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: single-, double- and quad-register.
1671 class N2VS<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,
1675 (outs DPR_VFP2:$Vd), (ins DPR_VFP2:$Vm),
1676 IIC_VUNAD, OpcodeStr, Dt, "$Vd, $Vm", "", []>;
1677 class N2VD<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1678 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
1679 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
1680 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
1681 (ins DPR:$Vm), IIC_VUNAD, OpcodeStr, Dt,"$Vd, $Vm", "",
1682 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vm))))]>;
1683 class N2VQ<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1684 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
1685 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
1686 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
1687 (ins QPR:$Vm), IIC_VUNAQ, OpcodeStr, Dt,"$Vd, $Vm", "",
1688 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vm))))]>;
1690 // Basic 2-register intrinsics, both double- and quad-register.
1691 class N2VDInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1692 bits<2> op17_16, bits<5> op11_7, bit op4,
1693 InstrItinClass itin, string OpcodeStr, string Dt,
1694 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1695 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
1696 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
1697 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm))))]>;
1698 class N2VQInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1699 bits<2> op17_16, bits<5> op11_7, bit op4,
1700 InstrItinClass itin, string OpcodeStr, string Dt,
1701 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1702 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
1703 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
1704 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm))))]>;
1706 // Narrow 2-register operations.
1707 class N2VN<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1708 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1709 InstrItinClass itin, string OpcodeStr, string Dt,
1710 ValueType TyD, ValueType TyQ, SDNode OpNode>
1711 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$Vd),
1712 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
1713 [(set DPR:$Vd, (TyD (OpNode (TyQ QPR:$Vm))))]>;
1715 // Narrow 2-register intrinsics.
1716 class N2VNInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1717 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1718 InstrItinClass itin, string OpcodeStr, string Dt,
1719 ValueType TyD, ValueType TyQ, Intrinsic IntOp>
1720 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$Vd),
1721 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
1722 [(set DPR:$Vd, (TyD (IntOp (TyQ QPR:$Vm))))]>;
1724 // Long 2-register operations (currently only used for VMOVL).
1725 class N2VL<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1726 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1727 InstrItinClass itin, string OpcodeStr, string Dt,
1728 ValueType TyQ, ValueType TyD, SDNode OpNode>
1729 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$Vd),
1730 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
1731 [(set QPR:$Vd, (TyQ (OpNode (TyD DPR:$Vm))))]>;
1733 // 2-register shuffles (VTRN/VZIP/VUZP), both double- and quad-register.
1734 class N2VDShuffle<bits<2> op19_18, bits<5> op11_7, string OpcodeStr, string Dt>
1735 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 0, 0, (outs DPR:$Vd, DPR:$Vm),
1736 (ins DPR:$src1, DPR:$src2), IIC_VPERMD,
1737 OpcodeStr, Dt, "$Vd, $Vm",
1738 "$src1 = $Vd, $src2 = $Vm", []>;
1739 class N2VQShuffle<bits<2> op19_18, bits<5> op11_7,
1740 InstrItinClass itin, string OpcodeStr, string Dt>
1741 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 1, 0, (outs QPR:$Vd, QPR:$Vm),
1742 (ins QPR:$src1, QPR:$src2), itin, OpcodeStr, Dt, "$Vd, $Vm",
1743 "$src1 = $Vd, $src2 = $Vm", []>;
1745 // Basic 3-register operations: single-, double- and quad-register.
1746 class N3VS<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1747 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1748 SDNode OpNode, bit Commutable>
1749 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1750 (outs DPR_VFP2:$Vd), (ins DPR_VFP2:$Vn, DPR_VFP2:$Vm), N3RegFrm,
1751 IIC_VBIND, OpcodeStr, Dt, "$Vd, $Vn, $Vm", "", []> {
1752 let isCommutable = Commutable;
1755 class N3VD<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1756 InstrItinClass itin, string OpcodeStr, string Dt,
1757 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
1758 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1759 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1760 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
1761 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
1762 let isCommutable = Commutable;
1764 // Same as N3VD but no data type.
1765 class N3VDX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1766 InstrItinClass itin, string OpcodeStr,
1767 ValueType ResTy, ValueType OpTy,
1768 SDNode OpNode, bit Commutable>
1769 : N3VX<op24, op23, op21_20, op11_8, 0, op4,
1770 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1771 OpcodeStr, "$Vd, $Vn, $Vm", "",
1772 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>{
1773 let isCommutable = Commutable;
1776 class N3VDSL<bits<2> op21_20, bits<4> op11_8,
1777 InstrItinClass itin, string OpcodeStr, string Dt,
1778 ValueType Ty, SDNode ShOp>
1779 : N3V<0, 1, op21_20, op11_8, 1, 0,
1780 (outs DPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, nohash_imm:$lane),
1781 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "",
1783 (Ty (ShOp (Ty DPR:$Vn),
1784 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),imm:$lane)))))]> {
1785 let isCommutable = 0;
1787 class N3VDSL16<bits<2> op21_20, bits<4> op11_8,
1788 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1789 : N3V<0, 1, op21_20, op11_8, 1, 0,
1790 (outs DPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, nohash_imm:$lane),
1791 NVMulSLFrm, IIC_VMULi16D, OpcodeStr, Dt,"$Vd, $Vn, $Vm[$lane]","",
1793 (Ty (ShOp (Ty DPR:$Vn),
1794 (Ty (NEONvduplane (Ty DPR_8:$Vm), imm:$lane)))))]> {
1795 let isCommutable = 0;
1798 class N3VQ<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1799 InstrItinClass itin, string OpcodeStr, string Dt,
1800 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
1801 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1802 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
1803 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
1804 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
1805 let isCommutable = Commutable;
1807 class N3VQX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1808 InstrItinClass itin, string OpcodeStr,
1809 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
1810 : N3VX<op24, op23, op21_20, op11_8, 1, op4,
1811 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
1812 OpcodeStr, "$Vd, $Vn, $Vm", "",
1813 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]>{
1814 let isCommutable = Commutable;
1816 class N3VQSL<bits<2> op21_20, bits<4> op11_8,
1817 InstrItinClass itin, string OpcodeStr, string Dt,
1818 ValueType ResTy, ValueType OpTy, SDNode ShOp>
1819 : N3V<1, 1, op21_20, op11_8, 1, 0,
1820 (outs QPR:$Vd), (ins QPR:$Vn, DPR_VFP2:$Vm, nohash_imm:$lane),
1821 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "",
1822 [(set (ResTy QPR:$Vd),
1823 (ResTy (ShOp (ResTy QPR:$Vn),
1824 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
1826 let isCommutable = 0;
1828 class N3VQSL16<bits<2> op21_20, bits<4> op11_8, string OpcodeStr, string Dt,
1829 ValueType ResTy, ValueType OpTy, SDNode ShOp>
1830 : N3V<1, 1, op21_20, op11_8, 1, 0,
1831 (outs QPR:$Vd), (ins QPR:$Vn, DPR_8:$Vm, nohash_imm:$lane),
1832 NVMulSLFrm, IIC_VMULi16Q, OpcodeStr, Dt,"$Vd, $Vn, $Vm[$lane]","",
1833 [(set (ResTy QPR:$Vd),
1834 (ResTy (ShOp (ResTy QPR:$Vn),
1835 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
1837 let isCommutable = 0;
1840 // Basic 3-register intrinsics, both double- and quad-register.
1841 class N3VDInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1842 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1843 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
1844 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1845 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), f, itin,
1846 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
1847 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
1848 let isCommutable = Commutable;
1850 class N3VDIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1851 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
1852 : N3V<0, 1, op21_20, op11_8, 1, 0,
1853 (outs DPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, nohash_imm:$lane),
1854 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "",
1856 (Ty (IntOp (Ty DPR:$Vn),
1857 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),
1859 let isCommutable = 0;
1861 class N3VDIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1862 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
1863 : N3V<0, 1, op21_20, op11_8, 1, 0,
1864 (outs DPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, nohash_imm:$lane),
1865 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "",
1867 (Ty (IntOp (Ty DPR:$Vn),
1868 (Ty (NEONvduplane (Ty DPR_8:$Vm), imm:$lane)))))]> {
1869 let isCommutable = 0;
1871 class N3VDIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1872 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1873 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1874 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1875 (outs DPR:$Vd), (ins DPR:$Vm, DPR:$Vn), f, itin,
1876 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
1877 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (OpTy DPR:$Vn))))]> {
1878 let isCommutable = 0;
1881 class N3VQInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1882 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1883 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
1884 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1885 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), f, itin,
1886 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
1887 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
1888 let isCommutable = Commutable;
1890 class N3VQIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1891 string OpcodeStr, string Dt,
1892 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1893 : N3V<1, 1, op21_20, op11_8, 1, 0,
1894 (outs QPR:$Vd), (ins QPR:$Vn, DPR_VFP2:$Vm, nohash_imm:$lane),
1895 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "",
1896 [(set (ResTy QPR:$Vd),
1897 (ResTy (IntOp (ResTy QPR:$Vn),
1898 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
1900 let isCommutable = 0;
1902 class N3VQIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1903 string OpcodeStr, string Dt,
1904 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1905 : N3V<1, 1, op21_20, op11_8, 1, 0,
1906 (outs QPR:$Vd), (ins QPR:$Vn, DPR_8:$Vm, nohash_imm:$lane),
1907 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "",
1908 [(set (ResTy QPR:$Vd),
1909 (ResTy (IntOp (ResTy QPR:$Vn),
1910 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
1912 let isCommutable = 0;
1914 class N3VQIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1915 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1916 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1917 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1918 (outs QPR:$Vd), (ins QPR:$Vm, QPR:$Vn), f, itin,
1919 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
1920 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (OpTy QPR:$Vn))))]> {
1921 let isCommutable = 0;
1924 // Multiply-Add/Sub operations: single-, double- and quad-register.
1925 class N3VSMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1926 InstrItinClass itin, string OpcodeStr, string Dt,
1927 ValueType Ty, SDPatternOperator MulOp, SDNode OpNode>
1928 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1929 (outs DPR_VFP2:$Vd),
1930 (ins DPR_VFP2:$src1, DPR_VFP2:$Vn, DPR_VFP2:$Vm), N3RegFrm, itin,
1931 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd", []>;
1933 class N3VDMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1934 InstrItinClass itin, string OpcodeStr, string Dt,
1935 ValueType Ty, SDPatternOperator MulOp, SDPatternOperator OpNode>
1936 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1937 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1938 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1939 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
1940 (Ty (MulOp DPR:$Vn, DPR:$Vm)))))]>;
1942 class N3VDMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1943 string OpcodeStr, string Dt,
1944 ValueType Ty, SDPatternOperator MulOp, SDPatternOperator ShOp>
1945 : N3V<0, 1, op21_20, op11_8, 1, 0,
1947 (ins DPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, nohash_imm:$lane),
1949 OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "$src1 = $Vd",
1951 (Ty (ShOp (Ty DPR:$src1),
1953 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),
1955 class N3VDMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1956 string OpcodeStr, string Dt,
1957 ValueType Ty, SDNode MulOp, SDNode ShOp>
1958 : N3V<0, 1, op21_20, op11_8, 1, 0,
1960 (ins DPR:$src1, DPR:$Vn, DPR_8:$Vm, nohash_imm:$lane),
1962 OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "$src1 = $Vd",
1964 (Ty (ShOp (Ty DPR:$src1),
1966 (Ty (NEONvduplane (Ty DPR_8:$Vm),
1969 class N3VQMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1970 InstrItinClass itin, string OpcodeStr, string Dt, ValueType Ty,
1971 SDPatternOperator MulOp, SDPatternOperator OpNode>
1972 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1973 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
1974 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1975 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
1976 (Ty (MulOp QPR:$Vn, QPR:$Vm)))))]>;
1977 class N3VQMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1978 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1979 SDPatternOperator MulOp, SDPatternOperator ShOp>
1980 : N3V<1, 1, op21_20, op11_8, 1, 0,
1982 (ins QPR:$src1, QPR:$Vn, DPR_VFP2:$Vm, nohash_imm:$lane),
1984 OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "$src1 = $Vd",
1985 [(set (ResTy QPR:$Vd),
1986 (ResTy (ShOp (ResTy QPR:$src1),
1987 (ResTy (MulOp QPR:$Vn,
1988 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
1990 class N3VQMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1991 string OpcodeStr, string Dt,
1992 ValueType ResTy, ValueType OpTy,
1993 SDNode MulOp, SDNode ShOp>
1994 : N3V<1, 1, op21_20, op11_8, 1, 0,
1996 (ins QPR:$src1, QPR:$Vn, DPR_8:$Vm, nohash_imm:$lane),
1998 OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "$src1 = $Vd",
1999 [(set (ResTy QPR:$Vd),
2000 (ResTy (ShOp (ResTy QPR:$src1),
2001 (ResTy (MulOp QPR:$Vn,
2002 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2005 // Neon Intrinsic-Op instructions (VABA): double- and quad-register.
2006 class N3VDIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2007 InstrItinClass itin, string OpcodeStr, string Dt,
2008 ValueType Ty, Intrinsic IntOp, SDNode OpNode>
2009 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2010 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2011 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2012 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
2013 (Ty (IntOp (Ty DPR:$Vn), (Ty DPR:$Vm))))))]>;
2014 class N3VQIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2015 InstrItinClass itin, string OpcodeStr, string Dt,
2016 ValueType Ty, Intrinsic IntOp, SDNode OpNode>
2017 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2018 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2019 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2020 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
2021 (Ty (IntOp (Ty QPR:$Vn), (Ty QPR:$Vm))))))]>;
2023 // Neon 3-argument intrinsics, both double- and quad-register.
2024 // The destination register is also used as the first source operand register.
2025 class N3VDInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2026 InstrItinClass itin, string OpcodeStr, string Dt,
2027 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2028 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2029 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2030 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2031 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$src1),
2032 (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>;
2033 class N3VQInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2034 InstrItinClass itin, string OpcodeStr, string Dt,
2035 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2036 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2037 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2038 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2039 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$src1),
2040 (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]>;
2042 // Long Multiply-Add/Sub operations.
2043 class N3VLMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2044 InstrItinClass itin, string OpcodeStr, string Dt,
2045 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2046 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2047 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2048 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2049 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
2050 (TyQ (MulOp (TyD DPR:$Vn),
2051 (TyD DPR:$Vm)))))]>;
2052 class N3VLMulOpSL<bit op24, bits<2> op21_20, bits<4> op11_8,
2053 InstrItinClass itin, string OpcodeStr, string Dt,
2054 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2055 : N3V<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$Vd),
2056 (ins QPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, nohash_imm:$lane),
2058 OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "$src1 = $Vd",
2060 (OpNode (TyQ QPR:$src1),
2061 (TyQ (MulOp (TyD DPR:$Vn),
2062 (TyD (NEONvduplane (TyD DPR_VFP2:$Vm),
2064 class N3VLMulOpSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2065 InstrItinClass itin, string OpcodeStr, string Dt,
2066 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2067 : N3V<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$Vd),
2068 (ins QPR:$src1, DPR:$Vn, DPR_8:$Vm, nohash_imm:$lane),
2070 OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "$src1 = $Vd",
2072 (OpNode (TyQ QPR:$src1),
2073 (TyQ (MulOp (TyD DPR:$Vn),
2074 (TyD (NEONvduplane (TyD DPR_8:$Vm),
2077 // Long Intrinsic-Op vector operations with explicit extend (VABAL).
2078 class N3VLIntExtOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2079 InstrItinClass itin, string OpcodeStr, string Dt,
2080 ValueType TyQ, ValueType TyD, Intrinsic IntOp, SDNode ExtOp,
2082 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2083 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2084 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2085 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
2086 (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
2087 (TyD DPR:$Vm)))))))]>;
2089 // Neon Long 3-argument intrinsic. The destination register is
2090 // a quad-register and is also used as the first source operand register.
2091 class N3VLInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2092 InstrItinClass itin, string OpcodeStr, string Dt,
2093 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
2094 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2095 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2096 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2098 (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$Vn), (TyD DPR:$Vm))))]>;
2099 class N3VLInt3SL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2100 string OpcodeStr, string Dt,
2101 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2102 : N3V<op24, 1, op21_20, op11_8, 1, 0,
2104 (ins QPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, nohash_imm:$lane),
2106 OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "$src1 = $Vd",
2107 [(set (ResTy QPR:$Vd),
2108 (ResTy (IntOp (ResTy QPR:$src1),
2110 (OpTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2112 class N3VLInt3SL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2113 InstrItinClass itin, string OpcodeStr, string Dt,
2114 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2115 : N3V<op24, 1, op21_20, op11_8, 1, 0,
2117 (ins QPR:$src1, DPR:$Vn, DPR_8:$Vm, nohash_imm:$lane),
2119 OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "$src1 = $Vd",
2120 [(set (ResTy QPR:$Vd),
2121 (ResTy (IntOp (ResTy QPR:$src1),
2123 (OpTy (NEONvduplane (OpTy DPR_8:$Vm),
2126 // Narrowing 3-register intrinsics.
2127 class N3VNInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2128 string OpcodeStr, string Dt, ValueType TyD, ValueType TyQ,
2129 Intrinsic IntOp, bit Commutable>
2130 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2131 (outs DPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINi4D,
2132 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2133 [(set DPR:$Vd, (TyD (IntOp (TyQ QPR:$Vn), (TyQ QPR:$Vm))))]> {
2134 let isCommutable = Commutable;
2137 // Long 3-register operations.
2138 class N3VL<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2139 InstrItinClass itin, string OpcodeStr, string Dt,
2140 ValueType TyQ, ValueType TyD, SDNode OpNode, bit Commutable>
2141 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2142 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2143 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2144 [(set QPR:$Vd, (TyQ (OpNode (TyD DPR:$Vn), (TyD DPR:$Vm))))]> {
2145 let isCommutable = Commutable;
2147 class N3VLSL<bit op24, bits<2> op21_20, bits<4> op11_8,
2148 InstrItinClass itin, string OpcodeStr, string Dt,
2149 ValueType TyQ, ValueType TyD, SDNode OpNode>
2150 : N3V<op24, 1, op21_20, op11_8, 1, 0,
2151 (outs QPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, nohash_imm:$lane),
2152 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "",
2154 (TyQ (OpNode (TyD DPR:$Vn),
2155 (TyD (NEONvduplane (TyD DPR_VFP2:$Vm),imm:$lane)))))]>;
2156 class N3VLSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2157 InstrItinClass itin, string OpcodeStr, string Dt,
2158 ValueType TyQ, ValueType TyD, SDNode OpNode>
2159 : N3V<op24, 1, op21_20, op11_8, 1, 0,
2160 (outs QPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, nohash_imm:$lane),
2161 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "",
2163 (TyQ (OpNode (TyD DPR:$Vn),
2164 (TyD (NEONvduplane (TyD DPR_8:$Vm), imm:$lane)))))]>;
2166 // Long 3-register operations with explicitly extended operands.
2167 class N3VLExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2168 InstrItinClass itin, string OpcodeStr, string Dt,
2169 ValueType TyQ, ValueType TyD, SDNode OpNode, SDNode ExtOp,
2171 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2172 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2173 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2174 [(set QPR:$Vd, (OpNode (TyQ (ExtOp (TyD DPR:$Vn))),
2175 (TyQ (ExtOp (TyD DPR:$Vm)))))]> {
2176 let isCommutable = Commutable;
2179 // Long 3-register intrinsics with explicit extend (VABDL).
2180 class N3VLIntExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2181 InstrItinClass itin, string OpcodeStr, string Dt,
2182 ValueType TyQ, ValueType TyD, Intrinsic IntOp, SDNode ExtOp,
2184 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2185 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2186 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2187 [(set QPR:$Vd, (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
2188 (TyD DPR:$Vm))))))]> {
2189 let isCommutable = Commutable;
2192 // Long 3-register intrinsics.
