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:$addr)),
532 def VLD1LNd8 : VLD1LN<0b0000, {?,?,?,0}, "8", v8i8, extloadi8> {
533 let Inst{7-5} = lane{2-0};
535 def VLD1LNd16 : VLD1LN<0b0100, {?,?,0,?}, "16", v4i16, extloadi16> {
536 let Inst{7-6} = lane{1-0};
539 def VLD1LNd32 : VLD1LN<0b1000, {?,0,?,?}, "32", v2i32, load> {
540 let Inst{7} = lane{0};
545 def VLD1LNq8Pseudo : VLD1QLNPseudo<v16i8, extloadi8>;
546 def VLD1LNq16Pseudo : VLD1QLNPseudo<v8i16, extloadi16>;
547 def VLD1LNq32Pseudo : VLD1QLNPseudo<v4i32, load>;
549 def : Pat<(vector_insert (v2f32 DPR:$src),
550 (f32 (load addrmode6:$addr)), imm:$lane),
551 (VLD1LNd32 addrmode6:$addr, DPR:$src, imm:$lane)>;
552 def : Pat<(vector_insert (v4f32 QPR:$src),
553 (f32 (load addrmode6:$addr)), imm:$lane),
554 (VLD1LNq32Pseudo addrmode6:$addr, QPR:$src, imm:$lane)>;
556 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
558 // ...with address register writeback:
559 class VLD1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
560 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, GPR:$wb),
561 (ins addrmode6:$Rn, am6offset:$Rm,
562 DPR:$src, nohash_imm:$lane), IIC_VLD1lnu, "vld1", Dt,
563 "\\{$Vd[$lane]\\}, $Rn$Rm",
564 "$src = $Vd, $Rn.addr = $wb", []>;
566 def VLD1LNd8_UPD : VLD1LNWB<0b0000, {?,?,?,0}, "8"> {
567 let Inst{7-5} = lane{2-0};
569 def VLD1LNd16_UPD : VLD1LNWB<0b0100, {?,?,0,?}, "16"> {
570 let Inst{7-6} = lane{1-0};
573 def VLD1LNd32_UPD : VLD1LNWB<0b1000, {?,0,?,?}, "32"> {
574 let Inst{7} = lane{0};
579 def VLD1LNq8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
580 def VLD1LNq16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
581 def VLD1LNq32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
583 // VLD2LN : Vector Load (single 2-element structure to one lane)
584 class VLD2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
585 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2),
586 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, nohash_imm:$lane),
587 IIC_VLD2ln, "vld2", Dt, "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn",
588 "$src1 = $Vd, $src2 = $dst2", []> {
593 def VLD2LNd8 : VLD2LN<0b0001, {?,?,?,?}, "8"> {
594 let Inst{7-5} = lane{2-0};
596 def VLD2LNd16 : VLD2LN<0b0101, {?,?,0,?}, "16"> {
597 let Inst{7-6} = lane{1-0};
599 def VLD2LNd32 : VLD2LN<0b1001, {?,0,0,?}, "32"> {
600 let Inst{7} = lane{0};
603 def VLD2LNd8Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
604 def VLD2LNd16Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
605 def VLD2LNd32Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
607 // ...with double-spaced registers:
608 def VLD2LNq16 : VLD2LN<0b0101, {?,?,1,?}, "16"> {
609 let Inst{7-6} = lane{1-0};
611 def VLD2LNq32 : VLD2LN<0b1001, {?,1,0,?}, "32"> {
612 let Inst{7} = lane{0};
615 def VLD2LNq16Pseudo : VLDQQLNPseudo<IIC_VLD2ln>;
616 def VLD2LNq32Pseudo : VLDQQLNPseudo<IIC_VLD2ln>;
618 // ...with address register writeback:
619 class VLD2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
620 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
621 (ins addrmode6:$Rn, am6offset:$Rm,
622 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VLD2lnu, "vld2", Dt,
623 "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn$Rm",
624 "$src1 = $Vd, $src2 = $dst2, $Rn.addr = $wb", []> {
628 def VLD2LNd8_UPD : VLD2LNWB<0b0001, {?,?,?,?}, "8"> {
629 let Inst{7-5} = lane{2-0};
631 def VLD2LNd16_UPD : VLD2LNWB<0b0101, {?,?,0,?}, "16"> {
632 let Inst{7-6} = lane{1-0};
634 def VLD2LNd32_UPD : VLD2LNWB<0b1001, {?,0,0,?}, "32"> {
635 let Inst{7} = lane{0};
638 def VLD2LNd8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
639 def VLD2LNd16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
640 def VLD2LNd32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
642 def VLD2LNq16_UPD : VLD2LNWB<0b0101, {?,?,1,?}, "16"> {
643 let Inst{7-6} = lane{1-0};
645 def VLD2LNq32_UPD : VLD2LNWB<0b1001, {?,1,0,?}, "32"> {
646 let Inst{7} = lane{0};
649 def VLD2LNq16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
650 def VLD2LNq32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
652 // VLD3LN : Vector Load (single 3-element structure to one lane)
653 class VLD3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
654 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
655 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3,
656 nohash_imm:$lane), IIC_VLD3ln, "vld3", Dt,
657 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn",
658 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3", []> {
662 def VLD3LNd8 : VLD3LN<0b0010, {?,?,?,0}, "8"> {
663 let Inst{7-5} = lane{2-0};
665 def VLD3LNd16 : VLD3LN<0b0110, {?,?,0,0}, "16"> {
666 let Inst{7-6} = lane{1-0};
668 def VLD3LNd32 : VLD3LN<0b1010, {?,0,0,0}, "32"> {
669 let Inst{7} = lane{0};
672 def VLD3LNd8Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
673 def VLD3LNd16Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
674 def VLD3LNd32Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
676 // ...with double-spaced registers:
677 def VLD3LNq16 : VLD3LN<0b0110, {?,?,1,0}, "16"> {
678 let Inst{7-6} = lane{1-0};
680 def VLD3LNq32 : VLD3LN<0b1010, {?,1,0,0}, "32"> {
681 let Inst{7} = lane{0};
684 def VLD3LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
685 def VLD3LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
687 // ...with address register writeback:
688 class VLD3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
689 : NLdStLn<1, 0b10, op11_8, op7_4,
690 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
691 (ins addrmode6:$Rn, am6offset:$Rm,
692 DPR:$src1, DPR:$src2, DPR:$src3, nohash_imm:$lane),
693 IIC_VLD3lnu, "vld3", Dt,
694 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn$Rm",
695 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $Rn.addr = $wb",
698 def VLD3LNd8_UPD : VLD3LNWB<0b0010, {?,?,?,0}, "8"> {
699 let Inst{7-5} = lane{2-0};
701 def VLD3LNd16_UPD : VLD3LNWB<0b0110, {?,?,0,0}, "16"> {
702 let Inst{7-6} = lane{1-0};
704 def VLD3LNd32_UPD : VLD3LNWB<0b1010, {?,0,0,0}, "32"> {
705 let Inst{7} = lane{0};
708 def VLD3LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
709 def VLD3LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
710 def VLD3LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
712 def VLD3LNq16_UPD : VLD3LNWB<0b0110, {?,?,1,0}, "16"> {
713 let Inst{7-6} = lane{1-0};
715 def VLD3LNq32_UPD : VLD3LNWB<0b1010, {?,1,0,0}, "32"> {
716 let Inst{7} = lane{0};
719 def VLD3LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
720 def VLD3LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
722 // VLD4LN : Vector Load (single 4-element structure to one lane)
723 class VLD4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
724 : NLdStLn<1, 0b10, op11_8, op7_4,
725 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
726 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
727 nohash_imm:$lane), IIC_VLD4ln, "vld4", Dt,
728 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn",
729 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4", []> {
734 def VLD4LNd8 : VLD4LN<0b0011, {?,?,?,?}, "8"> {
735 let Inst{7-5} = lane{2-0};
737 def VLD4LNd16 : VLD4LN<0b0111, {?,?,0,?}, "16"> {
738 let Inst{7-6} = lane{1-0};
740 def VLD4LNd32 : VLD4LN<0b1011, {?,0,?,?}, "32"> {
741 let Inst{7} = lane{0};
745 def VLD4LNd8Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
746 def VLD4LNd16Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
747 def VLD4LNd32Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
749 // ...with double-spaced registers:
750 def VLD4LNq16 : VLD4LN<0b0111, {?,?,1,?}, "16"> {
751 let Inst{7-6} = lane{1-0};
753 def VLD4LNq32 : VLD4LN<0b1011, {?,1,?,?}, "32"> {
754 let Inst{7} = lane{0};
758 def VLD4LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>;
759 def VLD4LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>;
761 // ...with address register writeback:
762 class VLD4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
763 : NLdStLn<1, 0b10, op11_8, op7_4,
764 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
765 (ins addrmode6:$Rn, am6offset:$Rm,
766 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
767 IIC_VLD4ln, "vld4", Dt,
768 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn$Rm",
769 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4, $Rn.addr = $wb",
774 def VLD4LNd8_UPD : VLD4LNWB<0b0011, {?,?,?,?}, "8"> {
775 let Inst{7-5} = lane{2-0};
777 def VLD4LNd16_UPD : VLD4LNWB<0b0111, {?,?,0,?}, "16"> {
778 let Inst{7-6} = lane{1-0};
780 def VLD4LNd32_UPD : VLD4LNWB<0b1011, {?,0,?,?}, "32"> {
781 let Inst{7} = lane{0};
785 def VLD4LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
786 def VLD4LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
787 def VLD4LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
789 def VLD4LNq16_UPD : VLD4LNWB<0b0111, {?,?,1,?}, "16"> {
790 let Inst{7-6} = lane{1-0};
792 def VLD4LNq32_UPD : VLD4LNWB<0b1011, {?,1,?,?}, "32"> {
793 let Inst{7} = lane{0};
797 def VLD4LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
798 def VLD4LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
800 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
802 // VLD1DUP : Vector Load (single element to all lanes)
803 class VLD1DUP<bits<4> op7_4, string Dt, ValueType Ty, PatFrag LoadOp>
804 : NLdSt<1, 0b10, 0b1100, op7_4, (outs DPR:$Vd), (ins addrmode6dup:$Rn),
805 IIC_VLD1dup, "vld1", Dt, "\\{$Vd[]\\}, $Rn", "",
806 [(set DPR:$Vd, (Ty (NEONvdup (i32 (LoadOp addrmode6dup:$Rn)))))]> {
810 class VLD1QDUPPseudo<ValueType Ty, PatFrag LoadOp> : VLDQPseudo<IIC_VLD1dup> {
811 let Pattern = [(set QPR:$dst,
812 (Ty (NEONvdup (i32 (LoadOp addrmode6dup:$addr)))))];
815 def VLD1DUPd8 : VLD1DUP<{0,0,0,?}, "8", v8i8, extloadi8>;
816 def VLD1DUPd16 : VLD1DUP<{0,1,0,?}, "16", v4i16, extloadi16>;
817 def VLD1DUPd32 : VLD1DUP<{1,0,0,?}, "32", v2i32, load>;
819 def VLD1DUPq8Pseudo : VLD1QDUPPseudo<v16i8, extloadi8>;
820 def VLD1DUPq16Pseudo : VLD1QDUPPseudo<v8i16, extloadi16>;
821 def VLD1DUPq32Pseudo : VLD1QDUPPseudo<v4i32, load>;
823 def : Pat<(v2f32 (NEONvdup (f32 (load addrmode6dup:$addr)))),
824 (VLD1DUPd32 addrmode6:$addr)>;
825 def : Pat<(v4f32 (NEONvdup (f32 (load addrmode6dup:$addr)))),
826 (VLD1DUPq32Pseudo addrmode6:$addr)>;
828 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
830 class VLD1QDUP<bits<4> op7_4, string Dt>
831 : NLdSt<1, 0b10, 0b1100, op7_4, (outs DPR:$Vd, DPR:$dst2),
832 (ins addrmode6dup:$Rn), IIC_VLD1dup,
833 "vld1", Dt, "\\{$Vd[], $dst2[]\\}, $Rn", "", []> {
838 def VLD1DUPq8 : VLD1QDUP<{0,0,1,0}, "8">;
839 def VLD1DUPq16 : VLD1QDUP<{0,1,1,?}, "16">;
840 def VLD1DUPq32 : VLD1QDUP<{1,0,1,?}, "32">;
842 // ...with address register writeback:
843 class VLD1DUPWB<bits<4> op7_4, string Dt>
844 : NLdSt<1, 0b10, 0b1100, op7_4, (outs DPR:$Vd, GPR:$wb),
845 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD1dupu,
846 "vld1", Dt, "\\{$Vd[]\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
849 class VLD1QDUPWB<bits<4> op7_4, string Dt>
850 : NLdSt<1, 0b10, 0b1100, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
851 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD1dupu,
852 "vld1", Dt, "\\{$Vd[], $dst2[]\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
856 def VLD1DUPd8_UPD : VLD1DUPWB<{0,0,0,0}, "8">;
857 def VLD1DUPd16_UPD : VLD1DUPWB<{0,1,0,?}, "16">;
858 def VLD1DUPd32_UPD : VLD1DUPWB<{1,0,0,?}, "32">;
860 def VLD1DUPq8_UPD : VLD1QDUPWB<{0,0,1,0}, "8">;
861 def VLD1DUPq16_UPD : VLD1QDUPWB<{0,1,1,?}, "16">;
862 def VLD1DUPq32_UPD : VLD1QDUPWB<{1,0,1,?}, "32">;
864 def VLD1DUPq8Pseudo_UPD : VLDQWBPseudo<IIC_VLD1dupu>;
865 def VLD1DUPq16Pseudo_UPD : VLDQWBPseudo<IIC_VLD1dupu>;
866 def VLD1DUPq32Pseudo_UPD : VLDQWBPseudo<IIC_VLD1dupu>;
868 // VLD2DUP : Vector Load (single 2-element structure to all lanes)
869 class VLD2DUP<bits<4> op7_4, string Dt>
870 : NLdSt<1, 0b10, 0b1101, op7_4, (outs DPR:$Vd, DPR:$dst2),
871 (ins addrmode6dup:$Rn), IIC_VLD2dup,
872 "vld2", Dt, "\\{$Vd[], $dst2[]\\}, $Rn", "", []> {
877 def VLD2DUPd8 : VLD2DUP<{0,0,0,?}, "8">;
878 def VLD2DUPd16 : VLD2DUP<{0,1,0,?}, "16">;
879 def VLD2DUPd32 : VLD2DUP<{1,0,0,?}, "32">;
881 def VLD2DUPd8Pseudo : VLDQPseudo<IIC_VLD2dup>;
882 def VLD2DUPd16Pseudo : VLDQPseudo<IIC_VLD2dup>;
883 def VLD2DUPd32Pseudo : VLDQPseudo<IIC_VLD2dup>;
885 // ...with double-spaced registers (not used for codegen):
886 def VLD2DUPd8x2 : VLD2DUP<{0,0,1,?}, "8">;
887 def VLD2DUPd16x2 : VLD2DUP<{0,1,1,?}, "16">;
888 def VLD2DUPd32x2 : VLD2DUP<{1,0,1,?}, "32">;
890 // ...with address register writeback:
891 class VLD2DUPWB<bits<4> op7_4, string Dt>
892 : NLdSt<1, 0b10, 0b1101, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
893 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD2dupu,
894 "vld2", Dt, "\\{$Vd[], $dst2[]\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
898 def VLD2DUPd8_UPD : VLD2DUPWB<{0,0,0,0}, "8">;
899 def VLD2DUPd16_UPD : VLD2DUPWB<{0,1,0,?}, "16">;
900 def VLD2DUPd32_UPD : VLD2DUPWB<{1,0,0,?}, "32">;
902 def VLD2DUPd8x2_UPD : VLD2DUPWB<{0,0,1,0}, "8">;
903 def VLD2DUPd16x2_UPD : VLD2DUPWB<{0,1,1,?}, "16">;
904 def VLD2DUPd32x2_UPD : VLD2DUPWB<{1,0,1,?}, "32">;
906 def VLD2DUPd8Pseudo_UPD : VLDQWBPseudo<IIC_VLD2dupu>;
907 def VLD2DUPd16Pseudo_UPD : VLDQWBPseudo<IIC_VLD2dupu>;
908 def VLD2DUPd32Pseudo_UPD : VLDQWBPseudo<IIC_VLD2dupu>;
910 // VLD3DUP : Vector Load (single 3-element structure to all lanes)
911 class VLD3DUP<bits<4> op7_4, string Dt>
912 : NLdSt<1, 0b10, 0b1110, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
913 (ins addrmode6dup:$Rn), IIC_VLD3dup,
914 "vld3", Dt, "\\{$Vd[], $dst2[], $dst3[]\\}, $Rn", "", []> {
919 def VLD3DUPd8 : VLD3DUP<{0,0,0,?}, "8">;
920 def VLD3DUPd16 : VLD3DUP<{0,1,0,?}, "16">;
921 def VLD3DUPd32 : VLD3DUP<{1,0,0,?}, "32">;
923 def VLD3DUPd8Pseudo : VLDQQPseudo<IIC_VLD3dup>;
924 def VLD3DUPd16Pseudo : VLDQQPseudo<IIC_VLD3dup>;
925 def VLD3DUPd32Pseudo : VLDQQPseudo<IIC_VLD3dup>;
927 // ...with double-spaced registers (not used for codegen):
928 def VLD3DUPd8x2 : VLD3DUP<{0,0,1,?}, "8">;
929 def VLD3DUPd16x2 : VLD3DUP<{0,1,1,?}, "16">;
930 def VLD3DUPd32x2 : VLD3DUP<{1,0,1,?}, "32">;
932 // ...with address register writeback:
933 class VLD3DUPWB<bits<4> op7_4, string Dt>
934 : NLdSt<1, 0b10, 0b1110, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
935 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD3dupu,
936 "vld3", Dt, "\\{$Vd[], $dst2[], $dst3[]\\}, $Rn$Rm",
937 "$Rn.addr = $wb", []> {
941 def VLD3DUPd8_UPD : VLD3DUPWB<{0,0,0,0}, "8">;
942 def VLD3DUPd16_UPD : VLD3DUPWB<{0,1,0,?}, "16">;
943 def VLD3DUPd32_UPD : VLD3DUPWB<{1,0,0,?}, "32">;
945 def VLD3DUPd8x2_UPD : VLD3DUPWB<{0,0,1,0}, "8">;
946 def VLD3DUPd16x2_UPD : VLD3DUPWB<{0,1,1,?}, "16">;
947 def VLD3DUPd32x2_UPD : VLD3DUPWB<{1,0,1,?}, "32">;
949 def VLD3DUPd8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
950 def VLD3DUPd16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
951 def VLD3DUPd32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
953 // VLD4DUP : Vector Load (single 4-element structure to all lanes)
954 class VLD4DUP<bits<4> op7_4, string Dt>
955 : NLdSt<1, 0b10, 0b1111, op7_4,
956 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
957 (ins addrmode6dup:$Rn), IIC_VLD4dup,
958 "vld4", Dt, "\\{$Vd[], $dst2[], $dst3[], $dst4[]\\}, $Rn", "", []> {
963 def VLD4DUPd8 : VLD4DUP<{0,0,0,?}, "8">;
964 def VLD4DUPd16 : VLD4DUP<{0,1,0,?}, "16">;
965 def VLD4DUPd32 : VLD4DUP<{1,?,0,?}, "32"> { let Inst{6} = Rn{5}; }
967 def VLD4DUPd8Pseudo : VLDQQPseudo<IIC_VLD4dup>;
968 def VLD4DUPd16Pseudo : VLDQQPseudo<IIC_VLD4dup>;
969 def VLD4DUPd32Pseudo : VLDQQPseudo<IIC_VLD4dup>;
971 // ...with double-spaced registers (not used for codegen):
972 def VLD4DUPd8x2 : VLD4DUP<{0,0,1,?}, "8">;
973 def VLD4DUPd16x2 : VLD4DUP<{0,1,1,?}, "16">;
974 def VLD4DUPd32x2 : VLD4DUP<{1,?,1,?}, "32"> { let Inst{6} = Rn{5}; }
976 // ...with address register writeback:
977 class VLD4DUPWB<bits<4> op7_4, string Dt>
978 : NLdSt<1, 0b10, 0b1111, op7_4,
979 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
980 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD4dupu,
981 "vld4", Dt, "\\{$Vd[], $dst2[], $dst3[], $dst4[]\\}, $Rn$Rm",
982 "$Rn.addr = $wb", []> {
986 def VLD4DUPd8_UPD : VLD4DUPWB<{0,0,0,0}, "8">;
987 def VLD4DUPd16_UPD : VLD4DUPWB<{0,1,0,?}, "16">;
988 def VLD4DUPd32_UPD : VLD4DUPWB<{1,?,0,?}, "32"> { let Inst{6} = Rn{5}; }
990 def VLD4DUPd8x2_UPD : VLD4DUPWB<{0,0,1,0}, "8">;
991 def VLD4DUPd16x2_UPD : VLD4DUPWB<{0,1,1,?}, "16">;
992 def VLD4DUPd32x2_UPD : VLD4DUPWB<{1,?,1,?}, "32"> { let Inst{6} = Rn{5}; }
994 def VLD4DUPd8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
995 def VLD4DUPd16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
996 def VLD4DUPd32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
998 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
1000 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
1002 // Classes for VST* pseudo-instructions with multi-register operands.
1003 // These are expanded to real instructions after register allocation.
