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 = $wb",
767 def VLD4LNd8_UPD : VLD4LNWB<0b0011, {?,?,?,?}, "8"> {
768 let Inst{7-5} = lane{2-0};
770 def VLD4LNd16_UPD : VLD4LNWB<0b0111, {?,?,0,?}, "16"> {
771 let Inst{7-6} = lane{1-0};
773 def VLD4LNd32_UPD : VLD4LNWB<0b1011, {?,0,?,?}, "32"> {
774 let Inst{7} = lane{0};
778 def VLD4LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
779 def VLD4LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
780 def VLD4LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
782 def VLD4LNq16_UPD : VLD4LNWB<0b0111, {?,?,1,?}, "16"> {
783 let Inst{7-6} = lane{1-0};
785 def VLD4LNq32_UPD : VLD4LNWB<0b1011, {?,1,?,?}, "32"> {
786 let Inst{7} = lane{0};
790 def VLD4LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
791 def VLD4LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
793 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
795 // VLD1DUP : Vector Load (single element to all lanes)
796 class VLD1DUP<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
798 : NLdSt<1, 0b10, op11_8, op7_4, (outs DPR:$Vd), (ins addrmode6:$Rn),
799 IIC_VLD1dup, "vld1", Dt, "\\{$Vd[]\\}, $Rn", "",
800 [(set DPR:$Vd, (Ty (NEONvdup (i32 (LoadOp addrmode6:$Rn)))))]> {
804 class VLD1QDUPPseudo<ValueType Ty, PatFrag LoadOp> : VLDQPseudo<IIC_VLD1dup> {
805 let Pattern = [(set QPR:$dst,
806 (Ty (NEONvdup (i32 (LoadOp addrmode6:$addr)))))];
809 def VLD1DUPd8 : VLD1DUP<0b1100, {0,0,0,?}, "8", v8i8, extloadi8>;
810 def VLD1DUPd16 : VLD1DUP<0b1100, {0,1,0,?}, "16", v4i16, extloadi16>;
811 def VLD1DUPd32 : VLD1DUP<0b1100, {1,0,0,?}, "32", v2i32, load>;
813 def VLD1DUPq8Pseudo : VLD1QDUPPseudo<v16i8, extloadi8>;
814 def VLD1DUPq16Pseudo : VLD1QDUPPseudo<v8i16, extloadi16>;
815 def VLD1DUPq32Pseudo : VLD1QDUPPseudo<v4i32, load>;
817 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
819 class VLD1QDUP<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
821 : NLdSt<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2),
822 (ins addrmode6:$Rn), IIC_VLD1dup,
823 "vld1", Dt, "\\{$Vd[], $dst2[]\\}, $Rn", "", []> {
828 def VLD1DUPq8 : VLD1QDUP<0b1100, {0,0,1,0}, "8", v16i8, extloadi8>;
829 def VLD1DUPq16 : VLD1QDUP<0b1100, {0,1,1,?}, "16", v8i16, extloadi16>;
830 def VLD1DUPq32 : VLD1QDUP<0b1100, {1,0,1,?}, "32", v4i32, load>;
832 // ...with address register writeback:
833 class VLD1DUPWB<bits<4> op11_8, bits<4> op7_4, string Dt>
834 : NLdSt<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, GPR:$wb),
835 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD1dupu,
836 "vld1", Dt, "\\{$Vd[]\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
839 class VLD1QDUPWB<bits<4> op11_8, bits<4> op7_4, string Dt>
840 : NLdSt<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
841 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD1dupu,
842 "vld1", Dt, "\\{$Vd[], $dst2[]\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
846 def VLD1DUPd8_UPD : VLD1DUPWB<0b1100, {0,0,0,0}, "8">;
847 def VLD1DUPd16_UPD : VLD1DUPWB<0b1100, {0,1,0,?}, "16">;
848 def VLD1DUPd32_UPD : VLD1DUPWB<0b1100, {1,0,0,?}, "32">;
850 def VLD1DUPq8_UPD : VLD1QDUPWB<0b1100, {0,0,1,0}, "8">;
851 def VLD1DUPq16_UPD : VLD1QDUPWB<0b1100, {0,1,1,?}, "16">;
852 def VLD1DUPq32_UPD : VLD1QDUPWB<0b1100, {1,0,1,?}, "32">;
854 def VLD1DUPq8Pseudo_UPD : VLDQWBPseudo<IIC_VLD1dupu>;
855 def VLD1DUPq16Pseudo_UPD : VLDQWBPseudo<IIC_VLD1dupu>;
856 def VLD1DUPq32Pseudo_UPD : VLDQWBPseudo<IIC_VLD1dupu>;
858 // VLD2DUP : Vector Load (single 2-element structure to all lanes)
859 // VLD3DUP : Vector Load (single 3-element structure to all lanes)
860 // VLD4DUP : Vector Load (single 4-element structure to all lanes)
861 // FIXME: Not yet implemented.
862 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
864 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
866 // Classes for VST* pseudo-instructions with multi-register operands.
867 // These are expanded to real instructions after register allocation.
868 class VSTQPseudo<InstrItinClass itin>
869 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src), itin, "">;
870 class VSTQWBPseudo<InstrItinClass itin>
871 : PseudoNLdSt<(outs GPR:$wb),
872 (ins addrmode6:$addr, am6offset:$offset, QPR:$src), itin,
874 class VSTQQPseudo<InstrItinClass itin>
875 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src), itin, "">;
876 class VSTQQWBPseudo<InstrItinClass itin>
877 : PseudoNLdSt<(outs GPR:$wb),
878 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src), itin,
880 class VSTQQQQWBPseudo<InstrItinClass itin>
881 : PseudoNLdSt<(outs GPR:$wb),
882 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
885 // VST1 : Vector Store (multiple single elements)
886 class VST1D<bits<4> op7_4, string Dt>
887 : NLdSt<0,0b00,0b0111,op7_4, (outs), (ins addrmode6:$Rn, DPR:$Vd),
888 IIC_VST1, "vst1", Dt, "\\{$Vd\\}, $Rn", "", []> {
892 class VST1Q<bits<4> op7_4, string Dt>
893 : NLdSt<0,0b00,0b1010,op7_4, (outs),
894 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2), IIC_VST1x2,
895 "vst1", Dt, "\\{$Vd, $src2\\}, $Rn", "", []> {
897 let Inst{5-4} = Rn{5-4};
900 def VST1d8 : VST1D<{0,0,0,?}, "8">;
901 def VST1d16 : VST1D<{0,1,0,?}, "16">;
902 def VST1d32 : VST1D<{1,0,0,?}, "32">;
903 def VST1d64 : VST1D<{1,1,0,?}, "64">;
905 def VST1q8 : VST1Q<{0,0,?,?}, "8">;
906 def VST1q16 : VST1Q<{0,1,?,?}, "16">;
907 def VST1q32 : VST1Q<{1,0,?,?}, "32">;
908 def VST1q64 : VST1Q<{1,1,?,?}, "64">;
910 def VST1q8Pseudo : VSTQPseudo<IIC_VST1x2>;
911 def VST1q16Pseudo : VSTQPseudo<IIC_VST1x2>;
912 def VST1q32Pseudo : VSTQPseudo<IIC_VST1x2>;
913 def VST1q64Pseudo : VSTQPseudo<IIC_VST1x2>;
915 // ...with address register writeback:
916 class VST1DWB<bits<4> op7_4, string Dt>
917 : NLdSt<0, 0b00, 0b0111, op7_4, (outs GPR:$wb),
918 (ins addrmode6:$Rn, am6offset:$Rm, DPR:$Vd), IIC_VST1u,
919 "vst1", Dt, "\\{$Vd\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
922 class VST1QWB<bits<4> op7_4, string Dt>
923 : NLdSt<0, 0b00, 0b1010, op7_4, (outs GPR:$wb),
924 (ins addrmode6:$Rn, am6offset:$Rm, DPR:$Vd, DPR:$src2),
925 IIC_VST1x2u, "vst1", Dt, "\\{$Vd, $src2\\}, $Rn$Rm",
926 "$Rn.addr = $wb", []> {
927 let Inst{5-4} = Rn{5-4};
930 def VST1d8_UPD : VST1DWB<{0,0,0,?}, "8">;
931 def VST1d16_UPD : VST1DWB<{0,1,0,?}, "16">;
932 def VST1d32_UPD : VST1DWB<{1,0,0,?}, "32">;
933 def VST1d64_UPD : VST1DWB<{1,1,0,?}, "64">;
935 def VST1q8_UPD : VST1QWB<{0,0,?,?}, "8">;
936 def VST1q16_UPD : VST1QWB<{0,1,?,?}, "16">;
937 def VST1q32_UPD : VST1QWB<{1,0,?,?}, "32">;
938 def VST1q64_UPD : VST1QWB<{1,1,?,?}, "64">;
940 def VST1q8Pseudo_UPD : VSTQWBPseudo<IIC_VST1x2u>;
941 def VST1q16Pseudo_UPD : VSTQWBPseudo<IIC_VST1x2u>;
942 def VST1q32Pseudo_UPD : VSTQWBPseudo<IIC_VST1x2u>;
943 def VST1q64Pseudo_UPD : VSTQWBPseudo<IIC_VST1x2u>;
945 // ...with 3 registers (some of these are only for the disassembler):
946 class VST1D3<bits<4> op7_4, string Dt>
947 : NLdSt<0, 0b00, 0b0110, op7_4, (outs),
948 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3),
949 IIC_VST1x3, "vst1", Dt, "\\{$Vd, $src2, $src3\\}, $Rn", "", []> {
953 class VST1D3WB<bits<4> op7_4, string Dt>
954 : NLdSt<0, 0b00, 0b0110, op7_4, (outs GPR:$wb),
955 (ins addrmode6:$Rn, am6offset:$Rm,
956 DPR:$Vd, DPR:$src2, DPR:$src3),
957 IIC_VST1x3u, "vst1", Dt, "\\{$Vd, $src2, $src3\\}, $Rn$Rm",
958 "$Rn.addr = $wb", []> {
962 def VST1d8T : VST1D3<{0,0,0,?}, "8">;
963 def VST1d16T : VST1D3<{0,1,0,?}, "16">;
964 def VST1d32T : VST1D3<{1,0,0,?}, "32">;
965 def VST1d64T : VST1D3<{1,1,0,?}, "64">;
967 def VST1d8T_UPD : VST1D3WB<{0,0,0,?}, "8">;
968 def VST1d16T_UPD : VST1D3WB<{0,1,0,?}, "16">;
969 def VST1d32T_UPD : VST1D3WB<{1,0,0,?}, "32">;
970 def VST1d64T_UPD : VST1D3WB<{1,1,0,?}, "64">;
972 def VST1d64TPseudo : VSTQQPseudo<IIC_VST1x3>;
973 def VST1d64TPseudo_UPD : VSTQQWBPseudo<IIC_VST1x3u>;
975 // ...with 4 registers (some of these are only for the disassembler):
976 class VST1D4<bits<4> op7_4, string Dt>
977 : NLdSt<0, 0b00, 0b0010, op7_4, (outs),
978 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
979 IIC_VST1x4, "vst1", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn", "",
982 let Inst{5-4} = Rn{5-4};
984 class VST1D4WB<bits<4> op7_4, string Dt>
985 : NLdSt<0, 0b00, 0b0010, op7_4, (outs GPR:$wb),
986 (ins addrmode6:$Rn, am6offset:$Rm,
987 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST1x4u,
988 "vst1", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
989 "$Rn.addr = $wb", []> {
990 let Inst{5-4} = Rn{5-4};
993 def VST1d8Q : VST1D4<{0,0,?,?}, "8">;
994 def VST1d16Q : VST1D4<{0,1,?,?}, "16">;
995 def VST1d32Q : VST1D4<{1,0,?,?}, "32">;
996 def VST1d64Q : VST1D4<{1,1,?,?}, "64">;
998 def VST1d8Q_UPD : VST1D4WB<{0,0,?,?}, "8">;
999 def VST1d16Q_UPD : VST1D4WB<{0,1,?,?}, "16">;
1000 def VST1d32Q_UPD : VST1D4WB<{1,0,?,?}, "32">;
1001 def VST1d64Q_UPD : VST1D4WB<{1,1,?,?}, "64">;
1003 def VST1d64QPseudo : VSTQQPseudo<IIC_VST1x4>;
1004 def VST1d64QPseudo_UPD : VSTQQWBPseudo<IIC_VST1x4u>;
1006 // VST2 : Vector Store (multiple 2-element structures)
1007 class VST2D<bits<4> op11_8, bits<4> op7_4, string Dt>
1008 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1009 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2),
1010 IIC_VST2, "vst2", Dt, "\\{$Vd, $src2\\}, $Rn", "", []> {
1012 let Inst{5-4} = Rn{5-4};
1014 class VST2Q<bits<4> op7_4, string Dt>
1015 : NLdSt<0, 0b00, 0b0011, op7_4, (outs),
1016 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
1017 IIC_VST2x2, "vst2", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn",
1020 let Inst{5-4} = Rn{5-4};
1023 def VST2d8 : VST2D<0b1000, {0,0,?,?}, "8">;
1024 def VST2d16 : VST2D<0b1000, {0,1,?,?}, "16">;
1025 def VST2d32 : VST2D<0b1000, {1,0,?,?}, "32">;
1027 def VST2q8 : VST2Q<{0,0,?,?}, "8">;
1028 def VST2q16 : VST2Q<{0,1,?,?}, "16">;
1029 def VST2q32 : VST2Q<{1,0,?,?}, "32">;
1031 def VST2d8Pseudo : VSTQPseudo<IIC_VST2>;
1032 def VST2d16Pseudo : VSTQPseudo<IIC_VST2>;
1033 def VST2d32Pseudo : VSTQPseudo<IIC_VST2>;
1035 def VST2q8Pseudo : VSTQQPseudo<IIC_VST2x2>;
1036 def VST2q16Pseudo : VSTQQPseudo<IIC_VST2x2>;
1037 def VST2q32Pseudo : VSTQQPseudo<IIC_VST2x2>;
1039 // ...with address register writeback:
1040 class VST2DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1041 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1042 (ins addrmode6:$Rn, am6offset:$Rm, DPR:$Vd, DPR:$src2),
1043 IIC_VST2u, "vst2", Dt, "\\{$Vd, $src2\\}, $Rn$Rm",
1044 "$Rn.addr = $wb", []> {
1045 let Inst{5-4} = Rn{5-4};
1047 class VST2QWB<bits<4> op7_4, string Dt>
1048 : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
1049 (ins addrmode6:$Rn, am6offset:$Rm,
1050 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST2x2u,
1051 "vst2", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
1052 "$Rn.addr = $wb", []> {
1053 let Inst{5-4} = Rn{5-4};
1056 def VST2d8_UPD : VST2DWB<0b1000, {0,0,?,?}, "8">;
1057 def VST2d16_UPD : VST2DWB<0b1000, {0,1,?,?}, "16">;
1058 def VST2d32_UPD : VST2DWB<0b1000, {1,0,?,?}, "32">;
1060 def VST2q8_UPD : VST2QWB<{0,0,?,?}, "8">;
1061 def VST2q16_UPD : VST2QWB<{0,1,?,?}, "16">;
1062 def VST2q32_UPD : VST2QWB<{1,0,?,?}, "32">;
1064 def VST2d8Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
1065 def VST2d16Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
1066 def VST2d32Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
1068 def VST2q8Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
1069 def VST2q16Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
1070 def VST2q32Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
1072 // ...with double-spaced registers (for disassembly only):
1073 def VST2b8 : VST2D<0b1001, {0,0,?,?}, "8">;
1074 def VST2b16 : VST2D<0b1001, {0,1,?,?}, "16">;
1075 def VST2b32 : VST2D<0b1001, {1,0,?,?}, "32">;
1076 def VST2b8_UPD : VST2DWB<0b1001, {0,0,?,?}, "8">;
1077 def VST2b16_UPD : VST2DWB<0b1001, {0,1,?,?}, "16">;
1078 def VST2b32_UPD : VST2DWB<0b1001, {1,0,?,?}, "32">;
1080 // VST3 : Vector Store (multiple 3-element structures)
1081 class VST3D<bits<4> op11_8, bits<4> op7_4, string Dt>
1082 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1083 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3,
1084 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn", "", []> {
1086 let Inst{4} = Rn{4};
1089 def VST3d8 : VST3D<0b0100, {0,0,0,?}, "8">;
1090 def VST3d16 : VST3D<0b0100, {0,1,0,?}, "16">;
1091 def VST3d32 : VST3D<0b0100, {1,0,0,?}, "32">;
1093 def VST3d8Pseudo : VSTQQPseudo<IIC_VST3>;
1094 def VST3d16Pseudo : VSTQQPseudo<IIC_VST3>;
1095 def VST3d32Pseudo : VSTQQPseudo<IIC_VST3>;
1097 // ...with address register writeback:
1098 class VST3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1099 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1100 (ins addrmode6:$Rn, am6offset:$Rm,
1101 DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3u,
1102 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn$Rm",
1103 "$Rn.addr = $wb", []> {
1104 let Inst{4} = Rn{4};
1107 def VST3d8_UPD : VST3DWB<0b0100, {0,0,0,?}, "8">;
1108 def VST3d16_UPD : VST3DWB<0b0100, {0,1,0,?}, "16">;
1109 def VST3d32_UPD : VST3DWB<0b0100, {1,0,0,?}, "32">;
1111 def VST3d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1112 def VST3d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1113 def VST3d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1115 // ...with double-spaced registers (non-updating versions for disassembly only):
1116 def VST3q8 : VST3D<0b0101, {0,0,0,?}, "8">;
1117 def VST3q16 : VST3D<0b0101, {0,1,0,?}, "16">;
1118 def VST3q32 : VST3D<0b0101, {1,0,0,?}, "32">;
1119 def VST3q8_UPD : VST3DWB<0b0101, {0,0,0,?}, "8">;
1120 def VST3q16_UPD : VST3DWB<0b0101, {0,1,0,?}, "16">;
1121 def VST3q32_UPD : VST3DWB<0b0101, {1,0,0,?}, "32">;
1123 def VST3q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1124 def VST3q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1125 def VST3q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1127 // ...alternate versions to be allocated odd register numbers:
1128 def VST3q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1129 def VST3q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1130 def VST3q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1132 // VST4 : Vector Store (multiple 4-element structures)
1133 class VST4D<bits<4> op11_8, bits<4> op7_4, string Dt>
1134 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1135 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
1136 IIC_VST4, "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn",
1139 let Inst{5-4} = Rn{5-4};
1142 def VST4d8 : VST4D<0b0000, {0,0,?,?}, "8">;
1143 def VST4d16 : VST4D<0b0000, {0,1,?,?}, "16">;
1144 def VST4d32 : VST4D<0b0000, {1,0,?,?}, "32">;
1146 def VST4d8Pseudo : VSTQQPseudo<IIC_VST4>;
1147 def VST4d16Pseudo : VSTQQPseudo<IIC_VST4>;
1148 def VST4d32Pseudo : VSTQQPseudo<IIC_VST4>;
1150 // ...with address register writeback:
1151 class VST4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1152 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1153 (ins addrmode6:$Rn, am6offset:$Rm,
1154 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST4u,
1155 "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
1156 "$Rn.addr = $wb", []> {
1157 let Inst{5-4} = Rn{5-4};
1160 def VST4d8_UPD : VST4DWB<0b0000, {0,0,?,?}, "8">;
1161 def VST4d16_UPD : VST4DWB<0b0000, {0,1,?,?}, "16">;
1162 def VST4d32_UPD : VST4DWB<0b0000, {1,0,?,?}, "32">;
1164 def VST4d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1165 def VST4d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1166 def VST4d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1168 // ...with double-spaced registers (non-updating versions for disassembly only):
1169 def VST4q8 : VST4D<0b0001, {0,0,?,?}, "8">;
1170 def VST4q16 : VST4D<0b0001, {0,1,?,?}, "16">;
1171 def VST4q32 : VST4D<0b0001, {1,0,?,?}, "32">;
1172 def VST4q8_UPD : VST4DWB<0b0001, {0,0,?,?}, "8">;
1173 def VST4q16_UPD : VST4DWB<0b0001, {0,1,?,?}, "16">;
1174 def VST4q32_UPD : VST4DWB<0b0001, {1,0,?,?}, "32">;
1176 def VST4q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1177 def VST4q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1178 def VST4q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1180 // ...alternate versions to be allocated odd register numbers:
1181 def VST4q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1182 def VST4q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1183 def VST4q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1185 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
1187 // Classes for VST*LN pseudo-instructions with multi-register operands.
1188 // These are expanded to real instructions after register allocation.
