1,0,?,?}, "32">;
341 def VLD2q8_UPD : VLD2QWB<{0,0,?,?}, "8">;
342 def VLD2q16_UPD : VLD2QWB<{0,1,?,?}, "16">;
343 def VLD2q32_UPD : VLD2QWB<{1,0,?,?}, "32">;
345 def VLD2d8Pseudo_UPD : VLDQWBPseudo<IIC_VLD2u>;
346 def VLD2d16Pseudo_UPD : VLDQWBPseudo<IIC_VLD2u>;
347 def VLD2d32Pseudo_UPD : VLDQWBPseudo<IIC_VLD2u>;
349 def VLD2q8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD2x2u>;
350 def VLD2q16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD2x2u>;
351 def VLD2q32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD2x2u>;
353 // ...with double-spaced registers (for disassembly only):
354 def VLD2b8 : VLD2D<0b1001, {0,0,?,?}, "8">;
355 def VLD2b16 : VLD2D<0b1001, {0,1,?,?}, "16">;
356 def VLD2b32 : VLD2D<0b1001, {1,0,?,?}, "32">;
357 def VLD2b8_UPD : VLD2DWB<0b1001, {0,0,?,?}, "8">;
358 def VLD2b16_UPD : VLD2DWB<0b1001, {0,1,?,?}, "16">;
359 def VLD2b32_UPD : VLD2DWB<0b1001, {1,0,?,?}, "32">;
361 // VLD3 : Vector Load (multiple 3-element structures)
362 class VLD3D<bits<4> op11_8, bits<4> op7_4, string Dt>
363 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
364 (ins addrmode6:$Rn), IIC_VLD3,
365 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn", "", []> {
370 def VLD3d8 : VLD3D<0b0100, {0,0,0,?}, "8">;
371 def VLD3d16 : VLD3D<0b0100, {0,1,0,?}, "16">;
372 def VLD3d32 : VLD3D<0b0100, {1,0,0,?}, "32">;
374 def VLD3d8Pseudo : VLDQQPseudo<IIC_VLD3>;
375 def VLD3d16Pseudo : VLDQQPseudo<IIC_VLD3>;
376 def VLD3d32Pseudo : VLDQQPseudo<IIC_VLD3>;
378 // ...with address register writeback:
379 class VLD3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
380 : NLdSt<0, 0b10, op11_8, op7_4,
381 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
382 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD3u,
383 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn$Rm",
384 "$Rn.addr = $wb", []> {
388 def VLD3d8_UPD : VLD3DWB<0b0100, {0,0,0,?}, "8">;
389 def VLD3d16_UPD : VLD3DWB<0b0100, {0,1,0,?}, "16">;
390 def VLD3d32_UPD : VLD3DWB<0b0100, {1,0,0,?}, "32">;
392 def VLD3d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
393 def VLD3d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
394 def VLD3d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
396 // ...with double-spaced registers (non-updating versions for disassembly only):
397 def VLD3q8 : VLD3D<0b0101, {0,0,0,?}, "8">;
398 def VLD3q16 : VLD3D<0b0101, {0,1,0,?}, "16">;
399 def VLD3q32 : VLD3D<0b0101, {1,0,0,?}, "32">;
400 def VLD3q8_UPD : VLD3DWB<0b0101, {0,0,0,?}, "8">;
401 def VLD3q16_UPD : VLD3DWB<0b0101, {0,1,0,?}, "16">;
402 def VLD3q32_UPD : VLD3DWB<0b0101, {1,0,0,?}, "32">;
404 def VLD3q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
405 def VLD3q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
406 def VLD3q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
408 // ...alternate versions to be allocated odd register numbers:
409 def VLD3q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
410 def VLD3q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
411 def VLD3q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
413 // VLD4 : Vector Load (multiple 4-element structures)
414 class VLD4D<bits<4> op11_8, bits<4> op7_4, string Dt>
415 : NLdSt<0, 0b10, op11_8, op7_4,
416 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
417 (ins addrmode6:$Rn), IIC_VLD4,
418 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn", "", []> {
420 let Inst{5-4} = Rn{5-4};
423 def VLD4d8 : VLD4D<0b0000, {0,0,?,?}, "8">;
424 def VLD4d16 : VLD4D<0b0000, {0,1,?,?}, "16">;
425 def VLD4d32 : VLD4D<0b0000, {1,0,?,?}, "32">;
427 def VLD4d8Pseudo : VLDQQPseudo<IIC_VLD4>;
428 def VLD4d16Pseudo : VLDQQPseudo<IIC_VLD4>;
429 def VLD4d32Pseudo : VLDQQPseudo<IIC_VLD4>;
431 // ...with address register writeback:
432 class VLD4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
433 : NLdSt<0, 0b10, op11_8, op7_4,
434 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
435 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD4,
436 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn$Rm",
437 "$Rn.addr = $wb", []> {
438 let Inst{5-4} = Rn{5-4};
441 def VLD4d8_UPD : VLD4DWB<0b0000, {0,0,?,?}, "8">;
442 def VLD4d16_UPD : VLD4DWB<0b0000, {0,1,?,?}, "16">;
443 def VLD4d32_UPD : VLD4DWB<0b0000, {1,0,?,?}, "32">;
445 def VLD4d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4>;
446 def VLD4d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4>;
447 def VLD4d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4>;
449 // ...with double-spaced registers (non-updating versions for disassembly only):
450 def VLD4q8 : VLD4D<0b0001, {0,0,?,?}, "8">;
451 def VLD4q16 : VLD4D<0b0001, {0,1,?,?}, "16">;
452 def VLD4q32 : VLD4D<0b0001, {1,0,?,?}, "32">;
453 def VLD4q8_UPD : VLD4DWB<0b0001, {0,0,?,?}, "8">;
454 def VLD4q16_UPD : VLD4DWB<0b0001, {0,1,?,?}, "16">;
455 def VLD4q32_UPD : VLD4DWB<0b0001, {1,0,?,?}, "32">;
457 def VLD4q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4>;
458 def VLD4q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4>;
459 def VLD4q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4>;
461 // ...alternate versions to be allocated odd register numbers:
462 def VLD4q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4>;
463 def VLD4q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4>;
464 def VLD4q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4>;
466 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
468 // Classes for VLD*LN pseudo-instructions with multi-register operands.
469 // These are expanded to real instructions after register allocation.
470 class VLDQLNPseudo<InstrItinClass itin>
471 : PseudoNLdSt<(outs QPR:$dst),
472 (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
473 itin, "$src = $dst">;
474 class VLDQLNWBPseudo<InstrItinClass itin>
475 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
476 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
477 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
478 class VLDQQLNPseudo<InstrItinClass itin>
479 : PseudoNLdSt<(outs QQPR:$dst),
480 (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
481 itin, "$src = $dst">;
482 class VLDQQLNWBPseudo<InstrItinClass itin>
483 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
484 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
485 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
486 class VLDQQQQLNPseudo<InstrItinClass itin>
487 : PseudoNLdSt<(outs QQQQPR:$dst),
488 (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
489 itin, "$src = $dst">;
490 class VLDQQQQLNWBPseudo<InstrItinClass itin>
491 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
492 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
493 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
495 // VLD1LN : Vector Load (single element to one lane)
496 class VLD1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
498 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd),
499 (ins addrmode6:$Rn, DPR:$src, nohash_imm:$lane),
500 IIC_VLD1ln, "vld1", Dt, "\\{$Vd[$lane]\\}, $Rn",
502 [(set DPR:$Vd, (vector_insert (Ty DPR:$src),
503 (i32 (LoadOp addrmode6:$Rn)),
507 class VLD1QLNPseudo<ValueType Ty, PatFrag LoadOp> : VLDQLNPseudo<IIC_VLD1ln> {
508 let Pattern = [(set QPR:$dst, (vector_insert (Ty QPR:$src),
509 (i32 (LoadOp addrmode6:$addr)),
513 def VLD1LNd8 : VLD1LN<0b0000, {?,?,?,0}, "8", v8i8, extloadi8> {
514 let Inst{7-5} = lane{2-0};
516 def VLD1LNd16 : VLD1LN<0b0100, {?,?,0,?}, "16", v4i16, extloadi16> {
517 let Inst{7-6} = lane{1-0};
520 def VLD1LNd32 : VLD1LN<0b1000, {?,0,?,?}, "32", v2i32, load> {
521 let Inst{7} = lane{0};
526 def VLD1LNq8Pseudo : VLD1QLNPseudo<v16i8, extloadi8>;
527 def VLD1LNq16Pseudo : VLD1QLNPseudo<v8i16, extloadi16>;
528 def VLD1LNq32Pseudo : VLD1QLNPseudo<v4i32, load>;
530 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
532 // ...with address register writeback:
533 class VLD1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
534 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, GPR:$wb),
535 (ins addrmode6:$Rn, am6offset:$Rm,
536 DPR:$src, nohash_imm:$lane), IIC_VLD1lnu, "vld1", Dt,
537 "\\{$Vd[$lane]\\}, $Rn$Rm",
538 "$src = $Vd, $Rn.addr = $wb", []>;
540 def VLD1LNd8_UPD : VLD1LNWB<0b0000, {?,?,?,0}, "8"> {
541 let Inst{7-5} = lane{2-0};
543 def VLD1LNd16_UPD : VLD1LNWB<0b0100, {?,?,0,?}, "16"> {
544 let Inst{7-6} = lane{1-0};
547 def VLD1LNd32_UPD : VLD1LNWB<0b1000, {?,0,?,?}, "32"> {
548 let Inst{7} = lane{0};
553 def VLD1LNq8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
554 def VLD1LNq16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
555 def VLD1LNq32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
557 // VLD2LN : Vector Load (single 2-element structure to one lane)
558 class VLD2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
559 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2),
560 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, nohash_imm:$lane),
561 IIC_VLD2ln, "vld2", Dt, "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn",
562 "$src1 = $Vd, $src2 = $dst2", []> {
567 def VLD2LNd8 : VLD2LN<0b0001, {?,?,?,?}, "8"> {
568 let Inst{7-5} = lane{2-0};
570 def VLD2LNd16 : VLD2LN<0b0101, {?,?,0,?}, "16"> {
571 let Inst{7-6} = lane{1-0};
573 def VLD2LNd32 : VLD2LN<0b1001, {?,0,0,?}, "32"> {
574 let Inst{7} = lane{0};
577 def VLD2LNd8Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
578 def VLD2LNd16Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
579 def VLD2LNd32Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
581 // ...with double-spaced registers:
582 def VLD2LNq16 : VLD2LN<0b0101, {?,?,1,?}, "16"> {
583 let Inst{7-6} = lane{1-0};
585 def VLD2LNq32 : VLD2LN<0b1001, {?,1,0,?}, "32"> {
586 let Inst{7} = lane{0};
589 def VLD2LNq16Pseudo : VLDQQLNPseudo<IIC_VLD2ln>;
590 def VLD2LNq32Pseudo : VLDQQLNPseudo<IIC_VLD2ln>;
592 // ...with address register writeback:
593 class VLD2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
594 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
595 (ins addrmode6:$Rn, am6offset:$Rm,
596 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VLD2lnu, "vld2", Dt,
597 "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn$Rm",
598 "$src1 = $Vd, $src2 = $dst2, $Rn.addr = $wb", []> {
602 def VLD2LNd8_UPD : VLD2LNWB<0b0001, {?,?,?,?}, "8"> {
603 let Inst{7-5} = lane{2-0};
605 def VLD2LNd16_UPD : VLD2LNWB<0b0101, {?,?,0,?}, "16"> {
606 let Inst{7-6} = lane{1-0};
608 def VLD2LNd32_UPD : VLD2LNWB<0b1001, {?,0,0,?}, "32"> {
609 let Inst{7} = lane{0};
612 def VLD2LNd8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
613 def VLD2LNd16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
614 def VLD2LNd32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
616 def VLD2LNq16_UPD : VLD2LNWB<0b0101, {?,?,1,?}, "16"> {
617 let Inst{7-6} = lane{1-0};
619 def VLD2LNq32_UPD : VLD2LNWB<0b1001, {?,1,0,?}, "32"> {
620 let Inst{7} = lane{0};
623 def VLD2LNq16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
624 def VLD2LNq32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
626 // VLD3LN : Vector Load (single 3-element structure to one lane)
627 class VLD3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
628 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
629 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3,
630 nohash_imm:$lane), IIC_VLD3ln, "vld3", Dt,
631 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn",
632 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3", []> {
636 def VLD3LNd8 : VLD3LN<0b0010, {?,?,?,0}, "8"> {
637 let Inst{7-5} = lane{2-0};
639 def VLD3LNd16 : VLD3LN<0b0110, {?,?,0,0}, "16"> {
640 let Inst{7-6} = lane{1-0};
642 def VLD3LNd32 : VLD3LN<0b1010, {?,0,0,0}, "32"> {
643 let Inst{7} = lane{0};
646 def VLD3LNd8Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
647 def VLD3LNd16Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
648 def VLD3LNd32Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
650 // ...with double-spaced registers:
651 def VLD3LNq16 : VLD3LN<0b0110, {?,?,1,0}, "16"> {
652 let Inst{7-6} = lane{1-0};
654 def VLD3LNq32 : VLD3LN<0b1010, {?,1,0,0}, "32"> {
655 let Inst{7} = lane{0};
658 def VLD3LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
659 def VLD3LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
661 // ...with address register writeback:
662 class VLD3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
663 : NLdStLn<1, 0b10, op11_8, op7_4,
664 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
665 (ins addrmode6:$Rn, am6offset:$Rm,
666 DPR:$src1, DPR:$src2, DPR:$src3, nohash_imm:$lane),
667 IIC_VLD3lnu, "vld3", Dt,
668 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn$Rm",
669 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $Rn.addr = $wb",
672 def VLD3LNd8_UPD : VLD3LNWB<0b0010, {?,?,?,0}, "8"> {
673 let Inst{7-5} = lane{2-0};
675 def VLD3LNd16_UPD : VLD3LNWB<0b0110, {?,?,0,0}, "16"> {
676 let Inst{7-6} = lane{1-0};
678 def VLD3LNd32_UPD : VLD3LNWB<0b1010, {?,0,0,0}, "32"> {
679 let Inst{7} = lane{0};
682 def VLD3LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
683 def VLD3LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
684 def VLD3LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
686 def VLD3LNq16_UPD : VLD3LNWB<0b0110, {?,?,1,0}, "16"> {
687 let Inst{7-6} = lane{1-0};
689 def VLD3LNq32_UPD : VLD3LNWB<0b1010, {?,1,0,0}, "32"> {
690 let Inst{7} = lane{0};
693 def VLD3LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
694 def VLD3LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
696 // VLD4LN : Vector Load (single 4-element structure to one lane)
697 class VLD4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
698 : NLdStLn<1, 0b10, op11_8, op7_4,
699 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
700 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
701 nohash_imm:$lane), IIC_VLD4ln, "vld4", Dt,
702 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn",
703 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4", []> {
708 def VLD4LNd8 : VLD4LN<0b0011, {?,?,?,?}, "8"> {
709 let Inst{7-5} = lane{2-0};
711 def VLD4LNd16 : VLD4LN<0b0111, {?,?,0,?}, "16"> {
712 let Inst{7-6} = lane{1-0};
714 def VLD4LNd32 : VLD4LN<0b1011, {?,0,?,?}, "32"> {
715 let Inst{7} = lane{0};
719 def VLD4LNd8Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
720 def VLD4LNd16Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
721 def VLD4LNd32Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
723 // ...with double-spaced registers:
724 def VLD4LNq16 : VLD4LN<0b0111, {?,?,1,?}, "16"> {
725 let Inst{7-6} = lane{1-0};
727 def VLD4LNq32 : VLD4LN<0b1011, {?,1,?,?}, "32"> {
728 let Inst{7} = lane{0};
732 def VLD4LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>;
733 def VLD4LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>;
735 // ...with address register writeback:
736 class VLD4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
737 : NLdStLn<1, 0b10, op11_8, op7_4,
738 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
739 (ins addrmode6:$Rn, am6offset:$Rm,
740 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
741 IIC_VLD4ln, "vld4", Dt,
742 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn$Rm",
743 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4, $Rn.addr = $wb",
748 def VLD4LNd8_UPD : VLD4LNWB<0b0011, {?,?,?,?}, "8"> {
749 let Inst{7-5} = lane{2-0};
751 def VLD4LNd16_UPD : VLD4LNWB<0b0111, {?,?,0,?}, "16"> {
752 let Inst{7-6} = lane{1-0};
754 def VLD4LNd32_UPD : VLD4LNWB<0b1011, {?,0,?,?}, "32"> {
755 let Inst{7} = lane{0};
759 def VLD4LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
760 def VLD4LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
761 def VLD4LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
763 def VLD4LNq16_UPD : VLD4LNWB<0b0111, {?,?,1,?}, "16"> {
764 let Inst{7-6} = lane{1-0};
766 def VLD4LNq32_UPD : VLD4LNWB<0b1011, {?,1,?,?}, "32"> {
767 let Inst{7} = lane{0};
771 def VLD4LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
772 def VLD4LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
774 // VLD1DUP : Vector Load (single element to all lanes)
775 // VLD2DUP : Vector Load (single 2-element structure to all lanes)
776 // VLD3DUP : Vector Load (single 3-element structure to all lanes)
777 // VLD4DUP : Vector Load (single 4-element structure to all lanes)
778 // FIXME: Not yet implemented.
779 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
781 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
783 // Classes for VST* pseudo-instructions with multi-register operands.
784 // These are expanded to real instructions after register allocation.
785 class VSTQPseudo<InstrItinClass itin>
786 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src), itin, "">;
787 class VSTQWBPseudo<InstrItinClass itin>
788 : PseudoNLdSt<(outs GPR:$wb),
789 (ins addrmode6:$addr, am6offset:$offset, QPR:$src), itin,
791 class VSTQQPseudo<InstrItinClass itin>
792 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src), itin, "">;
793 class VSTQQWBPseudo<InstrItinClass itin>
794 : PseudoNLdSt<(outs GPR:$wb),
795 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src), itin,
797 class VSTQQQQWBPseudo<InstrItinClass itin>
798 : PseudoNLdSt<(outs GPR:$wb),
799 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
802 // VST1 : Vector Store (multiple single elements)
803 class VST1D<bits<4> op7_4, string Dt>
804 : NLdSt<0,0b00,0b0111,op7_4, (outs), (ins addrmode6:$Rn, DPR:$Vd),
805 IIC_VST1, "vst1", Dt, "\\{$Vd\\}, $Rn", "", []> {
809 class VST1Q<bits<4> op7_4, string Dt>
810 : NLdSt<0,0b00,0b1010,op7_4, (outs),
811 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2), IIC_VST1x2,
812 "vst1", Dt, "\\{$Vd, $src2\\}, $Rn", "", []> {
814 let Inst{5-4} = Rn{5-4};
817 def VST1d8 : VST1D<{0,0,0,?}, "8">;
818 def VST1d16 : VST1D<{0,1,0,?}, "16">;
819 def VST1d32 : VST1D<{1,0,0,?}, "32">;
820 def VST1d64 : VST1D<{1,1,0,?}, "64">;
822 def VST1q8 : VST1Q<{0,0,?,?}, "8">;
823 def VST1q16 : VST1Q<{0,1,?,?}, "16">;
824 def VST1q32 : VST1Q<{1,0,?,?}, "32">;
825 def VST1q64 : VST1Q<{1,1,?,?}, "64">;
827 def VST1q8Pseudo : VSTQPseudo<IIC_VST1x2>;
828 def VST1q16Pseudo : VSTQPseudo<IIC_VST1x2>;
829 def VST1q32Pseudo : VSTQPseudo<IIC_VST1x2>;
830 def VST1q64Pseudo : VSTQPseudo<IIC_VST1x2>;
832 // ...with address register writeback:
833 class VST1DWB<bits<4> op7_4, string Dt>
834 : NLdSt<0, 0b00, 0b0111, op7_4, (outs GPR:$wb),
835 (ins addrmode6:$Rn, am6offset:$Rm, DPR:$Vd), IIC_VST1u,
836 "vst1", Dt, "\\{$Vd\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
839 class VST1QWB<bits<4> op7_4, string Dt>
840 : NLdSt<0, 0b00, 0b1010, op7_4, (outs GPR:$wb),
841 (ins addrmode6:$Rn, am6offset:$Rm, DPR:$Vd, DPR:$src2),
842 IIC_VST1x2u, "vst1", Dt, "\\{$Vd, $src2\\}, $Rn$Rm",
843 "$Rn.addr = $wb", []> {
844 let Inst{5-4} = Rn{5-4};
847 def VST1d8_UPD : VST1DWB<{0,0,0,?}, "8">;
848 def VST1d16_UPD : VST1DWB<{0,1,0,?}, "16">;
849 def VST1d32_UPD : VST1DWB<{1,0,0,?}, "32">;
850 def VST1d64_UPD : VST1DWB<{1,1,0,?}, "64">;
852 def VST1q8_UPD : VST1QWB<{0,0,?,?}, "8">;
853 def VST1q16_UPD : VST1QWB<{0,1,?,?}, "16">;
854 def VST1q32_UPD : VST1QWB<{1,0,?,?}, "32">;
855 def VST1q64_UPD : VST1QWB<{1,1,?,?}, "64">;
857 def VST1q8Pseudo_UPD : VSTQWBPseudo<IIC_VST1x2u>;
858 def VST1q16Pseudo_UPD : VSTQWBPseudo<IIC_VST1x2u>;
859 def VST1q32Pseudo_UPD : VSTQWBPseudo<IIC_VST1x2u>;
860 def VST1q64Pseudo_UPD : VSTQWBPseudo<IIC_VST1x2u>;
862 // ...with 3 registers (some of these are only for the disassembler):
863 class VST1D3<bits<4> op7_4, string Dt>
864 : NLdSt<0, 0b00, 0b0110, op7_4, (outs),
865 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3),
866 IIC_VST1x3, "vst1", Dt, "\\{$Vd, $src2, $src3\\}, $Rn", "", []> {
870 class VST1D3WB<bits<4> op7_4, string Dt>
871 : NLdSt<0, 0b00, 0b0110, op7_4, (outs GPR:$wb),
872 (ins addrmode6:$Rn, am6offset:$Rm,
873 DPR:$Vd, DPR:$src2, DPR:$src3),
874 IIC_VST1x3u, "vst1", Dt, "\\{$Vd, $src2, $src3\\}, $Rn$Rm",
875 "$Rn.addr = $wb", []> {
879 def VST1d8T : VST1D3<{0,0,0,?}, "8">;
880 def VST1d16T : VST1D3<{0,1,0,?}, "16">;
881 def VST1d32T : VST1D3<{1,0,0,?}, "32">;
882 def VST1d64T : VST1D3<{1,1,0,?}, "64">;
884 def VST1d8T_UPD : VST1D3WB<{0,0,0,?}, "8">;
885 def VST1d16T_UPD : VST1D3WB<{0,1,0,?}, "16">;
886 def VST1d32T_UPD : VST1D3WB<{1,0,0,?}, "32">;
887 def VST1d64T_UPD : VST1D3WB<{1,1,0,?}, "64">;
889 def VST1d64TPseudo : VSTQQPseudo<IIC_VST1x3>;
890 def VST1d64TPseudo_UPD : VSTQQWBPseudo<IIC_VST1x3u>;
892 // ...with 4 registers (some of these are only for the disassembler):
893 class VST1D4<bits<4> op7_4, string Dt>
894 : NLdSt<0, 0b00, 0b0010, op7_4, (outs),
895 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
896 IIC_VST1x4, "vst1", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn", "",
899 let Inst{5-4} = Rn{5-4};
901 class VST1D4WB<bits<4> op7_4, string Dt>
902 : NLdSt<0, 0b00, 0b0010, op7_4, (outs GPR:$wb),
903 (ins addrmode6:$Rn, am6offset:$Rm,
904 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST1x4u,
905 "vst1", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
906 "$Rn.addr = $wb", []> {
907 let Inst{5-4} = Rn{5-4};
910 def VST1d8Q : VST1D4<{0,0,?,?}, "8">;
911 def VST1d16Q : VST1D4<{0,1,?,?}, "16">;
912 def VST1d32Q : VST1D4<{1,0,?,?}, "32">;
913 def VST1d64Q : VST1D4<{1,1,?,?}, "64">;
915 def VST1d8Q_UPD : VST1D4WB<{0,0,?,?}, "8">;
916 def VST1d16Q_UPD : VST1D4WB<{0,1,?,?}, "16">;
917 def VST1d32Q_UPD : VST1D4WB<{1,0,?,?}, "32">;
918 def VST1d64Q_UPD : VST1D4WB<{1,1,?,?}, "64">;
920 def VST1d64QPseudo : VSTQQPseudo<IIC_VST1x4>;
921 def VST1d64QPseudo_UPD : VSTQQWBPseudo<IIC_VST1x4u>;
923 // VST2 : Vector Store (multiple 2-element structures)
924 class VST2D<bits<4> op11_8, bits<4> op7_4, string Dt>
925 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
926 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2),
927 IIC_VST2, "vst2", Dt, "\\{$Vd, $src2\\}, $Rn", "", []> {
929 let Inst{5-4} = Rn{5-4};
931 class VST2Q<bits<4> op7_4, string Dt>
932 : NLdSt<0, 0b00, 0b0011, op7_4, (outs),
933 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
934 IIC_VST2x2, "vst2", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn",
937 let Inst{5-4} = Rn{5-4};
940 def VST2d8 : VST2D<0b1000, {0,0,?,?}, "8">;
941 def VST2d16 : VST2D<0b1000, {0,1,?,?}, "16">;
942 def VST2d32 : VST2D<0b1000, {1,0,?,?}, "32">;
944 def VST2q8 : VST2Q<{0,0,?,?}, "8">;
945 def VST2q16 : VST2Q<{0,1,?,?}, "16">;
946 def VST2q32 : VST2Q<{1,0,?,?}, "32">;
948 def VST2d8Pseudo : VSTQPseudo<IIC_VST2>;
949 def VST2d16Pseudo : VSTQPseudo<IIC_VST2>;
950 def VST2d32Pseudo : VSTQPseudo<IIC_VST2>;
952 def VST2q8Pseudo : VSTQQPseudo<IIC_VST2x2>;
953 def VST2q16Pseudo : VSTQQPseudo<IIC_VST2x2>;
954 def VST2q32Pseudo : VSTQQPseudo<IIC_VST2x2>;
956 // ...with address register writeback:
957 class VST2DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
958 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
959 (ins addrmode6:$Rn, am6offset:$Rm, DPR:$Vd, DPR:$src2),
960 IIC_VST2u, "vst2", Dt, "\\{$Vd, $src2\\}, $Rn$Rm",
961 "$Rn.addr = $wb", []> {
962 let Inst{5-4} = Rn{5-4};
964 class VST2QWB<bits<4> op7_4, string Dt>