1004 class VSTQPseudo<InstrItinClass itin>
1005 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src), itin, "">;
1006 class VSTQWBPseudo<InstrItinClass itin>
1007 : PseudoNLdSt<(outs GPR:$wb),
1008 (ins addrmode6:$addr, am6offset:$offset, QPR:$src), itin,
1009 "$addr.addr = $wb">;
1010 class VSTQQPseudo<InstrItinClass itin>
1011 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src), itin, "">;
1012 class VSTQQWBPseudo<InstrItinClass itin>
1013 : PseudoNLdSt<(outs GPR:$wb),
1014 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src), itin,
1015 "$addr.addr = $wb">;
1016 class VSTQQQQWBPseudo<InstrItinClass itin>
1017 : PseudoNLdSt<(outs GPR:$wb),
1018 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
1019 "$addr.addr = $wb">;
1021 // VST1 : Vector Store (multiple single elements)
1022 class VST1D<bits<4> op7_4, string Dt>
1023 : NLdSt<0,0b00,0b0111,op7_4, (outs), (ins addrmode6:$Rn, DPR:$Vd),
1024 IIC_VST1, "vst1", Dt, "\\{$Vd\\}, $Rn", "", []> {
1026 let Inst{4} = Rn{4};
1028 class VST1Q<bits<4> op7_4, string Dt>
1029 : NLdSt<0,0b00,0b1010,op7_4, (outs),
1030 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2), IIC_VST1x2,
1031 "vst1", Dt, "\\{$Vd, $src2\\}, $Rn", "", []> {
1033 let Inst{5-4} = Rn{5-4};
1036 def VST1d8 : VST1D<{0,0,0,?}, "8">;
1037 def VST1d16 : VST1D<{0,1,0,?}, "16">;
1038 def VST1d32 : VST1D<{1,0,0,?}, "32">;
1039 def VST1d64 : VST1D<{1,1,0,?}, "64">;
1041 def VST1q8 : VST1Q<{0,0,?,?}, "8">;
1042 def VST1q16 : VST1Q<{0,1,?,?}, "16">;
1043 def VST1q32 : VST1Q<{1,0,?,?}, "32">;
1044 def VST1q64 : VST1Q<{1,1,?,?}, "64">;
1046 def VST1q8Pseudo : VSTQPseudo<IIC_VST1x2>;
1047 def VST1q16Pseudo : VSTQPseudo<IIC_VST1x2>;
1048 def VST1q32Pseudo : VSTQPseudo<IIC_VST1x2>;
1049 def VST1q64Pseudo : VSTQPseudo<IIC_VST1x2>;
1051 // ...with address register writeback:
1052 class VST1DWB<bits<4> op7_4, string Dt>
1053 : NLdSt<0, 0b00, 0b0111, op7_4, (outs GPR:$wb),
1054 (ins addrmode6:$Rn, am6offset:$Rm, DPR:$Vd), IIC_VST1u,
1055 "vst1", Dt, "\\{$Vd\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
1056 let Inst{4} = Rn{4};
1058 class VST1QWB<bits<4> op7_4, string Dt>
1059 : NLdSt<0, 0b00, 0b1010, op7_4, (outs GPR:$wb),
1060 (ins addrmode6:$Rn, am6offset:$Rm, DPR:$Vd, DPR:$src2),
1061 IIC_VST1x2u, "vst1", Dt, "\\{$Vd, $src2\\}, $Rn$Rm",
1062 "$Rn.addr = $wb", []> {
1063 let Inst{5-4} = Rn{5-4};
1066 def VST1d8_UPD : VST1DWB<{0,0,0,?}, "8">;
1067 def VST1d16_UPD : VST1DWB<{0,1,0,?}, "16">;
1068 def VST1d32_UPD : VST1DWB<{1,0,0,?}, "32">;
1069 def VST1d64_UPD : VST1DWB<{1,1,0,?}, "64">;
1071 def VST1q8_UPD : VST1QWB<{0,0,?,?}, "8">;
1072 def VST1q16_UPD : VST1QWB<{0,1,?,?}, "16">;
1073 def VST1q32_UPD : VST1QWB<{1,0,?,?}, "32">;
1074 def VST1q64_UPD : VST1QWB<{1,1,?,?}, "64">;
1076 def VST1q8Pseudo_UPD : VSTQWBPseudo<IIC_VST1x2u>;
1077 def VST1q16Pseudo_UPD : VSTQWBPseudo<IIC_VST1x2u>;
1078 def VST1q32Pseudo_UPD : VSTQWBPseudo<IIC_VST1x2u>;
1079 def VST1q64Pseudo_UPD : VSTQWBPseudo<IIC_VST1x2u>;
1081 // ...with 3 registers (some of these are only for the disassembler):
1082 class VST1D3<bits<4> op7_4, string Dt>
1083 : NLdSt<0, 0b00, 0b0110, op7_4, (outs),
1084 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3),
1085 IIC_VST1x3, "vst1", Dt, "\\{$Vd, $src2, $src3\\}, $Rn", "", []> {
1087 let Inst{4} = Rn{4};
1089 class VST1D3WB<bits<4> op7_4, string Dt>
1090 : NLdSt<0, 0b00, 0b0110, op7_4, (outs GPR:$wb),
1091 (ins addrmode6:$Rn, am6offset:$Rm,
1092 DPR:$Vd, DPR:$src2, DPR:$src3),
1093 IIC_VST1x3u, "vst1", Dt, "\\{$Vd, $src2, $src3\\}, $Rn$Rm",
1094 "$Rn.addr = $wb", []> {
1095 let Inst{4} = Rn{4};
1098 def VST1d8T : VST1D3<{0,0,0,?}, "8">;
1099 def VST1d16T : VST1D3<{0,1,0,?}, "16">;
1100 def VST1d32T : VST1D3<{1,0,0,?}, "32">;
1101 def VST1d64T : VST1D3<{1,1,0,?}, "64">;
1103 def VST1d8T_UPD : VST1D3WB<{0,0,0,?}, "8">;
1104 def VST1d16T_UPD : VST1D3WB<{0,1,0,?}, "16">;
1105 def VST1d32T_UPD : VST1D3WB<{1,0,0,?}, "32">;
1106 def VST1d64T_UPD : VST1D3WB<{1,1,0,?}, "64">;
1108 def VST1d64TPseudo : VSTQQPseudo<IIC_VST1x3>;
1109 def VST1d64TPseudo_UPD : VSTQQWBPseudo<IIC_VST1x3u>;
1111 // ...with 4 registers (some of these are only for the disassembler):
1112 class VST1D4<bits<4> op7_4, string Dt>
1113 : NLdSt<0, 0b00, 0b0010, op7_4, (outs),
1114 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
1115 IIC_VST1x4, "vst1", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn", "",
1118 let Inst{5-4} = Rn{5-4};
1120 class VST1D4WB<bits<4> op7_4, string Dt>
1121 : NLdSt<0, 0b00, 0b0010, op7_4, (outs GPR:$wb),
1122 (ins addrmode6:$Rn, am6offset:$Rm,
1123 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST1x4u,
1124 "vst1", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
1125 "$Rn.addr = $wb", []> {
1126 let Inst{5-4} = Rn{5-4};
1129 def VST1d8Q : VST1D4<{0,0,?,?}, "8">;
1130 def VST1d16Q : VST1D4<{0,1,?,?}, "16">;
1131 def VST1d32Q : VST1D4<{1,0,?,?}, "32">;
1132 def VST1d64Q : VST1D4<{1,1,?,?}, "64">;
1134 def VST1d8Q_UPD : VST1D4WB<{0,0,?,?}, "8">;
1135 def VST1d16Q_UPD : VST1D4WB<{0,1,?,?}, "16">;
1136 def VST1d32Q_UPD : VST1D4WB<{1,0,?,?}, "32">;
1137 def VST1d64Q_UPD : VST1D4WB<{1,1,?,?}, "64">;
1139 def VST1d64QPseudo : VSTQQPseudo<IIC_VST1x4>;
1140 def VST1d64QPseudo_UPD : VSTQQWBPseudo<IIC_VST1x4u>;
1142 // VST2 : Vector Store (multiple 2-element structures)
1143 class VST2D<bits<4> op11_8, bits<4> op7_4, string Dt>
1144 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1145 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2),
1146 IIC_VST2, "vst2", Dt, "\\{$Vd, $src2\\}, $Rn", "", []> {
1148 let Inst{5-4} = Rn{5-4};
1150 class VST2Q<bits<4> op7_4, string Dt>
1151 : NLdSt<0, 0b00, 0b0011, op7_4, (outs),
1152 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
1153 IIC_VST2x2, "vst2", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn",
1156 let Inst{5-4} = Rn{5-4};
1159 def VST2d8 : VST2D<0b1000, {0,0,?,?}, "8">;
1160 def VST2d16 : VST2D<0b1000, {0,1,?,?}, "16">;
1161 def VST2d32 : VST2D<0b1000, {1,0,?,?}, "32">;
1163 def VST2q8 : VST2Q<{0,0,?,?}, "8">;
1164 def VST2q16 : VST2Q<{0,1,?,?}, "16">;
1165 def VST2q32 : VST2Q<{1,0,?,?}, "32">;
1167 def VST2d8Pseudo : VSTQPseudo<IIC_VST2>;
1168 def VST2d16Pseudo : VSTQPseudo<IIC_VST2>;
1169 def VST2d32Pseudo : VSTQPseudo<IIC_VST2>;
1171 def VST2q8Pseudo : VSTQQPseudo<IIC_VST2x2>;
1172 def VST2q16Pseudo : VSTQQPseudo<IIC_VST2x2>;
1173 def VST2q32Pseudo : VSTQQPseudo<IIC_VST2x2>;
1175 // ...with address register writeback:
1176 class VST2DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1177 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1178 (ins addrmode6:$Rn, am6offset:$Rm, DPR:$Vd, DPR:$src2),
1179 IIC_VST2u, "vst2", Dt, "\\{$Vd, $src2\\}, $Rn$Rm",
1180 "$Rn.addr = $wb", []> {
1181 let Inst{5-4} = Rn{5-4};
1183 class VST2QWB<bits<4> op7_4, string Dt>
1184 : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
1185 (ins addrmode6:$Rn, am6offset:$Rm,
1186 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST2x2u,
1187 "vst2", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
1188 "$Rn.addr = $wb", []> {
1189 let Inst{5-4} = Rn{5-4};
1192 def VST2d8_UPD : VST2DWB<0b1000, {0,0,?,?}, "8">;
1193 def VST2d16_UPD : VST2DWB<0b1000, {0,1,?,?}, "16">;
1194 def VST2d32_UPD : VST2DWB<0b1000, {1,0,?,?}, "32">;
1196 def VST2q8_UPD : VST2QWB<{0,0,?,?}, "8">;
1197 def VST2q16_UPD : VST2QWB<{0,1,?,?}, "16">;
1198 def VST2q32_UPD : VST2QWB<{1,0,?,?}, "32">;
1200 def VST2d8Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
1201 def VST2d16Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
1202 def VST2d32Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
1204 def VST2q8Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
1205 def VST2q16Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
1206 def VST2q32Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
1208 // ...with double-spaced registers (for disassembly only):
1209 def VST2b8 : VST2D<0b1001, {0,0,?,?}, "8">;
1210 def VST2b16 : VST2D<0b1001, {0,1,?,?}, "16">;
1211 def VST2b32 : VST2D<0b1001, {1,0,?,?}, "32">;
1212 def VST2b8_UPD : VST2DWB<0b1001, {0,0,?,?}, "8">;
1213 def VST2b16_UPD : VST2DWB<0b1001, {0,1,?,?}, "16">;
1214 def VST2b32_UPD : VST2DWB<0b1001, {1,0,?,?}, "32">;
1216 // VST3 : Vector Store (multiple 3-element structures)
1217 class VST3D<bits<4> op11_8, bits<4> op7_4, string Dt>
1218 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1219 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3,
1220 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn", "", []> {
1222 let Inst{4} = Rn{4};
1225 def VST3d8 : VST3D<0b0100, {0,0,0,?}, "8">;
1226 def VST3d16 : VST3D<0b0100, {0,1,0,?}, "16">;
1227 def VST3d32 : VST3D<0b0100, {1,0,0,?}, "32">;
1229 def VST3d8Pseudo : VSTQQPseudo<IIC_VST3>;
1230 def VST3d16Pseudo : VSTQQPseudo<IIC_VST3>;
1231 def VST3d32Pseudo : VSTQQPseudo<IIC_VST3>;
1233 // ...with address register writeback:
1234 class VST3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1235 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1236 (ins addrmode6:$Rn, am6offset:$Rm,
1237 DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3u,
1238 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn$Rm",
1239 "$Rn.addr = $wb", []> {
1240 let Inst{4} = Rn{4};
1243 def VST3d8_UPD : VST3DWB<0b0100, {0,0,0,?}, "8">;
1244 def VST3d16_UPD : VST3DWB<0b0100, {0,1,0,?}, "16">;
1245 def VST3d32_UPD : VST3DWB<0b0100, {1,0,0,?}, "32">;
1247 def VST3d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1248 def VST3d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1249 def VST3d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1251 // ...with double-spaced registers (non-updating versions for disassembly only):
1252 def VST3q8 : VST3D<0b0101, {0,0,0,?}, "8">;
1253 def VST3q16 : VST3D<0b0101, {0,1,0,?}, "16">;
1254 def VST3q32 : VST3D<0b0101, {1,0,0,?}, "32">;
1255 def VST3q8_UPD : VST3DWB<0b0101, {0,0,0,?}, "8">;
1256 def VST3q16_UPD : VST3DWB<0b0101, {0,1,0,?}, "16">;
1257 def VST3q32_UPD : VST3DWB<0b0101, {1,0,0,?}, "32">;
1259 def VST3q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1260 def VST3q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1261 def VST3q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1263 // ...alternate versions to be allocated odd register numbers:
1264 def VST3q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1265 def VST3q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1266 def VST3q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1268 // VST4 : Vector Store (multiple 4-element structures)
1269 class VST4D<bits<4> op11_8, bits<4> op7_4, string Dt>
1270 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1271 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
1272 IIC_VST4, "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn",
1275 let Inst{5-4} = Rn{5-4};
1278 def VST4d8 : VST4D<0b0000, {0,0,?,?}, "8">;
1279 def VST4d16 : VST4D<0b0000, {0,1,?,?}, "16">;
1280 def VST4d32 : VST4D<0b0000, {1,0,?,?}, "32">;
1282 def VST4d8Pseudo : VSTQQPseudo<IIC_VST4>;
1283 def VST4d16Pseudo : VSTQQPseudo<IIC_VST4>;
1284 def VST4d32Pseudo : VSTQQPseudo<IIC_VST4>;
1286 // ...with address register writeback:
1287 class VST4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1288 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1289 (ins addrmode6:$Rn, am6offset:$Rm,
1290 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST4u,
1291 "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
1292 "$Rn.addr = $wb", []> {
1293 let Inst{5-4} = Rn{5-4};
1296 def VST4d8_UPD : VST4DWB<0b0000, {0,0,?,?}, "8">;
1297 def VST4d16_UPD : VST4DWB<0b0000, {0,1,?,?}, "16">;
1298 def VST4d32_UPD : VST4DWB<0b0000, {1,0,?,?}, "32">;
1300 def VST4d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1301 def VST4d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1302 def VST4d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1304 // ...with double-spaced registers (non-updating versions for disassembly only):
1305 def VST4q8 : VST4D<0b0001, {0,0,?,?}, "8">;
1306 def VST4q16 : VST4D<0b0001, {0,1,?,?}, "16">;
1307 def VST4q32 : VST4D<0b0001, {1,0,?,?}, "32">;
1308 def VST4q8_UPD : VST4DWB<0b0001, {0,0,?,?}, "8">;
1309 def VST4q16_UPD : VST4DWB<0b0001, {0,1,?,?}, "16">;
1310 def VST4q32_UPD : VST4DWB<0b0001, {1,0,?,?}, "32">;
1312 def VST4q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1313 def VST4q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1314 def VST4q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1316 // ...alternate versions to be allocated odd register numbers:
1317 def VST4q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1318 def VST4q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1319 def VST4q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1321 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
1323 // Classes for VST*LN pseudo-instructions with multi-register operands.
1324 // These are expanded to real instructions after register allocation.
1325 class VSTQLNPseudo<InstrItinClass itin>
1326 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
1328 class VSTQLNWBPseudo<InstrItinClass itin>
1329 : PseudoNLdSt<(outs GPR:$wb),
1330 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
1331 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1332 class VSTQQLNPseudo<InstrItinClass itin>
1333 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
1335 class VSTQQLNWBPseudo<InstrItinClass itin>
1336 : PseudoNLdSt<(outs GPR:$wb),
1337 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
1338 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1339 class VSTQQQQLNPseudo<InstrItinClass itin>
1340 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
1342 class VSTQQQQLNWBPseudo<InstrItinClass itin>
1343 : PseudoNLdSt<(outs GPR:$wb),
1344 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
1345 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1347 // VST1LN : Vector Store (single element from one lane)
1348 class VST1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1349 PatFrag StoreOp, SDNode ExtractOp>
1350 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1351 (ins addrmode6:$Rn, DPR:$Vd, nohash_imm:$lane),
1352 IIC_VST1ln, "vst1", Dt, "\\{$Vd[$lane]\\}, $Rn", "",
1353 [(StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane), addrmode6:$Rn)]> {
1356 class VST1QLNPseudo<ValueType Ty, PatFrag StoreOp, SDNode ExtractOp>
1357 : VSTQLNPseudo<IIC_VST1ln> {
1358 let Pattern = [(StoreOp (ExtractOp (Ty QPR:$src), imm:$lane),
1362 def VST1LNd8 : VST1LN<0b0000, {?,?,?,0}, "8", v8i8, truncstorei8,
1364 let Inst{7-5} = lane{2-0};
1366 def VST1LNd16 : VST1LN<0b0100, {?,?,0,?}, "16", v4i16, truncstorei16,
1368 let Inst{7-6} = lane{1-0};
1369 let Inst{4} = Rn{5};
1371 def VST1LNd32 : VST1LN<0b1000, {?,0,?,?}, "32", v2i32, store, extractelt> {
1372 let Inst{7} = lane{0};
1373 let Inst{5-4} = Rn{5-4};
1376 def VST1LNq8Pseudo : VST1QLNPseudo<v16i8, truncstorei8, NEONvgetlaneu>;
1377 def VST1LNq16Pseudo : VST1QLNPseudo<v8i16, truncstorei16, NEONvgetlaneu>;
1378 def VST1LNq32Pseudo : VST1QLNPseudo<v4i32, store, extractelt>;
1380 def : Pat<(store (extractelt (v2f32 DPR:$src), imm:$lane), addrmode6:$addr),
1381 (VST1LNd32 addrmode6:$addr, DPR:$src, imm:$lane)>;
1382 def : Pat<(store (extractelt (v4f32 QPR:$src), imm:$lane), addrmode6:$addr),
1383 (VST1LNq32Pseudo addrmode6:$addr, QPR:$src, imm:$lane)>;
1385 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
1387 // ...with address register writeback:
1388 class VST1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1389 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1390 (ins addrmode6:$Rn, am6offset:$Rm,
1391 DPR:$Vd, nohash_imm:$lane), IIC_VST1lnu, "vst1", Dt,
1392 "\\{$Vd[$lane]\\}, $Rn$Rm",
1393 "$Rn.addr = $wb", []>;
1395 def VST1LNd8_UPD : VST1LNWB<0b0000, {?,?,?,0}, "8"> {
1396 let Inst{7-5} = lane{2-0};
1398 def VST1LNd16_UPD : VST1LNWB<0b0100, {?,?,0,?}, "16"> {
1399 let Inst{7-6} = lane{1-0};
1400 let Inst{4} = Rn{5};
1402 def VST1LNd32_UPD : VST1LNWB<0b1000, {?,0,?,?}, "32"> {
1403 let Inst{7} = lane{0};
1404 let Inst{5-4} = Rn{5-4};
1407 def VST1LNq8Pseudo_UPD : VSTQLNWBPseudo<IIC_VST1lnu>;
1408 def VST1LNq16Pseudo_UPD : VSTQLNWBPseudo<IIC_VST1lnu>;
1409 def VST1LNq32Pseudo_UPD : VSTQLNWBPseudo<IIC_VST1lnu>;
1411 // VST2LN : Vector Store (single 2-element structure from one lane)
1412 class VST2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1413 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1414 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, nohash_imm:$lane),
1415 IIC_VST2ln, "vst2", Dt, "\\{$Vd[$lane], $src2[$lane]\\}, $Rn",
1418 let Inst{4} = Rn{4};
1421 def VST2LNd8 : VST2LN<0b0001, {?,?,?,?}, "8"> {
1422 let Inst{7-5} = lane{2-0};
1424 def VST2LNd16 : VST2LN<0b0101, {?,?,0,?}, "16"> {
1425 let Inst{7-6} = lane{1-0};
1427 def VST2LNd32 : VST2LN<0b1001, {?,0,0,?}, "32"> {
1428 let Inst{7} = lane{0};
1431 def VST2LNd8Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1432 def VST2LNd16Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1433 def VST2LNd32Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1435 // ...with double-spaced registers:
1436 def VST2LNq16 : VST2LN<0b0101, {?,?,1,?}, "16"> {
1437 let Inst{7-6} = lane{1-0};
1438 let Inst{4} = Rn{4};
1440 def VST2LNq32 : VST2LN<0b1001, {?,1,0,?}, "32"> {
1441 let Inst{7} = lane{0};
1442 let Inst{4} = Rn{4};
1445 def VST2LNq16Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
1446 def VST2LNq32Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
1448 // ...with address register writeback:
1449 class VST2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1450 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1451 (ins addrmode6:$addr, am6offset:$offset,
1452 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VST2lnu, "vst2", Dt,
1453 "\\{$src1[$lane], $src2[$lane]\\}, $addr$offset",
1454 "$addr.addr = $wb", []> {
1455 let Inst{4} = Rn{4};
1458 def VST2LNd8_UPD : VST2LNWB<0b0001, {?,?,?,?}, "8"> {
1459 let Inst{7-5} = lane{2-0};
1461 def VST2LNd16_UPD : VST2LNWB<0b0101, {?,?,0,?}, "16"> {
1462 let Inst{7-6} = lane{1-0};
1464 def VST2LNd32_UPD : VST2LNWB<0b1001, {?,0,0,?}, "32"> {
1465 let Inst{7} = lane{0};
1468 def VST2LNd8Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1469 def VST2LNd16Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1470 def VST2LNd32Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1472 def VST2LNq16_UPD : VST2LNWB<0b0101, {?,?,1,?}, "16"> {
1473 let Inst{7-6} = lane{1-0};
1475 def VST2LNq32_UPD : VST2LNWB<0b1001, {?,1,0,?}, "32"> {
1476 let Inst{7} = lane{0};
1479 def VST2LNq16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
1480 def VST2LNq32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
1482 // VST3LN : Vector Store (single 3-element structure from one lane)
1483 class VST3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1484 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1485 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3,
1486 nohash_imm:$lane), IIC_VST3ln, "vst3", Dt,
1487 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn", "", []> {
1491 def VST3LNd8 : VST3LN<0b0010, {?,?,?,0}, "8"> {
1492 let Inst{7-5} = lane{2-0};
1494 def VST3LNd16 : VST3LN<0b0110, {?,?,0,0}, "16"> {
1495 let Inst{7-6} = lane{1-0};
1497 def VST3LNd32 : VST3LN<0b1010, {?,0,0,0}, "32"> {
1498 let Inst{7} = lane{0};
1501 def VST3LNd8Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1502 def VST3LNd16Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1503 def VST3LNd32Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1505 // ...with double-spaced registers:
1506 def VST3LNq16 : VST3LN<0b0110, {?,?,1,0}, "16"> {
1507 let Inst{7-6} = lane{1-0};
1509 def VST3LNq32 : VST3LN<0b1010, {?,1,0,0}, "32"> {
1510 let Inst{7} = lane{0};
1513 def VST3LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
1514 def VST3LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
1516 // ...with address register writeback:
1517 class VST3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1518 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1519 (ins addrmode6:$Rn, am6offset:$Rm,
1520 DPR:$Vd, DPR:$src2, DPR:$src3, nohash_imm:$lane),
1521 IIC_VST3lnu, "vst3", Dt,
1522 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn$Rm",
1523 "$Rn.addr = $wb", []>;
1525 def VST3LNd8_UPD : VST3LNWB<0b0010, {?,?,?,0}, "8"> {
1526 let Inst{7-5} = lane{2-0};
1528 def VST3LNd16_UPD : VST3LNWB<0b0110, {?,?,0,0}, "16"> {
1529 let Inst{7-6} = lane{1-0};
1531 def VST3LNd32_UPD : VST3LNWB<0b1010, {?,0,0,0}, "32"> {
1532 let Inst{7} = lane{0};
1535 def VST3LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1536 def VST3LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1537 def VST3LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1539 def VST3LNq16_UPD : VST3LNWB<0b0110, {?,?,1,0}, "16"> {
1540 let Inst{7-6} = lane{1-0};
1542 def VST3LNq32_UPD : VST3LNWB<0b1010, {?,1,0,0}, "32"> {
1543 let Inst{7} = lane{0};
1546 def VST3LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
1547 def VST3LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
1549 // VST4LN : Vector Store (single 4-element structure from one lane)
1550 class VST4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1551 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1552 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4,
1553 nohash_imm:$lane), IIC_VST4ln, "vst4", Dt,
1554 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn",
1557 let Inst{4} = Rn{4};
1560 def VST4LNd8 : VST4LN<0b0011, {?,?,?,?}, "8"> {
1561 let Inst{7-5} = lane{2-0};
1563 def VST4LNd16 : VST4LN<0b0111, {?,?,0,?}, "16"> {
1564 let Inst{7-6} = lane{1-0};
1566 def VST4LNd32 : VST4LN<0b1011, {?,0,?,?}, "32"> {
1567 let Inst{7} = lane{0};
1568 let Inst{5} = Rn{5};
1571 def VST4LNd8Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1572 def VST4LNd16Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1573 def VST4LNd32Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1575 // ...with double-spaced registers:
1576 def VST4LNq16 : VST4LN<0b0111, {?,?,1,?}, "16"> {
1577 let Inst{7-6} = lane{1-0};
1579 def VST4LNq32 : VST4LN<0b1011, {?,1,?,?}, "32"> {
1580 let Inst{7} = lane{0};
1581 let Inst{5} = Rn{5};
1584 def VST4LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
1585 def VST4LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
1587 // ...with address register writeback:
1588 class VST4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1589 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1590 (ins addrmode6:$Rn, am6offset:$Rm,
1591 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
1592 IIC_VST4lnu, "vst4", Dt,
1593 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn$Rm",
1594 "$Rn.addr = $wb", []> {
1595 let Inst{4} = Rn{4};
1598 def VST4LNd8_UPD : VST4LNWB<0b0011, {?,?,?,?}, "8"> {
1599 let Inst{7-5} = lane{2-0};
1601 def VST4LNd16_UPD : VST4LNWB<0b0111, {?,?,0,?}, "16"> {
1602 let Inst{7-6} = lane{1-0};
1604 def VST4LNd32_UPD : VST4LNWB<0b1011, {?,0,?,?}, "32"> {
1605 let Inst{7} = lane{0};
1606 let Inst{5} = Rn{5};
1609 def VST4LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
1610 def VST4LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
1611 def VST4LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
1613 def VST4LNq16_UPD : VST4LNWB<0b0111, {?,?,1,?}, "16"> {
1614 let Inst{7-6} = lane{1-0};
1616 def VST4LNq32_UPD : VST4LNWB<0b1011, {?,1,?,?}, "32"> {
1617 let Inst{7} = lane{0};
1618 let Inst{5} = Rn{5};
1621 def VST4LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
1622 def VST4LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
1624 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
1627 //===----------------------------------------------------------------------===//
1628 // NEON pattern fragments
1629 //===----------------------------------------------------------------------===//
1631 // Extract D sub-registers of Q registers.
1632 def DSubReg_i8_reg : SDNodeXForm<imm, [{
1633 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1634 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/8, MVT::i32);
1636 def DSubReg_i16_reg : SDNodeXForm<imm, [{
1637 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1638 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/4, MVT::i32);
1640 def DSubReg_i32_reg : SDNodeXForm<imm, [{
1641 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1642 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/2, MVT::i32);
1644 def DSubReg_f64_reg : SDNodeXForm<imm, [{
1645 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1646 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue(), MVT::i32);
1649 // Extract S sub-registers of Q/D registers.
1650 def SSubReg_f32_reg : SDNodeXForm<imm, [{
1651 assert(ARM::ssub_3 == ARM::ssub_0+3 && "Unexpected subreg numbering");
1652 return CurDAG->getTargetConstant(ARM::ssub_0 + N->getZExtValue(), MVT::i32);
1655 // Translate lane numbers from Q registers to D subregs.
1656 def SubReg_i8_lane : SDNodeXForm<imm, [{
1657 return CurDAG->getTargetConstant(N->getZExtValue() & 7, MVT::i32);
1659 def SubReg_i16_lane : SDNodeXForm<imm, [{
1660 return CurDAG->getTargetConstant(N->getZExtValue() & 3, MVT::i32);
1662 def SubReg_i32_lane : SDNodeXForm<imm, [{
1663 return CurDAG->getTargetConstant(N->getZExtValue() & 1, MVT::i32);
1666 //===----------------------------------------------------------------------===//
1667 // Instruction Classes
1668 //===----------------------------------------------------------------------===//
1670 // Basic 2-register operations: double- and quad-register.
1671 class N2VD<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1672 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
1673 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
1674 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
1675 (ins DPR:$Vm), IIC_VUNAD, OpcodeStr, Dt,"$Vd, $Vm", "",
1676 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vm))))]>;
1677 class N2VQ<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1678 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
1679 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
1680 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
1681 (ins QPR:$Vm), IIC_VUNAQ, OpcodeStr, Dt,"$Vd, $Vm", "",
1682 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vm))))]>;
1684 // Basic 2-register intrinsics, both double- and quad-register.
1685 class N2VDInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1686 bits<2> op17_16, bits<5> op11_7, bit op4,
1687 InstrItinClass itin, string OpcodeStr, string Dt,
1688 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1689 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
1690 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
1691 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm))))]>;
1692 class N2VQInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1693 bits<2> op17_16, bits<5> op11_7, bit op4,
1694 InstrItinClass itin, string OpcodeStr, string Dt,
1695 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1696 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
1697 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
1698 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm))))]>;
1700 // Narrow 2-register operations.
1701 class N2VN<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1702 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1703 InstrItinClass itin, string OpcodeStr, string Dt,
1704 ValueType TyD, ValueType TyQ, SDNode OpNode>
1705 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$Vd),
1706 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
1707 [(set DPR:$Vd, (TyD (OpNode (TyQ QPR:$Vm))))]>;
1709 // Narrow 2-register intrinsics.