1189 class VSTQLNPseudo<InstrItinClass itin>
1190 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
1192 class VSTQLNWBPseudo<InstrItinClass itin>
1193 : PseudoNLdSt<(outs GPR:$wb),
1194 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
1195 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1196 class VSTQQLNPseudo<InstrItinClass itin>
1197 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
1199 class VSTQQLNWBPseudo<InstrItinClass itin>
1200 : PseudoNLdSt<(outs GPR:$wb),
1201 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
1202 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1203 class VSTQQQQLNPseudo<InstrItinClass itin>
1204 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
1206 class VSTQQQQLNWBPseudo<InstrItinClass itin>
1207 : PseudoNLdSt<(outs GPR:$wb),
1208 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
1209 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1211 // VST1LN : Vector Store (single element from one lane)
1212 class VST1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1213 PatFrag StoreOp, SDNode ExtractOp>
1214 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1215 (ins addrmode6:$Rn, DPR:$Vd, nohash_imm:$lane),
1216 IIC_VST1ln, "vst1", Dt, "\\{$Vd[$lane]\\}, $Rn", "",
1217 [(StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane), addrmode6:$Rn)]> {
1220 class VST1QLNPseudo<ValueType Ty, PatFrag StoreOp, SDNode ExtractOp>
1221 : VSTQLNPseudo<IIC_VST1ln> {
1222 let Pattern = [(StoreOp (ExtractOp (Ty QPR:$src), imm:$lane),
1226 def VST1LNd8 : VST1LN<0b0000, {?,?,?,0}, "8", v8i8, truncstorei8,
1228 let Inst{7-5} = lane{2-0};
1230 def VST1LNd16 : VST1LN<0b0100, {?,?,0,?}, "16", v4i16, truncstorei16,
1232 let Inst{7-6} = lane{1-0};
1233 let Inst{4} = Rn{5};
1235 def VST1LNd32 : VST1LN<0b1000, {?,0,?,?}, "32", v2i32, store, extractelt> {
1236 let Inst{7} = lane{0};
1237 let Inst{5-4} = Rn{5-4};
1240 def VST1LNq8Pseudo : VST1QLNPseudo<v16i8, truncstorei8, NEONvgetlaneu>;
1241 def VST1LNq16Pseudo : VST1QLNPseudo<v8i16, truncstorei16, NEONvgetlaneu>;
1242 def VST1LNq32Pseudo : VST1QLNPseudo<v4i32, store, extractelt>;
1244 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
1246 // ...with address register writeback:
1247 class VST1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1248 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1249 (ins addrmode6:$Rn, am6offset:$Rm,
1250 DPR:$Vd, nohash_imm:$lane), IIC_VST1lnu, "vst1", Dt,
1251 "\\{$Vd[$lane]\\}, $Rn$Rm",
1252 "$Rn.addr = $wb", []>;
1254 def VST1LNd8_UPD : VST1LNWB<0b0000, {?,?,?,0}, "8"> {
1255 let Inst{7-5} = lane{2-0};
1257 def VST1LNd16_UPD : VST1LNWB<0b0100, {?,?,0,?}, "16"> {
1258 let Inst{7-6} = lane{1-0};
1259 let Inst{4} = Rn{5};
1261 def VST1LNd32_UPD : VST1LNWB<0b1000, {?,0,?,?}, "32"> {
1262 let Inst{7} = lane{0};
1263 let Inst{5-4} = Rn{5-4};
1266 def VST1LNq8Pseudo_UPD : VSTQLNWBPseudo<IIC_VST1lnu>;
1267 def VST1LNq16Pseudo_UPD : VSTQLNWBPseudo<IIC_VST1lnu>;
1268 def VST1LNq32Pseudo_UPD : VSTQLNWBPseudo<IIC_VST1lnu>;
1270 // VST2LN : Vector Store (single 2-element structure from one lane)
1271 class VST2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1272 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1273 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, nohash_imm:$lane),
1274 IIC_VST2ln, "vst2", Dt, "\\{$Vd[$lane], $src2[$lane]\\}, $Rn",
1277 let Inst{4} = Rn{4};
1280 def VST2LNd8 : VST2LN<0b0001, {?,?,?,?}, "8"> {
1281 let Inst{7-5} = lane{2-0};
1283 def VST2LNd16 : VST2LN<0b0101, {?,?,0,?}, "16"> {
1284 let Inst{7-6} = lane{1-0};
1286 def VST2LNd32 : VST2LN<0b1001, {?,0,0,?}, "32"> {
1287 let Inst{7} = lane{0};
1290 def VST2LNd8Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1291 def VST2LNd16Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1292 def VST2LNd32Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1294 // ...with double-spaced registers:
1295 def VST2LNq16 : VST2LN<0b0101, {?,?,1,?}, "16"> {
1296 let Inst{7-6} = lane{1-0};
1297 let Inst{4} = Rn{4};
1299 def VST2LNq32 : VST2LN<0b1001, {?,1,0,?}, "32"> {
1300 let Inst{7} = lane{0};
1301 let Inst{4} = Rn{4};
1304 def VST2LNq16Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
1305 def VST2LNq32Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
1307 // ...with address register writeback:
1308 class VST2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1309 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1310 (ins addrmode6:$addr, am6offset:$offset,
1311 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VST2lnu, "vst2", Dt,
1312 "\\{$src1[$lane], $src2[$lane]\\}, $addr$offset",
1313 "$addr.addr = $wb", []> {
1314 let Inst{4} = Rn{4};
1317 def VST2LNd8_UPD : VST2LNWB<0b0001, {?,?,?,?}, "8"> {
1318 let Inst{7-5} = lane{2-0};
1320 def VST2LNd16_UPD : VST2LNWB<0b0101, {?,?,0,?}, "16"> {
1321 let Inst{7-6} = lane{1-0};
1323 def VST2LNd32_UPD : VST2LNWB<0b1001, {?,0,0,?}, "32"> {
1324 let Inst{7} = lane{0};
1327 def VST2LNd8Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1328 def VST2LNd16Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1329 def VST2LNd32Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1331 def VST2LNq16_UPD : VST2LNWB<0b0101, {?,?,1,?}, "16"> {
1332 let Inst{7-6} = lane{1-0};
1334 def VST2LNq32_UPD : VST2LNWB<0b1001, {?,1,0,?}, "32"> {
1335 let Inst{7} = lane{0};
1338 def VST2LNq16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
1339 def VST2LNq32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
1341 // VST3LN : Vector Store (single 3-element structure from one lane)
1342 class VST3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1343 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1344 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3,
1345 nohash_imm:$lane), IIC_VST3ln, "vst3", Dt,
1346 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn", "", []> {
1350 def VST3LNd8 : VST3LN<0b0010, {?,?,?,0}, "8"> {
1351 let Inst{7-5} = lane{2-0};
1353 def VST3LNd16 : VST3LN<0b0110, {?,?,0,0}, "16"> {
1354 let Inst{7-6} = lane{1-0};
1356 def VST3LNd32 : VST3LN<0b1010, {?,0,0,0}, "32"> {
1357 let Inst{7} = lane{0};
1360 def VST3LNd8Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1361 def VST3LNd16Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1362 def VST3LNd32Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1364 // ...with double-spaced registers:
1365 def VST3LNq16 : VST3LN<0b0110, {?,?,1,0}, "16"> {
1366 let Inst{7-6} = lane{1-0};
1368 def VST3LNq32 : VST3LN<0b1010, {?,1,0,0}, "32"> {
1369 let Inst{7} = lane{0};
1372 def VST3LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
1373 def VST3LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
1375 // ...with address register writeback:
1376 class VST3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1377 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1378 (ins addrmode6:$Rn, am6offset:$Rm,
1379 DPR:$Vd, DPR:$src2, DPR:$src3, nohash_imm:$lane),
1380 IIC_VST3lnu, "vst3", Dt,
1381 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn$Rm",
1382 "$Rn.addr = $wb", []>;
1384 def VST3LNd8_UPD : VST3LNWB<0b0010, {?,?,?,0}, "8"> {
1385 let Inst{7-5} = lane{2-0};
1387 def VST3LNd16_UPD : VST3LNWB<0b0110, {?,?,0,0}, "16"> {
1388 let Inst{7-6} = lane{1-0};
1390 def VST3LNd32_UPD : VST3LNWB<0b1010, {?,0,0,0}, "32"> {
1391 let Inst{7} = lane{0};
1394 def VST3LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1395 def VST3LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1396 def VST3LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1398 def VST3LNq16_UPD : VST3LNWB<0b0110, {?,?,1,0}, "16"> {
1399 let Inst{7-6} = lane{1-0};
1401 def VST3LNq32_UPD : VST3LNWB<0b1010, {?,1,0,0}, "32"> {
1402 let Inst{7} = lane{0};
1405 def VST3LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
1406 def VST3LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
1408 // VST4LN : Vector Store (single 4-element structure from one lane)
1409 class VST4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1410 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1411 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4,
1412 nohash_imm:$lane), IIC_VST4ln, "vst4", Dt,
1413 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn",
1416 let Inst{4} = Rn{4};
1419 def VST4LNd8 : VST4LN<0b0011, {?,?,?,?}, "8"> {
1420 let Inst{7-5} = lane{2-0};
1422 def VST4LNd16 : VST4LN<0b0111, {?,?,0,?}, "16"> {
1423 let Inst{7-6} = lane{1-0};
1425 def VST4LNd32 : VST4LN<0b1011, {?,0,?,?}, "32"> {
1426 let Inst{7} = lane{0};
1427 let Inst{5} = Rn{5};
1430 def VST4LNd8Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1431 def VST4LNd16Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1432 def VST4LNd32Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1434 // ...with double-spaced registers:
1435 def VST4LNq16 : VST4LN<0b0111, {?,?,1,?}, "16"> {
1436 let Inst{7-6} = lane{1-0};
1438 def VST4LNq32 : VST4LN<0b1011, {?,1,?,?}, "32"> {
1439 let Inst{7} = lane{0};
1440 let Inst{5} = Rn{5};
1443 def VST4LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
1444 def VST4LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
1446 // ...with address register writeback:
1447 class VST4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1448 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1449 (ins addrmode6:$Rn, am6offset:$Rm,
1450 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
1451 IIC_VST4lnu, "vst4", Dt,
1452 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn$Rm",
1453 "$Rn.addr = $wb", []> {
1454 let Inst{4} = Rn{4};
1457 def VST4LNd8_UPD : VST4LNWB<0b0011, {?,?,?,?}, "8"> {
1458 let Inst{7-5} = lane{2-0};
1460 def VST4LNd16_UPD : VST4LNWB<0b0111, {?,?,0,?}, "16"> {
1461 let Inst{7-6} = lane{1-0};
1463 def VST4LNd32_UPD : VST4LNWB<0b1011, {?,0,?,?}, "32"> {
1464 let Inst{7} = lane{0};
1465 let Inst{5} = Rn{5};
1468 def VST4LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
1469 def VST4LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
1470 def VST4LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
1472 def VST4LNq16_UPD : VST4LNWB<0b0111, {?,?,1,?}, "16"> {
1473 let Inst{7-6} = lane{1-0};
1475 def VST4LNq32_UPD : VST4LNWB<0b1011, {?,1,?,?}, "32"> {
1476 let Inst{7} = lane{0};
1477 let Inst{5} = Rn{5};
1480 def VST4LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
1481 def VST4LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
1483 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
1486 //===----------------------------------------------------------------------===//
1487 // NEON pattern fragments
1488 //===----------------------------------------------------------------------===//
1490 // Extract D sub-registers of Q registers.
1491 def DSubReg_i8_reg : SDNodeXForm<imm, [{
1492 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1493 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/8, MVT::i32);
1495 def DSubReg_i16_reg : SDNodeXForm<imm, [{
1496 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1497 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/4, MVT::i32);
1499 def DSubReg_i32_reg : SDNodeXForm<imm, [{
1500 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1501 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/2, MVT::i32);
1503 def DSubReg_f64_reg : SDNodeXForm<imm, [{
1504 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1505 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue(), MVT::i32);
1508 // Extract S sub-registers of Q/D registers.
1509 def SSubReg_f32_reg : SDNodeXForm<imm, [{
1510 assert(ARM::ssub_3 == ARM::ssub_0+3 && "Unexpected subreg numbering");
1511 return CurDAG->getTargetConstant(ARM::ssub_0 + N->getZExtValue(), MVT::i32);
1514 // Translate lane numbers from Q registers to D subregs.
1515 def SubReg_i8_lane : SDNodeXForm<imm, [{
1516 return CurDAG->getTargetConstant(N->getZExtValue() & 7, MVT::i32);
1518 def SubReg_i16_lane : SDNodeXForm<imm, [{
1519 return CurDAG->getTargetConstant(N->getZExtValue() & 3, MVT::i32);
1521 def SubReg_i32_lane : SDNodeXForm<imm, [{
1522 return CurDAG->getTargetConstant(N->getZExtValue() & 1, MVT::i32);
1525 //===----------------------------------------------------------------------===//
1526 // Instruction Classes
1527 //===----------------------------------------------------------------------===//
1529 // Basic 2-register operations: single-, double- and quad-register.
1530 class N2VS<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1531 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
1532 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
1533 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
1534 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src),
1535 IIC_VUNAD, OpcodeStr, Dt, "$dst, $src", "", []>;
1536 class N2VD<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1537 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
1538 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
1539 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
1540 (ins DPR:$src), IIC_VUNAD, OpcodeStr, Dt,"$dst, $src", "",
1541 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src))))]>;
1542 class N2VQ<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1543 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
1544 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
1545 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
1546 (ins QPR:$src), IIC_VUNAQ, OpcodeStr, Dt,"$dst, $src", "",
1547 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src))))]>;
1549 // Basic 2-register intrinsics, both double- and quad-register.
1550 class N2VDInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1551 bits<2> op17_16, bits<5> op11_7, bit op4,
1552 InstrItinClass itin, string OpcodeStr, string Dt,
1553 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1554 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
1555 (ins DPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
1556 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
1557 class N2VQInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1558 bits<2> op17_16, bits<5> op11_7, bit op4,
1559 InstrItinClass itin, string OpcodeStr, string Dt,
1560 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1561 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
1562 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
1563 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
1565 // Narrow 2-register operations.
1566 class N2VN<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1567 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1568 InstrItinClass itin, string OpcodeStr, string Dt,
1569 ValueType TyD, ValueType TyQ, SDNode OpNode>
1570 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$dst),
1571 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
1572 [(set DPR:$dst, (TyD (OpNode (TyQ QPR:$src))))]>;
1574 // Narrow 2-register intrinsics.
1575 class N2VNInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1576 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1577 InstrItinClass itin, string OpcodeStr, string Dt,
1578 ValueType TyD, ValueType TyQ, Intrinsic IntOp>
1579 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$dst),
1580 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
1581 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src))))]>;
1583 // Long 2-register operations (currently only used for VMOVL).
1584 class N2VL<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1585 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1586 InstrItinClass itin, string OpcodeStr, string Dt,
1587 ValueType TyQ, ValueType TyD, SDNode OpNode>
1588 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$dst),
1589 (ins DPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
1590 [(set QPR:$dst, (TyQ (OpNode (TyD DPR:$src))))]>;
1592 // 2-register shuffles (VTRN/VZIP/VUZP), both double- and quad-register.
1593 class N2VDShuffle<bits<2> op19_18, bits<5> op11_7, string OpcodeStr, string Dt>
1594 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 0, 0, (outs DPR:$dst1, DPR:$dst2),
1595 (ins DPR:$src1, DPR:$src2), IIC_VPERMD,
1596 OpcodeStr, Dt, "$dst1, $dst2",
1597 "$src1 = $dst1, $src2 = $dst2", []>;
1598 class N2VQShuffle<bits<2> op19_18, bits<5> op11_7,
1599 InstrItinClass itin, string OpcodeStr, string Dt>
1600 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 1, 0, (outs QPR:$dst1, QPR:$dst2),
1601 (ins QPR:$src1, QPR:$src2), itin, OpcodeStr, Dt, "$dst1, $dst2",
1602 "$src1 = $dst1, $src2 = $dst2", []>;
1604 // Basic 3-register operations: single-, double- and quad-register.
1605 class N3VS<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1606 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1607 SDNode OpNode, bit Commutable>
1608 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1609 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src1, DPR_VFP2:$src2), N3RegFrm,
1610 IIC_VBIND, OpcodeStr, Dt, "$dst, $src1, $src2", "", []> {
1611 let isCommutable = Commutable;
1614 class N3VD<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1615 InstrItinClass itin, string OpcodeStr, string Dt,
1616 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
1617 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1618 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1619 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
1620 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
1621 let isCommutable = Commutable;
1623 // Same as N3VD but no data type.
1624 class N3VDX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1625 InstrItinClass itin, string OpcodeStr,
1626 ValueType ResTy, ValueType OpTy,
1627 SDNode OpNode, bit Commutable>
1628 : N3VX<op24, op23, op21_20, op11_8, 0, op4,
1629 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1630 OpcodeStr, "$Vd, $Vn, $Vm", "",
1631 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>{
1632 let isCommutable = Commutable;
1635 class N3VDSL<bits<2> op21_20, bits<4> op11_8,
1636 InstrItinClass itin, string OpcodeStr, string Dt,
1637 ValueType Ty, SDNode ShOp>
1638 : N3V<0, 1, op21_20, op11_8, 1, 0,
1639 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1640 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1641 [(set (Ty DPR:$dst),
1642 (Ty (ShOp (Ty DPR:$src1),
1643 (Ty (NEONvduplane (Ty DPR_VFP2:$src2),imm:$lane)))))]> {
1644 let isCommutable = 0;
1646 class N3VDSL16<bits<2> op21_20, bits<4> op11_8,
1647 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1648 : N3V<0, 1, op21_20, op11_8, 1, 0,
1649 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1650 NVMulSLFrm, IIC_VMULi16D, OpcodeStr, Dt,"$dst, $src1, $src2[$lane]","",
1651 [(set (Ty DPR:$dst),
1652 (Ty (ShOp (Ty DPR:$src1),
1653 (Ty (NEONvduplane (Ty DPR_8:$src2), imm:$lane)))))]> {
1654 let isCommutable = 0;
1657 class N3VQ<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1658 InstrItinClass itin, string OpcodeStr, string Dt,
1659 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
1660 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1661 (outs QPR:$Qd), (ins QPR:$Qn, QPR:$Qm), N3RegFrm, itin,
1662 OpcodeStr, Dt, "$Qd, $Qn, $Qm", "",
1663 [(set QPR:$Qd, (ResTy (OpNode (OpTy QPR:$Qn), (OpTy QPR:$Qm))))]> {
1664 let isCommutable = Commutable;
1666 class N3VQX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1667 InstrItinClass itin, string OpcodeStr,
1668 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
1669 : N3VX<op24, op23, op21_20, op11_8, 1, op4,
1670 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), N3RegFrm, itin,
1671 OpcodeStr, "$dst, $src1, $src2", "",
1672 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src1), (OpTy QPR:$src2))))]>{
1673 let isCommutable = Commutable;
1675 class N3VQSL<bits<2> op21_20, bits<4> op11_8,
1676 InstrItinClass itin, string OpcodeStr, string Dt,
1677 ValueType ResTy, ValueType OpTy, SDNode ShOp>
1678 : N3V<1, 1, op21_20, op11_8, 1, 0,
1679 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1680 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1681 [(set (ResTy QPR:$dst),
1682 (ResTy (ShOp (ResTy QPR:$src1),
1683 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1685 let isCommutable = 0;
1687 class N3VQSL16<bits<2> op21_20, bits<4> op11_8, string OpcodeStr, string Dt,
1688 ValueType ResTy, ValueType OpTy, SDNode ShOp>
1689 : N3V<1, 1, op21_20, op11_8, 1, 0,
1690 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1691 NVMulSLFrm, IIC_VMULi16Q, OpcodeStr, Dt,"$dst, $src1, $src2[$lane]","",
1692 [(set (ResTy QPR:$dst),
1693 (ResTy (ShOp (ResTy QPR:$src1),
1694 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
1696 let isCommutable = 0;
1699 // Basic 3-register intrinsics, both double- and quad-register.
1700 class N3VDInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1701 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1702 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
1703 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1704 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), f, itin,
1705 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
1706 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
1707 let isCommutable = Commutable;
1709 class N3VDIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1710 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
1711 : N3V<0, 1, op21_20, op11_8, 1, 0,
1712 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1713 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1714 [(set (Ty DPR:$dst),
1715 (Ty (IntOp (Ty DPR:$src1),
1716 (Ty (NEONvduplane (Ty DPR_VFP2:$src2),
1718 let isCommutable = 0;
1720 class N3VDIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1721 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
1722 : N3V<0, 1, op21_20, op11_8, 1, 0,
1723 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1724 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1725 [(set (Ty DPR:$dst),
1726 (Ty (IntOp (Ty DPR:$src1),
1727 (Ty (NEONvduplane (Ty DPR_8:$src2), imm:$lane)))))]> {
1728 let isCommutable = 0;
1730 class N3VDIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1731 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1732 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1733 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1734 (outs DPR:$Vd), (ins DPR:$Vm, DPR:$Vn), f, itin,
1735 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
1736 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (OpTy DPR:$Vn))))]> {
1737 let isCommutable = 0;
1740 class N3VQInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1741 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1742 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
1743 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1744 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), f, itin,
1745 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
1746 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
1747 let isCommutable = Commutable;
1749 class N3VQIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1750 string OpcodeStr, string Dt,
1751 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1752 : N3V<1, 1, op21_20, op11_8, 1, 0,
1753 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1754 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1755 [(set (ResTy QPR:$dst),
1756 (ResTy (IntOp (ResTy QPR:$src1),
1757 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1759 let isCommutable = 0;
1761 class N3VQIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1762 string OpcodeStr, string Dt,
1763 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1764 : N3V<1, 1, op21_20, op11_8, 1, 0,
1765 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1766 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1767 [(set (ResTy QPR:$dst),
1768 (ResTy (IntOp (ResTy QPR:$src1),
1769 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
1771 let isCommutable = 0;
1773 class N3VQIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1774 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1775 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1776 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1777 (outs QPR:$Vd), (ins QPR:$Vm, QPR:$Vn), f, itin,
1778 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
1779 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (OpTy QPR:$Vn))))]> {
1780 let isCommutable = 0;
1783 // Multiply-Add/Sub operations: single-, double- and quad-register.
1784 class N3VSMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1785 InstrItinClass itin, string OpcodeStr, string Dt,
1786 ValueType Ty, SDNode MulOp, SDNode OpNode>
1787 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1788 (outs DPR_VFP2:$dst),
1789 (ins DPR_VFP2:$src1, DPR_VFP2:$src2, DPR_VFP2:$src3), N3RegFrm, itin,
1790 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst", []>;
1792 class N3VDMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1793 InstrItinClass itin, string OpcodeStr, string Dt,
1794 ValueType Ty, SDNode MulOp, SDNode OpNode>
1795 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1796 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1797 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1798 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
1799 (Ty (MulOp DPR:$Vn, DPR:$Vm)))))]>;
1801 class N3VDMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1802 string OpcodeStr, string Dt,
1803 ValueType Ty, SDNode MulOp, SDNode ShOp>
1804 : N3V<0, 1, op21_20, op11_8, 1, 0,
1806 (ins DPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1808 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1809 [(set (Ty DPR:$dst),
1810 (Ty (ShOp (Ty DPR:$src1),
1811 (Ty (MulOp DPR:$src2,
1812 (Ty (NEONvduplane (Ty DPR_VFP2:$src3),
1814 class N3VDMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1815 string OpcodeStr, string Dt,
1816 ValueType Ty, SDNode MulOp, SDNode ShOp>
1817 : N3V<0, 1, op21_20, op11_8, 1, 0,
1819 (ins DPR:$src1, DPR:$Vn, DPR_8:$Vm, nohash_imm:$lane),
1821 OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "$src1 = $Vd",
1823 (Ty (ShOp (Ty DPR:$src1),
1825 (Ty (NEONvduplane (Ty DPR_8:$Vm),
1828 class N3VQMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1829 InstrItinClass itin, string OpcodeStr, string Dt, ValueType Ty,
1830 SDNode MulOp, SDNode OpNode>
1831 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1832 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
1833 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1834 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
1835 (Ty (MulOp QPR:$Vn, QPR:$Vm)))))]>;
1836 class N3VQMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1837 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1838 SDNode MulOp, SDNode ShOp>
1839 : N3V<1, 1, op21_20, op11_8, 1, 0,
1841 (ins QPR:$src1, QPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1843 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1844 [(set (ResTy QPR:$dst),
1845 (ResTy (ShOp (ResTy QPR:$src1),
1846 (ResTy (MulOp QPR:$src2,
1847 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src3),
1849 class N3VQMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1850 string OpcodeStr, string Dt,
1851 ValueType ResTy, ValueType OpTy,
1852 SDNode MulOp, SDNode ShOp>
1853 : N3V<1, 1, op21_20, op11_8, 1, 0,
1855 (ins QPR:$src1, QPR:$src2, DPR_8:$src3, nohash_imm:$lane),
1857 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1858 [(set (ResTy QPR:$dst),
1859 (ResTy (ShOp (ResTy QPR:$src1),
1860 (ResTy (MulOp QPR:$src2,
1861 (ResTy (NEONvduplane (OpTy DPR_8:$src3),
1864 // Neon Intrinsic-Op instructions (VABA): double- and quad-register.
1865 class N3VDIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1866 InstrItinClass itin, string OpcodeStr, string Dt,
1867 ValueType Ty, Intrinsic IntOp, SDNode OpNode>
1868 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1869 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1870 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1871 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
1872 (Ty (IntOp (Ty DPR:$Vn), (Ty DPR:$Vm))))))]>;
1873 class N3VQIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1874 InstrItinClass itin, string OpcodeStr, string Dt,
1875 ValueType Ty, Intrinsic IntOp, SDNode OpNode>
1876 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1877 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
1878 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1879 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
1880 (Ty (IntOp (Ty QPR:$Vn), (Ty QPR:$Vm))))))]>;
1882 // Neon 3-argument intrinsics, both double- and quad-register.
1883 // The destination register is also used as the first source operand register.
1884 class N3VDInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1885 InstrItinClass itin, string OpcodeStr, string Dt,
1886 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1887 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1888 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3), N3RegFrm, itin,
1889 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1890 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src1),
1891 (OpTy DPR:$src2), (OpTy DPR:$src3))))]>;
1892 class N3VQInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1893 InstrItinClass itin, string OpcodeStr, string Dt,
1894 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1895 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1896 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3), N3RegFrm, itin,
1897 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1898 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src1),
1899 (OpTy QPR:$src2), (OpTy QPR:$src3))))]>;
1901 // Long Multiply-Add/Sub operations.
1902 class N3VLMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1903 InstrItinClass itin, string OpcodeStr, string Dt,
1904 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
1905 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1906 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1907 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1908 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
1909 (TyQ (MulOp (TyD DPR:$Vn),
1910 (TyD DPR:$Vm)))))]>;
1911 class N3VLMulOpSL<bit op24, bits<2> op21_20, bits<4> op11_8,
1912 InstrItinClass itin, string OpcodeStr, string Dt,
1913 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
1914 : N3V<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$dst),
1915 (ins QPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1917 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1919 (OpNode (TyQ QPR:$src1),
1920 (TyQ (MulOp (TyD DPR:$src2),
1921 (TyD (NEONvduplane (TyD DPR_VFP2:$src3),
1923 class N3VLMulOpSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1924 InstrItinClass itin, string OpcodeStr, string Dt,
1925 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
1926 : N3V<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$dst),
1927 (ins QPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane),
1929 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1931 (OpNode (TyQ QPR:$src1),
1932 (TyQ (MulOp (TyD DPR:$src2),
1933 (TyD (NEONvduplane (TyD DPR_8:$src3),
1936 // Long Intrinsic-Op vector operations with explicit extend (VABAL).
1937 class N3VLIntExtOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1938 InstrItinClass itin, string OpcodeStr, string Dt,
1939 ValueType TyQ, ValueType TyD, Intrinsic IntOp, SDNode ExtOp,
1941 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1942 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1943 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1944 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
1945 (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
1946 (TyD DPR:$Vm)))))))]>;
1948 // Neon Long 3-argument intrinsic. The destination register is
1949 // a quad-register and is also used as the first source operand register.
