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, ldstm_mode:$mode),
148 [(set QPR:$dst, (v2f64 (load GPR:$Rn)))]>;
150 // Use VSTM to store a Q register as a D register pair.
151 // This is a pseudo instruction that is expanded to VSTMD after reg alloc.
153 : PseudoVFPLdStM<(outs), (ins QPR:$src, GPR:$Rn, ldstm_mode:$mode),
155 [(store (v2f64 QPR:$src), GPR:$Rn)]>;
157 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
159 // Classes for VLD* pseudo-instructions with multi-register operands.
160 // These are expanded to real instructions after register allocation.
161 class VLDQPseudo<InstrItinClass itin>
162 : PseudoNLdSt<(outs QPR:$dst), (ins addrmode6:$addr), itin, "">;
163 class VLDQWBPseudo<InstrItinClass itin>
164 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
165 (ins addrmode6:$addr, am6offset:$offset), itin,
167 class VLDQQPseudo<InstrItinClass itin>
168 : PseudoNLdSt<(outs QQPR:$dst), (ins addrmode6:$addr), itin, "">;
169 class VLDQQWBPseudo<InstrItinClass itin>
170 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
171 (ins addrmode6:$addr, am6offset:$offset), itin,
173 class VLDQQQQWBPseudo<InstrItinClass itin>
174 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
175 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
176 "$addr.addr = $wb, $src = $dst">;
178 // VLD1 : Vector Load (multiple single elements)
179 class VLD1D<bits<4> op7_4, string Dt>
180 : NLdSt<0,0b10,0b0111,op7_4, (outs DPR:$Vd),
181 (ins addrmode6:$Rn), IIC_VLD1,
182 "vld1", Dt, "\\{$Vd\\}, $Rn", "", []> {
186 class VLD1Q<bits<4> op7_4, string Dt>
187 : NLdSt<0,0b10,0b1010,op7_4, (outs DPR:$Vd, DPR:$dst2),
188 (ins addrmode6:$Rn), IIC_VLD1x2,
189 "vld1", Dt, "\\{$Vd, $dst2\\}, $Rn", "", []> {
191 let Inst{5-4} = Rn{5-4};
194 def VLD1d8 : VLD1D<{0,0,0,?}, "8">;
195 def VLD1d16 : VLD1D<{0,1,0,?}, "16">;
196 def VLD1d32 : VLD1D<{1,0,0,?}, "32">;
197 def VLD1d64 : VLD1D<{1,1,0,?}, "64">;
199 def VLD1q8 : VLD1Q<{0,0,?,?}, "8">;
200 def VLD1q16 : VLD1Q<{0,1,?,?}, "16">;
201 def VLD1q32 : VLD1Q<{1,0,?,?}, "32">;
202 def VLD1q64 : VLD1Q<{1,1,?,?}, "64">;
204 def VLD1q8Pseudo : VLDQPseudo<IIC_VLD1x2>;
205 def VLD1q16Pseudo : VLDQPseudo<IIC_VLD1x2>;
206 def VLD1q32Pseudo : VLDQPseudo<IIC_VLD1x2>;
207 def VLD1q64Pseudo : VLDQPseudo<IIC_VLD1x2>;
209 // ...with address register writeback:
210 class VLD1DWB<bits<4> op7_4, string Dt>
211 : NLdSt<0,0b10,0b0111,op7_4, (outs DPR:$Vd, GPR:$wb),
212 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD1u,
213 "vld1", Dt, "\\{$Vd\\}, $Rn$Rm",
214 "$Rn.addr = $wb", []> {
217 class VLD1QWB<bits<4> op7_4, string Dt>
218 : NLdSt<0,0b10,0b1010,op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
219 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD1x2u,
220 "vld1", Dt, "\\{$Vd, $dst2\\}, $Rn$Rm",
221 "$Rn.addr = $wb", []> {
222 let Inst{5-4} = Rn{5-4};
225 def VLD1d8_UPD : VLD1DWB<{0,0,0,?}, "8">;
226 def VLD1d16_UPD : VLD1DWB<{0,1,0,?}, "16">;
227 def VLD1d32_UPD : VLD1DWB<{1,0,0,?}, "32">;
228 def VLD1d64_UPD : VLD1DWB<{1,1,0,?}, "64">;
230 def VLD1q8_UPD : VLD1QWB<{0,0,?,?}, "8">;
231 def VLD1q16_UPD : VLD1QWB<{0,1,?,?}, "16">;
232 def VLD1q32_UPD : VLD1QWB<{1,0,?,?}, "32">;
233 def VLD1q64_UPD : VLD1QWB<{1,1,?,?}, "64">;
235 def VLD1q8Pseudo_UPD : VLDQWBPseudo<IIC_VLD1x2u>;
236 def VLD1q16Pseudo_UPD : VLDQWBPseudo<IIC_VLD1x2u>;
237 def VLD1q32Pseudo_UPD : VLDQWBPseudo<IIC_VLD1x2u>;
238 def VLD1q64Pseudo_UPD : VLDQWBPseudo<IIC_VLD1x2u>;
240 // ...with 3 registers (some of these are only for the disassembler):
241 class VLD1D3<bits<4> op7_4, string Dt>
242 : NLdSt<0,0b10,0b0110,op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
243 (ins addrmode6:$Rn), IIC_VLD1x3, "vld1", Dt,
244 "\\{$Vd, $dst2, $dst3\\}, $Rn", "", []> {
248 class VLD1D3WB<bits<4> op7_4, string Dt>
249 : NLdSt<0,0b10,0b0110,op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
250 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD1x3u, "vld1", Dt,
251 "\\{$Vd, $dst2, $dst3\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
255 def VLD1d8T : VLD1D3<{0,0,0,?}, "8">;
256 def VLD1d16T : VLD1D3<{0,1,0,?}, "16">;
257 def VLD1d32T : VLD1D3<{1,0,0,?}, "32">;
258 def VLD1d64T : VLD1D3<{1,1,0,?}, "64">;
260 def VLD1d8T_UPD : VLD1D3WB<{0,0,0,?}, "8">;
261 def VLD1d16T_UPD : VLD1D3WB<{0,1,0,?}, "16">;
262 def VLD1d32T_UPD : VLD1D3WB<{1,0,0,?}, "32">;
263 def VLD1d64T_UPD : VLD1D3WB<{1,1,0,?}, "64">;
265 def VLD1d64TPseudo : VLDQQPseudo<IIC_VLD1x3>;
266 def VLD1d64TPseudo_UPD : VLDQQWBPseudo<IIC_VLD1x3u>;
268 // ...with 4 registers (some of these are only for the disassembler):
269 class VLD1D4<bits<4> op7_4, string Dt>
270 : NLdSt<0,0b10,0b0010,op7_4,(outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
271 (ins addrmode6:$Rn), IIC_VLD1x4, "vld1", Dt,
272 "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn", "", []> {
274 let Inst{5-4} = Rn{5-4};
276 class VLD1D4WB<bits<4> op7_4, string Dt>
277 : NLdSt<0,0b10,0b0010,op7_4,
278 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
279 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD4, "vld1", Dt,
280 "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn$Rm", "$Rn.addr = $wb",
282 let Inst{5-4} = Rn{5-4};
285 def VLD1d8Q : VLD1D4<{0,0,?,?}, "8">;
286 def VLD1d16Q : VLD1D4<{0,1,?,?}, "16">;
287 def VLD1d32Q : VLD1D4<{1,0,?,?}, "32">;
288 def VLD1d64Q : VLD1D4<{1,1,?,?}, "64">;
290 def VLD1d8Q_UPD : VLD1D4WB<{0,0,?,?}, "8">;
291 def VLD1d16Q_UPD : VLD1D4WB<{0,1,?,?}, "16">;
292 def VLD1d32Q_UPD : VLD1D4WB<{1,0,?,?}, "32">;
293 def VLD1d64Q_UPD : VLD1D4WB<{1,1,?,?}, "64">;
295 def VLD1d64QPseudo : VLDQQPseudo<IIC_VLD1x4>;
296 def VLD1d64QPseudo_UPD : VLDQQWBPseudo<IIC_VLD1x4u>;
298 // VLD2 : Vector Load (multiple 2-element structures)
299 class VLD2D<bits<4> op11_8, bits<4> op7_4, string Dt>
300 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2),
301 (ins addrmode6:$Rn), IIC_VLD2,
302 "vld2", Dt, "\\{$Vd, $dst2\\}, $Rn", "", []> {
304 let Inst{5-4} = Rn{5-4};
306 class VLD2Q<bits<4> op7_4, string Dt>
307 : NLdSt<0, 0b10, 0b0011, op7_4,
308 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
309 (ins addrmode6:$Rn), IIC_VLD2x2,
310 "vld2", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn", "", []> {
312 let Inst{5-4} = Rn{5-4};
315 def VLD2d8 : VLD2D<0b1000, {0,0,?,?}, "8">;
316 def VLD2d16 : VLD2D<0b1000, {0,1,?,?}, "16">;
317 def VLD2d32 : VLD2D<0b1000, {1,0,?,?}, "32">;
319 def VLD2q8 : VLD2Q<{0,0,?,?}, "8">;
320 def VLD2q16 : VLD2Q<{0,1,?,?}, "16">;
321 def VLD2q32 : VLD2Q<{1,0,?,?}, "32">;
323 def VLD2d8Pseudo : VLDQPseudo<IIC_VLD2>;
324 def VLD2d16Pseudo : VLDQPseudo<IIC_VLD2>;
325 def VLD2d32Pseudo : VLDQPseudo<IIC_VLD2>;
327 def VLD2q8Pseudo : VLDQQPseudo<IIC_VLD2x2>;
328 def VLD2q16Pseudo : VLDQQPseudo<IIC_VLD2x2>;
329 def VLD2q32Pseudo : VLDQQPseudo<IIC_VLD2x2>;
331 // ...with address register writeback:
332 class VLD2DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
333 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
334 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD2u,
335 "vld2", Dt, "\\{$Vd, $dst2\\}, $Rn$Rm",
336 "$Rn.addr = $wb", []> {
337 let Inst{5-4} = Rn{5-4};
339 class VLD2QWB<bits<4> op7_4, string Dt>
340 : NLdSt<0, 0b10, 0b0011, op7_4,
341 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
342 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD2x2u,
343 "vld2", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn$Rm",
344 "$Rn.addr = $wb", []> {
345 let Inst{5-4} = Rn{5-4};
348 def VLD2d8_UPD : VLD2DWB<0b1000, {0,0,?,?}, "8">;
349 def VLD2d16_UPD : VLD2DWB<0b1000, {0,1,?,?}, "16">;
350 def VLD2d32_UPD : VLD2DWB<0b1000, {1,0,?,?}, "32">;
352 def VLD2q8_UPD : VLD2QWB<{0,0,?,?}, "8">;
353 def VLD2q16_UPD : VLD2QWB<{0,1,?,?}, "16">;
354 def VLD2q32_UPD : VLD2QWB<{1,0,?,?}, "32">;
356 def VLD2d8Pseudo_UPD : VLDQWBPseudo<IIC_VLD2u>;
357 def VLD2d16Pseudo_UPD : VLDQWBPseudo<IIC_VLD2u>;
358 def VLD2d32Pseudo_UPD : VLDQWBPseudo<IIC_VLD2u>;
360 def VLD2q8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD2x2u>;
361 def VLD2q16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD2x2u>;
362 def VLD2q32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD2x2u>;
364 // ...with double-spaced registers (for disassembly only):
365 def VLD2b8 : VLD2D<0b1001, {0,0,?,?}, "8">;
366 def VLD2b16 : VLD2D<0b1001, {0,1,?,?}, "16">;
367 def VLD2b32 : VLD2D<0b1001, {1,0,?,?}, "32">;
368 def VLD2b8_UPD : VLD2DWB<0b1001, {0,0,?,?}, "8">;
369 def VLD2b16_UPD : VLD2DWB<0b1001, {0,1,?,?}, "16">;
370 def VLD2b32_UPD : VLD2DWB<0b1001, {1,0,?,?}, "32">;
372 // VLD3 : Vector Load (multiple 3-element structures)
373 class VLD3D<bits<4> op11_8, bits<4> op7_4, string Dt>
374 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
375 (ins addrmode6:$Rn), IIC_VLD3,
376 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn", "", []> {
381 def VLD3d8 : VLD3D<0b0100, {0,0,0,?}, "8">;
382 def VLD3d16 : VLD3D<0b0100, {0,1,0,?}, "16">;
383 def VLD3d32 : VLD3D<0b0100, {1,0,0,?}, "32">;
385 def VLD3d8Pseudo : VLDQQPseudo<IIC_VLD3>;
386 def VLD3d16Pseudo : VLDQQPseudo<IIC_VLD3>;
387 def VLD3d32Pseudo : VLDQQPseudo<IIC_VLD3>;
389 // ...with address register writeback:
390 class VLD3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
391 : NLdSt<0, 0b10, op11_8, op7_4,
392 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
393 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD3u,
394 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn$Rm",
395 "$Rn.addr = $wb", []> {
399 def VLD3d8_UPD : VLD3DWB<0b0100, {0,0,0,?}, "8">;
400 def VLD3d16_UPD : VLD3DWB<0b0100, {0,1,0,?}, "16">;
401 def VLD3d32_UPD : VLD3DWB<0b0100, {1,0,0,?}, "32">;
403 def VLD3d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
404 def VLD3d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
405 def VLD3d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
407 // ...with double-spaced registers (non-updating versions for disassembly only):
408 def VLD3q8 : VLD3D<0b0101, {0,0,0,?}, "8">;
409 def VLD3q16 : VLD3D<0b0101, {0,1,0,?}, "16">;
410 def VLD3q32 : VLD3D<0b0101, {1,0,0,?}, "32">;
411 def VLD3q8_UPD : VLD3DWB<0b0101, {0,0,0,?}, "8">;
412 def VLD3q16_UPD : VLD3DWB<0b0101, {0,1,0,?}, "16">;
413 def VLD3q32_UPD : VLD3DWB<0b0101, {1,0,0,?}, "32">;
415 def VLD3q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
416 def VLD3q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
417 def VLD3q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
419 // ...alternate versions to be allocated odd register numbers:
420 def VLD3q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
421 def VLD3q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
422 def VLD3q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
424 // VLD4 : Vector Load (multiple 4-element structures)
425 class VLD4D<bits<4> op11_8, bits<4> op7_4, string Dt>
426 : NLdSt<0, 0b10, op11_8, op7_4,
427 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
428 (ins addrmode6:$Rn), IIC_VLD4,
429 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn", "", []> {
431 let Inst{5-4} = Rn{5-4};
434 def VLD4d8 : VLD4D<0b0000, {0,0,?,?}, "8">;
435 def VLD4d16 : VLD4D<0b0000, {0,1,?,?}, "16">;
436 def VLD4d32 : VLD4D<0b0000, {1,0,?,?}, "32">;
438 def VLD4d8Pseudo : VLDQQPseudo<IIC_VLD4>;
439 def VLD4d16Pseudo : VLDQQPseudo<IIC_VLD4>;
440 def VLD4d32Pseudo : VLDQQPseudo<IIC_VLD4>;
442 // ...with address register writeback:
443 class VLD4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
444 : NLdSt<0, 0b10, op11_8, op7_4,
445 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
446 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD4,
447 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn$Rm",
448 "$Rn.addr = $wb", []> {
449 let Inst{5-4} = Rn{5-4};
452 def VLD4d8_UPD : VLD4DWB<0b0000, {0,0,?,?}, "8">;
453 def VLD4d16_UPD : VLD4DWB<0b0000, {0,1,?,?}, "16">;
454 def VLD4d32_UPD : VLD4DWB<0b0000, {1,0,?,?}, "32">;
456 def VLD4d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4>;
457 def VLD4d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4>;
458 def VLD4d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4>;
460 // ...with double-spaced registers (non-updating versions for disassembly only):
461 def VLD4q8 : VLD4D<0b0001, {0,0,?,?}, "8">;
462 def VLD4q16 : VLD4D<0b0001, {0,1,?,?}, "16">;
463 def VLD4q32 : VLD4D<0b0001, {1,0,?,?}, "32">;
464 def VLD4q8_UPD : VLD4DWB<0b0001, {0,0,?,?}, "8">;
465 def VLD4q16_UPD : VLD4DWB<0b0001, {0,1,?,?}, "16">;
466 def VLD4q32_UPD : VLD4DWB<0b0001, {1,0,?,?}, "32">;
468 def VLD4q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4>;
469 def VLD4q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4>;
470 def VLD4q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4>;
472 // ...alternate versions to be allocated odd register numbers:
473 def VLD4q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4>;
474 def VLD4q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4>;
475 def VLD4q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4>;
477 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
479 // Classes for VLD*LN pseudo-instructions with multi-register operands.
480 // These are expanded to real instructions after register allocation.
481 class VLDQLNPseudo<InstrItinClass itin>
482 : PseudoNLdSt<(outs QPR:$dst),
483 (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
484 itin, "$src = $dst">;
485 class VLDQLNWBPseudo<InstrItinClass itin>
486 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
487 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
488 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
489 class VLDQQLNPseudo<InstrItinClass itin>
490 : PseudoNLdSt<(outs QQPR:$dst),
491 (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
492 itin, "$src = $dst">;
493 class VLDQQLNWBPseudo<InstrItinClass itin>
494 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
495 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
496 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
497 class VLDQQQQLNPseudo<InstrItinClass itin>
498 : PseudoNLdSt<(outs QQQQPR:$dst),
499 (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
500 itin, "$src = $dst">;
501 class VLDQQQQLNWBPseudo<InstrItinClass itin>
502 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
503 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
504 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
506 // VLD1LN : Vector Load (single element to one lane)
507 class VLD1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
509 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd),
510 (ins addrmode6:$Rn, DPR:$src, nohash_imm:$lane),
511 IIC_VLD1ln, "vld1", Dt, "\\{$Vd[$lane]\\}, $Rn",
513 [(set DPR:$Vd, (vector_insert (Ty DPR:$src),
514 (i32 (LoadOp addrmode6:$Rn)),
518 class VLD1QLNPseudo<ValueType Ty, PatFrag LoadOp> : VLDQLNPseudo<IIC_VLD1ln> {
519 let Pattern = [(set QPR:$dst, (vector_insert (Ty QPR:$src),
520 (i32 (LoadOp addrmode6:$addr)),
524 def VLD1LNd8 : VLD1LN<0b0000, {?,?,?,0}, "8", v8i8, extloadi8> {
525 let Inst{7-5} = lane{2-0};
527 def VLD1LNd16 : VLD1LN<0b0100, {?,?,0,?}, "16", v4i16, extloadi16> {
528 let Inst{7-6} = lane{1-0};
531 def VLD1LNd32 : VLD1LN<0b1000, {?,0,?,?}, "32", v2i32, load> {
532 let Inst{7} = lane{0};
537 def VLD1LNq8Pseudo : VLD1QLNPseudo<v16i8, extloadi8>;
538 def VLD1LNq16Pseudo : VLD1QLNPseudo<v8i16, extloadi16>;
539 def VLD1LNq32Pseudo : VLD1QLNPseudo<v4i32, load>;
541 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
543 // ...with address register writeback:
544 class VLD1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
545 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, GPR:$wb),
546 (ins addrmode6:$Rn, am6offset:$Rm,
547 DPR:$src, nohash_imm:$lane), IIC_VLD1lnu, "vld1", Dt,
548 "\\{$Vd[$lane]\\}, $Rn$Rm",
549 "$src = $Vd, $Rn.addr = $wb", []>;
551 def VLD1LNd8_UPD : VLD1LNWB<0b0000, {?,?,?,0}, "8"> {
552 let Inst{7-5} = lane{2-0};
554 def VLD1LNd16_UPD : VLD1LNWB<0b0100, {?,?,0,?}, "16"> {
555 let Inst{7-6} = lane{1-0};
558 def VLD1LNd32_UPD : VLD1LNWB<0b1000, {?,0,?,?}, "32"> {
559 let Inst{7} = lane{0};
564 def VLD1LNq8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
565 def VLD1LNq16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
566 def VLD1LNq32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
568 // VLD2LN : Vector Load (single 2-element structure to one lane)
569 class VLD2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
570 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2),
571 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, nohash_imm:$lane),
572 IIC_VLD2ln, "vld2", Dt, "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn",
573 "$src1 = $Vd, $src2 = $dst2", []> {
578 def VLD2LNd8 : VLD2LN<0b0001, {?,?,?,?}, "8"> {
579 let Inst{7-5} = lane{2-0};
581 def VLD2LNd16 : VLD2LN<0b0101, {?,?,0,?}, "16"> {
582 let Inst{7-6} = lane{1-0};
584 def VLD2LNd32 : VLD2LN<0b1001, {?,0,0,?}, "32"> {
585 let Inst{7} = lane{0};
588 def VLD2LNd8Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
589 def VLD2LNd16Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
590 def VLD2LNd32Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
592 // ...with double-spaced registers:
593 def VLD2LNq16 : VLD2LN<0b0101, {?,?,1,?}, "16"> {
594 let Inst{7-6} = lane{1-0};
596 def VLD2LNq32 : VLD2LN<0b1001, {?,1,0,?}, "32"> {
597 let Inst{7} = lane{0};
600 def VLD2LNq16Pseudo : VLDQQLNPseudo<IIC_VLD2ln>;
601 def VLD2LNq32Pseudo : VLDQQLNPseudo<IIC_VLD2ln>;
603 // ...with address register writeback:
604 class VLD2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
605 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
606 (ins addrmode6:$Rn, am6offset:$Rm,
607 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VLD2lnu, "vld2", Dt,
608 "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn$Rm",
609 "$src1 = $Vd, $src2 = $dst2, $Rn.addr = $wb", []> {
613 def VLD2LNd8_UPD : VLD2LNWB<0b0001, {?,?,?,?}, "8"> {
614 let Inst{7-5} = lane{2-0};
616 def VLD2LNd16_UPD : VLD2LNWB<0b0101, {?,?,0,?}, "16"> {
617 let Inst{7-6} = lane{1-0};
619 def VLD2LNd32_UPD : VLD2LNWB<0b1001, {?,0,0,?}, "32"> {
620 let Inst{7} = lane{0};
623 def VLD2LNd8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
624 def VLD2LNd16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
625 def VLD2LNd32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
627 def VLD2LNq16_UPD : VLD2LNWB<0b0101, {?,?,1,?}, "16"> {
628 let Inst{7-6} = lane{1-0};
630 def VLD2LNq32_UPD : VLD2LNWB<0b1001, {?,1,0,?}, "32"> {
631 let Inst{7} = lane{0};
634 def VLD2LNq16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
635 def VLD2LNq32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
637 // VLD3LN : Vector Load (single 3-element structure to one lane)
638 class VLD3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
639 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
640 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3,
641 nohash_imm:$lane), IIC_VLD3ln, "vld3", Dt,
642 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn",
643 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3", []> {
647 def VLD3LNd8 : VLD3LN<0b0010, {?,?,?,0}, "8"> {
648 let Inst{7-5} = lane{2-0};
650 def VLD3LNd16 : VLD3LN<0b0110, {?,?,0,0}, "16"> {
651 let Inst{7-6} = lane{1-0};
653 def VLD3LNd32 : VLD3LN<0b1010, {?,0,0,0}, "32"> {
654 let Inst{7} = lane{0};
657 def VLD3LNd8Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
658 def VLD3LNd16Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
659 def VLD3LNd32Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
661 // ...with double-spaced registers:
662 def VLD3LNq16 : VLD3LN<0b0110, {?,?,1,0}, "16"> {
663 let Inst{7-6} = lane{1-0};
665 def VLD3LNq32 : VLD3LN<0b1010, {?,1,0,0}, "32"> {
666 let Inst{7} = lane{0};
669 def VLD3LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
670 def VLD3LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
672 // ...with address register writeback:
673 class VLD3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
674 : NLdStLn<1, 0b10, op11_8, op7_4,
675 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
676 (ins addrmode6:$Rn, am6offset:$Rm,
677 DPR:$src1, DPR:$src2, DPR:$src3, nohash_imm:$lane),
678 IIC_VLD3lnu, "vld3", Dt,
679 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn$Rm",
680 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $Rn.addr = $wb",
683 def VLD3LNd8_UPD : VLD3LNWB<0b0010, {?,?,?,0}, "8"> {
684 let Inst{7-5} = lane{2-0};
686 def VLD3LNd16_UPD : VLD3LNWB<0b0110, {?,?,0,0}, "16"> {
687 let Inst{7-6} = lane{1-0};
689 def VLD3LNd32_UPD : VLD3LNWB<0b1010, {?,0,0,0}, "32"> {
690 let Inst{7} = lane{0};
693 def VLD3LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
694 def VLD3LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
695 def VLD3LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
697 def VLD3LNq16_UPD : VLD3LNWB<0b0110, {?,?,1,0}, "16"> {
698 let Inst{7-6} = lane{1-0};
700 def VLD3LNq32_UPD : VLD3LNWB<0b1010, {?,1,0,0}, "32"> {
701 let Inst{7} = lane{0};
704 def VLD3LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
705 def VLD3LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
707 // VLD4LN : Vector Load (single 4-element structure to one lane)
708 class VLD4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
709 : NLdStLn<1, 0b10, op11_8, op7_4,
710 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
711 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
712 nohash_imm:$lane), IIC_VLD4ln, "vld4", Dt,
713 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn",
714 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4", []> {
719 def VLD4LNd8 : VLD4LN<0b0011, {?,?,?,?}, "8"> {
720 let Inst{7-5} = lane{2-0};
722 def VLD4LNd16 : VLD4LN<0b0111, {?,?,0,?}, "16"> {
723 let Inst{7-6} = lane{1-0};
725 def VLD4LNd32 : VLD4LN<0b1011, {?,0,?,?}, "32"> {
726 let Inst{7} = lane{0};
730 def VLD4LNd8Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
731 def VLD4LNd16Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
732 def VLD4LNd32Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
734 // ...with double-spaced registers:
735 def VLD4LNq16 : VLD4LN<0b0111, {?,?,1,?}, "16"> {
736 let Inst{7-6} = lane{1-0};
738 def VLD4LNq32 : VLD4LN<0b1011, {?,1,?,?}, "32"> {
739 let Inst{7} = lane{0};
743 def VLD4LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>;
744 def VLD4LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>;
746 // ...with address register writeback:
747 class VLD4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
748 : NLdStLn<1, 0b10, op11_8, op7_4,
749 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
750 (ins addrmode6:$Rn, am6offset:$Rm,
751 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
752 IIC_VLD4ln, "vld4", Dt,
753 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn$Rm",
754 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4, $Rn.addr = $wb",
759 def VLD4LNd8_UPD : VLD4LNWB<0b0011, {?,?,?,?}, "8"> {
760 let Inst{7-5} = lane{2-0};
762 def VLD4LNd16_UPD : VLD4LNWB<0b0111, {?,?,0,?}, "16"> {
763 let Inst{7-6} = lane{1-0};
765 def VLD4LNd32_UPD : VLD4LNWB<0b1011, {?,0,?,?}, "32"> {
766 let Inst{7} = lane{0};
770 def VLD4LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
771 def VLD4LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
772 def VLD4LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
774 def VLD4LNq16_UPD : VLD4LNWB<0b0111, {?,?,1,?}, "16"> {
775 let Inst{7-6} = lane{1-0};
777 def VLD4LNq32_UPD : VLD4LNWB<0b1011, {?,1,?,?}, "32"> {
778 let Inst{7} = lane{0};
782 def VLD4LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
783 def VLD4LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
785 // VLD1DUP : Vector Load (single element to all lanes)
786 // VLD2DUP : Vector Load (single 2-element structure to all lanes)
787 // VLD3DUP : Vector Load (single 3-element structure to all lanes)
788 // VLD4DUP : Vector Load (single 4-element structure to all lanes)
789 // FIXME: Not yet implemented.