965 : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
966 (ins addrmode6:$Rn, am6offset:$Rm,
967 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST2x2u,
968 "vst2", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
969 "$Rn.addr = $wb", []> {
970 let Inst{5-4} = Rn{5-4};
973 def VST2d8_UPD : VST2DWB<0b1000, {0,0,?,?}, "8">;
974 def VST2d16_UPD : VST2DWB<0b1000, {0,1,?,?}, "16">;
975 def VST2d32_UPD : VST2DWB<0b1000, {1,0,?,?}, "32">;
977 def VST2q8_UPD : VST2QWB<{0,0,?,?}, "8">;
978 def VST2q16_UPD : VST2QWB<{0,1,?,?}, "16">;
979 def VST2q32_UPD : VST2QWB<{1,0,?,?}, "32">;
981 def VST2d8Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
982 def VST2d16Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
983 def VST2d32Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
985 def VST2q8Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
986 def VST2q16Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
987 def VST2q32Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
989 // ...with double-spaced registers (for disassembly only):
990 def VST2b8 : VST2D<0b1001, {0,0,?,?}, "8">;
991 def VST2b16 : VST2D<0b1001, {0,1,?,?}, "16">;
992 def VST2b32 : VST2D<0b1001, {1,0,?,?}, "32">;
993 def VST2b8_UPD : VST2DWB<0b1001, {0,0,?,?}, "8">;
994 def VST2b16_UPD : VST2DWB<0b1001, {0,1,?,?}, "16">;
995 def VST2b32_UPD : VST2DWB<0b1001, {1,0,?,?}, "32">;
997 // VST3 : Vector Store (multiple 3-element structures)
998 class VST3D<bits<4> op11_8, bits<4> op7_4, string Dt>
999 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1000 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3,
1001 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn", "", []> {
1003 let Inst{4} = Rn{4};
1006 def VST3d8 : VST3D<0b0100, {0,0,0,?}, "8">;
1007 def VST3d16 : VST3D<0b0100, {0,1,0,?}, "16">;
1008 def VST3d32 : VST3D<0b0100, {1,0,0,?}, "32">;
1010 def VST3d8Pseudo : VSTQQPseudo<IIC_VST3>;
1011 def VST3d16Pseudo : VSTQQPseudo<IIC_VST3>;
1012 def VST3d32Pseudo : VSTQQPseudo<IIC_VST3>;
1014 // ...with address register writeback:
1015 class VST3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1016 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1017 (ins addrmode6:$Rn, am6offset:$Rm,
1018 DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3u,
1019 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn$Rm",
1020 "$Rn.addr = $wb", []> {
1021 let Inst{4} = Rn{4};
1024 def VST3d8_UPD : VST3DWB<0b0100, {0,0,0,?}, "8">;
1025 def VST3d16_UPD : VST3DWB<0b0100, {0,1,0,?}, "16">;
1026 def VST3d32_UPD : VST3DWB<0b0100, {1,0,0,?}, "32">;
1028 def VST3d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1029 def VST3d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1030 def VST3d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1032 // ...with double-spaced registers (non-updating versions for disassembly only):
1033 def VST3q8 : VST3D<0b0101, {0,0,0,?}, "8">;
1034 def VST3q16 : VST3D<0b0101, {0,1,0,?}, "16">;
1035 def VST3q32 : VST3D<0b0101, {1,0,0,?}, "32">;
1036 def VST3q8_UPD : VST3DWB<0b0101, {0,0,0,?}, "8">;
1037 def VST3q16_UPD : VST3DWB<0b0101, {0,1,0,?}, "16">;
1038 def VST3q32_UPD : VST3DWB<0b0101, {1,0,0,?}, "32">;
1040 def VST3q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1041 def VST3q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1042 def VST3q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1044 // ...alternate versions to be allocated odd register numbers:
1045 def VST3q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1046 def VST3q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1047 def VST3q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1049 // VST4 : Vector Store (multiple 4-element structures)
1050 class VST4D<bits<4> op11_8, bits<4> op7_4, string Dt>
1051 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1052 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
1053 IIC_VST4, "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn",
1056 let Inst{5-4} = Rn{5-4};
1059 def VST4d8 : VST4D<0b0000, {0,0,?,?}, "8">;
1060 def VST4d16 : VST4D<0b0000, {0,1,?,?}, "16">;
1061 def VST4d32 : VST4D<0b0000, {1,0,?,?}, "32">;
1063 def VST4d8Pseudo : VSTQQPseudo<IIC_VST4>;
1064 def VST4d16Pseudo : VSTQQPseudo<IIC_VST4>;
1065 def VST4d32Pseudo : VSTQQPseudo<IIC_VST4>;
1067 // ...with address register writeback:
1068 class VST4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1069 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1070 (ins addrmode6:$Rn, am6offset:$Rm,
1071 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST4u,
1072 "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
1073 "$Rn.addr = $wb", []> {
1074 let Inst{5-4} = Rn{5-4};
1077 def VST4d8_UPD : VST4DWB<0b0000, {0,0,?,?}, "8">;
1078 def VST4d16_UPD : VST4DWB<0b0000, {0,1,?,?}, "16">;
1079 def VST4d32_UPD : VST4DWB<0b0000, {1,0,?,?}, "32">;
1081 def VST4d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1082 def VST4d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1083 def VST4d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1085 // ...with double-spaced registers (non-updating versions for disassembly only):
1086 def VST4q8 : VST4D<0b0001, {0,0,?,?}, "8">;
1087 def VST4q16 : VST4D<0b0001, {0,1,?,?}, "16">;
1088 def VST4q32 : VST4D<0b0001, {1,0,?,?}, "32">;
1089 def VST4q8_UPD : VST4DWB<0b0001, {0,0,?,?}, "8">;
1090 def VST4q16_UPD : VST4DWB<0b0001, {0,1,?,?}, "16">;
1091 def VST4q32_UPD : VST4DWB<0b0001, {1,0,?,?}, "32">;
1093 def VST4q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1094 def VST4q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1095 def VST4q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1097 // ...alternate versions to be allocated odd register numbers:
1098 def VST4q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1099 def VST4q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1100 def VST4q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1102 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
1104 // Classes for VST*LN pseudo-instructions with multi-register operands.
1105 // These are expanded to real instructions after register allocation.
1106 class VSTQLNPseudo<InstrItinClass itin>
1107 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
1109 class VSTQLNWBPseudo<InstrItinClass itin>
1110 : PseudoNLdSt<(outs GPR:$wb),
1111 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
1112 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1113 class VSTQQLNPseudo<InstrItinClass itin>
1114 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
1116 class VSTQQLNWBPseudo<InstrItinClass itin>
1117 : PseudoNLdSt<(outs GPR:$wb),
1118 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
1119 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1120 class VSTQQQQLNPseudo<InstrItinClass itin>
1121 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
1123 class VSTQQQQLNWBPseudo<InstrItinClass itin>
1124 : PseudoNLdSt<(outs GPR:$wb),
1125 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
1126 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1128 // VST1LN : Vector Store (single element from one lane)
1129 class VST1LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1130 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1131 (ins addrmode6:$Rn, DPR:$Vd, nohash_imm:$lane),
1132 IIC_VST1ln, "vst1", Dt, "\\{$Vd[$lane]\\}, $Rn", "", []> {
1136 def VST1LNd8 : VST1LN<0b0000, {?,?,?,0}, "8"> {
1137 let Inst{7-5} = lane{2-0};
1139 def VST1LNd16 : VST1LN<0b0100, {?,?,0,?}, "16"> {
1140 let Inst{7-6} = lane{1-0};
1141 let Inst{4} = Rn{5};
1143 def VST1LNd32 : VST1LN<0b1000, {?,0,?,?}, "32"> {
1144 let Inst{7} = lane{0};
1145 let Inst{5-4} = Rn{5-4};
1148 def VST1LNq8Pseudo : VSTQLNPseudo<IIC_VST1ln>;
1149 def VST1LNq16Pseudo : VSTQLNPseudo<IIC_VST1ln>;
1150 def VST1LNq32Pseudo : VSTQLNPseudo<IIC_VST1ln>;
1152 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
1154 // ...with address register writeback:
1155 class VST1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1156 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1157 (ins addrmode6:$Rn, am6offset:$Rm,
1158 DPR:$Vd, nohash_imm:$lane), IIC_VST1lnu, "vst1", Dt,
1159 "\\{$Vd[$lane]\\}, $Rn$Rm",
1160 "$Rn.addr = $wb", []>;
1162 def VST1LNd8_UPD : VST1LNWB<0b0000, {?,?,?,0}, "8"> {
1163 let Inst{7-5} = lane{2-0};
1165 def VST1LNd16_UPD : VST1LNWB<0b0100, {?,?,0,?}, "16"> {
1166 let Inst{7-6} = lane{1-0};
1167 let Inst{4} = Rn{5};
1169 def VST1LNd32_UPD : VST1LNWB<0b1000, {?,0,?,?}, "32"> {
1170 let Inst{7} = lane{0};
1171 let Inst{5-4} = Rn{5-4};
1174 def VST1LNq8Pseudo_UPD : VSTQLNWBPseudo<IIC_VST1lnu>;
1175 def VST1LNq16Pseudo_UPD : VSTQLNWBPseudo<IIC_VST1lnu>;
1176 def VST1LNq32Pseudo_UPD : VSTQLNWBPseudo<IIC_VST1lnu>;
1178 // VST2LN : Vector Store (single 2-element structure from one lane)
1179 class VST2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1180 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1181 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, nohash_imm:$lane),
1182 IIC_VST2ln, "vst2", Dt, "\\{$Vd[$lane], $src2[$lane]\\}, $Rn",
1185 let Inst{4} = Rn{4};
1188 def VST2LNd8 : VST2LN<0b0001, {?,?,?,?}, "8"> {
1189 let Inst{7-5} = lane{2-0};
1191 def VST2LNd16 : VST2LN<0b0101, {?,?,0,?}, "16"> {
1192 let Inst{7-6} = lane{1-0};
1194 def VST2LNd32 : VST2LN<0b1001, {?,0,0,?}, "32"> {
1195 let Inst{7} = lane{0};
1198 def VST2LNd8Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1199 def VST2LNd16Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1200 def VST2LNd32Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1202 // ...with double-spaced registers:
1203 def VST2LNq16 : VST2LN<0b0101, {?,?,1,?}, "16"> {
1204 let Inst{7-6} = lane{1-0};
1205 let Inst{4} = Rn{4};
1207 def VST2LNq32 : VST2LN<0b1001, {?,1,0,?}, "32"> {
1208 let Inst{7} = lane{0};
1209 let Inst{4} = Rn{4};
1212 def VST2LNq16Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
1213 def VST2LNq32Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
1215 // ...with address register writeback:
1216 class VST2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1217 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1218 (ins addrmode6:$addr, am6offset:$offset,
1219 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VST2lnu, "vst2", Dt,
1220 "\\{$src1[$lane], $src2[$lane]\\}, $addr$offset",
1221 "$addr.addr = $wb", []> {
1222 let Inst{4} = Rn{4};
1225 def VST2LNd8_UPD : VST2LNWB<0b0001, {?,?,?,?}, "8"> {
1226 let Inst{7-5} = lane{2-0};
1228 def VST2LNd16_UPD : VST2LNWB<0b0101, {?,?,0,?}, "16"> {
1229 let Inst{7-6} = lane{1-0};
1231 def VST2LNd32_UPD : VST2LNWB<0b1001, {?,0,0,?}, "32"> {
1232 let Inst{7} = lane{0};
1235 def VST2LNd8Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1236 def VST2LNd16Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1237 def VST2LNd32Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1239 def VST2LNq16_UPD : VST2LNWB<0b0101, {?,?,1,?}, "16"> {
1240 let Inst{7-6} = lane{1-0};
1242 def VST2LNq32_UPD : VST2LNWB<0b1001, {?,1,0,?}, "32"> {
1243 let Inst{7} = lane{0};
1246 def VST2LNq16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
1247 def VST2LNq32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
1249 // VST3LN : Vector Store (single 3-element structure from one lane)
1250 class VST3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1251 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1252 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3,
1253 nohash_imm:$lane), IIC_VST3ln, "vst3", Dt,
1254 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn", "", []> {
1258 def VST3LNd8 : VST3LN<0b0010, {?,?,?,0}, "8"> {
1259 let Inst{7-5} = lane{2-0};
1261 def VST3LNd16 : VST3LN<0b0110, {?,?,0,0}, "16"> {
1262 let Inst{7-6} = lane{1-0};
1264 def VST3LNd32 : VST3LN<0b1010, {?,0,0,0}, "32"> {
1265 let Inst{7} = lane{0};
1268 def VST3LNd8Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1269 def VST3LNd16Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1270 def VST3LNd32Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1272 // ...with double-spaced registers:
1273 def VST3LNq16 : VST3LN<0b0110, {?,?,1,0}, "16"> {
1274 let Inst{7-6} = lane{1-0};
1276 def VST3LNq32 : VST3LN<0b1010, {?,1,0,0}, "32"> {
1277 let Inst{7} = lane{0};
1280 def VST3LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
1281 def VST3LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
1283 // ...with address register writeback:
1284 class VST3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1285 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1286 (ins addrmode6:$Rn, am6offset:$Rm,
1287 DPR:$Vd, DPR:$src2, DPR:$src3, nohash_imm:$lane),
1288 IIC_VST3lnu, "vst3", Dt,
1289 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn$Rm",
1290 "$Rn.addr = $wb", []>;
1292 def VST3LNd8_UPD : VST3LNWB<0b0010, {?,?,?,0}, "8"> {
1293 let Inst{7-5} = lane{2-0};
1295 def VST3LNd16_UPD : VST3LNWB<0b0110, {?,?,0,0}, "16"> {
1296 let Inst{7-6} = lane{1-0};
1298 def VST3LNd32_UPD : VST3LNWB<0b1010, {?,0,0,0}, "32"> {
1299 let Inst{7} = lane{0};
1302 def VST3LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1303 def VST3LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1304 def VST3LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1306 def VST3LNq16_UPD : VST3LNWB<0b0110, {?,?,1,0}, "16"> {
1307 let Inst{7-6} = lane{1-0};
1309 def VST3LNq32_UPD : VST3LNWB<0b1010, {?,1,0,0}, "32"> {
1310 let Inst{7} = lane{0};
1313 def VST3LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
1314 def VST3LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
1316 // VST4LN : Vector Store (single 4-element structure from one lane)
1317 class VST4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1318 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1319 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4,
1320 nohash_imm:$lane), IIC_VST4ln, "vst4", Dt,
1321 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn",
1324 let Inst{4} = Rn{4};
1327 def VST4LNd8 : VST4LN<0b0011, {?,?,?,?}, "8"> {
1328 let Inst{7-5} = lane{2-0};
1330 def VST4LNd16 : VST4LN<0b0111, {?,?,0,?}, "16"> {
1331 let Inst{7-6} = lane{1-0};
1333 def VST4LNd32 : VST4LN<0b1011, {?,0,?,?}, "32"> {
1334 let Inst{7} = lane{0};
1335 let Inst{5} = Rn{5};
1338 def VST4LNd8Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1339 def VST4LNd16Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1340 def VST4LNd32Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1342 // ...with double-spaced registers:
1343 def VST4LNq16 : VST4LN<0b0111, {?,?,1,?}, "16"> {
1344 let Inst{7-6} = lane{1-0};
1346 def VST4LNq32 : VST4LN<0b1011, {?,1,?,?}, "32"> {
1347 let Inst{7} = lane{0};
1348 let Inst{5} = Rn{5};
1351 def VST4LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
1352 def VST4LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
1354 // ...with address register writeback:
1355 class VST4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1356 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1357 (ins addrmode6:$Rn, am6offset:$Rm,
1358 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
1359 IIC_VST4lnu, "vst4", Dt,
1360 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn$Rm",
1361 "$Rn.addr = $wb", []> {
1362 let Inst{4} = Rn{4};
1365 def VST4LNd8_UPD : VST4LNWB<0b0011, {?,?,?,?}, "8"> {
1366 let Inst{7-5} = lane{2-0};
1368 def VST4LNd16_UPD : VST4LNWB<0b0111, {?,?,0,?}, "16"> {
1369 let Inst{7-6} = lane{1-0};
1371 def VST4LNd32_UPD : VST4LNWB<0b1011, {?,0,?,?}, "32"> {
1372 let Inst{7} = lane{0};
1373 let Inst{5} = Rn{5};
1376 def VST4LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
1377 def VST4LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
1378 def VST4LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
1380 def VST4LNq16_UPD : VST4LNWB<0b0111, {?,?,1,?}, "16"> {
1381 let Inst{7-6} = lane{1-0};
1383 def VST4LNq32_UPD : VST4LNWB<0b1011, {?,1,?,?}, "32"> {
1384 let Inst{7} = lane{0};
1385 let Inst{5} = Rn{5};
1388 def VST4LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
1389 def VST4LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
1391 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
1394 //===----------------------------------------------------------------------===//
1395 // NEON pattern fragments
1396 //===----------------------------------------------------------------------===//
1398 // Extract D sub-registers of Q registers.
1399 def DSubReg_i8_reg : SDNodeXForm<imm, [{
1400 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1401 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/8, MVT::i32);
1403 def DSubReg_i16_reg : SDNodeXForm<imm, [{
1404 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1405 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/4, MVT::i32);
1407 def DSubReg_i32_reg : SDNodeXForm<imm, [{
1408 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1409 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/2, MVT::i32);
1411 def DSubReg_f64_reg : SDNodeXForm<imm, [{
1412 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1413 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue(), MVT::i32);
1416 // Extract S sub-registers of Q/D registers.
1417 def SSubReg_f32_reg : SDNodeXForm<imm, [{
1418 assert(ARM::ssub_3 == ARM::ssub_0+3 && "Unexpected subreg numbering");
1419 return CurDAG->getTargetConstant(ARM::ssub_0 + N->getZExtValue(), MVT::i32);
1422 // Translate lane numbers from Q registers to D subregs.
1423 def SubReg_i8_lane : SDNodeXForm<imm, [{
1424 return CurDAG->getTargetConstant(N->getZExtValue() & 7, MVT::i32);
1426 def SubReg_i16_lane : SDNodeXForm<imm, [{
1427 return CurDAG->getTargetConstant(N->getZExtValue() & 3, MVT::i32);
1429 def SubReg_i32_lane : SDNodeXForm<imm, [{
1430 return CurDAG->getTargetConstant(N->getZExtValue() & 1, MVT::i32);
1433 //===----------------------------------------------------------------------===//
1434 // Instruction Classes
1435 //===----------------------------------------------------------------------===//
1437 // Basic 2-register operations: single-, double- and quad-register.
1438 class N2VS<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1439 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
1440 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
1441 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
1442 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src),
1443 IIC_VUNAD, OpcodeStr, Dt, "$dst, $src", "", []>;
1444 class N2VD<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1445 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
1446 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
1447 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
1448 (ins DPR:$src), IIC_VUNAD, OpcodeStr, Dt,"$dst, $src", "",
1449 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src))))]>;
1450 class N2VQ<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1451 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
1452 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
1453 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
1454 (ins QPR:$src), IIC_VUNAQ, OpcodeStr, Dt,"$dst, $src", "",
1455 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src))))]>;
1457 // Basic 2-register intrinsics, both double- and quad-register.
1458 class N2VDInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1459 bits<2> op17_16, bits<5> op11_7, bit op4,
1460 InstrItinClass itin, string OpcodeStr, string Dt,
1461 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1462 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
1463 (ins DPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
1464 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
1465 class N2VQInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1466 bits<2> op17_16, bits<5> op11_7, bit op4,
1467 InstrItinClass itin, string OpcodeStr, string Dt,
1468 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1469 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
1470 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
1471 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
1473 // Narrow 2-register operations.
1474 class N2VN<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1475 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1476 InstrItinClass itin, string OpcodeStr, string Dt,
1477 ValueType TyD, ValueType TyQ, SDNode OpNode>
1478 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$dst),
1479 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
1480 [(set DPR:$dst, (TyD (OpNode (TyQ QPR:$src))))]>;
1482 // Narrow 2-register intrinsics.
1483 class N2VNInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1484 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1485 InstrItinClass itin, string OpcodeStr, string Dt,
1486 ValueType TyD, ValueType TyQ, Intrinsic IntOp>
1487 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$dst),
1488 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
1489 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src))))]>;
1491 // Long 2-register operations (currently only used for VMOVL).
1492 class N2VL<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1493 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1494 InstrItinClass itin, string OpcodeStr, string Dt,
1495 ValueType TyQ, ValueType TyD, SDNode OpNode>
1496 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$dst),
1497 (ins DPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
1498 [(set QPR:$dst, (TyQ (OpNode (TyD DPR:$src))))]>;
1500 // 2-register shuffles (VTRN/VZIP/VUZP), both double- and quad-register.
1501 class N2VDShuffle<bits<2> op19_18, bits<5> op11_7, string OpcodeStr, string Dt>
1502 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 0, 0, (outs DPR:$dst1, DPR:$dst2),
1503 (ins DPR:$src1, DPR:$src2), IIC_VPERMD,
1504 OpcodeStr, Dt, "$dst1, $dst2",
1505 "$src1 = $dst1, $src2 = $dst2", []>;
1506 class N2VQShuffle<bits<2> op19_18, bits<5> op11_7,
1507 InstrItinClass itin, string OpcodeStr, string Dt>
1508 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 1, 0, (outs QPR:$dst1, QPR:$dst2),
1509 (ins QPR:$src1, QPR:$src2), itin, OpcodeStr, Dt, "$dst1, $dst2",
1510 "$src1 = $dst1, $src2 = $dst2", []>;
1512 // Basic 3-register operations: single-, double- and quad-register.
1513 class N3VS<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1514 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1515 SDNode OpNode, bit Commutable>
1516 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1517 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src1, DPR_VFP2:$src2), N3RegFrm,
1518 IIC_VBIND, OpcodeStr, Dt, "$dst, $src1, $src2", "", []> {
1519 let isCommutable = Commutable;
1522 class N3VD<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1523 InstrItinClass itin, string OpcodeStr, string Dt,
1524 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
1525 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1526 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1527 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
1528 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
1529 let isCommutable = Commutable;
1531 // Same as N3VD but no data type.