1950 class N3VLInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1951 InstrItinClass itin, string OpcodeStr, string Dt,
1952 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
1953 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1954 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1955 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1957 (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$Vn), (TyD DPR:$Vm))))]>;
1958 class N3VLInt3SL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1959 string OpcodeStr, string Dt,
1960 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1961 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1963 (ins QPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1965 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1966 [(set (ResTy QPR:$dst),
1967 (ResTy (IntOp (ResTy QPR:$src1),
1969 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src3),
1971 class N3VLInt3SL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1972 InstrItinClass itin, string OpcodeStr, string Dt,
1973 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1974 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1976 (ins QPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane),
1978 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1979 [(set (ResTy QPR:$dst),
1980 (ResTy (IntOp (ResTy QPR:$src1),
1982 (OpTy (NEONvduplane (OpTy DPR_8:$src3),
1985 // Narrowing 3-register intrinsics.
1986 class N3VNInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1987 string OpcodeStr, string Dt, ValueType TyD, ValueType TyQ,
1988 Intrinsic IntOp, bit Commutable>
1989 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1990 (outs DPR:$dst), (ins QPR:$src1, QPR:$src2), N3RegFrm, IIC_VBINi4D,
1991 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1992 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src1), (TyQ QPR:$src2))))]> {
1993 let isCommutable = Commutable;
1996 // Long 3-register operations.
1997 class N3VL<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1998 InstrItinClass itin, string OpcodeStr, string Dt,
1999 ValueType TyQ, ValueType TyD, SDNode OpNode, bit Commutable>
2000 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2001 (outs QPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
2002 OpcodeStr, Dt, "$dst, $src1, $src2", "",
2003 [(set QPR:$dst, (TyQ (OpNode (TyD DPR:$src1), (TyD DPR:$src2))))]> {
2004 let isCommutable = Commutable;
2006 class N3VLSL<bit op24, bits<2> op21_20, bits<4> op11_8,
2007 InstrItinClass itin, string OpcodeStr, string Dt,
2008 ValueType TyQ, ValueType TyD, SDNode OpNode>
2009 : N3V<op24, 1, op21_20, op11_8, 1, 0,
2010 (outs QPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
2011 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
2013 (TyQ (OpNode (TyD DPR:$src1),
2014 (TyD (NEONvduplane (TyD DPR_VFP2:$src2),imm:$lane)))))]>;
2015 class N3VLSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2016 InstrItinClass itin, string OpcodeStr, string Dt,
2017 ValueType TyQ, ValueType TyD, SDNode OpNode>
2018 : N3V<op24, 1, op21_20, op11_8, 1, 0,
2019 (outs QPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
2020 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
2022 (TyQ (OpNode (TyD DPR:$src1),
2023 (TyD (NEONvduplane (TyD DPR_8:$src2), imm:$lane)))))]>;
2025 // Long 3-register operations with explicitly extended operands.
2026 class N3VLExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2027 InstrItinClass itin, string OpcodeStr, string Dt,
2028 ValueType TyQ, ValueType TyD, SDNode OpNode, SDNode ExtOp,
2030 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2031 (outs QPR:$Qd), (ins DPR:$Dn, DPR:$Dm), N3RegFrm, itin,
2032 OpcodeStr, Dt, "$Qd, $Dn, $Dm", "",
2033 [(set QPR:$Qd, (OpNode (TyQ (ExtOp (TyD DPR:$Dn))),
2034 (TyQ (ExtOp (TyD DPR:$Dm)))))]> {
2035 let isCommutable = Commutable;
2038 // Long 3-register intrinsics with explicit extend (VABDL).
2039 class N3VLIntExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2040 InstrItinClass itin, string OpcodeStr, string Dt,
2041 ValueType TyQ, ValueType TyD, Intrinsic IntOp, SDNode ExtOp,
2043 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2044 (outs QPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
2045 OpcodeStr, Dt, "$dst, $src1, $src2", "",
2046 [(set QPR:$dst, (TyQ (ExtOp (TyD (IntOp (TyD DPR:$src1),
2047 (TyD DPR:$src2))))))]> {
2048 let isCommutable = Commutable;
2051 // Long 3-register intrinsics.
2052 class N3VLInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2053 InstrItinClass itin, string OpcodeStr, string Dt,
2054 ValueType TyQ, ValueType TyD, Intrinsic IntOp, bit Commutable>
2055 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2056 (outs QPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
2057 OpcodeStr, Dt, "$dst, $src1, $src2", "",
2058 [(set QPR:$dst, (TyQ (IntOp (TyD DPR:$src1), (TyD DPR:$src2))))]> {
2059 let isCommutable = Commutable;
2061 class N3VLIntSL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2062 string OpcodeStr, string Dt,
2063 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2064 : N3V<op24, 1, op21_20, op11_8, 1, 0,
2065 (outs QPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
2066 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
2067 [(set (ResTy QPR:$dst),
2068 (ResTy (IntOp (OpTy DPR:$src1),
2069 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src2),
2071 class N3VLIntSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2072 InstrItinClass itin, string OpcodeStr, string Dt,
2073 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2074 : N3V<op24, 1, op21_20, op11_8, 1, 0,
2075 (outs QPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
2076 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
2077 [(set (ResTy QPR:$dst),
2078 (ResTy (IntOp (OpTy DPR:$src1),
2079 (OpTy (NEONvduplane (OpTy DPR_8:$src2),
2082 // Wide 3-register operations.
2083 class N3VW<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2084 string OpcodeStr, string Dt, ValueType TyQ, ValueType TyD,
2085 SDNode OpNode, SDNode ExtOp, bit Commutable>
2086 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2087 (outs QPR:$Qd), (ins QPR:$Qn, DPR:$Dm), N3RegFrm, IIC_VSUBiD,
2088 OpcodeStr, Dt, "$Qd, $Qn, $Dm", "",
2089 [(set QPR:$Qd, (OpNode (TyQ QPR:$Qn),
2090 (TyQ (ExtOp (TyD DPR:$Dm)))))]> {
2091 let isCommutable = Commutable;
2094 // Pairwise long 2-register intrinsics, both double- and quad-register.
2095 class N2VDPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2096 bits<2> op17_16, bits<5> op11_7, bit op4,
2097 string OpcodeStr, string Dt,
2098 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2099 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
2100 (ins DPR:$src), IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
2101 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
2102 class N2VQPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2103 bits<2> op17_16, bits<5> op11_7, bit op4,
2104 string OpcodeStr, string Dt,
2105 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2106 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
2107 (ins QPR:$src), IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
2108 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
2110 // Pairwise long 2-register accumulate intrinsics,
2111 // both double- and quad-register.
2112 // The destination register is also used as the first source operand register.
2113 class N2VDPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2114 bits<2> op17_16, bits<5> op11_7, bit op4,
2115 string OpcodeStr, string Dt,
2116 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2117 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
2118 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vm), IIC_VPALiD,
2119 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
2120 [(set DPR:$Vd, (ResTy (IntOp (ResTy DPR:$src1), (OpTy DPR:$Vm))))]>;
2121 class N2VQPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2122 bits<2> op17_16, bits<5> op11_7, bit op4,
2123 string OpcodeStr, string Dt,
2124 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2125 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4,
2126 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vm), IIC_VPALiQ,
2127 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
2128 [(set QPR:$Vd, (ResTy (IntOp (ResTy QPR:$src1), (OpTy QPR:$Vm))))]>;
2130 // Shift by immediate,
2131 // both double- and quad-register.
2132 class N2VDSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2133 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2134 ValueType Ty, SDNode OpNode>
2135 : N2VImm<op24, op23, op11_8, op7, 0, op4,
2136 (outs DPR:$dst), (ins DPR:$src, i32imm:$SIMM), f, itin,
2137 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
2138 [(set DPR:$dst, (Ty (OpNode (Ty DPR:$src), (i32 imm:$SIMM))))]>;
2139 class N2VQSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2140 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2141 ValueType Ty, SDNode OpNode>
2142 : N2VImm<op24, op23, op11_8, op7, 1, op4,
2143 (outs QPR:$dst), (ins QPR:$src, i32imm:$SIMM), f, itin,
2144 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
2145 [(set QPR:$dst, (Ty (OpNode (Ty QPR:$src), (i32 imm:$SIMM))))]>;
2147 // Long shift by immediate.
2148 class N2VLSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
2149 string OpcodeStr, string Dt,
2150 ValueType ResTy, ValueType OpTy, SDNode OpNode>
2151 : N2VImm<op24, op23, op11_8, op7, op6, op4,
2152 (outs QPR:$dst), (ins DPR:$src, i32imm:$SIMM), N2RegVShLFrm,
2153 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src, $SIMM", "",
2154 [(set QPR:$dst, (ResTy (OpNode (OpTy DPR:$src),
2155 (i32 imm:$SIMM))))]>;
2157 // Narrow shift by immediate.
2158 class N2VNSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
2159 InstrItinClass itin, string OpcodeStr, string Dt,
2160 ValueType ResTy, ValueType OpTy, SDNode OpNode>
2161 : N2VImm<op24, op23, op11_8, op7, op6, op4,
2162 (outs DPR:$dst), (ins QPR:$src, i32imm:$SIMM), N2RegVShRFrm, itin,
2163 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
2164 [(set DPR:$dst, (ResTy (OpNode (OpTy QPR:$src),
2165 (i32 imm:$SIMM))))]>;
2167 // Shift right by immediate and accumulate,
2168 // both double- and quad-register.
2169 class N2VDShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2170 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
2171 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
2172 (ins DPR:$src1, DPR:$Vm, i32imm:$SIMM), N2RegVShRFrm, IIC_VPALiD,
2173 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2174 [(set DPR:$Vd, (Ty (add DPR:$src1,
2175 (Ty (ShOp DPR:$Vm, (i32 imm:$SIMM))))))]>;
2176 class N2VQShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2177 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
2178 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
2179 (ins QPR:$src1, QPR:$Vm, i32imm:$SIMM), N2RegVShRFrm, IIC_VPALiD,
2180 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2181 [(set QPR:$Vd, (Ty (add QPR:$src1,
2182 (Ty (ShOp QPR:$Vm, (i32 imm:$SIMM))))))]>;
2184 // Shift by immediate and insert,
2185 // both double- and quad-register.
2186 class N2VDShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2187 Format f, string OpcodeStr, string Dt, ValueType Ty,SDNode ShOp>
2188 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
2189 (ins DPR:$src1, DPR:$Vm, i32imm:$SIMM), f, IIC_VSHLiD,
2190 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2191 [(set DPR:$Vd, (Ty (ShOp DPR:$src1, DPR:$Vm, (i32 imm:$SIMM))))]>;
2192 class N2VQShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2193 Format f, string OpcodeStr, string Dt, ValueType Ty,SDNode ShOp>
2194 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
2195 (ins QPR:$src1, QPR:$Vm, i32imm:$SIMM), f, IIC_VSHLiQ,
2196 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2197 [(set QPR:$Vd, (Ty (ShOp QPR:$src1, QPR:$Vm, (i32 imm:$SIMM))))]>;
2199 // Convert, with fractional bits immediate,
2200 // both double- and quad-register.
2201 class N2VCvtD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2202 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2204 : N2VImm<op24, op23, op11_8, op7, 0, op4,
2205 (outs DPR:$Vd), (ins DPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
2206 IIC_VUNAD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2207 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (i32 imm:$SIMM))))]>;
2208 class N2VCvtQ<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2209 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2211 : N2VImm<op24, op23, op11_8, op7, 1, op4,
2212 (outs QPR:$Vd), (ins QPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
2213 IIC_VUNAQ, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2214 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (i32 imm:$SIMM))))]>;
2216 //===----------------------------------------------------------------------===//
2218 //===----------------------------------------------------------------------===//
2220 // Abbreviations used in multiclass suffixes:
2221 // Q = quarter int (8 bit) elements
2222 // H = half int (16 bit) elements
2223 // S = single int (32 bit) elements
2224 // D = double int (64 bit) elements
2226 // Neon 2-register vector operations -- for disassembly only.
2228 // First with only element sizes of 8, 16 and 32 bits:
2229 multiclass N2V_QHS_cmp<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2230 bits<5> op11_7, bit op4, string opc, string Dt,
2231 string asm, SDNode OpNode> {
2232 // 64-bit vector types.
2233 def v8i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 0, op4,
2234 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
2235 opc, !strconcat(Dt, "8"), asm, "",
2236 [(set DPR:$dst, (v8i8 (OpNode (v8i8 DPR:$src))))]>;
2237 def v4i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 0, op4,
2238 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
2239 opc, !strconcat(Dt, "16"), asm, "",
2240 [(set DPR:$dst, (v4i16 (OpNode (v4i16 DPR:$src))))]>;
2241 def v2i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
2242 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
2243 opc, !strconcat(Dt, "32"), asm, "",
2244 [(set DPR:$dst, (v2i32 (OpNode (v2i32 DPR:$src))))]>;
2245 def v2f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
2246 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
2247 opc, "f32", asm, "",
2248 [(set DPR:$dst, (v2f32 (OpNode (v2f32 DPR:$src))))]> {
2249 let Inst{10} = 1; // overwrite F = 1
2252 // 128-bit vector types.
2253 def v16i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 1, op4,
2254 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
2255 opc, !strconcat(Dt, "8"), asm, "",
2256 [(set QPR:$dst, (v16i8 (OpNode (v16i8 QPR:$src))))]>;
2257 def v8i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 1, op4,
2258 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
2259 opc, !strconcat(Dt, "16"), asm, "",
2260 [(set QPR:$dst, (v8i16 (OpNode (v8i16 QPR:$src))))]>;
2261 def v4i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
2262 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
2263 opc, !strconcat(Dt, "32"), asm, "",
2264 [(set QPR:$dst, (v4i32 (OpNode (v4i32 QPR:$src))))]>;
2265 def v4f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
2266 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
2267 opc, "f32", asm, "",
2268 [(set QPR:$dst, (v4f32 (OpNode (v4f32 QPR:$src))))]> {
2269 let Inst{10} = 1; // overwrite F = 1
2273 // Neon 3-register vector operations.
2275 // First with only element sizes of 8, 16 and 32 bits:
2276 multiclass N3V_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2277 InstrItinClass itinD16, InstrItinClass itinD32,
2278 InstrItinClass itinQ16, InstrItinClass itinQ32,
2279 string OpcodeStr, string Dt,
2280 SDNode OpNode, bit Commutable = 0> {
2281 // 64-bit vector types.
2282 def v8i8 : N3VD<op24, op23, 0b00, op11_8, op4, itinD16,
2283 OpcodeStr, !strconcat(Dt, "8"),
2284 v8i8, v8i8, OpNode, Commutable>;
2285 def v4i16 : N3VD<op24, op23, 0b01, op11_8, op4, itinD16,
2286 OpcodeStr, !strconcat(Dt, "16"),
2287 v4i16, v4i16, OpNode, Commutable>;
2288 def v2i32 : N3VD<op24, op23, 0b10, op11_8, op4, itinD32,
2289 OpcodeStr, !strconcat(Dt, "32"),
2290 v2i32, v2i32, OpNode, Commutable>;
2292 // 128-bit vector types.
2293 def v16i8 : N3VQ<op24, op23, 0b00, op11_8, op4, itinQ16,
2294 OpcodeStr, !strconcat(Dt, "8"),
2295 v16i8, v16i8, OpNode, Commutable>;
2296 def v8i16 : N3VQ<op24, op23, 0b01, op11_8, op4, itinQ16,
2297 OpcodeStr, !strconcat(Dt, "16"),
2298 v8i16, v8i16, OpNode, Commutable>;
2299 def v4i32 : N3VQ<op24, op23, 0b10, op11_8, op4, itinQ32,
2300 OpcodeStr, !strconcat(Dt, "32"),
2301 v4i32, v4i32, OpNode, Commutable>;
2304 multiclass N3VSL_HS<bits<4> op11_8, string OpcodeStr, string Dt, SDNode ShOp> {
2305 def v4i16 : N3VDSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
2307 def v2i32 : N3VDSL<0b10, op11_8, IIC_VMULi32D, OpcodeStr, !strconcat(Dt,"32"),
2309 def v8i16 : N3VQSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
2310 v8i16, v4i16, ShOp>;
2311 def v4i32 : N3VQSL<0b10, op11_8, IIC_VMULi32Q, OpcodeStr, !strconcat(Dt,"32"),
2312 v4i32, v2i32, ShOp>;
2315 // ....then also with element size 64 bits:
2316 multiclass N3V_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2317 InstrItinClass itinD, InstrItinClass itinQ,
2318 string OpcodeStr, string Dt,
2319 SDNode OpNode, bit Commutable = 0>
2320 : N3V_QHS<op24, op23, op11_8, op4, itinD, itinD, itinQ, itinQ,
2321 OpcodeStr, Dt, OpNode, Commutable> {
2322 def v1i64 : N3VD<op24, op23, 0b11, op11_8, op4, itinD,
2323 OpcodeStr, !strconcat(Dt, "64"),
2324 v1i64, v1i64, OpNode, Commutable>;
2325 def v2i64 : N3VQ<op24, op23, 0b11, op11_8, op4, itinQ,
2326 OpcodeStr, !strconcat(Dt, "64"),
2327 v2i64, v2i64, OpNode, Commutable>;
2331 // Neon Narrowing 2-register vector operations,
2332 // source operand element sizes of 16, 32 and 64 bits:
2333 multiclass N2VN_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2334 bits<5> op11_7, bit op6, bit op4,
2335 InstrItinClass itin, string OpcodeStr, string Dt,
2337 def v8i8 : N2VN<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
2338 itin, OpcodeStr, !strconcat(Dt, "16"),
2339 v8i8, v8i16, OpNode>;
2340 def v4i16 : N2VN<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
2341 itin, OpcodeStr, !strconcat(Dt, "32"),
2342 v4i16, v4i32, OpNode>;
2343 def v2i32 : N2VN<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
2344 itin, OpcodeStr, !strconcat(Dt, "64"),
2345 v2i32, v2i64, OpNode>;
2348 // Neon Narrowing 2-register vector intrinsics,
2349 // source operand element sizes of 16, 32 and 64 bits:
2350 multiclass N2VNInt_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2351 bits<5> op11_7, bit op6, bit op4,
2352 InstrItinClass itin, string OpcodeStr, string Dt,
2354 def v8i8 : N2VNInt<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
2355 itin, OpcodeStr, !strconcat(Dt, "16"),
2356 v8i8, v8i16, IntOp>;
2357 def v4i16 : N2VNInt<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
2358 itin, OpcodeStr, !strconcat(Dt, "32"),
2359 v4i16, v4i32, IntOp>;
2360 def v2i32 : N2VNInt<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
2361 itin, OpcodeStr, !strconcat(Dt, "64"),
2362 v2i32, v2i64, IntOp>;
2366 // Neon Lengthening 2-register vector intrinsic (currently specific to VMOVL).
2367 // source operand element sizes of 16, 32 and 64 bits:
2368 multiclass N2VL_QHS<bits<2> op24_23, bits<5> op11_7, bit op6, bit op4,
2369 string OpcodeStr, string Dt, SDNode OpNode> {
2370 def v8i16 : N2VL<op24_23, 0b00, 0b10, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2371 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode>;
2372 def v4i32 : N2VL<op24_23, 0b01, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2373 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
2374 def v2i64 : N2VL<op24_23, 0b10, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2375 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
2379 // Neon 3-register vector intrinsics.
2381 // First with only element sizes of 16 and 32 bits:
2382 multiclass N3VInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2383 InstrItinClass itinD16, InstrItinClass itinD32,
2384 InstrItinClass itinQ16, InstrItinClass itinQ32,
2385 string OpcodeStr, string Dt,
2386 Intrinsic IntOp, bit Commutable = 0> {
2387 // 64-bit vector types.
2388 def v4i16 : N3VDInt<op24, op23, 0b01, op11_8, op4, f, itinD16,
2389 OpcodeStr, !strconcat(Dt, "16"),
2390 v4i16, v4i16, IntOp, Commutable>;
2391 def v2i32 : N3VDInt<op24, op23, 0b10, op11_8, op4, f, itinD32,
2392 OpcodeStr, !strconcat(Dt, "32"),
2393 v2i32, v2i32, IntOp, Commutable>;
2395 // 128-bit vector types.
2396 def v8i16 : N3VQInt<op24, op23, 0b01, op11_8, op4, f, itinQ16,
2397 OpcodeStr, !strconcat(Dt, "16"),
2398 v8i16, v8i16, IntOp, Commutable>;
2399 def v4i32 : N3VQInt<op24, op23, 0b10, op11_8, op4, f, itinQ32,
2400 OpcodeStr, !strconcat(Dt, "32"),
2401 v4i32, v4i32, IntOp, Commutable>;
2403 multiclass N3VInt_HSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2404 InstrItinClass itinD16, InstrItinClass itinD32,
2405 InstrItinClass itinQ16, InstrItinClass itinQ32,
2406 string OpcodeStr, string Dt,
2408 // 64-bit vector types.
2409 def v4i16 : N3VDIntSh<op24, op23, 0b01, op11_8, op4, f, itinD16,
2410 OpcodeStr, !strconcat(Dt, "16"),
2411 v4i16, v4i16, IntOp>;
2412 def v2i32 : N3VDIntSh<op24, op23, 0b10, op11_8, op4, f, itinD32,
2413 OpcodeStr, !strconcat(Dt, "32"),
2414 v2i32, v2i32, IntOp>;
2416 // 128-bit vector types.
2417 def v8i16 : N3VQIntSh<op24, op23, 0b01, op11_8, op4, f, itinQ16,
2418 OpcodeStr, !strconcat(Dt, "16"),
2419 v8i16, v8i16, IntOp>;
2420 def v4i32 : N3VQIntSh<op24, op23, 0b10, op11_8, op4, f, itinQ32,
2421 OpcodeStr, !strconcat(Dt, "32"),
2422 v4i32, v4i32, IntOp>;
2425 multiclass N3VIntSL_HS<bits<4> op11_8,
2426 InstrItinClass itinD16, InstrItinClass itinD32,
2427 InstrItinClass itinQ16, InstrItinClass itinQ32,
2428 string OpcodeStr, string Dt, Intrinsic IntOp> {
2429 def v4i16 : N3VDIntSL16<0b01, op11_8, itinD16,
2430 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp>;
2431 def v2i32 : N3VDIntSL<0b10, op11_8, itinD32,
2432 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp>;
2433 def v8i16 : N3VQIntSL16<0b01, op11_8, itinQ16,
2434 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16, IntOp>;
2435 def v4i32 : N3VQIntSL<0b10, op11_8, itinQ32,
2436 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32, IntOp>;
2439 // ....then also with element size of 8 bits:
2440 multiclass N3VInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2441 InstrItinClass itinD16, InstrItinClass itinD32,
2442 InstrItinClass itinQ16, InstrItinClass itinQ32,
2443 string OpcodeStr, string Dt,
2444 Intrinsic IntOp, bit Commutable = 0>
2445 : N3VInt_HS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2446 OpcodeStr, Dt, IntOp, Commutable> {
2447 def v8i8 : N3VDInt<op24, op23, 0b00, op11_8, op4, f, itinD16,
2448 OpcodeStr, !strconcat(Dt, "8"),
2449 v8i8, v8i8, IntOp, Commutable>;
2450 def v16i8 : N3VQInt<op24, op23, 0b00, op11_8, op4, f, itinQ16,
2451 OpcodeStr, !strconcat(Dt, "8"),
2452 v16i8, v16i8, IntOp, Commutable>;
2454 multiclass N3VInt_QHSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2455 InstrItinClass itinD16, InstrItinClass itinD32,
2456 InstrItinClass itinQ16, InstrItinClass itinQ32,
2457 string OpcodeStr, string Dt,
2459 : N3VInt_HSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2460 OpcodeStr, Dt, IntOp> {
2461 def v8i8 : N3VDIntSh<op24, op23, 0b00, op11_8, op4, f, itinD16,
2462 OpcodeStr, !strconcat(Dt, "8"),
2464 def v16i8 : N3VQIntSh<op24, op23, 0b00, op11_8, op4, f, itinQ16,
2465 OpcodeStr, !strconcat(Dt, "8"),
2466 v16i8, v16i8, IntOp>;
2470 // ....then also with element size of 64 bits:
2471 multiclass N3VInt_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2472 InstrItinClass itinD16, InstrItinClass itinD32,
2473 InstrItinClass itinQ16, InstrItinClass itinQ32,
2474 string OpcodeStr, string Dt,
2475 Intrinsic IntOp, bit Commutable = 0>
2476 : N3VInt_QHS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2477 OpcodeStr, Dt, IntOp, Commutable> {
2478 def v1i64 : N3VDInt<op24, op23, 0b11, op11_8, op4, f, itinD32,
2479 OpcodeStr, !strconcat(Dt, "64"),
2480 v1i64, v1i64, IntOp, Commutable>;
2481 def v2i64 : N3VQInt<op24, op23, 0b11, op11_8, op4, f, itinQ32,
2482 OpcodeStr, !strconcat(Dt, "64"),
2483 v2i64, v2i64, IntOp, Commutable>;
2485 multiclass N3VInt_QHSDSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2486 InstrItinClass itinD16, InstrItinClass itinD32,
2487 InstrItinClass itinQ16, InstrItinClass itinQ32,
2488 string OpcodeStr, string Dt,
2490 : N3VInt_QHSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2491 OpcodeStr, Dt, IntOp> {
2492 def v1i64 : N3VDIntSh<op24, op23, 0b11, op11_8, op4, f, itinD32,
2493 OpcodeStr, !strconcat(Dt, "64"),
2494 v1i64, v1i64, IntOp>;
2495 def v2i64 : N3VQIntSh<op24, op23, 0b11, op11_8, op4, f, itinQ32,
2496 OpcodeStr, !strconcat(Dt, "64"),
2497 v2i64, v2i64, IntOp>;
2500 // Neon Narrowing 3-register vector intrinsics,
2501 // source operand element sizes of 16, 32 and 64 bits:
2502 multiclass N3VNInt_HSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2503 string OpcodeStr, string Dt,
2504 Intrinsic IntOp, bit Commutable = 0> {
2505 def v8i8 : N3VNInt<op24, op23, 0b00, op11_8, op4,
2506 OpcodeStr, !strconcat(Dt, "16"),
2507 v8i8, v8i16, IntOp, Commutable>;
2508 def v4i16 : N3VNInt<op24, op23, 0b01, op11_8, op4,
2509 OpcodeStr, !strconcat(Dt, "32"),
2510 v4i16, v4i32, IntOp, Commutable>;
2511 def v2i32 : N3VNInt<op24, op23, 0b10, op11_8, op4,
2512 OpcodeStr, !strconcat(Dt, "64"),
2513 v2i32, v2i64, IntOp, Commutable>;
2517 // Neon Long 3-register vector operations.