1710 class N2VNInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1711 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1712 InstrItinClass itin, string OpcodeStr, string Dt,
1713 ValueType TyD, ValueType TyQ, Intrinsic IntOp>
1714 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$Vd),
1715 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
1716 [(set DPR:$Vd, (TyD (IntOp (TyQ QPR:$Vm))))]>;
1718 // Long 2-register operations (currently only used for VMOVL).
1719 class N2VL<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1720 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1721 InstrItinClass itin, string OpcodeStr, string Dt,
1722 ValueType TyQ, ValueType TyD, SDNode OpNode>
1723 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$Vd),
1724 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
1725 [(set QPR:$Vd, (TyQ (OpNode (TyD DPR:$Vm))))]>;
1727 // 2-register shuffles (VTRN/VZIP/VUZP), both double- and quad-register.
1728 class N2VDShuffle<bits<2> op19_18, bits<5> op11_7, string OpcodeStr, string Dt>
1729 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 0, 0, (outs DPR:$Vd, DPR:$Vm),
1730 (ins DPR:$src1, DPR:$src2), IIC_VPERMD,
1731 OpcodeStr, Dt, "$Vd, $Vm",
1732 "$src1 = $Vd, $src2 = $Vm", []>;
1733 class N2VQShuffle<bits<2> op19_18, bits<5> op11_7,
1734 InstrItinClass itin, string OpcodeStr, string Dt>
1735 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 1, 0, (outs QPR:$Vd, QPR:$Vm),
1736 (ins QPR:$src1, QPR:$src2), itin, OpcodeStr, Dt, "$Vd, $Vm",
1737 "$src1 = $Vd, $src2 = $Vm", []>;
1739 // Basic 3-register operations: double- and quad-register.
1740 class N3VD<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1741 InstrItinClass itin, string OpcodeStr, string Dt,
1742 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
1743 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1744 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1745 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
1746 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
1747 let isCommutable = Commutable;
1749 // Same as N3VD but no data type.
1750 class N3VDX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1751 InstrItinClass itin, string OpcodeStr,
1752 ValueType ResTy, ValueType OpTy,
1753 SDNode OpNode, bit Commutable>
1754 : N3VX<op24, op23, op21_20, op11_8, 0, op4,
1755 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1756 OpcodeStr, "$Vd, $Vn, $Vm", "",
1757 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>{
1758 let isCommutable = Commutable;
1761 class N3VDSL<bits<2> op21_20, bits<4> op11_8,
1762 InstrItinClass itin, string OpcodeStr, string Dt,
1763 ValueType Ty, SDNode ShOp>
1764 : N3V<0, 1, op21_20, op11_8, 1, 0,
1765 (outs DPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, nohash_imm:$lane),
1766 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "",
1768 (Ty (ShOp (Ty DPR:$Vn),
1769 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),imm:$lane)))))]> {
1770 let isCommutable = 0;
1772 class N3VDSL16<bits<2> op21_20, bits<4> op11_8,
1773 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1774 : N3V<0, 1, op21_20, op11_8, 1, 0,
1775 (outs DPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, nohash_imm:$lane),
1776 NVMulSLFrm, IIC_VMULi16D, OpcodeStr, Dt,"$Vd, $Vn, $Vm[$lane]","",
1778 (Ty (ShOp (Ty DPR:$Vn),
1779 (Ty (NEONvduplane (Ty DPR_8:$Vm), imm:$lane)))))]> {
1780 let isCommutable = 0;
1783 class N3VQ<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1784 InstrItinClass itin, string OpcodeStr, string Dt,
1785 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
1786 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1787 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
1788 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
1789 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
1790 let isCommutable = Commutable;
1792 class N3VQX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1793 InstrItinClass itin, string OpcodeStr,
1794 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
1795 : N3VX<op24, op23, op21_20, op11_8, 1, op4,
1796 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
1797 OpcodeStr, "$Vd, $Vn, $Vm", "",
1798 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]>{
1799 let isCommutable = Commutable;
1801 class N3VQSL<bits<2> op21_20, bits<4> op11_8,
1802 InstrItinClass itin, string OpcodeStr, string Dt,
1803 ValueType ResTy, ValueType OpTy, SDNode ShOp>
1804 : N3V<1, 1, op21_20, op11_8, 1, 0,
1805 (outs QPR:$Vd), (ins QPR:$Vn, DPR_VFP2:$Vm, nohash_imm:$lane),
1806 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "",
1807 [(set (ResTy QPR:$Vd),
1808 (ResTy (ShOp (ResTy QPR:$Vn),
1809 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
1811 let isCommutable = 0;
1813 class N3VQSL16<bits<2> op21_20, bits<4> op11_8, string OpcodeStr, string Dt,
1814 ValueType ResTy, ValueType OpTy, SDNode ShOp>
1815 : N3V<1, 1, op21_20, op11_8, 1, 0,
1816 (outs QPR:$Vd), (ins QPR:$Vn, DPR_8:$Vm, nohash_imm:$lane),
1817 NVMulSLFrm, IIC_VMULi16Q, OpcodeStr, Dt,"$Vd, $Vn, $Vm[$lane]","",
1818 [(set (ResTy QPR:$Vd),
1819 (ResTy (ShOp (ResTy QPR:$Vn),
1820 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
1822 let isCommutable = 0;
1825 // Basic 3-register intrinsics, both double- and quad-register.
1826 class N3VDInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1827 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1828 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
1829 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1830 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), f, itin,
1831 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
1832 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
1833 let isCommutable = Commutable;
1835 class N3VDIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1836 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
1837 : N3V<0, 1, op21_20, op11_8, 1, 0,
1838 (outs DPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, nohash_imm:$lane),
1839 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "",
1841 (Ty (IntOp (Ty DPR:$Vn),
1842 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),
1844 let isCommutable = 0;
1846 class N3VDIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1847 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
1848 : N3V<0, 1, op21_20, op11_8, 1, 0,
1849 (outs DPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, nohash_imm:$lane),
1850 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "",
1852 (Ty (IntOp (Ty DPR:$Vn),
1853 (Ty (NEONvduplane (Ty DPR_8:$Vm), imm:$lane)))))]> {
1854 let isCommutable = 0;
1856 class N3VDIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1857 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1858 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1859 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1860 (outs DPR:$Vd), (ins DPR:$Vm, DPR:$Vn), f, itin,
1861 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
1862 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (OpTy DPR:$Vn))))]> {
1863 let isCommutable = 0;
1866 class N3VQInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1867 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1868 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
1869 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1870 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), f, itin,
1871 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
1872 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
1873 let isCommutable = Commutable;
1875 class N3VQIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1876 string OpcodeStr, string Dt,
1877 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1878 : N3V<1, 1, op21_20, op11_8, 1, 0,
1879 (outs QPR:$Vd), (ins QPR:$Vn, DPR_VFP2:$Vm, nohash_imm:$lane),
1880 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "",
1881 [(set (ResTy QPR:$Vd),
1882 (ResTy (IntOp (ResTy QPR:$Vn),
1883 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
1885 let isCommutable = 0;
1887 class N3VQIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1888 string OpcodeStr, string Dt,
1889 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1890 : N3V<1, 1, op21_20, op11_8, 1, 0,
1891 (outs QPR:$Vd), (ins QPR:$Vn, DPR_8:$Vm, nohash_imm:$lane),
1892 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "",
1893 [(set (ResTy QPR:$Vd),
1894 (ResTy (IntOp (ResTy QPR:$Vn),
1895 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
1897 let isCommutable = 0;
1899 class N3VQIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1900 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1901 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1902 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1903 (outs QPR:$Vd), (ins QPR:$Vm, QPR:$Vn), f, itin,
1904 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
1905 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (OpTy QPR:$Vn))))]> {
1906 let isCommutable = 0;
1909 // Multiply-Add/Sub operations: double- and quad-register.
1910 class N3VDMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1911 InstrItinClass itin, string OpcodeStr, string Dt,
1912 ValueType Ty, SDPatternOperator MulOp, SDPatternOperator OpNode>
1913 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1914 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1915 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1916 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
1917 (Ty (MulOp DPR:$Vn, DPR:$Vm)))))]>;
1919 class N3VDMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1920 string OpcodeStr, string Dt,
1921 ValueType Ty, SDPatternOperator MulOp, SDPatternOperator ShOp>
1922 : N3V<0, 1, op21_20, op11_8, 1, 0,
1924 (ins DPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, nohash_imm:$lane),
1926 OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "$src1 = $Vd",
1928 (Ty (ShOp (Ty DPR:$src1),
1930 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),
1932 class N3VDMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1933 string OpcodeStr, string Dt,
1934 ValueType Ty, SDNode MulOp, SDNode ShOp>
1935 : N3V<0, 1, op21_20, op11_8, 1, 0,
1937 (ins DPR:$src1, DPR:$Vn, DPR_8:$Vm, nohash_imm:$lane),
1939 OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "$src1 = $Vd",
1941 (Ty (ShOp (Ty DPR:$src1),
1943 (Ty (NEONvduplane (Ty DPR_8:$Vm),
1946 class N3VQMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1947 InstrItinClass itin, string OpcodeStr, string Dt, ValueType Ty,
1948 SDPatternOperator MulOp, SDPatternOperator OpNode>
1949 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1950 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
1951 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1952 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
1953 (Ty (MulOp QPR:$Vn, QPR:$Vm)))))]>;
1954 class N3VQMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1955 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1956 SDPatternOperator MulOp, SDPatternOperator ShOp>
1957 : N3V<1, 1, op21_20, op11_8, 1, 0,
1959 (ins QPR:$src1, QPR:$Vn, DPR_VFP2:$Vm, nohash_imm:$lane),
1961 OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "$src1 = $Vd",
1962 [(set (ResTy QPR:$Vd),
1963 (ResTy (ShOp (ResTy QPR:$src1),
1964 (ResTy (MulOp QPR:$Vn,
1965 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
1967 class N3VQMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1968 string OpcodeStr, string Dt,
1969 ValueType ResTy, ValueType OpTy,
1970 SDNode MulOp, SDNode ShOp>
1971 : N3V<1, 1, op21_20, op11_8, 1, 0,
1973 (ins QPR:$src1, QPR:$Vn, DPR_8:$Vm, nohash_imm:$lane),
1975 OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "$src1 = $Vd",
1976 [(set (ResTy QPR:$Vd),
1977 (ResTy (ShOp (ResTy QPR:$src1),
1978 (ResTy (MulOp QPR:$Vn,
1979 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
1982 // Neon Intrinsic-Op instructions (VABA): double- and quad-register.
1983 class N3VDIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1984 InstrItinClass itin, string OpcodeStr, string Dt,
1985 ValueType Ty, Intrinsic IntOp, SDNode OpNode>
1986 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1987 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1988 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1989 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
1990 (Ty (IntOp (Ty DPR:$Vn), (Ty DPR:$Vm))))))]>;
1991 class N3VQIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1992 InstrItinClass itin, string OpcodeStr, string Dt,
1993 ValueType Ty, Intrinsic IntOp, SDNode OpNode>
1994 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1995 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
1996 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1997 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
1998 (Ty (IntOp (Ty QPR:$Vn), (Ty QPR:$Vm))))))]>;
2000 // Neon 3-argument intrinsics, both double- and quad-register.
2001 // The destination register is also used as the first source operand register.
2002 class N3VDInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2003 InstrItinClass itin, string OpcodeStr, string Dt,
2004 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2005 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2006 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2007 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2008 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$src1),
2009 (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>;
2010 class N3VQInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2011 InstrItinClass itin, string OpcodeStr, string Dt,
2012 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2013 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2014 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2015 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2016 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$src1),
2017 (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]>;
2019 // Long Multiply-Add/Sub operations.
2020 class N3VLMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2021 InstrItinClass itin, string OpcodeStr, string Dt,
2022 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2023 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2024 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2025 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2026 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
2027 (TyQ (MulOp (TyD DPR:$Vn),
2028 (TyD DPR:$Vm)))))]>;
2029 class N3VLMulOpSL<bit op24, bits<2> op21_20, bits<4> op11_8,
2030 InstrItinClass itin, string OpcodeStr, string Dt,
2031 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2032 : N3V<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$Vd),
2033 (ins QPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, nohash_imm:$lane),
2035 OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "$src1 = $Vd",
2037 (OpNode (TyQ QPR:$src1),
2038 (TyQ (MulOp (TyD DPR:$Vn),
2039 (TyD (NEONvduplane (TyD DPR_VFP2:$Vm),
2041 class N3VLMulOpSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2042 InstrItinClass itin, string OpcodeStr, string Dt,
2043 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2044 : N3V<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$Vd),
2045 (ins QPR:$src1, DPR:$Vn, DPR_8:$Vm, nohash_imm:$lane),
2047 OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "$src1 = $Vd",
2049 (OpNode (TyQ QPR:$src1),
2050 (TyQ (MulOp (TyD DPR:$Vn),
2051 (TyD (NEONvduplane (TyD DPR_8:$Vm),
2054 // Long Intrinsic-Op vector operations with explicit extend (VABAL).
2055 class N3VLIntExtOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2056 InstrItinClass itin, string OpcodeStr, string Dt,
2057 ValueType TyQ, ValueType TyD, Intrinsic IntOp, SDNode ExtOp,
2059 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2060 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2061 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2062 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
2063 (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
2064 (TyD DPR:$Vm)))))))]>;
2066 // Neon Long 3-argument intrinsic. The destination register is
2067 // a quad-register and is also used as the first source operand register.
2068 class N3VLInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2069 InstrItinClass itin, string OpcodeStr, string Dt,
2070 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
2071 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2072 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2073 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2075 (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$Vn), (TyD DPR:$Vm))))]>;
2076 class N3VLInt3SL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2077 string OpcodeStr, string Dt,
2078 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2079 : N3V<op24, 1, op21_20, op11_8, 1, 0,
2081 (ins QPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, nohash_imm:$lane),
2083 OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "$src1 = $Vd",
2084 [(set (ResTy QPR:$Vd),
2085 (ResTy (IntOp (ResTy QPR:$src1),
2087 (OpTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2089 class N3VLInt3SL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2090 InstrItinClass itin, string OpcodeStr, string Dt,
2091 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2092 : N3V<op24, 1, op21_20, op11_8, 1, 0,
2094 (ins QPR:$src1, DPR:$Vn, DPR_8:$Vm, nohash_imm:$lane),
2096 OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "$src1 = $Vd",
2097 [(set (ResTy QPR:$Vd),
2098 (ResTy (IntOp (ResTy QPR:$src1),
2100 (OpTy (NEONvduplane (OpTy DPR_8:$Vm),
2103 // Narrowing 3-register intrinsics.
2104 class N3VNInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2105 string OpcodeStr, string Dt, ValueType TyD, ValueType TyQ,
2106 Intrinsic IntOp, bit Commutable>
2107 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2108 (outs DPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINi4D,
2109 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2110 [(set DPR:$Vd, (TyD (IntOp (TyQ QPR:$Vn), (TyQ QPR:$Vm))))]> {
2111 let isCommutable = Commutable;
2114 // Long 3-register operations.
2115 class N3VL<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2116 InstrItinClass itin, string OpcodeStr, string Dt,
2117 ValueType TyQ, ValueType TyD, SDNode OpNode, bit Commutable>
2118 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2119 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2120 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2121 [(set QPR:$Vd, (TyQ (OpNode (TyD DPR:$Vn), (TyD DPR:$Vm))))]> {
2122 let isCommutable = Commutable;
2124 class N3VLSL<bit op24, bits<2> op21_20, bits<4> op11_8,
2125 InstrItinClass itin, string OpcodeStr, string Dt,
2126 ValueType TyQ, ValueType TyD, SDNode OpNode>
2127 : N3V<op24, 1, op21_20, op11_8, 1, 0,
2128 (outs QPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, nohash_imm:$lane),
2129 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "",
2131 (TyQ (OpNode (TyD DPR:$Vn),
2132 (TyD (NEONvduplane (TyD DPR_VFP2:$Vm),imm:$lane)))))]>;
2133 class N3VLSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2134 InstrItinClass itin, string OpcodeStr, string Dt,
2135 ValueType TyQ, ValueType TyD, SDNode OpNode>
2136 : N3V<op24, 1, op21_20, op11_8, 1, 0,
2137 (outs QPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, nohash_imm:$lane),
2138 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "",
2140 (TyQ (OpNode (TyD DPR:$Vn),
2141 (TyD (NEONvduplane (TyD DPR_8:$Vm), imm:$lane)))))]>;
2143 // Long 3-register operations with explicitly extended operands.
2144 class N3VLExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2145 InstrItinClass itin, string OpcodeStr, string Dt,
2146 ValueType TyQ, ValueType TyD, SDNode OpNode, SDNode ExtOp,
2148 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2149 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2150 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2151 [(set QPR:$Vd, (OpNode (TyQ (ExtOp (TyD DPR:$Vn))),
2152 (TyQ (ExtOp (TyD DPR:$Vm)))))]> {
2153 let isCommutable = Commutable;
2156 // Long 3-register intrinsics with explicit extend (VABDL).
2157 class N3VLIntExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2158 InstrItinClass itin, string OpcodeStr, string Dt,
2159 ValueType TyQ, ValueType TyD, Intrinsic IntOp, SDNode ExtOp,
2161 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2162 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2163 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2164 [(set QPR:$Vd, (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
2165 (TyD DPR:$Vm))))))]> {
2166 let isCommutable = Commutable;
2169 // Long 3-register intrinsics.
2170 class N3VLInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2171 InstrItinClass itin, string OpcodeStr, string Dt,
2172 ValueType TyQ, ValueType TyD, Intrinsic IntOp, bit Commutable>
2173 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2174 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2175 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2176 [(set QPR:$Vd, (TyQ (IntOp (TyD DPR:$Vn), (TyD DPR:$Vm))))]> {
2177 let isCommutable = Commutable;
2179 class N3VLIntSL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2180 string OpcodeStr, string Dt,
2181 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2182 : N3V<op24, 1, op21_20, op11_8, 1, 0,
2183 (outs QPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, nohash_imm:$lane),
2184 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "",
2185 [(set (ResTy QPR:$Vd),
2186 (ResTy (IntOp (OpTy DPR:$Vn),
2187 (OpTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2189 class N3VLIntSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2190 InstrItinClass itin, string OpcodeStr, string Dt,
2191 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2192 : N3V<op24, 1, op21_20, op11_8, 1, 0,
2193 (outs QPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, nohash_imm:$lane),
2194 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "",
2195 [(set (ResTy QPR:$Vd),
2196 (ResTy (IntOp (OpTy DPR:$Vn),
2197 (OpTy (NEONvduplane (OpTy DPR_8:$Vm),
2200 // Wide 3-register operations.
2201 class N3VW<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2202 string OpcodeStr, string Dt, ValueType TyQ, ValueType TyD,
2203 SDNode OpNode, SDNode ExtOp, bit Commutable>
2204 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2205 (outs QPR:$Vd), (ins QPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VSUBiD,
2206 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2207 [(set QPR:$Vd, (OpNode (TyQ QPR:$Vn),
2208 (TyQ (ExtOp (TyD DPR:$Vm)))))]> {
2209 let isCommutable = Commutable;
2212 // Pairwise long 2-register intrinsics, both double- and quad-register.
2213 class N2VDPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2214 bits<2> op17_16, bits<5> op11_7, bit op4,
2215 string OpcodeStr, string Dt,
2216 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2217 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
2218 (ins DPR:$Vm), IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
2219 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm))))]>;
2220 class N2VQPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2221 bits<2> op17_16, bits<5> op11_7, bit op4,
2222 string OpcodeStr, string Dt,
2223 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2224 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
2225 (ins QPR:$Vm), IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
2226 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm))))]>;
2228 // Pairwise long 2-register accumulate intrinsics,
2229 // both double- and quad-register.
2230 // The destination register is also used as the first source operand register.
2231 class N2VDPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2232 bits<2> op17_16, bits<5> op11_7, bit op4,
2233 string OpcodeStr, string Dt,
2234 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2235 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
2236 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vm), IIC_VPALiD,
2237 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
2238 [(set DPR:$Vd, (ResTy (IntOp (ResTy DPR:$src1), (OpTy DPR:$Vm))))]>;
2239 class N2VQPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2240 bits<2> op17_16, bits<5> op11_7, bit op4,
2241 string OpcodeStr, string Dt,
2242 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2243 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4,
2244 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vm), IIC_VPALiQ,
2245 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
2246 [(set QPR:$Vd, (ResTy (IntOp (ResTy QPR:$src1), (OpTy QPR:$Vm))))]>;
2248 // Shift by immediate,
2249 // both double- and quad-register.
2250 class N2VDSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2251 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2252 ValueType Ty, SDNode OpNode>
2253 : N2VImm<op24, op23, op11_8, op7, 0, op4,
2254 (outs DPR:$Vd), (ins DPR:$Vm, i32imm:$SIMM), f, itin,
2255 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2256 [(set DPR:$Vd, (Ty (OpNode (Ty DPR:$Vm), (i32 imm:$SIMM))))]>;
2257 class N2VQSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2258 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2259 ValueType Ty, SDNode OpNode>
2260 : N2VImm<op24, op23, op11_8, op7, 1, op4,
2261 (outs QPR:$Vd), (ins QPR:$Vm, i32imm:$SIMM), f, itin,
2262 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2263 [(set QPR:$Vd, (Ty (OpNode (Ty QPR:$Vm), (i32 imm:$SIMM))))]>;
2265 // Long shift by immediate.
2266 class N2VLSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
2267 string OpcodeStr, string Dt,
2268 ValueType ResTy, ValueType OpTy, SDNode OpNode>
2269 : N2VImm<op24, op23, op11_8, op7, op6, op4,
2270 (outs QPR:$Vd), (ins DPR:$Vm, i32imm:$SIMM), N2RegVShLFrm,
2271 IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2272 [(set QPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vm),
2273 (i32 imm:$SIMM))))]>;
2275 // Narrow shift by immediate.
2276 class N2VNSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
2277 InstrItinClass itin, string OpcodeStr, string Dt,
2278 ValueType ResTy, ValueType OpTy, SDNode OpNode>
2279 : N2VImm<op24, op23, op11_8, op7, op6, op4,
2280 (outs DPR:$Vd), (ins QPR:$Vm, i32imm:$SIMM), N2RegVShRFrm, itin,
2281 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2282 [(set DPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vm),
2283 (i32 imm:$SIMM))))]>;
2285 // Shift right by immediate and accumulate,
2286 // both double- and quad-register.
2287 class N2VDShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2288 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
2289 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
2290 (ins DPR:$src1, DPR:$Vm, i32imm:$SIMM), N2RegVShRFrm, IIC_VPALiD,
2291 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2292 [(set DPR:$Vd, (Ty (add DPR:$src1,
2293 (Ty (ShOp DPR:$Vm, (i32 imm:$SIMM))))))]>;
2294 class N2VQShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2295 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
2296 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
2297 (ins QPR:$src1, QPR:$Vm, i32imm:$SIMM), N2RegVShRFrm, IIC_VPALiD,
2298 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2299 [(set QPR:$Vd, (Ty (add QPR:$src1,
2300 (Ty (ShOp QPR:$Vm, (i32 imm:$SIMM))))))]>;
2302 // Shift by immediate and insert,
2303 // both double- and quad-register.
2304 class N2VDShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2305 Format f, string OpcodeStr, string Dt, ValueType Ty,SDNode ShOp>
2306 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
2307 (ins DPR:$src1, DPR:$Vm, i32imm:$SIMM), f, IIC_VSHLiD,
2308 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2309 [(set DPR:$Vd, (Ty (ShOp DPR:$src1, DPR:$Vm, (i32 imm:$SIMM))))]>;
2310 class N2VQShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2311 Format f, string OpcodeStr, string Dt, ValueType Ty,SDNode ShOp>
2312 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
2313 (ins QPR:$src1, QPR:$Vm, i32imm:$SIMM), f, IIC_VSHLiQ,
2314 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2315 [(set QPR:$Vd, (Ty (ShOp QPR:$src1, QPR:$Vm, (i32 imm:$SIMM))))]>;
2317 // Convert, with fractional bits immediate,
2318 // both double- and quad-register.
2319 class N2VCvtD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2320 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2322 : N2VImm<op24, op23, op11_8, op7, 0, op4,
2323 (outs DPR:$Vd), (ins DPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
2324 IIC_VUNAD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2325 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (i32 imm:$SIMM))))]>;
2326 class N2VCvtQ<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2327 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2329 : N2VImm<op24, op23, op11_8, op7, 1, op4,
2330 (outs QPR:$Vd), (ins QPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
2331 IIC_VUNAQ, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2332 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (i32 imm:$SIMM))))]>;
2334 //===----------------------------------------------------------------------===//
2336 //===----------------------------------------------------------------------===//
2338 // Abbreviations used in multiclass suffixes:
2339 // Q = quarter int (8 bit) elements
2340 // H = half int (16 bit) elements
2341 // S = single int (32 bit) elements
2342 // D = double int (64 bit) elements
2344 // Neon 2-register vector operations -- for disassembly only.
2346 // First with only element sizes of 8, 16 and 32 bits:
2347 multiclass N2V_QHS_cmp<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2348 bits<5> op11_7, bit op4, string opc, string Dt,
2349 string asm, SDNode OpNode> {
2350 // 64-bit vector types.
2351 def v8i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 0, op4,
2352 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2353 opc, !strconcat(Dt, "8"), asm, "",
2354 [(set DPR:$Vd, (v8i8 (OpNode (v8i8 DPR:$Vm))))]>;
2355 def v4i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 0, op4,
2356 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2357 opc, !strconcat(Dt, "16"), asm, "",
2358 [(set DPR:$Vd, (v4i16 (OpNode (v4i16 DPR:$Vm))))]>;
2359 def v2i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
2360 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2361 opc, !strconcat(Dt, "32"), asm, "",
2362 [(set DPR:$Vd, (v2i32 (OpNode (v2i32 DPR:$Vm))))]>;
2363 def v2f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
2364 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2365 opc, "f32", asm, "",
2366 [(set DPR:$Vd, (v2f32 (OpNode (v2f32 DPR:$Vm))))]> {
2367 let Inst{10} = 1; // overwrite F = 1
2370 // 128-bit vector types.