1532 class N3VDX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1533 InstrItinClass itin, string OpcodeStr,
1534 ValueType ResTy, ValueType OpTy,
1535 SDNode OpNode, bit Commutable>
1536 : N3VX<op24, op23, op21_20, op11_8, 0, op4,
1537 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
1538 OpcodeStr, "$dst, $src1, $src2", "",
1539 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src1), (OpTy DPR:$src2))))]>{
1540 let isCommutable = Commutable;
1543 class N3VDSL<bits<2> op21_20, bits<4> op11_8,
1544 InstrItinClass itin, string OpcodeStr, string Dt,
1545 ValueType Ty, SDNode ShOp>
1546 : N3V<0, 1, op21_20, op11_8, 1, 0,
1547 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1548 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1549 [(set (Ty DPR:$dst),
1550 (Ty (ShOp (Ty DPR:$src1),
1551 (Ty (NEONvduplane (Ty DPR_VFP2:$src2),imm:$lane)))))]> {
1552 let isCommutable = 0;
1554 class N3VDSL16<bits<2> op21_20, bits<4> op11_8,
1555 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1556 : N3V<0, 1, op21_20, op11_8, 1, 0,
1557 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1558 NVMulSLFrm, IIC_VMULi16D, OpcodeStr, Dt,"$dst, $src1, $src2[$lane]","",
1559 [(set (Ty DPR:$dst),
1560 (Ty (ShOp (Ty DPR:$src1),
1561 (Ty (NEONvduplane (Ty DPR_8:$src2), imm:$lane)))))]> {
1562 let isCommutable = 0;
1565 class N3VQ<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1566 InstrItinClass itin, string OpcodeStr, string Dt,
1567 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
1568 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1569 (outs QPR:$Qd), (ins QPR:$Qn, QPR:$Qm), N3RegFrm, itin,
1570 OpcodeStr, Dt, "$Qd, $Qn, $Qm", "",
1571 [(set QPR:$Qd, (ResTy (OpNode (OpTy QPR:$Qn), (OpTy QPR:$Qm))))]> {
1572 let isCommutable = Commutable;
1574 class N3VQX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1575 InstrItinClass itin, string OpcodeStr,
1576 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
1577 : N3VX<op24, op23, op21_20, op11_8, 1, op4,
1578 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), N3RegFrm, itin,
1579 OpcodeStr, "$dst, $src1, $src2", "",
1580 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src1), (OpTy QPR:$src2))))]>{
1581 let isCommutable = Commutable;
1583 class N3VQSL<bits<2> op21_20, bits<4> op11_8,
1584 InstrItinClass itin, string OpcodeStr, string Dt,
1585 ValueType ResTy, ValueType OpTy, SDNode ShOp>
1586 : N3V<1, 1, op21_20, op11_8, 1, 0,
1587 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1588 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1589 [(set (ResTy QPR:$dst),
1590 (ResTy (ShOp (ResTy QPR:$src1),
1591 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1593 let isCommutable = 0;
1595 class N3VQSL16<bits<2> op21_20, bits<4> op11_8, string OpcodeStr, string Dt,
1596 ValueType ResTy, ValueType OpTy, SDNode ShOp>
1597 : N3V<1, 1, op21_20, op11_8, 1, 0,
1598 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1599 NVMulSLFrm, IIC_VMULi16Q, OpcodeStr, Dt,"$dst, $src1, $src2[$lane]","",
1600 [(set (ResTy QPR:$dst),
1601 (ResTy (ShOp (ResTy QPR:$src1),
1602 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
1604 let isCommutable = 0;
1607 // Basic 3-register intrinsics, both double- and quad-register.
1608 class N3VDInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1609 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1610 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
1611 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1612 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), f, itin,
1613 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
1614 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
1615 let isCommutable = Commutable;
1617 class N3VDIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1618 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
1619 : N3V<0, 1, op21_20, op11_8, 1, 0,
1620 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1621 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1622 [(set (Ty DPR:$dst),
1623 (Ty (IntOp (Ty DPR:$src1),
1624 (Ty (NEONvduplane (Ty DPR_VFP2:$src2),
1626 let isCommutable = 0;
1628 class N3VDIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1629 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
1630 : N3V<0, 1, op21_20, op11_8, 1, 0,
1631 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1632 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1633 [(set (Ty DPR:$dst),
1634 (Ty (IntOp (Ty DPR:$src1),
1635 (Ty (NEONvduplane (Ty DPR_8:$src2), imm:$lane)))))]> {
1636 let isCommutable = 0;
1638 class N3VDIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1639 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1640 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1641 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1642 (outs DPR:$Vd), (ins DPR:$Vm, DPR:$Vn), f, itin,
1643 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
1644 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (OpTy DPR:$Vn))))]> {
1645 let isCommutable = 0;
1648 class N3VQInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1649 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1650 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
1651 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1652 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), f, itin,
1653 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
1654 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
1655 let isCommutable = Commutable;
1657 class N3VQIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1658 string OpcodeStr, string Dt,
1659 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1660 : N3V<1, 1, op21_20, op11_8, 1, 0,
1661 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1662 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1663 [(set (ResTy QPR:$dst),
1664 (ResTy (IntOp (ResTy QPR:$src1),
1665 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1667 let isCommutable = 0;
1669 class N3VQIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1670 string OpcodeStr, string Dt,
1671 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1672 : N3V<1, 1, op21_20, op11_8, 1, 0,
1673 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1674 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1675 [(set (ResTy QPR:$dst),
1676 (ResTy (IntOp (ResTy QPR:$src1),
1677 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
1679 let isCommutable = 0;
1681 class N3VQIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1682 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1683 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1684 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1685 (outs QPR:$Vd), (ins QPR:$Vm, QPR:$Vn), f, itin,
1686 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
1687 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (OpTy QPR:$Vn))))]> {
1688 let isCommutable = 0;
1691 // Multiply-Add/Sub operations: single-, double- and quad-register.
1692 class N3VSMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1693 InstrItinClass itin, string OpcodeStr, string Dt,
1694 ValueType Ty, SDNode MulOp, SDNode OpNode>
1695 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1696 (outs DPR_VFP2:$dst),
1697 (ins DPR_VFP2:$src1, DPR_VFP2:$src2, DPR_VFP2:$src3), N3RegFrm, itin,
1698 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst", []>;
1700 class N3VDMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1701 InstrItinClass itin, string OpcodeStr, string Dt,
1702 ValueType Ty, SDNode MulOp, SDNode OpNode>
1703 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1704 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1705 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1706 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
1707 (Ty (MulOp DPR:$Vn, DPR:$Vm)))))]>;
1709 class N3VDMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1710 string OpcodeStr, string Dt,
1711 ValueType Ty, SDNode MulOp, SDNode ShOp>
1712 : N3V<0, 1, op21_20, op11_8, 1, 0,
1714 (ins DPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1716 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1717 [(set (Ty DPR:$dst),
1718 (Ty (ShOp (Ty DPR:$src1),
1719 (Ty (MulOp DPR:$src2,
1720 (Ty (NEONvduplane (Ty DPR_VFP2:$src3),
1722 class N3VDMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1723 string OpcodeStr, string Dt,
1724 ValueType Ty, SDNode MulOp, SDNode ShOp>
1725 : N3V<0, 1, op21_20, op11_8, 1, 0,
1727 (ins DPR:$src1, DPR:$Vn, DPR_8:$Vm, nohash_imm:$lane),
1729 OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "$src1 = $Vd",
1731 (Ty (ShOp (Ty DPR:$src1),
1733 (Ty (NEONvduplane (Ty DPR_8:$Vm),
1736 class N3VQMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1737 InstrItinClass itin, string OpcodeStr, string Dt, ValueType Ty,
1738 SDNode MulOp, SDNode OpNode>
1739 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1740 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
1741 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1742 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
1743 (Ty (MulOp QPR:$Vn, QPR:$Vm)))))]>;
1744 class N3VQMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1745 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1746 SDNode MulOp, SDNode ShOp>
1747 : N3V<1, 1, op21_20, op11_8, 1, 0,
1749 (ins QPR:$src1, QPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1751 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1752 [(set (ResTy QPR:$dst),
1753 (ResTy (ShOp (ResTy QPR:$src1),
1754 (ResTy (MulOp QPR:$src2,
1755 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src3),
1757 class N3VQMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1758 string OpcodeStr, string Dt,
1759 ValueType ResTy, ValueType OpTy,
1760 SDNode MulOp, SDNode ShOp>
1761 : N3V<1, 1, op21_20, op11_8, 1, 0,
1763 (ins QPR:$src1, QPR:$src2, DPR_8:$src3, nohash_imm:$lane),
1765 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1766 [(set (ResTy QPR:$dst),
1767 (ResTy (ShOp (ResTy QPR:$src1),
1768 (ResTy (MulOp QPR:$src2,
1769 (ResTy (NEONvduplane (OpTy DPR_8:$src3),
1772 // Neon Intrinsic-Op instructions (VABA): double- and quad-register.
1773 class N3VDIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1774 InstrItinClass itin, string OpcodeStr, string Dt,
1775 ValueType Ty, Intrinsic IntOp, SDNode OpNode>
1776 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1777 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1778 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1779 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
1780 (Ty (IntOp (Ty DPR:$Vn), (Ty DPR:$Vm))))))]>;
1781 class N3VQIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1782 InstrItinClass itin, string OpcodeStr, string Dt,
1783 ValueType Ty, Intrinsic IntOp, SDNode OpNode>
1784 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1785 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
1786 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1787 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
1788 (Ty (IntOp (Ty QPR:$Vn), (Ty QPR:$Vm))))))]>;
1790 // Neon 3-argument intrinsics, both double- and quad-register.
1791 // The destination register is also used as the first source operand register.
1792 class N3VDInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1793 InstrItinClass itin, string OpcodeStr, string Dt,
1794 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1795 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1796 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3), N3RegFrm, itin,
1797 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1798 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src1),
1799 (OpTy DPR:$src2), (OpTy DPR:$src3))))]>;
1800 class N3VQInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1801 InstrItinClass itin, string OpcodeStr, string Dt,
1802 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1803 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1804 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3), N3RegFrm, itin,
1805 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1806 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src1),
1807 (OpTy QPR:$src2), (OpTy QPR:$src3))))]>;
1809 // Long Multiply-Add/Sub operations.
1810 class N3VLMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1811 InstrItinClass itin, string OpcodeStr, string Dt,
1812 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
1813 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1814 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1815 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1816 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
1817 (TyQ (MulOp (TyD DPR:$Vn),
1818 (TyD DPR:$Vm)))))]>;
1819 class N3VLMulOpSL<bit op24, bits<2> op21_20, bits<4> op11_8,
1820 InstrItinClass itin, string OpcodeStr, string Dt,
1821 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
1822 : N3V<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$dst),
1823 (ins QPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1825 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1827 (OpNode (TyQ QPR:$src1),
1828 (TyQ (MulOp (TyD DPR:$src2),
1829 (TyD (NEONvduplane (TyD DPR_VFP2:$src3),
1831 class N3VLMulOpSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1832 InstrItinClass itin, string OpcodeStr, string Dt,
1833 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
1834 : N3V<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$dst),
1835 (ins QPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane),
1837 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1839 (OpNode (TyQ QPR:$src1),
1840 (TyQ (MulOp (TyD DPR:$src2),
1841 (TyD (NEONvduplane (TyD DPR_8:$src3),
1844 // Long Intrinsic-Op vector operations with explicit extend (VABAL).
1845 class N3VLIntExtOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1846 InstrItinClass itin, string OpcodeStr, string Dt,
1847 ValueType TyQ, ValueType TyD, Intrinsic IntOp, SDNode ExtOp,
1849 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1850 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1851 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1852 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
1853 (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
1854 (TyD DPR:$Vm)))))))]>;
1856 // Neon Long 3-argument intrinsic. The destination register is
1857 // a quad-register and is also used as the first source operand register.
1858 class N3VLInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1859 InstrItinClass itin, string OpcodeStr, string Dt,
1860 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
1861 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1862 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1863 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1865 (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$Vn), (TyD DPR:$Vm))))]>;
1866 class N3VLInt3SL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1867 string OpcodeStr, string Dt,
1868 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1869 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1871 (ins QPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1873 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1874 [(set (ResTy QPR:$dst),
1875 (ResTy (IntOp (ResTy QPR:$src1),
1877 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src3),
1879 class N3VLInt3SL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1880 InstrItinClass itin, 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_8:$src3, nohash_imm:$lane),
1886 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1887 [(set (ResTy QPR:$dst),
1888 (ResTy (IntOp (ResTy QPR:$src1),
1890 (OpTy (NEONvduplane (OpTy DPR_8:$src3),
1893 // Narrowing 3-register intrinsics.
1894 class N3VNInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1895 string OpcodeStr, string Dt, ValueType TyD, ValueType TyQ,
1896 Intrinsic IntOp, bit Commutable>
1897 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1898 (outs DPR:$dst), (ins QPR:$src1, QPR:$src2), N3RegFrm, IIC_VBINi4D,
1899 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1900 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src1), (TyQ QPR:$src2))))]> {
1901 let isCommutable = Commutable;
1904 // Long 3-register operations.
1905 class N3VL<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1906 InstrItinClass itin, string OpcodeStr, string Dt,
1907 ValueType TyQ, ValueType TyD, SDNode OpNode, bit Commutable>
1908 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1909 (outs QPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
1910 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1911 [(set QPR:$dst, (TyQ (OpNode (TyD DPR:$src1), (TyD DPR:$src2))))]> {
1912 let isCommutable = Commutable;
1914 class N3VLSL<bit op24, bits<2> op21_20, bits<4> op11_8,
1915 InstrItinClass itin, string OpcodeStr, string Dt,
1916 ValueType TyQ, ValueType TyD, SDNode OpNode>
1917 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1918 (outs QPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1919 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1921 (TyQ (OpNode (TyD DPR:$src1),
1922 (TyD (NEONvduplane (TyD DPR_VFP2:$src2),imm:$lane)))))]>;
1923 class N3VLSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1924 InstrItinClass itin, string OpcodeStr, string Dt,
1925 ValueType TyQ, ValueType TyD, SDNode OpNode>
1926 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1927 (outs QPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1928 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1930 (TyQ (OpNode (TyD DPR:$src1),
1931 (TyD (NEONvduplane (TyD DPR_8:$src2), imm:$lane)))))]>;
1933 // Long 3-register operations with explicitly extended operands.
1934 class N3VLExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1935 InstrItinClass itin, string OpcodeStr, string Dt,
1936 ValueType TyQ, ValueType TyD, SDNode OpNode, SDNode ExtOp,
1938 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1939 (outs QPR:$Qd), (ins DPR:$Dn, DPR:$Dm), N3RegFrm, itin,
1940 OpcodeStr, Dt, "$Qd, $Dn, $Dm", "",
1941 [(set QPR:$Qd, (OpNode (TyQ (ExtOp (TyD DPR:$Dn))),
1942 (TyQ (ExtOp (TyD DPR:$Dm)))))]> {
1943 let isCommutable = Commutable;
1946 // Long 3-register intrinsics with explicit extend (VABDL).
1947 class N3VLIntExt<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, Intrinsic IntOp, SDNode ExtOp,
1951 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1952 (outs QPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
1953 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1954 [(set QPR:$dst, (TyQ (ExtOp (TyD (IntOp (TyD DPR:$src1),
1955 (TyD DPR:$src2))))))]> {
1956 let isCommutable = Commutable;
1959 // Long 3-register intrinsics.
1960 class N3VLInt<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, bit Commutable>
1963 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1964 (outs QPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
1965 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1966 [(set QPR:$dst, (TyQ (IntOp (TyD DPR:$src1), (TyD DPR:$src2))))]> {
1967 let isCommutable = Commutable;
1969 class N3VLIntSL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1970 string OpcodeStr, string Dt,
1971 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1972 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1973 (outs QPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1974 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1975 [(set (ResTy QPR:$dst),
1976 (ResTy (IntOp (OpTy DPR:$src1),
1977 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1979 class N3VLIntSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1980 InstrItinClass itin, string OpcodeStr, string Dt,
1981 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1982 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1983 (outs QPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1984 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1985 [(set (ResTy QPR:$dst),
1986 (ResTy (IntOp (OpTy DPR:$src1),
1987 (OpTy (NEONvduplane (OpTy DPR_8:$src2),
1990 // Wide 3-register operations.
1991 class N3VW<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1992 string OpcodeStr, string Dt, ValueType TyQ, ValueType TyD,
1993 SDNode OpNode, SDNode ExtOp, bit Commutable>
1994 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1995 (outs QPR:$Qd), (ins QPR:$Qn, DPR:$Dm), N3RegFrm, IIC_VSUBiD,
1996 OpcodeStr, Dt, "$Qd, $Qn, $Dm", "",
1997 [(set QPR:$Qd, (OpNode (TyQ QPR:$Qn),
1998 (TyQ (ExtOp (TyD DPR:$Dm)))))]> {
1999 let isCommutable = Commutable;
2002 // Pairwise long 2-register intrinsics, both double- and quad-register.
2003 class N2VDPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2004 bits<2> op17_16, bits<5> op11_7, bit op4,
2005 string OpcodeStr, string Dt,
2006 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2007 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
2008 (ins DPR:$src), IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
2009 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
2010 class N2VQPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2011 bits<2> op17_16, bits<5> op11_7, bit op4,
2012 string OpcodeStr, string Dt,
2013 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2014 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
2015 (ins QPR:$src), IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
2016 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
2018 // Pairwise long 2-register accumulate intrinsics,
2019 // both double- and quad-register.
2020 // The destination register is also used as the first source operand register.
2021 class N2VDPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2022 bits<2> op17_16, bits<5> op11_7, bit op4,
2023 string OpcodeStr, string Dt,
2024 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2025 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
2026 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vm), IIC_VPALiD,
2027 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
2028 [(set DPR:$Vd, (ResTy (IntOp (ResTy DPR:$src1), (OpTy DPR:$Vm))))]>;
2029 class N2VQPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2030 bits<2> op17_16, bits<5> op11_7, bit op4,
2031 string OpcodeStr, string Dt,
2032 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2033 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4,
2034 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vm), IIC_VPALiQ,
2035 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
2036 [(set QPR:$Vd, (ResTy (IntOp (ResTy QPR:$src1), (OpTy QPR:$Vm))))]>;
2038 // Shift by immediate,
2039 // both double- and quad-register.
2040 class N2VDSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2041 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2042 ValueType Ty, SDNode OpNode>
2043 : N2VImm<op24, op23, op11_8, op7, 0, op4,
2044 (outs DPR:$dst), (ins DPR:$src, i32imm:$SIMM), f, itin,
2045 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
2046 [(set DPR:$dst, (Ty (OpNode (Ty DPR:$src), (i32 imm:$SIMM))))]>;
2047 class N2VQSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2048 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2049 ValueType Ty, SDNode OpNode>
2050 : N2VImm<op24, op23, op11_8, op7, 1, op4,
2051 (outs QPR:$dst), (ins QPR:$src, i32imm:$SIMM), f, itin,
2052 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
2053 [(set QPR:$dst, (Ty (OpNode (Ty QPR:$src), (i32 imm:$SIMM))))]>;
2055 // Long shift by immediate.
2056 class N2VLSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
2057 string OpcodeStr, string Dt,
2058 ValueType ResTy, ValueType OpTy, SDNode OpNode>
2059 : N2VImm<op24, op23, op11_8, op7, op6, op4,
2060 (outs QPR:$dst), (ins DPR:$src, i32imm:$SIMM), N2RegVShLFrm,
2061 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src, $SIMM", "",
2062 [(set QPR:$dst, (ResTy (OpNode (OpTy DPR:$src),
2063 (i32 imm:$SIMM))))]>;
2065 // Narrow shift by immediate.
2066 class N2VNSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
2067 InstrItinClass itin, string OpcodeStr, string Dt,
2068 ValueType ResTy, ValueType OpTy, SDNode OpNode>
2069 : N2VImm<op24, op23, op11_8, op7, op6, op4,
2070 (outs DPR:$dst), (ins QPR:$src, i32imm:$SIMM), N2RegVShRFrm, itin,
2071 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
2072 [(set DPR:$dst, (ResTy (OpNode (OpTy QPR:$src),
2073 (i32 imm:$SIMM))))]>;
2075 // Shift right by immediate and accumulate,
2076 // both double- and quad-register.
2077 class N2VDShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2078 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
2079 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
2080 (ins DPR:$src1, DPR:$Vm, i32imm:$SIMM), N2RegVShRFrm, IIC_VPALiD,
2081 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2082 [(set DPR:$Vd, (Ty (add DPR:$src1,
2083 (Ty (ShOp DPR:$Vm, (i32 imm:$SIMM))))))]>;
2084 class N2VQShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2085 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
2086 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
2087 (ins QPR:$src1, QPR:$Vm, i32imm:$SIMM), N2RegVShRFrm, IIC_VPALiD,
2088 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2089 [(set QPR:$Vd, (Ty (add QPR:$src1,
2090 (Ty (ShOp QPR:$Vm, (i32 imm:$SIMM))))))]>;
2092 // Shift by immediate and insert,
2093 // both double- and quad-register.
2094 class N2VDShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2095 Format f, string OpcodeStr, string Dt, ValueType Ty,SDNode ShOp>
2096 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
2097 (ins DPR:$src1, DPR:$Vm, i32imm:$SIMM), f, IIC_VSHLiD,
2098 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2099 [(set DPR:$Vd, (Ty (ShOp DPR:$src1, DPR:$Vm, (i32 imm:$SIMM))))]>;
2100 class N2VQShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2101 Format f, string OpcodeStr, string Dt, ValueType Ty,SDNode ShOp>
2102 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
2103 (ins QPR:$src1, QPR:$Vm, i32imm:$SIMM), f, IIC_VSHLiQ,
2104 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2105 [(set QPR:$Vd, (Ty (ShOp QPR:$src1, QPR:$Vm, (i32 imm:$SIMM))))]>;
2107 // Convert, with fractional bits immediate,
2108 // both double- and quad-register.
2109 class N2VCvtD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2110 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2112 : N2VImm<op24, op23, op11_8, op7, 0, op4,
2113 (outs DPR:$Vd), (ins DPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
2114 IIC_VUNAD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2115 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (i32 imm:$SIMM))))]>;
2116 class N2VCvtQ<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2117 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2119 : N2VImm<op24, op23, op11_8, op7, 1, op4,
2120 (outs QPR:$Vd), (ins QPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
2121 IIC_VUNAQ, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2122 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (i32 imm:$SIMM))))]>;
2124 //===----------------------------------------------------------------------===//
2126 //===----------------------------------------------------------------------===//
2128 // Abbreviations used in multiclass suffixes:
2129 // Q = quarter int (8 bit) elements
2130 // H = half int (16 bit) elements
2131 // S = single int (32 bit) elements
2132 // D = double int (64 bit) elements
2134 // Neon 2-register vector operations -- for disassembly only.
2136 // First with only element sizes of 8, 16 and 32 bits:
2137 multiclass N2V_QHS_cmp<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2138 bits<5> op11_7, bit op4, string opc, string Dt,
2140 // 64-bit vector types.
2141 def v8i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 0, op4,
2142 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
2143 opc, !strconcat(Dt, "8"), asm, "", []>;
2144 def v4i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 0, op4,
2145 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
2146 opc, !strconcat(Dt, "16"), asm, "", []>;
2147 def v2i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
2148 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
2149 opc, !strconcat(Dt, "32"), asm, "", []>;
2150 def v2f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
2151 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
2152 opc, "f32", asm, "", []> {
2153 let Inst{10} = 1; // overwrite F = 1
2156 // 128-bit vector types.
2157 def v16i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 1, op4,
2158 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
2159 opc, !strconcat(Dt, "8"), asm, "", []>;
2160 def v8i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 1, op4,
2161 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
2162 opc, !strconcat(Dt, "16"), asm, "", []>;
2163 def v4i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
2164 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
2165 opc, !strconcat(Dt, "32"), asm, "", []>;
2166 def v4f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
2167 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
2168 opc, "f32", asm, "", []> {
2169 let Inst{10} = 1; // overwrite F = 1
2173 // Neon 3-register vector operations.
2175 // First with only element sizes of 8, 16 and 32 bits:
2176 multiclass N3V_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2177 InstrItinClass itinD16, InstrItinClass itinD32,
2178 InstrItinClass itinQ16, InstrItinClass itinQ32,
2179 string OpcodeStr, string Dt,
2180 SDNode OpNode, bit Commutable = 0> {
2181 // 64-bit vector types.
2182 def v8i8 : N3VD<op24, op23, 0b00, op11_8, op4, itinD16,
2183 OpcodeStr, !strconcat(Dt, "8"),
2184 v8i8, v8i8, OpNode, Commutable>;
2185 def v4i16 : N3VD<op24, op23, 0b01, op11_8, op4, itinD16,
2186 OpcodeStr, !strconcat(Dt, "16"),
2187 v4i16, v4i16, OpNode, Commutable>;
2188 def v2i32 : N3VD<op24, op23, 0b10, op11_8, op4, itinD32,
2189 OpcodeStr, !strconcat(Dt, "32"),
2190 v2i32, v2i32, OpNode, Commutable>;
2192 // 128-bit vector types.