2519 multiclass N3VL_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2520 InstrItinClass itin16, InstrItinClass itin32,
2521 string OpcodeStr, string Dt,
2522 SDNode OpNode, bit Commutable = 0> {
2523 def v8i16 : N3VL<op24, op23, 0b00, op11_8, op4, itin16,
2524 OpcodeStr, !strconcat(Dt, "8"),
2525 v8i16, v8i8, OpNode, Commutable>;
2526 def v4i32 : N3VL<op24, op23, 0b01, op11_8, op4, itin16,
2527 OpcodeStr, !strconcat(Dt, "16"),
2528 v4i32, v4i16, OpNode, Commutable>;
2529 def v2i64 : N3VL<op24, op23, 0b10, op11_8, op4, itin32,
2530 OpcodeStr, !strconcat(Dt, "32"),
2531 v2i64, v2i32, OpNode, Commutable>;
2534 multiclass N3VLSL_HS<bit op24, bits<4> op11_8,
2535 InstrItinClass itin, string OpcodeStr, string Dt,
2537 def v4i16 : N3VLSL16<op24, 0b01, op11_8, itin, OpcodeStr,
2538 !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
2539 def v2i32 : N3VLSL<op24, 0b10, op11_8, itin, OpcodeStr,
2540 !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
2543 multiclass N3VLExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2544 InstrItinClass itin16, InstrItinClass itin32,
2545 string OpcodeStr, string Dt,
2546 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
2547 def v8i16 : N3VLExt<op24, op23, 0b00, op11_8, op4, itin16,
2548 OpcodeStr, !strconcat(Dt, "8"),
2549 v8i16, v8i8, OpNode, ExtOp, Commutable>;
2550 def v4i32 : N3VLExt<op24, op23, 0b01, op11_8, op4, itin16,
2551 OpcodeStr, !strconcat(Dt, "16"),
2552 v4i32, v4i16, OpNode, ExtOp, Commutable>;
2553 def v2i64 : N3VLExt<op24, op23, 0b10, op11_8, op4, itin32,
2554 OpcodeStr, !strconcat(Dt, "32"),
2555 v2i64, v2i32, OpNode, ExtOp, Commutable>;
2558 // Neon Long 3-register vector intrinsics.
2560 // First with only element sizes of 16 and 32 bits:
2561 multiclass N3VLInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
2562 InstrItinClass itin16, InstrItinClass itin32,
2563 string OpcodeStr, string Dt,
2564 Intrinsic IntOp, bit Commutable = 0> {
2565 def v4i32 : N3VLInt<op24, op23, 0b01, op11_8, op4, itin16,
2566 OpcodeStr, !strconcat(Dt, "16"),
2567 v4i32, v4i16, IntOp, Commutable>;
2568 def v2i64 : N3VLInt<op24, op23, 0b10, op11_8, op4, itin32,
2569 OpcodeStr, !strconcat(Dt, "32"),
2570 v2i64, v2i32, IntOp, Commutable>;
2573 multiclass N3VLIntSL_HS<bit op24, bits<4> op11_8,
2574 InstrItinClass itin, string OpcodeStr, string Dt,
2576 def v4i16 : N3VLIntSL16<op24, 0b01, op11_8, itin,
2577 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
2578 def v2i32 : N3VLIntSL<op24, 0b10, op11_8, itin,
2579 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
2582 // ....then also with element size of 8 bits:
2583 multiclass N3VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2584 InstrItinClass itin16, InstrItinClass itin32,
2585 string OpcodeStr, string Dt,
2586 Intrinsic IntOp, bit Commutable = 0>
2587 : N3VLInt_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt,
2588 IntOp, Commutable> {
2589 def v8i16 : N3VLInt<op24, op23, 0b00, op11_8, op4, itin16,
2590 OpcodeStr, !strconcat(Dt, "8"),
2591 v8i16, v8i8, IntOp, Commutable>;
2594 // ....with explicit extend (VABDL).
2595 multiclass N3VLIntExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2596 InstrItinClass itin, string OpcodeStr, string Dt,
2597 Intrinsic IntOp, SDNode ExtOp, bit Commutable = 0> {
2598 def v8i16 : N3VLIntExt<op24, op23, 0b00, op11_8, op4, itin,
2599 OpcodeStr, !strconcat(Dt, "8"),
2600 v8i16, v8i8, IntOp, ExtOp, Commutable>;
2601 def v4i32 : N3VLIntExt<op24, op23, 0b01, op11_8, op4, itin,
2602 OpcodeStr, !strconcat(Dt, "16"),
2603 v4i32, v4i16, IntOp, ExtOp, Commutable>;
2604 def v2i64 : N3VLIntExt<op24, op23, 0b10, op11_8, op4, itin,
2605 OpcodeStr, !strconcat(Dt, "32"),
2606 v2i64, v2i32, IntOp, ExtOp, Commutable>;
2610 // Neon Wide 3-register vector intrinsics,
2611 // source operand element sizes of 8, 16 and 32 bits:
2612 multiclass N3VW_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2613 string OpcodeStr, string Dt,
2614 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
2615 def v8i16 : N3VW<op24, op23, 0b00, op11_8, op4,
2616 OpcodeStr, !strconcat(Dt, "8"),
2617 v8i16, v8i8, OpNode, ExtOp, Commutable>;
2618 def v4i32 : N3VW<op24, op23, 0b01, op11_8, op4,
2619 OpcodeStr, !strconcat(Dt, "16"),
2620 v4i32, v4i16, OpNode, ExtOp, Commutable>;
2621 def v2i64 : N3VW<op24, op23, 0b10, op11_8, op4,
2622 OpcodeStr, !strconcat(Dt, "32"),
2623 v2i64, v2i32, OpNode, ExtOp, Commutable>;
2627 // Neon Multiply-Op vector operations,
2628 // element sizes of 8, 16 and 32 bits:
2629 multiclass N3VMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2630 InstrItinClass itinD16, InstrItinClass itinD32,
2631 InstrItinClass itinQ16, InstrItinClass itinQ32,
2632 string OpcodeStr, string Dt, SDNode OpNode> {
2633 // 64-bit vector types.
2634 def v8i8 : N3VDMulOp<op24, op23, 0b00, op11_8, op4, itinD16,
2635 OpcodeStr, !strconcat(Dt, "8"), v8i8, mul, OpNode>;
2636 def v4i16 : N3VDMulOp<op24, op23, 0b01, op11_8, op4, itinD16,
2637 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, OpNode>;
2638 def v2i32 : N3VDMulOp<op24, op23, 0b10, op11_8, op4, itinD32,
2639 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, OpNode>;
2641 // 128-bit vector types.
2642 def v16i8 : N3VQMulOp<op24, op23, 0b00, op11_8, op4, itinQ16,
2643 OpcodeStr, !strconcat(Dt, "8"), v16i8, mul, OpNode>;
2644 def v8i16 : N3VQMulOp<op24, op23, 0b01, op11_8, op4, itinQ16,
2645 OpcodeStr, !strconcat(Dt, "16"), v8i16, mul, OpNode>;
2646 def v4i32 : N3VQMulOp<op24, op23, 0b10, op11_8, op4, itinQ32,
2647 OpcodeStr, !strconcat(Dt, "32"), v4i32, mul, OpNode>;
2650 multiclass N3VMulOpSL_HS<bits<4> op11_8,
2651 InstrItinClass itinD16, InstrItinClass itinD32,
2652 InstrItinClass itinQ16, InstrItinClass itinQ32,
2653 string OpcodeStr, string Dt, SDNode ShOp> {
2654 def v4i16 : N3VDMulOpSL16<0b01, op11_8, itinD16,
2655 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, ShOp>;
2656 def v2i32 : N3VDMulOpSL<0b10, op11_8, itinD32,
2657 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, ShOp>;
2658 def v8i16 : N3VQMulOpSL16<0b01, op11_8, itinQ16,
2659 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16,
2661 def v4i32 : N3VQMulOpSL<0b10, op11_8, itinQ32,
2662 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32,
2666 // Neon Intrinsic-Op vector operations,
2667 // element sizes of 8, 16 and 32 bits:
2668 multiclass N3VIntOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2669 InstrItinClass itinD, InstrItinClass itinQ,
2670 string OpcodeStr, string Dt, Intrinsic IntOp,
2672 // 64-bit vector types.
2673 def v8i8 : N3VDIntOp<op24, op23, 0b00, op11_8, op4, itinD,
2674 OpcodeStr, !strconcat(Dt, "8"), v8i8, IntOp, OpNode>;
2675 def v4i16 : N3VDIntOp<op24, op23, 0b01, op11_8, op4, itinD,
2676 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp, OpNode>;
2677 def v2i32 : N3VDIntOp<op24, op23, 0b10, op11_8, op4, itinD,
2678 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp, OpNode>;
2680 // 128-bit vector types.
2681 def v16i8 : N3VQIntOp<op24, op23, 0b00, op11_8, op4, itinQ,
2682 OpcodeStr, !strconcat(Dt, "8"), v16i8, IntOp, OpNode>;
2683 def v8i16 : N3VQIntOp<op24, op23, 0b01, op11_8, op4, itinQ,
2684 OpcodeStr, !strconcat(Dt, "16"), v8i16, IntOp, OpNode>;
2685 def v4i32 : N3VQIntOp<op24, op23, 0b10, op11_8, op4, itinQ,
2686 OpcodeStr, !strconcat(Dt, "32"), v4i32, IntOp, OpNode>;
2689 // Neon 3-argument intrinsics,
2690 // element sizes of 8, 16 and 32 bits:
2691 multiclass N3VInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2692 InstrItinClass itinD, InstrItinClass itinQ,
2693 string OpcodeStr, string Dt, Intrinsic IntOp> {
2694 // 64-bit vector types.
2695 def v8i8 : N3VDInt3<op24, op23, 0b00, op11_8, op4, itinD,
2696 OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
2697 def v4i16 : N3VDInt3<op24, op23, 0b01, op11_8, op4, itinD,
2698 OpcodeStr, !strconcat(Dt, "16"), v4i16, v4i16, IntOp>;
2699 def v2i32 : N3VDInt3<op24, op23, 0b10, op11_8, op4, itinD,
2700 OpcodeStr, !strconcat(Dt, "32"), v2i32, v2i32, IntOp>;
2702 // 128-bit vector types.
2703 def v16i8 : N3VQInt3<op24, op23, 0b00, op11_8, op4, itinQ,
2704 OpcodeStr, !strconcat(Dt, "8"), v16i8, v16i8, IntOp>;
2705 def v8i16 : N3VQInt3<op24, op23, 0b01, op11_8, op4, itinQ,
2706 OpcodeStr, !strconcat(Dt, "16"), v8i16, v8i16, IntOp>;
2707 def v4i32 : N3VQInt3<op24, op23, 0b10, op11_8, op4, itinQ,
2708 OpcodeStr, !strconcat(Dt, "32"), v4i32, v4i32, IntOp>;
2712 // Neon Long Multiply-Op vector operations,
2713 // element sizes of 8, 16 and 32 bits:
2714 multiclass N3VLMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2715 InstrItinClass itin16, InstrItinClass itin32,
2716 string OpcodeStr, string Dt, SDNode MulOp,
2718 def v8i16 : N3VLMulOp<op24, op23, 0b00, op11_8, op4, itin16, OpcodeStr,
2719 !strconcat(Dt, "8"), v8i16, v8i8, MulOp, OpNode>;
2720 def v4i32 : N3VLMulOp<op24, op23, 0b01, op11_8, op4, itin16, OpcodeStr,
2721 !strconcat(Dt, "16"), v4i32, v4i16, MulOp, OpNode>;
2722 def v2i64 : N3VLMulOp<op24, op23, 0b10, op11_8, op4, itin32, OpcodeStr,
2723 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
2726 multiclass N3VLMulOpSL_HS<bit op24, bits<4> op11_8, string OpcodeStr,
2727 string Dt, SDNode MulOp, SDNode OpNode> {
2728 def v4i16 : N3VLMulOpSL16<op24, 0b01, op11_8, IIC_VMACi16D, OpcodeStr,
2729 !strconcat(Dt,"16"), v4i32, v4i16, MulOp, OpNode>;
2730 def v2i32 : N3VLMulOpSL<op24, 0b10, op11_8, IIC_VMACi32D, OpcodeStr,
2731 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
2735 // Neon Long 3-argument intrinsics.
2737 // First with only element sizes of 16 and 32 bits:
2738 multiclass N3VLInt3_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
2739 InstrItinClass itin16, InstrItinClass itin32,
2740 string OpcodeStr, string Dt, Intrinsic IntOp> {
2741 def v4i32 : N3VLInt3<op24, op23, 0b01, op11_8, op4, itin16,
2742 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
2743 def v2i64 : N3VLInt3<op24, op23, 0b10, op11_8, op4, itin32,
2744 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
2747 multiclass N3VLInt3SL_HS<bit op24, bits<4> op11_8,
2748 string OpcodeStr, string Dt, Intrinsic IntOp> {
2749 def v4i16 : N3VLInt3SL16<op24, 0b01, op11_8, IIC_VMACi16D,
2750 OpcodeStr, !strconcat(Dt,"16"), v4i32, v4i16, IntOp>;
2751 def v2i32 : N3VLInt3SL<op24, 0b10, op11_8, IIC_VMACi32D,
2752 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
2755 // ....then also with element size of 8 bits:
2756 multiclass N3VLInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2757 InstrItinClass itin16, InstrItinClass itin32,
2758 string OpcodeStr, string Dt, Intrinsic IntOp>
2759 : N3VLInt3_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt, IntOp> {
2760 def v8i16 : N3VLInt3<op24, op23, 0b00, op11_8, op4, itin16,
2761 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
2764 // ....with explicit extend (VABAL).
2765 multiclass N3VLIntExtOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2766 InstrItinClass itin, string OpcodeStr, string Dt,
2767 Intrinsic IntOp, SDNode ExtOp, SDNode OpNode> {
2768 def v8i16 : N3VLIntExtOp<op24, op23, 0b00, op11_8, op4, itin,
2769 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8,
2770 IntOp, ExtOp, OpNode>;
2771 def v4i32 : N3VLIntExtOp<op24, op23, 0b01, op11_8, op4, itin,
2772 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16,
2773 IntOp, ExtOp, OpNode>;
2774 def v2i64 : N3VLIntExtOp<op24, op23, 0b10, op11_8, op4, itin,
2775 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32,
2776 IntOp, ExtOp, OpNode>;
2780 // Neon 2-register vector intrinsics,
2781 // element sizes of 8, 16 and 32 bits:
2782 multiclass N2VInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2783 bits<5> op11_7, bit op4,
2784 InstrItinClass itinD, InstrItinClass itinQ,
2785 string OpcodeStr, string Dt, Intrinsic IntOp> {
2786 // 64-bit vector types.
2787 def v8i8 : N2VDInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2788 itinD, OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
2789 def v4i16 : N2VDInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2790 itinD, OpcodeStr, !strconcat(Dt, "16"),v4i16,v4i16,IntOp>;
2791 def v2i32 : N2VDInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2792 itinD, OpcodeStr, !strconcat(Dt, "32"),v2i32,v2i32,IntOp>;
2794 // 128-bit vector types.
2795 def v16i8 : N2VQInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2796 itinQ, OpcodeStr, !strconcat(Dt, "8"), v16i8,v16i8,IntOp>;
2797 def v8i16 : N2VQInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2798 itinQ, OpcodeStr, !strconcat(Dt, "16"),v8i16,v8i16,IntOp>;
2799 def v4i32 : N2VQInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2800 itinQ, OpcodeStr, !strconcat(Dt, "32"),v4i32,v4i32,IntOp>;
2804 // Neon Pairwise long 2-register intrinsics,
2805 // element sizes of 8, 16 and 32 bits:
2806 multiclass N2VPLInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2807 bits<5> op11_7, bit op4,
2808 string OpcodeStr, string Dt, Intrinsic IntOp> {
2809 // 64-bit vector types.
2810 def v8i8 : N2VDPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2811 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
2812 def v4i16 : N2VDPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2813 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
2814 def v2i32 : N2VDPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2815 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
2817 // 128-bit vector types.
2818 def v16i8 : N2VQPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2819 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
2820 def v8i16 : N2VQPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2821 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
2822 def v4i32 : N2VQPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2823 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
2827 // Neon Pairwise long 2-register accumulate intrinsics,
2828 // element sizes of 8, 16 and 32 bits:
2829 multiclass N2VPLInt2_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2830 bits<5> op11_7, bit op4,
2831 string OpcodeStr, string Dt, Intrinsic IntOp> {
2832 // 64-bit vector types.
2833 def v8i8 : N2VDPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2834 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
2835 def v4i16 : N2VDPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2836 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
2837 def v2i32 : N2VDPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2838 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
2840 // 128-bit vector types.
2841 def v16i8 : N2VQPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2842 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
2843 def v8i16 : N2VQPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2844 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
2845 def v4i32 : N2VQPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2846 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
2850 // Neon 2-register vector shift by immediate,
2851 // with f of either N2RegVShLFrm or N2RegVShRFrm
2852 // element sizes of 8, 16, 32 and 64 bits:
2853 multiclass N2VSh_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2854 InstrItinClass itin, string OpcodeStr, string Dt,
2855 SDNode OpNode, Format f> {
2856 // 64-bit vector types.
2857 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
2858 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
2859 let Inst{21-19} = 0b001; // imm6 = 001xxx
2861 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
2862 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
2863 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2865 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
2866 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
2867 let Inst{21} = 0b1; // imm6 = 1xxxxx
2869 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, f, itin,
2870 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
2873 // 128-bit vector types.
2874 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
2875 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
2876 let Inst{21-19} = 0b001; // imm6 = 001xxx
2878 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
2879 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
2880 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2882 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
2883 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
2884 let Inst{21} = 0b1; // imm6 = 1xxxxx
2886 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, f, itin,
2887 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
2891 // Neon Shift-Accumulate vector operations,
2892 // element sizes of 8, 16, 32 and 64 bits:
2893 multiclass N2VShAdd_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2894 string OpcodeStr, string Dt, SDNode ShOp> {
2895 // 64-bit vector types.
2896 def v8i8 : N2VDShAdd<op24, op23, op11_8, 0, op4,
2897 OpcodeStr, !strconcat(Dt, "8"), v8i8, ShOp> {
2898 let Inst{21-19} = 0b001; // imm6 = 001xxx
2900 def v4i16 : N2VDShAdd<op24, op23, op11_8, 0, op4,
2901 OpcodeStr, !strconcat(Dt, "16"), v4i16, ShOp> {
2902 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2904 def v2i32 : N2VDShAdd<op24, op23, op11_8, 0, op4,
2905 OpcodeStr, !strconcat(Dt, "32"), v2i32, ShOp> {
2906 let Inst{21} = 0b1; // imm6 = 1xxxxx
2908 def v1i64 : N2VDShAdd<op24, op23, op11_8, 1, op4,
2909 OpcodeStr, !strconcat(Dt, "64"), v1i64, ShOp>;
2912 // 128-bit vector types.
2913 def v16i8 : N2VQShAdd<op24, op23, op11_8, 0, op4,
2914 OpcodeStr, !strconcat(Dt, "8"), v16i8, ShOp> {
2915 let Inst{21-19} = 0b001; // imm6 = 001xxx
2917 def v8i16 : N2VQShAdd<op24, op23, op11_8, 0, op4,
2918 OpcodeStr, !strconcat(Dt, "16"), v8i16, ShOp> {
2919 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2921 def v4i32 : N2VQShAdd<op24, op23, op11_8, 0, op4,
2922 OpcodeStr, !strconcat(Dt, "32"), v4i32, ShOp> {
2923 let Inst{21} = 0b1; // imm6 = 1xxxxx
2925 def v2i64 : N2VQShAdd<op24, op23, op11_8, 1, op4,
2926 OpcodeStr, !strconcat(Dt, "64"), v2i64, ShOp>;
2931 // Neon Shift-Insert vector operations,
2932 // with f of either N2RegVShLFrm or N2RegVShRFrm
2933 // element sizes of 8, 16, 32 and 64 bits:
2934 multiclass N2VShIns_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2935 string OpcodeStr, SDNode ShOp,
2937 // 64-bit vector types.
2938 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4,
2939 f, OpcodeStr, "8", v8i8, ShOp> {
2940 let Inst{21-19} = 0b001; // imm6 = 001xxx
2942 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4,
2943 f, OpcodeStr, "16", v4i16, ShOp> {
2944 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2946 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4,
2947 f, OpcodeStr, "32", v2i32, ShOp> {
2948 let Inst{21} = 0b1; // imm6 = 1xxxxx
2950 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4,
2951 f, OpcodeStr, "64", v1i64, ShOp>;
2954 // 128-bit vector types.
2955 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4,
2956 f, OpcodeStr, "8", v16i8, ShOp> {
2957 let Inst{21-19} = 0b001; // imm6 = 001xxx
2959 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4,
2960 f, OpcodeStr, "16", v8i16, ShOp> {
2961 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2963 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4,
2964 f, OpcodeStr, "32", v4i32, ShOp> {
2965 let Inst{21} = 0b1; // imm6 = 1xxxxx
2967 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4,
2968 f, OpcodeStr, "64", v2i64, ShOp>;
2972 // Neon Shift Long operations,
2973 // element sizes of 8, 16, 32 bits:
2974 multiclass N2VLSh_QHS<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
2975 bit op4, string OpcodeStr, string Dt, SDNode OpNode> {
2976 def v8i16 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
2977 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode> {
2978 let Inst{21-19} = 0b001; // imm6 = 001xxx
2980 def v4i32 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
2981 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode> {
2982 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2984 def v2i64 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
2985 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode> {
2986 let Inst{21} = 0b1; // imm6 = 1xxxxx
2990 // Neon Shift Narrow operations,
2991 // element sizes of 16, 32, 64 bits:
2992 multiclass N2VNSh_HSD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
2993 bit op4, InstrItinClass itin, string OpcodeStr, string Dt,
2995 def v8i8 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
2996 OpcodeStr, !strconcat(Dt, "16"), v8i8, v8i16, OpNode> {
2997 let Inst{21-19} = 0b001; // imm6 = 001xxx
2999 def v4i16 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3000 OpcodeStr, !strconcat(Dt, "32"), v4i16, v4i32, OpNode> {
3001 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3003 def v2i32 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3004 OpcodeStr, !strconcat(Dt, "64"), v2i32, v2i64, OpNode> {
3005 let Inst{21} = 0b1; // imm6 = 1xxxxx
3009 //===----------------------------------------------------------------------===//
3010 // Instruction Definitions.