2371 def v16i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 1, op4,
2372 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2373 opc, !strconcat(Dt, "8"), asm, "",
2374 [(set QPR:$Vd, (v16i8 (OpNode (v16i8 QPR:$Vm))))]>;
2375 def v8i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 1, op4,
2376 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2377 opc, !strconcat(Dt, "16"), asm, "",
2378 [(set QPR:$Vd, (v8i16 (OpNode (v8i16 QPR:$Vm))))]>;
2379 def v4i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
2380 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2381 opc, !strconcat(Dt, "32"), asm, "",
2382 [(set QPR:$Vd, (v4i32 (OpNode (v4i32 QPR:$Vm))))]>;
2383 def v4f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
2384 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2385 opc, "f32", asm, "",
2386 [(set QPR:$Vd, (v4f32 (OpNode (v4f32 QPR:$Vm))))]> {
2387 let Inst{10} = 1; // overwrite F = 1
2391 // Neon 3-register vector operations.
2393 // First with only element sizes of 8, 16 and 32 bits:
2394 multiclass N3V_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2395 InstrItinClass itinD16, InstrItinClass itinD32,
2396 InstrItinClass itinQ16, InstrItinClass itinQ32,
2397 string OpcodeStr, string Dt,
2398 SDNode OpNode, bit Commutable = 0> {
2399 // 64-bit vector types.
2400 def v8i8 : N3VD<op24, op23, 0b00, op11_8, op4, itinD16,
2401 OpcodeStr, !strconcat(Dt, "8"),
2402 v8i8, v8i8, OpNode, Commutable>;
2403 def v4i16 : N3VD<op24, op23, 0b01, op11_8, op4, itinD16,
2404 OpcodeStr, !strconcat(Dt, "16"),
2405 v4i16, v4i16, OpNode, Commutable>;
2406 def v2i32 : N3VD<op24, op23, 0b10, op11_8, op4, itinD32,
2407 OpcodeStr, !strconcat(Dt, "32"),
2408 v2i32, v2i32, OpNode, Commutable>;
2410 // 128-bit vector types.
2411 def v16i8 : N3VQ<op24, op23, 0b00, op11_8, op4, itinQ16,
2412 OpcodeStr, !strconcat(Dt, "8"),
2413 v16i8, v16i8, OpNode, Commutable>;
2414 def v8i16 : N3VQ<op24, op23, 0b01, op11_8, op4, itinQ16,
2415 OpcodeStr, !strconcat(Dt, "16"),
2416 v8i16, v8i16, OpNode, Commutable>;
2417 def v4i32 : N3VQ<op24, op23, 0b10, op11_8, op4, itinQ32,
2418 OpcodeStr, !strconcat(Dt, "32"),
2419 v4i32, v4i32, OpNode, Commutable>;
2422 multiclass N3VSL_HS<bits<4> op11_8, string OpcodeStr, string Dt, SDNode ShOp> {
2423 def v4i16 : N3VDSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
2425 def v2i32 : N3VDSL<0b10, op11_8, IIC_VMULi32D, OpcodeStr, !strconcat(Dt,"32"),
2427 def v8i16 : N3VQSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
2428 v8i16, v4i16, ShOp>;
2429 def v4i32 : N3VQSL<0b10, op11_8, IIC_VMULi32Q, OpcodeStr, !strconcat(Dt,"32"),
2430 v4i32, v2i32, ShOp>;
2433 // ....then also with element size 64 bits:
2434 multiclass N3V_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2435 InstrItinClass itinD, InstrItinClass itinQ,
2436 string OpcodeStr, string Dt,
2437 SDNode OpNode, bit Commutable = 0>
2438 : N3V_QHS<op24, op23, op11_8, op4, itinD, itinD, itinQ, itinQ,
2439 OpcodeStr, Dt, OpNode, Commutable> {
2440 def v1i64 : N3VD<op24, op23, 0b11, op11_8, op4, itinD,
2441 OpcodeStr, !strconcat(Dt, "64"),
2442 v1i64, v1i64, OpNode, Commutable>;
2443 def v2i64 : N3VQ<op24, op23, 0b11, op11_8, op4, itinQ,
2444 OpcodeStr, !strconcat(Dt, "64"),
2445 v2i64, v2i64, OpNode, Commutable>;
2449 // Neon Narrowing 2-register vector operations,
2450 // source operand element sizes of 16, 32 and 64 bits:
2451 multiclass N2VN_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2452 bits<5> op11_7, bit op6, bit op4,
2453 InstrItinClass itin, string OpcodeStr, string Dt,
2455 def v8i8 : N2VN<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
2456 itin, OpcodeStr, !strconcat(Dt, "16"),
2457 v8i8, v8i16, OpNode>;
2458 def v4i16 : N2VN<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
2459 itin, OpcodeStr, !strconcat(Dt, "32"),
2460 v4i16, v4i32, OpNode>;
2461 def v2i32 : N2VN<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
2462 itin, OpcodeStr, !strconcat(Dt, "64"),
2463 v2i32, v2i64, OpNode>;
2466 // Neon Narrowing 2-register vector intrinsics,
2467 // source operand element sizes of 16, 32 and 64 bits:
2468 multiclass N2VNInt_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2469 bits<5> op11_7, bit op6, bit op4,
2470 InstrItinClass itin, string OpcodeStr, string Dt,
2472 def v8i8 : N2VNInt<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
2473 itin, OpcodeStr, !strconcat(Dt, "16"),
2474 v8i8, v8i16, IntOp>;
2475 def v4i16 : N2VNInt<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
2476 itin, OpcodeStr, !strconcat(Dt, "32"),
2477 v4i16, v4i32, IntOp>;
2478 def v2i32 : N2VNInt<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
2479 itin, OpcodeStr, !strconcat(Dt, "64"),
2480 v2i32, v2i64, IntOp>;
2484 // Neon Lengthening 2-register vector intrinsic (currently specific to VMOVL).
2485 // source operand element sizes of 16, 32 and 64 bits:
2486 multiclass N2VL_QHS<bits<2> op24_23, bits<5> op11_7, bit op6, bit op4,
2487 string OpcodeStr, string Dt, SDNode OpNode> {
2488 def v8i16 : N2VL<op24_23, 0b00, 0b10, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2489 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode>;
2490 def v4i32 : N2VL<op24_23, 0b01, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2491 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
2492 def v2i64 : N2VL<op24_23, 0b10, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2493 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
2497 // Neon 3-register vector intrinsics.
2499 // First with only element sizes of 16 and 32 bits:
2500 multiclass N3VInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2501 InstrItinClass itinD16, InstrItinClass itinD32,
2502 InstrItinClass itinQ16, InstrItinClass itinQ32,
2503 string OpcodeStr, string Dt,
2504 Intrinsic IntOp, bit Commutable = 0> {
2505 // 64-bit vector types.
2506 def v4i16 : N3VDInt<op24, op23, 0b01, op11_8, op4, f, itinD16,
2507 OpcodeStr, !strconcat(Dt, "16"),
2508 v4i16, v4i16, IntOp, Commutable>;
2509 def v2i32 : N3VDInt<op24, op23, 0b10, op11_8, op4, f, itinD32,
2510 OpcodeStr, !strconcat(Dt, "32"),
2511 v2i32, v2i32, IntOp, Commutable>;
2513 // 128-bit vector types.
2514 def v8i16 : N3VQInt<op24, op23, 0b01, op11_8, op4, f, itinQ16,
2515 OpcodeStr, !strconcat(Dt, "16"),
2516 v8i16, v8i16, IntOp, Commutable>;
2517 def v4i32 : N3VQInt<op24, op23, 0b10, op11_8, op4, f, itinQ32,
2518 OpcodeStr, !strconcat(Dt, "32"),
2519 v4i32, v4i32, IntOp, Commutable>;
2521 multiclass N3VInt_HSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2522 InstrItinClass itinD16, InstrItinClass itinD32,
2523 InstrItinClass itinQ16, InstrItinClass itinQ32,
2524 string OpcodeStr, string Dt,
2526 // 64-bit vector types.
2527 def v4i16 : N3VDIntSh<op24, op23, 0b01, op11_8, op4, f, itinD16,
2528 OpcodeStr, !strconcat(Dt, "16"),
2529 v4i16, v4i16, IntOp>;
2530 def v2i32 : N3VDIntSh<op24, op23, 0b10, op11_8, op4, f, itinD32,
2531 OpcodeStr, !strconcat(Dt, "32"),
2532 v2i32, v2i32, IntOp>;
2534 // 128-bit vector types.
2535 def v8i16 : N3VQIntSh<op24, op23, 0b01, op11_8, op4, f, itinQ16,
2536 OpcodeStr, !strconcat(Dt, "16"),
2537 v8i16, v8i16, IntOp>;
2538 def v4i32 : N3VQIntSh<op24, op23, 0b10, op11_8, op4, f, itinQ32,
2539 OpcodeStr, !strconcat(Dt, "32"),
2540 v4i32, v4i32, IntOp>;
2543 multiclass N3VIntSL_HS<bits<4> op11_8,
2544 InstrItinClass itinD16, InstrItinClass itinD32,
2545 InstrItinClass itinQ16, InstrItinClass itinQ32,
2546 string OpcodeStr, string Dt, Intrinsic IntOp> {
2547 def v4i16 : N3VDIntSL16<0b01, op11_8, itinD16,
2548 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp>;
2549 def v2i32 : N3VDIntSL<0b10, op11_8, itinD32,
2550 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp>;
2551 def v8i16 : N3VQIntSL16<0b01, op11_8, itinQ16,
2552 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16, IntOp>;
2553 def v4i32 : N3VQIntSL<0b10, op11_8, itinQ32,
2554 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32, IntOp>;
2557 // ....then also with element size of 8 bits:
2558 multiclass N3VInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2559 InstrItinClass itinD16, InstrItinClass itinD32,
2560 InstrItinClass itinQ16, InstrItinClass itinQ32,
2561 string OpcodeStr, string Dt,
2562 Intrinsic IntOp, bit Commutable = 0>
2563 : N3VInt_HS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2564 OpcodeStr, Dt, IntOp, Commutable> {
2565 def v8i8 : N3VDInt<op24, op23, 0b00, op11_8, op4, f, itinD16,
2566 OpcodeStr, !strconcat(Dt, "8"),
2567 v8i8, v8i8, IntOp, Commutable>;
2568 def v16i8 : N3VQInt<op24, op23, 0b00, op11_8, op4, f, itinQ16,
2569 OpcodeStr, !strconcat(Dt, "8"),
2570 v16i8, v16i8, IntOp, Commutable>;
2572 multiclass N3VInt_QHSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2573 InstrItinClass itinD16, InstrItinClass itinD32,
2574 InstrItinClass itinQ16, InstrItinClass itinQ32,
2575 string OpcodeStr, string Dt,
2577 : N3VInt_HSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2578 OpcodeStr, Dt, IntOp> {
2579 def v8i8 : N3VDIntSh<op24, op23, 0b00, op11_8, op4, f, itinD16,
2580 OpcodeStr, !strconcat(Dt, "8"),
2582 def v16i8 : N3VQIntSh<op24, op23, 0b00, op11_8, op4, f, itinQ16,
2583 OpcodeStr, !strconcat(Dt, "8"),
2584 v16i8, v16i8, IntOp>;
2588 // ....then also with element size of 64 bits:
2589 multiclass N3VInt_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2590 InstrItinClass itinD16, InstrItinClass itinD32,
2591 InstrItinClass itinQ16, InstrItinClass itinQ32,
2592 string OpcodeStr, string Dt,
2593 Intrinsic IntOp, bit Commutable = 0>
2594 : N3VInt_QHS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2595 OpcodeStr, Dt, IntOp, Commutable> {
2596 def v1i64 : N3VDInt<op24, op23, 0b11, op11_8, op4, f, itinD32,
2597 OpcodeStr, !strconcat(Dt, "64"),
2598 v1i64, v1i64, IntOp, Commutable>;
2599 def v2i64 : N3VQInt<op24, op23, 0b11, op11_8, op4, f, itinQ32,
2600 OpcodeStr, !strconcat(Dt, "64"),
2601 v2i64, v2i64, IntOp, Commutable>;
2603 multiclass N3VInt_QHSDSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2604 InstrItinClass itinD16, InstrItinClass itinD32,
2605 InstrItinClass itinQ16, InstrItinClass itinQ32,
2606 string OpcodeStr, string Dt,
2608 : N3VInt_QHSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2609 OpcodeStr, Dt, IntOp> {
2610 def v1i64 : N3VDIntSh<op24, op23, 0b11, op11_8, op4, f, itinD32,
2611 OpcodeStr, !strconcat(Dt, "64"),
2612 v1i64, v1i64, IntOp>;
2613 def v2i64 : N3VQIntSh<op24, op23, 0b11, op11_8, op4, f, itinQ32,
2614 OpcodeStr, !strconcat(Dt, "64"),
2615 v2i64, v2i64, IntOp>;
2618 // Neon Narrowing 3-register vector intrinsics,
2619 // source operand element sizes of 16, 32 and 64 bits:
2620 multiclass N3VNInt_HSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2621 string OpcodeStr, string Dt,
2622 Intrinsic IntOp, bit Commutable = 0> {
2623 def v8i8 : N3VNInt<op24, op23, 0b00, op11_8, op4,
2624 OpcodeStr, !strconcat(Dt, "16"),
2625 v8i8, v8i16, IntOp, Commutable>;
2626 def v4i16 : N3VNInt<op24, op23, 0b01, op11_8, op4,
2627 OpcodeStr, !strconcat(Dt, "32"),
2628 v4i16, v4i32, IntOp, Commutable>;
2629 def v2i32 : N3VNInt<op24, op23, 0b10, op11_8, op4,
2630 OpcodeStr, !strconcat(Dt, "64"),
2631 v2i32, v2i64, IntOp, Commutable>;
2635 // Neon Long 3-register vector operations.
2637 multiclass N3VL_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2638 InstrItinClass itin16, InstrItinClass itin32,
2639 string OpcodeStr, string Dt,
2640 SDNode OpNode, bit Commutable = 0> {
2641 def v8i16 : N3VL<op24, op23, 0b00, op11_8, op4, itin16,
2642 OpcodeStr, !strconcat(Dt, "8"),
2643 v8i16, v8i8, OpNode, Commutable>;
2644 def v4i32 : N3VL<op24, op23, 0b01, op11_8, op4, itin16,
2645 OpcodeStr, !strconcat(Dt, "16"),
2646 v4i32, v4i16, OpNode, Commutable>;
2647 def v2i64 : N3VL<op24, op23, 0b10, op11_8, op4, itin32,
2648 OpcodeStr, !strconcat(Dt, "32"),
2649 v2i64, v2i32, OpNode, Commutable>;
2652 multiclass N3VLSL_HS<bit op24, bits<4> op11_8,
2653 InstrItinClass itin, string OpcodeStr, string Dt,
2655 def v4i16 : N3VLSL16<op24, 0b01, op11_8, itin, OpcodeStr,
2656 !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
2657 def v2i32 : N3VLSL<op24, 0b10, op11_8, itin, OpcodeStr,
2658 !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
2661 multiclass N3VLExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2662 InstrItinClass itin16, InstrItinClass itin32,
2663 string OpcodeStr, string Dt,
2664 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
2665 def v8i16 : N3VLExt<op24, op23, 0b00, op11_8, op4, itin16,
2666 OpcodeStr, !strconcat(Dt, "8"),
2667 v8i16, v8i8, OpNode, ExtOp, Commutable>;
2668 def v4i32 : N3VLExt<op24, op23, 0b01, op11_8, op4, itin16,
2669 OpcodeStr, !strconcat(Dt, "16"),
2670 v4i32, v4i16, OpNode, ExtOp, Commutable>;
2671 def v2i64 : N3VLExt<op24, op23, 0b10, op11_8, op4, itin32,
2672 OpcodeStr, !strconcat(Dt, "32"),
2673 v2i64, v2i32, OpNode, ExtOp, Commutable>;
2676 // Neon Long 3-register vector intrinsics.
2678 // First with only element sizes of 16 and 32 bits:
2679 multiclass N3VLInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
2680 InstrItinClass itin16, InstrItinClass itin32,
2681 string OpcodeStr, string Dt,
2682 Intrinsic IntOp, bit Commutable = 0> {
2683 def v4i32 : N3VLInt<op24, op23, 0b01, op11_8, op4, itin16,
2684 OpcodeStr, !strconcat(Dt, "16"),
2685 v4i32, v4i16, IntOp, Commutable>;
2686 def v2i64 : N3VLInt<op24, op23, 0b10, op11_8, op4, itin32,
2687 OpcodeStr, !strconcat(Dt, "32"),
2688 v2i64, v2i32, IntOp, Commutable>;
2691 multiclass N3VLIntSL_HS<bit op24, bits<4> op11_8,
2692 InstrItinClass itin, string OpcodeStr, string Dt,
2694 def v4i16 : N3VLIntSL16<op24, 0b01, op11_8, itin,
2695 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
2696 def v2i32 : N3VLIntSL<op24, 0b10, op11_8, itin,
2697 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
2700 // ....then also with element size of 8 bits:
2701 multiclass N3VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2702 InstrItinClass itin16, InstrItinClass itin32,
2703 string OpcodeStr, string Dt,
2704 Intrinsic IntOp, bit Commutable = 0>
2705 : N3VLInt_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt,
2706 IntOp, Commutable> {
2707 def v8i16 : N3VLInt<op24, op23, 0b00, op11_8, op4, itin16,
2708 OpcodeStr, !strconcat(Dt, "8"),
2709 v8i16, v8i8, IntOp, Commutable>;
2712 // ....with explicit extend (VABDL).
2713 multiclass N3VLIntExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2714 InstrItinClass itin, string OpcodeStr, string Dt,
2715 Intrinsic IntOp, SDNode ExtOp, bit Commutable = 0> {
2716 def v8i16 : N3VLIntExt<op24, op23, 0b00, op11_8, op4, itin,
2717 OpcodeStr, !strconcat(Dt, "8"),
2718 v8i16, v8i8, IntOp, ExtOp, Commutable>;
2719 def v4i32 : N3VLIntExt<op24, op23, 0b01, op11_8, op4, itin,
2720 OpcodeStr, !strconcat(Dt, "16"),
2721 v4i32, v4i16, IntOp, ExtOp, Commutable>;
2722 def v2i64 : N3VLIntExt<op24, op23, 0b10, op11_8, op4, itin,
2723 OpcodeStr, !strconcat(Dt, "32"),
2724 v2i64, v2i32, IntOp, ExtOp, Commutable>;
2728 // Neon Wide 3-register vector intrinsics,
2729 // source operand element sizes of 8, 16 and 32 bits:
2730 multiclass N3VW_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2731 string OpcodeStr, string Dt,
2732 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
2733 def v8i16 : N3VW<op24, op23, 0b00, op11_8, op4,
2734 OpcodeStr, !strconcat(Dt, "8"),
2735 v8i16, v8i8, OpNode, ExtOp, Commutable>;
2736 def v4i32 : N3VW<op24, op23, 0b01, op11_8, op4,
2737 OpcodeStr, !strconcat(Dt, "16"),
2738 v4i32, v4i16, OpNode, ExtOp, Commutable>;
2739 def v2i64 : N3VW<op24, op23, 0b10, op11_8, op4,
2740 OpcodeStr, !strconcat(Dt, "32"),
2741 v2i64, v2i32, OpNode, ExtOp, Commutable>;
2745 // Neon Multiply-Op vector operations,
2746 // element sizes of 8, 16 and 32 bits:
2747 multiclass N3VMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2748 InstrItinClass itinD16, InstrItinClass itinD32,
2749 InstrItinClass itinQ16, InstrItinClass itinQ32,
2750 string OpcodeStr, string Dt, SDNode OpNode> {
2751 // 64-bit vector types.
2752 def v8i8 : N3VDMulOp<op24, op23, 0b00, op11_8, op4, itinD16,
2753 OpcodeStr, !strconcat(Dt, "8"), v8i8, mul, OpNode>;
2754 def v4i16 : N3VDMulOp<op24, op23, 0b01, op11_8, op4, itinD16,
2755 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, OpNode>;
2756 def v2i32 : N3VDMulOp<op24, op23, 0b10, op11_8, op4, itinD32,
2757 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, OpNode>;
2759 // 128-bit vector types.
2760 def v16i8 : N3VQMulOp<op24, op23, 0b00, op11_8, op4, itinQ16,
2761 OpcodeStr, !strconcat(Dt, "8"), v16i8, mul, OpNode>;
2762 def v8i16 : N3VQMulOp<op24, op23, 0b01, op11_8, op4, itinQ16,
2763 OpcodeStr, !strconcat(Dt, "16"), v8i16, mul, OpNode>;
2764 def v4i32 : N3VQMulOp<op24, op23, 0b10, op11_8, op4, itinQ32,
2765 OpcodeStr, !strconcat(Dt, "32"), v4i32, mul, OpNode>;
2768 multiclass N3VMulOpSL_HS<bits<4> op11_8,
2769 InstrItinClass itinD16, InstrItinClass itinD32,
2770 InstrItinClass itinQ16, InstrItinClass itinQ32,
2771 string OpcodeStr, string Dt, SDNode ShOp> {
2772 def v4i16 : N3VDMulOpSL16<0b01, op11_8, itinD16,
2773 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, ShOp>;
2774 def v2i32 : N3VDMulOpSL<0b10, op11_8, itinD32,
2775 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, ShOp>;
2776 def v8i16 : N3VQMulOpSL16<0b01, op11_8, itinQ16,
2777 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16,
2779 def v4i32 : N3VQMulOpSL<0b10, op11_8, itinQ32,
2780 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32,
2784 // Neon Intrinsic-Op vector operations,
2785 // element sizes of 8, 16 and 32 bits:
2786 multiclass N3VIntOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2787 InstrItinClass itinD, InstrItinClass itinQ,
2788 string OpcodeStr, string Dt, Intrinsic IntOp,
2790 // 64-bit vector types.
2791 def v8i8 : N3VDIntOp<op24, op23, 0b00, op11_8, op4, itinD,
2792 OpcodeStr, !strconcat(Dt, "8"), v8i8, IntOp, OpNode>;
2793 def v4i16 : N3VDIntOp<op24, op23, 0b01, op11_8, op4, itinD,
2794 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp, OpNode>;
2795 def v2i32 : N3VDIntOp<op24, op23, 0b10, op11_8, op4, itinD,
2796 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp, OpNode>;
2798 // 128-bit vector types.
2799 def v16i8 : N3VQIntOp<op24, op23, 0b00, op11_8, op4, itinQ,
2800 OpcodeStr, !strconcat(Dt, "8"), v16i8, IntOp, OpNode>;
2801 def v8i16 : N3VQIntOp<op24, op23, 0b01, op11_8, op4, itinQ,
2802 OpcodeStr, !strconcat(Dt, "16"), v8i16, IntOp, OpNode>;
2803 def v4i32 : N3VQIntOp<op24, op23, 0b10, op11_8, op4, itinQ,
2804 OpcodeStr, !strconcat(Dt, "32"), v4i32, IntOp, OpNode>;
2807 // Neon 3-argument intrinsics,
2808 // element sizes of 8, 16 and 32 bits:
2809 multiclass N3VInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2810 InstrItinClass itinD, InstrItinClass itinQ,
2811 string OpcodeStr, string Dt, Intrinsic IntOp> {
2812 // 64-bit vector types.
2813 def v8i8 : N3VDInt3<op24, op23, 0b00, op11_8, op4, itinD,
2814 OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
2815 def v4i16 : N3VDInt3<op24, op23, 0b01, op11_8, op4, itinD,
2816 OpcodeStr, !strconcat(Dt, "16"), v4i16, v4i16, IntOp>;
2817 def v2i32 : N3VDInt3<op24, op23, 0b10, op11_8, op4, itinD,
2818 OpcodeStr, !strconcat(Dt, "32"), v2i32, v2i32, IntOp>;
2820 // 128-bit vector types.
2821 def v16i8 : N3VQInt3<op24, op23, 0b00, op11_8, op4, itinQ,
2822 OpcodeStr, !strconcat(Dt, "8"), v16i8, v16i8, IntOp>;
2823 def v8i16 : N3VQInt3<op24, op23, 0b01, op11_8, op4, itinQ,
2824 OpcodeStr, !strconcat(Dt, "16"), v8i16, v8i16, IntOp>;
2825 def v4i32 : N3VQInt3<op24, op23, 0b10, op11_8, op4, itinQ,
2826 OpcodeStr, !strconcat(Dt, "32"), v4i32, v4i32, IntOp>;
2830 // Neon Long Multiply-Op vector operations,
2831 // element sizes of 8, 16 and 32 bits:
2832 multiclass N3VLMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2833 InstrItinClass itin16, InstrItinClass itin32,
2834 string OpcodeStr, string Dt, SDNode MulOp,
2836 def v8i16 : N3VLMulOp<op24, op23, 0b00, op11_8, op4, itin16, OpcodeStr,
2837 !strconcat(Dt, "8"), v8i16, v8i8, MulOp, OpNode>;
2838 def v4i32 : N3VLMulOp<op24, op23, 0b01, op11_8, op4, itin16, OpcodeStr,
2839 !strconcat(Dt, "16"), v4i32, v4i16, MulOp, OpNode>;
2840 def v2i64 : N3VLMulOp<op24, op23, 0b10, op11_8, op4, itin32, OpcodeStr,
2841 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
2844 multiclass N3VLMulOpSL_HS<bit op24, bits<4> op11_8, string OpcodeStr,
2845 string Dt, SDNode MulOp, SDNode OpNode> {
2846 def v4i16 : N3VLMulOpSL16<op24, 0b01, op11_8, IIC_VMACi16D, OpcodeStr,
2847 !strconcat(Dt,"16"), v4i32, v4i16, MulOp, OpNode>;
2848 def v2i32 : N3VLMulOpSL<op24, 0b10, op11_8, IIC_VMACi32D, OpcodeStr,
2849 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
2853 // Neon Long 3-argument intrinsics.