2193 def v16i8 : N3VQ<op24, op23, 0b00, op11_8, op4, itinQ16,
2194 OpcodeStr, !strconcat(Dt, "8"),
2195 v16i8, v16i8, OpNode, Commutable>;
2196 def v8i16 : N3VQ<op24, op23, 0b01, op11_8, op4, itinQ16,
2197 OpcodeStr, !strconcat(Dt, "16"),
2198 v8i16, v8i16, OpNode, Commutable>;
2199 def v4i32 : N3VQ<op24, op23, 0b10, op11_8, op4, itinQ32,
2200 OpcodeStr, !strconcat(Dt, "32"),
2201 v4i32, v4i32, OpNode, Commutable>;
2204 multiclass N3VSL_HS<bits<4> op11_8, string OpcodeStr, string Dt, SDNode ShOp> {
2205 def v4i16 : N3VDSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
2207 def v2i32 : N3VDSL<0b10, op11_8, IIC_VMULi32D, OpcodeStr, !strconcat(Dt,"32"),
2209 def v8i16 : N3VQSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
2210 v8i16, v4i16, ShOp>;
2211 def v4i32 : N3VQSL<0b10, op11_8, IIC_VMULi32Q, OpcodeStr, !strconcat(Dt,"32"),
2212 v4i32, v2i32, ShOp>;
2215 // ....then also with element size 64 bits:
2216 multiclass N3V_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2217 InstrItinClass itinD, InstrItinClass itinQ,
2218 string OpcodeStr, string Dt,
2219 SDNode OpNode, bit Commutable = 0>
2220 : N3V_QHS<op24, op23, op11_8, op4, itinD, itinD, itinQ, itinQ,
2221 OpcodeStr, Dt, OpNode, Commutable> {
2222 def v1i64 : N3VD<op24, op23, 0b11, op11_8, op4, itinD,
2223 OpcodeStr, !strconcat(Dt, "64"),
2224 v1i64, v1i64, OpNode, Commutable>;
2225 def v2i64 : N3VQ<op24, op23, 0b11, op11_8, op4, itinQ,
2226 OpcodeStr, !strconcat(Dt, "64"),
2227 v2i64, v2i64, OpNode, Commutable>;
2231 // Neon Narrowing 2-register vector operations,
2232 // source operand element sizes of 16, 32 and 64 bits:
2233 multiclass N2VN_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2234 bits<5> op11_7, bit op6, bit op4,
2235 InstrItinClass itin, string OpcodeStr, string Dt,
2237 def v8i8 : N2VN<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
2238 itin, OpcodeStr, !strconcat(Dt, "16"),
2239 v8i8, v8i16, OpNode>;
2240 def v4i16 : N2VN<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
2241 itin, OpcodeStr, !strconcat(Dt, "32"),
2242 v4i16, v4i32, OpNode>;
2243 def v2i32 : N2VN<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
2244 itin, OpcodeStr, !strconcat(Dt, "64"),
2245 v2i32, v2i64, OpNode>;
2248 // Neon Narrowing 2-register vector intrinsics,
2249 // source operand element sizes of 16, 32 and 64 bits:
2250 multiclass N2VNInt_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2251 bits<5> op11_7, bit op6, bit op4,
2252 InstrItinClass itin, string OpcodeStr, string Dt,
2254 def v8i8 : N2VNInt<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
2255 itin, OpcodeStr, !strconcat(Dt, "16"),
2256 v8i8, v8i16, IntOp>;
2257 def v4i16 : N2VNInt<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
2258 itin, OpcodeStr, !strconcat(Dt, "32"),
2259 v4i16, v4i32, IntOp>;
2260 def v2i32 : N2VNInt<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
2261 itin, OpcodeStr, !strconcat(Dt, "64"),
2262 v2i32, v2i64, IntOp>;
2266 // Neon Lengthening 2-register vector intrinsic (currently specific to VMOVL).
2267 // source operand element sizes of 16, 32 and 64 bits:
2268 multiclass N2VL_QHS<bits<2> op24_23, bits<5> op11_7, bit op6, bit op4,
2269 string OpcodeStr, string Dt, SDNode OpNode> {
2270 def v8i16 : N2VL<op24_23, 0b00, 0b10, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2271 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode>;
2272 def v4i32 : N2VL<op24_23, 0b01, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2273 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
2274 def v2i64 : N2VL<op24_23, 0b10, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2275 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
2279 // Neon 3-register vector intrinsics.
2281 // First with only element sizes of 16 and 32 bits:
2282 multiclass N3VInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2283 InstrItinClass itinD16, InstrItinClass itinD32,
2284 InstrItinClass itinQ16, InstrItinClass itinQ32,
2285 string OpcodeStr, string Dt,
2286 Intrinsic IntOp, bit Commutable = 0> {
2287 // 64-bit vector types.
2288 def v4i16 : N3VDInt<op24, op23, 0b01, op11_8, op4, f, itinD16,
2289 OpcodeStr, !strconcat(Dt, "16"),
2290 v4i16, v4i16, IntOp, Commutable>;
2291 def v2i32 : N3VDInt<op24, op23, 0b10, op11_8, op4, f, itinD32,
2292 OpcodeStr, !strconcat(Dt, "32"),
2293 v2i32, v2i32, IntOp, Commutable>;
2295 // 128-bit vector types.
2296 def v8i16 : N3VQInt<op24, op23, 0b01, op11_8, op4, f, itinQ16,
2297 OpcodeStr, !strconcat(Dt, "16"),
2298 v8i16, v8i16, IntOp, Commutable>;
2299 def v4i32 : N3VQInt<op24, op23, 0b10, op11_8, op4, f, itinQ32,
2300 OpcodeStr, !strconcat(Dt, "32"),
2301 v4i32, v4i32, IntOp, Commutable>;
2303 multiclass N3VInt_HSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2304 InstrItinClass itinD16, InstrItinClass itinD32,
2305 InstrItinClass itinQ16, InstrItinClass itinQ32,
2306 string OpcodeStr, string Dt,
2308 // 64-bit vector types.
2309 def v4i16 : N3VDIntSh<op24, op23, 0b01, op11_8, op4, f, itinD16,
2310 OpcodeStr, !strconcat(Dt, "16"),
2311 v4i16, v4i16, IntOp>;
2312 def v2i32 : N3VDIntSh<op24, op23, 0b10, op11_8, op4, f, itinD32,
2313 OpcodeStr, !strconcat(Dt, "32"),
2314 v2i32, v2i32, IntOp>;
2316 // 128-bit vector types.
2317 def v8i16 : N3VQIntSh<op24, op23, 0b01, op11_8, op4, f, itinQ16,
2318 OpcodeStr, !strconcat(Dt, "16"),
2319 v8i16, v8i16, IntOp>;
2320 def v4i32 : N3VQIntSh<op24, op23, 0b10, op11_8, op4, f, itinQ32,
2321 OpcodeStr, !strconcat(Dt, "32"),
2322 v4i32, v4i32, IntOp>;
2325 multiclass N3VIntSL_HS<bits<4> op11_8,
2326 InstrItinClass itinD16, InstrItinClass itinD32,
2327 InstrItinClass itinQ16, InstrItinClass itinQ32,
2328 string OpcodeStr, string Dt, Intrinsic IntOp> {
2329 def v4i16 : N3VDIntSL16<0b01, op11_8, itinD16,
2330 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp>;
2331 def v2i32 : N3VDIntSL<0b10, op11_8, itinD32,
2332 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp>;
2333 def v8i16 : N3VQIntSL16<0b01, op11_8, itinQ16,
2334 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16, IntOp>;
2335 def v4i32 : N3VQIntSL<0b10, op11_8, itinQ32,
2336 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32, IntOp>;
2339 // ....then also with element size of 8 bits:
2340 multiclass N3VInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2341 InstrItinClass itinD16, InstrItinClass itinD32,
2342 InstrItinClass itinQ16, InstrItinClass itinQ32,
2343 string OpcodeStr, string Dt,
2344 Intrinsic IntOp, bit Commutable = 0>
2345 : N3VInt_HS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2346 OpcodeStr, Dt, IntOp, Commutable> {
2347 def v8i8 : N3VDInt<op24, op23, 0b00, op11_8, op4, f, itinD16,
2348 OpcodeStr, !strconcat(Dt, "8"),
2349 v8i8, v8i8, IntOp, Commutable>;
2350 def v16i8 : N3VQInt<op24, op23, 0b00, op11_8, op4, f, itinQ16,
2351 OpcodeStr, !strconcat(Dt, "8"),
2352 v16i8, v16i8, IntOp, Commutable>;
2354 multiclass N3VInt_QHSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2355 InstrItinClass itinD16, InstrItinClass itinD32,
2356 InstrItinClass itinQ16, InstrItinClass itinQ32,
2357 string OpcodeStr, string Dt,
2359 : N3VInt_HSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2360 OpcodeStr, Dt, IntOp> {
2361 def v8i8 : N3VDIntSh<op24, op23, 0b00, op11_8, op4, f, itinD16,
2362 OpcodeStr, !strconcat(Dt, "8"),
2364 def v16i8 : N3VQIntSh<op24, op23, 0b00, op11_8, op4, f, itinQ16,
2365 OpcodeStr, !strconcat(Dt, "8"),
2366 v16i8, v16i8, IntOp>;
2370 // ....then also with element size of 64 bits:
2371 multiclass N3VInt_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2372 InstrItinClass itinD16, InstrItinClass itinD32,
2373 InstrItinClass itinQ16, InstrItinClass itinQ32,
2374 string OpcodeStr, string Dt,
2375 Intrinsic IntOp, bit Commutable = 0>
2376 : N3VInt_QHS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2377 OpcodeStr, Dt, IntOp, Commutable> {
2378 def v1i64 : N3VDInt<op24, op23, 0b11, op11_8, op4, f, itinD32,
2379 OpcodeStr, !strconcat(Dt, "64"),
2380 v1i64, v1i64, IntOp, Commutable>;
2381 def v2i64 : N3VQInt<op24, op23, 0b11, op11_8, op4, f, itinQ32,
2382 OpcodeStr, !strconcat(Dt, "64"),
2383 v2i64, v2i64, IntOp, Commutable>;
2385 multiclass N3VInt_QHSDSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2386 InstrItinClass itinD16, InstrItinClass itinD32,
2387 InstrItinClass itinQ16, InstrItinClass itinQ32,
2388 string OpcodeStr, string Dt,
2390 : N3VInt_QHSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2391 OpcodeStr, Dt, IntOp> {
2392 def v1i64 : N3VDIntSh<op24, op23, 0b11, op11_8, op4, f, itinD32,
2393 OpcodeStr, !strconcat(Dt, "64"),
2394 v1i64, v1i64, IntOp>;
2395 def v2i64 : N3VQIntSh<op24, op23, 0b11, op11_8, op4, f, itinQ32,
2396 OpcodeStr, !strconcat(Dt, "64"),
2397 v2i64, v2i64, IntOp>;
2400 // Neon Narrowing 3-register vector intrinsics,
2401 // source operand element sizes of 16, 32 and 64 bits:
2402 multiclass N3VNInt_HSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2403 string OpcodeStr, string Dt,
2404 Intrinsic IntOp, bit Commutable = 0> {
2405 def v8i8 : N3VNInt<op24, op23, 0b00, op11_8, op4,
2406 OpcodeStr, !strconcat(Dt, "16"),
2407 v8i8, v8i16, IntOp, Commutable>;
2408 def v4i16 : N3VNInt<op24, op23, 0b01, op11_8, op4,
2409 OpcodeStr, !strconcat(Dt, "32"),
2410 v4i16, v4i32, IntOp, Commutable>;
2411 def v2i32 : N3VNInt<op24, op23, 0b10, op11_8, op4,
2412 OpcodeStr, !strconcat(Dt, "64"),
2413 v2i32, v2i64, IntOp, Commutable>;
2417 // Neon Long 3-register vector operations.
2419 multiclass N3VL_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2420 InstrItinClass itin16, InstrItinClass itin32,
2421 string OpcodeStr, string Dt,
2422 SDNode OpNode, bit Commutable = 0> {
2423 def v8i16 : N3VL<op24, op23, 0b00, op11_8, op4, itin16,
2424 OpcodeStr, !strconcat(Dt, "8"),
2425 v8i16, v8i8, OpNode, Commutable>;
2426 def v4i32 : N3VL<op24, op23, 0b01, op11_8, op4, itin16,
2427 OpcodeStr, !strconcat(Dt, "16"),
2428 v4i32, v4i16, OpNode, Commutable>;
2429 def v2i64 : N3VL<op24, op23, 0b10, op11_8, op4, itin32,
2430 OpcodeStr, !strconcat(Dt, "32"),
2431 v2i64, v2i32, OpNode, Commutable>;
2434 multiclass N3VLSL_HS<bit op24, bits<4> op11_8,
2435 InstrItinClass itin, string OpcodeStr, string Dt,
2437 def v4i16 : N3VLSL16<op24, 0b01, op11_8, itin, OpcodeStr,
2438 !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
2439 def v2i32 : N3VLSL<op24, 0b10, op11_8, itin, OpcodeStr,
2440 !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
2443 multiclass N3VLExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2444 InstrItinClass itin16, InstrItinClass itin32,
2445 string OpcodeStr, string Dt,
2446 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
2447 def v8i16 : N3VLExt<op24, op23, 0b00, op11_8, op4, itin16,
2448 OpcodeStr, !strconcat(Dt, "8"),
2449 v8i16, v8i8, OpNode, ExtOp, Commutable>;
2450 def v4i32 : N3VLExt<op24, op23, 0b01, op11_8, op4, itin16,
2451 OpcodeStr, !strconcat(Dt, "16"),
2452 v4i32, v4i16, OpNode, ExtOp, Commutable>;
2453 def v2i64 : N3VLExt<op24, op23, 0b10, op11_8, op4, itin32,
2454 OpcodeStr, !strconcat(Dt, "32"),
2455 v2i64, v2i32, OpNode, ExtOp, Commutable>;
2458 // Neon Long 3-register vector intrinsics.
2460 // First with only element sizes of 16 and 32 bits:
2461 multiclass N3VLInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
2462 InstrItinClass itin16, InstrItinClass itin32,
2463 string OpcodeStr, string Dt,
2464 Intrinsic IntOp, bit Commutable = 0> {
2465 def v4i32 : N3VLInt<op24, op23, 0b01, op11_8, op4, itin16,
2466 OpcodeStr, !strconcat(Dt, "16"),
2467 v4i32, v4i16, IntOp, Commutable>;
2468 def v2i64 : N3VLInt<op24, op23, 0b10, op11_8, op4, itin32,
2469 OpcodeStr, !strconcat(Dt, "32"),
2470 v2i64, v2i32, IntOp, Commutable>;
2473 multiclass N3VLIntSL_HS<bit op24, bits<4> op11_8,
2474 InstrItinClass itin, string OpcodeStr, string Dt,
2476 def v4i16 : N3VLIntSL16<op24, 0b01, op11_8, itin,
2477 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
2478 def v2i32 : N3VLIntSL<op24, 0b10, op11_8, itin,
2479 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
2482 // ....then also with element size of 8 bits:
2483 multiclass N3VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2484 InstrItinClass itin16, InstrItinClass itin32,
2485 string OpcodeStr, string Dt,
2486 Intrinsic IntOp, bit Commutable = 0>
2487 : N3VLInt_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt,
2488 IntOp, Commutable> {
2489 def v8i16 : N3VLInt<op24, op23, 0b00, op11_8, op4, itin16,
2490 OpcodeStr, !strconcat(Dt, "8"),
2491 v8i16, v8i8, IntOp, Commutable>;
2494 // ....with explicit extend (VABDL).
2495 multiclass N3VLIntExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2496 InstrItinClass itin, string OpcodeStr, string Dt,
2497 Intrinsic IntOp, SDNode ExtOp, bit Commutable = 0> {
2498 def v8i16 : N3VLIntExt<op24, op23, 0b00, op11_8, op4, itin,
2499 OpcodeStr, !strconcat(Dt, "8"),
2500 v8i16, v8i8, IntOp, ExtOp, Commutable>;
2501 def v4i32 : N3VLIntExt<op24, op23, 0b01, op11_8, op4, itin,
2502 OpcodeStr, !strconcat(Dt, "16"),
2503 v4i32, v4i16, IntOp, ExtOp, Commutable>;
2504 def v2i64 : N3VLIntExt<op24, op23, 0b10, op11_8, op4, itin,
2505 OpcodeStr, !strconcat(Dt, "32"),
2506 v2i64, v2i32, IntOp, ExtOp, Commutable>;
2510 // Neon Wide 3-register vector intrinsics,
2511 // source operand element sizes of 8, 16 and 32 bits:
2512 multiclass N3VW_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2513 string OpcodeStr, string Dt,
2514 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
2515 def v8i16 : N3VW<op24, op23, 0b00, op11_8, op4,
2516 OpcodeStr, !strconcat(Dt, "8"),
2517 v8i16, v8i8, OpNode, ExtOp, Commutable>;
2518 def v4i32 : N3VW<op24, op23, 0b01, op11_8, op4,
2519 OpcodeStr, !strconcat(Dt, "16"),
2520 v4i32, v4i16, OpNode, ExtOp, Commutable>;
2521 def v2i64 : N3VW<op24, op23, 0b10, op11_8, op4,
2522 OpcodeStr, !strconcat(Dt, "32"),
2523 v2i64, v2i32, OpNode, ExtOp, Commutable>;
2527 // Neon Multiply-Op vector operations,
2528 // element sizes of 8, 16 and 32 bits:
2529 multiclass N3VMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2530 InstrItinClass itinD16, InstrItinClass itinD32,
2531 InstrItinClass itinQ16, InstrItinClass itinQ32,
2532 string OpcodeStr, string Dt, SDNode OpNode> {
2533 // 64-bit vector types.
2534 def v8i8 : N3VDMulOp<op24, op23, 0b00, op11_8, op4, itinD16,
2535 OpcodeStr, !strconcat(Dt, "8"), v8i8, mul, OpNode>;
2536 def v4i16 : N3VDMulOp<op24, op23, 0b01, op11_8, op4, itinD16,
2537 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, OpNode>;
2538 def v2i32 : N3VDMulOp<op24, op23, 0b10, op11_8, op4, itinD32,
2539 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, OpNode>;
2541 // 128-bit vector types.
2542 def v16i8 : N3VQMulOp<op24, op23, 0b00, op11_8, op4, itinQ16,
2543 OpcodeStr, !strconcat(Dt, "8"), v16i8, mul, OpNode>;
2544 def v8i16 : N3VQMulOp<op24, op23, 0b01, op11_8, op4, itinQ16,
2545 OpcodeStr, !strconcat(Dt, "16"), v8i16, mul, OpNode>;
2546 def v4i32 : N3VQMulOp<op24, op23, 0b10, op11_8, op4, itinQ32,
2547 OpcodeStr, !strconcat(Dt, "32"), v4i32, mul, OpNode>;
2550 multiclass N3VMulOpSL_HS<bits<4> op11_8,
2551 InstrItinClass itinD16, InstrItinClass itinD32,
2552 InstrItinClass itinQ16, InstrItinClass itinQ32,
2553 string OpcodeStr, string Dt, SDNode ShOp> {
2554 def v4i16 : N3VDMulOpSL16<0b01, op11_8, itinD16,
2555 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, ShOp>;
2556 def v2i32 : N3VDMulOpSL<0b10, op11_8, itinD32,
2557 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, ShOp>;
2558 def v8i16 : N3VQMulOpSL16<0b01, op11_8, itinQ16,
2559 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16,
2561 def v4i32 : N3VQMulOpSL<0b10, op11_8, itinQ32,
2562 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32,
2566 // Neon Intrinsic-Op vector operations,
2567 // element sizes of 8, 16 and 32 bits:
2568 multiclass N3VIntOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2569 InstrItinClass itinD, InstrItinClass itinQ,
2570 string OpcodeStr, string Dt, Intrinsic IntOp,
2572 // 64-bit vector types.
2573 def v8i8 : N3VDIntOp<op24, op23, 0b00, op11_8, op4, itinD,
2574 OpcodeStr, !strconcat(Dt, "8"), v8i8, IntOp, OpNode>;
2575 def v4i16 : N3VDIntOp<op24, op23, 0b01, op11_8, op4, itinD,
2576 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp, OpNode>;
2577 def v2i32 : N3VDIntOp<op24, op23, 0b10, op11_8, op4, itinD,
2578 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp, OpNode>;
2580 // 128-bit vector types.
2581 def v16i8 : N3VQIntOp<op24, op23, 0b00, op11_8, op4, itinQ,
2582 OpcodeStr, !strconcat(Dt, "8"), v16i8, IntOp, OpNode>;
2583 def v8i16 : N3VQIntOp<op24, op23, 0b01, op11_8, op4, itinQ,
2584 OpcodeStr, !strconcat(Dt, "16"), v8i16, IntOp, OpNode>;
2585 def v4i32 : N3VQIntOp<op24, op23, 0b10, op11_8, op4, itinQ,
2586 OpcodeStr, !strconcat(Dt, "32"), v4i32, IntOp, OpNode>;
2589 // Neon 3-argument intrinsics,
2590 // element sizes of 8, 16 and 32 bits:
2591 multiclass N3VInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2592 InstrItinClass itinD, InstrItinClass itinQ,
2593 string OpcodeStr, string Dt, Intrinsic IntOp> {
2594 // 64-bit vector types.
2595 def v8i8 : N3VDInt3<op24, op23, 0b00, op11_8, op4, itinD,
2596 OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
2597 def v4i16 : N3VDInt3<op24, op23, 0b01, op11_8, op4, itinD,
2598 OpcodeStr, !strconcat(Dt, "16"), v4i16, v4i16, IntOp>;
2599 def v2i32 : N3VDInt3<op24, op23, 0b10, op11_8, op4, itinD,
2600 OpcodeStr, !strconcat(Dt, "32"), v2i32, v2i32, IntOp>;
2602 // 128-bit vector types.
2603 def v16i8 : N3VQInt3<op24, op23, 0b00, op11_8, op4, itinQ,
2604 OpcodeStr, !strconcat(Dt, "8"), v16i8, v16i8, IntOp>;
2605 def v8i16 : N3VQInt3<op24, op23, 0b01, op11_8, op4, itinQ,
2606 OpcodeStr, !strconcat(Dt, "16"), v8i16, v8i16, IntOp>;
2607 def v4i32 : N3VQInt3<op24, op23, 0b10, op11_8, op4, itinQ,
2608 OpcodeStr, !strconcat(Dt, "32"), v4i32, v4i32, IntOp>;
2612 // Neon Long Multiply-Op vector operations,
2613 // element sizes of 8, 16 and 32 bits:
2614 multiclass N3VLMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2615 InstrItinClass itin16, InstrItinClass itin32,
2616 string OpcodeStr, string Dt, SDNode MulOp,
2618 def v8i16 : N3VLMulOp<op24, op23, 0b00, op11_8, op4, itin16, OpcodeStr,
2619 !strconcat(Dt, "8"), v8i16, v8i8, MulOp, OpNode>;
2620 def v4i32 : N3VLMulOp<op24, op23, 0b01, op11_8, op4, itin16, OpcodeStr,
2621 !strconcat(Dt, "16"), v4i32, v4i16, MulOp, OpNode>;
2622 def v2i64 : N3VLMulOp<op24, op23, 0b10, op11_8, op4, itin32, OpcodeStr,
2623 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
2626 multiclass N3VLMulOpSL_HS<bit op24, bits<4> op11_8, string OpcodeStr,
2627 string Dt, SDNode MulOp, SDNode OpNode> {
2628 def v4i16 : N3VLMulOpSL16<op24, 0b01, op11_8, IIC_VMACi16D, OpcodeStr,
2629 !strconcat(Dt,"16"), v4i32, v4i16, MulOp, OpNode>;
2630 def v2i32 : N3VLMulOpSL<op24, 0b10, op11_8, IIC_VMACi32D, OpcodeStr,
2631 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
2635 // Neon Long 3-argument intrinsics.
2637 // First with only element sizes of 16 and 32 bits:
2638 multiclass N3VLInt3_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
2639 InstrItinClass itin16, InstrItinClass itin32,
2640 string OpcodeStr, string Dt, Intrinsic IntOp> {
2641 def v4i32 : N3VLInt3<op24, op23, 0b01, op11_8, op4, itin16,
2642 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
2643 def v2i64 : N3VLInt3<op24, op23, 0b10, op11_8, op4, itin32,
2644 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
2647 multiclass N3VLInt3SL_HS<bit op24, bits<4> op11_8,
2648 string OpcodeStr, string Dt, Intrinsic IntOp> {
2649 def v4i16 : N3VLInt3SL16<op24, 0b01, op11_8, IIC_VMACi16D,
2650 OpcodeStr, !strconcat(Dt,"16"), v4i32, v4i16, IntOp>;
2651 def v2i32 : N3VLInt3SL<op24, 0b10, op11_8, IIC_VMACi32D,
2652 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
2655 // ....then also with element size of 8 bits:
2656 multiclass N3VLInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2657 InstrItinClass itin16, InstrItinClass itin32,
2658 string OpcodeStr, string Dt, Intrinsic IntOp>
2659 : N3VLInt3_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt, IntOp> {
2660 def v8i16 : N3VLInt3<op24, op23, 0b00, op11_8, op4, itin16,
2661 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
2664 // ....with explicit extend (VABAL).
2665 multiclass N3VLIntExtOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2666 InstrItinClass itin, string OpcodeStr, string Dt,
2667 Intrinsic IntOp, SDNode ExtOp, SDNode OpNode> {
2668 def v8i16 : N3VLIntExtOp<op24, op23, 0b00, op11_8, op4, itin,
2669 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8,
2670 IntOp, ExtOp, OpNode>;
2671 def v4i32 : N3VLIntExtOp<op24, op23, 0b01, op11_8, op4, itin,
2672 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16,
2673 IntOp, ExtOp, OpNode>;
2674 def v2i64 : N3VLIntExtOp<op24, op23, 0b10, op11_8, op4, itin,
2675 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32,
2676 IntOp, ExtOp, OpNode>;
2680 // Neon 2-register vector intrinsics,
2681 // element sizes of 8, 16 and 32 bits:
2682 multiclass N2VInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2683 bits<5> op11_7, bit op4,
2684 InstrItinClass itinD, InstrItinClass itinQ,
2685 string OpcodeStr, string Dt, Intrinsic IntOp> {
2686 // 64-bit vector types.
2687 def v8i8 : N2VDInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2688 itinD, OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
2689 def v4i16 : N2VDInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2690 itinD, OpcodeStr, !strconcat(Dt, "16"),v4i16,v4i16,IntOp>;
2691 def v2i32 : N2VDInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2692 itinD, OpcodeStr, !strconcat(Dt, "32"),v2i32,v2i32,IntOp>;
2694 // 128-bit vector types.