3011 //===----------------------------------------------------------------------===//
3013 // Vector Add Operations.
3015 // VADD : Vector Add (integer and floating-point)
3016 defm VADD : N3V_QHSD<0, 0, 0b1000, 0, IIC_VBINiD, IIC_VBINiQ, "vadd", "i",
3018 def VADDfd : N3VD<0, 0, 0b00, 0b1101, 0, IIC_VBIND, "vadd", "f32",
3019 v2f32, v2f32, fadd, 1>;
3020 def VADDfq : N3VQ<0, 0, 0b00, 0b1101, 0, IIC_VBINQ, "vadd", "f32",
3021 v4f32, v4f32, fadd, 1>;
3022 // VADDL : Vector Add Long (Q = D + D)
3023 defm VADDLs : N3VLExt_QHS<0,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
3024 "vaddl", "s", add, sext, 1>;
3025 defm VADDLu : N3VLExt_QHS<1,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
3026 "vaddl", "u", add, zext, 1>;
3027 // VADDW : Vector Add Wide (Q = Q + D)
3028 defm VADDWs : N3VW_QHS<0,1,0b0001,0, "vaddw", "s", add, sext, 0>;
3029 defm VADDWu : N3VW_QHS<1,1,0b0001,0, "vaddw", "u", add, zext, 0>;
3030 // VHADD : Vector Halving Add
3031 defm VHADDs : N3VInt_QHS<0, 0, 0b0000, 0, N3RegFrm,
3032 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3033 "vhadd", "s", int_arm_neon_vhadds, 1>;
3034 defm VHADDu : N3VInt_QHS<1, 0, 0b0000, 0, N3RegFrm,
3035 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3036 "vhadd", "u", int_arm_neon_vhaddu, 1>;
3037 // VRHADD : Vector Rounding Halving Add
3038 defm VRHADDs : N3VInt_QHS<0, 0, 0b0001, 0, N3RegFrm,
3039 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3040 "vrhadd", "s", int_arm_neon_vrhadds, 1>;
3041 defm VRHADDu : N3VInt_QHS<1, 0, 0b0001, 0, N3RegFrm,
3042 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3043 "vrhadd", "u", int_arm_neon_vrhaddu, 1>;
3044 // VQADD : Vector Saturating Add
3045 defm VQADDs : N3VInt_QHSD<0, 0, 0b0000, 1, N3RegFrm,
3046 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3047 "vqadd", "s", int_arm_neon_vqadds, 1>;
3048 defm VQADDu : N3VInt_QHSD<1, 0, 0b0000, 1, N3RegFrm,
3049 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3050 "vqadd", "u", int_arm_neon_vqaddu, 1>;
3051 // VADDHN : Vector Add and Narrow Returning High Half (D = Q + Q)
3052 defm VADDHN : N3VNInt_HSD<0,1,0b0100,0, "vaddhn", "i",
3053 int_arm_neon_vaddhn, 1>;
3054 // VRADDHN : Vector Rounding Add and Narrow Returning High Half (D = Q + Q)
3055 defm VRADDHN : N3VNInt_HSD<1,1,0b0100,0, "vraddhn", "i",
3056 int_arm_neon_vraddhn, 1>;
3058 // Vector Multiply Operations.
3060 // VMUL : Vector Multiply (integer, polynomial and floating-point)
3061 defm VMUL : N3V_QHS<0, 0, 0b1001, 1, IIC_VMULi16D, IIC_VMULi32D,
3062 IIC_VMULi16Q, IIC_VMULi32Q, "vmul", "i", mul, 1>;
3063 def VMULpd : N3VDInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16D, "vmul",
3064 "p8", v8i8, v8i8, int_arm_neon_vmulp, 1>;
3065 def VMULpq : N3VQInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16Q, "vmul",
3066 "p8", v16i8, v16i8, int_arm_neon_vmulp, 1>;
3067 def VMULfd : N3VD<1, 0, 0b00, 0b1101, 1, IIC_VFMULD, "vmul", "f32",
3068 v2f32, v2f32, fmul, 1>;
3069 def VMULfq : N3VQ<1, 0, 0b00, 0b1101, 1, IIC_VFMULQ, "vmul", "f32",
3070 v4f32, v4f32, fmul, 1>;
3071 defm VMULsl : N3VSL_HS<0b1000, "vmul", "i", mul>;
3072 def VMULslfd : N3VDSL<0b10, 0b1001, IIC_VBIND, "vmul", "f32", v2f32, fmul>;
3073 def VMULslfq : N3VQSL<0b10, 0b1001, IIC_VBINQ, "vmul", "f32", v4f32,
3076 def : Pat<(v8i16 (mul (v8i16 QPR:$src1),
3077 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
3078 (v8i16 (VMULslv8i16 (v8i16 QPR:$src1),
3079 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3080 (DSubReg_i16_reg imm:$lane))),
3081 (SubReg_i16_lane imm:$lane)))>;
3082 def : Pat<(v4i32 (mul (v4i32 QPR:$src1),
3083 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
3084 (v4i32 (VMULslv4i32 (v4i32 QPR:$src1),
3085 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3086 (DSubReg_i32_reg imm:$lane))),
3087 (SubReg_i32_lane imm:$lane)))>;
3088 def : Pat<(v4f32 (fmul (v4f32 QPR:$src1),
3089 (v4f32 (NEONvduplane (v4f32 QPR:$src2), imm:$lane)))),
3090 (v4f32 (VMULslfq (v4f32 QPR:$src1),
3091 (v2f32 (EXTRACT_SUBREG QPR:$src2,
3092 (DSubReg_i32_reg imm:$lane))),
3093 (SubReg_i32_lane imm:$lane)))>;
3095 // VQDMULH : Vector Saturating Doubling Multiply Returning High Half
3096 defm VQDMULH : N3VInt_HS<0, 0, 0b1011, 0, N3RegFrm, IIC_VMULi16D, IIC_VMULi32D,
3097 IIC_VMULi16Q, IIC_VMULi32Q,
3098 "vqdmulh", "s", int_arm_neon_vqdmulh, 1>;
3099 defm VQDMULHsl: N3VIntSL_HS<0b1100, IIC_VMULi16D, IIC_VMULi32D,
3100 IIC_VMULi16Q, IIC_VMULi32Q,
3101 "vqdmulh", "s", int_arm_neon_vqdmulh>;
3102 def : Pat<(v8i16 (int_arm_neon_vqdmulh (v8i16 QPR:$src1),
3103 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
3105 (v8i16 (VQDMULHslv8i16 (v8i16 QPR:$src1),
3106 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3107 (DSubReg_i16_reg imm:$lane))),
3108 (SubReg_i16_lane imm:$lane)))>;
3109 def : Pat<(v4i32 (int_arm_neon_vqdmulh (v4i32 QPR:$src1),
3110 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
3112 (v4i32 (VQDMULHslv4i32 (v4i32 QPR:$src1),
3113 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3114 (DSubReg_i32_reg imm:$lane))),
3115 (SubReg_i32_lane imm:$lane)))>;
3117 // VQRDMULH : Vector Rounding Saturating Doubling Multiply Returning High Half
3118 defm VQRDMULH : N3VInt_HS<1, 0, 0b1011, 0, N3RegFrm,
3119 IIC_VMULi16D,IIC_VMULi32D,IIC_VMULi16Q,IIC_VMULi32Q,
3120 "vqrdmulh", "s", int_arm_neon_vqrdmulh, 1>;
3121 defm VQRDMULHsl : N3VIntSL_HS<0b1101, IIC_VMULi16D, IIC_VMULi32D,
3122 IIC_VMULi16Q, IIC_VMULi32Q,
3123 "vqrdmulh", "s", int_arm_neon_vqrdmulh>;
3124 def : Pat<(v8i16 (int_arm_neon_vqrdmulh (v8i16 QPR:$src1),
3125 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
3127 (v8i16 (VQRDMULHslv8i16 (v8i16 QPR:$src1),
3128 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3129 (DSubReg_i16_reg imm:$lane))),
3130 (SubReg_i16_lane imm:$lane)))>;
3131 def : Pat<(v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$src1),
3132 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
3134 (v4i32 (VQRDMULHslv4i32 (v4i32 QPR:$src1),
3135 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3136 (DSubReg_i32_reg imm:$lane))),
3137 (SubReg_i32_lane imm:$lane)))>;
3139 // VMULL : Vector Multiply Long (integer and polynomial) (Q = D * D)
3140 defm VMULLs : N3VL_QHS<0,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
3141 "vmull", "s", NEONvmulls, 1>;
3142 defm VMULLu : N3VL_QHS<1,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
3143 "vmull", "u", NEONvmullu, 1>;
3144 def VMULLp : N3VLInt<0, 1, 0b00, 0b1110, 0, IIC_VMULi16D, "vmull", "p8",
3145 v8i16, v8i8, int_arm_neon_vmullp, 1>;
3146 defm VMULLsls : N3VLSL_HS<0, 0b1010, IIC_VMULi16D, "vmull", "s", NEONvmulls>;
3147 defm VMULLslu : N3VLSL_HS<1, 0b1010, IIC_VMULi16D, "vmull", "u", NEONvmullu>;
3149 // VQDMULL : Vector Saturating Doubling Multiply Long (Q = D * D)
3150 defm VQDMULL : N3VLInt_HS<0,1,0b1101,0, IIC_VMULi16D, IIC_VMULi32D,
3151 "vqdmull", "s", int_arm_neon_vqdmull, 1>;
3152 defm VQDMULLsl: N3VLIntSL_HS<0, 0b1011, IIC_VMULi16D,
3153 "vqdmull", "s", int_arm_neon_vqdmull>;
3155 // Vector Multiply-Accumulate and Multiply-Subtract Operations.
3157 // VMLA : Vector Multiply Accumulate (integer and floating-point)
3158 defm VMLA : N3VMulOp_QHS<0, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3159 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
3160 def VMLAfd : N3VDMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACD, "vmla", "f32",
3162 def VMLAfq : N3VQMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACQ, "vmla", "f32",
3164 defm VMLAsl : N3VMulOpSL_HS<0b0000, IIC_VMACi16D, IIC_VMACi32D,
3165 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
3166 def VMLAslfd : N3VDMulOpSL<0b10, 0b0001, IIC_VMACD, "vmla", "f32",
3168 def VMLAslfq : N3VQMulOpSL<0b10, 0b0001, IIC_VMACQ, "vmla", "f32",
3169 v4f32, v2f32, fmul, fadd>;
3171 def : Pat<(v8i16 (add (v8i16 QPR:$src1),
3172 (mul (v8i16 QPR:$src2),
3173 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
3174 (v8i16 (VMLAslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
3175 (v4i16 (EXTRACT_SUBREG QPR:$src3,
3176 (DSubReg_i16_reg imm:$lane))),
3177 (SubReg_i16_lane imm:$lane)))>;
3179 def : Pat<(v4i32 (add (v4i32 QPR:$src1),
3180 (mul (v4i32 QPR:$src2),
3181 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
3182 (v4i32 (VMLAslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
3183 (v2i32 (EXTRACT_SUBREG QPR:$src3,
3184 (DSubReg_i32_reg imm:$lane))),
3185 (SubReg_i32_lane imm:$lane)))>;
3187 def : Pat<(v4f32 (fadd (v4f32 QPR:$src1),
3188 (fmul (v4f32 QPR:$src2),
3189 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
3190 (v4f32 (VMLAslfq (v4f32 QPR:$src1),
3192 (v2f32 (EXTRACT_SUBREG QPR:$src3,
3193 (DSubReg_i32_reg imm:$lane))),
3194 (SubReg_i32_lane imm:$lane)))>;
3196 // VMLAL : Vector Multiply Accumulate Long (Q += D * D)
3197 defm VMLALs : N3VLMulOp_QHS<0,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
3198 "vmlal", "s", NEONvmulls, add>;
3199 defm VMLALu : N3VLMulOp_QHS<1,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
3200 "vmlal", "u", NEONvmullu, add>;
3202 defm VMLALsls : N3VLMulOpSL_HS<0, 0b0010, "vmlal", "s", NEONvmulls, add>;
3203 defm VMLALslu : N3VLMulOpSL_HS<1, 0b0010, "vmlal", "u", NEONvmullu, add>;
3205 // VQDMLAL : Vector Saturating Doubling Multiply Accumulate Long (Q += D * D)
3206 defm VQDMLAL : N3VLInt3_HS<0, 1, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3207 "vqdmlal", "s", int_arm_neon_vqdmlal>;
3208 defm VQDMLALsl: N3VLInt3SL_HS<0, 0b0011, "vqdmlal", "s", int_arm_neon_vqdmlal>;
3210 // VMLS : Vector Multiply Subtract (integer and floating-point)
3211 defm VMLS : N3VMulOp_QHS<1, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3212 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
3213 def VMLSfd : N3VDMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACD, "vmls", "f32",
3215 def VMLSfq : N3VQMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACQ, "vmls", "f32",
3217 defm VMLSsl : N3VMulOpSL_HS<0b0100, IIC_VMACi16D, IIC_VMACi32D,
3218 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
3219 def VMLSslfd : N3VDMulOpSL<0b10, 0b0101, IIC_VMACD, "vmls", "f32",
3221 def VMLSslfq : N3VQMulOpSL<0b10, 0b0101, IIC_VMACQ, "vmls", "f32",
3222 v4f32, v2f32, fmul, fsub>;
3224 def : Pat<(v8i16 (sub (v8i16 QPR:$src1),
3225 (mul (v8i16 QPR:$src2),
3226 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
3227 (v8i16 (VMLSslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
3228 (v4i16 (EXTRACT_SUBREG QPR:$src3,
3229 (DSubReg_i16_reg imm:$lane))),
3230 (SubReg_i16_lane imm:$lane)))>;
3232 def : Pat<(v4i32 (sub (v4i32 QPR:$src1),
3233 (mul (v4i32 QPR:$src2),
3234 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
3235 (v4i32 (VMLSslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
3236 (v2i32 (EXTRACT_SUBREG QPR:$src3,
3237 (DSubReg_i32_reg imm:$lane))),
3238 (SubReg_i32_lane imm:$lane)))>;
3240 def : Pat<(v4f32 (fsub (v4f32 QPR:$src1),
3241 (fmul (v4f32 QPR:$src2),
3242 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
3243 (v4f32 (VMLSslfq (v4f32 QPR:$src1), (v4f32 QPR:$src2),
3244 (v2f32 (EXTRACT_SUBREG QPR:$src3,
3245 (DSubReg_i32_reg imm:$lane))),
3246 (SubReg_i32_lane imm:$lane)))>;
3248 // VMLSL : Vector Multiply Subtract Long (Q -= D * D)
3249 defm VMLSLs : N3VLMulOp_QHS<0,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
3250 "vmlsl", "s", NEONvmulls, sub>;
3251 defm VMLSLu : N3VLMulOp_QHS<1,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
3252 "vmlsl", "u", NEONvmullu, sub>;
3254 defm VMLSLsls : N3VLMulOpSL_HS<0, 0b0110, "vmlsl", "s", NEONvmulls, sub>;
3255 defm VMLSLslu : N3VLMulOpSL_HS<1, 0b0110, "vmlsl", "u", NEONvmullu, sub>;
3257 // VQDMLSL : Vector Saturating Doubling Multiply Subtract Long (Q -= D * D)
3258 defm VQDMLSL : N3VLInt3_HS<0, 1, 0b1011, 0, IIC_VMACi16D, IIC_VMACi32D,
3259 "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
3260 defm VQDMLSLsl: N3VLInt3SL_HS<0, 0b111, "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
3262 // Vector Subtract Operations.
3264 // VSUB : Vector Subtract (integer and floating-point)
3265 defm VSUB : N3V_QHSD<1, 0, 0b1000, 0, IIC_VSUBiD, IIC_VSUBiQ,
3266 "vsub", "i", sub, 0>;
3267 def VSUBfd : N3VD<0, 0, 0b10, 0b1101, 0, IIC_VBIND, "vsub", "f32",
3268 v2f32, v2f32, fsub, 0>;
3269 def VSUBfq : N3VQ<0, 0, 0b10, 0b1101, 0, IIC_VBINQ, "vsub", "f32",
3270 v4f32, v4f32, fsub, 0>;
3271 // VSUBL : Vector Subtract Long (Q = D - D)
3272 defm VSUBLs : N3VLExt_QHS<0,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
3273 "vsubl", "s", sub, sext, 0>;
3274 defm VSUBLu : N3VLExt_QHS<1,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
3275 "vsubl", "u", sub, zext, 0>;
3276 // VSUBW : Vector Subtract Wide (Q = Q - D)
3277 defm VSUBWs : N3VW_QHS<0,1,0b0011,0, "vsubw", "s", sub, sext, 0>;
3278 defm VSUBWu : N3VW_QHS<1,1,0b0011,0, "vsubw", "u", sub, zext, 0>;
3279 // VHSUB : Vector Halving Subtract
3280 defm VHSUBs : N3VInt_QHS<0, 0, 0b0010, 0, N3RegFrm,
3281 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3282 "vhsub", "s", int_arm_neon_vhsubs, 0>;
3283 defm VHSUBu : N3VInt_QHS<1, 0, 0b0010, 0, N3RegFrm,
3284 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3285 "vhsub", "u", int_arm_neon_vhsubu, 0>;
3286 // VQSUB : Vector Saturing Subtract
3287 defm VQSUBs : N3VInt_QHSD<0, 0, 0b0010, 1, N3RegFrm,
3288 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3289 "vqsub", "s", int_arm_neon_vqsubs, 0>;
3290 defm VQSUBu : N3VInt_QHSD<1, 0, 0b0010, 1, N3RegFrm,
3291 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3292 "vqsub", "u", int_arm_neon_vqsubu, 0>;
3293 // VSUBHN : Vector Subtract and Narrow Returning High Half (D = Q - Q)
3294 defm VSUBHN : N3VNInt_HSD<0,1,0b0110,0, "vsubhn", "i",
3295 int_arm_neon_vsubhn, 0>;
3296 // VRSUBHN : Vector Rounding Subtract and Narrow Returning High Half (D=Q-Q)
3297 defm VRSUBHN : N3VNInt_HSD<1,1,0b0110,0, "vrsubhn", "i",
3298 int_arm_neon_vrsubhn, 0>;
3300 // Vector Comparisons.
3302 // VCEQ : Vector Compare Equal
3303 defm VCEQ : N3V_QHS<1, 0, 0b1000, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3304 IIC_VSUBi4Q, "vceq", "i", NEONvceq, 1>;
3305 def VCEQfd : N3VD<0,0,0b00,0b1110,0, IIC_VBIND, "vceq", "f32", v2i32, v2f32,
3307 def VCEQfq : N3VQ<0,0,0b00,0b1110,0, IIC_VBINQ, "vceq", "f32", v4i32, v4f32,
3310 defm VCEQz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00010, 0, "vceq", "i",
3311 "$dst, $src, #0", NEONvceqz>;
3313 // VCGE : Vector Compare Greater Than or Equal
3314 defm VCGEs : N3V_QHS<0, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3315 IIC_VSUBi4Q, "vcge", "s", NEONvcge, 0>;
3316 defm VCGEu : N3V_QHS<1, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3317 IIC_VSUBi4Q, "vcge", "u", NEONvcgeu, 0>;
3318 def VCGEfd : N3VD<1,0,0b00,0b1110,0, IIC_VBIND, "vcge", "f32", v2i32, v2f32,
3320 def VCGEfq : N3VQ<1,0,0b00,0b1110,0, IIC_VBINQ, "vcge", "f32", v4i32, v4f32,
3323 defm VCGEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00001, 0, "vcge", "s",
3324 "$dst, $src, #0", NEONvcgez>;
3325 defm VCLEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00011, 0, "vcle", "s",
3326 "$dst, $src, #0", NEONvclez>;
3328 // VCGT : Vector Compare Greater Than
3329 defm VCGTs : N3V_QHS<0, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3330 IIC_VSUBi4Q, "vcgt", "s", NEONvcgt, 0>;
3331 defm VCGTu : N3V_QHS<1, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3332 IIC_VSUBi4Q, "vcgt", "u", NEONvcgtu, 0>;
3333 def VCGTfd : N3VD<1,0,0b10,0b1110,0, IIC_VBIND, "vcgt", "f32", v2i32, v2f32,
3335 def VCGTfq : N3VQ<1,0,0b10,0b1110,0, IIC_VBINQ, "vcgt", "f32", v4i32, v4f32,
3338 defm VCGTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00000, 0, "vcgt", "s",
3339 "$dst, $src, #0", NEONvcgtz>;
3340 defm VCLTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00100, 0, "vclt", "s",
3341 "$dst, $src, #0", NEONvcltz>;
3343 // VACGE : Vector Absolute Compare Greater Than or Equal (aka VCAGE)
3344 def VACGEd : N3VDInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacge",
3345 "f32", v2i32, v2f32, int_arm_neon_vacged, 0>;
3346 def VACGEq : N3VQInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacge",
3347 "f32", v4i32, v4f32, int_arm_neon_vacgeq, 0>;
3348 // VACGT : Vector Absolute Compare Greater Than (aka VCAGT)
3349 def VACGTd : N3VDInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacgt",
3350 "f32", v2i32, v2f32, int_arm_neon_vacgtd, 0>;
3351 def VACGTq : N3VQInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacgt",
3352 "f32", v4i32, v4f32, int_arm_neon_vacgtq, 0>;
3353 // VTST : Vector Test Bits
3354 defm VTST : N3V_QHS<0, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
3355 IIC_VBINi4Q, "vtst", "", NEONvtst, 1>;
3357 // Vector Bitwise Operations.
