1 //===- ARMInstrNEON.td - NEON support for ARM -----------------------------===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file describes the ARM NEON instruction set.
12 //===----------------------------------------------------------------------===//
14 //===----------------------------------------------------------------------===//
15 // NEON-specific DAG Nodes.
16 //===----------------------------------------------------------------------===//
18 def SDTARMVCMP : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisSameAs<1, 2>]>;
19 def SDTARMVCMPZ : SDTypeProfile<1, 1, []>;
21 def NEONvceq : SDNode<"ARMISD::VCEQ", SDTARMVCMP>;
22 def NEONvceqz : SDNode<"ARMISD::VCEQZ", SDTARMVCMPZ>;
23 def NEONvcge : SDNode<"ARMISD::VCGE", SDTARMVCMP>;
24 def NEONvcgez : SDNode<"ARMISD::VCGEZ", SDTARMVCMPZ>;
25 def NEONvclez : SDNode<"ARMISD::VCLEZ", SDTARMVCMPZ>;
26 def NEONvcgeu : SDNode<"ARMISD::VCGEU", SDTARMVCMP>;
27 def NEONvcgt : SDNode<"ARMISD::VCGT", SDTARMVCMP>;
28 def NEONvcgtz : SDNode<"ARMISD::VCGTZ", SDTARMVCMPZ>;
29 def NEONvcltz : SDNode<"ARMISD::VCLTZ", SDTARMVCMPZ>;
30 def NEONvcgtu : SDNode<"ARMISD::VCGTU", SDTARMVCMP>;
31 def NEONvtst : SDNode<"ARMISD::VTST", SDTARMVCMP>;
33 // Types for vector shift by immediates. The "SHX" version is for long and
34 // narrow operations where the source and destination vectors have different
35 // types. The "SHINS" version is for shift and insert operations.
36 def SDTARMVSH : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
38 def SDTARMVSHX : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
40 def SDTARMVSHINS : SDTypeProfile<1, 3, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
41 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
43 def NEONvshl : SDNode<"ARMISD::VSHL", SDTARMVSH>;
44 def NEONvshrs : SDNode<"ARMISD::VSHRs", SDTARMVSH>;
45 def NEONvshru : SDNode<"ARMISD::VSHRu", SDTARMVSH>;
46 def NEONvshlls : SDNode<"ARMISD::VSHLLs", SDTARMVSHX>;
47 def NEONvshllu : SDNode<"ARMISD::VSHLLu", SDTARMVSHX>;
48 def NEONvshlli : SDNode<"ARMISD::VSHLLi", SDTARMVSHX>;
49 def NEONvshrn : SDNode<"ARMISD::VSHRN", SDTARMVSHX>;
51 def NEONvrshrs : SDNode<"ARMISD::VRSHRs", SDTARMVSH>;
52 def NEONvrshru : SDNode<"ARMISD::VRSHRu", SDTARMVSH>;
53 def NEONvrshrn : SDNode<"ARMISD::VRSHRN", SDTARMVSHX>;
55 def NEONvqshls : SDNode<"ARMISD::VQSHLs", SDTARMVSH>;
56 def NEONvqshlu : SDNode<"ARMISD::VQSHLu", SDTARMVSH>;
57 def NEONvqshlsu : SDNode<"ARMISD::VQSHLsu", SDTARMVSH>;
58 def NEONvqshrns : SDNode<"ARMISD::VQSHRNs", SDTARMVSHX>;
59 def NEONvqshrnu : SDNode<"ARMISD::VQSHRNu", SDTARMVSHX>;
60 def NEONvqshrnsu : SDNode<"ARMISD::VQSHRNsu", SDTARMVSHX>;
62 def NEONvqrshrns : SDNode<"ARMISD::VQRSHRNs", SDTARMVSHX>;
63 def NEONvqrshrnu : SDNode<"ARMISD::VQRSHRNu", SDTARMVSHX>;
64 def NEONvqrshrnsu : SDNode<"ARMISD::VQRSHRNsu", SDTARMVSHX>;
66 def NEONvsli : SDNode<"ARMISD::VSLI", SDTARMVSHINS>;
67 def NEONvsri : SDNode<"ARMISD::VSRI", SDTARMVSHINS>;
69 def SDTARMVGETLN : SDTypeProfile<1, 2, [SDTCisVT<0, i32>, SDTCisInt<1>,
71 def NEONvgetlaneu : SDNode<"ARMISD::VGETLANEu", SDTARMVGETLN>;
72 def NEONvgetlanes : SDNode<"ARMISD::VGETLANEs", SDTARMVGETLN>;
74 def SDTARMVMOVIMM : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisVT<1, i32>]>;
75 def NEONvmovImm : SDNode<"ARMISD::VMOVIMM", SDTARMVMOVIMM>;
76 def NEONvmvnImm : SDNode<"ARMISD::VMVNIMM", SDTARMVMOVIMM>;
78 def SDTARMVORRIMM : SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
80 def NEONvorrImm : SDNode<"ARMISD::VORRIMM", SDTARMVORRIMM>;
81 def NEONvbicImm : SDNode<"ARMISD::VBICIMM", SDTARMVORRIMM>;
83 def NEONvdup : SDNode<"ARMISD::VDUP", SDTypeProfile<1, 1, [SDTCisVec<0>]>>;
85 // VDUPLANE can produce a quad-register result from a double-register source,
86 // so the result is not constrained to match the source.
87 def NEONvduplane : SDNode<"ARMISD::VDUPLANE",
88 SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisVec<1>,
91 def SDTARMVEXT : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
92 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
93 def NEONvext : SDNode<"ARMISD::VEXT", SDTARMVEXT>;
95 def SDTARMVSHUF : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisSameAs<0, 1>]>;
96 def NEONvrev64 : SDNode<"ARMISD::VREV64", SDTARMVSHUF>;
97 def NEONvrev32 : SDNode<"ARMISD::VREV32", SDTARMVSHUF>;
98 def NEONvrev16 : SDNode<"ARMISD::VREV16", SDTARMVSHUF>;
100 def SDTARMVSHUF2 : SDTypeProfile<2, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
102 SDTCisSameAs<0, 3>]>;
103 def NEONzip : SDNode<"ARMISD::VZIP", SDTARMVSHUF2>;
104 def NEONuzp : SDNode<"ARMISD::VUZP", SDTARMVSHUF2>;
105 def NEONtrn : SDNode<"ARMISD::VTRN", SDTARMVSHUF2>;
107 def SDTARMVMULL : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
108 SDTCisSameAs<1, 2>]>;
109 def NEONvmulls : SDNode<"ARMISD::VMULLs", SDTARMVMULL>;
110 def NEONvmullu : SDNode<"ARMISD::VMULLu", SDTARMVMULL>;
112 def SDTARMFMAX : SDTypeProfile<1, 2, [SDTCisVT<0, f32>, SDTCisSameAs<0, 1>,
113 SDTCisSameAs<0, 2>]>;
114 def NEONfmax : SDNode<"ARMISD::FMAX", SDTARMFMAX>;
115 def NEONfmin : SDNode<"ARMISD::FMIN", SDTARMFMAX>;
117 def NEONimmAllZerosV: PatLeaf<(NEONvmovImm (i32 timm)), [{
118 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
119 unsigned EltBits = 0;
120 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
121 return (EltBits == 32 && EltVal == 0);
124 def NEONimmAllOnesV: PatLeaf<(NEONvmovImm (i32 timm)), [{
125 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
126 unsigned EltBits = 0;
127 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
128 return (EltBits == 8 && EltVal == 0xff);
131 //===----------------------------------------------------------------------===//
132 // NEON operand definitions
133 //===----------------------------------------------------------------------===//
135 def nModImm : Operand<i32> {
136 let PrintMethod = "printNEONModImmOperand";
139 //===----------------------------------------------------------------------===//
140 // NEON load / store instructions
141 //===----------------------------------------------------------------------===//
143 // Use VLDM to load a Q register as a D register pair.
144 // This is a pseudo instruction that is expanded to VLDMD after reg alloc.
146 : PseudoVFPLdStM<(outs QPR:$dst), (ins GPR:$Rn),
148 [(set QPR:$dst, (v2f64 (load GPR:$Rn)))]>;
150 : PseudoVFPLdStM<(outs QPR:$dst), (ins GPR:$Rn),
152 [(set QPR:$dst, (v2f64 (load GPR:$Rn)))]>;
154 // Use VSTM to store a Q register as a D register pair.
155 // This is a pseudo instruction that is expanded to VSTMD after reg alloc.
157 : PseudoVFPLdStM<(outs), (ins QPR:$src, GPR:$Rn),
159 [(store (v2f64 QPR:$src), GPR:$Rn)]>;
161 : PseudoVFPLdStM<(outs), (ins QPR:$src, GPR:$Rn),
163 [(store (v2f64 QPR:$src), GPR:$Rn)]>;
165 // Classes for VLD* pseudo-instructions with multi-register operands.
166 // These are expanded to real instructions after register allocation.
167 class VLDQPseudo<InstrItinClass itin>
168 : PseudoNLdSt<(outs QPR:$dst), (ins addrmode6:$addr), itin, "">;
169 class VLDQWBPseudo<InstrItinClass itin>
170 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
171 (ins addrmode6:$addr, am6offset:$offset), itin,
173 class VLDQQPseudo<InstrItinClass itin>
174 : PseudoNLdSt<(outs QQPR:$dst), (ins addrmode6:$addr), itin, "">;
175 class VLDQQWBPseudo<InstrItinClass itin>
176 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
177 (ins addrmode6:$addr, am6offset:$offset), itin,
179 class VLDQQQQWBPseudo<InstrItinClass itin>
180 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
181 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
182 "$addr.addr = $wb, $src = $dst">;
184 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
186 // VLD1 : Vector Load (multiple single elements)
187 class VLD1D<bits<4> op7_4, string Dt>
188 : NLdSt<0,0b10,0b0111,op7_4, (outs DPR:$Vd),
189 (ins addrmode6:$Rn), IIC_VLD1,
190 "vld1", Dt, "\\{$Vd\\}, $Rn", "", []> {
194 class VLD1Q<bits<4> op7_4, string Dt>
195 : NLdSt<0,0b10,0b1010,op7_4, (outs DPR:$Vd, DPR:$dst2),
196 (ins addrmode6:$Rn), IIC_VLD1x2,
197 "vld1", Dt, "\\{$Vd, $dst2\\}, $Rn", "", []> {
199 let Inst{5-4} = Rn{5-4};
202 def VLD1d8 : VLD1D<{0,0,0,?}, "8">;
203 def VLD1d16 : VLD1D<{0,1,0,?}, "16">;
204 def VLD1d32 : VLD1D<{1,0,0,?}, "32">;
205 def VLD1d64 : VLD1D<{1,1,0,?}, "64">;
207 def VLD1q8 : VLD1Q<{0,0,?,?}, "8">;
208 def VLD1q16 : VLD1Q<{0,1,?,?}, "16">;
209 def VLD1q32 : VLD1Q<{1,0,?,?}, "32">;
210 def VLD1q64 : VLD1Q<{1,1,?,?}, "64">;
212 def VLD1q8Pseudo : VLDQPseudo<IIC_VLD1x2>;
213 def VLD1q16Pseudo : VLDQPseudo<IIC_VLD1x2>;
214 def VLD1q32Pseudo : VLDQPseudo<IIC_VLD1x2>;
215 def VLD1q64Pseudo : VLDQPseudo<IIC_VLD1x2>;
217 // ...with address register writeback:
218 class VLD1DWB<bits<4> op7_4, string Dt>
219 : NLdSt<0,0b10,0b0111,op7_4, (outs DPR:$Vd, GPR:$wb),
220 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD1u,
221 "vld1", Dt, "\\{$Vd\\}, $Rn$Rm",
222 "$Rn.addr = $wb", []> {
225 class VLD1QWB<bits<4> op7_4, string Dt>
226 : NLdSt<0,0b10,0b1010,op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
227 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD1x2u,
228 "vld1", Dt, "\\{$Vd, $dst2\\}, $Rn$Rm",
229 "$Rn.addr = $wb", []> {
230 let Inst{5-4} = Rn{5-4};
233 def VLD1d8_UPD : VLD1DWB<{0,0,0,?}, "8">;
234 def VLD1d16_UPD : VLD1DWB<{0,1,0,?}, "16">;
235 def VLD1d32_UPD : VLD1DWB<{1,0,0,?}, "32">;
236 def VLD1d64_UPD : VLD1DWB<{1,1,0,?}, "64">;
238 def VLD1q8_UPD : VLD1QWB<{0,0,?,?}, "8">;
239 def VLD1q16_UPD : VLD1QWB<{0,1,?,?}, "16">;
240 def VLD1q32_UPD : VLD1QWB<{1,0,?,?}, "32">;
241 def VLD1q64_UPD : VLD1QWB<{1,1,?,?}, "64">;
243 def VLD1q8Pseudo_UPD : VLDQWBPseudo<IIC_VLD1x2u>;
244 def VLD1q16Pseudo_UPD : VLDQWBPseudo<IIC_VLD1x2u>;
245 def VLD1q32Pseudo_UPD : VLDQWBPseudo<IIC_VLD1x2u>;
246 def VLD1q64Pseudo_UPD : VLDQWBPseudo<IIC_VLD1x2u>;
248 // ...with 3 registers (some of these are only for the disassembler):
249 class VLD1D3<bits<4> op7_4, string Dt>
250 : NLdSt<0,0b10,0b0110,op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
251 (ins addrmode6:$Rn), IIC_VLD1x3, "vld1", Dt,
252 "\\{$Vd, $dst2, $dst3\\}, $Rn", "", []> {
256 class VLD1D3WB<bits<4> op7_4, string Dt>
257 : NLdSt<0,0b10,0b0110,op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
258 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD1x3u, "vld1", Dt,
259 "\\{$Vd, $dst2, $dst3\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
263 def VLD1d8T : VLD1D3<{0,0,0,?}, "8">;
264 def VLD1d16T : VLD1D3<{0,1,0,?}, "16">;
265 def VLD1d32T : VLD1D3<{1,0,0,?}, "32">;
266 def VLD1d64T : VLD1D3<{1,1,0,?}, "64">;
268 def VLD1d8T_UPD : VLD1D3WB<{0,0,0,?}, "8">;
269 def VLD1d16T_UPD : VLD1D3WB<{0,1,0,?}, "16">;
270 def VLD1d32T_UPD : VLD1D3WB<{1,0,0,?}, "32">;
271 def VLD1d64T_UPD : VLD1D3WB<{1,1,0,?}, "64">;
273 def VLD1d64TPseudo : VLDQQPseudo<IIC_VLD1x3>;
274 def VLD1d64TPseudo_UPD : VLDQQWBPseudo<IIC_VLD1x3u>;
276 // ...with 4 registers (some of these are only for the disassembler):
277 class VLD1D4<bits<4> op7_4, string Dt>
278 : NLdSt<0,0b10,0b0010,op7_4,(outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
279 (ins addrmode6:$Rn), IIC_VLD1x4, "vld1", Dt,
280 "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn", "", []> {
282 let Inst{5-4} = Rn{5-4};
284 class VLD1D4WB<bits<4> op7_4, string Dt>
285 : NLdSt<0,0b10,0b0010,op7_4,
286 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
287 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD4, "vld1", Dt,
288 "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn$Rm", "$Rn.addr = $wb",
290 let Inst{5-4} = Rn{5-4};
293 def VLD1d8Q : VLD1D4<{0,0,?,?}, "8">;
294 def VLD1d16Q : VLD1D4<{0,1,?,?}, "16">;
295 def VLD1d32Q : VLD1D4<{1,0,?,?}, "32">;
296 def VLD1d64Q : VLD1D4<{1,1,?,?}, "64">;
298 def VLD1d8Q_UPD : VLD1D4WB<{0,0,?,?}, "8">;
299 def VLD1d16Q_UPD : VLD1D4WB<{0,1,?,?}, "16">;
300 def VLD1d32Q_UPD : VLD1D4WB<{1,0,?,?}, "32">;
301 def VLD1d64Q_UPD : VLD1D4WB<{1,1,?,?}, "64">;
303 def VLD1d64QPseudo : VLDQQPseudo<IIC_VLD1x4>;
304 def VLD1d64QPseudo_UPD : VLDQQWBPseudo<IIC_VLD1x4u>;
306 // VLD2 : Vector Load (multiple 2-element structures)
307 class VLD2D<bits<4> op11_8, bits<4> op7_4, string Dt>
308 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2),
309 (ins addrmode6:$Rn), IIC_VLD2,
310 "vld2", Dt, "\\{$Vd, $dst2\\}, $Rn", "", []> {
312 let Inst{5-4} = Rn{5-4};
314 class VLD2Q<bits<4> op7_4, string Dt>
315 : NLdSt<0, 0b10, 0b0011, op7_4,
316 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
317 (ins addrmode6:$Rn), IIC_VLD2x2,
318 "vld2", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn", "", []> {
320 let Inst{5-4} = Rn{5-4};
323 def VLD2d8 : VLD2D<0b1000, {0,0,?,?}, "8">;
324 def VLD2d16 : VLD2D<0b1000, {0,1,?,?}, "16">;
325 def VLD2d32 : VLD2D<0b1000, {1,0,?,?}, "32">;
327 def VLD2q8 : VLD2Q<{0,0,?,?}, "8">;
328 def VLD2q16 : VLD2Q<{0,1,?,?}, "16">;
329 def VLD2q32 : VLD2Q<{1,0,?,?}, "32">;
331 def VLD2d8Pseudo : VLDQPseudo<IIC_VLD2>;
332 def VLD2d16Pseudo : VLDQPseudo<IIC_VLD2>;
333 def VLD2d32Pseudo : VLDQPseudo<IIC_VLD2>;
335 def VLD2q8Pseudo : VLDQQPseudo<IIC_VLD2x2>;
336 def VLD2q16Pseudo : VLDQQPseudo<IIC_VLD2x2>;
337 def VLD2q32Pseudo : VLDQQPseudo<IIC_VLD2x2>;
339 // ...with address register writeback:
340 class VLD2DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
341 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
342 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD2u,
343 "vld2", Dt, "\\{$Vd, $dst2\\}, $Rn$Rm",
344 "$Rn.addr = $wb", []> {
345 let Inst{5-4} = Rn{5-4};
347 class VLD2QWB<bits<4> op7_4, string Dt>
348 : NLdSt<0, 0b10, 0b0011, op7_4,
349 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
350 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD2x2u,
351 "vld2", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn$Rm",
352 "$Rn.addr = $wb", []> {
353 let Inst{5-4} = Rn{5-4};
356 def VLD2d8_UPD : VLD2DWB<0b1000, {0,0,?,?}, "8">;
357 def VLD2d16_UPD : VLD2DWB<0b1000, {0,1,?,?}, "16">;
358 def VLD2d32_UPD : VLD2DWB<0b1000, {1,0,?,?}, "32">;
360 def VLD2q8_UPD : VLD2QWB<{0,0,?,?}, "8">;
361 def VLD2q16_UPD : VLD2QWB<{0,1,?,?}, "16">;
362 def VLD2q32_UPD : VLD2QWB<{1,0,?,?}, "32">;
364 def VLD2d8Pseudo_UPD : VLDQWBPseudo<IIC_VLD2u>;
365 def VLD2d16Pseudo_UPD : VLDQWBPseudo<IIC_VLD2u>;
366 def VLD2d32Pseudo_UPD : VLDQWBPseudo<IIC_VLD2u>;
368 def VLD2q8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD2x2u>;
369 def VLD2q16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD2x2u>;
370 def VLD2q32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD2x2u>;
372 // ...with double-spaced registers (for disassembly only):
373 def VLD2b8 : VLD2D<0b1001, {0,0,?,?}, "8">;
374 def VLD2b16 : VLD2D<0b1001, {0,1,?,?}, "16">;
375 def VLD2b32 : VLD2D<0b1001, {1,0,?,?}, "32">;
376 def VLD2b8_UPD : VLD2DWB<0b1001, {0,0,?,?}, "8">;
377 def VLD2b16_UPD : VLD2DWB<0b1001, {0,1,?,?}, "16">;
378 def VLD2b32_UPD : VLD2DWB<0b1001, {1,0,?,?}, "32">;
380 // VLD3 : Vector Load (multiple 3-element structures)
381 class VLD3D<bits<4> op11_8, bits<4> op7_4, string Dt>
382 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
383 (ins addrmode6:$Rn), IIC_VLD3,
384 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn", "", []> {
389 def VLD3d8 : VLD3D<0b0100, {0,0,0,?}, "8">;
390 def VLD3d16 : VLD3D<0b0100, {0,1,0,?}, "16">;
391 def VLD3d32 : VLD3D<0b0100, {1,0,0,?}, "32">;
393 def VLD3d8Pseudo : VLDQQPseudo<IIC_VLD3>;
394 def VLD3d16Pseudo : VLDQQPseudo<IIC_VLD3>;
395 def VLD3d32Pseudo : VLDQQPseudo<IIC_VLD3>;
397 // ...with address register writeback:
398 class VLD3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
399 : NLdSt<0, 0b10, op11_8, op7_4,
400 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
401 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD3u,
402 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn$Rm",
403 "$Rn.addr = $wb", []> {
407 def VLD3d8_UPD : VLD3DWB<0b0100, {0,0,0,?}, "8">;
408 def VLD3d16_UPD : VLD3DWB<0b0100, {0,1,0,?}, "16">;
409 def VLD3d32_UPD : VLD3DWB<0b0100, {1,0,0,?}, "32">;
411 def VLD3d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
412 def VLD3d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
413 def VLD3d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
415 // ...with double-spaced registers (non-updating versions for disassembly only):
416 def VLD3q8 : VLD3D<0b0101, {0,0,0,?}, "8">;
417 def VLD3q16 : VLD3D<0b0101, {0,1,0,?}, "16">;
418 def VLD3q32 : VLD3D<0b0101, {1,0,0,?}, "32">;
419 def VLD3q8_UPD : VLD3DWB<0b0101, {0,0,0,?}, "8">;
420 def VLD3q16_UPD : VLD3DWB<0b0101, {0,1,0,?}, "16">;
421 def VLD3q32_UPD : VLD3DWB<0b0101, {1,0,0,?}, "32">;
423 def VLD3q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
424 def VLD3q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
425 def VLD3q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
427 // ...alternate versions to be allocated odd register numbers:
428 def VLD3q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
429 def VLD3q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
430 def VLD3q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
432 // VLD4 : Vector Load (multiple 4-element structures)
433 class VLD4D<bits<4> op11_8, bits<4> op7_4, string Dt>
434 : NLdSt<0, 0b10, op11_8, op7_4,
435 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
436 (ins addrmode6:$Rn), IIC_VLD4,
437 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn", "", []> {
439 let Inst{5-4} = Rn{5-4};
442 def VLD4d8 : VLD4D<0b0000, {0,0,?,?}, "8">;
443 def VLD4d16 : VLD4D<0b0000, {0,1,?,?}, "16">;
444 def VLD4d32 : VLD4D<0b0000, {1,0,?,?}, "32">;
446 def VLD4d8Pseudo : VLDQQPseudo<IIC_VLD4>;
447 def VLD4d16Pseudo : VLDQQPseudo<IIC_VLD4>;
448 def VLD4d32Pseudo : VLDQQPseudo<IIC_VLD4>;
450 // ...with address register writeback:
451 class VLD4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
452 : NLdSt<0, 0b10, op11_8, op7_4,
453 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
454 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD4,
455 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn$Rm",
456 "$Rn.addr = $wb", []> {
457 let Inst{5-4} = Rn{5-4};
460 def VLD4d8_UPD : VLD4DWB<0b0000, {0,0,?,?}, "8">;
461 def VLD4d16_UPD : VLD4DWB<0b0000, {0,1,?,?}, "16">;
462 def VLD4d32_UPD : VLD4DWB<0b0000, {1,0,?,?}, "32">;
464 def VLD4d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4>;
465 def VLD4d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4>;
466 def VLD4d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4>;
468 // ...with double-spaced registers (non-updating versions for disassembly only):
469 def VLD4q8 : VLD4D<0b0001, {0,0,?,?}, "8">;
470 def VLD4q16 : VLD4D<0b0001, {0,1,?,?}, "16">;
471 def VLD4q32 : VLD4D<0b0001, {1,0,?,?}, "32">;
472 def VLD4q8_UPD : VLD4DWB<0b0001, {0,0,?,?}, "8">;
473 def VLD4q16_UPD : VLD4DWB<0b0001, {0,1,?,?}, "16">;
474 def VLD4q32_UPD : VLD4DWB<0b0001, {1,0,?,?}, "32">;
476 def VLD4q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4>;
477 def VLD4q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4>;
478 def VLD4q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4>;
480 // ...alternate versions to be allocated odd register numbers:
481 def VLD4q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4>;
482 def VLD4q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4>;
483 def VLD4q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4>;
485 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
487 // Classes for VLD*LN pseudo-instructions with multi-register operands.
488 // These are expanded to real instructions after register allocation.
