1 //===- ARMInstrNEON.td - NEON support for ARM -----------------------------===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file describes the ARM NEON instruction set.
12 //===----------------------------------------------------------------------===//
14 //===----------------------------------------------------------------------===//
15 // NEON-specific DAG Nodes.
16 //===----------------------------------------------------------------------===//
18 def SDTARMVCMP : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisSameAs<1, 2>]>;
19 def SDTARMVCMPZ : SDTypeProfile<1, 1, []>;
21 def NEONvceq : SDNode<"ARMISD::VCEQ", SDTARMVCMP>;
22 def NEONvceqz : SDNode<"ARMISD::VCEQZ", SDTARMVCMPZ>;
23 def NEONvcge : SDNode<"ARMISD::VCGE", SDTARMVCMP>;
24 def NEONvcgez : SDNode<"ARMISD::VCGEZ", SDTARMVCMPZ>;
25 def NEONvclez : SDNode<"ARMISD::VCLEZ", SDTARMVCMPZ>;
26 def NEONvcgeu : SDNode<"ARMISD::VCGEU", SDTARMVCMP>;
27 def NEONvcgt : SDNode<"ARMISD::VCGT", SDTARMVCMP>;
28 def NEONvcgtz : SDNode<"ARMISD::VCGTZ", SDTARMVCMPZ>;
29 def NEONvcltz : SDNode<"ARMISD::VCLTZ", SDTARMVCMPZ>;
30 def NEONvcgtu : SDNode<"ARMISD::VCGTU", SDTARMVCMP>;
31 def NEONvtst : SDNode<"ARMISD::VTST", SDTARMVCMP>;
33 // Types for vector shift by immediates. The "SHX" version is for long and
34 // narrow operations where the source and destination vectors have different
35 // types. The "SHINS" version is for shift and insert operations.
36 def SDTARMVSH : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
38 def SDTARMVSHX : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
40 def SDTARMVSHINS : SDTypeProfile<1, 3, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
41 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
43 def NEONvshl : SDNode<"ARMISD::VSHL", SDTARMVSH>;
44 def NEONvshrs : SDNode<"ARMISD::VSHRs", SDTARMVSH>;
45 def NEONvshru : SDNode<"ARMISD::VSHRu", SDTARMVSH>;
46 def NEONvshlls : SDNode<"ARMISD::VSHLLs", SDTARMVSHX>;
47 def NEONvshllu : SDNode<"ARMISD::VSHLLu", SDTARMVSHX>;
48 def NEONvshlli : SDNode<"ARMISD::VSHLLi", SDTARMVSHX>;
49 def NEONvshrn : SDNode<"ARMISD::VSHRN", SDTARMVSHX>;
51 def NEONvrshrs : SDNode<"ARMISD::VRSHRs", SDTARMVSH>;
52 def NEONvrshru : SDNode<"ARMISD::VRSHRu", SDTARMVSH>;
53 def NEONvrshrn : SDNode<"ARMISD::VRSHRN", SDTARMVSHX>;
55 def NEONvqshls : SDNode<"ARMISD::VQSHLs", SDTARMVSH>;
56 def NEONvqshlu : SDNode<"ARMISD::VQSHLu", SDTARMVSH>;
57 def NEONvqshlsu : SDNode<"ARMISD::VQSHLsu", SDTARMVSH>;
58 def NEONvqshrns : SDNode<"ARMISD::VQSHRNs", SDTARMVSHX>;
59 def NEONvqshrnu : SDNode<"ARMISD::VQSHRNu", SDTARMVSHX>;
60 def NEONvqshrnsu : SDNode<"ARMISD::VQSHRNsu", SDTARMVSHX>;
62 def NEONvqrshrns : SDNode<"ARMISD::VQRSHRNs", SDTARMVSHX>;
63 def NEONvqrshrnu : SDNode<"ARMISD::VQRSHRNu", SDTARMVSHX>;
64 def NEONvqrshrnsu : SDNode<"ARMISD::VQRSHRNsu", SDTARMVSHX>;
66 def NEONvsli : SDNode<"ARMISD::VSLI", SDTARMVSHINS>;
67 def NEONvsri : SDNode<"ARMISD::VSRI", SDTARMVSHINS>;
69 def SDTARMVGETLN : SDTypeProfile<1, 2, [SDTCisVT<0, i32>, SDTCisInt<1>,
71 def NEONvgetlaneu : SDNode<"ARMISD::VGETLANEu", SDTARMVGETLN>;
72 def NEONvgetlanes : SDNode<"ARMISD::VGETLANEs", SDTARMVGETLN>;
74 def SDTARMVMOVIMM : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisVT<1, i32>]>;
75 def NEONvmovImm : SDNode<"ARMISD::VMOVIMM", SDTARMVMOVIMM>;
76 def NEONvmvnImm : SDNode<"ARMISD::VMVNIMM", SDTARMVMOVIMM>;
78 def SDTARMVORRIMM : SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
80 def NEONvorrImm : SDNode<"ARMISD::VORRIMM", SDTARMVORRIMM>;
81 def NEONvbicImm : SDNode<"ARMISD::VBICIMM", SDTARMVORRIMM>;
83 def NEONvdup : SDNode<"ARMISD::VDUP", SDTypeProfile<1, 1, [SDTCisVec<0>]>>;
85 // VDUPLANE can produce a quad-register result from a double-register source,
86 // so the result is not constrained to match the source.
87 def NEONvduplane : SDNode<"ARMISD::VDUPLANE",
88 SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisVec<1>,
91 def SDTARMVEXT : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
92 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
93 def NEONvext : SDNode<"ARMISD::VEXT", SDTARMVEXT>;
95 def SDTARMVSHUF : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisSameAs<0, 1>]>;
96 def NEONvrev64 : SDNode<"ARMISD::VREV64", SDTARMVSHUF>;
97 def NEONvrev32 : SDNode<"ARMISD::VREV32", SDTARMVSHUF>;
98 def NEONvrev16 : SDNode<"ARMISD::VREV16", SDTARMVSHUF>;
100 def SDTARMVSHUF2 : SDTypeProfile<2, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
102 SDTCisSameAs<0, 3>]>;
103 def NEONzip : SDNode<"ARMISD::VZIP", SDTARMVSHUF2>;
104 def NEONuzp : SDNode<"ARMISD::VUZP", SDTARMVSHUF2>;
105 def NEONtrn : SDNode<"ARMISD::VTRN", SDTARMVSHUF2>;
107 def SDTARMVMULL : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
108 SDTCisSameAs<1, 2>]>;
109 def NEONvmulls : SDNode<"ARMISD::VMULLs", SDTARMVMULL>;
110 def NEONvmullu : SDNode<"ARMISD::VMULLu", SDTARMVMULL>;
112 def SDTARMFMAX : SDTypeProfile<1, 2, [SDTCisVT<0, f32>, SDTCisSameAs<0, 1>,
113 SDTCisSameAs<0, 2>]>;
114 def NEONfmax : SDNode<"ARMISD::FMAX", SDTARMFMAX>;
115 def NEONfmin : SDNode<"ARMISD::FMIN", SDTARMFMAX>;
117 def NEONimmAllZerosV: PatLeaf<(NEONvmovImm (i32 timm)), [{
118 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
119 unsigned EltBits = 0;
120 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
121 return (EltBits == 32 && EltVal == 0);
124 def NEONimmAllOnesV: PatLeaf<(NEONvmovImm (i32 timm)), [{
125 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
126 unsigned EltBits = 0;
127 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
128 return (EltBits == 8 && EltVal == 0xff);
131 //===----------------------------------------------------------------------===//
132 // NEON operand definitions
133 //===----------------------------------------------------------------------===//
135 def nModImm : Operand<i32> {
136 let PrintMethod = "printNEONModImmOperand";
139 //===----------------------------------------------------------------------===//
140 // NEON load / store instructions
141 //===----------------------------------------------------------------------===//
143 // Use VLDM to load a Q register as a D register pair.
144 // This is a pseudo instruction that is expanded to VLDMD after reg alloc.
146 : PseudoVFPLdStM<(outs QPR:$dst), (ins GPR:$Rn),
148 [(set QPR:$dst, (v2f64 (load GPR:$Rn)))]>;
150 : PseudoVFPLdStM<(outs QPR:$dst), (ins GPR:$Rn),
152 [(set QPR:$dst, (v2f64 (load GPR:$Rn)))]>;
154 // Use VSTM to store a Q register as a D register pair.
155 // This is a pseudo instruction that is expanded to VSTMD after reg alloc.
157 : PseudoVFPLdStM<(outs), (ins QPR:$src, GPR:$Rn),
159 [(store (v2f64 QPR:$src), GPR:$Rn)]>;
161 : PseudoVFPLdStM<(outs), (ins QPR:$src, GPR:$Rn),
163 [(store (v2f64 QPR:$src), GPR:$Rn)]>;
165 // Classes for VLD* pseudo-instructions with multi-register operands.
166 // These are expanded to real instructions after register allocation.
167 class VLDQPseudo<InstrItinClass itin>
168 : PseudoNLdSt<(outs QPR:$dst), (ins addrmode6:$addr), itin, "">;
169 class VLDQWBPseudo<InstrItinClass itin>
170 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
171 (ins addrmode6:$addr, am6offset:$offset), itin,
173 class VLDQQPseudo<InstrItinClass itin>
174 : PseudoNLdSt<(outs QQPR:$dst), (ins addrmode6:$addr), itin, "">;
175 class VLDQQWBPseudo<InstrItinClass itin>
176 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
177 (ins addrmode6:$addr, am6offset:$offset), itin,
179 class VLDQQQQWBPseudo<InstrItinClass itin>
180 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
181 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
182 "$addr.addr = $wb, $src = $dst">;
184 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
186 // VLD1 : Vector Load (multiple single elements)
187 class VLD1D<bits<4> op7_4, string Dt>
188 : NLdSt<0,0b10,0b0111,op7_4, (outs DPR:$Vd),
189 (ins addrmode6:$Rn), IIC_VLD1,
190 "vld1", Dt, "\\{$Vd\\}, $Rn", "", []> {
194 class VLD1Q<bits<4> op7_4, string Dt>
195 : NLdSt<0,0b10,0b1010,op7_4, (outs DPR:$Vd, DPR:$dst2),
196 (ins addrmode6:$Rn), IIC_VLD1x2,
197 "vld1", Dt, "\\{$Vd, $dst2\\}, $Rn", "", []> {
199 let Inst{5-4} = Rn{5-4};
202 def VLD1d8 : VLD1D<{0,0,0,?}, "8">;
203 def VLD1d16 : VLD1D<{0,1,0,?}, "16">;
204 def VLD1d32 : VLD1D<{1,0,0,?}, "32">;
205 def VLD1d64 : VLD1D<{1,1,0,?}, "64">;
207 def VLD1q8 : VLD1Q<{0,0,?,?}, "8">;
208 def VLD1q16 : VLD1Q<{0,1,?,?}, "16">;
209 def VLD1q32 : VLD1Q<{1,0,?,?}, "32">;
210 def VLD1q64 : VLD1Q<{1,1,?,?}, "64">;
212 def VLD1q8Pseudo : VLDQPseudo<IIC_VLD1x2>;
213 def VLD1q16Pseudo : VLDQPseudo<IIC_VLD1x2>;
214 def VLD1q32Pseudo : VLDQPseudo<IIC_VLD1x2>;
215 def VLD1q64Pseudo : VLDQPseudo<IIC_VLD1x2>;
217 // ...with address register writeback:
218 class VLD1DWB<bits<4> op7_4, string Dt>
219 : NLdSt<0,0b10,0b0111,op7_4, (outs DPR:$Vd, GPR:$wb),
220 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD1u,
221 "vld1", Dt, "\\{$Vd\\}, $Rn$Rm",
222 "$Rn.addr = $wb", []> {
225 class VLD1QWB<bits<4> op7_4, string Dt>
226 : NLdSt<0,0b10,0b1010,op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
227 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD1x2u,
228 "vld1", Dt, "\\{$Vd, $dst2\\}, $Rn$Rm",
229 "$Rn.addr = $wb", []> {
230 let Inst{5-4} = Rn{5-4};
233 def VLD1d8_UPD : VLD1DWB<{0,0,0,?}, "8">;
234 def VLD1d16_UPD : VLD1DWB<{0,1,0,?}, "16">;
235 def VLD1d32_UPD : VLD1DWB<{1,0,0,?}, "32">;
236 def VLD1d64_UPD : VLD1DWB<{1,1,0,?}, "64">;
238 def VLD1q8_UPD : VLD1QWB<{0,0,?,?}, "8">;
239 def VLD1q16_UPD : VLD1QWB<{0,1,?,?}, "16">;
240 def VLD1q32_UPD : VLD1QWB<{1,0,?,?}, "32">;
241 def VLD1q64_UPD : VLD1QWB<{1,1,?,?}, "64">;
243 def VLD1q8Pseudo_UPD : VLDQWBPseudo<IIC_VLD1x2u>;
244 def VLD1q16Pseudo_UPD : VLDQWBPseudo<IIC_VLD1x2u>;
245 def VLD1q32Pseudo_UPD : VLDQWBPseudo<IIC_VLD1x2u>;
246 def VLD1q64Pseudo_UPD : VLDQWBPseudo<IIC_VLD1x2u>;
248 // ...with 3 registers (some of these are only for the disassembler):
249 class VLD1D3<bits<4> op7_4, string Dt>
250 : NLdSt<0,0b10,0b0110,op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
251 (ins addrmode6:$Rn), IIC_VLD1x3, "vld1", Dt,
252 "\\{$Vd, $dst2, $dst3\\}, $Rn", "", []> {
256 class VLD1D3WB<bits<4> op7_4, string Dt>
257 : NLdSt<0,0b10,0b0110,op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
258 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD1x3u, "vld1", Dt,
259 "\\{$Vd, $dst2, $dst3\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
263 def VLD1d8T : VLD1D3<{0,0,0,?}, "8">;
264 def VLD1d16T : VLD1D3<{0,1,0,?}, "16">;
265 def VLD1d32T : VLD1D3<{1,0,0,?}, "32">;
266 def VLD1d64T : VLD1D3<{1,1,0,?}, "64">;
268 def VLD1d8T_UPD : VLD1D3WB<{0,0,0,?}, "8">;
269 def VLD1d16T_UPD : VLD1D3WB<{0,1,0,?}, "16">;
270 def VLD1d32T_UPD : VLD1D3WB<{1,0,0,?}, "32">;
271 def VLD1d64T_UPD : VLD1D3WB<{1,1,0,?}, "64">;
273 def VLD1d64TPseudo : VLDQQPseudo<IIC_VLD1x3>;
274 def VLD1d64TPseudo_UPD : VLDQQWBPseudo<IIC_VLD1x3u>;
276 // ...with 4 registers (some of these are only for the disassembler):
277 class VLD1D4<bits<4> op7_4, string Dt>
278 : NLdSt<0,0b10,0b0010,op7_4,(outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
279 (ins addrmode6:$Rn), IIC_VLD1x4, "vld1", Dt,
280 "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn", "", []> {
282 let Inst{5-4} = Rn{5-4};
284 class VLD1D4WB<bits<4> op7_4, string Dt>
285 : NLdSt<0,0b10,0b0010,op7_4,
286 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
287 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD4, "vld1", Dt,
288 "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn$Rm", "$Rn.addr = $wb",
290 let Inst{5-4} = Rn{5-4};
293 def VLD1d8Q : VLD1D4<{0,0,?,?}, "8">;
294 def VLD1d16Q : VLD1D4<{0,1,?,?}, "16">;
295 def VLD1d32Q : VLD1D4<{1,0,?,?}, "32">;
296 def VLD1d64Q : VLD1D4<{1,1,?,?}, "64">;
298 def VLD1d8Q_UPD : VLD1D4WB<{0,0,?,?}, "8">;
299 def VLD1d16Q_UPD : VLD1D4WB<{0,1,?,?}, "16">;
300 def VLD1d32Q_UPD : VLD1D4WB<{1,0,?,?}, "32">;
301 def VLD1d64Q_UPD : VLD1D4WB<{1,1,?,?}, "64">;
303 def VLD1d64QPseudo : VLDQQPseudo<IIC_VLD1x4>;
304 def VLD1d64QPseudo_UPD : VLDQQWBPseudo<IIC_VLD1x4u>;
306 // VLD2 : Vector Load (multiple 2-element structures)
307 class VLD2D<bits<4> op11_8, bits<4> op7_4, string Dt>
308 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2),
309 (ins addrmode6:$Rn), IIC_VLD2,
310 "vld2", Dt, "\\{$Vd, $dst2\\}, $Rn", "", []> {
312 let Inst{5-4} = Rn{5-4};
314 class VLD2Q<bits<4> op7_4, string Dt>
315 : NLdSt<0, 0b10, 0b0011, op7_4,
316 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
317 (ins addrmode6:$Rn), IIC_VLD2x2,
318 "vld2", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn", "", []> {
320 let Inst{5-4} = Rn{5-4};
323 def VLD2d8 : VLD2D<0b1000, {0,0,?,?}, "8">;
324 def VLD2d16 : VLD2D<0b1000, {0,1,?,?}, "16">;
325 def VLD2d32 : VLD2D<0b1000, {1,0,?,?}, "32">;
327 def VLD2q8 : VLD2Q<{0,0,?,?}, "8">;
328 def VLD2q16 : VLD2Q<{0,1,?,?}, "16">;
329 def VLD2q32 : VLD2Q<{1,0,?,?}, "32">;
331 def VLD2d8Pseudo : VLDQPseudo<IIC_VLD2>;
332 def VLD2d16Pseudo : VLDQPseudo<IIC_VLD2>;
333 def VLD2d32Pseudo : VLDQPseudo<IIC_VLD2>;
335 def VLD2q8Pseudo : VLDQQPseudo<IIC_VLD2x2>;
336 def VLD2q16Pseudo : VLDQQPseudo<IIC_VLD2x2>;
337 def VLD2q32Pseudo : VLDQQPseudo<IIC_VLD2x2>;
339 // ...with address register writeback:
340 class VLD2DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
341 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
342 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD2u,
343 "vld2", Dt, "\\{$Vd, $dst2\\}, $Rn$Rm",
344 "$Rn.addr = $wb", []> {
345 let Inst{5-4} = Rn{5-4};
347 class VLD2QWB<bits<4> op7_4, string Dt>
348 : NLdSt<0, 0b10, 0b0011, op7_4,
349 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
350 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD2x2u,
351 "vld2", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn$Rm",
352 "$Rn.addr = $wb", []> {
353 let Inst{5-4} = Rn{5-4};
356 def VLD2d8_UPD : VLD2DWB<0b1000, {0,0,?,?}, "8">;
357 def VLD2d16_UPD : VLD2DWB<0b1000, {0,1,?,?}, "16">;
358 def VLD2d32_UPD : VLD2DWB<0b1000, {1,0,?,?}, "32">;
360 def VLD2q8_UPD : VLD2QWB<{0,0,?,?}, "8">;
361 def VLD2q16_UPD : VLD2QWB<{0,1,?,?}, "16">;
362 def VLD2q32_UPD : VLD2QWB<{1,0,?,?}, "32">;
364 def VLD2d8Pseudo_UPD : VLDQWBPseudo<IIC_VLD2u>;
365 def VLD2d16Pseudo_UPD : VLDQWBPseudo<IIC_VLD2u>;
366 def VLD2d32Pseudo_UPD : VLDQWBPseudo<IIC_VLD2u>;
368 def VLD2q8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD2x2u>;
369 def VLD2q16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD2x2u>;
370 def VLD2q32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD2x2u>;
372 // ...with double-spaced registers (for disassembly only):
373 def VLD2b8 : VLD2D<0b1001, {0,0,?,?}, "8">;
374 def VLD2b16 : VLD2D<0b1001, {0,1,?,?}, "16">;
375 def VLD2b32 : VLD2D<0b1001, {1,0,?,?}, "32">;
376 def VLD2b8_UPD : VLD2DWB<0b1001, {0,0,?,?}, "8">;
377 def VLD2b16_UPD : VLD2DWB<0b1001, {0,1,?,?}, "16">;
378 def VLD2b32_UPD : VLD2DWB<0b1001, {1,0,?,?}, "32">;
380 // VLD3 : Vector Load (multiple 3-element structures)
381 class VLD3D<bits<4> op11_8, bits<4> op7_4, string Dt>
382 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
383 (ins addrmode6:$Rn), IIC_VLD3,
384 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn", "", []> {
389 def VLD3d8 : VLD3D<0b0100, {0,0,0,?}, "8">;
390 def VLD3d16 : VLD3D<0b0100, {0,1,0,?}, "16">;
391 def VLD3d32 : VLD3D<0b0100, {1,0,0,?}, "32">;
393 def VLD3d8Pseudo : VLDQQPseudo<IIC_VLD3>;
394 def VLD3d16Pseudo : VLDQQPseudo<IIC_VLD3>;
395 def VLD3d32Pseudo : VLDQQPseudo<IIC_VLD3>;
397 // ...with address register writeback:
398 class VLD3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
399 : NLdSt<0, 0b10, op11_8, op7_4,
400 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
401 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD3u,
402 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn$Rm",
403 "$Rn.addr = $wb", []> {
407 def VLD3d8_UPD : VLD3DWB<0b0100, {0,0,0,?}, "8">;
408 def VLD3d16_UPD : VLD3DWB<0b0100, {0,1,0,?}, "16">;
409 def VLD3d32_UPD : VLD3DWB<0b0100, {1,0,0,?}, "32">;
411 def VLD3d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
412 def VLD3d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
413 def VLD3d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
415 // ...with double-spaced registers (non-updating versions for disassembly only):
416 def VLD3q8 : VLD3D<0b0101, {0,0,0,?}, "8">;
417 def VLD3q16 : VLD3D<0b0101, {0,1,0,?}, "16">;
418 def VLD3q32 : VLD3D<0b0101, {1,0,0,?}, "32">;
419 def VLD3q8_UPD : VLD3DWB<0b0101, {0,0,0,?}, "8">;
420 def VLD3q16_UPD : VLD3DWB<0b0101, {0,1,0,?}, "16">;
421 def VLD3q32_UPD : VLD3DWB<0b0101, {1,0,0,?}, "32">;
423 def VLD3q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
424 def VLD3q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
425 def VLD3q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
427 // ...alternate versions to be allocated odd register numbers:
428 def VLD3q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
429 def VLD3q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
430 def VLD3q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
432 // VLD4 : Vector Load (multiple 4-element structures)
433 class VLD4D<bits<4> op11_8, bits<4> op7_4, string Dt>
434 : NLdSt<0, 0b10, op11_8, op7_4,
435 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
436 (ins addrmode6:$Rn), IIC_VLD4,
437 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn", "", []> {
439 let Inst{5-4} = Rn{5-4};
442 def VLD4d8 : VLD4D<0b0000, {0,0,?,?}, "8">;
443 def VLD4d16 : VLD4D<0b0000, {0,1,?,?}, "16">;
444 def VLD4d32 : VLD4D<0b0000, {1,0,?,?}, "32">;
446 def VLD4d8Pseudo : VLDQQPseudo<IIC_VLD4>;
447 def VLD4d16Pseudo : VLDQQPseudo<IIC_VLD4>;
448 def VLD4d32Pseudo : VLDQQPseudo<IIC_VLD4>;
450 // ...with address register writeback:
451 class VLD4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
452 : NLdSt<0, 0b10, op11_8, op7_4,
453 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
454 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD4,
455 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn$Rm",
456 "$Rn.addr = $wb", []> {
457 let Inst{5-4} = Rn{5-4};
460 def VLD4d8_UPD : VLD4DWB<0b0000, {0,0,?,?}, "8">;
461 def VLD4d16_UPD : VLD4DWB<0b0000, {0,1,?,?}, "16">;
462 def VLD4d32_UPD : VLD4DWB<0b0000, {1,0,?,?}, "32">;
464 def VLD4d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4>;
465 def VLD4d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4>;
466 def VLD4d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4>;
468 // ...with double-spaced registers (non-updating versions for disassembly only):
469 def VLD4q8 : VLD4D<0b0001, {0,0,?,?}, "8">;
470 def VLD4q16 : VLD4D<0b0001, {0,1,?,?}, "16">;
471 def VLD4q32 : VLD4D<0b0001, {1,0,?,?}, "32">;
472 def VLD4q8_UPD : VLD4DWB<0b0001, {0,0,?,?}, "8">;
473 def VLD4q16_UPD : VLD4DWB<0b0001, {0,1,?,?}, "16">;
474 def VLD4q32_UPD : VLD4DWB<0b0001, {1,0,?,?}, "32">;
476 def VLD4q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4>;
477 def VLD4q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4>;
478 def VLD4q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4>;
480 // ...alternate versions to be allocated odd register numbers:
481 def VLD4q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4>;
482 def VLD4q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4>;
483 def VLD4q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4>;
485 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
487 // Classes for VLD*LN pseudo-instructions with multi-register operands.
488 // These are expanded to real instructions after register allocation.