790 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
792 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
794 // Classes for VST* pseudo-instructions with multi-register operands.
795 // These are expanded to real instructions after register allocation.
796 class VSTQPseudo<InstrItinClass itin>
797 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src), itin, "">;
798 class VSTQWBPseudo<InstrItinClass itin>
799 : PseudoNLdSt<(outs GPR:$wb),
800 (ins addrmode6:$addr, am6offset:$offset, QPR:$src), itin,
802 class VSTQQPseudo<InstrItinClass itin>
803 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src), itin, "">;
804 class VSTQQWBPseudo<InstrItinClass itin>
805 : PseudoNLdSt<(outs GPR:$wb),
806 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src), itin,
808 class VSTQQQQWBPseudo<InstrItinClass itin>
809 : PseudoNLdSt<(outs GPR:$wb),
810 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
813 // VST1 : Vector Store (multiple single elements)
814 class VST1D<bits<4> op7_4, string Dt>
815 : NLdSt<0,0b00,0b0111,op7_4, (outs), (ins addrmode6:$Rn, DPR:$Vd),
816 IIC_VST1, "vst1", Dt, "\\{$Vd\\}, $Rn", "", []> {
820 class VST1Q<bits<4> op7_4, string Dt>
821 : NLdSt<0,0b00,0b1010,op7_4, (outs),
822 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2), IIC_VST1x2,
823 "vst1", Dt, "\\{$Vd, $src2\\}, $Rn", "", []> {
825 let Inst{5-4} = Rn{5-4};
828 def VST1d8 : VST1D<{0,0,0,?}, "8">;
829 def VST1d16 : VST1D<{0,1,0,?}, "16">;
830 def VST1d32 : VST1D<{1,0,0,?}, "32">;
831 def VST1d64 : VST1D<{1,1,0,?}, "64">;
833 def VST1q8 : VST1Q<{0,0,?,?}, "8">;
834 def VST1q16 : VST1Q<{0,1,?,?}, "16">;
835 def VST1q32 : VST1Q<{1,0,?,?}, "32">;
836 def VST1q64 : VST1Q<{1,1,?,?}, "64">;
838 def VST1q8Pseudo : VSTQPseudo<IIC_VST1x2>;
839 def VST1q16Pseudo : VSTQPseudo<IIC_VST1x2>;
840 def VST1q32Pseudo : VSTQPseudo<IIC_VST1x2>;
841 def VST1q64Pseudo : VSTQPseudo<IIC_VST1x2>;
843 // ...with address register writeback:
844 class VST1DWB<bits<4> op7_4, string Dt>
845 : NLdSt<0, 0b00, 0b0111, op7_4, (outs GPR:$wb),
846 (ins addrmode6:$Rn, am6offset:$Rm, DPR:$Vd), IIC_VST1u,
847 "vst1", Dt, "\\{$Vd\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
850 class VST1QWB<bits<4> op7_4, string Dt>
851 : NLdSt<0, 0b00, 0b1010, op7_4, (outs GPR:$wb),
852 (ins addrmode6:$Rn, am6offset:$Rm, DPR:$Vd, DPR:$src2),
853 IIC_VST1x2u, "vst1", Dt, "\\{$Vd, $src2\\}, $Rn$Rm",
854 "$Rn.addr = $wb", []> {
855 let Inst{5-4} = Rn{5-4};
858 def VST1d8_UPD : VST1DWB<{0,0,0,?}, "8">;
859 def VST1d16_UPD : VST1DWB<{0,1,0,?}, "16">;
860 def VST1d32_UPD : VST1DWB<{1,0,0,?}, "32">;
861 def VST1d64_UPD : VST1DWB<{1,1,0,?}, "64">;
863 def VST1q8_UPD : VST1QWB<{0,0,?,?}, "8">;
864 def VST1q16_UPD : VST1QWB<{0,1,?,?}, "16">;
865 def VST1q32_UPD : VST1QWB<{1,0,?,?}, "32">;
866 def VST1q64_UPD : VST1QWB<{1,1,?,?}, "64">;
868 def VST1q8Pseudo_UPD : VSTQWBPseudo<IIC_VST1x2u>;
869 def VST1q16Pseudo_UPD : VSTQWBPseudo<IIC_VST1x2u>;
870 def VST1q32Pseudo_UPD : VSTQWBPseudo<IIC_VST1x2u>;
871 def VST1q64Pseudo_UPD : VSTQWBPseudo<IIC_VST1x2u>;
873 // ...with 3 registers (some of these are only for the disassembler):
874 class VST1D3<bits<4> op7_4, string Dt>
875 : NLdSt<0, 0b00, 0b0110, op7_4, (outs),
876 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3),
877 IIC_VST1x3, "vst1", Dt, "\\{$Vd, $src2, $src3\\}, $Rn", "", []> {
881 class VST1D3WB<bits<4> op7_4, string Dt>
882 : NLdSt<0, 0b00, 0b0110, op7_4, (outs GPR:$wb),
883 (ins addrmode6:$Rn, am6offset:$Rm,
884 DPR:$Vd, DPR:$src2, DPR:$src3),
885 IIC_VST1x3u, "vst1", Dt, "\\{$Vd, $src2, $src3\\}, $Rn$Rm",
886 "$Rn.addr = $wb", []> {
890 def VST1d8T : VST1D3<{0,0,0,?}, "8">;
891 def VST1d16T : VST1D3<{0,1,0,?}, "16">;
892 def VST1d32T : VST1D3<{1,0,0,?}, "32">;
893 def VST1d64T : VST1D3<{1,1,0,?}, "64">;
895 def VST1d8T_UPD : VST1D3WB<{0,0,0,?}, "8">;
896 def VST1d16T_UPD : VST1D3WB<{0,1,0,?}, "16">;
897 def VST1d32T_UPD : VST1D3WB<{1,0,0,?}, "32">;
898 def VST1d64T_UPD : VST1D3WB<{1,1,0,?}, "64">;
900 def VST1d64TPseudo : VSTQQPseudo<IIC_VST1x3>;
901 def VST1d64TPseudo_UPD : VSTQQWBPseudo<IIC_VST1x3u>;
903 // ...with 4 registers (some of these are only for the disassembler):
904 class VST1D4<bits<4> op7_4, string Dt>
905 : NLdSt<0, 0b00, 0b0010, op7_4, (outs),
906 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
907 IIC_VST1x4, "vst1", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn", "",
910 let Inst{5-4} = Rn{5-4};
912 class VST1D4WB<bits<4> op7_4, string Dt>
913 : NLdSt<0, 0b00, 0b0010, op7_4, (outs GPR:$wb),
914 (ins addrmode6:$Rn, am6offset:$Rm,
915 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST1x4u,
916 "vst1", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
917 "$Rn.addr = $wb", []> {
918 let Inst{5-4} = Rn{5-4};
921 def VST1d8Q : VST1D4<{0,0,?,?}, "8">;
922 def VST1d16Q : VST1D4<{0,1,?,?}, "16">;
923 def VST1d32Q : VST1D4<{1,0,?,?}, "32">;
924 def VST1d64Q : VST1D4<{1,1,?,?}, "64">;
926 def VST1d8Q_UPD : VST1D4WB<{0,0,?,?}, "8">;
927 def VST1d16Q_UPD : VST1D4WB<{0,1,?,?}, "16">;
928 def VST1d32Q_UPD : VST1D4WB<{1,0,?,?}, "32">;
929 def VST1d64Q_UPD : VST1D4WB<{1,1,?,?}, "64">;
931 def VST1d64QPseudo : VSTQQPseudo<IIC_VST1x4>;
932 def VST1d64QPseudo_UPD : VSTQQWBPseudo<IIC_VST1x4u>;
934 // VST2 : Vector Store (multiple 2-element structures)
935 class VST2D<bits<4> op11_8, bits<4> op7_4, string Dt>
936 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
937 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2),
938 IIC_VST2, "vst2", Dt, "\\{$Vd, $src2\\}, $Rn", "", []> {
940 let Inst{5-4} = Rn{5-4};
942 class VST2Q<bits<4> op7_4, string Dt>
943 : NLdSt<0, 0b00, 0b0011, op7_4, (outs),
944 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
945 IIC_VST2x2, "vst2", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn",
948 let Inst{5-4} = Rn{5-4};
951 def VST2d8 : VST2D<0b1000, {0,0,?,?}, "8">;
952 def VST2d16 : VST2D<0b1000, {0,1,?,?}, "16">;
953 def VST2d32 : VST2D<0b1000, {1,0,?,?}, "32">;
955 def VST2q8 : VST2Q<{0,0,?,?}, "8">;
956 def VST2q16 : VST2Q<{0,1,?,?}, "16">;
957 def VST2q32 : VST2Q<{1,0,?,?}, "32">;
959 def VST2d8Pseudo : VSTQPseudo<IIC_VST2>;
960 def VST2d16Pseudo : VSTQPseudo<IIC_VST2>;
961 def VST2d32Pseudo : VSTQPseudo<IIC_VST2>;
963 def VST2q8Pseudo : VSTQQPseudo<IIC_VST2x2>;
964 def VST2q16Pseudo : VSTQQPseudo<IIC_VST2x2>;
965 def VST2q32Pseudo : VSTQQPseudo<IIC_VST2x2>;
967 // ...with address register writeback:
968 class VST2DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
969 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
970 (ins addrmode6:$Rn, am6offset:$Rm, DPR:$Vd, DPR:$src2),
971 IIC_VST2u, "vst2", Dt, "\\{$Vd, $src2\\}, $Rn$Rm",
972 "$Rn.addr = $wb", []> {
973 let Inst{5-4} = Rn{5-4};
975 class VST2QWB<bits<4> op7_4, string Dt>
976 : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
977 (ins addrmode6:$Rn, am6offset:$Rm,
978 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST2x2u,
979 "vst2", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
980 "$Rn.addr = $wb", []> {
981 let Inst{5-4} = Rn{5-4};
984 def VST2d8_UPD : VST2DWB<0b1000, {0,0,?,?}, "8">;
985 def VST2d16_UPD : VST2DWB<0b1000, {0,1,?,?}, "16">;
986 def VST2d32_UPD : VST2DWB<0b1000, {1,0,?,?}, "32">;
988 def VST2q8_UPD : VST2QWB<{0,0,?,?}, "8">;
989 def VST2q16_UPD : VST2QWB<{0,1,?,?}, "16">;
990 def VST2q32_UPD : VST2QWB<{1,0,?,?}, "32">;
992 def VST2d8Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
993 def VST2d16Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
994 def VST2d32Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
996 def VST2q8Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
997 def VST2q16Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
998 def VST2q32Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
1000 // ...with double-spaced registers (for disassembly only):
1001 def VST2b8 : VST2D<0b1001, {0,0,?,?}, "8">;
1002 def VST2b16 : VST2D<0b1001, {0,1,?,?}, "16">;
1003 def VST2b32 : VST2D<0b1001, {1,0,?,?}, "32">;
1004 def VST2b8_UPD : VST2DWB<0b1001, {0,0,?,?}, "8">;
1005 def VST2b16_UPD : VST2DWB<0b1001, {0,1,?,?}, "16">;
1006 def VST2b32_UPD : VST2DWB<0b1001, {1,0,?,?}, "32">;
1008 // VST3 : Vector Store (multiple 3-element structures)
1009 class VST3D<bits<4> op11_8, bits<4> op7_4, string Dt>
1010 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1011 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3,
1012 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn", "", []> {
1014 let Inst{4} = Rn{4};
1017 def VST3d8 : VST3D<0b0100, {0,0,0,?}, "8">;
1018 def VST3d16 : VST3D<0b0100, {0,1,0,?}, "16">;
1019 def VST3d32 : VST3D<0b0100, {1,0,0,?}, "32">;
1021 def VST3d8Pseudo : VSTQQPseudo<IIC_VST3>;
1022 def VST3d16Pseudo : VSTQQPseudo<IIC_VST3>;
1023 def VST3d32Pseudo : VSTQQPseudo<IIC_VST3>;
1025 // ...with address register writeback:
1026 class VST3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1027 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1028 (ins addrmode6:$Rn, am6offset:$Rm,
1029 DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3u,
1030 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn$Rm",
1031 "$Rn.addr = $wb", []> {
1032 let Inst{4} = Rn{4};
1035 def VST3d8_UPD : VST3DWB<0b0100, {0,0,0,?}, "8">;
1036 def VST3d16_UPD : VST3DWB<0b0100, {0,1,0,?}, "16">;
1037 def VST3d32_UPD : VST3DWB<0b0100, {1,0,0,?}, "32">;
1039 def VST3d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1040 def VST3d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1041 def VST3d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1043 // ...with double-spaced registers (non-updating versions for disassembly only):
1044 def VST3q8 : VST3D<0b0101, {0,0,0,?}, "8">;
1045 def VST3q16 : VST3D<0b0101, {0,1,0,?}, "16">;
1046 def VST3q32 : VST3D<0b0101, {1,0,0,?}, "32">;
1047 def VST3q8_UPD : VST3DWB<0b0101, {0,0,0,?}, "8">;
1048 def VST3q16_UPD : VST3DWB<0b0101, {0,1,0,?}, "16">;
1049 def VST3q32_UPD : VST3DWB<0b0101, {1,0,0,?}, "32">;
1051 def VST3q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1052 def VST3q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1053 def VST3q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1055 // ...alternate versions to be allocated odd register numbers:
1056 def VST3q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1057 def VST3q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1058 def VST3q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1060 // VST4 : Vector Store (multiple 4-element structures)
1061 class VST4D<bits<4> op11_8, bits<4> op7_4, string Dt>
1062 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1063 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
1064 IIC_VST4, "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn",
1067 let Inst{5-4} = Rn{5-4};
1070 def VST4d8 : VST4D<0b0000, {0,0,?,?}, "8">;
1071 def VST4d16 : VST4D<0b0000, {0,1,?,?}, "16">;
1072 def VST4d32 : VST4D<0b0000, {1,0,?,?}, "32">;
1074 def VST4d8Pseudo : VSTQQPseudo<IIC_VST4>;
1075 def VST4d16Pseudo : VSTQQPseudo<IIC_VST4>;
1076 def VST4d32Pseudo : VSTQQPseudo<IIC_VST4>;
1078 // ...with address register writeback:
1079 class VST4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1080 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1081 (ins addrmode6:$Rn, am6offset:$Rm,
1082 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST4u,
1083 "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
1084 "$Rn.addr = $wb", []> {
1085 let Inst{5-4} = Rn{5-4};
1088 def VST4d8_UPD : VST4DWB<0b0000, {0,0,?,?}, "8">;
1089 def VST4d16_UPD : VST4DWB<0b0000, {0,1,?,?}, "16">;
1090 def VST4d32_UPD : VST4DWB<0b0000, {1,0,?,?}, "32">;
1092 def VST4d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1093 def VST4d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1094 def VST4d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1096 // ...with double-spaced registers (non-updating versions for disassembly only):
1097 def VST4q8 : VST4D<0b0001, {0,0,?,?}, "8">;
1098 def VST4q16 : VST4D<0b0001, {0,1,?,?}, "16">;
1099 def VST4q32 : VST4D<0b0001, {1,0,?,?}, "32">;
1100 def VST4q8_UPD : VST4DWB<0b0001, {0,0,?,?}, "8">;
1101 def VST4q16_UPD : VST4DWB<0b0001, {0,1,?,?}, "16">;
1102 def VST4q32_UPD : VST4DWB<0b0001, {1,0,?,?}, "32">;
1104 def VST4q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1105 def VST4q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1106 def VST4q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1108 // ...alternate versions to be allocated odd register numbers:
1109 def VST4q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1110 def VST4q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1111 def VST4q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1113 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
1115 // Classes for VST*LN pseudo-instructions with multi-register operands.
1116 // These are expanded to real instructions after register allocation.
1117 class VSTQLNPseudo<InstrItinClass itin>
1118 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
1120 class VSTQLNWBPseudo<InstrItinClass itin>
1121 : PseudoNLdSt<(outs GPR:$wb),
1122 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
1123 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1124 class VSTQQLNPseudo<InstrItinClass itin>
1125 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
1127 class VSTQQLNWBPseudo<InstrItinClass itin>
1128 : PseudoNLdSt<(outs GPR:$wb),
1129 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
1130 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1131 class VSTQQQQLNPseudo<InstrItinClass itin>
1132 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
1134 class VSTQQQQLNWBPseudo<InstrItinClass itin>
1135 : PseudoNLdSt<(outs GPR:$wb),
1136 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
1137 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1139 // VST1LN : Vector Store (single element from one lane)
1140 class VST1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1141 PatFrag StoreOp, SDNode ExtractOp>
1142 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1143 (ins addrmode6:$Rn, DPR:$Vd, nohash_imm:$lane),
1144 IIC_VST1ln, "vst1", Dt, "\\{$Vd[$lane]\\}, $Rn", "",
1145 [(StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane), addrmode6:$Rn)]> {
1148 class VST1QLNPseudo<ValueType Ty, PatFrag StoreOp, SDNode ExtractOp>
1149 : VSTQLNPseudo<IIC_VST1ln> {
1150 let Pattern = [(StoreOp (ExtractOp (Ty QPR:$src), imm:$lane),
1154 def VST1LNd8 : VST1LN<0b0000, {?,?,?,0}, "8", v8i8, truncstorei8,
1156 let Inst{7-5} = lane{2-0};
1158 def VST1LNd16 : VST1LN<0b0100, {?,?,0,?}, "16", v4i16, truncstorei16,
1160 let Inst{7-6} = lane{1-0};
1161 let Inst{4} = Rn{5};
1163 def VST1LNd32 : VST1LN<0b1000, {?,0,?,?}, "32", v2i32, store, extractelt> {
1164 let Inst{7} = lane{0};
1165 let Inst{5-4} = Rn{5-4};
1168 def VST1LNq8Pseudo : VST1QLNPseudo<v16i8, truncstorei8, NEONvgetlaneu>;
1169 def VST1LNq16Pseudo : VST1QLNPseudo<v8i16, truncstorei16, NEONvgetlaneu>;
1170 def VST1LNq32Pseudo : VST1QLNPseudo<v4i32, store, extractelt>;
1172 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
1174 // ...with address register writeback:
1175 class VST1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1176 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1177 (ins addrmode6:$Rn, am6offset:$Rm,
1178 DPR:$Vd, nohash_imm:$lane), IIC_VST1lnu, "vst1", Dt,
1179 "\\{$Vd[$lane]\\}, $Rn$Rm",
1180 "$Rn.addr = $wb", []>;
1182 def VST1LNd8_UPD : VST1LNWB<0b0000, {?,?,?,0}, "8"> {
1183 let Inst{7-5} = lane{2-0};
1185 def VST1LNd16_UPD : VST1LNWB<0b0100, {?,?,0,?}, "16"> {
1186 let Inst{7-6} = lane{1-0};
1187 let Inst{4} = Rn{5};
1189 def VST1LNd32_UPD : VST1LNWB<0b1000, {?,0,?,?}, "32"> {
1190 let Inst{7} = lane{0};
1191 let Inst{5-4} = Rn{5-4};
1194 def VST1LNq8Pseudo_UPD : VSTQLNWBPseudo<IIC_VST1lnu>;
1195 def VST1LNq16Pseudo_UPD : VSTQLNWBPseudo<IIC_VST1lnu>;
1196 def VST1LNq32Pseudo_UPD : VSTQLNWBPseudo<IIC_VST1lnu>;
1198 // VST2LN : Vector Store (single 2-element structure from one lane)
1199 class VST2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1200 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1201 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, nohash_imm:$lane),
1202 IIC_VST2ln, "vst2", Dt, "\\{$Vd[$lane], $src2[$lane]\\}, $Rn",
1205 let Inst{4} = Rn{4};
1208 def VST2LNd8 : VST2LN<0b0001, {?,?,?,?}, "8"> {
1209 let Inst{7-5} = lane{2-0};
1211 def VST2LNd16 : VST2LN<0b0101, {?,?,0,?}, "16"> {
1212 let Inst{7-6} = lane{1-0};
1214 def VST2LNd32 : VST2LN<0b1001, {?,0,0,?}, "32"> {
1215 let Inst{7} = lane{0};
1218 def VST2LNd8Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1219 def VST2LNd16Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1220 def VST2LNd32Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1222 // ...with double-spaced registers:
1223 def VST2LNq16 : VST2LN<0b0101, {?,?,1,?}, "16"> {
1224 let Inst{7-6} = lane{1-0};
1225 let Inst{4} = Rn{4};
1227 def VST2LNq32 : VST2LN<0b1001, {?,1,0,?}, "32"> {
1228 let Inst{7} = lane{0};
1229 let Inst{4} = Rn{4};
1232 def VST2LNq16Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
1233 def VST2LNq32Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
1235 // ...with address register writeback:
1236 class VST2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1237 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1238 (ins addrmode6:$addr, am6offset:$offset,
1239 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VST2lnu, "vst2", Dt,
1240 "\\{$src1[$lane], $src2[$lane]\\}, $addr$offset",
1241 "$addr.addr = $wb", []> {
1242 let Inst{4} = Rn{4};
1245 def VST2LNd8_UPD : VST2LNWB<0b0001, {?,?,?,?}, "8"> {
1246 let Inst{7-5} = lane{2-0};
1248 def VST2LNd16_UPD : VST2LNWB<0b0101, {?,?,0,?}, "16"> {
1249 let Inst{7-6} = lane{1-0};
1251 def VST2LNd32_UPD : VST2LNWB<0b1001, {?,0,0,?}, "32"> {
1252 let Inst{7} = lane{0};
1255 def VST2LNd8Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1256 def VST2LNd16Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1257 def VST2LNd32Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1259 def VST2LNq16_UPD : VST2LNWB<0b0101, {?,?,1,?}, "16"> {
1260 let Inst{7-6} = lane{1-0};
1262 def VST2LNq32_UPD : VST2LNWB<0b1001, {?,1,0,?}, "32"> {
1263 let Inst{7} = lane{0};
1266 def VST2LNq16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
1267 def VST2LNq32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
1269 // VST3LN : Vector Store (single 3-element structure from one lane)
1270 class VST3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1271 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1272 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3,
1273 nohash_imm:$lane), IIC_VST3ln, "vst3", Dt,
1274 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn", "", []> {
1278 def VST3LNd8 : VST3LN<0b0010, {?,?,?,0}, "8"> {
1279 let Inst{7-5} = lane{2-0};
1281 def VST3LNd16 : VST3LN<0b0110, {?,?,0,0}, "16"> {
1282 let Inst{7-6} = lane{1-0};
1284 def VST3LNd32 : VST3LN<0b1010, {?,0,0,0}, "32"> {
1285 let Inst{7} = lane{0};
1288 def VST3LNd8Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1289 def VST3LNd16Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1290 def VST3LNd32Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1292 // ...with double-spaced registers:
1293 def VST3LNq16 : VST3LN<0b0110, {?,?,1,0}, "16"> {
1294 let Inst{7-6} = lane{1-0};
1296 def VST3LNq32 : VST3LN<0b1010, {?,1,0,0}, "32"> {
1297 let Inst{7} = lane{0};
1300 def VST3LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
1301 def VST3LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
1303 // ...with address register writeback:
1304 class VST3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1305 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1306 (ins addrmode6:$Rn, am6offset:$Rm,
1307 DPR:$Vd, DPR:$src2, DPR:$src3, nohash_imm:$lane),
1308 IIC_VST3lnu, "vst3", Dt,
1309 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn$Rm",
1310 "$Rn.addr = $wb", []>;
1312 def VST3LNd8_UPD : VST3LNWB<0b0010, {?,?,?,0}, "8"> {
1313 let Inst{7-5} = lane{2-0};
1315 def VST3LNd16_UPD : VST3LNWB<0b0110, {?,?,0,0}, "16"> {
1316 let Inst{7-6} = lane{1-0};
1318 def VST3LNd32_UPD : VST3LNWB<0b1010, {?,0,0,0}, "32"> {
1319 let Inst{7} = lane{0};
1322 def VST3LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1323 def VST3LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1324 def VST3LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1326 def VST3LNq16_UPD : VST3LNWB<0b0110, {?,?,1,0}, "16"> {
1327 let Inst{7-6} = lane{1-0};
1329 def VST3LNq32_UPD : VST3LNWB<0b1010, {?,1,0,0}, "32"> {
1330 let Inst{7} = lane{0};
1333 def VST3LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
1334 def VST3LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
1336 // VST4LN : Vector Store (single 4-element structure from one lane)
1337 class VST4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1338 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1339 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4,
1340 nohash_imm:$lane), IIC_VST4ln, "vst4", Dt,
1341 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn",
1344 let Inst{4} = Rn{4};
1347 def VST4LNd8 : VST4LN<0b0011, {?,?,?,?}, "8"> {
1348 let Inst{7-5} = lane{2-0};
1350 def VST4LNd16 : VST4LN<0b0111, {?,?,0,?}, "16"> {
1351 let Inst{7-6} = lane{1-0};
1353 def VST4LNd32 : VST4LN<0b1011, {?,0,?,?}, "32"> {
1354 let Inst{7} = lane{0};
1355 let Inst{5} = Rn{5};
1358 def VST4LNd8Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1359 def VST4LNd16Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1360 def VST4LNd32Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1362 // ...with double-spaced registers:
1363 def VST4LNq16 : VST4LN<0b0111, {?,?,1,?}, "16"> {
1364 let Inst{7-6} = lane{1-0};
1366 def VST4LNq32 : VST4LN<0b1011, {?,1,?,?}, "32"> {
1367 let Inst{7} = lane{0};
1368 let Inst{5} = Rn{5};
1371 def VST4LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
1372 def VST4LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
1374 // ...with address register writeback:
1375 class VST4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1376 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1377 (ins addrmode6:$Rn, am6offset:$Rm,
1378 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
1379 IIC_VST4lnu, "vst4", Dt,
1380 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn$Rm",
1381 "$Rn.addr = $wb", []> {
1382 let Inst{4} = Rn{4};
1385 def VST4LNd8_UPD : VST4LNWB<0b0011, {?,?,?,?}, "8"> {
1386 let Inst{7-5} = lane{2-0};
1388 def VST4LNd16_UPD : VST4LNWB<0b0111, {?,?,0,?}, "16"> {
1389 let Inst{7-6} = lane{1-0};
1391 def VST4LNd32_UPD : VST4LNWB<0b1011, {?,0,?,?}, "32"> {
1392 let Inst{7} = lane{0};
1393 let Inst{5} = Rn{5};
1396 def VST4LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
1397 def VST4LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
1398 def VST4LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
1400 def VST4LNq16_UPD : VST4LNWB<0b0111, {?,?,1,?}, "16"> {
1401 let Inst{7-6} = lane{1-0};
1403 def VST4LNq32_UPD : VST4LNWB<0b1011, {?,1,?,?}, "32"> {
1404 let Inst{7} = lane{0};
1405 let Inst{5} = Rn{5};
1408 def VST4LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
1409 def VST4LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
1411 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
1414 //===----------------------------------------------------------------------===//
1415 // NEON pattern fragments
1416 //===----------------------------------------------------------------------===//
1418 // Extract D sub-registers of Q registers.