2695 def v16i8 : N2VQInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2696 itinQ, OpcodeStr, !strconcat(Dt, "8"), v16i8,v16i8,IntOp>;
2697 def v8i16 : N2VQInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2698 itinQ, OpcodeStr, !strconcat(Dt, "16"),v8i16,v8i16,IntOp>;
2699 def v4i32 : N2VQInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2700 itinQ, OpcodeStr, !strconcat(Dt, "32"),v4i32,v4i32,IntOp>;
2704 // Neon Pairwise long 2-register intrinsics,
2705 // element sizes of 8, 16 and 32 bits:
2706 multiclass N2VPLInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2707 bits<5> op11_7, bit op4,
2708 string OpcodeStr, string Dt, Intrinsic IntOp> {
2709 // 64-bit vector types.
2710 def v8i8 : N2VDPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2711 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
2712 def v4i16 : N2VDPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2713 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
2714 def v2i32 : N2VDPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2715 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
2717 // 128-bit vector types.
2718 def v16i8 : N2VQPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2719 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
2720 def v8i16 : N2VQPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2721 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
2722 def v4i32 : N2VQPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2723 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
2727 // Neon Pairwise long 2-register accumulate intrinsics,
2728 // element sizes of 8, 16 and 32 bits:
2729 multiclass N2VPLInt2_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2730 bits<5> op11_7, bit op4,
2731 string OpcodeStr, string Dt, Intrinsic IntOp> {
2732 // 64-bit vector types.
2733 def v8i8 : N2VDPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2734 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
2735 def v4i16 : N2VDPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2736 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
2737 def v2i32 : N2VDPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2738 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
2740 // 128-bit vector types.
2741 def v16i8 : N2VQPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2742 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
2743 def v8i16 : N2VQPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2744 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
2745 def v4i32 : N2VQPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2746 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
2750 // Neon 2-register vector shift by immediate,
2751 // with f of either N2RegVShLFrm or N2RegVShRFrm
2752 // element sizes of 8, 16, 32 and 64 bits:
2753 multiclass N2VSh_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2754 InstrItinClass itin, string OpcodeStr, string Dt,
2755 SDNode OpNode, Format f> {
2756 // 64-bit vector types.
2757 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
2758 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
2759 let Inst{21-19} = 0b001; // imm6 = 001xxx
2761 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
2762 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
2763 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2765 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
2766 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
2767 let Inst{21} = 0b1; // imm6 = 1xxxxx
2769 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, f, itin,
2770 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
2773 // 128-bit vector types.
2774 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
2775 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
2776 let Inst{21-19} = 0b001; // imm6 = 001xxx
2778 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
2779 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
2780 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2782 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
2783 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
2784 let Inst{21} = 0b1; // imm6 = 1xxxxx
2786 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, f, itin,
2787 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
2791 // Neon Shift-Accumulate vector operations,
2792 // element sizes of 8, 16, 32 and 64 bits:
2793 multiclass N2VShAdd_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2794 string OpcodeStr, string Dt, SDNode ShOp> {
2795 // 64-bit vector types.
2796 def v8i8 : N2VDShAdd<op24, op23, op11_8, 0, op4,
2797 OpcodeStr, !strconcat(Dt, "8"), v8i8, ShOp> {
2798 let Inst{21-19} = 0b001; // imm6 = 001xxx
2800 def v4i16 : N2VDShAdd<op24, op23, op11_8, 0, op4,
2801 OpcodeStr, !strconcat(Dt, "16"), v4i16, ShOp> {
2802 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2804 def v2i32 : N2VDShAdd<op24, op23, op11_8, 0, op4,
2805 OpcodeStr, !strconcat(Dt, "32"), v2i32, ShOp> {
2806 let Inst{21} = 0b1; // imm6 = 1xxxxx
2808 def v1i64 : N2VDShAdd<op24, op23, op11_8, 1, op4,
2809 OpcodeStr, !strconcat(Dt, "64"), v1i64, ShOp>;
2812 // 128-bit vector types.
2813 def v16i8 : N2VQShAdd<op24, op23, op11_8, 0, op4,
2814 OpcodeStr, !strconcat(Dt, "8"), v16i8, ShOp> {
2815 let Inst{21-19} = 0b001; // imm6 = 001xxx
2817 def v8i16 : N2VQShAdd<op24, op23, op11_8, 0, op4,
2818 OpcodeStr, !strconcat(Dt, "16"), v8i16, ShOp> {
2819 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2821 def v4i32 : N2VQShAdd<op24, op23, op11_8, 0, op4,
2822 OpcodeStr, !strconcat(Dt, "32"), v4i32, ShOp> {
2823 let Inst{21} = 0b1; // imm6 = 1xxxxx
2825 def v2i64 : N2VQShAdd<op24, op23, op11_8, 1, op4,
2826 OpcodeStr, !strconcat(Dt, "64"), v2i64, ShOp>;
2831 // Neon Shift-Insert vector operations,
2832 // with f of either N2RegVShLFrm or N2RegVShRFrm
2833 // element sizes of 8, 16, 32 and 64 bits:
2834 multiclass N2VShIns_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2835 string OpcodeStr, SDNode ShOp,
2837 // 64-bit vector types.
2838 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4,
2839 f, OpcodeStr, "8", v8i8, ShOp> {
2840 let Inst{21-19} = 0b001; // imm6 = 001xxx
2842 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4,
2843 f, OpcodeStr, "16", v4i16, ShOp> {
2844 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2846 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4,
2847 f, OpcodeStr, "32", v2i32, ShOp> {
2848 let Inst{21} = 0b1; // imm6 = 1xxxxx
2850 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4,
2851 f, OpcodeStr, "64", v1i64, ShOp>;
2854 // 128-bit vector types.
2855 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4,
2856 f, OpcodeStr, "8", v16i8, ShOp> {
2857 let Inst{21-19} = 0b001; // imm6 = 001xxx
2859 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4,
2860 f, OpcodeStr, "16", v8i16, ShOp> {
2861 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2863 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4,
2864 f, OpcodeStr, "32", v4i32, ShOp> {
2865 let Inst{21} = 0b1; // imm6 = 1xxxxx
2867 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4,
2868 f, OpcodeStr, "64", v2i64, ShOp>;
2872 // Neon Shift Long operations,
2873 // element sizes of 8, 16, 32 bits:
2874 multiclass N2VLSh_QHS<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
2875 bit op4, string OpcodeStr, string Dt, SDNode OpNode> {
2876 def v8i16 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
2877 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode> {
2878 let Inst{21-19} = 0b001; // imm6 = 001xxx
2880 def v4i32 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
2881 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode> {
2882 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2884 def v2i64 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
2885 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode> {
2886 let Inst{21} = 0b1; // imm6 = 1xxxxx
2890 // Neon Shift Narrow operations,
2891 // element sizes of 16, 32, 64 bits:
2892 multiclass N2VNSh_HSD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
2893 bit op4, InstrItinClass itin, string OpcodeStr, string Dt,
2895 def v8i8 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
2896 OpcodeStr, !strconcat(Dt, "16"), v8i8, v8i16, OpNode> {
2897 let Inst{21-19} = 0b001; // imm6 = 001xxx
2899 def v4i16 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
2900 OpcodeStr, !strconcat(Dt, "32"), v4i16, v4i32, OpNode> {
2901 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2903 def v2i32 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
2904 OpcodeStr, !strconcat(Dt, "64"), v2i32, v2i64, OpNode> {
2905 let Inst{21} = 0b1; // imm6 = 1xxxxx
2909 //===----------------------------------------------------------------------===//
2910 // Instruction Definitions.
2911 //===----------------------------------------------------------------------===//
2913 // Vector Add Operations.
2915 // VADD : Vector Add (integer and floating-point)
2916 defm VADD : N3V_QHSD<0, 0, 0b1000, 0, IIC_VBINiD, IIC_VBINiQ, "vadd", "i",
2918 def VADDfd : N3VD<0, 0, 0b00, 0b1101, 0, IIC_VBIND, "vadd", "f32",
2919 v2f32, v2f32, fadd, 1>;
2920 def VADDfq : N3VQ<0, 0, 0b00, 0b1101, 0, IIC_VBINQ, "vadd", "f32",
2921 v4f32, v4f32, fadd, 1>;
2922 // VADDL : Vector Add Long (Q = D + D)
2923 defm VADDLs : N3VLExt_QHS<0,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
2924 "vaddl", "s", add, sext, 1>;
2925 defm VADDLu : N3VLExt_QHS<1,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
2926 "vaddl", "u", add, zext, 1>;
2927 // VADDW : Vector Add Wide (Q = Q + D)
2928 defm VADDWs : N3VW_QHS<0,1,0b0001,0, "vaddw", "s", add, sext, 0>;
2929 defm VADDWu : N3VW_QHS<1,1,0b0001,0, "vaddw", "u", add, zext, 0>;
2930 // VHADD : Vector Halving Add
2931 defm VHADDs : N3VInt_QHS<0, 0, 0b0000, 0, N3RegFrm,
2932 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2933 "vhadd", "s", int_arm_neon_vhadds, 1>;
2934 defm VHADDu : N3VInt_QHS<1, 0, 0b0000, 0, N3RegFrm,
2935 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2936 "vhadd", "u", int_arm_neon_vhaddu, 1>;
2937 // VRHADD : Vector Rounding Halving Add
2938 defm VRHADDs : N3VInt_QHS<0, 0, 0b0001, 0, N3RegFrm,
2939 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2940 "vrhadd", "s", int_arm_neon_vrhadds, 1>;
2941 defm VRHADDu : N3VInt_QHS<1, 0, 0b0001, 0, N3RegFrm,
2942 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2943 "vrhadd", "u", int_arm_neon_vrhaddu, 1>;
2944 // VQADD : Vector Saturating Add
2945 defm VQADDs : N3VInt_QHSD<0, 0, 0b0000, 1, N3RegFrm,
2946 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2947 "vqadd", "s", int_arm_neon_vqadds, 1>;
2948 defm VQADDu : N3VInt_QHSD<1, 0, 0b0000, 1, N3RegFrm,
2949 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2950 "vqadd", "u", int_arm_neon_vqaddu, 1>;
2951 // VADDHN : Vector Add and Narrow Returning High Half (D = Q + Q)
2952 defm VADDHN : N3VNInt_HSD<0,1,0b0100,0, "vaddhn", "i",
2953 int_arm_neon_vaddhn, 1>;
2954 // VRADDHN : Vector Rounding Add and Narrow Returning High Half (D = Q + Q)
2955 defm VRADDHN : N3VNInt_HSD<1,1,0b0100,0, "vraddhn", "i",
2956 int_arm_neon_vraddhn, 1>;
2958 // Vector Multiply Operations.
2960 // VMUL : Vector Multiply (integer, polynomial and floating-point)
2961 defm VMUL : N3V_QHS<0, 0, 0b1001, 1, IIC_VMULi16D, IIC_VMULi32D,
2962 IIC_VMULi16Q, IIC_VMULi32Q, "vmul", "i", mul, 1>;
2963 def VMULpd : N3VDInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16D, "vmul",
2964 "p8", v8i8, v8i8, int_arm_neon_vmulp, 1>;
2965 def VMULpq : N3VQInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16Q, "vmul",
2966 "p8", v16i8, v16i8, int_arm_neon_vmulp, 1>;
2967 def VMULfd : N3VD<1, 0, 0b00, 0b1101, 1, IIC_VFMULD, "vmul", "f32",
2968 v2f32, v2f32, fmul, 1>;
2969 def VMULfq : N3VQ<1, 0, 0b00, 0b1101, 1, IIC_VFMULQ, "vmul", "f32",
2970 v4f32, v4f32, fmul, 1>;
2971 defm VMULsl : N3VSL_HS<0b1000, "vmul", "i", mul>;
2972 def VMULslfd : N3VDSL<0b10, 0b1001, IIC_VBIND, "vmul", "f32", v2f32, fmul>;
2973 def VMULslfq : N3VQSL<0b10, 0b1001, IIC_VBINQ, "vmul", "f32", v4f32,
2976 def : Pat<(v8i16 (mul (v8i16 QPR:$src1),
2977 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
2978 (v8i16 (VMULslv8i16 (v8i16 QPR:$src1),
2979 (v4i16 (EXTRACT_SUBREG QPR:$src2,
2980 (DSubReg_i16_reg imm:$lane))),
2981 (SubReg_i16_lane imm:$lane)))>;
2982 def : Pat<(v4i32 (mul (v4i32 QPR:$src1),
2983 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
2984 (v4i32 (VMULslv4i32 (v4i32 QPR:$src1),
2985 (v2i32 (EXTRACT_SUBREG QPR:$src2,
2986 (DSubReg_i32_reg imm:$lane))),
2987 (SubReg_i32_lane imm:$lane)))>;
2988 def : Pat<(v4f32 (fmul (v4f32 QPR:$src1),
2989 (v4f32 (NEONvduplane (v4f32 QPR:$src2), imm:$lane)))),
2990 (v4f32 (VMULslfq (v4f32 QPR:$src1),
2991 (v2f32 (EXTRACT_SUBREG QPR:$src2,
2992 (DSubReg_i32_reg imm:$lane))),
2993 (SubReg_i32_lane imm:$lane)))>;
2995 // VQDMULH : Vector Saturating Doubling Multiply Returning High Half
2996 defm VQDMULH : N3VInt_HS<0, 0, 0b1011, 0, N3RegFrm, IIC_VMULi16D, IIC_VMULi32D,
2997 IIC_VMULi16Q, IIC_VMULi32Q,
2998 "vqdmulh", "s", int_arm_neon_vqdmulh, 1>;
2999 defm VQDMULHsl: N3VIntSL_HS<0b1100, IIC_VMULi16D, IIC_VMULi32D,
3000 IIC_VMULi16Q, IIC_VMULi32Q,
3001 "vqdmulh", "s", int_arm_neon_vqdmulh>;
3002 def : Pat<(v8i16 (int_arm_neon_vqdmulh (v8i16 QPR:$src1),
3003 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
3005 (v8i16 (VQDMULHslv8i16 (v8i16 QPR:$src1),
3006 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3007 (DSubReg_i16_reg imm:$lane))),
3008 (SubReg_i16_lane imm:$lane)))>;
3009 def : Pat<(v4i32 (int_arm_neon_vqdmulh (v4i32 QPR:$src1),
3010 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
3012 (v4i32 (VQDMULHslv4i32 (v4i32 QPR:$src1),
3013 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3014 (DSubReg_i32_reg imm:$lane))),
3015 (SubReg_i32_lane imm:$lane)))>;
3017 // VQRDMULH : Vector Rounding Saturating Doubling Multiply Returning High Half
3018 defm VQRDMULH : N3VInt_HS<1, 0, 0b1011, 0, N3RegFrm,
3019 IIC_VMULi16D,IIC_VMULi32D,IIC_VMULi16Q,IIC_VMULi32Q,
3020 "vqrdmulh", "s", int_arm_neon_vqrdmulh, 1>;
3021 defm VQRDMULHsl : N3VIntSL_HS<0b1101, IIC_VMULi16D, IIC_VMULi32D,
3022 IIC_VMULi16Q, IIC_VMULi32Q,
3023 "vqrdmulh", "s", int_arm_neon_vqrdmulh>;
3024 def : Pat<(v8i16 (int_arm_neon_vqrdmulh (v8i16 QPR:$src1),
3025 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
3027 (v8i16 (VQRDMULHslv8i16 (v8i16 QPR:$src1),
3028 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3029 (DSubReg_i16_reg imm:$lane))),
3030 (SubReg_i16_lane imm:$lane)))>;
3031 def : Pat<(v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$src1),
3032 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
3034 (v4i32 (VQRDMULHslv4i32 (v4i32 QPR:$src1),
3035 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3036 (DSubReg_i32_reg imm:$lane))),
3037 (SubReg_i32_lane imm:$lane)))>;
3039 // VMULL : Vector Multiply Long (integer and polynomial) (Q = D * D)
3040 defm VMULLs : N3VL_QHS<0,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
3041 "vmull", "s", NEONvmulls, 1>;
3042 defm VMULLu : N3VL_QHS<1,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
3043 "vmull", "u", NEONvmullu, 1>;
3044 def VMULLp : N3VLInt<0, 1, 0b00, 0b1110, 0, IIC_VMULi16D, "vmull", "p8",
3045 v8i16, v8i8, int_arm_neon_vmullp, 1>;
3046 defm VMULLsls : N3VLSL_HS<0, 0b1010, IIC_VMULi16D, "vmull", "s", NEONvmulls>;
3047 defm VMULLslu : N3VLSL_HS<1, 0b1010, IIC_VMULi16D, "vmull", "u", NEONvmullu>;
3049 // VQDMULL : Vector Saturating Doubling Multiply Long (Q = D * D)
3050 defm VQDMULL : N3VLInt_HS<0,1,0b1101,0, IIC_VMULi16D, IIC_VMULi32D,
3051 "vqdmull", "s", int_arm_neon_vqdmull, 1>;
3052 defm VQDMULLsl: N3VLIntSL_HS<0, 0b1011, IIC_VMULi16D,
3053 "vqdmull", "s", int_arm_neon_vqdmull>;
3055 // Vector Multiply-Accumulate and Multiply-Subtract Operations.
3057 // VMLA : Vector Multiply Accumulate (integer and floating-point)
3058 defm VMLA : N3VMulOp_QHS<0, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3059 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
3060 def VMLAfd : N3VDMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACD, "vmla", "f32",
3062 def VMLAfq : N3VQMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACQ, "vmla", "f32",
3064 defm VMLAsl : N3VMulOpSL_HS<0b0000, IIC_VMACi16D, IIC_VMACi32D,
3065 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
3066 def VMLAslfd : N3VDMulOpSL<0b10, 0b0001, IIC_VMACD, "vmla", "f32",
3068 def VMLAslfq : N3VQMulOpSL<0b10, 0b0001, IIC_VMACQ, "vmla", "f32",
3069 v4f32, v2f32, fmul, fadd>;
3071 def : Pat<(v8i16 (add (v8i16 QPR:$src1),
3072 (mul (v8i16 QPR:$src2),
3073 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
3074 (v8i16 (VMLAslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
3075 (v4i16 (EXTRACT_SUBREG QPR:$src3,
3076 (DSubReg_i16_reg imm:$lane))),
3077 (SubReg_i16_lane imm:$lane)))>;
3079 def : Pat<(v4i32 (add (v4i32 QPR:$src1),
3080 (mul (v4i32 QPR:$src2),
3081 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
3082 (v4i32 (VMLAslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
3083 (v2i32 (EXTRACT_SUBREG QPR:$src3,
3084 (DSubReg_i32_reg imm:$lane))),
3085 (SubReg_i32_lane imm:$lane)))>;
3087 def : Pat<(v4f32 (fadd (v4f32 QPR:$src1),
3088 (fmul (v4f32 QPR:$src2),
3089 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
3090 (v4f32 (VMLAslfq (v4f32 QPR:$src1),
3092 (v2f32 (EXTRACT_SUBREG QPR:$src3,
3093 (DSubReg_i32_reg imm:$lane))),
3094 (SubReg_i32_lane imm:$lane)))>;
3096 // VMLAL : Vector Multiply Accumulate Long (Q += D * D)
3097 defm VMLALs : N3VLMulOp_QHS<0,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
3098 "vmlal", "s", NEONvmulls, add>;
3099 defm VMLALu : N3VLMulOp_QHS<1,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
3100 "vmlal", "u", NEONvmullu, add>;
3102 defm VMLALsls : N3VLMulOpSL_HS<0, 0b0010, "vmlal", "s", NEONvmulls, add>;
3103 defm VMLALslu : N3VLMulOpSL_HS<1, 0b0010, "vmlal", "u", NEONvmullu, add>;
3105 // VQDMLAL : Vector Saturating Doubling Multiply Accumulate Long (Q += D * D)
3106 defm VQDMLAL : N3VLInt3_HS<0, 1, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3107 "vqdmlal", "s", int_arm_neon_vqdmlal>;
3108 defm VQDMLALsl: N3VLInt3SL_HS<0, 0b0011, "vqdmlal", "s", int_arm_neon_vqdmlal>;
3110 // VMLS : Vector Multiply Subtract (integer and floating-point)
3111 defm VMLS : N3VMulOp_QHS<1, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3112 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
3113 def VMLSfd : N3VDMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACD, "vmls", "f32",
3115 def VMLSfq : N3VQMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACQ, "vmls", "f32",
3117 defm VMLSsl : N3VMulOpSL_HS<0b0100, IIC_VMACi16D, IIC_VMACi32D,
3118 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
3119 def VMLSslfd : N3VDMulOpSL<0b10, 0b0101, IIC_VMACD, "vmls", "f32",
3121 def VMLSslfq : N3VQMulOpSL<0b10, 0b0101, IIC_VMACQ, "vmls", "f32",
3122 v4f32, v2f32, fmul, fsub>;
3124 def : Pat<(v8i16 (sub (v8i16 QPR:$src1),
3125 (mul (v8i16 QPR:$src2),
3126 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
3127 (v8i16 (VMLSslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
3128 (v4i16 (EXTRACT_SUBREG QPR:$src3,
3129 (DSubReg_i16_reg imm:$lane))),
3130 (SubReg_i16_lane imm:$lane)))>;
3132 def : Pat<(v4i32 (sub (v4i32 QPR:$src1),
3133 (mul (v4i32 QPR:$src2),
3134 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
3135 (v4i32 (VMLSslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
3136 (v2i32 (EXTRACT_SUBREG QPR:$src3,
3137 (DSubReg_i32_reg imm:$lane))),
3138 (SubReg_i32_lane imm:$lane)))>;
3140 def : Pat<(v4f32 (fsub (v4f32 QPR:$src1),
3141 (fmul (v4f32 QPR:$src2),
3142 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
3143 (v4f32 (VMLSslfq (v4f32 QPR:$src1), (v4f32 QPR:$src2),
3144 (v2f32 (EXTRACT_SUBREG QPR:$src3,
3145 (DSubReg_i32_reg imm:$lane))),
3146 (SubReg_i32_lane imm:$lane)))>;
3148 // VMLSL : Vector Multiply Subtract Long (Q -= D * D)
3149 defm VMLSLs : N3VLMulOp_QHS<0,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
3150 "vmlsl", "s", NEONvmulls, sub>;
3151 defm VMLSLu : N3VLMulOp_QHS<1,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
3152 "vmlsl", "u", NEONvmullu, sub>;
3154 defm VMLSLsls : N3VLMulOpSL_HS<0, 0b0110, "vmlsl", "s", NEONvmulls, sub>;
3155 defm VMLSLslu : N3VLMulOpSL_HS<1, 0b0110, "vmlsl", "u", NEONvmullu, sub>;
3157 // VQDMLSL : Vector Saturating Doubling Multiply Subtract Long (Q -= D * D)
3158 defm VQDMLSL : N3VLInt3_HS<0, 1, 0b1011, 0, IIC_VMACi16D, IIC_VMACi32D,
3159 "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
3160 defm VQDMLSLsl: N3VLInt3SL_HS<0, 0b111, "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
3162 // Vector Subtract Operations.
3164 // VSUB : Vector Subtract (integer and floating-point)
3165 defm VSUB : N3V_QHSD<1, 0, 0b1000, 0, IIC_VSUBiD, IIC_VSUBiQ,
3166 "vsub", "i", sub, 0>;
3167 def VSUBfd : N3VD<0, 0, 0b10, 0b1101, 0, IIC_VBIND, "vsub", "f32",
3168 v2f32, v2f32, fsub, 0>;
3169 def VSUBfq : N3VQ<0, 0, 0b10, 0b1101, 0, IIC_VBINQ, "vsub", "f32",
3170 v4f32, v4f32, fsub, 0>;
3171 // VSUBL : Vector Subtract Long (Q = D - D)
3172 defm VSUBLs : N3VLExt_QHS<0,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
3173 "vsubl", "s", sub, sext, 0>;
3174 defm VSUBLu : N3VLExt_QHS<1,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
3175 "vsubl", "u", sub, zext, 0>;
3176 // VSUBW : Vector Subtract Wide (Q = Q - D)
3177 defm VSUBWs : N3VW_QHS<0,1,0b0011,0, "vsubw", "s", sub, sext, 0>;
3178 defm VSUBWu : N3VW_QHS<1,1,0b0011,0, "vsubw", "u", sub, zext, 0>;
3179 // VHSUB : Vector Halving Subtract
3180 defm VHSUBs : N3VInt_QHS<0, 0, 0b0010, 0, N3RegFrm,
3181 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3182 "vhsub", "s", int_arm_neon_vhsubs, 0>;
3183 defm VHSUBu : N3VInt_QHS<1, 0, 0b0010, 0, N3RegFrm,
3184 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3185 "vhsub", "u", int_arm_neon_vhsubu, 0>;
3186 // VQSUB : Vector Saturing Subtract
3187 defm VQSUBs : N3VInt_QHSD<0, 0, 0b0010, 1, N3RegFrm,
3188 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3189 "vqsub", "s", int_arm_neon_vqsubs, 0>;
3190 defm VQSUBu : N3VInt_QHSD<1, 0, 0b0010, 1, N3RegFrm,
3191 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3192 "vqsub", "u", int_arm_neon_vqsubu, 0>;
3193 // VSUBHN : Vector Subtract and Narrow Returning High Half (D = Q - Q)
3194 defm VSUBHN : N3VNInt_HSD<0,1,0b0110,0, "vsubhn", "i",
3195 int_arm_neon_vsubhn, 0>;
3196 // VRSUBHN : Vector Rounding Subtract and Narrow Returning High Half (D=Q-Q)
3197 defm VRSUBHN : N3VNInt_HSD<1,1,0b0110,0, "vrsubhn", "i",
3198 int_arm_neon_vrsubhn, 0>;
3200 // Vector Comparisons.