489 class VLDQLNPseudo<InstrItinClass itin>
490 : PseudoNLdSt<(outs QPR:$dst),
491 (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
492 itin, "$src = $dst">;
493 class VLDQLNWBPseudo<InstrItinClass itin>
494 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
495 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
496 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
497 class VLDQQLNPseudo<InstrItinClass itin>
498 : PseudoNLdSt<(outs QQPR:$dst),
499 (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
500 itin, "$src = $dst">;
501 class VLDQQLNWBPseudo<InstrItinClass itin>
502 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
503 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
504 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
505 class VLDQQQQLNPseudo<InstrItinClass itin>
506 : PseudoNLdSt<(outs QQQQPR:$dst),
507 (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
508 itin, "$src = $dst">;
509 class VLDQQQQLNWBPseudo<InstrItinClass itin>
510 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
511 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
512 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
514 // VLD1LN : Vector Load (single element to one lane)
515 class VLD1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
517 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd),
518 (ins addrmode6:$Rn, DPR:$src, nohash_imm:$lane),
519 IIC_VLD1ln, "vld1", Dt, "\\{$Vd[$lane]\\}, $Rn",
521 [(set DPR:$Vd, (vector_insert (Ty DPR:$src),
522 (i32 (LoadOp addrmode6:$Rn)),
526 class VLD1QLNPseudo<ValueType Ty, PatFrag LoadOp> : VLDQLNPseudo<IIC_VLD1ln> {
527 let Pattern = [(set QPR:$dst, (vector_insert (Ty QPR:$src),
528 (i32 (LoadOp addrmode6:$addr)),
532 def VLD1LNd8 : VLD1LN<0b0000, {?,?,?,0}, "8", v8i8, extloadi8> {
533 let Inst{7-5} = lane{2-0};
535 def VLD1LNd16 : VLD1LN<0b0100, {?,?,0,?}, "16", v4i16, extloadi16> {
536 let Inst{7-6} = lane{1-0};
539 def VLD1LNd32 : VLD1LN<0b1000, {?,0,?,?}, "32", v2i32, load> {
540 let Inst{7} = lane{0};
545 def VLD1LNq8Pseudo : VLD1QLNPseudo<v16i8, extloadi8>;
546 def VLD1LNq16Pseudo : VLD1QLNPseudo<v8i16, extloadi16>;
547 def VLD1LNq32Pseudo : VLD1QLNPseudo<v4i32, load>;
549 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
551 // ...with address register writeback:
552 class VLD1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
553 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, GPR:$wb),
554 (ins addrmode6:$Rn, am6offset:$Rm,
555 DPR:$src, nohash_imm:$lane), IIC_VLD1lnu, "vld1", Dt,
556 "\\{$Vd[$lane]\\}, $Rn$Rm",
557 "$src = $Vd, $Rn.addr = $wb", []>;
559 def VLD1LNd8_UPD : VLD1LNWB<0b0000, {?,?,?,0}, "8"> {
560 let Inst{7-5} = lane{2-0};
562 def VLD1LNd16_UPD : VLD1LNWB<0b0100, {?,?,0,?}, "16"> {
563 let Inst{7-6} = lane{1-0};
566 def VLD1LNd32_UPD : VLD1LNWB<0b1000, {?,0,?,?}, "32"> {
567 let Inst{7} = lane{0};
572 def VLD1LNq8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
573 def VLD1LNq16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
574 def VLD1LNq32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
576 // VLD2LN : Vector Load (single 2-element structure to one lane)
577 class VLD2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
578 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2),
579 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, nohash_imm:$lane),
580 IIC_VLD2ln, "vld2", Dt, "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn",
581 "$src1 = $Vd, $src2 = $dst2", []> {
586 def VLD2LNd8 : VLD2LN<0b0001, {?,?,?,?}, "8"> {
587 let Inst{7-5} = lane{2-0};
589 def VLD2LNd16 : VLD2LN<0b0101, {?,?,0,?}, "16"> {
590 let Inst{7-6} = lane{1-0};
592 def VLD2LNd32 : VLD2LN<0b1001, {?,0,0,?}, "32"> {
593 let Inst{7} = lane{0};
596 def VLD2LNd8Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
597 def VLD2LNd16Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
598 def VLD2LNd32Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
600 // ...with double-spaced registers:
601 def VLD2LNq16 : VLD2LN<0b0101, {?,?,1,?}, "16"> {
602 let Inst{7-6} = lane{1-0};
604 def VLD2LNq32 : VLD2LN<0b1001, {?,1,0,?}, "32"> {
605 let Inst{7} = lane{0};
608 def VLD2LNq16Pseudo : VLDQQLNPseudo<IIC_VLD2ln>;
609 def VLD2LNq32Pseudo : VLDQQLNPseudo<IIC_VLD2ln>;
611 // ...with address register writeback:
612 class VLD2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
613 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
614 (ins addrmode6:$Rn, am6offset:$Rm,
615 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VLD2lnu, "vld2", Dt,
616 "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn$Rm",
617 "$src1 = $Vd, $src2 = $dst2, $Rn.addr = $wb", []> {
621 def VLD2LNd8_UPD : VLD2LNWB<0b0001, {?,?,?,?}, "8"> {
622 let Inst{7-5} = lane{2-0};
624 def VLD2LNd16_UPD : VLD2LNWB<0b0101, {?,?,0,?}, "16"> {
625 let Inst{7-6} = lane{1-0};
627 def VLD2LNd32_UPD : VLD2LNWB<0b1001, {?,0,0,?}, "32"> {
628 let Inst{7} = lane{0};
631 def VLD2LNd8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
632 def VLD2LNd16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
633 def VLD2LNd32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
635 def VLD2LNq16_UPD : VLD2LNWB<0b0101, {?,?,1,?}, "16"> {
636 let Inst{7-6} = lane{1-0};
638 def VLD2LNq32_UPD : VLD2LNWB<0b1001, {?,1,0,?}, "32"> {
639 let Inst{7} = lane{0};
642 def VLD2LNq16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
643 def VLD2LNq32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
645 // VLD3LN : Vector Load (single 3-element structure to one lane)
646 class VLD3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
647 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
648 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3,
649 nohash_imm:$lane), IIC_VLD3ln, "vld3", Dt,
650 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn",
651 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3", []> {
655 def VLD3LNd8 : VLD3LN<0b0010, {?,?,?,0}, "8"> {
656 let Inst{7-5} = lane{2-0};
658 def VLD3LNd16 : VLD3LN<0b0110, {?,?,0,0}, "16"> {
659 let Inst{7-6} = lane{1-0};
661 def VLD3LNd32 : VLD3LN<0b1010, {?,0,0,0}, "32"> {
662 let Inst{7} = lane{0};
665 def VLD3LNd8Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
666 def VLD3LNd16Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
667 def VLD3LNd32Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
669 // ...with double-spaced registers:
670 def VLD3LNq16 : VLD3LN<0b0110, {?,?,1,0}, "16"> {
671 let Inst{7-6} = lane{1-0};
673 def VLD3LNq32 : VLD3LN<0b1010, {?,1,0,0}, "32"> {
674 let Inst{7} = lane{0};
677 def VLD3LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
678 def VLD3LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
680 // ...with address register writeback:
681 class VLD3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
682 : NLdStLn<1, 0b10, op11_8, op7_4,
683 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
684 (ins addrmode6:$Rn, am6offset:$Rm,
685 DPR:$src1, DPR:$src2, DPR:$src3, nohash_imm:$lane),
686 IIC_VLD3lnu, "vld3", Dt,
687 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn$Rm",
688 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $Rn.addr = $wb",
691 def VLD3LNd8_UPD : VLD3LNWB<0b0010, {?,?,?,0}, "8"> {
692 let Inst{7-5} = lane{2-0};
694 def VLD3LNd16_UPD : VLD3LNWB<0b0110, {?,?,0,0}, "16"> {
695 let Inst{7-6} = lane{1-0};
697 def VLD3LNd32_UPD : VLD3LNWB<0b1010, {?,0,0,0}, "32"> {
698 let Inst{7} = lane{0};
701 def VLD3LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
702 def VLD3LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
703 def VLD3LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
705 def VLD3LNq16_UPD : VLD3LNWB<0b0110, {?,?,1,0}, "16"> {
706 let Inst{7-6} = lane{1-0};
708 def VLD3LNq32_UPD : VLD3LNWB<0b1010, {?,1,0,0}, "32"> {
709 let Inst{7} = lane{0};
712 def VLD3LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
713 def VLD3LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
715 // VLD4LN : Vector Load (single 4-element structure to one lane)
716 class VLD4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
717 : NLdStLn<1, 0b10, op11_8, op7_4,
718 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
719 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
720 nohash_imm:$lane), IIC_VLD4ln, "vld4", Dt,
721 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn",
722 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4", []> {
727 def VLD4LNd8 : VLD4LN<0b0011, {?,?,?,?}, "8"> {
728 let Inst{7-5} = lane{2-0};
730 def VLD4LNd16 : VLD4LN<0b0111, {?,?,0,?}, "16"> {
731 let Inst{7-6} = lane{1-0};
733 def VLD4LNd32 : VLD4LN<0b1011, {?,0,?,?}, "32"> {
734 let Inst{7} = lane{0};
738 def VLD4LNd8Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
739 def VLD4LNd16Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
740 def VLD4LNd32Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
742 // ...with double-spaced registers:
743 def VLD4LNq16 : VLD4LN<0b0111, {?,?,1,?}, "16"> {
744 let Inst{7-6} = lane{1-0};
746 def VLD4LNq32 : VLD4LN<0b1011, {?,1,?,?}, "32"> {
747 let Inst{7} = lane{0};
751 def VLD4LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>;
752 def VLD4LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>;
754 // ...with address register writeback:
755 class VLD4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
756 : NLdStLn<1, 0b10, op11_8, op7_4,
757 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
758 (ins addrmode6:$Rn, am6offset:$Rm,
759 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
760 IIC_VLD4ln, "vld4", Dt,
761 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn$Rm",
762 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4, $Rn.addr = $wb",
767 def VLD4LNd8_UPD : VLD4LNWB<0b0011, {?,?,?,?}, "8"> {
768 let Inst{7-5} = lane{2-0};
770 def VLD4LNd16_UPD : VLD4LNWB<0b0111, {?,?,0,?}, "16"> {
771 let Inst{7-6} = lane{1-0};
773 def VLD4LNd32_UPD : VLD4LNWB<0b1011, {?,0,?,?}, "32"> {
774 let Inst{7} = lane{0};
778 def VLD4LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
779 def VLD4LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
780 def VLD4LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
782 def VLD4LNq16_UPD : VLD4LNWB<0b0111, {?,?,1,?}, "16"> {
783 let Inst{7-6} = lane{1-0};
785 def VLD4LNq32_UPD : VLD4LNWB<0b1011, {?,1,?,?}, "32"> {
786 let Inst{7} = lane{0};
790 def VLD4LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
791 def VLD4LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
793 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
795 // VLD1DUP : Vector Load (single element to all lanes)
796 class VLD1DUP<bits<4> op7_4, string Dt, ValueType Ty, PatFrag LoadOp>
797 : NLdSt<1, 0b10, 0b1100, op7_4, (outs DPR:$Vd), (ins addrmode6:$Rn),
798 IIC_VLD1dup, "vld1", Dt, "\\{$Vd[]\\}, $Rn", "",
799 [(set DPR:$Vd, (Ty (NEONvdup (i32 (LoadOp addrmode6:$Rn)))))]> {
803 class VLD1QDUPPseudo<ValueType Ty, PatFrag LoadOp> : VLDQPseudo<IIC_VLD1dup> {
804 let Pattern = [(set QPR:$dst,
805 (Ty (NEONvdup (i32 (LoadOp addrmode6:$addr)))))];
808 def VLD1DUPd8 : VLD1DUP<{0,0,0,?}, "8", v8i8, extloadi8>;
809 def VLD1DUPd16 : VLD1DUP<{0,1,0,?}, "16", v4i16, extloadi16>;
810 def VLD1DUPd32 : VLD1DUP<{1,0,0,?}, "32", v2i32, load>;
812 def VLD1DUPq8Pseudo : VLD1QDUPPseudo<v16i8, extloadi8>;
813 def VLD1DUPq16Pseudo : VLD1QDUPPseudo<v8i16, extloadi16>;
814 def VLD1DUPq32Pseudo : VLD1QDUPPseudo<v4i32, load>;
816 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
818 class VLD1QDUP<bits<4> op7_4, string Dt, ValueType Ty, PatFrag LoadOp>
819 : NLdSt<1, 0b10, 0b1100, op7_4, (outs DPR:$Vd, DPR:$dst2),
820 (ins addrmode6:$Rn), IIC_VLD1dup,
821 "vld1", Dt, "\\{$Vd[], $dst2[]\\}, $Rn", "", []> {
826 def VLD1DUPq8 : VLD1QDUP<{0,0,1,0}, "8", v16i8, extloadi8>;
827 def VLD1DUPq16 : VLD1QDUP<{0,1,1,?}, "16", v8i16, extloadi16>;
828 def VLD1DUPq32 : VLD1QDUP<{1,0,1,?}, "32", v4i32, load>;
830 // ...with address register writeback:
831 class VLD1DUPWB<bits<4> op7_4, string Dt>
832 : NLdSt<1, 0b10, 0b1100, op7_4, (outs DPR:$Vd, GPR:$wb),
833 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD1dupu,
834 "vld1", Dt, "\\{$Vd[]\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
837 class VLD1QDUPWB<bits<4> op7_4, string Dt>
838 : NLdSt<1, 0b10, 0b1100, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
839 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD1dupu,
840 "vld1", Dt, "\\{$Vd[], $dst2[]\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
844 def VLD1DUPd8_UPD : VLD1DUPWB<{0,0,0,0}, "8">;
845 def VLD1DUPd16_UPD : VLD1DUPWB<{0,1,0,?}, "16">;
846 def VLD1DUPd32_UPD : VLD1DUPWB<{1,0,0,?}, "32">;
848 def VLD1DUPq8_UPD : VLD1QDUPWB<{0,0,1,0}, "8">;
849 def VLD1DUPq16_UPD : VLD1QDUPWB<{0,1,1,?}, "16">;
850 def VLD1DUPq32_UPD : VLD1QDUPWB<{1,0,1,?}, "32">;
852 def VLD1DUPq8Pseudo_UPD : VLDQWBPseudo<IIC_VLD1dupu>;
853 def VLD1DUPq16Pseudo_UPD : VLDQWBPseudo<IIC_VLD1dupu>;
854 def VLD1DUPq32Pseudo_UPD : VLDQWBPseudo<IIC_VLD1dupu>;
856 // VLD2DUP : Vector Load (single 2-element structure to all lanes)
857 // VLD3DUP : Vector Load (single 3-element structure to all lanes)
858 // VLD4DUP : Vector Load (single 4-element structure to all lanes)
859 // FIXME: Not yet implemented.
860 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
862 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
864 // Classes for VST* pseudo-instructions with multi-register operands.
865 // These are expanded to real instructions after register allocation.
866 class VSTQPseudo<InstrItinClass itin>
867 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src), itin, "">;
868 class VSTQWBPseudo<InstrItinClass itin>
869 : PseudoNLdSt<(outs GPR:$wb),
870 (ins addrmode6:$addr, am6offset:$offset, QPR:$src), itin,
872 class VSTQQPseudo<InstrItinClass itin>
873 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src), itin, "">;
874 class VSTQQWBPseudo<InstrItinClass itin>
875 : PseudoNLdSt<(outs GPR:$wb),
876 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src), itin,
878 class VSTQQQQWBPseudo<InstrItinClass itin>
879 : PseudoNLdSt<(outs GPR:$wb),
880 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
883 // VST1 : Vector Store (multiple single elements)
884 class VST1D<bits<4> op7_4, string Dt>
885 : NLdSt<0,0b00,0b0111,op7_4, (outs), (ins addrmode6:$Rn, DPR:$Vd),
886 IIC_VST1, "vst1", Dt, "\\{$Vd\\}, $Rn", "", []> {
890 class VST1Q<bits<4> op7_4, string Dt>
891 : NLdSt<0,0b00,0b1010,op7_4, (outs),
892 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2), IIC_VST1x2,
893 "vst1", Dt, "\\{$Vd, $src2\\}, $Rn", "", []> {
895 let Inst{5-4} = Rn{5-4};
898 def VST1d8 : VST1D<{0,0,0,?}, "8">;
899 def VST1d16 : VST1D<{0,1,0,?}, "16">;
900 def VST1d32 : VST1D<{1,0,0,?}, "32">;
901 def VST1d64 : VST1D<{1,1,0,?}, "64">;
903 def VST1q8 : VST1Q<{0,0,?,?}, "8">;
904 def VST1q16 : VST1Q<{0,1,?,?}, "16">;
905 def VST1q32 : VST1Q<{1,0,?,?}, "32">;
906 def VST1q64 : VST1Q<{1,1,?,?}, "64">;
908 def VST1q8Pseudo : VSTQPseudo<IIC_VST1x2>;
909 def VST1q16Pseudo : VSTQPseudo<IIC_VST1x2>;
910 def VST1q32Pseudo : VSTQPseudo<IIC_VST1x2>;
911 def VST1q64Pseudo : VSTQPseudo<IIC_VST1x2>;
913 // ...with address register writeback:
914 class VST1DWB<bits<4> op7_4, string Dt>
915 : NLdSt<0, 0b00, 0b0111, op7_4, (outs GPR:$wb),
916 (ins addrmode6:$Rn, am6offset:$Rm, DPR:$Vd), IIC_VST1u,
917 "vst1", Dt, "\\{$Vd\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
920 class VST1QWB<bits<4> op7_4, string Dt>
921 : NLdSt<0, 0b00, 0b1010, op7_4, (outs GPR:$wb),
922 (ins addrmode6:$Rn, am6offset:$Rm, DPR:$Vd, DPR:$src2),
923 IIC_VST1x2u, "vst1", Dt, "\\{$Vd, $src2\\}, $Rn$Rm",
924 "$Rn.addr = $wb", []> {
925 let Inst{5-4} = Rn{5-4};
928 def VST1d8_UPD : VST1DWB<{0,0,0,?}, "8">;
929 def VST1d16_UPD : VST1DWB<{0,1,0,?}, "16">;
930 def VST1d32_UPD : VST1DWB<{1,0,0,?}, "32">;
931 def VST1d64_UPD : VST1DWB<{1,1,0,?}, "64">;
933 def VST1q8_UPD : VST1QWB<{0,0,?,?}, "8">;
934 def VST1q16_UPD : VST1QWB<{0,1,?,?}, "16">;
935 def VST1q32_UPD : VST1QWB<{1,0,?,?}, "32">;
936 def VST1q64_UPD : VST1QWB<{1,1,?,?}, "64">;
938 def VST1q8Pseudo_UPD : VSTQWBPseudo<IIC_VST1x2u>;
939 def VST1q16Pseudo_UPD : VSTQWBPseudo<IIC_VST1x2u>;
940 def VST1q32Pseudo_UPD : VSTQWBPseudo<IIC_VST1x2u>;
941 def VST1q64Pseudo_UPD : VSTQWBPseudo<IIC_VST1x2u>;
943 // ...with 3 registers (some of these are only for the disassembler):
944 class VST1D3<bits<4> op7_4, string Dt>
945 : NLdSt<0, 0b00, 0b0110, op7_4, (outs),
946 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3),
947 IIC_VST1x3, "vst1", Dt, "\\{$Vd, $src2, $src3\\}, $Rn", "", []> {
951 class VST1D3WB<bits<4> op7_4, string Dt>
952 : NLdSt<0, 0b00, 0b0110, op7_4, (outs GPR:$wb),
953 (ins addrmode6:$Rn, am6offset:$Rm,
954 DPR:$Vd, DPR:$src2, DPR:$src3),
955 IIC_VST1x3u, "vst1", Dt, "\\{$Vd, $src2, $src3\\}, $Rn$Rm",
956 "$Rn.addr = $wb", []> {
960 def VST1d8T : VST1D3<{0,0,0,?}, "8">;
961 def VST1d16T : VST1D3<{0,1,0,?}, "16">;
962 def VST1d32T : VST1D3<{1,0,0,?}, "32">;
963 def VST1d64T : VST1D3<{1,1,0,?}, "64">;
965 def VST1d8T_UPD : VST1D3WB<{0,0,0,?}, "8">;
966 def VST1d16T_UPD : VST1D3WB<{0,1,0,?}, "16">;
967 def VST1d32T_UPD : VST1D3WB<{1,0,0,?}, "32">;
968 def VST1d64T_UPD : VST1D3WB<{1,1,0,?}, "64">;
970 def VST1d64TPseudo : VSTQQPseudo<IIC_VST1x3>;
971 def VST1d64TPseudo_UPD : VSTQQWBPseudo<IIC_VST1x3u>;
973 // ...with 4 registers (some of these are only for the disassembler):
974 class VST1D4<bits<4> op7_4, string Dt>
975 : NLdSt<0, 0b00, 0b0010, op7_4, (outs),
976 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
977 IIC_VST1x4, "vst1", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn", "",
980 let Inst{5-4} = Rn{5-4};
982 class VST1D4WB<bits<4> op7_4, string Dt>
983 : NLdSt<0, 0b00, 0b0010, op7_4, (outs GPR:$wb),
984 (ins addrmode6:$Rn, am6offset:$Rm,
985 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST1x4u,
986 "vst1", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
987 "$Rn.addr = $wb", []> {
988 let Inst{5-4} = Rn{5-4};
991 def VST1d8Q : VST1D4<{0,0,?,?}, "8">;
992 def VST1d16Q : VST1D4<{0,1,?,?}, "16">;
993 def VST1d32Q : VST1D4<{1,0,?,?}, "32">;
994 def VST1d64Q : VST1D4<{1,1,?,?}, "64">;
996 def VST1d8Q_UPD : VST1D4WB<{0,0,?,?}, "8">;
997 def VST1d16Q_UPD : VST1D4WB<{0,1,?,?}, "16">;
998 def VST1d32Q_UPD : VST1D4WB<{1,0,?,?}, "32">;
999 def VST1d64Q_UPD : VST1D4WB<{1,1,?,?}, "64">;
1001 def VST1d64QPseudo : VSTQQPseudo<IIC_VST1x4>;
1002 def VST1d64QPseudo_UPD : VSTQQWBPseudo<IIC_VST1x4u>;
1004 // VST2 : Vector Store (multiple 2-element structures)
1005 class VST2D<bits<4> op11_8, bits<4> op7_4, string Dt>
1006 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1007 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2),
1008 IIC_VST2, "vst2", Dt, "\\{$Vd, $src2\\}, $Rn", "", []> {
1010 let Inst{5-4} = Rn{5-4};
1012 class VST2Q<bits<4> op7_4, string Dt>
1013 : NLdSt<0, 0b00, 0b0011, op7_4, (outs),
1014 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
1015 IIC_VST2x2, "vst2", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn",
1018 let Inst{5-4} = Rn{5-4};
1021 def VST2d8 : VST2D<0b1000, {0,0,?,?}, "8">;
1022 def VST2d16 : VST2D<0b1000, {0,1,?,?}, "16">;
1023 def VST2d32 : VST2D<0b1000, {1,0,?,?}, "32">;
1025 def VST2q8 : VST2Q<{0,0,?,?}, "8">;
1026 def VST2q16 : VST2Q<{0,1,?,?}, "16">;
1027 def VST2q32 : VST2Q<{1,0,?,?}, "32">;
1029 def VST2d8Pseudo : VSTQPseudo<IIC_VST2>;
1030 def VST2d16Pseudo : VSTQPseudo<IIC_VST2>;
1031 def VST2d32Pseudo : VSTQPseudo<IIC_VST2>;
1033 def VST2q8Pseudo : VSTQQPseudo<IIC_VST2x2>;
1034 def VST2q16Pseudo : VSTQQPseudo<IIC_VST2x2>;
1035 def VST2q32Pseudo : VSTQQPseudo<IIC_VST2x2>;
1037 // ...with address register writeback:
1038 class VST2DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1039 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1040 (ins addrmode6:$Rn, am6offset:$Rm, DPR:$Vd, DPR:$src2),
1041 IIC_VST2u, "vst2", Dt, "\\{$Vd, $src2\\}, $Rn$Rm",
1042 "$Rn.addr = $wb", []> {
1043 let Inst{5-4} = Rn{5-4};
1045 class VST2QWB<bits<4> op7_4, string Dt>
1046 : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
1047 (ins addrmode6:$Rn, am6offset:$Rm,
1048 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST2x2u,
1049 "vst2", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
1050 "$Rn.addr = $wb", []> {
1051 let Inst{5-4} = Rn{5-4};
1054 def VST2d8_UPD : VST2DWB<0b1000, {0,0,?,?}, "8">;
1055 def VST2d16_UPD : VST2DWB<0b1000, {0,1,?,?}, "16">;
1056 def VST2d32_UPD : VST2DWB<0b1000, {1,0,?,?}, "32">;
1058 def VST2q8_UPD : VST2QWB<{0,0,?,?}, "8">;
1059 def VST2q16_UPD : VST2QWB<{0,1,?,?}, "16">;
1060 def VST2q32_UPD : VST2QWB<{1,0,?,?}, "32">;
1062 def VST2d8Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
1063 def VST2d16Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
1064 def VST2d32Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
1066 def VST2q8Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
1067 def VST2q16Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
1068 def VST2q32Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
1070 // ...with double-spaced registers (for disassembly only):
1071 def VST2b8 : VST2D<0b1001, {0,0,?,?}, "8">;
1072 def VST2b16 : VST2D<0b1001, {0,1,?,?}, "16">;
1073 def VST2b32 : VST2D<0b1001, {1,0,?,?}, "32">;
1074 def VST2b8_UPD : VST2DWB<0b1001, {0,0,?,?}, "8">;
1075 def VST2b16_UPD : VST2DWB<0b1001, {0,1,?,?}, "16">;
1076 def VST2b32_UPD : VST2DWB<0b1001, {1,0,?,?}, "32">;
1078 // VST3 : Vector Store (multiple 3-element structures)
1079 class VST3D<bits<4> op11_8, bits<4> op7_4, string Dt>
1080 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1081 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3,
1082 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn", "", []> {
1084 let Inst{4} = Rn{4};
1087 def VST3d8 : VST3D<0b0100, {0,0,0,?}, "8">;
1088 def VST3d16 : VST3D<0b0100, {0,1,0,?}, "16">;
1089 def VST3d32 : VST3D<0b0100, {1,0,0,?}, "32">;
1091 def VST3d8Pseudo : VSTQQPseudo<IIC_VST3>;
1092 def VST3d16Pseudo : VSTQQPseudo<IIC_VST3>;
1093 def VST3d32Pseudo : VSTQQPseudo<IIC_VST3>;
1095 // ...with address register writeback:
1096 class VST3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1097 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1098 (ins addrmode6:$Rn, am6offset:$Rm,
1099 DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3u,
1100 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn$Rm",
1101 "$Rn.addr = $wb", []> {
1102 let Inst{4} = Rn{4};
1105 def VST3d8_UPD : VST3DWB<0b0100, {0,0,0,?}, "8">;
1106 def VST3d16_UPD : VST3DWB<0b0100, {0,1,0,?}, "16">;
1107 def VST3d32_UPD : VST3DWB<0b0100, {1,0,0,?}, "32">;
1109 def VST3d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1110 def VST3d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1111 def VST3d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1113 // ...with double-spaced registers (non-updating versions for disassembly only):
1114 def VST3q8 : VST3D<0b0101, {0,0,0,?}, "8">;
1115 def VST3q16 : VST3D<0b0101, {0,1,0,?}, "16">;
1116 def VST3q32 : VST3D<0b0101, {1,0,0,?}, "32">;
1117 def VST3q8_UPD : VST3DWB<0b0101, {0,0,0,?}, "8">;
1118 def VST3q16_UPD : VST3DWB<0b0101, {0,1,0,?}, "16">;
1119 def VST3q32_UPD : VST3DWB<0b0101, {1,0,0,?}, "32">;
1121 def VST3q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1122 def VST3q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1123 def VST3q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1125 // ...alternate versions to be allocated odd register numbers:
1126 def VST3q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1127 def VST3q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1128 def VST3q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1130 // VST4 : Vector Store (multiple 4-element structures)
1131 class VST4D<bits<4> op11_8, bits<4> op7_4, string Dt>
1132 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1133 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
1134 IIC_VST4, "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn",
1137 let Inst{5-4} = Rn{5-4};
1140 def VST4d8 : VST4D<0b0000, {0,0,?,?}, "8">;
1141 def VST4d16 : VST4D<0b0000, {0,1,?,?}, "16">;
1142 def VST4d32 : VST4D<0b0000, {1,0,?,?}, "32">;
1144 def VST4d8Pseudo : VSTQQPseudo<IIC_VST4>;
1145 def VST4d16Pseudo : VSTQQPseudo<IIC_VST4>;
1146 def VST4d32Pseudo : VSTQQPseudo<IIC_VST4>;
1148 // ...with address register writeback:
1149 class VST4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1150 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1151 (ins addrmode6:$Rn, am6offset:$Rm,
1152 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST4u,
1153 "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
1154 "$Rn.addr = $wb", []> {
1155 let Inst{5-4} = Rn{5-4};
1158 def VST4d8_UPD : VST4DWB<0b0000, {0,0,?,?}, "8">;
1159 def VST4d16_UPD : VST4DWB<0b0000, {0,1,?,?}, "16">;
1160 def VST4d32_UPD : VST4DWB<0b0000, {1,0,?,?}, "32">;
1162 def VST4d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1163 def VST4d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1164 def VST4d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1166 // ...with double-spaced registers (non-updating versions for disassembly only):
1167 def VST4q8 : VST4D<0b0001, {0,0,?,?}, "8">;
1168 def VST4q16 : VST4D<0b0001, {0,1,?,?}, "16">;
1169 def VST4q32 : VST4D<0b0001, {1,0,?,?}, "32">;
1170 def VST4q8_UPD : VST4DWB<0b0001, {0,0,?,?}, "8">;
1171 def VST4q16_UPD : VST4DWB<0b0001, {0,1,?,?}, "16">;
1172 def VST4q32_UPD : VST4DWB<0b0001, {1,0,?,?}, "32">;
1174 def VST4q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1175 def VST4q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1176 def VST4q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1178 // ...alternate versions to be allocated odd register numbers:
1179 def VST4q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1180 def VST4q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1181 def VST4q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1183 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
1185 // Classes for VST*LN pseudo-instructions with multi-register operands.
1186 // These are expanded to real instructions after register allocation.