1419 def DSubReg_i8_reg : SDNodeXForm<imm, [{
1420 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1421 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/8, MVT::i32);
1423 def DSubReg_i16_reg : SDNodeXForm<imm, [{
1424 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1425 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/4, MVT::i32);
1427 def DSubReg_i32_reg : SDNodeXForm<imm, [{
1428 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1429 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/2, MVT::i32);
1431 def DSubReg_f64_reg : SDNodeXForm<imm, [{
1432 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1433 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue(), MVT::i32);
1436 // Extract S sub-registers of Q/D registers.
1437 def SSubReg_f32_reg : SDNodeXForm<imm, [{
1438 assert(ARM::ssub_3 == ARM::ssub_0+3 && "Unexpected subreg numbering");
1439 return CurDAG->getTargetConstant(ARM::ssub_0 + N->getZExtValue(), MVT::i32);
1442 // Translate lane numbers from Q registers to D subregs.
1443 def SubReg_i8_lane : SDNodeXForm<imm, [{
1444 return CurDAG->getTargetConstant(N->getZExtValue() & 7, MVT::i32);
1446 def SubReg_i16_lane : SDNodeXForm<imm, [{
1447 return CurDAG->getTargetConstant(N->getZExtValue() & 3, MVT::i32);
1449 def SubReg_i32_lane : SDNodeXForm<imm, [{
1450 return CurDAG->getTargetConstant(N->getZExtValue() & 1, MVT::i32);
1453 //===----------------------------------------------------------------------===//
1454 // Instruction Classes
1455 //===----------------------------------------------------------------------===//
1457 // Basic 2-register operations: single-, double- and quad-register.
1458 class N2VS<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1459 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
1460 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
1461 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
1462 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src),
1463 IIC_VUNAD, OpcodeStr, Dt, "$dst, $src", "", []>;
1464 class N2VD<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1465 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
1466 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
1467 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
1468 (ins DPR:$src), IIC_VUNAD, OpcodeStr, Dt,"$dst, $src", "",
1469 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src))))]>;
1470 class N2VQ<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1471 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
1472 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
1473 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
1474 (ins QPR:$src), IIC_VUNAQ, OpcodeStr, Dt,"$dst, $src", "",
1475 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src))))]>;
1477 // Basic 2-register intrinsics, both double- and quad-register.
1478 class N2VDInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1479 bits<2> op17_16, bits<5> op11_7, bit op4,
1480 InstrItinClass itin, string OpcodeStr, string Dt,
1481 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1482 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
1483 (ins DPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
1484 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
1485 class N2VQInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1486 bits<2> op17_16, bits<5> op11_7, bit op4,
1487 InstrItinClass itin, string OpcodeStr, string Dt,
1488 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1489 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
1490 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
1491 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
1493 // Narrow 2-register operations.
1494 class N2VN<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1495 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1496 InstrItinClass itin, string OpcodeStr, string Dt,
1497 ValueType TyD, ValueType TyQ, SDNode OpNode>
1498 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$dst),
1499 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
1500 [(set DPR:$dst, (TyD (OpNode (TyQ QPR:$src))))]>;
1502 // Narrow 2-register intrinsics.
1503 class N2VNInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1504 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1505 InstrItinClass itin, string OpcodeStr, string Dt,
1506 ValueType TyD, ValueType TyQ, Intrinsic IntOp>
1507 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$dst),
1508 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
1509 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src))))]>;
1511 // Long 2-register operations (currently only used for VMOVL).
1512 class N2VL<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1513 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1514 InstrItinClass itin, string OpcodeStr, string Dt,
1515 ValueType TyQ, ValueType TyD, SDNode OpNode>
1516 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$dst),
1517 (ins DPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
1518 [(set QPR:$dst, (TyQ (OpNode (TyD DPR:$src))))]>;
1520 // 2-register shuffles (VTRN/VZIP/VUZP), both double- and quad-register.
1521 class N2VDShuffle<bits<2> op19_18, bits<5> op11_7, string OpcodeStr, string Dt>
1522 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 0, 0, (outs DPR:$dst1, DPR:$dst2),
1523 (ins DPR:$src1, DPR:$src2), IIC_VPERMD,
1524 OpcodeStr, Dt, "$dst1, $dst2",
1525 "$src1 = $dst1, $src2 = $dst2", []>;
1526 class N2VQShuffle<bits<2> op19_18, bits<5> op11_7,
1527 InstrItinClass itin, string OpcodeStr, string Dt>
1528 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 1, 0, (outs QPR:$dst1, QPR:$dst2),
1529 (ins QPR:$src1, QPR:$src2), itin, OpcodeStr, Dt, "$dst1, $dst2",
1530 "$src1 = $dst1, $src2 = $dst2", []>;
1532 // Basic 3-register operations: single-, double- and quad-register.
1533 class N3VS<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1534 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1535 SDNode OpNode, bit Commutable>
1536 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1537 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src1, DPR_VFP2:$src2), N3RegFrm,
1538 IIC_VBIND, OpcodeStr, Dt, "$dst, $src1, $src2", "", []> {
1539 let isCommutable = Commutable;
1542 class N3VD<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1543 InstrItinClass itin, string OpcodeStr, string Dt,
1544 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
1545 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1546 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1547 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
1548 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
1549 let isCommutable = Commutable;
1551 // Same as N3VD but no data type.
1552 class N3VDX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1553 InstrItinClass itin, string OpcodeStr,
1554 ValueType ResTy, ValueType OpTy,
1555 SDNode OpNode, bit Commutable>
1556 : N3VX<op24, op23, op21_20, op11_8, 0, op4,
1557 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
1558 OpcodeStr, "$dst, $src1, $src2", "",
1559 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src1), (OpTy DPR:$src2))))]>{
1560 let isCommutable = Commutable;
1563 class N3VDSL<bits<2> op21_20, bits<4> op11_8,
1564 InstrItinClass itin, string OpcodeStr, string Dt,
1565 ValueType Ty, SDNode ShOp>
1566 : N3V<0, 1, op21_20, op11_8, 1, 0,
1567 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1568 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1569 [(set (Ty DPR:$dst),
1570 (Ty (ShOp (Ty DPR:$src1),
1571 (Ty (NEONvduplane (Ty DPR_VFP2:$src2),imm:$lane)))))]> {
1572 let isCommutable = 0;
1574 class N3VDSL16<bits<2> op21_20, bits<4> op11_8,
1575 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1576 : N3V<0, 1, op21_20, op11_8, 1, 0,
1577 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1578 NVMulSLFrm, IIC_VMULi16D, OpcodeStr, Dt,"$dst, $src1, $src2[$lane]","",
1579 [(set (Ty DPR:$dst),
1580 (Ty (ShOp (Ty DPR:$src1),
1581 (Ty (NEONvduplane (Ty DPR_8:$src2), imm:$lane)))))]> {
1582 let isCommutable = 0;
1585 class N3VQ<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1586 InstrItinClass itin, string OpcodeStr, string Dt,
1587 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
1588 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1589 (outs QPR:$Qd), (ins QPR:$Qn, QPR:$Qm), N3RegFrm, itin,
1590 OpcodeStr, Dt, "$Qd, $Qn, $Qm", "",
1591 [(set QPR:$Qd, (ResTy (OpNode (OpTy QPR:$Qn), (OpTy QPR:$Qm))))]> {
1592 let isCommutable = Commutable;
1594 class N3VQX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1595 InstrItinClass itin, string OpcodeStr,
1596 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
1597 : N3VX<op24, op23, op21_20, op11_8, 1, op4,
1598 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), N3RegFrm, itin,
1599 OpcodeStr, "$dst, $src1, $src2", "",
1600 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src1), (OpTy QPR:$src2))))]>{
1601 let isCommutable = Commutable;
1603 class N3VQSL<bits<2> op21_20, bits<4> op11_8,
1604 InstrItinClass itin, string OpcodeStr, string Dt,
1605 ValueType ResTy, ValueType OpTy, SDNode ShOp>
1606 : N3V<1, 1, op21_20, op11_8, 1, 0,
1607 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1608 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1609 [(set (ResTy QPR:$dst),
1610 (ResTy (ShOp (ResTy QPR:$src1),
1611 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1613 let isCommutable = 0;
1615 class N3VQSL16<bits<2> op21_20, bits<4> op11_8, string OpcodeStr, string Dt,
1616 ValueType ResTy, ValueType OpTy, SDNode ShOp>
1617 : N3V<1, 1, op21_20, op11_8, 1, 0,
1618 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1619 NVMulSLFrm, IIC_VMULi16Q, OpcodeStr, Dt,"$dst, $src1, $src2[$lane]","",
1620 [(set (ResTy QPR:$dst),
1621 (ResTy (ShOp (ResTy QPR:$src1),
1622 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
1624 let isCommutable = 0;
1627 // Basic 3-register intrinsics, both double- and quad-register.
1628 class N3VDInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1629 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1630 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
1631 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1632 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), f, itin,
1633 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
1634 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
1635 let isCommutable = Commutable;
1637 class N3VDIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1638 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
1639 : N3V<0, 1, op21_20, op11_8, 1, 0,
1640 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1641 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1642 [(set (Ty DPR:$dst),
1643 (Ty (IntOp (Ty DPR:$src1),
1644 (Ty (NEONvduplane (Ty DPR_VFP2:$src2),
1646 let isCommutable = 0;
1648 class N3VDIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1649 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
1650 : N3V<0, 1, op21_20, op11_8, 1, 0,
1651 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1652 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1653 [(set (Ty DPR:$dst),
1654 (Ty (IntOp (Ty DPR:$src1),
1655 (Ty (NEONvduplane (Ty DPR_8:$src2), imm:$lane)))))]> {
1656 let isCommutable = 0;
1658 class N3VDIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1659 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1660 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1661 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1662 (outs DPR:$Vd), (ins DPR:$Vm, DPR:$Vn), f, itin,
1663 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
1664 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (OpTy DPR:$Vn))))]> {
1665 let isCommutable = 0;
1668 class N3VQInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1669 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1670 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
1671 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1672 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), f, itin,
1673 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
1674 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
1675 let isCommutable = Commutable;
1677 class N3VQIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1678 string OpcodeStr, string Dt,
1679 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1680 : N3V<1, 1, op21_20, op11_8, 1, 0,
1681 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1682 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1683 [(set (ResTy QPR:$dst),
1684 (ResTy (IntOp (ResTy QPR:$src1),
1685 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1687 let isCommutable = 0;
1689 class N3VQIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1690 string OpcodeStr, string Dt,
1691 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1692 : N3V<1, 1, op21_20, op11_8, 1, 0,
1693 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1694 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1695 [(set (ResTy QPR:$dst),
1696 (ResTy (IntOp (ResTy QPR:$src1),
1697 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
1699 let isCommutable = 0;
1701 class N3VQIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1702 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1703 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1704 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1705 (outs QPR:$Vd), (ins QPR:$Vm, QPR:$Vn), f, itin,
1706 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
1707 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (OpTy QPR:$Vn))))]> {
1708 let isCommutable = 0;
1711 // Multiply-Add/Sub operations: single-, double- and quad-register.
1712 class N3VSMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1713 InstrItinClass itin, string OpcodeStr, string Dt,
1714 ValueType Ty, SDNode MulOp, SDNode OpNode>
1715 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1716 (outs DPR_VFP2:$dst),
1717 (ins DPR_VFP2:$src1, DPR_VFP2:$src2, DPR_VFP2:$src3), N3RegFrm, itin,
1718 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst", []>;
1720 class N3VDMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1721 InstrItinClass itin, string OpcodeStr, string Dt,
1722 ValueType Ty, SDNode MulOp, SDNode OpNode>
1723 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1724 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1725 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1726 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
1727 (Ty (MulOp DPR:$Vn, DPR:$Vm)))))]>;
1729 class N3VDMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1730 string OpcodeStr, string Dt,
1731 ValueType Ty, SDNode MulOp, SDNode ShOp>
1732 : N3V<0, 1, op21_20, op11_8, 1, 0,
1734 (ins DPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1736 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1737 [(set (Ty DPR:$dst),
1738 (Ty (ShOp (Ty DPR:$src1),
1739 (Ty (MulOp DPR:$src2,
1740 (Ty (NEONvduplane (Ty DPR_VFP2:$src3),
1742 class N3VDMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1743 string OpcodeStr, string Dt,
1744 ValueType Ty, SDNode MulOp, SDNode ShOp>
1745 : N3V<0, 1, op21_20, op11_8, 1, 0,
1747 (ins DPR:$src1, DPR:$Vn, DPR_8:$Vm, nohash_imm:$lane),
1749 OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "$src1 = $Vd",
1751 (Ty (ShOp (Ty DPR:$src1),
1753 (Ty (NEONvduplane (Ty DPR_8:$Vm),
1756 class N3VQMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1757 InstrItinClass itin, string OpcodeStr, string Dt, ValueType Ty,
1758 SDNode MulOp, SDNode OpNode>
1759 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1760 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
1761 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1762 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
1763 (Ty (MulOp QPR:$Vn, QPR:$Vm)))))]>;
1764 class N3VQMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1765 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1766 SDNode MulOp, SDNode ShOp>
1767 : N3V<1, 1, op21_20, op11_8, 1, 0,
1769 (ins QPR:$src1, QPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1771 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1772 [(set (ResTy QPR:$dst),
1773 (ResTy (ShOp (ResTy QPR:$src1),
1774 (ResTy (MulOp QPR:$src2,
1775 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src3),
1777 class N3VQMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1778 string OpcodeStr, string Dt,
1779 ValueType ResTy, ValueType OpTy,
1780 SDNode MulOp, SDNode ShOp>
1781 : N3V<1, 1, op21_20, op11_8, 1, 0,
1783 (ins QPR:$src1, QPR:$src2, DPR_8:$src3, nohash_imm:$lane),
1785 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1786 [(set (ResTy QPR:$dst),
1787 (ResTy (ShOp (ResTy QPR:$src1),
1788 (ResTy (MulOp QPR:$src2,
1789 (ResTy (NEONvduplane (OpTy DPR_8:$src3),
1792 // Neon Intrinsic-Op instructions (VABA): double- and quad-register.
1793 class N3VDIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1794 InstrItinClass itin, string OpcodeStr, string Dt,
1795 ValueType Ty, Intrinsic IntOp, SDNode OpNode>
1796 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1797 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1798 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1799 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
1800 (Ty (IntOp (Ty DPR:$Vn), (Ty DPR:$Vm))))))]>;
1801 class N3VQIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1802 InstrItinClass itin, string OpcodeStr, string Dt,
1803 ValueType Ty, Intrinsic IntOp, SDNode OpNode>
1804 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1805 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
1806 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1807 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
1808 (Ty (IntOp (Ty QPR:$Vn), (Ty QPR:$Vm))))))]>;
1810 // Neon 3-argument intrinsics, both double- and quad-register.
1811 // The destination register is also used as the first source operand register.
1812 class N3VDInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1813 InstrItinClass itin, string OpcodeStr, string Dt,
1814 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1815 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1816 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3), N3RegFrm, itin,
1817 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1818 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src1),
1819 (OpTy DPR:$src2), (OpTy DPR:$src3))))]>;
1820 class N3VQInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1821 InstrItinClass itin, string OpcodeStr, string Dt,
1822 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1823 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1824 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3), N3RegFrm, itin,
1825 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1826 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src1),
1827 (OpTy QPR:$src2), (OpTy QPR:$src3))))]>;
1829 // Long Multiply-Add/Sub operations.
1830 class N3VLMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1831 InstrItinClass itin, string OpcodeStr, string Dt,
1832 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
1833 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1834 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1835 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1836 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
1837 (TyQ (MulOp (TyD DPR:$Vn),
1838 (TyD DPR:$Vm)))))]>;
1839 class N3VLMulOpSL<bit op24, bits<2> op21_20, bits<4> op11_8,
1840 InstrItinClass itin, string OpcodeStr, string Dt,
1841 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
1842 : N3V<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$dst),
1843 (ins QPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1845 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1847 (OpNode (TyQ QPR:$src1),
1848 (TyQ (MulOp (TyD DPR:$src2),
1849 (TyD (NEONvduplane (TyD DPR_VFP2:$src3),
1851 class N3VLMulOpSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1852 InstrItinClass itin, string OpcodeStr, string Dt,
1853 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
1854 : N3V<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$dst),
1855 (ins QPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane),
1857 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1859 (OpNode (TyQ QPR:$src1),
1860 (TyQ (MulOp (TyD DPR:$src2),
1861 (TyD (NEONvduplane (TyD DPR_8:$src3),
1864 // Long Intrinsic-Op vector operations with explicit extend (VABAL).
1865 class N3VLIntExtOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1866 InstrItinClass itin, string OpcodeStr, string Dt,
1867 ValueType TyQ, ValueType TyD, Intrinsic IntOp, SDNode ExtOp,
1869 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1870 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1871 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1872 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
1873 (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
1874 (TyD DPR:$Vm)))))))]>;
1876 // Neon Long 3-argument intrinsic. The destination register is
1877 // a quad-register and is also used as the first source operand register.
1878 class N3VLInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1879 InstrItinClass itin, string OpcodeStr, string Dt,
1880 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
1881 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1882 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1883 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1885 (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$Vn), (TyD DPR:$Vm))))]>;
1886 class N3VLInt3SL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1887 string OpcodeStr, string Dt,
1888 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1889 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1891 (ins QPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1893 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1894 [(set (ResTy QPR:$dst),
1895 (ResTy (IntOp (ResTy QPR:$src1),
1897 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src3),
1899 class N3VLInt3SL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1900 InstrItinClass itin, string OpcodeStr, string Dt,
1901 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1902 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1904 (ins QPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane),
1906 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1907 [(set (ResTy QPR:$dst),
1908 (ResTy (IntOp (ResTy QPR:$src1),
1910 (OpTy (NEONvduplane (OpTy DPR_8:$src3),
1913 // Narrowing 3-register intrinsics.
1914 class N3VNInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1915 string OpcodeStr, string Dt, ValueType TyD, ValueType TyQ,
1916 Intrinsic IntOp, bit Commutable>
1917 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1918 (outs DPR:$dst), (ins QPR:$src1, QPR:$src2), N3RegFrm, IIC_VBINi4D,
1919 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1920 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src1), (TyQ QPR:$src2))))]> {
1921 let isCommutable = Commutable;
1924 // Long 3-register operations.
1925 class N3VL<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1926 InstrItinClass itin, string OpcodeStr, string Dt,
1927 ValueType TyQ, ValueType TyD, SDNode OpNode, bit Commutable>
1928 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1929 (outs QPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
1930 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1931 [(set QPR:$dst, (TyQ (OpNode (TyD DPR:$src1), (TyD DPR:$src2))))]> {
1932 let isCommutable = Commutable;
1934 class N3VLSL<bit op24, bits<2> op21_20, bits<4> op11_8,
1935 InstrItinClass itin, string OpcodeStr, string Dt,
1936 ValueType TyQ, ValueType TyD, SDNode OpNode>
1937 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1938 (outs QPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1939 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1941 (TyQ (OpNode (TyD DPR:$src1),
1942 (TyD (NEONvduplane (TyD DPR_VFP2:$src2),imm:$lane)))))]>;
1943 class N3VLSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1944 InstrItinClass itin, string OpcodeStr, string Dt,
1945 ValueType TyQ, ValueType TyD, SDNode OpNode>
1946 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1947 (outs QPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1948 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1950 (TyQ (OpNode (TyD DPR:$src1),
1951 (TyD (NEONvduplane (TyD DPR_8:$src2), imm:$lane)))))]>;
1953 // Long 3-register operations with explicitly extended operands.
1954 class N3VLExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1955 InstrItinClass itin, string OpcodeStr, string Dt,
1956 ValueType TyQ, ValueType TyD, SDNode OpNode, SDNode ExtOp,
1958 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1959 (outs QPR:$Qd), (ins DPR:$Dn, DPR:$Dm), N3RegFrm, itin,
1960 OpcodeStr, Dt, "$Qd, $Dn, $Dm", "",
1961 [(set QPR:$Qd, (OpNode (TyQ (ExtOp (TyD DPR:$Dn))),
1962 (TyQ (ExtOp (TyD DPR:$Dm)))))]> {
1963 let isCommutable = Commutable;
1966 // Long 3-register intrinsics with explicit extend (VABDL).
1967 class N3VLIntExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1968 InstrItinClass itin, string OpcodeStr, string Dt,
1969 ValueType TyQ, ValueType TyD, Intrinsic IntOp, SDNode ExtOp,
1971 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1972 (outs QPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
1973 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1974 [(set QPR:$dst, (TyQ (ExtOp (TyD (IntOp (TyD DPR:$src1),
1975 (TyD DPR:$src2))))))]> {
1976 let isCommutable = Commutable;
1979 // Long 3-register intrinsics.
1980 class N3VLInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1981 InstrItinClass itin, string OpcodeStr, string Dt,
1982 ValueType TyQ, ValueType TyD, Intrinsic IntOp, bit Commutable>
1983 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1984 (outs QPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
1985 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1986 [(set QPR:$dst, (TyQ (IntOp (TyD DPR:$src1), (TyD DPR:$src2))))]> {
1987 let isCommutable = Commutable;
1989 class N3VLIntSL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1990 string OpcodeStr, string Dt,
1991 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1992 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1993 (outs QPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1994 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1995 [(set (ResTy QPR:$dst),
1996 (ResTy (IntOp (OpTy DPR:$src1),
1997 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1999 class N3VLIntSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2000 InstrItinClass itin, string OpcodeStr, string Dt,
2001 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2002 : N3V<op24, 1, op21_20, op11_8, 1, 0,
2003 (outs QPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
2004 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
2005 [(set (ResTy QPR:$dst),
2006 (ResTy (IntOp (OpTy DPR:$src1),
2007 (OpTy (NEONvduplane (OpTy DPR_8:$src2),
2010 // Wide 3-register operations.
2011 class N3VW<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2012 string OpcodeStr, string Dt, ValueType TyQ, ValueType TyD,
2013 SDNode OpNode, SDNode ExtOp, bit Commutable>
2014 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2015 (outs QPR:$Qd), (ins QPR:$Qn, DPR:$Dm), N3RegFrm, IIC_VSUBiD,
2016 OpcodeStr, Dt, "$Qd, $Qn, $Dm", "",
2017 [(set QPR:$Qd, (OpNode (TyQ QPR:$Qn),
2018 (TyQ (ExtOp (TyD DPR:$Dm)))))]> {
2019 let isCommutable = Commutable;
2022 // Pairwise long 2-register intrinsics, both double- and quad-register.
2023 class N2VDPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2024 bits<2> op17_16, bits<5> op11_7, bit op4,
2025 string OpcodeStr, string Dt,
2026 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2027 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
2028 (ins DPR:$src), IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
2029 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
2030 class N2VQPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2031 bits<2> op17_16, bits<5> op11_7, bit op4,
2032 string OpcodeStr, string Dt,
2033 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2034 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
2035 (ins QPR:$src), IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
2036 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
2038 // Pairwise long 2-register accumulate intrinsics,
2039 // both double- and quad-register.
2040 // The destination register is also used as the first source operand register.
2041 class N2VDPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2042 bits<2> op17_16, bits<5> op11_7, bit op4,
2043 string OpcodeStr, string Dt,
2044 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2045 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
2046 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vm), IIC_VPALiD,
2047 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
2048 [(set DPR:$Vd, (ResTy (IntOp (ResTy DPR:$src1), (OpTy DPR:$Vm))))]>;
2049 class N2VQPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2050 bits<2> op17_16, bits<5> op11_7, bit op4,
2051 string OpcodeStr, string Dt,
2052 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2053 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4,
2054 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vm), IIC_VPALiQ,
2055 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
2056 [(set QPR:$Vd, (ResTy (IntOp (ResTy QPR:$src1), (OpTy QPR:$Vm))))]>;
2058 // Shift by immediate,
2059 // both double- and quad-register.
2060 class N2VDSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2061 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2062 ValueType Ty, SDNode OpNode>
2063 : N2VImm<op24, op23, op11_8, op7, 0, op4,
2064 (outs DPR:$dst), (ins DPR:$src, i32imm:$SIMM), f, itin,
2065 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
2066 [(set DPR:$dst, (Ty (OpNode (Ty DPR:$src), (i32 imm:$SIMM))))]>;
2067 class N2VQSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2068 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2069 ValueType Ty, SDNode OpNode>
2070 : N2VImm<op24, op23, op11_8, op7, 1, op4,
2071 (outs QPR:$dst), (ins QPR:$src, i32imm:$SIMM), f, itin,
2072 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
2073 [(set QPR:$dst, (Ty (OpNode (Ty QPR:$src), (i32 imm:$SIMM))))]>;
2075 // Long shift by immediate.
2076 class N2VLSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
2077 string OpcodeStr, string Dt,
2078 ValueType ResTy, ValueType OpTy, SDNode OpNode>
2079 : N2VImm<op24, op23, op11_8, op7, op6, op4,
2080 (outs QPR:$dst), (ins DPR:$src, i32imm:$SIMM), N2RegVShLFrm,
2081 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src, $SIMM", "",
2082 [(set QPR:$dst, (ResTy (OpNode (OpTy DPR:$src),
2083 (i32 imm:$SIMM))))]>;
2085 // Narrow shift by immediate.
2086 class N2VNSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
2087 InstrItinClass itin, string OpcodeStr, string Dt,
2088 ValueType ResTy, ValueType OpTy, SDNode OpNode>
2089 : N2VImm<op24, op23, op11_8, op7, op6, op4,
2090 (outs DPR:$dst), (ins QPR:$src, i32imm:$SIMM), N2RegVShRFrm, itin,
2091 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
2092 [(set DPR:$dst, (ResTy (OpNode (OpTy QPR:$src),
2093 (i32 imm:$SIMM))))]>;
2095 // Shift right by immediate and accumulate,
2096 // both double- and quad-register.
2097 class N2VDShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2098 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
2099 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
2100 (ins DPR:$src1, DPR:$Vm, i32imm:$SIMM), N2RegVShRFrm, IIC_VPALiD,
2101 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2102 [(set DPR:$Vd, (Ty (add DPR:$src1,
2103 (Ty (ShOp DPR:$Vm, (i32 imm:$SIMM))))))]>;
2104 class N2VQShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2105 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
2106 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
2107 (ins QPR:$src1, QPR:$Vm, i32imm:$SIMM), N2RegVShRFrm, IIC_VPALiD,
2108 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2109 [(set QPR:$Vd, (Ty (add QPR:$src1,
2110 (Ty (ShOp QPR:$Vm, (i32 imm:$SIMM))))))]>;
2112 // Shift by immediate and insert,
2113 // both double- and quad-register.