489 class VLDQLNPseudo<InstrItinClass itin>
490 : PseudoNLdSt<(outs QPR:$dst),
491 (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
492 itin, "$src = $dst">;
493 class VLDQLNWBPseudo<InstrItinClass itin>
494 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
495 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
496 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
497 class VLDQQLNPseudo<InstrItinClass itin>
498 : PseudoNLdSt<(outs QQPR:$dst),
499 (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
500 itin, "$src = $dst">;
501 class VLDQQLNWBPseudo<InstrItinClass itin>
502 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
503 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
504 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
505 class VLDQQQQLNPseudo<InstrItinClass itin>
506 : PseudoNLdSt<(outs QQQQPR:$dst),
507 (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
508 itin, "$src = $dst">;
509 class VLDQQQQLNWBPseudo<InstrItinClass itin>
510 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
511 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
512 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
514 // VLD1LN : Vector Load (single element to one lane)
515 class VLD1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
517 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd),
518 (ins addrmode6:$Rn, DPR:$src, nohash_imm:$lane),
519 IIC_VLD1ln, "vld1", Dt, "\\{$Vd[$lane]\\}, $Rn",
521 [(set DPR:$Vd, (vector_insert (Ty DPR:$src),
522 (i32 (LoadOp addrmode6:$Rn)),
526 class VLD1QLNPseudo<ValueType Ty, PatFrag LoadOp> : VLDQLNPseudo<IIC_VLD1ln> {
527 let Pattern = [(set QPR:$dst, (vector_insert (Ty QPR:$src),
528 (i32 (LoadOp addrmode6:$addr)),
532 def VLD1LNd8 : VLD1LN<0b0000, {?,?,?,0}, "8", v8i8, extloadi8> {
533 let Inst{7-5} = lane{2-0};
535 def VLD1LNd16 : VLD1LN<0b0100, {?,?,0,?}, "16", v4i16, extloadi16> {
536 let Inst{7-6} = lane{1-0};
539 def VLD1LNd32 : VLD1LN<0b1000, {?,0,?,?}, "32", v2i32, load> {
540 let Inst{7} = lane{0};
545 def VLD1LNq8Pseudo : VLD1QLNPseudo<v16i8, extloadi8>;
546 def VLD1LNq16Pseudo : VLD1QLNPseudo<v8i16, extloadi16>;
547 def VLD1LNq32Pseudo : VLD1QLNPseudo<v4i32, load>;
549 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
551 // ...with address register writeback:
552 class VLD1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
553 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, GPR:$wb),
554 (ins addrmode6:$Rn, am6offset:$Rm,
555 DPR:$src, nohash_imm:$lane), IIC_VLD1lnu, "vld1", Dt,
556 "\\{$Vd[$lane]\\}, $Rn$Rm",
557 "$src = $Vd, $Rn.addr = $wb", []>;
559 def VLD1LNd8_UPD : VLD1LNWB<0b0000, {?,?,?,0}, "8"> {
560 let Inst{7-5} = lane{2-0};
562 def VLD1LNd16_UPD : VLD1LNWB<0b0100, {?,?,0,?}, "16"> {
563 let Inst{7-6} = lane{1-0};
566 def VLD1LNd32_UPD : VLD1LNWB<0b1000, {?,0,?,?}, "32"> {
567 let Inst{7} = lane{0};
572 def VLD1LNq8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
573 def VLD1LNq16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
574 def VLD1LNq32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
576 // VLD2LN : Vector Load (single 2-element structure to one lane)
577 class VLD2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
578 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2),
579 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, nohash_imm:$lane),
580 IIC_VLD2ln, "vld2", Dt, "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn",
581 "$src1 = $Vd, $src2 = $dst2", []> {
586 def VLD2LNd8 : VLD2LN<0b0001, {?,?,?,?}, "8"> {
587 let Inst{7-5} = lane{2-0};
589 def VLD2LNd16 : VLD2LN<0b0101, {?,?,0,?}, "16"> {
590 let Inst{7-6} = lane{1-0};
592 def VLD2LNd32 : VLD2LN<0b1001, {?,0,0,?}, "32"> {
593 let Inst{7} = lane{0};
596 def VLD2LNd8Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
597 def VLD2LNd16Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
598 def VLD2LNd32Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
600 // ...with double-spaced registers:
601 def VLD2LNq16 : VLD2LN<0b0101, {?,?,1,?}, "16"> {
602 let Inst{7-6} = lane{1-0};
604 def VLD2LNq32 : VLD2LN<0b1001, {?,1,0,?}, "32"> {
605 let Inst{7} = lane{0};
608 def VLD2LNq16Pseudo : VLDQQLNPseudo<IIC_VLD2ln>;
609 def VLD2LNq32Pseudo : VLDQQLNPseudo<IIC_VLD2ln>;
611 // ...with address register writeback:
612 class VLD2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
613 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
614 (ins addrmode6:$Rn, am6offset:$Rm,
615 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VLD2lnu, "vld2", Dt,
616 "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn$Rm",
617 "$src1 = $Vd, $src2 = $dst2, $Rn.addr = $wb", []> {
621 def VLD2LNd8_UPD : VLD2LNWB<0b0001, {?,?,?,?}, "8"> {
622 let Inst{7-5} = lane{2-0};
624 def VLD2LNd16_UPD : VLD2LNWB<0b0101, {?,?,0,?}, "16"> {
625 let Inst{7-6} = lane{1-0};
627 def VLD2LNd32_UPD : VLD2LNWB<0b1001, {?,0,0,?}, "32"> {
628 let Inst{7} = lane{0};
631 def VLD2LNd8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
632 def VLD2LNd16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
633 def VLD2LNd32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
635 def VLD2LNq16_UPD : VLD2LNWB<0b0101, {?,?,1,?}, "16"> {
636 let Inst{7-6} = lane{1-0};
638 def VLD2LNq32_UPD : VLD2LNWB<0b1001, {?,1,0,?}, "32"> {
639 let Inst{7} = lane{0};
642 def VLD2LNq16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
643 def VLD2LNq32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
645 // VLD3LN : Vector Load (single 3-element structure to one lane)
646 class VLD3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
647 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
648 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3,
649 nohash_imm:$lane), IIC_VLD3ln, "vld3", Dt,
650 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn",
651 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3", []> {
655 def VLD3LNd8 : VLD3LN<0b0010, {?,?,?,0}, "8"> {
656 let Inst{7-5} = lane{2-0};
658 def VLD3LNd16 : VLD3LN<0b0110, {?,?,0,0}, "16"> {
659 let Inst{7-6} = lane{1-0};
661 def VLD3LNd32 : VLD3LN<0b1010, {?,0,0,0}, "32"> {
662 let Inst{7} = lane{0};
665 def VLD3LNd8Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
666 def VLD3LNd16Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
667 def VLD3LNd32Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
669 // ...with double-spaced registers:
670 def VLD3LNq16 : VLD3LN<0b0110, {?,?,1,0}, "16"> {
671 let Inst{7-6} = lane{1-0};
673 def VLD3LNq32 : VLD3LN<0b1010, {?,1,0,0}, "32"> {
674 let Inst{7} = lane{0};
677 def VLD3LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
678 def VLD3LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
680 // ...with address register writeback:
681 class VLD3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
682 : NLdStLn<1, 0b10, op11_8, op7_4,
683 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
684 (ins addrmode6:$Rn, am6offset:$Rm,
685 DPR:$src1, DPR:$src2, DPR:$src3, nohash_imm:$lane),
686 IIC_VLD3lnu, "vld3", Dt,
687 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn$Rm",
688 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $Rn.addr = $wb",
691 def VLD3LNd8_UPD : VLD3LNWB<0b0010, {?,?,?,0}, "8"> {
692 let Inst{7-5} = lane{2-0};
694 def VLD3LNd16_UPD : VLD3LNWB<0b0110, {?,?,0,0}, "16"> {
695 let Inst{7-6} = lane{1-0};
697 def VLD3LNd32_UPD : VLD3LNWB<0b1010, {?,0,0,0}, "32"> {
698 let Inst{7} = lane{0};
701 def VLD3LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
702 def VLD3LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
703 def VLD3LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
705 def VLD3LNq16_UPD : VLD3LNWB<0b0110, {?,?,1,0}, "16"> {
706 let Inst{7-6} = lane{1-0};
708 def VLD3LNq32_UPD : VLD3LNWB<0b1010, {?,1,0,0}, "32"> {
709 let Inst{7} = lane{0};
712 def VLD3LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
713 def VLD3LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
715 // VLD4LN : Vector Load (single 4-element structure to one lane)
716 class VLD4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
717 : NLdStLn<1, 0b10, op11_8, op7_4,
718 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
719 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
720 nohash_imm:$lane), IIC_VLD4ln, "vld4", Dt,
721 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn",
722 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4", []> {
727 def VLD4LNd8 : VLD4LN<0b0011, {?,?,?,?}, "8"> {
728 let Inst{7-5} = lane{2-0};
730 def VLD4LNd16 : VLD4LN<0b0111, {?,?,0,?}, "16"> {
731 let Inst{7-6} = lane{1-0};
733 def VLD4LNd32 : VLD4LN<0b1011, {?,0,?,?}, "32"> {
734 let Inst{7} = lane{0};
738 def VLD4LNd8Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
739 def VLD4LNd16Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
740 def VLD4LNd32Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
742 // ...with double-spaced registers:
743 def VLD4LNq16 : VLD4LN<0b0111, {?,?,1,?}, "16"> {
744 let Inst{7-6} = lane{1-0};
746 def VLD4LNq32 : VLD4LN<0b1011, {?,1,?,?}, "32"> {
747 let Inst{7} = lane{0};
751 def VLD4LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>;
752 def VLD4LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>;
754 // ...with address register writeback:
755 class VLD4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
756 : NLdStLn<1, 0b10, op11_8, op7_4,
757 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
758 (ins addrmode6:$Rn, am6offset:$Rm,
759 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
760 IIC_VLD4ln, "vld4", Dt,
761 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn$Rm",
762 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4, $Rn.addr = $wb",
767 def VLD4LNd8_UPD : VLD4LNWB<0b0011, {?,?,?,?}, "8"> {
768 let Inst{7-5} = lane{2-0};
770 def VLD4LNd16_UPD : VLD4LNWB<0b0111, {?,?,0,?}, "16"> {
771 let Inst{7-6} = lane{1-0};
773 def VLD4LNd32_UPD : VLD4LNWB<0b1011, {?,0,?,?}, "32"> {
774 let Inst{7} = lane{0};
778 def VLD4LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
779 def VLD4LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
780 def VLD4LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
782 def VLD4LNq16_UPD : VLD4LNWB<0b0111, {?,?,1,?}, "16"> {
783 let Inst{7-6} = lane{1-0};
785 def VLD4LNq32_UPD : VLD4LNWB<0b1011, {?,1,?,?}, "32"> {
786 let Inst{7} = lane{0};
790 def VLD4LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
791 def VLD4LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
793 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
795 // VLD1DUP : Vector Load (single element to all lanes)
796 class VLD1DUP<bits<4> op7_4, string Dt, ValueType Ty, PatFrag LoadOp>
797 : NLdSt<1, 0b10, 0b1100, op7_4, (outs DPR:$Vd), (ins addrmode6:$Rn),
798 IIC_VLD1dup, "vld1", Dt, "\\{$Vd[]\\}, $Rn", "",
799 [(set DPR:$Vd, (Ty (NEONvdup (i32 (LoadOp addrmode6:$Rn)))))]> {
803 class VLD1QDUPPseudo<ValueType Ty, PatFrag LoadOp> : VLDQPseudo<IIC_VLD1dup> {
804 let Pattern = [(set QPR:$dst,
805 (Ty (NEONvdup (i32 (LoadOp addrmode6:$addr)))))];
808 def VLD1DUPd8 : VLD1DUP<{0,0,0,?}, "8", v8i8, extloadi8>;
809 def VLD1DUPd16 : VLD1DUP<{0,1,0,?}, "16", v4i16, extloadi16>;
810 def VLD1DUPd32 : VLD1DUP<{1,0,0,?}, "32", v2i32, load>;
812 def VLD1DUPq8Pseudo : VLD1QDUPPseudo<v16i8, extloadi8>;
813 def VLD1DUPq16Pseudo : VLD1QDUPPseudo<v8i16, extloadi16>;
814 def VLD1DUPq32Pseudo : VLD1QDUPPseudo<v4i32, load>;
816 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
818 class VLD1QDUP<bits<4> op7_4, string Dt, ValueType Ty, PatFrag LoadOp>
819 : NLdSt<1, 0b10, 0b1100, op7_4, (outs DPR:$Vd, DPR:$dst2),
820 (ins addrmode6:$Rn), IIC_VLD1dup,
821 "vld1", Dt, "\\{$Vd[], $dst2[]\\}, $Rn", "", []> {
826 def VLD1DUPq8 : VLD1QDUP<{0,0,1,0}, "8", v16i8, extloadi8>;
827 def VLD1DUPq16 : VLD1QDUP<{0,1,1,?}, "16", v8i16, extloadi16>;
828 def VLD1DUPq32 : VLD1QDUP<{1,0,1,?}, "32", v4i32, load>;
830 // ...with address register writeback:
831 class VLD1DUPWB<bits<4> op7_4, string Dt>
832 : NLdSt<1, 0b10, 0b1100, op7_4, (outs DPR:$Vd, GPR:$wb),
833 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD1dupu,
834 "vld1", Dt, "\\{$Vd[]\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
837 class VLD1QDUPWB<bits<4> op7_4, string Dt>
838 : NLdSt<1, 0b10, 0b1100, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
839 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD1dupu,
840 "vld1", Dt, "\\{$Vd[], $dst2[]\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
844 def VLD1DUPd8_UPD : VLD1DUPWB<{0,0,0,0}, "8">;
845 def VLD1DUPd16_UPD : VLD1DUPWB<{0,1,0,?}, "16">;
846 def VLD1DUPd32_UPD : VLD1DUPWB<{1,0,0,?}, "32">;
848 def VLD1DUPq8_UPD : VLD1QDUPWB<{0,0,1,0}, "8">;
849 def VLD1DUPq16_UPD : VLD1QDUPWB<{0,1,1,?}, "16">;
850 def VLD1DUPq32_UPD : VLD1QDUPWB<{1,0,1,?}, "32">;
852 def VLD1DUPq8Pseudo_UPD : VLDQWBPseudo<IIC_VLD1dupu>;
853 def VLD1DUPq16Pseudo_UPD : VLDQWBPseudo<IIC_VLD1dupu>;
854 def VLD1DUPq32Pseudo_UPD : VLDQWBPseudo<IIC_VLD1dupu>;
856 // VLD2DUP : Vector Load (single 2-element structure to all lanes)
857 class VLD2DUP<bits<4> op7_4, string Dt>
858 : NLdSt<1, 0b10, 0b1101, op7_4, (outs DPR:$Vd, DPR:$dst2),
859 (ins addrmode6:$Rn), IIC_VLD2dup,
860 "vld2", Dt, "\\{$Vd[], $dst2[]\\}, $Rn", "", []> {
865 def VLD2DUPd8 : VLD2DUP<{0,0,0,?}, "8">;
866 def VLD2DUPd16 : VLD2DUP<{0,1,0,?}, "16">;
867 def VLD2DUPd32 : VLD2DUP<{1,0,0,?}, "32">;
869 def VLD2DUPd8Pseudo : VLDQPseudo<IIC_VLD2dup>;
870 def VLD2DUPd16Pseudo : VLDQPseudo<IIC_VLD2dup>;
871 def VLD2DUPd32Pseudo : VLDQPseudo<IIC_VLD2dup>;
873 // ...with double-spaced registers (not used for codegen):
874 def VLD2DUPd8Q : VLD2DUP<{0,0,1,?}, "8">;
875 def VLD2DUPd16Q : VLD2DUP<{0,1,1,?}, "16">;
876 def VLD2DUPd32Q : VLD2DUP<{1,0,1,?}, "32">;
878 // ...with address register writeback:
879 class VLD2DUPWB<bits<4> op7_4, string Dt>
880 : NLdSt<1, 0b10, 0b1101, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
881 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD2dupu,
882 "vld2", Dt, "\\{$Vd[], $dst2[]\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
886 def VLD2DUPd8_UPD : VLD2DUPWB<{0,0,0,0}, "8">;
887 def VLD2DUPd16_UPD : VLD2DUPWB<{0,1,0,?}, "16">;
888 def VLD2DUPd32_UPD : VLD2DUPWB<{1,0,0,?}, "32">;
890 def VLD2DUPd8Q_UPD : VLD2DUPWB<{0,0,1,0}, "8">;
891 def VLD2DUPd16Q_UPD : VLD2DUPWB<{0,1,1,?}, "16">;
892 def VLD2DUPd32Q_UPD : VLD2DUPWB<{1,0,1,?}, "32">;
894 def VLD2DUPd8Pseudo_UPD : VLDQWBPseudo<IIC_VLD2dupu>;
895 def VLD2DUPd16Pseudo_UPD : VLDQWBPseudo<IIC_VLD2dupu>;
896 def VLD2DUPd32Pseudo_UPD : VLDQWBPseudo<IIC_VLD2dupu>;
898 // VLD3DUP : Vector Load (single 3-element structure to all lanes)
899 class VLD3DUP<bits<4> op7_4, string Dt>
900 : NLdSt<1, 0b10, 0b1110, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
901 (ins addrmode6:$Rn), IIC_VLD3dup,
902 "vld3", Dt, "\\{$Vd[], $dst2[], $dst3[]\\}, $Rn", "", []> {
907 def VLD3DUPd8 : VLD3DUP<{0,0,0,?}, "8">;
908 def VLD3DUPd16 : VLD3DUP<{0,1,0,?}, "16">;
909 def VLD3DUPd32 : VLD3DUP<{1,0,0,?}, "32">;
911 def VLD3DUPd8Pseudo : VLDQQPseudo<IIC_VLD3dup>;
912 def VLD3DUPd16Pseudo : VLDQQPseudo<IIC_VLD3dup>;
913 def VLD3DUPd32Pseudo : VLDQQPseudo<IIC_VLD3dup>;
915 // ...with double-spaced registers (not used for codegen):
916 def VLD3DUPd8T : VLD3DUP<{0,0,1,?}, "8">;
917 def VLD3DUPd16T : VLD3DUP<{0,1,1,?}, "16">;
918 def VLD3DUPd32T : VLD3DUP<{1,0,1,?}, "32">;
920 // ...with address register writeback:
921 class VLD3DUPWB<bits<4> op7_4, string Dt>
922 : NLdSt<1, 0b10, 0b1110, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
923 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD3dupu,
924 "vld3", Dt, "\\{$Vd[], $dst2[], $dst3[]\\}, $Rn$Rm",
925 "$Rn.addr = $wb", []> {
929 def VLD3DUPd8_UPD : VLD3DUPWB<{0,0,0,0}, "8">;
930 def VLD3DUPd16_UPD : VLD3DUPWB<{0,1,0,?}, "16">;
931 def VLD3DUPd32_UPD : VLD3DUPWB<{1,0,0,?}, "32">;
933 def VLD3DUPd8T_UPD : VLD3DUPWB<{0,0,1,0}, "8">;
934 def VLD3DUPd16T_UPD : VLD3DUPWB<{0,1,1,?}, "16">;
935 def VLD3DUPd32T_UPD : VLD3DUPWB<{1,0,1,?}, "32">;
937 def VLD3DUPd8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
938 def VLD3DUPd16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
939 def VLD3DUPd32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
941 // VLD4DUP : Vector Load (single 4-element structure to all lanes)
942 // FIXME: Not yet implemented.
943 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
945 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
947 // Classes for VST* pseudo-instructions with multi-register operands.
948 // These are expanded to real instructions after register allocation.
949 class VSTQPseudo<InstrItinClass itin>
950 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src), itin, "">;
951 class VSTQWBPseudo<InstrItinClass itin>
952 : PseudoNLdSt<(outs GPR:$wb),
953 (ins addrmode6:$addr, am6offset:$offset, QPR:$src), itin,
955 class VSTQQPseudo<InstrItinClass itin>
956 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src), itin, "">;
957 class VSTQQWBPseudo<InstrItinClass itin>
958 : PseudoNLdSt<(outs GPR:$wb),
959 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src), itin,
961 class VSTQQQQWBPseudo<InstrItinClass itin>
962 : PseudoNLdSt<(outs GPR:$wb),
963 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
966 // VST1 : Vector Store (multiple single elements)
967 class VST1D<bits<4> op7_4, string Dt>
968 : NLdSt<0,0b00,0b0111,op7_4, (outs), (ins addrmode6:$Rn, DPR:$Vd),
969 IIC_VST1, "vst1", Dt, "\\{$Vd\\}, $Rn", "", []> {
973 class VST1Q<bits<4> op7_4, string Dt>
974 : NLdSt<0,0b00,0b1010,op7_4, (outs),
975 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2), IIC_VST1x2,
976 "vst1", Dt, "\\{$Vd, $src2\\}, $Rn", "", []> {
978 let Inst{5-4} = Rn{5-4};
981 def VST1d8 : VST1D<{0,0,0,?}, "8">;
982 def VST1d16 : VST1D<{0,1,0,?}, "16">;
983 def VST1d32 : VST1D<{1,0,0,?}, "32">;
984 def VST1d64 : VST1D<{1,1,0,?}, "64">;
986 def VST1q8 : VST1Q<{0,0,?,?}, "8">;
987 def VST1q16 : VST1Q<{0,1,?,?}, "16">;
988 def VST1q32 : VST1Q<{1,0,?,?}, "32">;
989 def VST1q64 : VST1Q<{1,1,?,?}, "64">;
991 def VST1q8Pseudo : VSTQPseudo<IIC_VST1x2>;
992 def VST1q16Pseudo : VSTQPseudo<IIC_VST1x2>;
993 def VST1q32Pseudo : VSTQPseudo<IIC_VST1x2>;
994 def VST1q64Pseudo : VSTQPseudo<IIC_VST1x2>;
996 // ...with address register writeback:
997 class VST1DWB<bits<4> op7_4, string Dt>
998 : NLdSt<0, 0b00, 0b0111, op7_4, (outs GPR:$wb),
999 (ins addrmode6:$Rn, am6offset:$Rm, DPR:$Vd), IIC_VST1u,
1000 "vst1", Dt, "\\{$Vd\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
1001 let Inst{4} = Rn{4};
1003 class VST1QWB<bits<4> op7_4, string Dt>
1004 : NLdSt<0, 0b00, 0b1010, op7_4, (outs GPR:$wb),
1005 (ins addrmode6:$Rn, am6offset:$Rm, DPR:$Vd, DPR:$src2),
1006 IIC_VST1x2u, "vst1", Dt, "\\{$Vd, $src2\\}, $Rn$Rm",
1007 "$Rn.addr = $wb", []> {
1008 let Inst{5-4} = Rn{5-4};
1011 def VST1d8_UPD : VST1DWB<{0,0,0,?}, "8">;
1012 def VST1d16_UPD : VST1DWB<{0,1,0,?}, "16">;
1013 def VST1d32_UPD : VST1DWB<{1,0,0,?}, "32">;
1014 def VST1d64_UPD : VST1DWB<{1,1,0,?}, "64">;
1016 def VST1q8_UPD : VST1QWB<{0,0,?,?}, "8">;
1017 def VST1q16_UPD : VST1QWB<{0,1,?,?}, "16">;
1018 def VST1q32_UPD : VST1QWB<{1,0,?,?}, "32">;
1019 def VST1q64_UPD : VST1QWB<{1,1,?,?}, "64">;
1021 def VST1q8Pseudo_UPD : VSTQWBPseudo<IIC_VST1x2u>;
1022 def VST1q16Pseudo_UPD : VSTQWBPseudo<IIC_VST1x2u>;
1023 def VST1q32Pseudo_UPD : VSTQWBPseudo<IIC_VST1x2u>;
1024 def VST1q64Pseudo_UPD : VSTQWBPseudo<IIC_VST1x2u>;
1026 // ...with 3 registers (some of these are only for the disassembler):
1027 class VST1D3<bits<4> op7_4, string Dt>
1028 : NLdSt<0, 0b00, 0b0110, op7_4, (outs),
1029 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3),
1030 IIC_VST1x3, "vst1", Dt, "\\{$Vd, $src2, $src3\\}, $Rn", "", []> {
1032 let Inst{4} = Rn{4};
1034 class VST1D3WB<bits<4> op7_4, string Dt>
1035 : NLdSt<0, 0b00, 0b0110, op7_4, (outs GPR:$wb),
1036 (ins addrmode6:$Rn, am6offset:$Rm,
1037 DPR:$Vd, DPR:$src2, DPR:$src3),
1038 IIC_VST1x3u, "vst1", Dt, "\\{$Vd, $src2, $src3\\}, $Rn$Rm",
1039 "$Rn.addr = $wb", []> {
1040 let Inst{4} = Rn{4};
1043 def VST1d8T : VST1D3<{0,0,0,?}, "8">;
1044 def VST1d16T : VST1D3<{0,1,0,?}, "16">;
1045 def VST1d32T : VST1D3<{1,0,0,?}, "32">;
1046 def VST1d64T : VST1D3<{1,1,0,?}, "64">;
1048 def VST1d8T_UPD : VST1D3WB<{0,0,0,?}, "8">;
1049 def VST1d16T_UPD : VST1D3WB<{0,1,0,?}, "16">;
1050 def VST1d32T_UPD : VST1D3WB<{1,0,0,?}, "32">;
1051 def VST1d64T_UPD : VST1D3WB<{1,1,0,?}, "64">;
1053 def VST1d64TPseudo : VSTQQPseudo<IIC_VST1x3>;
1054 def VST1d64TPseudo_UPD : VSTQQWBPseudo<IIC_VST1x3u>;
1056 // ...with 4 registers (some of these are only for the disassembler):
1057 class VST1D4<bits<4> op7_4, string Dt>
1058 : NLdSt<0, 0b00, 0b0010, op7_4, (outs),
1059 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
1060 IIC_VST1x4, "vst1", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn", "",
1063 let Inst{5-4} = Rn{5-4};
1065 class VST1D4WB<bits<4> op7_4, string Dt>
1066 : NLdSt<0, 0b00, 0b0010, op7_4, (outs GPR:$wb),
1067 (ins addrmode6:$Rn, am6offset:$Rm,
1068 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST1x4u,
1069 "vst1", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
1070 "$Rn.addr = $wb", []> {
1071 let Inst{5-4} = Rn{5-4};
1074 def VST1d8Q : VST1D4<{0,0,?,?}, "8">;
1075 def VST1d16Q : VST1D4<{0,1,?,?}, "16">;
1076 def VST1d32Q : VST1D4<{1,0,?,?}, "32">;
1077 def VST1d64Q : VST1D4<{1,1,?,?}, "64">;
1079 def VST1d8Q_UPD : VST1D4WB<{0,0,?,?}, "8">;
1080 def VST1d16Q_UPD : VST1D4WB<{0,1,?,?}, "16">;
1081 def VST1d32Q_UPD : VST1D4WB<{1,0,?,?}, "32">;
1082 def VST1d64Q_UPD : VST1D4WB<{1,1,?,?}, "64">;
1084 def VST1d64QPseudo : VSTQQPseudo<IIC_VST1x4>;
1085 def VST1d64QPseudo_UPD : VSTQQWBPseudo<IIC_VST1x4u>;
1087 // VST2 : Vector Store (multiple 2-element structures)
1088 class VST2D<bits<4> op11_8, bits<4> op7_4, string Dt>
1089 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1090 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2),
1091 IIC_VST2, "vst2", Dt, "\\{$Vd, $src2\\}, $Rn", "", []> {
1093 let Inst{5-4} = Rn{5-4};
1095 class VST2Q<bits<4> op7_4, string Dt>
1096 : NLdSt<0, 0b00, 0b0011, op7_4, (outs),
1097 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
1098 IIC_VST2x2, "vst2", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn",
1101 let Inst{5-4} = Rn{5-4};
1104 def VST2d8 : VST2D<0b1000, {0,0,?,?}, "8">;
1105 def VST2d16 : VST2D<0b1000, {0,1,?,?}, "16">;
1106 def VST2d32 : VST2D<0b1000, {1,0,?,?}, "32">;
1108 def VST2q8 : VST2Q<{0,0,?,?}, "8">;
1109 def VST2q16 : VST2Q<{0,1,?,?}, "16">;
1110 def VST2q32 : VST2Q<{1,0,?,?}, "32">;
1112 def VST2d8Pseudo : VSTQPseudo<IIC_VST2>;
1113 def VST2d16Pseudo : VSTQPseudo<IIC_VST2>;
1114 def VST2d32Pseudo : VSTQPseudo<IIC_VST2>;
1116 def VST2q8Pseudo : VSTQQPseudo<IIC_VST2x2>;
1117 def VST2q16Pseudo : VSTQQPseudo<IIC_VST2x2>;
1118 def VST2q32Pseudo : VSTQQPseudo<IIC_VST2x2>;
1120 // ...with address register writeback:
1121 class VST2DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1122 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1123 (ins addrmode6:$Rn, am6offset:$Rm, DPR:$Vd, DPR:$src2),
1124 IIC_VST2u, "vst2", Dt, "\\{$Vd, $src2\\}, $Rn$Rm",
1125 "$Rn.addr = $wb", []> {
1126 let Inst{5-4} = Rn{5-4};
1128 class VST2QWB<bits<4> op7_4, string Dt>
1129 : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
1130 (ins addrmode6:$Rn, am6offset:$Rm,
1131 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST2x2u,
1132 "vst2", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
1133 "$Rn.addr = $wb", []> {
1134 let Inst{5-4} = Rn{5-4};
1137 def VST2d8_UPD : VST2DWB<0b1000, {0,0,?,?}, "8">;
1138 def VST2d16_UPD : VST2DWB<0b1000, {0,1,?,?}, "16">;
1139 def VST2d32_UPD : VST2DWB<0b1000, {1,0,?,?}, "32">;
1141 def VST2q8_UPD : VST2QWB<{0,0,?,?}, "8">;
1142 def VST2q16_UPD : VST2QWB<{0,1,?,?}, "16">;
1143 def VST2q32_UPD : VST2QWB<{1,0,?,?}, "32">;
1145 def VST2d8Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
1146 def VST2d16Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
1147 def VST2d32Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
1149 def VST2q8Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
1150 def VST2q16Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
1151 def VST2q32Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
1153 // ...with double-spaced registers (for disassembly only):
1154 def VST2b8 : VST2D<0b1001, {0,0,?,?}, "8">;
1155 def VST2b16 : VST2D<0b1001, {0,1,?,?}, "16">;
1156 def VST2b32 : VST2D<0b1001, {1,0,?,?}, "32">;
1157 def VST2b8_UPD : VST2DWB<0b1001, {0,0,?,?}, "8">;
1158 def VST2b16_UPD : VST2DWB<0b1001, {0,1,?,?}, "16">;
1159 def VST2b32_UPD : VST2DWB<0b1001, {1,0,?,?}, "32">;
1161 // VST3 : Vector Store (multiple 3-element structures)
1162 class VST3D<bits<4> op11_8, bits<4> op7_4, string Dt>
1163 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1164 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3,
1165 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn", "", []> {
1167 let Inst{4} = Rn{4};
1170 def VST3d8 : VST3D<0b0100, {0,0,0,?}, "8">;
1171 def VST3d16 : VST3D<0b0100, {0,1,0,?}, "16">;
1172 def VST3d32 : VST3D<0b0100, {1,0,0,?}, "32">;
1174 def VST3d8Pseudo : VSTQQPseudo<IIC_VST3>;
1175 def VST3d16Pseudo : VSTQQPseudo<IIC_VST3>;
1176 def VST3d32Pseudo : VSTQQPseudo<IIC_VST3>;
1178 // ...with address register writeback:
1179 class VST3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1180 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1181 (ins addrmode6:$Rn, am6offset:$Rm,
1182 DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3u,
1183 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn$Rm",
1184 "$Rn.addr = $wb", []> {
1185 let Inst{4} = Rn{4};
1188 def VST3d8_UPD : VST3DWB<0b0100, {0,0,0,?}, "8">;
1189 def VST3d16_UPD : VST3DWB<0b0100, {0,1,0,?}, "16">;
1190 def VST3d32_UPD : VST3DWB<0b0100, {1,0,0,?}, "32">;
1192 def VST3d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1193 def VST3d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1194 def VST3d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1196 // ...with double-spaced registers (non-updating versions for disassembly only):
1197 def VST3q8 : VST3D<0b0101, {0,0,0,?}, "8">;
1198 def VST3q16 : VST3D<0b0101, {0,1,0,?}, "16">;
1199 def VST3q32 : VST3D<0b0101, {1,0,0,?}, "32">;
1200 def VST3q8_UPD : VST3DWB<0b0101, {0,0,0,?}, "8">;
1201 def VST3q16_UPD : VST3DWB<0b0101, {0,1,0,?}, "16">;
1202 def VST3q32_UPD : VST3DWB<0b0101, {1,0,0,?}, "32">;
1204 def VST3q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1205 def VST3q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1206 def VST3q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1208 // ...alternate versions to be allocated odd register numbers:
1209 def VST3q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1210 def VST3q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1211 def VST3q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1213 // VST4 : Vector Store (multiple 4-element structures)
1214 class VST4D<bits<4> op11_8, bits<4> op7_4, string Dt>
1215 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1216 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
1217 IIC_VST4, "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn",
1220 let Inst{5-4} = Rn{5-4};
1223 def VST4d8 : VST4D<0b0000, {0,0,?,?}, "8">;
1224 def VST4d16 : VST4D<0b0000, {0,1,?,?}, "16">;
1225 def VST4d32 : VST4D<0b0000, {1,0,?,?}, "32">;
1227 def VST4d8Pseudo : VSTQQPseudo<IIC_VST4>;
1228 def VST4d16Pseudo : VSTQQPseudo<IIC_VST4>;
1229 def VST4d32Pseudo : VSTQQPseudo<IIC_VST4>;
1231 // ...with address register writeback:
1232 class VST4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1233 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1234 (ins addrmode6:$Rn, am6offset:$Rm,
1235 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST4u,
1236 "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
1237 "$Rn.addr = $wb", []> {
1238 let Inst{5-4} = Rn{5-4};
1241 def VST4d8_UPD : VST4DWB<0b0000, {0,0,?,?}, "8">;
1242 def VST4d16_UPD : VST4DWB<0b0000, {0,1,?,?}, "16">;
1243 def VST4d32_UPD : VST4DWB<0b0000, {1,0,?,?}, "32">;
1245 def VST4d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1246 def VST4d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1247 def VST4d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1249 // ...with double-spaced registers (non-updating versions for disassembly only):
1250 def VST4q8 : VST4D<0b0001, {0,0,?,?}, "8">;
1251 def VST4q16 : VST4D<0b0001, {0,1,?,?}, "16">;
1252 def VST4q32 : VST4D<0b0001, {1,0,?,?}, "32">;
1253 def VST4q8_UPD : VST4DWB<0b0001, {0,0,?,?}, "8">;
1254 def VST4q16_UPD : VST4DWB<0b0001, {0,1,?,?}, "16">;
1255 def VST4q32_UPD : VST4DWB<0b0001, {1,0,?,?}, "32">;
1257 def VST4q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1258 def VST4q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1259 def VST4q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1261 // ...alternate versions to be allocated odd register numbers:
1262 def VST4q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1263 def VST4q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1264 def VST4q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1266 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
1268 // Classes for VST*LN pseudo-instructions with multi-register operands.