1187 class VSTQLNPseudo<InstrItinClass itin>
1188 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
1190 class VSTQLNWBPseudo<InstrItinClass itin>
1191 : PseudoNLdSt<(outs GPR:$wb),
1192 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
1193 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1194 class VSTQQLNPseudo<InstrItinClass itin>
1195 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
1197 class VSTQQLNWBPseudo<InstrItinClass itin>
1198 : PseudoNLdSt<(outs GPR:$wb),
1199 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
1200 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1201 class VSTQQQQLNPseudo<InstrItinClass itin>
1202 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
1204 class VSTQQQQLNWBPseudo<InstrItinClass itin>
1205 : PseudoNLdSt<(outs GPR:$wb),
1206 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
1207 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1209 // VST1LN : Vector Store (single element from one lane)
1210 class VST1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1211 PatFrag StoreOp, SDNode ExtractOp>
1212 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1213 (ins addrmode6:$Rn, DPR:$Vd, nohash_imm:$lane),
1214 IIC_VST1ln, "vst1", Dt, "\\{$Vd[$lane]\\}, $Rn", "",
1215 [(StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane), addrmode6:$Rn)]> {
1218 class VST1QLNPseudo<ValueType Ty, PatFrag StoreOp, SDNode ExtractOp>
1219 : VSTQLNPseudo<IIC_VST1ln> {
1220 let Pattern = [(StoreOp (ExtractOp (Ty QPR:$src), imm:$lane),
1224 def VST1LNd8 : VST1LN<0b0000, {?,?,?,0}, "8", v8i8, truncstorei8,
1226 let Inst{7-5} = lane{2-0};
1228 def VST1LNd16 : VST1LN<0b0100, {?,?,0,?}, "16", v4i16, truncstorei16,
1230 let Inst{7-6} = lane{1-0};
1231 let Inst{4} = Rn{5};
1233 def VST1LNd32 : VST1LN<0b1000, {?,0,?,?}, "32", v2i32, store, extractelt> {
1234 let Inst{7} = lane{0};
1235 let Inst{5-4} = Rn{5-4};
1238 def VST1LNq8Pseudo : VST1QLNPseudo<v16i8, truncstorei8, NEONvgetlaneu>;
1239 def VST1LNq16Pseudo : VST1QLNPseudo<v8i16, truncstorei16, NEONvgetlaneu>;
1240 def VST1LNq32Pseudo : VST1QLNPseudo<v4i32, store, extractelt>;
1242 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
1244 // ...with address register writeback:
1245 class VST1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1246 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1247 (ins addrmode6:$Rn, am6offset:$Rm,
1248 DPR:$Vd, nohash_imm:$lane), IIC_VST1lnu, "vst1", Dt,
1249 "\\{$Vd[$lane]\\}, $Rn$Rm",
1250 "$Rn.addr = $wb", []>;
1252 def VST1LNd8_UPD : VST1LNWB<0b0000, {?,?,?,0}, "8"> {
1253 let Inst{7-5} = lane{2-0};
1255 def VST1LNd16_UPD : VST1LNWB<0b0100, {?,?,0,?}, "16"> {
1256 let Inst{7-6} = lane{1-0};
1257 let Inst{4} = Rn{5};
1259 def VST1LNd32_UPD : VST1LNWB<0b1000, {?,0,?,?}, "32"> {
1260 let Inst{7} = lane{0};
1261 let Inst{5-4} = Rn{5-4};
1264 def VST1LNq8Pseudo_UPD : VSTQLNWBPseudo<IIC_VST1lnu>;
1265 def VST1LNq16Pseudo_UPD : VSTQLNWBPseudo<IIC_VST1lnu>;
1266 def VST1LNq32Pseudo_UPD : VSTQLNWBPseudo<IIC_VST1lnu>;
1268 // VST2LN : Vector Store (single 2-element structure from one lane)
1269 class VST2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1270 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1271 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, nohash_imm:$lane),
1272 IIC_VST2ln, "vst2", Dt, "\\{$Vd[$lane], $src2[$lane]\\}, $Rn",
1275 let Inst{4} = Rn{4};
1278 def VST2LNd8 : VST2LN<0b0001, {?,?,?,?}, "8"> {
1279 let Inst{7-5} = lane{2-0};
1281 def VST2LNd16 : VST2LN<0b0101, {?,?,0,?}, "16"> {
1282 let Inst{7-6} = lane{1-0};
1284 def VST2LNd32 : VST2LN<0b1001, {?,0,0,?}, "32"> {
1285 let Inst{7} = lane{0};
1288 def VST2LNd8Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1289 def VST2LNd16Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1290 def VST2LNd32Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1292 // ...with double-spaced registers:
1293 def VST2LNq16 : VST2LN<0b0101, {?,?,1,?}, "16"> {
1294 let Inst{7-6} = lane{1-0};
1295 let Inst{4} = Rn{4};
1297 def VST2LNq32 : VST2LN<0b1001, {?,1,0,?}, "32"> {
1298 let Inst{7} = lane{0};
1299 let Inst{4} = Rn{4};
1302 def VST2LNq16Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
1303 def VST2LNq32Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
1305 // ...with address register writeback:
1306 class VST2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1307 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1308 (ins addrmode6:$addr, am6offset:$offset,
1309 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VST2lnu, "vst2", Dt,
1310 "\\{$src1[$lane], $src2[$lane]\\}, $addr$offset",
1311 "$addr.addr = $wb", []> {
1312 let Inst{4} = Rn{4};
1315 def VST2LNd8_UPD : VST2LNWB<0b0001, {?,?,?,?}, "8"> {
1316 let Inst{7-5} = lane{2-0};
1318 def VST2LNd16_UPD : VST2LNWB<0b0101, {?,?,0,?}, "16"> {
1319 let Inst{7-6} = lane{1-0};
1321 def VST2LNd32_UPD : VST2LNWB<0b1001, {?,0,0,?}, "32"> {
1322 let Inst{7} = lane{0};
1325 def VST2LNd8Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1326 def VST2LNd16Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1327 def VST2LNd32Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1329 def VST2LNq16_UPD : VST2LNWB<0b0101, {?,?,1,?}, "16"> {
1330 let Inst{7-6} = lane{1-0};
1332 def VST2LNq32_UPD : VST2LNWB<0b1001, {?,1,0,?}, "32"> {
1333 let Inst{7} = lane{0};
1336 def VST2LNq16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
1337 def VST2LNq32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
1339 // VST3LN : Vector Store (single 3-element structure from one lane)
1340 class VST3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1341 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1342 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3,
1343 nohash_imm:$lane), IIC_VST3ln, "vst3", Dt,
1344 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn", "", []> {
1348 def VST3LNd8 : VST3LN<0b0010, {?,?,?,0}, "8"> {
1349 let Inst{7-5} = lane{2-0};
1351 def VST3LNd16 : VST3LN<0b0110, {?,?,0,0}, "16"> {
1352 let Inst{7-6} = lane{1-0};
1354 def VST3LNd32 : VST3LN<0b1010, {?,0,0,0}, "32"> {
1355 let Inst{7} = lane{0};
1358 def VST3LNd8Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1359 def VST3LNd16Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1360 def VST3LNd32Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1362 // ...with double-spaced registers:
1363 def VST3LNq16 : VST3LN<0b0110, {?,?,1,0}, "16"> {
1364 let Inst{7-6} = lane{1-0};
1366 def VST3LNq32 : VST3LN<0b1010, {?,1,0,0}, "32"> {
1367 let Inst{7} = lane{0};
1370 def VST3LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
1371 def VST3LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
1373 // ...with address register writeback:
1374 class VST3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1375 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1376 (ins addrmode6:$Rn, am6offset:$Rm,
1377 DPR:$Vd, DPR:$src2, DPR:$src3, nohash_imm:$lane),
1378 IIC_VST3lnu, "vst3", Dt,
1379 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn$Rm",
1380 "$Rn.addr = $wb", []>;
1382 def VST3LNd8_UPD : VST3LNWB<0b0010, {?,?,?,0}, "8"> {
1383 let Inst{7-5} = lane{2-0};
1385 def VST3LNd16_UPD : VST3LNWB<0b0110, {?,?,0,0}, "16"> {
1386 let Inst{7-6} = lane{1-0};
1388 def VST3LNd32_UPD : VST3LNWB<0b1010, {?,0,0,0}, "32"> {
1389 let Inst{7} = lane{0};
1392 def VST3LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1393 def VST3LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1394 def VST3LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1396 def VST3LNq16_UPD : VST3LNWB<0b0110, {?,?,1,0}, "16"> {
1397 let Inst{7-6} = lane{1-0};
1399 def VST3LNq32_UPD : VST3LNWB<0b1010, {?,1,0,0}, "32"> {
1400 let Inst{7} = lane{0};
1403 def VST3LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
1404 def VST3LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
1406 // VST4LN : Vector Store (single 4-element structure from one lane)
1407 class VST4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1408 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1409 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4,
1410 nohash_imm:$lane), IIC_VST4ln, "vst4", Dt,
1411 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn",
1414 let Inst{4} = Rn{4};
1417 def VST4LNd8 : VST4LN<0b0011, {?,?,?,?}, "8"> {
1418 let Inst{7-5} = lane{2-0};
1420 def VST4LNd16 : VST4LN<0b0111, {?,?,0,?}, "16"> {
1421 let Inst{7-6} = lane{1-0};
1423 def VST4LNd32 : VST4LN<0b1011, {?,0,?,?}, "32"> {
1424 let Inst{7} = lane{0};
1425 let Inst{5} = Rn{5};
1428 def VST4LNd8Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1429 def VST4LNd16Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1430 def VST4LNd32Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1432 // ...with double-spaced registers:
1433 def VST4LNq16 : VST4LN<0b0111, {?,?,1,?}, "16"> {
1434 let Inst{7-6} = lane{1-0};
1436 def VST4LNq32 : VST4LN<0b1011, {?,1,?,?}, "32"> {
1437 let Inst{7} = lane{0};
1438 let Inst{5} = Rn{5};
1441 def VST4LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
1442 def VST4LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
1444 // ...with address register writeback:
1445 class VST4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1446 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1447 (ins addrmode6:$Rn, am6offset:$Rm,
1448 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
1449 IIC_VST4lnu, "vst4", Dt,
1450 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn$Rm",
1451 "$Rn.addr = $wb", []> {
1452 let Inst{4} = Rn{4};
1455 def VST4LNd8_UPD : VST4LNWB<0b0011, {?,?,?,?}, "8"> {
1456 let Inst{7-5} = lane{2-0};
1458 def VST4LNd16_UPD : VST4LNWB<0b0111, {?,?,0,?}, "16"> {
1459 let Inst{7-6} = lane{1-0};
1461 def VST4LNd32_UPD : VST4LNWB<0b1011, {?,0,?,?}, "32"> {
1462 let Inst{7} = lane{0};
1463 let Inst{5} = Rn{5};
1466 def VST4LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
1467 def VST4LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
1468 def VST4LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
1470 def VST4LNq16_UPD : VST4LNWB<0b0111, {?,?,1,?}, "16"> {
1471 let Inst{7-6} = lane{1-0};
1473 def VST4LNq32_UPD : VST4LNWB<0b1011, {?,1,?,?}, "32"> {
1474 let Inst{7} = lane{0};
1475 let Inst{5} = Rn{5};
1478 def VST4LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
1479 def VST4LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
1481 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
1484 //===----------------------------------------------------------------------===//
1485 // NEON pattern fragments
1486 //===----------------------------------------------------------------------===//
1488 // Extract D sub-registers of Q registers.
1489 def DSubReg_i8_reg : SDNodeXForm<imm, [{
1490 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1491 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/8, MVT::i32);
1493 def DSubReg_i16_reg : SDNodeXForm<imm, [{
1494 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1495 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/4, MVT::i32);
1497 def DSubReg_i32_reg : SDNodeXForm<imm, [{
1498 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1499 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/2, MVT::i32);
1501 def DSubReg_f64_reg : SDNodeXForm<imm, [{
1502 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1503 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue(), MVT::i32);
1506 // Extract S sub-registers of Q/D registers.
1507 def SSubReg_f32_reg : SDNodeXForm<imm, [{
1508 assert(ARM::ssub_3 == ARM::ssub_0+3 && "Unexpected subreg numbering");
1509 return CurDAG->getTargetConstant(ARM::ssub_0 + N->getZExtValue(), MVT::i32);
1512 // Translate lane numbers from Q registers to D subregs.
1513 def SubReg_i8_lane : SDNodeXForm<imm, [{
1514 return CurDAG->getTargetConstant(N->getZExtValue() & 7, MVT::i32);
1516 def SubReg_i16_lane : SDNodeXForm<imm, [{
1517 return CurDAG->getTargetConstant(N->getZExtValue() & 3, MVT::i32);
1519 def SubReg_i32_lane : SDNodeXForm<imm, [{
1520 return CurDAG->getTargetConstant(N->getZExtValue() & 1, MVT::i32);
1523 //===----------------------------------------------------------------------===//
1524 // Instruction Classes
1525 //===----------------------------------------------------------------------===//
1527 // Basic 2-register operations: single-, double- and quad-register.
1528 class N2VS<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1529 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
1530 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
1531 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
1532 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src),
1533 IIC_VUNAD, OpcodeStr, Dt, "$dst, $src", "", []>;
1534 class N2VD<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1535 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
1536 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
1537 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
1538 (ins DPR:$src), IIC_VUNAD, OpcodeStr, Dt,"$dst, $src", "",
1539 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src))))]>;
1540 class N2VQ<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1541 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
1542 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
1543 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
1544 (ins QPR:$src), IIC_VUNAQ, OpcodeStr, Dt,"$dst, $src", "",
1545 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src))))]>;
1547 // Basic 2-register intrinsics, both double- and quad-register.
1548 class N2VDInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1549 bits<2> op17_16, bits<5> op11_7, bit op4,
1550 InstrItinClass itin, string OpcodeStr, string Dt,
1551 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1552 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
1553 (ins DPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
1554 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
1555 class N2VQInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1556 bits<2> op17_16, bits<5> op11_7, bit op4,
1557 InstrItinClass itin, string OpcodeStr, string Dt,
1558 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1559 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
1560 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
1561 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
1563 // Narrow 2-register operations.
1564 class N2VN<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1565 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1566 InstrItinClass itin, string OpcodeStr, string Dt,
1567 ValueType TyD, ValueType TyQ, SDNode OpNode>
1568 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$dst),
1569 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
1570 [(set DPR:$dst, (TyD (OpNode (TyQ QPR:$src))))]>;
1572 // Narrow 2-register intrinsics.
1573 class N2VNInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1574 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1575 InstrItinClass itin, string OpcodeStr, string Dt,
1576 ValueType TyD, ValueType TyQ, Intrinsic IntOp>
1577 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$dst),
1578 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
1579 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src))))]>;
1581 // Long 2-register operations (currently only used for VMOVL).
1582 class N2VL<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1583 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1584 InstrItinClass itin, string OpcodeStr, string Dt,
1585 ValueType TyQ, ValueType TyD, SDNode OpNode>
1586 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$dst),
1587 (ins DPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
1588 [(set QPR:$dst, (TyQ (OpNode (TyD DPR:$src))))]>;
1590 // 2-register shuffles (VTRN/VZIP/VUZP), both double- and quad-register.
1591 class N2VDShuffle<bits<2> op19_18, bits<5> op11_7, string OpcodeStr, string Dt>
1592 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 0, 0, (outs DPR:$dst1, DPR:$dst2),
1593 (ins DPR:$src1, DPR:$src2), IIC_VPERMD,
1594 OpcodeStr, Dt, "$dst1, $dst2",
1595 "$src1 = $dst1, $src2 = $dst2", []>;
1596 class N2VQShuffle<bits<2> op19_18, bits<5> op11_7,
1597 InstrItinClass itin, string OpcodeStr, string Dt>
1598 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 1, 0, (outs QPR:$dst1, QPR:$dst2),
1599 (ins QPR:$src1, QPR:$src2), itin, OpcodeStr, Dt, "$dst1, $dst2",
1600 "$src1 = $dst1, $src2 = $dst2", []>;
1602 // Basic 3-register operations: single-, double- and quad-register.
1603 class N3VS<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1604 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1605 SDNode OpNode, bit Commutable>
1606 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1607 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src1, DPR_VFP2:$src2), N3RegFrm,
1608 IIC_VBIND, OpcodeStr, Dt, "$dst, $src1, $src2", "", []> {
1609 let isCommutable = Commutable;
1612 class N3VD<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1613 InstrItinClass itin, string OpcodeStr, string Dt,
1614 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
1615 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1616 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1617 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
1618 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
1619 let isCommutable = Commutable;
1621 // Same as N3VD but no data type.
1622 class N3VDX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1623 InstrItinClass itin, string OpcodeStr,
1624 ValueType ResTy, ValueType OpTy,
1625 SDNode OpNode, bit Commutable>
1626 : N3VX<op24, op23, op21_20, op11_8, 0, op4,
1627 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1628 OpcodeStr, "$Vd, $Vn, $Vm", "",
1629 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>{
1630 let isCommutable = Commutable;
1633 class N3VDSL<bits<2> op21_20, bits<4> op11_8,
1634 InstrItinClass itin, string OpcodeStr, string Dt,
1635 ValueType Ty, SDNode ShOp>
1636 : N3V<0, 1, op21_20, op11_8, 1, 0,
1637 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1638 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1639 [(set (Ty DPR:$dst),
1640 (Ty (ShOp (Ty DPR:$src1),
1641 (Ty (NEONvduplane (Ty DPR_VFP2:$src2),imm:$lane)))))]> {
1642 let isCommutable = 0;
1644 class N3VDSL16<bits<2> op21_20, bits<4> op11_8,
1645 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1646 : N3V<0, 1, op21_20, op11_8, 1, 0,
1647 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1648 NVMulSLFrm, IIC_VMULi16D, OpcodeStr, Dt,"$dst, $src1, $src2[$lane]","",
1649 [(set (Ty DPR:$dst),
1650 (Ty (ShOp (Ty DPR:$src1),
1651 (Ty (NEONvduplane (Ty DPR_8:$src2), imm:$lane)))))]> {
1652 let isCommutable = 0;
1655 class N3VQ<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1656 InstrItinClass itin, string OpcodeStr, string Dt,
1657 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
1658 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1659 (outs QPR:$Qd), (ins QPR:$Qn, QPR:$Qm), N3RegFrm, itin,
1660 OpcodeStr, Dt, "$Qd, $Qn, $Qm", "",
1661 [(set QPR:$Qd, (ResTy (OpNode (OpTy QPR:$Qn), (OpTy QPR:$Qm))))]> {
1662 let isCommutable = Commutable;
1664 class N3VQX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1665 InstrItinClass itin, string OpcodeStr,
1666 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
1667 : N3VX<op24, op23, op21_20, op11_8, 1, op4,
1668 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), N3RegFrm, itin,
1669 OpcodeStr, "$dst, $src1, $src2", "",
1670 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src1), (OpTy QPR:$src2))))]>{
1671 let isCommutable = Commutable;
1673 class N3VQSL<bits<2> op21_20, bits<4> op11_8,
1674 InstrItinClass itin, string OpcodeStr, string Dt,
1675 ValueType ResTy, ValueType OpTy, SDNode ShOp>
1676 : N3V<1, 1, op21_20, op11_8, 1, 0,
1677 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1678 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1679 [(set (ResTy QPR:$dst),
1680 (ResTy (ShOp (ResTy QPR:$src1),
1681 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1683 let isCommutable = 0;
1685 class N3VQSL16<bits<2> op21_20, bits<4> op11_8, string OpcodeStr, string Dt,
1686 ValueType ResTy, ValueType OpTy, SDNode ShOp>
1687 : N3V<1, 1, op21_20, op11_8, 1, 0,
1688 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1689 NVMulSLFrm, IIC_VMULi16Q, OpcodeStr, Dt,"$dst, $src1, $src2[$lane]","",
1690 [(set (ResTy QPR:$dst),
1691 (ResTy (ShOp (ResTy QPR:$src1),
1692 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
1694 let isCommutable = 0;
1697 // Basic 3-register intrinsics, both double- and quad-register.
1698 class N3VDInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1699 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1700 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
1701 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1702 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), f, itin,
1703 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
1704 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
1705 let isCommutable = Commutable;
1707 class N3VDIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1708 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
1709 : N3V<0, 1, op21_20, op11_8, 1, 0,
1710 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1711 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1712 [(set (Ty DPR:$dst),
1713 (Ty (IntOp (Ty DPR:$src1),
1714 (Ty (NEONvduplane (Ty DPR_VFP2:$src2),
1716 let isCommutable = 0;
1718 class N3VDIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1719 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
1720 : N3V<0, 1, op21_20, op11_8, 1, 0,
1721 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1722 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1723 [(set (Ty DPR:$dst),
1724 (Ty (IntOp (Ty DPR:$src1),
1725 (Ty (NEONvduplane (Ty DPR_8:$src2), imm:$lane)))))]> {
1726 let isCommutable = 0;
1728 class N3VDIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1729 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1730 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1731 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1732 (outs DPR:$Vd), (ins DPR:$Vm, DPR:$Vn), f, itin,
1733 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
1734 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (OpTy DPR:$Vn))))]> {
1735 let isCommutable = 0;
1738 class N3VQInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1739 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1740 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
1741 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1742 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), f, itin,
1743 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
1744 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
1745 let isCommutable = Commutable;
1747 class N3VQIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1748 string OpcodeStr, string Dt,
1749 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1750 : N3V<1, 1, op21_20, op11_8, 1, 0,
1751 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1752 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1753 [(set (ResTy QPR:$dst),
1754 (ResTy (IntOp (ResTy QPR:$src1),
1755 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1757 let isCommutable = 0;
1759 class N3VQIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1760 string OpcodeStr, string Dt,
1761 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1762 : N3V<1, 1, op21_20, op11_8, 1, 0,
1763 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1764 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1765 [(set (ResTy QPR:$dst),
1766 (ResTy (IntOp (ResTy QPR:$src1),
1767 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
1769 let isCommutable = 0;
1771 class N3VQIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1772 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1773 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1774 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1775 (outs QPR:$Vd), (ins QPR:$Vm, QPR:$Vn), f, itin,
1776 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
1777 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (OpTy QPR:$Vn))))]> {
1778 let isCommutable = 0;
1781 // Multiply-Add/Sub operations: single-, double- and quad-register.
1782 class N3VSMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1783 InstrItinClass itin, string OpcodeStr, string Dt,
1784 ValueType Ty, SDNode MulOp, SDNode OpNode>
1785 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1786 (outs DPR_VFP2:$dst),
1787 (ins DPR_VFP2:$src1, DPR_VFP2:$src2, DPR_VFP2:$src3), N3RegFrm, itin,
1788 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst", []>;
1790 class N3VDMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1791 InstrItinClass itin, string OpcodeStr, string Dt,
1792 ValueType Ty, SDNode MulOp, SDNode OpNode>
1793 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1794 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1795 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1796 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
1797 (Ty (MulOp DPR:$Vn, DPR:$Vm)))))]>;
1799 class N3VDMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1800 string OpcodeStr, string Dt,
1801 ValueType Ty, SDNode MulOp, SDNode ShOp>
1802 : N3V<0, 1, op21_20, op11_8, 1, 0,
1804 (ins DPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1806 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1807 [(set (Ty DPR:$dst),
1808 (Ty (ShOp (Ty DPR:$src1),
1809 (Ty (MulOp DPR:$src2,
1810 (Ty (NEONvduplane (Ty DPR_VFP2:$src3),
1812 class N3VDMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1813 string OpcodeStr, string Dt,
1814 ValueType Ty, SDNode MulOp, SDNode ShOp>
1815 : N3V<0, 1, op21_20, op11_8, 1, 0,
1817 (ins DPR:$src1, DPR:$Vn, DPR_8:$Vm, nohash_imm:$lane),
1819 OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "$src1 = $Vd",
1821 (Ty (ShOp (Ty DPR:$src1),
1823 (Ty (NEONvduplane (Ty DPR_8:$Vm),
1826 class N3VQMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1827 InstrItinClass itin, string OpcodeStr, string Dt, ValueType Ty,
1828 SDNode MulOp, SDNode OpNode>
1829 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1830 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
1831 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1832 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
1833 (Ty (MulOp QPR:$Vn, QPR:$Vm)))))]>;
1834 class N3VQMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1835 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1836 SDNode MulOp, SDNode ShOp>
1837 : N3V<1, 1, op21_20, op11_8, 1, 0,
1839 (ins QPR:$src1, QPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1841 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1842 [(set (ResTy QPR:$dst),
1843 (ResTy (ShOp (ResTy QPR:$src1),
1844 (ResTy (MulOp QPR:$src2,
1845 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src3),
1847 class N3VQMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1848 string OpcodeStr, string Dt,
1849 ValueType ResTy, ValueType OpTy,
1850 SDNode MulOp, SDNode ShOp>
1851 : N3V<1, 1, op21_20, op11_8, 1, 0,
1853 (ins QPR:$src1, QPR:$src2, DPR_8:$src3, nohash_imm:$lane),
1855 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1856 [(set (ResTy QPR:$dst),
1857 (ResTy (ShOp (ResTy QPR:$src1),
1858 (ResTy (MulOp QPR:$src2,
1859 (ResTy (NEONvduplane (OpTy DPR_8:$src3),
1862 // Neon Intrinsic-Op instructions (VABA): double- and quad-register.
1863 class N3VDIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1864 InstrItinClass itin, string OpcodeStr, string Dt,
1865 ValueType Ty, Intrinsic IntOp, SDNode OpNode>
1866 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1867 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1868 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1869 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
1870 (Ty (IntOp (Ty DPR:$Vn), (Ty DPR:$Vm))))))]>;
1871 class N3VQIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1872 InstrItinClass itin, string OpcodeStr, string Dt,
1873 ValueType Ty, Intrinsic IntOp, SDNode OpNode>
1874 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1875 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
1876 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1877 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
1878 (Ty (IntOp (Ty QPR:$Vn), (Ty QPR:$Vm))))))]>;
1880 // Neon 3-argument intrinsics, both double- and quad-register.
1881 // The destination register is also used as the first source operand register.
1882 class N3VDInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1883 InstrItinClass itin, string OpcodeStr, string Dt,
1884 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1885 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1886 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3), N3RegFrm, itin,
1887 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1888 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src1),
1889 (OpTy DPR:$src2), (OpTy DPR:$src3))))]>;
1890 class N3VQInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1891 InstrItinClass itin, string OpcodeStr, string Dt,
1892 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1893 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1894 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3), N3RegFrm, itin,
1895 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1896 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src1),
1897 (OpTy QPR:$src2), (OpTy QPR:$src3))))]>;
1899 // Long Multiply-Add/Sub operations.
1900 class N3VLMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1901 InstrItinClass itin, string OpcodeStr, string Dt,
1902 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
1903 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1904 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1905 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1906 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
1907 (TyQ (MulOp (TyD DPR:$Vn),
1908 (TyD DPR:$Vm)))))]>;
1909 class N3VLMulOpSL<bit op24, bits<2> op21_20, bits<4> op11_8,
1910 InstrItinClass itin, string OpcodeStr, string Dt,
1911 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
1912 : N3V<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$dst),
1913 (ins QPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1915 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1917 (OpNode (TyQ QPR:$src1),
1918 (TyQ (MulOp (TyD DPR:$src2),
1919 (TyD (NEONvduplane (TyD DPR_VFP2:$src3),
1921 class N3VLMulOpSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1922 InstrItinClass itin, string OpcodeStr, string Dt,
1923 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
1924 : N3V<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$dst),
1925 (ins QPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane),
1927 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1929 (OpNode (TyQ QPR:$src1),
1930 (TyQ (MulOp (TyD DPR:$src2),
1931 (TyD (NEONvduplane (TyD DPR_8:$src3),
1934 // Long Intrinsic-Op vector operations with explicit extend (VABAL).
1935 class N3VLIntExtOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1936 InstrItinClass itin, string OpcodeStr, string Dt,
1937 ValueType TyQ, ValueType TyD, Intrinsic IntOp, SDNode ExtOp,
1939 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1940 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1941 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1942 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
1943 (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
1944 (TyD DPR:$Vm)))))))]>;
1946 // Neon Long 3-argument intrinsic. The destination register is
1947 // a quad-register and is also used as the first source operand register.
1948 class N3VLInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1949 InstrItinClass itin, string OpcodeStr, string Dt,
1950 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
1951 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1952 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1953 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1955 (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$Vn), (TyD DPR:$Vm))))]>;
1956 class N3VLInt3SL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1957 string OpcodeStr, string Dt,
1958 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1959 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1961 (ins QPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1963 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1964 [(set (ResTy QPR:$dst),
1965 (ResTy (IntOp (ResTy QPR:$src1),
1967 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src3),
1969 class N3VLInt3SL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1970 InstrItinClass itin, string OpcodeStr, string Dt,
1971 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1972 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1974 (ins QPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane),
1976 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1977 [(set (ResTy QPR:$dst),
1978 (ResTy (IntOp (ResTy QPR:$src1),
1980 (OpTy (NEONvduplane (OpTy DPR_8:$src3),
1983 // Narrowing 3-register intrinsics.
1984 class N3VNInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1985 string OpcodeStr, string Dt, ValueType TyD, ValueType TyQ,
1986 Intrinsic IntOp, bit Commutable>
1987 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1988 (outs DPR:$dst), (ins QPR:$src1, QPR:$src2), N3RegFrm, IIC_VBINi4D,
1989 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1990 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src1), (TyQ QPR:$src2))))]> {
1991 let isCommutable = Commutable;
1994 // Long 3-register operations.
1995 class N3VL<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1996 InstrItinClass itin, string OpcodeStr, string Dt,
1997 ValueType TyQ, ValueType TyD, SDNode OpNode, bit Commutable>
1998 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1999 (outs QPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
2000 OpcodeStr, Dt, "$dst, $src1, $src2", "",
2001 [(set QPR:$dst, (TyQ (OpNode (TyD DPR:$src1), (TyD DPR:$src2))))]> {
2002 let isCommutable = Commutable;
2004 class N3VLSL<bit op24, bits<2> op21_20, bits<4> op11_8,
2005 InstrItinClass itin, string OpcodeStr, string Dt,
2006 ValueType TyQ, ValueType TyD, SDNode OpNode>
2007 : N3V<op24, 1, op21_20, op11_8, 1, 0,
2008 (outs QPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
2009 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
2011 (TyQ (OpNode (TyD DPR:$src1),
2012 (TyD (NEONvduplane (TyD DPR_VFP2:$src2),imm:$lane)))))]>;
2013 class N3VLSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2014 InstrItinClass itin, string OpcodeStr, string Dt,
2015 ValueType TyQ, ValueType TyD, SDNode OpNode>
2016 : N3V<op24, 1, op21_20, op11_8, 1, 0,
2017 (outs QPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
2018 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
2020 (TyQ (OpNode (TyD DPR:$src1),
2021 (TyD (NEONvduplane (TyD DPR_8:$src2), imm:$lane)))))]>;
2023 // Long 3-register operations with explicitly extended operands.