2193 class N3VLInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2194 InstrItinClass itin, string OpcodeStr, string Dt,
2195 ValueType TyQ, ValueType TyD, Intrinsic IntOp, bit Commutable>
2196 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2197 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2198 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2199 [(set QPR:$Vd, (TyQ (IntOp (TyD DPR:$Vn), (TyD DPR:$Vm))))]> {
2200 let isCommutable = Commutable;
2202 class N3VLIntSL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2203 string OpcodeStr, string Dt,
2204 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2205 : N3V<op24, 1, op21_20, op11_8, 1, 0,
2206 (outs QPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, nohash_imm:$lane),
2207 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "",
2208 [(set (ResTy QPR:$Vd),
2209 (ResTy (IntOp (OpTy DPR:$Vn),
2210 (OpTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2212 class N3VLIntSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2213 InstrItinClass itin, string OpcodeStr, string Dt,
2214 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2215 : N3V<op24, 1, op21_20, op11_8, 1, 0,
2216 (outs QPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, nohash_imm:$lane),
2217 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "",
2218 [(set (ResTy QPR:$Vd),
2219 (ResTy (IntOp (OpTy DPR:$Vn),
2220 (OpTy (NEONvduplane (OpTy DPR_8:$Vm),
2223 // Wide 3-register operations.
2224 class N3VW<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2225 string OpcodeStr, string Dt, ValueType TyQ, ValueType TyD,
2226 SDNode OpNode, SDNode ExtOp, bit Commutable>
2227 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2228 (outs QPR:$Vd), (ins QPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VSUBiD,
2229 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2230 [(set QPR:$Vd, (OpNode (TyQ QPR:$Vn),
2231 (TyQ (ExtOp (TyD DPR:$Vm)))))]> {
2232 let isCommutable = Commutable;
2235 // Pairwise long 2-register intrinsics, both double- and quad-register.
2236 class N2VDPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2237 bits<2> op17_16, bits<5> op11_7, bit op4,
2238 string OpcodeStr, string Dt,
2239 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2240 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
2241 (ins DPR:$Vm), IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
2242 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm))))]>;
2243 class N2VQPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2244 bits<2> op17_16, bits<5> op11_7, bit op4,
2245 string OpcodeStr, string Dt,
2246 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2247 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
2248 (ins QPR:$Vm), IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
2249 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm))))]>;
2251 // Pairwise long 2-register accumulate intrinsics,
2252 // both double- and quad-register.
2253 // The destination register is also used as the first source operand register.
2254 class N2VDPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2255 bits<2> op17_16, bits<5> op11_7, bit op4,
2256 string OpcodeStr, string Dt,
2257 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2258 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
2259 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vm), IIC_VPALiD,
2260 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
2261 [(set DPR:$Vd, (ResTy (IntOp (ResTy DPR:$src1), (OpTy DPR:$Vm))))]>;
2262 class N2VQPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2263 bits<2> op17_16, bits<5> op11_7, bit op4,
2264 string OpcodeStr, string Dt,
2265 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2266 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4,
2267 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vm), IIC_VPALiQ,
2268 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
2269 [(set QPR:$Vd, (ResTy (IntOp (ResTy QPR:$src1), (OpTy QPR:$Vm))))]>;
2271 // Shift by immediate,
2272 // both double- and quad-register.
2273 class N2VDSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2274 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2275 ValueType Ty, SDNode OpNode>
2276 : N2VImm<op24, op23, op11_8, op7, 0, op4,
2277 (outs DPR:$Vd), (ins DPR:$Vm, i32imm:$SIMM), f, itin,
2278 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2279 [(set DPR:$Vd, (Ty (OpNode (Ty DPR:$Vm), (i32 imm:$SIMM))))]>;
2280 class N2VQSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2281 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2282 ValueType Ty, SDNode OpNode>
2283 : N2VImm<op24, op23, op11_8, op7, 1, op4,
2284 (outs QPR:$Vd), (ins QPR:$Vm, i32imm:$SIMM), f, itin,
2285 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2286 [(set QPR:$Vd, (Ty (OpNode (Ty QPR:$Vm), (i32 imm:$SIMM))))]>;
2288 // Long shift by immediate.
2289 class N2VLSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
2290 string OpcodeStr, string Dt,
2291 ValueType ResTy, ValueType OpTy, SDNode OpNode>
2292 : N2VImm<op24, op23, op11_8, op7, op6, op4,
2293 (outs QPR:$Vd), (ins DPR:$Vm, i32imm:$SIMM), N2RegVShLFrm,
2294 IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2295 [(set QPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vm),
2296 (i32 imm:$SIMM))))]>;
2298 // Narrow shift by immediate.
2299 class N2VNSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
2300 InstrItinClass itin, string OpcodeStr, string Dt,
2301 ValueType ResTy, ValueType OpTy, SDNode OpNode>
2302 : N2VImm<op24, op23, op11_8, op7, op6, op4,
2303 (outs DPR:$Vd), (ins QPR:$Vm, i32imm:$SIMM), N2RegVShRFrm, itin,
2304 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2305 [(set DPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vm),
2306 (i32 imm:$SIMM))))]>;
2308 // Shift right by immediate and accumulate,
2309 // both double- and quad-register.
2310 class N2VDShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2311 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
2312 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
2313 (ins DPR:$src1, DPR:$Vm, i32imm:$SIMM), N2RegVShRFrm, IIC_VPALiD,
2314 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2315 [(set DPR:$Vd, (Ty (add DPR:$src1,
2316 (Ty (ShOp DPR:$Vm, (i32 imm:$SIMM))))))]>;
2317 class N2VQShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2318 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
2319 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
2320 (ins QPR:$src1, QPR:$Vm, i32imm:$SIMM), N2RegVShRFrm, IIC_VPALiD,
2321 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2322 [(set QPR:$Vd, (Ty (add QPR:$src1,
2323 (Ty (ShOp QPR:$Vm, (i32 imm:$SIMM))))))]>;
2325 // Shift by immediate and insert,
2326 // both double- and quad-register.
2327 class N2VDShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2328 Format f, string OpcodeStr, string Dt, ValueType Ty,SDNode ShOp>
2329 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
2330 (ins DPR:$src1, DPR:$Vm, i32imm:$SIMM), f, IIC_VSHLiD,
2331 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2332 [(set DPR:$Vd, (Ty (ShOp DPR:$src1, DPR:$Vm, (i32 imm:$SIMM))))]>;
2333 class N2VQShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2334 Format f, string OpcodeStr, string Dt, ValueType Ty,SDNode ShOp>
2335 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
2336 (ins QPR:$src1, QPR:$Vm, i32imm:$SIMM), f, IIC_VSHLiQ,
2337 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2338 [(set QPR:$Vd, (Ty (ShOp QPR:$src1, QPR:$Vm, (i32 imm:$SIMM))))]>;
2340 // Convert, with fractional bits immediate,
2341 // both double- and quad-register.
2342 class N2VCvtD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2343 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2345 : N2VImm<op24, op23, op11_8, op7, 0, op4,
2346 (outs DPR:$Vd), (ins DPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
2347 IIC_VUNAD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2348 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (i32 imm:$SIMM))))]>;
2349 class N2VCvtQ<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2350 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2352 : N2VImm<op24, op23, op11_8, op7, 1, op4,
2353 (outs QPR:$Vd), (ins QPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
2354 IIC_VUNAQ, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2355 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (i32 imm:$SIMM))))]>;
2357 //===----------------------------------------------------------------------===//
2359 //===----------------------------------------------------------------------===//
2361 // Abbreviations used in multiclass suffixes:
2362 // Q = quarter int (8 bit) elements
2363 // H = half int (16 bit) elements
2364 // S = single int (32 bit) elements
2365 // D = double int (64 bit) elements
2367 // Neon 2-register vector operations -- for disassembly only.
2369 // First with only element sizes of 8, 16 and 32 bits:
2370 multiclass N2V_QHS_cmp<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2371 bits<5> op11_7, bit op4, string opc, string Dt,
2372 string asm, SDNode OpNode> {
2373 // 64-bit vector types.
2374 def v8i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 0, op4,
2375 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2376 opc, !strconcat(Dt, "8"), asm, "",
2377 [(set DPR:$Vd, (v8i8 (OpNode (v8i8 DPR:$Vm))))]>;
2378 def v4i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 0, op4,
2379 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2380 opc, !strconcat(Dt, "16"), asm, "",
2381 [(set DPR:$Vd, (v4i16 (OpNode (v4i16 DPR:$Vm))))]>;
2382 def v2i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
2383 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2384 opc, !strconcat(Dt, "32"), asm, "",
2385 [(set DPR:$Vd, (v2i32 (OpNode (v2i32 DPR:$Vm))))]>;
2386 def v2f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
2387 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2388 opc, "f32", asm, "",
2389 [(set DPR:$Vd, (v2f32 (OpNode (v2f32 DPR:$Vm))))]> {
2390 let Inst{10} = 1; // overwrite F = 1
2393 // 128-bit vector types.
2394 def v16i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 1, op4,
2395 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2396 opc, !strconcat(Dt, "8"), asm, "",
2397 [(set QPR:$Vd, (v16i8 (OpNode (v16i8 QPR:$Vm))))]>;
2398 def v8i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 1, op4,
2399 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2400 opc, !strconcat(Dt, "16"), asm, "",
2401 [(set QPR:$Vd, (v8i16 (OpNode (v8i16 QPR:$Vm))))]>;
2402 def v4i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
2403 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2404 opc, !strconcat(Dt, "32"), asm, "",
2405 [(set QPR:$Vd, (v4i32 (OpNode (v4i32 QPR:$Vm))))]>;
2406 def v4f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
2407 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2408 opc, "f32", asm, "",
2409 [(set QPR:$Vd, (v4f32 (OpNode (v4f32 QPR:$Vm))))]> {
2410 let Inst{10} = 1; // overwrite F = 1
2414 // Neon 3-register vector operations.
2416 // First with only element sizes of 8, 16 and 32 bits:
2417 multiclass N3V_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2418 InstrItinClass itinD16, InstrItinClass itinD32,
2419 InstrItinClass itinQ16, InstrItinClass itinQ32,
2420 string OpcodeStr, string Dt,
2421 SDNode OpNode, bit Commutable = 0> {
2422 // 64-bit vector types.
2423 def v8i8 : N3VD<op24, op23, 0b00, op11_8, op4, itinD16,
2424 OpcodeStr, !strconcat(Dt, "8"),
2425 v8i8, v8i8, OpNode, Commutable>;
2426 def v4i16 : N3VD<op24, op23, 0b01, op11_8, op4, itinD16,
2427 OpcodeStr, !strconcat(Dt, "16"),
2428 v4i16, v4i16, OpNode, Commutable>;
2429 def v2i32 : N3VD<op24, op23, 0b10, op11_8, op4, itinD32,
2430 OpcodeStr, !strconcat(Dt, "32"),
2431 v2i32, v2i32, OpNode, Commutable>;
2433 // 128-bit vector types.
2434 def v16i8 : N3VQ<op24, op23, 0b00, op11_8, op4, itinQ16,
2435 OpcodeStr, !strconcat(Dt, "8"),
2436 v16i8, v16i8, OpNode, Commutable>;
2437 def v8i16 : N3VQ<op24, op23, 0b01, op11_8, op4, itinQ16,
2438 OpcodeStr, !strconcat(Dt, "16"),
2439 v8i16, v8i16, OpNode, Commutable>;
2440 def v4i32 : N3VQ<op24, op23, 0b10, op11_8, op4, itinQ32,
2441 OpcodeStr, !strconcat(Dt, "32"),
2442 v4i32, v4i32, OpNode, Commutable>;
2445 multiclass N3VSL_HS<bits<4> op11_8, string OpcodeStr, string Dt, SDNode ShOp> {
2446 def v4i16 : N3VDSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
2448 def v2i32 : N3VDSL<0b10, op11_8, IIC_VMULi32D, OpcodeStr, !strconcat(Dt,"32"),
2450 def v8i16 : N3VQSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
2451 v8i16, v4i16, ShOp>;
2452 def v4i32 : N3VQSL<0b10, op11_8, IIC_VMULi32Q, OpcodeStr, !strconcat(Dt,"32"),
2453 v4i32, v2i32, ShOp>;
2456 // ....then also with element size 64 bits:
2457 multiclass N3V_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2458 InstrItinClass itinD, InstrItinClass itinQ,
2459 string OpcodeStr, string Dt,
2460 SDNode OpNode, bit Commutable = 0>
2461 : N3V_QHS<op24, op23, op11_8, op4, itinD, itinD, itinQ, itinQ,
2462 OpcodeStr, Dt, OpNode, Commutable> {
2463 def v1i64 : N3VD<op24, op23, 0b11, op11_8, op4, itinD,
2464 OpcodeStr, !strconcat(Dt, "64"),
2465 v1i64, v1i64, OpNode, Commutable>;
2466 def v2i64 : N3VQ<op24, op23, 0b11, op11_8, op4, itinQ,
2467 OpcodeStr, !strconcat(Dt, "64"),
2468 v2i64, v2i64, OpNode, Commutable>;
2472 // Neon Narrowing 2-register vector operations,
2473 // source operand element sizes of 16, 32 and 64 bits:
2474 multiclass N2VN_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2475 bits<5> op11_7, bit op6, bit op4,
2476 InstrItinClass itin, string OpcodeStr, string Dt,
2478 def v8i8 : N2VN<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
2479 itin, OpcodeStr, !strconcat(Dt, "16"),
2480 v8i8, v8i16, OpNode>;
2481 def v4i16 : N2VN<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
2482 itin, OpcodeStr, !strconcat(Dt, "32"),
2483 v4i16, v4i32, OpNode>;
2484 def v2i32 : N2VN<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
2485 itin, OpcodeStr, !strconcat(Dt, "64"),
2486 v2i32, v2i64, OpNode>;
2489 // Neon Narrowing 2-register vector intrinsics,
2490 // source operand element sizes of 16, 32 and 64 bits:
2491 multiclass N2VNInt_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2492 bits<5> op11_7, bit op6, bit op4,
2493 InstrItinClass itin, string OpcodeStr, string Dt,
2495 def v8i8 : N2VNInt<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
2496 itin, OpcodeStr, !strconcat(Dt, "16"),
2497 v8i8, v8i16, IntOp>;
2498 def v4i16 : N2VNInt<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
2499 itin, OpcodeStr, !strconcat(Dt, "32"),
2500 v4i16, v4i32, IntOp>;
2501 def v2i32 : N2VNInt<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
2502 itin, OpcodeStr, !strconcat(Dt, "64"),
2503 v2i32, v2i64, IntOp>;
2507 // Neon Lengthening 2-register vector intrinsic (currently specific to VMOVL).
2508 // source operand element sizes of 16, 32 and 64 bits:
2509 multiclass N2VL_QHS<bits<2> op24_23, bits<5> op11_7, bit op6, bit op4,
2510 string OpcodeStr, string Dt, SDNode OpNode> {
2511 def v8i16 : N2VL<op24_23, 0b00, 0b10, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2512 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode>;
2513 def v4i32 : N2VL<op24_23, 0b01, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2514 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
2515 def v2i64 : N2VL<op24_23, 0b10, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2516 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
2520 // Neon 3-register vector intrinsics.
2522 // First with only element sizes of 16 and 32 bits:
2523 multiclass N3VInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2524 InstrItinClass itinD16, InstrItinClass itinD32,
2525 InstrItinClass itinQ16, InstrItinClass itinQ32,
2526 string OpcodeStr, string Dt,
2527 Intrinsic IntOp, bit Commutable = 0> {
2528 // 64-bit vector types.
2529 def v4i16 : N3VDInt<op24, op23, 0b01, op11_8, op4, f, itinD16,
2530 OpcodeStr, !strconcat(Dt, "16"),
2531 v4i16, v4i16, IntOp, Commutable>;
2532 def v2i32 : N3VDInt<op24, op23, 0b10, op11_8, op4, f, itinD32,
2533 OpcodeStr, !strconcat(Dt, "32"),
2534 v2i32, v2i32, IntOp, Commutable>;
2536 // 128-bit vector types.
2537 def v8i16 : N3VQInt<op24, op23, 0b01, op11_8, op4, f, itinQ16,
2538 OpcodeStr, !strconcat(Dt, "16"),
2539 v8i16, v8i16, IntOp, Commutable>;
2540 def v4i32 : N3VQInt<op24, op23, 0b10, op11_8, op4, f, itinQ32,
2541 OpcodeStr, !strconcat(Dt, "32"),
2542 v4i32, v4i32, IntOp, Commutable>;
2544 multiclass N3VInt_HSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2545 InstrItinClass itinD16, InstrItinClass itinD32,
2546 InstrItinClass itinQ16, InstrItinClass itinQ32,
2547 string OpcodeStr, string Dt,
2549 // 64-bit vector types.
2550 def v4i16 : N3VDIntSh<op24, op23, 0b01, op11_8, op4, f, itinD16,
2551 OpcodeStr, !strconcat(Dt, "16"),
2552 v4i16, v4i16, IntOp>;
2553 def v2i32 : N3VDIntSh<op24, op23, 0b10, op11_8, op4, f, itinD32,
2554 OpcodeStr, !strconcat(Dt, "32"),
2555 v2i32, v2i32, IntOp>;
2557 // 128-bit vector types.
2558 def v8i16 : N3VQIntSh<op24, op23, 0b01, op11_8, op4, f, itinQ16,
2559 OpcodeStr, !strconcat(Dt, "16"),
2560 v8i16, v8i16, IntOp>;
2561 def v4i32 : N3VQIntSh<op24, op23, 0b10, op11_8, op4, f, itinQ32,
2562 OpcodeStr, !strconcat(Dt, "32"),
2563 v4i32, v4i32, IntOp>;
2566 multiclass N3VIntSL_HS<bits<4> op11_8,
2567 InstrItinClass itinD16, InstrItinClass itinD32,
2568 InstrItinClass itinQ16, InstrItinClass itinQ32,
2569 string OpcodeStr, string Dt, Intrinsic IntOp> {
2570 def v4i16 : N3VDIntSL16<0b01, op11_8, itinD16,
2571 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp>;
2572 def v2i32 : N3VDIntSL<0b10, op11_8, itinD32,
2573 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp>;
2574 def v8i16 : N3VQIntSL16<0b01, op11_8, itinQ16,
2575 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16, IntOp>;
2576 def v4i32 : N3VQIntSL<0b10, op11_8, itinQ32,
2577 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32, IntOp>;
2580 // ....then also with element size of 8 bits:
2581 multiclass N3VInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2582 InstrItinClass itinD16, InstrItinClass itinD32,
2583 InstrItinClass itinQ16, InstrItinClass itinQ32,
2584 string OpcodeStr, string Dt,
2585 Intrinsic IntOp, bit Commutable = 0>
2586 : N3VInt_HS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2587 OpcodeStr, Dt, IntOp, Commutable> {
2588 def v8i8 : N3VDInt<op24, op23, 0b00, op11_8, op4, f, itinD16,
2589 OpcodeStr, !strconcat(Dt, "8"),
2590 v8i8, v8i8, IntOp, Commutable>;
2591 def v16i8 : N3VQInt<op24, op23, 0b00, op11_8, op4, f, itinQ16,
2592 OpcodeStr, !strconcat(Dt, "8"),
2593 v16i8, v16i8, IntOp, Commutable>;
2595 multiclass N3VInt_QHSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2596 InstrItinClass itinD16, InstrItinClass itinD32,
2597 InstrItinClass itinQ16, InstrItinClass itinQ32,
2598 string OpcodeStr, string Dt,
2600 : N3VInt_HSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2601 OpcodeStr, Dt, IntOp> {
2602 def v8i8 : N3VDIntSh<op24, op23, 0b00, op11_8, op4, f, itinD16,
2603 OpcodeStr, !strconcat(Dt, "8"),
2605 def v16i8 : N3VQIntSh<op24, op23, 0b00, op11_8, op4, f, itinQ16,
2606 OpcodeStr, !strconcat(Dt, "8"),
2607 v16i8, v16i8, IntOp>;
2611 // ....then also with element size of 64 bits:
2612 multiclass N3VInt_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2613 InstrItinClass itinD16, InstrItinClass itinD32,
2614 InstrItinClass itinQ16, InstrItinClass itinQ32,
2615 string OpcodeStr, string Dt,
2616 Intrinsic IntOp, bit Commutable = 0>
2617 : N3VInt_QHS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2618 OpcodeStr, Dt, IntOp, Commutable> {
2619 def v1i64 : N3VDInt<op24, op23, 0b11, op11_8, op4, f, itinD32,
2620 OpcodeStr, !strconcat(Dt, "64"),
2621 v1i64, v1i64, IntOp, Commutable>;
2622 def v2i64 : N3VQInt<op24, op23, 0b11, op11_8, op4, f, itinQ32,
2623 OpcodeStr, !strconcat(Dt, "64"),
2624 v2i64, v2i64, IntOp, Commutable>;
2626 multiclass N3VInt_QHSDSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2627 InstrItinClass itinD16, InstrItinClass itinD32,
2628 InstrItinClass itinQ16, InstrItinClass itinQ32,
2629 string OpcodeStr, string Dt,
2631 : N3VInt_QHSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2632 OpcodeStr, Dt, IntOp> {
2633 def v1i64 : N3VDIntSh<op24, op23, 0b11, op11_8, op4, f, itinD32,
2634 OpcodeStr, !strconcat(Dt, "64"),
2635 v1i64, v1i64, IntOp>;
2636 def v2i64 : N3VQIntSh<op24, op23, 0b11, op11_8, op4, f, itinQ32,
2637 OpcodeStr, !strconcat(Dt, "64"),
2638 v2i64, v2i64, IntOp>;
2641 // Neon Narrowing 3-register vector intrinsics,
2642 // source operand element sizes of 16, 32 and 64 bits:
2643 multiclass N3VNInt_HSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2644 string OpcodeStr, string Dt,
2645 Intrinsic IntOp, bit Commutable = 0> {
2646 def v8i8 : N3VNInt<op24, op23, 0b00, op11_8, op4,
2647 OpcodeStr, !strconcat(Dt, "16"),
2648 v8i8, v8i16, IntOp, Commutable>;
2649 def v4i16 : N3VNInt<op24, op23, 0b01, op11_8, op4,
2650 OpcodeStr, !strconcat(Dt, "32"),
2651 v4i16, v4i32, IntOp, Commutable>;
2652 def v2i32 : N3VNInt<op24, op23, 0b10, op11_8, op4,
2653 OpcodeStr, !strconcat(Dt, "64"),
2654 v2i32, v2i64, IntOp, Commutable>;
2658 // Neon Long 3-register vector operations.