1269 // These are expanded to real instructions after register allocation.
1270 class VSTQLNPseudo<InstrItinClass itin>
1271 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
1273 class VSTQLNWBPseudo<InstrItinClass itin>
1274 : PseudoNLdSt<(outs GPR:$wb),
1275 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
1276 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1277 class VSTQQLNPseudo<InstrItinClass itin>
1278 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
1280 class VSTQQLNWBPseudo<InstrItinClass itin>
1281 : PseudoNLdSt<(outs GPR:$wb),
1282 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
1283 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1284 class VSTQQQQLNPseudo<InstrItinClass itin>
1285 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
1287 class VSTQQQQLNWBPseudo<InstrItinClass itin>
1288 : PseudoNLdSt<(outs GPR:$wb),
1289 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
1290 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1292 // VST1LN : Vector Store (single element from one lane)
1293 class VST1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1294 PatFrag StoreOp, SDNode ExtractOp>
1295 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1296 (ins addrmode6:$Rn, DPR:$Vd, nohash_imm:$lane),
1297 IIC_VST1ln, "vst1", Dt, "\\{$Vd[$lane]\\}, $Rn", "",
1298 [(StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane), addrmode6:$Rn)]> {
1301 class VST1QLNPseudo<ValueType Ty, PatFrag StoreOp, SDNode ExtractOp>
1302 : VSTQLNPseudo<IIC_VST1ln> {
1303 let Pattern = [(StoreOp (ExtractOp (Ty QPR:$src), imm:$lane),
1307 def VST1LNd8 : VST1LN<0b0000, {?,?,?,0}, "8", v8i8, truncstorei8,
1309 let Inst{7-5} = lane{2-0};
1311 def VST1LNd16 : VST1LN<0b0100, {?,?,0,?}, "16", v4i16, truncstorei16,
1313 let Inst{7-6} = lane{1-0};
1314 let Inst{4} = Rn{5};
1316 def VST1LNd32 : VST1LN<0b1000, {?,0,?,?}, "32", v2i32, store, extractelt> {
1317 let Inst{7} = lane{0};
1318 let Inst{5-4} = Rn{5-4};
1321 def VST1LNq8Pseudo : VST1QLNPseudo<v16i8, truncstorei8, NEONvgetlaneu>;
1322 def VST1LNq16Pseudo : VST1QLNPseudo<v8i16, truncstorei16, NEONvgetlaneu>;
1323 def VST1LNq32Pseudo : VST1QLNPseudo<v4i32, store, extractelt>;
1325 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
1327 // ...with address register writeback:
1328 class VST1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1329 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1330 (ins addrmode6:$Rn, am6offset:$Rm,
1331 DPR:$Vd, nohash_imm:$lane), IIC_VST1lnu, "vst1", Dt,
1332 "\\{$Vd[$lane]\\}, $Rn$Rm",
1333 "$Rn.addr = $wb", []>;
1335 def VST1LNd8_UPD : VST1LNWB<0b0000, {?,?,?,0}, "8"> {
1336 let Inst{7-5} = lane{2-0};
1338 def VST1LNd16_UPD : VST1LNWB<0b0100, {?,?,0,?}, "16"> {
1339 let Inst{7-6} = lane{1-0};
1340 let Inst{4} = Rn{5};
1342 def VST1LNd32_UPD : VST1LNWB<0b1000, {?,0,?,?}, "32"> {
1343 let Inst{7} = lane{0};
1344 let Inst{5-4} = Rn{5-4};
1347 def VST1LNq8Pseudo_UPD : VSTQLNWBPseudo<IIC_VST1lnu>;
1348 def VST1LNq16Pseudo_UPD : VSTQLNWBPseudo<IIC_VST1lnu>;
1349 def VST1LNq32Pseudo_UPD : VSTQLNWBPseudo<IIC_VST1lnu>;
1351 // VST2LN : Vector Store (single 2-element structure from one lane)
1352 class VST2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1353 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1354 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, nohash_imm:$lane),
1355 IIC_VST2ln, "vst2", Dt, "\\{$Vd[$lane], $src2[$lane]\\}, $Rn",
1358 let Inst{4} = Rn{4};
1361 def VST2LNd8 : VST2LN<0b0001, {?,?,?,?}, "8"> {
1362 let Inst{7-5} = lane{2-0};
1364 def VST2LNd16 : VST2LN<0b0101, {?,?,0,?}, "16"> {
1365 let Inst{7-6} = lane{1-0};
1367 def VST2LNd32 : VST2LN<0b1001, {?,0,0,?}, "32"> {
1368 let Inst{7} = lane{0};
1371 def VST2LNd8Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1372 def VST2LNd16Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1373 def VST2LNd32Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1375 // ...with double-spaced registers:
1376 def VST2LNq16 : VST2LN<0b0101, {?,?,1,?}, "16"> {
1377 let Inst{7-6} = lane{1-0};
1378 let Inst{4} = Rn{4};
1380 def VST2LNq32 : VST2LN<0b1001, {?,1,0,?}, "32"> {
1381 let Inst{7} = lane{0};
1382 let Inst{4} = Rn{4};
1385 def VST2LNq16Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
1386 def VST2LNq32Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
1388 // ...with address register writeback:
1389 class VST2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1390 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1391 (ins addrmode6:$addr, am6offset:$offset,
1392 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VST2lnu, "vst2", Dt,
1393 "\\{$src1[$lane], $src2[$lane]\\}, $addr$offset",
1394 "$addr.addr = $wb", []> {
1395 let Inst{4} = Rn{4};
1398 def VST2LNd8_UPD : VST2LNWB<0b0001, {?,?,?,?}, "8"> {
1399 let Inst{7-5} = lane{2-0};
1401 def VST2LNd16_UPD : VST2LNWB<0b0101, {?,?,0,?}, "16"> {
1402 let Inst{7-6} = lane{1-0};
1404 def VST2LNd32_UPD : VST2LNWB<0b1001, {?,0,0,?}, "32"> {
1405 let Inst{7} = lane{0};
1408 def VST2LNd8Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1409 def VST2LNd16Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1410 def VST2LNd32Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1412 def VST2LNq16_UPD : VST2LNWB<0b0101, {?,?,1,?}, "16"> {
1413 let Inst{7-6} = lane{1-0};
1415 def VST2LNq32_UPD : VST2LNWB<0b1001, {?,1,0,?}, "32"> {
1416 let Inst{7} = lane{0};
1419 def VST2LNq16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
1420 def VST2LNq32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
1422 // VST3LN : Vector Store (single 3-element structure from one lane)
1423 class VST3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1424 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1425 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3,
1426 nohash_imm:$lane), IIC_VST3ln, "vst3", Dt,
1427 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn", "", []> {
1431 def VST3LNd8 : VST3LN<0b0010, {?,?,?,0}, "8"> {
1432 let Inst{7-5} = lane{2-0};
1434 def VST3LNd16 : VST3LN<0b0110, {?,?,0,0}, "16"> {
1435 let Inst{7-6} = lane{1-0};
1437 def VST3LNd32 : VST3LN<0b1010, {?,0,0,0}, "32"> {
1438 let Inst{7} = lane{0};
1441 def VST3LNd8Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1442 def VST3LNd16Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1443 def VST3LNd32Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1445 // ...with double-spaced registers:
1446 def VST3LNq16 : VST3LN<0b0110, {?,?,1,0}, "16"> {
1447 let Inst{7-6} = lane{1-0};
1449 def VST3LNq32 : VST3LN<0b1010, {?,1,0,0}, "32"> {
1450 let Inst{7} = lane{0};
1453 def VST3LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
1454 def VST3LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
1456 // ...with address register writeback:
1457 class VST3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1458 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1459 (ins addrmode6:$Rn, am6offset:$Rm,
1460 DPR:$Vd, DPR:$src2, DPR:$src3, nohash_imm:$lane),
1461 IIC_VST3lnu, "vst3", Dt,
1462 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn$Rm",
1463 "$Rn.addr = $wb", []>;
1465 def VST3LNd8_UPD : VST3LNWB<0b0010, {?,?,?,0}, "8"> {
1466 let Inst{7-5} = lane{2-0};
1468 def VST3LNd16_UPD : VST3LNWB<0b0110, {?,?,0,0}, "16"> {
1469 let Inst{7-6} = lane{1-0};
1471 def VST3LNd32_UPD : VST3LNWB<0b1010, {?,0,0,0}, "32"> {
1472 let Inst{7} = lane{0};
1475 def VST3LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1476 def VST3LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1477 def VST3LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1479 def VST3LNq16_UPD : VST3LNWB<0b0110, {?,?,1,0}, "16"> {
1480 let Inst{7-6} = lane{1-0};
1482 def VST3LNq32_UPD : VST3LNWB<0b1010, {?,1,0,0}, "32"> {
1483 let Inst{7} = lane{0};
1486 def VST3LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
1487 def VST3LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
1489 // VST4LN : Vector Store (single 4-element structure from one lane)
1490 class VST4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1491 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1492 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4,
1493 nohash_imm:$lane), IIC_VST4ln, "vst4", Dt,
1494 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn",
1497 let Inst{4} = Rn{4};
1500 def VST4LNd8 : VST4LN<0b0011, {?,?,?,?}, "8"> {
1501 let Inst{7-5} = lane{2-0};
1503 def VST4LNd16 : VST4LN<0b0111, {?,?,0,?}, "16"> {
1504 let Inst{7-6} = lane{1-0};
1506 def VST4LNd32 : VST4LN<0b1011, {?,0,?,?}, "32"> {
1507 let Inst{7} = lane{0};
1508 let Inst{5} = Rn{5};
1511 def VST4LNd8Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1512 def VST4LNd16Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1513 def VST4LNd32Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1515 // ...with double-spaced registers:
1516 def VST4LNq16 : VST4LN<0b0111, {?,?,1,?}, "16"> {
1517 let Inst{7-6} = lane{1-0};
1519 def VST4LNq32 : VST4LN<0b1011, {?,1,?,?}, "32"> {
1520 let Inst{7} = lane{0};
1521 let Inst{5} = Rn{5};
1524 def VST4LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
1525 def VST4LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
1527 // ...with address register writeback:
1528 class VST4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1529 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1530 (ins addrmode6:$Rn, am6offset:$Rm,
1531 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
1532 IIC_VST4lnu, "vst4", Dt,
1533 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn$Rm",
1534 "$Rn.addr = $wb", []> {
1535 let Inst{4} = Rn{4};
1538 def VST4LNd8_UPD : VST4LNWB<0b0011, {?,?,?,?}, "8"> {
1539 let Inst{7-5} = lane{2-0};
1541 def VST4LNd16_UPD : VST4LNWB<0b0111, {?,?,0,?}, "16"> {
1542 let Inst{7-6} = lane{1-0};
1544 def VST4LNd32_UPD : VST4LNWB<0b1011, {?,0,?,?}, "32"> {
1545 let Inst{7} = lane{0};
1546 let Inst{5} = Rn{5};
1549 def VST4LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
1550 def VST4LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
1551 def VST4LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
1553 def VST4LNq16_UPD : VST4LNWB<0b0111, {?,?,1,?}, "16"> {
1554 let Inst{7-6} = lane{1-0};
1556 def VST4LNq32_UPD : VST4LNWB<0b1011, {?,1,?,?}, "32"> {
1557 let Inst{7} = lane{0};
1558 let Inst{5} = Rn{5};
1561 def VST4LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
1562 def VST4LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
1564 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
1567 //===----------------------------------------------------------------------===//
1568 // NEON pattern fragments
1569 //===----------------------------------------------------------------------===//
1571 // Extract D sub-registers of Q registers.
1572 def DSubReg_i8_reg : SDNodeXForm<imm, [{
1573 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1574 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/8, MVT::i32);
1576 def DSubReg_i16_reg : SDNodeXForm<imm, [{
1577 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1578 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/4, MVT::i32);
1580 def DSubReg_i32_reg : SDNodeXForm<imm, [{
1581 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1582 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/2, MVT::i32);
1584 def DSubReg_f64_reg : SDNodeXForm<imm, [{
1585 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1586 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue(), MVT::i32);
1589 // Extract S sub-registers of Q/D registers.
1590 def SSubReg_f32_reg : SDNodeXForm<imm, [{
1591 assert(ARM::ssub_3 == ARM::ssub_0+3 && "Unexpected subreg numbering");
1592 return CurDAG->getTargetConstant(ARM::ssub_0 + N->getZExtValue(), MVT::i32);
1595 // Translate lane numbers from Q registers to D subregs.
1596 def SubReg_i8_lane : SDNodeXForm<imm, [{
1597 return CurDAG->getTargetConstant(N->getZExtValue() & 7, MVT::i32);
1599 def SubReg_i16_lane : SDNodeXForm<imm, [{
1600 return CurDAG->getTargetConstant(N->getZExtValue() & 3, MVT::i32);
1602 def SubReg_i32_lane : SDNodeXForm<imm, [{
1603 return CurDAG->getTargetConstant(N->getZExtValue() & 1, MVT::i32);
1606 //===----------------------------------------------------------------------===//
1607 // Instruction Classes
1608 //===----------------------------------------------------------------------===//
1610 // Basic 2-register operations: single-, double- and quad-register.
1611 class N2VS<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1612 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
1613 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
1614 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
1615 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src),
1616 IIC_VUNAD, OpcodeStr, Dt, "$dst, $src", "", []>;
1617 class N2VD<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1618 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
1619 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
1620 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
1621 (ins DPR:$src), IIC_VUNAD, OpcodeStr, Dt,"$dst, $src", "",
1622 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src))))]>;
1623 class N2VQ<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1624 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
1625 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
1626 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
1627 (ins QPR:$src), IIC_VUNAQ, OpcodeStr, Dt,"$dst, $src", "",
1628 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src))))]>;
1630 // Basic 2-register intrinsics, both double- and quad-register.
1631 class N2VDInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1632 bits<2> op17_16, bits<5> op11_7, bit op4,
1633 InstrItinClass itin, string OpcodeStr, string Dt,
1634 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1635 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
1636 (ins DPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
1637 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
1638 class N2VQInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1639 bits<2> op17_16, bits<5> op11_7, bit op4,
1640 InstrItinClass itin, string OpcodeStr, string Dt,
1641 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1642 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
1643 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
1644 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
1646 // Narrow 2-register operations.
1647 class N2VN<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1648 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1649 InstrItinClass itin, string OpcodeStr, string Dt,
1650 ValueType TyD, ValueType TyQ, SDNode OpNode>
1651 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$dst),
1652 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
1653 [(set DPR:$dst, (TyD (OpNode (TyQ QPR:$src))))]>;
1655 // Narrow 2-register intrinsics.
1656 class N2VNInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1657 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1658 InstrItinClass itin, string OpcodeStr, string Dt,
1659 ValueType TyD, ValueType TyQ, Intrinsic IntOp>
1660 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$dst),
1661 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
1662 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src))))]>;
1664 // Long 2-register operations (currently only used for VMOVL).
1665 class N2VL<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1666 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1667 InstrItinClass itin, string OpcodeStr, string Dt,
1668 ValueType TyQ, ValueType TyD, SDNode OpNode>
1669 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$dst),
1670 (ins DPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
1671 [(set QPR:$dst, (TyQ (OpNode (TyD DPR:$src))))]>;
1673 // 2-register shuffles (VTRN/VZIP/VUZP), both double- and quad-register.
1674 class N2VDShuffle<bits<2> op19_18, bits<5> op11_7, string OpcodeStr, string Dt>
1675 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 0, 0, (outs DPR:$dst1, DPR:$dst2),
1676 (ins DPR:$src1, DPR:$src2), IIC_VPERMD,
1677 OpcodeStr, Dt, "$dst1, $dst2",
1678 "$src1 = $dst1, $src2 = $dst2", []>;
1679 class N2VQShuffle<bits<2> op19_18, bits<5> op11_7,
1680 InstrItinClass itin, string OpcodeStr, string Dt>
1681 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 1, 0, (outs QPR:$dst1, QPR:$dst2),
1682 (ins QPR:$src1, QPR:$src2), itin, OpcodeStr, Dt, "$dst1, $dst2",
1683 "$src1 = $dst1, $src2 = $dst2", []>;
1685 // Basic 3-register operations: single-, double- and quad-register.
1686 class N3VS<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1687 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1688 SDNode OpNode, bit Commutable>
1689 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1690 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src1, DPR_VFP2:$src2), N3RegFrm,
1691 IIC_VBIND, OpcodeStr, Dt, "$dst, $src1, $src2", "", []> {
1692 let isCommutable = Commutable;
1695 class N3VD<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1696 InstrItinClass itin, string OpcodeStr, string Dt,
1697 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
1698 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1699 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1700 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
1701 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
1702 let isCommutable = Commutable;
1704 // Same as N3VD but no data type.
1705 class N3VDX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1706 InstrItinClass itin, string OpcodeStr,
1707 ValueType ResTy, ValueType OpTy,
1708 SDNode OpNode, bit Commutable>
1709 : N3VX<op24, op23, op21_20, op11_8, 0, op4,
1710 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1711 OpcodeStr, "$Vd, $Vn, $Vm", "",
1712 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>{
1713 let isCommutable = Commutable;
1716 class N3VDSL<bits<2> op21_20, bits<4> op11_8,
1717 InstrItinClass itin, string OpcodeStr, string Dt,
1718 ValueType Ty, SDNode ShOp>
1719 : N3V<0, 1, op21_20, op11_8, 1, 0,
1720 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1721 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1722 [(set (Ty DPR:$dst),
1723 (Ty (ShOp (Ty DPR:$src1),
1724 (Ty (NEONvduplane (Ty DPR_VFP2:$src2),imm:$lane)))))]> {
1725 let isCommutable = 0;
1727 class N3VDSL16<bits<2> op21_20, bits<4> op11_8,
1728 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1729 : N3V<0, 1, op21_20, op11_8, 1, 0,
1730 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1731 NVMulSLFrm, IIC_VMULi16D, OpcodeStr, Dt,"$dst, $src1, $src2[$lane]","",
1732 [(set (Ty DPR:$dst),
1733 (Ty (ShOp (Ty DPR:$src1),
1734 (Ty (NEONvduplane (Ty DPR_8:$src2), imm:$lane)))))]> {
1735 let isCommutable = 0;
1738 class N3VQ<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1739 InstrItinClass itin, string OpcodeStr, string Dt,
1740 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
1741 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1742 (outs QPR:$Qd), (ins QPR:$Qn, QPR:$Qm), N3RegFrm, itin,
1743 OpcodeStr, Dt, "$Qd, $Qn, $Qm", "",
1744 [(set QPR:$Qd, (ResTy (OpNode (OpTy QPR:$Qn), (OpTy QPR:$Qm))))]> {
1745 let isCommutable = Commutable;
1747 class N3VQX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1748 InstrItinClass itin, string OpcodeStr,
1749 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
1750 : N3VX<op24, op23, op21_20, op11_8, 1, op4,
1751 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), N3RegFrm, itin,
1752 OpcodeStr, "$dst, $src1, $src2", "",
1753 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src1), (OpTy QPR:$src2))))]>{
1754 let isCommutable = Commutable;
1756 class N3VQSL<bits<2> op21_20, bits<4> op11_8,
1757 InstrItinClass itin, string OpcodeStr, string Dt,
1758 ValueType ResTy, ValueType OpTy, SDNode ShOp>
1759 : N3V<1, 1, op21_20, op11_8, 1, 0,
1760 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1761 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1762 [(set (ResTy QPR:$dst),
1763 (ResTy (ShOp (ResTy QPR:$src1),
1764 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1766 let isCommutable = 0;
1768 class N3VQSL16<bits<2> op21_20, bits<4> op11_8, string OpcodeStr, string Dt,
1769 ValueType ResTy, ValueType OpTy, SDNode ShOp>
1770 : N3V<1, 1, op21_20, op11_8, 1, 0,
1771 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1772 NVMulSLFrm, IIC_VMULi16Q, OpcodeStr, Dt,"$dst, $src1, $src2[$lane]","",
1773 [(set (ResTy QPR:$dst),
1774 (ResTy (ShOp (ResTy QPR:$src1),
1775 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
1777 let isCommutable = 0;
1780 // Basic 3-register intrinsics, both double- and quad-register.
1781 class N3VDInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1782 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1783 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
1784 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1785 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), f, itin,
1786 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
1787 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
1788 let isCommutable = Commutable;
1790 class N3VDIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1791 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
1792 : N3V<0, 1, op21_20, op11_8, 1, 0,
1793 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1794 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1795 [(set (Ty DPR:$dst),
1796 (Ty (IntOp (Ty DPR:$src1),
1797 (Ty (NEONvduplane (Ty DPR_VFP2:$src2),
1799 let isCommutable = 0;
1801 class N3VDIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1802 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
1803 : N3V<0, 1, op21_20, op11_8, 1, 0,
1804 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1805 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1806 [(set (Ty DPR:$dst),
1807 (Ty (IntOp (Ty DPR:$src1),
1808 (Ty (NEONvduplane (Ty DPR_8:$src2), imm:$lane)))))]> {
1809 let isCommutable = 0;
1811 class N3VDIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1812 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1813 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1814 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1815 (outs DPR:$Vd), (ins DPR:$Vm, DPR:$Vn), f, itin,
1816 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
1817 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (OpTy DPR:$Vn))))]> {
1818 let isCommutable = 0;
1821 class N3VQInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1822 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1823 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
1824 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1825 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), f, itin,
1826 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
1827 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
1828 let isCommutable = Commutable;
1830 class N3VQIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1831 string OpcodeStr, string Dt,
1832 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1833 : N3V<1, 1, op21_20, op11_8, 1, 0,
1834 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1835 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1836 [(set (ResTy QPR:$dst),
1837 (ResTy (IntOp (ResTy QPR:$src1),
1838 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1840 let isCommutable = 0;
1842 class N3VQIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1843 string OpcodeStr, string Dt,
1844 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1845 : N3V<1, 1, op21_20, op11_8, 1, 0,
1846 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1847 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1848 [(set (ResTy QPR:$dst),
1849 (ResTy (IntOp (ResTy QPR:$src1),
1850 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
1852 let isCommutable = 0;
1854 class N3VQIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1855 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1856 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1857 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1858 (outs QPR:$Vd), (ins QPR:$Vm, QPR:$Vn), f, itin,
1859 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
1860 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (OpTy QPR:$Vn))))]> {
1861 let isCommutable = 0;
1864 // Multiply-Add/Sub operations: single-, double- and quad-register.
1865 class N3VSMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1866 InstrItinClass itin, string OpcodeStr, string Dt,
1867 ValueType Ty, SDNode MulOp, SDNode OpNode>
1868 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1869 (outs DPR_VFP2:$dst),
1870 (ins DPR_VFP2:$src1, DPR_VFP2:$src2, DPR_VFP2:$src3), N3RegFrm, itin,
1871 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst", []>;
1873 class N3VDMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1874 InstrItinClass itin, string OpcodeStr, string Dt,
1875 ValueType Ty, SDNode MulOp, SDNode OpNode>
1876 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1877 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1878 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1879 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
1880 (Ty (MulOp DPR:$Vn, DPR:$Vm)))))]>;
1882 class N3VDMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1883 string OpcodeStr, string Dt,
1884 ValueType Ty, SDNode MulOp, SDNode ShOp>
1885 : N3V<0, 1, op21_20, op11_8, 1, 0,
1887 (ins DPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1889 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1890 [(set (Ty DPR:$dst),
1891 (Ty (ShOp (Ty DPR:$src1),
1892 (Ty (MulOp DPR:$src2,
1893 (Ty (NEONvduplane (Ty DPR_VFP2:$src3),
1895 class N3VDMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1896 string OpcodeStr, string Dt,
1897 ValueType Ty, SDNode MulOp, SDNode ShOp>
1898 : N3V<0, 1, op21_20, op11_8, 1, 0,
1900 (ins DPR:$src1, DPR:$Vn, DPR_8:$Vm, nohash_imm:$lane),
1902 OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "$src1 = $Vd",
1904 (Ty (ShOp (Ty DPR:$src1),
1906 (Ty (NEONvduplane (Ty DPR_8:$Vm),
1909 class N3VQMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1910 InstrItinClass itin, string OpcodeStr, string Dt, ValueType Ty,
1911 SDNode MulOp, SDNode OpNode>
1912 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1913 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
1914 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1915 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
1916 (Ty (MulOp QPR:$Vn, QPR:$Vm)))))]>;
1917 class N3VQMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1918 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1919 SDNode MulOp, SDNode ShOp>
1920 : N3V<1, 1, op21_20, op11_8, 1, 0,
1922 (ins QPR:$src1, QPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1924 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1925 [(set (ResTy QPR:$dst),
1926 (ResTy (ShOp (ResTy QPR:$src1),
1927 (ResTy (MulOp QPR:$src2,
1928 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src3),
1930 class N3VQMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1931 string OpcodeStr, string Dt,
1932 ValueType ResTy, ValueType OpTy,
1933 SDNode MulOp, SDNode ShOp>
1934 : N3V<1, 1, op21_20, op11_8, 1, 0,
1936 (ins QPR:$src1, QPR:$src2, DPR_8:$src3, nohash_imm:$lane),
1938 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1939 [(set (ResTy QPR:$dst),
1940 (ResTy (ShOp (ResTy QPR:$src1),
1941 (ResTy (MulOp QPR:$src2,
1942 (ResTy (NEONvduplane (OpTy DPR_8:$src3),
1945 // Neon Intrinsic-Op instructions (VABA): double- and quad-register.
1946 class N3VDIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1947 InstrItinClass itin, string OpcodeStr, string Dt,
1948 ValueType Ty, Intrinsic IntOp, SDNode OpNode>
1949 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1950 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1951 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1952 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
1953 (Ty (IntOp (Ty DPR:$Vn), (Ty DPR:$Vm))))))]>;
1954 class N3VQIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1955 InstrItinClass itin, string OpcodeStr, string Dt,
1956 ValueType Ty, Intrinsic IntOp, SDNode OpNode>
1957 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1958 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
1959 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1960 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
1961 (Ty (IntOp (Ty QPR:$Vn), (Ty QPR:$Vm))))))]>;
1963 // Neon 3-argument intrinsics, both double- and quad-register.
1964 // The destination register is also used as the first source operand register.
1965 class N3VDInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1966 InstrItinClass itin, string OpcodeStr, string Dt,
1967 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1968 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1969 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3), N3RegFrm, itin,
1970 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1971 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src1),
1972 (OpTy DPR:$src2), (OpTy DPR:$src3))))]>;
1973 class N3VQInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1974 InstrItinClass itin, string OpcodeStr, string Dt,
1975 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1976 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1977 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3), N3RegFrm, itin,
1978 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1979 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src1),
1980 (OpTy QPR:$src2), (OpTy QPR:$src3))))]>;
1982 // Long Multiply-Add/Sub operations.
1983 class N3VLMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1984 InstrItinClass itin, string OpcodeStr, string Dt,
1985 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
1986 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1987 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1988 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1989 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
1990 (TyQ (MulOp (TyD DPR:$Vn),
1991 (TyD DPR:$Vm)))))]>;
1992 class N3VLMulOpSL<bit op24, bits<2> op21_20, bits<4> op11_8,
1993 InstrItinClass itin, string OpcodeStr, string Dt,
1994 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
1995 : N3V<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$dst),
1996 (ins QPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1998 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
2000 (OpNode (TyQ QPR:$src1),
2001 (TyQ (MulOp (TyD DPR:$src2),
2002 (TyD (NEONvduplane (TyD DPR_VFP2:$src3),
2004 class N3VLMulOpSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2005 InstrItinClass itin, string OpcodeStr, string Dt,
2006 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2007 : N3V<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$dst),
2008 (ins QPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane),
2010 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
2012 (OpNode (TyQ QPR:$src1),
2013 (TyQ (MulOp (TyD DPR:$src2),
2014 (TyD (NEONvduplane (TyD DPR_8:$src3),
2017 // Long Intrinsic-Op vector operations with explicit extend (VABAL).