2855 // First with only element sizes of 16 and 32 bits:
2856 multiclass N3VLInt3_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
2857 InstrItinClass itin16, InstrItinClass itin32,
2858 string OpcodeStr, string Dt, Intrinsic IntOp> {
2859 def v4i32 : N3VLInt3<op24, op23, 0b01, op11_8, op4, itin16,
2860 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
2861 def v2i64 : N3VLInt3<op24, op23, 0b10, op11_8, op4, itin32,
2862 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
2865 multiclass N3VLInt3SL_HS<bit op24, bits<4> op11_8,
2866 string OpcodeStr, string Dt, Intrinsic IntOp> {
2867 def v4i16 : N3VLInt3SL16<op24, 0b01, op11_8, IIC_VMACi16D,
2868 OpcodeStr, !strconcat(Dt,"16"), v4i32, v4i16, IntOp>;
2869 def v2i32 : N3VLInt3SL<op24, 0b10, op11_8, IIC_VMACi32D,
2870 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
2873 // ....then also with element size of 8 bits:
2874 multiclass N3VLInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2875 InstrItinClass itin16, InstrItinClass itin32,
2876 string OpcodeStr, string Dt, Intrinsic IntOp>
2877 : N3VLInt3_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt, IntOp> {
2878 def v8i16 : N3VLInt3<op24, op23, 0b00, op11_8, op4, itin16,
2879 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
2882 // ....with explicit extend (VABAL).
2883 multiclass N3VLIntExtOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2884 InstrItinClass itin, string OpcodeStr, string Dt,
2885 Intrinsic IntOp, SDNode ExtOp, SDNode OpNode> {
2886 def v8i16 : N3VLIntExtOp<op24, op23, 0b00, op11_8, op4, itin,
2887 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8,
2888 IntOp, ExtOp, OpNode>;
2889 def v4i32 : N3VLIntExtOp<op24, op23, 0b01, op11_8, op4, itin,
2890 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16,
2891 IntOp, ExtOp, OpNode>;
2892 def v2i64 : N3VLIntExtOp<op24, op23, 0b10, op11_8, op4, itin,
2893 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32,
2894 IntOp, ExtOp, OpNode>;
2898 // Neon 2-register vector intrinsics,
2899 // element sizes of 8, 16 and 32 bits:
2900 multiclass N2VInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2901 bits<5> op11_7, bit op4,
2902 InstrItinClass itinD, InstrItinClass itinQ,
2903 string OpcodeStr, string Dt, Intrinsic IntOp> {
2904 // 64-bit vector types.
2905 def v8i8 : N2VDInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2906 itinD, OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
2907 def v4i16 : N2VDInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2908 itinD, OpcodeStr, !strconcat(Dt, "16"),v4i16,v4i16,IntOp>;
2909 def v2i32 : N2VDInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2910 itinD, OpcodeStr, !strconcat(Dt, "32"),v2i32,v2i32,IntOp>;
2912 // 128-bit vector types.
2913 def v16i8 : N2VQInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2914 itinQ, OpcodeStr, !strconcat(Dt, "8"), v16i8,v16i8,IntOp>;
2915 def v8i16 : N2VQInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2916 itinQ, OpcodeStr, !strconcat(Dt, "16"),v8i16,v8i16,IntOp>;
2917 def v4i32 : N2VQInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2918 itinQ, OpcodeStr, !strconcat(Dt, "32"),v4i32,v4i32,IntOp>;
2922 // Neon Pairwise long 2-register intrinsics,
2923 // element sizes of 8, 16 and 32 bits:
2924 multiclass N2VPLInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2925 bits<5> op11_7, bit op4,
2926 string OpcodeStr, string Dt, Intrinsic IntOp> {
2927 // 64-bit vector types.
2928 def v8i8 : N2VDPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2929 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
2930 def v4i16 : N2VDPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2931 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
2932 def v2i32 : N2VDPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2933 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
2935 // 128-bit vector types.
2936 def v16i8 : N2VQPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2937 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
2938 def v8i16 : N2VQPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2939 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
2940 def v4i32 : N2VQPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2941 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
2945 // Neon Pairwise long 2-register accumulate intrinsics,
2946 // element sizes of 8, 16 and 32 bits:
2947 multiclass N2VPLInt2_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2948 bits<5> op11_7, bit op4,
2949 string OpcodeStr, string Dt, Intrinsic IntOp> {
2950 // 64-bit vector types.
2951 def v8i8 : N2VDPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2952 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
2953 def v4i16 : N2VDPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2954 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
2955 def v2i32 : N2VDPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2956 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
2958 // 128-bit vector types.
2959 def v16i8 : N2VQPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2960 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
2961 def v8i16 : N2VQPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2962 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
2963 def v4i32 : N2VQPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2964 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
2968 // Neon 2-register vector shift by immediate,
2969 // with f of either N2RegVShLFrm or N2RegVShRFrm
2970 // element sizes of 8, 16, 32 and 64 bits:
2971 multiclass N2VSh_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2972 InstrItinClass itin, string OpcodeStr, string Dt,
2973 SDNode OpNode, Format f> {
2974 // 64-bit vector types.
2975 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
2976 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
2977 let Inst{21-19} = 0b001; // imm6 = 001xxx
2979 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
2980 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
2981 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2983 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
2984 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
2985 let Inst{21} = 0b1; // imm6 = 1xxxxx
2987 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, f, itin,
2988 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
2991 // 128-bit vector types.
2992 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
2993 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
2994 let Inst{21-19} = 0b001; // imm6 = 001xxx
2996 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
2997 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
2998 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3000 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
3001 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
3002 let Inst{21} = 0b1; // imm6 = 1xxxxx
3004 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, f, itin,
3005 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
3009 // Neon Shift-Accumulate vector operations,
3010 // element sizes of 8, 16, 32 and 64 bits:
3011 multiclass N2VShAdd_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3012 string OpcodeStr, string Dt, SDNode ShOp> {
3013 // 64-bit vector types.
3014 def v8i8 : N2VDShAdd<op24, op23, op11_8, 0, op4,
3015 OpcodeStr, !strconcat(Dt, "8"), v8i8, ShOp> {
3016 let Inst{21-19} = 0b001; // imm6 = 001xxx
3018 def v4i16 : N2VDShAdd<op24, op23, op11_8, 0, op4,
3019 OpcodeStr, !strconcat(Dt, "16"), v4i16, ShOp> {
3020 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3022 def v2i32 : N2VDShAdd<op24, op23, op11_8, 0, op4,
3023 OpcodeStr, !strconcat(Dt, "32"), v2i32, ShOp> {
3024 let Inst{21} = 0b1; // imm6 = 1xxxxx
3026 def v1i64 : N2VDShAdd<op24, op23, op11_8, 1, op4,
3027 OpcodeStr, !strconcat(Dt, "64"), v1i64, ShOp>;
3030 // 128-bit vector types.
3031 def v16i8 : N2VQShAdd<op24, op23, op11_8, 0, op4,
3032 OpcodeStr, !strconcat(Dt, "8"), v16i8, ShOp> {
3033 let Inst{21-19} = 0b001; // imm6 = 001xxx
3035 def v8i16 : N2VQShAdd<op24, op23, op11_8, 0, op4,
3036 OpcodeStr, !strconcat(Dt, "16"), v8i16, ShOp> {
3037 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3039 def v4i32 : N2VQShAdd<op24, op23, op11_8, 0, op4,
3040 OpcodeStr, !strconcat(Dt, "32"), v4i32, ShOp> {
3041 let Inst{21} = 0b1; // imm6 = 1xxxxx
3043 def v2i64 : N2VQShAdd<op24, op23, op11_8, 1, op4,
3044 OpcodeStr, !strconcat(Dt, "64"), v2i64, ShOp>;
3049 // Neon Shift-Insert vector operations,
3050 // with f of either N2RegVShLFrm or N2RegVShRFrm
3051 // element sizes of 8, 16, 32 and 64 bits:
3052 multiclass N2VShIns_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3053 string OpcodeStr, SDNode ShOp,
3055 // 64-bit vector types.
3056 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4,
3057 f, OpcodeStr, "8", v8i8, ShOp> {
3058 let Inst{21-19} = 0b001; // imm6 = 001xxx
3060 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4,
3061 f, OpcodeStr, "16", v4i16, ShOp> {
3062 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3064 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4,
3065 f, OpcodeStr, "32", v2i32, ShOp> {
3066 let Inst{21} = 0b1; // imm6 = 1xxxxx
3068 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4,
3069 f, OpcodeStr, "64", v1i64, ShOp>;
3072 // 128-bit vector types.
3073 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4,
3074 f, OpcodeStr, "8", v16i8, ShOp> {
3075 let Inst{21-19} = 0b001; // imm6 = 001xxx
3077 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4,
3078 f, OpcodeStr, "16", v8i16, ShOp> {
3079 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3081 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4,
3082 f, OpcodeStr, "32", v4i32, ShOp> {
3083 let Inst{21} = 0b1; // imm6 = 1xxxxx
3085 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4,
3086 f, OpcodeStr, "64", v2i64, ShOp>;
3090 // Neon Shift Long operations,
3091 // element sizes of 8, 16, 32 bits:
3092 multiclass N2VLSh_QHS<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
3093 bit op4, string OpcodeStr, string Dt, SDNode OpNode> {
3094 def v8i16 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3095 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode> {
3096 let Inst{21-19} = 0b001; // imm6 = 001xxx
3098 def v4i32 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3099 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode> {
3100 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3102 def v2i64 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3103 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode> {
3104 let Inst{21} = 0b1; // imm6 = 1xxxxx
3108 // Neon Shift Narrow operations,
3109 // element sizes of 16, 32, 64 bits:
3110 multiclass N2VNSh_HSD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
3111 bit op4, InstrItinClass itin, string OpcodeStr, string Dt,
3113 def v8i8 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3114 OpcodeStr, !strconcat(Dt, "16"), v8i8, v8i16, OpNode> {
3115 let Inst{21-19} = 0b001; // imm6 = 001xxx
3117 def v4i16 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3118 OpcodeStr, !strconcat(Dt, "32"), v4i16, v4i32, OpNode> {
3119 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3121 def v2i32 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3122 OpcodeStr, !strconcat(Dt, "64"), v2i32, v2i64, OpNode> {
3123 let Inst{21} = 0b1; // imm6 = 1xxxxx
3127 //===----------------------------------------------------------------------===//
3128 // Instruction Definitions.
3129 //===----------------------------------------------------------------------===//
3131 // Vector Add Operations.
3133 // VADD : Vector Add (integer and floating-point)
3134 defm VADD : N3V_QHSD<0, 0, 0b1000, 0, IIC_VBINiD, IIC_VBINiQ, "vadd", "i",
3136 def VADDfd : N3VD<0, 0, 0b00, 0b1101, 0, IIC_VBIND, "vadd", "f32",
3137 v2f32, v2f32, fadd, 1>;
3138 def VADDfq : N3VQ<0, 0, 0b00, 0b1101, 0, IIC_VBINQ, "vadd", "f32",
3139 v4f32, v4f32, fadd, 1>;
3140 // VADDL : Vector Add Long (Q = D + D)
3141 defm VADDLs : N3VLExt_QHS<0,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
3142 "vaddl", "s", add, sext, 1>;
3143 defm VADDLu : N3VLExt_QHS<1,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
3144 "vaddl", "u", add, zext, 1>;
3145 // VADDW : Vector Add Wide (Q = Q + D)
3146 defm VADDWs : N3VW_QHS<0,1,0b0001,0, "vaddw", "s", add, sext, 0>;
3147 defm VADDWu : N3VW_QHS<1,1,0b0001,0, "vaddw", "u", add, zext, 0>;
3148 // VHADD : Vector Halving Add
3149 defm VHADDs : N3VInt_QHS<0, 0, 0b0000, 0, N3RegFrm,
3150 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3151 "vhadd", "s", int_arm_neon_vhadds, 1>;
3152 defm VHADDu : N3VInt_QHS<1, 0, 0b0000, 0, N3RegFrm,
3153 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3154 "vhadd", "u", int_arm_neon_vhaddu, 1>;
3155 // VRHADD : Vector Rounding Halving Add
3156 defm VRHADDs : N3VInt_QHS<0, 0, 0b0001, 0, N3RegFrm,
3157 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3158 "vrhadd", "s", int_arm_neon_vrhadds, 1>;
3159 defm VRHADDu : N3VInt_QHS<1, 0, 0b0001, 0, N3RegFrm,
3160 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3161 "vrhadd", "u", int_arm_neon_vrhaddu, 1>;
3162 // VQADD : Vector Saturating Add
3163 defm VQADDs : N3VInt_QHSD<0, 0, 0b0000, 1, N3RegFrm,
3164 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3165 "vqadd", "s", int_arm_neon_vqadds, 1>;
3166 defm VQADDu : N3VInt_QHSD<1, 0, 0b0000, 1, N3RegFrm,
3167 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3168 "vqadd", "u", int_arm_neon_vqaddu, 1>;
3169 // VADDHN : Vector Add and Narrow Returning High Half (D = Q + Q)
3170 defm VADDHN : N3VNInt_HSD<0,1,0b0100,0, "vaddhn", "i",
3171 int_arm_neon_vaddhn, 1>;
3172 // VRADDHN : Vector Rounding Add and Narrow Returning High Half (D = Q + Q)
3173 defm VRADDHN : N3VNInt_HSD<1,1,0b0100,0, "vraddhn", "i",
3174 int_arm_neon_vraddhn, 1>;
3176 // Vector Multiply Operations.
3178 // VMUL : Vector Multiply (integer, polynomial and floating-point)
3179 defm VMUL : N3V_QHS<0, 0, 0b1001, 1, IIC_VMULi16D, IIC_VMULi32D,
3180 IIC_VMULi16Q, IIC_VMULi32Q, "vmul", "i", mul, 1>;
3181 def VMULpd : N3VDInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16D, "vmul",
3182 "p8", v8i8, v8i8, int_arm_neon_vmulp, 1>;
3183 def VMULpq : N3VQInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16Q, "vmul",
3184 "p8", v16i8, v16i8, int_arm_neon_vmulp, 1>;
3185 def VMULfd : N3VD<1, 0, 0b00, 0b1101, 1, IIC_VFMULD, "vmul", "f32",
3186 v2f32, v2f32, fmul, 1>;
3187 def VMULfq : N3VQ<1, 0, 0b00, 0b1101, 1, IIC_VFMULQ, "vmul", "f32",
3188 v4f32, v4f32, fmul, 1>;
3189 defm VMULsl : N3VSL_HS<0b1000, "vmul", "i", mul>;
3190 def VMULslfd : N3VDSL<0b10, 0b1001, IIC_VBIND, "vmul", "f32", v2f32, fmul>;
3191 def VMULslfq : N3VQSL<0b10, 0b1001, IIC_VBINQ, "vmul", "f32", v4f32,
3194 def : Pat<(v8i16 (mul (v8i16 QPR:$src1),
3195 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
3196 (v8i16 (VMULslv8i16 (v8i16 QPR:$src1),
3197 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3198 (DSubReg_i16_reg imm:$lane))),
3199 (SubReg_i16_lane imm:$lane)))>;
3200 def : Pat<(v4i32 (mul (v4i32 QPR:$src1),
3201 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
3202 (v4i32 (VMULslv4i32 (v4i32 QPR:$src1),
3203 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3204 (DSubReg_i32_reg imm:$lane))),
3205 (SubReg_i32_lane imm:$lane)))>;
3206 def : Pat<(v4f32 (fmul (v4f32 QPR:$src1),
3207 (v4f32 (NEONvduplane (v4f32 QPR:$src2), imm:$lane)))),
3208 (v4f32 (VMULslfq (v4f32 QPR:$src1),
3209 (v2f32 (EXTRACT_SUBREG QPR:$src2,
3210 (DSubReg_i32_reg imm:$lane))),
3211 (SubReg_i32_lane imm:$lane)))>;
3213 // VQDMULH : Vector Saturating Doubling Multiply Returning High Half
3214 defm VQDMULH : N3VInt_HS<0, 0, 0b1011, 0, N3RegFrm, IIC_VMULi16D, IIC_VMULi32D,
3215 IIC_VMULi16Q, IIC_VMULi32Q,
3216 "vqdmulh", "s", int_arm_neon_vqdmulh, 1>;
3217 defm VQDMULHsl: N3VIntSL_HS<0b1100, IIC_VMULi16D, IIC_VMULi32D,
3218 IIC_VMULi16Q, IIC_VMULi32Q,
3219 "vqdmulh", "s", int_arm_neon_vqdmulh>;
3220 def : Pat<(v8i16 (int_arm_neon_vqdmulh (v8i16 QPR:$src1),
3221 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
3223 (v8i16 (VQDMULHslv8i16 (v8i16 QPR:$src1),
3224 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3225 (DSubReg_i16_reg imm:$lane))),
3226 (SubReg_i16_lane imm:$lane)))>;
3227 def : Pat<(v4i32 (int_arm_neon_vqdmulh (v4i32 QPR:$src1),
3228 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
3230 (v4i32 (VQDMULHslv4i32 (v4i32 QPR:$src1),
3231 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3232 (DSubReg_i32_reg imm:$lane))),
3233 (SubReg_i32_lane imm:$lane)))>;
3235 // VQRDMULH : Vector Rounding Saturating Doubling Multiply Returning High Half
3236 defm VQRDMULH : N3VInt_HS<1, 0, 0b1011, 0, N3RegFrm,
3237 IIC_VMULi16D,IIC_VMULi32D,IIC_VMULi16Q,IIC_VMULi32Q,
3238 "vqrdmulh", "s", int_arm_neon_vqrdmulh, 1>;
3239 defm VQRDMULHsl : N3VIntSL_HS<0b1101, IIC_VMULi16D, IIC_VMULi32D,
3240 IIC_VMULi16Q, IIC_VMULi32Q,
3241 "vqrdmulh", "s", int_arm_neon_vqrdmulh>;
3242 def : Pat<(v8i16 (int_arm_neon_vqrdmulh (v8i16 QPR:$src1),
3243 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
3245 (v8i16 (VQRDMULHslv8i16 (v8i16 QPR:$src1),
3246 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3247 (DSubReg_i16_reg imm:$lane))),
3248 (SubReg_i16_lane imm:$lane)))>;
3249 def : Pat<(v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$src1),
3250 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
3252 (v4i32 (VQRDMULHslv4i32 (v4i32 QPR:$src1),
3253 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3254 (DSubReg_i32_reg imm:$lane))),
3255 (SubReg_i32_lane imm:$lane)))>;
3257 // VMULL : Vector Multiply Long (integer and polynomial) (Q = D * D)
3258 defm VMULLs : N3VL_QHS<0,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
3259 "vmull", "s", NEONvmulls, 1>;
3260 defm VMULLu : N3VL_QHS<1,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
3261 "vmull", "u", NEONvmullu, 1>;
3262 def VMULLp : N3VLInt<0, 1, 0b00, 0b1110, 0, IIC_VMULi16D, "vmull", "p8",
3263 v8i16, v8i8, int_arm_neon_vmullp, 1>;
3264 defm VMULLsls : N3VLSL_HS<0, 0b1010, IIC_VMULi16D, "vmull", "s", NEONvmulls>;
3265 defm VMULLslu : N3VLSL_HS<1, 0b1010, IIC_VMULi16D, "vmull", "u", NEONvmullu>;
3267 // VQDMULL : Vector Saturating Doubling Multiply Long (Q = D * D)
3268 defm VQDMULL : N3VLInt_HS<0,1,0b1101,0, IIC_VMULi16D, IIC_VMULi32D,
3269 "vqdmull", "s", int_arm_neon_vqdmull, 1>;
3270 defm VQDMULLsl: N3VLIntSL_HS<0, 0b1011, IIC_VMULi16D,
3271 "vqdmull", "s", int_arm_neon_vqdmull>;
3273 // Vector Multiply-Accumulate and Multiply-Subtract Operations.
3275 // VMLA : Vector Multiply Accumulate (integer and floating-point)
3276 defm VMLA : N3VMulOp_QHS<0, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3277 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
3278 def VMLAfd : N3VDMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACD, "vmla", "f32",
3279 v2f32, fmul_su, fadd_mlx>,
3280 Requires<[HasNEON, UseFPVMLx]>;
3281 def VMLAfq : N3VQMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACQ, "vmla", "f32",
3282 v4f32, fmul_su, fadd_mlx>,
3283 Requires<[HasNEON, UseFPVMLx]>;
3284 defm VMLAsl : N3VMulOpSL_HS<0b0000, IIC_VMACi16D, IIC_VMACi32D,
3285 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
3286 def VMLAslfd : N3VDMulOpSL<0b10, 0b0001, IIC_VMACD, "vmla", "f32",
3287 v2f32, fmul_su, fadd_mlx>,
3288 Requires<[HasNEON, UseFPVMLx]>;
3289 def VMLAslfq : N3VQMulOpSL<0b10, 0b0001, IIC_VMACQ, "vmla", "f32",
3290 v4f32, v2f32, fmul_su, fadd_mlx>,
3291 Requires<[HasNEON, UseFPVMLx]>;
3293 def : Pat<(v8i16 (add (v8i16 QPR:$src1),
3294 (mul (v8i16 QPR:$src2),
3295 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
3296 (v8i16 (VMLAslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
3297 (v4i16 (EXTRACT_SUBREG QPR:$src3,
3298 (DSubReg_i16_reg imm:$lane))),
3299 (SubReg_i16_lane imm:$lane)))>;
3301 def : Pat<(v4i32 (add (v4i32 QPR:$src1),
3302 (mul (v4i32 QPR:$src2),
3303 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
3304 (v4i32 (VMLAslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
3305 (v2i32 (EXTRACT_SUBREG QPR:$src3,
3306 (DSubReg_i32_reg imm:$lane))),
3307 (SubReg_i32_lane imm:$lane)))>;
3309 def : Pat<(v4f32 (fadd_mlx (v4f32 QPR:$src1),
3310 (fmul_su (v4f32 QPR:$src2),
3311 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
3312 (v4f32 (VMLAslfq (v4f32 QPR:$src1),
3314 (v2f32 (EXTRACT_SUBREG QPR:$src3,
3315 (DSubReg_i32_reg imm:$lane))),
3316 (SubReg_i32_lane imm:$lane)))>,
3317 Requires<[HasNEON, UseFPVMLx]>;
3319 // VMLAL : Vector Multiply Accumulate Long (Q += D * D)
3320 defm VMLALs : N3VLMulOp_QHS<0,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
3321 "vmlal", "s", NEONvmulls, add>;
3322 defm VMLALu : N3VLMulOp_QHS<1,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
3323 "vmlal", "u", NEONvmullu, add>;
3325 defm VMLALsls : N3VLMulOpSL_HS<0, 0b0010, "vmlal", "s", NEONvmulls, add>;
3326 defm VMLALslu : N3VLMulOpSL_HS<1, 0b0010, "vmlal", "u", NEONvmullu, add>;
3328 // VQDMLAL : Vector Saturating Doubling Multiply Accumulate Long (Q += D * D)
3329 defm VQDMLAL : N3VLInt3_HS<0, 1, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3330 "vqdmlal", "s", int_arm_neon_vqdmlal>;
3331 defm VQDMLALsl: N3VLInt3SL_HS<0, 0b0011, "vqdmlal", "s", int_arm_neon_vqdmlal>;
3333 // VMLS : Vector Multiply Subtract (integer and floating-point)
3334 defm VMLS : N3VMulOp_QHS<1, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3335 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
3336 def VMLSfd : N3VDMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACD, "vmls", "f32",
3337 v2f32, fmul_su, fsub_mlx>,
3338 Requires<[HasNEON, UseFPVMLx]>;
3339 def VMLSfq : N3VQMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACQ, "vmls", "f32",
3340 v4f32, fmul_su, fsub_mlx>,
3341 Requires<[HasNEON, UseFPVMLx]>;
3342 defm VMLSsl : N3VMulOpSL_HS<0b0100, IIC_VMACi16D, IIC_VMACi32D,
3343 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
3344 def VMLSslfd : N3VDMulOpSL<0b10, 0b0101, IIC_VMACD, "vmls", "f32",
3345 v2f32, fmul_su, fsub_mlx>,
3346 Requires<[HasNEON, UseFPVMLx]>;
3347 def VMLSslfq : N3VQMulOpSL<0b10, 0b0101, IIC_VMACQ, "vmls", "f32",
3348 v4f32, v2f32, fmul_su, fsub_mlx>,
3349 Requires<[HasNEON, UseFPVMLx]>;
3351 def : Pat<(v8i16 (sub (v8i16 QPR:$src1),
3352 (mul (v8i16 QPR:$src2),
3353 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
3354 (v8i16 (VMLSslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
3355 (v4i16 (EXTRACT_SUBREG QPR:$src3,
3356 (DSubReg_i16_reg imm:$lane))),
3357 (SubReg_i16_lane imm:$lane)))>;
3359 def : Pat<(v4i32 (sub (v4i32 QPR:$src1),
3360 (mul (v4i32 QPR:$src2),
3361 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
3362 (v4i32 (VMLSslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
3363 (v2i32 (EXTRACT_SUBREG QPR:$src3,
3364 (DSubReg_i32_reg imm:$lane))),
3365 (SubReg_i32_lane imm:$lane)))>;
3367 def : Pat<(v4f32 (fsub_mlx (v4f32 QPR:$src1),
3368 (fmul_su (v4f32 QPR:$src2),
3369 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
3370 (v4f32 (VMLSslfq (v4f32 QPR:$src1), (v4f32 QPR:$src2),
3371 (v2f32 (EXTRACT_SUBREG QPR:$src3,
3372 (DSubReg_i32_reg imm:$lane))),
3373 (SubReg_i32_lane imm:$lane)))>,
3374 Requires<[HasNEON, UseFPVMLx]>;
3376 // VMLSL : Vector Multiply Subtract Long (Q -= D * D)
3377 defm VMLSLs : N3VLMulOp_QHS<0,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
3378 "vmlsl", "s", NEONvmulls, sub>;
3379 defm VMLSLu : N3VLMulOp_QHS<1,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
3380 "vmlsl", "u", NEONvmullu, sub>;
3382 defm VMLSLsls : N3VLMulOpSL_HS<0, 0b0110, "vmlsl", "s", NEONvmulls, sub>;
3383 defm VMLSLslu : N3VLMulOpSL_HS<1, 0b0110, "vmlsl", "u", NEONvmullu, sub>;
3385 // VQDMLSL : Vector Saturating Doubling Multiply Subtract Long (Q -= D * D)
3386 defm VQDMLSL : N3VLInt3_HS<0, 1, 0b1011, 0, IIC_VMACi16D, IIC_VMACi32D,
3387 "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
3388 defm VQDMLSLsl: N3VLInt3SL_HS<0, 0b111, "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
3390 // Vector Subtract Operations.