2660 multiclass N3VL_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2661 InstrItinClass itin16, InstrItinClass itin32,
2662 string OpcodeStr, string Dt,
2663 SDNode OpNode, bit Commutable = 0> {
2664 def v8i16 : N3VL<op24, op23, 0b00, op11_8, op4, itin16,
2665 OpcodeStr, !strconcat(Dt, "8"),
2666 v8i16, v8i8, OpNode, Commutable>;
2667 def v4i32 : N3VL<op24, op23, 0b01, op11_8, op4, itin16,
2668 OpcodeStr, !strconcat(Dt, "16"),
2669 v4i32, v4i16, OpNode, Commutable>;
2670 def v2i64 : N3VL<op24, op23, 0b10, op11_8, op4, itin32,
2671 OpcodeStr, !strconcat(Dt, "32"),
2672 v2i64, v2i32, OpNode, Commutable>;
2675 multiclass N3VLSL_HS<bit op24, bits<4> op11_8,
2676 InstrItinClass itin, string OpcodeStr, string Dt,
2678 def v4i16 : N3VLSL16<op24, 0b01, op11_8, itin, OpcodeStr,
2679 !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
2680 def v2i32 : N3VLSL<op24, 0b10, op11_8, itin, OpcodeStr,
2681 !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
2684 multiclass N3VLExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2685 InstrItinClass itin16, InstrItinClass itin32,
2686 string OpcodeStr, string Dt,
2687 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
2688 def v8i16 : N3VLExt<op24, op23, 0b00, op11_8, op4, itin16,
2689 OpcodeStr, !strconcat(Dt, "8"),
2690 v8i16, v8i8, OpNode, ExtOp, Commutable>;
2691 def v4i32 : N3VLExt<op24, op23, 0b01, op11_8, op4, itin16,
2692 OpcodeStr, !strconcat(Dt, "16"),
2693 v4i32, v4i16, OpNode, ExtOp, Commutable>;
2694 def v2i64 : N3VLExt<op24, op23, 0b10, op11_8, op4, itin32,
2695 OpcodeStr, !strconcat(Dt, "32"),
2696 v2i64, v2i32, OpNode, ExtOp, Commutable>;
2699 // Neon Long 3-register vector intrinsics.
2701 // First with only element sizes of 16 and 32 bits:
2702 multiclass N3VLInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
2703 InstrItinClass itin16, InstrItinClass itin32,
2704 string OpcodeStr, string Dt,
2705 Intrinsic IntOp, bit Commutable = 0> {
2706 def v4i32 : N3VLInt<op24, op23, 0b01, op11_8, op4, itin16,
2707 OpcodeStr, !strconcat(Dt, "16"),
2708 v4i32, v4i16, IntOp, Commutable>;
2709 def v2i64 : N3VLInt<op24, op23, 0b10, op11_8, op4, itin32,
2710 OpcodeStr, !strconcat(Dt, "32"),
2711 v2i64, v2i32, IntOp, Commutable>;
2714 multiclass N3VLIntSL_HS<bit op24, bits<4> op11_8,
2715 InstrItinClass itin, string OpcodeStr, string Dt,
2717 def v4i16 : N3VLIntSL16<op24, 0b01, op11_8, itin,
2718 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
2719 def v2i32 : N3VLIntSL<op24, 0b10, op11_8, itin,
2720 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
2723 // ....then also with element size of 8 bits:
2724 multiclass N3VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2725 InstrItinClass itin16, InstrItinClass itin32,
2726 string OpcodeStr, string Dt,
2727 Intrinsic IntOp, bit Commutable = 0>
2728 : N3VLInt_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt,
2729 IntOp, Commutable> {
2730 def v8i16 : N3VLInt<op24, op23, 0b00, op11_8, op4, itin16,
2731 OpcodeStr, !strconcat(Dt, "8"),
2732 v8i16, v8i8, IntOp, Commutable>;
2735 // ....with explicit extend (VABDL).
2736 multiclass N3VLIntExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2737 InstrItinClass itin, string OpcodeStr, string Dt,
2738 Intrinsic IntOp, SDNode ExtOp, bit Commutable = 0> {
2739 def v8i16 : N3VLIntExt<op24, op23, 0b00, op11_8, op4, itin,
2740 OpcodeStr, !strconcat(Dt, "8"),
2741 v8i16, v8i8, IntOp, ExtOp, Commutable>;
2742 def v4i32 : N3VLIntExt<op24, op23, 0b01, op11_8, op4, itin,
2743 OpcodeStr, !strconcat(Dt, "16"),
2744 v4i32, v4i16, IntOp, ExtOp, Commutable>;
2745 def v2i64 : N3VLIntExt<op24, op23, 0b10, op11_8, op4, itin,
2746 OpcodeStr, !strconcat(Dt, "32"),
2747 v2i64, v2i32, IntOp, ExtOp, Commutable>;
2751 // Neon Wide 3-register vector intrinsics,
2752 // source operand element sizes of 8, 16 and 32 bits:
2753 multiclass N3VW_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2754 string OpcodeStr, string Dt,
2755 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
2756 def v8i16 : N3VW<op24, op23, 0b00, op11_8, op4,
2757 OpcodeStr, !strconcat(Dt, "8"),
2758 v8i16, v8i8, OpNode, ExtOp, Commutable>;
2759 def v4i32 : N3VW<op24, op23, 0b01, op11_8, op4,
2760 OpcodeStr, !strconcat(Dt, "16"),
2761 v4i32, v4i16, OpNode, ExtOp, Commutable>;
2762 def v2i64 : N3VW<op24, op23, 0b10, op11_8, op4,
2763 OpcodeStr, !strconcat(Dt, "32"),
2764 v2i64, v2i32, OpNode, ExtOp, Commutable>;
2768 // Neon Multiply-Op vector operations,
2769 // element sizes of 8, 16 and 32 bits:
2770 multiclass N3VMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2771 InstrItinClass itinD16, InstrItinClass itinD32,
2772 InstrItinClass itinQ16, InstrItinClass itinQ32,
2773 string OpcodeStr, string Dt, SDNode OpNode> {
2774 // 64-bit vector types.
2775 def v8i8 : N3VDMulOp<op24, op23, 0b00, op11_8, op4, itinD16,
2776 OpcodeStr, !strconcat(Dt, "8"), v8i8, mul, OpNode>;
2777 def v4i16 : N3VDMulOp<op24, op23, 0b01, op11_8, op4, itinD16,
2778 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, OpNode>;
2779 def v2i32 : N3VDMulOp<op24, op23, 0b10, op11_8, op4, itinD32,
2780 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, OpNode>;
2782 // 128-bit vector types.
2783 def v16i8 : N3VQMulOp<op24, op23, 0b00, op11_8, op4, itinQ16,
2784 OpcodeStr, !strconcat(Dt, "8"), v16i8, mul, OpNode>;
2785 def v8i16 : N3VQMulOp<op24, op23, 0b01, op11_8, op4, itinQ16,
2786 OpcodeStr, !strconcat(Dt, "16"), v8i16, mul, OpNode>;
2787 def v4i32 : N3VQMulOp<op24, op23, 0b10, op11_8, op4, itinQ32,
2788 OpcodeStr, !strconcat(Dt, "32"), v4i32, mul, OpNode>;
2791 multiclass N3VMulOpSL_HS<bits<4> op11_8,
2792 InstrItinClass itinD16, InstrItinClass itinD32,
2793 InstrItinClass itinQ16, InstrItinClass itinQ32,
2794 string OpcodeStr, string Dt, SDNode ShOp> {
2795 def v4i16 : N3VDMulOpSL16<0b01, op11_8, itinD16,
2796 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, ShOp>;
2797 def v2i32 : N3VDMulOpSL<0b10, op11_8, itinD32,
2798 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, ShOp>;
2799 def v8i16 : N3VQMulOpSL16<0b01, op11_8, itinQ16,
2800 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16,
2802 def v4i32 : N3VQMulOpSL<0b10, op11_8, itinQ32,
2803 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32,
2807 // Neon Intrinsic-Op vector operations,
2808 // element sizes of 8, 16 and 32 bits:
2809 multiclass N3VIntOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2810 InstrItinClass itinD, InstrItinClass itinQ,
2811 string OpcodeStr, string Dt, Intrinsic IntOp,
2813 // 64-bit vector types.
2814 def v8i8 : N3VDIntOp<op24, op23, 0b00, op11_8, op4, itinD,
2815 OpcodeStr, !strconcat(Dt, "8"), v8i8, IntOp, OpNode>;
2816 def v4i16 : N3VDIntOp<op24, op23, 0b01, op11_8, op4, itinD,
2817 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp, OpNode>;
2818 def v2i32 : N3VDIntOp<op24, op23, 0b10, op11_8, op4, itinD,
2819 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp, OpNode>;
2821 // 128-bit vector types.
2822 def v16i8 : N3VQIntOp<op24, op23, 0b00, op11_8, op4, itinQ,
2823 OpcodeStr, !strconcat(Dt, "8"), v16i8, IntOp, OpNode>;
2824 def v8i16 : N3VQIntOp<op24, op23, 0b01, op11_8, op4, itinQ,
2825 OpcodeStr, !strconcat(Dt, "16"), v8i16, IntOp, OpNode>;
2826 def v4i32 : N3VQIntOp<op24, op23, 0b10, op11_8, op4, itinQ,
2827 OpcodeStr, !strconcat(Dt, "32"), v4i32, IntOp, OpNode>;
2830 // Neon 3-argument intrinsics,
2831 // element sizes of 8, 16 and 32 bits:
2832 multiclass N3VInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2833 InstrItinClass itinD, InstrItinClass itinQ,
2834 string OpcodeStr, string Dt, Intrinsic IntOp> {
2835 // 64-bit vector types.
2836 def v8i8 : N3VDInt3<op24, op23, 0b00, op11_8, op4, itinD,
2837 OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
2838 def v4i16 : N3VDInt3<op24, op23, 0b01, op11_8, op4, itinD,
2839 OpcodeStr, !strconcat(Dt, "16"), v4i16, v4i16, IntOp>;
2840 def v2i32 : N3VDInt3<op24, op23, 0b10, op11_8, op4, itinD,
2841 OpcodeStr, !strconcat(Dt, "32"), v2i32, v2i32, IntOp>;
2843 // 128-bit vector types.
2844 def v16i8 : N3VQInt3<op24, op23, 0b00, op11_8, op4, itinQ,
2845 OpcodeStr, !strconcat(Dt, "8"), v16i8, v16i8, IntOp>;
2846 def v8i16 : N3VQInt3<op24, op23, 0b01, op11_8, op4, itinQ,
2847 OpcodeStr, !strconcat(Dt, "16"), v8i16, v8i16, IntOp>;
2848 def v4i32 : N3VQInt3<op24, op23, 0b10, op11_8, op4, itinQ,
2849 OpcodeStr, !strconcat(Dt, "32"), v4i32, v4i32, IntOp>;
2853 // Neon Long Multiply-Op vector operations,
2854 // element sizes of 8, 16 and 32 bits:
2855 multiclass N3VLMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2856 InstrItinClass itin16, InstrItinClass itin32,
2857 string OpcodeStr, string Dt, SDNode MulOp,
2859 def v8i16 : N3VLMulOp<op24, op23, 0b00, op11_8, op4, itin16, OpcodeStr,
2860 !strconcat(Dt, "8"), v8i16, v8i8, MulOp, OpNode>;
2861 def v4i32 : N3VLMulOp<op24, op23, 0b01, op11_8, op4, itin16, OpcodeStr,
2862 !strconcat(Dt, "16"), v4i32, v4i16, MulOp, OpNode>;
2863 def v2i64 : N3VLMulOp<op24, op23, 0b10, op11_8, op4, itin32, OpcodeStr,
2864 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
2867 multiclass N3VLMulOpSL_HS<bit op24, bits<4> op11_8, string OpcodeStr,
2868 string Dt, SDNode MulOp, SDNode OpNode> {
2869 def v4i16 : N3VLMulOpSL16<op24, 0b01, op11_8, IIC_VMACi16D, OpcodeStr,
2870 !strconcat(Dt,"16"), v4i32, v4i16, MulOp, OpNode>;
2871 def v2i32 : N3VLMulOpSL<op24, 0b10, op11_8, IIC_VMACi32D, OpcodeStr,
2872 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
2876 // Neon Long 3-argument intrinsics.
2878 // First with only element sizes of 16 and 32 bits:
2879 multiclass N3VLInt3_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
2880 InstrItinClass itin16, InstrItinClass itin32,
2881 string OpcodeStr, string Dt, Intrinsic IntOp> {
2882 def v4i32 : N3VLInt3<op24, op23, 0b01, op11_8, op4, itin16,
2883 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
2884 def v2i64 : N3VLInt3<op24, op23, 0b10, op11_8, op4, itin32,
2885 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
2888 multiclass N3VLInt3SL_HS<bit op24, bits<4> op11_8,
2889 string OpcodeStr, string Dt, Intrinsic IntOp> {
2890 def v4i16 : N3VLInt3SL16<op24, 0b01, op11_8, IIC_VMACi16D,
2891 OpcodeStr, !strconcat(Dt,"16"), v4i32, v4i16, IntOp>;
2892 def v2i32 : N3VLInt3SL<op24, 0b10, op11_8, IIC_VMACi32D,
2893 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
2896 // ....then also with element size of 8 bits:
2897 multiclass N3VLInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2898 InstrItinClass itin16, InstrItinClass itin32,
2899 string OpcodeStr, string Dt, Intrinsic IntOp>
2900 : N3VLInt3_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt, IntOp> {
2901 def v8i16 : N3VLInt3<op24, op23, 0b00, op11_8, op4, itin16,
2902 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
2905 // ....with explicit extend (VABAL).
2906 multiclass N3VLIntExtOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2907 InstrItinClass itin, string OpcodeStr, string Dt,
2908 Intrinsic IntOp, SDNode ExtOp, SDNode OpNode> {
2909 def v8i16 : N3VLIntExtOp<op24, op23, 0b00, op11_8, op4, itin,
2910 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8,
2911 IntOp, ExtOp, OpNode>;
2912 def v4i32 : N3VLIntExtOp<op24, op23, 0b01, op11_8, op4, itin,
2913 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16,
2914 IntOp, ExtOp, OpNode>;
2915 def v2i64 : N3VLIntExtOp<op24, op23, 0b10, op11_8, op4, itin,
2916 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32,
2917 IntOp, ExtOp, OpNode>;
2921 // Neon 2-register vector intrinsics,
2922 // element sizes of 8, 16 and 32 bits:
2923 multiclass N2VInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2924 bits<5> op11_7, bit op4,
2925 InstrItinClass itinD, InstrItinClass itinQ,
2926 string OpcodeStr, string Dt, Intrinsic IntOp> {
2927 // 64-bit vector types.
2928 def v8i8 : N2VDInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2929 itinD, OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
2930 def v4i16 : N2VDInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2931 itinD, OpcodeStr, !strconcat(Dt, "16"),v4i16,v4i16,IntOp>;
2932 def v2i32 : N2VDInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2933 itinD, OpcodeStr, !strconcat(Dt, "32"),v2i32,v2i32,IntOp>;
2935 // 128-bit vector types.
2936 def v16i8 : N2VQInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2937 itinQ, OpcodeStr, !strconcat(Dt, "8"), v16i8,v16i8,IntOp>;
2938 def v8i16 : N2VQInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2939 itinQ, OpcodeStr, !strconcat(Dt, "16"),v8i16,v8i16,IntOp>;
2940 def v4i32 : N2VQInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2941 itinQ, OpcodeStr, !strconcat(Dt, "32"),v4i32,v4i32,IntOp>;
2945 // Neon Pairwise long 2-register intrinsics,
2946 // element sizes of 8, 16 and 32 bits:
2947 multiclass N2VPLInt_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 : N2VDPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2952 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
2953 def v4i16 : N2VDPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2954 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
2955 def v2i32 : N2VDPLInt<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 : N2VQPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2960 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
2961 def v8i16 : N2VQPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2962 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
2963 def v4i32 : N2VQPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2964 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
2968 // Neon Pairwise long 2-register accumulate intrinsics,
2969 // element sizes of 8, 16 and 32 bits:
2970 multiclass N2VPLInt2_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2971 bits<5> op11_7, bit op4,
2972 string OpcodeStr, string Dt, Intrinsic IntOp> {
2973 // 64-bit vector types.
2974 def v8i8 : N2VDPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2975 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
2976 def v4i16 : N2VDPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2977 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
2978 def v2i32 : N2VDPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2979 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
2981 // 128-bit vector types.
2982 def v16i8 : N2VQPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2983 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
2984 def v8i16 : N2VQPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2985 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
2986 def v4i32 : N2VQPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2987 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
2991 // Neon 2-register vector shift by immediate,
2992 // with f of either N2RegVShLFrm or N2RegVShRFrm
2993 // element sizes of 8, 16, 32 and 64 bits:
2994 multiclass N2VSh_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2995 InstrItinClass itin, string OpcodeStr, string Dt,
2996 SDNode OpNode, Format f> {
2997 // 64-bit vector types.
2998 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
2999 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
3000 let Inst{21-19} = 0b001; // imm6 = 001xxx
3002 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
3003 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
3004 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3006 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
3007 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
3008 let Inst{21} = 0b1; // imm6 = 1xxxxx
3010 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, f, itin,
3011 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
3014 // 128-bit vector types.
3015 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
3016 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
3017 let Inst{21-19} = 0b001; // imm6 = 001xxx
3019 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
3020 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
3021 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3023 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
3024 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
3025 let Inst{21} = 0b1; // imm6 = 1xxxxx
3027 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, f, itin,
3028 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
3032 // Neon Shift-Accumulate vector operations,
3033 // element sizes of 8, 16, 32 and 64 bits:
3034 multiclass N2VShAdd_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3035 string OpcodeStr, string Dt, SDNode ShOp> {
3036 // 64-bit vector types.
3037 def v8i8 : N2VDShAdd<op24, op23, op11_8, 0, op4,
3038 OpcodeStr, !strconcat(Dt, "8"), v8i8, ShOp> {
3039 let Inst{21-19} = 0b001; // imm6 = 001xxx
3041 def v4i16 : N2VDShAdd<op24, op23, op11_8, 0, op4,
3042 OpcodeStr, !strconcat(Dt, "16"), v4i16, ShOp> {
3043 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3045 def v2i32 : N2VDShAdd<op24, op23, op11_8, 0, op4,
3046 OpcodeStr, !strconcat(Dt, "32"), v2i32, ShOp> {
3047 let Inst{21} = 0b1; // imm6 = 1xxxxx
3049 def v1i64 : N2VDShAdd<op24, op23, op11_8, 1, op4,
3050 OpcodeStr, !strconcat(Dt, "64"), v1i64, ShOp>;
3053 // 128-bit vector types.