3202 // VCEQ : Vector Compare Equal
3203 defm VCEQ : N3V_QHS<1, 0, 0b1000, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3204 IIC_VSUBi4Q, "vceq", "i", NEONvceq, 1>;
3205 def VCEQfd : N3VD<0,0,0b00,0b1110,0, IIC_VBIND, "vceq", "f32", v2i32, v2f32,
3207 def VCEQfq : N3VQ<0,0,0b00,0b1110,0, IIC_VBINQ, "vceq", "f32", v4i32, v4f32,
3209 // For disassembly only.
3210 defm VCEQz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00010, 0, "vceq", "i",
3213 // VCGE : Vector Compare Greater Than or Equal
3214 defm VCGEs : N3V_QHS<0, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3215 IIC_VSUBi4Q, "vcge", "s", NEONvcge, 0>;
3216 defm VCGEu : N3V_QHS<1, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3217 IIC_VSUBi4Q, "vcge", "u", NEONvcgeu, 0>;
3218 def VCGEfd : N3VD<1,0,0b00,0b1110,0, IIC_VBIND, "vcge", "f32", v2i32, v2f32,
3220 def VCGEfq : N3VQ<1,0,0b00,0b1110,0, IIC_VBINQ, "vcge", "f32", v4i32, v4f32,
3222 // For disassembly only.
3223 // FIXME: This instruction's encoding MAY NOT BE correct.
3224 defm VCGEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00001, 0, "vcge", "s",
3226 // For disassembly only.
3227 // FIXME: This instruction's encoding MAY NOT BE correct.
3228 defm VCLEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00011, 0, "vcle", "s",
3231 // VCGT : Vector Compare Greater Than
3232 defm VCGTs : N3V_QHS<0, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3233 IIC_VSUBi4Q, "vcgt", "s", NEONvcgt, 0>;
3234 defm VCGTu : N3V_QHS<1, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3235 IIC_VSUBi4Q, "vcgt", "u", NEONvcgtu, 0>;
3236 def VCGTfd : N3VD<1,0,0b10,0b1110,0, IIC_VBIND, "vcgt", "f32", v2i32, v2f32,
3238 def VCGTfq : N3VQ<1,0,0b10,0b1110,0, IIC_VBINQ, "vcgt", "f32", v4i32, v4f32,
3240 // For disassembly only.
3241 // FIXME: This instruction's encoding MAY NOT BE correct.
3242 defm VCGTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00000, 0, "vcgt", "s",
3244 // For disassembly only.
3245 // FIXME: This instruction's encoding MAY NOT BE correct.
3246 defm VCLTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00100, 0, "vclt", "s",
3249 // VACGE : Vector Absolute Compare Greater Than or Equal (aka VCAGE)
3250 def VACGEd : N3VDInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacge",
3251 "f32", v2i32, v2f32, int_arm_neon_vacged, 0>;
3252 def VACGEq : N3VQInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacge",
3253 "f32", v4i32, v4f32, int_arm_neon_vacgeq, 0>;
3254 // VACGT : Vector Absolute Compare Greater Than (aka VCAGT)
3255 def VACGTd : N3VDInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacgt",
3256 "f32", v2i32, v2f32, int_arm_neon_vacgtd, 0>;
3257 def VACGTq : N3VQInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacgt",
3258 "f32", v4i32, v4f32, int_arm_neon_vacgtq, 0>;
3259 // VTST : Vector Test Bits
3260 defm VTST : N3V_QHS<0, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
3261 IIC_VBINi4Q, "vtst", "", NEONvtst, 1>;
3263 // Vector Bitwise Operations.
3265 def vnotd : PatFrag<(ops node:$in),
3266 (xor node:$in, (bitconvert (v8i8 NEONimmAllOnesV)))>;
3267 def vnotq : PatFrag<(ops node:$in),
3268 (xor node:$in, (bitconvert (v16i8 NEONimmAllOnesV)))>;
3271 // VAND : Vector Bitwise AND
3272 def VANDd : N3VDX<0, 0, 0b00, 0b0001, 1, IIC_VBINiD, "vand",
3273 v2i32, v2i32, and, 1>;
3274 def VANDq : N3VQX<0, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "vand",
3275 v4i32, v4i32, and, 1>;
3277 // VEOR : Vector Bitwise Exclusive OR
3278 def VEORd : N3VDX<1, 0, 0b00, 0b0001, 1, IIC_VBINiD, "veor",
3279 v2i32, v2i32, xor, 1>;
3280 def VEORq : N3VQX<1, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "veor",
3281 v4i32, v4i32, xor, 1>;
3283 // VORR : Vector Bitwise OR
3284 def VORRd : N3VDX<0, 0, 0b10, 0b0001, 1, IIC_VBINiD, "vorr",
3285 v2i32, v2i32, or, 1>;
3286 def VORRq : N3VQX<0, 0, 0b10, 0b0001, 1, IIC_VBINiQ, "vorr",
3287 v4i32, v4i32, or, 1>;
3289 // VBIC : Vector Bitwise Bit Clear (AND NOT)
3290 def VBICd : N3VX<0, 0, 0b01, 0b0001, 0, 1, (outs DPR:$dst),
3291 (ins DPR:$src1, DPR:$src2), N3RegFrm, IIC_VBINiD,
3292 "vbic", "$dst, $src1, $src2", "",
3293 [(set DPR:$dst, (v2i32 (and DPR:$src1,
3294 (vnotd DPR:$src2))))]>;
3295 def VBICq : N3VX<0, 0, 0b01, 0b0001, 1, 1, (outs QPR:$dst),
3296 (ins QPR:$src1, QPR:$src2), N3RegFrm, IIC_VBINiQ,
3297 "vbic", "$dst, $src1, $src2", "",
3298 [(set QPR:$dst, (v4i32 (and QPR:$src1,
3299 (vnotq QPR:$src2))))]>;
3301 // VORN : Vector Bitwise OR NOT
3302 def VORNd : N3VX<0, 0, 0b11, 0b0001, 0, 1, (outs DPR:$dst),
3303 (ins DPR:$src1, DPR:$src2), N3RegFrm, IIC_VBINiD,
3304 "vorn", "$dst, $src1, $src2", "",
3305 [(set DPR:$dst, (v2i32 (or DPR:$src1,
3306 (vnotd DPR:$src2))))]>;
3307 def VORNq : N3VX<0, 0, 0b11, 0b0001, 1, 1, (outs QPR:$dst),
3308 (ins QPR:$src1, QPR:$src2), N3RegFrm, IIC_VBINiQ,
3309 "vorn", "$dst, $src1, $src2", "",
3310 [(set QPR:$dst, (v4i32 (or QPR:$src1,
3311 (vnotq QPR:$src2))))]>;
3313 // VMVN : Vector Bitwise NOT (Immediate)
3315 let isReMaterializable = 1 in {
3317 def VMVNv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 1, 1, (outs DPR:$dst),
3318 (ins nModImm:$SIMM), IIC_VMOVImm,
3319 "vmvn", "i16", "$dst, $SIMM", "",
3320 [(set DPR:$dst, (v4i16 (NEONvmvnImm timm:$SIMM)))]> {
3321 let Inst{9} = SIMM{9};
3324 def VMVNv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 1, 1, (outs QPR:$dst),
3325 (ins nModImm:$SIMM), IIC_VMOVImm,
3326 "vmvn", "i16", "$dst, $SIMM", "",
3327 [(set QPR:$dst, (v8i16 (NEONvmvnImm timm:$SIMM)))]> {
3328 let Inst{9} = SIMM{9};
3331 def VMVNv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 1, 1, (outs DPR:$dst),
3332 (ins nModImm:$SIMM), IIC_VMOVImm,
3333 "vmvn", "i32", "$dst, $SIMM", "",
3334 [(set DPR:$dst, (v2i32 (NEONvmvnImm timm:$SIMM)))]> {
3335 let Inst{11-8} = SIMM{11-8};
3338 def VMVNv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 1, 1, (outs QPR:$dst),
3339 (ins nModImm:$SIMM), IIC_VMOVImm,
3340 "vmvn", "i32", "$dst, $SIMM", "",
3341 [(set QPR:$dst, (v4i32 (NEONvmvnImm timm:$SIMM)))]> {
3342 let Inst{11-8} = SIMM{11-8};
3346 // VMVN : Vector Bitwise NOT
3347 def VMVNd : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 0, 0,
3348 (outs DPR:$dst), (ins DPR:$src), IIC_VSUBiD,
3349 "vmvn", "$dst, $src", "",
3350 [(set DPR:$dst, (v2i32 (vnotd DPR:$src)))]>;
3351 def VMVNq : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 1, 0,
3352 (outs QPR:$dst), (ins QPR:$src), IIC_VSUBiD,
3353 "vmvn", "$dst, $src", "",
3354 [(set QPR:$dst, (v4i32 (vnotq QPR:$src)))]>;
3355 def : Pat<(v2i32 (vnotd DPR:$src)), (VMVNd DPR:$src)>;
3356 def : Pat<(v4i32 (vnotq QPR:$src)), (VMVNq QPR:$src)>;
3358 // VBSL : Vector Bitwise Select
3359 def VBSLd : N3VX<1, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
3360 (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
3361 N3RegFrm, IIC_VCNTiD,
3362 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3364 (v2i32 (or (and DPR:$Vn, DPR:$src1),
3365 (and DPR:$Vm, (vnotd DPR:$src1)))))]>;
3366 def VBSLq : N3VX<1, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
3367 (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
3368 N3RegFrm, IIC_VCNTiQ,
3369 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3371 (v4i32 (or (and QPR:$Vn, QPR:$src1),
3372 (and QPR:$Vm, (vnotq QPR:$src1)))))]>;
3374 // VBIF : Vector Bitwise Insert if False
3375 // like VBSL but with: "vbif $dst, $src3, $src1", "$src2 = $dst",
3376 // FIXME: This instruction's encoding MAY NOT BE correct.
3377 def VBIFd : N3VX<1, 0, 0b11, 0b0001, 0, 1,
3378 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
3379 N3RegFrm, IIC_VBINiD,
3380 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3381 [/* For disassembly only; pattern left blank */]>;
3382 def VBIFq : N3VX<1, 0, 0b11, 0b0001, 1, 1,
3383 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
3384 N3RegFrm, IIC_VBINiQ,
3385 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3386 [/* For disassembly only; pattern left blank */]>;
3388 // VBIT : Vector Bitwise Insert if True
3389 // like VBSL but with: "vbit $dst, $src2, $src1", "$src3 = $dst",
3390 // FIXME: This instruction's encoding MAY NOT BE correct.
3391 def VBITd : N3VX<1, 0, 0b10, 0b0001, 0, 1,
3392 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
3393 N3RegFrm, IIC_VBINiD,
3394 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3395 [/* For disassembly only; pattern left blank */]>;
3396 def VBITq : N3VX<1, 0, 0b10, 0b0001, 1, 1,
3397 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
3398 N3RegFrm, IIC_VBINiQ,
3399 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3400 [/* For disassembly only; pattern left blank */]>;
3402 // VBIT/VBIF are not yet implemented. The TwoAddress pass will not go looking
3403 // for equivalent operations with different register constraints; it just
3406 // Vector Absolute Differences.
3408 // VABD : Vector Absolute Difference
3409 defm VABDs : N3VInt_QHS<0, 0, 0b0111, 0, N3RegFrm,
3410 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3411 "vabd", "s", int_arm_neon_vabds, 1>;
3412 defm VABDu : N3VInt_QHS<1, 0, 0b0111, 0, N3RegFrm,
3413 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3414 "vabd", "u", int_arm_neon_vabdu, 1>;
3415 def VABDfd : N3VDInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBIND,
3416 "vabd", "f32", v2f32, v2f32, int_arm_neon_vabds, 1>;
3417 def VABDfq : N3VQInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBINQ,
3418 "vabd", "f32", v4f32, v4f32, int_arm_neon_vabds, 1>;
3420 // VABDL : Vector Absolute Difference Long (Q = | D - D |)
3421 defm VABDLs : N3VLIntExt_QHS<0,1,0b0111,0, IIC_VSUBi4Q,
3422 "vabdl", "s", int_arm_neon_vabds, zext, 1>;
3423 defm VABDLu : N3VLIntExt_QHS<1,1,0b0111,0, IIC_VSUBi4Q,
3424 "vabdl", "u", int_arm_neon_vabdu, zext, 1>;
3426 // VABA : Vector Absolute Difference and Accumulate
3427 defm VABAs : N3VIntOp_QHS<0,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
3428 "vaba", "s", int_arm_neon_vabds, add>;
3429 defm VABAu : N3VIntOp_QHS<1,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
3430 "vaba", "u", int_arm_neon_vabdu, add>;
3432 // VABAL : Vector Absolute Difference and Accumulate Long (Q += | D - D |)
3433 defm VABALs : N3VLIntExtOp_QHS<0,1,0b0101,0, IIC_VABAD,
3434 "vabal", "s", int_arm_neon_vabds, zext, add>;
3435 defm VABALu : N3VLIntExtOp_QHS<1,1,0b0101,0, IIC_VABAD,
3436 "vabal", "u", int_arm_neon_vabdu, zext, add>;
3438 // Vector Maximum and Minimum.
3440 // VMAX : Vector Maximum
3441 defm VMAXs : N3VInt_QHS<0, 0, 0b0110, 0, N3RegFrm,
3442 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3443 "vmax", "s", int_arm_neon_vmaxs, 1>;
3444 defm VMAXu : N3VInt_QHS<1, 0, 0b0110, 0, N3RegFrm,
3445 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3446 "vmax", "u", int_arm_neon_vmaxu, 1>;
3447 def VMAXfd : N3VDInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBIND,
3449 v2f32, v2f32, int_arm_neon_vmaxs, 1>;
3450 def VMAXfq : N3VQInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBINQ,
3452 v4f32, v4f32, int_arm_neon_vmaxs, 1>;
3454 // VMIN : Vector Minimum
3455 defm VMINs : N3VInt_QHS<0, 0, 0b0110, 1, N3RegFrm,
3456 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3457 "vmin", "s", int_arm_neon_vmins, 1>;
3458 defm VMINu : N3VInt_QHS<1, 0, 0b0110, 1, N3RegFrm,
3459 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3460 "vmin", "u", int_arm_neon_vminu, 1>;
3461 def VMINfd : N3VDInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBIND,
3463 v2f32, v2f32, int_arm_neon_vmins, 1>;
3464 def VMINfq : N3VQInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBINQ,
3466 v4f32, v4f32, int_arm_neon_vmins, 1>;
3468 // Vector Pairwise Operations.
3470 // VPADD : Vector Pairwise Add
3471 def VPADDi8 : N3VDInt<0, 0, 0b00, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
3473 v8i8, v8i8, int_arm_neon_vpadd, 0>;
3474 def VPADDi16 : N3VDInt<0, 0, 0b01, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
3476 v4i16, v4i16, int_arm_neon_vpadd, 0>;
3477 def VPADDi32 : N3VDInt<0, 0, 0b10, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
3479 v2i32, v2i32, int_arm_neon_vpadd, 0>;
3480 def VPADDf : N3VDInt<1, 0, 0b00, 0b1101, 0, N3RegFrm,
3481 IIC_VPBIND, "vpadd", "f32",
3482 v2f32, v2f32, int_arm_neon_vpadd, 0>;
3484 // VPADDL : Vector Pairwise Add Long
3485 defm VPADDLs : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00100, 0, "vpaddl", "s",
3486 int_arm_neon_vpaddls>;
3487 defm VPADDLu : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00101, 0, "vpaddl", "u",
3488 int_arm_neon_vpaddlu>;
3490 // VPADAL : Vector Pairwise Add and Accumulate Long
3491 defm VPADALs : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01100, 0, "vpadal", "s",
3492 int_arm_neon_vpadals>;
3493 defm VPADALu : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01101, 0, "vpadal", "u",
3494 int_arm_neon_vpadalu>;
3496 // VPMAX : Vector Pairwise Maximum
3497 def VPMAXs8 : N3VDInt<0, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3498 "s8", v8i8, v8i8, int_arm_neon_vpmaxs, 0>;
3499 def VPMAXs16 : N3VDInt<0, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3500 "s16", v4i16, v4i16, int_arm_neon_vpmaxs, 0>;
3501 def VPMAXs32 : N3VDInt<0, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3502 "s32", v2i32, v2i32, int_arm_neon_vpmaxs, 0>;
3503 def VPMAXu8 : N3VDInt<1, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3504 "u8", v8i8, v8i8, int_arm_neon_vpmaxu, 0>;
3505 def VPMAXu16 : N3VDInt<1, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3506 "u16", v4i16, v4i16, int_arm_neon_vpmaxu, 0>;
3507 def VPMAXu32 : N3VDInt<1, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3508 "u32", v2i32, v2i32, int_arm_neon_vpmaxu, 0>;
3509 def VPMAXf : N3VDInt<1, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmax",
3510 "f32", v2f32, v2f32, int_arm_neon_vpmaxs, 0>;
3512 // VPMIN : Vector Pairwise Minimum
3513 def VPMINs8 : N3VDInt<0, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3514 "s8", v8i8, v8i8, int_arm_neon_vpmins, 0>;
3515 def VPMINs16 : N3VDInt<0, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3516 "s16", v4i16, v4i16, int_arm_neon_vpmins, 0>;
3517 def VPMINs32 : N3VDInt<0, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3518 "s32", v2i32, v2i32, int_arm_neon_vpmins, 0>;
3519 def VPMINu8 : N3VDInt<1, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3520 "u8", v8i8, v8i8, int_arm_neon_vpminu, 0>;
3521 def VPMINu16 : N3VDInt<1, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3522 "u16", v4i16, v4i16, int_arm_neon_vpminu, 0>;
3523 def VPMINu32 : N3VDInt<1, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3524 "u32", v2i32, v2i32, int_arm_neon_vpminu, 0>;
3525 def VPMINf : N3VDInt<1, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmin",
3526 "f32", v2f32, v2f32, int_arm_neon_vpmins, 0>;
3528 // Vector Reciprocal and Reciprocal Square Root Estimate and Step.
3530 // VRECPE : Vector Reciprocal Estimate
3531 def VRECPEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
3532 IIC_VUNAD, "vrecpe", "u32",
3533 v2i32, v2i32, int_arm_neon_vrecpe>;
3534 def VRECPEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
3535 IIC_VUNAQ, "vrecpe", "u32",
3536 v4i32, v4i32, int_arm_neon_vrecpe>;
3537 def VRECPEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
3538 IIC_VUNAD, "vrecpe", "f32",
3539 v2f32, v2f32, int_arm_neon_vrecpe>;
3540 def VRECPEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
3541 IIC_VUNAQ, "vrecpe", "f32",
3542 v4f32, v4f32, int_arm_neon_vrecpe>;
3544 // VRECPS : Vector Reciprocal Step
3545 def VRECPSfd : N3VDInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
3546 IIC_VRECSD, "vrecps", "f32",
3547 v2f32, v2f32, int_arm_neon_vrecps, 1>;
3548 def VRECPSfq : N3VQInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
3549 IIC_VRECSQ, "vrecps", "f32",
3550 v4f32, v4f32, int_arm_neon_vrecps, 1>;
3552 // VRSQRTE : Vector Reciprocal Square Root Estimate
3553 def VRSQRTEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
3554 IIC_VUNAD, "vrsqrte", "u32",
3555 v2i32, v2i32, int_arm_neon_vrsqrte>;
3556 def VRSQRTEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
3557 IIC_VUNAQ, "vrsqrte", "u32",
3558 v4i32, v4i32, int_arm_neon_vrsqrte>;
3559 def VRSQRTEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
3560 IIC_VUNAD, "vrsqrte", "f32",
3561 v2f32, v2f32, int_arm_neon_vrsqrte>;
3562 def VRSQRTEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
3563 IIC_VUNAQ, "vrsqrte", "f32",
3564 v4f32, v4f32, int_arm_neon_vrsqrte>;
3566 // VRSQRTS : Vector Reciprocal Square Root Step
3567 def VRSQRTSfd : N3VDInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
3568 IIC_VRECSD, "vrsqrts", "f32",
3569 v2f32, v2f32, int_arm_neon_vrsqrts, 1>;
3570 def VRSQRTSfq : N3VQInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
3571 IIC_VRECSQ, "vrsqrts", "f32",
3572 v4f32, v4f32, int_arm_neon_vrsqrts, 1>;
3576 // VSHL : Vector Shift
3577 defm VSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 0, N3RegVShFrm,
3578 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
3579 "vshl", "s", int_arm_neon_vshifts>;
3580 defm VSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 0, N3RegVShFrm,
3581 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
3582 "vshl", "u", int_arm_neon_vshiftu>;
3583 // VSHL : Vector Shift Left (Immediate)
3584 defm VSHLi : N2VSh_QHSD<0, 1, 0b0101, 1, IIC_VSHLiD, "vshl", "i", NEONvshl,
3586 // VSHR : Vector Shift Right (Immediate)
3587 defm VSHRs : N2VSh_QHSD<0, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "s", NEONvshrs,
3589 defm VSHRu : N2VSh_QHSD<1, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "u", NEONvshru,
3592 // VSHLL : Vector Shift Left Long
3593 defm VSHLLs : N2VLSh_QHS<0, 1, 0b1010, 0, 0, 1, "vshll", "s", NEONvshlls>;
3594 defm VSHLLu : N2VLSh_QHS<1, 1, 0b1010, 0, 0, 1, "vshll", "u", NEONvshllu>;
3596 // VSHLL : Vector Shift Left Long (with maximum shift count)
3597 class N2VLShMax<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
3598 bit op6, bit op4, string OpcodeStr, string Dt, ValueType ResTy,
3599 ValueType OpTy, SDNode OpNode>
3600 : N2VLSh<op24, op23, op11_8, op7, op6, op4, OpcodeStr, Dt,
3601 ResTy, OpTy, OpNode> {
3602 let Inst{21-16} = op21_16;
3604 def VSHLLi8 : N2VLShMax<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll", "i8",
3605 v8i16, v8i8, NEONvshlli>;
3606 def VSHLLi16 : N2VLShMax<1, 1, 0b110110, 0b0011, 0, 0, 0, "vshll", "i16",
3607 v4i32, v4i16, NEONvshlli>;
3608 def VSHLLi32 : N2VLShMax<1, 1, 0b111010, 0b0011, 0, 0, 0, "vshll", "i32",
3609 v2i64, v2i32, NEONvshlli>;
3611 // VSHRN : Vector Shift Right and Narrow
3612 defm VSHRN : N2VNSh_HSD<0,1,0b1000,0,0,1, IIC_VSHLiD, "vshrn", "i",
3615 // VRSHL : Vector Rounding Shift
3616 defm VRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 0, N3RegVShFrm,
3617 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3618 "vrshl", "s", int_arm_neon_vrshifts>;
3619 defm VRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 0, N3RegVShFrm,
3620 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3621 "vrshl", "u", int_arm_neon_vrshiftu>;
3622 // VRSHR : Vector Rounding Shift Right
3623 defm VRSHRs : N2VSh_QHSD<0,1,0b0010,1, IIC_VSHLi4D, "vrshr", "s", NEONvrshrs,
3625 defm VRSHRu : N2VSh_QHSD<1,1,0b0010,1, IIC_VSHLi4D, "vrshr", "u", NEONvrshru,
3628 // VRSHRN : Vector Rounding Shift Right and Narrow
3629 defm VRSHRN : N2VNSh_HSD<0, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vrshrn", "i",
3632 // VQSHL : Vector Saturating Shift
3633 defm VQSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 1, N3RegVShFrm,
3634 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3635 "vqshl", "s", int_arm_neon_vqshifts>;
3636 defm VQSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 1, N3RegVShFrm,
3637 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3638 "vqshl", "u", int_arm_neon_vqshiftu>;
3639 // VQSHL : Vector Saturating Shift Left (Immediate)
3640 defm VQSHLsi : N2VSh_QHSD<0,1,0b0111,1, IIC_VSHLi4D, "vqshl", "s",NEONvqshls,
3642 defm VQSHLui : N2VSh_QHSD<1,1,0b0111,1, IIC_VSHLi4D, "vqshl", "u",NEONvqshlu,
3644 // VQSHLU : Vector Saturating Shift Left (Immediate, Unsigned)
3645 defm VQSHLsu : N2VSh_QHSD<1,1,0b0110,1, IIC_VSHLi4D,"vqshlu","s",NEONvqshlsu,
3648 // VQSHRN : Vector Saturating Shift Right and Narrow
3649 defm VQSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "s",
3651 defm VQSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "u",
3654 // VQSHRUN : Vector Saturating Shift Right and Narrow (Unsigned)
3655 defm VQSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 0, 1, IIC_VSHLi4D, "vqshrun", "s",
3658 // VQRSHL : Vector Saturating Rounding Shift
3659 defm VQRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 1, N3RegVShFrm,
3660 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3661 "vqrshl", "s", int_arm_neon_vqrshifts>;
3662 defm VQRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 1, N3RegVShFrm,
3663 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3664 "vqrshl", "u", int_arm_neon_vqrshiftu>;
3666 // VQRSHRN : Vector Saturating Rounding Shift Right and Narrow
3667 defm VQRSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "s",
3669 defm VQRSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "u",
3672 // VQRSHRUN : Vector Saturating Rounding Shift Right and Narrow (Unsigned)
3673 defm VQRSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vqrshrun", "s",
3676 // VSRA : Vector Shift Right and Accumulate
3677 defm VSRAs : N2VShAdd_QHSD<0, 1, 0b0001, 1, "vsra", "s", NEONvshrs>;
3678 defm VSRAu : N2VShAdd_QHSD<1, 1, 0b0001, 1, "vsra", "u", NEONvshru>;
3679 // VRSRA : Vector Rounding Shift Right and Accumulate
3680 defm VRSRAs : N2VShAdd_QHSD<0, 1, 0b0011, 1, "vrsra", "s", NEONvrshrs>;
3681 defm VRSRAu : N2VShAdd_QHSD<1, 1, 0b0011, 1, "vrsra", "u", NEONvrshru>;
3683 // VSLI : Vector Shift Left and Insert
3684 defm VSLI : N2VShIns_QHSD<1, 1, 0b0101, 1, "vsli", NEONvsli, N2RegVShLFrm>;
3685 // VSRI : Vector Shift Right and Insert
3686 defm VSRI : N2VShIns_QHSD<1, 1, 0b0100, 1, "vsri", NEONvsri, N2RegVShRFrm>;
3688 // Vector Absolute and Saturating Absolute.