3392 // VSUB : Vector Subtract (integer and floating-point)
3393 defm VSUB : N3V_QHSD<1, 0, 0b1000, 0, IIC_VSUBiD, IIC_VSUBiQ,
3394 "vsub", "i", sub, 0>;
3395 def VSUBfd : N3VD<0, 0, 0b10, 0b1101, 0, IIC_VBIND, "vsub", "f32",
3396 v2f32, v2f32, fsub, 0>;
3397 def VSUBfq : N3VQ<0, 0, 0b10, 0b1101, 0, IIC_VBINQ, "vsub", "f32",
3398 v4f32, v4f32, fsub, 0>;
3399 // VSUBL : Vector Subtract Long (Q = D - D)
3400 defm VSUBLs : N3VLExt_QHS<0,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
3401 "vsubl", "s", sub, sext, 0>;
3402 defm VSUBLu : N3VLExt_QHS<1,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
3403 "vsubl", "u", sub, zext, 0>;
3404 // VSUBW : Vector Subtract Wide (Q = Q - D)
3405 defm VSUBWs : N3VW_QHS<0,1,0b0011,0, "vsubw", "s", sub, sext, 0>;
3406 defm VSUBWu : N3VW_QHS<1,1,0b0011,0, "vsubw", "u", sub, zext, 0>;
3407 // VHSUB : Vector Halving Subtract
3408 defm VHSUBs : N3VInt_QHS<0, 0, 0b0010, 0, N3RegFrm,
3409 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3410 "vhsub", "s", int_arm_neon_vhsubs, 0>;
3411 defm VHSUBu : N3VInt_QHS<1, 0, 0b0010, 0, N3RegFrm,
3412 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3413 "vhsub", "u", int_arm_neon_vhsubu, 0>;
3414 // VQSUB : Vector Saturing Subtract
3415 defm VQSUBs : N3VInt_QHSD<0, 0, 0b0010, 1, N3RegFrm,
3416 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3417 "vqsub", "s", int_arm_neon_vqsubs, 0>;
3418 defm VQSUBu : N3VInt_QHSD<1, 0, 0b0010, 1, N3RegFrm,
3419 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3420 "vqsub", "u", int_arm_neon_vqsubu, 0>;
3421 // VSUBHN : Vector Subtract and Narrow Returning High Half (D = Q - Q)
3422 defm VSUBHN : N3VNInt_HSD<0,1,0b0110,0, "vsubhn", "i",
3423 int_arm_neon_vsubhn, 0>;
3424 // VRSUBHN : Vector Rounding Subtract and Narrow Returning High Half (D=Q-Q)
3425 defm VRSUBHN : N3VNInt_HSD<1,1,0b0110,0, "vrsubhn", "i",
3426 int_arm_neon_vrsubhn, 0>;
3428 // Vector Comparisons.
3430 // VCEQ : Vector Compare Equal
3431 defm VCEQ : N3V_QHS<1, 0, 0b1000, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3432 IIC_VSUBi4Q, "vceq", "i", NEONvceq, 1>;
3433 def VCEQfd : N3VD<0,0,0b00,0b1110,0, IIC_VBIND, "vceq", "f32", v2i32, v2f32,
3435 def VCEQfq : N3VQ<0,0,0b00,0b1110,0, IIC_VBINQ, "vceq", "f32", v4i32, v4f32,
3438 defm VCEQz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00010, 0, "vceq", "i",
3439 "$Vd, $Vm, #0", NEONvceqz>;
3441 // VCGE : Vector Compare Greater Than or Equal
3442 defm VCGEs : N3V_QHS<0, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3443 IIC_VSUBi4Q, "vcge", "s", NEONvcge, 0>;
3444 defm VCGEu : N3V_QHS<1, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3445 IIC_VSUBi4Q, "vcge", "u", NEONvcgeu, 0>;
3446 def VCGEfd : N3VD<1,0,0b00,0b1110,0, IIC_VBIND, "vcge", "f32", v2i32, v2f32,
3448 def VCGEfq : N3VQ<1,0,0b00,0b1110,0, IIC_VBINQ, "vcge", "f32", v4i32, v4f32,
3451 defm VCGEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00001, 0, "vcge", "s",
3452 "$Vd, $Vm, #0", NEONvcgez>;
3453 defm VCLEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00011, 0, "vcle", "s",
3454 "$Vd, $Vm, #0", NEONvclez>;
3456 // VCGT : Vector Compare Greater Than
3457 defm VCGTs : N3V_QHS<0, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3458 IIC_VSUBi4Q, "vcgt", "s", NEONvcgt, 0>;
3459 defm VCGTu : N3V_QHS<1, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3460 IIC_VSUBi4Q, "vcgt", "u", NEONvcgtu, 0>;
3461 def VCGTfd : N3VD<1,0,0b10,0b1110,0, IIC_VBIND, "vcgt", "f32", v2i32, v2f32,
3463 def VCGTfq : N3VQ<1,0,0b10,0b1110,0, IIC_VBINQ, "vcgt", "f32", v4i32, v4f32,
3466 defm VCGTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00000, 0, "vcgt", "s",
3467 "$Vd, $Vm, #0", NEONvcgtz>;
3468 defm VCLTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00100, 0, "vclt", "s",
3469 "$Vd, $Vm, #0", NEONvcltz>;
3471 // VACGE : Vector Absolute Compare Greater Than or Equal (aka VCAGE)
3472 def VACGEd : N3VDInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacge",
3473 "f32", v2i32, v2f32, int_arm_neon_vacged, 0>;
3474 def VACGEq : N3VQInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacge",
3475 "f32", v4i32, v4f32, int_arm_neon_vacgeq, 0>;
3476 // VACGT : Vector Absolute Compare Greater Than (aka VCAGT)
3477 def VACGTd : N3VDInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacgt",
3478 "f32", v2i32, v2f32, int_arm_neon_vacgtd, 0>;
3479 def VACGTq : N3VQInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacgt",
3480 "f32", v4i32, v4f32, int_arm_neon_vacgtq, 0>;
3481 // VTST : Vector Test Bits
3482 defm VTST : N3V_QHS<0, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
3483 IIC_VBINi4Q, "vtst", "", NEONvtst, 1>;
3485 // Vector Bitwise Operations.
3487 def vnotd : PatFrag<(ops node:$in),
3488 (xor node:$in, (bitconvert (v8i8 NEONimmAllOnesV)))>;
3489 def vnotq : PatFrag<(ops node:$in),
3490 (xor node:$in, (bitconvert (v16i8 NEONimmAllOnesV)))>;
3493 // VAND : Vector Bitwise AND
3494 def VANDd : N3VDX<0, 0, 0b00, 0b0001, 1, IIC_VBINiD, "vand",
3495 v2i32, v2i32, and, 1>;
3496 def VANDq : N3VQX<0, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "vand",
3497 v4i32, v4i32, and, 1>;
3499 // VEOR : Vector Bitwise Exclusive OR
3500 def VEORd : N3VDX<1, 0, 0b00, 0b0001, 1, IIC_VBINiD, "veor",
3501 v2i32, v2i32, xor, 1>;
3502 def VEORq : N3VQX<1, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "veor",
3503 v4i32, v4i32, xor, 1>;
3505 // VORR : Vector Bitwise OR
3506 def VORRd : N3VDX<0, 0, 0b10, 0b0001, 1, IIC_VBINiD, "vorr",
3507 v2i32, v2i32, or, 1>;
3508 def VORRq : N3VQX<0, 0, 0b10, 0b0001, 1, IIC_VBINiQ, "vorr",
3509 v4i32, v4i32, or, 1>;
3511 def VORRiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 0, 1,
3512 (outs DPR:$Vd), (ins nModImm:$SIMM, DPR:$src),
3514 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
3516 (v4i16 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
3517 let Inst{9} = SIMM{9};
3520 def VORRiv2i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 0, 0, 1,
3521 (outs DPR:$Vd), (ins nModImm:$SIMM, DPR:$src),
3523 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
3525 (v2i32 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
3526 let Inst{10-9} = SIMM{10-9};
3529 def VORRiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 0, 1,
3530 (outs QPR:$Vd), (ins nModImm:$SIMM, QPR:$src),
3532 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
3534 (v8i16 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
3535 let Inst{9} = SIMM{9};
3538 def VORRiv4i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 1, 0, 1,
3539 (outs QPR:$Vd), (ins nModImm:$SIMM, QPR:$src),
3541 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
3543 (v4i32 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
3544 let Inst{10-9} = SIMM{10-9};
3548 // VBIC : Vector Bitwise Bit Clear (AND NOT)
3549 def VBICd : N3VX<0, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
3550 (ins DPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VBINiD,
3551 "vbic", "$Vd, $Vn, $Vm", "",
3552 [(set DPR:$Vd, (v2i32 (and DPR:$Vn,
3553 (vnotd DPR:$Vm))))]>;
3554 def VBICq : N3VX<0, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
3555 (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINiQ,
3556 "vbic", "$Vd, $Vn, $Vm", "",
3557 [(set QPR:$Vd, (v4i32 (and QPR:$Vn,
3558 (vnotq QPR:$Vm))))]>;
3560 def VBICiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 1, 1,
3561 (outs DPR:$Vd), (ins nModImm:$SIMM, DPR:$src),
3563 "vbic", "i16", "$Vd, $SIMM", "$src = $Vd",
3565 (v4i16 (NEONvbicImm DPR:$src, timm:$SIMM)))]> {
3566 let Inst{9} = SIMM{9};
3569 def VBICiv2i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 0, 1, 1,
3570 (outs DPR:$Vd), (ins nModImm:$SIMM, DPR:$src),
3572 "vbic", "i32", "$Vd, $SIMM", "$src = $Vd",
3574 (v2i32 (NEONvbicImm DPR:$src, timm:$SIMM)))]> {
3575 let Inst{10-9} = SIMM{10-9};
3578 def VBICiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 1, 1,
3579 (outs QPR:$Vd), (ins nModImm:$SIMM, QPR:$src),
3581 "vbic", "i16", "$Vd, $SIMM", "$src = $Vd",
3583 (v8i16 (NEONvbicImm QPR:$src, timm:$SIMM)))]> {
3584 let Inst{9} = SIMM{9};
3587 def VBICiv4i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 1, 1, 1,
3588 (outs QPR:$Vd), (ins nModImm:$SIMM, QPR:$src),
3590 "vbic", "i32", "$Vd, $SIMM", "$src = $Vd",
3592 (v4i32 (NEONvbicImm QPR:$src, timm:$SIMM)))]> {
3593 let Inst{10-9} = SIMM{10-9};
3596 // VORN : Vector Bitwise OR NOT
3597 def VORNd : N3VX<0, 0, 0b11, 0b0001, 0, 1, (outs DPR:$Vd),
3598 (ins DPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VBINiD,
3599 "vorn", "$Vd, $Vn, $Vm", "",
3600 [(set DPR:$Vd, (v2i32 (or DPR:$Vn,
3601 (vnotd DPR:$Vm))))]>;
3602 def VORNq : N3VX<0, 0, 0b11, 0b0001, 1, 1, (outs QPR:$Vd),
3603 (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINiQ,
3604 "vorn", "$Vd, $Vn, $Vm", "",
3605 [(set QPR:$Vd, (v4i32 (or QPR:$Vn,
3606 (vnotq QPR:$Vm))))]>;
3608 // VMVN : Vector Bitwise NOT (Immediate)
3610 let isReMaterializable = 1 in {
3612 def VMVNv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 1, 1, (outs DPR:$Vd),
3613 (ins nModImm:$SIMM), IIC_VMOVImm,
3614 "vmvn", "i16", "$Vd, $SIMM", "",
3615 [(set DPR:$Vd, (v4i16 (NEONvmvnImm timm:$SIMM)))]> {
3616 let Inst{9} = SIMM{9};
3619 def VMVNv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 1, 1, (outs QPR:$Vd),
3620 (ins nModImm:$SIMM), IIC_VMOVImm,
3621 "vmvn", "i16", "$Vd, $SIMM", "",
3622 [(set QPR:$Vd, (v8i16 (NEONvmvnImm timm:$SIMM)))]> {
3623 let Inst{9} = SIMM{9};
3626 def VMVNv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 1, 1, (outs DPR:$Vd),
3627 (ins nModImm:$SIMM), IIC_VMOVImm,
3628 "vmvn", "i32", "$Vd, $SIMM", "",
3629 [(set DPR:$Vd, (v2i32 (NEONvmvnImm timm:$SIMM)))]> {
3630 let Inst{11-8} = SIMM{11-8};
3633 def VMVNv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 1, 1, (outs QPR:$Vd),
3634 (ins nModImm:$SIMM), IIC_VMOVImm,
3635 "vmvn", "i32", "$Vd, $SIMM", "",
3636 [(set QPR:$Vd, (v4i32 (NEONvmvnImm timm:$SIMM)))]> {
3637 let Inst{11-8} = SIMM{11-8};
3641 // VMVN : Vector Bitwise NOT
3642 def VMVNd : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 0, 0,
3643 (outs DPR:$Vd), (ins DPR:$Vm), IIC_VSUBiD,
3644 "vmvn", "$Vd, $Vm", "",
3645 [(set DPR:$Vd, (v2i32 (vnotd DPR:$Vm)))]>;
3646 def VMVNq : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 1, 0,
3647 (outs QPR:$Vd), (ins QPR:$Vm), IIC_VSUBiD,
3648 "vmvn", "$Vd, $Vm", "",
3649 [(set QPR:$Vd, (v4i32 (vnotq QPR:$Vm)))]>;
3650 def : Pat<(v2i32 (vnotd DPR:$src)), (VMVNd DPR:$src)>;
3651 def : Pat<(v4i32 (vnotq QPR:$src)), (VMVNq QPR:$src)>;
3653 // VBSL : Vector Bitwise Select
3654 def VBSLd : N3VX<1, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
3655 (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
3656 N3RegFrm, IIC_VCNTiD,
3657 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3659 (v2i32 (or (and DPR:$Vn, DPR:$src1),
3660 (and DPR:$Vm, (vnotd DPR:$src1)))))]>;
3661 def VBSLq : N3VX<1, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
3662 (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
3663 N3RegFrm, IIC_VCNTiQ,
3664 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3666 (v4i32 (or (and QPR:$Vn, QPR:$src1),
3667 (and QPR:$Vm, (vnotq QPR:$src1)))))]>;
3669 // VBIF : Vector Bitwise Insert if False
3670 // like VBSL but with: "vbif $dst, $src3, $src1", "$src2 = $dst",
3671 // FIXME: This instruction's encoding MAY NOT BE correct.
3672 def VBIFd : N3VX<1, 0, 0b11, 0b0001, 0, 1,
3673 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
3674 N3RegFrm, IIC_VBINiD,
3675 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3676 [/* For disassembly only; pattern left blank */]>;
3677 def VBIFq : N3VX<1, 0, 0b11, 0b0001, 1, 1,
3678 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
3679 N3RegFrm, IIC_VBINiQ,
3680 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3681 [/* For disassembly only; pattern left blank */]>;
3683 // VBIT : Vector Bitwise Insert if True
3684 // like VBSL but with: "vbit $dst, $src2, $src1", "$src3 = $dst",
3685 // FIXME: This instruction's encoding MAY NOT BE correct.
3686 def VBITd : N3VX<1, 0, 0b10, 0b0001, 0, 1,
3687 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
3688 N3RegFrm, IIC_VBINiD,
3689 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3690 [/* For disassembly only; pattern left blank */]>;
3691 def VBITq : N3VX<1, 0, 0b10, 0b0001, 1, 1,
3692 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
3693 N3RegFrm, IIC_VBINiQ,
3694 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3695 [/* For disassembly only; pattern left blank */]>;
3697 // VBIT/VBIF are not yet implemented. The TwoAddress pass will not go looking
3698 // for equivalent operations with different register constraints; it just
3701 // Vector Absolute Differences.
3703 // VABD : Vector Absolute Difference
3704 defm VABDs : N3VInt_QHS<0, 0, 0b0111, 0, N3RegFrm,
3705 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3706 "vabd", "s", int_arm_neon_vabds, 1>;
3707 defm VABDu : N3VInt_QHS<1, 0, 0b0111, 0, N3RegFrm,
3708 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3709 "vabd", "u", int_arm_neon_vabdu, 1>;
3710 def VABDfd : N3VDInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBIND,
3711 "vabd", "f32", v2f32, v2f32, int_arm_neon_vabds, 1>;
3712 def VABDfq : N3VQInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBINQ,
3713 "vabd", "f32", v4f32, v4f32, int_arm_neon_vabds, 1>;
3715 // VABDL : Vector Absolute Difference Long (Q = | D - D |)
3716 defm VABDLs : N3VLIntExt_QHS<0,1,0b0111,0, IIC_VSUBi4Q,
3717 "vabdl", "s", int_arm_neon_vabds, zext, 1>;
3718 defm VABDLu : N3VLIntExt_QHS<1,1,0b0111,0, IIC_VSUBi4Q,
3719 "vabdl", "u", int_arm_neon_vabdu, zext, 1>;
3721 // VABA : Vector Absolute Difference and Accumulate
3722 defm VABAs : N3VIntOp_QHS<0,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
3723 "vaba", "s", int_arm_neon_vabds, add>;
3724 defm VABAu : N3VIntOp_QHS<1,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
3725 "vaba", "u", int_arm_neon_vabdu, add>;
3727 // VABAL : Vector Absolute Difference and Accumulate Long (Q += | D - D |)
3728 defm VABALs : N3VLIntExtOp_QHS<0,1,0b0101,0, IIC_VABAD,
3729 "vabal", "s", int_arm_neon_vabds, zext, add>;
3730 defm VABALu : N3VLIntExtOp_QHS<1,1,0b0101,0, IIC_VABAD,
3731 "vabal", "u", int_arm_neon_vabdu, zext, add>;
3733 // Vector Maximum and Minimum.
3735 // VMAX : Vector Maximum
3736 defm VMAXs : N3VInt_QHS<0, 0, 0b0110, 0, N3RegFrm,
3737 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3738 "vmax", "s", int_arm_neon_vmaxs, 1>;
3739 defm VMAXu : N3VInt_QHS<1, 0, 0b0110, 0, N3RegFrm,
3740 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3741 "vmax", "u", int_arm_neon_vmaxu, 1>;
3742 def VMAXfd : N3VDInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBIND,
3744 v2f32, v2f32, int_arm_neon_vmaxs, 1>;
3745 def VMAXfq : N3VQInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBINQ,
3747 v4f32, v4f32, int_arm_neon_vmaxs, 1>;
3749 // VMIN : Vector Minimum
3750 defm VMINs : N3VInt_QHS<0, 0, 0b0110, 1, N3RegFrm,
3751 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3752 "vmin", "s", int_arm_neon_vmins, 1>;
3753 defm VMINu : N3VInt_QHS<1, 0, 0b0110, 1, N3RegFrm,
3754 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3755 "vmin", "u", int_arm_neon_vminu, 1>;
3756 def VMINfd : N3VDInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBIND,
3758 v2f32, v2f32, int_arm_neon_vmins, 1>;
3759 def VMINfq : N3VQInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBINQ,
3761 v4f32, v4f32, int_arm_neon_vmins, 1>;
3763 // Vector Pairwise Operations.
3765 // VPADD : Vector Pairwise Add
3766 def VPADDi8 : N3VDInt<0, 0, 0b00, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
3768 v8i8, v8i8, int_arm_neon_vpadd, 0>;
3769 def VPADDi16 : N3VDInt<0, 0, 0b01, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
3771 v4i16, v4i16, int_arm_neon_vpadd, 0>;
3772 def VPADDi32 : N3VDInt<0, 0, 0b10, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
3774 v2i32, v2i32, int_arm_neon_vpadd, 0>;
3775 def VPADDf : N3VDInt<1, 0, 0b00, 0b1101, 0, N3RegFrm,
3776 IIC_VPBIND, "vpadd", "f32",
3777 v2f32, v2f32, int_arm_neon_vpadd, 0>;
3779 // VPADDL : Vector Pairwise Add Long
3780 defm VPADDLs : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00100, 0, "vpaddl", "s",
3781 int_arm_neon_vpaddls>;
3782 defm VPADDLu : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00101, 0, "vpaddl", "u",
3783 int_arm_neon_vpaddlu>;
3785 // VPADAL : Vector Pairwise Add and Accumulate Long
3786 defm VPADALs : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01100, 0, "vpadal", "s",
3787 int_arm_neon_vpadals>;
3788 defm VPADALu : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01101, 0, "vpadal", "u",
3789 int_arm_neon_vpadalu>;
3791 // VPMAX : Vector Pairwise Maximum
3792 def VPMAXs8 : N3VDInt<0, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3793 "s8", v8i8, v8i8, int_arm_neon_vpmaxs, 0>;
3794 def VPMAXs16 : N3VDInt<0, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3795 "s16", v4i16, v4i16, int_arm_neon_vpmaxs, 0>;
3796 def VPMAXs32 : N3VDInt<0, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3797 "s32", v2i32, v2i32, int_arm_neon_vpmaxs, 0>;
3798 def VPMAXu8 : N3VDInt<1, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3799 "u8", v8i8, v8i8, int_arm_neon_vpmaxu, 0>;
3800 def VPMAXu16 : N3VDInt<1, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3801 "u16", v4i16, v4i16, int_arm_neon_vpmaxu, 0>;
3802 def VPMAXu32 : N3VDInt<1, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3803 "u32", v2i32, v2i32, int_arm_neon_vpmaxu, 0>;
3804 def VPMAXf : N3VDInt<1, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmax",
3805 "f32", v2f32, v2f32, int_arm_neon_vpmaxs, 0>;
3807 // VPMIN : Vector Pairwise Minimum
3808 def VPMINs8 : N3VDInt<0, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3809 "s8", v8i8, v8i8, int_arm_neon_vpmins, 0>;
3810 def VPMINs16 : N3VDInt<0, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3811 "s16", v4i16, v4i16, int_arm_neon_vpmins, 0>;
3812 def VPMINs32 : N3VDInt<0, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3813 "s32", v2i32, v2i32, int_arm_neon_vpmins, 0>;
3814 def VPMINu8 : N3VDInt<1, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3815 "u8", v8i8, v8i8, int_arm_neon_vpminu, 0>;
3816 def VPMINu16 : N3VDInt<1, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3817 "u16", v4i16, v4i16, int_arm_neon_vpminu, 0>;
3818 def VPMINu32 : N3VDInt<1, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3819 "u32", v2i32, v2i32, int_arm_neon_vpminu, 0>;
3820 def VPMINf : N3VDInt<1, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmin",
3821 "f32", v2f32, v2f32, int_arm_neon_vpmins, 0>;
3823 // Vector Reciprocal and Reciprocal Square Root Estimate and Step.