3054 def v16i8 : N2VQShAdd<op24, op23, op11_8, 0, op4,
3055 OpcodeStr, !strconcat(Dt, "8"), v16i8, ShOp> {
3056 let Inst{21-19} = 0b001; // imm6 = 001xxx
3058 def v8i16 : N2VQShAdd<op24, op23, op11_8, 0, op4,
3059 OpcodeStr, !strconcat(Dt, "16"), v8i16, ShOp> {
3060 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3062 def v4i32 : N2VQShAdd<op24, op23, op11_8, 0, op4,
3063 OpcodeStr, !strconcat(Dt, "32"), v4i32, ShOp> {
3064 let Inst{21} = 0b1; // imm6 = 1xxxxx
3066 def v2i64 : N2VQShAdd<op24, op23, op11_8, 1, op4,
3067 OpcodeStr, !strconcat(Dt, "64"), v2i64, ShOp>;
3072 // Neon Shift-Insert vector operations,
3073 // with f of either N2RegVShLFrm or N2RegVShRFrm
3074 // element sizes of 8, 16, 32 and 64 bits:
3075 multiclass N2VShIns_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3076 string OpcodeStr, SDNode ShOp,
3078 // 64-bit vector types.
3079 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4,
3080 f, OpcodeStr, "8", v8i8, ShOp> {
3081 let Inst{21-19} = 0b001; // imm6 = 001xxx
3083 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4,
3084 f, OpcodeStr, "16", v4i16, ShOp> {
3085 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3087 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4,
3088 f, OpcodeStr, "32", v2i32, ShOp> {
3089 let Inst{21} = 0b1; // imm6 = 1xxxxx
3091 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4,
3092 f, OpcodeStr, "64", v1i64, ShOp>;
3095 // 128-bit vector types.
3096 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4,
3097 f, OpcodeStr, "8", v16i8, ShOp> {
3098 let Inst{21-19} = 0b001; // imm6 = 001xxx
3100 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4,
3101 f, OpcodeStr, "16", v8i16, ShOp> {
3102 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3104 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4,
3105 f, OpcodeStr, "32", v4i32, ShOp> {
3106 let Inst{21} = 0b1; // imm6 = 1xxxxx
3108 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4,
3109 f, OpcodeStr, "64", v2i64, ShOp>;
3113 // Neon Shift Long operations,
3114 // element sizes of 8, 16, 32 bits:
3115 multiclass N2VLSh_QHS<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
3116 bit op4, string OpcodeStr, string Dt, SDNode OpNode> {
3117 def v8i16 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3118 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode> {
3119 let Inst{21-19} = 0b001; // imm6 = 001xxx
3121 def v4i32 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3122 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode> {
3123 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3125 def v2i64 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3126 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode> {
3127 let Inst{21} = 0b1; // imm6 = 1xxxxx
3131 // Neon Shift Narrow operations,
3132 // element sizes of 16, 32, 64 bits:
3133 multiclass N2VNSh_HSD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
3134 bit op4, InstrItinClass itin, string OpcodeStr, string Dt,
3136 def v8i8 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3137 OpcodeStr, !strconcat(Dt, "16"), v8i8, v8i16, OpNode> {
3138 let Inst{21-19} = 0b001; // imm6 = 001xxx
3140 def v4i16 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3141 OpcodeStr, !strconcat(Dt, "32"), v4i16, v4i32, OpNode> {
3142 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3144 def v2i32 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3145 OpcodeStr, !strconcat(Dt, "64"), v2i32, v2i64, OpNode> {
3146 let Inst{21} = 0b1; // imm6 = 1xxxxx
3150 //===----------------------------------------------------------------------===//
3151 // Instruction Definitions.
3152 //===----------------------------------------------------------------------===//
3154 // Vector Add Operations.
3156 // VADD : Vector Add (integer and floating-point)
3157 defm VADD : N3V_QHSD<0, 0, 0b1000, 0, IIC_VBINiD, IIC_VBINiQ, "vadd", "i",
3159 def VADDfd : N3VD<0, 0, 0b00, 0b1101, 0, IIC_VBIND, "vadd", "f32",
3160 v2f32, v2f32, fadd, 1>;
3161 def VADDfq : N3VQ<0, 0, 0b00, 0b1101, 0, IIC_VBINQ, "vadd", "f32",
3162 v4f32, v4f32, fadd, 1>;
3163 // VADDL : Vector Add Long (Q = D + D)
3164 defm VADDLs : N3VLExt_QHS<0,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
3165 "vaddl", "s", add, sext, 1>;
3166 defm VADDLu : N3VLExt_QHS<1,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
3167 "vaddl", "u", add, zext, 1>;
3168 // VADDW : Vector Add Wide (Q = Q + D)
3169 defm VADDWs : N3VW_QHS<0,1,0b0001,0, "vaddw", "s", add, sext, 0>;
3170 defm VADDWu : N3VW_QHS<1,1,0b0001,0, "vaddw", "u", add, zext, 0>;
3171 // VHADD : Vector Halving Add
3172 defm VHADDs : N3VInt_QHS<0, 0, 0b0000, 0, N3RegFrm,
3173 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3174 "vhadd", "s", int_arm_neon_vhadds, 1>;
3175 defm VHADDu : N3VInt_QHS<1, 0, 0b0000, 0, N3RegFrm,
3176 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3177 "vhadd", "u", int_arm_neon_vhaddu, 1>;
3178 // VRHADD : Vector Rounding Halving Add
3179 defm VRHADDs : N3VInt_QHS<0, 0, 0b0001, 0, N3RegFrm,
3180 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3181 "vrhadd", "s", int_arm_neon_vrhadds, 1>;
3182 defm VRHADDu : N3VInt_QHS<1, 0, 0b0001, 0, N3RegFrm,
3183 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3184 "vrhadd", "u", int_arm_neon_vrhaddu, 1>;
3185 // VQADD : Vector Saturating Add
3186 defm VQADDs : N3VInt_QHSD<0, 0, 0b0000, 1, N3RegFrm,
3187 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3188 "vqadd", "s", int_arm_neon_vqadds, 1>;
3189 defm VQADDu : N3VInt_QHSD<1, 0, 0b0000, 1, N3RegFrm,
3190 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3191 "vqadd", "u", int_arm_neon_vqaddu, 1>;
3192 // VADDHN : Vector Add and Narrow Returning High Half (D = Q + Q)
3193 defm VADDHN : N3VNInt_HSD<0,1,0b0100,0, "vaddhn", "i",
3194 int_arm_neon_vaddhn, 1>;
3195 // VRADDHN : Vector Rounding Add and Narrow Returning High Half (D = Q + Q)
3196 defm VRADDHN : N3VNInt_HSD<1,1,0b0100,0, "vraddhn", "i",
3197 int_arm_neon_vraddhn, 1>;
3199 // Vector Multiply Operations.
3201 // VMUL : Vector Multiply (integer, polynomial and floating-point)
3202 defm VMUL : N3V_QHS<0, 0, 0b1001, 1, IIC_VMULi16D, IIC_VMULi32D,
3203 IIC_VMULi16Q, IIC_VMULi32Q, "vmul", "i", mul, 1>;
3204 def VMULpd : N3VDInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16D, "vmul",
3205 "p8", v8i8, v8i8, int_arm_neon_vmulp, 1>;
3206 def VMULpq : N3VQInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16Q, "vmul",
3207 "p8", v16i8, v16i8, int_arm_neon_vmulp, 1>;
3208 def VMULfd : N3VD<1, 0, 0b00, 0b1101, 1, IIC_VFMULD, "vmul", "f32",
3209 v2f32, v2f32, fmul, 1>;
3210 def VMULfq : N3VQ<1, 0, 0b00, 0b1101, 1, IIC_VFMULQ, "vmul", "f32",
3211 v4f32, v4f32, fmul, 1>;
3212 defm VMULsl : N3VSL_HS<0b1000, "vmul", "i", mul>;
3213 def VMULslfd : N3VDSL<0b10, 0b1001, IIC_VBIND, "vmul", "f32", v2f32, fmul>;
3214 def VMULslfq : N3VQSL<0b10, 0b1001, IIC_VBINQ, "vmul", "f32", v4f32,
3217 def : Pat<(v8i16 (mul (v8i16 QPR:$src1),
3218 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
3219 (v8i16 (VMULslv8i16 (v8i16 QPR:$src1),
3220 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3221 (DSubReg_i16_reg imm:$lane))),
3222 (SubReg_i16_lane imm:$lane)))>;
3223 def : Pat<(v4i32 (mul (v4i32 QPR:$src1),
3224 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
3225 (v4i32 (VMULslv4i32 (v4i32 QPR:$src1),
3226 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3227 (DSubReg_i32_reg imm:$lane))),
3228 (SubReg_i32_lane imm:$lane)))>;
3229 def : Pat<(v4f32 (fmul (v4f32 QPR:$src1),
3230 (v4f32 (NEONvduplane (v4f32 QPR:$src2), imm:$lane)))),
3231 (v4f32 (VMULslfq (v4f32 QPR:$src1),
3232 (v2f32 (EXTRACT_SUBREG QPR:$src2,
3233 (DSubReg_i32_reg imm:$lane))),
3234 (SubReg_i32_lane imm:$lane)))>;
3236 // VQDMULH : Vector Saturating Doubling Multiply Returning High Half
3237 defm VQDMULH : N3VInt_HS<0, 0, 0b1011, 0, N3RegFrm, IIC_VMULi16D, IIC_VMULi32D,
3238 IIC_VMULi16Q, IIC_VMULi32Q,
3239 "vqdmulh", "s", int_arm_neon_vqdmulh, 1>;
3240 defm VQDMULHsl: N3VIntSL_HS<0b1100, IIC_VMULi16D, IIC_VMULi32D,
3241 IIC_VMULi16Q, IIC_VMULi32Q,
3242 "vqdmulh", "s", int_arm_neon_vqdmulh>;
3243 def : Pat<(v8i16 (int_arm_neon_vqdmulh (v8i16 QPR:$src1),
3244 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
3246 (v8i16 (VQDMULHslv8i16 (v8i16 QPR:$src1),
3247 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3248 (DSubReg_i16_reg imm:$lane))),
3249 (SubReg_i16_lane imm:$lane)))>;
3250 def : Pat<(v4i32 (int_arm_neon_vqdmulh (v4i32 QPR:$src1),
3251 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
3253 (v4i32 (VQDMULHslv4i32 (v4i32 QPR:$src1),
3254 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3255 (DSubReg_i32_reg imm:$lane))),
3256 (SubReg_i32_lane imm:$lane)))>;
3258 // VQRDMULH : Vector Rounding Saturating Doubling Multiply Returning High Half
3259 defm VQRDMULH : N3VInt_HS<1, 0, 0b1011, 0, N3RegFrm,
3260 IIC_VMULi16D,IIC_VMULi32D,IIC_VMULi16Q,IIC_VMULi32Q,
3261 "vqrdmulh", "s", int_arm_neon_vqrdmulh, 1>;
3262 defm VQRDMULHsl : N3VIntSL_HS<0b1101, IIC_VMULi16D, IIC_VMULi32D,
3263 IIC_VMULi16Q, IIC_VMULi32Q,
3264 "vqrdmulh", "s", int_arm_neon_vqrdmulh>;
3265 def : Pat<(v8i16 (int_arm_neon_vqrdmulh (v8i16 QPR:$src1),
3266 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
3268 (v8i16 (VQRDMULHslv8i16 (v8i16 QPR:$src1),
3269 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3270 (DSubReg_i16_reg imm:$lane))),
3271 (SubReg_i16_lane imm:$lane)))>;
3272 def : Pat<(v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$src1),
3273 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
3275 (v4i32 (VQRDMULHslv4i32 (v4i32 QPR:$src1),
3276 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3277 (DSubReg_i32_reg imm:$lane))),
3278 (SubReg_i32_lane imm:$lane)))>;
3280 // VMULL : Vector Multiply Long (integer and polynomial) (Q = D * D)
3281 defm VMULLs : N3VL_QHS<0,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
3282 "vmull", "s", NEONvmulls, 1>;
3283 defm VMULLu : N3VL_QHS<1,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
3284 "vmull", "u", NEONvmullu, 1>;
3285 def VMULLp : N3VLInt<0, 1, 0b00, 0b1110, 0, IIC_VMULi16D, "vmull", "p8",
3286 v8i16, v8i8, int_arm_neon_vmullp, 1>;
3287 defm VMULLsls : N3VLSL_HS<0, 0b1010, IIC_VMULi16D, "vmull", "s", NEONvmulls>;
3288 defm VMULLslu : N3VLSL_HS<1, 0b1010, IIC_VMULi16D, "vmull", "u", NEONvmullu>;
3290 // VQDMULL : Vector Saturating Doubling Multiply Long (Q = D * D)
3291 defm VQDMULL : N3VLInt_HS<0,1,0b1101,0, IIC_VMULi16D, IIC_VMULi32D,
3292 "vqdmull", "s", int_arm_neon_vqdmull, 1>;
3293 defm VQDMULLsl: N3VLIntSL_HS<0, 0b1011, IIC_VMULi16D,
3294 "vqdmull", "s", int_arm_neon_vqdmull>;
3296 // Vector Multiply-Accumulate and Multiply-Subtract Operations.
3298 // VMLA : Vector Multiply Accumulate (integer and floating-point)
3299 defm VMLA : N3VMulOp_QHS<0, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3300 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
3301 def VMLAfd : N3VDMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACD, "vmla", "f32",
3302 v2f32, fmul_su, fadd_mlx>,
3303 Requires<[HasNEON, UseFPVMLx]>;
3304 def VMLAfq : N3VQMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACQ, "vmla", "f32",
3305 v4f32, fmul_su, fadd_mlx>,
3306 Requires<[HasNEON, UseFPVMLx]>;
3307 defm VMLAsl : N3VMulOpSL_HS<0b0000, IIC_VMACi16D, IIC_VMACi32D,
3308 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
3309 def VMLAslfd : N3VDMulOpSL<0b10, 0b0001, IIC_VMACD, "vmla", "f32",
3310 v2f32, fmul_su, fadd_mlx>,
3311 Requires<[HasNEON, UseFPVMLx]>;
3312 def VMLAslfq : N3VQMulOpSL<0b10, 0b0001, IIC_VMACQ, "vmla", "f32",
3313 v4f32, v2f32, fmul_su, fadd_mlx>,
3314 Requires<[HasNEON, UseFPVMLx]>;
3316 def : Pat<(v8i16 (add (v8i16 QPR:$src1),
3317 (mul (v8i16 QPR:$src2),
3318 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
3319 (v8i16 (VMLAslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
3320 (v4i16 (EXTRACT_SUBREG QPR:$src3,
3321 (DSubReg_i16_reg imm:$lane))),
3322 (SubReg_i16_lane imm:$lane)))>;
3324 def : Pat<(v4i32 (add (v4i32 QPR:$src1),
3325 (mul (v4i32 QPR:$src2),
3326 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
3327 (v4i32 (VMLAslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
3328 (v2i32 (EXTRACT_SUBREG QPR:$src3,
3329 (DSubReg_i32_reg imm:$lane))),
3330 (SubReg_i32_lane imm:$lane)))>;
3332 def : Pat<(v4f32 (fadd_mlx (v4f32 QPR:$src1),
3333 (fmul_su (v4f32 QPR:$src2),
3334 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
3335 (v4f32 (VMLAslfq (v4f32 QPR:$src1),
3337 (v2f32 (EXTRACT_SUBREG QPR:$src3,
3338 (DSubReg_i32_reg imm:$lane))),
3339 (SubReg_i32_lane imm:$lane)))>,
3340 Requires<[HasNEON, UseFPVMLx]>;
3342 // VMLAL : Vector Multiply Accumulate Long (Q += D * D)
3343 defm VMLALs : N3VLMulOp_QHS<0,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
3344 "vmlal", "s", NEONvmulls, add>;
3345 defm VMLALu : N3VLMulOp_QHS<1,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
3346 "vmlal", "u", NEONvmullu, add>;
3348 defm VMLALsls : N3VLMulOpSL_HS<0, 0b0010, "vmlal", "s", NEONvmulls, add>;
3349 defm VMLALslu : N3VLMulOpSL_HS<1, 0b0010, "vmlal", "u", NEONvmullu, add>;
3351 // VQDMLAL : Vector Saturating Doubling Multiply Accumulate Long (Q += D * D)
3352 defm VQDMLAL : N3VLInt3_HS<0, 1, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3353 "vqdmlal", "s", int_arm_neon_vqdmlal>;
3354 defm VQDMLALsl: N3VLInt3SL_HS<0, 0b0011, "vqdmlal", "s", int_arm_neon_vqdmlal>;
3356 // VMLS : Vector Multiply Subtract (integer and floating-point)
3357 defm VMLS : N3VMulOp_QHS<1, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3358 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
3359 def VMLSfd : N3VDMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACD, "vmls", "f32",
3360 v2f32, fmul_su, fsub_mlx>,
3361 Requires<[HasNEON, UseFPVMLx]>;
3362 def VMLSfq : N3VQMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACQ, "vmls", "f32",
3363 v4f32, fmul_su, fsub_mlx>,
3364 Requires<[HasNEON, UseFPVMLx]>;
3365 defm VMLSsl : N3VMulOpSL_HS<0b0100, IIC_VMACi16D, IIC_VMACi32D,
3366 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
3367 def VMLSslfd : N3VDMulOpSL<0b10, 0b0101, IIC_VMACD, "vmls", "f32",
3368 v2f32, fmul_su, fsub_mlx>,
3369 Requires<[HasNEON, UseFPVMLx]>;
3370 def VMLSslfq : N3VQMulOpSL<0b10, 0b0101, IIC_VMACQ, "vmls", "f32",
3371 v4f32, v2f32, fmul_su, fsub_mlx>,
3372 Requires<[HasNEON, UseFPVMLx]>;
3374 def : Pat<(v8i16 (sub (v8i16 QPR:$src1),
3375 (mul (v8i16 QPR:$src2),
3376 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
3377 (v8i16 (VMLSslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
3378 (v4i16 (EXTRACT_SUBREG QPR:$src3,
3379 (DSubReg_i16_reg imm:$lane))),
3380 (SubReg_i16_lane imm:$lane)))>;
3382 def : Pat<(v4i32 (sub (v4i32 QPR:$src1),
3383 (mul (v4i32 QPR:$src2),
3384 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
3385 (v4i32 (VMLSslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
3386 (v2i32 (EXTRACT_SUBREG QPR:$src3,
3387 (DSubReg_i32_reg imm:$lane))),
3388 (SubReg_i32_lane imm:$lane)))>;
3390 def : Pat<(v4f32 (fsub_mlx (v4f32 QPR:$src1),
3391 (fmul_su (v4f32 QPR:$src2),
3392 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
3393 (v4f32 (VMLSslfq (v4f32 QPR:$src1), (v4f32 QPR:$src2),
3394 (v2f32 (EXTRACT_SUBREG QPR:$src3,
3395 (DSubReg_i32_reg imm:$lane))),
3396 (SubReg_i32_lane imm:$lane)))>,
3397 Requires<[HasNEON, UseFPVMLx]>;
3399 // VMLSL : Vector Multiply Subtract Long (Q -= D * D)
3400 defm VMLSLs : N3VLMulOp_QHS<0,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
3401 "vmlsl", "s", NEONvmulls, sub>;
3402 defm VMLSLu : N3VLMulOp_QHS<1,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
3403 "vmlsl", "u", NEONvmullu, sub>;
3405 defm VMLSLsls : N3VLMulOpSL_HS<0, 0b0110, "vmlsl", "s", NEONvmulls, sub>;
3406 defm VMLSLslu : N3VLMulOpSL_HS<1, 0b0110, "vmlsl", "u", NEONvmullu, sub>;
3408 // VQDMLSL : Vector Saturating Doubling Multiply Subtract Long (Q -= D * D)
3409 defm VQDMLSL : N3VLInt3_HS<0, 1, 0b1011, 0, IIC_VMACi16D, IIC_VMACi32D,
3410 "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
3411 defm VQDMLSLsl: N3VLInt3SL_HS<0, 0b111, "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
3413 // Vector Subtract Operations.