3690 // VABS : Vector Absolute Value
3691 defm VABS : N2VInt_QHS<0b11, 0b11, 0b01, 0b00110, 0,
3692 IIC_VUNAiD, IIC_VUNAiQ, "vabs", "s",
3694 def VABSfd : N2VDInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
3695 IIC_VUNAD, "vabs", "f32",
3696 v2f32, v2f32, int_arm_neon_vabs>;
3697 def VABSfq : N2VQInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
3698 IIC_VUNAQ, "vabs", "f32",
3699 v4f32, v4f32, int_arm_neon_vabs>;
3701 // VQABS : Vector Saturating Absolute Value
3702 defm VQABS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01110, 0,
3703 IIC_VQUNAiD, IIC_VQUNAiQ, "vqabs", "s",
3704 int_arm_neon_vqabs>;
3708 def vnegd : PatFrag<(ops node:$in),
3709 (sub (bitconvert (v2i32 NEONimmAllZerosV)), node:$in)>;
3710 def vnegq : PatFrag<(ops node:$in),
3711 (sub (bitconvert (v4i32 NEONimmAllZerosV)), node:$in)>;
3713 class VNEGD<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
3714 : N2V<0b11, 0b11, size, 0b01, 0b00111, 0, 0, (outs DPR:$dst), (ins DPR:$src),
3715 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
3716 [(set DPR:$dst, (Ty (vnegd DPR:$src)))]>;
3717 class VNEGQ<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
3718 : N2V<0b11, 0b11, size, 0b01, 0b00111, 1, 0, (outs QPR:$dst), (ins QPR:$src),
3719 IIC_VSHLiQ, OpcodeStr, Dt, "$dst, $src", "",
3720 [(set QPR:$dst, (Ty (vnegq QPR:$src)))]>;
3722 // VNEG : Vector Negate (integer)
3723 def VNEGs8d : VNEGD<0b00, "vneg", "s8", v8i8>;
3724 def VNEGs16d : VNEGD<0b01, "vneg", "s16", v4i16>;
3725 def VNEGs32d : VNEGD<0b10, "vneg", "s32", v2i32>;
3726 def VNEGs8q : VNEGQ<0b00, "vneg", "s8", v16i8>;
3727 def VNEGs16q : VNEGQ<0b01, "vneg", "s16", v8i16>;
3728 def VNEGs32q : VNEGQ<0b10, "vneg", "s32", v4i32>;
3730 // VNEG : Vector Negate (floating-point)
3731 def VNEGfd : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
3732 (outs DPR:$dst), (ins DPR:$src), IIC_VUNAD,
3733 "vneg", "f32", "$dst, $src", "",
3734 [(set DPR:$dst, (v2f32 (fneg DPR:$src)))]>;
3735 def VNEGf32q : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 1, 0,
3736 (outs QPR:$dst), (ins QPR:$src), IIC_VUNAQ,
3737 "vneg", "f32", "$dst, $src", "",
3738 [(set QPR:$dst, (v4f32 (fneg QPR:$src)))]>;
3740 def : Pat<(v8i8 (vnegd DPR:$src)), (VNEGs8d DPR:$src)>;
3741 def : Pat<(v4i16 (vnegd DPR:$src)), (VNEGs16d DPR:$src)>;
3742 def : Pat<(v2i32 (vnegd DPR:$src)), (VNEGs32d DPR:$src)>;
3743 def : Pat<(v16i8 (vnegq QPR:$src)), (VNEGs8q QPR:$src)>;
3744 def : Pat<(v8i16 (vnegq QPR:$src)), (VNEGs16q QPR:$src)>;
3745 def : Pat<(v4i32 (vnegq QPR:$src)), (VNEGs32q QPR:$src)>;
3747 // VQNEG : Vector Saturating Negate
3748 defm VQNEG : N2VInt_QHS<0b11, 0b11, 0b00, 0b01111, 0,
3749 IIC_VQUNAiD, IIC_VQUNAiQ, "vqneg", "s",
3750 int_arm_neon_vqneg>;
3752 // Vector Bit Counting Operations.
3754 // VCLS : Vector Count Leading Sign Bits
3755 defm VCLS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01000, 0,
3756 IIC_VCNTiD, IIC_VCNTiQ, "vcls", "s",
3758 // VCLZ : Vector Count Leading Zeros
3759 defm VCLZ : N2VInt_QHS<0b11, 0b11, 0b00, 0b01001, 0,
3760 IIC_VCNTiD, IIC_VCNTiQ, "vclz", "i",
3762 // VCNT : Vector Count One Bits
3763 def VCNTd : N2VDInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
3764 IIC_VCNTiD, "vcnt", "8",
3765 v8i8, v8i8, int_arm_neon_vcnt>;
3766 def VCNTq : N2VQInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
3767 IIC_VCNTiQ, "vcnt", "8",
3768 v16i8, v16i8, int_arm_neon_vcnt>;
3770 // Vector Swap -- for disassembly only.
3771 def VSWPd : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 0, 0,
3772 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
3773 "vswp", "$dst, $src", "", []>;
3774 def VSWPq : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 1, 0,
3775 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
3776 "vswp", "$dst, $src", "", []>;
3778 // Vector Move Operations.
3780 // VMOV : Vector Move (Register)
3782 let neverHasSideEffects = 1 in {
3783 def VMOVDneon: N3VX<0, 0, 0b10, 0b0001, 0, 1, (outs DPR:$dst), (ins DPR:$src),
3784 N3RegFrm, IIC_VMOV, "vmov", "$dst, $src", "", []>;
3785 def VMOVQ : N3VX<0, 0, 0b10, 0b0001, 1, 1, (outs QPR:$dst), (ins QPR:$src),
3786 N3RegFrm, IIC_VMOV, "vmov", "$dst, $src", "", []>;
3788 // Pseudo vector move instructions for QQ and QQQQ registers. This should
3789 // be expanded after register allocation is completed.
3790 def VMOVQQ : PseudoInst<(outs QQPR:$dst), (ins QQPR:$src),
3791 NoItinerary, "", []>;
3793 def VMOVQQQQ : PseudoInst<(outs QQQQPR:$dst), (ins QQQQPR:$src),
3794 NoItinerary, "", []>;
3795 } // neverHasSideEffects
3797 // VMOV : Vector Move (Immediate)
3799 let isReMaterializable = 1 in {
3800 def VMOVv8i8 : N1ModImm<1, 0b000, 0b1110, 0, 0, 0, 1, (outs DPR:$dst),
3801 (ins nModImm:$SIMM), IIC_VMOVImm,
3802 "vmov", "i8", "$dst, $SIMM", "",
3803 [(set DPR:$dst, (v8i8 (NEONvmovImm timm:$SIMM)))]>;
3804 def VMOVv16i8 : N1ModImm<1, 0b000, 0b1110, 0, 1, 0, 1, (outs QPR:$dst),
3805 (ins nModImm:$SIMM), IIC_VMOVImm,
3806 "vmov", "i8", "$dst, $SIMM", "",
3807 [(set QPR:$dst, (v16i8 (NEONvmovImm timm:$SIMM)))]>;
3809 def VMOVv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 0, 1, (outs DPR:$dst),
3810 (ins nModImm:$SIMM), IIC_VMOVImm,
3811 "vmov", "i16", "$dst, $SIMM", "",
3812 [(set DPR:$dst, (v4i16 (NEONvmovImm timm:$SIMM)))]> {
3813 let Inst{9} = SIMM{9};
3816 def VMOVv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 0, 1, (outs QPR:$dst),
3817 (ins nModImm:$SIMM), IIC_VMOVImm,
3818 "vmov", "i16", "$dst, $SIMM", "",
3819 [(set QPR:$dst, (v8i16 (NEONvmovImm timm:$SIMM)))]> {
3820 let Inst{9} = SIMM{9};
3823 def VMOVv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 0, 1, (outs DPR:$dst),
3824 (ins nModImm:$SIMM), IIC_VMOVImm,
3825 "vmov", "i32", "$dst, $SIMM", "",
3826 [(set DPR:$dst, (v2i32 (NEONvmovImm timm:$SIMM)))]> {
3827 let Inst{11-8} = SIMM{11-8};
3830 def VMOVv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 0, 1, (outs QPR:$dst),
3831 (ins nModImm:$SIMM), IIC_VMOVImm,
3832 "vmov", "i32", "$dst, $SIMM", "",
3833 [(set QPR:$dst, (v4i32 (NEONvmovImm timm:$SIMM)))]> {
3834 let Inst{11-8} = SIMM{11-8};
3837 def VMOVv1i64 : N1ModImm<1, 0b000, 0b1110, 0, 0, 1, 1, (outs DPR:$dst),
3838 (ins nModImm:$SIMM), IIC_VMOVImm,
3839 "vmov", "i64", "$dst, $SIMM", "",
3840 [(set DPR:$dst, (v1i64 (NEONvmovImm timm:$SIMM)))]>;
3841 def VMOVv2i64 : N1ModImm<1, 0b000, 0b1110, 0, 1, 1, 1, (outs QPR:$dst),
3842 (ins nModImm:$SIMM), IIC_VMOVImm,
3843 "vmov", "i64", "$dst, $SIMM", "",
3844 [(set QPR:$dst, (v2i64 (NEONvmovImm timm:$SIMM)))]>;
3845 } // isReMaterializable
3847 // VMOV : Vector Get Lane (move scalar to ARM core register)
3849 def VGETLNs8 : NVGetLane<{1,1,1,0,0,1,?,1}, 0b1011, {?,?},
3850 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
3851 IIC_VMOVSI, "vmov", "s8", "$R, $V[$lane]",
3852 [(set GPR:$R, (NEONvgetlanes (v8i8 DPR:$V),
3854 let Inst{21} = lane{2};
3855 let Inst{6-5} = lane{1-0};
3857 def VGETLNs16 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, {?,1},
3858 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
3859 IIC_VMOVSI, "vmov", "s16", "$R, $V[$lane]",
3860 [(set GPR:$R, (NEONvgetlanes (v4i16 DPR:$V),
3862 let Inst{21} = lane{1};
3863 let Inst{6} = lane{0};
3865 def VGETLNu8 : NVGetLane<{1,1,1,0,1,1,?,1}, 0b1011, {?,?},
3866 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
3867 IIC_VMOVSI, "vmov", "u8", "$R, $V[$lane]",
3868 [(set GPR:$R, (NEONvgetlaneu (v8i8 DPR:$V),
3870 let Inst{21} = lane{2};
3871 let Inst{6-5} = lane{1-0};
3873 def VGETLNu16 : NVGetLane<{1,1,1,0,1,0,?,1}, 0b1011, {?,1},
3874 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
3875 IIC_VMOVSI, "vmov", "u16", "$R, $V[$lane]",
3876 [(set GPR:$R, (NEONvgetlaneu (v4i16 DPR:$V),
3878 let Inst{21} = lane{1};
3879 let Inst{6} = lane{0};
3881 def VGETLNi32 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, 0b00,
3882 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
3883 IIC_VMOVSI, "vmov", "32", "$R, $V[$lane]",
3884 [(set GPR:$R, (extractelt (v2i32 DPR:$V),
3886 let Inst{21} = lane{0};
3888 // def VGETLNf32: see FMRDH and FMRDL in ARMInstrVFP.td
3889 def : Pat<(NEONvgetlanes (v16i8 QPR:$src), imm:$lane),
3890 (VGETLNs8 (v8i8 (EXTRACT_SUBREG QPR:$src,
3891 (DSubReg_i8_reg imm:$lane))),
3892 (SubReg_i8_lane imm:$lane))>;
3893 def : Pat<(NEONvgetlanes (v8i16 QPR:$src), imm:$lane),
3894 (VGETLNs16 (v4i16 (EXTRACT_SUBREG QPR:$src,
3895 (DSubReg_i16_reg imm:$lane))),
3896 (SubReg_i16_lane imm:$lane))>;
3897 def : Pat<(NEONvgetlaneu (v16i8 QPR:$src), imm:$lane),
3898 (VGETLNu8 (v8i8 (EXTRACT_SUBREG QPR:$src,
3899 (DSubReg_i8_reg imm:$lane))),
3900 (SubReg_i8_lane imm:$lane))>;
3901 def : Pat<(NEONvgetlaneu (v8i16 QPR:$src), imm:$lane),
3902 (VGETLNu16 (v4i16 (EXTRACT_SUBREG QPR:$src,
3903 (DSubReg_i16_reg imm:$lane))),
3904 (SubReg_i16_lane imm:$lane))>;
3905 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
3906 (VGETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src,
3907 (DSubReg_i32_reg imm:$lane))),
3908 (SubReg_i32_lane imm:$lane))>;
3909 def : Pat<(extractelt (v2f32 DPR:$src1), imm:$src2),
3910 (EXTRACT_SUBREG (v2f32 (COPY_TO_REGCLASS (v2f32 DPR:$src1),DPR_VFP2)),
3911 (SSubReg_f32_reg imm:$src2))>;
3912 def : Pat<(extractelt (v4f32 QPR:$src1), imm:$src2),
3913 (EXTRACT_SUBREG (v4f32 (COPY_TO_REGCLASS (v4f32 QPR:$src1),QPR_VFP2)),
3914 (SSubReg_f32_reg imm:$src2))>;
3915 //def : Pat<(extractelt (v2i64 QPR:$src1), imm:$src2),
3916 // (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
3917 def : Pat<(extractelt (v2f64 QPR:$src1), imm:$src2),
3918 (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
3921 // VMOV : Vector Set Lane (move ARM core register to scalar)
3923 let Constraints = "$src1 = $V" in {
3924 def VSETLNi8 : NVSetLane<{1,1,1,0,0,1,?,0}, 0b1011, {?,?}, (outs DPR:$V),
3925 (ins DPR:$src1, GPR:$R, nohash_imm:$lane),
3926 IIC_VMOVISL, "vmov", "8", "$V[$lane], $R",
3927 [(set DPR:$V, (vector_insert (v8i8 DPR:$src1),
3928 GPR:$R, imm:$lane))]> {
3929 let Inst{21} = lane{2};
3930 let Inst{6-5} = lane{1-0};
3932 def VSETLNi16 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, {?,1}, (outs DPR:$V),
3933 (ins DPR:$src1, GPR:$R, nohash_imm:$lane),
3934 IIC_VMOVISL, "vmov", "16", "$V[$lane], $R",
3935 [(set DPR:$V, (vector_insert (v4i16 DPR:$src1),
3936 GPR:$R, imm:$lane))]> {
3937 let Inst{21} = lane{1};
3938 let Inst{6} = lane{0};
3940 def VSETLNi32 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, 0b00, (outs DPR:$V),
3941 (ins DPR:$src1, GPR:$R, nohash_imm:$lane),
3942 IIC_VMOVISL, "vmov", "32", "$V[$lane], $R",
3943 [(set DPR:$V, (insertelt (v2i32 DPR:$src1),
3944 GPR:$R, imm:$lane))]> {
3945 let Inst{21} = lane{0};
3948 def : Pat<(vector_insert (v16i8 QPR:$src1), GPR:$src2, imm:$lane),
3949 (v16i8 (INSERT_SUBREG QPR:$src1,
3950 (v8i8 (VSETLNi8 (v8i8 (EXTRACT_SUBREG QPR:$src1,
3951 (DSubReg_i8_reg imm:$lane))),
3952 GPR:$src2, (SubReg_i8_lane imm:$lane))),
3953 (DSubReg_i8_reg imm:$lane)))>;
3954 def : Pat<(vector_insert (v8i16 QPR:$src1), GPR:$src2, imm:$lane),
3955 (v8i16 (INSERT_SUBREG QPR:$src1,
3956 (v4i16 (VSETLNi16 (v4i16 (EXTRACT_SUBREG QPR:$src1,
3957 (DSubReg_i16_reg imm:$lane))),
3958 GPR:$src2, (SubReg_i16_lane imm:$lane))),
3959 (DSubReg_i16_reg imm:$lane)))>;
3960 def : Pat<(insertelt (v4i32 QPR:$src1), GPR:$src2, imm:$lane),
3961 (v4i32 (INSERT_SUBREG QPR:$src1,
3962 (v2i32 (VSETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src1,
3963 (DSubReg_i32_reg imm:$lane))),
3964 GPR:$src2, (SubReg_i32_lane imm:$lane))),
3965 (DSubReg_i32_reg imm:$lane)))>;
3967 def : Pat<(v2f32 (insertelt DPR:$src1, SPR:$src2, imm:$src3)),
3968 (INSERT_SUBREG (v2f32 (COPY_TO_REGCLASS DPR:$src1, DPR_VFP2)),
3969 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
3970 def : Pat<(v4f32 (insertelt QPR:$src1, SPR:$src2, imm:$src3)),
3971 (INSERT_SUBREG (v4f32 (COPY_TO_REGCLASS QPR:$src1, QPR_VFP2)),
3972 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
3974 //def : Pat<(v2i64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
3975 // (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
3976 def : Pat<(v2f64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
3977 (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
3979 def : Pat<(v2f32 (scalar_to_vector SPR:$src)),
3980 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
3981 def : Pat<(v2f64 (scalar_to_vector (f64 DPR:$src))),
3982 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
3983 def : Pat<(v4f32 (scalar_to_vector SPR:$src)),
3984 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
3986 def : Pat<(v8i8 (scalar_to_vector GPR:$src)),
3987 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
3988 def : Pat<(v4i16 (scalar_to_vector GPR:$src)),
3989 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
3990 def : Pat<(v2i32 (scalar_to_vector GPR:$src)),
3991 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
3993 def : Pat<(v16i8 (scalar_to_vector GPR:$src)),
3994 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3995 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
3997 def : Pat<(v8i16 (scalar_to_vector GPR:$src)),
3998 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
3999 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4001 def : Pat<(v4i32 (scalar_to_vector GPR:$src)),
4002 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
4003 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4006 // VDUP : Vector Duplicate (from ARM core register to all elements)
4008 class VDUPD<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
4009 : NVDup<opcod1, 0b1011, opcod3, (outs DPR:$dst), (ins GPR:$src),
4010 IIC_VMOVIS, "vdup", Dt, "$dst, $src",
4011 [(set DPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
4012 class VDUPQ<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
4013 : NVDup<opcod1, 0b1011, opcod3, (outs QPR:$dst), (ins GPR:$src),
4014 IIC_VMOVIS, "vdup", Dt, "$dst, $src",
4015 [(set QPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
4017 def VDUP8d : VDUPD<0b11101100, 0b00, "8", v8i8>;
4018 def VDUP16d : VDUPD<0b11101000, 0b01, "16", v4i16>;
4019 def VDUP32d : VDUPD<0b11101000, 0b00, "32", v2i32>;
4020 def VDUP8q : VDUPQ<0b11101110, 0b00, "8", v16i8>;
4021 def VDUP16q : VDUPQ<0b11101010, 0b01, "16", v8i16>;
4022 def VDUP32q : VDUPQ<0b11101010, 0b00, "32", v4i32>;
4024 def VDUPfd : NVDup<0b11101000, 0b1011, 0b00, (outs DPR:$dst), (ins GPR:$src),
4025 IIC_VMOVIS, "vdup", "32", "$dst, $src",
4026 [(set DPR:$dst, (v2f32 (NEONvdup
4027 (f32 (bitconvert GPR:$src)))))]>;
4028 def VDUPfq : NVDup<0b11101010, 0b1011, 0b00, (outs QPR:$dst), (ins GPR:$src),
4029 IIC_VMOVIS, "vdup", "32", "$dst, $src",
4030 [(set QPR:$dst, (v4f32 (NEONvdup
4031 (f32 (bitconvert GPR:$src)))))]>;
4033 // VDUP : Vector Duplicate Lane (from scalar to all elements)
4035 class VDUPLND<bits<4> op19_16, string OpcodeStr, string Dt,
4037 : NVDupLane<op19_16, 0, (outs DPR:$dst), (ins DPR:$src, nohash_imm:$lane),
4038 IIC_VMOVD, OpcodeStr, Dt, "$dst, $src[$lane]",
4039 [(set DPR:$dst, (Ty (NEONvduplane (Ty DPR:$src), imm:$lane)))]>;
4041 class VDUPLNQ<bits<4> op19_16, string OpcodeStr, string Dt,
4042 ValueType ResTy, ValueType OpTy>
4043 : NVDupLane<op19_16, 1, (outs QPR:$dst), (ins DPR:$src, nohash_imm:$lane),
4044 IIC_VMOVQ, OpcodeStr, Dt, "$dst, $src[$lane]",
4045 [(set QPR:$dst, (ResTy (NEONvduplane (OpTy DPR:$src),
4048 // Inst{19-16} is partially specified depending on the element size.
4050 def VDUPLN8d : VDUPLND<{?,?,?,1}, "vdup", "8", v8i8> {
4051 let Inst{19-17} = lane{2-0};
4053 def VDUPLN16d : VDUPLND<{?,?,1,0}, "vdup", "16", v4i16> {
4054 let Inst{19-18} = lane{1-0};
4056 def VDUPLN32d : VDUPLND<{?,1,0,0}, "vdup", "32", v2i32> {
4057 let Inst{19} = lane{0};
4059 def VDUPLNfd : VDUPLND<{?,1,0,0}, "vdup", "32", v2f32> {
4060 let Inst{19} = lane{0};
4062 def VDUPLN8q : VDUPLNQ<{?,?,?,1}, "vdup", "8", v16i8, v8i8> {
4063 let Inst{19-17} = lane{2-0};
4065 def VDUPLN16q : VDUPLNQ<{?,?,1,0}, "vdup", "16", v8i16, v4i16> {
4066 let Inst{19-18} = lane{1-0};
4068 def VDUPLN32q : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4i32, v2i32> {
4069 let Inst{19} = lane{0};
4071 def VDUPLNfq : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4f32, v2f32> {
4072 let Inst{19} = lane{0};
4075 def : Pat<(v16i8 (NEONvduplane (v16i8 QPR:$src), imm:$lane)),
4076 (v16i8 (VDUPLN8q (v8i8 (EXTRACT_SUBREG QPR:$src,
4077 (DSubReg_i8_reg imm:$lane))),
4078 (SubReg_i8_lane imm:$lane)))>;
4079 def : Pat<(v8i16 (NEONvduplane (v8i16 QPR:$src), imm:$lane)),
4080 (v8i16 (VDUPLN16q (v4i16 (EXTRACT_SUBREG QPR:$src,
4081 (DSubReg_i16_reg imm:$lane))),
4082 (SubReg_i16_lane imm:$lane)))>;
4083 def : Pat<(v4i32 (NEONvduplane (v4i32 QPR:$src), imm:$lane)),
4084 (v4i32 (VDUPLN32q (v2i32 (EXTRACT_SUBREG QPR:$src,
4085 (DSubReg_i32_reg imm:$lane))),
4086 (SubReg_i32_lane imm:$lane)))>;
4087 def : Pat<(v4f32 (NEONvduplane (v4f32 QPR:$src), imm:$lane)),
4088 (v4f32 (VDUPLNfq (v2f32 (EXTRACT_SUBREG QPR:$src,
4089 (DSubReg_i32_reg imm:$lane))),
4090 (SubReg_i32_lane imm:$lane)))>;
4092 def VDUPfdf : PseudoNeonI<(outs DPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
4093 [(set DPR:$dst, (v2f32 (NEONvdup (f32 SPR:$src))))]>;
4094 def VDUPfqf : PseudoNeonI<(outs QPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
4095 [(set QPR:$dst, (v4f32 (NEONvdup (f32 SPR:$src))))]>;
4097 // VMOVN : Vector Narrowing Move
4098 defm VMOVN : N2VN_HSD<0b11,0b11,0b10,0b00100,0,0, IIC_VMOVN,
4099 "vmovn", "i", trunc>;
4100 // VQMOVN : Vector Saturating Narrowing Move
4101 defm VQMOVNs : N2VNInt_HSD<0b11,0b11,0b10,0b00101,0,0, IIC_VQUNAiD,
4102 "vqmovn", "s", int_arm_neon_vqmovns>;
4103 defm VQMOVNu : N2VNInt_HSD<0b11,0b11,0b10,0b00101,1,0, IIC_VQUNAiD,
4104 "vqmovn", "u", int_arm_neon_vqmovnu>;
4105 defm VQMOVNsu : N2VNInt_HSD<0b11,0b11,0b10,0b00100,1,0, IIC_VQUNAiD,
4106 "vqmovun", "s", int_arm_neon_vqmovnsu>;
4107 // VMOVL : Vector Lengthening Move
4108 defm VMOVLs : N2VL_QHS<0b01,0b10100,0,1, "vmovl", "s", sext>;
4109 defm VMOVLu : N2VL_QHS<0b11,0b10100,0,1, "vmovl", "u", zext>;
4111 // Vector Conversions.