2114 class N2VDShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2115 Format f, string OpcodeStr, string Dt, ValueType Ty,SDNode ShOp>
2116 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
2117 (ins DPR:$src1, DPR:$Vm, i32imm:$SIMM), f, IIC_VSHLiD,
2118 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2119 [(set DPR:$Vd, (Ty (ShOp DPR:$src1, DPR:$Vm, (i32 imm:$SIMM))))]>;
2120 class N2VQShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2121 Format f, string OpcodeStr, string Dt, ValueType Ty,SDNode ShOp>
2122 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
2123 (ins QPR:$src1, QPR:$Vm, i32imm:$SIMM), f, IIC_VSHLiQ,
2124 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2125 [(set QPR:$Vd, (Ty (ShOp QPR:$src1, QPR:$Vm, (i32 imm:$SIMM))))]>;
2127 // Convert, with fractional bits immediate,
2128 // both double- and quad-register.
2129 class N2VCvtD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2130 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2132 : N2VImm<op24, op23, op11_8, op7, 0, op4,
2133 (outs DPR:$Vd), (ins DPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
2134 IIC_VUNAD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2135 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (i32 imm:$SIMM))))]>;
2136 class N2VCvtQ<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2137 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2139 : N2VImm<op24, op23, op11_8, op7, 1, op4,
2140 (outs QPR:$Vd), (ins QPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
2141 IIC_VUNAQ, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2142 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (i32 imm:$SIMM))))]>;
2144 //===----------------------------------------------------------------------===//
2146 //===----------------------------------------------------------------------===//
2148 // Abbreviations used in multiclass suffixes:
2149 // Q = quarter int (8 bit) elements
2150 // H = half int (16 bit) elements
2151 // S = single int (32 bit) elements
2152 // D = double int (64 bit) elements
2154 // Neon 2-register vector operations -- for disassembly only.
2156 // First with only element sizes of 8, 16 and 32 bits:
2157 multiclass N2V_QHS_cmp<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2158 bits<5> op11_7, bit op4, string opc, string Dt,
2159 string asm, SDNode OpNode> {
2160 // 64-bit vector types.
2161 def v8i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 0, op4,
2162 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
2163 opc, !strconcat(Dt, "8"), asm, "",
2164 [(set DPR:$dst, (v8i8 (OpNode (v8i8 DPR:$src))))]>;
2165 def v4i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 0, op4,
2166 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
2167 opc, !strconcat(Dt, "16"), asm, "",
2168 [(set DPR:$dst, (v4i16 (OpNode (v4i16 DPR:$src))))]>;
2169 def v2i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
2170 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
2171 opc, !strconcat(Dt, "32"), asm, "",
2172 [(set DPR:$dst, (v2i32 (OpNode (v2i32 DPR:$src))))]>;
2173 def v2f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
2174 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
2175 opc, "f32", asm, "",
2176 [(set DPR:$dst, (v2f32 (OpNode (v2f32 DPR:$src))))]> {
2177 let Inst{10} = 1; // overwrite F = 1
2180 // 128-bit vector types.
2181 def v16i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 1, op4,
2182 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
2183 opc, !strconcat(Dt, "8"), asm, "",
2184 [(set QPR:$dst, (v16i8 (OpNode (v16i8 QPR:$src))))]>;
2185 def v8i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 1, op4,
2186 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
2187 opc, !strconcat(Dt, "16"), asm, "",
2188 [(set QPR:$dst, (v8i16 (OpNode (v8i16 QPR:$src))))]>;
2189 def v4i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
2190 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
2191 opc, !strconcat(Dt, "32"), asm, "",
2192 [(set QPR:$dst, (v4i32 (OpNode (v4i32 QPR:$src))))]>;
2193 def v4f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
2194 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
2195 opc, "f32", asm, "",
2196 [(set QPR:$dst, (v4f32 (OpNode (v4f32 QPR:$src))))]> {
2197 let Inst{10} = 1; // overwrite F = 1
2201 // Neon 3-register vector operations.
2203 // First with only element sizes of 8, 16 and 32 bits:
2204 multiclass N3V_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2205 InstrItinClass itinD16, InstrItinClass itinD32,
2206 InstrItinClass itinQ16, InstrItinClass itinQ32,
2207 string OpcodeStr, string Dt,
2208 SDNode OpNode, bit Commutable = 0> {
2209 // 64-bit vector types.
2210 def v8i8 : N3VD<op24, op23, 0b00, op11_8, op4, itinD16,
2211 OpcodeStr, !strconcat(Dt, "8"),
2212 v8i8, v8i8, OpNode, Commutable>;
2213 def v4i16 : N3VD<op24, op23, 0b01, op11_8, op4, itinD16,
2214 OpcodeStr, !strconcat(Dt, "16"),
2215 v4i16, v4i16, OpNode, Commutable>;
2216 def v2i32 : N3VD<op24, op23, 0b10, op11_8, op4, itinD32,
2217 OpcodeStr, !strconcat(Dt, "32"),
2218 v2i32, v2i32, OpNode, Commutable>;
2220 // 128-bit vector types.
2221 def v16i8 : N3VQ<op24, op23, 0b00, op11_8, op4, itinQ16,
2222 OpcodeStr, !strconcat(Dt, "8"),
2223 v16i8, v16i8, OpNode, Commutable>;
2224 def v8i16 : N3VQ<op24, op23, 0b01, op11_8, op4, itinQ16,
2225 OpcodeStr, !strconcat(Dt, "16"),
2226 v8i16, v8i16, OpNode, Commutable>;
2227 def v4i32 : N3VQ<op24, op23, 0b10, op11_8, op4, itinQ32,
2228 OpcodeStr, !strconcat(Dt, "32"),
2229 v4i32, v4i32, OpNode, Commutable>;
2232 multiclass N3VSL_HS<bits<4> op11_8, string OpcodeStr, string Dt, SDNode ShOp> {
2233 def v4i16 : N3VDSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
2235 def v2i32 : N3VDSL<0b10, op11_8, IIC_VMULi32D, OpcodeStr, !strconcat(Dt,"32"),
2237 def v8i16 : N3VQSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
2238 v8i16, v4i16, ShOp>;
2239 def v4i32 : N3VQSL<0b10, op11_8, IIC_VMULi32Q, OpcodeStr, !strconcat(Dt,"32"),
2240 v4i32, v2i32, ShOp>;
2243 // ....then also with element size 64 bits:
2244 multiclass N3V_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2245 InstrItinClass itinD, InstrItinClass itinQ,
2246 string OpcodeStr, string Dt,
2247 SDNode OpNode, bit Commutable = 0>
2248 : N3V_QHS<op24, op23, op11_8, op4, itinD, itinD, itinQ, itinQ,
2249 OpcodeStr, Dt, OpNode, Commutable> {
2250 def v1i64 : N3VD<op24, op23, 0b11, op11_8, op4, itinD,
2251 OpcodeStr, !strconcat(Dt, "64"),
2252 v1i64, v1i64, OpNode, Commutable>;
2253 def v2i64 : N3VQ<op24, op23, 0b11, op11_8, op4, itinQ,
2254 OpcodeStr, !strconcat(Dt, "64"),
2255 v2i64, v2i64, OpNode, Commutable>;
2259 // Neon Narrowing 2-register vector operations,
2260 // source operand element sizes of 16, 32 and 64 bits:
2261 multiclass N2VN_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2262 bits<5> op11_7, bit op6, bit op4,
2263 InstrItinClass itin, string OpcodeStr, string Dt,
2265 def v8i8 : N2VN<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
2266 itin, OpcodeStr, !strconcat(Dt, "16"),
2267 v8i8, v8i16, OpNode>;
2268 def v4i16 : N2VN<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
2269 itin, OpcodeStr, !strconcat(Dt, "32"),
2270 v4i16, v4i32, OpNode>;
2271 def v2i32 : N2VN<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
2272 itin, OpcodeStr, !strconcat(Dt, "64"),
2273 v2i32, v2i64, OpNode>;
2276 // Neon Narrowing 2-register vector intrinsics,
2277 // source operand element sizes of 16, 32 and 64 bits:
2278 multiclass N2VNInt_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2279 bits<5> op11_7, bit op6, bit op4,
2280 InstrItinClass itin, string OpcodeStr, string Dt,
2282 def v8i8 : N2VNInt<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
2283 itin, OpcodeStr, !strconcat(Dt, "16"),
2284 v8i8, v8i16, IntOp>;
2285 def v4i16 : N2VNInt<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
2286 itin, OpcodeStr, !strconcat(Dt, "32"),
2287 v4i16, v4i32, IntOp>;
2288 def v2i32 : N2VNInt<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
2289 itin, OpcodeStr, !strconcat(Dt, "64"),
2290 v2i32, v2i64, IntOp>;
2294 // Neon Lengthening 2-register vector intrinsic (currently specific to VMOVL).
2295 // source operand element sizes of 16, 32 and 64 bits:
2296 multiclass N2VL_QHS<bits<2> op24_23, bits<5> op11_7, bit op6, bit op4,
2297 string OpcodeStr, string Dt, SDNode OpNode> {
2298 def v8i16 : N2VL<op24_23, 0b00, 0b10, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2299 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode>;
2300 def v4i32 : N2VL<op24_23, 0b01, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2301 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
2302 def v2i64 : N2VL<op24_23, 0b10, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2303 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
2307 // Neon 3-register vector intrinsics.
2309 // First with only element sizes of 16 and 32 bits:
2310 multiclass N3VInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2311 InstrItinClass itinD16, InstrItinClass itinD32,
2312 InstrItinClass itinQ16, InstrItinClass itinQ32,
2313 string OpcodeStr, string Dt,
2314 Intrinsic IntOp, bit Commutable = 0> {
2315 // 64-bit vector types.
2316 def v4i16 : N3VDInt<op24, op23, 0b01, op11_8, op4, f, itinD16,
2317 OpcodeStr, !strconcat(Dt, "16"),
2318 v4i16, v4i16, IntOp, Commutable>;
2319 def v2i32 : N3VDInt<op24, op23, 0b10, op11_8, op4, f, itinD32,
2320 OpcodeStr, !strconcat(Dt, "32"),
2321 v2i32, v2i32, IntOp, Commutable>;
2323 // 128-bit vector types.
2324 def v8i16 : N3VQInt<op24, op23, 0b01, op11_8, op4, f, itinQ16,
2325 OpcodeStr, !strconcat(Dt, "16"),
2326 v8i16, v8i16, IntOp, Commutable>;
2327 def v4i32 : N3VQInt<op24, op23, 0b10, op11_8, op4, f, itinQ32,
2328 OpcodeStr, !strconcat(Dt, "32"),
2329 v4i32, v4i32, IntOp, Commutable>;
2331 multiclass N3VInt_HSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2332 InstrItinClass itinD16, InstrItinClass itinD32,
2333 InstrItinClass itinQ16, InstrItinClass itinQ32,
2334 string OpcodeStr, string Dt,
2336 // 64-bit vector types.
2337 def v4i16 : N3VDIntSh<op24, op23, 0b01, op11_8, op4, f, itinD16,
2338 OpcodeStr, !strconcat(Dt, "16"),
2339 v4i16, v4i16, IntOp>;
2340 def v2i32 : N3VDIntSh<op24, op23, 0b10, op11_8, op4, f, itinD32,
2341 OpcodeStr, !strconcat(Dt, "32"),
2342 v2i32, v2i32, IntOp>;
2344 // 128-bit vector types.
2345 def v8i16 : N3VQIntSh<op24, op23, 0b01, op11_8, op4, f, itinQ16,
2346 OpcodeStr, !strconcat(Dt, "16"),
2347 v8i16, v8i16, IntOp>;
2348 def v4i32 : N3VQIntSh<op24, op23, 0b10, op11_8, op4, f, itinQ32,
2349 OpcodeStr, !strconcat(Dt, "32"),
2350 v4i32, v4i32, IntOp>;
2353 multiclass N3VIntSL_HS<bits<4> op11_8,
2354 InstrItinClass itinD16, InstrItinClass itinD32,
2355 InstrItinClass itinQ16, InstrItinClass itinQ32,
2356 string OpcodeStr, string Dt, Intrinsic IntOp> {
2357 def v4i16 : N3VDIntSL16<0b01, op11_8, itinD16,
2358 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp>;
2359 def v2i32 : N3VDIntSL<0b10, op11_8, itinD32,
2360 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp>;
2361 def v8i16 : N3VQIntSL16<0b01, op11_8, itinQ16,
2362 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16, IntOp>;
2363 def v4i32 : N3VQIntSL<0b10, op11_8, itinQ32,
2364 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32, IntOp>;
2367 // ....then also with element size of 8 bits:
2368 multiclass N3VInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2369 InstrItinClass itinD16, InstrItinClass itinD32,
2370 InstrItinClass itinQ16, InstrItinClass itinQ32,
2371 string OpcodeStr, string Dt,
2372 Intrinsic IntOp, bit Commutable = 0>
2373 : N3VInt_HS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2374 OpcodeStr, Dt, IntOp, Commutable> {
2375 def v8i8 : N3VDInt<op24, op23, 0b00, op11_8, op4, f, itinD16,
2376 OpcodeStr, !strconcat(Dt, "8"),
2377 v8i8, v8i8, IntOp, Commutable>;
2378 def v16i8 : N3VQInt<op24, op23, 0b00, op11_8, op4, f, itinQ16,
2379 OpcodeStr, !strconcat(Dt, "8"),
2380 v16i8, v16i8, IntOp, Commutable>;
2382 multiclass N3VInt_QHSSh<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,
2387 : N3VInt_HSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2388 OpcodeStr, Dt, IntOp> {
2389 def v8i8 : N3VDIntSh<op24, op23, 0b00, op11_8, op4, f, itinD16,
2390 OpcodeStr, !strconcat(Dt, "8"),
2392 def v16i8 : N3VQIntSh<op24, op23, 0b00, op11_8, op4, f, itinQ16,
2393 OpcodeStr, !strconcat(Dt, "8"),
2394 v16i8, v16i8, IntOp>;
2398 // ....then also with element size of 64 bits:
2399 multiclass N3VInt_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2400 InstrItinClass itinD16, InstrItinClass itinD32,
2401 InstrItinClass itinQ16, InstrItinClass itinQ32,
2402 string OpcodeStr, string Dt,
2403 Intrinsic IntOp, bit Commutable = 0>
2404 : N3VInt_QHS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2405 OpcodeStr, Dt, IntOp, Commutable> {
2406 def v1i64 : N3VDInt<op24, op23, 0b11, op11_8, op4, f, itinD32,
2407 OpcodeStr, !strconcat(Dt, "64"),
2408 v1i64, v1i64, IntOp, Commutable>;
2409 def v2i64 : N3VQInt<op24, op23, 0b11, op11_8, op4, f, itinQ32,
2410 OpcodeStr, !strconcat(Dt, "64"),
2411 v2i64, v2i64, IntOp, Commutable>;
2413 multiclass N3VInt_QHSDSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2414 InstrItinClass itinD16, InstrItinClass itinD32,
2415 InstrItinClass itinQ16, InstrItinClass itinQ32,
2416 string OpcodeStr, string Dt,
2418 : N3VInt_QHSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2419 OpcodeStr, Dt, IntOp> {
2420 def v1i64 : N3VDIntSh<op24, op23, 0b11, op11_8, op4, f, itinD32,
2421 OpcodeStr, !strconcat(Dt, "64"),
2422 v1i64, v1i64, IntOp>;
2423 def v2i64 : N3VQIntSh<op24, op23, 0b11, op11_8, op4, f, itinQ32,
2424 OpcodeStr, !strconcat(Dt, "64"),
2425 v2i64, v2i64, IntOp>;
2428 // Neon Narrowing 3-register vector intrinsics,
2429 // source operand element sizes of 16, 32 and 64 bits:
2430 multiclass N3VNInt_HSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2431 string OpcodeStr, string Dt,
2432 Intrinsic IntOp, bit Commutable = 0> {
2433 def v8i8 : N3VNInt<op24, op23, 0b00, op11_8, op4,
2434 OpcodeStr, !strconcat(Dt, "16"),
2435 v8i8, v8i16, IntOp, Commutable>;
2436 def v4i16 : N3VNInt<op24, op23, 0b01, op11_8, op4,
2437 OpcodeStr, !strconcat(Dt, "32"),
2438 v4i16, v4i32, IntOp, Commutable>;
2439 def v2i32 : N3VNInt<op24, op23, 0b10, op11_8, op4,
2440 OpcodeStr, !strconcat(Dt, "64"),
2441 v2i32, v2i64, IntOp, Commutable>;
2445 // Neon Long 3-register vector operations.
2447 multiclass N3VL_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2448 InstrItinClass itin16, InstrItinClass itin32,
2449 string OpcodeStr, string Dt,
2450 SDNode OpNode, bit Commutable = 0> {
2451 def v8i16 : N3VL<op24, op23, 0b00, op11_8, op4, itin16,
2452 OpcodeStr, !strconcat(Dt, "8"),
2453 v8i16, v8i8, OpNode, Commutable>;
2454 def v4i32 : N3VL<op24, op23, 0b01, op11_8, op4, itin16,
2455 OpcodeStr, !strconcat(Dt, "16"),
2456 v4i32, v4i16, OpNode, Commutable>;
2457 def v2i64 : N3VL<op24, op23, 0b10, op11_8, op4, itin32,
2458 OpcodeStr, !strconcat(Dt, "32"),
2459 v2i64, v2i32, OpNode, Commutable>;
2462 multiclass N3VLSL_HS<bit op24, bits<4> op11_8,
2463 InstrItinClass itin, string OpcodeStr, string Dt,
2465 def v4i16 : N3VLSL16<op24, 0b01, op11_8, itin, OpcodeStr,
2466 !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
2467 def v2i32 : N3VLSL<op24, 0b10, op11_8, itin, OpcodeStr,
2468 !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
2471 multiclass N3VLExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2472 InstrItinClass itin16, InstrItinClass itin32,
2473 string OpcodeStr, string Dt,
2474 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
2475 def v8i16 : N3VLExt<op24, op23, 0b00, op11_8, op4, itin16,
2476 OpcodeStr, !strconcat(Dt, "8"),
2477 v8i16, v8i8, OpNode, ExtOp, Commutable>;
2478 def v4i32 : N3VLExt<op24, op23, 0b01, op11_8, op4, itin16,
2479 OpcodeStr, !strconcat(Dt, "16"),
2480 v4i32, v4i16, OpNode, ExtOp, Commutable>;
2481 def v2i64 : N3VLExt<op24, op23, 0b10, op11_8, op4, itin32,
2482 OpcodeStr, !strconcat(Dt, "32"),
2483 v2i64, v2i32, OpNode, ExtOp, Commutable>;
2486 // Neon Long 3-register vector intrinsics.
2488 // First with only element sizes of 16 and 32 bits:
2489 multiclass N3VLInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
2490 InstrItinClass itin16, InstrItinClass itin32,
2491 string OpcodeStr, string Dt,
2492 Intrinsic IntOp, bit Commutable = 0> {
2493 def v4i32 : N3VLInt<op24, op23, 0b01, op11_8, op4, itin16,
2494 OpcodeStr, !strconcat(Dt, "16"),
2495 v4i32, v4i16, IntOp, Commutable>;
2496 def v2i64 : N3VLInt<op24, op23, 0b10, op11_8, op4, itin32,
2497 OpcodeStr, !strconcat(Dt, "32"),
2498 v2i64, v2i32, IntOp, Commutable>;
2501 multiclass N3VLIntSL_HS<bit op24, bits<4> op11_8,
2502 InstrItinClass itin, string OpcodeStr, string Dt,
2504 def v4i16 : N3VLIntSL16<op24, 0b01, op11_8, itin,
2505 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
2506 def v2i32 : N3VLIntSL<op24, 0b10, op11_8, itin,
2507 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
2510 // ....then also with element size of 8 bits:
2511 multiclass N3VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2512 InstrItinClass itin16, InstrItinClass itin32,
2513 string OpcodeStr, string Dt,
2514 Intrinsic IntOp, bit Commutable = 0>
2515 : N3VLInt_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt,
2516 IntOp, Commutable> {
2517 def v8i16 : N3VLInt<op24, op23, 0b00, op11_8, op4, itin16,
2518 OpcodeStr, !strconcat(Dt, "8"),
2519 v8i16, v8i8, IntOp, Commutable>;
2522 // ....with explicit extend (VABDL).
2523 multiclass N3VLIntExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2524 InstrItinClass itin, string OpcodeStr, string Dt,
2525 Intrinsic IntOp, SDNode ExtOp, bit Commutable = 0> {
2526 def v8i16 : N3VLIntExt<op24, op23, 0b00, op11_8, op4, itin,
2527 OpcodeStr, !strconcat(Dt, "8"),
2528 v8i16, v8i8, IntOp, ExtOp, Commutable>;
2529 def v4i32 : N3VLIntExt<op24, op23, 0b01, op11_8, op4, itin,
2530 OpcodeStr, !strconcat(Dt, "16"),
2531 v4i32, v4i16, IntOp, ExtOp, Commutable>;
2532 def v2i64 : N3VLIntExt<op24, op23, 0b10, op11_8, op4, itin,
2533 OpcodeStr, !strconcat(Dt, "32"),
2534 v2i64, v2i32, IntOp, ExtOp, Commutable>;
2538 // Neon Wide 3-register vector intrinsics,
2539 // source operand element sizes of 8, 16 and 32 bits:
2540 multiclass N3VW_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2541 string OpcodeStr, string Dt,
2542 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
2543 def v8i16 : N3VW<op24, op23, 0b00, op11_8, op4,
2544 OpcodeStr, !strconcat(Dt, "8"),
2545 v8i16, v8i8, OpNode, ExtOp, Commutable>;
2546 def v4i32 : N3VW<op24, op23, 0b01, op11_8, op4,
2547 OpcodeStr, !strconcat(Dt, "16"),
2548 v4i32, v4i16, OpNode, ExtOp, Commutable>;
2549 def v2i64 : N3VW<op24, op23, 0b10, op11_8, op4,
2550 OpcodeStr, !strconcat(Dt, "32"),
2551 v2i64, v2i32, OpNode, ExtOp, Commutable>;
2555 // Neon Multiply-Op vector operations,
2556 // element sizes of 8, 16 and 32 bits:
2557 multiclass N3VMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2558 InstrItinClass itinD16, InstrItinClass itinD32,
2559 InstrItinClass itinQ16, InstrItinClass itinQ32,
2560 string OpcodeStr, string Dt, SDNode OpNode> {
2561 // 64-bit vector types.
2562 def v8i8 : N3VDMulOp<op24, op23, 0b00, op11_8, op4, itinD16,
2563 OpcodeStr, !strconcat(Dt, "8"), v8i8, mul, OpNode>;
2564 def v4i16 : N3VDMulOp<op24, op23, 0b01, op11_8, op4, itinD16,
2565 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, OpNode>;
2566 def v2i32 : N3VDMulOp<op24, op23, 0b10, op11_8, op4, itinD32,
2567 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, OpNode>;
2569 // 128-bit vector types.
2570 def v16i8 : N3VQMulOp<op24, op23, 0b00, op11_8, op4, itinQ16,
2571 OpcodeStr, !strconcat(Dt, "8"), v16i8, mul, OpNode>;
2572 def v8i16 : N3VQMulOp<op24, op23, 0b01, op11_8, op4, itinQ16,
2573 OpcodeStr, !strconcat(Dt, "16"), v8i16, mul, OpNode>;
2574 def v4i32 : N3VQMulOp<op24, op23, 0b10, op11_8, op4, itinQ32,
2575 OpcodeStr, !strconcat(Dt, "32"), v4i32, mul, OpNode>;
2578 multiclass N3VMulOpSL_HS<bits<4> op11_8,
2579 InstrItinClass itinD16, InstrItinClass itinD32,
2580 InstrItinClass itinQ16, InstrItinClass itinQ32,
2581 string OpcodeStr, string Dt, SDNode ShOp> {
2582 def v4i16 : N3VDMulOpSL16<0b01, op11_8, itinD16,
2583 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, ShOp>;
2584 def v2i32 : N3VDMulOpSL<0b10, op11_8, itinD32,
2585 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, ShOp>;
2586 def v8i16 : N3VQMulOpSL16<0b01, op11_8, itinQ16,
2587 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16,
2589 def v4i32 : N3VQMulOpSL<0b10, op11_8, itinQ32,
2590 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32,
2594 // Neon Intrinsic-Op vector operations,
2595 // element sizes of 8, 16 and 32 bits:
2596 multiclass N3VIntOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2597 InstrItinClass itinD, InstrItinClass itinQ,
2598 string OpcodeStr, string Dt, Intrinsic IntOp,
2600 // 64-bit vector types.
2601 def v8i8 : N3VDIntOp<op24, op23, 0b00, op11_8, op4, itinD,
2602 OpcodeStr, !strconcat(Dt, "8"), v8i8, IntOp, OpNode>;
2603 def v4i16 : N3VDIntOp<op24, op23, 0b01, op11_8, op4, itinD,
2604 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp, OpNode>;
2605 def v2i32 : N3VDIntOp<op24, op23, 0b10, op11_8, op4, itinD,
2606 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp, OpNode>;
2608 // 128-bit vector types.
2609 def v16i8 : N3VQIntOp<op24, op23, 0b00, op11_8, op4, itinQ,
2610 OpcodeStr, !strconcat(Dt, "8"), v16i8, IntOp, OpNode>;
2611 def v8i16 : N3VQIntOp<op24, op23, 0b01, op11_8, op4, itinQ,
2612 OpcodeStr, !strconcat(Dt, "16"), v8i16, IntOp, OpNode>;
2613 def v4i32 : N3VQIntOp<op24, op23, 0b10, op11_8, op4, itinQ,
2614 OpcodeStr, !strconcat(Dt, "32"), v4i32, IntOp, OpNode>;
2617 // Neon 3-argument intrinsics,
2618 // element sizes of 8, 16 and 32 bits:
2619 multiclass N3VInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2620 InstrItinClass itinD, InstrItinClass itinQ,
2621 string OpcodeStr, string Dt, Intrinsic IntOp> {
2622 // 64-bit vector types.
2623 def v8i8 : N3VDInt3<op24, op23, 0b00, op11_8, op4, itinD,
2624 OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
2625 def v4i16 : N3VDInt3<op24, op23, 0b01, op11_8, op4, itinD,
2626 OpcodeStr, !strconcat(Dt, "16"), v4i16, v4i16, IntOp>;
2627 def v2i32 : N3VDInt3<op24, op23, 0b10, op11_8, op4, itinD,
2628 OpcodeStr, !strconcat(Dt, "32"), v2i32, v2i32, IntOp>;
2630 // 128-bit vector types.