3415 // VSUB : Vector Subtract (integer and floating-point)
3416 defm VSUB : N3V_QHSD<1, 0, 0b1000, 0, IIC_VSUBiD, IIC_VSUBiQ,
3417 "vsub", "i", sub, 0>;
3418 def VSUBfd : N3VD<0, 0, 0b10, 0b1101, 0, IIC_VBIND, "vsub", "f32",
3419 v2f32, v2f32, fsub, 0>;
3420 def VSUBfq : N3VQ<0, 0, 0b10, 0b1101, 0, IIC_VBINQ, "vsub", "f32",
3421 v4f32, v4f32, fsub, 0>;
3422 // VSUBL : Vector Subtract Long (Q = D - D)
3423 defm VSUBLs : N3VLExt_QHS<0,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
3424 "vsubl", "s", sub, sext, 0>;
3425 defm VSUBLu : N3VLExt_QHS<1,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
3426 "vsubl", "u", sub, zext, 0>;
3427 // VSUBW : Vector Subtract Wide (Q = Q - D)
3428 defm VSUBWs : N3VW_QHS<0,1,0b0011,0, "vsubw", "s", sub, sext, 0>;
3429 defm VSUBWu : N3VW_QHS<1,1,0b0011,0, "vsubw", "u", sub, zext, 0>;
3430 // VHSUB : Vector Halving Subtract
3431 defm VHSUBs : N3VInt_QHS<0, 0, 0b0010, 0, N3RegFrm,
3432 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3433 "vhsub", "s", int_arm_neon_vhsubs, 0>;
3434 defm VHSUBu : N3VInt_QHS<1, 0, 0b0010, 0, N3RegFrm,
3435 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3436 "vhsub", "u", int_arm_neon_vhsubu, 0>;
3437 // VQSUB : Vector Saturing Subtract
3438 defm VQSUBs : N3VInt_QHSD<0, 0, 0b0010, 1, N3RegFrm,
3439 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3440 "vqsub", "s", int_arm_neon_vqsubs, 0>;
3441 defm VQSUBu : N3VInt_QHSD<1, 0, 0b0010, 1, N3RegFrm,
3442 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3443 "vqsub", "u", int_arm_neon_vqsubu, 0>;
3444 // VSUBHN : Vector Subtract and Narrow Returning High Half (D = Q - Q)
3445 defm VSUBHN : N3VNInt_HSD<0,1,0b0110,0, "vsubhn", "i",
3446 int_arm_neon_vsubhn, 0>;
3447 // VRSUBHN : Vector Rounding Subtract and Narrow Returning High Half (D=Q-Q)
3448 defm VRSUBHN : N3VNInt_HSD<1,1,0b0110,0, "vrsubhn", "i",
3449 int_arm_neon_vrsubhn, 0>;
3451 // Vector Comparisons.
3453 // VCEQ : Vector Compare Equal
3454 defm VCEQ : N3V_QHS<1, 0, 0b1000, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3455 IIC_VSUBi4Q, "vceq", "i", NEONvceq, 1>;
3456 def VCEQfd : N3VD<0,0,0b00,0b1110,0, IIC_VBIND, "vceq", "f32", v2i32, v2f32,
3458 def VCEQfq : N3VQ<0,0,0b00,0b1110,0, IIC_VBINQ, "vceq", "f32", v4i32, v4f32,
3461 defm VCEQz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00010, 0, "vceq", "i",
3462 "$Vd, $Vm, #0", NEONvceqz>;
3464 // VCGE : Vector Compare Greater Than or Equal
3465 defm VCGEs : N3V_QHS<0, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3466 IIC_VSUBi4Q, "vcge", "s", NEONvcge, 0>;
3467 defm VCGEu : N3V_QHS<1, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3468 IIC_VSUBi4Q, "vcge", "u", NEONvcgeu, 0>;
3469 def VCGEfd : N3VD<1,0,0b00,0b1110,0, IIC_VBIND, "vcge", "f32", v2i32, v2f32,
3471 def VCGEfq : N3VQ<1,0,0b00,0b1110,0, IIC_VBINQ, "vcge", "f32", v4i32, v4f32,
3474 defm VCGEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00001, 0, "vcge", "s",
3475 "$Vd, $Vm, #0", NEONvcgez>;
3476 defm VCLEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00011, 0, "vcle", "s",
3477 "$Vd, $Vm, #0", NEONvclez>;
3479 // VCGT : Vector Compare Greater Than
3480 defm VCGTs : N3V_QHS<0, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3481 IIC_VSUBi4Q, "vcgt", "s", NEONvcgt, 0>;
3482 defm VCGTu : N3V_QHS<1, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3483 IIC_VSUBi4Q, "vcgt", "u", NEONvcgtu, 0>;
3484 def VCGTfd : N3VD<1,0,0b10,0b1110,0, IIC_VBIND, "vcgt", "f32", v2i32, v2f32,
3486 def VCGTfq : N3VQ<1,0,0b10,0b1110,0, IIC_VBINQ, "vcgt", "f32", v4i32, v4f32,
3489 defm VCGTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00000, 0, "vcgt", "s",
3490 "$Vd, $Vm, #0", NEONvcgtz>;
3491 defm VCLTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00100, 0, "vclt", "s",
3492 "$Vd, $Vm, #0", NEONvcltz>;
3494 // VACGE : Vector Absolute Compare Greater Than or Equal (aka VCAGE)
3495 def VACGEd : N3VDInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacge",
3496 "f32", v2i32, v2f32, int_arm_neon_vacged, 0>;
3497 def VACGEq : N3VQInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacge",
3498 "f32", v4i32, v4f32, int_arm_neon_vacgeq, 0>;
3499 // VACGT : Vector Absolute Compare Greater Than (aka VCAGT)
3500 def VACGTd : N3VDInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacgt",
3501 "f32", v2i32, v2f32, int_arm_neon_vacgtd, 0>;
3502 def VACGTq : N3VQInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacgt",
3503 "f32", v4i32, v4f32, int_arm_neon_vacgtq, 0>;
3504 // VTST : Vector Test Bits
3505 defm VTST : N3V_QHS<0, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
3506 IIC_VBINi4Q, "vtst", "", NEONvtst, 1>;
3508 // Vector Bitwise Operations.
3510 def vnotd : PatFrag<(ops node:$in),
3511 (xor node:$in, (bitconvert (v8i8 NEONimmAllOnesV)))>;
3512 def vnotq : PatFrag<(ops node:$in),
3513 (xor node:$in, (bitconvert (v16i8 NEONimmAllOnesV)))>;
3516 // VAND : Vector Bitwise AND
3517 def VANDd : N3VDX<0, 0, 0b00, 0b0001, 1, IIC_VBINiD, "vand",
3518 v2i32, v2i32, and, 1>;
3519 def VANDq : N3VQX<0, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "vand",
3520 v4i32, v4i32, and, 1>;
3522 // VEOR : Vector Bitwise Exclusive OR
3523 def VEORd : N3VDX<1, 0, 0b00, 0b0001, 1, IIC_VBINiD, "veor",
3524 v2i32, v2i32, xor, 1>;
3525 def VEORq : N3VQX<1, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "veor",
3526 v4i32, v4i32, xor, 1>;
3528 // VORR : Vector Bitwise OR
3529 def VORRd : N3VDX<0, 0, 0b10, 0b0001, 1, IIC_VBINiD, "vorr",
3530 v2i32, v2i32, or, 1>;
3531 def VORRq : N3VQX<0, 0, 0b10, 0b0001, 1, IIC_VBINiQ, "vorr",
3532 v4i32, v4i32, or, 1>;
3534 def VORRiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 0, 1,
3535 (outs DPR:$Vd), (ins nModImm:$SIMM, DPR:$src),
3537 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
3539 (v4i16 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
3540 let Inst{9} = SIMM{9};
3543 def VORRiv2i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 0, 0, 1,
3544 (outs DPR:$Vd), (ins nModImm:$SIMM, DPR:$src),
3546 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
3548 (v2i32 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
3549 let Inst{10-9} = SIMM{10-9};
3552 def VORRiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 0, 1,
3553 (outs QPR:$Vd), (ins nModImm:$SIMM, QPR:$src),
3555 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
3557 (v8i16 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
3558 let Inst{9} = SIMM{9};
3561 def VORRiv4i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 1, 0, 1,
3562 (outs QPR:$Vd), (ins nModImm:$SIMM, QPR:$src),
3564 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
3566 (v4i32 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
3567 let Inst{10-9} = SIMM{10-9};
3571 // VBIC : Vector Bitwise Bit Clear (AND NOT)
3572 def VBICd : N3VX<0, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
3573 (ins DPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VBINiD,
3574 "vbic", "$Vd, $Vn, $Vm", "",
3575 [(set DPR:$Vd, (v2i32 (and DPR:$Vn,
3576 (vnotd DPR:$Vm))))]>;
3577 def VBICq : N3VX<0, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
3578 (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINiQ,
3579 "vbic", "$Vd, $Vn, $Vm", "",
3580 [(set QPR:$Vd, (v4i32 (and QPR:$Vn,
3581 (vnotq QPR:$Vm))))]>;
3583 def VBICiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 1, 1,
3584 (outs DPR:$Vd), (ins nModImm:$SIMM, DPR:$src),
3586 "vbic", "i16", "$Vd, $SIMM", "$src = $Vd",
3588 (v4i16 (NEONvbicImm DPR:$src, timm:$SIMM)))]> {
3589 let Inst{9} = SIMM{9};
3592 def VBICiv2i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 0, 1, 1,
3593 (outs DPR:$Vd), (ins nModImm:$SIMM, DPR:$src),
3595 "vbic", "i32", "$Vd, $SIMM", "$src = $Vd",
3597 (v2i32 (NEONvbicImm DPR:$src, timm:$SIMM)))]> {
3598 let Inst{10-9} = SIMM{10-9};
3601 def VBICiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 1, 1,
3602 (outs QPR:$Vd), (ins nModImm:$SIMM, QPR:$src),
3604 "vbic", "i16", "$Vd, $SIMM", "$src = $Vd",
3606 (v8i16 (NEONvbicImm QPR:$src, timm:$SIMM)))]> {
3607 let Inst{9} = SIMM{9};
3610 def VBICiv4i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 1, 1, 1,
3611 (outs QPR:$Vd), (ins nModImm:$SIMM, QPR:$src),
3613 "vbic", "i32", "$Vd, $SIMM", "$src = $Vd",
3615 (v4i32 (NEONvbicImm QPR:$src, timm:$SIMM)))]> {
3616 let Inst{10-9} = SIMM{10-9};
3619 // VORN : Vector Bitwise OR NOT
3620 def VORNd : N3VX<0, 0, 0b11, 0b0001, 0, 1, (outs DPR:$Vd),
3621 (ins DPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VBINiD,
3622 "vorn", "$Vd, $Vn, $Vm", "",
3623 [(set DPR:$Vd, (v2i32 (or DPR:$Vn,
3624 (vnotd DPR:$Vm))))]>;
3625 def VORNq : N3VX<0, 0, 0b11, 0b0001, 1, 1, (outs QPR:$Vd),
3626 (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINiQ,
3627 "vorn", "$Vd, $Vn, $Vm", "",
3628 [(set QPR:$Vd, (v4i32 (or QPR:$Vn,
3629 (vnotq QPR:$Vm))))]>;
3631 // VMVN : Vector Bitwise NOT (Immediate)
3633 let isReMaterializable = 1 in {
3635 def VMVNv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 1, 1, (outs DPR:$Vd),
3636 (ins nModImm:$SIMM), IIC_VMOVImm,
3637 "vmvn", "i16", "$Vd, $SIMM", "",
3638 [(set DPR:$Vd, (v4i16 (NEONvmvnImm timm:$SIMM)))]> {
3639 let Inst{9} = SIMM{9};
3642 def VMVNv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 1, 1, (outs QPR:$Vd),
3643 (ins nModImm:$SIMM), IIC_VMOVImm,
3644 "vmvn", "i16", "$Vd, $SIMM", "",
3645 [(set QPR:$Vd, (v8i16 (NEONvmvnImm timm:$SIMM)))]> {
3646 let Inst{9} = SIMM{9};
3649 def VMVNv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 1, 1, (outs DPR:$Vd),
3650 (ins nModImm:$SIMM), IIC_VMOVImm,
3651 "vmvn", "i32", "$Vd, $SIMM", "",
3652 [(set DPR:$Vd, (v2i32 (NEONvmvnImm timm:$SIMM)))]> {
3653 let Inst{11-8} = SIMM{11-8};
3656 def VMVNv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 1, 1, (outs QPR:$Vd),
3657 (ins nModImm:$SIMM), IIC_VMOVImm,
3658 "vmvn", "i32", "$Vd, $SIMM", "",
3659 [(set QPR:$Vd, (v4i32 (NEONvmvnImm timm:$SIMM)))]> {
3660 let Inst{11-8} = SIMM{11-8};
3664 // VMVN : Vector Bitwise NOT
3665 def VMVNd : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 0, 0,
3666 (outs DPR:$Vd), (ins DPR:$Vm), IIC_VSUBiD,
3667 "vmvn", "$Vd, $Vm", "",
3668 [(set DPR:$Vd, (v2i32 (vnotd DPR:$Vm)))]>;
3669 def VMVNq : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 1, 0,
3670 (outs QPR:$Vd), (ins QPR:$Vm), IIC_VSUBiD,
3671 "vmvn", "$Vd, $Vm", "",
3672 [(set QPR:$Vd, (v4i32 (vnotq QPR:$Vm)))]>;
3673 def : Pat<(v2i32 (vnotd DPR:$src)), (VMVNd DPR:$src)>;
3674 def : Pat<(v4i32 (vnotq QPR:$src)), (VMVNq QPR:$src)>;
3676 // VBSL : Vector Bitwise Select
3677 def VBSLd : N3VX<1, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
3678 (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
3679 N3RegFrm, IIC_VCNTiD,
3680 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3682 (v2i32 (or (and DPR:$Vn, DPR:$src1),
3683 (and DPR:$Vm, (vnotd DPR:$src1)))))]>;
3684 def VBSLq : N3VX<1, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
3685 (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
3686 N3RegFrm, IIC_VCNTiQ,
3687 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3689 (v4i32 (or (and QPR:$Vn, QPR:$src1),
3690 (and QPR:$Vm, (vnotq QPR:$src1)))))]>;
3692 // VBIF : Vector Bitwise Insert if False
3693 // like VBSL but with: "vbif $dst, $src3, $src1", "$src2 = $dst",
3694 // FIXME: This instruction's encoding MAY NOT BE correct.
3695 def VBIFd : N3VX<1, 0, 0b11, 0b0001, 0, 1,
3696 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
3697 N3RegFrm, IIC_VBINiD,
3698 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3699 [/* For disassembly only; pattern left blank */]>;
3700 def VBIFq : N3VX<1, 0, 0b11, 0b0001, 1, 1,
3701 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
3702 N3RegFrm, IIC_VBINiQ,
3703 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3704 [/* For disassembly only; pattern left blank */]>;
3706 // VBIT : Vector Bitwise Insert if True
3707 // like VBSL but with: "vbit $dst, $src2, $src1", "$src3 = $dst",
3708 // FIXME: This instruction's encoding MAY NOT BE correct.
3709 def VBITd : N3VX<1, 0, 0b10, 0b0001, 0, 1,
3710 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
3711 N3RegFrm, IIC_VBINiD,
3712 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3713 [/* For disassembly only; pattern left blank */]>;
3714 def VBITq : N3VX<1, 0, 0b10, 0b0001, 1, 1,
3715 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
3716 N3RegFrm, IIC_VBINiQ,
3717 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3718 [/* For disassembly only; pattern left blank */]>;
3720 // VBIT/VBIF are not yet implemented. The TwoAddress pass will not go looking
3721 // for equivalent operations with different register constraints; it just
3724 // Vector Absolute Differences.
3726 // VABD : Vector Absolute Difference
3727 defm VABDs : N3VInt_QHS<0, 0, 0b0111, 0, N3RegFrm,
3728 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3729 "vabd", "s", int_arm_neon_vabds, 1>;
3730 defm VABDu : N3VInt_QHS<1, 0, 0b0111, 0, N3RegFrm,
3731 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3732 "vabd", "u", int_arm_neon_vabdu, 1>;
3733 def VABDfd : N3VDInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBIND,
3734 "vabd", "f32", v2f32, v2f32, int_arm_neon_vabds, 1>;
3735 def VABDfq : N3VQInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBINQ,
3736 "vabd", "f32", v4f32, v4f32, int_arm_neon_vabds, 1>;
3738 // VABDL : Vector Absolute Difference Long (Q = | D - D |)
3739 defm VABDLs : N3VLIntExt_QHS<0,1,0b0111,0, IIC_VSUBi4Q,
3740 "vabdl", "s", int_arm_neon_vabds, zext, 1>;
3741 defm VABDLu : N3VLIntExt_QHS<1,1,0b0111,0, IIC_VSUBi4Q,
3742 "vabdl", "u", int_arm_neon_vabdu, zext, 1>;
3744 // VABA : Vector Absolute Difference and Accumulate
3745 defm VABAs : N3VIntOp_QHS<0,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
3746 "vaba", "s", int_arm_neon_vabds, add>;
3747 defm VABAu : N3VIntOp_QHS<1,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
3748 "vaba", "u", int_arm_neon_vabdu, add>;
3750 // VABAL : Vector Absolute Difference and Accumulate Long (Q += | D - D |)
3751 defm VABALs : N3VLIntExtOp_QHS<0,1,0b0101,0, IIC_VABAD,
3752 "vabal", "s", int_arm_neon_vabds, zext, add>;
3753 defm VABALu : N3VLIntExtOp_QHS<1,1,0b0101,0, IIC_VABAD,
3754 "vabal", "u", int_arm_neon_vabdu, zext, add>;
3756 // Vector Maximum and Minimum.
3758 // VMAX : Vector Maximum
3759 defm VMAXs : N3VInt_QHS<0, 0, 0b0110, 0, N3RegFrm,
3760 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3761 "vmax", "s", int_arm_neon_vmaxs, 1>;
3762 defm VMAXu : N3VInt_QHS<1, 0, 0b0110, 0, N3RegFrm,
3763 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3764 "vmax", "u", int_arm_neon_vmaxu, 1>;
3765 def VMAXfd : N3VDInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBIND,
3767 v2f32, v2f32, int_arm_neon_vmaxs, 1>;
3768 def VMAXfq : N3VQInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBINQ,
3770 v4f32, v4f32, int_arm_neon_vmaxs, 1>;
3772 // VMIN : Vector Minimum
3773 defm VMINs : N3VInt_QHS<0, 0, 0b0110, 1, N3RegFrm,
3774 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3775 "vmin", "s", int_arm_neon_vmins, 1>;
3776 defm VMINu : N3VInt_QHS<1, 0, 0b0110, 1, N3RegFrm,
3777 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3778 "vmin", "u", int_arm_neon_vminu, 1>;
3779 def VMINfd : N3VDInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBIND,
3781 v2f32, v2f32, int_arm_neon_vmins, 1>;
3782 def VMINfq : N3VQInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBINQ,
3784 v4f32, v4f32, int_arm_neon_vmins, 1>;
3786 // Vector Pairwise Operations.