3359 def vnotd : PatFrag<(ops node:$in),
3360 (xor node:$in, (bitconvert (v8i8 NEONimmAllOnesV)))>;
3361 def vnotq : PatFrag<(ops node:$in),
3362 (xor node:$in, (bitconvert (v16i8 NEONimmAllOnesV)))>;
3365 // VAND : Vector Bitwise AND
3366 def VANDd : N3VDX<0, 0, 0b00, 0b0001, 1, IIC_VBINiD, "vand",
3367 v2i32, v2i32, and, 1>;
3368 def VANDq : N3VQX<0, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "vand",
3369 v4i32, v4i32, and, 1>;
3371 // VEOR : Vector Bitwise Exclusive OR
3372 def VEORd : N3VDX<1, 0, 0b00, 0b0001, 1, IIC_VBINiD, "veor",
3373 v2i32, v2i32, xor, 1>;
3374 def VEORq : N3VQX<1, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "veor",
3375 v4i32, v4i32, xor, 1>;
3377 // VORR : Vector Bitwise OR
3378 def VORRd : N3VDX<0, 0, 0b10, 0b0001, 1, IIC_VBINiD, "vorr",
3379 v2i32, v2i32, or, 1>;
3380 def VORRq : N3VQX<0, 0, 0b10, 0b0001, 1, IIC_VBINiQ, "vorr",
3381 v4i32, v4i32, or, 1>;
3383 def VORRiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 0, 1,
3384 (outs DPR:$Vd), (ins nModImm:$SIMM, DPR:$src),
3386 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
3388 (v4i16 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
3389 let Inst{9} = SIMM{9};
3392 def VORRiv2i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 0, 0, 1,
3393 (outs DPR:$Vd), (ins nModImm:$SIMM, DPR:$src),
3395 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
3397 (v2i32 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
3398 let Inst{10-9} = SIMM{10-9};
3401 def VORRiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 0, 1,
3402 (outs QPR:$Vd), (ins nModImm:$SIMM, QPR:$src),
3404 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
3406 (v8i16 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
3407 let Inst{9} = SIMM{9};
3410 def VORRiv4i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 1, 0, 1,
3411 (outs QPR:$Vd), (ins nModImm:$SIMM, QPR:$src),
3413 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
3415 (v4i32 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
3416 let Inst{10-9} = SIMM{10-9};
3420 // VBIC : Vector Bitwise Bit Clear (AND NOT)
3421 def VBICd : N3VX<0, 0, 0b01, 0b0001, 0, 1, (outs DPR:$dst),
3422 (ins DPR:$src1, DPR:$src2), N3RegFrm, IIC_VBINiD,
3423 "vbic", "$dst, $src1, $src2", "",
3424 [(set DPR:$dst, (v2i32 (and DPR:$src1,
3425 (vnotd DPR:$src2))))]>;
3426 def VBICq : N3VX<0, 0, 0b01, 0b0001, 1, 1, (outs QPR:$dst),
3427 (ins QPR:$src1, QPR:$src2), N3RegFrm, IIC_VBINiQ,
3428 "vbic", "$dst, $src1, $src2", "",
3429 [(set QPR:$dst, (v4i32 (and QPR:$src1,
3430 (vnotq QPR:$src2))))]>;
3432 def VBICiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 1, 1,
3433 (outs DPR:$Vd), (ins nModImm:$SIMM, DPR:$src),
3435 "vbic", "i16", "$Vd, $SIMM", "$src = $Vd",
3437 (v4i16 (NEONvbicImm DPR:$src, timm:$SIMM)))]> {
3438 let Inst{9} = SIMM{9};
3441 def VBICiv2i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 0, 1, 1,
3442 (outs DPR:$Vd), (ins nModImm:$SIMM, DPR:$src),
3444 "vbic", "i32", "$Vd, $SIMM", "$src = $Vd",
3446 (v2i32 (NEONvbicImm DPR:$src, timm:$SIMM)))]> {
3447 let Inst{10-9} = SIMM{10-9};
3450 def VBICiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 1, 1,
3451 (outs QPR:$Vd), (ins nModImm:$SIMM, QPR:$src),
3453 "vbic", "i16", "$Vd, $SIMM", "$src = $Vd",
3455 (v8i16 (NEONvbicImm QPR:$src, timm:$SIMM)))]> {
3456 let Inst{9} = SIMM{9};
3459 def VBICiv4i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 1, 1, 1,
3460 (outs QPR:$Vd), (ins nModImm:$SIMM, QPR:$src),
3462 "vbic", "i32", "$Vd, $SIMM", "$src = $Vd",
3464 (v4i32 (NEONvbicImm QPR:$src, timm:$SIMM)))]> {
3465 let Inst{10-9} = SIMM{10-9};
3468 // VORN : Vector Bitwise OR NOT
3469 def VORNd : N3VX<0, 0, 0b11, 0b0001, 0, 1, (outs DPR:$dst),
3470 (ins DPR:$src1, DPR:$src2), N3RegFrm, IIC_VBINiD,
3471 "vorn", "$dst, $src1, $src2", "",
3472 [(set DPR:$dst, (v2i32 (or DPR:$src1,
3473 (vnotd DPR:$src2))))]>;
3474 def VORNq : N3VX<0, 0, 0b11, 0b0001, 1, 1, (outs QPR:$dst),
3475 (ins QPR:$src1, QPR:$src2), N3RegFrm, IIC_VBINiQ,
3476 "vorn", "$dst, $src1, $src2", "",
3477 [(set QPR:$dst, (v4i32 (or QPR:$src1,
3478 (vnotq QPR:$src2))))]>;
3480 // VMVN : Vector Bitwise NOT (Immediate)
3482 let isReMaterializable = 1 in {
3484 def VMVNv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 1, 1, (outs DPR:$dst),
3485 (ins nModImm:$SIMM), IIC_VMOVImm,
3486 "vmvn", "i16", "$dst, $SIMM", "",
3487 [(set DPR:$dst, (v4i16 (NEONvmvnImm timm:$SIMM)))]> {
3488 let Inst{9} = SIMM{9};
3491 def VMVNv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 1, 1, (outs QPR:$dst),
3492 (ins nModImm:$SIMM), IIC_VMOVImm,
3493 "vmvn", "i16", "$dst, $SIMM", "",
3494 [(set QPR:$dst, (v8i16 (NEONvmvnImm timm:$SIMM)))]> {
3495 let Inst{9} = SIMM{9};
3498 def VMVNv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 1, 1, (outs DPR:$dst),
3499 (ins nModImm:$SIMM), IIC_VMOVImm,
3500 "vmvn", "i32", "$dst, $SIMM", "",
3501 [(set DPR:$dst, (v2i32 (NEONvmvnImm timm:$SIMM)))]> {
3502 let Inst{11-8} = SIMM{11-8};
3505 def VMVNv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 1, 1, (outs QPR:$dst),
3506 (ins nModImm:$SIMM), IIC_VMOVImm,
3507 "vmvn", "i32", "$dst, $SIMM", "",
3508 [(set QPR:$dst, (v4i32 (NEONvmvnImm timm:$SIMM)))]> {
3509 let Inst{11-8} = SIMM{11-8};
3513 // VMVN : Vector Bitwise NOT
3514 def VMVNd : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 0, 0,
3515 (outs DPR:$dst), (ins DPR:$src), IIC_VSUBiD,
3516 "vmvn", "$dst, $src", "",
3517 [(set DPR:$dst, (v2i32 (vnotd DPR:$src)))]>;
3518 def VMVNq : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 1, 0,
3519 (outs QPR:$dst), (ins QPR:$src), IIC_VSUBiD,
3520 "vmvn", "$dst, $src", "",
3521 [(set QPR:$dst, (v4i32 (vnotq QPR:$src)))]>;
3522 def : Pat<(v2i32 (vnotd DPR:$src)), (VMVNd DPR:$src)>;
3523 def : Pat<(v4i32 (vnotq QPR:$src)), (VMVNq QPR:$src)>;
3525 // VBSL : Vector Bitwise Select
3526 def VBSLd : N3VX<1, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
3527 (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
3528 N3RegFrm, IIC_VCNTiD,
3529 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3531 (v2i32 (or (and DPR:$Vn, DPR:$src1),
3532 (and DPR:$Vm, (vnotd DPR:$src1)))))]>;
3533 def VBSLq : N3VX<1, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
3534 (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
3535 N3RegFrm, IIC_VCNTiQ,
3536 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3538 (v4i32 (or (and QPR:$Vn, QPR:$src1),
3539 (and QPR:$Vm, (vnotq QPR:$src1)))))]>;
3541 // VBIF : Vector Bitwise Insert if False
3542 // like VBSL but with: "vbif $dst, $src3, $src1", "$src2 = $dst",
3543 // FIXME: This instruction's encoding MAY NOT BE correct.
3544 def VBIFd : N3VX<1, 0, 0b11, 0b0001, 0, 1,
3545 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
3546 N3RegFrm, IIC_VBINiD,
3547 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3548 [/* For disassembly only; pattern left blank */]>;
3549 def VBIFq : N3VX<1, 0, 0b11, 0b0001, 1, 1,
3550 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
3551 N3RegFrm, IIC_VBINiQ,
3552 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3553 [/* For disassembly only; pattern left blank */]>;
3555 // VBIT : Vector Bitwise Insert if True
3556 // like VBSL but with: "vbit $dst, $src2, $src1", "$src3 = $dst",
3557 // FIXME: This instruction's encoding MAY NOT BE correct.
3558 def VBITd : N3VX<1, 0, 0b10, 0b0001, 0, 1,
3559 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
3560 N3RegFrm, IIC_VBINiD,
3561 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3562 [/* For disassembly only; pattern left blank */]>;
3563 def VBITq : N3VX<1, 0, 0b10, 0b0001, 1, 1,
3564 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
3565 N3RegFrm, IIC_VBINiQ,
3566 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3567 [/* For disassembly only; pattern left blank */]>;
3569 // VBIT/VBIF are not yet implemented. The TwoAddress pass will not go looking
3570 // for equivalent operations with different register constraints; it just
3573 // Vector Absolute Differences.
3575 // VABD : Vector Absolute Difference
3576 defm VABDs : N3VInt_QHS<0, 0, 0b0111, 0, N3RegFrm,
3577 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3578 "vabd", "s", int_arm_neon_vabds, 1>;
3579 defm VABDu : N3VInt_QHS<1, 0, 0b0111, 0, N3RegFrm,
3580 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3581 "vabd", "u", int_arm_neon_vabdu, 1>;
3582 def VABDfd : N3VDInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBIND,
3583 "vabd", "f32", v2f32, v2f32, int_arm_neon_vabds, 1>;
3584 def VABDfq : N3VQInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBINQ,
3585 "vabd", "f32", v4f32, v4f32, int_arm_neon_vabds, 1>;
3587 // VABDL : Vector Absolute Difference Long (Q = | D - D |)
3588 defm VABDLs : N3VLIntExt_QHS<0,1,0b0111,0, IIC_VSUBi4Q,
3589 "vabdl", "s", int_arm_neon_vabds, zext, 1>;
3590 defm VABDLu : N3VLIntExt_QHS<1,1,0b0111,0, IIC_VSUBi4Q,
3591 "vabdl", "u", int_arm_neon_vabdu, zext, 1>;
3593 // VABA : Vector Absolute Difference and Accumulate
3594 defm VABAs : N3VIntOp_QHS<0,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
3595 "vaba", "s", int_arm_neon_vabds, add>;
3596 defm VABAu : N3VIntOp_QHS<1,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
3597 "vaba", "u", int_arm_neon_vabdu, add>;
3599 // VABAL : Vector Absolute Difference and Accumulate Long (Q += | D - D |)
3600 defm VABALs : N3VLIntExtOp_QHS<0,1,0b0101,0, IIC_VABAD,
3601 "vabal", "s", int_arm_neon_vabds, zext, add>;
3602 defm VABALu : N3VLIntExtOp_QHS<1,1,0b0101,0, IIC_VABAD,
3603 "vabal", "u", int_arm_neon_vabdu, zext, add>;
3605 // Vector Maximum and Minimum.
3607 // VMAX : Vector Maximum
3608 defm VMAXs : N3VInt_QHS<0, 0, 0b0110, 0, N3RegFrm,
3609 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3610 "vmax", "s", int_arm_neon_vmaxs, 1>;
3611 defm VMAXu : N3VInt_QHS<1, 0, 0b0110, 0, N3RegFrm,
3612 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3613 "vmax", "u", int_arm_neon_vmaxu, 1>;
3614 def VMAXfd : N3VDInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBIND,
3616 v2f32, v2f32, int_arm_neon_vmaxs, 1>;
3617 def VMAXfq : N3VQInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBINQ,
3619 v4f32, v4f32, int_arm_neon_vmaxs, 1>;
3621 // VMIN : Vector Minimum
3622 defm VMINs : N3VInt_QHS<0, 0, 0b0110, 1, N3RegFrm,
3623 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3624 "vmin", "s", int_arm_neon_vmins, 1>;
3625 defm VMINu : N3VInt_QHS<1, 0, 0b0110, 1, N3RegFrm,
3626 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3627 "vmin", "u", int_arm_neon_vminu, 1>;
3628 def VMINfd : N3VDInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBIND,
3630 v2f32, v2f32, int_arm_neon_vmins, 1>;
3631 def VMINfq : N3VQInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBINQ,
3633 v4f32, v4f32, int_arm_neon_vmins, 1>;
3635 // Vector Pairwise Operations.
3637 // VPADD : Vector Pairwise Add
3638 def VPADDi8 : N3VDInt<0, 0, 0b00, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
3640 v8i8, v8i8, int_arm_neon_vpadd, 0>;
3641 def VPADDi16 : N3VDInt<0, 0, 0b01, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
3643 v4i16, v4i16, int_arm_neon_vpadd, 0>;
3644 def VPADDi32 : N3VDInt<0, 0, 0b10, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
3646 v2i32, v2i32, int_arm_neon_vpadd, 0>;
3647 def VPADDf : N3VDInt<1, 0, 0b00, 0b1101, 0, N3RegFrm,
3648 IIC_VPBIND, "vpadd", "f32",
3649 v2f32, v2f32, int_arm_neon_vpadd, 0>;
3651 // VPADDL : Vector Pairwise Add Long
3652 defm VPADDLs : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00100, 0, "vpaddl", "s",
3653 int_arm_neon_vpaddls>;
3654 defm VPADDLu : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00101, 0, "vpaddl", "u",
3655 int_arm_neon_vpaddlu>;
3657 // VPADAL : Vector Pairwise Add and Accumulate Long
3658 defm VPADALs : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01100, 0, "vpadal", "s",
3659 int_arm_neon_vpadals>;
3660 defm VPADALu : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01101, 0, "vpadal", "u",
3661 int_arm_neon_vpadalu>;
3663 // VPMAX : Vector Pairwise Maximum
3664 def VPMAXs8 : N3VDInt<0, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3665 "s8", v8i8, v8i8, int_arm_neon_vpmaxs, 0>;
3666 def VPMAXs16 : N3VDInt<0, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3667 "s16", v4i16, v4i16, int_arm_neon_vpmaxs, 0>;
3668 def VPMAXs32 : N3VDInt<0, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3669 "s32", v2i32, v2i32, int_arm_neon_vpmaxs, 0>;
3670 def VPMAXu8 : N3VDInt<1, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3671 "u8", v8i8, v8i8, int_arm_neon_vpmaxu, 0>;
3672 def VPMAXu16 : N3VDInt<1, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3673 "u16", v4i16, v4i16, int_arm_neon_vpmaxu, 0>;
3674 def VPMAXu32 : N3VDInt<1, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3675 "u32", v2i32, v2i32, int_arm_neon_vpmaxu, 0>;
3676 def VPMAXf : N3VDInt<1, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmax",
3677 "f32", v2f32, v2f32, int_arm_neon_vpmaxs, 0>;
3679 // VPMIN : Vector Pairwise Minimum
3680 def VPMINs8 : N3VDInt<0, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3681 "s8", v8i8, v8i8, int_arm_neon_vpmins, 0>;
3682 def VPMINs16 : N3VDInt<0, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3683 "s16", v4i16, v4i16, int_arm_neon_vpmins, 0>;
3684 def VPMINs32 : N3VDInt<0, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3685 "s32", v2i32, v2i32, int_arm_neon_vpmins, 0>;
3686 def VPMINu8 : N3VDInt<1, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3687 "u8", v8i8, v8i8, int_arm_neon_vpminu, 0>;
3688 def VPMINu16 : N3VDInt<1, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3689 "u16", v4i16, v4i16, int_arm_neon_vpminu, 0>;
3690 def VPMINu32 : N3VDInt<1, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3691 "u32", v2i32, v2i32, int_arm_neon_vpminu, 0>;
3692 def VPMINf : N3VDInt<1, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmin",
3693 "f32", v2f32, v2f32, int_arm_neon_vpmins, 0>;
3695 // Vector Reciprocal and Reciprocal Square Root Estimate and Step.
3697 // VRECPE : Vector Reciprocal Estimate
3698 def VRECPEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
3699 IIC_VUNAD, "vrecpe", "u32",
3700 v2i32, v2i32, int_arm_neon_vrecpe>;
3701 def VRECPEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
3702 IIC_VUNAQ, "vrecpe", "u32",
3703 v4i32, v4i32, int_arm_neon_vrecpe>;
3704 def VRECPEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
3705 IIC_VUNAD, "vrecpe", "f32",
3706 v2f32, v2f32, int_arm_neon_vrecpe>;
3707 def VRECPEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
3708 IIC_VUNAQ, "vrecpe", "f32",
3709 v4f32, v4f32, int_arm_neon_vrecpe>;
3711 // VRECPS : Vector Reciprocal Step
3712 def VRECPSfd : N3VDInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
3713 IIC_VRECSD, "vrecps", "f32",
3714 v2f32, v2f32, int_arm_neon_vrecps, 1>;
3715 def VRECPSfq : N3VQInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
3716 IIC_VRECSQ, "vrecps", "f32",
3717 v4f32, v4f32, int_arm_neon_vrecps, 1>;
3719 // VRSQRTE : Vector Reciprocal Square Root Estimate
3720 def VRSQRTEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
3721 IIC_VUNAD, "vrsqrte", "u32",
3722 v2i32, v2i32, int_arm_neon_vrsqrte>;
3723 def VRSQRTEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
3724 IIC_VUNAQ, "vrsqrte", "u32",
3725 v4i32, v4i32, int_arm_neon_vrsqrte>;
3726 def VRSQRTEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
3727 IIC_VUNAD, "vrsqrte", "f32",
3728 v2f32, v2f32, int_arm_neon_vrsqrte>;
3729 def VRSQRTEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
3730 IIC_VUNAQ, "vrsqrte", "f32",
3731 v4f32, v4f32, int_arm_neon_vrsqrte>;
3733 // VRSQRTS : Vector Reciprocal Square Root Step
3734 def VRSQRTSfd : N3VDInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
3735 IIC_VRECSD, "vrsqrts", "f32",
3736 v2f32, v2f32, int_arm_neon_vrsqrts, 1>;
3737 def VRSQRTSfq : N3VQInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
3738 IIC_VRECSQ, "vrsqrts", "f32",
3739 v4f32, v4f32, int_arm_neon_vrsqrts, 1>;
3743 // VSHL : Vector Shift
3744 defm VSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 0, N3RegVShFrm,
3745 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
3746 "vshl", "s", int_arm_neon_vshifts>;
3747 defm VSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 0, N3RegVShFrm,
3748 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
3749 "vshl", "u", int_arm_neon_vshiftu>;
3750 // VSHL : Vector Shift Left (Immediate)
3751 defm VSHLi : N2VSh_QHSD<0, 1, 0b0101, 1, IIC_VSHLiD, "vshl", "i", NEONvshl,
3753 // VSHR : Vector Shift Right (Immediate)
3754 defm VSHRs : N2VSh_QHSD<0, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "s", NEONvshrs,
3756 defm VSHRu : N2VSh_QHSD<1, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "u", NEONvshru,
3759 // VSHLL : Vector Shift Left Long
3760 defm VSHLLs : N2VLSh_QHS<0, 1, 0b1010, 0, 0, 1, "vshll", "s", NEONvshlls>;
3761 defm VSHLLu : N2VLSh_QHS<1, 1, 0b1010, 0, 0, 1, "vshll", "u", NEONvshllu>;
3763 // VSHLL : Vector Shift Left Long (with maximum shift count)
3764 class N2VLShMax<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
3765 bit op6, bit op4, string OpcodeStr, string Dt, ValueType ResTy,
3766 ValueType OpTy, SDNode OpNode>
3767 : N2VLSh<op24, op23, op11_8, op7, op6, op4, OpcodeStr, Dt,
3768 ResTy, OpTy, OpNode> {
3769 let Inst{21-16} = op21_16;
3771 def VSHLLi8 : N2VLShMax<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll", "i8",
3772 v8i16, v8i8, NEONvshlli>;
3773 def VSHLLi16 : N2VLShMax<1, 1, 0b110110, 0b0011, 0, 0, 0, "vshll", "i16",
3774 v4i32, v4i16, NEONvshlli>;
3775 def VSHLLi32 : N2VLShMax<1, 1, 0b111010, 0b0011, 0, 0, 0, "vshll", "i32",
3776 v2i64, v2i32, NEONvshlli>;
3778 // VSHRN : Vector Shift Right and Narrow
3779 defm VSHRN : N2VNSh_HSD<0,1,0b1000,0,0,1, IIC_VSHLiD, "vshrn", "i",
3782 // VRSHL : Vector Rounding Shift
3783 defm VRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 0, N3RegVShFrm,
3784 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3785 "vrshl", "s", int_arm_neon_vrshifts>;
3786 defm VRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 0, N3RegVShFrm,
3787 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3788 "vrshl", "u", int_arm_neon_vrshiftu>;
3789 // VRSHR : Vector Rounding Shift Right
3790 defm VRSHRs : N2VSh_QHSD<0,1,0b0010,1, IIC_VSHLi4D, "vrshr", "s", NEONvrshrs,
3792 defm VRSHRu : N2VSh_QHSD<1,1,0b0010,1, IIC_VSHLi4D, "vrshr", "u", NEONvrshru,
3795 // VRSHRN : Vector Rounding Shift Right and Narrow
3796 defm VRSHRN : N2VNSh_HSD<0, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vrshrn", "i",
3799 // VQSHL : Vector Saturating Shift
3800 defm VQSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 1, N3RegVShFrm,
3801 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3802 "vqshl", "s", int_arm_neon_vqshifts>;
3803 defm VQSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 1, N3RegVShFrm,
3804 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3805 "vqshl", "u", int_arm_neon_vqshiftu>;
3806 // VQSHL : Vector Saturating Shift Left (Immediate)
3807 defm VQSHLsi : N2VSh_QHSD<0,1,0b0111,1, IIC_VSHLi4D, "vqshl", "s",NEONvqshls,
3809 defm VQSHLui : N2VSh_QHSD<1,1,0b0111,1, IIC_VSHLi4D, "vqshl", "u",NEONvqshlu,
3811 // VQSHLU : Vector Saturating Shift Left (Immediate, Unsigned)
3812 defm VQSHLsu : N2VSh_QHSD<1,1,0b0110,1, IIC_VSHLi4D,"vqshlu","s",NEONvqshlsu,
3815 // VQSHRN : Vector Saturating Shift Right and Narrow
3816 defm VQSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "s",
3818 defm VQSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "u",
3821 // VQSHRUN : Vector Saturating Shift Right and Narrow (Unsigned)
3822 defm VQSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 0, 1, IIC_VSHLi4D, "vqshrun", "s",
3825 // VQRSHL : Vector Saturating Rounding Shift
3826 defm VQRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 1, N3RegVShFrm,
3827 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3828 "vqrshl", "s", int_arm_neon_vqrshifts>;
3829 defm VQRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 1, N3RegVShFrm,
3830 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3831 "vqrshl", "u", int_arm_neon_vqrshiftu>;
3833 // VQRSHRN : Vector Saturating Rounding Shift Right and Narrow
3834 defm VQRSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "s",
3836 defm VQRSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "u",
3839 // VQRSHRUN : Vector Saturating Rounding Shift Right and Narrow (Unsigned)
3840 defm VQRSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vqrshrun", "s",
3843 // VSRA : Vector Shift Right and Accumulate
3844 defm VSRAs : N2VShAdd_QHSD<0, 1, 0b0001, 1, "vsra", "s", NEONvshrs>;
3845 defm VSRAu : N2VShAdd_QHSD<1, 1, 0b0001, 1, "vsra", "u", NEONvshru>;
3846 // VRSRA : Vector Rounding Shift Right and Accumulate
3847 defm VRSRAs : N2VShAdd_QHSD<0, 1, 0b0011, 1, "vrsra", "s", NEONvrshrs>;
3848 defm VRSRAu : N2VShAdd_QHSD<1, 1, 0b0011, 1, "vrsra", "u", NEONvrshru>;
3850 // VSLI : Vector Shift Left and Insert
3851 defm VSLI : N2VShIns_QHSD<1, 1, 0b0101, 1, "vsli", NEONvsli, N2RegVShLFrm>;
3852 // VSRI : Vector Shift Right and Insert
3853 defm VSRI : N2VShIns_QHSD<1, 1, 0b0100, 1, "vsri", NEONvsri, N2RegVShRFrm>;
3855 // Vector Absolute and Saturating Absolute.