2631 def v16i8 : N3VQInt3<op24, op23, 0b00, op11_8, op4, itinQ,
2632 OpcodeStr, !strconcat(Dt, "8"), v16i8, v16i8, IntOp>;
2633 def v8i16 : N3VQInt3<op24, op23, 0b01, op11_8, op4, itinQ,
2634 OpcodeStr, !strconcat(Dt, "16"), v8i16, v8i16, IntOp>;
2635 def v4i32 : N3VQInt3<op24, op23, 0b10, op11_8, op4, itinQ,
2636 OpcodeStr, !strconcat(Dt, "32"), v4i32, v4i32, IntOp>;
2640 // Neon Long Multiply-Op vector operations,
2641 // element sizes of 8, 16 and 32 bits:
2642 multiclass N3VLMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2643 InstrItinClass itin16, InstrItinClass itin32,
2644 string OpcodeStr, string Dt, SDNode MulOp,
2646 def v8i16 : N3VLMulOp<op24, op23, 0b00, op11_8, op4, itin16, OpcodeStr,
2647 !strconcat(Dt, "8"), v8i16, v8i8, MulOp, OpNode>;
2648 def v4i32 : N3VLMulOp<op24, op23, 0b01, op11_8, op4, itin16, OpcodeStr,
2649 !strconcat(Dt, "16"), v4i32, v4i16, MulOp, OpNode>;
2650 def v2i64 : N3VLMulOp<op24, op23, 0b10, op11_8, op4, itin32, OpcodeStr,
2651 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
2654 multiclass N3VLMulOpSL_HS<bit op24, bits<4> op11_8, string OpcodeStr,
2655 string Dt, SDNode MulOp, SDNode OpNode> {
2656 def v4i16 : N3VLMulOpSL16<op24, 0b01, op11_8, IIC_VMACi16D, OpcodeStr,
2657 !strconcat(Dt,"16"), v4i32, v4i16, MulOp, OpNode>;
2658 def v2i32 : N3VLMulOpSL<op24, 0b10, op11_8, IIC_VMACi32D, OpcodeStr,
2659 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
2663 // Neon Long 3-argument intrinsics.
2665 // First with only element sizes of 16 and 32 bits:
2666 multiclass N3VLInt3_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
2667 InstrItinClass itin16, InstrItinClass itin32,
2668 string OpcodeStr, string Dt, Intrinsic IntOp> {
2669 def v4i32 : N3VLInt3<op24, op23, 0b01, op11_8, op4, itin16,
2670 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
2671 def v2i64 : N3VLInt3<op24, op23, 0b10, op11_8, op4, itin32,
2672 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
2675 multiclass N3VLInt3SL_HS<bit op24, bits<4> op11_8,
2676 string OpcodeStr, string Dt, Intrinsic IntOp> {
2677 def v4i16 : N3VLInt3SL16<op24, 0b01, op11_8, IIC_VMACi16D,
2678 OpcodeStr, !strconcat(Dt,"16"), v4i32, v4i16, IntOp>;
2679 def v2i32 : N3VLInt3SL<op24, 0b10, op11_8, IIC_VMACi32D,
2680 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
2683 // ....then also with element size of 8 bits:
2684 multiclass N3VLInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2685 InstrItinClass itin16, InstrItinClass itin32,
2686 string OpcodeStr, string Dt, Intrinsic IntOp>
2687 : N3VLInt3_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt, IntOp> {
2688 def v8i16 : N3VLInt3<op24, op23, 0b00, op11_8, op4, itin16,
2689 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
2692 // ....with explicit extend (VABAL).
2693 multiclass N3VLIntExtOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2694 InstrItinClass itin, string OpcodeStr, string Dt,
2695 Intrinsic IntOp, SDNode ExtOp, SDNode OpNode> {
2696 def v8i16 : N3VLIntExtOp<op24, op23, 0b00, op11_8, op4, itin,
2697 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8,
2698 IntOp, ExtOp, OpNode>;
2699 def v4i32 : N3VLIntExtOp<op24, op23, 0b01, op11_8, op4, itin,
2700 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16,
2701 IntOp, ExtOp, OpNode>;
2702 def v2i64 : N3VLIntExtOp<op24, op23, 0b10, op11_8, op4, itin,
2703 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32,
2704 IntOp, ExtOp, OpNode>;
2708 // Neon 2-register vector intrinsics,
2709 // element sizes of 8, 16 and 32 bits:
2710 multiclass N2VInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2711 bits<5> op11_7, bit op4,
2712 InstrItinClass itinD, InstrItinClass itinQ,
2713 string OpcodeStr, string Dt, Intrinsic IntOp> {
2714 // 64-bit vector types.
2715 def v8i8 : N2VDInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2716 itinD, OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
2717 def v4i16 : N2VDInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2718 itinD, OpcodeStr, !strconcat(Dt, "16"),v4i16,v4i16,IntOp>;
2719 def v2i32 : N2VDInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2720 itinD, OpcodeStr, !strconcat(Dt, "32"),v2i32,v2i32,IntOp>;
2722 // 128-bit vector types.
2723 def v16i8 : N2VQInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2724 itinQ, OpcodeStr, !strconcat(Dt, "8"), v16i8,v16i8,IntOp>;
2725 def v8i16 : N2VQInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2726 itinQ, OpcodeStr, !strconcat(Dt, "16"),v8i16,v8i16,IntOp>;
2727 def v4i32 : N2VQInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2728 itinQ, OpcodeStr, !strconcat(Dt, "32"),v4i32,v4i32,IntOp>;
2732 // Neon Pairwise long 2-register intrinsics,
2733 // element sizes of 8, 16 and 32 bits:
2734 multiclass N2VPLInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2735 bits<5> op11_7, bit op4,
2736 string OpcodeStr, string Dt, Intrinsic IntOp> {
2737 // 64-bit vector types.
2738 def v8i8 : N2VDPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2739 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
2740 def v4i16 : N2VDPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2741 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
2742 def v2i32 : N2VDPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2743 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
2745 // 128-bit vector types.
2746 def v16i8 : N2VQPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2747 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
2748 def v8i16 : N2VQPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2749 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
2750 def v4i32 : N2VQPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2751 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
2755 // Neon Pairwise long 2-register accumulate intrinsics,
2756 // element sizes of 8, 16 and 32 bits:
2757 multiclass N2VPLInt2_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2758 bits<5> op11_7, bit op4,
2759 string OpcodeStr, string Dt, Intrinsic IntOp> {
2760 // 64-bit vector types.
2761 def v8i8 : N2VDPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2762 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
2763 def v4i16 : N2VDPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2764 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
2765 def v2i32 : N2VDPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2766 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
2768 // 128-bit vector types.
2769 def v16i8 : N2VQPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2770 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
2771 def v8i16 : N2VQPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2772 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
2773 def v4i32 : N2VQPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2774 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
2778 // Neon 2-register vector shift by immediate,
2779 // with f of either N2RegVShLFrm or N2RegVShRFrm
2780 // element sizes of 8, 16, 32 and 64 bits:
2781 multiclass N2VSh_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2782 InstrItinClass itin, string OpcodeStr, string Dt,
2783 SDNode OpNode, Format f> {
2784 // 64-bit vector types.
2785 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
2786 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
2787 let Inst{21-19} = 0b001; // imm6 = 001xxx
2789 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
2790 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
2791 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2793 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
2794 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
2795 let Inst{21} = 0b1; // imm6 = 1xxxxx
2797 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, f, itin,
2798 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
2801 // 128-bit vector types.
2802 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
2803 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
2804 let Inst{21-19} = 0b001; // imm6 = 001xxx
2806 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
2807 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
2808 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2810 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
2811 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
2812 let Inst{21} = 0b1; // imm6 = 1xxxxx
2814 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, f, itin,
2815 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
2819 // Neon Shift-Accumulate vector operations,
2820 // element sizes of 8, 16, 32 and 64 bits:
2821 multiclass N2VShAdd_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2822 string OpcodeStr, string Dt, SDNode ShOp> {
2823 // 64-bit vector types.
2824 def v8i8 : N2VDShAdd<op24, op23, op11_8, 0, op4,
2825 OpcodeStr, !strconcat(Dt, "8"), v8i8, ShOp> {
2826 let Inst{21-19} = 0b001; // imm6 = 001xxx
2828 def v4i16 : N2VDShAdd<op24, op23, op11_8, 0, op4,
2829 OpcodeStr, !strconcat(Dt, "16"), v4i16, ShOp> {
2830 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2832 def v2i32 : N2VDShAdd<op24, op23, op11_8, 0, op4,
2833 OpcodeStr, !strconcat(Dt, "32"), v2i32, ShOp> {
2834 let Inst{21} = 0b1; // imm6 = 1xxxxx
2836 def v1i64 : N2VDShAdd<op24, op23, op11_8, 1, op4,
2837 OpcodeStr, !strconcat(Dt, "64"), v1i64, ShOp>;
2840 // 128-bit vector types.
2841 def v16i8 : N2VQShAdd<op24, op23, op11_8, 0, op4,
2842 OpcodeStr, !strconcat(Dt, "8"), v16i8, ShOp> {
2843 let Inst{21-19} = 0b001; // imm6 = 001xxx
2845 def v8i16 : N2VQShAdd<op24, op23, op11_8, 0, op4,
2846 OpcodeStr, !strconcat(Dt, "16"), v8i16, ShOp> {
2847 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2849 def v4i32 : N2VQShAdd<op24, op23, op11_8, 0, op4,
2850 OpcodeStr, !strconcat(Dt, "32"), v4i32, ShOp> {
2851 let Inst{21} = 0b1; // imm6 = 1xxxxx
2853 def v2i64 : N2VQShAdd<op24, op23, op11_8, 1, op4,
2854 OpcodeStr, !strconcat(Dt, "64"), v2i64, ShOp>;
2859 // Neon Shift-Insert vector operations,
2860 // with f of either N2RegVShLFrm or N2RegVShRFrm
2861 // element sizes of 8, 16, 32 and 64 bits:
2862 multiclass N2VShIns_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2863 string OpcodeStr, SDNode ShOp,
2865 // 64-bit vector types.
2866 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4,
2867 f, OpcodeStr, "8", v8i8, ShOp> {
2868 let Inst{21-19} = 0b001; // imm6 = 001xxx
2870 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4,
2871 f, OpcodeStr, "16", v4i16, ShOp> {
2872 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2874 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4,
2875 f, OpcodeStr, "32", v2i32, ShOp> {
2876 let Inst{21} = 0b1; // imm6 = 1xxxxx
2878 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4,
2879 f, OpcodeStr, "64", v1i64, ShOp>;
2882 // 128-bit vector types.
2883 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4,
2884 f, OpcodeStr, "8", v16i8, ShOp> {
2885 let Inst{21-19} = 0b001; // imm6 = 001xxx
2887 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4,
2888 f, OpcodeStr, "16", v8i16, ShOp> {
2889 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2891 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4,
2892 f, OpcodeStr, "32", v4i32, ShOp> {
2893 let Inst{21} = 0b1; // imm6 = 1xxxxx
2895 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4,
2896 f, OpcodeStr, "64", v2i64, ShOp>;
2900 // Neon Shift Long operations,
2901 // element sizes of 8, 16, 32 bits:
2902 multiclass N2VLSh_QHS<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
2903 bit op4, string OpcodeStr, string Dt, SDNode OpNode> {
2904 def v8i16 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
2905 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode> {
2906 let Inst{21-19} = 0b001; // imm6 = 001xxx
2908 def v4i32 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
2909 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode> {
2910 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2912 def v2i64 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
2913 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode> {
2914 let Inst{21} = 0b1; // imm6 = 1xxxxx
2918 // Neon Shift Narrow operations,
2919 // element sizes of 16, 32, 64 bits:
2920 multiclass N2VNSh_HSD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
2921 bit op4, InstrItinClass itin, string OpcodeStr, string Dt,
2923 def v8i8 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
2924 OpcodeStr, !strconcat(Dt, "16"), v8i8, v8i16, OpNode> {
2925 let Inst{21-19} = 0b001; // imm6 = 001xxx
2927 def v4i16 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
2928 OpcodeStr, !strconcat(Dt, "32"), v4i16, v4i32, OpNode> {
2929 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2931 def v2i32 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
2932 OpcodeStr, !strconcat(Dt, "64"), v2i32, v2i64, OpNode> {
2933 let Inst{21} = 0b1; // imm6 = 1xxxxx
2937 //===----------------------------------------------------------------------===//
2938 // Instruction Definitions.
2939 //===----------------------------------------------------------------------===//
2941 // Vector Add Operations.
2943 // VADD : Vector Add (integer and floating-point)
2944 defm VADD : N3V_QHSD<0, 0, 0b1000, 0, IIC_VBINiD, IIC_VBINiQ, "vadd", "i",
2946 def VADDfd : N3VD<0, 0, 0b00, 0b1101, 0, IIC_VBIND, "vadd", "f32",
2947 v2f32, v2f32, fadd, 1>;
2948 def VADDfq : N3VQ<0, 0, 0b00, 0b1101, 0, IIC_VBINQ, "vadd", "f32",
2949 v4f32, v4f32, fadd, 1>;
2950 // VADDL : Vector Add Long (Q = D + D)
2951 defm VADDLs : N3VLExt_QHS<0,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
2952 "vaddl", "s", add, sext, 1>;
2953 defm VADDLu : N3VLExt_QHS<1,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
2954 "vaddl", "u", add, zext, 1>;
2955 // VADDW : Vector Add Wide (Q = Q + D)
2956 defm VADDWs : N3VW_QHS<0,1,0b0001,0, "vaddw", "s", add, sext, 0>;
2957 defm VADDWu : N3VW_QHS<1,1,0b0001,0, "vaddw", "u", add, zext, 0>;
2958 // VHADD : Vector Halving Add
2959 defm VHADDs : N3VInt_QHS<0, 0, 0b0000, 0, N3RegFrm,
2960 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2961 "vhadd", "s", int_arm_neon_vhadds, 1>;
2962 defm VHADDu : N3VInt_QHS<1, 0, 0b0000, 0, N3RegFrm,
2963 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2964 "vhadd", "u", int_arm_neon_vhaddu, 1>;
2965 // VRHADD : Vector Rounding Halving Add
2966 defm VRHADDs : N3VInt_QHS<0, 0, 0b0001, 0, N3RegFrm,
2967 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2968 "vrhadd", "s", int_arm_neon_vrhadds, 1>;
2969 defm VRHADDu : N3VInt_QHS<1, 0, 0b0001, 0, N3RegFrm,
2970 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2971 "vrhadd", "u", int_arm_neon_vrhaddu, 1>;
2972 // VQADD : Vector Saturating Add
2973 defm VQADDs : N3VInt_QHSD<0, 0, 0b0000, 1, N3RegFrm,
2974 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2975 "vqadd", "s", int_arm_neon_vqadds, 1>;
2976 defm VQADDu : N3VInt_QHSD<1, 0, 0b0000, 1, N3RegFrm,
2977 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2978 "vqadd", "u", int_arm_neon_vqaddu, 1>;
2979 // VADDHN : Vector Add and Narrow Returning High Half (D = Q + Q)
2980 defm VADDHN : N3VNInt_HSD<0,1,0b0100,0, "vaddhn", "i",
2981 int_arm_neon_vaddhn, 1>;
2982 // VRADDHN : Vector Rounding Add and Narrow Returning High Half (D = Q + Q)
2983 defm VRADDHN : N3VNInt_HSD<1,1,0b0100,0, "vraddhn", "i",
2984 int_arm_neon_vraddhn, 1>;
2986 // Vector Multiply Operations.
2988 // VMUL : Vector Multiply (integer, polynomial and floating-point)
2989 defm VMUL : N3V_QHS<0, 0, 0b1001, 1, IIC_VMULi16D, IIC_VMULi32D,
2990 IIC_VMULi16Q, IIC_VMULi32Q, "vmul", "i", mul, 1>;
2991 def VMULpd : N3VDInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16D, "vmul",
2992 "p8", v8i8, v8i8, int_arm_neon_vmulp, 1>;
2993 def VMULpq : N3VQInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16Q, "vmul",
2994 "p8", v16i8, v16i8, int_arm_neon_vmulp, 1>;
2995 def VMULfd : N3VD<1, 0, 0b00, 0b1101, 1, IIC_VFMULD, "vmul", "f32",
2996 v2f32, v2f32, fmul, 1>;
2997 def VMULfq : N3VQ<1, 0, 0b00, 0b1101, 1, IIC_VFMULQ, "vmul", "f32",
2998 v4f32, v4f32, fmul, 1>;
2999 defm VMULsl : N3VSL_HS<0b1000, "vmul", "i", mul>;
3000 def VMULslfd : N3VDSL<0b10, 0b1001, IIC_VBIND, "vmul", "f32", v2f32, fmul>;
3001 def VMULslfq : N3VQSL<0b10, 0b1001, IIC_VBINQ, "vmul", "f32", v4f32,
3004 def : Pat<(v8i16 (mul (v8i16 QPR:$src1),
3005 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
3006 (v8i16 (VMULslv8i16 (v8i16 QPR:$src1),
3007 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3008 (DSubReg_i16_reg imm:$lane))),
3009 (SubReg_i16_lane imm:$lane)))>;
3010 def : Pat<(v4i32 (mul (v4i32 QPR:$src1),
3011 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
3012 (v4i32 (VMULslv4i32 (v4i32 QPR:$src1),
3013 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3014 (DSubReg_i32_reg imm:$lane))),
3015 (SubReg_i32_lane imm:$lane)))>;
3016 def : Pat<(v4f32 (fmul (v4f32 QPR:$src1),
3017 (v4f32 (NEONvduplane (v4f32 QPR:$src2), imm:$lane)))),
3018 (v4f32 (VMULslfq (v4f32 QPR:$src1),
3019 (v2f32 (EXTRACT_SUBREG QPR:$src2,
3020 (DSubReg_i32_reg imm:$lane))),
3021 (SubReg_i32_lane imm:$lane)))>;
3023 // VQDMULH : Vector Saturating Doubling Multiply Returning High Half
3024 defm VQDMULH : N3VInt_HS<0, 0, 0b1011, 0, N3RegFrm, IIC_VMULi16D, IIC_VMULi32D,
3025 IIC_VMULi16Q, IIC_VMULi32Q,
3026 "vqdmulh", "s", int_arm_neon_vqdmulh, 1>;
3027 defm VQDMULHsl: N3VIntSL_HS<0b1100, IIC_VMULi16D, IIC_VMULi32D,
3028 IIC_VMULi16Q, IIC_VMULi32Q,
3029 "vqdmulh", "s", int_arm_neon_vqdmulh>;
3030 def : Pat<(v8i16 (int_arm_neon_vqdmulh (v8i16 QPR:$src1),
3031 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
3033 (v8i16 (VQDMULHslv8i16 (v8i16 QPR:$src1),
3034 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3035 (DSubReg_i16_reg imm:$lane))),
3036 (SubReg_i16_lane imm:$lane)))>;
3037 def : Pat<(v4i32 (int_arm_neon_vqdmulh (v4i32 QPR:$src1),
3038 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
3040 (v4i32 (VQDMULHslv4i32 (v4i32 QPR:$src1),
3041 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3042 (DSubReg_i32_reg imm:$lane))),
3043 (SubReg_i32_lane imm:$lane)))>;
3045 // VQRDMULH : Vector Rounding Saturating Doubling Multiply Returning High Half
3046 defm VQRDMULH : N3VInt_HS<1, 0, 0b1011, 0, N3RegFrm,
3047 IIC_VMULi16D,IIC_VMULi32D,IIC_VMULi16Q,IIC_VMULi32Q,
3048 "vqrdmulh", "s", int_arm_neon_vqrdmulh, 1>;
3049 defm VQRDMULHsl : N3VIntSL_HS<0b1101, IIC_VMULi16D, IIC_VMULi32D,
3050 IIC_VMULi16Q, IIC_VMULi32Q,
3051 "vqrdmulh", "s", int_arm_neon_vqrdmulh>;
3052 def : Pat<(v8i16 (int_arm_neon_vqrdmulh (v8i16 QPR:$src1),
3053 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
3055 (v8i16 (VQRDMULHslv8i16 (v8i16 QPR:$src1),
3056 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3057 (DSubReg_i16_reg imm:$lane))),
3058 (SubReg_i16_lane imm:$lane)))>;
3059 def : Pat<(v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$src1),
3060 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
3062 (v4i32 (VQRDMULHslv4i32 (v4i32 QPR:$src1),
3063 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3064 (DSubReg_i32_reg imm:$lane))),
3065 (SubReg_i32_lane imm:$lane)))>;
3067 // VMULL : Vector Multiply Long (integer and polynomial) (Q = D * D)
3068 defm VMULLs : N3VL_QHS<0,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
3069 "vmull", "s", NEONvmulls, 1>;
3070 defm VMULLu : N3VL_QHS<1,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
3071 "vmull", "u", NEONvmullu, 1>;
3072 def VMULLp : N3VLInt<0, 1, 0b00, 0b1110, 0, IIC_VMULi16D, "vmull", "p8",
3073 v8i16, v8i8, int_arm_neon_vmullp, 1>;
3074 defm VMULLsls : N3VLSL_HS<0, 0b1010, IIC_VMULi16D, "vmull", "s", NEONvmulls>;
3075 defm VMULLslu : N3VLSL_HS<1, 0b1010, IIC_VMULi16D, "vmull", "u", NEONvmullu>;
3077 // VQDMULL : Vector Saturating Doubling Multiply Long (Q = D * D)
3078 defm VQDMULL : N3VLInt_HS<0,1,0b1101,0, IIC_VMULi16D, IIC_VMULi32D,
3079 "vqdmull", "s", int_arm_neon_vqdmull, 1>;
3080 defm VQDMULLsl: N3VLIntSL_HS<0, 0b1011, IIC_VMULi16D,
3081 "vqdmull", "s", int_arm_neon_vqdmull>;
3083 // Vector Multiply-Accumulate and Multiply-Subtract Operations.
3085 // VMLA : Vector Multiply Accumulate (integer and floating-point)
3086 defm VMLA : N3VMulOp_QHS<0, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3087 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
3088 def VMLAfd : N3VDMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACD, "vmla", "f32",
3090 def VMLAfq : N3VQMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACQ, "vmla", "f32",
3092 defm VMLAsl : N3VMulOpSL_HS<0b0000, IIC_VMACi16D, IIC_VMACi32D,
3093 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
3094 def VMLAslfd : N3VDMulOpSL<0b10, 0b0001, IIC_VMACD, "vmla", "f32",
3096 def VMLAslfq : N3VQMulOpSL<0b10, 0b0001, IIC_VMACQ, "vmla", "f32",
3097 v4f32, v2f32, fmul, fadd>;
3099 def : Pat<(v8i16 (add (v8i16 QPR:$src1),
3100 (mul (v8i16 QPR:$src2),
3101 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
3102 (v8i16 (VMLAslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
3103 (v4i16 (EXTRACT_SUBREG QPR:$src3,
3104 (DSubReg_i16_reg imm:$lane))),
3105 (SubReg_i16_lane imm:$lane)))>;
3107 def : Pat<(v4i32 (add (v4i32 QPR:$src1),
3108 (mul (v4i32 QPR:$src2),
3109 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
3110 (v4i32 (VMLAslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
3111 (v2i32 (EXTRACT_SUBREG QPR:$src3,
3112 (DSubReg_i32_reg imm:$lane))),
3113 (SubReg_i32_lane imm:$lane)))>;
3115 def : Pat<(v4f32 (fadd (v4f32 QPR:$src1),
3116 (fmul (v4f32 QPR:$src2),
3117 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
3118 (v4f32 (VMLAslfq (v4f32 QPR:$src1),
3120 (v2f32 (EXTRACT_SUBREG QPR:$src3,
3121 (DSubReg_i32_reg imm:$lane))),
3122 (SubReg_i32_lane imm:$lane)))>;
3124 // VMLAL : Vector Multiply Accumulate Long (Q += D * D)
3125 defm VMLALs : N3VLMulOp_QHS<0,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
3126 "vmlal", "s", NEONvmulls, add>;
3127 defm VMLALu : N3VLMulOp_QHS<1,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
3128 "vmlal", "u", NEONvmullu, add>;
3130 defm VMLALsls : N3VLMulOpSL_HS<0, 0b0010, "vmlal", "s", NEONvmulls, add>;
3131 defm VMLALslu : N3VLMulOpSL_HS<1, 0b0010, "vmlal", "u", NEONvmullu, add>;
3133 // VQDMLAL : Vector Saturating Doubling Multiply Accumulate Long (Q += D * D)
3134 defm VQDMLAL : N3VLInt3_HS<0, 1, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3135 "vqdmlal", "s", int_arm_neon_vqdmlal>;
3136 defm VQDMLALsl: N3VLInt3SL_HS<0, 0b0011, "vqdmlal", "s", int_arm_neon_vqdmlal>;
3138 // VMLS : Vector Multiply Subtract (integer and floating-point)
3139 defm VMLS : N3VMulOp_QHS<1, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3140 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
3141 def VMLSfd : N3VDMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACD, "vmls", "f32",
3143 def VMLSfq : N3VQMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACQ, "vmls", "f32",
3145 defm VMLSsl : N3VMulOpSL_HS<0b0100, IIC_VMACi16D, IIC_VMACi32D,
3146 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
3147 def VMLSslfd : N3VDMulOpSL<0b10, 0b0101, IIC_VMACD, "vmls", "f32",
3149 def VMLSslfq : N3VQMulOpSL<0b10, 0b0101, IIC_VMACQ, "vmls", "f32",
3150 v4f32, v2f32, fmul, fsub>;
3152 def : Pat<(v8i16 (sub (v8i16 QPR:$src1),
3153 (mul (v8i16 QPR:$src2),
3154 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
3155 (v8i16 (VMLSslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
3156 (v4i16 (EXTRACT_SUBREG QPR:$src3,
3157 (DSubReg_i16_reg imm:$lane))),
3158 (SubReg_i16_lane imm:$lane)))>;
3160 def : Pat<(v4i32 (sub (v4i32 QPR:$src1),
3161 (mul (v4i32 QPR:$src2),
3162 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
3163 (v4i32 (VMLSslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
3164 (v2i32 (EXTRACT_SUBREG QPR:$src3,
3165 (DSubReg_i32_reg imm:$lane))),
3166 (SubReg_i32_lane imm:$lane)))>;
3168 def : Pat<(v4f32 (fsub (v4f32 QPR:$src1),
3169 (fmul (v4f32 QPR:$src2),
3170 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
3171 (v4f32 (VMLSslfq (v4f32 QPR:$src1), (v4f32 QPR:$src2),
3172 (v2f32 (EXTRACT_SUBREG QPR:$src3,
3173 (DSubReg_i32_reg imm:$lane))),
3174 (SubReg_i32_lane imm:$lane)))>;
3176 // VMLSL : Vector Multiply Subtract Long (Q -= D * D)
3177 defm VMLSLs : N3VLMulOp_QHS<0,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
3178 "vmlsl", "s", NEONvmulls, sub>;
3179 defm VMLSLu : N3VLMulOp_QHS<1,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
3180 "vmlsl", "u", NEONvmullu, sub>;
3182 defm VMLSLsls : N3VLMulOpSL_HS<0, 0b0110, "vmlsl", "s", NEONvmulls, sub>;
3183 defm VMLSLslu : N3VLMulOpSL_HS<1, 0b0110, "vmlsl", "u", NEONvmullu, sub>;
3185 // VQDMLSL : Vector Saturating Doubling Multiply Subtract Long (Q -= D * D)
3186 defm VQDMLSL : N3VLInt3_HS<0, 1, 0b1011, 0, IIC_VMACi16D, IIC_VMACi32D,
3187 "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
3188 defm VQDMLSLsl: N3VLInt3SL_HS<0, 0b111, "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
3190 // Vector Subtract Operations.