3788 // VPADD : Vector Pairwise Add
3789 def VPADDi8 : N3VDInt<0, 0, 0b00, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
3791 v8i8, v8i8, int_arm_neon_vpadd, 0>;
3792 def VPADDi16 : N3VDInt<0, 0, 0b01, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
3794 v4i16, v4i16, int_arm_neon_vpadd, 0>;
3795 def VPADDi32 : N3VDInt<0, 0, 0b10, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
3797 v2i32, v2i32, int_arm_neon_vpadd, 0>;
3798 def VPADDf : N3VDInt<1, 0, 0b00, 0b1101, 0, N3RegFrm,
3799 IIC_VPBIND, "vpadd", "f32",
3800 v2f32, v2f32, int_arm_neon_vpadd, 0>;
3802 // VPADDL : Vector Pairwise Add Long
3803 defm VPADDLs : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00100, 0, "vpaddl", "s",
3804 int_arm_neon_vpaddls>;
3805 defm VPADDLu : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00101, 0, "vpaddl", "u",
3806 int_arm_neon_vpaddlu>;
3808 // VPADAL : Vector Pairwise Add and Accumulate Long
3809 defm VPADALs : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01100, 0, "vpadal", "s",
3810 int_arm_neon_vpadals>;
3811 defm VPADALu : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01101, 0, "vpadal", "u",
3812 int_arm_neon_vpadalu>;
3814 // VPMAX : Vector Pairwise Maximum
3815 def VPMAXs8 : N3VDInt<0, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3816 "s8", v8i8, v8i8, int_arm_neon_vpmaxs, 0>;
3817 def VPMAXs16 : N3VDInt<0, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3818 "s16", v4i16, v4i16, int_arm_neon_vpmaxs, 0>;
3819 def VPMAXs32 : N3VDInt<0, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3820 "s32", v2i32, v2i32, int_arm_neon_vpmaxs, 0>;
3821 def VPMAXu8 : N3VDInt<1, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3822 "u8", v8i8, v8i8, int_arm_neon_vpmaxu, 0>;
3823 def VPMAXu16 : N3VDInt<1, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3824 "u16", v4i16, v4i16, int_arm_neon_vpmaxu, 0>;
3825 def VPMAXu32 : N3VDInt<1, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3826 "u32", v2i32, v2i32, int_arm_neon_vpmaxu, 0>;
3827 def VPMAXf : N3VDInt<1, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmax",
3828 "f32", v2f32, v2f32, int_arm_neon_vpmaxs, 0>;
3830 // VPMIN : Vector Pairwise Minimum
3831 def VPMINs8 : N3VDInt<0, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3832 "s8", v8i8, v8i8, int_arm_neon_vpmins, 0>;
3833 def VPMINs16 : N3VDInt<0, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3834 "s16", v4i16, v4i16, int_arm_neon_vpmins, 0>;
3835 def VPMINs32 : N3VDInt<0, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3836 "s32", v2i32, v2i32, int_arm_neon_vpmins, 0>;
3837 def VPMINu8 : N3VDInt<1, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3838 "u8", v8i8, v8i8, int_arm_neon_vpminu, 0>;
3839 def VPMINu16 : N3VDInt<1, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3840 "u16", v4i16, v4i16, int_arm_neon_vpminu, 0>;
3841 def VPMINu32 : N3VDInt<1, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3842 "u32", v2i32, v2i32, int_arm_neon_vpminu, 0>;
3843 def VPMINf : N3VDInt<1, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmin",
3844 "f32", v2f32, v2f32, int_arm_neon_vpmins, 0>;
3846 // Vector Reciprocal and Reciprocal Square Root Estimate and Step.
3848 // VRECPE : Vector Reciprocal Estimate
3849 def VRECPEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
3850 IIC_VUNAD, "vrecpe", "u32",
3851 v2i32, v2i32, int_arm_neon_vrecpe>;
3852 def VRECPEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
3853 IIC_VUNAQ, "vrecpe", "u32",
3854 v4i32, v4i32, int_arm_neon_vrecpe>;
3855 def VRECPEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
3856 IIC_VUNAD, "vrecpe", "f32",
3857 v2f32, v2f32, int_arm_neon_vrecpe>;
3858 def VRECPEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
3859 IIC_VUNAQ, "vrecpe", "f32",
3860 v4f32, v4f32, int_arm_neon_vrecpe>;
3862 // VRECPS : Vector Reciprocal Step
3863 def VRECPSfd : N3VDInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
3864 IIC_VRECSD, "vrecps", "f32",
3865 v2f32, v2f32, int_arm_neon_vrecps, 1>;
3866 def VRECPSfq : N3VQInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
3867 IIC_VRECSQ, "vrecps", "f32",
3868 v4f32, v4f32, int_arm_neon_vrecps, 1>;
3870 // VRSQRTE : Vector Reciprocal Square Root Estimate
3871 def VRSQRTEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
3872 IIC_VUNAD, "vrsqrte", "u32",
3873 v2i32, v2i32, int_arm_neon_vrsqrte>;
3874 def VRSQRTEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
3875 IIC_VUNAQ, "vrsqrte", "u32",
3876 v4i32, v4i32, int_arm_neon_vrsqrte>;
3877 def VRSQRTEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
3878 IIC_VUNAD, "vrsqrte", "f32",
3879 v2f32, v2f32, int_arm_neon_vrsqrte>;
3880 def VRSQRTEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
3881 IIC_VUNAQ, "vrsqrte", "f32",
3882 v4f32, v4f32, int_arm_neon_vrsqrte>;
3884 // VRSQRTS : Vector Reciprocal Square Root Step
3885 def VRSQRTSfd : N3VDInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
3886 IIC_VRECSD, "vrsqrts", "f32",
3887 v2f32, v2f32, int_arm_neon_vrsqrts, 1>;
3888 def VRSQRTSfq : N3VQInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
3889 IIC_VRECSQ, "vrsqrts", "f32",
3890 v4f32, v4f32, int_arm_neon_vrsqrts, 1>;
3894 // VSHL : Vector Shift
3895 defm VSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 0, N3RegVShFrm,
3896 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
3897 "vshl", "s", int_arm_neon_vshifts>;
3898 defm VSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 0, N3RegVShFrm,
3899 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
3900 "vshl", "u", int_arm_neon_vshiftu>;
3901 // VSHL : Vector Shift Left (Immediate)
3902 defm VSHLi : N2VSh_QHSD<0, 1, 0b0101, 1, IIC_VSHLiD, "vshl", "i", NEONvshl,
3904 // VSHR : Vector Shift Right (Immediate)
3905 defm VSHRs : N2VSh_QHSD<0, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "s", NEONvshrs,
3907 defm VSHRu : N2VSh_QHSD<1, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "u", NEONvshru,
3910 // VSHLL : Vector Shift Left Long
3911 defm VSHLLs : N2VLSh_QHS<0, 1, 0b1010, 0, 0, 1, "vshll", "s", NEONvshlls>;
3912 defm VSHLLu : N2VLSh_QHS<1, 1, 0b1010, 0, 0, 1, "vshll", "u", NEONvshllu>;
3914 // VSHLL : Vector Shift Left Long (with maximum shift count)
3915 class N2VLShMax<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
3916 bit op6, bit op4, string OpcodeStr, string Dt, ValueType ResTy,
3917 ValueType OpTy, SDNode OpNode>
3918 : N2VLSh<op24, op23, op11_8, op7, op6, op4, OpcodeStr, Dt,
3919 ResTy, OpTy, OpNode> {
3920 let Inst{21-16} = op21_16;
3922 def VSHLLi8 : N2VLShMax<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll", "i8",
3923 v8i16, v8i8, NEONvshlli>;
3924 def VSHLLi16 : N2VLShMax<1, 1, 0b110110, 0b0011, 0, 0, 0, "vshll", "i16",
3925 v4i32, v4i16, NEONvshlli>;
3926 def VSHLLi32 : N2VLShMax<1, 1, 0b111010, 0b0011, 0, 0, 0, "vshll", "i32",
3927 v2i64, v2i32, NEONvshlli>;
3929 // VSHRN : Vector Shift Right and Narrow
3930 defm VSHRN : N2VNSh_HSD<0,1,0b1000,0,0,1, IIC_VSHLiD, "vshrn", "i",
3933 // VRSHL : Vector Rounding Shift
3934 defm VRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 0, N3RegVShFrm,
3935 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3936 "vrshl", "s", int_arm_neon_vrshifts>;
3937 defm VRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 0, N3RegVShFrm,
3938 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3939 "vrshl", "u", int_arm_neon_vrshiftu>;
3940 // VRSHR : Vector Rounding Shift Right
3941 defm VRSHRs : N2VSh_QHSD<0,1,0b0010,1, IIC_VSHLi4D, "vrshr", "s", NEONvrshrs,
3943 defm VRSHRu : N2VSh_QHSD<1,1,0b0010,1, IIC_VSHLi4D, "vrshr", "u", NEONvrshru,
3946 // VRSHRN : Vector Rounding Shift Right and Narrow
3947 defm VRSHRN : N2VNSh_HSD<0, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vrshrn", "i",
3950 // VQSHL : Vector Saturating Shift
3951 defm VQSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 1, N3RegVShFrm,
3952 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3953 "vqshl", "s", int_arm_neon_vqshifts>;
3954 defm VQSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 1, N3RegVShFrm,
3955 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3956 "vqshl", "u", int_arm_neon_vqshiftu>;
3957 // VQSHL : Vector Saturating Shift Left (Immediate)
3958 defm VQSHLsi : N2VSh_QHSD<0,1,0b0111,1, IIC_VSHLi4D, "vqshl", "s",NEONvqshls,
3960 defm VQSHLui : N2VSh_QHSD<1,1,0b0111,1, IIC_VSHLi4D, "vqshl", "u",NEONvqshlu,
3962 // VQSHLU : Vector Saturating Shift Left (Immediate, Unsigned)
3963 defm VQSHLsu : N2VSh_QHSD<1,1,0b0110,1, IIC_VSHLi4D,"vqshlu","s",NEONvqshlsu,
3966 // VQSHRN : Vector Saturating Shift Right and Narrow
3967 defm VQSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "s",
3969 defm VQSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "u",
3972 // VQSHRUN : Vector Saturating Shift Right and Narrow (Unsigned)
3973 defm VQSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 0, 1, IIC_VSHLi4D, "vqshrun", "s",
3976 // VQRSHL : Vector Saturating Rounding Shift
3977 defm VQRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 1, N3RegVShFrm,
3978 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3979 "vqrshl", "s", int_arm_neon_vqrshifts>;
3980 defm VQRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 1, N3RegVShFrm,
3981 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3982 "vqrshl", "u", int_arm_neon_vqrshiftu>;
3984 // VQRSHRN : Vector Saturating Rounding Shift Right and Narrow
3985 defm VQRSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "s",
3987 defm VQRSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "u",
3990 // VQRSHRUN : Vector Saturating Rounding Shift Right and Narrow (Unsigned)
3991 defm VQRSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vqrshrun", "s",
3994 // VSRA : Vector Shift Right and Accumulate
3995 defm VSRAs : N2VShAdd_QHSD<0, 1, 0b0001, 1, "vsra", "s", NEONvshrs>;
3996 defm VSRAu : N2VShAdd_QHSD<1, 1, 0b0001, 1, "vsra", "u", NEONvshru>;
3997 // VRSRA : Vector Rounding Shift Right and Accumulate
3998 defm VRSRAs : N2VShAdd_QHSD<0, 1, 0b0011, 1, "vrsra", "s", NEONvrshrs>;
3999 defm VRSRAu : N2VShAdd_QHSD<1, 1, 0b0011, 1, "vrsra", "u", NEONvrshru>;
4001 // VSLI : Vector Shift Left and Insert
4002 defm VSLI : N2VShIns_QHSD<1, 1, 0b0101, 1, "vsli", NEONvsli, N2RegVShLFrm>;
4003 // VSRI : Vector Shift Right and Insert
4004 defm VSRI : N2VShIns_QHSD<1, 1, 0b0100, 1, "vsri", NEONvsri, N2RegVShRFrm>;
4006 // Vector Absolute and Saturating Absolute.
4008 // VABS : Vector Absolute Value
4009 defm VABS : N2VInt_QHS<0b11, 0b11, 0b01, 0b00110, 0,
4010 IIC_VUNAiD, IIC_VUNAiQ, "vabs", "s",
4012 def VABSfd : N2VDInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
4013 IIC_VUNAD, "vabs", "f32",
4014 v2f32, v2f32, int_arm_neon_vabs>;
4015 def VABSfq : N2VQInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
4016 IIC_VUNAQ, "vabs", "f32",
4017 v4f32, v4f32, int_arm_neon_vabs>;
4019 // VQABS : Vector Saturating Absolute Value
4020 defm VQABS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01110, 0,
4021 IIC_VQUNAiD, IIC_VQUNAiQ, "vqabs", "s",
4022 int_arm_neon_vqabs>;
4026 def vnegd : PatFrag<(ops node:$in),
4027 (sub (bitconvert (v2i32 NEONimmAllZerosV)), node:$in)>;
4028 def vnegq : PatFrag<(ops node:$in),
4029 (sub (bitconvert (v4i32 NEONimmAllZerosV)), node:$in)>;
4031 class VNEGD<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
4032 : N2V<0b11, 0b11, size, 0b01, 0b00111, 0, 0, (outs DPR:$Vd), (ins DPR:$Vm),
4033 IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
4034 [(set DPR:$Vd, (Ty (vnegd DPR:$Vm)))]>;
4035 class VNEGQ<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
4036 : N2V<0b11, 0b11, size, 0b01, 0b00111, 1, 0, (outs QPR:$Vd), (ins QPR:$Vm),
4037 IIC_VSHLiQ, OpcodeStr, Dt, "$Vd, $Vm", "",
4038 [(set QPR:$Vd, (Ty (vnegq QPR:$Vm)))]>;
4040 // VNEG : Vector Negate (integer)
4041 def VNEGs8d : VNEGD<0b00, "vneg", "s8", v8i8>;
4042 def VNEGs16d : VNEGD<0b01, "vneg", "s16", v4i16>;
4043 def VNEGs32d : VNEGD<0b10, "vneg", "s32", v2i32>;
4044 def VNEGs8q : VNEGQ<0b00, "vneg", "s8", v16i8>;
4045 def VNEGs16q : VNEGQ<0b01, "vneg", "s16", v8i16>;
4046 def VNEGs32q : VNEGQ<0b10, "vneg", "s32", v4i32>;
4048 // VNEG : Vector Negate (floating-point)
4049 def VNEGfd : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
4050 (outs DPR:$Vd), (ins DPR:$Vm), IIC_VUNAD,
4051 "vneg", "f32", "$Vd, $Vm", "",
4052 [(set DPR:$Vd, (v2f32 (fneg DPR:$Vm)))]>;
4053 def VNEGf32q : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 1, 0,
4054 (outs QPR:$Vd), (ins QPR:$Vm), IIC_VUNAQ,
4055 "vneg", "f32", "$Vd, $Vm", "",
4056 [(set QPR:$Vd, (v4f32 (fneg QPR:$Vm)))]>;
4058 def : Pat<(v8i8 (vnegd DPR:$src)), (VNEGs8d DPR:$src)>;
4059 def : Pat<(v4i16 (vnegd DPR:$src)), (VNEGs16d DPR:$src)>;
4060 def : Pat<(v2i32 (vnegd DPR:$src)), (VNEGs32d DPR:$src)>;
4061 def : Pat<(v16i8 (vnegq QPR:$src)), (VNEGs8q QPR:$src)>;
4062 def : Pat<(v8i16 (vnegq QPR:$src)), (VNEGs16q QPR:$src)>;
4063 def : Pat<(v4i32 (vnegq QPR:$src)), (VNEGs32q QPR:$src)>;
4065 // VQNEG : Vector Saturating Negate
4066 defm VQNEG : N2VInt_QHS<0b11, 0b11, 0b00, 0b01111, 0,
4067 IIC_VQUNAiD, IIC_VQUNAiQ, "vqneg", "s",
4068 int_arm_neon_vqneg>;
4070 // Vector Bit Counting Operations.
4072 // VCLS : Vector Count Leading Sign Bits
4073 defm VCLS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01000, 0,
4074 IIC_VCNTiD, IIC_VCNTiQ, "vcls", "s",
4076 // VCLZ : Vector Count Leading Zeros
4077 defm VCLZ : N2VInt_QHS<0b11, 0b11, 0b00, 0b01001, 0,
4078 IIC_VCNTiD, IIC_VCNTiQ, "vclz", "i",
4080 // VCNT : Vector Count One Bits
4081 def VCNTd : N2VDInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
4082 IIC_VCNTiD, "vcnt", "8",
4083 v8i8, v8i8, int_arm_neon_vcnt>;
4084 def VCNTq : N2VQInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
4085 IIC_VCNTiQ, "vcnt", "8",
4086 v16i8, v16i8, int_arm_neon_vcnt>;
4088 // Vector Swap -- for disassembly only.
4089 def VSWPd : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 0, 0,
4090 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
4091 "vswp", "$Vd, $Vm", "", []>;
4092 def VSWPq : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 1, 0,
4093 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
4094 "vswp", "$Vd, $Vm", "", []>;
4096 // Vector Move Operations.
4098 // VMOV : Vector Move (Register)
4100 let neverHasSideEffects = 1 in {
4101 def VMOVDneon: N3VX<0, 0, 0b10, 0b0001, 0, 1, (outs DPR:$Vd), (ins DPR:$Vm),
4102 N3RegFrm, IIC_VMOV, "vmov", "$Vd, $Vm", "", []> {
4103 let Vn{4-0} = Vm{4-0};
4105 def VMOVQ : N3VX<0, 0, 0b10, 0b0001, 1, 1, (outs QPR:$Vd), (ins QPR:$Vm),
4106 N3RegFrm, IIC_VMOV, "vmov", "$Vd, $Vm", "", []> {
4107 let Vn{4-0} = Vm{4-0};
4110 // Pseudo vector move instructions for QQ and QQQQ registers. This should
4111 // be expanded after register allocation is completed.