3857 // VABS : Vector Absolute Value
3858 defm VABS : N2VInt_QHS<0b11, 0b11, 0b01, 0b00110, 0,
3859 IIC_VUNAiD, IIC_VUNAiQ, "vabs", "s",
3861 def VABSfd : N2VDInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
3862 IIC_VUNAD, "vabs", "f32",
3863 v2f32, v2f32, int_arm_neon_vabs>;
3864 def VABSfq : N2VQInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
3865 IIC_VUNAQ, "vabs", "f32",
3866 v4f32, v4f32, int_arm_neon_vabs>;
3868 // VQABS : Vector Saturating Absolute Value
3869 defm VQABS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01110, 0,
3870 IIC_VQUNAiD, IIC_VQUNAiQ, "vqabs", "s",
3871 int_arm_neon_vqabs>;
3875 def vnegd : PatFrag<(ops node:$in),
3876 (sub (bitconvert (v2i32 NEONimmAllZerosV)), node:$in)>;
3877 def vnegq : PatFrag<(ops node:$in),
3878 (sub (bitconvert (v4i32 NEONimmAllZerosV)), node:$in)>;
3880 class VNEGD<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
3881 : N2V<0b11, 0b11, size, 0b01, 0b00111, 0, 0, (outs DPR:$dst), (ins DPR:$src),
3882 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
3883 [(set DPR:$dst, (Ty (vnegd DPR:$src)))]>;
3884 class VNEGQ<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
3885 : N2V<0b11, 0b11, size, 0b01, 0b00111, 1, 0, (outs QPR:$dst), (ins QPR:$src),
3886 IIC_VSHLiQ, OpcodeStr, Dt, "$dst, $src", "",
3887 [(set QPR:$dst, (Ty (vnegq QPR:$src)))]>;
3889 // VNEG : Vector Negate (integer)
3890 def VNEGs8d : VNEGD<0b00, "vneg", "s8", v8i8>;
3891 def VNEGs16d : VNEGD<0b01, "vneg", "s16", v4i16>;
3892 def VNEGs32d : VNEGD<0b10, "vneg", "s32", v2i32>;
3893 def VNEGs8q : VNEGQ<0b00, "vneg", "s8", v16i8>;
3894 def VNEGs16q : VNEGQ<0b01, "vneg", "s16", v8i16>;
3895 def VNEGs32q : VNEGQ<0b10, "vneg", "s32", v4i32>;
3897 // VNEG : Vector Negate (floating-point)
3898 def VNEGfd : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
3899 (outs DPR:$dst), (ins DPR:$src), IIC_VUNAD,
3900 "vneg", "f32", "$dst, $src", "",
3901 [(set DPR:$dst, (v2f32 (fneg DPR:$src)))]>;
3902 def VNEGf32q : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 1, 0,
3903 (outs QPR:$dst), (ins QPR:$src), IIC_VUNAQ,
3904 "vneg", "f32", "$dst, $src", "",
3905 [(set QPR:$dst, (v4f32 (fneg QPR:$src)))]>;
3907 def : Pat<(v8i8 (vnegd DPR:$src)), (VNEGs8d DPR:$src)>;
3908 def : Pat<(v4i16 (vnegd DPR:$src)), (VNEGs16d DPR:$src)>;
3909 def : Pat<(v2i32 (vnegd DPR:$src)), (VNEGs32d DPR:$src)>;
3910 def : Pat<(v16i8 (vnegq QPR:$src)), (VNEGs8q QPR:$src)>;
3911 def : Pat<(v8i16 (vnegq QPR:$src)), (VNEGs16q QPR:$src)>;
3912 def : Pat<(v4i32 (vnegq QPR:$src)), (VNEGs32q QPR:$src)>;
3914 // VQNEG : Vector Saturating Negate
3915 defm VQNEG : N2VInt_QHS<0b11, 0b11, 0b00, 0b01111, 0,
3916 IIC_VQUNAiD, IIC_VQUNAiQ, "vqneg", "s",
3917 int_arm_neon_vqneg>;
3919 // Vector Bit Counting Operations.
3921 // VCLS : Vector Count Leading Sign Bits
3922 defm VCLS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01000, 0,
3923 IIC_VCNTiD, IIC_VCNTiQ, "vcls", "s",
3925 // VCLZ : Vector Count Leading Zeros
3926 defm VCLZ : N2VInt_QHS<0b11, 0b11, 0b00, 0b01001, 0,
3927 IIC_VCNTiD, IIC_VCNTiQ, "vclz", "i",
3929 // VCNT : Vector Count One Bits
3930 def VCNTd : N2VDInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
3931 IIC_VCNTiD, "vcnt", "8",
3932 v8i8, v8i8, int_arm_neon_vcnt>;
3933 def VCNTq : N2VQInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
3934 IIC_VCNTiQ, "vcnt", "8",
3935 v16i8, v16i8, int_arm_neon_vcnt>;
3937 // Vector Swap -- for disassembly only.
3938 def VSWPd : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 0, 0,
3939 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
3940 "vswp", "$dst, $src", "", []>;
3941 def VSWPq : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 1, 0,
3942 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
3943 "vswp", "$dst, $src", "", []>;
3945 // Vector Move Operations.
3947 // VMOV : Vector Move (Register)
3949 let neverHasSideEffects = 1 in {
3950 def VMOVDneon: N3VX<0, 0, 0b10, 0b0001, 0, 1, (outs DPR:$Vd), (ins DPR:$Vm),
3951 N3RegFrm, IIC_VMOV, "vmov", "$Vd, $Vm", "", []> {
3952 let Vn{4-0} = Vm{4-0};
3954 def VMOVQ : N3VX<0, 0, 0b10, 0b0001, 1, 1, (outs QPR:$Vd), (ins QPR:$Vm),
3955 N3RegFrm, IIC_VMOV, "vmov", "$Vd, $Vm", "", []> {
3956 let Vn{4-0} = Vm{4-0};
3959 // Pseudo vector move instructions for QQ and QQQQ registers. This should
3960 // be expanded after register allocation is completed.
3961 def VMOVQQ : PseudoInst<(outs QQPR:$dst), (ins QQPR:$src),
3964 def VMOVQQQQ : PseudoInst<(outs QQQQPR:$dst), (ins QQQQPR:$src),
3966 } // neverHasSideEffects
3968 // VMOV : Vector Move (Immediate)
3970 let isReMaterializable = 1 in {
3971 def VMOVv8i8 : N1ModImm<1, 0b000, 0b1110, 0, 0, 0, 1, (outs DPR:$dst),
3972 (ins nModImm:$SIMM), IIC_VMOVImm,
3973 "vmov", "i8", "$dst, $SIMM", "",
3974 [(set DPR:$dst, (v8i8 (NEONvmovImm timm:$SIMM)))]>;
3975 def VMOVv16i8 : N1ModImm<1, 0b000, 0b1110, 0, 1, 0, 1, (outs QPR:$dst),
3976 (ins nModImm:$SIMM), IIC_VMOVImm,
3977 "vmov", "i8", "$dst, $SIMM", "",
3978 [(set QPR:$dst, (v16i8 (NEONvmovImm timm:$SIMM)))]>;
3980 def VMOVv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 0, 1, (outs DPR:$dst),
3981 (ins nModImm:$SIMM), IIC_VMOVImm,
3982 "vmov", "i16", "$dst, $SIMM", "",
3983 [(set DPR:$dst, (v4i16 (NEONvmovImm timm:$SIMM)))]> {
3984 let Inst{9} = SIMM{9};
3987 def VMOVv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 0, 1, (outs QPR:$dst),
3988 (ins nModImm:$SIMM), IIC_VMOVImm,
3989 "vmov", "i16", "$dst, $SIMM", "",
3990 [(set QPR:$dst, (v8i16 (NEONvmovImm timm:$SIMM)))]> {
3991 let Inst{9} = SIMM{9};
3994 def VMOVv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 0, 1, (outs DPR:$dst),
3995 (ins nModImm:$SIMM), IIC_VMOVImm,
3996 "vmov", "i32", "$dst, $SIMM", "",
3997 [(set DPR:$dst, (v2i32 (NEONvmovImm timm:$SIMM)))]> {
3998 let Inst{11-8} = SIMM{11-8};
4001 def VMOVv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 0, 1, (outs QPR:$dst),
4002 (ins nModImm:$SIMM), IIC_VMOVImm,
4003 "vmov", "i32", "$dst, $SIMM", "",
4004 [(set QPR:$dst, (v4i32 (NEONvmovImm timm:$SIMM)))]> {
4005 let Inst{11-8} = SIMM{11-8};
4008 def VMOVv1i64 : N1ModImm<1, 0b000, 0b1110, 0, 0, 1, 1, (outs DPR:$dst),
4009 (ins nModImm:$SIMM), IIC_VMOVImm,
4010 "vmov", "i64", "$dst, $SIMM", "",
4011 [(set DPR:$dst, (v1i64 (NEONvmovImm timm:$SIMM)))]>;
4012 def VMOVv2i64 : N1ModImm<1, 0b000, 0b1110, 0, 1, 1, 1, (outs QPR:$dst),
4013 (ins nModImm:$SIMM), IIC_VMOVImm,
4014 "vmov", "i64", "$dst, $SIMM", "",
4015 [(set QPR:$dst, (v2i64 (NEONvmovImm timm:$SIMM)))]>;
4016 } // isReMaterializable
4018 // VMOV : Vector Get Lane (move scalar to ARM core register)
4020 def VGETLNs8 : NVGetLane<{1,1,1,0,0,1,?,1}, 0b1011, {?,?},
4021 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
4022 IIC_VMOVSI, "vmov", "s8", "$R, $V[$lane]",
4023 [(set GPR:$R, (NEONvgetlanes (v8i8 DPR:$V),
4025 let Inst{21} = lane{2};
4026 let Inst{6-5} = lane{1-0};
4028 def VGETLNs16 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, {?,1},
4029 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
4030 IIC_VMOVSI, "vmov", "s16", "$R, $V[$lane]",
4031 [(set GPR:$R, (NEONvgetlanes (v4i16 DPR:$V),
4033 let Inst{21} = lane{1};
4034 let Inst{6} = lane{0};
4036 def VGETLNu8 : NVGetLane<{1,1,1,0,1,1,?,1}, 0b1011, {?,?},
4037 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
4038 IIC_VMOVSI, "vmov", "u8", "$R, $V[$lane]",
4039 [(set GPR:$R, (NEONvgetlaneu (v8i8 DPR:$V),
4041 let Inst{21} = lane{2};
4042 let Inst{6-5} = lane{1-0};
4044 def VGETLNu16 : NVGetLane<{1,1,1,0,1,0,?,1}, 0b1011, {?,1},
4045 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
4046 IIC_VMOVSI, "vmov", "u16", "$R, $V[$lane]",
4047 [(set GPR:$R, (NEONvgetlaneu (v4i16 DPR:$V),
4049 let Inst{21} = lane{1};
4050 let Inst{6} = lane{0};
4052 def VGETLNi32 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, 0b00,
4053 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
4054 IIC_VMOVSI, "vmov", "32", "$R, $V[$lane]",
4055 [(set GPR:$R, (extractelt (v2i32 DPR:$V),
4057 let Inst{21} = lane{0};
4059 // def VGETLNf32: see FMRDH and FMRDL in ARMInstrVFP.td
4060 def : Pat<(NEONvgetlanes (v16i8 QPR:$src), imm:$lane),
4061 (VGETLNs8 (v8i8 (EXTRACT_SUBREG QPR:$src,
4062 (DSubReg_i8_reg imm:$lane))),
4063 (SubReg_i8_lane imm:$lane))>;
4064 def : Pat<(NEONvgetlanes (v8i16 QPR:$src), imm:$lane),
4065 (VGETLNs16 (v4i16 (EXTRACT_SUBREG QPR:$src,
4066 (DSubReg_i16_reg imm:$lane))),
4067 (SubReg_i16_lane imm:$lane))>;
4068 def : Pat<(NEONvgetlaneu (v16i8 QPR:$src), imm:$lane),
4069 (VGETLNu8 (v8i8 (EXTRACT_SUBREG QPR:$src,
4070 (DSubReg_i8_reg imm:$lane))),
4071 (SubReg_i8_lane imm:$lane))>;
4072 def : Pat<(NEONvgetlaneu (v8i16 QPR:$src), imm:$lane),
4073 (VGETLNu16 (v4i16 (EXTRACT_SUBREG QPR:$src,
4074 (DSubReg_i16_reg imm:$lane))),
4075 (SubReg_i16_lane imm:$lane))>;
4076 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
4077 (VGETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src,
4078 (DSubReg_i32_reg imm:$lane))),
4079 (SubReg_i32_lane imm:$lane))>;
4080 def : Pat<(extractelt (v2f32 DPR:$src1), imm:$src2),
4081 (EXTRACT_SUBREG (v2f32 (COPY_TO_REGCLASS (v2f32 DPR:$src1),DPR_VFP2)),
4082 (SSubReg_f32_reg imm:$src2))>;
4083 def : Pat<(extractelt (v4f32 QPR:$src1), imm:$src2),
4084 (EXTRACT_SUBREG (v4f32 (COPY_TO_REGCLASS (v4f32 QPR:$src1),QPR_VFP2)),
4085 (SSubReg_f32_reg imm:$src2))>;
4086 //def : Pat<(extractelt (v2i64 QPR:$src1), imm:$src2),
4087 // (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
4088 def : Pat<(extractelt (v2f64 QPR:$src1), imm:$src2),
4089 (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
4092 // VMOV : Vector Set Lane (move ARM core register to scalar)
4094 let Constraints = "$src1 = $V" in {
4095 def VSETLNi8 : NVSetLane<{1,1,1,0,0,1,?,0}, 0b1011, {?,?}, (outs DPR:$V),
4096 (ins DPR:$src1, GPR:$R, nohash_imm:$lane),
4097 IIC_VMOVISL, "vmov", "8", "$V[$lane], $R",
4098 [(set DPR:$V, (vector_insert (v8i8 DPR:$src1),
4099 GPR:$R, imm:$lane))]> {
4100 let Inst{21} = lane{2};
4101 let Inst{6-5} = lane{1-0};
4103 def VSETLNi16 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, {?,1}, (outs DPR:$V),
4104 (ins DPR:$src1, GPR:$R, nohash_imm:$lane),
4105 IIC_VMOVISL, "vmov", "16", "$V[$lane], $R",
4106 [(set DPR:$V, (vector_insert (v4i16 DPR:$src1),
4107 GPR:$R, imm:$lane))]> {
4108 let Inst{21} = lane{1};
4109 let Inst{6} = lane{0};
4111 def VSETLNi32 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, 0b00, (outs DPR:$V),
4112 (ins DPR:$src1, GPR:$R, nohash_imm:$lane),
4113 IIC_VMOVISL, "vmov", "32", "$V[$lane], $R",
4114 [(set DPR:$V, (insertelt (v2i32 DPR:$src1),
4115 GPR:$R, imm:$lane))]> {
4116 let Inst{21} = lane{0};
4119 def : Pat<(vector_insert (v16i8 QPR:$src1), GPR:$src2, imm:$lane),
4120 (v16i8 (INSERT_SUBREG QPR:$src1,
4121 (v8i8 (VSETLNi8 (v8i8 (EXTRACT_SUBREG QPR:$src1,
4122 (DSubReg_i8_reg imm:$lane))),
4123 GPR:$src2, (SubReg_i8_lane imm:$lane))),
4124 (DSubReg_i8_reg imm:$lane)))>;
4125 def : Pat<(vector_insert (v8i16 QPR:$src1), GPR:$src2, imm:$lane),
4126 (v8i16 (INSERT_SUBREG QPR:$src1,
4127 (v4i16 (VSETLNi16 (v4i16 (EXTRACT_SUBREG QPR:$src1,
4128 (DSubReg_i16_reg imm:$lane))),
4129 GPR:$src2, (SubReg_i16_lane imm:$lane))),
4130 (DSubReg_i16_reg imm:$lane)))>;
4131 def : Pat<(insertelt (v4i32 QPR:$src1), GPR:$src2, imm:$lane),
4132 (v4i32 (INSERT_SUBREG QPR:$src1,
4133 (v2i32 (VSETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src1,
4134 (DSubReg_i32_reg imm:$lane))),
4135 GPR:$src2, (SubReg_i32_lane imm:$lane))),
4136 (DSubReg_i32_reg imm:$lane)))>;
4138 def : Pat<(v2f32 (insertelt DPR:$src1, SPR:$src2, imm:$src3)),
4139 (INSERT_SUBREG (v2f32 (COPY_TO_REGCLASS DPR:$src1, DPR_VFP2)),
4140 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
4141 def : Pat<(v4f32 (insertelt QPR:$src1, SPR:$src2, imm:$src3)),
4142 (INSERT_SUBREG (v4f32 (COPY_TO_REGCLASS QPR:$src1, QPR_VFP2)),
4143 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
4145 //def : Pat<(v2i64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
4146 // (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
4147 def : Pat<(v2f64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
4148 (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
4150 def : Pat<(v2f32 (scalar_to_vector SPR:$src)),
4151 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
4152 def : Pat<(v2f64 (scalar_to_vector (f64 DPR:$src))),
4153 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
4154 def : Pat<(v4f32 (scalar_to_vector SPR:$src)),
4155 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
4157 def : Pat<(v8i8 (scalar_to_vector GPR:$src)),
4158 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4159 def : Pat<(v4i16 (scalar_to_vector GPR:$src)),
4160 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4161 def : Pat<(v2i32 (scalar_to_vector GPR:$src)),
4162 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4164 def : Pat<(v16i8 (scalar_to_vector GPR:$src)),
4165 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4166 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4168 def : Pat<(v8i16 (scalar_to_vector GPR:$src)),
4169 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
4170 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4172 def : Pat<(v4i32 (scalar_to_vector GPR:$src)),
4173 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
4174 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4177 // VDUP : Vector Duplicate (from ARM core register to all elements)
4179 class VDUPD<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
4180 : NVDup<opcod1, 0b1011, opcod3, (outs DPR:$dst), (ins GPR:$src),
4181 IIC_VMOVIS, "vdup", Dt, "$dst, $src",
4182 [(set DPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
4183 class VDUPQ<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
4184 : NVDup<opcod1, 0b1011, opcod3, (outs QPR:$dst), (ins GPR:$src),
4185 IIC_VMOVIS, "vdup", Dt, "$dst, $src",
4186 [(set QPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
4188 def VDUP8d : VDUPD<0b11101100, 0b00, "8", v8i8>;
4189 def VDUP16d : VDUPD<0b11101000, 0b01, "16", v4i16>;
4190 def VDUP32d : VDUPD<0b11101000, 0b00, "32", v2i32>;
4191 def VDUP8q : VDUPQ<0b11101110, 0b00, "8", v16i8>;
4192 def VDUP16q : VDUPQ<0b11101010, 0b01, "16", v8i16>;
4193 def VDUP32q : VDUPQ<0b11101010, 0b00, "32", v4i32>;
4195 def VDUPfd : NVDup<0b11101000, 0b1011, 0b00, (outs DPR:$dst), (ins GPR:$src),
4196 IIC_VMOVIS, "vdup", "32", "$dst, $src",
4197 [(set DPR:$dst, (v2f32 (NEONvdup
4198 (f32 (bitconvert GPR:$src)))))]>;
4199 def VDUPfq : NVDup<0b11101010, 0b1011, 0b00, (outs QPR:$dst), (ins GPR:$src),
4200 IIC_VMOVIS, "vdup", "32", "$dst, $src",
4201 [(set QPR:$dst, (v4f32 (NEONvdup
4202 (f32 (bitconvert GPR:$src)))))]>;
4204 // VDUP : Vector Duplicate Lane (from scalar to all elements)
4206 class VDUPLND<bits<4> op19_16, string OpcodeStr, string Dt,
4208 : NVDupLane<op19_16, 0, (outs DPR:$dst), (ins DPR:$src, nohash_imm:$lane),
4209 IIC_VMOVD, OpcodeStr, Dt, "$dst, $src[$lane]",
4210 [(set DPR:$dst, (Ty (NEONvduplane (Ty DPR:$src), imm:$lane)))]>;
4212 class VDUPLNQ<bits<4> op19_16, string OpcodeStr, string Dt,
4213 ValueType ResTy, ValueType OpTy>
4214 : NVDupLane<op19_16, 1, (outs QPR:$dst), (ins DPR:$src, nohash_imm:$lane),
4215 IIC_VMOVQ, OpcodeStr, Dt, "$dst, $src[$lane]",
4216 [(set QPR:$dst, (ResTy (NEONvduplane (OpTy DPR:$src),
4219 // Inst{19-16} is partially specified depending on the element size.
4221 def VDUPLN8d : VDUPLND<{?,?,?,1}, "vdup", "8", v8i8> {
4222 let Inst{19-17} = lane{2-0};
4224 def VDUPLN16d : VDUPLND<{?,?,1,0}, "vdup", "16", v4i16> {
4225 let Inst{19-18} = lane{1-0};
4227 def VDUPLN32d : VDUPLND<{?,1,0,0}, "vdup", "32", v2i32> {
4228 let Inst{19} = lane{0};
4230 def VDUPLNfd : VDUPLND<{?,1,0,0}, "vdup", "32", v2f32> {
4231 let Inst{19} = lane{0};
4233 def VDUPLN8q : VDUPLNQ<{?,?,?,1}, "vdup", "8", v16i8, v8i8> {
4234 let Inst{19-17} = lane{2-0};
4236 def VDUPLN16q : VDUPLNQ<{?,?,1,0}, "vdup", "16", v8i16, v4i16> {
4237 let Inst{19-18} = lane{1-0};
4239 def VDUPLN32q : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4i32, v2i32> {
4240 let Inst{19} = lane{0};
4242 def VDUPLNfq : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4f32, v2f32> {
4243 let Inst{19} = lane{0};
4246 def : Pat<(v16i8 (NEONvduplane (v16i8 QPR:$src), imm:$lane)),
4247 (v16i8 (VDUPLN8q (v8i8 (EXTRACT_SUBREG QPR:$src,
4248 (DSubReg_i8_reg imm:$lane))),
4249 (SubReg_i8_lane imm:$lane)))>;
4250 def : Pat<(v8i16 (NEONvduplane (v8i16 QPR:$src), imm:$lane)),
4251 (v8i16 (VDUPLN16q (v4i16 (EXTRACT_SUBREG QPR:$src,
4252 (DSubReg_i16_reg imm:$lane))),
4253 (SubReg_i16_lane imm:$lane)))>;
4254 def : Pat<(v4i32 (NEONvduplane (v4i32 QPR:$src), imm:$lane)),
4255 (v4i32 (VDUPLN32q (v2i32 (EXTRACT_SUBREG QPR:$src,
4256 (DSubReg_i32_reg imm:$lane))),
4257 (SubReg_i32_lane imm:$lane)))>;
4258 def : Pat<(v4f32 (NEONvduplane (v4f32 QPR:$src), imm:$lane)),
4259 (v4f32 (VDUPLNfq (v2f32 (EXTRACT_SUBREG QPR:$src,
4260 (DSubReg_i32_reg imm:$lane))),
4261 (SubReg_i32_lane imm:$lane)))>;
4263 def VDUPfdf : PseudoNeonI<(outs DPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
4264 [(set DPR:$dst, (v2f32 (NEONvdup (f32 SPR:$src))))]>;
4265 def VDUPfqf : PseudoNeonI<(outs QPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
4266 [(set QPR:$dst, (v4f32 (NEONvdup (f32 SPR:$src))))]>;
4268 // VMOVN : Vector Narrowing Move
4269 defm VMOVN : N2VN_HSD<0b11,0b11,0b10,0b00100,0,0, IIC_VMOVN,
4270 "vmovn", "i", trunc>;
4271 // VQMOVN : Vector Saturating Narrowing Move
4272 defm VQMOVNs : N2VNInt_HSD<0b11,0b11,0b10,0b00101,0,0, IIC_VQUNAiD,
4273 "vqmovn", "s", int_arm_neon_vqmovns>;
4274 defm VQMOVNu : N2VNInt_HSD<0b11,0b11,0b10,0b00101,1,0, IIC_VQUNAiD,
4275 "vqmovn", "u", int_arm_neon_vqmovnu>;
4276 defm VQMOVNsu : N2VNInt_HSD<0b11,0b11,0b10,0b00100,1,0, IIC_VQUNAiD,
4277 "vqmovun", "s", int_arm_neon_vqmovnsu>;
4278 // VMOVL : Vector Lengthening Move
4279 defm VMOVLs : N2VL_QHS<0b01,0b10100,0,1, "vmovl", "s", sext>;
4280 defm VMOVLu : N2VL_QHS<0b11,0b10100,0,1, "vmovl", "u", zext>;
4282 // Vector Conversions.
4284 // VCVT : Vector Convert Between Floating-Point and Integers
4285 def VCVTf2sd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
4286 v2i32, v2f32, fp_to_sint>;
4287 def VCVTf2ud : N2VD<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
4288 v2i32, v2f32, fp_to_uint>;
4289 def VCVTs2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
4290 v2f32, v2i32, sint_to_fp>;
4291 def VCVTu2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
4292 v2f32, v2i32, uint_to_fp>;
4294 def VCVTf2sq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
4295 v4i32, v4f32, fp_to_sint>;
4296 def VCVTf2uq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
4297 v4i32, v4f32, fp_to_uint>;
4298 def VCVTs2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
4299 v4f32, v4i32, sint_to_fp>;
4300 def VCVTu2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
4301 v4f32, v4i32, uint_to_fp>;
4303 // VCVT : Vector Convert Between Floating-Point and Fixed-Point.