2018 class N3VLIntExtOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2019 InstrItinClass itin, string OpcodeStr, string Dt,
2020 ValueType TyQ, ValueType TyD, Intrinsic IntOp, SDNode ExtOp,
2022 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2023 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2024 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2025 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
2026 (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
2027 (TyD DPR:$Vm)))))))]>;
2029 // Neon Long 3-argument intrinsic. The destination register is
2030 // a quad-register and is also used as the first source operand register.
2031 class N3VLInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2032 InstrItinClass itin, string OpcodeStr, string Dt,
2033 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
2034 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2035 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2036 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2038 (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$Vn), (TyD DPR:$Vm))))]>;
2039 class N3VLInt3SL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2040 string OpcodeStr, string Dt,
2041 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2042 : N3V<op24, 1, op21_20, op11_8, 1, 0,
2044 (ins QPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
2046 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
2047 [(set (ResTy QPR:$dst),
2048 (ResTy (IntOp (ResTy QPR:$src1),
2050 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src3),
2052 class N3VLInt3SL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2053 InstrItinClass itin, string OpcodeStr, string Dt,
2054 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2055 : N3V<op24, 1, op21_20, op11_8, 1, 0,
2057 (ins QPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane),
2059 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
2060 [(set (ResTy QPR:$dst),
2061 (ResTy (IntOp (ResTy QPR:$src1),
2063 (OpTy (NEONvduplane (OpTy DPR_8:$src3),
2066 // Narrowing 3-register intrinsics.
2067 class N3VNInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2068 string OpcodeStr, string Dt, ValueType TyD, ValueType TyQ,
2069 Intrinsic IntOp, bit Commutable>
2070 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2071 (outs DPR:$dst), (ins QPR:$src1, QPR:$src2), N3RegFrm, IIC_VBINi4D,
2072 OpcodeStr, Dt, "$dst, $src1, $src2", "",
2073 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src1), (TyQ QPR:$src2))))]> {
2074 let isCommutable = Commutable;
2077 // Long 3-register operations.
2078 class N3VL<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, SDNode OpNode, bit Commutable>
2081 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2082 (outs QPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
2083 OpcodeStr, Dt, "$dst, $src1, $src2", "",
2084 [(set QPR:$dst, (TyQ (OpNode (TyD DPR:$src1), (TyD DPR:$src2))))]> {
2085 let isCommutable = Commutable;
2087 class N3VLSL<bit op24, bits<2> op21_20, bits<4> op11_8,
2088 InstrItinClass itin, string OpcodeStr, string Dt,
2089 ValueType TyQ, ValueType TyD, SDNode OpNode>
2090 : N3V<op24, 1, op21_20, op11_8, 1, 0,
2091 (outs QPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
2092 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
2094 (TyQ (OpNode (TyD DPR:$src1),
2095 (TyD (NEONvduplane (TyD DPR_VFP2:$src2),imm:$lane)))))]>;
2096 class N3VLSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2097 InstrItinClass itin, string OpcodeStr, string Dt,
2098 ValueType TyQ, ValueType TyD, SDNode OpNode>
2099 : N3V<op24, 1, op21_20, op11_8, 1, 0,
2100 (outs QPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
2101 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
2103 (TyQ (OpNode (TyD DPR:$src1),
2104 (TyD (NEONvduplane (TyD DPR_8:$src2), imm:$lane)))))]>;
2106 // Long 3-register operations with explicitly extended operands.
2107 class N3VLExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2108 InstrItinClass itin, string OpcodeStr, string Dt,
2109 ValueType TyQ, ValueType TyD, SDNode OpNode, SDNode ExtOp,
2111 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2112 (outs QPR:$Qd), (ins DPR:$Dn, DPR:$Dm), N3RegFrm, itin,
2113 OpcodeStr, Dt, "$Qd, $Dn, $Dm", "",
2114 [(set QPR:$Qd, (OpNode (TyQ (ExtOp (TyD DPR:$Dn))),
2115 (TyQ (ExtOp (TyD DPR:$Dm)))))]> {
2116 let isCommutable = Commutable;
2119 // Long 3-register intrinsics with explicit extend (VABDL).
2120 class N3VLIntExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2121 InstrItinClass itin, string OpcodeStr, string Dt,
2122 ValueType TyQ, ValueType TyD, Intrinsic IntOp, SDNode ExtOp,
2124 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2125 (outs QPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
2126 OpcodeStr, Dt, "$dst, $src1, $src2", "",
2127 [(set QPR:$dst, (TyQ (ExtOp (TyD (IntOp (TyD DPR:$src1),
2128 (TyD DPR:$src2))))))]> {
2129 let isCommutable = Commutable;
2132 // Long 3-register intrinsics.
2133 class N3VLInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2134 InstrItinClass itin, string OpcodeStr, string Dt,
2135 ValueType TyQ, ValueType TyD, Intrinsic IntOp, bit Commutable>
2136 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2137 (outs QPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
2138 OpcodeStr, Dt, "$dst, $src1, $src2", "",
2139 [(set QPR:$dst, (TyQ (IntOp (TyD DPR:$src1), (TyD DPR:$src2))))]> {
2140 let isCommutable = Commutable;
2142 class N3VLIntSL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2143 string OpcodeStr, string Dt,
2144 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2145 : N3V<op24, 1, op21_20, op11_8, 1, 0,
2146 (outs QPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
2147 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
2148 [(set (ResTy QPR:$dst),
2149 (ResTy (IntOp (OpTy DPR:$src1),
2150 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src2),
2152 class N3VLIntSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2153 InstrItinClass itin, string OpcodeStr, string Dt,
2154 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2155 : N3V<op24, 1, op21_20, op11_8, 1, 0,
2156 (outs QPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
2157 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
2158 [(set (ResTy QPR:$dst),
2159 (ResTy (IntOp (OpTy DPR:$src1),
2160 (OpTy (NEONvduplane (OpTy DPR_8:$src2),
2163 // Wide 3-register operations.
2164 class N3VW<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2165 string OpcodeStr, string Dt, ValueType TyQ, ValueType TyD,
2166 SDNode OpNode, SDNode ExtOp, bit Commutable>
2167 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2168 (outs QPR:$Qd), (ins QPR:$Qn, DPR:$Dm), N3RegFrm, IIC_VSUBiD,
2169 OpcodeStr, Dt, "$Qd, $Qn, $Dm", "",
2170 [(set QPR:$Qd, (OpNode (TyQ QPR:$Qn),
2171 (TyQ (ExtOp (TyD DPR:$Dm)))))]> {
2172 let isCommutable = Commutable;
2175 // Pairwise long 2-register intrinsics, both double- and quad-register.
2176 class N2VDPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2177 bits<2> op17_16, bits<5> op11_7, bit op4,
2178 string OpcodeStr, string Dt,
2179 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2180 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
2181 (ins DPR:$src), IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
2182 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
2183 class N2VQPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2184 bits<2> op17_16, bits<5> op11_7, bit op4,
2185 string OpcodeStr, string Dt,
2186 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2187 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
2188 (ins QPR:$src), IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
2189 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
2191 // Pairwise long 2-register accumulate intrinsics,
2192 // both double- and quad-register.
2193 // The destination register is also used as the first source operand register.
2194 class N2VDPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2195 bits<2> op17_16, bits<5> op11_7, bit op4,
2196 string OpcodeStr, string Dt,
2197 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2198 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
2199 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vm), IIC_VPALiD,
2200 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
2201 [(set DPR:$Vd, (ResTy (IntOp (ResTy DPR:$src1), (OpTy DPR:$Vm))))]>;
2202 class N2VQPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2203 bits<2> op17_16, bits<5> op11_7, bit op4,
2204 string OpcodeStr, string Dt,
2205 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2206 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4,
2207 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vm), IIC_VPALiQ,
2208 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
2209 [(set QPR:$Vd, (ResTy (IntOp (ResTy QPR:$src1), (OpTy QPR:$Vm))))]>;
2211 // Shift by immediate,
2212 // both double- and quad-register.
2213 class N2VDSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2214 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2215 ValueType Ty, SDNode OpNode>
2216 : N2VImm<op24, op23, op11_8, op7, 0, op4,
2217 (outs DPR:$dst), (ins DPR:$src, i32imm:$SIMM), f, itin,
2218 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
2219 [(set DPR:$dst, (Ty (OpNode (Ty DPR:$src), (i32 imm:$SIMM))))]>;
2220 class N2VQSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2221 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2222 ValueType Ty, SDNode OpNode>
2223 : N2VImm<op24, op23, op11_8, op7, 1, op4,
2224 (outs QPR:$dst), (ins QPR:$src, i32imm:$SIMM), f, itin,
2225 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
2226 [(set QPR:$dst, (Ty (OpNode (Ty QPR:$src), (i32 imm:$SIMM))))]>;
2228 // Long shift by immediate.
2229 class N2VLSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
2230 string OpcodeStr, string Dt,
2231 ValueType ResTy, ValueType OpTy, SDNode OpNode>
2232 : N2VImm<op24, op23, op11_8, op7, op6, op4,
2233 (outs QPR:$dst), (ins DPR:$src, i32imm:$SIMM), N2RegVShLFrm,
2234 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src, $SIMM", "",
2235 [(set QPR:$dst, (ResTy (OpNode (OpTy DPR:$src),
2236 (i32 imm:$SIMM))))]>;
2238 // Narrow shift by immediate.
2239 class N2VNSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
2240 InstrItinClass itin, string OpcodeStr, string Dt,
2241 ValueType ResTy, ValueType OpTy, SDNode OpNode>
2242 : N2VImm<op24, op23, op11_8, op7, op6, op4,
2243 (outs DPR:$dst), (ins QPR:$src, i32imm:$SIMM), N2RegVShRFrm, itin,
2244 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
2245 [(set DPR:$dst, (ResTy (OpNode (OpTy QPR:$src),
2246 (i32 imm:$SIMM))))]>;
2248 // Shift right by immediate and accumulate,
2249 // both double- and quad-register.
2250 class N2VDShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2251 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
2252 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
2253 (ins DPR:$src1, DPR:$Vm, i32imm:$SIMM), N2RegVShRFrm, IIC_VPALiD,
2254 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2255 [(set DPR:$Vd, (Ty (add DPR:$src1,
2256 (Ty (ShOp DPR:$Vm, (i32 imm:$SIMM))))))]>;
2257 class N2VQShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2258 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
2259 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
2260 (ins QPR:$src1, QPR:$Vm, i32imm:$SIMM), N2RegVShRFrm, IIC_VPALiD,
2261 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2262 [(set QPR:$Vd, (Ty (add QPR:$src1,
2263 (Ty (ShOp QPR:$Vm, (i32 imm:$SIMM))))))]>;
2265 // Shift by immediate and insert,
2266 // both double- and quad-register.
2267 class N2VDShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2268 Format f, string OpcodeStr, string Dt, ValueType Ty,SDNode ShOp>
2269 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
2270 (ins DPR:$src1, DPR:$Vm, i32imm:$SIMM), f, IIC_VSHLiD,
2271 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2272 [(set DPR:$Vd, (Ty (ShOp DPR:$src1, DPR:$Vm, (i32 imm:$SIMM))))]>;
2273 class N2VQShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2274 Format f, string OpcodeStr, string Dt, ValueType Ty,SDNode ShOp>
2275 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
2276 (ins QPR:$src1, QPR:$Vm, i32imm:$SIMM), f, IIC_VSHLiQ,
2277 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2278 [(set QPR:$Vd, (Ty (ShOp QPR:$src1, QPR:$Vm, (i32 imm:$SIMM))))]>;
2280 // Convert, with fractional bits immediate,
2281 // both double- and quad-register.
2282 class N2VCvtD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2283 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2285 : N2VImm<op24, op23, op11_8, op7, 0, op4,
2286 (outs DPR:$Vd), (ins DPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
2287 IIC_VUNAD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2288 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (i32 imm:$SIMM))))]>;
2289 class N2VCvtQ<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2290 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2292 : N2VImm<op24, op23, op11_8, op7, 1, op4,
2293 (outs QPR:$Vd), (ins QPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
2294 IIC_VUNAQ, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2295 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (i32 imm:$SIMM))))]>;
2297 //===----------------------------------------------------------------------===//
2299 //===----------------------------------------------------------------------===//
2301 // Abbreviations used in multiclass suffixes:
2302 // Q = quarter int (8 bit) elements
2303 // H = half int (16 bit) elements
2304 // S = single int (32 bit) elements
2305 // D = double int (64 bit) elements
2307 // Neon 2-register vector operations -- for disassembly only.
2309 // First with only element sizes of 8, 16 and 32 bits:
2310 multiclass N2V_QHS_cmp<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2311 bits<5> op11_7, bit op4, string opc, string Dt,
2312 string asm, SDNode OpNode> {
2313 // 64-bit vector types.
2314 def v8i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 0, op4,
2315 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
2316 opc, !strconcat(Dt, "8"), asm, "",
2317 [(set DPR:$dst, (v8i8 (OpNode (v8i8 DPR:$src))))]>;
2318 def v4i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 0, op4,
2319 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
2320 opc, !strconcat(Dt, "16"), asm, "",
2321 [(set DPR:$dst, (v4i16 (OpNode (v4i16 DPR:$src))))]>;
2322 def v2i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
2323 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
2324 opc, !strconcat(Dt, "32"), asm, "",
2325 [(set DPR:$dst, (v2i32 (OpNode (v2i32 DPR:$src))))]>;
2326 def v2f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
2327 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
2328 opc, "f32", asm, "",
2329 [(set DPR:$dst, (v2f32 (OpNode (v2f32 DPR:$src))))]> {
2330 let Inst{10} = 1; // overwrite F = 1
2333 // 128-bit vector types.
2334 def v16i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 1, op4,
2335 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
2336 opc, !strconcat(Dt, "8"), asm, "",
2337 [(set QPR:$dst, (v16i8 (OpNode (v16i8 QPR:$src))))]>;
2338 def v8i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 1, op4,
2339 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
2340 opc, !strconcat(Dt, "16"), asm, "",
2341 [(set QPR:$dst, (v8i16 (OpNode (v8i16 QPR:$src))))]>;
2342 def v4i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
2343 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
2344 opc, !strconcat(Dt, "32"), asm, "",
2345 [(set QPR:$dst, (v4i32 (OpNode (v4i32 QPR:$src))))]>;
2346 def v4f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
2347 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
2348 opc, "f32", asm, "",
2349 [(set QPR:$dst, (v4f32 (OpNode (v4f32 QPR:$src))))]> {
2350 let Inst{10} = 1; // overwrite F = 1
2354 // Neon 3-register vector operations.
2356 // First with only element sizes of 8, 16 and 32 bits:
2357 multiclass N3V_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2358 InstrItinClass itinD16, InstrItinClass itinD32,
2359 InstrItinClass itinQ16, InstrItinClass itinQ32,
2360 string OpcodeStr, string Dt,
2361 SDNode OpNode, bit Commutable = 0> {
2362 // 64-bit vector types.
2363 def v8i8 : N3VD<op24, op23, 0b00, op11_8, op4, itinD16,
2364 OpcodeStr, !strconcat(Dt, "8"),
2365 v8i8, v8i8, OpNode, Commutable>;
2366 def v4i16 : N3VD<op24, op23, 0b01, op11_8, op4, itinD16,
2367 OpcodeStr, !strconcat(Dt, "16"),
2368 v4i16, v4i16, OpNode, Commutable>;
2369 def v2i32 : N3VD<op24, op23, 0b10, op11_8, op4, itinD32,
2370 OpcodeStr, !strconcat(Dt, "32"),
2371 v2i32, v2i32, OpNode, Commutable>;
2373 // 128-bit vector types.
2374 def v16i8 : N3VQ<op24, op23, 0b00, op11_8, op4, itinQ16,
2375 OpcodeStr, !strconcat(Dt, "8"),
2376 v16i8, v16i8, OpNode, Commutable>;
2377 def v8i16 : N3VQ<op24, op23, 0b01, op11_8, op4, itinQ16,
2378 OpcodeStr, !strconcat(Dt, "16"),
2379 v8i16, v8i16, OpNode, Commutable>;
2380 def v4i32 : N3VQ<op24, op23, 0b10, op11_8, op4, itinQ32,
2381 OpcodeStr, !strconcat(Dt, "32"),
2382 v4i32, v4i32, OpNode, Commutable>;
2385 multiclass N3VSL_HS<bits<4> op11_8, string OpcodeStr, string Dt, SDNode ShOp> {
2386 def v4i16 : N3VDSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
2388 def v2i32 : N3VDSL<0b10, op11_8, IIC_VMULi32D, OpcodeStr, !strconcat(Dt,"32"),
2390 def v8i16 : N3VQSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
2391 v8i16, v4i16, ShOp>;
2392 def v4i32 : N3VQSL<0b10, op11_8, IIC_VMULi32Q, OpcodeStr, !strconcat(Dt,"32"),
2393 v4i32, v2i32, ShOp>;
2396 // ....then also with element size 64 bits:
2397 multiclass N3V_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2398 InstrItinClass itinD, InstrItinClass itinQ,
2399 string OpcodeStr, string Dt,
2400 SDNode OpNode, bit Commutable = 0>
2401 : N3V_QHS<op24, op23, op11_8, op4, itinD, itinD, itinQ, itinQ,
2402 OpcodeStr, Dt, OpNode, Commutable> {
2403 def v1i64 : N3VD<op24, op23, 0b11, op11_8, op4, itinD,
2404 OpcodeStr, !strconcat(Dt, "64"),
2405 v1i64, v1i64, OpNode, Commutable>;
2406 def v2i64 : N3VQ<op24, op23, 0b11, op11_8, op4, itinQ,
2407 OpcodeStr, !strconcat(Dt, "64"),
2408 v2i64, v2i64, OpNode, Commutable>;
2412 // Neon Narrowing 2-register vector operations,
2413 // source operand element sizes of 16, 32 and 64 bits:
2414 multiclass N2VN_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2415 bits<5> op11_7, bit op6, bit op4,
2416 InstrItinClass itin, string OpcodeStr, string Dt,
2418 def v8i8 : N2VN<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
2419 itin, OpcodeStr, !strconcat(Dt, "16"),
2420 v8i8, v8i16, OpNode>;
2421 def v4i16 : N2VN<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
2422 itin, OpcodeStr, !strconcat(Dt, "32"),
2423 v4i16, v4i32, OpNode>;
2424 def v2i32 : N2VN<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
2425 itin, OpcodeStr, !strconcat(Dt, "64"),
2426 v2i32, v2i64, OpNode>;
2429 // Neon Narrowing 2-register vector intrinsics,
2430 // source operand element sizes of 16, 32 and 64 bits:
2431 multiclass N2VNInt_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2432 bits<5> op11_7, bit op6, bit op4,
2433 InstrItinClass itin, string OpcodeStr, string Dt,
2435 def v8i8 : N2VNInt<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
2436 itin, OpcodeStr, !strconcat(Dt, "16"),
2437 v8i8, v8i16, IntOp>;
2438 def v4i16 : N2VNInt<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
2439 itin, OpcodeStr, !strconcat(Dt, "32"),
2440 v4i16, v4i32, IntOp>;
2441 def v2i32 : N2VNInt<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
2442 itin, OpcodeStr, !strconcat(Dt, "64"),
2443 v2i32, v2i64, IntOp>;
2447 // Neon Lengthening 2-register vector intrinsic (currently specific to VMOVL).
2448 // source operand element sizes of 16, 32 and 64 bits:
2449 multiclass N2VL_QHS<bits<2> op24_23, bits<5> op11_7, bit op6, bit op4,
2450 string OpcodeStr, string Dt, SDNode OpNode> {
2451 def v8i16 : N2VL<op24_23, 0b00, 0b10, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2452 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode>;
2453 def v4i32 : N2VL<op24_23, 0b01, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2454 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
2455 def v2i64 : N2VL<op24_23, 0b10, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2456 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
2460 // Neon 3-register vector intrinsics.
2462 // First with only element sizes of 16 and 32 bits:
2463 multiclass N3VInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2464 InstrItinClass itinD16, InstrItinClass itinD32,
2465 InstrItinClass itinQ16, InstrItinClass itinQ32,
2466 string OpcodeStr, string Dt,
2467 Intrinsic IntOp, bit Commutable = 0> {
2468 // 64-bit vector types.
2469 def v4i16 : N3VDInt<op24, op23, 0b01, op11_8, op4, f, itinD16,
2470 OpcodeStr, !strconcat(Dt, "16"),
2471 v4i16, v4i16, IntOp, Commutable>;
2472 def v2i32 : N3VDInt<op24, op23, 0b10, op11_8, op4, f, itinD32,
2473 OpcodeStr, !strconcat(Dt, "32"),
2474 v2i32, v2i32, IntOp, Commutable>;
2476 // 128-bit vector types.
2477 def v8i16 : N3VQInt<op24, op23, 0b01, op11_8, op4, f, itinQ16,
2478 OpcodeStr, !strconcat(Dt, "16"),
2479 v8i16, v8i16, IntOp, Commutable>;
2480 def v4i32 : N3VQInt<op24, op23, 0b10, op11_8, op4, f, itinQ32,
2481 OpcodeStr, !strconcat(Dt, "32"),
2482 v4i32, v4i32, IntOp, Commutable>;
2484 multiclass N3VInt_HSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2485 InstrItinClass itinD16, InstrItinClass itinD32,
2486 InstrItinClass itinQ16, InstrItinClass itinQ32,
2487 string OpcodeStr, string Dt,
2489 // 64-bit vector types.
2490 def v4i16 : N3VDIntSh<op24, op23, 0b01, op11_8, op4, f, itinD16,
2491 OpcodeStr, !strconcat(Dt, "16"),
2492 v4i16, v4i16, IntOp>;
2493 def v2i32 : N3VDIntSh<op24, op23, 0b10, op11_8, op4, f, itinD32,
2494 OpcodeStr, !strconcat(Dt, "32"),
2495 v2i32, v2i32, IntOp>;
2497 // 128-bit vector types.
2498 def v8i16 : N3VQIntSh<op24, op23, 0b01, op11_8, op4, f, itinQ16,
2499 OpcodeStr, !strconcat(Dt, "16"),
2500 v8i16, v8i16, IntOp>;
2501 def v4i32 : N3VQIntSh<op24, op23, 0b10, op11_8, op4, f, itinQ32,
2502 OpcodeStr, !strconcat(Dt, "32"),
2503 v4i32, v4i32, IntOp>;
2506 multiclass N3VIntSL_HS<bits<4> op11_8,
2507 InstrItinClass itinD16, InstrItinClass itinD32,
2508 InstrItinClass itinQ16, InstrItinClass itinQ32,
2509 string OpcodeStr, string Dt, Intrinsic IntOp> {
2510 def v4i16 : N3VDIntSL16<0b01, op11_8, itinD16,
2511 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp>;
2512 def v2i32 : N3VDIntSL<0b10, op11_8, itinD32,
2513 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp>;
2514 def v8i16 : N3VQIntSL16<0b01, op11_8, itinQ16,
2515 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16, IntOp>;
2516 def v4i32 : N3VQIntSL<0b10, op11_8, itinQ32,
2517 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32, IntOp>;
2520 // ....then also with element size of 8 bits:
2521 multiclass N3VInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2522 InstrItinClass itinD16, InstrItinClass itinD32,
2523 InstrItinClass itinQ16, InstrItinClass itinQ32,
2524 string OpcodeStr, string Dt,
2525 Intrinsic IntOp, bit Commutable = 0>
2526 : N3VInt_HS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2527 OpcodeStr, Dt, IntOp, Commutable> {
2528 def v8i8 : N3VDInt<op24, op23, 0b00, op11_8, op4, f, itinD16,
2529 OpcodeStr, !strconcat(Dt, "8"),
2530 v8i8, v8i8, IntOp, Commutable>;
2531 def v16i8 : N3VQInt<op24, op23, 0b00, op11_8, op4, f, itinQ16,
2532 OpcodeStr, !strconcat(Dt, "8"),
2533 v16i8, v16i8, IntOp, Commutable>;
2535 multiclass N3VInt_QHSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2536 InstrItinClass itinD16, InstrItinClass itinD32,
2537 InstrItinClass itinQ16, InstrItinClass itinQ32,
2538 string OpcodeStr, string Dt,
2540 : N3VInt_HSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2541 OpcodeStr, Dt, IntOp> {
2542 def v8i8 : N3VDIntSh<op24, op23, 0b00, op11_8, op4, f, itinD16,
2543 OpcodeStr, !strconcat(Dt, "8"),
2545 def v16i8 : N3VQIntSh<op24, op23, 0b00, op11_8, op4, f, itinQ16,
2546 OpcodeStr, !strconcat(Dt, "8"),
2547 v16i8, v16i8, IntOp>;
2551 // ....then also with element size of 64 bits:
2552 multiclass N3VInt_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2553 InstrItinClass itinD16, InstrItinClass itinD32,
2554 InstrItinClass itinQ16, InstrItinClass itinQ32,
2555 string OpcodeStr, string Dt,
2556 Intrinsic IntOp, bit Commutable = 0>
2557 : N3VInt_QHS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2558 OpcodeStr, Dt, IntOp, Commutable> {
2559 def v1i64 : N3VDInt<op24, op23, 0b11, op11_8, op4, f, itinD32,
2560 OpcodeStr, !strconcat(Dt, "64"),
2561 v1i64, v1i64, IntOp, Commutable>;
2562 def v2i64 : N3VQInt<op24, op23, 0b11, op11_8, op4, f, itinQ32,
2563 OpcodeStr, !strconcat(Dt, "64"),
2564 v2i64, v2i64, IntOp, Commutable>;
2566 multiclass N3VInt_QHSDSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2567 InstrItinClass itinD16, InstrItinClass itinD32,
2568 InstrItinClass itinQ16, InstrItinClass itinQ32,
2569 string OpcodeStr, string Dt,
2571 : N3VInt_QHSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2572 OpcodeStr, Dt, IntOp> {
2573 def v1i64 : N3VDIntSh<op24, op23, 0b11, op11_8, op4, f, itinD32,
2574 OpcodeStr, !strconcat(Dt, "64"),
2575 v1i64, v1i64, IntOp>;
2576 def v2i64 : N3VQIntSh<op24, op23, 0b11, op11_8, op4, f, itinQ32,
2577 OpcodeStr, !strconcat(Dt, "64"),
2578 v2i64, v2i64, IntOp>;
2581 // Neon Narrowing 3-register vector intrinsics,
2582 // source operand element sizes of 16, 32 and 64 bits:
2583 multiclass N3VNInt_HSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2584 string OpcodeStr, string Dt,
2585 Intrinsic IntOp, bit Commutable = 0> {
2586 def v8i8 : N3VNInt<op24, op23, 0b00, op11_8, op4,
2587 OpcodeStr, !strconcat(Dt, "16"),
2588 v8i8, v8i16, IntOp, Commutable>;
2589 def v4i16 : N3VNInt<op24, op23, 0b01, op11_8, op4,
2590 OpcodeStr, !strconcat(Dt, "32"),
2591 v4i16, v4i32, IntOp, Commutable>;
2592 def v2i32 : N3VNInt<op24, op23, 0b10, op11_8, op4,
2593 OpcodeStr, !strconcat(Dt, "64"),
2594 v2i32, v2i64, IntOp, Commutable>;
2598 // Neon Long 3-register vector operations.
2600 multiclass N3VL_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2601 InstrItinClass itin16, InstrItinClass itin32,
2602 string OpcodeStr, string Dt,
2603 SDNode OpNode, bit Commutable = 0> {
2604 def v8i16 : N3VL<op24, op23, 0b00, op11_8, op4, itin16,
2605 OpcodeStr, !strconcat(Dt, "8"),
2606 v8i16, v8i8, OpNode, Commutable>;
2607 def v4i32 : N3VL<op24, op23, 0b01, op11_8, op4, itin16,
2608 OpcodeStr, !strconcat(Dt, "16"),
2609 v4i32, v4i16, OpNode, Commutable>;
2610 def v2i64 : N3VL<op24, op23, 0b10, op11_8, op4, itin32,
2611 OpcodeStr, !strconcat(Dt, "32"),
2612 v2i64, v2i32, OpNode, Commutable>;
2615 multiclass N3VLSL_HS<bit op24, bits<4> op11_8,
2616 InstrItinClass itin, string OpcodeStr, string Dt,
2618 def v4i16 : N3VLSL16<op24, 0b01, op11_8, itin, OpcodeStr,
2619 !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
2620 def v2i32 : N3VLSL<op24, 0b10, op11_8, itin, OpcodeStr,
2621 !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
2624 multiclass N3VLExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2625 InstrItinClass itin16, InstrItinClass itin32,
2626 string OpcodeStr, string Dt,
2627 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
2628 def v8i16 : N3VLExt<op24, op23, 0b00, op11_8, op4, itin16,
2629 OpcodeStr, !strconcat(Dt, "8"),
2630 v8i16, v8i8, OpNode, ExtOp, Commutable>;
2631 def v4i32 : N3VLExt<op24, op23, 0b01, op11_8, op4, itin16,
2632 OpcodeStr, !strconcat(Dt, "16"),
2633 v4i32, v4i16, OpNode, ExtOp, Commutable>;
2634 def v2i64 : N3VLExt<op24, op23, 0b10, op11_8, op4, itin32,
2635 OpcodeStr, !strconcat(Dt, "32"),
2636 v2i64, v2i32, OpNode, ExtOp, Commutable>;
2639 // Neon Long 3-register vector intrinsics.