3192 // VSUB : Vector Subtract (integer and floating-point)
3193 defm VSUB : N3V_QHSD<1, 0, 0b1000, 0, IIC_VSUBiD, IIC_VSUBiQ,
3194 "vsub", "i", sub, 0>;
3195 def VSUBfd : N3VD<0, 0, 0b10, 0b1101, 0, IIC_VBIND, "vsub", "f32",
3196 v2f32, v2f32, fsub, 0>;
3197 def VSUBfq : N3VQ<0, 0, 0b10, 0b1101, 0, IIC_VBINQ, "vsub", "f32",
3198 v4f32, v4f32, fsub, 0>;
3199 // VSUBL : Vector Subtract Long (Q = D - D)
3200 defm VSUBLs : N3VLExt_QHS<0,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
3201 "vsubl", "s", sub, sext, 0>;
3202 defm VSUBLu : N3VLExt_QHS<1,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
3203 "vsubl", "u", sub, zext, 0>;
3204 // VSUBW : Vector Subtract Wide (Q = Q - D)
3205 defm VSUBWs : N3VW_QHS<0,1,0b0011,0, "vsubw", "s", sub, sext, 0>;
3206 defm VSUBWu : N3VW_QHS<1,1,0b0011,0, "vsubw", "u", sub, zext, 0>;
3207 // VHSUB : Vector Halving Subtract
3208 defm VHSUBs : N3VInt_QHS<0, 0, 0b0010, 0, N3RegFrm,
3209 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3210 "vhsub", "s", int_arm_neon_vhsubs, 0>;
3211 defm VHSUBu : N3VInt_QHS<1, 0, 0b0010, 0, N3RegFrm,
3212 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3213 "vhsub", "u", int_arm_neon_vhsubu, 0>;
3214 // VQSUB : Vector Saturing Subtract
3215 defm VQSUBs : N3VInt_QHSD<0, 0, 0b0010, 1, N3RegFrm,
3216 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3217 "vqsub", "s", int_arm_neon_vqsubs, 0>;
3218 defm VQSUBu : N3VInt_QHSD<1, 0, 0b0010, 1, N3RegFrm,
3219 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3220 "vqsub", "u", int_arm_neon_vqsubu, 0>;
3221 // VSUBHN : Vector Subtract and Narrow Returning High Half (D = Q - Q)
3222 defm VSUBHN : N3VNInt_HSD<0,1,0b0110,0, "vsubhn", "i",
3223 int_arm_neon_vsubhn, 0>;
3224 // VRSUBHN : Vector Rounding Subtract and Narrow Returning High Half (D=Q-Q)
3225 defm VRSUBHN : N3VNInt_HSD<1,1,0b0110,0, "vrsubhn", "i",
3226 int_arm_neon_vrsubhn, 0>;
3228 // Vector Comparisons.
3230 // VCEQ : Vector Compare Equal
3231 defm VCEQ : N3V_QHS<1, 0, 0b1000, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3232 IIC_VSUBi4Q, "vceq", "i", NEONvceq, 1>;
3233 def VCEQfd : N3VD<0,0,0b00,0b1110,0, IIC_VBIND, "vceq", "f32", v2i32, v2f32,
3235 def VCEQfq : N3VQ<0,0,0b00,0b1110,0, IIC_VBINQ, "vceq", "f32", v4i32, v4f32,
3238 defm VCEQz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00010, 0, "vceq", "i",
3239 "$dst, $src, #0", NEONvceqz>;
3241 // VCGE : Vector Compare Greater Than or Equal
3242 defm VCGEs : N3V_QHS<0, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3243 IIC_VSUBi4Q, "vcge", "s", NEONvcge, 0>;
3244 defm VCGEu : N3V_QHS<1, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3245 IIC_VSUBi4Q, "vcge", "u", NEONvcgeu, 0>;
3246 def VCGEfd : N3VD<1,0,0b00,0b1110,0, IIC_VBIND, "vcge", "f32", v2i32, v2f32,
3248 def VCGEfq : N3VQ<1,0,0b00,0b1110,0, IIC_VBINQ, "vcge", "f32", v4i32, v4f32,
3251 defm VCGEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00001, 0, "vcge", "s",
3252 "$dst, $src, #0", NEONvcgez>;
3253 defm VCLEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00011, 0, "vcle", "s",
3254 "$dst, $src, #0", NEONvclez>;
3256 // VCGT : Vector Compare Greater Than
3257 defm VCGTs : N3V_QHS<0, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3258 IIC_VSUBi4Q, "vcgt", "s", NEONvcgt, 0>;
3259 defm VCGTu : N3V_QHS<1, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3260 IIC_VSUBi4Q, "vcgt", "u", NEONvcgtu, 0>;
3261 def VCGTfd : N3VD<1,0,0b10,0b1110,0, IIC_VBIND, "vcgt", "f32", v2i32, v2f32,
3263 def VCGTfq : N3VQ<1,0,0b10,0b1110,0, IIC_VBINQ, "vcgt", "f32", v4i32, v4f32,
3266 defm VCGTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00000, 0, "vcgt", "s",
3267 "$dst, $src, #0", NEONvcgtz>;
3268 defm VCLTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00100, 0, "vclt", "s",
3269 "$dst, $src, #0", NEONvcltz>;
3271 // VACGE : Vector Absolute Compare Greater Than or Equal (aka VCAGE)
3272 def VACGEd : N3VDInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacge",
3273 "f32", v2i32, v2f32, int_arm_neon_vacged, 0>;
3274 def VACGEq : N3VQInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacge",
3275 "f32", v4i32, v4f32, int_arm_neon_vacgeq, 0>;
3276 // VACGT : Vector Absolute Compare Greater Than (aka VCAGT)
3277 def VACGTd : N3VDInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacgt",
3278 "f32", v2i32, v2f32, int_arm_neon_vacgtd, 0>;
3279 def VACGTq : N3VQInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacgt",
3280 "f32", v4i32, v4f32, int_arm_neon_vacgtq, 0>;
3281 // VTST : Vector Test Bits
3282 defm VTST : N3V_QHS<0, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
3283 IIC_VBINi4Q, "vtst", "", NEONvtst, 1>;
3285 // Vector Bitwise Operations.
3287 def vnotd : PatFrag<(ops node:$in),
3288 (xor node:$in, (bitconvert (v8i8 NEONimmAllOnesV)))>;
3289 def vnotq : PatFrag<(ops node:$in),
3290 (xor node:$in, (bitconvert (v16i8 NEONimmAllOnesV)))>;
3293 // VAND : Vector Bitwise AND
3294 def VANDd : N3VDX<0, 0, 0b00, 0b0001, 1, IIC_VBINiD, "vand",
3295 v2i32, v2i32, and, 1>;
3296 def VANDq : N3VQX<0, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "vand",
3297 v4i32, v4i32, and, 1>;
3299 // VEOR : Vector Bitwise Exclusive OR
3300 def VEORd : N3VDX<1, 0, 0b00, 0b0001, 1, IIC_VBINiD, "veor",
3301 v2i32, v2i32, xor, 1>;
3302 def VEORq : N3VQX<1, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "veor",
3303 v4i32, v4i32, xor, 1>;
3305 // VORR : Vector Bitwise OR
3306 def VORRd : N3VDX<0, 0, 0b10, 0b0001, 1, IIC_VBINiD, "vorr",
3307 v2i32, v2i32, or, 1>;
3308 def VORRq : N3VQX<0, 0, 0b10, 0b0001, 1, IIC_VBINiQ, "vorr",
3309 v4i32, v4i32, or, 1>;
3311 def VORRiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 0, 1,
3312 (outs DPR:$Vd), (ins nModImm:$SIMM, DPR:$src),
3314 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
3316 (v4i16 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
3317 let Inst{9} = SIMM{9};
3320 def VORRiv2i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 0, 0, 1,
3321 (outs DPR:$Vd), (ins nModImm:$SIMM, DPR:$src),
3323 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
3325 (v2i32 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
3326 let Inst{10-9} = SIMM{10-9};
3329 def VORRiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 0, 1,
3330 (outs QPR:$Vd), (ins nModImm:$SIMM, QPR:$src),
3332 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
3334 (v8i16 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
3335 let Inst{9} = SIMM{9};
3338 def VORRiv4i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 1, 0, 1,
3339 (outs QPR:$Vd), (ins nModImm:$SIMM, QPR:$src),
3341 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
3343 (v4i32 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
3344 let Inst{10-9} = SIMM{10-9};
3348 // VBIC : Vector Bitwise Bit Clear (AND NOT)
3349 def VBICd : N3VX<0, 0, 0b01, 0b0001, 0, 1, (outs DPR:$dst),
3350 (ins DPR:$src1, DPR:$src2), N3RegFrm, IIC_VBINiD,
3351 "vbic", "$dst, $src1, $src2", "",
3352 [(set DPR:$dst, (v2i32 (and DPR:$src1,
3353 (vnotd DPR:$src2))))]>;
3354 def VBICq : N3VX<0, 0, 0b01, 0b0001, 1, 1, (outs QPR:$dst),
3355 (ins QPR:$src1, QPR:$src2), N3RegFrm, IIC_VBINiQ,
3356 "vbic", "$dst, $src1, $src2", "",
3357 [(set QPR:$dst, (v4i32 (and QPR:$src1,
3358 (vnotq QPR:$src2))))]>;
3360 def VBICiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 1, 1,
3361 (outs DPR:$Vd), (ins nModImm:$SIMM, DPR:$src),
3363 "vbic", "i16", "$Vd, $SIMM", "$src = $Vd",
3365 (v4i16 (NEONvbicImm DPR:$src, timm:$SIMM)))]> {
3366 let Inst{9} = SIMM{9};
3369 def VBICiv2i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 0, 1, 1,
3370 (outs DPR:$Vd), (ins nModImm:$SIMM, DPR:$src),
3372 "vbic", "i32", "$Vd, $SIMM", "$src = $Vd",
3374 (v2i32 (NEONvbicImm DPR:$src, timm:$SIMM)))]> {
3375 let Inst{10-9} = SIMM{10-9};
3378 def VBICiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 1, 1,
3379 (outs QPR:$Vd), (ins nModImm:$SIMM, QPR:$src),
3381 "vbic", "i16", "$Vd, $SIMM", "$src = $Vd",
3383 (v8i16 (NEONvbicImm QPR:$src, timm:$SIMM)))]> {
3384 let Inst{9} = SIMM{9};
3387 def VBICiv4i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 1, 1, 1,
3388 (outs QPR:$Vd), (ins nModImm:$SIMM, QPR:$src),
3390 "vbic", "i32", "$Vd, $SIMM", "$src = $Vd",
3392 (v4i32 (NEONvbicImm QPR:$src, timm:$SIMM)))]> {
3393 let Inst{10-9} = SIMM{10-9};
3396 // VORN : Vector Bitwise OR NOT
3397 def VORNd : N3VX<0, 0, 0b11, 0b0001, 0, 1, (outs DPR:$dst),
3398 (ins DPR:$src1, DPR:$src2), N3RegFrm, IIC_VBINiD,
3399 "vorn", "$dst, $src1, $src2", "",
3400 [(set DPR:$dst, (v2i32 (or DPR:$src1,
3401 (vnotd DPR:$src2))))]>;
3402 def VORNq : N3VX<0, 0, 0b11, 0b0001, 1, 1, (outs QPR:$dst),
3403 (ins QPR:$src1, QPR:$src2), N3RegFrm, IIC_VBINiQ,
3404 "vorn", "$dst, $src1, $src2", "",
3405 [(set QPR:$dst, (v4i32 (or QPR:$src1,
3406 (vnotq QPR:$src2))))]>;
3408 // VMVN : Vector Bitwise NOT (Immediate)
3410 let isReMaterializable = 1 in {
3412 def VMVNv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 1, 1, (outs DPR:$dst),
3413 (ins nModImm:$SIMM), IIC_VMOVImm,
3414 "vmvn", "i16", "$dst, $SIMM", "",
3415 [(set DPR:$dst, (v4i16 (NEONvmvnImm timm:$SIMM)))]> {
3416 let Inst{9} = SIMM{9};
3419 def VMVNv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 1, 1, (outs QPR:$dst),
3420 (ins nModImm:$SIMM), IIC_VMOVImm,
3421 "vmvn", "i16", "$dst, $SIMM", "",
3422 [(set QPR:$dst, (v8i16 (NEONvmvnImm timm:$SIMM)))]> {
3423 let Inst{9} = SIMM{9};
3426 def VMVNv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 1, 1, (outs DPR:$dst),
3427 (ins nModImm:$SIMM), IIC_VMOVImm,
3428 "vmvn", "i32", "$dst, $SIMM", "",
3429 [(set DPR:$dst, (v2i32 (NEONvmvnImm timm:$SIMM)))]> {
3430 let Inst{11-8} = SIMM{11-8};
3433 def VMVNv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 1, 1, (outs QPR:$dst),
3434 (ins nModImm:$SIMM), IIC_VMOVImm,
3435 "vmvn", "i32", "$dst, $SIMM", "",
3436 [(set QPR:$dst, (v4i32 (NEONvmvnImm timm:$SIMM)))]> {
3437 let Inst{11-8} = SIMM{11-8};
3441 // VMVN : Vector Bitwise NOT
3442 def VMVNd : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 0, 0,
3443 (outs DPR:$dst), (ins DPR:$src), IIC_VSUBiD,
3444 "vmvn", "$dst, $src", "",
3445 [(set DPR:$dst, (v2i32 (vnotd DPR:$src)))]>;
3446 def VMVNq : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 1, 0,
3447 (outs QPR:$dst), (ins QPR:$src), IIC_VSUBiD,
3448 "vmvn", "$dst, $src", "",
3449 [(set QPR:$dst, (v4i32 (vnotq QPR:$src)))]>;
3450 def : Pat<(v2i32 (vnotd DPR:$src)), (VMVNd DPR:$src)>;
3451 def : Pat<(v4i32 (vnotq QPR:$src)), (VMVNq QPR:$src)>;
3453 // VBSL : Vector Bitwise Select
3454 def VBSLd : N3VX<1, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
3455 (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
3456 N3RegFrm, IIC_VCNTiD,
3457 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3459 (v2i32 (or (and DPR:$Vn, DPR:$src1),
3460 (and DPR:$Vm, (vnotd DPR:$src1)))))]>;
3461 def VBSLq : N3VX<1, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
3462 (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
3463 N3RegFrm, IIC_VCNTiQ,
3464 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3466 (v4i32 (or (and QPR:$Vn, QPR:$src1),
3467 (and QPR:$Vm, (vnotq QPR:$src1)))))]>;
3469 // VBIF : Vector Bitwise Insert if False
3470 // like VBSL but with: "vbif $dst, $src3, $src1", "$src2 = $dst",
3471 // FIXME: This instruction's encoding MAY NOT BE correct.
3472 def VBIFd : N3VX<1, 0, 0b11, 0b0001, 0, 1,
3473 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
3474 N3RegFrm, IIC_VBINiD,
3475 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3476 [/* For disassembly only; pattern left blank */]>;
3477 def VBIFq : N3VX<1, 0, 0b11, 0b0001, 1, 1,
3478 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
3479 N3RegFrm, IIC_VBINiQ,
3480 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3481 [/* For disassembly only; pattern left blank */]>;
3483 // VBIT : Vector Bitwise Insert if True
3484 // like VBSL but with: "vbit $dst, $src2, $src1", "$src3 = $dst",
3485 // FIXME: This instruction's encoding MAY NOT BE correct.
3486 def VBITd : N3VX<1, 0, 0b10, 0b0001, 0, 1,
3487 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
3488 N3RegFrm, IIC_VBINiD,
3489 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3490 [/* For disassembly only; pattern left blank */]>;
3491 def VBITq : N3VX<1, 0, 0b10, 0b0001, 1, 1,
3492 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
3493 N3RegFrm, IIC_VBINiQ,
3494 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3495 [/* For disassembly only; pattern left blank */]>;
3497 // VBIT/VBIF are not yet implemented. The TwoAddress pass will not go looking
3498 // for equivalent operations with different register constraints; it just
3501 // Vector Absolute Differences.
3503 // VABD : Vector Absolute Difference
3504 defm VABDs : N3VInt_QHS<0, 0, 0b0111, 0, N3RegFrm,
3505 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3506 "vabd", "s", int_arm_neon_vabds, 1>;
3507 defm VABDu : N3VInt_QHS<1, 0, 0b0111, 0, N3RegFrm,
3508 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3509 "vabd", "u", int_arm_neon_vabdu, 1>;
3510 def VABDfd : N3VDInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBIND,
3511 "vabd", "f32", v2f32, v2f32, int_arm_neon_vabds, 1>;
3512 def VABDfq : N3VQInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBINQ,
3513 "vabd", "f32", v4f32, v4f32, int_arm_neon_vabds, 1>;
3515 // VABDL : Vector Absolute Difference Long (Q = | D - D |)
3516 defm VABDLs : N3VLIntExt_QHS<0,1,0b0111,0, IIC_VSUBi4Q,
3517 "vabdl", "s", int_arm_neon_vabds, zext, 1>;
3518 defm VABDLu : N3VLIntExt_QHS<1,1,0b0111,0, IIC_VSUBi4Q,
3519 "vabdl", "u", int_arm_neon_vabdu, zext, 1>;
3521 // VABA : Vector Absolute Difference and Accumulate
3522 defm VABAs : N3VIntOp_QHS<0,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
3523 "vaba", "s", int_arm_neon_vabds, add>;
3524 defm VABAu : N3VIntOp_QHS<1,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
3525 "vaba", "u", int_arm_neon_vabdu, add>;
3527 // VABAL : Vector Absolute Difference and Accumulate Long (Q += | D - D |)
3528 defm VABALs : N3VLIntExtOp_QHS<0,1,0b0101,0, IIC_VABAD,
3529 "vabal", "s", int_arm_neon_vabds, zext, add>;
3530 defm VABALu : N3VLIntExtOp_QHS<1,1,0b0101,0, IIC_VABAD,
3531 "vabal", "u", int_arm_neon_vabdu, zext, add>;
3533 // Vector Maximum and Minimum.
3535 // VMAX : Vector Maximum
3536 defm VMAXs : N3VInt_QHS<0, 0, 0b0110, 0, N3RegFrm,
3537 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3538 "vmax", "s", int_arm_neon_vmaxs, 1>;
3539 defm VMAXu : N3VInt_QHS<1, 0, 0b0110, 0, N3RegFrm,
3540 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3541 "vmax", "u", int_arm_neon_vmaxu, 1>;
3542 def VMAXfd : N3VDInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBIND,
3544 v2f32, v2f32, int_arm_neon_vmaxs, 1>;
3545 def VMAXfq : N3VQInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBINQ,
3547 v4f32, v4f32, int_arm_neon_vmaxs, 1>;
3549 // VMIN : Vector Minimum
3550 defm VMINs : N3VInt_QHS<0, 0, 0b0110, 1, N3RegFrm,
3551 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3552 "vmin", "s", int_arm_neon_vmins, 1>;
3553 defm VMINu : N3VInt_QHS<1, 0, 0b0110, 1, N3RegFrm,
3554 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3555 "vmin", "u", int_arm_neon_vminu, 1>;
3556 def VMINfd : N3VDInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBIND,
3558 v2f32, v2f32, int_arm_neon_vmins, 1>;
3559 def VMINfq : N3VQInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBINQ,
3561 v4f32, v4f32, int_arm_neon_vmins, 1>;
3563 // Vector Pairwise Operations.
3565 // VPADD : Vector Pairwise Add
3566 def VPADDi8 : N3VDInt<0, 0, 0b00, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
3568 v8i8, v8i8, int_arm_neon_vpadd, 0>;
3569 def VPADDi16 : N3VDInt<0, 0, 0b01, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
3571 v4i16, v4i16, int_arm_neon_vpadd, 0>;
3572 def VPADDi32 : N3VDInt<0, 0, 0b10, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
3574 v2i32, v2i32, int_arm_neon_vpadd, 0>;
3575 def VPADDf : N3VDInt<1, 0, 0b00, 0b1101, 0, N3RegFrm,
3576 IIC_VPBIND, "vpadd", "f32",
3577 v2f32, v2f32, int_arm_neon_vpadd, 0>;
3579 // VPADDL : Vector Pairwise Add Long
3580 defm VPADDLs : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00100, 0, "vpaddl", "s",
3581 int_arm_neon_vpaddls>;
3582 defm VPADDLu : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00101, 0, "vpaddl", "u",
3583 int_arm_neon_vpaddlu>;
3585 // VPADAL : Vector Pairwise Add and Accumulate Long
3586 defm VPADALs : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01100, 0, "vpadal", "s",
3587 int_arm_neon_vpadals>;
3588 defm VPADALu : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01101, 0, "vpadal", "u",
3589 int_arm_neon_vpadalu>;
3591 // VPMAX : Vector Pairwise Maximum
3592 def VPMAXs8 : N3VDInt<0, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3593 "s8", v8i8, v8i8, int_arm_neon_vpmaxs, 0>;
3594 def VPMAXs16 : N3VDInt<0, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3595 "s16", v4i16, v4i16, int_arm_neon_vpmaxs, 0>;
3596 def VPMAXs32 : N3VDInt<0, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3597 "s32", v2i32, v2i32, int_arm_neon_vpmaxs, 0>;
3598 def VPMAXu8 : N3VDInt<1, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3599 "u8", v8i8, v8i8, int_arm_neon_vpmaxu, 0>;
3600 def VPMAXu16 : N3VDInt<1, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3601 "u16", v4i16, v4i16, int_arm_neon_vpmaxu, 0>;
3602 def VPMAXu32 : N3VDInt<1, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3603 "u32", v2i32, v2i32, int_arm_neon_vpmaxu, 0>;
3604 def VPMAXf : N3VDInt<1, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmax",
3605 "f32", v2f32, v2f32, int_arm_neon_vpmaxs, 0>;
3607 // VPMIN : Vector Pairwise Minimum
3608 def VPMINs8 : N3VDInt<0, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3609 "s8", v8i8, v8i8, int_arm_neon_vpmins, 0>;
3610 def VPMINs16 : N3VDInt<0, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3611 "s16", v4i16, v4i16, int_arm_neon_vpmins, 0>;
3612 def VPMINs32 : N3VDInt<0, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3613 "s32", v2i32, v2i32, int_arm_neon_vpmins, 0>;
3614 def VPMINu8 : N3VDInt<1, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3615 "u8", v8i8, v8i8, int_arm_neon_vpminu, 0>;
3616 def VPMINu16 : N3VDInt<1, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3617 "u16", v4i16, v4i16, int_arm_neon_vpminu, 0>;
3618 def VPMINu32 : N3VDInt<1, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3619 "u32", v2i32, v2i32, int_arm_neon_vpminu, 0>;
3620 def VPMINf : N3VDInt<1, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmin",
3621 "f32", v2f32, v2f32, int_arm_neon_vpmins, 0>;
3623 // Vector Reciprocal and Reciprocal Square Root Estimate and Step.