3825 // VRECPE : Vector Reciprocal Estimate
3826 def VRECPEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
3827 IIC_VUNAD, "vrecpe", "u32",
3828 v2i32, v2i32, int_arm_neon_vrecpe>;
3829 def VRECPEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
3830 IIC_VUNAQ, "vrecpe", "u32",
3831 v4i32, v4i32, int_arm_neon_vrecpe>;
3832 def VRECPEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
3833 IIC_VUNAD, "vrecpe", "f32",
3834 v2f32, v2f32, int_arm_neon_vrecpe>;
3835 def VRECPEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
3836 IIC_VUNAQ, "vrecpe", "f32",
3837 v4f32, v4f32, int_arm_neon_vrecpe>;
3839 // VRECPS : Vector Reciprocal Step
3840 def VRECPSfd : N3VDInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
3841 IIC_VRECSD, "vrecps", "f32",
3842 v2f32, v2f32, int_arm_neon_vrecps, 1>;
3843 def VRECPSfq : N3VQInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
3844 IIC_VRECSQ, "vrecps", "f32",
3845 v4f32, v4f32, int_arm_neon_vrecps, 1>;
3847 // VRSQRTE : Vector Reciprocal Square Root Estimate
3848 def VRSQRTEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
3849 IIC_VUNAD, "vrsqrte", "u32",
3850 v2i32, v2i32, int_arm_neon_vrsqrte>;
3851 def VRSQRTEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
3852 IIC_VUNAQ, "vrsqrte", "u32",
3853 v4i32, v4i32, int_arm_neon_vrsqrte>;
3854 def VRSQRTEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
3855 IIC_VUNAD, "vrsqrte", "f32",
3856 v2f32, v2f32, int_arm_neon_vrsqrte>;
3857 def VRSQRTEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
3858 IIC_VUNAQ, "vrsqrte", "f32",
3859 v4f32, v4f32, int_arm_neon_vrsqrte>;
3861 // VRSQRTS : Vector Reciprocal Square Root Step
3862 def VRSQRTSfd : N3VDInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
3863 IIC_VRECSD, "vrsqrts", "f32",
3864 v2f32, v2f32, int_arm_neon_vrsqrts, 1>;
3865 def VRSQRTSfq : N3VQInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
3866 IIC_VRECSQ, "vrsqrts", "f32",
3867 v4f32, v4f32, int_arm_neon_vrsqrts, 1>;
3871 // VSHL : Vector Shift
3872 defm VSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 0, N3RegVShFrm,
3873 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
3874 "vshl", "s", int_arm_neon_vshifts>;
3875 defm VSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 0, N3RegVShFrm,
3876 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
3877 "vshl", "u", int_arm_neon_vshiftu>;
3878 // VSHL : Vector Shift Left (Immediate)
3879 defm VSHLi : N2VSh_QHSD<0, 1, 0b0101, 1, IIC_VSHLiD, "vshl", "i", NEONvshl,
3881 // VSHR : Vector Shift Right (Immediate)
3882 defm VSHRs : N2VSh_QHSD<0, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "s", NEONvshrs,
3884 defm VSHRu : N2VSh_QHSD<1, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "u", NEONvshru,
3887 // VSHLL : Vector Shift Left Long
3888 defm VSHLLs : N2VLSh_QHS<0, 1, 0b1010, 0, 0, 1, "vshll", "s", NEONvshlls>;
3889 defm VSHLLu : N2VLSh_QHS<1, 1, 0b1010, 0, 0, 1, "vshll", "u", NEONvshllu>;
3891 // VSHLL : Vector Shift Left Long (with maximum shift count)
3892 class N2VLShMax<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
3893 bit op6, bit op4, string OpcodeStr, string Dt, ValueType ResTy,
3894 ValueType OpTy, SDNode OpNode>
3895 : N2VLSh<op24, op23, op11_8, op7, op6, op4, OpcodeStr, Dt,
3896 ResTy, OpTy, OpNode> {
3897 let Inst{21-16} = op21_16;
3899 def VSHLLi8 : N2VLShMax<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll", "i8",
3900 v8i16, v8i8, NEONvshlli>;
3901 def VSHLLi16 : N2VLShMax<1, 1, 0b110110, 0b0011, 0, 0, 0, "vshll", "i16",
3902 v4i32, v4i16, NEONvshlli>;
3903 def VSHLLi32 : N2VLShMax<1, 1, 0b111010, 0b0011, 0, 0, 0, "vshll", "i32",
3904 v2i64, v2i32, NEONvshlli>;
3906 // VSHRN : Vector Shift Right and Narrow
3907 defm VSHRN : N2VNSh_HSD<0,1,0b1000,0,0,1, IIC_VSHLiD, "vshrn", "i",
3910 // VRSHL : Vector Rounding Shift
3911 defm VRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 0, N3RegVShFrm,
3912 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3913 "vrshl", "s", int_arm_neon_vrshifts>;
3914 defm VRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 0, N3RegVShFrm,
3915 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3916 "vrshl", "u", int_arm_neon_vrshiftu>;
3917 // VRSHR : Vector Rounding Shift Right
3918 defm VRSHRs : N2VSh_QHSD<0,1,0b0010,1, IIC_VSHLi4D, "vrshr", "s", NEONvrshrs,
3920 defm VRSHRu : N2VSh_QHSD<1,1,0b0010,1, IIC_VSHLi4D, "vrshr", "u", NEONvrshru,
3923 // VRSHRN : Vector Rounding Shift Right and Narrow
3924 defm VRSHRN : N2VNSh_HSD<0, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vrshrn", "i",
3927 // VQSHL : Vector Saturating Shift
3928 defm VQSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 1, N3RegVShFrm,
3929 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3930 "vqshl", "s", int_arm_neon_vqshifts>;
3931 defm VQSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 1, N3RegVShFrm,
3932 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3933 "vqshl", "u", int_arm_neon_vqshiftu>;
3934 // VQSHL : Vector Saturating Shift Left (Immediate)
3935 defm VQSHLsi : N2VSh_QHSD<0,1,0b0111,1, IIC_VSHLi4D, "vqshl", "s",NEONvqshls,
3937 defm VQSHLui : N2VSh_QHSD<1,1,0b0111,1, IIC_VSHLi4D, "vqshl", "u",NEONvqshlu,
3939 // VQSHLU : Vector Saturating Shift Left (Immediate, Unsigned)
3940 defm VQSHLsu : N2VSh_QHSD<1,1,0b0110,1, IIC_VSHLi4D,"vqshlu","s",NEONvqshlsu,
3943 // VQSHRN : Vector Saturating Shift Right and Narrow
3944 defm VQSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "s",
3946 defm VQSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "u",
3949 // VQSHRUN : Vector Saturating Shift Right and Narrow (Unsigned)
3950 defm VQSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 0, 1, IIC_VSHLi4D, "vqshrun", "s",
3953 // VQRSHL : Vector Saturating Rounding Shift
3954 defm VQRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 1, N3RegVShFrm,
3955 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3956 "vqrshl", "s", int_arm_neon_vqrshifts>;
3957 defm VQRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 1, N3RegVShFrm,
3958 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3959 "vqrshl", "u", int_arm_neon_vqrshiftu>;
3961 // VQRSHRN : Vector Saturating Rounding Shift Right and Narrow
3962 defm VQRSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "s",
3964 defm VQRSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "u",
3967 // VQRSHRUN : Vector Saturating Rounding Shift Right and Narrow (Unsigned)
3968 defm VQRSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vqrshrun", "s",
3971 // VSRA : Vector Shift Right and Accumulate
3972 defm VSRAs : N2VShAdd_QHSD<0, 1, 0b0001, 1, "vsra", "s", NEONvshrs>;
3973 defm VSRAu : N2VShAdd_QHSD<1, 1, 0b0001, 1, "vsra", "u", NEONvshru>;
3974 // VRSRA : Vector Rounding Shift Right and Accumulate
3975 defm VRSRAs : N2VShAdd_QHSD<0, 1, 0b0011, 1, "vrsra", "s", NEONvrshrs>;
3976 defm VRSRAu : N2VShAdd_QHSD<1, 1, 0b0011, 1, "vrsra", "u", NEONvrshru>;
3978 // VSLI : Vector Shift Left and Insert
3979 defm VSLI : N2VShIns_QHSD<1, 1, 0b0101, 1, "vsli", NEONvsli, N2RegVShLFrm>;
3980 // VSRI : Vector Shift Right and Insert
3981 defm VSRI : N2VShIns_QHSD<1, 1, 0b0100, 1, "vsri", NEONvsri, N2RegVShRFrm>;
3983 // Vector Absolute and Saturating Absolute.
3985 // VABS : Vector Absolute Value
3986 defm VABS : N2VInt_QHS<0b11, 0b11, 0b01, 0b00110, 0,
3987 IIC_VUNAiD, IIC_VUNAiQ, "vabs", "s",
3989 def VABSfd : N2VDInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
3990 IIC_VUNAD, "vabs", "f32",
3991 v2f32, v2f32, int_arm_neon_vabs>;
3992 def VABSfq : N2VQInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
3993 IIC_VUNAQ, "vabs", "f32",
3994 v4f32, v4f32, int_arm_neon_vabs>;
3996 // VQABS : Vector Saturating Absolute Value
3997 defm VQABS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01110, 0,
3998 IIC_VQUNAiD, IIC_VQUNAiQ, "vqabs", "s",
3999 int_arm_neon_vqabs>;
4003 def vnegd : PatFrag<(ops node:$in),
4004 (sub (bitconvert (v2i32 NEONimmAllZerosV)), node:$in)>;
4005 def vnegq : PatFrag<(ops node:$in),
4006 (sub (bitconvert (v4i32 NEONimmAllZerosV)), node:$in)>;
4008 class VNEGD<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
4009 : N2V<0b11, 0b11, size, 0b01, 0b00111, 0, 0, (outs DPR:$Vd), (ins DPR:$Vm),
4010 IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
4011 [(set DPR:$Vd, (Ty (vnegd DPR:$Vm)))]>;
4012 class VNEGQ<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
4013 : N2V<0b11, 0b11, size, 0b01, 0b00111, 1, 0, (outs QPR:$Vd), (ins QPR:$Vm),
4014 IIC_VSHLiQ, OpcodeStr, Dt, "$Vd, $Vm", "",
4015 [(set QPR:$Vd, (Ty (vnegq QPR:$Vm)))]>;
4017 // VNEG : Vector Negate (integer)
4018 def VNEGs8d : VNEGD<0b00, "vneg", "s8", v8i8>;
4019 def VNEGs16d : VNEGD<0b01, "vneg", "s16", v4i16>;
4020 def VNEGs32d : VNEGD<0b10, "vneg", "s32", v2i32>;
4021 def VNEGs8q : VNEGQ<0b00, "vneg", "s8", v16i8>;
4022 def VNEGs16q : VNEGQ<0b01, "vneg", "s16", v8i16>;
4023 def VNEGs32q : VNEGQ<0b10, "vneg", "s32", v4i32>;
4025 // VNEG : Vector Negate (floating-point)
4026 def VNEGfd : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
4027 (outs DPR:$Vd), (ins DPR:$Vm), IIC_VUNAD,
4028 "vneg", "f32", "$Vd, $Vm", "",
4029 [(set DPR:$Vd, (v2f32 (fneg DPR:$Vm)))]>;
4030 def VNEGf32q : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 1, 0,
4031 (outs QPR:$Vd), (ins QPR:$Vm), IIC_VUNAQ,
4032 "vneg", "f32", "$Vd, $Vm", "",
4033 [(set QPR:$Vd, (v4f32 (fneg QPR:$Vm)))]>;
4035 def : Pat<(v8i8 (vnegd DPR:$src)), (VNEGs8d DPR:$src)>;
4036 def : Pat<(v4i16 (vnegd DPR:$src)), (VNEGs16d DPR:$src)>;
4037 def : Pat<(v2i32 (vnegd DPR:$src)), (VNEGs32d DPR:$src)>;
4038 def : Pat<(v16i8 (vnegq QPR:$src)), (VNEGs8q QPR:$src)>;
4039 def : Pat<(v8i16 (vnegq QPR:$src)), (VNEGs16q QPR:$src)>;
4040 def : Pat<(v4i32 (vnegq QPR:$src)), (VNEGs32q QPR:$src)>;
4042 // VQNEG : Vector Saturating Negate
4043 defm VQNEG : N2VInt_QHS<0b11, 0b11, 0b00, 0b01111, 0,
4044 IIC_VQUNAiD, IIC_VQUNAiQ, "vqneg", "s",
4045 int_arm_neon_vqneg>;
4047 // Vector Bit Counting Operations.
4049 // VCLS : Vector Count Leading Sign Bits
4050 defm VCLS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01000, 0,
4051 IIC_VCNTiD, IIC_VCNTiQ, "vcls", "s",
4053 // VCLZ : Vector Count Leading Zeros
4054 defm VCLZ : N2VInt_QHS<0b11, 0b11, 0b00, 0b01001, 0,
4055 IIC_VCNTiD, IIC_VCNTiQ, "vclz", "i",
4057 // VCNT : Vector Count One Bits
4058 def VCNTd : N2VDInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
4059 IIC_VCNTiD, "vcnt", "8",
4060 v8i8, v8i8, int_arm_neon_vcnt>;
4061 def VCNTq : N2VQInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
4062 IIC_VCNTiQ, "vcnt", "8",
4063 v16i8, v16i8, int_arm_neon_vcnt>;
4065 // Vector Swap -- for disassembly only.
4066 def VSWPd : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 0, 0,
4067 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
4068 "vswp", "$Vd, $Vm", "", []>;
4069 def VSWPq : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 1, 0,
4070 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
4071 "vswp", "$Vd, $Vm", "", []>;
4073 // Vector Move Operations.
4075 // VMOV : Vector Move (Register)
4077 let neverHasSideEffects = 1 in {
4078 def VMOVDneon: N3VX<0, 0, 0b10, 0b0001, 0, 1, (outs DPR:$Vd), (ins DPR:$Vm),
4079 N3RegFrm, IIC_VMOV, "vmov", "$Vd, $Vm", "", []> {
4080 let Vn{4-0} = Vm{4-0};
4082 def VMOVQ : N3VX<0, 0, 0b10, 0b0001, 1, 1, (outs QPR:$Vd), (ins QPR:$Vm),
4083 N3RegFrm, IIC_VMOV, "vmov", "$Vd, $Vm", "", []> {
4084 let Vn{4-0} = Vm{4-0};
4087 // Pseudo vector move instructions for QQ and QQQQ registers. This should
4088 // be expanded after register allocation is completed.
4089 def VMOVQQ : PseudoInst<(outs QQPR:$dst), (ins QQPR:$src),
4092 def VMOVQQQQ : PseudoInst<(outs QQQQPR:$dst), (ins QQQQPR:$src),
4094 } // neverHasSideEffects
4096 // VMOV : Vector Move (Immediate)
4098 let isReMaterializable = 1 in {
4099 def VMOVv8i8 : N1ModImm<1, 0b000, 0b1110, 0, 0, 0, 1, (outs DPR:$Vd),
4100 (ins nModImm:$SIMM), IIC_VMOVImm,
4101 "vmov", "i8", "$Vd, $SIMM", "",
4102 [(set DPR:$Vd, (v8i8 (NEONvmovImm timm:$SIMM)))]>;
4103 def VMOVv16i8 : N1ModImm<1, 0b000, 0b1110, 0, 1, 0, 1, (outs QPR:$Vd),
4104 (ins nModImm:$SIMM), IIC_VMOVImm,
4105 "vmov", "i8", "$Vd, $SIMM", "",
4106 [(set QPR:$Vd, (v16i8 (NEONvmovImm timm:$SIMM)))]>;
4108 def VMOVv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 0, 1, (outs DPR:$Vd),
4109 (ins nModImm:$SIMM), IIC_VMOVImm,
4110 "vmov", "i16", "$Vd, $SIMM", "",
4111 [(set DPR:$Vd, (v4i16 (NEONvmovImm timm:$SIMM)))]> {
4112 let Inst{9} = SIMM{9};
4115 def VMOVv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 0, 1, (outs QPR:$Vd),
4116 (ins nModImm:$SIMM), IIC_VMOVImm,
4117 "vmov", "i16", "$Vd, $SIMM", "",
4118 [(set QPR:$Vd, (v8i16 (NEONvmovImm timm:$SIMM)))]> {
4119 let Inst{9} = SIMM{9};
4122 def VMOVv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 0, 1, (outs DPR:$Vd),
4123 (ins nModImm:$SIMM), IIC_VMOVImm,
4124 "vmov", "i32", "$Vd, $SIMM", "",
4125 [(set DPR:$Vd, (v2i32 (NEONvmovImm timm:$SIMM)))]> {
4126 let Inst{11-8} = SIMM{11-8};
4129 def VMOVv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 0, 1, (outs QPR:$Vd),
4130 (ins nModImm:$SIMM), IIC_VMOVImm,
4131 "vmov", "i32", "$Vd, $SIMM", "",
4132 [(set QPR:$Vd, (v4i32 (NEONvmovImm timm:$SIMM)))]> {
4133 let Inst{11-8} = SIMM{11-8};
4136 def VMOVv1i64 : N1ModImm<1, 0b000, 0b1110, 0, 0, 1, 1, (outs DPR:$Vd),
4137 (ins nModImm:$SIMM), IIC_VMOVImm,
4138 "vmov", "i64", "$Vd, $SIMM", "",
4139 [(set DPR:$Vd, (v1i64 (NEONvmovImm timm:$SIMM)))]>;
4140 def VMOVv2i64 : N1ModImm<1, 0b000, 0b1110, 0, 1, 1, 1, (outs QPR:$Vd),
4141 (ins nModImm:$SIMM), IIC_VMOVImm,
4142 "vmov", "i64", "$Vd, $SIMM", "",
4143 [(set QPR:$Vd, (v2i64 (NEONvmovImm timm:$SIMM)))]>;
4144 } // isReMaterializable
4146 // VMOV : Vector Get Lane (move scalar to ARM core register)
4148 def VGETLNs8 : NVGetLane<{1,1,1,0,0,1,?,1}, 0b1011, {?,?},
4149 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
4150 IIC_VMOVSI, "vmov", "s8", "$R, $V[$lane]",
4151 [(set GPR:$R, (NEONvgetlanes (v8i8 DPR:$V),
4153 let Inst{21} = lane{2};
4154 let Inst{6-5} = lane{1-0};
4156 def VGETLNs16 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, {?,1},
4157 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
4158 IIC_VMOVSI, "vmov", "s16", "$R, $V[$lane]",
4159 [(set GPR:$R, (NEONvgetlanes (v4i16 DPR:$V),
4161 let Inst{21} = lane{1};
4162 let Inst{6} = lane{0};
4164 def VGETLNu8 : NVGetLane<{1,1,1,0,1,1,?,1}, 0b1011, {?,?},
4165 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
4166 IIC_VMOVSI, "vmov", "u8", "$R, $V[$lane]",
4167 [(set GPR:$R, (NEONvgetlaneu (v8i8 DPR:$V),
4169 let Inst{21} = lane{2};
4170 let Inst{6-5} = lane{1-0};
4172 def VGETLNu16 : NVGetLane<{1,1,1,0,1,0,?,1}, 0b1011, {?,1},
4173 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
4174 IIC_VMOVSI, "vmov", "u16", "$R, $V[$lane]",
4175 [(set GPR:$R, (NEONvgetlaneu (v4i16 DPR:$V),
4177 let Inst{21} = lane{1};
4178 let Inst{6} = lane{0};
4180 def VGETLNi32 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, 0b00,
4181 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
4182 IIC_VMOVSI, "vmov", "32", "$R, $V[$lane]",
4183 [(set GPR:$R, (extractelt (v2i32 DPR:$V),
4185 let Inst{21} = lane{0};
4187 // def VGETLNf32: see FMRDH and FMRDL in ARMInstrVFP.td
4188 def : Pat<(NEONvgetlanes (v16i8 QPR:$src), imm:$lane),
4189 (VGETLNs8 (v8i8 (EXTRACT_SUBREG QPR:$src,
4190 (DSubReg_i8_reg imm:$lane))),
4191 (SubReg_i8_lane imm:$lane))>;
4192 def : Pat<(NEONvgetlanes (v8i16 QPR:$src), imm:$lane),
4193 (VGETLNs16 (v4i16 (EXTRACT_SUBREG QPR:$src,
4194 (DSubReg_i16_reg imm:$lane))),
4195 (SubReg_i16_lane imm:$lane))>;
4196 def : Pat<(NEONvgetlaneu (v16i8 QPR:$src), imm:$lane),
4197 (VGETLNu8 (v8i8 (EXTRACT_SUBREG QPR:$src,
4198 (DSubReg_i8_reg imm:$lane))),
4199 (SubReg_i8_lane imm:$lane))>;
4200 def : Pat<(NEONvgetlaneu (v8i16 QPR:$src), imm:$lane),
4201 (VGETLNu16 (v4i16 (EXTRACT_SUBREG QPR:$src,
4202 (DSubReg_i16_reg imm:$lane))),
4203 (SubReg_i16_lane imm:$lane))>;
4204 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
4205 (VGETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src,
4206 (DSubReg_i32_reg imm:$lane))),
4207 (SubReg_i32_lane imm:$lane))>;
4208 def : Pat<(extractelt (v2f32 DPR:$src1), imm:$src2),
4209 (EXTRACT_SUBREG (v2f32 (COPY_TO_REGCLASS (v2f32 DPR:$src1),DPR_VFP2)),
4210 (SSubReg_f32_reg imm:$src2))>;
4211 def : Pat<(extractelt (v4f32 QPR:$src1), imm:$src2),
4212 (EXTRACT_SUBREG (v4f32 (COPY_TO_REGCLASS (v4f32 QPR:$src1),QPR_VFP2)),
4213 (SSubReg_f32_reg imm:$src2))>;
4214 //def : Pat<(extractelt (v2i64 QPR:$src1), imm:$src2),
4215 // (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
4216 def : Pat<(extractelt (v2f64 QPR:$src1), imm:$src2),
4217 (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
4220 // VMOV : Vector Set Lane (move ARM core register to scalar)
4222 let Constraints = "$src1 = $V" in {
4223 def VSETLNi8 : NVSetLane<{1,1,1,0,0,1,?,0}, 0b1011, {?,?}, (outs DPR:$V),
4224 (ins DPR:$src1, GPR:$R, nohash_imm:$lane),
4225 IIC_VMOVISL, "vmov", "8", "$V[$lane], $R",
4226 [(set DPR:$V, (vector_insert (v8i8 DPR:$src1),
4227 GPR:$R, imm:$lane))]> {
4228 let Inst{21} = lane{2};
4229 let Inst{6-5} = lane{1-0};
4231 def VSETLNi16 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, {?,1}, (outs DPR:$V),
4232 (ins DPR:$src1, GPR:$R, nohash_imm:$lane),
4233 IIC_VMOVISL, "vmov", "16", "$V[$lane], $R",
4234 [(set DPR:$V, (vector_insert (v4i16 DPR:$src1),
4235 GPR:$R, imm:$lane))]> {
4236 let Inst{21} = lane{1};
4237 let Inst{6} = lane{0};
4239 def VSETLNi32 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, 0b00, (outs DPR:$V),
4240 (ins DPR:$src1, GPR:$R, nohash_imm:$lane),
4241 IIC_VMOVISL, "vmov", "32", "$V[$lane], $R",
4242 [(set DPR:$V, (insertelt (v2i32 DPR:$src1),
4243 GPR:$R, imm:$lane))]> {
4244 let Inst{21} = lane{0};
4247 def : Pat<(vector_insert (v16i8 QPR:$src1), GPR:$src2, imm:$lane),
4248 (v16i8 (INSERT_SUBREG QPR:$src1,
4249 (v8i8 (VSETLNi8 (v8i8 (EXTRACT_SUBREG QPR:$src1,
4250 (DSubReg_i8_reg imm:$lane))),
4251 GPR:$src2, (SubReg_i8_lane imm:$lane))),
4252 (DSubReg_i8_reg imm:$lane)))>;
4253 def : Pat<(vector_insert (v8i16 QPR:$src1), GPR:$src2, imm:$lane),
4254 (v8i16 (INSERT_SUBREG QPR:$src1,
4255 (v4i16 (VSETLNi16 (v4i16 (EXTRACT_SUBREG QPR:$src1,
4256 (DSubReg_i16_reg imm:$lane))),
4257 GPR:$src2, (SubReg_i16_lane imm:$lane))),
4258 (DSubReg_i16_reg imm:$lane)))>;
4259 def : Pat<(insertelt (v4i32 QPR:$src1), GPR:$src2, imm:$lane),
4260 (v4i32 (INSERT_SUBREG QPR:$src1,
4261 (v2i32 (VSETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src1,
4262 (DSubReg_i32_reg imm:$lane))),
4263 GPR:$src2, (SubReg_i32_lane imm:$lane))),
4264 (DSubReg_i32_reg imm:$lane)))>;
4266 def : Pat<(v2f32 (insertelt DPR:$src1, SPR:$src2, imm:$src3)),
4267 (INSERT_SUBREG (v2f32 (COPY_TO_REGCLASS DPR:$src1, DPR_VFP2)),
4268 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
4269 def : Pat<(v4f32 (insertelt QPR:$src1, SPR:$src2, imm:$src3)),
4270 (INSERT_SUBREG (v4f32 (COPY_TO_REGCLASS QPR:$src1, QPR_VFP2)),
4271 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
4273 //def : Pat<(v2i64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
4274 // (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
4275 def : Pat<(v2f64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
4276 (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
4278 def : Pat<(v2f32 (scalar_to_vector SPR:$src)),
4279 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
4280 def : Pat<(v2f64 (scalar_to_vector (f64 DPR:$src))),
4281 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
4282 def : Pat<(v4f32 (scalar_to_vector SPR:$src)),
4283 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
4285 def : Pat<(v8i8 (scalar_to_vector GPR:$src)),
4286 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4287 def : Pat<(v4i16 (scalar_to_vector GPR:$src)),
4288 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4289 def : Pat<(v2i32 (scalar_to_vector GPR:$src)),
4290 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4292 def : Pat<(v16i8 (scalar_to_vector GPR:$src)),
4293 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4294 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4296 def : Pat<(v8i16 (scalar_to_vector GPR:$src)),
4297 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
4298 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4300 def : Pat<(v4i32 (scalar_to_vector GPR:$src)),
4301 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
4302 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4305 // VDUP : Vector Duplicate (from ARM core register to all elements)
4307 class VDUPD<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
4308 : NVDup<opcod1, 0b1011, opcod3, (outs DPR:$V), (ins GPR:$R),
4309 IIC_VMOVIS, "vdup", Dt, "$V, $R",
4310 [(set DPR:$V, (Ty (NEONvdup (i32 GPR:$R))))]>;
4311 class VDUPQ<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
4312 : NVDup<opcod1, 0b1011, opcod3, (outs QPR:$V), (ins GPR:$R),
4313 IIC_VMOVIS, "vdup", Dt, "$V, $R",
4314 [(set QPR:$V, (Ty (NEONvdup (i32 GPR:$R))))]>;
4316 def VDUP8d : VDUPD<0b11101100, 0b00, "8", v8i8>;
4317 def VDUP16d : VDUPD<0b11101000, 0b01, "16", v4i16>;
4318 def VDUP32d : VDUPD<0b11101000, 0b00, "32", v2i32>;
4319 def VDUP8q : VDUPQ<0b11101110, 0b00, "8", v16i8>;
4320 def VDUP16q : VDUPQ<0b11101010, 0b01, "16", v8i16>;
4321 def VDUP32q : VDUPQ<0b11101010, 0b00, "32", v4i32>;
4323 def VDUPfd : NVDup<0b11101000, 0b1011, 0b00, (outs DPR:$V), (ins GPR:$R),
4324 IIC_VMOVIS, "vdup", "32", "$V, $R",
4325 [(set DPR:$V, (v2f32 (NEONvdup
4326 (f32 (bitconvert GPR:$R)))))]>;
4327 def VDUPfq : NVDup<0b11101010, 0b1011, 0b00, (outs QPR:$V), (ins GPR:$R),
4328 IIC_VMOVIS, "vdup", "32", "$V, $R",
4329 [(set QPR:$V, (v4f32 (NEONvdup
4330 (f32 (bitconvert GPR:$R)))))]>;
4332 // VDUP : Vector Duplicate Lane (from scalar to all elements)
4334 class VDUPLND<bits<4> op19_16, string OpcodeStr, string Dt,
4336 : NVDupLane<op19_16, 0, (outs DPR:$Vd), (ins DPR:$Vm, nohash_imm:$lane),
4337 IIC_VMOVD, OpcodeStr, Dt, "$Vd, $Vm[$lane]",
4338 [(set DPR:$Vd, (Ty (NEONvduplane (Ty DPR:$Vm), imm:$lane)))]>;
4340 class VDUPLNQ<bits<4> op19_16, string OpcodeStr, string Dt,
4341 ValueType ResTy, ValueType OpTy>
4342 : NVDupLane<op19_16, 1, (outs QPR:$Vd), (ins DPR:$Vm, nohash_imm:$lane),
4343 IIC_VMOVQ, OpcodeStr, Dt, "$Vd, $Vm[$lane]",
4344 [(set QPR:$Vd, (ResTy (NEONvduplane (OpTy DPR:$Vm),
4347 // Inst{19-16} is partially specified depending on the element size.