4304 def VCVTf2xsd : N2VCvtD<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
4305 v2i32, v2f32, int_arm_neon_vcvtfp2fxs>;
4306 def VCVTf2xud : N2VCvtD<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
4307 v2i32, v2f32, int_arm_neon_vcvtfp2fxu>;
4308 def VCVTxs2fd : N2VCvtD<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
4309 v2f32, v2i32, int_arm_neon_vcvtfxs2fp>;
4310 def VCVTxu2fd : N2VCvtD<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
4311 v2f32, v2i32, int_arm_neon_vcvtfxu2fp>;
4313 def VCVTf2xsq : N2VCvtQ<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
4314 v4i32, v4f32, int_arm_neon_vcvtfp2fxs>;
4315 def VCVTf2xuq : N2VCvtQ<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
4316 v4i32, v4f32, int_arm_neon_vcvtfp2fxu>;
4317 def VCVTxs2fq : N2VCvtQ<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
4318 v4f32, v4i32, int_arm_neon_vcvtfxs2fp>;
4319 def VCVTxu2fq : N2VCvtQ<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
4320 v4f32, v4i32, int_arm_neon_vcvtfxu2fp>;
4324 // VREV64 : Vector Reverse elements within 64-bit doublewords
4326 class VREV64D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4327 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 0, 0, (outs DPR:$Vd),
4328 (ins DPR:$Vm), IIC_VMOVD,
4329 OpcodeStr, Dt, "$Vd, $Vm", "",
4330 [(set DPR:$Vd, (Ty (NEONvrev64 (Ty DPR:$Vm))))]>;
4331 class VREV64Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4332 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 1, 0, (outs QPR:$Vd),
4333 (ins QPR:$Vm), IIC_VMOVQ,
4334 OpcodeStr, Dt, "$Vd, $Vm", "",
4335 [(set QPR:$Vd, (Ty (NEONvrev64 (Ty QPR:$Vm))))]>;
4337 def VREV64d8 : VREV64D<0b00, "vrev64", "8", v8i8>;
4338 def VREV64d16 : VREV64D<0b01, "vrev64", "16", v4i16>;
4339 def VREV64d32 : VREV64D<0b10, "vrev64", "32", v2i32>;
4340 def VREV64df : VREV64D<0b10, "vrev64", "32", v2f32>;
4342 def VREV64q8 : VREV64Q<0b00, "vrev64", "8", v16i8>;
4343 def VREV64q16 : VREV64Q<0b01, "vrev64", "16", v8i16>;
4344 def VREV64q32 : VREV64Q<0b10, "vrev64", "32", v4i32>;
4345 def VREV64qf : VREV64Q<0b10, "vrev64", "32", v4f32>;
4347 // VREV32 : Vector Reverse elements within 32-bit words
4349 class VREV32D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4350 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 0, 0, (outs DPR:$Vd),
4351 (ins DPR:$Vm), IIC_VMOVD,
4352 OpcodeStr, Dt, "$Vd, $Vm", "",
4353 [(set DPR:$Vd, (Ty (NEONvrev32 (Ty DPR:$Vm))))]>;
4354 class VREV32Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4355 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 1, 0, (outs QPR:$Vd),
4356 (ins QPR:$Vm), IIC_VMOVQ,
4357 OpcodeStr, Dt, "$Vd, $Vm", "",
4358 [(set QPR:$Vd, (Ty (NEONvrev32 (Ty QPR:$Vm))))]>;
4360 def VREV32d8 : VREV32D<0b00, "vrev32", "8", v8i8>;
4361 def VREV32d16 : VREV32D<0b01, "vrev32", "16", v4i16>;
4363 def VREV32q8 : VREV32Q<0b00, "vrev32", "8", v16i8>;
4364 def VREV32q16 : VREV32Q<0b01, "vrev32", "16", v8i16>;
4366 // VREV16 : Vector Reverse elements within 16-bit halfwords
4368 class VREV16D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4369 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 0, 0, (outs DPR:$Vd),
4370 (ins DPR:$Vm), IIC_VMOVD,
4371 OpcodeStr, Dt, "$Vd, $Vm", "",
4372 [(set DPR:$Vd, (Ty (NEONvrev16 (Ty DPR:$Vm))))]>;
4373 class VREV16Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4374 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 1, 0, (outs QPR:$Vd),
4375 (ins QPR:$Vm), IIC_VMOVQ,
4376 OpcodeStr, Dt, "$Vd, $Vm", "",
4377 [(set QPR:$Vd, (Ty (NEONvrev16 (Ty QPR:$Vm))))]>;
4379 def VREV16d8 : VREV16D<0b00, "vrev16", "8", v8i8>;
4380 def VREV16q8 : VREV16Q<0b00, "vrev16", "8", v16i8>;
4382 // Other Vector Shuffles.
4384 // VEXT : Vector Extract
4386 class VEXTd<string OpcodeStr, string Dt, ValueType Ty>
4387 : N3V<0,1,0b11,{?,?,?,?},0,0, (outs DPR:$Vd),
4388 (ins DPR:$Vn, DPR:$Vm, i32imm:$index), NVExtFrm,
4389 IIC_VEXTD, OpcodeStr, Dt, "$Vd, $Vn, $Vm, $index", "",
4390 [(set DPR:$Vd, (Ty (NEONvext (Ty DPR:$Vn),
4391 (Ty DPR:$Vm), imm:$index)))]> {
4393 let Inst{11-8} = index{3-0};
4396 class VEXTq<string OpcodeStr, string Dt, ValueType Ty>
4397 : N3V<0,1,0b11,{?,?,?,?},1,0, (outs QPR:$Vd),
4398 (ins QPR:$Vn, QPR:$Vm, i32imm:$index), NVExtFrm,
4399 IIC_VEXTQ, OpcodeStr, Dt, "$Vd, $Vn, $Vm, $index", "",
4400 [(set QPR:$Vd, (Ty (NEONvext (Ty QPR:$Vn),
4401 (Ty QPR:$Vm), imm:$index)))]> {
4403 let Inst{11-8} = index{3-0};
4406 def VEXTd8 : VEXTd<"vext", "8", v8i8> {
4407 let Inst{11-8} = index{3-0};
4409 def VEXTd16 : VEXTd<"vext", "16", v4i16> {
4410 let Inst{11-9} = index{2-0};
4413 def VEXTd32 : VEXTd<"vext", "32", v2i32> {
4414 let Inst{11-10} = index{1-0};
4415 let Inst{9-8} = 0b00;
4417 def VEXTdf : VEXTd<"vext", "32", v2f32> {
4418 let Inst{11} = index{0};
4419 let Inst{10-8} = 0b000;
4422 def VEXTq8 : VEXTq<"vext", "8", v16i8> {
4423 let Inst{11-8} = index{3-0};
4425 def VEXTq16 : VEXTq<"vext", "16", v8i16> {
4426 let Inst{11-9} = index{2-0};
4429 def VEXTq32 : VEXTq<"vext", "32", v4i32> {
4430 let Inst{11-10} = index{1-0};
4431 let Inst{9-8} = 0b00;
4433 def VEXTqf : VEXTq<"vext", "32", v4f32> {
4434 let Inst{11} = index{0};
4435 let Inst{10-8} = 0b000;
4438 // VTRN : Vector Transpose
4440 def VTRNd8 : N2VDShuffle<0b00, 0b00001, "vtrn", "8">;
4441 def VTRNd16 : N2VDShuffle<0b01, 0b00001, "vtrn", "16">;
4442 def VTRNd32 : N2VDShuffle<0b10, 0b00001, "vtrn", "32">;
4444 def VTRNq8 : N2VQShuffle<0b00, 0b00001, IIC_VPERMQ, "vtrn", "8">;
4445 def VTRNq16 : N2VQShuffle<0b01, 0b00001, IIC_VPERMQ, "vtrn", "16">;
4446 def VTRNq32 : N2VQShuffle<0b10, 0b00001, IIC_VPERMQ, "vtrn", "32">;
4448 // VUZP : Vector Unzip (Deinterleave)
4450 def VUZPd8 : N2VDShuffle<0b00, 0b00010, "vuzp", "8">;
4451 def VUZPd16 : N2VDShuffle<0b01, 0b00010, "vuzp", "16">;
4452 def VUZPd32 : N2VDShuffle<0b10, 0b00010, "vuzp", "32">;
4454 def VUZPq8 : N2VQShuffle<0b00, 0b00010, IIC_VPERMQ3, "vuzp", "8">;
4455 def VUZPq16 : N2VQShuffle<0b01, 0b00010, IIC_VPERMQ3, "vuzp", "16">;
4456 def VUZPq32 : N2VQShuffle<0b10, 0b00010, IIC_VPERMQ3, "vuzp", "32">;
4458 // VZIP : Vector Zip (Interleave)
4460 def VZIPd8 : N2VDShuffle<0b00, 0b00011, "vzip", "8">;
4461 def VZIPd16 : N2VDShuffle<0b01, 0b00011, "vzip", "16">;
4462 def VZIPd32 : N2VDShuffle<0b10, 0b00011, "vzip", "32">;
4464 def VZIPq8 : N2VQShuffle<0b00, 0b00011, IIC_VPERMQ3, "vzip", "8">;
4465 def VZIPq16 : N2VQShuffle<0b01, 0b00011, IIC_VPERMQ3, "vzip", "16">;
4466 def VZIPq32 : N2VQShuffle<0b10, 0b00011, IIC_VPERMQ3, "vzip", "32">;
4468 // Vector Table Lookup and Table Extension.
4470 // VTBL : Vector Table Lookup
4472 : N3V<1,1,0b11,0b1000,0,0, (outs DPR:$Vd),
4473 (ins DPR:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTB1,
4474 "vtbl", "8", "$Vd, \\{$Vn\\}, $Vm", "",
4475 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbl1 DPR:$Vn, DPR:$Vm)))]>;
4476 let hasExtraSrcRegAllocReq = 1 in {
4478 : N3V<1,1,0b11,0b1001,0,0, (outs DPR:$Vd),
4479 (ins DPR:$Vn, DPR:$tbl2, DPR:$Vm), NVTBLFrm, IIC_VTB2,
4480 "vtbl", "8", "$Vd, \\{$Vn, $tbl2\\}, $Vm", "", []>;
4482 : N3V<1,1,0b11,0b1010,0,0, (outs DPR:$Vd),
4483 (ins DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$Vm), NVTBLFrm, IIC_VTB3,
4484 "vtbl", "8", "$Vd, \\{$Vn, $tbl2, $tbl3\\}, $Vm", "", []>;
4486 : N3V<1,1,0b11,0b1011,0,0, (outs DPR:$Vd),
4487 (ins DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$Vm),
4489 "vtbl", "8", "$Vd, \\{$Vn, $tbl2, $tbl3, $tbl4\\}, $Vm", "", []>;
4490 } // hasExtraSrcRegAllocReq = 1
4493 : PseudoNeonI<(outs DPR:$dst), (ins QPR:$tbl, DPR:$src), IIC_VTB2, "", []>;
4495 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB3, "", []>;
4497 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB4, "", []>;
4499 // VTBX : Vector Table Extension
4501 : N3V<1,1,0b11,0b1000,1,0, (outs DPR:$Vd),
4502 (ins DPR:$orig, DPR:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTBX1,
4503 "vtbx", "8", "$Vd, \\{$Vn\\}, $Vm", "$orig = $Vd",
4504 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbx1
4505 DPR:$orig, DPR:$Vn, DPR:$Vm)))]>;
4506 let hasExtraSrcRegAllocReq = 1 in {
4508 : N3V<1,1,0b11,0b1001,1,0, (outs DPR:$Vd),
4509 (ins DPR:$orig, DPR:$Vn, DPR:$tbl2, DPR:$Vm), NVTBLFrm, IIC_VTBX2,
4510 "vtbx", "8", "$Vd, \\{$Vn, $tbl2\\}, $Vm", "$orig = $Vd", []>;
4512 : N3V<1,1,0b11,0b1010,1,0, (outs DPR:$Vd),
4513 (ins DPR:$orig, DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$Vm),
4514 NVTBLFrm, IIC_VTBX3,
4515 "vtbx", "8", "$Vd, \\{$Vn, $tbl2, $tbl3\\}, $Vm",
4518 : N3V<1,1,0b11,0b1011,1,0, (outs DPR:$Vd), (ins DPR:$orig, DPR:$Vn,
4519 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$Vm), NVTBLFrm, IIC_VTBX4,
4520 "vtbx", "8", "$Vd, \\{$Vn, $tbl2, $tbl3, $tbl4\\}, $Vm",
4522 } // hasExtraSrcRegAllocReq = 1
4525 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QPR:$tbl, DPR:$src),
4526 IIC_VTBX2, "$orig = $dst", []>;
4528 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
4529 IIC_VTBX3, "$orig = $dst", []>;
4531 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
4532 IIC_VTBX4, "$orig = $dst", []>;
4534 //===----------------------------------------------------------------------===//
4535 // NEON instructions for single-precision FP math
4536 //===----------------------------------------------------------------------===//
4538 class N2VSPat<SDNode OpNode, ValueType ResTy, ValueType OpTy, NeonI Inst>
4539 : NEONFPPat<(ResTy (OpNode SPR:$a)),
4540 (EXTRACT_SUBREG (OpTy (Inst (INSERT_SUBREG (OpTy (IMPLICIT_DEF)),
4544 class N3VSPat<SDNode OpNode, NeonI Inst>
4545 : NEONFPPat<(f32 (OpNode SPR:$a, SPR:$b)),
4546 (EXTRACT_SUBREG (v2f32
4547 (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
4549 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
4553 class N3VSMulOpPat<SDNode MulNode, SDNode OpNode, NeonI Inst>
4554 : NEONFPPat<(f32 (OpNode SPR:$acc, (f32 (MulNode SPR:$a, SPR:$b)))),
4555 (EXTRACT_SUBREG (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
4557 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
4559 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
4563 // These need separate instructions because they must use DPR_VFP2 register
4564 // class which have SPR sub-registers.
4566 // Vector Add Operations used for single-precision FP
4567 let neverHasSideEffects = 1 in
4568 def VADDfd_sfp : N3VS<0,0,0b00,0b1101,0, "vadd", "f32", v2f32, v2f32, fadd, 1>;
4569 def : N3VSPat<fadd, VADDfd_sfp>;
4571 // Vector Sub Operations used for single-precision FP
4572 let neverHasSideEffects = 1 in
4573 def VSUBfd_sfp : N3VS<0,0,0b10,0b1101,0, "vsub", "f32", v2f32, v2f32, fsub, 0>;
4574 def : N3VSPat<fsub, VSUBfd_sfp>;
4576 // Vector Multiply Operations used for single-precision FP
4577 let neverHasSideEffects = 1 in
4578 def VMULfd_sfp : N3VS<1,0,0b00,0b1101,1, "vmul", "f32", v2f32, v2f32, fmul, 1>;
4579 def : N3VSPat<fmul, VMULfd_sfp>;
4581 // Vector Multiply-Accumulate/Subtract used for single-precision FP
4582 // vml[as].f32 can cause 4-8 cycle stalls in following ASIMD instructions, so
4583 // we want to avoid them for now. e.g., alternating vmla/vadd instructions.
4585 //let neverHasSideEffects = 1 in
4586 //def VMLAfd_sfp : N3VSMulOp<0,0,0b00,0b1101,1, IIC_VMACD, "vmla", "f32",
4587 // v2f32, fmul, fadd>;
4588 //def : N3VSMulOpPat<fmul, fadd, VMLAfd_sfp>;
4590 //let neverHasSideEffects = 1 in
4591 //def VMLSfd_sfp : N3VSMulOp<0,0,0b10,0b1101,1, IIC_VMACD, "vmls", "f32",
4592 // v2f32, fmul, fsub>;
4593 //def : N3VSMulOpPat<fmul, fsub, VMLSfd_sfp>;
4595 // Vector Absolute used for single-precision FP
4596 let neverHasSideEffects = 1 in
4597 def VABSfd_sfp : N2V<0b11, 0b11, 0b10, 0b01, 0b01110, 0, 0,
4598 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), IIC_VUNAD,
4599 "vabs", "f32", "$dst, $src", "", []>;
4600 def : N2VSPat<fabs, f32, v2f32, VABSfd_sfp>;
4602 // Vector Negate used for single-precision FP
4603 let neverHasSideEffects = 1 in
4604 def VNEGfd_sfp : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
4605 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), IIC_VUNAD,
4606 "vneg", "f32", "$dst, $src", "", []>;
4607 def : N2VSPat<fneg, f32, v2f32, VNEGfd_sfp>;
4609 // Vector Maximum used for single-precision FP
4610 let neverHasSideEffects = 1 in
4611 def VMAXfd_sfp : N3V<0, 0, 0b00, 0b1111, 0, 0, (outs DPR_VFP2:$dst),
4612 (ins DPR_VFP2:$src1, DPR_VFP2:$src2), N3RegFrm, IIC_VBIND,
4613 "vmax", "f32", "$dst, $src1, $src2", "", []>;
4614 def : N3VSPat<NEONfmax, VMAXfd_sfp>;
4616 // Vector Minimum used for single-precision FP
4617 let neverHasSideEffects = 1 in
4618 def VMINfd_sfp : N3V<0, 0, 0b00, 0b1111, 0, 0, (outs DPR_VFP2:$dst),
4619 (ins DPR_VFP2:$src1, DPR_VFP2:$src2), N3RegFrm, IIC_VBIND,
4620 "vmin", "f32", "$dst, $src1, $src2", "", []>;
4621 def : N3VSPat<NEONfmin, VMINfd_sfp>;
4623 // Vector Convert between single-precision FP and integer
4624 let neverHasSideEffects = 1 in
4625 def VCVTf2sd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
4626 v2i32, v2f32, fp_to_sint>;
4627 def : N2VSPat<arm_ftosi, f32, v2f32, VCVTf2sd_sfp>;
4629 let neverHasSideEffects = 1 in
4630 def VCVTf2ud_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
4631 v2i32, v2f32, fp_to_uint>;
4632 def : N2VSPat<arm_ftoui, f32, v2f32, VCVTf2ud_sfp>;
4634 let neverHasSideEffects = 1 in
4635 def VCVTs2fd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
4636 v2f32, v2i32, sint_to_fp>;
4637 def : N2VSPat<arm_sitof, f32, v2i32, VCVTs2fd_sfp>;
4639 let neverHasSideEffects = 1 in
4640 def VCVTu2fd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
4641 v2f32, v2i32, uint_to_fp>;
4642 def : N2VSPat<arm_uitof, f32, v2i32, VCVTu2fd_sfp>;
4644 //===----------------------------------------------------------------------===//
4645 // Non-Instruction Patterns
4646 //===----------------------------------------------------------------------===//
4649 def : Pat<(v1i64 (bitconvert (v2i32 DPR:$src))), (v1i64 DPR:$src)>;
4650 def : Pat<(v1i64 (bitconvert (v4i16 DPR:$src))), (v1i64 DPR:$src)>;
4651 def : Pat<(v1i64 (bitconvert (v8i8 DPR:$src))), (v1i64 DPR:$src)>;
4652 def : Pat<(v1i64 (bitconvert (f64 DPR:$src))), (v1i64 DPR:$src)>;
4653 def : Pat<(v1i64 (bitconvert (v2f32 DPR:$src))), (v1i64 DPR:$src)>;
4654 def : Pat<(v2i32 (bitconvert (v1i64 DPR:$src))), (v2i32 DPR:$src)>;
4655 def : Pat<(v2i32 (bitconvert (v4i16 DPR:$src))), (v2i32 DPR:$src)>;
4656 def : Pat<(v2i32 (bitconvert (v8i8 DPR:$src))), (v2i32 DPR:$src)>;
4657 def : Pat<(v2i32 (bitconvert (f64 DPR:$src))), (v2i32 DPR:$src)>;
4658 def : Pat<(v2i32 (bitconvert (v2f32 DPR:$src))), (v2i32 DPR:$src)>;
4659 def : Pat<(v4i16 (bitconvert (v1i64 DPR:$src))), (v4i16 DPR:$src)>;
4660 def : Pat<(v4i16 (bitconvert (v2i32 DPR:$src))), (v4i16 DPR:$src)>;
4661 def : Pat<(v4i16 (bitconvert (v8i8 DPR:$src))), (v4i16 DPR:$src)>;
4662 def : Pat<(v4i16 (bitconvert (f64 DPR:$src))), (v4i16 DPR:$src)>;
4663 def : Pat<(v4i16 (bitconvert (v2f32 DPR:$src))), (v4i16 DPR:$src)>;
4664 def : Pat<(v8i8 (bitconvert (v1i64 DPR:$src))), (v8i8 DPR:$src)>;
4665 def : Pat<(v8i8 (bitconvert (v2i32 DPR:$src))), (v8i8 DPR:$src)>;
4666 def : Pat<(v8i8 (bitconvert (v4i16 DPR:$src))), (v8i8 DPR:$src)>;
4667 def : Pat<(v8i8 (bitconvert (f64 DPR:$src))), (v8i8 DPR:$src)>;
4668 def : Pat<(v8i8 (bitconvert (v2f32 DPR:$src))), (v8i8 DPR:$src)>;
4669 def : Pat<(f64 (bitconvert (v1i64 DPR:$src))), (f64 DPR:$src)>;
4670 def : Pat<(f64 (bitconvert (v2i32 DPR:$src))), (f64 DPR:$src)>;
4671 def : Pat<(f64 (bitconvert (v4i16 DPR:$src))), (f64 DPR:$src)>;
4672 def : Pat<(f64 (bitconvert (v8i8 DPR:$src))), (f64 DPR:$src)>;
4673 def : Pat<(f64 (bitconvert (v2f32 DPR:$src))), (f64 DPR:$src)>;
4674 def : Pat<(v2f32 (bitconvert (f64 DPR:$src))), (v2f32 DPR:$src)>;
4675 def : Pat<(v2f32 (bitconvert (v1i64 DPR:$src))), (v2f32 DPR:$src)>;
4676 def : Pat<(v2f32 (bitconvert (v2i32 DPR:$src))), (v2f32 DPR:$src)>;
4677 def : Pat<(v2f32 (bitconvert (v4i16 DPR:$src))), (v2f32 DPR:$src)>;
4678 def : Pat<(v2f32 (bitconvert (v8i8 DPR:$src))), (v2f32 DPR:$src)>;
4680 def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (v2i64 QPR:$src)>;
4681 def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (v2i64 QPR:$src)>;
4682 def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (v2i64 QPR:$src)>;
4683 def : Pat<(v2i64 (bitconvert (v2f64 QPR:$src))), (v2i64 QPR:$src)>;
4684 def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (v2i64 QPR:$src)>;
4685 def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (v4i32 QPR:$src)>;
4686 def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (v4i32 QPR:$src)>;
4687 def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (v4i32 QPR:$src)>;
4688 def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (v4i32 QPR:$src)>;
4689 def : Pat<(v4i32 (bitconvert (v4f32 QPR:$src))), (v4i32 QPR:$src)>;
4690 def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (v8i16 QPR:$src)>;
4691 def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (v8i16 QPR:$src)>;
4692 def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (v8i16 QPR:$src)>;
4693 def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (v8i16 QPR:$src)>;
4694 def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (v8i16 QPR:$src)>;
4695 def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (v16i8 QPR:$src)>;
4696 def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (v16i8 QPR:$src)>;
4697 def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (v16i8 QPR:$src)>;
4698 def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (v16i8 QPR:$src)>;
4699 def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (v16i8 QPR:$src)>;
4700 def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (v4f32 QPR:$src)>;
4701 def : Pat<(v4f32 (bitconvert (v4i32 QPR:$src))), (v4f32 QPR:$src)>;
4702 def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (v4f32 QPR:$src)>;
4703 def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (v4f32 QPR:$src)>;
4704 def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (v4f32 QPR:$src)>;
4705 def : Pat<(v2f64 (bitconvert (v2i64 QPR:$src))), (v2f64 QPR:$src)>;
4706 def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (v2f64 QPR:$src)>;
4707 def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 QPR:$src)>;
4708 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 QPR:$src)>;
4709 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 QPR:$src)>;