3625 // VRECPE : Vector Reciprocal Estimate
3626 def VRECPEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
3627 IIC_VUNAD, "vrecpe", "u32",
3628 v2i32, v2i32, int_arm_neon_vrecpe>;
3629 def VRECPEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
3630 IIC_VUNAQ, "vrecpe", "u32",
3631 v4i32, v4i32, int_arm_neon_vrecpe>;
3632 def VRECPEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
3633 IIC_VUNAD, "vrecpe", "f32",
3634 v2f32, v2f32, int_arm_neon_vrecpe>;
3635 def VRECPEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
3636 IIC_VUNAQ, "vrecpe", "f32",
3637 v4f32, v4f32, int_arm_neon_vrecpe>;
3639 // VRECPS : Vector Reciprocal Step
3640 def VRECPSfd : N3VDInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
3641 IIC_VRECSD, "vrecps", "f32",
3642 v2f32, v2f32, int_arm_neon_vrecps, 1>;
3643 def VRECPSfq : N3VQInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
3644 IIC_VRECSQ, "vrecps", "f32",
3645 v4f32, v4f32, int_arm_neon_vrecps, 1>;
3647 // VRSQRTE : Vector Reciprocal Square Root Estimate
3648 def VRSQRTEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
3649 IIC_VUNAD, "vrsqrte", "u32",
3650 v2i32, v2i32, int_arm_neon_vrsqrte>;
3651 def VRSQRTEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
3652 IIC_VUNAQ, "vrsqrte", "u32",
3653 v4i32, v4i32, int_arm_neon_vrsqrte>;
3654 def VRSQRTEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
3655 IIC_VUNAD, "vrsqrte", "f32",
3656 v2f32, v2f32, int_arm_neon_vrsqrte>;
3657 def VRSQRTEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
3658 IIC_VUNAQ, "vrsqrte", "f32",
3659 v4f32, v4f32, int_arm_neon_vrsqrte>;
3661 // VRSQRTS : Vector Reciprocal Square Root Step
3662 def VRSQRTSfd : N3VDInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
3663 IIC_VRECSD, "vrsqrts", "f32",
3664 v2f32, v2f32, int_arm_neon_vrsqrts, 1>;
3665 def VRSQRTSfq : N3VQInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
3666 IIC_VRECSQ, "vrsqrts", "f32",
3667 v4f32, v4f32, int_arm_neon_vrsqrts, 1>;
3671 // VSHL : Vector Shift
3672 defm VSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 0, N3RegVShFrm,
3673 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
3674 "vshl", "s", int_arm_neon_vshifts>;
3675 defm VSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 0, N3RegVShFrm,
3676 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
3677 "vshl", "u", int_arm_neon_vshiftu>;
3678 // VSHL : Vector Shift Left (Immediate)
3679 defm VSHLi : N2VSh_QHSD<0, 1, 0b0101, 1, IIC_VSHLiD, "vshl", "i", NEONvshl,
3681 // VSHR : Vector Shift Right (Immediate)
3682 defm VSHRs : N2VSh_QHSD<0, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "s", NEONvshrs,
3684 defm VSHRu : N2VSh_QHSD<1, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "u", NEONvshru,
3687 // VSHLL : Vector Shift Left Long
3688 defm VSHLLs : N2VLSh_QHS<0, 1, 0b1010, 0, 0, 1, "vshll", "s", NEONvshlls>;
3689 defm VSHLLu : N2VLSh_QHS<1, 1, 0b1010, 0, 0, 1, "vshll", "u", NEONvshllu>;
3691 // VSHLL : Vector Shift Left Long (with maximum shift count)
3692 class N2VLShMax<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
3693 bit op6, bit op4, string OpcodeStr, string Dt, ValueType ResTy,
3694 ValueType OpTy, SDNode OpNode>
3695 : N2VLSh<op24, op23, op11_8, op7, op6, op4, OpcodeStr, Dt,
3696 ResTy, OpTy, OpNode> {
3697 let Inst{21-16} = op21_16;
3699 def VSHLLi8 : N2VLShMax<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll", "i8",
3700 v8i16, v8i8, NEONvshlli>;
3701 def VSHLLi16 : N2VLShMax<1, 1, 0b110110, 0b0011, 0, 0, 0, "vshll", "i16",
3702 v4i32, v4i16, NEONvshlli>;
3703 def VSHLLi32 : N2VLShMax<1, 1, 0b111010, 0b0011, 0, 0, 0, "vshll", "i32",
3704 v2i64, v2i32, NEONvshlli>;
3706 // VSHRN : Vector Shift Right and Narrow
3707 defm VSHRN : N2VNSh_HSD<0,1,0b1000,0,0,1, IIC_VSHLiD, "vshrn", "i",
3710 // VRSHL : Vector Rounding Shift
3711 defm VRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 0, N3RegVShFrm,
3712 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3713 "vrshl", "s", int_arm_neon_vrshifts>;
3714 defm VRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 0, N3RegVShFrm,
3715 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3716 "vrshl", "u", int_arm_neon_vrshiftu>;
3717 // VRSHR : Vector Rounding Shift Right
3718 defm VRSHRs : N2VSh_QHSD<0,1,0b0010,1, IIC_VSHLi4D, "vrshr", "s", NEONvrshrs,
3720 defm VRSHRu : N2VSh_QHSD<1,1,0b0010,1, IIC_VSHLi4D, "vrshr", "u", NEONvrshru,
3723 // VRSHRN : Vector Rounding Shift Right and Narrow
3724 defm VRSHRN : N2VNSh_HSD<0, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vrshrn", "i",
3727 // VQSHL : Vector Saturating Shift
3728 defm VQSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 1, N3RegVShFrm,
3729 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3730 "vqshl", "s", int_arm_neon_vqshifts>;
3731 defm VQSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 1, N3RegVShFrm,
3732 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3733 "vqshl", "u", int_arm_neon_vqshiftu>;
3734 // VQSHL : Vector Saturating Shift Left (Immediate)
3735 defm VQSHLsi : N2VSh_QHSD<0,1,0b0111,1, IIC_VSHLi4D, "vqshl", "s",NEONvqshls,
3737 defm VQSHLui : N2VSh_QHSD<1,1,0b0111,1, IIC_VSHLi4D, "vqshl", "u",NEONvqshlu,
3739 // VQSHLU : Vector Saturating Shift Left (Immediate, Unsigned)
3740 defm VQSHLsu : N2VSh_QHSD<1,1,0b0110,1, IIC_VSHLi4D,"vqshlu","s",NEONvqshlsu,
3743 // VQSHRN : Vector Saturating Shift Right and Narrow
3744 defm VQSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "s",
3746 defm VQSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "u",
3749 // VQSHRUN : Vector Saturating Shift Right and Narrow (Unsigned)
3750 defm VQSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 0, 1, IIC_VSHLi4D, "vqshrun", "s",
3753 // VQRSHL : Vector Saturating Rounding Shift
3754 defm VQRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 1, N3RegVShFrm,
3755 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3756 "vqrshl", "s", int_arm_neon_vqrshifts>;
3757 defm VQRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 1, N3RegVShFrm,
3758 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3759 "vqrshl", "u", int_arm_neon_vqrshiftu>;
3761 // VQRSHRN : Vector Saturating Rounding Shift Right and Narrow
3762 defm VQRSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "s",
3764 defm VQRSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "u",
3767 // VQRSHRUN : Vector Saturating Rounding Shift Right and Narrow (Unsigned)
3768 defm VQRSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vqrshrun", "s",
3771 // VSRA : Vector Shift Right and Accumulate
3772 defm VSRAs : N2VShAdd_QHSD<0, 1, 0b0001, 1, "vsra", "s", NEONvshrs>;
3773 defm VSRAu : N2VShAdd_QHSD<1, 1, 0b0001, 1, "vsra", "u", NEONvshru>;
3774 // VRSRA : Vector Rounding Shift Right and Accumulate
3775 defm VRSRAs : N2VShAdd_QHSD<0, 1, 0b0011, 1, "vrsra", "s", NEONvrshrs>;
3776 defm VRSRAu : N2VShAdd_QHSD<1, 1, 0b0011, 1, "vrsra", "u", NEONvrshru>;
3778 // VSLI : Vector Shift Left and Insert
3779 defm VSLI : N2VShIns_QHSD<1, 1, 0b0101, 1, "vsli", NEONvsli, N2RegVShLFrm>;
3780 // VSRI : Vector Shift Right and Insert
3781 defm VSRI : N2VShIns_QHSD<1, 1, 0b0100, 1, "vsri", NEONvsri, N2RegVShRFrm>;
3783 // Vector Absolute and Saturating Absolute.
3785 // VABS : Vector Absolute Value
3786 defm VABS : N2VInt_QHS<0b11, 0b11, 0b01, 0b00110, 0,
3787 IIC_VUNAiD, IIC_VUNAiQ, "vabs", "s",
3789 def VABSfd : N2VDInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
3790 IIC_VUNAD, "vabs", "f32",
3791 v2f32, v2f32, int_arm_neon_vabs>;
3792 def VABSfq : N2VQInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
3793 IIC_VUNAQ, "vabs", "f32",
3794 v4f32, v4f32, int_arm_neon_vabs>;
3796 // VQABS : Vector Saturating Absolute Value
3797 defm VQABS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01110, 0,
3798 IIC_VQUNAiD, IIC_VQUNAiQ, "vqabs", "s",
3799 int_arm_neon_vqabs>;
3803 def vnegd : PatFrag<(ops node:$in),
3804 (sub (bitconvert (v2i32 NEONimmAllZerosV)), node:$in)>;
3805 def vnegq : PatFrag<(ops node:$in),
3806 (sub (bitconvert (v4i32 NEONimmAllZerosV)), node:$in)>;
3808 class VNEGD<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
3809 : N2V<0b11, 0b11, size, 0b01, 0b00111, 0, 0, (outs DPR:$dst), (ins DPR:$src),
3810 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
3811 [(set DPR:$dst, (Ty (vnegd DPR:$src)))]>;
3812 class VNEGQ<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
3813 : N2V<0b11, 0b11, size, 0b01, 0b00111, 1, 0, (outs QPR:$dst), (ins QPR:$src),
3814 IIC_VSHLiQ, OpcodeStr, Dt, "$dst, $src", "",
3815 [(set QPR:$dst, (Ty (vnegq QPR:$src)))]>;
3817 // VNEG : Vector Negate (integer)
3818 def VNEGs8d : VNEGD<0b00, "vneg", "s8", v8i8>;
3819 def VNEGs16d : VNEGD<0b01, "vneg", "s16", v4i16>;
3820 def VNEGs32d : VNEGD<0b10, "vneg", "s32", v2i32>;
3821 def VNEGs8q : VNEGQ<0b00, "vneg", "s8", v16i8>;
3822 def VNEGs16q : VNEGQ<0b01, "vneg", "s16", v8i16>;
3823 def VNEGs32q : VNEGQ<0b10, "vneg", "s32", v4i32>;
3825 // VNEG : Vector Negate (floating-point)
3826 def VNEGfd : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
3827 (outs DPR:$dst), (ins DPR:$src), IIC_VUNAD,
3828 "vneg", "f32", "$dst, $src", "",
3829 [(set DPR:$dst, (v2f32 (fneg DPR:$src)))]>;
3830 def VNEGf32q : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 1, 0,
3831 (outs QPR:$dst), (ins QPR:$src), IIC_VUNAQ,
3832 "vneg", "f32", "$dst, $src", "",
3833 [(set QPR:$dst, (v4f32 (fneg QPR:$src)))]>;
3835 def : Pat<(v8i8 (vnegd DPR:$src)), (VNEGs8d DPR:$src)>;
3836 def : Pat<(v4i16 (vnegd DPR:$src)), (VNEGs16d DPR:$src)>;
3837 def : Pat<(v2i32 (vnegd DPR:$src)), (VNEGs32d DPR:$src)>;
3838 def : Pat<(v16i8 (vnegq QPR:$src)), (VNEGs8q QPR:$src)>;
3839 def : Pat<(v8i16 (vnegq QPR:$src)), (VNEGs16q QPR:$src)>;
3840 def : Pat<(v4i32 (vnegq QPR:$src)), (VNEGs32q QPR:$src)>;
3842 // VQNEG : Vector Saturating Negate
3843 defm VQNEG : N2VInt_QHS<0b11, 0b11, 0b00, 0b01111, 0,
3844 IIC_VQUNAiD, IIC_VQUNAiQ, "vqneg", "s",
3845 int_arm_neon_vqneg>;
3847 // Vector Bit Counting Operations.
3849 // VCLS : Vector Count Leading Sign Bits
3850 defm VCLS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01000, 0,
3851 IIC_VCNTiD, IIC_VCNTiQ, "vcls", "s",
3853 // VCLZ : Vector Count Leading Zeros
3854 defm VCLZ : N2VInt_QHS<0b11, 0b11, 0b00, 0b01001, 0,
3855 IIC_VCNTiD, IIC_VCNTiQ, "vclz", "i",
3857 // VCNT : Vector Count One Bits
3858 def VCNTd : N2VDInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
3859 IIC_VCNTiD, "vcnt", "8",
3860 v8i8, v8i8, int_arm_neon_vcnt>;
3861 def VCNTq : N2VQInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
3862 IIC_VCNTiQ, "vcnt", "8",
3863 v16i8, v16i8, int_arm_neon_vcnt>;
3865 // Vector Swap -- for disassembly only.
3866 def VSWPd : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 0, 0,
3867 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
3868 "vswp", "$dst, $src", "", []>;
3869 def VSWPq : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 1, 0,
3870 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
3871 "vswp", "$dst, $src", "", []>;
3873 // Vector Move Operations.
3875 // VMOV : Vector Move (Register)
3877 let neverHasSideEffects = 1 in {
3878 def VMOVDneon: N3VX<0, 0, 0b10, 0b0001, 0, 1, (outs DPR:$dst), (ins DPR:$src),
3879 N3RegFrm, IIC_VMOV, "vmov", "$dst, $src", "", []>;
3880 def VMOVQ : N3VX<0, 0, 0b10, 0b0001, 1, 1, (outs QPR:$dst), (ins QPR:$src),
3881 N3RegFrm, IIC_VMOV, "vmov", "$dst, $src", "", []>;
3883 // Pseudo vector move instructions for QQ and QQQQ registers. This should
3884 // be expanded after register allocation is completed.
3885 def VMOVQQ : PseudoInst<(outs QQPR:$dst), (ins QQPR:$src),
3886 NoItinerary, "", []>;
3888 def VMOVQQQQ : PseudoInst<(outs QQQQPR:$dst), (ins QQQQPR:$src),
3889 NoItinerary, "", []>;
3890 } // neverHasSideEffects
3892 // VMOV : Vector Move (Immediate)
3894 let isReMaterializable = 1 in {
3895 def VMOVv8i8 : N1ModImm<1, 0b000, 0b1110, 0, 0, 0, 1, (outs DPR:$dst),
3896 (ins nModImm:$SIMM), IIC_VMOVImm,
3897 "vmov", "i8", "$dst, $SIMM", "",
3898 [(set DPR:$dst, (v8i8 (NEONvmovImm timm:$SIMM)))]>;
3899 def VMOVv16i8 : N1ModImm<1, 0b000, 0b1110, 0, 1, 0, 1, (outs QPR:$dst),
3900 (ins nModImm:$SIMM), IIC_VMOVImm,
3901 "vmov", "i8", "$dst, $SIMM", "",
3902 [(set QPR:$dst, (v16i8 (NEONvmovImm timm:$SIMM)))]>;
3904 def VMOVv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 0, 1, (outs DPR:$dst),
3905 (ins nModImm:$SIMM), IIC_VMOVImm,
3906 "vmov", "i16", "$dst, $SIMM", "",
3907 [(set DPR:$dst, (v4i16 (NEONvmovImm timm:$SIMM)))]> {
3908 let Inst{9} = SIMM{9};
3911 def VMOVv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 0, 1, (outs QPR:$dst),
3912 (ins nModImm:$SIMM), IIC_VMOVImm,
3913 "vmov", "i16", "$dst, $SIMM", "",
3914 [(set QPR:$dst, (v8i16 (NEONvmovImm timm:$SIMM)))]> {
3915 let Inst{9} = SIMM{9};
3918 def VMOVv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 0, 1, (outs DPR:$dst),
3919 (ins nModImm:$SIMM), IIC_VMOVImm,
3920 "vmov", "i32", "$dst, $SIMM", "",
3921 [(set DPR:$dst, (v2i32 (NEONvmovImm timm:$SIMM)))]> {
3922 let Inst{11-8} = SIMM{11-8};
3925 def VMOVv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 0, 1, (outs QPR:$dst),
3926 (ins nModImm:$SIMM), IIC_VMOVImm,
3927 "vmov", "i32", "$dst, $SIMM", "",
3928 [(set QPR:$dst, (v4i32 (NEONvmovImm timm:$SIMM)))]> {
3929 let Inst{11-8} = SIMM{11-8};
3932 def VMOVv1i64 : N1ModImm<1, 0b000, 0b1110, 0, 0, 1, 1, (outs DPR:$dst),
3933 (ins nModImm:$SIMM), IIC_VMOVImm,
3934 "vmov", "i64", "$dst, $SIMM", "",
3935 [(set DPR:$dst, (v1i64 (NEONvmovImm timm:$SIMM)))]>;
3936 def VMOVv2i64 : N1ModImm<1, 0b000, 0b1110, 0, 1, 1, 1, (outs QPR:$dst),
3937 (ins nModImm:$SIMM), IIC_VMOVImm,
3938 "vmov", "i64", "$dst, $SIMM", "",
3939 [(set QPR:$dst, (v2i64 (NEONvmovImm timm:$SIMM)))]>;
3940 } // isReMaterializable
3942 // VMOV : Vector Get Lane (move scalar to ARM core register)
3944 def VGETLNs8 : NVGetLane<{1,1,1,0,0,1,?,1}, 0b1011, {?,?},
3945 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
3946 IIC_VMOVSI, "vmov", "s8", "$R, $V[$lane]",
3947 [(set GPR:$R, (NEONvgetlanes (v8i8 DPR:$V),
3949 let Inst{21} = lane{2};
3950 let Inst{6-5} = lane{1-0};
3952 def VGETLNs16 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, {?,1},
3953 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
3954 IIC_VMOVSI, "vmov", "s16", "$R, $V[$lane]",
3955 [(set GPR:$R, (NEONvgetlanes (v4i16 DPR:$V),
3957 let Inst{21} = lane{1};
3958 let Inst{6} = lane{0};
3960 def VGETLNu8 : NVGetLane<{1,1,1,0,1,1,?,1}, 0b1011, {?,?},
3961 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
3962 IIC_VMOVSI, "vmov", "u8", "$R, $V[$lane]",
3963 [(set GPR:$R, (NEONvgetlaneu (v8i8 DPR:$V),
3965 let Inst{21} = lane{2};
3966 let Inst{6-5} = lane{1-0};
3968 def VGETLNu16 : NVGetLane<{1,1,1,0,1,0,?,1}, 0b1011, {?,1},
3969 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
3970 IIC_VMOVSI, "vmov", "u16", "$R, $V[$lane]",
3971 [(set GPR:$R, (NEONvgetlaneu (v4i16 DPR:$V),
3973 let Inst{21} = lane{1};
3974 let Inst{6} = lane{0};
3976 def VGETLNi32 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, 0b00,
3977 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
3978 IIC_VMOVSI, "vmov", "32", "$R, $V[$lane]",
3979 [(set GPR:$R, (extractelt (v2i32 DPR:$V),
3981 let Inst{21} = lane{0};
3983 // def VGETLNf32: see FMRDH and FMRDL in ARMInstrVFP.td
3984 def : Pat<(NEONvgetlanes (v16i8 QPR:$src), imm:$lane),
3985 (VGETLNs8 (v8i8 (EXTRACT_SUBREG QPR:$src,
3986 (DSubReg_i8_reg imm:$lane))),
3987 (SubReg_i8_lane imm:$lane))>;
3988 def : Pat<(NEONvgetlanes (v8i16 QPR:$src), imm:$lane),
3989 (VGETLNs16 (v4i16 (EXTRACT_SUBREG QPR:$src,
3990 (DSubReg_i16_reg imm:$lane))),
3991 (SubReg_i16_lane imm:$lane))>;
3992 def : Pat<(NEONvgetlaneu (v16i8 QPR:$src), imm:$lane),
3993 (VGETLNu8 (v8i8 (EXTRACT_SUBREG QPR:$src,
3994 (DSubReg_i8_reg imm:$lane))),
3995 (SubReg_i8_lane imm:$lane))>;
3996 def : Pat<(NEONvgetlaneu (v8i16 QPR:$src), imm:$lane),
3997 (VGETLNu16 (v4i16 (EXTRACT_SUBREG QPR:$src,
3998 (DSubReg_i16_reg imm:$lane))),
3999 (SubReg_i16_lane imm:$lane))>;
4000 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
4001 (VGETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src,
4002 (DSubReg_i32_reg imm:$lane))),
4003 (SubReg_i32_lane imm:$lane))>;
4004 def : Pat<(extractelt (v2f32 DPR:$src1), imm:$src2),
4005 (EXTRACT_SUBREG (v2f32 (COPY_TO_REGCLASS (v2f32 DPR:$src1),DPR_VFP2)),
4006 (SSubReg_f32_reg imm:$src2))>;
4007 def : Pat<(extractelt (v4f32 QPR:$src1), imm:$src2),
4008 (EXTRACT_SUBREG (v4f32 (COPY_TO_REGCLASS (v4f32 QPR:$src1),QPR_VFP2)),
4009 (SSubReg_f32_reg imm:$src2))>;
4010 //def : Pat<(extractelt (v2i64 QPR:$src1), imm:$src2),
4011 // (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
4012 def : Pat<(extractelt (v2f64 QPR:$src1), imm:$src2),
4013 (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
4016 // VMOV : Vector Set Lane (move ARM core register to scalar)
4018 let Constraints = "$src1 = $V" in {
4019 def VSETLNi8 : NVSetLane<{1,1,1,0,0,1,?,0}, 0b1011, {?,?}, (outs DPR:$V),
4020 (ins DPR:$src1, GPR:$R, nohash_imm:$lane),
4021 IIC_VMOVISL, "vmov", "8", "$V[$lane], $R",
4022 [(set DPR:$V, (vector_insert (v8i8 DPR:$src1),
4023 GPR:$R, imm:$lane))]> {
4024 let Inst{21} = lane{2};
4025 let Inst{6-5} = lane{1-0};
4027 def VSETLNi16 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, {?,1}, (outs DPR:$V),
4028 (ins DPR:$src1, GPR:$R, nohash_imm:$lane),
4029 IIC_VMOVISL, "vmov", "16", "$V[$lane], $R",
4030 [(set DPR:$V, (vector_insert (v4i16 DPR:$src1),
4031 GPR:$R, imm:$lane))]> {
4032 let Inst{21} = lane{1};
4033 let Inst{6} = lane{0};
4035 def VSETLNi32 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, 0b00, (outs DPR:$V),
4036 (ins DPR:$src1, GPR:$R, nohash_imm:$lane),
4037 IIC_VMOVISL, "vmov", "32", "$V[$lane], $R",
4038 [(set DPR:$V, (insertelt (v2i32 DPR:$src1),
4039 GPR:$R, imm:$lane))]> {
4040 let Inst{21} = lane{0};
4043 def : Pat<(vector_insert (v16i8 QPR:$src1), GPR:$src2, imm:$lane),
4044 (v16i8 (INSERT_SUBREG QPR:$src1,
4045 (v8i8 (VSETLNi8 (v8i8 (EXTRACT_SUBREG QPR:$src1,
4046 (DSubReg_i8_reg imm:$lane))),
4047 GPR:$src2, (SubReg_i8_lane imm:$lane))),
4048 (DSubReg_i8_reg imm:$lane)))>;
4049 def : Pat<(vector_insert (v8i16 QPR:$src1), GPR:$src2, imm:$lane),
4050 (v8i16 (INSERT_SUBREG QPR:$src1,
4051 (v4i16 (VSETLNi16 (v4i16 (EXTRACT_SUBREG QPR:$src1,
4052 (DSubReg_i16_reg imm:$lane))),
4053 GPR:$src2, (SubReg_i16_lane imm:$lane))),
4054 (DSubReg_i16_reg imm:$lane)))>;
4055 def : Pat<(insertelt (v4i32 QPR:$src1), GPR:$src2, imm:$lane),
4056 (v4i32 (INSERT_SUBREG QPR:$src1,
4057 (v2i32 (VSETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src1,
4058 (DSubReg_i32_reg imm:$lane))),
4059 GPR:$src2, (SubReg_i32_lane imm:$lane))),
4060 (DSubReg_i32_reg imm:$lane)))>;
4062 def : Pat<(v2f32 (insertelt DPR:$src1, SPR:$src2, imm:$src3)),
4063 (INSERT_SUBREG (v2f32 (COPY_TO_REGCLASS DPR:$src1, DPR_VFP2)),
4064 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
4065 def : Pat<(v4f32 (insertelt QPR:$src1, SPR:$src2, imm:$src3)),
4066 (INSERT_SUBREG (v4f32 (COPY_TO_REGCLASS QPR:$src1, QPR_VFP2)),
4067 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
4069 //def : Pat<(v2i64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
4070 // (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
4071 def : Pat<(v2f64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
4072 (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
4074 def : Pat<(v2f32 (scalar_to_vector SPR:$src)),
4075 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
4076 def : Pat<(v2f64 (scalar_to_vector (f64 DPR:$src))),
4077 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
4078 def : Pat<(v4f32 (scalar_to_vector SPR:$src)),
4079 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
4081 def : Pat<(v8i8 (scalar_to_vector GPR:$src)),
4082 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4083 def : Pat<(v4i16 (scalar_to_vector GPR:$src)),
4084 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4085 def : Pat<(v2i32 (scalar_to_vector GPR:$src)),
4086 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4088 def : Pat<(v16i8 (scalar_to_vector GPR:$src)),
4089 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4090 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4092 def : Pat<(v8i16 (scalar_to_vector GPR:$src)),
4093 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
4094 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4096 def : Pat<(v4i32 (scalar_to_vector GPR:$src)),
4097 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
4098 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4101 // VDUP : Vector Duplicate (from ARM core register to all elements)
4103 class VDUPD<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
4104 : NVDup<opcod1, 0b1011, opcod3, (outs DPR:$dst), (ins GPR:$src),
4105 IIC_VMOVIS, "vdup", Dt, "$dst, $src",
4106 [(set DPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
4107 class VDUPQ<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
4108 : NVDup<opcod1, 0b1011, opcod3, (outs QPR:$dst), (ins GPR:$src),
4109 IIC_VMOVIS, "vdup", Dt, "$dst, $src",
4110 [(set QPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
4112 def VDUP8d : VDUPD<0b11101100, 0b00, "8", v8i8>;
4113 def VDUP16d : VDUPD<0b11101000, 0b01, "16", v4i16>;
4114 def VDUP32d : VDUPD<0b11101000, 0b00, "32", v2i32>;
4115 def VDUP8q : VDUPQ<0b11101110, 0b00, "8", v16i8>;
4116 def VDUP16q : VDUPQ<0b11101010, 0b01, "16", v8i16>;
4117 def VDUP32q : VDUPQ<0b11101010, 0b00, "32", v4i32>;
4119 def VDUPfd : NVDup<0b11101000, 0b1011, 0b00, (outs DPR:$dst), (ins GPR:$src),
4120 IIC_VMOVIS, "vdup", "32", "$dst, $src",
4121 [(set DPR:$dst, (v2f32 (NEONvdup
4122 (f32 (bitconvert GPR:$src)))))]>;
4123 def VDUPfq : NVDup<0b11101010, 0b1011, 0b00, (outs QPR:$dst), (ins GPR:$src),
4124 IIC_VMOVIS, "vdup", "32", "$dst, $src",
4125 [(set QPR:$dst, (v4f32 (NEONvdup
4126 (f32 (bitconvert GPR:$src)))))]>;
4128 // VDUP : Vector Duplicate Lane (from scalar to all elements)
4130 class VDUPLND<bits<4> op19_16, string OpcodeStr, string Dt,
4132 : NVDupLane<op19_16, 0, (outs DPR:$dst), (ins DPR:$src, nohash_imm:$lane),
4133 IIC_VMOVD, OpcodeStr, Dt, "$dst, $src[$lane]",
4134 [(set DPR:$dst, (Ty (NEONvduplane (Ty DPR:$src), imm:$lane)))]>;
4136 class VDUPLNQ<bits<4> op19_16, string OpcodeStr, string Dt,
4137 ValueType ResTy, ValueType OpTy>
4138 : NVDupLane<op19_16, 1, (outs QPR:$dst), (ins DPR:$src, nohash_imm:$lane),
4139 IIC_VMOVQ, OpcodeStr, Dt, "$dst, $src[$lane]",
4140 [(set QPR:$dst, (ResTy (NEONvduplane (OpTy DPR:$src),
4143 // Inst{19-16} is partially specified depending on the element size.