2024 class N3VLExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2025 InstrItinClass itin, string OpcodeStr, string Dt,
2026 ValueType TyQ, ValueType TyD, SDNode OpNode, SDNode ExtOp,
2028 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2029 (outs QPR:$Qd), (ins DPR:$Dn, DPR:$Dm), N3RegFrm, itin,
2030 OpcodeStr, Dt, "$Qd, $Dn, $Dm", "",
2031 [(set QPR:$Qd, (OpNode (TyQ (ExtOp (TyD DPR:$Dn))),
2032 (TyQ (ExtOp (TyD DPR:$Dm)))))]> {
2033 let isCommutable = Commutable;
2036 // Long 3-register intrinsics with explicit extend (VABDL).
2037 class N3VLIntExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2038 InstrItinClass itin, string OpcodeStr, string Dt,
2039 ValueType TyQ, ValueType TyD, Intrinsic IntOp, SDNode ExtOp,
2041 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2042 (outs QPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
2043 OpcodeStr, Dt, "$dst, $src1, $src2", "",
2044 [(set QPR:$dst, (TyQ (ExtOp (TyD (IntOp (TyD DPR:$src1),
2045 (TyD DPR:$src2))))))]> {
2046 let isCommutable = Commutable;
2049 // Long 3-register intrinsics.
2050 class N3VLInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2051 InstrItinClass itin, string OpcodeStr, string Dt,
2052 ValueType TyQ, ValueType TyD, Intrinsic IntOp, bit Commutable>
2053 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2054 (outs QPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
2055 OpcodeStr, Dt, "$dst, $src1, $src2", "",
2056 [(set QPR:$dst, (TyQ (IntOp (TyD DPR:$src1), (TyD DPR:$src2))))]> {
2057 let isCommutable = Commutable;
2059 class N3VLIntSL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2060 string OpcodeStr, string Dt,
2061 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2062 : N3V<op24, 1, op21_20, op11_8, 1, 0,
2063 (outs QPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
2064 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
2065 [(set (ResTy QPR:$dst),
2066 (ResTy (IntOp (OpTy DPR:$src1),
2067 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src2),
2069 class N3VLIntSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2070 InstrItinClass itin, string OpcodeStr, string Dt,
2071 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2072 : N3V<op24, 1, op21_20, op11_8, 1, 0,
2073 (outs QPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
2074 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
2075 [(set (ResTy QPR:$dst),
2076 (ResTy (IntOp (OpTy DPR:$src1),
2077 (OpTy (NEONvduplane (OpTy DPR_8:$src2),
2080 // Wide 3-register operations.
2081 class N3VW<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2082 string OpcodeStr, string Dt, ValueType TyQ, ValueType TyD,
2083 SDNode OpNode, SDNode ExtOp, bit Commutable>
2084 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2085 (outs QPR:$Qd), (ins QPR:$Qn, DPR:$Dm), N3RegFrm, IIC_VSUBiD,
2086 OpcodeStr, Dt, "$Qd, $Qn, $Dm", "",
2087 [(set QPR:$Qd, (OpNode (TyQ QPR:$Qn),
2088 (TyQ (ExtOp (TyD DPR:$Dm)))))]> {
2089 let isCommutable = Commutable;
2092 // Pairwise long 2-register intrinsics, both double- and quad-register.
2093 class N2VDPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2094 bits<2> op17_16, bits<5> op11_7, bit op4,
2095 string OpcodeStr, string Dt,
2096 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2097 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
2098 (ins DPR:$src), IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
2099 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
2100 class N2VQPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2101 bits<2> op17_16, bits<5> op11_7, bit op4,
2102 string OpcodeStr, string Dt,
2103 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2104 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
2105 (ins QPR:$src), IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
2106 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
2108 // Pairwise long 2-register accumulate intrinsics,
2109 // both double- and quad-register.
2110 // The destination register is also used as the first source operand register.
2111 class N2VDPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2112 bits<2> op17_16, bits<5> op11_7, bit op4,
2113 string OpcodeStr, string Dt,
2114 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2115 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
2116 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vm), IIC_VPALiD,
2117 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
2118 [(set DPR:$Vd, (ResTy (IntOp (ResTy DPR:$src1), (OpTy DPR:$Vm))))]>;
2119 class N2VQPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2120 bits<2> op17_16, bits<5> op11_7, bit op4,
2121 string OpcodeStr, string Dt,
2122 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2123 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4,
2124 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vm), IIC_VPALiQ,
2125 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
2126 [(set QPR:$Vd, (ResTy (IntOp (ResTy QPR:$src1), (OpTy QPR:$Vm))))]>;
2128 // Shift by immediate,
2129 // both double- and quad-register.
2130 class N2VDSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2131 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2132 ValueType Ty, SDNode OpNode>
2133 : N2VImm<op24, op23, op11_8, op7, 0, op4,
2134 (outs DPR:$dst), (ins DPR:$src, i32imm:$SIMM), f, itin,
2135 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
2136 [(set DPR:$dst, (Ty (OpNode (Ty DPR:$src), (i32 imm:$SIMM))))]>;
2137 class N2VQSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2138 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2139 ValueType Ty, SDNode OpNode>
2140 : N2VImm<op24, op23, op11_8, op7, 1, op4,
2141 (outs QPR:$dst), (ins QPR:$src, i32imm:$SIMM), f, itin,
2142 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
2143 [(set QPR:$dst, (Ty (OpNode (Ty QPR:$src), (i32 imm:$SIMM))))]>;
2145 // Long shift by immediate.
2146 class N2VLSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
2147 string OpcodeStr, string Dt,
2148 ValueType ResTy, ValueType OpTy, SDNode OpNode>
2149 : N2VImm<op24, op23, op11_8, op7, op6, op4,
2150 (outs QPR:$dst), (ins DPR:$src, i32imm:$SIMM), N2RegVShLFrm,
2151 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src, $SIMM", "",
2152 [(set QPR:$dst, (ResTy (OpNode (OpTy DPR:$src),
2153 (i32 imm:$SIMM))))]>;
2155 // Narrow shift by immediate.
2156 class N2VNSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
2157 InstrItinClass itin, string OpcodeStr, string Dt,
2158 ValueType ResTy, ValueType OpTy, SDNode OpNode>
2159 : N2VImm<op24, op23, op11_8, op7, op6, op4,
2160 (outs DPR:$dst), (ins QPR:$src, i32imm:$SIMM), N2RegVShRFrm, itin,
2161 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
2162 [(set DPR:$dst, (ResTy (OpNode (OpTy QPR:$src),
2163 (i32 imm:$SIMM))))]>;
2165 // Shift right by immediate and accumulate,
2166 // both double- and quad-register.
2167 class N2VDShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2168 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
2169 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
2170 (ins DPR:$src1, DPR:$Vm, i32imm:$SIMM), N2RegVShRFrm, IIC_VPALiD,
2171 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2172 [(set DPR:$Vd, (Ty (add DPR:$src1,
2173 (Ty (ShOp DPR:$Vm, (i32 imm:$SIMM))))))]>;
2174 class N2VQShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2175 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
2176 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
2177 (ins QPR:$src1, QPR:$Vm, i32imm:$SIMM), N2RegVShRFrm, IIC_VPALiD,
2178 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2179 [(set QPR:$Vd, (Ty (add QPR:$src1,
2180 (Ty (ShOp QPR:$Vm, (i32 imm:$SIMM))))))]>;
2182 // Shift by immediate and insert,
2183 // both double- and quad-register.
2184 class N2VDShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2185 Format f, string OpcodeStr, string Dt, ValueType Ty,SDNode ShOp>
2186 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
2187 (ins DPR:$src1, DPR:$Vm, i32imm:$SIMM), f, IIC_VSHLiD,
2188 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2189 [(set DPR:$Vd, (Ty (ShOp DPR:$src1, DPR:$Vm, (i32 imm:$SIMM))))]>;
2190 class N2VQShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2191 Format f, string OpcodeStr, string Dt, ValueType Ty,SDNode ShOp>
2192 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
2193 (ins QPR:$src1, QPR:$Vm, i32imm:$SIMM), f, IIC_VSHLiQ,
2194 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2195 [(set QPR:$Vd, (Ty (ShOp QPR:$src1, QPR:$Vm, (i32 imm:$SIMM))))]>;
2197 // Convert, with fractional bits immediate,
2198 // both double- and quad-register.
2199 class N2VCvtD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2200 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2202 : N2VImm<op24, op23, op11_8, op7, 0, op4,
2203 (outs DPR:$Vd), (ins DPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
2204 IIC_VUNAD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2205 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (i32 imm:$SIMM))))]>;
2206 class N2VCvtQ<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2207 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2209 : N2VImm<op24, op23, op11_8, op7, 1, op4,
2210 (outs QPR:$Vd), (ins QPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
2211 IIC_VUNAQ, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2212 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (i32 imm:$SIMM))))]>;
2214 //===----------------------------------------------------------------------===//
2216 //===----------------------------------------------------------------------===//
2218 // Abbreviations used in multiclass suffixes:
2219 // Q = quarter int (8 bit) elements
2220 // H = half int (16 bit) elements
2221 // S = single int (32 bit) elements
2222 // D = double int (64 bit) elements
2224 // Neon 2-register vector operations -- for disassembly only.
2226 // First with only element sizes of 8, 16 and 32 bits:
2227 multiclass N2V_QHS_cmp<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2228 bits<5> op11_7, bit op4, string opc, string Dt,
2229 string asm, SDNode OpNode> {
2230 // 64-bit vector types.
2231 def v8i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 0, op4,
2232 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
2233 opc, !strconcat(Dt, "8"), asm, "",
2234 [(set DPR:$dst, (v8i8 (OpNode (v8i8 DPR:$src))))]>;
2235 def v4i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 0, op4,
2236 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
2237 opc, !strconcat(Dt, "16"), asm, "",
2238 [(set DPR:$dst, (v4i16 (OpNode (v4i16 DPR:$src))))]>;
2239 def v2i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
2240 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
2241 opc, !strconcat(Dt, "32"), asm, "",
2242 [(set DPR:$dst, (v2i32 (OpNode (v2i32 DPR:$src))))]>;
2243 def v2f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
2244 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
2245 opc, "f32", asm, "",
2246 [(set DPR:$dst, (v2f32 (OpNode (v2f32 DPR:$src))))]> {
2247 let Inst{10} = 1; // overwrite F = 1
2250 // 128-bit vector types.
2251 def v16i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 1, op4,
2252 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
2253 opc, !strconcat(Dt, "8"), asm, "",
2254 [(set QPR:$dst, (v16i8 (OpNode (v16i8 QPR:$src))))]>;
2255 def v8i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 1, op4,
2256 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
2257 opc, !strconcat(Dt, "16"), asm, "",
2258 [(set QPR:$dst, (v8i16 (OpNode (v8i16 QPR:$src))))]>;
2259 def v4i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
2260 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
2261 opc, !strconcat(Dt, "32"), asm, "",
2262 [(set QPR:$dst, (v4i32 (OpNode (v4i32 QPR:$src))))]>;
2263 def v4f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
2264 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
2265 opc, "f32", asm, "",
2266 [(set QPR:$dst, (v4f32 (OpNode (v4f32 QPR:$src))))]> {
2267 let Inst{10} = 1; // overwrite F = 1
2271 // Neon 3-register vector operations.
2273 // First with only element sizes of 8, 16 and 32 bits:
2274 multiclass N3V_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2275 InstrItinClass itinD16, InstrItinClass itinD32,
2276 InstrItinClass itinQ16, InstrItinClass itinQ32,
2277 string OpcodeStr, string Dt,
2278 SDNode OpNode, bit Commutable = 0> {
2279 // 64-bit vector types.
2280 def v8i8 : N3VD<op24, op23, 0b00, op11_8, op4, itinD16,
2281 OpcodeStr, !strconcat(Dt, "8"),
2282 v8i8, v8i8, OpNode, Commutable>;
2283 def v4i16 : N3VD<op24, op23, 0b01, op11_8, op4, itinD16,
2284 OpcodeStr, !strconcat(Dt, "16"),
2285 v4i16, v4i16, OpNode, Commutable>;
2286 def v2i32 : N3VD<op24, op23, 0b10, op11_8, op4, itinD32,
2287 OpcodeStr, !strconcat(Dt, "32"),
2288 v2i32, v2i32, OpNode, Commutable>;
2290 // 128-bit vector types.
2291 def v16i8 : N3VQ<op24, op23, 0b00, op11_8, op4, itinQ16,
2292 OpcodeStr, !strconcat(Dt, "8"),
2293 v16i8, v16i8, OpNode, Commutable>;
2294 def v8i16 : N3VQ<op24, op23, 0b01, op11_8, op4, itinQ16,
2295 OpcodeStr, !strconcat(Dt, "16"),
2296 v8i16, v8i16, OpNode, Commutable>;
2297 def v4i32 : N3VQ<op24, op23, 0b10, op11_8, op4, itinQ32,
2298 OpcodeStr, !strconcat(Dt, "32"),
2299 v4i32, v4i32, OpNode, Commutable>;
2302 multiclass N3VSL_HS<bits<4> op11_8, string OpcodeStr, string Dt, SDNode ShOp> {
2303 def v4i16 : N3VDSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
2305 def v2i32 : N3VDSL<0b10, op11_8, IIC_VMULi32D, OpcodeStr, !strconcat(Dt,"32"),
2307 def v8i16 : N3VQSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
2308 v8i16, v4i16, ShOp>;
2309 def v4i32 : N3VQSL<0b10, op11_8, IIC_VMULi32Q, OpcodeStr, !strconcat(Dt,"32"),
2310 v4i32, v2i32, ShOp>;
2313 // ....then also with element size 64 bits:
2314 multiclass N3V_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2315 InstrItinClass itinD, InstrItinClass itinQ,
2316 string OpcodeStr, string Dt,
2317 SDNode OpNode, bit Commutable = 0>
2318 : N3V_QHS<op24, op23, op11_8, op4, itinD, itinD, itinQ, itinQ,
2319 OpcodeStr, Dt, OpNode, Commutable> {
2320 def v1i64 : N3VD<op24, op23, 0b11, op11_8, op4, itinD,
2321 OpcodeStr, !strconcat(Dt, "64"),
2322 v1i64, v1i64, OpNode, Commutable>;
2323 def v2i64 : N3VQ<op24, op23, 0b11, op11_8, op4, itinQ,
2324 OpcodeStr, !strconcat(Dt, "64"),
2325 v2i64, v2i64, OpNode, Commutable>;
2329 // Neon Narrowing 2-register vector operations,
2330 // source operand element sizes of 16, 32 and 64 bits:
2331 multiclass N2VN_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2332 bits<5> op11_7, bit op6, bit op4,
2333 InstrItinClass itin, string OpcodeStr, string Dt,
2335 def v8i8 : N2VN<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
2336 itin, OpcodeStr, !strconcat(Dt, "16"),
2337 v8i8, v8i16, OpNode>;
2338 def v4i16 : N2VN<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
2339 itin, OpcodeStr, !strconcat(Dt, "32"),
2340 v4i16, v4i32, OpNode>;
2341 def v2i32 : N2VN<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
2342 itin, OpcodeStr, !strconcat(Dt, "64"),
2343 v2i32, v2i64, OpNode>;
2346 // Neon Narrowing 2-register vector intrinsics,
2347 // source operand element sizes of 16, 32 and 64 bits:
2348 multiclass N2VNInt_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2349 bits<5> op11_7, bit op6, bit op4,
2350 InstrItinClass itin, string OpcodeStr, string Dt,
2352 def v8i8 : N2VNInt<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
2353 itin, OpcodeStr, !strconcat(Dt, "16"),
2354 v8i8, v8i16, IntOp>;
2355 def v4i16 : N2VNInt<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
2356 itin, OpcodeStr, !strconcat(Dt, "32"),
2357 v4i16, v4i32, IntOp>;
2358 def v2i32 : N2VNInt<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
2359 itin, OpcodeStr, !strconcat(Dt, "64"),
2360 v2i32, v2i64, IntOp>;
2364 // Neon Lengthening 2-register vector intrinsic (currently specific to VMOVL).
2365 // source operand element sizes of 16, 32 and 64 bits:
2366 multiclass N2VL_QHS<bits<2> op24_23, bits<5> op11_7, bit op6, bit op4,
2367 string OpcodeStr, string Dt, SDNode OpNode> {
2368 def v8i16 : N2VL<op24_23, 0b00, 0b10, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2369 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode>;
2370 def v4i32 : N2VL<op24_23, 0b01, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2371 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
2372 def v2i64 : N2VL<op24_23, 0b10, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2373 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
2377 // Neon 3-register vector intrinsics.
2379 // First with only element sizes of 16 and 32 bits:
2380 multiclass N3VInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2381 InstrItinClass itinD16, InstrItinClass itinD32,
2382 InstrItinClass itinQ16, InstrItinClass itinQ32,
2383 string OpcodeStr, string Dt,
2384 Intrinsic IntOp, bit Commutable = 0> {
2385 // 64-bit vector types.
2386 def v4i16 : N3VDInt<op24, op23, 0b01, op11_8, op4, f, itinD16,
2387 OpcodeStr, !strconcat(Dt, "16"),
2388 v4i16, v4i16, IntOp, Commutable>;
2389 def v2i32 : N3VDInt<op24, op23, 0b10, op11_8, op4, f, itinD32,
2390 OpcodeStr, !strconcat(Dt, "32"),
2391 v2i32, v2i32, IntOp, Commutable>;
2393 // 128-bit vector types.
2394 def v8i16 : N3VQInt<op24, op23, 0b01, op11_8, op4, f, itinQ16,
2395 OpcodeStr, !strconcat(Dt, "16"),
2396 v8i16, v8i16, IntOp, Commutable>;
2397 def v4i32 : N3VQInt<op24, op23, 0b10, op11_8, op4, f, itinQ32,
2398 OpcodeStr, !strconcat(Dt, "32"),
2399 v4i32, v4i32, IntOp, Commutable>;
2401 multiclass N3VInt_HSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2402 InstrItinClass itinD16, InstrItinClass itinD32,
2403 InstrItinClass itinQ16, InstrItinClass itinQ32,
2404 string OpcodeStr, string Dt,
2406 // 64-bit vector types.
2407 def v4i16 : N3VDIntSh<op24, op23, 0b01, op11_8, op4, f, itinD16,
2408 OpcodeStr, !strconcat(Dt, "16"),
2409 v4i16, v4i16, IntOp>;
2410 def v2i32 : N3VDIntSh<op24, op23, 0b10, op11_8, op4, f, itinD32,
2411 OpcodeStr, !strconcat(Dt, "32"),
2412 v2i32, v2i32, IntOp>;
2414 // 128-bit vector types.
2415 def v8i16 : N3VQIntSh<op24, op23, 0b01, op11_8, op4, f, itinQ16,
2416 OpcodeStr, !strconcat(Dt, "16"),
2417 v8i16, v8i16, IntOp>;
2418 def v4i32 : N3VQIntSh<op24, op23, 0b10, op11_8, op4, f, itinQ32,
2419 OpcodeStr, !strconcat(Dt, "32"),
2420 v4i32, v4i32, IntOp>;
2423 multiclass N3VIntSL_HS<bits<4> op11_8,
2424 InstrItinClass itinD16, InstrItinClass itinD32,
2425 InstrItinClass itinQ16, InstrItinClass itinQ32,
2426 string OpcodeStr, string Dt, Intrinsic IntOp> {
2427 def v4i16 : N3VDIntSL16<0b01, op11_8, itinD16,
2428 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp>;
2429 def v2i32 : N3VDIntSL<0b10, op11_8, itinD32,
2430 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp>;
2431 def v8i16 : N3VQIntSL16<0b01, op11_8, itinQ16,
2432 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16, IntOp>;
2433 def v4i32 : N3VQIntSL<0b10, op11_8, itinQ32,
2434 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32, IntOp>;
2437 // ....then also with element size of 8 bits:
2438 multiclass N3VInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2439 InstrItinClass itinD16, InstrItinClass itinD32,
2440 InstrItinClass itinQ16, InstrItinClass itinQ32,
2441 string OpcodeStr, string Dt,
2442 Intrinsic IntOp, bit Commutable = 0>
2443 : N3VInt_HS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2444 OpcodeStr, Dt, IntOp, Commutable> {
2445 def v8i8 : N3VDInt<op24, op23, 0b00, op11_8, op4, f, itinD16,
2446 OpcodeStr, !strconcat(Dt, "8"),
2447 v8i8, v8i8, IntOp, Commutable>;
2448 def v16i8 : N3VQInt<op24, op23, 0b00, op11_8, op4, f, itinQ16,
2449 OpcodeStr, !strconcat(Dt, "8"),
2450 v16i8, v16i8, IntOp, Commutable>;
2452 multiclass N3VInt_QHSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2453 InstrItinClass itinD16, InstrItinClass itinD32,
2454 InstrItinClass itinQ16, InstrItinClass itinQ32,
2455 string OpcodeStr, string Dt,
2457 : N3VInt_HSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2458 OpcodeStr, Dt, IntOp> {
2459 def v8i8 : N3VDIntSh<op24, op23, 0b00, op11_8, op4, f, itinD16,
2460 OpcodeStr, !strconcat(Dt, "8"),
2462 def v16i8 : N3VQIntSh<op24, op23, 0b00, op11_8, op4, f, itinQ16,
2463 OpcodeStr, !strconcat(Dt, "8"),
2464 v16i8, v16i8, IntOp>;
2468 // ....then also with element size of 64 bits:
2469 multiclass N3VInt_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2470 InstrItinClass itinD16, InstrItinClass itinD32,
2471 InstrItinClass itinQ16, InstrItinClass itinQ32,
2472 string OpcodeStr, string Dt,
2473 Intrinsic IntOp, bit Commutable = 0>
2474 : N3VInt_QHS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2475 OpcodeStr, Dt, IntOp, Commutable> {
2476 def v1i64 : N3VDInt<op24, op23, 0b11, op11_8, op4, f, itinD32,
2477 OpcodeStr, !strconcat(Dt, "64"),
2478 v1i64, v1i64, IntOp, Commutable>;
2479 def v2i64 : N3VQInt<op24, op23, 0b11, op11_8, op4, f, itinQ32,
2480 OpcodeStr, !strconcat(Dt, "64"),
2481 v2i64, v2i64, IntOp, Commutable>;
2483 multiclass N3VInt_QHSDSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2484 InstrItinClass itinD16, InstrItinClass itinD32,
2485 InstrItinClass itinQ16, InstrItinClass itinQ32,
2486 string OpcodeStr, string Dt,
2488 : N3VInt_QHSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2489 OpcodeStr, Dt, IntOp> {
2490 def v1i64 : N3VDIntSh<op24, op23, 0b11, op11_8, op4, f, itinD32,
2491 OpcodeStr, !strconcat(Dt, "64"),
2492 v1i64, v1i64, IntOp>;
2493 def v2i64 : N3VQIntSh<op24, op23, 0b11, op11_8, op4, f, itinQ32,
2494 OpcodeStr, !strconcat(Dt, "64"),
2495 v2i64, v2i64, IntOp>;
2498 // Neon Narrowing 3-register vector intrinsics,
2499 // source operand element sizes of 16, 32 and 64 bits:
2500 multiclass N3VNInt_HSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2501 string OpcodeStr, string Dt,
2502 Intrinsic IntOp, bit Commutable = 0> {
2503 def v8i8 : N3VNInt<op24, op23, 0b00, op11_8, op4,
2504 OpcodeStr, !strconcat(Dt, "16"),
2505 v8i8, v8i16, IntOp, Commutable>;
2506 def v4i16 : N3VNInt<op24, op23, 0b01, op11_8, op4,
2507 OpcodeStr, !strconcat(Dt, "32"),
2508 v4i16, v4i32, IntOp, Commutable>;
2509 def v2i32 : N3VNInt<op24, op23, 0b10, op11_8, op4,
2510 OpcodeStr, !strconcat(Dt, "64"),
2511 v2i32, v2i64, IntOp, Commutable>;
2515 // Neon Long 3-register vector operations.
2517 multiclass N3VL_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2518 InstrItinClass itin16, InstrItinClass itin32,
2519 string OpcodeStr, string Dt,
2520 SDNode OpNode, bit Commutable = 0> {
2521 def v8i16 : N3VL<op24, op23, 0b00, op11_8, op4, itin16,
2522 OpcodeStr, !strconcat(Dt, "8"),
2523 v8i16, v8i8, OpNode, Commutable>;
2524 def v4i32 : N3VL<op24, op23, 0b01, op11_8, op4, itin16,
2525 OpcodeStr, !strconcat(Dt, "16"),
2526 v4i32, v4i16, OpNode, Commutable>;
2527 def v2i64 : N3VL<op24, op23, 0b10, op11_8, op4, itin32,
2528 OpcodeStr, !strconcat(Dt, "32"),
2529 v2i64, v2i32, OpNode, Commutable>;
2532 multiclass N3VLSL_HS<bit op24, bits<4> op11_8,
2533 InstrItinClass itin, string OpcodeStr, string Dt,
2535 def v4i16 : N3VLSL16<op24, 0b01, op11_8, itin, OpcodeStr,
2536 !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
2537 def v2i32 : N3VLSL<op24, 0b10, op11_8, itin, OpcodeStr,
2538 !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
2541 multiclass N3VLExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2542 InstrItinClass itin16, InstrItinClass itin32,
2543 string OpcodeStr, string Dt,
2544 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
2545 def v8i16 : N3VLExt<op24, op23, 0b00, op11_8, op4, itin16,
2546 OpcodeStr, !strconcat(Dt, "8"),
2547 v8i16, v8i8, OpNode, ExtOp, Commutable>;
2548 def v4i32 : N3VLExt<op24, op23, 0b01, op11_8, op4, itin16,
2549 OpcodeStr, !strconcat(Dt, "16"),
2550 v4i32, v4i16, OpNode, ExtOp, Commutable>;
2551 def v2i64 : N3VLExt<op24, op23, 0b10, op11_8, op4, itin32,
2552 OpcodeStr, !strconcat(Dt, "32"),
2553 v2i64, v2i32, OpNode, ExtOp, Commutable>;
2556 // Neon Long 3-register vector intrinsics.
2558 // First with only element sizes of 16 and 32 bits:
2559 multiclass N3VLInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
2560 InstrItinClass itin16, InstrItinClass itin32,
2561 string OpcodeStr, string Dt,
2562 Intrinsic IntOp, bit Commutable = 0> {
2563 def v4i32 : N3VLInt<op24, op23, 0b01, op11_8, op4, itin16,
2564 OpcodeStr, !strconcat(Dt, "16"),
2565 v4i32, v4i16, IntOp, Commutable>;
2566 def v2i64 : N3VLInt<op24, op23, 0b10, op11_8, op4, itin32,
2567 OpcodeStr, !strconcat(Dt, "32"),
2568 v2i64, v2i32, IntOp, Commutable>;
2571 multiclass N3VLIntSL_HS<bit op24, bits<4> op11_8,
2572 InstrItinClass itin, string OpcodeStr, string Dt,
2574 def v4i16 : N3VLIntSL16<op24, 0b01, op11_8, itin,
2575 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
2576 def v2i32 : N3VLIntSL<op24, 0b10, op11_8, itin,
2577 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
2580 // ....then also with element size of 8 bits:
2581 multiclass N3VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2582 InstrItinClass itin16, InstrItinClass itin32,
2583 string OpcodeStr, string Dt,
2584 Intrinsic IntOp, bit Commutable = 0>
2585 : N3VLInt_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt,
2586 IntOp, Commutable> {
2587 def v8i16 : N3VLInt<op24, op23, 0b00, op11_8, op4, itin16,
2588 OpcodeStr, !strconcat(Dt, "8"),
2589 v8i16, v8i8, IntOp, Commutable>;
2592 // ....with explicit extend (VABDL).