4112 def VMOVQQ : PseudoInst<(outs QQPR:$dst), (ins QQPR:$src),
4115 def VMOVQQQQ : PseudoInst<(outs QQQQPR:$dst), (ins QQQQPR:$src),
4117 } // neverHasSideEffects
4119 // VMOV : Vector Move (Immediate)
4121 let isReMaterializable = 1 in {
4122 def VMOVv8i8 : N1ModImm<1, 0b000, 0b1110, 0, 0, 0, 1, (outs DPR:$Vd),
4123 (ins nModImm:$SIMM), IIC_VMOVImm,
4124 "vmov", "i8", "$Vd, $SIMM", "",
4125 [(set DPR:$Vd, (v8i8 (NEONvmovImm timm:$SIMM)))]>;
4126 def VMOVv16i8 : N1ModImm<1, 0b000, 0b1110, 0, 1, 0, 1, (outs QPR:$Vd),
4127 (ins nModImm:$SIMM), IIC_VMOVImm,
4128 "vmov", "i8", "$Vd, $SIMM", "",
4129 [(set QPR:$Vd, (v16i8 (NEONvmovImm timm:$SIMM)))]>;
4131 def VMOVv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 0, 1, (outs DPR:$Vd),
4132 (ins nModImm:$SIMM), IIC_VMOVImm,
4133 "vmov", "i16", "$Vd, $SIMM", "",
4134 [(set DPR:$Vd, (v4i16 (NEONvmovImm timm:$SIMM)))]> {
4135 let Inst{9} = SIMM{9};
4138 def VMOVv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 0, 1, (outs QPR:$Vd),
4139 (ins nModImm:$SIMM), IIC_VMOVImm,
4140 "vmov", "i16", "$Vd, $SIMM", "",
4141 [(set QPR:$Vd, (v8i16 (NEONvmovImm timm:$SIMM)))]> {
4142 let Inst{9} = SIMM{9};
4145 def VMOVv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 0, 1, (outs DPR:$Vd),
4146 (ins nModImm:$SIMM), IIC_VMOVImm,
4147 "vmov", "i32", "$Vd, $SIMM", "",
4148 [(set DPR:$Vd, (v2i32 (NEONvmovImm timm:$SIMM)))]> {
4149 let Inst{11-8} = SIMM{11-8};
4152 def VMOVv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 0, 1, (outs QPR:$Vd),
4153 (ins nModImm:$SIMM), IIC_VMOVImm,
4154 "vmov", "i32", "$Vd, $SIMM", "",
4155 [(set QPR:$Vd, (v4i32 (NEONvmovImm timm:$SIMM)))]> {
4156 let Inst{11-8} = SIMM{11-8};
4159 def VMOVv1i64 : N1ModImm<1, 0b000, 0b1110, 0, 0, 1, 1, (outs DPR:$Vd),
4160 (ins nModImm:$SIMM), IIC_VMOVImm,
4161 "vmov", "i64", "$Vd, $SIMM", "",
4162 [(set DPR:$Vd, (v1i64 (NEONvmovImm timm:$SIMM)))]>;
4163 def VMOVv2i64 : N1ModImm<1, 0b000, 0b1110, 0, 1, 1, 1, (outs QPR:$Vd),
4164 (ins nModImm:$SIMM), IIC_VMOVImm,
4165 "vmov", "i64", "$Vd, $SIMM", "",
4166 [(set QPR:$Vd, (v2i64 (NEONvmovImm timm:$SIMM)))]>;
4167 } // isReMaterializable
4169 // VMOV : Vector Get Lane (move scalar to ARM core register)
4171 def VGETLNs8 : NVGetLane<{1,1,1,0,0,1,?,1}, 0b1011, {?,?},
4172 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
4173 IIC_VMOVSI, "vmov", "s8", "$R, $V[$lane]",
4174 [(set GPR:$R, (NEONvgetlanes (v8i8 DPR:$V),
4176 let Inst{21} = lane{2};
4177 let Inst{6-5} = lane{1-0};
4179 def VGETLNs16 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, {?,1},
4180 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
4181 IIC_VMOVSI, "vmov", "s16", "$R, $V[$lane]",
4182 [(set GPR:$R, (NEONvgetlanes (v4i16 DPR:$V),
4184 let Inst{21} = lane{1};
4185 let Inst{6} = lane{0};
4187 def VGETLNu8 : NVGetLane<{1,1,1,0,1,1,?,1}, 0b1011, {?,?},
4188 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
4189 IIC_VMOVSI, "vmov", "u8", "$R, $V[$lane]",
4190 [(set GPR:$R, (NEONvgetlaneu (v8i8 DPR:$V),
4192 let Inst{21} = lane{2};
4193 let Inst{6-5} = lane{1-0};
4195 def VGETLNu16 : NVGetLane<{1,1,1,0,1,0,?,1}, 0b1011, {?,1},
4196 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
4197 IIC_VMOVSI, "vmov", "u16", "$R, $V[$lane]",
4198 [(set GPR:$R, (NEONvgetlaneu (v4i16 DPR:$V),
4200 let Inst{21} = lane{1};
4201 let Inst{6} = lane{0};
4203 def VGETLNi32 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, 0b00,
4204 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
4205 IIC_VMOVSI, "vmov", "32", "$R, $V[$lane]",
4206 [(set GPR:$R, (extractelt (v2i32 DPR:$V),
4208 let Inst{21} = lane{0};
4210 // def VGETLNf32: see FMRDH and FMRDL in ARMInstrVFP.td
4211 def : Pat<(NEONvgetlanes (v16i8 QPR:$src), imm:$lane),
4212 (VGETLNs8 (v8i8 (EXTRACT_SUBREG QPR:$src,
4213 (DSubReg_i8_reg imm:$lane))),
4214 (SubReg_i8_lane imm:$lane))>;
4215 def : Pat<(NEONvgetlanes (v8i16 QPR:$src), imm:$lane),
4216 (VGETLNs16 (v4i16 (EXTRACT_SUBREG QPR:$src,
4217 (DSubReg_i16_reg imm:$lane))),
4218 (SubReg_i16_lane imm:$lane))>;
4219 def : Pat<(NEONvgetlaneu (v16i8 QPR:$src), imm:$lane),
4220 (VGETLNu8 (v8i8 (EXTRACT_SUBREG QPR:$src,
4221 (DSubReg_i8_reg imm:$lane))),
4222 (SubReg_i8_lane imm:$lane))>;
4223 def : Pat<(NEONvgetlaneu (v8i16 QPR:$src), imm:$lane),
4224 (VGETLNu16 (v4i16 (EXTRACT_SUBREG QPR:$src,
4225 (DSubReg_i16_reg imm:$lane))),
4226 (SubReg_i16_lane imm:$lane))>;
4227 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
4228 (VGETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src,
4229 (DSubReg_i32_reg imm:$lane))),
4230 (SubReg_i32_lane imm:$lane))>;
4231 def : Pat<(extractelt (v2f32 DPR:$src1), imm:$src2),
4232 (EXTRACT_SUBREG (v2f32 (COPY_TO_REGCLASS (v2f32 DPR:$src1),DPR_VFP2)),
4233 (SSubReg_f32_reg imm:$src2))>;
4234 def : Pat<(extractelt (v4f32 QPR:$src1), imm:$src2),
4235 (EXTRACT_SUBREG (v4f32 (COPY_TO_REGCLASS (v4f32 QPR:$src1),QPR_VFP2)),
4236 (SSubReg_f32_reg imm:$src2))>;
4237 //def : Pat<(extractelt (v2i64 QPR:$src1), imm:$src2),
4238 // (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
4239 def : Pat<(extractelt (v2f64 QPR:$src1), imm:$src2),
4240 (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
4243 // VMOV : Vector Set Lane (move ARM core register to scalar)
4245 let Constraints = "$src1 = $V" in {
4246 def VSETLNi8 : NVSetLane<{1,1,1,0,0,1,?,0}, 0b1011, {?,?}, (outs DPR:$V),
4247 (ins DPR:$src1, GPR:$R, nohash_imm:$lane),
4248 IIC_VMOVISL, "vmov", "8", "$V[$lane], $R",
4249 [(set DPR:$V, (vector_insert (v8i8 DPR:$src1),
4250 GPR:$R, imm:$lane))]> {
4251 let Inst{21} = lane{2};
4252 let Inst{6-5} = lane{1-0};
4254 def VSETLNi16 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, {?,1}, (outs DPR:$V),
4255 (ins DPR:$src1, GPR:$R, nohash_imm:$lane),
4256 IIC_VMOVISL, "vmov", "16", "$V[$lane], $R",
4257 [(set DPR:$V, (vector_insert (v4i16 DPR:$src1),
4258 GPR:$R, imm:$lane))]> {
4259 let Inst{21} = lane{1};
4260 let Inst{6} = lane{0};
4262 def VSETLNi32 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, 0b00, (outs DPR:$V),
4263 (ins DPR:$src1, GPR:$R, nohash_imm:$lane),
4264 IIC_VMOVISL, "vmov", "32", "$V[$lane], $R",
4265 [(set DPR:$V, (insertelt (v2i32 DPR:$src1),
4266 GPR:$R, imm:$lane))]> {
4267 let Inst{21} = lane{0};
4270 def : Pat<(vector_insert (v16i8 QPR:$src1), GPR:$src2, imm:$lane),
4271 (v16i8 (INSERT_SUBREG QPR:$src1,
4272 (v8i8 (VSETLNi8 (v8i8 (EXTRACT_SUBREG QPR:$src1,
4273 (DSubReg_i8_reg imm:$lane))),
4274 GPR:$src2, (SubReg_i8_lane imm:$lane))),
4275 (DSubReg_i8_reg imm:$lane)))>;
4276 def : Pat<(vector_insert (v8i16 QPR:$src1), GPR:$src2, imm:$lane),
4277 (v8i16 (INSERT_SUBREG QPR:$src1,
4278 (v4i16 (VSETLNi16 (v4i16 (EXTRACT_SUBREG QPR:$src1,
4279 (DSubReg_i16_reg imm:$lane))),
4280 GPR:$src2, (SubReg_i16_lane imm:$lane))),
4281 (DSubReg_i16_reg imm:$lane)))>;
4282 def : Pat<(insertelt (v4i32 QPR:$src1), GPR:$src2, imm:$lane),
4283 (v4i32 (INSERT_SUBREG QPR:$src1,
4284 (v2i32 (VSETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src1,
4285 (DSubReg_i32_reg imm:$lane))),
4286 GPR:$src2, (SubReg_i32_lane imm:$lane))),
4287 (DSubReg_i32_reg imm:$lane)))>;
4289 def : Pat<(v2f32 (insertelt DPR:$src1, SPR:$src2, imm:$src3)),
4290 (INSERT_SUBREG (v2f32 (COPY_TO_REGCLASS DPR:$src1, DPR_VFP2)),
4291 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
4292 def : Pat<(v4f32 (insertelt QPR:$src1, SPR:$src2, imm:$src3)),
4293 (INSERT_SUBREG (v4f32 (COPY_TO_REGCLASS QPR:$src1, QPR_VFP2)),
4294 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
4296 //def : Pat<(v2i64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
4297 // (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
4298 def : Pat<(v2f64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
4299 (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
4301 def : Pat<(v2f32 (scalar_to_vector SPR:$src)),
4302 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
4303 def : Pat<(v2f64 (scalar_to_vector (f64 DPR:$src))),
4304 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
4305 def : Pat<(v4f32 (scalar_to_vector SPR:$src)),
4306 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
4308 def : Pat<(v8i8 (scalar_to_vector GPR:$src)),
4309 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4310 def : Pat<(v4i16 (scalar_to_vector GPR:$src)),
4311 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4312 def : Pat<(v2i32 (scalar_to_vector GPR:$src)),
4313 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4315 def : Pat<(v16i8 (scalar_to_vector GPR:$src)),
4316 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4317 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4319 def : Pat<(v8i16 (scalar_to_vector GPR:$src)),
4320 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
4321 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4323 def : Pat<(v4i32 (scalar_to_vector GPR:$src)),
4324 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
4325 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4328 // VDUP : Vector Duplicate (from ARM core register to all elements)
4330 class VDUPD<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
4331 : NVDup<opcod1, 0b1011, opcod3, (outs DPR:$V), (ins GPR:$R),
4332 IIC_VMOVIS, "vdup", Dt, "$V, $R",
4333 [(set DPR:$V, (Ty (NEONvdup (i32 GPR:$R))))]>;
4334 class VDUPQ<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
4335 : NVDup<opcod1, 0b1011, opcod3, (outs QPR:$V), (ins GPR:$R),
4336 IIC_VMOVIS, "vdup", Dt, "$V, $R",
4337 [(set QPR:$V, (Ty (NEONvdup (i32 GPR:$R))))]>;
4339 def VDUP8d : VDUPD<0b11101100, 0b00, "8", v8i8>;
4340 def VDUP16d : VDUPD<0b11101000, 0b01, "16", v4i16>;
4341 def VDUP32d : VDUPD<0b11101000, 0b00, "32", v2i32>;
4342 def VDUP8q : VDUPQ<0b11101110, 0b00, "8", v16i8>;
4343 def VDUP16q : VDUPQ<0b11101010, 0b01, "16", v8i16>;
4344 def VDUP32q : VDUPQ<0b11101010, 0b00, "32", v4i32>;
4346 def VDUPfd : NVDup<0b11101000, 0b1011, 0b00, (outs DPR:$V), (ins GPR:$R),
4347 IIC_VMOVIS, "vdup", "32", "$V, $R",
4348 [(set DPR:$V, (v2f32 (NEONvdup
4349 (f32 (bitconvert GPR:$R)))))]>;
4350 def VDUPfq : NVDup<0b11101010, 0b1011, 0b00, (outs QPR:$V), (ins GPR:$R),
4351 IIC_VMOVIS, "vdup", "32", "$V, $R",
4352 [(set QPR:$V, (v4f32 (NEONvdup
4353 (f32 (bitconvert GPR:$R)))))]>;
4355 // VDUP : Vector Duplicate Lane (from scalar to all elements)
4357 class VDUPLND<bits<4> op19_16, string OpcodeStr, string Dt,
4359 : NVDupLane<op19_16, 0, (outs DPR:$Vd), (ins DPR:$Vm, nohash_imm:$lane),
4360 IIC_VMOVD, OpcodeStr, Dt, "$Vd, $Vm[$lane]",
4361 [(set DPR:$Vd, (Ty (NEONvduplane (Ty DPR:$Vm), imm:$lane)))]>;
4363 class VDUPLNQ<bits<4> op19_16, string OpcodeStr, string Dt,
4364 ValueType ResTy, ValueType OpTy>
4365 : NVDupLane<op19_16, 1, (outs QPR:$Vd), (ins DPR:$Vm, nohash_imm:$lane),
4366 IIC_VMOVQ, OpcodeStr, Dt, "$Vd, $Vm[$lane]",
4367 [(set QPR:$Vd, (ResTy (NEONvduplane (OpTy DPR:$Vm),
4370 // Inst{19-16} is partially specified depending on the element size.
4372 def VDUPLN8d : VDUPLND<{?,?,?,1}, "vdup", "8", v8i8> {
4373 let Inst{19-17} = lane{2-0};
4375 def VDUPLN16d : VDUPLND<{?,?,1,0}, "vdup", "16", v4i16> {
4376 let Inst{19-18} = lane{1-0};
4378 def VDUPLN32d : VDUPLND<{?,1,0,0}, "vdup", "32", v2i32> {
4379 let Inst{19} = lane{0};
4381 def VDUPLNfd : VDUPLND<{?,1,0,0}, "vdup", "32", v2f32> {
4382 let Inst{19} = lane{0};
4384 def VDUPLN8q : VDUPLNQ<{?,?,?,1}, "vdup", "8", v16i8, v8i8> {
4385 let Inst{19-17} = lane{2-0};
4387 def VDUPLN16q : VDUPLNQ<{?,?,1,0}, "vdup", "16", v8i16, v4i16> {
4388 let Inst{19-18} = lane{1-0};
4390 def VDUPLN32q : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4i32, v2i32> {
4391 let Inst{19} = lane{0};
4393 def VDUPLNfq : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4f32, v2f32> {
4394 let Inst{19} = lane{0};
4397 def : Pat<(v16i8 (NEONvduplane (v16i8 QPR:$src), imm:$lane)),
4398 (v16i8 (VDUPLN8q (v8i8 (EXTRACT_SUBREG QPR:$src,
4399 (DSubReg_i8_reg imm:$lane))),
4400 (SubReg_i8_lane imm:$lane)))>;
4401 def : Pat<(v8i16 (NEONvduplane (v8i16 QPR:$src), imm:$lane)),
4402 (v8i16 (VDUPLN16q (v4i16 (EXTRACT_SUBREG QPR:$src,
4403 (DSubReg_i16_reg imm:$lane))),
4404 (SubReg_i16_lane imm:$lane)))>;
4405 def : Pat<(v4i32 (NEONvduplane (v4i32 QPR:$src), imm:$lane)),
4406 (v4i32 (VDUPLN32q (v2i32 (EXTRACT_SUBREG QPR:$src,
4407 (DSubReg_i32_reg imm:$lane))),
4408 (SubReg_i32_lane imm:$lane)))>;
4409 def : Pat<(v4f32 (NEONvduplane (v4f32 QPR:$src), imm:$lane)),
4410 (v4f32 (VDUPLNfq (v2f32 (EXTRACT_SUBREG QPR:$src,
4411 (DSubReg_i32_reg imm:$lane))),
4412 (SubReg_i32_lane imm:$lane)))>;
4414 def VDUPfdf : PseudoNeonI<(outs DPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
4415 [(set DPR:$dst, (v2f32 (NEONvdup (f32 SPR:$src))))]>;
4416 def VDUPfqf : PseudoNeonI<(outs QPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
4417 [(set QPR:$dst, (v4f32 (NEONvdup (f32 SPR:$src))))]>;
4419 // VMOVN : Vector Narrowing Move
4420 defm VMOVN : N2VN_HSD<0b11,0b11,0b10,0b00100,0,0, IIC_VMOVN,
4421 "vmovn", "i", trunc>;
4422 // VQMOVN : Vector Saturating Narrowing Move
4423 defm VQMOVNs : N2VNInt_HSD<0b11,0b11,0b10,0b00101,0,0, IIC_VQUNAiD,
4424 "vqmovn", "s", int_arm_neon_vqmovns>;
4425 defm VQMOVNu : N2VNInt_HSD<0b11,0b11,0b10,0b00101,1,0, IIC_VQUNAiD,
4426 "vqmovn", "u", int_arm_neon_vqmovnu>;
4427 defm VQMOVNsu : N2VNInt_HSD<0b11,0b11,0b10,0b00100,1,0, IIC_VQUNAiD,
4428 "vqmovun", "s", int_arm_neon_vqmovnsu>;
4429 // VMOVL : Vector Lengthening Move
4430 defm VMOVLs : N2VL_QHS<0b01,0b10100,0,1, "vmovl", "s", sext>;
4431 defm VMOVLu : N2VL_QHS<0b11,0b10100,0,1, "vmovl", "u", zext>;
4433 // Vector Conversions.
4435 // VCVT : Vector Convert Between Floating-Point and Integers
4436 def VCVTf2sd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
4437 v2i32, v2f32, fp_to_sint>;
4438 def VCVTf2ud : N2VD<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
4439 v2i32, v2f32, fp_to_uint>;
4440 def VCVTs2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
4441 v2f32, v2i32, sint_to_fp>;
4442 def VCVTu2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
4443 v2f32, v2i32, uint_to_fp>;
4445 def VCVTf2sq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
4446 v4i32, v4f32, fp_to_sint>;
4447 def VCVTf2uq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
4448 v4i32, v4f32, fp_to_uint>;
4449 def VCVTs2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
4450 v4f32, v4i32, sint_to_fp>;
4451 def VCVTu2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
4452 v4f32, v4i32, uint_to_fp>;
4454 // VCVT : Vector Convert Between Floating-Point and Fixed-Point.
4455 def VCVTf2xsd : N2VCvtD<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
4456 v2i32, v2f32, int_arm_neon_vcvtfp2fxs>;
4457 def VCVTf2xud : N2VCvtD<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
4458 v2i32, v2f32, int_arm_neon_vcvtfp2fxu>;
4459 def VCVTxs2fd : N2VCvtD<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
4460 v2f32, v2i32, int_arm_neon_vcvtfxs2fp>;
4461 def VCVTxu2fd : N2VCvtD<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
4462 v2f32, v2i32, int_arm_neon_vcvtfxu2fp>;
4464 def VCVTf2xsq : N2VCvtQ<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
4465 v4i32, v4f32, int_arm_neon_vcvtfp2fxs>;
4466 def VCVTf2xuq : N2VCvtQ<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
4467 v4i32, v4f32, int_arm_neon_vcvtfp2fxu>;
4468 def VCVTxs2fq : N2VCvtQ<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
4469 v4f32, v4i32, int_arm_neon_vcvtfxs2fp>;
4470 def VCVTxu2fq : N2VCvtQ<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
4471 v4f32, v4i32, int_arm_neon_vcvtfxu2fp>;
4475 // VREV64 : Vector Reverse elements within 64-bit doublewords
4477 class VREV64D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4478 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 0, 0, (outs DPR:$Vd),
4479 (ins DPR:$Vm), IIC_VMOVD,
4480 OpcodeStr, Dt, "$Vd, $Vm", "",
4481 [(set DPR:$Vd, (Ty (NEONvrev64 (Ty DPR:$Vm))))]>;
4482 class VREV64Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4483 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 1, 0, (outs QPR:$Vd),
4484 (ins QPR:$Vm), IIC_VMOVQ,
4485 OpcodeStr, Dt, "$Vd, $Vm", "",
4486 [(set QPR:$Vd, (Ty (NEONvrev64 (Ty QPR:$Vm))))]>;
4488 def VREV64d8 : VREV64D<0b00, "vrev64", "8", v8i8>;
4489 def VREV64d16 : VREV64D<0b01, "vrev64", "16", v4i16>;
4490 def VREV64d32 : VREV64D<0b10, "vrev64", "32", v2i32>;
4491 def VREV64df : VREV64D<0b10, "vrev64", "32", v2f32>;
4493 def VREV64q8 : VREV64Q<0b00, "vrev64", "8", v16i8>;
4494 def VREV64q16 : VREV64Q<0b01, "vrev64", "16", v8i16>;
4495 def VREV64q32 : VREV64Q<0b10, "vrev64", "32", v4i32>;
4496 def VREV64qf : VREV64Q<0b10, "vrev64", "32", v4f32>;
4498 // VREV32 : Vector Reverse elements within 32-bit words
4500 class VREV32D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4501 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 0, 0, (outs DPR:$Vd),
4502 (ins DPR:$Vm), IIC_VMOVD,
4503 OpcodeStr, Dt, "$Vd, $Vm", "",
4504 [(set DPR:$Vd, (Ty (NEONvrev32 (Ty DPR:$Vm))))]>;
4505 class VREV32Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4506 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 1, 0, (outs QPR:$Vd),
4507 (ins QPR:$Vm), IIC_VMOVQ,
4508 OpcodeStr, Dt, "$Vd, $Vm", "",
4509 [(set QPR:$Vd, (Ty (NEONvrev32 (Ty QPR:$Vm))))]>;
4511 def VREV32d8 : VREV32D<0b00, "vrev32", "8", v8i8>;
4512 def VREV32d16 : VREV32D<0b01, "vrev32", "16", v4i16>;
4514 def VREV32q8 : VREV32Q<0b00, "vrev32", "8", v16i8>;
4515 def VREV32q16 : VREV32Q<0b01, "vrev32", "16", v8i16>;
4517 // VREV16 : Vector Reverse elements within 16-bit halfwords
4519 class VREV16D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4520 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 0, 0, (outs DPR:$Vd),
4521 (ins DPR:$Vm), IIC_VMOVD,
4522 OpcodeStr, Dt, "$Vd, $Vm", "",
4523 [(set DPR:$Vd, (Ty (NEONvrev16 (Ty DPR:$Vm))))]>;
4524 class VREV16Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4525 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 1, 0, (outs QPR:$Vd),
4526 (ins QPR:$Vm), IIC_VMOVQ,
4527 OpcodeStr, Dt, "$Vd, $Vm", "",
4528 [(set QPR:$Vd, (Ty (NEONvrev16 (Ty QPR:$Vm))))]>;
4530 def VREV16d8 : VREV16D<0b00, "vrev16", "8", v8i8>;
4531 def VREV16q8 : VREV16Q<0b00, "vrev16", "8", v16i8>;
4533 // Other Vector Shuffles.