4113 // VCVT : Vector Convert Between Floating-Point and Integers
4114 def VCVTf2sd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
4115 v2i32, v2f32, fp_to_sint>;
4116 def VCVTf2ud : N2VD<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
4117 v2i32, v2f32, fp_to_uint>;
4118 def VCVTs2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
4119 v2f32, v2i32, sint_to_fp>;
4120 def VCVTu2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
4121 v2f32, v2i32, uint_to_fp>;
4123 def VCVTf2sq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
4124 v4i32, v4f32, fp_to_sint>;
4125 def VCVTf2uq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
4126 v4i32, v4f32, fp_to_uint>;
4127 def VCVTs2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
4128 v4f32, v4i32, sint_to_fp>;
4129 def VCVTu2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
4130 v4f32, v4i32, uint_to_fp>;
4132 // VCVT : Vector Convert Between Floating-Point and Fixed-Point.
4133 def VCVTf2xsd : N2VCvtD<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
4134 v2i32, v2f32, int_arm_neon_vcvtfp2fxs>;
4135 def VCVTf2xud : N2VCvtD<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
4136 v2i32, v2f32, int_arm_neon_vcvtfp2fxu>;
4137 def VCVTxs2fd : N2VCvtD<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
4138 v2f32, v2i32, int_arm_neon_vcvtfxs2fp>;
4139 def VCVTxu2fd : N2VCvtD<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
4140 v2f32, v2i32, int_arm_neon_vcvtfxu2fp>;
4142 def VCVTf2xsq : N2VCvtQ<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
4143 v4i32, v4f32, int_arm_neon_vcvtfp2fxs>;
4144 def VCVTf2xuq : N2VCvtQ<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
4145 v4i32, v4f32, int_arm_neon_vcvtfp2fxu>;
4146 def VCVTxs2fq : N2VCvtQ<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
4147 v4f32, v4i32, int_arm_neon_vcvtfxs2fp>;
4148 def VCVTxu2fq : N2VCvtQ<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
4149 v4f32, v4i32, int_arm_neon_vcvtfxu2fp>;
4153 // VREV64 : Vector Reverse elements within 64-bit doublewords
4155 class VREV64D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4156 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 0, 0, (outs DPR:$dst),
4157 (ins DPR:$src), IIC_VMOVD,
4158 OpcodeStr, Dt, "$dst, $src", "",
4159 [(set DPR:$dst, (Ty (NEONvrev64 (Ty DPR:$src))))]>;
4160 class VREV64Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4161 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 1, 0, (outs QPR:$dst),
4162 (ins QPR:$src), IIC_VMOVQ,
4163 OpcodeStr, Dt, "$dst, $src", "",
4164 [(set QPR:$dst, (Ty (NEONvrev64 (Ty QPR:$src))))]>;
4166 def VREV64d8 : VREV64D<0b00, "vrev64", "8", v8i8>;
4167 def VREV64d16 : VREV64D<0b01, "vrev64", "16", v4i16>;
4168 def VREV64d32 : VREV64D<0b10, "vrev64", "32", v2i32>;
4169 def VREV64df : VREV64D<0b10, "vrev64", "32", v2f32>;
4171 def VREV64q8 : VREV64Q<0b00, "vrev64", "8", v16i8>;
4172 def VREV64q16 : VREV64Q<0b01, "vrev64", "16", v8i16>;
4173 def VREV64q32 : VREV64Q<0b10, "vrev64", "32", v4i32>;
4174 def VREV64qf : VREV64Q<0b10, "vrev64", "32", v4f32>;
4176 // VREV32 : Vector Reverse elements within 32-bit words
4178 class VREV32D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4179 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 0, 0, (outs DPR:$dst),
4180 (ins DPR:$src), IIC_VMOVD,
4181 OpcodeStr, Dt, "$dst, $src", "",
4182 [(set DPR:$dst, (Ty (NEONvrev32 (Ty DPR:$src))))]>;
4183 class VREV32Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4184 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 1, 0, (outs QPR:$dst),
4185 (ins QPR:$src), IIC_VMOVQ,
4186 OpcodeStr, Dt, "$dst, $src", "",
4187 [(set QPR:$dst, (Ty (NEONvrev32 (Ty QPR:$src))))]>;
4189 def VREV32d8 : VREV32D<0b00, "vrev32", "8", v8i8>;
4190 def VREV32d16 : VREV32D<0b01, "vrev32", "16", v4i16>;
4192 def VREV32q8 : VREV32Q<0b00, "vrev32", "8", v16i8>;
4193 def VREV32q16 : VREV32Q<0b01, "vrev32", "16", v8i16>;
4195 // VREV16 : Vector Reverse elements within 16-bit halfwords
4197 class VREV16D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4198 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 0, 0, (outs DPR:$dst),
4199 (ins DPR:$src), IIC_VMOVD,
4200 OpcodeStr, Dt, "$dst, $src", "",
4201 [(set DPR:$dst, (Ty (NEONvrev16 (Ty DPR:$src))))]>;
4202 class VREV16Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4203 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 1, 0, (outs QPR:$dst),
4204 (ins QPR:$src), IIC_VMOVQ,
4205 OpcodeStr, Dt, "$dst, $src", "",
4206 [(set QPR:$dst, (Ty (NEONvrev16 (Ty QPR:$src))))]>;
4208 def VREV16d8 : VREV16D<0b00, "vrev16", "8", v8i8>;
4209 def VREV16q8 : VREV16Q<0b00, "vrev16", "8", v16i8>;
4211 // Other Vector Shuffles.
4213 // VEXT : Vector Extract
4215 class VEXTd<string OpcodeStr, string Dt, ValueType Ty>
4216 : N3V<0,1,0b11,{?,?,?,?},0,0, (outs DPR:$dst),
4217 (ins DPR:$lhs, DPR:$rhs, i32imm:$index), NVExtFrm,
4218 IIC_VEXTD, OpcodeStr, Dt, "$dst, $lhs, $rhs, $index", "",
4219 [(set DPR:$dst, (Ty (NEONvext (Ty DPR:$lhs),
4220 (Ty DPR:$rhs), imm:$index)))]> {
4222 let Inst{11-8} = index{3-0};
4225 class VEXTq<string OpcodeStr, string Dt, ValueType Ty>
4226 : N3V<0,1,0b11,{?,?,?,?},1,0, (outs QPR:$dst),
4227 (ins QPR:$lhs, QPR:$rhs, i32imm:$index), NVExtFrm,
4228 IIC_VEXTQ, OpcodeStr, Dt, "$dst, $lhs, $rhs, $index", "",
4229 [(set QPR:$dst, (Ty (NEONvext (Ty QPR:$lhs),
4230 (Ty QPR:$rhs), imm:$index)))]> {
4232 let Inst{11-8} = index{3-0};
4235 def VEXTd8 : VEXTd<"vext", "8", v8i8>;
4236 def VEXTd16 : VEXTd<"vext", "16", v4i16>;
4237 def VEXTd32 : VEXTd<"vext", "32", v2i32>;
4238 def VEXTdf : VEXTd<"vext", "32", v2f32>;
4240 def VEXTq8 : VEXTq<"vext", "8", v16i8>;
4241 def VEXTq16 : VEXTq<"vext", "16", v8i16>;
4242 def VEXTq32 : VEXTq<"vext", "32", v4i32>;
4243 def VEXTqf : VEXTq<"vext", "32", v4f32>;
4245 // VTRN : Vector Transpose
4247 def VTRNd8 : N2VDShuffle<0b00, 0b00001, "vtrn", "8">;
4248 def VTRNd16 : N2VDShuffle<0b01, 0b00001, "vtrn", "16">;
4249 def VTRNd32 : N2VDShuffle<0b10, 0b00001, "vtrn", "32">;
4251 def VTRNq8 : N2VQShuffle<0b00, 0b00001, IIC_VPERMQ, "vtrn", "8">;
4252 def VTRNq16 : N2VQShuffle<0b01, 0b00001, IIC_VPERMQ, "vtrn", "16">;
4253 def VTRNq32 : N2VQShuffle<0b10, 0b00001, IIC_VPERMQ, "vtrn", "32">;
4255 // VUZP : Vector Unzip (Deinterleave)
4257 def VUZPd8 : N2VDShuffle<0b00, 0b00010, "vuzp", "8">;
4258 def VUZPd16 : N2VDShuffle<0b01, 0b00010, "vuzp", "16">;
4259 def VUZPd32 : N2VDShuffle<0b10, 0b00010, "vuzp", "32">;
4261 def VUZPq8 : N2VQShuffle<0b00, 0b00010, IIC_VPERMQ3, "vuzp", "8">;
4262 def VUZPq16 : N2VQShuffle<0b01, 0b00010, IIC_VPERMQ3, "vuzp", "16">;
4263 def VUZPq32 : N2VQShuffle<0b10, 0b00010, IIC_VPERMQ3, "vuzp", "32">;
4265 // VZIP : Vector Zip (Interleave)
4267 def VZIPd8 : N2VDShuffle<0b00, 0b00011, "vzip", "8">;
4268 def VZIPd16 : N2VDShuffle<0b01, 0b00011, "vzip", "16">;
4269 def VZIPd32 : N2VDShuffle<0b10, 0b00011, "vzip", "32">;
4271 def VZIPq8 : N2VQShuffle<0b00, 0b00011, IIC_VPERMQ3, "vzip", "8">;
4272 def VZIPq16 : N2VQShuffle<0b01, 0b00011, IIC_VPERMQ3, "vzip", "16">;
4273 def VZIPq32 : N2VQShuffle<0b10, 0b00011, IIC_VPERMQ3, "vzip", "32">;
4275 // Vector Table Lookup and Table Extension.
4277 // VTBL : Vector Table Lookup
4279 : N3V<1,1,0b11,0b1000,0,0, (outs DPR:$Vd),
4280 (ins DPR:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTB1,
4281 "vtbl", "8", "$Vd, \\{$Vn\\}, $Vm", "",
4282 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbl1 DPR:$Vn, DPR:$Vm)))]>;
4283 let hasExtraSrcRegAllocReq = 1 in {
4285 : N3V<1,1,0b11,0b1001,0,0, (outs DPR:$Vd),
4286 (ins DPR:$Vn, DPR:$tbl2, DPR:$Vm), NVTBLFrm, IIC_VTB2,
4287 "vtbl", "8", "$Vd, \\{$Vn, $tbl2\\}, $Vm", "", []>;
4289 : N3V<1,1,0b11,0b1010,0,0, (outs DPR:$Vd),
4290 (ins DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$Vm), NVTBLFrm, IIC_VTB3,
4291 "vtbl", "8", "$Vd, \\{$Vn, $tbl2, $tbl3\\}, $Vm", "", []>;
4293 : N3V<1,1,0b11,0b1011,0,0, (outs DPR:$Vd),
4294 (ins DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$Vm),
4296 "vtbl", "8", "$Vd, \\{$Vn, $tbl2, $tbl3, $tbl4\\}, $Vm", "", []>;
4297 } // hasExtraSrcRegAllocReq = 1
4300 : PseudoNeonI<(outs DPR:$dst), (ins QPR:$tbl, DPR:$src), IIC_VTB2, "", []>;
4302 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB3, "", []>;
4304 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB4, "", []>;
4306 // VTBX : Vector Table Extension
4308 : N3V<1,1,0b11,0b1000,1,0, (outs DPR:$Vd),
4309 (ins DPR:$orig, DPR:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTBX1,
4310 "vtbx", "8", "$Vd, \\{$Vn\\}, $Vm", "$orig = $Vd",
4311 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbx1
4312 DPR:$orig, DPR:$Vn, DPR:$Vm)))]>;
4313 let hasExtraSrcRegAllocReq = 1 in {
4315 : N3V<1,1,0b11,0b1001,1,0, (outs DPR:$Vd),
4316 (ins DPR:$orig, DPR:$Vn, DPR:$tbl2, DPR:$Vm), NVTBLFrm, IIC_VTBX2,
4317 "vtbx", "8", "$Vd, \\{$Vn, $tbl2\\}, $Vm", "$orig = $Vd", []>;
4319 : N3V<1,1,0b11,0b1010,1,0, (outs DPR:$Vd),
4320 (ins DPR:$orig, DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$Vm),
4321 NVTBLFrm, IIC_VTBX3,
4322 "vtbx", "8", "$Vd, \\{$Vn, $tbl2, $tbl3\\}, $Vm",
4325 : N3V<1,1,0b11,0b1011,1,0, (outs DPR:$Vd), (ins DPR:$orig, DPR:$Vn,
4326 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$Vm), NVTBLFrm, IIC_VTBX4,
4327 "vtbx", "8", "$Vd, \\{$Vn, $tbl2, $tbl3, $tbl4\\}, $Vm",
4329 } // hasExtraSrcRegAllocReq = 1
4332 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QPR:$tbl, DPR:$src),
4333 IIC_VTBX2, "$orig = $dst", []>;
4335 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
4336 IIC_VTBX3, "$orig = $dst", []>;
4338 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
4339 IIC_VTBX4, "$orig = $dst", []>;
4341 //===----------------------------------------------------------------------===//
4342 // NEON instructions for single-precision FP math
4343 //===----------------------------------------------------------------------===//
4345 class N2VSPat<SDNode OpNode, ValueType ResTy, ValueType OpTy, NeonI Inst>
4346 : NEONFPPat<(ResTy (OpNode SPR:$a)),
4347 (EXTRACT_SUBREG (OpTy (Inst (INSERT_SUBREG (OpTy (IMPLICIT_DEF)),
4351 class N3VSPat<SDNode OpNode, NeonI Inst>
4352 : NEONFPPat<(f32 (OpNode SPR:$a, SPR:$b)),
4353 (EXTRACT_SUBREG (v2f32
4354 (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
4356 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
4360 class N3VSMulOpPat<SDNode MulNode, SDNode OpNode, NeonI Inst>
4361 : NEONFPPat<(f32 (OpNode SPR:$acc, (f32 (MulNode SPR:$a, SPR:$b)))),
4362 (EXTRACT_SUBREG (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
4364 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
4366 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
4370 // These need separate instructions because they must use DPR_VFP2 register
4371 // class which have SPR sub-registers.
4373 // Vector Add Operations used for single-precision FP
4374 let neverHasSideEffects = 1 in
4375 def VADDfd_sfp : N3VS<0,0,0b00,0b1101,0, "vadd", "f32", v2f32, v2f32, fadd, 1>;
4376 def : N3VSPat<fadd, VADDfd_sfp>;
4378 // Vector Sub Operations used for single-precision FP
4379 let neverHasSideEffects = 1 in
4380 def VSUBfd_sfp : N3VS<0,0,0b10,0b1101,0, "vsub", "f32", v2f32, v2f32, fsub, 0>;
4381 def : N3VSPat<fsub, VSUBfd_sfp>;
4383 // Vector Multiply Operations used for single-precision FP
4384 let neverHasSideEffects = 1 in
4385 def VMULfd_sfp : N3VS<1,0,0b00,0b1101,1, "vmul", "f32", v2f32, v2f32, fmul, 1>;
4386 def : N3VSPat<fmul, VMULfd_sfp>;
4388 // Vector Multiply-Accumulate/Subtract used for single-precision FP
4389 // vml[as].f32 can cause 4-8 cycle stalls in following ASIMD instructions, so
4390 // we want to avoid them for now. e.g., alternating vmla/vadd instructions.
4392 //let neverHasSideEffects = 1 in
4393 //def VMLAfd_sfp : N3VSMulOp<0,0,0b00,0b1101,1, IIC_VMACD, "vmla", "f32",
4394 // v2f32, fmul, fadd>;
4395 //def : N3VSMulOpPat<fmul, fadd, VMLAfd_sfp>;
4397 //let neverHasSideEffects = 1 in
4398 //def VMLSfd_sfp : N3VSMulOp<0,0,0b10,0b1101,1, IIC_VMACD, "vmls", "f32",
4399 // v2f32, fmul, fsub>;
4400 //def : N3VSMulOpPat<fmul, fsub, VMLSfd_sfp>;
4402 // Vector Absolute used for single-precision FP
4403 let neverHasSideEffects = 1 in
4404 def VABSfd_sfp : N2V<0b11, 0b11, 0b10, 0b01, 0b01110, 0, 0,
4405 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), IIC_VUNAD,
4406 "vabs", "f32", "$dst, $src", "", []>;
4407 def : N2VSPat<fabs, f32, v2f32, VABSfd_sfp>;
4409 // Vector Negate used for single-precision FP
4410 let neverHasSideEffects = 1 in
4411 def VNEGfd_sfp : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
4412 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), IIC_VUNAD,
4413 "vneg", "f32", "$dst, $src", "", []>;
4414 def : N2VSPat<fneg, f32, v2f32, VNEGfd_sfp>;
4416 // Vector Maximum used for single-precision FP
4417 let neverHasSideEffects = 1 in
4418 def VMAXfd_sfp : N3V<0, 0, 0b00, 0b1111, 0, 0, (outs DPR_VFP2:$dst),
4419 (ins DPR_VFP2:$src1, DPR_VFP2:$src2), N3RegFrm, IIC_VBIND,
4420 "vmax", "f32", "$dst, $src1, $src2", "", []>;
4421 def : N3VSPat<NEONfmax, VMAXfd_sfp>;
4423 // Vector Minimum used for single-precision FP
4424 let neverHasSideEffects = 1 in
4425 def VMINfd_sfp : N3V<0, 0, 0b00, 0b1111, 0, 0, (outs DPR_VFP2:$dst),
4426 (ins DPR_VFP2:$src1, DPR_VFP2:$src2), N3RegFrm, IIC_VBIND,
4427 "vmin", "f32", "$dst, $src1, $src2", "", []>;
4428 def : N3VSPat<NEONfmin, VMINfd_sfp>;
4430 // Vector Convert between single-precision FP and integer
4431 let neverHasSideEffects = 1 in
4432 def VCVTf2sd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
4433 v2i32, v2f32, fp_to_sint>;
4434 def : N2VSPat<arm_ftosi, f32, v2f32, VCVTf2sd_sfp>;
4436 let neverHasSideEffects = 1 in
4437 def VCVTf2ud_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
4438 v2i32, v2f32, fp_to_uint>;
4439 def : N2VSPat<arm_ftoui, f32, v2f32, VCVTf2ud_sfp>;
4441 let neverHasSideEffects = 1 in
4442 def VCVTs2fd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
4443 v2f32, v2i32, sint_to_fp>;
4444 def : N2VSPat<arm_sitof, f32, v2i32, VCVTs2fd_sfp>;
4446 let neverHasSideEffects = 1 in
4447 def VCVTu2fd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
4448 v2f32, v2i32, uint_to_fp>;
4449 def : N2VSPat<arm_uitof, f32, v2i32, VCVTu2fd_sfp>;
4451 //===----------------------------------------------------------------------===//
4452 // Non-Instruction Patterns
4453 //===----------------------------------------------------------------------===//
4456 def : Pat<(v1i64 (bitconvert (v2i32 DPR:$src))), (v1i64 DPR:$src)>;
4457 def : Pat<(v1i64 (bitconvert (v4i16 DPR:$src))), (v1i64 DPR:$src)>;
4458 def : Pat<(v1i64 (bitconvert (v8i8 DPR:$src))), (v1i64 DPR:$src)>;
4459 def : Pat<(v1i64 (bitconvert (f64 DPR:$src))), (v1i64 DPR:$src)>;
4460 def : Pat<(v1i64 (bitconvert (v2f32 DPR:$src))), (v1i64 DPR:$src)>;
4461 def : Pat<(v2i32 (bitconvert (v1i64 DPR:$src))), (v2i32 DPR:$src)>;
4462 def : Pat<(v2i32 (bitconvert (v4i16 DPR:$src))), (v2i32 DPR:$src)>;
4463 def : Pat<(v2i32 (bitconvert (v8i8 DPR:$src))), (v2i32 DPR:$src)>;
4464 def : Pat<(v2i32 (bitconvert (f64 DPR:$src))), (v2i32 DPR:$src)>;
4465 def : Pat<(v2i32 (bitconvert (v2f32 DPR:$src))), (v2i32 DPR:$src)>;
4466 def : Pat<(v4i16 (bitconvert (v1i64 DPR:$src))), (v4i16 DPR:$src)>;
4467 def : Pat<(v4i16 (bitconvert (v2i32 DPR:$src))), (v4i16 DPR:$src)>;
4468 def : Pat<(v4i16 (bitconvert (v8i8 DPR:$src))), (v4i16 DPR:$src)>;
4469 def : Pat<(v4i16 (bitconvert (f64 DPR:$src))), (v4i16 DPR:$src)>;
4470 def : Pat<(v4i16 (bitconvert (v2f32 DPR:$src))), (v4i16 DPR:$src)>;
4471 def : Pat<(v8i8 (bitconvert (v1i64 DPR:$src))), (v8i8 DPR:$src)>;
4472 def : Pat<(v8i8 (bitconvert (v2i32 DPR:$src))), (v8i8 DPR:$src)>;
4473 def : Pat<(v8i8 (bitconvert (v4i16 DPR:$src))), (v8i8 DPR:$src)>;
4474 def : Pat<(v8i8 (bitconvert (f64 DPR:$src))), (v8i8 DPR:$src)>;
4475 def : Pat<(v8i8 (bitconvert (v2f32 DPR:$src))), (v8i8 DPR:$src)>;
4476 def : Pat<(f64 (bitconvert (v1i64 DPR:$src))), (f64 DPR:$src)>;
4477 def : Pat<(f64 (bitconvert (v2i32 DPR:$src))), (f64 DPR:$src)>;
4478 def : Pat<(f64 (bitconvert (v4i16 DPR:$src))), (f64 DPR:$src)>;
4479 def : Pat<(f64 (bitconvert (v8i8 DPR:$src))), (f64 DPR:$src)>;
4480 def : Pat<(f64 (bitconvert (v2f32 DPR:$src))), (f64 DPR:$src)>;
4481 def : Pat<(v2f32 (bitconvert (f64 DPR:$src))), (v2f32 DPR:$src)>;
4482 def : Pat<(v2f32 (bitconvert (v1i64 DPR:$src))), (v2f32 DPR:$src)>;
4483 def : Pat<(v2f32 (bitconvert (v2i32 DPR:$src))), (v2f32 DPR:$src)>;
4484 def : Pat<(v2f32 (bitconvert (v4i16 DPR:$src))), (v2f32 DPR:$src)>;
4485 def : Pat<(v2f32 (bitconvert (v8i8 DPR:$src))), (v2f32 DPR:$src)>;
4487 def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (v2i64 QPR:$src)>;
4488 def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (v2i64 QPR:$src)>;
4489 def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (v2i64 QPR:$src)>;
4490 def : Pat<(v2i64 (bitconvert (v2f64 QPR:$src))), (v2i64 QPR:$src)>;
4491 def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (v2i64 QPR:$src)>;
4492 def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (v4i32 QPR:$src)>;
4493 def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (v4i32 QPR:$src)>;
4494 def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (v4i32 QPR:$src)>;
4495 def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (v4i32 QPR:$src)>;
4496 def : Pat<(v4i32 (bitconvert (v4f32 QPR:$src))), (v4i32 QPR:$src)>;
4497 def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (v8i16 QPR:$src)>;
4498 def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (v8i16 QPR:$src)>;
4499 def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (v8i16 QPR:$src)>;
4500 def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (v8i16 QPR:$src)>;
4501 def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (v8i16 QPR:$src)>;
4502 def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (v16i8 QPR:$src)>;
4503 def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (v16i8 QPR:$src)>;
4504 def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (v16i8 QPR:$src)>;
4505 def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (v16i8 QPR:$src)>;
4506 def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (v16i8 QPR:$src)>;
4507 def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (v4f32 QPR:$src)>;
4508 def : Pat<(v4f32 (bitconvert (v4i32 QPR:$src))), (v4f32 QPR:$src)>;
4509 def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (v4f32 QPR:$src)>;
4510 def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (v4f32 QPR:$src)>;
4511 def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (v4f32 QPR:$src)>;
4512 def : Pat<(v2f64 (bitconvert (v2i64 QPR:$src))), (v2f64 QPR:$src)>;
4513 def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (v2f64 QPR:$src)>;
4514 def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 QPR:$src)>;
4515 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 QPR:$src)>;
4516 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 QPR:$src)>;