2593 multiclass N3VLIntExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2594 InstrItinClass itin, string OpcodeStr, string Dt,
2595 Intrinsic IntOp, SDNode ExtOp, bit Commutable = 0> {
2596 def v8i16 : N3VLIntExt<op24, op23, 0b00, op11_8, op4, itin,
2597 OpcodeStr, !strconcat(Dt, "8"),
2598 v8i16, v8i8, IntOp, ExtOp, Commutable>;
2599 def v4i32 : N3VLIntExt<op24, op23, 0b01, op11_8, op4, itin,
2600 OpcodeStr, !strconcat(Dt, "16"),
2601 v4i32, v4i16, IntOp, ExtOp, Commutable>;
2602 def v2i64 : N3VLIntExt<op24, op23, 0b10, op11_8, op4, itin,
2603 OpcodeStr, !strconcat(Dt, "32"),
2604 v2i64, v2i32, IntOp, ExtOp, Commutable>;
2608 // Neon Wide 3-register vector intrinsics,
2609 // source operand element sizes of 8, 16 and 32 bits:
2610 multiclass N3VW_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2611 string OpcodeStr, string Dt,
2612 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
2613 def v8i16 : N3VW<op24, op23, 0b00, op11_8, op4,
2614 OpcodeStr, !strconcat(Dt, "8"),
2615 v8i16, v8i8, OpNode, ExtOp, Commutable>;
2616 def v4i32 : N3VW<op24, op23, 0b01, op11_8, op4,
2617 OpcodeStr, !strconcat(Dt, "16"),
2618 v4i32, v4i16, OpNode, ExtOp, Commutable>;
2619 def v2i64 : N3VW<op24, op23, 0b10, op11_8, op4,
2620 OpcodeStr, !strconcat(Dt, "32"),
2621 v2i64, v2i32, OpNode, ExtOp, Commutable>;
2625 // Neon Multiply-Op vector operations,
2626 // element sizes of 8, 16 and 32 bits:
2627 multiclass N3VMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2628 InstrItinClass itinD16, InstrItinClass itinD32,
2629 InstrItinClass itinQ16, InstrItinClass itinQ32,
2630 string OpcodeStr, string Dt, SDNode OpNode> {
2631 // 64-bit vector types.
2632 def v8i8 : N3VDMulOp<op24, op23, 0b00, op11_8, op4, itinD16,
2633 OpcodeStr, !strconcat(Dt, "8"), v8i8, mul, OpNode>;
2634 def v4i16 : N3VDMulOp<op24, op23, 0b01, op11_8, op4, itinD16,
2635 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, OpNode>;
2636 def v2i32 : N3VDMulOp<op24, op23, 0b10, op11_8, op4, itinD32,
2637 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, OpNode>;
2639 // 128-bit vector types.
2640 def v16i8 : N3VQMulOp<op24, op23, 0b00, op11_8, op4, itinQ16,
2641 OpcodeStr, !strconcat(Dt, "8"), v16i8, mul, OpNode>;
2642 def v8i16 : N3VQMulOp<op24, op23, 0b01, op11_8, op4, itinQ16,
2643 OpcodeStr, !strconcat(Dt, "16"), v8i16, mul, OpNode>;
2644 def v4i32 : N3VQMulOp<op24, op23, 0b10, op11_8, op4, itinQ32,
2645 OpcodeStr, !strconcat(Dt, "32"), v4i32, mul, OpNode>;
2648 multiclass N3VMulOpSL_HS<bits<4> op11_8,
2649 InstrItinClass itinD16, InstrItinClass itinD32,
2650 InstrItinClass itinQ16, InstrItinClass itinQ32,
2651 string OpcodeStr, string Dt, SDNode ShOp> {
2652 def v4i16 : N3VDMulOpSL16<0b01, op11_8, itinD16,
2653 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, ShOp>;
2654 def v2i32 : N3VDMulOpSL<0b10, op11_8, itinD32,
2655 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, ShOp>;
2656 def v8i16 : N3VQMulOpSL16<0b01, op11_8, itinQ16,
2657 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16,
2659 def v4i32 : N3VQMulOpSL<0b10, op11_8, itinQ32,
2660 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32,
2664 // Neon Intrinsic-Op vector operations,
2665 // element sizes of 8, 16 and 32 bits:
2666 multiclass N3VIntOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2667 InstrItinClass itinD, InstrItinClass itinQ,
2668 string OpcodeStr, string Dt, Intrinsic IntOp,
2670 // 64-bit vector types.
2671 def v8i8 : N3VDIntOp<op24, op23, 0b00, op11_8, op4, itinD,
2672 OpcodeStr, !strconcat(Dt, "8"), v8i8, IntOp, OpNode>;
2673 def v4i16 : N3VDIntOp<op24, op23, 0b01, op11_8, op4, itinD,
2674 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp, OpNode>;
2675 def v2i32 : N3VDIntOp<op24, op23, 0b10, op11_8, op4, itinD,
2676 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp, OpNode>;
2678 // 128-bit vector types.
2679 def v16i8 : N3VQIntOp<op24, op23, 0b00, op11_8, op4, itinQ,
2680 OpcodeStr, !strconcat(Dt, "8"), v16i8, IntOp, OpNode>;
2681 def v8i16 : N3VQIntOp<op24, op23, 0b01, op11_8, op4, itinQ,
2682 OpcodeStr, !strconcat(Dt, "16"), v8i16, IntOp, OpNode>;
2683 def v4i32 : N3VQIntOp<op24, op23, 0b10, op11_8, op4, itinQ,
2684 OpcodeStr, !strconcat(Dt, "32"), v4i32, IntOp, OpNode>;
2687 // Neon 3-argument intrinsics,
2688 // element sizes of 8, 16 and 32 bits:
2689 multiclass N3VInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2690 InstrItinClass itinD, InstrItinClass itinQ,
2691 string OpcodeStr, string Dt, Intrinsic IntOp> {
2692 // 64-bit vector types.
2693 def v8i8 : N3VDInt3<op24, op23, 0b00, op11_8, op4, itinD,
2694 OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
2695 def v4i16 : N3VDInt3<op24, op23, 0b01, op11_8, op4, itinD,
2696 OpcodeStr, !strconcat(Dt, "16"), v4i16, v4i16, IntOp>;
2697 def v2i32 : N3VDInt3<op24, op23, 0b10, op11_8, op4, itinD,
2698 OpcodeStr, !strconcat(Dt, "32"), v2i32, v2i32, IntOp>;
2700 // 128-bit vector types.
2701 def v16i8 : N3VQInt3<op24, op23, 0b00, op11_8, op4, itinQ,
2702 OpcodeStr, !strconcat(Dt, "8"), v16i8, v16i8, IntOp>;
2703 def v8i16 : N3VQInt3<op24, op23, 0b01, op11_8, op4, itinQ,
2704 OpcodeStr, !strconcat(Dt, "16"), v8i16, v8i16, IntOp>;
2705 def v4i32 : N3VQInt3<op24, op23, 0b10, op11_8, op4, itinQ,
2706 OpcodeStr, !strconcat(Dt, "32"), v4i32, v4i32, IntOp>;
2710 // Neon Long Multiply-Op vector operations,
2711 // element sizes of 8, 16 and 32 bits:
2712 multiclass N3VLMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2713 InstrItinClass itin16, InstrItinClass itin32,
2714 string OpcodeStr, string Dt, SDNode MulOp,
2716 def v8i16 : N3VLMulOp<op24, op23, 0b00, op11_8, op4, itin16, OpcodeStr,
2717 !strconcat(Dt, "8"), v8i16, v8i8, MulOp, OpNode>;
2718 def v4i32 : N3VLMulOp<op24, op23, 0b01, op11_8, op4, itin16, OpcodeStr,
2719 !strconcat(Dt, "16"), v4i32, v4i16, MulOp, OpNode>;
2720 def v2i64 : N3VLMulOp<op24, op23, 0b10, op11_8, op4, itin32, OpcodeStr,
2721 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
2724 multiclass N3VLMulOpSL_HS<bit op24, bits<4> op11_8, string OpcodeStr,
2725 string Dt, SDNode MulOp, SDNode OpNode> {
2726 def v4i16 : N3VLMulOpSL16<op24, 0b01, op11_8, IIC_VMACi16D, OpcodeStr,
2727 !strconcat(Dt,"16"), v4i32, v4i16, MulOp, OpNode>;
2728 def v2i32 : N3VLMulOpSL<op24, 0b10, op11_8, IIC_VMACi32D, OpcodeStr,
2729 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
2733 // Neon Long 3-argument intrinsics.
2735 // First with only element sizes of 16 and 32 bits:
2736 multiclass N3VLInt3_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
2737 InstrItinClass itin16, InstrItinClass itin32,
2738 string OpcodeStr, string Dt, Intrinsic IntOp> {
2739 def v4i32 : N3VLInt3<op24, op23, 0b01, op11_8, op4, itin16,
2740 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
2741 def v2i64 : N3VLInt3<op24, op23, 0b10, op11_8, op4, itin32,
2742 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
2745 multiclass N3VLInt3SL_HS<bit op24, bits<4> op11_8,
2746 string OpcodeStr, string Dt, Intrinsic IntOp> {
2747 def v4i16 : N3VLInt3SL16<op24, 0b01, op11_8, IIC_VMACi16D,
2748 OpcodeStr, !strconcat(Dt,"16"), v4i32, v4i16, IntOp>;
2749 def v2i32 : N3VLInt3SL<op24, 0b10, op11_8, IIC_VMACi32D,
2750 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
2753 // ....then also with element size of 8 bits:
2754 multiclass N3VLInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2755 InstrItinClass itin16, InstrItinClass itin32,
2756 string OpcodeStr, string Dt, Intrinsic IntOp>
2757 : N3VLInt3_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt, IntOp> {
2758 def v8i16 : N3VLInt3<op24, op23, 0b00, op11_8, op4, itin16,
2759 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
2762 // ....with explicit extend (VABAL).
2763 multiclass N3VLIntExtOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2764 InstrItinClass itin, string OpcodeStr, string Dt,
2765 Intrinsic IntOp, SDNode ExtOp, SDNode OpNode> {
2766 def v8i16 : N3VLIntExtOp<op24, op23, 0b00, op11_8, op4, itin,
2767 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8,
2768 IntOp, ExtOp, OpNode>;
2769 def v4i32 : N3VLIntExtOp<op24, op23, 0b01, op11_8, op4, itin,
2770 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16,
2771 IntOp, ExtOp, OpNode>;
2772 def v2i64 : N3VLIntExtOp<op24, op23, 0b10, op11_8, op4, itin,
2773 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32,
2774 IntOp, ExtOp, OpNode>;
2778 // Neon 2-register vector intrinsics,
2779 // element sizes of 8, 16 and 32 bits:
2780 multiclass N2VInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2781 bits<5> op11_7, bit op4,
2782 InstrItinClass itinD, InstrItinClass itinQ,
2783 string OpcodeStr, string Dt, Intrinsic IntOp> {
2784 // 64-bit vector types.
2785 def v8i8 : N2VDInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2786 itinD, OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
2787 def v4i16 : N2VDInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2788 itinD, OpcodeStr, !strconcat(Dt, "16"),v4i16,v4i16,IntOp>;
2789 def v2i32 : N2VDInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2790 itinD, OpcodeStr, !strconcat(Dt, "32"),v2i32,v2i32,IntOp>;
2792 // 128-bit vector types.
2793 def v16i8 : N2VQInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2794 itinQ, OpcodeStr, !strconcat(Dt, "8"), v16i8,v16i8,IntOp>;
2795 def v8i16 : N2VQInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2796 itinQ, OpcodeStr, !strconcat(Dt, "16"),v8i16,v8i16,IntOp>;
2797 def v4i32 : N2VQInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2798 itinQ, OpcodeStr, !strconcat(Dt, "32"),v4i32,v4i32,IntOp>;
2802 // Neon Pairwise long 2-register intrinsics,
2803 // element sizes of 8, 16 and 32 bits:
2804 multiclass N2VPLInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2805 bits<5> op11_7, bit op4,
2806 string OpcodeStr, string Dt, Intrinsic IntOp> {
2807 // 64-bit vector types.
2808 def v8i8 : N2VDPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2809 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
2810 def v4i16 : N2VDPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2811 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
2812 def v2i32 : N2VDPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2813 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
2815 // 128-bit vector types.
2816 def v16i8 : N2VQPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2817 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
2818 def v8i16 : N2VQPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2819 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
2820 def v4i32 : N2VQPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2821 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
2825 // Neon Pairwise long 2-register accumulate intrinsics,
2826 // element sizes of 8, 16 and 32 bits:
2827 multiclass N2VPLInt2_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2828 bits<5> op11_7, bit op4,
2829 string OpcodeStr, string Dt, Intrinsic IntOp> {
2830 // 64-bit vector types.
2831 def v8i8 : N2VDPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2832 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
2833 def v4i16 : N2VDPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2834 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
2835 def v2i32 : N2VDPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2836 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
2838 // 128-bit vector types.
2839 def v16i8 : N2VQPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2840 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
2841 def v8i16 : N2VQPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2842 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
2843 def v4i32 : N2VQPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2844 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
2848 // Neon 2-register vector shift by immediate,
2849 // with f of either N2RegVShLFrm or N2RegVShRFrm
2850 // element sizes of 8, 16, 32 and 64 bits:
2851 multiclass N2VSh_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2852 InstrItinClass itin, string OpcodeStr, string Dt,
2853 SDNode OpNode, Format f> {
2854 // 64-bit vector types.
2855 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
2856 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
2857 let Inst{21-19} = 0b001; // imm6 = 001xxx
2859 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
2860 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
2861 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2863 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
2864 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
2865 let Inst{21} = 0b1; // imm6 = 1xxxxx
2867 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, f, itin,
2868 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
2871 // 128-bit vector types.
2872 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
2873 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
2874 let Inst{21-19} = 0b001; // imm6 = 001xxx
2876 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
2877 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
2878 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2880 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
2881 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
2882 let Inst{21} = 0b1; // imm6 = 1xxxxx
2884 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, f, itin,
2885 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
2889 // Neon Shift-Accumulate vector operations,
2890 // element sizes of 8, 16, 32 and 64 bits:
2891 multiclass N2VShAdd_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2892 string OpcodeStr, string Dt, SDNode ShOp> {
2893 // 64-bit vector types.
2894 def v8i8 : N2VDShAdd<op24, op23, op11_8, 0, op4,
2895 OpcodeStr, !strconcat(Dt, "8"), v8i8, ShOp> {
2896 let Inst{21-19} = 0b001; // imm6 = 001xxx
2898 def v4i16 : N2VDShAdd<op24, op23, op11_8, 0, op4,
2899 OpcodeStr, !strconcat(Dt, "16"), v4i16, ShOp> {
2900 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2902 def v2i32 : N2VDShAdd<op24, op23, op11_8, 0, op4,
2903 OpcodeStr, !strconcat(Dt, "32"), v2i32, ShOp> {
2904 let Inst{21} = 0b1; // imm6 = 1xxxxx
2906 def v1i64 : N2VDShAdd<op24, op23, op11_8, 1, op4,
2907 OpcodeStr, !strconcat(Dt, "64"), v1i64, ShOp>;
2910 // 128-bit vector types.
2911 def v16i8 : N2VQShAdd<op24, op23, op11_8, 0, op4,
2912 OpcodeStr, !strconcat(Dt, "8"), v16i8, ShOp> {
2913 let Inst{21-19} = 0b001; // imm6 = 001xxx
2915 def v8i16 : N2VQShAdd<op24, op23, op11_8, 0, op4,
2916 OpcodeStr, !strconcat(Dt, "16"), v8i16, ShOp> {
2917 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2919 def v4i32 : N2VQShAdd<op24, op23, op11_8, 0, op4,
2920 OpcodeStr, !strconcat(Dt, "32"), v4i32, ShOp> {
2921 let Inst{21} = 0b1; // imm6 = 1xxxxx
2923 def v2i64 : N2VQShAdd<op24, op23, op11_8, 1, op4,
2924 OpcodeStr, !strconcat(Dt, "64"), v2i64, ShOp>;
2929 // Neon Shift-Insert vector operations,
2930 // with f of either N2RegVShLFrm or N2RegVShRFrm
2931 // element sizes of 8, 16, 32 and 64 bits:
2932 multiclass N2VShIns_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2933 string OpcodeStr, SDNode ShOp,
2935 // 64-bit vector types.
2936 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4,
2937 f, OpcodeStr, "8", v8i8, ShOp> {
2938 let Inst{21-19} = 0b001; // imm6 = 001xxx
2940 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4,
2941 f, OpcodeStr, "16", v4i16, ShOp> {
2942 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2944 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4,
2945 f, OpcodeStr, "32", v2i32, ShOp> {
2946 let Inst{21} = 0b1; // imm6 = 1xxxxx
2948 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4,
2949 f, OpcodeStr, "64", v1i64, ShOp>;
2952 // 128-bit vector types.
2953 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4,
2954 f, OpcodeStr, "8", v16i8, ShOp> {
2955 let Inst{21-19} = 0b001; // imm6 = 001xxx
2957 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4,
2958 f, OpcodeStr, "16", v8i16, ShOp> {
2959 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2961 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4,
2962 f, OpcodeStr, "32", v4i32, ShOp> {
2963 let Inst{21} = 0b1; // imm6 = 1xxxxx
2965 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4,
2966 f, OpcodeStr, "64", v2i64, ShOp>;
2970 // Neon Shift Long operations,
2971 // element sizes of 8, 16, 32 bits:
2972 multiclass N2VLSh_QHS<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
2973 bit op4, string OpcodeStr, string Dt, SDNode OpNode> {
2974 def v8i16 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
2975 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode> {
2976 let Inst{21-19} = 0b001; // imm6 = 001xxx
2978 def v4i32 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
2979 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode> {
2980 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2982 def v2i64 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
2983 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode> {
2984 let Inst{21} = 0b1; // imm6 = 1xxxxx
2988 // Neon Shift Narrow operations,
2989 // element sizes of 16, 32, 64 bits:
2990 multiclass N2VNSh_HSD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
2991 bit op4, InstrItinClass itin, string OpcodeStr, string Dt,
2993 def v8i8 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
2994 OpcodeStr, !strconcat(Dt, "16"), v8i8, v8i16, OpNode> {
2995 let Inst{21-19} = 0b001; // imm6 = 001xxx
2997 def v4i16 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
2998 OpcodeStr, !strconcat(Dt, "32"), v4i16, v4i32, OpNode> {
2999 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3001 def v2i32 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3002 OpcodeStr, !strconcat(Dt, "64"), v2i32, v2i64, OpNode> {
3003 let Inst{21} = 0b1; // imm6 = 1xxxxx
3007 //===----------------------------------------------------------------------===//
3008 // Instruction Definitions.
3009 //===----------------------------------------------------------------------===//
3011 // Vector Add Operations.
3013 // VADD : Vector Add (integer and floating-point)
3014 defm VADD : N3V_QHSD<0, 0, 0b1000, 0, IIC_VBINiD, IIC_VBINiQ, "vadd", "i",
3016 def VADDfd : N3VD<0, 0, 0b00, 0b1101, 0, IIC_VBIND, "vadd", "f32",
3017 v2f32, v2f32, fadd, 1>;
3018 def VADDfq : N3VQ<0, 0, 0b00, 0b1101, 0, IIC_VBINQ, "vadd", "f32",
3019 v4f32, v4f32, fadd, 1>;
3020 // VADDL : Vector Add Long (Q = D + D)
3021 defm VADDLs : N3VLExt_QHS<0,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
3022 "vaddl", "s", add, sext, 1>;
3023 defm VADDLu : N3VLExt_QHS<1,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
3024 "vaddl", "u", add, zext, 1>;
3025 // VADDW : Vector Add Wide (Q = Q + D)
3026 defm VADDWs : N3VW_QHS<0,1,0b0001,0, "vaddw", "s", add, sext, 0>;
3027 defm VADDWu : N3VW_QHS<1,1,0b0001,0, "vaddw", "u", add, zext, 0>;
3028 // VHADD : Vector Halving Add
3029 defm VHADDs : N3VInt_QHS<0, 0, 0b0000, 0, N3RegFrm,
3030 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3031 "vhadd", "s", int_arm_neon_vhadds, 1>;
3032 defm VHADDu : N3VInt_QHS<1, 0, 0b0000, 0, N3RegFrm,
3033 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3034 "vhadd", "u", int_arm_neon_vhaddu, 1>;
3035 // VRHADD : Vector Rounding Halving Add
3036 defm VRHADDs : N3VInt_QHS<0, 0, 0b0001, 0, N3RegFrm,
3037 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3038 "vrhadd", "s", int_arm_neon_vrhadds, 1>;
3039 defm VRHADDu : N3VInt_QHS<1, 0, 0b0001, 0, N3RegFrm,
3040 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3041 "vrhadd", "u", int_arm_neon_vrhaddu, 1>;
3042 // VQADD : Vector Saturating Add
3043 defm VQADDs : N3VInt_QHSD<0, 0, 0b0000, 1, N3RegFrm,
3044 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3045 "vqadd", "s", int_arm_neon_vqadds, 1>;
3046 defm VQADDu : N3VInt_QHSD<1, 0, 0b0000, 1, N3RegFrm,
3047 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3048 "vqadd", "u", int_arm_neon_vqaddu, 1>;
3049 // VADDHN : Vector Add and Narrow Returning High Half (D = Q + Q)
3050 defm VADDHN : N3VNInt_HSD<0,1,0b0100,0, "vaddhn", "i",
3051 int_arm_neon_vaddhn, 1>;
3052 // VRADDHN : Vector Rounding Add and Narrow Returning High Half (D = Q + Q)
3053 defm VRADDHN : N3VNInt_HSD<1,1,0b0100,0, "vraddhn", "i",
3054 int_arm_neon_vraddhn, 1>;
3056 // Vector Multiply Operations.
3058 // VMUL : Vector Multiply (integer, polynomial and floating-point)
3059 defm VMUL : N3V_QHS<0, 0, 0b1001, 1, IIC_VMULi16D, IIC_VMULi32D,
3060 IIC_VMULi16Q, IIC_VMULi32Q, "vmul", "i", mul, 1>;
3061 def VMULpd : N3VDInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16D, "vmul",
3062 "p8", v8i8, v8i8, int_arm_neon_vmulp, 1>;
3063 def VMULpq : N3VQInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16Q, "vmul",
3064 "p8", v16i8, v16i8, int_arm_neon_vmulp, 1>;
3065 def VMULfd : N3VD<1, 0, 0b00, 0b1101, 1, IIC_VFMULD, "vmul", "f32",
3066 v2f32, v2f32, fmul, 1>;
3067 def VMULfq : N3VQ<1, 0, 0b00, 0b1101, 1, IIC_VFMULQ, "vmul", "f32",
3068 v4f32, v4f32, fmul, 1>;
3069 defm VMULsl : N3VSL_HS<0b1000, "vmul", "i", mul>;
3070 def VMULslfd : N3VDSL<0b10, 0b1001, IIC_VBIND, "vmul", "f32", v2f32, fmul>;
3071 def VMULslfq : N3VQSL<0b10, 0b1001, IIC_VBINQ, "vmul", "f32", v4f32,
3074 def : Pat<(v8i16 (mul (v8i16 QPR:$src1),
3075 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
3076 (v8i16 (VMULslv8i16 (v8i16 QPR:$src1),
3077 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3078 (DSubReg_i16_reg imm:$lane))),
3079 (SubReg_i16_lane imm:$lane)))>;
3080 def : Pat<(v4i32 (mul (v4i32 QPR:$src1),
3081 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
3082 (v4i32 (VMULslv4i32 (v4i32 QPR:$src1),
3083 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3084 (DSubReg_i32_reg imm:$lane))),
3085 (SubReg_i32_lane imm:$lane)))>;
3086 def : Pat<(v4f32 (fmul (v4f32 QPR:$src1),
3087 (v4f32 (NEONvduplane (v4f32 QPR:$src2), imm:$lane)))),
3088 (v4f32 (VMULslfq (v4f32 QPR:$src1),
3089 (v2f32 (EXTRACT_SUBREG QPR:$src2,
3090 (DSubReg_i32_reg imm:$lane))),
3091 (SubReg_i32_lane imm:$lane)))>;
3093 // VQDMULH : Vector Saturating Doubling Multiply Returning High Half
3094 defm VQDMULH : N3VInt_HS<0, 0, 0b1011, 0, N3RegFrm, IIC_VMULi16D, IIC_VMULi32D,
3095 IIC_VMULi16Q, IIC_VMULi32Q,
3096 "vqdmulh", "s", int_arm_neon_vqdmulh, 1>;
3097 defm VQDMULHsl: N3VIntSL_HS<0b1100, IIC_VMULi16D, IIC_VMULi32D,
3098 IIC_VMULi16Q, IIC_VMULi32Q,
3099 "vqdmulh", "s", int_arm_neon_vqdmulh>;
3100 def : Pat<(v8i16 (int_arm_neon_vqdmulh (v8i16 QPR:$src1),
3101 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
3103 (v8i16 (VQDMULHslv8i16 (v8i16 QPR:$src1),
3104 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3105 (DSubReg_i16_reg imm:$lane))),
3106 (SubReg_i16_lane imm:$lane)))>;
3107 def : Pat<(v4i32 (int_arm_neon_vqdmulh (v4i32 QPR:$src1),
3108 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
3110 (v4i32 (VQDMULHslv4i32 (v4i32 QPR:$src1),
3111 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3112 (DSubReg_i32_reg imm:$lane))),
3113 (SubReg_i32_lane imm:$lane)))>;
3115 // VQRDMULH : Vector Rounding Saturating Doubling Multiply Returning High Half
3116 defm VQRDMULH : N3VInt_HS<1, 0, 0b1011, 0, N3RegFrm,
3117 IIC_VMULi16D,IIC_VMULi32D,IIC_VMULi16Q,IIC_VMULi32Q,
3118 "vqrdmulh", "s", int_arm_neon_vqrdmulh, 1>;
3119 defm VQRDMULHsl : N3VIntSL_HS<0b1101, IIC_VMULi16D, IIC_VMULi32D,
3120 IIC_VMULi16Q, IIC_VMULi32Q,
3121 "vqrdmulh", "s", int_arm_neon_vqrdmulh>;
3122 def : Pat<(v8i16 (int_arm_neon_vqrdmulh (v8i16 QPR:$src1),
3123 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
3125 (v8i16 (VQRDMULHslv8i16 (v8i16 QPR:$src1),
3126 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3127 (DSubReg_i16_reg imm:$lane))),
3128 (SubReg_i16_lane imm:$lane)))>;
3129 def : Pat<(v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$src1),
3130 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
3132 (v4i32 (VQRDMULHslv4i32 (v4i32 QPR:$src1),
3133 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3134 (DSubReg_i32_reg imm:$lane))),
3135 (SubReg_i32_lane imm:$lane)))>;
3137 // VMULL : Vector Multiply Long (integer and polynomial) (Q = D * D)
3138 defm VMULLs : N3VL_QHS<0,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
3139 "vmull", "s", NEONvmulls, 1>;
3140 defm VMULLu : N3VL_QHS<1,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
3141 "vmull", "u", NEONvmullu, 1>;
3142 def VMULLp : N3VLInt<0, 1, 0b00, 0b1110, 0, IIC_VMULi16D, "vmull", "p8",
3143 v8i16, v8i8, int_arm_neon_vmullp, 1>;
3144 defm VMULLsls : N3VLSL_HS<0, 0b1010, IIC_VMULi16D, "vmull", "s", NEONvmulls>;
3145 defm VMULLslu : N3VLSL_HS<1, 0b1010, IIC_VMULi16D, "vmull", "u", NEONvmullu>;
3147 // VQDMULL : Vector Saturating Doubling Multiply Long (Q = D * D)
3148 defm VQDMULL : N3VLInt_HS<0,1,0b1101,0, IIC_VMULi16D, IIC_VMULi32D,
3149 "vqdmull", "s", int_arm_neon_vqdmull, 1>;
3150 defm VQDMULLsl: N3VLIntSL_HS<0, 0b1011, IIC_VMULi16D,
3151 "vqdmull", "s", int_arm_neon_vqdmull>;
3153 // Vector Multiply-Accumulate and Multiply-Subtract Operations.