2641 // First with only element sizes of 16 and 32 bits:
2642 multiclass N3VLInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
2643 InstrItinClass itin16, InstrItinClass itin32,
2644 string OpcodeStr, string Dt,
2645 Intrinsic IntOp, bit Commutable = 0> {
2646 def v4i32 : N3VLInt<op24, op23, 0b01, op11_8, op4, itin16,
2647 OpcodeStr, !strconcat(Dt, "16"),
2648 v4i32, v4i16, IntOp, Commutable>;
2649 def v2i64 : N3VLInt<op24, op23, 0b10, op11_8, op4, itin32,
2650 OpcodeStr, !strconcat(Dt, "32"),
2651 v2i64, v2i32, IntOp, Commutable>;
2654 multiclass N3VLIntSL_HS<bit op24, bits<4> op11_8,
2655 InstrItinClass itin, string OpcodeStr, string Dt,
2657 def v4i16 : N3VLIntSL16<op24, 0b01, op11_8, itin,
2658 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
2659 def v2i32 : N3VLIntSL<op24, 0b10, op11_8, itin,
2660 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
2663 // ....then also with element size of 8 bits:
2664 multiclass N3VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2665 InstrItinClass itin16, InstrItinClass itin32,
2666 string OpcodeStr, string Dt,
2667 Intrinsic IntOp, bit Commutable = 0>
2668 : N3VLInt_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt,
2669 IntOp, Commutable> {
2670 def v8i16 : N3VLInt<op24, op23, 0b00, op11_8, op4, itin16,
2671 OpcodeStr, !strconcat(Dt, "8"),
2672 v8i16, v8i8, IntOp, Commutable>;
2675 // ....with explicit extend (VABDL).
2676 multiclass N3VLIntExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2677 InstrItinClass itin, string OpcodeStr, string Dt,
2678 Intrinsic IntOp, SDNode ExtOp, bit Commutable = 0> {
2679 def v8i16 : N3VLIntExt<op24, op23, 0b00, op11_8, op4, itin,
2680 OpcodeStr, !strconcat(Dt, "8"),
2681 v8i16, v8i8, IntOp, ExtOp, Commutable>;
2682 def v4i32 : N3VLIntExt<op24, op23, 0b01, op11_8, op4, itin,
2683 OpcodeStr, !strconcat(Dt, "16"),
2684 v4i32, v4i16, IntOp, ExtOp, Commutable>;
2685 def v2i64 : N3VLIntExt<op24, op23, 0b10, op11_8, op4, itin,
2686 OpcodeStr, !strconcat(Dt, "32"),
2687 v2i64, v2i32, IntOp, ExtOp, Commutable>;
2691 // Neon Wide 3-register vector intrinsics,
2692 // source operand element sizes of 8, 16 and 32 bits:
2693 multiclass N3VW_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2694 string OpcodeStr, string Dt,
2695 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
2696 def v8i16 : N3VW<op24, op23, 0b00, op11_8, op4,
2697 OpcodeStr, !strconcat(Dt, "8"),
2698 v8i16, v8i8, OpNode, ExtOp, Commutable>;
2699 def v4i32 : N3VW<op24, op23, 0b01, op11_8, op4,
2700 OpcodeStr, !strconcat(Dt, "16"),
2701 v4i32, v4i16, OpNode, ExtOp, Commutable>;
2702 def v2i64 : N3VW<op24, op23, 0b10, op11_8, op4,
2703 OpcodeStr, !strconcat(Dt, "32"),
2704 v2i64, v2i32, OpNode, ExtOp, Commutable>;
2708 // Neon Multiply-Op vector operations,
2709 // element sizes of 8, 16 and 32 bits:
2710 multiclass N3VMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2711 InstrItinClass itinD16, InstrItinClass itinD32,
2712 InstrItinClass itinQ16, InstrItinClass itinQ32,
2713 string OpcodeStr, string Dt, SDNode OpNode> {
2714 // 64-bit vector types.
2715 def v8i8 : N3VDMulOp<op24, op23, 0b00, op11_8, op4, itinD16,
2716 OpcodeStr, !strconcat(Dt, "8"), v8i8, mul, OpNode>;
2717 def v4i16 : N3VDMulOp<op24, op23, 0b01, op11_8, op4, itinD16,
2718 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, OpNode>;
2719 def v2i32 : N3VDMulOp<op24, op23, 0b10, op11_8, op4, itinD32,
2720 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, OpNode>;
2722 // 128-bit vector types.
2723 def v16i8 : N3VQMulOp<op24, op23, 0b00, op11_8, op4, itinQ16,
2724 OpcodeStr, !strconcat(Dt, "8"), v16i8, mul, OpNode>;
2725 def v8i16 : N3VQMulOp<op24, op23, 0b01, op11_8, op4, itinQ16,
2726 OpcodeStr, !strconcat(Dt, "16"), v8i16, mul, OpNode>;
2727 def v4i32 : N3VQMulOp<op24, op23, 0b10, op11_8, op4, itinQ32,
2728 OpcodeStr, !strconcat(Dt, "32"), v4i32, mul, OpNode>;
2731 multiclass N3VMulOpSL_HS<bits<4> op11_8,
2732 InstrItinClass itinD16, InstrItinClass itinD32,
2733 InstrItinClass itinQ16, InstrItinClass itinQ32,
2734 string OpcodeStr, string Dt, SDNode ShOp> {
2735 def v4i16 : N3VDMulOpSL16<0b01, op11_8, itinD16,
2736 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, ShOp>;
2737 def v2i32 : N3VDMulOpSL<0b10, op11_8, itinD32,
2738 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, ShOp>;
2739 def v8i16 : N3VQMulOpSL16<0b01, op11_8, itinQ16,
2740 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16,
2742 def v4i32 : N3VQMulOpSL<0b10, op11_8, itinQ32,
2743 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32,
2747 // Neon Intrinsic-Op vector operations,
2748 // element sizes of 8, 16 and 32 bits:
2749 multiclass N3VIntOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2750 InstrItinClass itinD, InstrItinClass itinQ,
2751 string OpcodeStr, string Dt, Intrinsic IntOp,
2753 // 64-bit vector types.
2754 def v8i8 : N3VDIntOp<op24, op23, 0b00, op11_8, op4, itinD,
2755 OpcodeStr, !strconcat(Dt, "8"), v8i8, IntOp, OpNode>;
2756 def v4i16 : N3VDIntOp<op24, op23, 0b01, op11_8, op4, itinD,
2757 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp, OpNode>;
2758 def v2i32 : N3VDIntOp<op24, op23, 0b10, op11_8, op4, itinD,
2759 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp, OpNode>;
2761 // 128-bit vector types.
2762 def v16i8 : N3VQIntOp<op24, op23, 0b00, op11_8, op4, itinQ,
2763 OpcodeStr, !strconcat(Dt, "8"), v16i8, IntOp, OpNode>;
2764 def v8i16 : N3VQIntOp<op24, op23, 0b01, op11_8, op4, itinQ,
2765 OpcodeStr, !strconcat(Dt, "16"), v8i16, IntOp, OpNode>;
2766 def v4i32 : N3VQIntOp<op24, op23, 0b10, op11_8, op4, itinQ,
2767 OpcodeStr, !strconcat(Dt, "32"), v4i32, IntOp, OpNode>;
2770 // Neon 3-argument intrinsics,
2771 // element sizes of 8, 16 and 32 bits:
2772 multiclass N3VInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2773 InstrItinClass itinD, InstrItinClass itinQ,
2774 string OpcodeStr, string Dt, Intrinsic IntOp> {
2775 // 64-bit vector types.
2776 def v8i8 : N3VDInt3<op24, op23, 0b00, op11_8, op4, itinD,
2777 OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
2778 def v4i16 : N3VDInt3<op24, op23, 0b01, op11_8, op4, itinD,
2779 OpcodeStr, !strconcat(Dt, "16"), v4i16, v4i16, IntOp>;
2780 def v2i32 : N3VDInt3<op24, op23, 0b10, op11_8, op4, itinD,
2781 OpcodeStr, !strconcat(Dt, "32"), v2i32, v2i32, IntOp>;
2783 // 128-bit vector types.
2784 def v16i8 : N3VQInt3<op24, op23, 0b00, op11_8, op4, itinQ,
2785 OpcodeStr, !strconcat(Dt, "8"), v16i8, v16i8, IntOp>;
2786 def v8i16 : N3VQInt3<op24, op23, 0b01, op11_8, op4, itinQ,
2787 OpcodeStr, !strconcat(Dt, "16"), v8i16, v8i16, IntOp>;
2788 def v4i32 : N3VQInt3<op24, op23, 0b10, op11_8, op4, itinQ,
2789 OpcodeStr, !strconcat(Dt, "32"), v4i32, v4i32, IntOp>;
2793 // Neon Long Multiply-Op vector operations,
2794 // element sizes of 8, 16 and 32 bits:
2795 multiclass N3VLMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2796 InstrItinClass itin16, InstrItinClass itin32,
2797 string OpcodeStr, string Dt, SDNode MulOp,
2799 def v8i16 : N3VLMulOp<op24, op23, 0b00, op11_8, op4, itin16, OpcodeStr,
2800 !strconcat(Dt, "8"), v8i16, v8i8, MulOp, OpNode>;
2801 def v4i32 : N3VLMulOp<op24, op23, 0b01, op11_8, op4, itin16, OpcodeStr,
2802 !strconcat(Dt, "16"), v4i32, v4i16, MulOp, OpNode>;
2803 def v2i64 : N3VLMulOp<op24, op23, 0b10, op11_8, op4, itin32, OpcodeStr,
2804 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
2807 multiclass N3VLMulOpSL_HS<bit op24, bits<4> op11_8, string OpcodeStr,
2808 string Dt, SDNode MulOp, SDNode OpNode> {
2809 def v4i16 : N3VLMulOpSL16<op24, 0b01, op11_8, IIC_VMACi16D, OpcodeStr,
2810 !strconcat(Dt,"16"), v4i32, v4i16, MulOp, OpNode>;
2811 def v2i32 : N3VLMulOpSL<op24, 0b10, op11_8, IIC_VMACi32D, OpcodeStr,
2812 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
2816 // Neon Long 3-argument intrinsics.
2818 // First with only element sizes of 16 and 32 bits:
2819 multiclass N3VLInt3_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
2820 InstrItinClass itin16, InstrItinClass itin32,
2821 string OpcodeStr, string Dt, Intrinsic IntOp> {
2822 def v4i32 : N3VLInt3<op24, op23, 0b01, op11_8, op4, itin16,
2823 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
2824 def v2i64 : N3VLInt3<op24, op23, 0b10, op11_8, op4, itin32,
2825 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
2828 multiclass N3VLInt3SL_HS<bit op24, bits<4> op11_8,
2829 string OpcodeStr, string Dt, Intrinsic IntOp> {
2830 def v4i16 : N3VLInt3SL16<op24, 0b01, op11_8, IIC_VMACi16D,
2831 OpcodeStr, !strconcat(Dt,"16"), v4i32, v4i16, IntOp>;
2832 def v2i32 : N3VLInt3SL<op24, 0b10, op11_8, IIC_VMACi32D,
2833 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
2836 // ....then also with element size of 8 bits:
2837 multiclass N3VLInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2838 InstrItinClass itin16, InstrItinClass itin32,
2839 string OpcodeStr, string Dt, Intrinsic IntOp>
2840 : N3VLInt3_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt, IntOp> {
2841 def v8i16 : N3VLInt3<op24, op23, 0b00, op11_8, op4, itin16,
2842 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
2845 // ....with explicit extend (VABAL).
2846 multiclass N3VLIntExtOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2847 InstrItinClass itin, string OpcodeStr, string Dt,
2848 Intrinsic IntOp, SDNode ExtOp, SDNode OpNode> {
2849 def v8i16 : N3VLIntExtOp<op24, op23, 0b00, op11_8, op4, itin,
2850 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8,
2851 IntOp, ExtOp, OpNode>;
2852 def v4i32 : N3VLIntExtOp<op24, op23, 0b01, op11_8, op4, itin,
2853 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16,
2854 IntOp, ExtOp, OpNode>;
2855 def v2i64 : N3VLIntExtOp<op24, op23, 0b10, op11_8, op4, itin,
2856 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32,
2857 IntOp, ExtOp, OpNode>;
2861 // Neon 2-register vector intrinsics,
2862 // element sizes of 8, 16 and 32 bits:
2863 multiclass N2VInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2864 bits<5> op11_7, bit op4,
2865 InstrItinClass itinD, InstrItinClass itinQ,
2866 string OpcodeStr, string Dt, Intrinsic IntOp> {
2867 // 64-bit vector types.
2868 def v8i8 : N2VDInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2869 itinD, OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
2870 def v4i16 : N2VDInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2871 itinD, OpcodeStr, !strconcat(Dt, "16"),v4i16,v4i16,IntOp>;
2872 def v2i32 : N2VDInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2873 itinD, OpcodeStr, !strconcat(Dt, "32"),v2i32,v2i32,IntOp>;
2875 // 128-bit vector types.
2876 def v16i8 : N2VQInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2877 itinQ, OpcodeStr, !strconcat(Dt, "8"), v16i8,v16i8,IntOp>;
2878 def v8i16 : N2VQInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2879 itinQ, OpcodeStr, !strconcat(Dt, "16"),v8i16,v8i16,IntOp>;
2880 def v4i32 : N2VQInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2881 itinQ, OpcodeStr, !strconcat(Dt, "32"),v4i32,v4i32,IntOp>;
2885 // Neon Pairwise long 2-register intrinsics,
2886 // element sizes of 8, 16 and 32 bits:
2887 multiclass N2VPLInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2888 bits<5> op11_7, bit op4,
2889 string OpcodeStr, string Dt, Intrinsic IntOp> {
2890 // 64-bit vector types.
2891 def v8i8 : N2VDPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2892 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
2893 def v4i16 : N2VDPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2894 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
2895 def v2i32 : N2VDPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2896 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
2898 // 128-bit vector types.
2899 def v16i8 : N2VQPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2900 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
2901 def v8i16 : N2VQPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2902 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
2903 def v4i32 : N2VQPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2904 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
2908 // Neon Pairwise long 2-register accumulate intrinsics,
2909 // element sizes of 8, 16 and 32 bits:
2910 multiclass N2VPLInt2_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2911 bits<5> op11_7, bit op4,
2912 string OpcodeStr, string Dt, Intrinsic IntOp> {
2913 // 64-bit vector types.
2914 def v8i8 : N2VDPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2915 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
2916 def v4i16 : N2VDPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2917 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
2918 def v2i32 : N2VDPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2919 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
2921 // 128-bit vector types.
2922 def v16i8 : N2VQPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2923 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
2924 def v8i16 : N2VQPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2925 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
2926 def v4i32 : N2VQPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2927 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
2931 // Neon 2-register vector shift by immediate,
2932 // with f of either N2RegVShLFrm or N2RegVShRFrm
2933 // element sizes of 8, 16, 32 and 64 bits:
2934 multiclass N2VSh_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2935 InstrItinClass itin, string OpcodeStr, string Dt,
2936 SDNode OpNode, Format f> {
2937 // 64-bit vector types.
2938 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
2939 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
2940 let Inst{21-19} = 0b001; // imm6 = 001xxx
2942 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
2943 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
2944 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2946 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
2947 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
2948 let Inst{21} = 0b1; // imm6 = 1xxxxx
2950 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, f, itin,
2951 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
2954 // 128-bit vector types.
2955 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
2956 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
2957 let Inst{21-19} = 0b001; // imm6 = 001xxx
2959 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
2960 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
2961 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2963 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
2964 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
2965 let Inst{21} = 0b1; // imm6 = 1xxxxx
2967 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, f, itin,
2968 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
2972 // Neon Shift-Accumulate vector operations,
2973 // element sizes of 8, 16, 32 and 64 bits:
2974 multiclass N2VShAdd_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2975 string OpcodeStr, string Dt, SDNode ShOp> {
2976 // 64-bit vector types.
2977 def v8i8 : N2VDShAdd<op24, op23, op11_8, 0, op4,
2978 OpcodeStr, !strconcat(Dt, "8"), v8i8, ShOp> {
2979 let Inst{21-19} = 0b001; // imm6 = 001xxx
2981 def v4i16 : N2VDShAdd<op24, op23, op11_8, 0, op4,
2982 OpcodeStr, !strconcat(Dt, "16"), v4i16, ShOp> {
2983 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2985 def v2i32 : N2VDShAdd<op24, op23, op11_8, 0, op4,
2986 OpcodeStr, !strconcat(Dt, "32"), v2i32, ShOp> {
2987 let Inst{21} = 0b1; // imm6 = 1xxxxx
2989 def v1i64 : N2VDShAdd<op24, op23, op11_8, 1, op4,
2990 OpcodeStr, !strconcat(Dt, "64"), v1i64, ShOp>;
2993 // 128-bit vector types.
2994 def v16i8 : N2VQShAdd<op24, op23, op11_8, 0, op4,
2995 OpcodeStr, !strconcat(Dt, "8"), v16i8, ShOp> {
2996 let Inst{21-19} = 0b001; // imm6 = 001xxx
2998 def v8i16 : N2VQShAdd<op24, op23, op11_8, 0, op4,
2999 OpcodeStr, !strconcat(Dt, "16"), v8i16, ShOp> {
3000 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3002 def v4i32 : N2VQShAdd<op24, op23, op11_8, 0, op4,
3003 OpcodeStr, !strconcat(Dt, "32"), v4i32, ShOp> {
3004 let Inst{21} = 0b1; // imm6 = 1xxxxx
3006 def v2i64 : N2VQShAdd<op24, op23, op11_8, 1, op4,
3007 OpcodeStr, !strconcat(Dt, "64"), v2i64, ShOp>;
3012 // Neon Shift-Insert vector operations,
3013 // with f of either N2RegVShLFrm or N2RegVShRFrm
3014 // element sizes of 8, 16, 32 and 64 bits:
3015 multiclass N2VShIns_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3016 string OpcodeStr, SDNode ShOp,
3018 // 64-bit vector types.
3019 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4,
3020 f, OpcodeStr, "8", v8i8, ShOp> {
3021 let Inst{21-19} = 0b001; // imm6 = 001xxx
3023 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4,
3024 f, OpcodeStr, "16", v4i16, ShOp> {
3025 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3027 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4,
3028 f, OpcodeStr, "32", v2i32, ShOp> {
3029 let Inst{21} = 0b1; // imm6 = 1xxxxx
3031 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4,
3032 f, OpcodeStr, "64", v1i64, ShOp>;
3035 // 128-bit vector types.
3036 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4,
3037 f, OpcodeStr, "8", v16i8, ShOp> {
3038 let Inst{21-19} = 0b001; // imm6 = 001xxx
3040 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4,
3041 f, OpcodeStr, "16", v8i16, ShOp> {
3042 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3044 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4,
3045 f, OpcodeStr, "32", v4i32, ShOp> {
3046 let Inst{21} = 0b1; // imm6 = 1xxxxx
3048 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4,
3049 f, OpcodeStr, "64", v2i64, ShOp>;
3053 // Neon Shift Long operations,
3054 // element sizes of 8, 16, 32 bits:
3055 multiclass N2VLSh_QHS<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
3056 bit op4, string OpcodeStr, string Dt, SDNode OpNode> {
3057 def v8i16 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3058 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode> {
3059 let Inst{21-19} = 0b001; // imm6 = 001xxx
3061 def v4i32 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3062 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode> {
3063 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3065 def v2i64 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3066 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode> {
3067 let Inst{21} = 0b1; // imm6 = 1xxxxx
3071 // Neon Shift Narrow operations,
3072 // element sizes of 16, 32, 64 bits:
3073 multiclass N2VNSh_HSD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
3074 bit op4, InstrItinClass itin, string OpcodeStr, string Dt,
3076 def v8i8 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3077 OpcodeStr, !strconcat(Dt, "16"), v8i8, v8i16, OpNode> {
3078 let Inst{21-19} = 0b001; // imm6 = 001xxx
3080 def v4i16 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3081 OpcodeStr, !strconcat(Dt, "32"), v4i16, v4i32, OpNode> {
3082 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3084 def v2i32 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3085 OpcodeStr, !strconcat(Dt, "64"), v2i32, v2i64, OpNode> {
3086 let Inst{21} = 0b1; // imm6 = 1xxxxx
3090 //===----------------------------------------------------------------------===//
3091 // Instruction Definitions.
3092 //===----------------------------------------------------------------------===//
3094 // Vector Add Operations.
3096 // VADD : Vector Add (integer and floating-point)
3097 defm VADD : N3V_QHSD<0, 0, 0b1000, 0, IIC_VBINiD, IIC_VBINiQ, "vadd", "i",
3099 def VADDfd : N3VD<0, 0, 0b00, 0b1101, 0, IIC_VBIND, "vadd", "f32",
3100 v2f32, v2f32, fadd, 1>;
3101 def VADDfq : N3VQ<0, 0, 0b00, 0b1101, 0, IIC_VBINQ, "vadd", "f32",
3102 v4f32, v4f32, fadd, 1>;
3103 // VADDL : Vector Add Long (Q = D + D)
3104 defm VADDLs : N3VLExt_QHS<0,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
3105 "vaddl", "s", add, sext, 1>;
3106 defm VADDLu : N3VLExt_QHS<1,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
3107 "vaddl", "u", add, zext, 1>;
3108 // VADDW : Vector Add Wide (Q = Q + D)
3109 defm VADDWs : N3VW_QHS<0,1,0b0001,0, "vaddw", "s", add, sext, 0>;
3110 defm VADDWu : N3VW_QHS<1,1,0b0001,0, "vaddw", "u", add, zext, 0>;
3111 // VHADD : Vector Halving Add
3112 defm VHADDs : N3VInt_QHS<0, 0, 0b0000, 0, N3RegFrm,
3113 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3114 "vhadd", "s", int_arm_neon_vhadds, 1>;
3115 defm VHADDu : N3VInt_QHS<1, 0, 0b0000, 0, N3RegFrm,
3116 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3117 "vhadd", "u", int_arm_neon_vhaddu, 1>;
3118 // VRHADD : Vector Rounding Halving Add
3119 defm VRHADDs : N3VInt_QHS<0, 0, 0b0001, 0, N3RegFrm,
3120 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3121 "vrhadd", "s", int_arm_neon_vrhadds, 1>;
3122 defm VRHADDu : N3VInt_QHS<1, 0, 0b0001, 0, N3RegFrm,
3123 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3124 "vrhadd", "u", int_arm_neon_vrhaddu, 1>;
3125 // VQADD : Vector Saturating Add
3126 defm VQADDs : N3VInt_QHSD<0, 0, 0b0000, 1, N3RegFrm,
3127 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3128 "vqadd", "s", int_arm_neon_vqadds, 1>;
3129 defm VQADDu : N3VInt_QHSD<1, 0, 0b0000, 1, N3RegFrm,
3130 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3131 "vqadd", "u", int_arm_neon_vqaddu, 1>;
3132 // VADDHN : Vector Add and Narrow Returning High Half (D = Q + Q)
3133 defm VADDHN : N3VNInt_HSD<0,1,0b0100,0, "vaddhn", "i",
3134 int_arm_neon_vaddhn, 1>;
3135 // VRADDHN : Vector Rounding Add and Narrow Returning High Half (D = Q + Q)
3136 defm VRADDHN : N3VNInt_HSD<1,1,0b0100,0, "vraddhn", "i",
3137 int_arm_neon_vraddhn, 1>;
3139 // Vector Multiply Operations.
3141 // VMUL : Vector Multiply (integer, polynomial and floating-point)
3142 defm VMUL : N3V_QHS<0, 0, 0b1001, 1, IIC_VMULi16D, IIC_VMULi32D,
3143 IIC_VMULi16Q, IIC_VMULi32Q, "vmul", "i", mul, 1>;
3144 def VMULpd : N3VDInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16D, "vmul",
3145 "p8", v8i8, v8i8, int_arm_neon_vmulp, 1>;
3146 def VMULpq : N3VQInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16Q, "vmul",
3147 "p8", v16i8, v16i8, int_arm_neon_vmulp, 1>;
3148 def VMULfd : N3VD<1, 0, 0b00, 0b1101, 1, IIC_VFMULD, "vmul", "f32",
3149 v2f32, v2f32, fmul, 1>;
3150 def VMULfq : N3VQ<1, 0, 0b00, 0b1101, 1, IIC_VFMULQ, "vmul", "f32",
3151 v4f32, v4f32, fmul, 1>;
3152 defm VMULsl : N3VSL_HS<0b1000, "vmul", "i", mul>;
3153 def VMULslfd : N3VDSL<0b10, 0b1001, IIC_VBIND, "vmul", "f32", v2f32, fmul>;
3154 def VMULslfq : N3VQSL<0b10, 0b1001, IIC_VBINQ, "vmul", "f32", v4f32,
3157 def : Pat<(v8i16 (mul (v8i16 QPR:$src1),
3158 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
3159 (v8i16 (VMULslv8i16 (v8i16 QPR:$src1),
3160 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3161 (DSubReg_i16_reg imm:$lane))),
3162 (SubReg_i16_lane imm:$lane)))>;
3163 def : Pat<(v4i32 (mul (v4i32 QPR:$src1),
3164 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
3165 (v4i32 (VMULslv4i32 (v4i32 QPR:$src1),
3166 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3167 (DSubReg_i32_reg imm:$lane))),
3168 (SubReg_i32_lane imm:$lane)))>;
3169 def : Pat<(v4f32 (fmul (v4f32 QPR:$src1),
3170 (v4f32 (NEONvduplane (v4f32 QPR:$src2), imm:$lane)))),
3171 (v4f32 (VMULslfq (v4f32 QPR:$src1),
3172 (v2f32 (EXTRACT_SUBREG QPR:$src2,
3173 (DSubReg_i32_reg imm:$lane))),
3174 (SubReg_i32_lane imm:$lane)))>;
3176 // VQDMULH : Vector Saturating Doubling Multiply Returning High Half
3177 defm VQDMULH : N3VInt_HS<0, 0, 0b1011, 0, N3RegFrm, IIC_VMULi16D, IIC_VMULi32D,
3178 IIC_VMULi16Q, IIC_VMULi32Q,
3179 "vqdmulh", "s", int_arm_neon_vqdmulh, 1>;
3180 defm VQDMULHsl: N3VIntSL_HS<0b1100, IIC_VMULi16D, IIC_VMULi32D,
3181 IIC_VMULi16Q, IIC_VMULi32Q,
3182 "vqdmulh", "s", int_arm_neon_vqdmulh>;
3183 def : Pat<(v8i16 (int_arm_neon_vqdmulh (v8i16 QPR:$src1),
3184 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
3186 (v8i16 (VQDMULHslv8i16 (v8i16 QPR:$src1),
3187 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3188 (DSubReg_i16_reg imm:$lane))),
3189 (SubReg_i16_lane imm:$lane)))>;
3190 def : Pat<(v4i32 (int_arm_neon_vqdmulh (v4i32 QPR:$src1),
3191 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
3193 (v4i32 (VQDMULHslv4i32 (v4i32 QPR:$src1),
3194 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3195 (DSubReg_i32_reg imm:$lane))),
3196 (SubReg_i32_lane imm:$lane)))>;
3198 // VQRDMULH : Vector Rounding Saturating Doubling Multiply Returning High Half
3199 defm VQRDMULH : N3VInt_HS<1, 0, 0b1011, 0, N3RegFrm,
3200 IIC_VMULi16D,IIC_VMULi32D,IIC_VMULi16Q,IIC_VMULi32Q,
3201 "vqrdmulh", "s", int_arm_neon_vqrdmulh, 1>;
3202 defm VQRDMULHsl : N3VIntSL_HS<0b1101, IIC_VMULi16D, IIC_VMULi32D,
3203 IIC_VMULi16Q, IIC_VMULi32Q,
3204 "vqrdmulh", "s", int_arm_neon_vqrdmulh>;
3205 def : Pat<(v8i16 (int_arm_neon_vqrdmulh (v8i16 QPR:$src1),
3206 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
3208 (v8i16 (VQRDMULHslv8i16 (v8i16 QPR:$src1),
3209 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3210 (DSubReg_i16_reg imm:$lane))),
3211 (SubReg_i16_lane imm:$lane)))>;
3212 def : Pat<(v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$src1),
3213 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
3215 (v4i32 (VQRDMULHslv4i32 (v4i32 QPR:$src1),
3216 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3217 (DSubReg_i32_reg imm:$lane))),
3218 (SubReg_i32_lane imm:$lane)))>;
3220 // VMULL : Vector Multiply Long (integer and polynomial) (Q = D * D)
3221 defm VMULLs : N3VL_QHS<0,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
3222 "vmull", "s", NEONvmulls, 1>;
3223 defm VMULLu : N3VL_QHS<1,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
3224 "vmull", "u", NEONvmullu, 1>;
3225 def VMULLp : N3VLInt<0, 1, 0b00, 0b1110, 0, IIC_VMULi16D, "vmull", "p8",
3226 v8i16, v8i8, int_arm_neon_vmullp, 1>;
3227 defm VMULLsls : N3VLSL_HS<0, 0b1010, IIC_VMULi16D, "vmull", "s", NEONvmulls>;
3228 defm VMULLslu : N3VLSL_HS<1, 0b1010, IIC_VMULi16D, "vmull", "u", NEONvmullu>;
3230 // VQDMULL : Vector Saturating Doubling Multiply Long (Q = D * D)
3231 defm VQDMULL : N3VLInt_HS<0,1,0b1101,0, IIC_VMULi16D, IIC_VMULi32D,
3232 "vqdmull", "s", int_arm_neon_vqdmull, 1>;
3233 defm VQDMULLsl: N3VLIntSL_HS<0, 0b1011, IIC_VMULi16D,
3234 "vqdmull", "s", int_arm_neon_vqdmull>;
3236 // Vector Multiply-Accumulate and Multiply-Subtract Operations.