4349 def VDUPLN8d : VDUPLND<{?,?,?,1}, "vdup", "8", v8i8> {
4350 let Inst{19-17} = lane{2-0};
4352 def VDUPLN16d : VDUPLND<{?,?,1,0}, "vdup", "16", v4i16> {
4353 let Inst{19-18} = lane{1-0};
4355 def VDUPLN32d : VDUPLND<{?,1,0,0}, "vdup", "32", v2i32> {
4356 let Inst{19} = lane{0};
4358 def VDUPLNfd : VDUPLND<{?,1,0,0}, "vdup", "32", v2f32> {
4359 let Inst{19} = lane{0};
4361 def VDUPLN8q : VDUPLNQ<{?,?,?,1}, "vdup", "8", v16i8, v8i8> {
4362 let Inst{19-17} = lane{2-0};
4364 def VDUPLN16q : VDUPLNQ<{?,?,1,0}, "vdup", "16", v8i16, v4i16> {
4365 let Inst{19-18} = lane{1-0};
4367 def VDUPLN32q : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4i32, v2i32> {
4368 let Inst{19} = lane{0};
4370 def VDUPLNfq : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4f32, v2f32> {
4371 let Inst{19} = lane{0};
4374 def : Pat<(v16i8 (NEONvduplane (v16i8 QPR:$src), imm:$lane)),
4375 (v16i8 (VDUPLN8q (v8i8 (EXTRACT_SUBREG QPR:$src,
4376 (DSubReg_i8_reg imm:$lane))),
4377 (SubReg_i8_lane imm:$lane)))>;
4378 def : Pat<(v8i16 (NEONvduplane (v8i16 QPR:$src), imm:$lane)),
4379 (v8i16 (VDUPLN16q (v4i16 (EXTRACT_SUBREG QPR:$src,
4380 (DSubReg_i16_reg imm:$lane))),
4381 (SubReg_i16_lane imm:$lane)))>;
4382 def : Pat<(v4i32 (NEONvduplane (v4i32 QPR:$src), imm:$lane)),
4383 (v4i32 (VDUPLN32q (v2i32 (EXTRACT_SUBREG QPR:$src,
4384 (DSubReg_i32_reg imm:$lane))),
4385 (SubReg_i32_lane imm:$lane)))>;
4386 def : Pat<(v4f32 (NEONvduplane (v4f32 QPR:$src), imm:$lane)),
4387 (v4f32 (VDUPLNfq (v2f32 (EXTRACT_SUBREG QPR:$src,
4388 (DSubReg_i32_reg imm:$lane))),
4389 (SubReg_i32_lane imm:$lane)))>;
4391 def VDUPfdf : PseudoNeonI<(outs DPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
4392 [(set DPR:$dst, (v2f32 (NEONvdup (f32 SPR:$src))))]>;
4393 def VDUPfqf : PseudoNeonI<(outs QPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
4394 [(set QPR:$dst, (v4f32 (NEONvdup (f32 SPR:$src))))]>;
4396 // VMOVN : Vector Narrowing Move
4397 defm VMOVN : N2VN_HSD<0b11,0b11,0b10,0b00100,0,0, IIC_VMOVN,
4398 "vmovn", "i", trunc>;
4399 // VQMOVN : Vector Saturating Narrowing Move
4400 defm VQMOVNs : N2VNInt_HSD<0b11,0b11,0b10,0b00101,0,0, IIC_VQUNAiD,
4401 "vqmovn", "s", int_arm_neon_vqmovns>;
4402 defm VQMOVNu : N2VNInt_HSD<0b11,0b11,0b10,0b00101,1,0, IIC_VQUNAiD,
4403 "vqmovn", "u", int_arm_neon_vqmovnu>;
4404 defm VQMOVNsu : N2VNInt_HSD<0b11,0b11,0b10,0b00100,1,0, IIC_VQUNAiD,
4405 "vqmovun", "s", int_arm_neon_vqmovnsu>;
4406 // VMOVL : Vector Lengthening Move
4407 defm VMOVLs : N2VL_QHS<0b01,0b10100,0,1, "vmovl", "s", sext>;
4408 defm VMOVLu : N2VL_QHS<0b11,0b10100,0,1, "vmovl", "u", zext>;
4410 // Vector Conversions.
4412 // VCVT : Vector Convert Between Floating-Point and Integers
4413 def VCVTf2sd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
4414 v2i32, v2f32, fp_to_sint>;
4415 def VCVTf2ud : N2VD<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
4416 v2i32, v2f32, fp_to_uint>;
4417 def VCVTs2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
4418 v2f32, v2i32, sint_to_fp>;
4419 def VCVTu2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
4420 v2f32, v2i32, uint_to_fp>;
4422 def VCVTf2sq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
4423 v4i32, v4f32, fp_to_sint>;
4424 def VCVTf2uq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
4425 v4i32, v4f32, fp_to_uint>;
4426 def VCVTs2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
4427 v4f32, v4i32, sint_to_fp>;
4428 def VCVTu2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
4429 v4f32, v4i32, uint_to_fp>;
4431 // VCVT : Vector Convert Between Floating-Point and Fixed-Point.
4432 def VCVTf2xsd : N2VCvtD<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
4433 v2i32, v2f32, int_arm_neon_vcvtfp2fxs>;
4434 def VCVTf2xud : N2VCvtD<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
4435 v2i32, v2f32, int_arm_neon_vcvtfp2fxu>;
4436 def VCVTxs2fd : N2VCvtD<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
4437 v2f32, v2i32, int_arm_neon_vcvtfxs2fp>;
4438 def VCVTxu2fd : N2VCvtD<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
4439 v2f32, v2i32, int_arm_neon_vcvtfxu2fp>;
4441 def VCVTf2xsq : N2VCvtQ<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
4442 v4i32, v4f32, int_arm_neon_vcvtfp2fxs>;
4443 def VCVTf2xuq : N2VCvtQ<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
4444 v4i32, v4f32, int_arm_neon_vcvtfp2fxu>;
4445 def VCVTxs2fq : N2VCvtQ<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
4446 v4f32, v4i32, int_arm_neon_vcvtfxs2fp>;
4447 def VCVTxu2fq : N2VCvtQ<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
4448 v4f32, v4i32, int_arm_neon_vcvtfxu2fp>;
4452 // VREV64 : Vector Reverse elements within 64-bit doublewords
4454 class VREV64D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4455 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 0, 0, (outs DPR:$Vd),
4456 (ins DPR:$Vm), IIC_VMOVD,
4457 OpcodeStr, Dt, "$Vd, $Vm", "",
4458 [(set DPR:$Vd, (Ty (NEONvrev64 (Ty DPR:$Vm))))]>;
4459 class VREV64Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4460 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 1, 0, (outs QPR:$Vd),
4461 (ins QPR:$Vm), IIC_VMOVQ,
4462 OpcodeStr, Dt, "$Vd, $Vm", "",
4463 [(set QPR:$Vd, (Ty (NEONvrev64 (Ty QPR:$Vm))))]>;
4465 def VREV64d8 : VREV64D<0b00, "vrev64", "8", v8i8>;
4466 def VREV64d16 : VREV64D<0b01, "vrev64", "16", v4i16>;
4467 def VREV64d32 : VREV64D<0b10, "vrev64", "32", v2i32>;
4468 def VREV64df : VREV64D<0b10, "vrev64", "32", v2f32>;
4470 def VREV64q8 : VREV64Q<0b00, "vrev64", "8", v16i8>;
4471 def VREV64q16 : VREV64Q<0b01, "vrev64", "16", v8i16>;
4472 def VREV64q32 : VREV64Q<0b10, "vrev64", "32", v4i32>;
4473 def VREV64qf : VREV64Q<0b10, "vrev64", "32", v4f32>;
4475 // VREV32 : Vector Reverse elements within 32-bit words
4477 class VREV32D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4478 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 0, 0, (outs DPR:$Vd),
4479 (ins DPR:$Vm), IIC_VMOVD,
4480 OpcodeStr, Dt, "$Vd, $Vm", "",
4481 [(set DPR:$Vd, (Ty (NEONvrev32 (Ty DPR:$Vm))))]>;
4482 class VREV32Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4483 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 1, 0, (outs QPR:$Vd),
4484 (ins QPR:$Vm), IIC_VMOVQ,
4485 OpcodeStr, Dt, "$Vd, $Vm", "",
4486 [(set QPR:$Vd, (Ty (NEONvrev32 (Ty QPR:$Vm))))]>;
4488 def VREV32d8 : VREV32D<0b00, "vrev32", "8", v8i8>;
4489 def VREV32d16 : VREV32D<0b01, "vrev32", "16", v4i16>;
4491 def VREV32q8 : VREV32Q<0b00, "vrev32", "8", v16i8>;
4492 def VREV32q16 : VREV32Q<0b01, "vrev32", "16", v8i16>;
4494 // VREV16 : Vector Reverse elements within 16-bit halfwords
4496 class VREV16D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4497 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 0, 0, (outs DPR:$Vd),
4498 (ins DPR:$Vm), IIC_VMOVD,
4499 OpcodeStr, Dt, "$Vd, $Vm", "",
4500 [(set DPR:$Vd, (Ty (NEONvrev16 (Ty DPR:$Vm))))]>;
4501 class VREV16Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4502 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 1, 0, (outs QPR:$Vd),
4503 (ins QPR:$Vm), IIC_VMOVQ,
4504 OpcodeStr, Dt, "$Vd, $Vm", "",
4505 [(set QPR:$Vd, (Ty (NEONvrev16 (Ty QPR:$Vm))))]>;
4507 def VREV16d8 : VREV16D<0b00, "vrev16", "8", v8i8>;
4508 def VREV16q8 : VREV16Q<0b00, "vrev16", "8", v16i8>;
4510 // Other Vector Shuffles.
4512 // VEXT : Vector Extract
4514 class VEXTd<string OpcodeStr, string Dt, ValueType Ty>
4515 : N3V<0,1,0b11,{?,?,?,?},0,0, (outs DPR:$Vd),
4516 (ins DPR:$Vn, DPR:$Vm, i32imm:$index), NVExtFrm,
4517 IIC_VEXTD, OpcodeStr, Dt, "$Vd, $Vn, $Vm, $index", "",
4518 [(set DPR:$Vd, (Ty (NEONvext (Ty DPR:$Vn),
4519 (Ty DPR:$Vm), imm:$index)))]> {
4521 let Inst{11-8} = index{3-0};
4524 class VEXTq<string OpcodeStr, string Dt, ValueType Ty>
4525 : N3V<0,1,0b11,{?,?,?,?},1,0, (outs QPR:$Vd),
4526 (ins QPR:$Vn, QPR:$Vm, i32imm:$index), NVExtFrm,
4527 IIC_VEXTQ, OpcodeStr, Dt, "$Vd, $Vn, $Vm, $index", "",
4528 [(set QPR:$Vd, (Ty (NEONvext (Ty QPR:$Vn),
4529 (Ty QPR:$Vm), imm:$index)))]> {
4531 let Inst{11-8} = index{3-0};
4534 def VEXTd8 : VEXTd<"vext", "8", v8i8> {
4535 let Inst{11-8} = index{3-0};
4537 def VEXTd16 : VEXTd<"vext", "16", v4i16> {
4538 let Inst{11-9} = index{2-0};
4541 def VEXTd32 : VEXTd<"vext", "32", v2i32> {
4542 let Inst{11-10} = index{1-0};
4543 let Inst{9-8} = 0b00;
4545 def VEXTdf : VEXTd<"vext", "32", v2f32> {
4546 let Inst{11} = index{0};
4547 let Inst{10-8} = 0b000;
4550 def VEXTq8 : VEXTq<"vext", "8", v16i8> {
4551 let Inst{11-8} = index{3-0};
4553 def VEXTq16 : VEXTq<"vext", "16", v8i16> {
4554 let Inst{11-9} = index{2-0};
4557 def VEXTq32 : VEXTq<"vext", "32", v4i32> {
4558 let Inst{11-10} = index{1-0};
4559 let Inst{9-8} = 0b00;
4561 def VEXTqf : VEXTq<"vext", "32", v4f32> {
4562 let Inst{11} = index{0};
4563 let Inst{10-8} = 0b000;
4566 // VTRN : Vector Transpose
4568 def VTRNd8 : N2VDShuffle<0b00, 0b00001, "vtrn", "8">;
4569 def VTRNd16 : N2VDShuffle<0b01, 0b00001, "vtrn", "16">;
4570 def VTRNd32 : N2VDShuffle<0b10, 0b00001, "vtrn", "32">;
4572 def VTRNq8 : N2VQShuffle<0b00, 0b00001, IIC_VPERMQ, "vtrn", "8">;
4573 def VTRNq16 : N2VQShuffle<0b01, 0b00001, IIC_VPERMQ, "vtrn", "16">;
4574 def VTRNq32 : N2VQShuffle<0b10, 0b00001, IIC_VPERMQ, "vtrn", "32">;
4576 // VUZP : Vector Unzip (Deinterleave)
4578 def VUZPd8 : N2VDShuffle<0b00, 0b00010, "vuzp", "8">;
4579 def VUZPd16 : N2VDShuffle<0b01, 0b00010, "vuzp", "16">;
4580 def VUZPd32 : N2VDShuffle<0b10, 0b00010, "vuzp", "32">;
4582 def VUZPq8 : N2VQShuffle<0b00, 0b00010, IIC_VPERMQ3, "vuzp", "8">;
4583 def VUZPq16 : N2VQShuffle<0b01, 0b00010, IIC_VPERMQ3, "vuzp", "16">;
4584 def VUZPq32 : N2VQShuffle<0b10, 0b00010, IIC_VPERMQ3, "vuzp", "32">;
4586 // VZIP : Vector Zip (Interleave)
4588 def VZIPd8 : N2VDShuffle<0b00, 0b00011, "vzip", "8">;
4589 def VZIPd16 : N2VDShuffle<0b01, 0b00011, "vzip", "16">;
4590 def VZIPd32 : N2VDShuffle<0b10, 0b00011, "vzip", "32">;
4592 def VZIPq8 : N2VQShuffle<0b00, 0b00011, IIC_VPERMQ3, "vzip", "8">;
4593 def VZIPq16 : N2VQShuffle<0b01, 0b00011, IIC_VPERMQ3, "vzip", "16">;
4594 def VZIPq32 : N2VQShuffle<0b10, 0b00011, IIC_VPERMQ3, "vzip", "32">;
4596 // Vector Table Lookup and Table Extension.
4598 // VTBL : Vector Table Lookup
4600 : N3V<1,1,0b11,0b1000,0,0, (outs DPR:$Vd),
4601 (ins DPR:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTB1,
4602 "vtbl", "8", "$Vd, \\{$Vn\\}, $Vm", "",
4603 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbl1 DPR:$Vn, DPR:$Vm)))]>;
4604 let hasExtraSrcRegAllocReq = 1 in {
4606 : N3V<1,1,0b11,0b1001,0,0, (outs DPR:$Vd),
4607 (ins DPR:$Vn, DPR:$tbl2, DPR:$Vm), NVTBLFrm, IIC_VTB2,
4608 "vtbl", "8", "$Vd, \\{$Vn, $tbl2\\}, $Vm", "", []>;
4610 : N3V<1,1,0b11,0b1010,0,0, (outs DPR:$Vd),
4611 (ins DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$Vm), NVTBLFrm, IIC_VTB3,
4612 "vtbl", "8", "$Vd, \\{$Vn, $tbl2, $tbl3\\}, $Vm", "", []>;
4614 : N3V<1,1,0b11,0b1011,0,0, (outs DPR:$Vd),
4615 (ins DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$Vm),
4617 "vtbl", "8", "$Vd, \\{$Vn, $tbl2, $tbl3, $tbl4\\}, $Vm", "", []>;
4618 } // hasExtraSrcRegAllocReq = 1
4621 : PseudoNeonI<(outs DPR:$dst), (ins QPR:$tbl, DPR:$src), IIC_VTB2, "", []>;
4623 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB3, "", []>;
4625 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB4, "", []>;
4627 // VTBX : Vector Table Extension
4629 : N3V<1,1,0b11,0b1000,1,0, (outs DPR:$Vd),
4630 (ins DPR:$orig, DPR:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTBX1,
4631 "vtbx", "8", "$Vd, \\{$Vn\\}, $Vm", "$orig = $Vd",
4632 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbx1
4633 DPR:$orig, DPR:$Vn, DPR:$Vm)))]>;
4634 let hasExtraSrcRegAllocReq = 1 in {
4636 : N3V<1,1,0b11,0b1001,1,0, (outs DPR:$Vd),
4637 (ins DPR:$orig, DPR:$Vn, DPR:$tbl2, DPR:$Vm), NVTBLFrm, IIC_VTBX2,
4638 "vtbx", "8", "$Vd, \\{$Vn, $tbl2\\}, $Vm", "$orig = $Vd", []>;
4640 : N3V<1,1,0b11,0b1010,1,0, (outs DPR:$Vd),
4641 (ins DPR:$orig, DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$Vm),
4642 NVTBLFrm, IIC_VTBX3,
4643 "vtbx", "8", "$Vd, \\{$Vn, $tbl2, $tbl3\\}, $Vm",
4646 : N3V<1,1,0b11,0b1011,1,0, (outs DPR:$Vd), (ins DPR:$orig, DPR:$Vn,
4647 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$Vm), NVTBLFrm, IIC_VTBX4,
4648 "vtbx", "8", "$Vd, \\{$Vn, $tbl2, $tbl3, $tbl4\\}, $Vm",
4650 } // hasExtraSrcRegAllocReq = 1
4653 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QPR:$tbl, DPR:$src),
4654 IIC_VTBX2, "$orig = $dst", []>;
4656 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
4657 IIC_VTBX3, "$orig = $dst", []>;
4659 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
4660 IIC_VTBX4, "$orig = $dst", []>;
4662 //===----------------------------------------------------------------------===//
4663 // NEON instructions for single-precision FP math
4664 //===----------------------------------------------------------------------===//
4666 class N2VSPat<SDNode OpNode, NeonI Inst>
4667 : NEONFPPat<(f32 (OpNode SPR:$a)),
4669 (v2f32 (COPY_TO_REGCLASS (Inst
4671 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
4672 SPR:$a, ssub_0)), DPR_VFP2)), ssub_0)>;
4674 class N3VSPat<SDNode OpNode, NeonI Inst>
4675 : NEONFPPat<(f32 (OpNode SPR:$a, SPR:$b)),
4677 (v2f32 (COPY_TO_REGCLASS (Inst
4679 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
4682 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
4683 SPR:$b, ssub_0)), DPR_VFP2)), ssub_0)>;
4685 class N3VSMulOpPat<SDNode MulNode, SDNode OpNode, NeonI Inst>
4686 : NEONFPPat<(f32 (OpNode SPR:$acc, (f32 (MulNode SPR:$a, SPR:$b)))),
4688 (v2f32 (COPY_TO_REGCLASS (Inst
4690 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
4693 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
4696 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
4697 SPR:$b, ssub_0)), DPR_VFP2)), ssub_0)>;
4699 def : N3VSPat<fadd, VADDfd>;
4700 def : N3VSPat<fsub, VSUBfd>;
4701 def : N3VSPat<fmul, VMULfd>;
4702 def : N3VSMulOpPat<fmul, fadd, VMLAfd>,
4703 Requires<[HasNEON, UseNEONForFP, UseFPVMLx]>;
4704 def : N3VSMulOpPat<fmul, fsub, VMLSfd>,
4705 Requires<[HasNEON, UseNEONForFP, UseFPVMLx]>;
4706 def : N2VSPat<fabs, VABSfd>;
4707 def : N2VSPat<fneg, VNEGfd>;
4708 def : N3VSPat<NEONfmax, VMAXfd>;
4709 def : N3VSPat<NEONfmin, VMINfd>;
4710 def : N2VSPat<arm_ftosi, VCVTf2sd>;
4711 def : N2VSPat<arm_ftoui, VCVTf2ud>;
4712 def : N2VSPat<arm_sitof, VCVTs2fd>;
4713 def : N2VSPat<arm_uitof, VCVTu2fd>;
4715 //===----------------------------------------------------------------------===//
4716 // Non-Instruction Patterns
4717 //===----------------------------------------------------------------------===//
4720 def : Pat<(v1i64 (bitconvert (v2i32 DPR:$src))), (v1i64 DPR:$src)>;
4721 def : Pat<(v1i64 (bitconvert (v4i16 DPR:$src))), (v1i64 DPR:$src)>;
4722 def : Pat<(v1i64 (bitconvert (v8i8 DPR:$src))), (v1i64 DPR:$src)>;
4723 def : Pat<(v1i64 (bitconvert (f64 DPR:$src))), (v1i64 DPR:$src)>;
4724 def : Pat<(v1i64 (bitconvert (v2f32 DPR:$src))), (v1i64 DPR:$src)>;
4725 def : Pat<(v2i32 (bitconvert (v1i64 DPR:$src))), (v2i32 DPR:$src)>;
4726 def : Pat<(v2i32 (bitconvert (v4i16 DPR:$src))), (v2i32 DPR:$src)>;
4727 def : Pat<(v2i32 (bitconvert (v8i8 DPR:$src))), (v2i32 DPR:$src)>;
4728 def : Pat<(v2i32 (bitconvert (f64 DPR:$src))), (v2i32 DPR:$src)>;
4729 def : Pat<(v2i32 (bitconvert (v2f32 DPR:$src))), (v2i32 DPR:$src)>;
4730 def : Pat<(v4i16 (bitconvert (v1i64 DPR:$src))), (v4i16 DPR:$src)>;
4731 def : Pat<(v4i16 (bitconvert (v2i32 DPR:$src))), (v4i16 DPR:$src)>;
4732 def : Pat<(v4i16 (bitconvert (v8i8 DPR:$src))), (v4i16 DPR:$src)>;
4733 def : Pat<(v4i16 (bitconvert (f64 DPR:$src))), (v4i16 DPR:$src)>;
4734 def : Pat<(v4i16 (bitconvert (v2f32 DPR:$src))), (v4i16 DPR:$src)>;
4735 def : Pat<(v8i8 (bitconvert (v1i64 DPR:$src))), (v8i8 DPR:$src)>;
4736 def : Pat<(v8i8 (bitconvert (v2i32 DPR:$src))), (v8i8 DPR:$src)>;
4737 def : Pat<(v8i8 (bitconvert (v4i16 DPR:$src))), (v8i8 DPR:$src)>;
4738 def : Pat<(v8i8 (bitconvert (f64 DPR:$src))), (v8i8 DPR:$src)>;
4739 def : Pat<(v8i8 (bitconvert (v2f32 DPR:$src))), (v8i8 DPR:$src)>;
4740 def : Pat<(f64 (bitconvert (v1i64 DPR:$src))), (f64 DPR:$src)>;
4741 def : Pat<(f64 (bitconvert (v2i32 DPR:$src))), (f64 DPR:$src)>;
4742 def : Pat<(f64 (bitconvert (v4i16 DPR:$src))), (f64 DPR:$src)>;
4743 def : Pat<(f64 (bitconvert (v8i8 DPR:$src))), (f64 DPR:$src)>;
4744 def : Pat<(f64 (bitconvert (v2f32 DPR:$src))), (f64 DPR:$src)>;
4745 def : Pat<(v2f32 (bitconvert (f64 DPR:$src))), (v2f32 DPR:$src)>;
4746 def : Pat<(v2f32 (bitconvert (v1i64 DPR:$src))), (v2f32 DPR:$src)>;
4747 def : Pat<(v2f32 (bitconvert (v2i32 DPR:$src))), (v2f32 DPR:$src)>;
4748 def : Pat<(v2f32 (bitconvert (v4i16 DPR:$src))), (v2f32 DPR:$src)>;
4749 def : Pat<(v2f32 (bitconvert (v8i8 DPR:$src))), (v2f32 DPR:$src)>;
4751 def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (v2i64 QPR:$src)>;
4752 def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (v2i64 QPR:$src)>;
4753 def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (v2i64 QPR:$src)>;
4754 def : Pat<(v2i64 (bitconvert (v2f64 QPR:$src))), (v2i64 QPR:$src)>;
4755 def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (v2i64 QPR:$src)>;
4756 def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (v4i32 QPR:$src)>;
4757 def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (v4i32 QPR:$src)>;
4758 def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (v4i32 QPR:$src)>;
4759 def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (v4i32 QPR:$src)>;
4760 def : Pat<(v4i32 (bitconvert (v4f32 QPR:$src))), (v4i32 QPR:$src)>;
4761 def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (v8i16 QPR:$src)>;
4762 def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (v8i16 QPR:$src)>;
4763 def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (v8i16 QPR:$src)>;
4764 def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (v8i16 QPR:$src)>;
4765 def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (v8i16 QPR:$src)>;
4766 def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (v16i8 QPR:$src)>;
4767 def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (v16i8 QPR:$src)>;
4768 def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (v16i8 QPR:$src)>;
4769 def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (v16i8 QPR:$src)>;
4770 def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (v16i8 QPR:$src)>;
4771 def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (v4f32 QPR:$src)>;
4772 def : Pat<(v4f32 (bitconvert (v4i32 QPR:$src))), (v4f32 QPR:$src)>;
4773 def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (v4f32 QPR:$src)>;
4774 def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (v4f32 QPR:$src)>;
4775 def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (v4f32 QPR:$src)>;
4776 def : Pat<(v2f64 (bitconvert (v2i64 QPR:$src))), (v2f64 QPR:$src)>;
4777 def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (v2f64 QPR:$src)>;
4778 def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 QPR:$src)>;
4779 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 QPR:$src)>;
4780 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 QPR:$src)>;