3155 // VMLA : Vector Multiply Accumulate (integer and floating-point)
3156 defm VMLA : N3VMulOp_QHS<0, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3157 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
3158 def VMLAfd : N3VDMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACD, "vmla", "f32",
3160 def VMLAfq : N3VQMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACQ, "vmla", "f32",
3162 defm VMLAsl : N3VMulOpSL_HS<0b0000, IIC_VMACi16D, IIC_VMACi32D,
3163 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
3164 def VMLAslfd : N3VDMulOpSL<0b10, 0b0001, IIC_VMACD, "vmla", "f32",
3166 def VMLAslfq : N3VQMulOpSL<0b10, 0b0001, IIC_VMACQ, "vmla", "f32",
3167 v4f32, v2f32, fmul, fadd>;
3169 def : Pat<(v8i16 (add (v8i16 QPR:$src1),
3170 (mul (v8i16 QPR:$src2),
3171 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
3172 (v8i16 (VMLAslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
3173 (v4i16 (EXTRACT_SUBREG QPR:$src3,
3174 (DSubReg_i16_reg imm:$lane))),
3175 (SubReg_i16_lane imm:$lane)))>;
3177 def : Pat<(v4i32 (add (v4i32 QPR:$src1),
3178 (mul (v4i32 QPR:$src2),
3179 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
3180 (v4i32 (VMLAslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
3181 (v2i32 (EXTRACT_SUBREG QPR:$src3,
3182 (DSubReg_i32_reg imm:$lane))),
3183 (SubReg_i32_lane imm:$lane)))>;
3185 def : Pat<(v4f32 (fadd (v4f32 QPR:$src1),
3186 (fmul (v4f32 QPR:$src2),
3187 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
3188 (v4f32 (VMLAslfq (v4f32 QPR:$src1),
3190 (v2f32 (EXTRACT_SUBREG QPR:$src3,
3191 (DSubReg_i32_reg imm:$lane))),
3192 (SubReg_i32_lane imm:$lane)))>;
3194 // VMLAL : Vector Multiply Accumulate Long (Q += D * D)
3195 defm VMLALs : N3VLMulOp_QHS<0,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
3196 "vmlal", "s", NEONvmulls, add>;
3197 defm VMLALu : N3VLMulOp_QHS<1,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
3198 "vmlal", "u", NEONvmullu, add>;
3200 defm VMLALsls : N3VLMulOpSL_HS<0, 0b0010, "vmlal", "s", NEONvmulls, add>;
3201 defm VMLALslu : N3VLMulOpSL_HS<1, 0b0010, "vmlal", "u", NEONvmullu, add>;
3203 // VQDMLAL : Vector Saturating Doubling Multiply Accumulate Long (Q += D * D)
3204 defm VQDMLAL : N3VLInt3_HS<0, 1, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3205 "vqdmlal", "s", int_arm_neon_vqdmlal>;
3206 defm VQDMLALsl: N3VLInt3SL_HS<0, 0b0011, "vqdmlal", "s", int_arm_neon_vqdmlal>;
3208 // VMLS : Vector Multiply Subtract (integer and floating-point)
3209 defm VMLS : N3VMulOp_QHS<1, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3210 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
3211 def VMLSfd : N3VDMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACD, "vmls", "f32",
3213 def VMLSfq : N3VQMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACQ, "vmls", "f32",
3215 defm VMLSsl : N3VMulOpSL_HS<0b0100, IIC_VMACi16D, IIC_VMACi32D,
3216 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
3217 def VMLSslfd : N3VDMulOpSL<0b10, 0b0101, IIC_VMACD, "vmls", "f32",
3219 def VMLSslfq : N3VQMulOpSL<0b10, 0b0101, IIC_VMACQ, "vmls", "f32",
3220 v4f32, v2f32, fmul, fsub>;
3222 def : Pat<(v8i16 (sub (v8i16 QPR:$src1),
3223 (mul (v8i16 QPR:$src2),
3224 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
3225 (v8i16 (VMLSslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
3226 (v4i16 (EXTRACT_SUBREG QPR:$src3,
3227 (DSubReg_i16_reg imm:$lane))),
3228 (SubReg_i16_lane imm:$lane)))>;
3230 def : Pat<(v4i32 (sub (v4i32 QPR:$src1),
3231 (mul (v4i32 QPR:$src2),
3232 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
3233 (v4i32 (VMLSslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
3234 (v2i32 (EXTRACT_SUBREG QPR:$src3,
3235 (DSubReg_i32_reg imm:$lane))),
3236 (SubReg_i32_lane imm:$lane)))>;
3238 def : Pat<(v4f32 (fsub (v4f32 QPR:$src1),
3239 (fmul (v4f32 QPR:$src2),
3240 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
3241 (v4f32 (VMLSslfq (v4f32 QPR:$src1), (v4f32 QPR:$src2),
3242 (v2f32 (EXTRACT_SUBREG QPR:$src3,
3243 (DSubReg_i32_reg imm:$lane))),
3244 (SubReg_i32_lane imm:$lane)))>;
3246 // VMLSL : Vector Multiply Subtract Long (Q -= D * D)
3247 defm VMLSLs : N3VLMulOp_QHS<0,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
3248 "vmlsl", "s", NEONvmulls, sub>;
3249 defm VMLSLu : N3VLMulOp_QHS<1,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
3250 "vmlsl", "u", NEONvmullu, sub>;
3252 defm VMLSLsls : N3VLMulOpSL_HS<0, 0b0110, "vmlsl", "s", NEONvmulls, sub>;
3253 defm VMLSLslu : N3VLMulOpSL_HS<1, 0b0110, "vmlsl", "u", NEONvmullu, sub>;
3255 // VQDMLSL : Vector Saturating Doubling Multiply Subtract Long (Q -= D * D)
3256 defm VQDMLSL : N3VLInt3_HS<0, 1, 0b1011, 0, IIC_VMACi16D, IIC_VMACi32D,
3257 "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
3258 defm VQDMLSLsl: N3VLInt3SL_HS<0, 0b111, "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
3260 // Vector Subtract Operations.
3262 // VSUB : Vector Subtract (integer and floating-point)
3263 defm VSUB : N3V_QHSD<1, 0, 0b1000, 0, IIC_VSUBiD, IIC_VSUBiQ,
3264 "vsub", "i", sub, 0>;
3265 def VSUBfd : N3VD<0, 0, 0b10, 0b1101, 0, IIC_VBIND, "vsub", "f32",
3266 v2f32, v2f32, fsub, 0>;
3267 def VSUBfq : N3VQ<0, 0, 0b10, 0b1101, 0, IIC_VBINQ, "vsub", "f32",
3268 v4f32, v4f32, fsub, 0>;
3269 // VSUBL : Vector Subtract Long (Q = D - D)
3270 defm VSUBLs : N3VLExt_QHS<0,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
3271 "vsubl", "s", sub, sext, 0>;
3272 defm VSUBLu : N3VLExt_QHS<1,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
3273 "vsubl", "u", sub, zext, 0>;
3274 // VSUBW : Vector Subtract Wide (Q = Q - D)
3275 defm VSUBWs : N3VW_QHS<0,1,0b0011,0, "vsubw", "s", sub, sext, 0>;
3276 defm VSUBWu : N3VW_QHS<1,1,0b0011,0, "vsubw", "u", sub, zext, 0>;
3277 // VHSUB : Vector Halving Subtract
3278 defm VHSUBs : N3VInt_QHS<0, 0, 0b0010, 0, N3RegFrm,
3279 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3280 "vhsub", "s", int_arm_neon_vhsubs, 0>;
3281 defm VHSUBu : N3VInt_QHS<1, 0, 0b0010, 0, N3RegFrm,
3282 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3283 "vhsub", "u", int_arm_neon_vhsubu, 0>;
3284 // VQSUB : Vector Saturing Subtract
3285 defm VQSUBs : N3VInt_QHSD<0, 0, 0b0010, 1, N3RegFrm,
3286 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3287 "vqsub", "s", int_arm_neon_vqsubs, 0>;
3288 defm VQSUBu : N3VInt_QHSD<1, 0, 0b0010, 1, N3RegFrm,
3289 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3290 "vqsub", "u", int_arm_neon_vqsubu, 0>;
3291 // VSUBHN : Vector Subtract and Narrow Returning High Half (D = Q - Q)
3292 defm VSUBHN : N3VNInt_HSD<0,1,0b0110,0, "vsubhn", "i",
3293 int_arm_neon_vsubhn, 0>;
3294 // VRSUBHN : Vector Rounding Subtract and Narrow Returning High Half (D=Q-Q)
3295 defm VRSUBHN : N3VNInt_HSD<1,1,0b0110,0, "vrsubhn", "i",
3296 int_arm_neon_vrsubhn, 0>;
3298 // Vector Comparisons.
3300 // VCEQ : Vector Compare Equal
3301 defm VCEQ : N3V_QHS<1, 0, 0b1000, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3302 IIC_VSUBi4Q, "vceq", "i", NEONvceq, 1>;
3303 def VCEQfd : N3VD<0,0,0b00,0b1110,0, IIC_VBIND, "vceq", "f32", v2i32, v2f32,
3305 def VCEQfq : N3VQ<0,0,0b00,0b1110,0, IIC_VBINQ, "vceq", "f32", v4i32, v4f32,
3308 defm VCEQz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00010, 0, "vceq", "i",
3309 "$dst, $src, #0", NEONvceqz>;
3311 // VCGE : Vector Compare Greater Than or Equal
3312 defm VCGEs : N3V_QHS<0, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3313 IIC_VSUBi4Q, "vcge", "s", NEONvcge, 0>;
3314 defm VCGEu : N3V_QHS<1, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3315 IIC_VSUBi4Q, "vcge", "u", NEONvcgeu, 0>;
3316 def VCGEfd : N3VD<1,0,0b00,0b1110,0, IIC_VBIND, "vcge", "f32", v2i32, v2f32,
3318 def VCGEfq : N3VQ<1,0,0b00,0b1110,0, IIC_VBINQ, "vcge", "f32", v4i32, v4f32,
3321 defm VCGEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00001, 0, "vcge", "s",
3322 "$dst, $src, #0", NEONvcgez>;
3323 defm VCLEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00011, 0, "vcle", "s",
3324 "$dst, $src, #0", NEONvclez>;
3326 // VCGT : Vector Compare Greater Than
3327 defm VCGTs : N3V_QHS<0, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3328 IIC_VSUBi4Q, "vcgt", "s", NEONvcgt, 0>;
3329 defm VCGTu : N3V_QHS<1, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3330 IIC_VSUBi4Q, "vcgt", "u", NEONvcgtu, 0>;
3331 def VCGTfd : N3VD<1,0,0b10,0b1110,0, IIC_VBIND, "vcgt", "f32", v2i32, v2f32,
3333 def VCGTfq : N3VQ<1,0,0b10,0b1110,0, IIC_VBINQ, "vcgt", "f32", v4i32, v4f32,
3336 defm VCGTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00000, 0, "vcgt", "s",
3337 "$dst, $src, #0", NEONvcgtz>;
3338 defm VCLTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00100, 0, "vclt", "s",
3339 "$dst, $src, #0", NEONvcltz>;
3341 // VACGE : Vector Absolute Compare Greater Than or Equal (aka VCAGE)
3342 def VACGEd : N3VDInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacge",
3343 "f32", v2i32, v2f32, int_arm_neon_vacged, 0>;
3344 def VACGEq : N3VQInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacge",
3345 "f32", v4i32, v4f32, int_arm_neon_vacgeq, 0>;
3346 // VACGT : Vector Absolute Compare Greater Than (aka VCAGT)
3347 def VACGTd : N3VDInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacgt",
3348 "f32", v2i32, v2f32, int_arm_neon_vacgtd, 0>;
3349 def VACGTq : N3VQInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacgt",
3350 "f32", v4i32, v4f32, int_arm_neon_vacgtq, 0>;
3351 // VTST : Vector Test Bits
3352 defm VTST : N3V_QHS<0, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
3353 IIC_VBINi4Q, "vtst", "", NEONvtst, 1>;
3355 // Vector Bitwise Operations.
3357 def vnotd : PatFrag<(ops node:$in),
3358 (xor node:$in, (bitconvert (v8i8 NEONimmAllOnesV)))>;
3359 def vnotq : PatFrag<(ops node:$in),
3360 (xor node:$in, (bitconvert (v16i8 NEONimmAllOnesV)))>;
3363 // VAND : Vector Bitwise AND
3364 def VANDd : N3VDX<0, 0, 0b00, 0b0001, 1, IIC_VBINiD, "vand",
3365 v2i32, v2i32, and, 1>;
3366 def VANDq : N3VQX<0, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "vand",
3367 v4i32, v4i32, and, 1>;
3369 // VEOR : Vector Bitwise Exclusive OR
3370 def VEORd : N3VDX<1, 0, 0b00, 0b0001, 1, IIC_VBINiD, "veor",
3371 v2i32, v2i32, xor, 1>;
3372 def VEORq : N3VQX<1, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "veor",
3373 v4i32, v4i32, xor, 1>;
3375 // VORR : Vector Bitwise OR
3376 def VORRd : N3VDX<0, 0, 0b10, 0b0001, 1, IIC_VBINiD, "vorr",
3377 v2i32, v2i32, or, 1>;
3378 def VORRq : N3VQX<0, 0, 0b10, 0b0001, 1, IIC_VBINiQ, "vorr",
3379 v4i32, v4i32, or, 1>;
3381 def VORRiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 0, 1,
3382 (outs DPR:$Vd), (ins nModImm:$SIMM, DPR:$src),
3384 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
3386 (v4i16 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
3387 let Inst{9} = SIMM{9};
3390 def VORRiv2i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 0, 0, 1,
3391 (outs DPR:$Vd), (ins nModImm:$SIMM, DPR:$src),
3393 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
3395 (v2i32 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
3396 let Inst{10-9} = SIMM{10-9};
3399 def VORRiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 0, 1,
3400 (outs QPR:$Vd), (ins nModImm:$SIMM, QPR:$src),
3402 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
3404 (v8i16 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
3405 let Inst{9} = SIMM{9};
3408 def VORRiv4i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 1, 0, 1,
3409 (outs QPR:$Vd), (ins nModImm:$SIMM, QPR:$src),
3411 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
3413 (v4i32 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
3414 let Inst{10-9} = SIMM{10-9};
3418 // VBIC : Vector Bitwise Bit Clear (AND NOT)
3419 def VBICd : N3VX<0, 0, 0b01, 0b0001, 0, 1, (outs DPR:$dst),
3420 (ins DPR:$src1, DPR:$src2), N3RegFrm, IIC_VBINiD,
3421 "vbic", "$dst, $src1, $src2", "",
3422 [(set DPR:$dst, (v2i32 (and DPR:$src1,
3423 (vnotd DPR:$src2))))]>;
3424 def VBICq : N3VX<0, 0, 0b01, 0b0001, 1, 1, (outs QPR:$dst),
3425 (ins QPR:$src1, QPR:$src2), N3RegFrm, IIC_VBINiQ,
3426 "vbic", "$dst, $src1, $src2", "",
3427 [(set QPR:$dst, (v4i32 (and QPR:$src1,
3428 (vnotq QPR:$src2))))]>;
3430 def VBICiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 1, 1,
3431 (outs DPR:$Vd), (ins nModImm:$SIMM, DPR:$src),
3433 "vbic", "i16", "$Vd, $SIMM", "$src = $Vd",
3435 (v4i16 (NEONvbicImm DPR:$src, timm:$SIMM)))]> {
3436 let Inst{9} = SIMM{9};
3439 def VBICiv2i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 0, 1, 1,
3440 (outs DPR:$Vd), (ins nModImm:$SIMM, DPR:$src),
3442 "vbic", "i32", "$Vd, $SIMM", "$src = $Vd",
3444 (v2i32 (NEONvbicImm DPR:$src, timm:$SIMM)))]> {
3445 let Inst{10-9} = SIMM{10-9};
3448 def VBICiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 1, 1,
3449 (outs QPR:$Vd), (ins nModImm:$SIMM, QPR:$src),
3451 "vbic", "i16", "$Vd, $SIMM", "$src = $Vd",
3453 (v8i16 (NEONvbicImm QPR:$src, timm:$SIMM)))]> {
3454 let Inst{9} = SIMM{9};
3457 def VBICiv4i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 1, 1, 1,
3458 (outs QPR:$Vd), (ins nModImm:$SIMM, QPR:$src),
3460 "vbic", "i32", "$Vd, $SIMM", "$src = $Vd",
3462 (v4i32 (NEONvbicImm QPR:$src, timm:$SIMM)))]> {
3463 let Inst{10-9} = SIMM{10-9};
3466 // VORN : Vector Bitwise OR NOT
3467 def VORNd : N3VX<0, 0, 0b11, 0b0001, 0, 1, (outs DPR:$dst),
3468 (ins DPR:$src1, DPR:$src2), N3RegFrm, IIC_VBINiD,
3469 "vorn", "$dst, $src1, $src2", "",
3470 [(set DPR:$dst, (v2i32 (or DPR:$src1,
3471 (vnotd DPR:$src2))))]>;
3472 def VORNq : N3VX<0, 0, 0b11, 0b0001, 1, 1, (outs QPR:$dst),
3473 (ins QPR:$src1, QPR:$src2), N3RegFrm, IIC_VBINiQ,
3474 "vorn", "$dst, $src1, $src2", "",
3475 [(set QPR:$dst, (v4i32 (or QPR:$src1,
3476 (vnotq QPR:$src2))))]>;
3478 // VMVN : Vector Bitwise NOT (Immediate)
3480 let isReMaterializable = 1 in {
3482 def VMVNv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 1, 1, (outs DPR:$dst),
3483 (ins nModImm:$SIMM), IIC_VMOVImm,
3484 "vmvn", "i16", "$dst, $SIMM", "",
3485 [(set DPR:$dst, (v4i16 (NEONvmvnImm timm:$SIMM)))]> {
3486 let Inst{9} = SIMM{9};
3489 def VMVNv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 1, 1, (outs QPR:$dst),
3490 (ins nModImm:$SIMM), IIC_VMOVImm,
3491 "vmvn", "i16", "$dst, $SIMM", "",
3492 [(set QPR:$dst, (v8i16 (NEONvmvnImm timm:$SIMM)))]> {
3493 let Inst{9} = SIMM{9};
3496 def VMVNv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 1, 1, (outs DPR:$dst),
3497 (ins nModImm:$SIMM), IIC_VMOVImm,
3498 "vmvn", "i32", "$dst, $SIMM", "",
3499 [(set DPR:$dst, (v2i32 (NEONvmvnImm timm:$SIMM)))]> {
3500 let Inst{11-8} = SIMM{11-8};
3503 def VMVNv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 1, 1, (outs QPR:$dst),
3504 (ins nModImm:$SIMM), IIC_VMOVImm,
3505 "vmvn", "i32", "$dst, $SIMM", "",
3506 [(set QPR:$dst, (v4i32 (NEONvmvnImm timm:$SIMM)))]> {
3507 let Inst{11-8} = SIMM{11-8};
3511 // VMVN : Vector Bitwise NOT
3512 def VMVNd : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 0, 0,
3513 (outs DPR:$dst), (ins DPR:$src), IIC_VSUBiD,
3514 "vmvn", "$dst, $src", "",
3515 [(set DPR:$dst, (v2i32 (vnotd DPR:$src)))]>;
3516 def VMVNq : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 1, 0,
3517 (outs QPR:$dst), (ins QPR:$src), IIC_VSUBiD,
3518 "vmvn", "$dst, $src", "",
3519 [(set QPR:$dst, (v4i32 (vnotq QPR:$src)))]>;
3520 def : Pat<(v2i32 (vnotd DPR:$src)), (VMVNd DPR:$src)>;
3521 def : Pat<(v4i32 (vnotq QPR:$src)), (VMVNq QPR:$src)>;
3523 // VBSL : Vector Bitwise Select
3524 def VBSLd : N3VX<1, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
3525 (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
3526 N3RegFrm, IIC_VCNTiD,
3527 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3529 (v2i32 (or (and DPR:$Vn, DPR:$src1),
3530 (and DPR:$Vm, (vnotd DPR:$src1)))))]>;
3531 def VBSLq : N3VX<1, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
3532 (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
3533 N3RegFrm, IIC_VCNTiQ,
3534 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3536 (v4i32 (or (and QPR:$Vn, QPR:$src1),
3537 (and QPR:$Vm, (vnotq QPR:$src1)))))]>;
3539 // VBIF : Vector Bitwise Insert if False
3540 // like VBSL but with: "vbif $dst, $src3, $src1", "$src2 = $dst",
3541 // FIXME: This instruction's encoding MAY NOT BE correct.
3542 def VBIFd : N3VX<1, 0, 0b11, 0b0001, 0, 1,
3543 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
3544 N3RegFrm, IIC_VBINiD,
3545 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3546 [/* For disassembly only; pattern left blank */]>;
3547 def VBIFq : N3VX<1, 0, 0b11, 0b0001, 1, 1,
3548 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
3549 N3RegFrm, IIC_VBINiQ,
3550 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3551 [/* For disassembly only; pattern left blank */]>;
3553 // VBIT : Vector Bitwise Insert if True
3554 // like VBSL but with: "vbit $dst, $src2, $src1", "$src3 = $dst",
3555 // FIXME: This instruction's encoding MAY NOT BE correct.
3556 def VBITd : N3VX<1, 0, 0b10, 0b0001, 0, 1,
3557 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
3558 N3RegFrm, IIC_VBINiD,
3559 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3560 [/* For disassembly only; pattern left blank */]>;
3561 def VBITq : N3VX<1, 0, 0b10, 0b0001, 1, 1,
3562 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
3563 N3RegFrm, IIC_VBINiQ,
3564 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3565 [/* For disassembly only; pattern left blank */]>;
3567 // VBIT/VBIF are not yet implemented. The TwoAddress pass will not go looking
3568 // for equivalent operations with different register constraints; it just
3571 // Vector Absolute Differences.
3573 // VABD : Vector Absolute Difference
3574 defm VABDs : N3VInt_QHS<0, 0, 0b0111, 0, N3RegFrm,
3575 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3576 "vabd", "s", int_arm_neon_vabds, 1>;
3577 defm VABDu : N3VInt_QHS<1, 0, 0b0111, 0, N3RegFrm,
3578 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3579 "vabd", "u", int_arm_neon_vabdu, 1>;
3580 def VABDfd : N3VDInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBIND,
3581 "vabd", "f32", v2f32, v2f32, int_arm_neon_vabds, 1>;
3582 def VABDfq : N3VQInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBINQ,
3583 "vabd", "f32", v4f32, v4f32, int_arm_neon_vabds, 1>;
3585 // VABDL : Vector Absolute Difference Long (Q = | D - D |)
3586 defm VABDLs : N3VLIntExt_QHS<0,1,0b0111,0, IIC_VSUBi4Q,
3587 "vabdl", "s", int_arm_neon_vabds, zext, 1>;
3588 defm VABDLu : N3VLIntExt_QHS<1,1,0b0111,0, IIC_VSUBi4Q,
3589 "vabdl", "u", int_arm_neon_vabdu, zext, 1>;
3591 // VABA : Vector Absolute Difference and Accumulate
3592 defm VABAs : N3VIntOp_QHS<0,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
3593 "vaba", "s", int_arm_neon_vabds, add>;
3594 defm VABAu : N3VIntOp_QHS<1,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
3595 "vaba", "u", int_arm_neon_vabdu, add>;
3597 // VABAL : Vector Absolute Difference and Accumulate Long (Q += | D - D |)
3598 defm VABALs : N3VLIntExtOp_QHS<0,1,0b0101,0, IIC_VABAD,
3599 "vabal", "s", int_arm_neon_vabds, zext, add>;
3600 defm VABALu : N3VLIntExtOp_QHS<1,1,0b0101,0, IIC_VABAD,
3601 "vabal", "u", int_arm_neon_vabdu, zext, add>;
3603 // Vector Maximum and Minimum.
3605 // VMAX : Vector Maximum
3606 defm VMAXs : N3VInt_QHS<0, 0, 0b0110, 0, N3RegFrm,
3607 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3608 "vmax", "s", int_arm_neon_vmaxs, 1>;
3609 defm VMAXu : N3VInt_QHS<1, 0, 0b0110, 0, N3RegFrm,
3610 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3611 "vmax", "u", int_arm_neon_vmaxu, 1>;
3612 def VMAXfd : N3VDInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBIND,
3614 v2f32, v2f32, int_arm_neon_vmaxs, 1>;
3615 def VMAXfq : N3VQInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBINQ,
3617 v4f32, v4f32, int_arm_neon_vmaxs, 1>;
3619 // VMIN : Vector Minimum
3620 defm VMINs : N3VInt_QHS<0, 0, 0b0110, 1, N3RegFrm,
3621 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3622 "vmin", "s", int_arm_neon_vmins, 1>;
3623 defm VMINu : N3VInt_QHS<1, 0, 0b0110, 1, N3RegFrm,
3624 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3625 "vmin", "u", int_arm_neon_vminu, 1>;
3626 def VMINfd : N3VDInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBIND,
3628 v2f32, v2f32, int_arm_neon_vmins, 1>;
3629 def VMINfq : N3VQInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBINQ,
3631 v4f32, v4f32, int_arm_neon_vmins, 1>;
3633 // Vector Pairwise Operations.
3635 // VPADD : Vector Pairwise Add
3636 def VPADDi8 : N3VDInt<0, 0, 0b00, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
3638 v8i8, v8i8, int_arm_neon_vpadd, 0>;
3639 def VPADDi16 : N3VDInt<0, 0, 0b01, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
3641 v4i16, v4i16, int_arm_neon_vpadd, 0>;
3642 def VPADDi32 : N3VDInt<0, 0, 0b10, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
3644 v2i32, v2i32, int_arm_neon_vpadd, 0>;
3645 def VPADDf : N3VDInt<1, 0, 0b00, 0b1101, 0, N3RegFrm,
3646 IIC_VPBIND, "vpadd", "f32",
3647 v2f32, v2f32, int_arm_neon_vpadd, 0>;
3649 // VPADDL : Vector Pairwise Add Long
3650 defm VPADDLs : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00100, 0, "vpaddl", "s",
3651 int_arm_neon_vpaddls>;
3652 defm VPADDLu : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00101, 0, "vpaddl", "u",
3653 int_arm_neon_vpaddlu>;
3655 // VPADAL : Vector Pairwise Add and Accumulate Long
3656 defm VPADALs : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01100, 0, "vpadal", "s",
3657 int_arm_neon_vpadals>;
3658 defm VPADALu : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01101, 0, "vpadal", "u",
3659 int_arm_neon_vpadalu>;
3661 // VPMAX : Vector Pairwise Maximum
3662 def VPMAXs8 : N3VDInt<0, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3663 "s8", v8i8, v8i8, int_arm_neon_vpmaxs, 0>;
3664 def VPMAXs16 : N3VDInt<0, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3665 "s16", v4i16, v4i16, int_arm_neon_vpmaxs, 0>;
3666 def VPMAXs32 : N3VDInt<0, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3667 "s32", v2i32, v2i32, int_arm_neon_vpmaxs, 0>;
3668 def VPMAXu8 : N3VDInt<1, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3669 "u8", v8i8, v8i8, int_arm_neon_vpmaxu, 0>;
3670 def VPMAXu16 : N3VDInt<1, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3671 "u16", v4i16, v4i16, int_arm_neon_vpmaxu, 0>;
3672 def VPMAXu32 : N3VDInt<1, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3673 "u32", v2i32, v2i32, int_arm_neon_vpmaxu, 0>;
3674 def VPMAXf : N3VDInt<1, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmax",
3675 "f32", v2f32, v2f32, int_arm_neon_vpmaxs, 0>;
3677 // VPMIN : Vector Pairwise Minimum
3678 def VPMINs8 : N3VDInt<0, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3679 "s8", v8i8, v8i8, int_arm_neon_vpmins, 0>;
3680 def VPMINs16 : N3VDInt<0, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3681 "s16", v4i16, v4i16, int_arm_neon_vpmins, 0>;
3682 def VPMINs32 : N3VDInt<0, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3683 "s32", v2i32, v2i32, int_arm_neon_vpmins, 0>;
3684 def VPMINu8 : N3VDInt<1, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3685 "u8", v8i8, v8i8, int_arm_neon_vpminu, 0>;
3686 def VPMINu16 : N3VDInt<1, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3687 "u16", v4i16, v4i16, int_arm_neon_vpminu, 0>;
3688 def VPMINu32 : N3VDInt<1, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3689 "u32", v2i32, v2i32, int_arm_neon_vpminu, 0>;
3690 def VPMINf : N3VDInt<1, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmin",
3691 "f32", v2f32, v2f32, int_arm_neon_vpmins, 0>;
3693 // Vector Reciprocal and Reciprocal Square Root Estimate and Step.