3238 // VMLA : Vector Multiply Accumulate (integer and floating-point)
3239 defm VMLA : N3VMulOp_QHS<0, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3240 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
3241 def VMLAfd : N3VDMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACD, "vmla", "f32",
3243 def VMLAfq : N3VQMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACQ, "vmla", "f32",
3245 defm VMLAsl : N3VMulOpSL_HS<0b0000, IIC_VMACi16D, IIC_VMACi32D,
3246 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
3247 def VMLAslfd : N3VDMulOpSL<0b10, 0b0001, IIC_VMACD, "vmla", "f32",
3249 def VMLAslfq : N3VQMulOpSL<0b10, 0b0001, IIC_VMACQ, "vmla", "f32",
3250 v4f32, v2f32, fmul, fadd>;
3252 def : Pat<(v8i16 (add (v8i16 QPR:$src1),
3253 (mul (v8i16 QPR:$src2),
3254 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
3255 (v8i16 (VMLAslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
3256 (v4i16 (EXTRACT_SUBREG QPR:$src3,
3257 (DSubReg_i16_reg imm:$lane))),
3258 (SubReg_i16_lane imm:$lane)))>;
3260 def : Pat<(v4i32 (add (v4i32 QPR:$src1),
3261 (mul (v4i32 QPR:$src2),
3262 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
3263 (v4i32 (VMLAslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
3264 (v2i32 (EXTRACT_SUBREG QPR:$src3,
3265 (DSubReg_i32_reg imm:$lane))),
3266 (SubReg_i32_lane imm:$lane)))>;
3268 def : Pat<(v4f32 (fadd (v4f32 QPR:$src1),
3269 (fmul (v4f32 QPR:$src2),
3270 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
3271 (v4f32 (VMLAslfq (v4f32 QPR:$src1),
3273 (v2f32 (EXTRACT_SUBREG QPR:$src3,
3274 (DSubReg_i32_reg imm:$lane))),
3275 (SubReg_i32_lane imm:$lane)))>;
3277 // VMLAL : Vector Multiply Accumulate Long (Q += D * D)
3278 defm VMLALs : N3VLMulOp_QHS<0,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
3279 "vmlal", "s", NEONvmulls, add>;
3280 defm VMLALu : N3VLMulOp_QHS<1,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
3281 "vmlal", "u", NEONvmullu, add>;
3283 defm VMLALsls : N3VLMulOpSL_HS<0, 0b0010, "vmlal", "s", NEONvmulls, add>;
3284 defm VMLALslu : N3VLMulOpSL_HS<1, 0b0010, "vmlal", "u", NEONvmullu, add>;
3286 // VQDMLAL : Vector Saturating Doubling Multiply Accumulate Long (Q += D * D)
3287 defm VQDMLAL : N3VLInt3_HS<0, 1, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3288 "vqdmlal", "s", int_arm_neon_vqdmlal>;
3289 defm VQDMLALsl: N3VLInt3SL_HS<0, 0b0011, "vqdmlal", "s", int_arm_neon_vqdmlal>;
3291 // VMLS : Vector Multiply Subtract (integer and floating-point)
3292 defm VMLS : N3VMulOp_QHS<1, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3293 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
3294 def VMLSfd : N3VDMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACD, "vmls", "f32",
3296 def VMLSfq : N3VQMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACQ, "vmls", "f32",
3298 defm VMLSsl : N3VMulOpSL_HS<0b0100, IIC_VMACi16D, IIC_VMACi32D,
3299 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
3300 def VMLSslfd : N3VDMulOpSL<0b10, 0b0101, IIC_VMACD, "vmls", "f32",
3302 def VMLSslfq : N3VQMulOpSL<0b10, 0b0101, IIC_VMACQ, "vmls", "f32",
3303 v4f32, v2f32, fmul, fsub>;
3305 def : Pat<(v8i16 (sub (v8i16 QPR:$src1),
3306 (mul (v8i16 QPR:$src2),
3307 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
3308 (v8i16 (VMLSslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
3309 (v4i16 (EXTRACT_SUBREG QPR:$src3,
3310 (DSubReg_i16_reg imm:$lane))),
3311 (SubReg_i16_lane imm:$lane)))>;
3313 def : Pat<(v4i32 (sub (v4i32 QPR:$src1),
3314 (mul (v4i32 QPR:$src2),
3315 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
3316 (v4i32 (VMLSslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
3317 (v2i32 (EXTRACT_SUBREG QPR:$src3,
3318 (DSubReg_i32_reg imm:$lane))),
3319 (SubReg_i32_lane imm:$lane)))>;
3321 def : Pat<(v4f32 (fsub (v4f32 QPR:$src1),
3322 (fmul (v4f32 QPR:$src2),
3323 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
3324 (v4f32 (VMLSslfq (v4f32 QPR:$src1), (v4f32 QPR:$src2),
3325 (v2f32 (EXTRACT_SUBREG QPR:$src3,
3326 (DSubReg_i32_reg imm:$lane))),
3327 (SubReg_i32_lane imm:$lane)))>;
3329 // VMLSL : Vector Multiply Subtract Long (Q -= D * D)
3330 defm VMLSLs : N3VLMulOp_QHS<0,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
3331 "vmlsl", "s", NEONvmulls, sub>;
3332 defm VMLSLu : N3VLMulOp_QHS<1,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
3333 "vmlsl", "u", NEONvmullu, sub>;
3335 defm VMLSLsls : N3VLMulOpSL_HS<0, 0b0110, "vmlsl", "s", NEONvmulls, sub>;
3336 defm VMLSLslu : N3VLMulOpSL_HS<1, 0b0110, "vmlsl", "u", NEONvmullu, sub>;
3338 // VQDMLSL : Vector Saturating Doubling Multiply Subtract Long (Q -= D * D)
3339 defm VQDMLSL : N3VLInt3_HS<0, 1, 0b1011, 0, IIC_VMACi16D, IIC_VMACi32D,
3340 "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
3341 defm VQDMLSLsl: N3VLInt3SL_HS<0, 0b111, "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
3343 // Vector Subtract Operations.
3345 // VSUB : Vector Subtract (integer and floating-point)
3346 defm VSUB : N3V_QHSD<1, 0, 0b1000, 0, IIC_VSUBiD, IIC_VSUBiQ,
3347 "vsub", "i", sub, 0>;
3348 def VSUBfd : N3VD<0, 0, 0b10, 0b1101, 0, IIC_VBIND, "vsub", "f32",
3349 v2f32, v2f32, fsub, 0>;
3350 def VSUBfq : N3VQ<0, 0, 0b10, 0b1101, 0, IIC_VBINQ, "vsub", "f32",
3351 v4f32, v4f32, fsub, 0>;
3352 // VSUBL : Vector Subtract Long (Q = D - D)
3353 defm VSUBLs : N3VLExt_QHS<0,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
3354 "vsubl", "s", sub, sext, 0>;
3355 defm VSUBLu : N3VLExt_QHS<1,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
3356 "vsubl", "u", sub, zext, 0>;
3357 // VSUBW : Vector Subtract Wide (Q = Q - D)
3358 defm VSUBWs : N3VW_QHS<0,1,0b0011,0, "vsubw", "s", sub, sext, 0>;
3359 defm VSUBWu : N3VW_QHS<1,1,0b0011,0, "vsubw", "u", sub, zext, 0>;
3360 // VHSUB : Vector Halving Subtract
3361 defm VHSUBs : N3VInt_QHS<0, 0, 0b0010, 0, N3RegFrm,
3362 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3363 "vhsub", "s", int_arm_neon_vhsubs, 0>;
3364 defm VHSUBu : N3VInt_QHS<1, 0, 0b0010, 0, N3RegFrm,
3365 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3366 "vhsub", "u", int_arm_neon_vhsubu, 0>;
3367 // VQSUB : Vector Saturing Subtract
3368 defm VQSUBs : N3VInt_QHSD<0, 0, 0b0010, 1, N3RegFrm,
3369 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3370 "vqsub", "s", int_arm_neon_vqsubs, 0>;
3371 defm VQSUBu : N3VInt_QHSD<1, 0, 0b0010, 1, N3RegFrm,
3372 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3373 "vqsub", "u", int_arm_neon_vqsubu, 0>;
3374 // VSUBHN : Vector Subtract and Narrow Returning High Half (D = Q - Q)
3375 defm VSUBHN : N3VNInt_HSD<0,1,0b0110,0, "vsubhn", "i",
3376 int_arm_neon_vsubhn, 0>;
3377 // VRSUBHN : Vector Rounding Subtract and Narrow Returning High Half (D=Q-Q)
3378 defm VRSUBHN : N3VNInt_HSD<1,1,0b0110,0, "vrsubhn", "i",
3379 int_arm_neon_vrsubhn, 0>;
3381 // Vector Comparisons.
3383 // VCEQ : Vector Compare Equal
3384 defm VCEQ : N3V_QHS<1, 0, 0b1000, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3385 IIC_VSUBi4Q, "vceq", "i", NEONvceq, 1>;
3386 def VCEQfd : N3VD<0,0,0b00,0b1110,0, IIC_VBIND, "vceq", "f32", v2i32, v2f32,
3388 def VCEQfq : N3VQ<0,0,0b00,0b1110,0, IIC_VBINQ, "vceq", "f32", v4i32, v4f32,
3391 defm VCEQz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00010, 0, "vceq", "i",
3392 "$dst, $src, #0", NEONvceqz>;
3394 // VCGE : Vector Compare Greater Than or Equal
3395 defm VCGEs : N3V_QHS<0, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3396 IIC_VSUBi4Q, "vcge", "s", NEONvcge, 0>;
3397 defm VCGEu : N3V_QHS<1, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3398 IIC_VSUBi4Q, "vcge", "u", NEONvcgeu, 0>;
3399 def VCGEfd : N3VD<1,0,0b00,0b1110,0, IIC_VBIND, "vcge", "f32", v2i32, v2f32,
3401 def VCGEfq : N3VQ<1,0,0b00,0b1110,0, IIC_VBINQ, "vcge", "f32", v4i32, v4f32,
3404 defm VCGEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00001, 0, "vcge", "s",
3405 "$dst, $src, #0", NEONvcgez>;
3406 defm VCLEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00011, 0, "vcle", "s",
3407 "$dst, $src, #0", NEONvclez>;
3409 // VCGT : Vector Compare Greater Than
3410 defm VCGTs : N3V_QHS<0, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3411 IIC_VSUBi4Q, "vcgt", "s", NEONvcgt, 0>;
3412 defm VCGTu : N3V_QHS<1, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3413 IIC_VSUBi4Q, "vcgt", "u", NEONvcgtu, 0>;
3414 def VCGTfd : N3VD<1,0,0b10,0b1110,0, IIC_VBIND, "vcgt", "f32", v2i32, v2f32,
3416 def VCGTfq : N3VQ<1,0,0b10,0b1110,0, IIC_VBINQ, "vcgt", "f32", v4i32, v4f32,
3419 defm VCGTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00000, 0, "vcgt", "s",
3420 "$dst, $src, #0", NEONvcgtz>;
3421 defm VCLTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00100, 0, "vclt", "s",
3422 "$dst, $src, #0", NEONvcltz>;
3424 // VACGE : Vector Absolute Compare Greater Than or Equal (aka VCAGE)
3425 def VACGEd : N3VDInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacge",
3426 "f32", v2i32, v2f32, int_arm_neon_vacged, 0>;
3427 def VACGEq : N3VQInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacge",
3428 "f32", v4i32, v4f32, int_arm_neon_vacgeq, 0>;
3429 // VACGT : Vector Absolute Compare Greater Than (aka VCAGT)
3430 def VACGTd : N3VDInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacgt",
3431 "f32", v2i32, v2f32, int_arm_neon_vacgtd, 0>;
3432 def VACGTq : N3VQInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacgt",
3433 "f32", v4i32, v4f32, int_arm_neon_vacgtq, 0>;
3434 // VTST : Vector Test Bits
3435 defm VTST : N3V_QHS<0, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
3436 IIC_VBINi4Q, "vtst", "", NEONvtst, 1>;
3438 // Vector Bitwise Operations.
3440 def vnotd : PatFrag<(ops node:$in),
3441 (xor node:$in, (bitconvert (v8i8 NEONimmAllOnesV)))>;
3442 def vnotq : PatFrag<(ops node:$in),
3443 (xor node:$in, (bitconvert (v16i8 NEONimmAllOnesV)))>;
3446 // VAND : Vector Bitwise AND
3447 def VANDd : N3VDX<0, 0, 0b00, 0b0001, 1, IIC_VBINiD, "vand",
3448 v2i32, v2i32, and, 1>;
3449 def VANDq : N3VQX<0, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "vand",
3450 v4i32, v4i32, and, 1>;
3452 // VEOR : Vector Bitwise Exclusive OR
3453 def VEORd : N3VDX<1, 0, 0b00, 0b0001, 1, IIC_VBINiD, "veor",
3454 v2i32, v2i32, xor, 1>;
3455 def VEORq : N3VQX<1, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "veor",
3456 v4i32, v4i32, xor, 1>;
3458 // VORR : Vector Bitwise OR
3459 def VORRd : N3VDX<0, 0, 0b10, 0b0001, 1, IIC_VBINiD, "vorr",
3460 v2i32, v2i32, or, 1>;
3461 def VORRq : N3VQX<0, 0, 0b10, 0b0001, 1, IIC_VBINiQ, "vorr",
3462 v4i32, v4i32, or, 1>;
3464 def VORRiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 0, 1,
3465 (outs DPR:$Vd), (ins nModImm:$SIMM, DPR:$src),
3467 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
3469 (v4i16 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
3470 let Inst{9} = SIMM{9};
3473 def VORRiv2i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 0, 0, 1,
3474 (outs DPR:$Vd), (ins nModImm:$SIMM, DPR:$src),
3476 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
3478 (v2i32 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
3479 let Inst{10-9} = SIMM{10-9};
3482 def VORRiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 0, 1,
3483 (outs QPR:$Vd), (ins nModImm:$SIMM, QPR:$src),
3485 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
3487 (v8i16 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
3488 let Inst{9} = SIMM{9};
3491 def VORRiv4i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 1, 0, 1,
3492 (outs QPR:$Vd), (ins nModImm:$SIMM, QPR:$src),
3494 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
3496 (v4i32 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
3497 let Inst{10-9} = SIMM{10-9};
3501 // VBIC : Vector Bitwise Bit Clear (AND NOT)
3502 def VBICd : N3VX<0, 0, 0b01, 0b0001, 0, 1, (outs DPR:$dst),
3503 (ins DPR:$src1, DPR:$src2), N3RegFrm, IIC_VBINiD,
3504 "vbic", "$dst, $src1, $src2", "",
3505 [(set DPR:$dst, (v2i32 (and DPR:$src1,
3506 (vnotd DPR:$src2))))]>;
3507 def VBICq : N3VX<0, 0, 0b01, 0b0001, 1, 1, (outs QPR:$dst),
3508 (ins QPR:$src1, QPR:$src2), N3RegFrm, IIC_VBINiQ,
3509 "vbic", "$dst, $src1, $src2", "",
3510 [(set QPR:$dst, (v4i32 (and QPR:$src1,
3511 (vnotq QPR:$src2))))]>;
3513 def VBICiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 1, 1,
3514 (outs DPR:$Vd), (ins nModImm:$SIMM, DPR:$src),
3516 "vbic", "i16", "$Vd, $SIMM", "$src = $Vd",
3518 (v4i16 (NEONvbicImm DPR:$src, timm:$SIMM)))]> {
3519 let Inst{9} = SIMM{9};
3522 def VBICiv2i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 0, 1, 1,
3523 (outs DPR:$Vd), (ins nModImm:$SIMM, DPR:$src),
3525 "vbic", "i32", "$Vd, $SIMM", "$src = $Vd",
3527 (v2i32 (NEONvbicImm DPR:$src, timm:$SIMM)))]> {
3528 let Inst{10-9} = SIMM{10-9};
3531 def VBICiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 1, 1,
3532 (outs QPR:$Vd), (ins nModImm:$SIMM, QPR:$src),
3534 "vbic", "i16", "$Vd, $SIMM", "$src = $Vd",
3536 (v8i16 (NEONvbicImm QPR:$src, timm:$SIMM)))]> {
3537 let Inst{9} = SIMM{9};
3540 def VBICiv4i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 1, 1, 1,
3541 (outs QPR:$Vd), (ins nModImm:$SIMM, QPR:$src),
3543 "vbic", "i32", "$Vd, $SIMM", "$src = $Vd",
3545 (v4i32 (NEONvbicImm QPR:$src, timm:$SIMM)))]> {
3546 let Inst{10-9} = SIMM{10-9};
3549 // VORN : Vector Bitwise OR NOT
3550 def VORNd : N3VX<0, 0, 0b11, 0b0001, 0, 1, (outs DPR:$dst),
3551 (ins DPR:$src1, DPR:$src2), N3RegFrm, IIC_VBINiD,
3552 "vorn", "$dst, $src1, $src2", "",
3553 [(set DPR:$dst, (v2i32 (or DPR:$src1,
3554 (vnotd DPR:$src2))))]>;
3555 def VORNq : N3VX<0, 0, 0b11, 0b0001, 1, 1, (outs QPR:$dst),
3556 (ins QPR:$src1, QPR:$src2), N3RegFrm, IIC_VBINiQ,
3557 "vorn", "$dst, $src1, $src2", "",
3558 [(set QPR:$dst, (v4i32 (or QPR:$src1,
3559 (vnotq QPR:$src2))))]>;
3561 // VMVN : Vector Bitwise NOT (Immediate)
3563 let isReMaterializable = 1 in {
3565 def VMVNv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 1, 1, (outs DPR:$dst),
3566 (ins nModImm:$SIMM), IIC_VMOVImm,
3567 "vmvn", "i16", "$dst, $SIMM", "",
3568 [(set DPR:$dst, (v4i16 (NEONvmvnImm timm:$SIMM)))]> {
3569 let Inst{9} = SIMM{9};
3572 def VMVNv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 1, 1, (outs QPR:$dst),
3573 (ins nModImm:$SIMM), IIC_VMOVImm,
3574 "vmvn", "i16", "$dst, $SIMM", "",
3575 [(set QPR:$dst, (v8i16 (NEONvmvnImm timm:$SIMM)))]> {
3576 let Inst{9} = SIMM{9};
3579 def VMVNv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 1, 1, (outs DPR:$dst),
3580 (ins nModImm:$SIMM), IIC_VMOVImm,
3581 "vmvn", "i32", "$dst, $SIMM", "",
3582 [(set DPR:$dst, (v2i32 (NEONvmvnImm timm:$SIMM)))]> {
3583 let Inst{11-8} = SIMM{11-8};
3586 def VMVNv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 1, 1, (outs QPR:$dst),
3587 (ins nModImm:$SIMM), IIC_VMOVImm,
3588 "vmvn", "i32", "$dst, $SIMM", "",
3589 [(set QPR:$dst, (v4i32 (NEONvmvnImm timm:$SIMM)))]> {
3590 let Inst{11-8} = SIMM{11-8};
3594 // VMVN : Vector Bitwise NOT
3595 def VMVNd : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 0, 0,
3596 (outs DPR:$dst), (ins DPR:$src), IIC_VSUBiD,
3597 "vmvn", "$dst, $src", "",
3598 [(set DPR:$dst, (v2i32 (vnotd DPR:$src)))]>;
3599 def VMVNq : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 1, 0,
3600 (outs QPR:$dst), (ins QPR:$src), IIC_VSUBiD,
3601 "vmvn", "$dst, $src", "",
3602 [(set QPR:$dst, (v4i32 (vnotq QPR:$src)))]>;
3603 def : Pat<(v2i32 (vnotd DPR:$src)), (VMVNd DPR:$src)>;
3604 def : Pat<(v4i32 (vnotq QPR:$src)), (VMVNq QPR:$src)>;
3606 // VBSL : Vector Bitwise Select
3607 def VBSLd : N3VX<1, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
3608 (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
3609 N3RegFrm, IIC_VCNTiD,
3610 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3612 (v2i32 (or (and DPR:$Vn, DPR:$src1),
3613 (and DPR:$Vm, (vnotd DPR:$src1)))))]>;
3614 def VBSLq : N3VX<1, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
3615 (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
3616 N3RegFrm, IIC_VCNTiQ,
3617 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3619 (v4i32 (or (and QPR:$Vn, QPR:$src1),
3620 (and QPR:$Vm, (vnotq QPR:$src1)))))]>;
3622 // VBIF : Vector Bitwise Insert if False
3623 // like VBSL but with: "vbif $dst, $src3, $src1", "$src2 = $dst",
3624 // FIXME: This instruction's encoding MAY NOT BE correct.
3625 def VBIFd : N3VX<1, 0, 0b11, 0b0001, 0, 1,
3626 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
3627 N3RegFrm, IIC_VBINiD,
3628 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3629 [/* For disassembly only; pattern left blank */]>;
3630 def VBIFq : N3VX<1, 0, 0b11, 0b0001, 1, 1,
3631 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
3632 N3RegFrm, IIC_VBINiQ,
3633 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3634 [/* For disassembly only; pattern left blank */]>;
3636 // VBIT : Vector Bitwise Insert if True
3637 // like VBSL but with: "vbit $dst, $src2, $src1", "$src3 = $dst",
3638 // FIXME: This instruction's encoding MAY NOT BE correct.
3639 def VBITd : N3VX<1, 0, 0b10, 0b0001, 0, 1,
3640 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
3641 N3RegFrm, IIC_VBINiD,
3642 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3643 [/* For disassembly only; pattern left blank */]>;
3644 def VBITq : N3VX<1, 0, 0b10, 0b0001, 1, 1,
3645 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
3646 N3RegFrm, IIC_VBINiQ,
3647 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3648 [/* For disassembly only; pattern left blank */]>;
3650 // VBIT/VBIF are not yet implemented. The TwoAddress pass will not go looking
3651 // for equivalent operations with different register constraints; it just
3654 // Vector Absolute Differences.
3656 // VABD : Vector Absolute Difference
3657 defm VABDs : N3VInt_QHS<0, 0, 0b0111, 0, N3RegFrm,
3658 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3659 "vabd", "s", int_arm_neon_vabds, 1>;
3660 defm VABDu : N3VInt_QHS<1, 0, 0b0111, 0, N3RegFrm,
3661 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3662 "vabd", "u", int_arm_neon_vabdu, 1>;
3663 def VABDfd : N3VDInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBIND,
3664 "vabd", "f32", v2f32, v2f32, int_arm_neon_vabds, 1>;
3665 def VABDfq : N3VQInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBINQ,
3666 "vabd", "f32", v4f32, v4f32, int_arm_neon_vabds, 1>;
3668 // VABDL : Vector Absolute Difference Long (Q = | D - D |)
3669 defm VABDLs : N3VLIntExt_QHS<0,1,0b0111,0, IIC_VSUBi4Q,
3670 "vabdl", "s", int_arm_neon_vabds, zext, 1>;
3671 defm VABDLu : N3VLIntExt_QHS<1,1,0b0111,0, IIC_VSUBi4Q,
3672 "vabdl", "u", int_arm_neon_vabdu, zext, 1>;
3674 // VABA : Vector Absolute Difference and Accumulate
3675 defm VABAs : N3VIntOp_QHS<0,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
3676 "vaba", "s", int_arm_neon_vabds, add>;
3677 defm VABAu : N3VIntOp_QHS<1,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
3678 "vaba", "u", int_arm_neon_vabdu, add>;
3680 // VABAL : Vector Absolute Difference and Accumulate Long (Q += | D - D |)
3681 defm VABALs : N3VLIntExtOp_QHS<0,1,0b0101,0, IIC_VABAD,
3682 "vabal", "s", int_arm_neon_vabds, zext, add>;
3683 defm VABALu : N3VLIntExtOp_QHS<1,1,0b0101,0, IIC_VABAD,
3684 "vabal", "u", int_arm_neon_vabdu, zext, add>;
3686 // Vector Maximum and Minimum.
3688 // VMAX : Vector Maximum
3689 defm VMAXs : N3VInt_QHS<0, 0, 0b0110, 0, N3RegFrm,
3690 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3691 "vmax", "s", int_arm_neon_vmaxs, 1>;
3692 defm VMAXu : N3VInt_QHS<1, 0, 0b0110, 0, N3RegFrm,
3693 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3694 "vmax", "u", int_arm_neon_vmaxu, 1>;
3695 def VMAXfd : N3VDInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBIND,
3697 v2f32, v2f32, int_arm_neon_vmaxs, 1>;
3698 def VMAXfq : N3VQInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBINQ,
3700 v4f32, v4f32, int_arm_neon_vmaxs, 1>;
3702 // VMIN : Vector Minimum
3703 defm VMINs : N3VInt_QHS<0, 0, 0b0110, 1, N3RegFrm,
3704 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3705 "vmin", "s", int_arm_neon_vmins, 1>;
3706 defm VMINu : N3VInt_QHS<1, 0, 0b0110, 1, N3RegFrm,
3707 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3708 "vmin", "u", int_arm_neon_vminu, 1>;
3709 def VMINfd : N3VDInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBIND,
3711 v2f32, v2f32, int_arm_neon_vmins, 1>;
3712 def VMINfq : N3VQInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBINQ,
3714 v4f32, v4f32, int_arm_neon_vmins, 1>;
3716 // Vector Pairwise Operations.
3718 // VPADD : Vector Pairwise Add
3719 def VPADDi8 : N3VDInt<0, 0, 0b00, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
3721 v8i8, v8i8, int_arm_neon_vpadd, 0>;
3722 def VPADDi16 : N3VDInt<0, 0, 0b01, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
3724 v4i16, v4i16, int_arm_neon_vpadd, 0>;
3725 def VPADDi32 : N3VDInt<0, 0, 0b10, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
3727 v2i32, v2i32, int_arm_neon_vpadd, 0>;
3728 def VPADDf : N3VDInt<1, 0, 0b00, 0b1101, 0, N3RegFrm,
3729 IIC_VPBIND, "vpadd", "f32",
3730 v2f32, v2f32, int_arm_neon_vpadd, 0>;
3732 // VPADDL : Vector Pairwise Add Long
3733 defm VPADDLs : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00100, 0, "vpaddl", "s",
3734 int_arm_neon_vpaddls>;
3735 defm VPADDLu : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00101, 0, "vpaddl", "u",
3736 int_arm_neon_vpaddlu>;
3738 // VPADAL : Vector Pairwise Add and Accumulate Long
3739 defm VPADALs : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01100, 0, "vpadal", "s",
3740 int_arm_neon_vpadals>;
3741 defm VPADALu : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01101, 0, "vpadal", "u",
3742 int_arm_neon_vpadalu>;
3744 // VPMAX : Vector Pairwise Maximum
3745 def VPMAXs8 : N3VDInt<0, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3746 "s8", v8i8, v8i8, int_arm_neon_vpmaxs, 0>;
3747 def VPMAXs16 : N3VDInt<0, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3748 "s16", v4i16, v4i16, int_arm_neon_vpmaxs, 0>;
3749 def VPMAXs32 : N3VDInt<0, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3750 "s32", v2i32, v2i32, int_arm_neon_vpmaxs, 0>;
3751 def VPMAXu8 : N3VDInt<1, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3752 "u8", v8i8, v8i8, int_arm_neon_vpmaxu, 0>;
3753 def VPMAXu16 : N3VDInt<1, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3754 "u16", v4i16, v4i16, int_arm_neon_vpmaxu, 0>;
3755 def VPMAXu32 : N3VDInt<1, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3756 "u32", v2i32, v2i32, int_arm_neon_vpmaxu, 0>;
3757 def VPMAXf : N3VDInt<1, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmax",
3758 "f32", v2f32, v2f32, int_arm_neon_vpmaxs, 0>;
3760 // VPMIN : Vector Pairwise Minimum
3761 def VPMINs8 : N3VDInt<0, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3762 "s8", v8i8, v8i8, int_arm_neon_vpmins, 0>;
3763 def VPMINs16 : N3VDInt<0, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3764 "s16", v4i16, v4i16, int_arm_neon_vpmins, 0>;
3765 def VPMINs32 : N3VDInt<0, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3766 "s32", v2i32, v2i32, int_arm_neon_vpmins, 0>;
3767 def VPMINu8 : N3VDInt<1, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3768 "u8", v8i8, v8i8, int_arm_neon_vpminu, 0>;
3769 def VPMINu16 : N3VDInt<1, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3770 "u16", v4i16, v4i16, int_arm_neon_vpminu, 0>;
3771 def VPMINu32 : N3VDInt<1, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3772 "u32", v2i32, v2i32, int_arm_neon_vpminu, 0>;
3773 def VPMINf : N3VDInt<1, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmin",
3774 "f32", v2f32, v2f32, int_arm_neon_vpmins, 0>;
3776 // Vector Reciprocal and Reciprocal Square Root Estimate and Step.