4535 // VEXT : Vector Extract
4537 class VEXTd<string OpcodeStr, string Dt, ValueType Ty>
4538 : N3V<0,1,0b11,{?,?,?,?},0,0, (outs DPR:$Vd),
4539 (ins DPR:$Vn, DPR:$Vm, i32imm:$index), NVExtFrm,
4540 IIC_VEXTD, OpcodeStr, Dt, "$Vd, $Vn, $Vm, $index", "",
4541 [(set DPR:$Vd, (Ty (NEONvext (Ty DPR:$Vn),
4542 (Ty DPR:$Vm), imm:$index)))]> {
4544 let Inst{11-8} = index{3-0};
4547 class VEXTq<string OpcodeStr, string Dt, ValueType Ty>
4548 : N3V<0,1,0b11,{?,?,?,?},1,0, (outs QPR:$Vd),
4549 (ins QPR:$Vn, QPR:$Vm, i32imm:$index), NVExtFrm,
4550 IIC_VEXTQ, OpcodeStr, Dt, "$Vd, $Vn, $Vm, $index", "",
4551 [(set QPR:$Vd, (Ty (NEONvext (Ty QPR:$Vn),
4552 (Ty QPR:$Vm), imm:$index)))]> {
4554 let Inst{11-8} = index{3-0};
4557 def VEXTd8 : VEXTd<"vext", "8", v8i8> {
4558 let Inst{11-8} = index{3-0};
4560 def VEXTd16 : VEXTd<"vext", "16", v4i16> {
4561 let Inst{11-9} = index{2-0};
4564 def VEXTd32 : VEXTd<"vext", "32", v2i32> {
4565 let Inst{11-10} = index{1-0};
4566 let Inst{9-8} = 0b00;
4568 def VEXTdf : VEXTd<"vext", "32", v2f32> {
4569 let Inst{11} = index{0};
4570 let Inst{10-8} = 0b000;
4573 def VEXTq8 : VEXTq<"vext", "8", v16i8> {
4574 let Inst{11-8} = index{3-0};
4576 def VEXTq16 : VEXTq<"vext", "16", v8i16> {
4577 let Inst{11-9} = index{2-0};
4580 def VEXTq32 : VEXTq<"vext", "32", v4i32> {
4581 let Inst{11-10} = index{1-0};
4582 let Inst{9-8} = 0b00;
4584 def VEXTqf : VEXTq<"vext", "32", v4f32> {
4585 let Inst{11} = index{0};
4586 let Inst{10-8} = 0b000;
4589 // VTRN : Vector Transpose
4591 def VTRNd8 : N2VDShuffle<0b00, 0b00001, "vtrn", "8">;
4592 def VTRNd16 : N2VDShuffle<0b01, 0b00001, "vtrn", "16">;
4593 def VTRNd32 : N2VDShuffle<0b10, 0b00001, "vtrn", "32">;
4595 def VTRNq8 : N2VQShuffle<0b00, 0b00001, IIC_VPERMQ, "vtrn", "8">;
4596 def VTRNq16 : N2VQShuffle<0b01, 0b00001, IIC_VPERMQ, "vtrn", "16">;
4597 def VTRNq32 : N2VQShuffle<0b10, 0b00001, IIC_VPERMQ, "vtrn", "32">;
4599 // VUZP : Vector Unzip (Deinterleave)
4601 def VUZPd8 : N2VDShuffle<0b00, 0b00010, "vuzp", "8">;
4602 def VUZPd16 : N2VDShuffle<0b01, 0b00010, "vuzp", "16">;
4603 def VUZPd32 : N2VDShuffle<0b10, 0b00010, "vuzp", "32">;
4605 def VUZPq8 : N2VQShuffle<0b00, 0b00010, IIC_VPERMQ3, "vuzp", "8">;
4606 def VUZPq16 : N2VQShuffle<0b01, 0b00010, IIC_VPERMQ3, "vuzp", "16">;
4607 def VUZPq32 : N2VQShuffle<0b10, 0b00010, IIC_VPERMQ3, "vuzp", "32">;
4609 // VZIP : Vector Zip (Interleave)
4611 def VZIPd8 : N2VDShuffle<0b00, 0b00011, "vzip", "8">;
4612 def VZIPd16 : N2VDShuffle<0b01, 0b00011, "vzip", "16">;
4613 def VZIPd32 : N2VDShuffle<0b10, 0b00011, "vzip", "32">;
4615 def VZIPq8 : N2VQShuffle<0b00, 0b00011, IIC_VPERMQ3, "vzip", "8">;
4616 def VZIPq16 : N2VQShuffle<0b01, 0b00011, IIC_VPERMQ3, "vzip", "16">;
4617 def VZIPq32 : N2VQShuffle<0b10, 0b00011, IIC_VPERMQ3, "vzip", "32">;
4619 // Vector Table Lookup and Table Extension.
4621 // VTBL : Vector Table Lookup
4623 : N3V<1,1,0b11,0b1000,0,0, (outs DPR:$Vd),
4624 (ins DPR:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTB1,
4625 "vtbl", "8", "$Vd, \\{$Vn\\}, $Vm", "",
4626 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbl1 DPR:$Vn, DPR:$Vm)))]>;
4627 let hasExtraSrcRegAllocReq = 1 in {
4629 : N3V<1,1,0b11,0b1001,0,0, (outs DPR:$Vd),
4630 (ins DPR:$Vn, DPR:$tbl2, DPR:$Vm), NVTBLFrm, IIC_VTB2,
4631 "vtbl", "8", "$Vd, \\{$Vn, $tbl2\\}, $Vm", "", []>;
4633 : N3V<1,1,0b11,0b1010,0,0, (outs DPR:$Vd),
4634 (ins DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$Vm), NVTBLFrm, IIC_VTB3,
4635 "vtbl", "8", "$Vd, \\{$Vn, $tbl2, $tbl3\\}, $Vm", "", []>;
4637 : N3V<1,1,0b11,0b1011,0,0, (outs DPR:$Vd),
4638 (ins DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$Vm),
4640 "vtbl", "8", "$Vd, \\{$Vn, $tbl2, $tbl3, $tbl4\\}, $Vm", "", []>;
4641 } // hasExtraSrcRegAllocReq = 1
4644 : PseudoNeonI<(outs DPR:$dst), (ins QPR:$tbl, DPR:$src), IIC_VTB2, "", []>;
4646 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB3, "", []>;
4648 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB4, "", []>;
4650 // VTBX : Vector Table Extension
4652 : N3V<1,1,0b11,0b1000,1,0, (outs DPR:$Vd),
4653 (ins DPR:$orig, DPR:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTBX1,
4654 "vtbx", "8", "$Vd, \\{$Vn\\}, $Vm", "$orig = $Vd",
4655 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbx1
4656 DPR:$orig, DPR:$Vn, DPR:$Vm)))]>;
4657 let hasExtraSrcRegAllocReq = 1 in {
4659 : N3V<1,1,0b11,0b1001,1,0, (outs DPR:$Vd),
4660 (ins DPR:$orig, DPR:$Vn, DPR:$tbl2, DPR:$Vm), NVTBLFrm, IIC_VTBX2,
4661 "vtbx", "8", "$Vd, \\{$Vn, $tbl2\\}, $Vm", "$orig = $Vd", []>;
4663 : N3V<1,1,0b11,0b1010,1,0, (outs DPR:$Vd),
4664 (ins DPR:$orig, DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$Vm),
4665 NVTBLFrm, IIC_VTBX3,
4666 "vtbx", "8", "$Vd, \\{$Vn, $tbl2, $tbl3\\}, $Vm",
4669 : N3V<1,1,0b11,0b1011,1,0, (outs DPR:$Vd), (ins DPR:$orig, DPR:$Vn,
4670 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$Vm), NVTBLFrm, IIC_VTBX4,
4671 "vtbx", "8", "$Vd, \\{$Vn, $tbl2, $tbl3, $tbl4\\}, $Vm",
4673 } // hasExtraSrcRegAllocReq = 1
4676 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QPR:$tbl, DPR:$src),
4677 IIC_VTBX2, "$orig = $dst", []>;
4679 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
4680 IIC_VTBX3, "$orig = $dst", []>;
4682 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
4683 IIC_VTBX4, "$orig = $dst", []>;
4685 //===----------------------------------------------------------------------===//
4686 // NEON instructions for single-precision FP math
4687 //===----------------------------------------------------------------------===//
4689 class N2VSPat<SDNode OpNode, ValueType ResTy, ValueType OpTy, NeonI Inst>
4690 : NEONFPPat<(ResTy (OpNode SPR:$a)),
4691 (EXTRACT_SUBREG (OpTy (Inst (INSERT_SUBREG (OpTy (IMPLICIT_DEF)),
4695 class N3VSPat<SDNode OpNode, NeonI Inst>
4696 : NEONFPPat<(f32 (OpNode SPR:$a, SPR:$b)),
4697 (EXTRACT_SUBREG (v2f32
4698 (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
4700 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
4704 class N3VSMulOpPat<SDNode MulNode, SDNode OpNode, NeonI Inst>
4705 : NEONFPPat<(f32 (OpNode SPR:$acc, (f32 (MulNode SPR:$a, SPR:$b)))),
4706 (EXTRACT_SUBREG (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
4708 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
4710 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
4714 // These need separate instructions because they must use DPR_VFP2 register
4715 // class which have SPR sub-registers.
4717 // Vector Add Operations used for single-precision FP
4718 let neverHasSideEffects = 1 in
4719 def VADDfd_sfp : N3VS<0,0,0b00,0b1101,0, "vadd", "f32", v2f32, v2f32, fadd, 1>;
4720 def : N3VSPat<fadd, VADDfd_sfp>;
4722 // Vector Sub Operations used for single-precision FP
4723 let neverHasSideEffects = 1 in
4724 def VSUBfd_sfp : N3VS<0,0,0b10,0b1101,0, "vsub", "f32", v2f32, v2f32, fsub, 0>;
4725 def : N3VSPat<fsub, VSUBfd_sfp>;
4727 // Vector Multiply Operations used for single-precision FP
4728 let neverHasSideEffects = 1 in
4729 def VMULfd_sfp : N3VS<1,0,0b00,0b1101,1, "vmul", "f32", v2f32, v2f32, fmul, 1>;
4730 def : N3VSPat<fmul, VMULfd_sfp>;
4732 // Vector Multiply-Accumulate/Subtract used for single-precision FP
4733 // vml[as].f32 can cause 4-8 cycle stalls in following ASIMD instructions, so
4734 // we want to avoid them for now. e.g., alternating vmla/vadd instructions.
4736 let neverHasSideEffects = 1 in
4737 def VMLAfd_sfp : N3VSMulOp<0,0,0b00,0b1101,1, IIC_VMACD, "vmla", "f32",
4738 v2f32, fmul_su, fadd>;
4739 def : N3VSMulOpPat<fmul, fadd, VMLAfd_sfp>,
4740 Requires<[HasNEON, UseNEONForFP, UseFPVMLx]>;
4742 let neverHasSideEffects = 1 in
4743 def VMLSfd_sfp : N3VSMulOp<0,0,0b10,0b1101,1, IIC_VMACD, "vmls", "f32",
4744 v2f32, fmul_su, fsub>;
4745 def : N3VSMulOpPat<fmul, fsub, VMLSfd_sfp>,
4746 Requires<[HasNEON, UseNEONForFP, UseFPVMLx]>;
4748 // Vector Absolute used for single-precision FP
4749 let neverHasSideEffects = 1 in
4750 def VABSfd_sfp : N2V<0b11, 0b11, 0b10, 0b01, 0b01110, 0, 0,
4751 (outs DPR_VFP2:$Vd), (ins DPR_VFP2:$Vm), IIC_VUNAD,
4752 "vabs", "f32", "$Vd, $Vm", "", []>;
4753 def : N2VSPat<fabs, f32, v2f32, VABSfd_sfp>;
4755 // Vector Negate used for single-precision FP
4756 let neverHasSideEffects = 1 in
4757 def VNEGfd_sfp : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
4758 (outs DPR_VFP2:$Vd), (ins DPR_VFP2:$Vm), IIC_VUNAD,
4759 "vneg", "f32", "$Vd, $Vm", "", []>;
4760 def : N2VSPat<fneg, f32, v2f32, VNEGfd_sfp>;
4762 // Vector Maximum used for single-precision FP
4763 let neverHasSideEffects = 1 in
4764 def VMAXfd_sfp : N3V<0, 0, 0b00, 0b1111, 0, 0, (outs DPR_VFP2:$Vd),
4765 (ins DPR_VFP2:$Vn, DPR_VFP2:$Vm), N3RegFrm, IIC_VBIND,
4766 "vmax", "f32", "$Vd, $Vn, $Vm", "", []>;
4767 def : N3VSPat<NEONfmax, VMAXfd_sfp>;
4769 // Vector Minimum used for single-precision FP
4770 let neverHasSideEffects = 1 in
4771 def VMINfd_sfp : N3V<0, 0, 0b10, 0b1111, 0, 0, (outs DPR_VFP2:$Vd),
4772 (ins DPR_VFP2:$Vn, DPR_VFP2:$Vm), N3RegFrm, IIC_VBIND,
4773 "vmin", "f32", "$Vd, $Vn, $Vm", "", []>;
4774 def : N3VSPat<NEONfmin, VMINfd_sfp>;
4776 // Vector Convert between single-precision FP and integer
4777 let neverHasSideEffects = 1 in
4778 def VCVTf2sd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
4779 v2i32, v2f32, fp_to_sint>;
4780 def : N2VSPat<arm_ftosi, f32, v2f32, VCVTf2sd_sfp>;
4782 let neverHasSideEffects = 1 in
4783 def VCVTf2ud_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
4784 v2i32, v2f32, fp_to_uint>;
4785 def : N2VSPat<arm_ftoui, f32, v2f32, VCVTf2ud_sfp>;
4787 let neverHasSideEffects = 1 in
4788 def VCVTs2fd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
4789 v2f32, v2i32, sint_to_fp>;
4790 def : N2VSPat<arm_sitof, f32, v2i32, VCVTs2fd_sfp>;
4792 let neverHasSideEffects = 1 in
4793 def VCVTu2fd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
4794 v2f32, v2i32, uint_to_fp>;
4795 def : N2VSPat<arm_uitof, f32, v2i32, VCVTu2fd_sfp>;
4797 //===----------------------------------------------------------------------===//
4798 // Non-Instruction Patterns
4799 //===----------------------------------------------------------------------===//
4802 def : Pat<(v1i64 (bitconvert (v2i32 DPR:$src))), (v1i64 DPR:$src)>;
4803 def : Pat<(v1i64 (bitconvert (v4i16 DPR:$src))), (v1i64 DPR:$src)>;
4804 def : Pat<(v1i64 (bitconvert (v8i8 DPR:$src))), (v1i64 DPR:$src)>;
4805 def : Pat<(v1i64 (bitconvert (f64 DPR:$src))), (v1i64 DPR:$src)>;
4806 def : Pat<(v1i64 (bitconvert (v2f32 DPR:$src))), (v1i64 DPR:$src)>;
4807 def : Pat<(v2i32 (bitconvert (v1i64 DPR:$src))), (v2i32 DPR:$src)>;
4808 def : Pat<(v2i32 (bitconvert (v4i16 DPR:$src))), (v2i32 DPR:$src)>;
4809 def : Pat<(v2i32 (bitconvert (v8i8 DPR:$src))), (v2i32 DPR:$src)>;
4810 def : Pat<(v2i32 (bitconvert (f64 DPR:$src))), (v2i32 DPR:$src)>;
4811 def : Pat<(v2i32 (bitconvert (v2f32 DPR:$src))), (v2i32 DPR:$src)>;
4812 def : Pat<(v4i16 (bitconvert (v1i64 DPR:$src))), (v4i16 DPR:$src)>;
4813 def : Pat<(v4i16 (bitconvert (v2i32 DPR:$src))), (v4i16 DPR:$src)>;
4814 def : Pat<(v4i16 (bitconvert (v8i8 DPR:$src))), (v4i16 DPR:$src)>;
4815 def : Pat<(v4i16 (bitconvert (f64 DPR:$src))), (v4i16 DPR:$src)>;
4816 def : Pat<(v4i16 (bitconvert (v2f32 DPR:$src))), (v4i16 DPR:$src)>;
4817 def : Pat<(v8i8 (bitconvert (v1i64 DPR:$src))), (v8i8 DPR:$src)>;
4818 def : Pat<(v8i8 (bitconvert (v2i32 DPR:$src))), (v8i8 DPR:$src)>;
4819 def : Pat<(v8i8 (bitconvert (v4i16 DPR:$src))), (v8i8 DPR:$src)>;
4820 def : Pat<(v8i8 (bitconvert (f64 DPR:$src))), (v8i8 DPR:$src)>;
4821 def : Pat<(v8i8 (bitconvert (v2f32 DPR:$src))), (v8i8 DPR:$src)>;
4822 def : Pat<(f64 (bitconvert (v1i64 DPR:$src))), (f64 DPR:$src)>;
4823 def : Pat<(f64 (bitconvert (v2i32 DPR:$src))), (f64 DPR:$src)>;
4824 def : Pat<(f64 (bitconvert (v4i16 DPR:$src))), (f64 DPR:$src)>;
4825 def : Pat<(f64 (bitconvert (v8i8 DPR:$src))), (f64 DPR:$src)>;
4826 def : Pat<(f64 (bitconvert (v2f32 DPR:$src))), (f64 DPR:$src)>;
4827 def : Pat<(v2f32 (bitconvert (f64 DPR:$src))), (v2f32 DPR:$src)>;
4828 def : Pat<(v2f32 (bitconvert (v1i64 DPR:$src))), (v2f32 DPR:$src)>;
4829 def : Pat<(v2f32 (bitconvert (v2i32 DPR:$src))), (v2f32 DPR:$src)>;
4830 def : Pat<(v2f32 (bitconvert (v4i16 DPR:$src))), (v2f32 DPR:$src)>;
4831 def : Pat<(v2f32 (bitconvert (v8i8 DPR:$src))), (v2f32 DPR:$src)>;
4833 def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (v2i64 QPR:$src)>;
4834 def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (v2i64 QPR:$src)>;
4835 def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (v2i64 QPR:$src)>;
4836 def : Pat<(v2i64 (bitconvert (v2f64 QPR:$src))), (v2i64 QPR:$src)>;
4837 def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (v2i64 QPR:$src)>;
4838 def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (v4i32 QPR:$src)>;
4839 def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (v4i32 QPR:$src)>;
4840 def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (v4i32 QPR:$src)>;
4841 def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (v4i32 QPR:$src)>;
4842 def : Pat<(v4i32 (bitconvert (v4f32 QPR:$src))), (v4i32 QPR:$src)>;
4843 def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (v8i16 QPR:$src)>;
4844 def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (v8i16 QPR:$src)>;
4845 def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (v8i16 QPR:$src)>;
4846 def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (v8i16 QPR:$src)>;
4847 def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (v8i16 QPR:$src)>;
4848 def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (v16i8 QPR:$src)>;
4849 def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (v16i8 QPR:$src)>;
4850 def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (v16i8 QPR:$src)>;
4851 def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (v16i8 QPR:$src)>;
4852 def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (v16i8 QPR:$src)>;
4853 def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (v4f32 QPR:$src)>;
4854 def : Pat<(v4f32 (bitconvert (v4i32 QPR:$src))), (v4f32 QPR:$src)>;
4855 def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (v4f32 QPR:$src)>;
4856 def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (v4f32 QPR:$src)>;
4857 def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (v4f32 QPR:$src)>;
4858 def : Pat<(v2f64 (bitconvert (v2i64 QPR:$src))), (v2f64 QPR:$src)>;
4859 def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (v2f64 QPR:$src)>;
4860 def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 QPR:$src)>;
4861 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 QPR:$src)>;
4862 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 QPR:$src)>;