3695 // VRECPE : Vector Reciprocal Estimate
3696 def VRECPEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
3697 IIC_VUNAD, "vrecpe", "u32",
3698 v2i32, v2i32, int_arm_neon_vrecpe>;
3699 def VRECPEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
3700 IIC_VUNAQ, "vrecpe", "u32",
3701 v4i32, v4i32, int_arm_neon_vrecpe>;
3702 def VRECPEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
3703 IIC_VUNAD, "vrecpe", "f32",
3704 v2f32, v2f32, int_arm_neon_vrecpe>;
3705 def VRECPEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
3706 IIC_VUNAQ, "vrecpe", "f32",
3707 v4f32, v4f32, int_arm_neon_vrecpe>;
3709 // VRECPS : Vector Reciprocal Step
3710 def VRECPSfd : N3VDInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
3711 IIC_VRECSD, "vrecps", "f32",
3712 v2f32, v2f32, int_arm_neon_vrecps, 1>;
3713 def VRECPSfq : N3VQInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
3714 IIC_VRECSQ, "vrecps", "f32",
3715 v4f32, v4f32, int_arm_neon_vrecps, 1>;
3717 // VRSQRTE : Vector Reciprocal Square Root Estimate
3718 def VRSQRTEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
3719 IIC_VUNAD, "vrsqrte", "u32",
3720 v2i32, v2i32, int_arm_neon_vrsqrte>;
3721 def VRSQRTEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
3722 IIC_VUNAQ, "vrsqrte", "u32",
3723 v4i32, v4i32, int_arm_neon_vrsqrte>;
3724 def VRSQRTEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
3725 IIC_VUNAD, "vrsqrte", "f32",
3726 v2f32, v2f32, int_arm_neon_vrsqrte>;
3727 def VRSQRTEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
3728 IIC_VUNAQ, "vrsqrte", "f32",
3729 v4f32, v4f32, int_arm_neon_vrsqrte>;
3731 // VRSQRTS : Vector Reciprocal Square Root Step
3732 def VRSQRTSfd : N3VDInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
3733 IIC_VRECSD, "vrsqrts", "f32",
3734 v2f32, v2f32, int_arm_neon_vrsqrts, 1>;
3735 def VRSQRTSfq : N3VQInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
3736 IIC_VRECSQ, "vrsqrts", "f32",
3737 v4f32, v4f32, int_arm_neon_vrsqrts, 1>;
3741 // VSHL : Vector Shift
3742 defm VSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 0, N3RegVShFrm,
3743 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
3744 "vshl", "s", int_arm_neon_vshifts>;
3745 defm VSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 0, N3RegVShFrm,
3746 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
3747 "vshl", "u", int_arm_neon_vshiftu>;
3748 // VSHL : Vector Shift Left (Immediate)
3749 defm VSHLi : N2VSh_QHSD<0, 1, 0b0101, 1, IIC_VSHLiD, "vshl", "i", NEONvshl,
3751 // VSHR : Vector Shift Right (Immediate)
3752 defm VSHRs : N2VSh_QHSD<0, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "s", NEONvshrs,
3754 defm VSHRu : N2VSh_QHSD<1, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "u", NEONvshru,
3757 // VSHLL : Vector Shift Left Long
3758 defm VSHLLs : N2VLSh_QHS<0, 1, 0b1010, 0, 0, 1, "vshll", "s", NEONvshlls>;
3759 defm VSHLLu : N2VLSh_QHS<1, 1, 0b1010, 0, 0, 1, "vshll", "u", NEONvshllu>;
3761 // VSHLL : Vector Shift Left Long (with maximum shift count)
3762 class N2VLShMax<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
3763 bit op6, bit op4, string OpcodeStr, string Dt, ValueType ResTy,
3764 ValueType OpTy, SDNode OpNode>
3765 : N2VLSh<op24, op23, op11_8, op7, op6, op4, OpcodeStr, Dt,
3766 ResTy, OpTy, OpNode> {
3767 let Inst{21-16} = op21_16;
3769 def VSHLLi8 : N2VLShMax<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll", "i8",
3770 v8i16, v8i8, NEONvshlli>;
3771 def VSHLLi16 : N2VLShMax<1, 1, 0b110110, 0b0011, 0, 0, 0, "vshll", "i16",
3772 v4i32, v4i16, NEONvshlli>;
3773 def VSHLLi32 : N2VLShMax<1, 1, 0b111010, 0b0011, 0, 0, 0, "vshll", "i32",
3774 v2i64, v2i32, NEONvshlli>;
3776 // VSHRN : Vector Shift Right and Narrow
3777 defm VSHRN : N2VNSh_HSD<0,1,0b1000,0,0,1, IIC_VSHLiD, "vshrn", "i",
3780 // VRSHL : Vector Rounding Shift
3781 defm VRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 0, N3RegVShFrm,
3782 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3783 "vrshl", "s", int_arm_neon_vrshifts>;
3784 defm VRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 0, N3RegVShFrm,
3785 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3786 "vrshl", "u", int_arm_neon_vrshiftu>;
3787 // VRSHR : Vector Rounding Shift Right
3788 defm VRSHRs : N2VSh_QHSD<0,1,0b0010,1, IIC_VSHLi4D, "vrshr", "s", NEONvrshrs,
3790 defm VRSHRu : N2VSh_QHSD<1,1,0b0010,1, IIC_VSHLi4D, "vrshr", "u", NEONvrshru,
3793 // VRSHRN : Vector Rounding Shift Right and Narrow
3794 defm VRSHRN : N2VNSh_HSD<0, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vrshrn", "i",
3797 // VQSHL : Vector Saturating Shift
3798 defm VQSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 1, N3RegVShFrm,
3799 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3800 "vqshl", "s", int_arm_neon_vqshifts>;
3801 defm VQSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 1, N3RegVShFrm,
3802 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3803 "vqshl", "u", int_arm_neon_vqshiftu>;
3804 // VQSHL : Vector Saturating Shift Left (Immediate)
3805 defm VQSHLsi : N2VSh_QHSD<0,1,0b0111,1, IIC_VSHLi4D, "vqshl", "s",NEONvqshls,
3807 defm VQSHLui : N2VSh_QHSD<1,1,0b0111,1, IIC_VSHLi4D, "vqshl", "u",NEONvqshlu,
3809 // VQSHLU : Vector Saturating Shift Left (Immediate, Unsigned)
3810 defm VQSHLsu : N2VSh_QHSD<1,1,0b0110,1, IIC_VSHLi4D,"vqshlu","s",NEONvqshlsu,
3813 // VQSHRN : Vector Saturating Shift Right and Narrow
3814 defm VQSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "s",
3816 defm VQSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "u",
3819 // VQSHRUN : Vector Saturating Shift Right and Narrow (Unsigned)
3820 defm VQSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 0, 1, IIC_VSHLi4D, "vqshrun", "s",
3823 // VQRSHL : Vector Saturating Rounding Shift
3824 defm VQRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 1, N3RegVShFrm,
3825 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3826 "vqrshl", "s", int_arm_neon_vqrshifts>;
3827 defm VQRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 1, N3RegVShFrm,
3828 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3829 "vqrshl", "u", int_arm_neon_vqrshiftu>;
3831 // VQRSHRN : Vector Saturating Rounding Shift Right and Narrow
3832 defm VQRSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "s",
3834 defm VQRSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "u",
3837 // VQRSHRUN : Vector Saturating Rounding Shift Right and Narrow (Unsigned)
3838 defm VQRSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vqrshrun", "s",
3841 // VSRA : Vector Shift Right and Accumulate
3842 defm VSRAs : N2VShAdd_QHSD<0, 1, 0b0001, 1, "vsra", "s", NEONvshrs>;
3843 defm VSRAu : N2VShAdd_QHSD<1, 1, 0b0001, 1, "vsra", "u", NEONvshru>;
3844 // VRSRA : Vector Rounding Shift Right and Accumulate
3845 defm VRSRAs : N2VShAdd_QHSD<0, 1, 0b0011, 1, "vrsra", "s", NEONvrshrs>;
3846 defm VRSRAu : N2VShAdd_QHSD<1, 1, 0b0011, 1, "vrsra", "u", NEONvrshru>;
3848 // VSLI : Vector Shift Left and Insert
3849 defm VSLI : N2VShIns_QHSD<1, 1, 0b0101, 1, "vsli", NEONvsli, N2RegVShLFrm>;
3850 // VSRI : Vector Shift Right and Insert
3851 defm VSRI : N2VShIns_QHSD<1, 1, 0b0100, 1, "vsri", NEONvsri, N2RegVShRFrm>;
3853 // Vector Absolute and Saturating Absolute.
3855 // VABS : Vector Absolute Value
3856 defm VABS : N2VInt_QHS<0b11, 0b11, 0b01, 0b00110, 0,
3857 IIC_VUNAiD, IIC_VUNAiQ, "vabs", "s",
3859 def VABSfd : N2VDInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
3860 IIC_VUNAD, "vabs", "f32",
3861 v2f32, v2f32, int_arm_neon_vabs>;
3862 def VABSfq : N2VQInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
3863 IIC_VUNAQ, "vabs", "f32",
3864 v4f32, v4f32, int_arm_neon_vabs>;
3866 // VQABS : Vector Saturating Absolute Value
3867 defm VQABS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01110, 0,
3868 IIC_VQUNAiD, IIC_VQUNAiQ, "vqabs", "s",
3869 int_arm_neon_vqabs>;
3873 def vnegd : PatFrag<(ops node:$in),
3874 (sub (bitconvert (v2i32 NEONimmAllZerosV)), node:$in)>;
3875 def vnegq : PatFrag<(ops node:$in),
3876 (sub (bitconvert (v4i32 NEONimmAllZerosV)), node:$in)>;
3878 class VNEGD<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
3879 : N2V<0b11, 0b11, size, 0b01, 0b00111, 0, 0, (outs DPR:$dst), (ins DPR:$src),
3880 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
3881 [(set DPR:$dst, (Ty (vnegd DPR:$src)))]>;
3882 class VNEGQ<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
3883 : N2V<0b11, 0b11, size, 0b01, 0b00111, 1, 0, (outs QPR:$dst), (ins QPR:$src),
3884 IIC_VSHLiQ, OpcodeStr, Dt, "$dst, $src", "",
3885 [(set QPR:$dst, (Ty (vnegq QPR:$src)))]>;
3887 // VNEG : Vector Negate (integer)
3888 def VNEGs8d : VNEGD<0b00, "vneg", "s8", v8i8>;
3889 def VNEGs16d : VNEGD<0b01, "vneg", "s16", v4i16>;
3890 def VNEGs32d : VNEGD<0b10, "vneg", "s32", v2i32>;
3891 def VNEGs8q : VNEGQ<0b00, "vneg", "s8", v16i8>;
3892 def VNEGs16q : VNEGQ<0b01, "vneg", "s16", v8i16>;
3893 def VNEGs32q : VNEGQ<0b10, "vneg", "s32", v4i32>;
3895 // VNEG : Vector Negate (floating-point)
3896 def VNEGfd : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
3897 (outs DPR:$dst), (ins DPR:$src), IIC_VUNAD,
3898 "vneg", "f32", "$dst, $src", "",
3899 [(set DPR:$dst, (v2f32 (fneg DPR:$src)))]>;
3900 def VNEGf32q : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 1, 0,
3901 (outs QPR:$dst), (ins QPR:$src), IIC_VUNAQ,
3902 "vneg", "f32", "$dst, $src", "",
3903 [(set QPR:$dst, (v4f32 (fneg QPR:$src)))]>;
3905 def : Pat<(v8i8 (vnegd DPR:$src)), (VNEGs8d DPR:$src)>;
3906 def : Pat<(v4i16 (vnegd DPR:$src)), (VNEGs16d DPR:$src)>;
3907 def : Pat<(v2i32 (vnegd DPR:$src)), (VNEGs32d DPR:$src)>;
3908 def : Pat<(v16i8 (vnegq QPR:$src)), (VNEGs8q QPR:$src)>;
3909 def : Pat<(v8i16 (vnegq QPR:$src)), (VNEGs16q QPR:$src)>;
3910 def : Pat<(v4i32 (vnegq QPR:$src)), (VNEGs32q QPR:$src)>;
3912 // VQNEG : Vector Saturating Negate
3913 defm VQNEG : N2VInt_QHS<0b11, 0b11, 0b00, 0b01111, 0,
3914 IIC_VQUNAiD, IIC_VQUNAiQ, "vqneg", "s",
3915 int_arm_neon_vqneg>;
3917 // Vector Bit Counting Operations.
3919 // VCLS : Vector Count Leading Sign Bits
3920 defm VCLS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01000, 0,
3921 IIC_VCNTiD, IIC_VCNTiQ, "vcls", "s",
3923 // VCLZ : Vector Count Leading Zeros
3924 defm VCLZ : N2VInt_QHS<0b11, 0b11, 0b00, 0b01001, 0,
3925 IIC_VCNTiD, IIC_VCNTiQ, "vclz", "i",
3927 // VCNT : Vector Count One Bits
3928 def VCNTd : N2VDInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
3929 IIC_VCNTiD, "vcnt", "8",
3930 v8i8, v8i8, int_arm_neon_vcnt>;
3931 def VCNTq : N2VQInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
3932 IIC_VCNTiQ, "vcnt", "8",
3933 v16i8, v16i8, int_arm_neon_vcnt>;
3935 // Vector Swap -- for disassembly only.
3936 def VSWPd : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 0, 0,
3937 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
3938 "vswp", "$dst, $src", "", []>;
3939 def VSWPq : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 1, 0,
3940 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
3941 "vswp", "$dst, $src", "", []>;
3943 // Vector Move Operations.
3945 // VMOV : Vector Move (Register)
3947 let neverHasSideEffects = 1 in {
3948 def VMOVDneon: N3VX<0, 0, 0b10, 0b0001, 0, 1, (outs DPR:$Vd), (ins DPR:$Vm),
3949 N3RegFrm, IIC_VMOV, "vmov", "$Vd, $Vm", "", []> {
3950 let Vn{4-0} = Vm{4-0};
3952 def VMOVQ : N3VX<0, 0, 0b10, 0b0001, 1, 1, (outs QPR:$Vd), (ins QPR:$Vm),
3953 N3RegFrm, IIC_VMOV, "vmov", "$Vd, $Vm", "", []> {
3954 let Vn{4-0} = Vm{4-0};
3957 // Pseudo vector move instructions for QQ and QQQQ registers. This should
3958 // be expanded after register allocation is completed.
3959 def VMOVQQ : PseudoInst<(outs QQPR:$dst), (ins QQPR:$src),
3962 def VMOVQQQQ : PseudoInst<(outs QQQQPR:$dst), (ins QQQQPR:$src),
3964 } // neverHasSideEffects
3966 // VMOV : Vector Move (Immediate)
3968 let isReMaterializable = 1 in {
3969 def VMOVv8i8 : N1ModImm<1, 0b000, 0b1110, 0, 0, 0, 1, (outs DPR:$dst),
3970 (ins nModImm:$SIMM), IIC_VMOVImm,
3971 "vmov", "i8", "$dst, $SIMM", "",
3972 [(set DPR:$dst, (v8i8 (NEONvmovImm timm:$SIMM)))]>;
3973 def VMOVv16i8 : N1ModImm<1, 0b000, 0b1110, 0, 1, 0, 1, (outs QPR:$dst),
3974 (ins nModImm:$SIMM), IIC_VMOVImm,
3975 "vmov", "i8", "$dst, $SIMM", "",
3976 [(set QPR:$dst, (v16i8 (NEONvmovImm timm:$SIMM)))]>;
3978 def VMOVv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 0, 1, (outs DPR:$dst),
3979 (ins nModImm:$SIMM), IIC_VMOVImm,
3980 "vmov", "i16", "$dst, $SIMM", "",
3981 [(set DPR:$dst, (v4i16 (NEONvmovImm timm:$SIMM)))]> {
3982 let Inst{9} = SIMM{9};
3985 def VMOVv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 0, 1, (outs QPR:$dst),
3986 (ins nModImm:$SIMM), IIC_VMOVImm,
3987 "vmov", "i16", "$dst, $SIMM", "",
3988 [(set QPR:$dst, (v8i16 (NEONvmovImm timm:$SIMM)))]> {
3989 let Inst{9} = SIMM{9};
3992 def VMOVv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 0, 1, (outs DPR:$dst),
3993 (ins nModImm:$SIMM), IIC_VMOVImm,
3994 "vmov", "i32", "$dst, $SIMM", "",
3995 [(set DPR:$dst, (v2i32 (NEONvmovImm timm:$SIMM)))]> {
3996 let Inst{11-8} = SIMM{11-8};
3999 def VMOVv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 0, 1, (outs QPR:$dst),
4000 (ins nModImm:$SIMM), IIC_VMOVImm,
4001 "vmov", "i32", "$dst, $SIMM", "",
4002 [(set QPR:$dst, (v4i32 (NEONvmovImm timm:$SIMM)))]> {
4003 let Inst{11-8} = SIMM{11-8};
4006 def VMOVv1i64 : N1ModImm<1, 0b000, 0b1110, 0, 0, 1, 1, (outs DPR:$dst),
4007 (ins nModImm:$SIMM), IIC_VMOVImm,
4008 "vmov", "i64", "$dst, $SIMM", "",
4009 [(set DPR:$dst, (v1i64 (NEONvmovImm timm:$SIMM)))]>;
4010 def VMOVv2i64 : N1ModImm<1, 0b000, 0b1110, 0, 1, 1, 1, (outs QPR:$dst),
4011 (ins nModImm:$SIMM), IIC_VMOVImm,
4012 "vmov", "i64", "$dst, $SIMM", "",
4013 [(set QPR:$dst, (v2i64 (NEONvmovImm timm:$SIMM)))]>;
4014 } // isReMaterializable
4016 // VMOV : Vector Get Lane (move scalar to ARM core register)
4018 def VGETLNs8 : NVGetLane<{1,1,1,0,0,1,?,1}, 0b1011, {?,?},
4019 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
4020 IIC_VMOVSI, "vmov", "s8", "$R, $V[$lane]",
4021 [(set GPR:$R, (NEONvgetlanes (v8i8 DPR:$V),
4023 let Inst{21} = lane{2};
4024 let Inst{6-5} = lane{1-0};
4026 def VGETLNs16 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, {?,1},
4027 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
4028 IIC_VMOVSI, "vmov", "s16", "$R, $V[$lane]",
4029 [(set GPR:$R, (NEONvgetlanes (v4i16 DPR:$V),
4031 let Inst{21} = lane{1};
4032 let Inst{6} = lane{0};
4034 def VGETLNu8 : NVGetLane<{1,1,1,0,1,1,?,1}, 0b1011, {?,?},
4035 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
4036 IIC_VMOVSI, "vmov", "u8", "$R, $V[$lane]",
4037 [(set GPR:$R, (NEONvgetlaneu (v8i8 DPR:$V),
4039 let Inst{21} = lane{2};
4040 let Inst{6-5} = lane{1-0};
4042 def VGETLNu16 : NVGetLane<{1,1,1,0,1,0,?,1}, 0b1011, {?,1},
4043 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
4044 IIC_VMOVSI, "vmov", "u16", "$R, $V[$lane]",
4045 [(set GPR:$R, (NEONvgetlaneu (v4i16 DPR:$V),
4047 let Inst{21} = lane{1};
4048 let Inst{6} = lane{0};
4050 def VGETLNi32 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, 0b00,
4051 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
4052 IIC_VMOVSI, "vmov", "32", "$R, $V[$lane]",
4053 [(set GPR:$R, (extractelt (v2i32 DPR:$V),
4055 let Inst{21} = lane{0};
4057 // def VGETLNf32: see FMRDH and FMRDL in ARMInstrVFP.td
4058 def : Pat<(NEONvgetlanes (v16i8 QPR:$src), imm:$lane),
4059 (VGETLNs8 (v8i8 (EXTRACT_SUBREG QPR:$src,
4060 (DSubReg_i8_reg imm:$lane))),
4061 (SubReg_i8_lane imm:$lane))>;
4062 def : Pat<(NEONvgetlanes (v8i16 QPR:$src), imm:$lane),
4063 (VGETLNs16 (v4i16 (EXTRACT_SUBREG QPR:$src,
4064 (DSubReg_i16_reg imm:$lane))),
4065 (SubReg_i16_lane imm:$lane))>;
4066 def : Pat<(NEONvgetlaneu (v16i8 QPR:$src), imm:$lane),
4067 (VGETLNu8 (v8i8 (EXTRACT_SUBREG QPR:$src,
4068 (DSubReg_i8_reg imm:$lane))),
4069 (SubReg_i8_lane imm:$lane))>;
4070 def : Pat<(NEONvgetlaneu (v8i16 QPR:$src), imm:$lane),
4071 (VGETLNu16 (v4i16 (EXTRACT_SUBREG QPR:$src,
4072 (DSubReg_i16_reg imm:$lane))),
4073 (SubReg_i16_lane imm:$lane))>;
4074 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
4075 (VGETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src,
4076 (DSubReg_i32_reg imm:$lane))),
4077 (SubReg_i32_lane imm:$lane))>;
4078 def : Pat<(extractelt (v2f32 DPR:$src1), imm:$src2),
4079 (EXTRACT_SUBREG (v2f32 (COPY_TO_REGCLASS (v2f32 DPR:$src1),DPR_VFP2)),
4080 (SSubReg_f32_reg imm:$src2))>;
4081 def : Pat<(extractelt (v4f32 QPR:$src1), imm:$src2),
4082 (EXTRACT_SUBREG (v4f32 (COPY_TO_REGCLASS (v4f32 QPR:$src1),QPR_VFP2)),
4083 (SSubReg_f32_reg imm:$src2))>;
4084 //def : Pat<(extractelt (v2i64 QPR:$src1), imm:$src2),
4085 // (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
4086 def : Pat<(extractelt (v2f64 QPR:$src1), imm:$src2),
4087 (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
4090 // VMOV : Vector Set Lane (move ARM core register to scalar)
4092 let Constraints = "$src1 = $V" in {
4093 def VSETLNi8 : NVSetLane<{1,1,1,0,0,1,?,0}, 0b1011, {?,?}, (outs DPR:$V),
4094 (ins DPR:$src1, GPR:$R, nohash_imm:$lane),
4095 IIC_VMOVISL, "vmov", "8", "$V[$lane], $R",
4096 [(set DPR:$V, (vector_insert (v8i8 DPR:$src1),
4097 GPR:$R, imm:$lane))]> {
4098 let Inst{21} = lane{2};
4099 let Inst{6-5} = lane{1-0};
4101 def VSETLNi16 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, {?,1}, (outs DPR:$V),
4102 (ins DPR:$src1, GPR:$R, nohash_imm:$lane),
4103 IIC_VMOVISL, "vmov", "16", "$V[$lane], $R",
4104 [(set DPR:$V, (vector_insert (v4i16 DPR:$src1),
4105 GPR:$R, imm:$lane))]> {
4106 let Inst{21} = lane{1};
4107 let Inst{6} = lane{0};
4109 def VSETLNi32 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, 0b00, (outs DPR:$V),
4110 (ins DPR:$src1, GPR:$R, nohash_imm:$lane),
4111 IIC_VMOVISL, "vmov", "32", "$V[$lane], $R",
4112 [(set DPR:$V, (insertelt (v2i32 DPR:$src1),
4113 GPR:$R, imm:$lane))]> {
4114 let Inst{21} = lane{0};
4117 def : Pat<(vector_insert (v16i8 QPR:$src1), GPR:$src2, imm:$lane),
4118 (v16i8 (INSERT_SUBREG QPR:$src1,
4119 (v8i8 (VSETLNi8 (v8i8 (EXTRACT_SUBREG QPR:$src1,
4120 (DSubReg_i8_reg imm:$lane))),
4121 GPR:$src2, (SubReg_i8_lane imm:$lane))),
4122 (DSubReg_i8_reg imm:$lane)))>;
4123 def : Pat<(vector_insert (v8i16 QPR:$src1), GPR:$src2, imm:$lane),
4124 (v8i16 (INSERT_SUBREG QPR:$src1,
4125 (v4i16 (VSETLNi16 (v4i16 (EXTRACT_SUBREG QPR:$src1,
4126 (DSubReg_i16_reg imm:$lane))),
4127 GPR:$src2, (SubReg_i16_lane imm:$lane))),
4128 (DSubReg_i16_reg imm:$lane)))>;
4129 def : Pat<(insertelt (v4i32 QPR:$src1), GPR:$src2, imm:$lane),
4130 (v4i32 (INSERT_SUBREG QPR:$src1,
4131 (v2i32 (VSETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src1,
4132 (DSubReg_i32_reg imm:$lane))),
4133 GPR:$src2, (SubReg_i32_lane imm:$lane))),
4134 (DSubReg_i32_reg imm:$lane)))>;
4136 def : Pat<(v2f32 (insertelt DPR:$src1, SPR:$src2, imm:$src3)),
4137 (INSERT_SUBREG (v2f32 (COPY_TO_REGCLASS DPR:$src1, DPR_VFP2)),
4138 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
4139 def : Pat<(v4f32 (insertelt QPR:$src1, SPR:$src2, imm:$src3)),
4140 (INSERT_SUBREG (v4f32 (COPY_TO_REGCLASS QPR:$src1, QPR_VFP2)),
4141 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
4143 //def : Pat<(v2i64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
4144 // (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
4145 def : Pat<(v2f64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
4146 (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
4148 def : Pat<(v2f32 (scalar_to_vector SPR:$src)),
4149 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
4150 def : Pat<(v2f64 (scalar_to_vector (f64 DPR:$src))),
4151 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
4152 def : Pat<(v4f32 (scalar_to_vector SPR:$src)),
4153 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
4155 def : Pat<(v8i8 (scalar_to_vector GPR:$src)),
4156 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4157 def : Pat<(v4i16 (scalar_to_vector GPR:$src)),
4158 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4159 def : Pat<(v2i32 (scalar_to_vector GPR:$src)),
4160 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4162 def : Pat<(v16i8 (scalar_to_vector GPR:$src)),
4163 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4164 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4166 def : Pat<(v8i16 (scalar_to_vector GPR:$src)),
4167 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
4168 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4170 def : Pat<(v4i32 (scalar_to_vector GPR:$src)),
4171 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
4172 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4175 // VDUP : Vector Duplicate (from ARM core register to all elements)
4177 class VDUPD<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
4178 : NVDup<opcod1, 0b1011, opcod3, (outs DPR:$dst), (ins GPR:$src),
4179 IIC_VMOVIS, "vdup", Dt, "$dst, $src",
4180 [(set DPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
4181 class VDUPQ<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
4182 : NVDup<opcod1, 0b1011, opcod3, (outs QPR:$dst), (ins GPR:$src),
4183 IIC_VMOVIS, "vdup", Dt, "$dst, $src",
4184 [(set QPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
4186 def VDUP8d : VDUPD<0b11101100, 0b00, "8", v8i8>;
4187 def VDUP16d : VDUPD<0b11101000, 0b01, "16", v4i16>;
4188 def VDUP32d : VDUPD<0b11101000, 0b00, "32", v2i32>;
4189 def VDUP8q : VDUPQ<0b11101110, 0b00, "8", v16i8>;
4190 def VDUP16q : VDUPQ<0b11101010, 0b01, "16", v8i16>;
4191 def VDUP32q : VDUPQ<0b11101010, 0b00, "32", v4i32>;
4193 def VDUPfd : NVDup<0b11101000, 0b1011, 0b00, (outs DPR:$dst), (ins GPR:$src),
4194 IIC_VMOVIS, "vdup", "32", "$dst, $src",
4195 [(set DPR:$dst, (v2f32 (NEONvdup
4196 (f32 (bitconvert GPR:$src)))))]>;
4197 def VDUPfq : NVDup<0b11101010, 0b1011, 0b00, (outs QPR:$dst), (ins GPR:$src),
4198 IIC_VMOVIS, "vdup", "32", "$dst, $src",
4199 [(set QPR:$dst, (v4f32 (NEONvdup
4200 (f32 (bitconvert GPR:$src)))))]>;
4202 // VDUP : Vector Duplicate Lane (from scalar to all elements)
4204 class VDUPLND<bits<4> op19_16, string OpcodeStr, string Dt,
4206 : NVDupLane<op19_16, 0, (outs DPR:$dst), (ins DPR:$src, nohash_imm:$lane),
4207 IIC_VMOVD, OpcodeStr, Dt, "$dst, $src[$lane]",
4208 [(set DPR:$dst, (Ty (NEONvduplane (Ty DPR:$src), imm:$lane)))]>;
4210 class VDUPLNQ<bits<4> op19_16, string OpcodeStr, string Dt,
4211 ValueType ResTy, ValueType OpTy>
4212 : NVDupLane<op19_16, 1, (outs QPR:$dst), (ins DPR:$src, nohash_imm:$lane),
4213 IIC_VMOVQ, OpcodeStr, Dt, "$dst, $src[$lane]",
4214 [(set QPR:$dst, (ResTy (NEONvduplane (OpTy DPR:$src),
4217 // Inst{19-16} is partially specified depending on the element size.