3778 // VRECPE : Vector Reciprocal Estimate
3779 def VRECPEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
3780 IIC_VUNAD, "vrecpe", "u32",
3781 v2i32, v2i32, int_arm_neon_vrecpe>;
3782 def VRECPEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
3783 IIC_VUNAQ, "vrecpe", "u32",
3784 v4i32, v4i32, int_arm_neon_vrecpe>;
3785 def VRECPEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
3786 IIC_VUNAD, "vrecpe", "f32",
3787 v2f32, v2f32, int_arm_neon_vrecpe>;
3788 def VRECPEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
3789 IIC_VUNAQ, "vrecpe", "f32",
3790 v4f32, v4f32, int_arm_neon_vrecpe>;
3792 // VRECPS : Vector Reciprocal Step
3793 def VRECPSfd : N3VDInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
3794 IIC_VRECSD, "vrecps", "f32",
3795 v2f32, v2f32, int_arm_neon_vrecps, 1>;
3796 def VRECPSfq : N3VQInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
3797 IIC_VRECSQ, "vrecps", "f32",
3798 v4f32, v4f32, int_arm_neon_vrecps, 1>;
3800 // VRSQRTE : Vector Reciprocal Square Root Estimate
3801 def VRSQRTEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
3802 IIC_VUNAD, "vrsqrte", "u32",
3803 v2i32, v2i32, int_arm_neon_vrsqrte>;
3804 def VRSQRTEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
3805 IIC_VUNAQ, "vrsqrte", "u32",
3806 v4i32, v4i32, int_arm_neon_vrsqrte>;
3807 def VRSQRTEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
3808 IIC_VUNAD, "vrsqrte", "f32",
3809 v2f32, v2f32, int_arm_neon_vrsqrte>;
3810 def VRSQRTEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
3811 IIC_VUNAQ, "vrsqrte", "f32",
3812 v4f32, v4f32, int_arm_neon_vrsqrte>;
3814 // VRSQRTS : Vector Reciprocal Square Root Step
3815 def VRSQRTSfd : N3VDInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
3816 IIC_VRECSD, "vrsqrts", "f32",
3817 v2f32, v2f32, int_arm_neon_vrsqrts, 1>;
3818 def VRSQRTSfq : N3VQInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
3819 IIC_VRECSQ, "vrsqrts", "f32",
3820 v4f32, v4f32, int_arm_neon_vrsqrts, 1>;
3824 // VSHL : Vector Shift
3825 defm VSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 0, N3RegVShFrm,
3826 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
3827 "vshl", "s", int_arm_neon_vshifts>;
3828 defm VSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 0, N3RegVShFrm,
3829 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
3830 "vshl", "u", int_arm_neon_vshiftu>;
3831 // VSHL : Vector Shift Left (Immediate)
3832 defm VSHLi : N2VSh_QHSD<0, 1, 0b0101, 1, IIC_VSHLiD, "vshl", "i", NEONvshl,
3834 // VSHR : Vector Shift Right (Immediate)
3835 defm VSHRs : N2VSh_QHSD<0, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "s", NEONvshrs,
3837 defm VSHRu : N2VSh_QHSD<1, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "u", NEONvshru,
3840 // VSHLL : Vector Shift Left Long
3841 defm VSHLLs : N2VLSh_QHS<0, 1, 0b1010, 0, 0, 1, "vshll", "s", NEONvshlls>;
3842 defm VSHLLu : N2VLSh_QHS<1, 1, 0b1010, 0, 0, 1, "vshll", "u", NEONvshllu>;
3844 // VSHLL : Vector Shift Left Long (with maximum shift count)
3845 class N2VLShMax<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
3846 bit op6, bit op4, string OpcodeStr, string Dt, ValueType ResTy,
3847 ValueType OpTy, SDNode OpNode>
3848 : N2VLSh<op24, op23, op11_8, op7, op6, op4, OpcodeStr, Dt,
3849 ResTy, OpTy, OpNode> {
3850 let Inst{21-16} = op21_16;
3852 def VSHLLi8 : N2VLShMax<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll", "i8",
3853 v8i16, v8i8, NEONvshlli>;
3854 def VSHLLi16 : N2VLShMax<1, 1, 0b110110, 0b0011, 0, 0, 0, "vshll", "i16",
3855 v4i32, v4i16, NEONvshlli>;
3856 def VSHLLi32 : N2VLShMax<1, 1, 0b111010, 0b0011, 0, 0, 0, "vshll", "i32",
3857 v2i64, v2i32, NEONvshlli>;
3859 // VSHRN : Vector Shift Right and Narrow
3860 defm VSHRN : N2VNSh_HSD<0,1,0b1000,0,0,1, IIC_VSHLiD, "vshrn", "i",
3863 // VRSHL : Vector Rounding Shift
3864 defm VRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 0, N3RegVShFrm,
3865 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3866 "vrshl", "s", int_arm_neon_vrshifts>;
3867 defm VRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 0, N3RegVShFrm,
3868 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3869 "vrshl", "u", int_arm_neon_vrshiftu>;
3870 // VRSHR : Vector Rounding Shift Right
3871 defm VRSHRs : N2VSh_QHSD<0,1,0b0010,1, IIC_VSHLi4D, "vrshr", "s", NEONvrshrs,
3873 defm VRSHRu : N2VSh_QHSD<1,1,0b0010,1, IIC_VSHLi4D, "vrshr", "u", NEONvrshru,
3876 // VRSHRN : Vector Rounding Shift Right and Narrow
3877 defm VRSHRN : N2VNSh_HSD<0, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vrshrn", "i",
3880 // VQSHL : Vector Saturating Shift
3881 defm VQSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 1, N3RegVShFrm,
3882 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3883 "vqshl", "s", int_arm_neon_vqshifts>;
3884 defm VQSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 1, N3RegVShFrm,
3885 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3886 "vqshl", "u", int_arm_neon_vqshiftu>;
3887 // VQSHL : Vector Saturating Shift Left (Immediate)
3888 defm VQSHLsi : N2VSh_QHSD<0,1,0b0111,1, IIC_VSHLi4D, "vqshl", "s",NEONvqshls,
3890 defm VQSHLui : N2VSh_QHSD<1,1,0b0111,1, IIC_VSHLi4D, "vqshl", "u",NEONvqshlu,
3892 // VQSHLU : Vector Saturating Shift Left (Immediate, Unsigned)
3893 defm VQSHLsu : N2VSh_QHSD<1,1,0b0110,1, IIC_VSHLi4D,"vqshlu","s",NEONvqshlsu,
3896 // VQSHRN : Vector Saturating Shift Right and Narrow
3897 defm VQSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "s",
3899 defm VQSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "u",
3902 // VQSHRUN : Vector Saturating Shift Right and Narrow (Unsigned)
3903 defm VQSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 0, 1, IIC_VSHLi4D, "vqshrun", "s",
3906 // VQRSHL : Vector Saturating Rounding Shift
3907 defm VQRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 1, N3RegVShFrm,
3908 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3909 "vqrshl", "s", int_arm_neon_vqrshifts>;
3910 defm VQRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 1, N3RegVShFrm,
3911 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3912 "vqrshl", "u", int_arm_neon_vqrshiftu>;
3914 // VQRSHRN : Vector Saturating Rounding Shift Right and Narrow
3915 defm VQRSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "s",
3917 defm VQRSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "u",
3920 // VQRSHRUN : Vector Saturating Rounding Shift Right and Narrow (Unsigned)
3921 defm VQRSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vqrshrun", "s",
3924 // VSRA : Vector Shift Right and Accumulate
3925 defm VSRAs : N2VShAdd_QHSD<0, 1, 0b0001, 1, "vsra", "s", NEONvshrs>;
3926 defm VSRAu : N2VShAdd_QHSD<1, 1, 0b0001, 1, "vsra", "u", NEONvshru>;
3927 // VRSRA : Vector Rounding Shift Right and Accumulate
3928 defm VRSRAs : N2VShAdd_QHSD<0, 1, 0b0011, 1, "vrsra", "s", NEONvrshrs>;
3929 defm VRSRAu : N2VShAdd_QHSD<1, 1, 0b0011, 1, "vrsra", "u", NEONvrshru>;
3931 // VSLI : Vector Shift Left and Insert
3932 defm VSLI : N2VShIns_QHSD<1, 1, 0b0101, 1, "vsli", NEONvsli, N2RegVShLFrm>;
3933 // VSRI : Vector Shift Right and Insert
3934 defm VSRI : N2VShIns_QHSD<1, 1, 0b0100, 1, "vsri", NEONvsri, N2RegVShRFrm>;
3936 // Vector Absolute and Saturating Absolute.
3938 // VABS : Vector Absolute Value
3939 defm VABS : N2VInt_QHS<0b11, 0b11, 0b01, 0b00110, 0,
3940 IIC_VUNAiD, IIC_VUNAiQ, "vabs", "s",
3942 def VABSfd : N2VDInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
3943 IIC_VUNAD, "vabs", "f32",
3944 v2f32, v2f32, int_arm_neon_vabs>;
3945 def VABSfq : N2VQInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
3946 IIC_VUNAQ, "vabs", "f32",
3947 v4f32, v4f32, int_arm_neon_vabs>;
3949 // VQABS : Vector Saturating Absolute Value
3950 defm VQABS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01110, 0,
3951 IIC_VQUNAiD, IIC_VQUNAiQ, "vqabs", "s",
3952 int_arm_neon_vqabs>;
3956 def vnegd : PatFrag<(ops node:$in),
3957 (sub (bitconvert (v2i32 NEONimmAllZerosV)), node:$in)>;
3958 def vnegq : PatFrag<(ops node:$in),
3959 (sub (bitconvert (v4i32 NEONimmAllZerosV)), node:$in)>;
3961 class VNEGD<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
3962 : N2V<0b11, 0b11, size, 0b01, 0b00111, 0, 0, (outs DPR:$dst), (ins DPR:$src),
3963 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
3964 [(set DPR:$dst, (Ty (vnegd DPR:$src)))]>;
3965 class VNEGQ<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
3966 : N2V<0b11, 0b11, size, 0b01, 0b00111, 1, 0, (outs QPR:$dst), (ins QPR:$src),
3967 IIC_VSHLiQ, OpcodeStr, Dt, "$dst, $src", "",
3968 [(set QPR:$dst, (Ty (vnegq QPR:$src)))]>;
3970 // VNEG : Vector Negate (integer)
3971 def VNEGs8d : VNEGD<0b00, "vneg", "s8", v8i8>;
3972 def VNEGs16d : VNEGD<0b01, "vneg", "s16", v4i16>;
3973 def VNEGs32d : VNEGD<0b10, "vneg", "s32", v2i32>;
3974 def VNEGs8q : VNEGQ<0b00, "vneg", "s8", v16i8>;
3975 def VNEGs16q : VNEGQ<0b01, "vneg", "s16", v8i16>;
3976 def VNEGs32q : VNEGQ<0b10, "vneg", "s32", v4i32>;
3978 // VNEG : Vector Negate (floating-point)
3979 def VNEGfd : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
3980 (outs DPR:$dst), (ins DPR:$src), IIC_VUNAD,
3981 "vneg", "f32", "$dst, $src", "",
3982 [(set DPR:$dst, (v2f32 (fneg DPR:$src)))]>;
3983 def VNEGf32q : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 1, 0,
3984 (outs QPR:$dst), (ins QPR:$src), IIC_VUNAQ,
3985 "vneg", "f32", "$dst, $src", "",
3986 [(set QPR:$dst, (v4f32 (fneg QPR:$src)))]>;
3988 def : Pat<(v8i8 (vnegd DPR:$src)), (VNEGs8d DPR:$src)>;
3989 def : Pat<(v4i16 (vnegd DPR:$src)), (VNEGs16d DPR:$src)>;
3990 def : Pat<(v2i32 (vnegd DPR:$src)), (VNEGs32d DPR:$src)>;
3991 def : Pat<(v16i8 (vnegq QPR:$src)), (VNEGs8q QPR:$src)>;
3992 def : Pat<(v8i16 (vnegq QPR:$src)), (VNEGs16q QPR:$src)>;
3993 def : Pat<(v4i32 (vnegq QPR:$src)), (VNEGs32q QPR:$src)>;
3995 // VQNEG : Vector Saturating Negate
3996 defm VQNEG : N2VInt_QHS<0b11, 0b11, 0b00, 0b01111, 0,
3997 IIC_VQUNAiD, IIC_VQUNAiQ, "vqneg", "s",
3998 int_arm_neon_vqneg>;
4000 // Vector Bit Counting Operations.
4002 // VCLS : Vector Count Leading Sign Bits
4003 defm VCLS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01000, 0,
4004 IIC_VCNTiD, IIC_VCNTiQ, "vcls", "s",
4006 // VCLZ : Vector Count Leading Zeros
4007 defm VCLZ : N2VInt_QHS<0b11, 0b11, 0b00, 0b01001, 0,
4008 IIC_VCNTiD, IIC_VCNTiQ, "vclz", "i",
4010 // VCNT : Vector Count One Bits
4011 def VCNTd : N2VDInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
4012 IIC_VCNTiD, "vcnt", "8",
4013 v8i8, v8i8, int_arm_neon_vcnt>;
4014 def VCNTq : N2VQInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
4015 IIC_VCNTiQ, "vcnt", "8",
4016 v16i8, v16i8, int_arm_neon_vcnt>;
4018 // Vector Swap -- for disassembly only.
4019 def VSWPd : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 0, 0,
4020 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
4021 "vswp", "$dst, $src", "", []>;
4022 def VSWPq : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 1, 0,
4023 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
4024 "vswp", "$dst, $src", "", []>;
4026 // Vector Move Operations.
4028 // VMOV : Vector Move (Register)
4030 let neverHasSideEffects = 1 in {
4031 def VMOVDneon: N3VX<0, 0, 0b10, 0b0001, 0, 1, (outs DPR:$Vd), (ins DPR:$Vm),
4032 N3RegFrm, IIC_VMOV, "vmov", "$Vd, $Vm", "", []> {
4033 let Vn{4-0} = Vm{4-0};
4035 def VMOVQ : N3VX<0, 0, 0b10, 0b0001, 1, 1, (outs QPR:$Vd), (ins QPR:$Vm),
4036 N3RegFrm, IIC_VMOV, "vmov", "$Vd, $Vm", "", []> {
4037 let Vn{4-0} = Vm{4-0};
4040 // Pseudo vector move instructions for QQ and QQQQ registers. This should
4041 // be expanded after register allocation is completed.
4042 def VMOVQQ : PseudoInst<(outs QQPR:$dst), (ins QQPR:$src),
4045 def VMOVQQQQ : PseudoInst<(outs QQQQPR:$dst), (ins QQQQPR:$src),
4047 } // neverHasSideEffects
4049 // VMOV : Vector Move (Immediate)
4051 let isReMaterializable = 1 in {
4052 def VMOVv8i8 : N1ModImm<1, 0b000, 0b1110, 0, 0, 0, 1, (outs DPR:$dst),
4053 (ins nModImm:$SIMM), IIC_VMOVImm,
4054 "vmov", "i8", "$dst, $SIMM", "",
4055 [(set DPR:$dst, (v8i8 (NEONvmovImm timm:$SIMM)))]>;
4056 def VMOVv16i8 : N1ModImm<1, 0b000, 0b1110, 0, 1, 0, 1, (outs QPR:$dst),
4057 (ins nModImm:$SIMM), IIC_VMOVImm,
4058 "vmov", "i8", "$dst, $SIMM", "",
4059 [(set QPR:$dst, (v16i8 (NEONvmovImm timm:$SIMM)))]>;
4061 def VMOVv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 0, 1, (outs DPR:$dst),
4062 (ins nModImm:$SIMM), IIC_VMOVImm,
4063 "vmov", "i16", "$dst, $SIMM", "",
4064 [(set DPR:$dst, (v4i16 (NEONvmovImm timm:$SIMM)))]> {
4065 let Inst{9} = SIMM{9};
4068 def VMOVv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 0, 1, (outs QPR:$dst),
4069 (ins nModImm:$SIMM), IIC_VMOVImm,
4070 "vmov", "i16", "$dst, $SIMM", "",
4071 [(set QPR:$dst, (v8i16 (NEONvmovImm timm:$SIMM)))]> {
4072 let Inst{9} = SIMM{9};
4075 def VMOVv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 0, 1, (outs DPR:$dst),
4076 (ins nModImm:$SIMM), IIC_VMOVImm,
4077 "vmov", "i32", "$dst, $SIMM", "",
4078 [(set DPR:$dst, (v2i32 (NEONvmovImm timm:$SIMM)))]> {
4079 let Inst{11-8} = SIMM{11-8};
4082 def VMOVv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 0, 1, (outs QPR:$dst),
4083 (ins nModImm:$SIMM), IIC_VMOVImm,
4084 "vmov", "i32", "$dst, $SIMM", "",
4085 [(set QPR:$dst, (v4i32 (NEONvmovImm timm:$SIMM)))]> {
4086 let Inst{11-8} = SIMM{11-8};
4089 def VMOVv1i64 : N1ModImm<1, 0b000, 0b1110, 0, 0, 1, 1, (outs DPR:$dst),
4090 (ins nModImm:$SIMM), IIC_VMOVImm,
4091 "vmov", "i64", "$dst, $SIMM", "",
4092 [(set DPR:$dst, (v1i64 (NEONvmovImm timm:$SIMM)))]>;
4093 def VMOVv2i64 : N1ModImm<1, 0b000, 0b1110, 0, 1, 1, 1, (outs QPR:$dst),
4094 (ins nModImm:$SIMM), IIC_VMOVImm,
4095 "vmov", "i64", "$dst, $SIMM", "",
4096 [(set QPR:$dst, (v2i64 (NEONvmovImm timm:$SIMM)))]>;
4097 } // isReMaterializable
4099 // VMOV : Vector Get Lane (move scalar to ARM core register)
4101 def VGETLNs8 : NVGetLane<{1,1,1,0,0,1,?,1}, 0b1011, {?,?},
4102 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
4103 IIC_VMOVSI, "vmov", "s8", "$R, $V[$lane]",
4104 [(set GPR:$R, (NEONvgetlanes (v8i8 DPR:$V),
4106 let Inst{21} = lane{2};
4107 let Inst{6-5} = lane{1-0};
4109 def VGETLNs16 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, {?,1},
4110 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
4111 IIC_VMOVSI, "vmov", "s16", "$R, $V[$lane]",
4112 [(set GPR:$R, (NEONvgetlanes (v4i16 DPR:$V),
4114 let Inst{21} = lane{1};
4115 let Inst{6} = lane{0};
4117 def VGETLNu8 : NVGetLane<{1,1,1,0,1,1,?,1}, 0b1011, {?,?},
4118 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
4119 IIC_VMOVSI, "vmov", "u8", "$R, $V[$lane]",
4120 [(set GPR:$R, (NEONvgetlaneu (v8i8 DPR:$V),
4122 let Inst{21} = lane{2};
4123 let Inst{6-5} = lane{1-0};
4125 def VGETLNu16 : NVGetLane<{1,1,1,0,1,0,?,1}, 0b1011, {?,1},
4126 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
4127 IIC_VMOVSI, "vmov", "u16", "$R, $V[$lane]",
4128 [(set GPR:$R, (NEONvgetlaneu (v4i16 DPR:$V),
4130 let Inst{21} = lane{1};
4131 let Inst{6} = lane{0};
4133 def VGETLNi32 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, 0b00,
4134 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
4135 IIC_VMOVSI, "vmov", "32", "$R, $V[$lane]",
4136 [(set GPR:$R, (extractelt (v2i32 DPR:$V),
4138 let Inst{21} = lane{0};
4140 // def VGETLNf32: see FMRDH and FMRDL in ARMInstrVFP.td
4141 def : Pat<(NEONvgetlanes (v16i8 QPR:$src), imm:$lane),
4142 (VGETLNs8 (v8i8 (EXTRACT_SUBREG QPR:$src,
4143 (DSubReg_i8_reg imm:$lane))),
4144 (SubReg_i8_lane imm:$lane))>;
4145 def : Pat<(NEONvgetlanes (v8i16 QPR:$src), imm:$lane),
4146 (VGETLNs16 (v4i16 (EXTRACT_SUBREG QPR:$src,
4147 (DSubReg_i16_reg imm:$lane))),
4148 (SubReg_i16_lane imm:$lane))>;
4149 def : Pat<(NEONvgetlaneu (v16i8 QPR:$src), imm:$lane),
4150 (VGETLNu8 (v8i8 (EXTRACT_SUBREG QPR:$src,
4151 (DSubReg_i8_reg imm:$lane))),
4152 (SubReg_i8_lane imm:$lane))>;
4153 def : Pat<(NEONvgetlaneu (v8i16 QPR:$src), imm:$lane),
4154 (VGETLNu16 (v4i16 (EXTRACT_SUBREG QPR:$src,
4155 (DSubReg_i16_reg imm:$lane))),
4156 (SubReg_i16_lane imm:$lane))>;
4157 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
4158 (VGETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src,
4159 (DSubReg_i32_reg imm:$lane))),
4160 (SubReg_i32_lane imm:$lane))>;
4161 def : Pat<(extractelt (v2f32 DPR:$src1), imm:$src2),
4162 (EXTRACT_SUBREG (v2f32 (COPY_TO_REGCLASS (v2f32 DPR:$src1),DPR_VFP2)),
4163 (SSubReg_f32_reg imm:$src2))>;
4164 def : Pat<(extractelt (v4f32 QPR:$src1), imm:$src2),
4165 (EXTRACT_SUBREG (v4f32 (COPY_TO_REGCLASS (v4f32 QPR:$src1),QPR_VFP2)),
4166 (SSubReg_f32_reg imm:$src2))>;
4167 //def : Pat<(extractelt (v2i64 QPR:$src1), imm:$src2),
4168 // (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
4169 def : Pat<(extractelt (v2f64 QPR:$src1), imm:$src2),
4170 (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
4173 // VMOV : Vector Set Lane (move ARM core register to scalar)
4175 let Constraints = "$src1 = $V" in {
4176 def VSETLNi8 : NVSetLane<{1,1,1,0,0,1,?,0}, 0b1011, {?,?}, (outs DPR:$V),
4177 (ins DPR:$src1, GPR:$R, nohash_imm:$lane),
4178 IIC_VMOVISL, "vmov", "8", "$V[$lane], $R",
4179 [(set DPR:$V, (vector_insert (v8i8 DPR:$src1),
4180 GPR:$R, imm:$lane))]> {
4181 let Inst{21} = lane{2};
4182 let Inst{6-5} = lane{1-0};
4184 def VSETLNi16 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, {?,1}, (outs DPR:$V),
4185 (ins DPR:$src1, GPR:$R, nohash_imm:$lane),
4186 IIC_VMOVISL, "vmov", "16", "$V[$lane], $R",
4187 [(set DPR:$V, (vector_insert (v4i16 DPR:$src1),
4188 GPR:$R, imm:$lane))]> {
4189 let Inst{21} = lane{1};
4190 let Inst{6} = lane{0};
4192 def VSETLNi32 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, 0b00, (outs DPR:$V),
4193 (ins DPR:$src1, GPR:$R, nohash_imm:$lane),
4194 IIC_VMOVISL, "vmov", "32", "$V[$lane], $R",
4195 [(set DPR:$V, (insertelt (v2i32 DPR:$src1),
4196 GPR:$R, imm:$lane))]> {
4197 let Inst{21} = lane{0};
4200 def : Pat<(vector_insert (v16i8 QPR:$src1), GPR:$src2, imm:$lane),
4201 (v16i8 (INSERT_SUBREG QPR:$src1,
4202 (v8i8 (VSETLNi8 (v8i8 (EXTRACT_SUBREG QPR:$src1,
4203 (DSubReg_i8_reg imm:$lane))),
4204 GPR:$src2, (SubReg_i8_lane imm:$lane))),
4205 (DSubReg_i8_reg imm:$lane)))>;
4206 def : Pat<(vector_insert (v8i16 QPR:$src1), GPR:$src2, imm:$lane),
4207 (v8i16 (INSERT_SUBREG QPR:$src1,
4208 (v4i16 (VSETLNi16 (v4i16 (EXTRACT_SUBREG QPR:$src1,
4209 (DSubReg_i16_reg imm:$lane))),
4210 GPR:$src2, (SubReg_i16_lane imm:$lane))),
4211 (DSubReg_i16_reg imm:$lane)))>;
4212 def : Pat<(insertelt (v4i32 QPR:$src1), GPR:$src2, imm:$lane),
4213 (v4i32 (INSERT_SUBREG QPR:$src1,
4214 (v2i32 (VSETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src1,
4215 (DSubReg_i32_reg imm:$lane))),
4216 GPR:$src2, (SubReg_i32_lane imm:$lane))),
4217 (DSubReg_i32_reg imm:$lane)))>;
4219 def : Pat<(v2f32 (insertelt DPR:$src1, SPR:$src2, imm:$src3)),
4220 (INSERT_SUBREG (v2f32 (COPY_TO_REGCLASS DPR:$src1, DPR_VFP2)),
4221 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
4222 def : Pat<(v4f32 (insertelt QPR:$src1, SPR:$src2, imm:$src3)),
4223 (INSERT_SUBREG (v4f32 (COPY_TO_REGCLASS QPR:$src1, QPR_VFP2)),
4224 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
4226 //def : Pat<(v2i64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
4227 // (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
4228 def : Pat<(v2f64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
4229 (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
4231 def : Pat<(v2f32 (scalar_to_vector SPR:$src)),
4232 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
4233 def : Pat<(v2f64 (scalar_to_vector (f64 DPR:$src))),
4234 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
4235 def : Pat<(v4f32 (scalar_to_vector SPR:$src)),
4236 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
4238 def : Pat<(v8i8 (scalar_to_vector GPR:$src)),
4239 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4240 def : Pat<(v4i16 (scalar_to_vector GPR:$src)),
4241 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4242 def : Pat<(v2i32 (scalar_to_vector GPR:$src)),
4243 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4245 def : Pat<(v16i8 (scalar_to_vector GPR:$src)),
4246 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4247 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4249 def : Pat<(v8i16 (scalar_to_vector GPR:$src)),
4250 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
4251 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4253 def : Pat<(v4i32 (scalar_to_vector GPR:$src)),
4254 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
4255 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4258 // VDUP : Vector Duplicate (from ARM core register to all elements)
4260 class VDUPD<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
4261 : NVDup<opcod1, 0b1011, opcod3, (outs DPR:$dst), (ins GPR:$src),
4262 IIC_VMOVIS, "vdup", Dt, "$dst, $src",
4263 [(set DPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
4264 class VDUPQ<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
4265 : NVDup<opcod1, 0b1011, opcod3, (outs QPR:$dst), (ins GPR:$src),
4266 IIC_VMOVIS, "vdup", Dt, "$dst, $src",
4267 [(set QPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
4269 def VDUP8d : VDUPD<0b11101100, 0b00, "8", v8i8>;
4270 def VDUP16d : VDUPD<0b11101000, 0b01, "16", v4i16>;
4271 def VDUP32d : VDUPD<0b11101000, 0b00, "32", v2i32>;
4272 def VDUP8q : VDUPQ<0b11101110, 0b00, "8", v16i8>;
4273 def VDUP16q : VDUPQ<0b11101010, 0b01, "16", v8i16>;
4274 def VDUP32q : VDUPQ<0b11101010, 0b00, "32", v4i32>;
4276 def VDUPfd : NVDup<0b11101000, 0b1011, 0b00, (outs DPR:$dst), (ins GPR:$src),
4277 IIC_VMOVIS, "vdup", "32", "$dst, $src",
4278 [(set DPR:$dst, (v2f32 (NEONvdup
4279 (f32 (bitconvert GPR:$src)))))]>;
4280 def VDUPfq : NVDup<0b11101010, 0b1011, 0b00, (outs QPR:$dst), (ins GPR:$src),
4281 IIC_VMOVIS, "vdup", "32", "$dst, $src",
4282 [(set QPR:$dst, (v4f32 (NEONvdup
4283 (f32 (bitconvert GPR:$src)))))]>;
4285 // VDUP : Vector Duplicate Lane (from scalar to all elements)
4287 class VDUPLND<bits<4> op19_16, string OpcodeStr, string Dt,
4289 : NVDupLane<op19_16, 0, (outs DPR:$dst), (ins DPR:$src, nohash_imm:$lane),
4290 IIC_VMOVD, OpcodeStr, Dt, "$dst, $src[$lane]",
4291 [(set DPR:$dst, (Ty (NEONvduplane (Ty DPR:$src), imm:$lane)))]>;
4293 class VDUPLNQ<bits<4> op19_16, string OpcodeStr, string Dt,
4294 ValueType ResTy, ValueType OpTy>
4295 : NVDupLane<op19_16, 1, (outs QPR:$dst), (ins DPR:$src, nohash_imm:$lane),
4296 IIC_VMOVQ, OpcodeStr, Dt, "$dst, $src[$lane]",
4297 [(set QPR:$dst, (ResTy (NEONvduplane (OpTy DPR:$src),
4300 // Inst{19-16} is partially specified depending on the element size.