4145 def VDUPLN8d : VDUPLND<{?,?,?,1}, "vdup", "8", v8i8> {
4146 let Inst{19-17} = lane{2-0};
4148 def VDUPLN16d : VDUPLND<{?,?,1,0}, "vdup", "16", v4i16> {
4149 let Inst{19-18} = lane{1-0};
4151 def VDUPLN32d : VDUPLND<{?,1,0,0}, "vdup", "32", v2i32> {
4152 let Inst{19} = lane{0};
4154 def VDUPLNfd : VDUPLND<{?,1,0,0}, "vdup", "32", v2f32> {
4155 let Inst{19} = lane{0};
4157 def VDUPLN8q : VDUPLNQ<{?,?,?,1}, "vdup", "8", v16i8, v8i8> {
4158 let Inst{19-17} = lane{2-0};
4160 def VDUPLN16q : VDUPLNQ<{?,?,1,0}, "vdup", "16", v8i16, v4i16> {
4161 let Inst{19-18} = lane{1-0};
4163 def VDUPLN32q : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4i32, v2i32> {
4164 let Inst{19} = lane{0};
4166 def VDUPLNfq : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4f32, v2f32> {
4167 let Inst{19} = lane{0};
4170 def : Pat<(v16i8 (NEONvduplane (v16i8 QPR:$src), imm:$lane)),
4171 (v16i8 (VDUPLN8q (v8i8 (EXTRACT_SUBREG QPR:$src,
4172 (DSubReg_i8_reg imm:$lane))),
4173 (SubReg_i8_lane imm:$lane)))>;
4174 def : Pat<(v8i16 (NEONvduplane (v8i16 QPR:$src), imm:$lane)),
4175 (v8i16 (VDUPLN16q (v4i16 (EXTRACT_SUBREG QPR:$src,
4176 (DSubReg_i16_reg imm:$lane))),
4177 (SubReg_i16_lane imm:$lane)))>;
4178 def : Pat<(v4i32 (NEONvduplane (v4i32 QPR:$src), imm:$lane)),
4179 (v4i32 (VDUPLN32q (v2i32 (EXTRACT_SUBREG QPR:$src,
4180 (DSubReg_i32_reg imm:$lane))),
4181 (SubReg_i32_lane imm:$lane)))>;
4182 def : Pat<(v4f32 (NEONvduplane (v4f32 QPR:$src), imm:$lane)),
4183 (v4f32 (VDUPLNfq (v2f32 (EXTRACT_SUBREG QPR:$src,
4184 (DSubReg_i32_reg imm:$lane))),
4185 (SubReg_i32_lane imm:$lane)))>;
4187 def VDUPfdf : PseudoNeonI<(outs DPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
4188 [(set DPR:$dst, (v2f32 (NEONvdup (f32 SPR:$src))))]>;
4189 def VDUPfqf : PseudoNeonI<(outs QPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
4190 [(set QPR:$dst, (v4f32 (NEONvdup (f32 SPR:$src))))]>;
4192 // VMOVN : Vector Narrowing Move
4193 defm VMOVN : N2VN_HSD<0b11,0b11,0b10,0b00100,0,0, IIC_VMOVN,
4194 "vmovn", "i", trunc>;
4195 // VQMOVN : Vector Saturating Narrowing Move
4196 defm VQMOVNs : N2VNInt_HSD<0b11,0b11,0b10,0b00101,0,0, IIC_VQUNAiD,
4197 "vqmovn", "s", int_arm_neon_vqmovns>;
4198 defm VQMOVNu : N2VNInt_HSD<0b11,0b11,0b10,0b00101,1,0, IIC_VQUNAiD,
4199 "vqmovn", "u", int_arm_neon_vqmovnu>;
4200 defm VQMOVNsu : N2VNInt_HSD<0b11,0b11,0b10,0b00100,1,0, IIC_VQUNAiD,
4201 "vqmovun", "s", int_arm_neon_vqmovnsu>;
4202 // VMOVL : Vector Lengthening Move
4203 defm VMOVLs : N2VL_QHS<0b01,0b10100,0,1, "vmovl", "s", sext>;
4204 defm VMOVLu : N2VL_QHS<0b11,0b10100,0,1, "vmovl", "u", zext>;
4206 // Vector Conversions.
4208 // VCVT : Vector Convert Between Floating-Point and Integers
4209 def VCVTf2sd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
4210 v2i32, v2f32, fp_to_sint>;
4211 def VCVTf2ud : N2VD<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
4212 v2i32, v2f32, fp_to_uint>;
4213 def VCVTs2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
4214 v2f32, v2i32, sint_to_fp>;
4215 def VCVTu2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
4216 v2f32, v2i32, uint_to_fp>;
4218 def VCVTf2sq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
4219 v4i32, v4f32, fp_to_sint>;
4220 def VCVTf2uq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
4221 v4i32, v4f32, fp_to_uint>;
4222 def VCVTs2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
4223 v4f32, v4i32, sint_to_fp>;
4224 def VCVTu2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
4225 v4f32, v4i32, uint_to_fp>;
4227 // VCVT : Vector Convert Between Floating-Point and Fixed-Point.
4228 def VCVTf2xsd : N2VCvtD<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
4229 v2i32, v2f32, int_arm_neon_vcvtfp2fxs>;
4230 def VCVTf2xud : N2VCvtD<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
4231 v2i32, v2f32, int_arm_neon_vcvtfp2fxu>;
4232 def VCVTxs2fd : N2VCvtD<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
4233 v2f32, v2i32, int_arm_neon_vcvtfxs2fp>;
4234 def VCVTxu2fd : N2VCvtD<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
4235 v2f32, v2i32, int_arm_neon_vcvtfxu2fp>;
4237 def VCVTf2xsq : N2VCvtQ<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
4238 v4i32, v4f32, int_arm_neon_vcvtfp2fxs>;
4239 def VCVTf2xuq : N2VCvtQ<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
4240 v4i32, v4f32, int_arm_neon_vcvtfp2fxu>;
4241 def VCVTxs2fq : N2VCvtQ<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
4242 v4f32, v4i32, int_arm_neon_vcvtfxs2fp>;
4243 def VCVTxu2fq : N2VCvtQ<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
4244 v4f32, v4i32, int_arm_neon_vcvtfxu2fp>;
4248 // VREV64 : Vector Reverse elements within 64-bit doublewords
4250 class VREV64D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4251 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 0, 0, (outs DPR:$dst),
4252 (ins DPR:$src), IIC_VMOVD,
4253 OpcodeStr, Dt, "$dst, $src", "",
4254 [(set DPR:$dst, (Ty (NEONvrev64 (Ty DPR:$src))))]>;
4255 class VREV64Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4256 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 1, 0, (outs QPR:$dst),
4257 (ins QPR:$src), IIC_VMOVQ,
4258 OpcodeStr, Dt, "$dst, $src", "",
4259 [(set QPR:$dst, (Ty (NEONvrev64 (Ty QPR:$src))))]>;
4261 def VREV64d8 : VREV64D<0b00, "vrev64", "8", v8i8>;
4262 def VREV64d16 : VREV64D<0b01, "vrev64", "16", v4i16>;
4263 def VREV64d32 : VREV64D<0b10, "vrev64", "32", v2i32>;
4264 def VREV64df : VREV64D<0b10, "vrev64", "32", v2f32>;
4266 def VREV64q8 : VREV64Q<0b00, "vrev64", "8", v16i8>;
4267 def VREV64q16 : VREV64Q<0b01, "vrev64", "16", v8i16>;
4268 def VREV64q32 : VREV64Q<0b10, "vrev64", "32", v4i32>;
4269 def VREV64qf : VREV64Q<0b10, "vrev64", "32", v4f32>;
4271 // VREV32 : Vector Reverse elements within 32-bit words
4273 class VREV32D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4274 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 0, 0, (outs DPR:$dst),
4275 (ins DPR:$src), IIC_VMOVD,
4276 OpcodeStr, Dt, "$dst, $src", "",
4277 [(set DPR:$dst, (Ty (NEONvrev32 (Ty DPR:$src))))]>;
4278 class VREV32Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4279 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 1, 0, (outs QPR:$dst),
4280 (ins QPR:$src), IIC_VMOVQ,
4281 OpcodeStr, Dt, "$dst, $src", "",
4282 [(set QPR:$dst, (Ty (NEONvrev32 (Ty QPR:$src))))]>;
4284 def VREV32d8 : VREV32D<0b00, "vrev32", "8", v8i8>;
4285 def VREV32d16 : VREV32D<0b01, "vrev32", "16", v4i16>;
4287 def VREV32q8 : VREV32Q<0b00, "vrev32", "8", v16i8>;
4288 def VREV32q16 : VREV32Q<0b01, "vrev32", "16", v8i16>;
4290 // VREV16 : Vector Reverse elements within 16-bit halfwords
4292 class VREV16D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4293 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 0, 0, (outs DPR:$dst),
4294 (ins DPR:$src), IIC_VMOVD,
4295 OpcodeStr, Dt, "$dst, $src", "",
4296 [(set DPR:$dst, (Ty (NEONvrev16 (Ty DPR:$src))))]>;
4297 class VREV16Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4298 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 1, 0, (outs QPR:$dst),
4299 (ins QPR:$src), IIC_VMOVQ,
4300 OpcodeStr, Dt, "$dst, $src", "",
4301 [(set QPR:$dst, (Ty (NEONvrev16 (Ty QPR:$src))))]>;
4303 def VREV16d8 : VREV16D<0b00, "vrev16", "8", v8i8>;
4304 def VREV16q8 : VREV16Q<0b00, "vrev16", "8", v16i8>;
4306 // Other Vector Shuffles.
4308 // VEXT : Vector Extract
4310 class VEXTd<string OpcodeStr, string Dt, ValueType Ty>
4311 : N3V<0,1,0b11,{?,?,?,?},0,0, (outs DPR:$dst),
4312 (ins DPR:$lhs, DPR:$rhs, i32imm:$index), NVExtFrm,
4313 IIC_VEXTD, OpcodeStr, Dt, "$dst, $lhs, $rhs, $index", "",
4314 [(set DPR:$dst, (Ty (NEONvext (Ty DPR:$lhs),
4315 (Ty DPR:$rhs), imm:$index)))]> {
4317 let Inst{11-8} = index{3-0};
4320 class VEXTq<string OpcodeStr, string Dt, ValueType Ty>
4321 : N3V<0,1,0b11,{?,?,?,?},1,0, (outs QPR:$dst),
4322 (ins QPR:$lhs, QPR:$rhs, i32imm:$index), NVExtFrm,
4323 IIC_VEXTQ, OpcodeStr, Dt, "$dst, $lhs, $rhs, $index", "",
4324 [(set QPR:$dst, (Ty (NEONvext (Ty QPR:$lhs),
4325 (Ty QPR:$rhs), imm:$index)))]> {
4327 let Inst{11-8} = index{3-0};
4330 def VEXTd8 : VEXTd<"vext", "8", v8i8> {
4331 let Inst{11-8} = index{3-0};
4333 def VEXTd16 : VEXTd<"vext", "16", v4i16> {
4334 let Inst{11-9} = index{2-0};
4337 def VEXTd32 : VEXTd<"vext", "32", v2i32> {
4338 let Inst{11-10} = index{1-0};
4339 let Inst{9-8} = 0b00;
4341 def VEXTdf : VEXTd<"vext", "32", v2f32> {
4342 let Inst{11} = index{0};
4343 let Inst{10-8} = 0b000;
4346 def VEXTq8 : VEXTq<"vext", "8", v16i8> {
4347 let Inst{11-8} = index{3-0};
4349 def VEXTq16 : VEXTq<"vext", "16", v8i16> {
4350 let Inst{11-9} = index{2-0};
4353 def VEXTq32 : VEXTq<"vext", "32", v4i32> {
4354 let Inst{11-10} = index{1-0};
4355 let Inst{9-8} = 0b00;
4357 def VEXTqf : VEXTq<"vext", "32", v4f32> {
4358 let Inst{11} = index{0};
4359 let Inst{10-8} = 0b000;
4362 // VTRN : Vector Transpose
4364 def VTRNd8 : N2VDShuffle<0b00, 0b00001, "vtrn", "8">;
4365 def VTRNd16 : N2VDShuffle<0b01, 0b00001, "vtrn", "16">;
4366 def VTRNd32 : N2VDShuffle<0b10, 0b00001, "vtrn", "32">;
4368 def VTRNq8 : N2VQShuffle<0b00, 0b00001, IIC_VPERMQ, "vtrn", "8">;
4369 def VTRNq16 : N2VQShuffle<0b01, 0b00001, IIC_VPERMQ, "vtrn", "16">;
4370 def VTRNq32 : N2VQShuffle<0b10, 0b00001, IIC_VPERMQ, "vtrn", "32">;
4372 // VUZP : Vector Unzip (Deinterleave)
4374 def VUZPd8 : N2VDShuffle<0b00, 0b00010, "vuzp", "8">;
4375 def VUZPd16 : N2VDShuffle<0b01, 0b00010, "vuzp", "16">;
4376 def VUZPd32 : N2VDShuffle<0b10, 0b00010, "vuzp", "32">;
4378 def VUZPq8 : N2VQShuffle<0b00, 0b00010, IIC_VPERMQ3, "vuzp", "8">;
4379 def VUZPq16 : N2VQShuffle<0b01, 0b00010, IIC_VPERMQ3, "vuzp", "16">;
4380 def VUZPq32 : N2VQShuffle<0b10, 0b00010, IIC_VPERMQ3, "vuzp", "32">;
4382 // VZIP : Vector Zip (Interleave)
4384 def VZIPd8 : N2VDShuffle<0b00, 0b00011, "vzip", "8">;
4385 def VZIPd16 : N2VDShuffle<0b01, 0b00011, "vzip", "16">;
4386 def VZIPd32 : N2VDShuffle<0b10, 0b00011, "vzip", "32">;
4388 def VZIPq8 : N2VQShuffle<0b00, 0b00011, IIC_VPERMQ3, "vzip", "8">;
4389 def VZIPq16 : N2VQShuffle<0b01, 0b00011, IIC_VPERMQ3, "vzip", "16">;
4390 def VZIPq32 : N2VQShuffle<0b10, 0b00011, IIC_VPERMQ3, "vzip", "32">;
4392 // Vector Table Lookup and Table Extension.
4394 // VTBL : Vector Table Lookup
4396 : N3V<1,1,0b11,0b1000,0,0, (outs DPR:$Vd),
4397 (ins DPR:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTB1,
4398 "vtbl", "8", "$Vd, \\{$Vn\\}, $Vm", "",
4399 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbl1 DPR:$Vn, DPR:$Vm)))]>;
4400 let hasExtraSrcRegAllocReq = 1 in {
4402 : N3V<1,1,0b11,0b1001,0,0, (outs DPR:$Vd),
4403 (ins DPR:$Vn, DPR:$tbl2, DPR:$Vm), NVTBLFrm, IIC_VTB2,
4404 "vtbl", "8", "$Vd, \\{$Vn, $tbl2\\}, $Vm", "", []>;
4406 : N3V<1,1,0b11,0b1010,0,0, (outs DPR:$Vd),
4407 (ins DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$Vm), NVTBLFrm, IIC_VTB3,
4408 "vtbl", "8", "$Vd, \\{$Vn, $tbl2, $tbl3\\}, $Vm", "", []>;
4410 : N3V<1,1,0b11,0b1011,0,0, (outs DPR:$Vd),
4411 (ins DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$Vm),
4413 "vtbl", "8", "$Vd, \\{$Vn, $tbl2, $tbl3, $tbl4\\}, $Vm", "", []>;
4414 } // hasExtraSrcRegAllocReq = 1
4417 : PseudoNeonI<(outs DPR:$dst), (ins QPR:$tbl, DPR:$src), IIC_VTB2, "", []>;
4419 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB3, "", []>;
4421 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB4, "", []>;
4423 // VTBX : Vector Table Extension
4425 : N3V<1,1,0b11,0b1000,1,0, (outs DPR:$Vd),
4426 (ins DPR:$orig, DPR:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTBX1,
4427 "vtbx", "8", "$Vd, \\{$Vn\\}, $Vm", "$orig = $Vd",
4428 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbx1
4429 DPR:$orig, DPR:$Vn, DPR:$Vm)))]>;
4430 let hasExtraSrcRegAllocReq = 1 in {
4432 : N3V<1,1,0b11,0b1001,1,0, (outs DPR:$Vd),
4433 (ins DPR:$orig, DPR:$Vn, DPR:$tbl2, DPR:$Vm), NVTBLFrm, IIC_VTBX2,
4434 "vtbx", "8", "$Vd, \\{$Vn, $tbl2\\}, $Vm", "$orig = $Vd", []>;
4436 : N3V<1,1,0b11,0b1010,1,0, (outs DPR:$Vd),
4437 (ins DPR:$orig, DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$Vm),
4438 NVTBLFrm, IIC_VTBX3,
4439 "vtbx", "8", "$Vd, \\{$Vn, $tbl2, $tbl3\\}, $Vm",
4442 : N3V<1,1,0b11,0b1011,1,0, (outs DPR:$Vd), (ins DPR:$orig, DPR:$Vn,
4443 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$Vm), NVTBLFrm, IIC_VTBX4,
4444 "vtbx", "8", "$Vd, \\{$Vn, $tbl2, $tbl3, $tbl4\\}, $Vm",
4446 } // hasExtraSrcRegAllocReq = 1
4449 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QPR:$tbl, DPR:$src),
4450 IIC_VTBX2, "$orig = $dst", []>;
4452 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
4453 IIC_VTBX3, "$orig = $dst", []>;
4455 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
4456 IIC_VTBX4, "$orig = $dst", []>;
4458 //===----------------------------------------------------------------------===//
4459 // NEON instructions for single-precision FP math
4460 //===----------------------------------------------------------------------===//
4462 class N2VSPat<SDNode OpNode, ValueType ResTy, ValueType OpTy, NeonI Inst>
4463 : NEONFPPat<(ResTy (OpNode SPR:$a)),
4464 (EXTRACT_SUBREG (OpTy (Inst (INSERT_SUBREG (OpTy (IMPLICIT_DEF)),
4468 class N3VSPat<SDNode OpNode, NeonI Inst>
4469 : NEONFPPat<(f32 (OpNode SPR:$a, SPR:$b)),
4470 (EXTRACT_SUBREG (v2f32
4471 (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
4473 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
4477 class N3VSMulOpPat<SDNode MulNode, SDNode OpNode, NeonI Inst>
4478 : NEONFPPat<(f32 (OpNode SPR:$acc, (f32 (MulNode SPR:$a, SPR:$b)))),
4479 (EXTRACT_SUBREG (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
4481 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
4483 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
4487 // These need separate instructions because they must use DPR_VFP2 register
4488 // class which have SPR sub-registers.
4490 // Vector Add Operations used for single-precision FP
4491 let neverHasSideEffects = 1 in
4492 def VADDfd_sfp : N3VS<0,0,0b00,0b1101,0, "vadd", "f32", v2f32, v2f32, fadd, 1>;
4493 def : N3VSPat<fadd, VADDfd_sfp>;
4495 // Vector Sub Operations used for single-precision FP
4496 let neverHasSideEffects = 1 in
4497 def VSUBfd_sfp : N3VS<0,0,0b10,0b1101,0, "vsub", "f32", v2f32, v2f32, fsub, 0>;
4498 def : N3VSPat<fsub, VSUBfd_sfp>;
4500 // Vector Multiply Operations used for single-precision FP
4501 let neverHasSideEffects = 1 in
4502 def VMULfd_sfp : N3VS<1,0,0b00,0b1101,1, "vmul", "f32", v2f32, v2f32, fmul, 1>;
4503 def : N3VSPat<fmul, VMULfd_sfp>;
4505 // Vector Multiply-Accumulate/Subtract used for single-precision FP
4506 // vml[as].f32 can cause 4-8 cycle stalls in following ASIMD instructions, so
4507 // we want to avoid them for now. e.g., alternating vmla/vadd instructions.
4509 //let neverHasSideEffects = 1 in
4510 //def VMLAfd_sfp : N3VSMulOp<0,0,0b00,0b1101,1, IIC_VMACD, "vmla", "f32",
4511 // v2f32, fmul, fadd>;
4512 //def : N3VSMulOpPat<fmul, fadd, VMLAfd_sfp>;
4514 //let neverHasSideEffects = 1 in
4515 //def VMLSfd_sfp : N3VSMulOp<0,0,0b10,0b1101,1, IIC_VMACD, "vmls", "f32",
4516 // v2f32, fmul, fsub>;
4517 //def : N3VSMulOpPat<fmul, fsub, VMLSfd_sfp>;
4519 // Vector Absolute used for single-precision FP
4520 let neverHasSideEffects = 1 in
4521 def VABSfd_sfp : N2V<0b11, 0b11, 0b10, 0b01, 0b01110, 0, 0,
4522 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), IIC_VUNAD,
4523 "vabs", "f32", "$dst, $src", "", []>;
4524 def : N2VSPat<fabs, f32, v2f32, VABSfd_sfp>;
4526 // Vector Negate used for single-precision FP
4527 let neverHasSideEffects = 1 in
4528 def VNEGfd_sfp : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
4529 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), IIC_VUNAD,
4530 "vneg", "f32", "$dst, $src", "", []>;
4531 def : N2VSPat<fneg, f32, v2f32, VNEGfd_sfp>;
4533 // Vector Maximum used for single-precision FP
4534 let neverHasSideEffects = 1 in
4535 def VMAXfd_sfp : N3V<0, 0, 0b00, 0b1111, 0, 0, (outs DPR_VFP2:$dst),
4536 (ins DPR_VFP2:$src1, DPR_VFP2:$src2), N3RegFrm, IIC_VBIND,
4537 "vmax", "f32", "$dst, $src1, $src2", "", []>;
4538 def : N3VSPat<NEONfmax, VMAXfd_sfp>;
4540 // Vector Minimum used for single-precision FP
4541 let neverHasSideEffects = 1 in
4542 def VMINfd_sfp : N3V<0, 0, 0b00, 0b1111, 0, 0, (outs DPR_VFP2:$dst),
4543 (ins DPR_VFP2:$src1, DPR_VFP2:$src2), N3RegFrm, IIC_VBIND,
4544 "vmin", "f32", "$dst, $src1, $src2", "", []>;
4545 def : N3VSPat<NEONfmin, VMINfd_sfp>;
4547 // Vector Convert between single-precision FP and integer
4548 let neverHasSideEffects = 1 in
4549 def VCVTf2sd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
4550 v2i32, v2f32, fp_to_sint>;
4551 def : N2VSPat<arm_ftosi, f32, v2f32, VCVTf2sd_sfp>;
4553 let neverHasSideEffects = 1 in
4554 def VCVTf2ud_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
4555 v2i32, v2f32, fp_to_uint>;
4556 def : N2VSPat<arm_ftoui, f32, v2f32, VCVTf2ud_sfp>;
4558 let neverHasSideEffects = 1 in
4559 def VCVTs2fd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
4560 v2f32, v2i32, sint_to_fp>;
4561 def : N2VSPat<arm_sitof, f32, v2i32, VCVTs2fd_sfp>;
4563 let neverHasSideEffects = 1 in
4564 def VCVTu2fd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
4565 v2f32, v2i32, uint_to_fp>;
4566 def : N2VSPat<arm_uitof, f32, v2i32, VCVTu2fd_sfp>;
4568 //===----------------------------------------------------------------------===//
4569 // Non-Instruction Patterns
4570 //===----------------------------------------------------------------------===//
4573 def : Pat<(v1i64 (bitconvert (v2i32 DPR:$src))), (v1i64 DPR:$src)>;
4574 def : Pat<(v1i64 (bitconvert (v4i16 DPR:$src))), (v1i64 DPR:$src)>;
4575 def : Pat<(v1i64 (bitconvert (v8i8 DPR:$src))), (v1i64 DPR:$src)>;
4576 def : Pat<(v1i64 (bitconvert (f64 DPR:$src))), (v1i64 DPR:$src)>;
4577 def : Pat<(v1i64 (bitconvert (v2f32 DPR:$src))), (v1i64 DPR:$src)>;
4578 def : Pat<(v2i32 (bitconvert (v1i64 DPR:$src))), (v2i32 DPR:$src)>;
4579 def : Pat<(v2i32 (bitconvert (v4i16 DPR:$src))), (v2i32 DPR:$src)>;
4580 def : Pat<(v2i32 (bitconvert (v8i8 DPR:$src))), (v2i32 DPR:$src)>;
4581 def : Pat<(v2i32 (bitconvert (f64 DPR:$src))), (v2i32 DPR:$src)>;
4582 def : Pat<(v2i32 (bitconvert (v2f32 DPR:$src))), (v2i32 DPR:$src)>;
4583 def : Pat<(v4i16 (bitconvert (v1i64 DPR:$src))), (v4i16 DPR:$src)>;
4584 def : Pat<(v4i16 (bitconvert (v2i32 DPR:$src))), (v4i16 DPR:$src)>;
4585 def : Pat<(v4i16 (bitconvert (v8i8 DPR:$src))), (v4i16 DPR:$src)>;
4586 def : Pat<(v4i16 (bitconvert (f64 DPR:$src))), (v4i16 DPR:$src)>;
4587 def : Pat<(v4i16 (bitconvert (v2f32 DPR:$src))), (v4i16 DPR:$src)>;
4588 def : Pat<(v8i8 (bitconvert (v1i64 DPR:$src))), (v8i8 DPR:$src)>;
4589 def : Pat<(v8i8 (bitconvert (v2i32 DPR:$src))), (v8i8 DPR:$src)>;
4590 def : Pat<(v8i8 (bitconvert (v4i16 DPR:$src))), (v8i8 DPR:$src)>;
4591 def : Pat<(v8i8 (bitconvert (f64 DPR:$src))), (v8i8 DPR:$src)>;
4592 def : Pat<(v8i8 (bitconvert (v2f32 DPR:$src))), (v8i8 DPR:$src)>;
4593 def : Pat<(f64 (bitconvert (v1i64 DPR:$src))), (f64 DPR:$src)>;
4594 def : Pat<(f64 (bitconvert (v2i32 DPR:$src))), (f64 DPR:$src)>;
4595 def : Pat<(f64 (bitconvert (v4i16 DPR:$src))), (f64 DPR:$src)>;
4596 def : Pat<(f64 (bitconvert (v8i8 DPR:$src))), (f64 DPR:$src)>;
4597 def : Pat<(f64 (bitconvert (v2f32 DPR:$src))), (f64 DPR:$src)>;
4598 def : Pat<(v2f32 (bitconvert (f64 DPR:$src))), (v2f32 DPR:$src)>;
4599 def : Pat<(v2f32 (bitconvert (v1i64 DPR:$src))), (v2f32 DPR:$src)>;
4600 def : Pat<(v2f32 (bitconvert (v2i32 DPR:$src))), (v2f32 DPR:$src)>;
4601 def : Pat<(v2f32 (bitconvert (v4i16 DPR:$src))), (v2f32 DPR:$src)>;
4602 def : Pat<(v2f32 (bitconvert (v8i8 DPR:$src))), (v2f32 DPR:$src)>;
4604 def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (v2i64 QPR:$src)>;
4605 def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (v2i64 QPR:$src)>;
4606 def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (v2i64 QPR:$src)>;
4607 def : Pat<(v2i64 (bitconvert (v2f64 QPR:$src))), (v2i64 QPR:$src)>;
4608 def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (v2i64 QPR:$src)>;
4609 def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (v4i32 QPR:$src)>;
4610 def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (v4i32 QPR:$src)>;
4611 def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (v4i32 QPR:$src)>;
4612 def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (v4i32 QPR:$src)>;
4613 def : Pat<(v4i32 (bitconvert (v4f32 QPR:$src))), (v4i32 QPR:$src)>;
4614 def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (v8i16 QPR:$src)>;
4615 def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (v8i16 QPR:$src)>;
4616 def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (v8i16 QPR:$src)>;
4617 def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (v8i16 QPR:$src)>;
4618 def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (v8i16 QPR:$src)>;
4619 def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (v16i8 QPR:$src)>;
4620 def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (v16i8 QPR:$src)>;
4621 def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (v16i8 QPR:$src)>;
4622 def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (v16i8 QPR:$src)>;
4623 def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (v16i8 QPR:$src)>;
4624 def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (v4f32 QPR:$src)>;
4625 def : Pat<(v4f32 (bitconvert (v4i32 QPR:$src))), (v4f32 QPR:$src)>;
4626 def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (v4f32 QPR:$src)>;
4627 def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (v4f32 QPR:$src)>;
4628 def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (v4f32 QPR:$src)>;
4629 def : Pat<(v2f64 (bitconvert (v2i64 QPR:$src))), (v2f64 QPR:$src)>;
4630 def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (v2f64 QPR:$src)>;
4631 def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 QPR:$src)>;
4632 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 QPR:$src)>;
4633 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 QPR:$src)>;