4219 def VDUPLN8d : VDUPLND<{?,?,?,1}, "vdup", "8", v8i8> {
4220 let Inst{19-17} = lane{2-0};
4222 def VDUPLN16d : VDUPLND<{?,?,1,0}, "vdup", "16", v4i16> {
4223 let Inst{19-18} = lane{1-0};
4225 def VDUPLN32d : VDUPLND<{?,1,0,0}, "vdup", "32", v2i32> {
4226 let Inst{19} = lane{0};
4228 def VDUPLNfd : VDUPLND<{?,1,0,0}, "vdup", "32", v2f32> {
4229 let Inst{19} = lane{0};
4231 def VDUPLN8q : VDUPLNQ<{?,?,?,1}, "vdup", "8", v16i8, v8i8> {
4232 let Inst{19-17} = lane{2-0};
4234 def VDUPLN16q : VDUPLNQ<{?,?,1,0}, "vdup", "16", v8i16, v4i16> {
4235 let Inst{19-18} = lane{1-0};
4237 def VDUPLN32q : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4i32, v2i32> {
4238 let Inst{19} = lane{0};
4240 def VDUPLNfq : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4f32, v2f32> {
4241 let Inst{19} = lane{0};
4244 def : Pat<(v16i8 (NEONvduplane (v16i8 QPR:$src), imm:$lane)),
4245 (v16i8 (VDUPLN8q (v8i8 (EXTRACT_SUBREG QPR:$src,
4246 (DSubReg_i8_reg imm:$lane))),
4247 (SubReg_i8_lane imm:$lane)))>;
4248 def : Pat<(v8i16 (NEONvduplane (v8i16 QPR:$src), imm:$lane)),
4249 (v8i16 (VDUPLN16q (v4i16 (EXTRACT_SUBREG QPR:$src,
4250 (DSubReg_i16_reg imm:$lane))),
4251 (SubReg_i16_lane imm:$lane)))>;
4252 def : Pat<(v4i32 (NEONvduplane (v4i32 QPR:$src), imm:$lane)),
4253 (v4i32 (VDUPLN32q (v2i32 (EXTRACT_SUBREG QPR:$src,
4254 (DSubReg_i32_reg imm:$lane))),
4255 (SubReg_i32_lane imm:$lane)))>;
4256 def : Pat<(v4f32 (NEONvduplane (v4f32 QPR:$src), imm:$lane)),
4257 (v4f32 (VDUPLNfq (v2f32 (EXTRACT_SUBREG QPR:$src,
4258 (DSubReg_i32_reg imm:$lane))),
4259 (SubReg_i32_lane imm:$lane)))>;
4261 def VDUPfdf : PseudoNeonI<(outs DPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
4262 [(set DPR:$dst, (v2f32 (NEONvdup (f32 SPR:$src))))]>;
4263 def VDUPfqf : PseudoNeonI<(outs QPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
4264 [(set QPR:$dst, (v4f32 (NEONvdup (f32 SPR:$src))))]>;
4266 // VMOVN : Vector Narrowing Move
4267 defm VMOVN : N2VN_HSD<0b11,0b11,0b10,0b00100,0,0, IIC_VMOVN,
4268 "vmovn", "i", trunc>;
4269 // VQMOVN : Vector Saturating Narrowing Move
4270 defm VQMOVNs : N2VNInt_HSD<0b11,0b11,0b10,0b00101,0,0, IIC_VQUNAiD,
4271 "vqmovn", "s", int_arm_neon_vqmovns>;
4272 defm VQMOVNu : N2VNInt_HSD<0b11,0b11,0b10,0b00101,1,0, IIC_VQUNAiD,
4273 "vqmovn", "u", int_arm_neon_vqmovnu>;
4274 defm VQMOVNsu : N2VNInt_HSD<0b11,0b11,0b10,0b00100,1,0, IIC_VQUNAiD,
4275 "vqmovun", "s", int_arm_neon_vqmovnsu>;
4276 // VMOVL : Vector Lengthening Move
4277 defm VMOVLs : N2VL_QHS<0b01,0b10100,0,1, "vmovl", "s", sext>;
4278 defm VMOVLu : N2VL_QHS<0b11,0b10100,0,1, "vmovl", "u", zext>;
4280 // Vector Conversions.
4282 // VCVT : Vector Convert Between Floating-Point and Integers
4283 def VCVTf2sd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
4284 v2i32, v2f32, fp_to_sint>;
4285 def VCVTf2ud : N2VD<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
4286 v2i32, v2f32, fp_to_uint>;
4287 def VCVTs2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
4288 v2f32, v2i32, sint_to_fp>;
4289 def VCVTu2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
4290 v2f32, v2i32, uint_to_fp>;
4292 def VCVTf2sq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
4293 v4i32, v4f32, fp_to_sint>;
4294 def VCVTf2uq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
4295 v4i32, v4f32, fp_to_uint>;
4296 def VCVTs2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
4297 v4f32, v4i32, sint_to_fp>;
4298 def VCVTu2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
4299 v4f32, v4i32, uint_to_fp>;
4301 // VCVT : Vector Convert Between Floating-Point and Fixed-Point.
4302 def VCVTf2xsd : N2VCvtD<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
4303 v2i32, v2f32, int_arm_neon_vcvtfp2fxs>;
4304 def VCVTf2xud : N2VCvtD<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
4305 v2i32, v2f32, int_arm_neon_vcvtfp2fxu>;
4306 def VCVTxs2fd : N2VCvtD<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
4307 v2f32, v2i32, int_arm_neon_vcvtfxs2fp>;
4308 def VCVTxu2fd : N2VCvtD<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
4309 v2f32, v2i32, int_arm_neon_vcvtfxu2fp>;
4311 def VCVTf2xsq : N2VCvtQ<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
4312 v4i32, v4f32, int_arm_neon_vcvtfp2fxs>;
4313 def VCVTf2xuq : N2VCvtQ<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
4314 v4i32, v4f32, int_arm_neon_vcvtfp2fxu>;
4315 def VCVTxs2fq : N2VCvtQ<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
4316 v4f32, v4i32, int_arm_neon_vcvtfxs2fp>;
4317 def VCVTxu2fq : N2VCvtQ<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
4318 v4f32, v4i32, int_arm_neon_vcvtfxu2fp>;
4322 // VREV64 : Vector Reverse elements within 64-bit doublewords
4324 class VREV64D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4325 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 0, 0, (outs DPR:$Vd),
4326 (ins DPR:$Vm), IIC_VMOVD,
4327 OpcodeStr, Dt, "$Vd, $Vm", "",
4328 [(set DPR:$Vd, (Ty (NEONvrev64 (Ty DPR:$Vm))))]>;
4329 class VREV64Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4330 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 1, 0, (outs QPR:$Vd),
4331 (ins QPR:$Vm), IIC_VMOVQ,
4332 OpcodeStr, Dt, "$Vd, $Vm", "",
4333 [(set QPR:$Vd, (Ty (NEONvrev64 (Ty QPR:$Vm))))]>;
4335 def VREV64d8 : VREV64D<0b00, "vrev64", "8", v8i8>;
4336 def VREV64d16 : VREV64D<0b01, "vrev64", "16", v4i16>;
4337 def VREV64d32 : VREV64D<0b10, "vrev64", "32", v2i32>;
4338 def VREV64df : VREV64D<0b10, "vrev64", "32", v2f32>;
4340 def VREV64q8 : VREV64Q<0b00, "vrev64", "8", v16i8>;
4341 def VREV64q16 : VREV64Q<0b01, "vrev64", "16", v8i16>;
4342 def VREV64q32 : VREV64Q<0b10, "vrev64", "32", v4i32>;
4343 def VREV64qf : VREV64Q<0b10, "vrev64", "32", v4f32>;
4345 // VREV32 : Vector Reverse elements within 32-bit words
4347 class VREV32D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4348 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 0, 0, (outs DPR:$Vd),
4349 (ins DPR:$Vm), IIC_VMOVD,
4350 OpcodeStr, Dt, "$Vd, $Vm", "",
4351 [(set DPR:$Vd, (Ty (NEONvrev32 (Ty DPR:$Vm))))]>;
4352 class VREV32Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4353 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 1, 0, (outs QPR:$Vd),
4354 (ins QPR:$Vm), IIC_VMOVQ,
4355 OpcodeStr, Dt, "$Vd, $Vm", "",
4356 [(set QPR:$Vd, (Ty (NEONvrev32 (Ty QPR:$Vm))))]>;
4358 def VREV32d8 : VREV32D<0b00, "vrev32", "8", v8i8>;
4359 def VREV32d16 : VREV32D<0b01, "vrev32", "16", v4i16>;
4361 def VREV32q8 : VREV32Q<0b00, "vrev32", "8", v16i8>;
4362 def VREV32q16 : VREV32Q<0b01, "vrev32", "16", v8i16>;
4364 // VREV16 : Vector Reverse elements within 16-bit halfwords
4366 class VREV16D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4367 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 0, 0, (outs DPR:$Vd),
4368 (ins DPR:$Vm), IIC_VMOVD,
4369 OpcodeStr, Dt, "$Vd, $Vm", "",
4370 [(set DPR:$Vd, (Ty (NEONvrev16 (Ty DPR:$Vm))))]>;
4371 class VREV16Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4372 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 1, 0, (outs QPR:$Vd),
4373 (ins QPR:$Vm), IIC_VMOVQ,
4374 OpcodeStr, Dt, "$Vd, $Vm", "",
4375 [(set QPR:$Vd, (Ty (NEONvrev16 (Ty QPR:$Vm))))]>;
4377 def VREV16d8 : VREV16D<0b00, "vrev16", "8", v8i8>;
4378 def VREV16q8 : VREV16Q<0b00, "vrev16", "8", v16i8>;
4380 // Other Vector Shuffles.
4382 // VEXT : Vector Extract
4384 class VEXTd<string OpcodeStr, string Dt, ValueType Ty>
4385 : N3V<0,1,0b11,{?,?,?,?},0,0, (outs DPR:$Vd),
4386 (ins DPR:$Vn, DPR:$Vm, i32imm:$index), NVExtFrm,
4387 IIC_VEXTD, OpcodeStr, Dt, "$Vd, $Vn, $Vm, $index", "",
4388 [(set DPR:$Vd, (Ty (NEONvext (Ty DPR:$Vn),
4389 (Ty DPR:$Vm), imm:$index)))]> {
4391 let Inst{11-8} = index{3-0};
4394 class VEXTq<string OpcodeStr, string Dt, ValueType Ty>
4395 : N3V<0,1,0b11,{?,?,?,?},1,0, (outs QPR:$Vd),
4396 (ins QPR:$Vn, QPR:$Vm, i32imm:$index), NVExtFrm,
4397 IIC_VEXTQ, OpcodeStr, Dt, "$Vd, $Vn, $Vm, $index", "",
4398 [(set QPR:$Vd, (Ty (NEONvext (Ty QPR:$Vn),
4399 (Ty QPR:$Vm), imm:$index)))]> {
4401 let Inst{11-8} = index{3-0};
4404 def VEXTd8 : VEXTd<"vext", "8", v8i8> {
4405 let Inst{11-8} = index{3-0};
4407 def VEXTd16 : VEXTd<"vext", "16", v4i16> {
4408 let Inst{11-9} = index{2-0};
4411 def VEXTd32 : VEXTd<"vext", "32", v2i32> {
4412 let Inst{11-10} = index{1-0};
4413 let Inst{9-8} = 0b00;
4415 def VEXTdf : VEXTd<"vext", "32", v2f32> {
4416 let Inst{11} = index{0};
4417 let Inst{10-8} = 0b000;
4420 def VEXTq8 : VEXTq<"vext", "8", v16i8> {
4421 let Inst{11-8} = index{3-0};
4423 def VEXTq16 : VEXTq<"vext", "16", v8i16> {
4424 let Inst{11-9} = index{2-0};
4427 def VEXTq32 : VEXTq<"vext", "32", v4i32> {
4428 let Inst{11-10} = index{1-0};
4429 let Inst{9-8} = 0b00;
4431 def VEXTqf : VEXTq<"vext", "32", v4f32> {
4432 let Inst{11} = index{0};
4433 let Inst{10-8} = 0b000;
4436 // VTRN : Vector Transpose
4438 def VTRNd8 : N2VDShuffle<0b00, 0b00001, "vtrn", "8">;
4439 def VTRNd16 : N2VDShuffle<0b01, 0b00001, "vtrn", "16">;
4440 def VTRNd32 : N2VDShuffle<0b10, 0b00001, "vtrn", "32">;
4442 def VTRNq8 : N2VQShuffle<0b00, 0b00001, IIC_VPERMQ, "vtrn", "8">;
4443 def VTRNq16 : N2VQShuffle<0b01, 0b00001, IIC_VPERMQ, "vtrn", "16">;
4444 def VTRNq32 : N2VQShuffle<0b10, 0b00001, IIC_VPERMQ, "vtrn", "32">;
4446 // VUZP : Vector Unzip (Deinterleave)
4448 def VUZPd8 : N2VDShuffle<0b00, 0b00010, "vuzp", "8">;
4449 def VUZPd16 : N2VDShuffle<0b01, 0b00010, "vuzp", "16">;
4450 def VUZPd32 : N2VDShuffle<0b10, 0b00010, "vuzp", "32">;
4452 def VUZPq8 : N2VQShuffle<0b00, 0b00010, IIC_VPERMQ3, "vuzp", "8">;
4453 def VUZPq16 : N2VQShuffle<0b01, 0b00010, IIC_VPERMQ3, "vuzp", "16">;
4454 def VUZPq32 : N2VQShuffle<0b10, 0b00010, IIC_VPERMQ3, "vuzp", "32">;
4456 // VZIP : Vector Zip (Interleave)
4458 def VZIPd8 : N2VDShuffle<0b00, 0b00011, "vzip", "8">;
4459 def VZIPd16 : N2VDShuffle<0b01, 0b00011, "vzip", "16">;
4460 def VZIPd32 : N2VDShuffle<0b10, 0b00011, "vzip", "32">;
4462 def VZIPq8 : N2VQShuffle<0b00, 0b00011, IIC_VPERMQ3, "vzip", "8">;
4463 def VZIPq16 : N2VQShuffle<0b01, 0b00011, IIC_VPERMQ3, "vzip", "16">;
4464 def VZIPq32 : N2VQShuffle<0b10, 0b00011, IIC_VPERMQ3, "vzip", "32">;
4466 // Vector Table Lookup and Table Extension.
4468 // VTBL : Vector Table Lookup
4470 : N3V<1,1,0b11,0b1000,0,0, (outs DPR:$Vd),
4471 (ins DPR:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTB1,
4472 "vtbl", "8", "$Vd, \\{$Vn\\}, $Vm", "",
4473 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbl1 DPR:$Vn, DPR:$Vm)))]>;
4474 let hasExtraSrcRegAllocReq = 1 in {
4476 : N3V<1,1,0b11,0b1001,0,0, (outs DPR:$Vd),
4477 (ins DPR:$Vn, DPR:$tbl2, DPR:$Vm), NVTBLFrm, IIC_VTB2,
4478 "vtbl", "8", "$Vd, \\{$Vn, $tbl2\\}, $Vm", "", []>;
4480 : N3V<1,1,0b11,0b1010,0,0, (outs DPR:$Vd),
4481 (ins DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$Vm), NVTBLFrm, IIC_VTB3,
4482 "vtbl", "8", "$Vd, \\{$Vn, $tbl2, $tbl3\\}, $Vm", "", []>;
4484 : N3V<1,1,0b11,0b1011,0,0, (outs DPR:$Vd),
4485 (ins DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$Vm),
4487 "vtbl", "8", "$Vd, \\{$Vn, $tbl2, $tbl3, $tbl4\\}, $Vm", "", []>;
4488 } // hasExtraSrcRegAllocReq = 1
4491 : PseudoNeonI<(outs DPR:$dst), (ins QPR:$tbl, DPR:$src), IIC_VTB2, "", []>;
4493 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB3, "", []>;
4495 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB4, "", []>;
4497 // VTBX : Vector Table Extension
4499 : N3V<1,1,0b11,0b1000,1,0, (outs DPR:$Vd),
4500 (ins DPR:$orig, DPR:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTBX1,
4501 "vtbx", "8", "$Vd, \\{$Vn\\}, $Vm", "$orig = $Vd",
4502 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbx1
4503 DPR:$orig, DPR:$Vn, DPR:$Vm)))]>;
4504 let hasExtraSrcRegAllocReq = 1 in {
4506 : N3V<1,1,0b11,0b1001,1,0, (outs DPR:$Vd),
4507 (ins DPR:$orig, DPR:$Vn, DPR:$tbl2, DPR:$Vm), NVTBLFrm, IIC_VTBX2,
4508 "vtbx", "8", "$Vd, \\{$Vn, $tbl2\\}, $Vm", "$orig = $Vd", []>;
4510 : N3V<1,1,0b11,0b1010,1,0, (outs DPR:$Vd),
4511 (ins DPR:$orig, DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$Vm),
4512 NVTBLFrm, IIC_VTBX3,
4513 "vtbx", "8", "$Vd, \\{$Vn, $tbl2, $tbl3\\}, $Vm",
4516 : N3V<1,1,0b11,0b1011,1,0, (outs DPR:$Vd), (ins DPR:$orig, DPR:$Vn,
4517 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$Vm), NVTBLFrm, IIC_VTBX4,
4518 "vtbx", "8", "$Vd, \\{$Vn, $tbl2, $tbl3, $tbl4\\}, $Vm",
4520 } // hasExtraSrcRegAllocReq = 1
4523 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QPR:$tbl, DPR:$src),
4524 IIC_VTBX2, "$orig = $dst", []>;
4526 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
4527 IIC_VTBX3, "$orig = $dst", []>;
4529 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
4530 IIC_VTBX4, "$orig = $dst", []>;
4532 //===----------------------------------------------------------------------===//
4533 // NEON instructions for single-precision FP math
4534 //===----------------------------------------------------------------------===//
4536 class N2VSPat<SDNode OpNode, ValueType ResTy, ValueType OpTy, NeonI Inst>
4537 : NEONFPPat<(ResTy (OpNode SPR:$a)),
4538 (EXTRACT_SUBREG (OpTy (Inst (INSERT_SUBREG (OpTy (IMPLICIT_DEF)),
4542 class N3VSPat<SDNode OpNode, NeonI Inst>
4543 : NEONFPPat<(f32 (OpNode SPR:$a, SPR:$b)),
4544 (EXTRACT_SUBREG (v2f32
4545 (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
4547 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
4551 class N3VSMulOpPat<SDNode MulNode, SDNode OpNode, NeonI Inst>
4552 : NEONFPPat<(f32 (OpNode SPR:$acc, (f32 (MulNode SPR:$a, SPR:$b)))),
4553 (EXTRACT_SUBREG (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
4555 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
4557 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
4561 // These need separate instructions because they must use DPR_VFP2 register
4562 // class which have SPR sub-registers.
4564 // Vector Add Operations used for single-precision FP
4565 let neverHasSideEffects = 1 in
4566 def VADDfd_sfp : N3VS<0,0,0b00,0b1101,0, "vadd", "f32", v2f32, v2f32, fadd, 1>;
4567 def : N3VSPat<fadd, VADDfd_sfp>;
4569 // Vector Sub Operations used for single-precision FP
4570 let neverHasSideEffects = 1 in
4571 def VSUBfd_sfp : N3VS<0,0,0b10,0b1101,0, "vsub", "f32", v2f32, v2f32, fsub, 0>;
4572 def : N3VSPat<fsub, VSUBfd_sfp>;
4574 // Vector Multiply Operations used for single-precision FP
4575 let neverHasSideEffects = 1 in
4576 def VMULfd_sfp : N3VS<1,0,0b00,0b1101,1, "vmul", "f32", v2f32, v2f32, fmul, 1>;
4577 def : N3VSPat<fmul, VMULfd_sfp>;
4579 // Vector Multiply-Accumulate/Subtract used for single-precision FP
4580 // vml[as].f32 can cause 4-8 cycle stalls in following ASIMD instructions, so
4581 // we want to avoid them for now. e.g., alternating vmla/vadd instructions.
4583 //let neverHasSideEffects = 1 in
4584 //def VMLAfd_sfp : N3VSMulOp<0,0,0b00,0b1101,1, IIC_VMACD, "vmla", "f32",
4585 // v2f32, fmul, fadd>;
4586 //def : N3VSMulOpPat<fmul, fadd, VMLAfd_sfp>;
4588 //let neverHasSideEffects = 1 in
4589 //def VMLSfd_sfp : N3VSMulOp<0,0,0b10,0b1101,1, IIC_VMACD, "vmls", "f32",
4590 // v2f32, fmul, fsub>;
4591 //def : N3VSMulOpPat<fmul, fsub, VMLSfd_sfp>;
4593 // Vector Absolute used for single-precision FP
4594 let neverHasSideEffects = 1 in
4595 def VABSfd_sfp : N2V<0b11, 0b11, 0b10, 0b01, 0b01110, 0, 0,
4596 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), IIC_VUNAD,
4597 "vabs", "f32", "$dst, $src", "", []>;
4598 def : N2VSPat<fabs, f32, v2f32, VABSfd_sfp>;
4600 // Vector Negate used for single-precision FP
4601 let neverHasSideEffects = 1 in
4602 def VNEGfd_sfp : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
4603 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), IIC_VUNAD,
4604 "vneg", "f32", "$dst, $src", "", []>;
4605 def : N2VSPat<fneg, f32, v2f32, VNEGfd_sfp>;
4607 // Vector Maximum used for single-precision FP
4608 let neverHasSideEffects = 1 in
4609 def VMAXfd_sfp : N3V<0, 0, 0b00, 0b1111, 0, 0, (outs DPR_VFP2:$dst),
4610 (ins DPR_VFP2:$src1, DPR_VFP2:$src2), N3RegFrm, IIC_VBIND,
4611 "vmax", "f32", "$dst, $src1, $src2", "", []>;
4612 def : N3VSPat<NEONfmax, VMAXfd_sfp>;
4614 // Vector Minimum used for single-precision FP
4615 let neverHasSideEffects = 1 in
4616 def VMINfd_sfp : N3V<0, 0, 0b00, 0b1111, 0, 0, (outs DPR_VFP2:$dst),
4617 (ins DPR_VFP2:$src1, DPR_VFP2:$src2), N3RegFrm, IIC_VBIND,
4618 "vmin", "f32", "$dst, $src1, $src2", "", []>;
4619 def : N3VSPat<NEONfmin, VMINfd_sfp>;
4621 // Vector Convert between single-precision FP and integer
4622 let neverHasSideEffects = 1 in
4623 def VCVTf2sd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
4624 v2i32, v2f32, fp_to_sint>;
4625 def : N2VSPat<arm_ftosi, f32, v2f32, VCVTf2sd_sfp>;
4627 let neverHasSideEffects = 1 in
4628 def VCVTf2ud_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
4629 v2i32, v2f32, fp_to_uint>;
4630 def : N2VSPat<arm_ftoui, f32, v2f32, VCVTf2ud_sfp>;
4632 let neverHasSideEffects = 1 in
4633 def VCVTs2fd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
4634 v2f32, v2i32, sint_to_fp>;
4635 def : N2VSPat<arm_sitof, f32, v2i32, VCVTs2fd_sfp>;
4637 let neverHasSideEffects = 1 in
4638 def VCVTu2fd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
4639 v2f32, v2i32, uint_to_fp>;
4640 def : N2VSPat<arm_uitof, f32, v2i32, VCVTu2fd_sfp>;
4642 //===----------------------------------------------------------------------===//
4643 // Non-Instruction Patterns
4644 //===----------------------------------------------------------------------===//
4647 def : Pat<(v1i64 (bitconvert (v2i32 DPR:$src))), (v1i64 DPR:$src)>;
4648 def : Pat<(v1i64 (bitconvert (v4i16 DPR:$src))), (v1i64 DPR:$src)>;
4649 def : Pat<(v1i64 (bitconvert (v8i8 DPR:$src))), (v1i64 DPR:$src)>;
4650 def : Pat<(v1i64 (bitconvert (f64 DPR:$src))), (v1i64 DPR:$src)>;
4651 def : Pat<(v1i64 (bitconvert (v2f32 DPR:$src))), (v1i64 DPR:$src)>;
4652 def : Pat<(v2i32 (bitconvert (v1i64 DPR:$src))), (v2i32 DPR:$src)>;
4653 def : Pat<(v2i32 (bitconvert (v4i16 DPR:$src))), (v2i32 DPR:$src)>;
4654 def : Pat<(v2i32 (bitconvert (v8i8 DPR:$src))), (v2i32 DPR:$src)>;
4655 def : Pat<(v2i32 (bitconvert (f64 DPR:$src))), (v2i32 DPR:$src)>;
4656 def : Pat<(v2i32 (bitconvert (v2f32 DPR:$src))), (v2i32 DPR:$src)>;
4657 def : Pat<(v4i16 (bitconvert (v1i64 DPR:$src))), (v4i16 DPR:$src)>;
4658 def : Pat<(v4i16 (bitconvert (v2i32 DPR:$src))), (v4i16 DPR:$src)>;
4659 def : Pat<(v4i16 (bitconvert (v8i8 DPR:$src))), (v4i16 DPR:$src)>;
4660 def : Pat<(v4i16 (bitconvert (f64 DPR:$src))), (v4i16 DPR:$src)>;
4661 def : Pat<(v4i16 (bitconvert (v2f32 DPR:$src))), (v4i16 DPR:$src)>;
4662 def : Pat<(v8i8 (bitconvert (v1i64 DPR:$src))), (v8i8 DPR:$src)>;
4663 def : Pat<(v8i8 (bitconvert (v2i32 DPR:$src))), (v8i8 DPR:$src)>;
4664 def : Pat<(v8i8 (bitconvert (v4i16 DPR:$src))), (v8i8 DPR:$src)>;
4665 def : Pat<(v8i8 (bitconvert (f64 DPR:$src))), (v8i8 DPR:$src)>;
4666 def : Pat<(v8i8 (bitconvert (v2f32 DPR:$src))), (v8i8 DPR:$src)>;
4667 def : Pat<(f64 (bitconvert (v1i64 DPR:$src))), (f64 DPR:$src)>;
4668 def : Pat<(f64 (bitconvert (v2i32 DPR:$src))), (f64 DPR:$src)>;
4669 def : Pat<(f64 (bitconvert (v4i16 DPR:$src))), (f64 DPR:$src)>;
4670 def : Pat<(f64 (bitconvert (v8i8 DPR:$src))), (f64 DPR:$src)>;
4671 def : Pat<(f64 (bitconvert (v2f32 DPR:$src))), (f64 DPR:$src)>;
4672 def : Pat<(v2f32 (bitconvert (f64 DPR:$src))), (v2f32 DPR:$src)>;
4673 def : Pat<(v2f32 (bitconvert (v1i64 DPR:$src))), (v2f32 DPR:$src)>;
4674 def : Pat<(v2f32 (bitconvert (v2i32 DPR:$src))), (v2f32 DPR:$src)>;
4675 def : Pat<(v2f32 (bitconvert (v4i16 DPR:$src))), (v2f32 DPR:$src)>;
4676 def : Pat<(v2f32 (bitconvert (v8i8 DPR:$src))), (v2f32 DPR:$src)>;
4678 def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (v2i64 QPR:$src)>;
4679 def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (v2i64 QPR:$src)>;
4680 def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (v2i64 QPR:$src)>;
4681 def : Pat<(v2i64 (bitconvert (v2f64 QPR:$src))), (v2i64 QPR:$src)>;
4682 def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (v2i64 QPR:$src)>;
4683 def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (v4i32 QPR:$src)>;
4684 def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (v4i32 QPR:$src)>;
4685 def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (v4i32 QPR:$src)>;
4686 def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (v4i32 QPR:$src)>;
4687 def : Pat<(v4i32 (bitconvert (v4f32 QPR:$src))), (v4i32 QPR:$src)>;
4688 def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (v8i16 QPR:$src)>;
4689 def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (v8i16 QPR:$src)>;
4690 def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (v8i16 QPR:$src)>;
4691 def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (v8i16 QPR:$src)>;
4692 def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (v8i16 QPR:$src)>;
4693 def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (v16i8 QPR:$src)>;
4694 def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (v16i8 QPR:$src)>;
4695 def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (v16i8 QPR:$src)>;
4696 def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (v16i8 QPR:$src)>;
4697 def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (v16i8 QPR:$src)>;
4698 def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (v4f32 QPR:$src)>;
4699 def : Pat<(v4f32 (bitconvert (v4i32 QPR:$src))), (v4f32 QPR:$src)>;
4700 def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (v4f32 QPR:$src)>;
4701 def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (v4f32 QPR:$src)>;
4702 def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (v4f32 QPR:$src)>;
4703 def : Pat<(v2f64 (bitconvert (v2i64 QPR:$src))), (v2f64 QPR:$src)>;
4704 def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (v2f64 QPR:$src)>;
4705 def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 QPR:$src)>;
4706 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 QPR:$src)>;
4707 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 QPR:$src)>;