4302 def VDUPLN8d : VDUPLND<{?,?,?,1}, "vdup", "8", v8i8> {
4303 let Inst{19-17} = lane{2-0};
4305 def VDUPLN16d : VDUPLND<{?,?,1,0}, "vdup", "16", v4i16> {
4306 let Inst{19-18} = lane{1-0};
4308 def VDUPLN32d : VDUPLND<{?,1,0,0}, "vdup", "32", v2i32> {
4309 let Inst{19} = lane{0};
4311 def VDUPLNfd : VDUPLND<{?,1,0,0}, "vdup", "32", v2f32> {
4312 let Inst{19} = lane{0};
4314 def VDUPLN8q : VDUPLNQ<{?,?,?,1}, "vdup", "8", v16i8, v8i8> {
4315 let Inst{19-17} = lane{2-0};
4317 def VDUPLN16q : VDUPLNQ<{?,?,1,0}, "vdup", "16", v8i16, v4i16> {
4318 let Inst{19-18} = lane{1-0};
4320 def VDUPLN32q : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4i32, v2i32> {
4321 let Inst{19} = lane{0};
4323 def VDUPLNfq : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4f32, v2f32> {
4324 let Inst{19} = lane{0};
4327 def : Pat<(v16i8 (NEONvduplane (v16i8 QPR:$src), imm:$lane)),
4328 (v16i8 (VDUPLN8q (v8i8 (EXTRACT_SUBREG QPR:$src,
4329 (DSubReg_i8_reg imm:$lane))),
4330 (SubReg_i8_lane imm:$lane)))>;
4331 def : Pat<(v8i16 (NEONvduplane (v8i16 QPR:$src), imm:$lane)),
4332 (v8i16 (VDUPLN16q (v4i16 (EXTRACT_SUBREG QPR:$src,
4333 (DSubReg_i16_reg imm:$lane))),
4334 (SubReg_i16_lane imm:$lane)))>;
4335 def : Pat<(v4i32 (NEONvduplane (v4i32 QPR:$src), imm:$lane)),
4336 (v4i32 (VDUPLN32q (v2i32 (EXTRACT_SUBREG QPR:$src,
4337 (DSubReg_i32_reg imm:$lane))),
4338 (SubReg_i32_lane imm:$lane)))>;
4339 def : Pat<(v4f32 (NEONvduplane (v4f32 QPR:$src), imm:$lane)),
4340 (v4f32 (VDUPLNfq (v2f32 (EXTRACT_SUBREG QPR:$src,
4341 (DSubReg_i32_reg imm:$lane))),
4342 (SubReg_i32_lane imm:$lane)))>;
4344 def VDUPfdf : PseudoNeonI<(outs DPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
4345 [(set DPR:$dst, (v2f32 (NEONvdup (f32 SPR:$src))))]>;
4346 def VDUPfqf : PseudoNeonI<(outs QPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
4347 [(set QPR:$dst, (v4f32 (NEONvdup (f32 SPR:$src))))]>;
4349 // VMOVN : Vector Narrowing Move
4350 defm VMOVN : N2VN_HSD<0b11,0b11,0b10,0b00100,0,0, IIC_VMOVN,
4351 "vmovn", "i", trunc>;
4352 // VQMOVN : Vector Saturating Narrowing Move
4353 defm VQMOVNs : N2VNInt_HSD<0b11,0b11,0b10,0b00101,0,0, IIC_VQUNAiD,
4354 "vqmovn", "s", int_arm_neon_vqmovns>;
4355 defm VQMOVNu : N2VNInt_HSD<0b11,0b11,0b10,0b00101,1,0, IIC_VQUNAiD,
4356 "vqmovn", "u", int_arm_neon_vqmovnu>;
4357 defm VQMOVNsu : N2VNInt_HSD<0b11,0b11,0b10,0b00100,1,0, IIC_VQUNAiD,
4358 "vqmovun", "s", int_arm_neon_vqmovnsu>;
4359 // VMOVL : Vector Lengthening Move
4360 defm VMOVLs : N2VL_QHS<0b01,0b10100,0,1, "vmovl", "s", sext>;
4361 defm VMOVLu : N2VL_QHS<0b11,0b10100,0,1, "vmovl", "u", zext>;
4363 // Vector Conversions.
4365 // VCVT : Vector Convert Between Floating-Point and Integers
4366 def VCVTf2sd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
4367 v2i32, v2f32, fp_to_sint>;
4368 def VCVTf2ud : N2VD<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
4369 v2i32, v2f32, fp_to_uint>;
4370 def VCVTs2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
4371 v2f32, v2i32, sint_to_fp>;
4372 def VCVTu2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
4373 v2f32, v2i32, uint_to_fp>;
4375 def VCVTf2sq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
4376 v4i32, v4f32, fp_to_sint>;
4377 def VCVTf2uq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
4378 v4i32, v4f32, fp_to_uint>;
4379 def VCVTs2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
4380 v4f32, v4i32, sint_to_fp>;
4381 def VCVTu2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
4382 v4f32, v4i32, uint_to_fp>;
4384 // VCVT : Vector Convert Between Floating-Point and Fixed-Point.
4385 def VCVTf2xsd : N2VCvtD<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
4386 v2i32, v2f32, int_arm_neon_vcvtfp2fxs>;
4387 def VCVTf2xud : N2VCvtD<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
4388 v2i32, v2f32, int_arm_neon_vcvtfp2fxu>;
4389 def VCVTxs2fd : N2VCvtD<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
4390 v2f32, v2i32, int_arm_neon_vcvtfxs2fp>;
4391 def VCVTxu2fd : N2VCvtD<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
4392 v2f32, v2i32, int_arm_neon_vcvtfxu2fp>;
4394 def VCVTf2xsq : N2VCvtQ<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
4395 v4i32, v4f32, int_arm_neon_vcvtfp2fxs>;
4396 def VCVTf2xuq : N2VCvtQ<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
4397 v4i32, v4f32, int_arm_neon_vcvtfp2fxu>;
4398 def VCVTxs2fq : N2VCvtQ<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
4399 v4f32, v4i32, int_arm_neon_vcvtfxs2fp>;
4400 def VCVTxu2fq : N2VCvtQ<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
4401 v4f32, v4i32, int_arm_neon_vcvtfxu2fp>;
4405 // VREV64 : Vector Reverse elements within 64-bit doublewords
4407 class VREV64D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4408 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 0, 0, (outs DPR:$Vd),
4409 (ins DPR:$Vm), IIC_VMOVD,
4410 OpcodeStr, Dt, "$Vd, $Vm", "",
4411 [(set DPR:$Vd, (Ty (NEONvrev64 (Ty DPR:$Vm))))]>;
4412 class VREV64Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4413 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 1, 0, (outs QPR:$Vd),
4414 (ins QPR:$Vm), IIC_VMOVQ,
4415 OpcodeStr, Dt, "$Vd, $Vm", "",
4416 [(set QPR:$Vd, (Ty (NEONvrev64 (Ty QPR:$Vm))))]>;
4418 def VREV64d8 : VREV64D<0b00, "vrev64", "8", v8i8>;
4419 def VREV64d16 : VREV64D<0b01, "vrev64", "16", v4i16>;
4420 def VREV64d32 : VREV64D<0b10, "vrev64", "32", v2i32>;
4421 def VREV64df : VREV64D<0b10, "vrev64", "32", v2f32>;
4423 def VREV64q8 : VREV64Q<0b00, "vrev64", "8", v16i8>;
4424 def VREV64q16 : VREV64Q<0b01, "vrev64", "16", v8i16>;
4425 def VREV64q32 : VREV64Q<0b10, "vrev64", "32", v4i32>;
4426 def VREV64qf : VREV64Q<0b10, "vrev64", "32", v4f32>;
4428 // VREV32 : Vector Reverse elements within 32-bit words
4430 class VREV32D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4431 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 0, 0, (outs DPR:$Vd),
4432 (ins DPR:$Vm), IIC_VMOVD,
4433 OpcodeStr, Dt, "$Vd, $Vm", "",
4434 [(set DPR:$Vd, (Ty (NEONvrev32 (Ty DPR:$Vm))))]>;
4435 class VREV32Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4436 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 1, 0, (outs QPR:$Vd),
4437 (ins QPR:$Vm), IIC_VMOVQ,
4438 OpcodeStr, Dt, "$Vd, $Vm", "",
4439 [(set QPR:$Vd, (Ty (NEONvrev32 (Ty QPR:$Vm))))]>;
4441 def VREV32d8 : VREV32D<0b00, "vrev32", "8", v8i8>;
4442 def VREV32d16 : VREV32D<0b01, "vrev32", "16", v4i16>;
4444 def VREV32q8 : VREV32Q<0b00, "vrev32", "8", v16i8>;
4445 def VREV32q16 : VREV32Q<0b01, "vrev32", "16", v8i16>;
4447 // VREV16 : Vector Reverse elements within 16-bit halfwords
4449 class VREV16D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4450 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 0, 0, (outs DPR:$Vd),
4451 (ins DPR:$Vm), IIC_VMOVD,
4452 OpcodeStr, Dt, "$Vd, $Vm", "",
4453 [(set DPR:$Vd, (Ty (NEONvrev16 (Ty DPR:$Vm))))]>;
4454 class VREV16Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4455 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 1, 0, (outs QPR:$Vd),
4456 (ins QPR:$Vm), IIC_VMOVQ,
4457 OpcodeStr, Dt, "$Vd, $Vm", "",
4458 [(set QPR:$Vd, (Ty (NEONvrev16 (Ty QPR:$Vm))))]>;
4460 def VREV16d8 : VREV16D<0b00, "vrev16", "8", v8i8>;
4461 def VREV16q8 : VREV16Q<0b00, "vrev16", "8", v16i8>;
4463 // Other Vector Shuffles.
4465 // VEXT : Vector Extract
4467 class VEXTd<string OpcodeStr, string Dt, ValueType Ty>
4468 : N3V<0,1,0b11,{?,?,?,?},0,0, (outs DPR:$Vd),
4469 (ins DPR:$Vn, DPR:$Vm, i32imm:$index), NVExtFrm,
4470 IIC_VEXTD, OpcodeStr, Dt, "$Vd, $Vn, $Vm, $index", "",
4471 [(set DPR:$Vd, (Ty (NEONvext (Ty DPR:$Vn),
4472 (Ty DPR:$Vm), imm:$index)))]> {
4474 let Inst{11-8} = index{3-0};
4477 class VEXTq<string OpcodeStr, string Dt, ValueType Ty>
4478 : N3V<0,1,0b11,{?,?,?,?},1,0, (outs QPR:$Vd),
4479 (ins QPR:$Vn, QPR:$Vm, i32imm:$index), NVExtFrm,
4480 IIC_VEXTQ, OpcodeStr, Dt, "$Vd, $Vn, $Vm, $index", "",
4481 [(set QPR:$Vd, (Ty (NEONvext (Ty QPR:$Vn),
4482 (Ty QPR:$Vm), imm:$index)))]> {
4484 let Inst{11-8} = index{3-0};
4487 def VEXTd8 : VEXTd<"vext", "8", v8i8> {
4488 let Inst{11-8} = index{3-0};
4490 def VEXTd16 : VEXTd<"vext", "16", v4i16> {
4491 let Inst{11-9} = index{2-0};
4494 def VEXTd32 : VEXTd<"vext", "32", v2i32> {
4495 let Inst{11-10} = index{1-0};
4496 let Inst{9-8} = 0b00;
4498 def VEXTdf : VEXTd<"vext", "32", v2f32> {
4499 let Inst{11} = index{0};
4500 let Inst{10-8} = 0b000;
4503 def VEXTq8 : VEXTq<"vext", "8", v16i8> {
4504 let Inst{11-8} = index{3-0};
4506 def VEXTq16 : VEXTq<"vext", "16", v8i16> {
4507 let Inst{11-9} = index{2-0};
4510 def VEXTq32 : VEXTq<"vext", "32", v4i32> {
4511 let Inst{11-10} = index{1-0};
4512 let Inst{9-8} = 0b00;
4514 def VEXTqf : VEXTq<"vext", "32", v4f32> {
4515 let Inst{11} = index{0};
4516 let Inst{10-8} = 0b000;
4519 // VTRN : Vector Transpose
4521 def VTRNd8 : N2VDShuffle<0b00, 0b00001, "vtrn", "8">;
4522 def VTRNd16 : N2VDShuffle<0b01, 0b00001, "vtrn", "16">;
4523 def VTRNd32 : N2VDShuffle<0b10, 0b00001, "vtrn", "32">;
4525 def VTRNq8 : N2VQShuffle<0b00, 0b00001, IIC_VPERMQ, "vtrn", "8">;
4526 def VTRNq16 : N2VQShuffle<0b01, 0b00001, IIC_VPERMQ, "vtrn", "16">;
4527 def VTRNq32 : N2VQShuffle<0b10, 0b00001, IIC_VPERMQ, "vtrn", "32">;
4529 // VUZP : Vector Unzip (Deinterleave)
4531 def VUZPd8 : N2VDShuffle<0b00, 0b00010, "vuzp", "8">;
4532 def VUZPd16 : N2VDShuffle<0b01, 0b00010, "vuzp", "16">;
4533 def VUZPd32 : N2VDShuffle<0b10, 0b00010, "vuzp", "32">;
4535 def VUZPq8 : N2VQShuffle<0b00, 0b00010, IIC_VPERMQ3, "vuzp", "8">;
4536 def VUZPq16 : N2VQShuffle<0b01, 0b00010, IIC_VPERMQ3, "vuzp", "16">;
4537 def VUZPq32 : N2VQShuffle<0b10, 0b00010, IIC_VPERMQ3, "vuzp", "32">;
4539 // VZIP : Vector Zip (Interleave)
4541 def VZIPd8 : N2VDShuffle<0b00, 0b00011, "vzip", "8">;
4542 def VZIPd16 : N2VDShuffle<0b01, 0b00011, "vzip", "16">;
4543 def VZIPd32 : N2VDShuffle<0b10, 0b00011, "vzip", "32">;
4545 def VZIPq8 : N2VQShuffle<0b00, 0b00011, IIC_VPERMQ3, "vzip", "8">;
4546 def VZIPq16 : N2VQShuffle<0b01, 0b00011, IIC_VPERMQ3, "vzip", "16">;
4547 def VZIPq32 : N2VQShuffle<0b10, 0b00011, IIC_VPERMQ3, "vzip", "32">;
4549 // Vector Table Lookup and Table Extension.
4551 // VTBL : Vector Table Lookup
4553 : N3V<1,1,0b11,0b1000,0,0, (outs DPR:$Vd),
4554 (ins DPR:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTB1,
4555 "vtbl", "8", "$Vd, \\{$Vn\\}, $Vm", "",
4556 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbl1 DPR:$Vn, DPR:$Vm)))]>;
4557 let hasExtraSrcRegAllocReq = 1 in {
4559 : N3V<1,1,0b11,0b1001,0,0, (outs DPR:$Vd),
4560 (ins DPR:$Vn, DPR:$tbl2, DPR:$Vm), NVTBLFrm, IIC_VTB2,
4561 "vtbl", "8", "$Vd, \\{$Vn, $tbl2\\}, $Vm", "", []>;
4563 : N3V<1,1,0b11,0b1010,0,0, (outs DPR:$Vd),
4564 (ins DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$Vm), NVTBLFrm, IIC_VTB3,
4565 "vtbl", "8", "$Vd, \\{$Vn, $tbl2, $tbl3\\}, $Vm", "", []>;
4567 : N3V<1,1,0b11,0b1011,0,0, (outs DPR:$Vd),
4568 (ins DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$Vm),
4570 "vtbl", "8", "$Vd, \\{$Vn, $tbl2, $tbl3, $tbl4\\}, $Vm", "", []>;
4571 } // hasExtraSrcRegAllocReq = 1
4574 : PseudoNeonI<(outs DPR:$dst), (ins QPR:$tbl, DPR:$src), IIC_VTB2, "", []>;
4576 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB3, "", []>;
4578 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB4, "", []>;
4580 // VTBX : Vector Table Extension
4582 : N3V<1,1,0b11,0b1000,1,0, (outs DPR:$Vd),
4583 (ins DPR:$orig, DPR:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTBX1,
4584 "vtbx", "8", "$Vd, \\{$Vn\\}, $Vm", "$orig = $Vd",
4585 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbx1
4586 DPR:$orig, DPR:$Vn, DPR:$Vm)))]>;
4587 let hasExtraSrcRegAllocReq = 1 in {
4589 : N3V<1,1,0b11,0b1001,1,0, (outs DPR:$Vd),
4590 (ins DPR:$orig, DPR:$Vn, DPR:$tbl2, DPR:$Vm), NVTBLFrm, IIC_VTBX2,
4591 "vtbx", "8", "$Vd, \\{$Vn, $tbl2\\}, $Vm", "$orig = $Vd", []>;
4593 : N3V<1,1,0b11,0b1010,1,0, (outs DPR:$Vd),
4594 (ins DPR:$orig, DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$Vm),
4595 NVTBLFrm, IIC_VTBX3,
4596 "vtbx", "8", "$Vd, \\{$Vn, $tbl2, $tbl3\\}, $Vm",
4599 : N3V<1,1,0b11,0b1011,1,0, (outs DPR:$Vd), (ins DPR:$orig, DPR:$Vn,
4600 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$Vm), NVTBLFrm, IIC_VTBX4,
4601 "vtbx", "8", "$Vd, \\{$Vn, $tbl2, $tbl3, $tbl4\\}, $Vm",
4603 } // hasExtraSrcRegAllocReq = 1
4606 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QPR:$tbl, DPR:$src),
4607 IIC_VTBX2, "$orig = $dst", []>;
4609 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
4610 IIC_VTBX3, "$orig = $dst", []>;
4612 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
4613 IIC_VTBX4, "$orig = $dst", []>;
4615 //===----------------------------------------------------------------------===//
4616 // NEON instructions for single-precision FP math
4617 //===----------------------------------------------------------------------===//
4619 class N2VSPat<SDNode OpNode, ValueType ResTy, ValueType OpTy, NeonI Inst>
4620 : NEONFPPat<(ResTy (OpNode SPR:$a)),
4621 (EXTRACT_SUBREG (OpTy (Inst (INSERT_SUBREG (OpTy (IMPLICIT_DEF)),
4625 class N3VSPat<SDNode OpNode, NeonI Inst>
4626 : NEONFPPat<(f32 (OpNode SPR:$a, SPR:$b)),
4627 (EXTRACT_SUBREG (v2f32
4628 (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
4630 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
4634 class N3VSMulOpPat<SDNode MulNode, SDNode OpNode, NeonI Inst>
4635 : NEONFPPat<(f32 (OpNode SPR:$acc, (f32 (MulNode SPR:$a, SPR:$b)))),
4636 (EXTRACT_SUBREG (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
4638 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
4640 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
4644 // These need separate instructions because they must use DPR_VFP2 register
4645 // class which have SPR sub-registers.
4647 // Vector Add Operations used for single-precision FP
4648 let neverHasSideEffects = 1 in
4649 def VADDfd_sfp : N3VS<0,0,0b00,0b1101,0, "vadd", "f32", v2f32, v2f32, fadd, 1>;
4650 def : N3VSPat<fadd, VADDfd_sfp>;
4652 // Vector Sub Operations used for single-precision FP
4653 let neverHasSideEffects = 1 in
4654 def VSUBfd_sfp : N3VS<0,0,0b10,0b1101,0, "vsub", "f32", v2f32, v2f32, fsub, 0>;
4655 def : N3VSPat<fsub, VSUBfd_sfp>;
4657 // Vector Multiply Operations used for single-precision FP
4658 let neverHasSideEffects = 1 in
4659 def VMULfd_sfp : N3VS<1,0,0b00,0b1101,1, "vmul", "f32", v2f32, v2f32, fmul, 1>;
4660 def : N3VSPat<fmul, VMULfd_sfp>;
4662 // Vector Multiply-Accumulate/Subtract used for single-precision FP
4663 // vml[as].f32 can cause 4-8 cycle stalls in following ASIMD instructions, so
4664 // we want to avoid them for now. e.g., alternating vmla/vadd instructions.
4666 //let neverHasSideEffects = 1 in
4667 //def VMLAfd_sfp : N3VSMulOp<0,0,0b00,0b1101,1, IIC_VMACD, "vmla", "f32",
4668 // v2f32, fmul, fadd>;
4669 //def : N3VSMulOpPat<fmul, fadd, VMLAfd_sfp>;
4671 //let neverHasSideEffects = 1 in
4672 //def VMLSfd_sfp : N3VSMulOp<0,0,0b10,0b1101,1, IIC_VMACD, "vmls", "f32",
4673 // v2f32, fmul, fsub>;
4674 //def : N3VSMulOpPat<fmul, fsub, VMLSfd_sfp>;
4676 // Vector Absolute used for single-precision FP
4677 let neverHasSideEffects = 1 in
4678 def VABSfd_sfp : N2V<0b11, 0b11, 0b10, 0b01, 0b01110, 0, 0,
4679 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), IIC_VUNAD,
4680 "vabs", "f32", "$dst, $src", "", []>;
4681 def : N2VSPat<fabs, f32, v2f32, VABSfd_sfp>;
4683 // Vector Negate used for single-precision FP
4684 let neverHasSideEffects = 1 in
4685 def VNEGfd_sfp : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
4686 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), IIC_VUNAD,
4687 "vneg", "f32", "$dst, $src", "", []>;
4688 def : N2VSPat<fneg, f32, v2f32, VNEGfd_sfp>;
4690 // Vector Maximum used for single-precision FP
4691 let neverHasSideEffects = 1 in
4692 def VMAXfd_sfp : N3V<0, 0, 0b00, 0b1111, 0, 0, (outs DPR_VFP2:$dst),
4693 (ins DPR_VFP2:$src1, DPR_VFP2:$src2), N3RegFrm, IIC_VBIND,
4694 "vmax", "f32", "$dst, $src1, $src2", "", []>;
4695 def : N3VSPat<NEONfmax, VMAXfd_sfp>;
4697 // Vector Minimum used for single-precision FP
4698 let neverHasSideEffects = 1 in
4699 def VMINfd_sfp : N3V<0, 0, 0b00, 0b1111, 0, 0, (outs DPR_VFP2:$dst),
4700 (ins DPR_VFP2:$src1, DPR_VFP2:$src2), N3RegFrm, IIC_VBIND,
4701 "vmin", "f32", "$dst, $src1, $src2", "", []>;
4702 def : N3VSPat<NEONfmin, VMINfd_sfp>;
4704 // Vector Convert between single-precision FP and integer
4705 let neverHasSideEffects = 1 in
4706 def VCVTf2sd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
4707 v2i32, v2f32, fp_to_sint>;
4708 def : N2VSPat<arm_ftosi, f32, v2f32, VCVTf2sd_sfp>;
4710 let neverHasSideEffects = 1 in
4711 def VCVTf2ud_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
4712 v2i32, v2f32, fp_to_uint>;
4713 def : N2VSPat<arm_ftoui, f32, v2f32, VCVTf2ud_sfp>;
4715 let neverHasSideEffects = 1 in
4716 def VCVTs2fd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
4717 v2f32, v2i32, sint_to_fp>;
4718 def : N2VSPat<arm_sitof, f32, v2i32, VCVTs2fd_sfp>;
4720 let neverHasSideEffects = 1 in
4721 def VCVTu2fd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
4722 v2f32, v2i32, uint_to_fp>;
4723 def : N2VSPat<arm_uitof, f32, v2i32, VCVTu2fd_sfp>;
4725 //===----------------------------------------------------------------------===//
4726 // Non-Instruction Patterns
4727 //===----------------------------------------------------------------------===//
4730 def : Pat<(v1i64 (bitconvert (v2i32 DPR:$src))), (v1i64 DPR:$src)>;
4731 def : Pat<(v1i64 (bitconvert (v4i16 DPR:$src))), (v1i64 DPR:$src)>;
4732 def : Pat<(v1i64 (bitconvert (v8i8 DPR:$src))), (v1i64 DPR:$src)>;
4733 def : Pat<(v1i64 (bitconvert (f64 DPR:$src))), (v1i64 DPR:$src)>;
4734 def : Pat<(v1i64 (bitconvert (v2f32 DPR:$src))), (v1i64 DPR:$src)>;
4735 def : Pat<(v2i32 (bitconvert (v1i64 DPR:$src))), (v2i32 DPR:$src)>;
4736 def : Pat<(v2i32 (bitconvert (v4i16 DPR:$src))), (v2i32 DPR:$src)>;
4737 def : Pat<(v2i32 (bitconvert (v8i8 DPR:$src))), (v2i32 DPR:$src)>;
4738 def : Pat<(v2i32 (bitconvert (f64 DPR:$src))), (v2i32 DPR:$src)>;
4739 def : Pat<(v2i32 (bitconvert (v2f32 DPR:$src))), (v2i32 DPR:$src)>;
4740 def : Pat<(v4i16 (bitconvert (v1i64 DPR:$src))), (v4i16 DPR:$src)>;
4741 def : Pat<(v4i16 (bitconvert (v2i32 DPR:$src))), (v4i16 DPR:$src)>;
4742 def : Pat<(v4i16 (bitconvert (v8i8 DPR:$src))), (v4i16 DPR:$src)>;
4743 def : Pat<(v4i16 (bitconvert (f64 DPR:$src))), (v4i16 DPR:$src)>;
4744 def : Pat<(v4i16 (bitconvert (v2f32 DPR:$src))), (v4i16 DPR:$src)>;
4745 def : Pat<(v8i8 (bitconvert (v1i64 DPR:$src))), (v8i8 DPR:$src)>;
4746 def : Pat<(v8i8 (bitconvert (v2i32 DPR:$src))), (v8i8 DPR:$src)>;
4747 def : Pat<(v8i8 (bitconvert (v4i16 DPR:$src))), (v8i8 DPR:$src)>;
4748 def : Pat<(v8i8 (bitconvert (f64 DPR:$src))), (v8i8 DPR:$src)>;
4749 def : Pat<(v8i8 (bitconvert (v2f32 DPR:$src))), (v8i8 DPR:$src)>;
4750 def : Pat<(f64 (bitconvert (v1i64 DPR:$src))), (f64 DPR:$src)>;
4751 def : Pat<(f64 (bitconvert (v2i32 DPR:$src))), (f64 DPR:$src)>;
4752 def : Pat<(f64 (bitconvert (v4i16 DPR:$src))), (f64 DPR:$src)>;
4753 def : Pat<(f64 (bitconvert (v8i8 DPR:$src))), (f64 DPR:$src)>;
4754 def : Pat<(f64 (bitconvert (v2f32 DPR:$src))), (f64 DPR:$src)>;
4755 def : Pat<(v2f32 (bitconvert (f64 DPR:$src))), (v2f32 DPR:$src)>;
4756 def : Pat<(v2f32 (bitconvert (v1i64 DPR:$src))), (v2f32 DPR:$src)>;
4757 def : Pat<(v2f32 (bitconvert (v2i32 DPR:$src))), (v2f32 DPR:$src)>;
4758 def : Pat<(v2f32 (bitconvert (v4i16 DPR:$src))), (v2f32 DPR:$src)>;
4759 def : Pat<(v2f32 (bitconvert (v8i8 DPR:$src))), (v2f32 DPR:$src)>;
4761 def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (v2i64 QPR:$src)>;
4762 def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (v2i64 QPR:$src)>;
4763 def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (v2i64 QPR:$src)>;
4764 def : Pat<(v2i64 (bitconvert (v2f64 QPR:$src))), (v2i64 QPR:$src)>;
4765 def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (v2i64 QPR:$src)>;
4766 def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (v4i32 QPR:$src)>;
4767 def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (v4i32 QPR:$src)>;
4768 def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (v4i32 QPR:$src)>;
4769 def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (v4i32 QPR:$src)>;
4770 def : Pat<(v4i32 (bitconvert (v4f32 QPR:$src))), (v4i32 QPR:$src)>;
4771 def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (v8i16 QPR:$src)>;
4772 def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (v8i16 QPR:$src)>;
4773 def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (v8i16 QPR:$src)>;
4774 def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (v8i16 QPR:$src)>;
4775 def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (v8i16 QPR:$src)>;
4776 def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (v16i8 QPR:$src)>;
4777 def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (v16i8 QPR:$src)>;
4778 def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (v16i8 QPR:$src)>;
4779 def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (v16i8 QPR:$src)>;
4780 def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (v16i8 QPR:$src)>;
4781 def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (v4f32 QPR:$src)>;
4782 def : Pat<(v4f32 (bitconvert (v4i32 QPR:$src))), (v4f32 QPR:$src)>;
4783 def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (v4f32 QPR:$src)>;
4784 def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (v4f32 QPR:$src)>;
4785 def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (v4f32 QPR:$src)>;
4786 def : Pat<(v2f64 (bitconvert (v2i64 QPR:$src))), (v2f64 QPR:$src)>;
4787 def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (v2f64 QPR:$src)>;
4788 def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 QPR:$src)>;
4789 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 QPR:$src)>;
4790 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 QPR:$src)>;