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>]>;
20 def NEONvceq : SDNode<"ARMISD::VCEQ", SDTARMVCMP>;
21 def NEONvcge : SDNode<"ARMISD::VCGE", SDTARMVCMP>;
22 def NEONvcgeu : SDNode<"ARMISD::VCGEU", SDTARMVCMP>;
23 def NEONvcgt : SDNode<"ARMISD::VCGT", SDTARMVCMP>;
24 def NEONvcgtu : SDNode<"ARMISD::VCGTU", SDTARMVCMP>;
25 def NEONvtst : SDNode<"ARMISD::VTST", SDTARMVCMP>;
27 // Types for vector shift by immediates. The "SHX" version is for long and
28 // narrow operations where the source and destination vectors have different
29 // types. The "SHINS" version is for shift and insert operations.
30 def SDTARMVSH : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
32 def SDTARMVSHX : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
34 def SDTARMVSHINS : SDTypeProfile<1, 3, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
35 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
37 def NEONvshl : SDNode<"ARMISD::VSHL", SDTARMVSH>;
38 def NEONvshrs : SDNode<"ARMISD::VSHRs", SDTARMVSH>;
39 def NEONvshru : SDNode<"ARMISD::VSHRu", SDTARMVSH>;
40 def NEONvshlls : SDNode<"ARMISD::VSHLLs", SDTARMVSHX>;
41 def NEONvshllu : SDNode<"ARMISD::VSHLLu", SDTARMVSHX>;
42 def NEONvshlli : SDNode<"ARMISD::VSHLLi", SDTARMVSHX>;
43 def NEONvshrn : SDNode<"ARMISD::VSHRN", SDTARMVSHX>;
45 def NEONvrshrs : SDNode<"ARMISD::VRSHRs", SDTARMVSH>;
46 def NEONvrshru : SDNode<"ARMISD::VRSHRu", SDTARMVSH>;
47 def NEONvrshrn : SDNode<"ARMISD::VRSHRN", SDTARMVSHX>;
49 def NEONvqshls : SDNode<"ARMISD::VQSHLs", SDTARMVSH>;
50 def NEONvqshlu : SDNode<"ARMISD::VQSHLu", SDTARMVSH>;
51 def NEONvqshlsu : SDNode<"ARMISD::VQSHLsu", SDTARMVSH>;
52 def NEONvqshrns : SDNode<"ARMISD::VQSHRNs", SDTARMVSHX>;
53 def NEONvqshrnu : SDNode<"ARMISD::VQSHRNu", SDTARMVSHX>;
54 def NEONvqshrnsu : SDNode<"ARMISD::VQSHRNsu", SDTARMVSHX>;
56 def NEONvqrshrns : SDNode<"ARMISD::VQRSHRNs", SDTARMVSHX>;
57 def NEONvqrshrnu : SDNode<"ARMISD::VQRSHRNu", SDTARMVSHX>;
58 def NEONvqrshrnsu : SDNode<"ARMISD::VQRSHRNsu", SDTARMVSHX>;
60 def NEONvsli : SDNode<"ARMISD::VSLI", SDTARMVSHINS>;
61 def NEONvsri : SDNode<"ARMISD::VSRI", SDTARMVSHINS>;
63 def SDTARMVGETLN : SDTypeProfile<1, 2, [SDTCisVT<0, i32>, SDTCisInt<1>,
65 def NEONvgetlaneu : SDNode<"ARMISD::VGETLANEu", SDTARMVGETLN>;
66 def NEONvgetlanes : SDNode<"ARMISD::VGETLANEs", SDTARMVGETLN>;
68 def NEONvdup : SDNode<"ARMISD::VDUP", SDTypeProfile<1, 1, [SDTCisVec<0>]>>;
70 // VDUPLANE can produce a quad-register result from a double-register source,
71 // so the result is not constrained to match the source.
72 def NEONvduplane : SDNode<"ARMISD::VDUPLANE",
73 SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisVec<1>,
76 def SDTARMVEXT : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
77 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
78 def NEONvext : SDNode<"ARMISD::VEXT", SDTARMVEXT>;
80 def SDTARMVSHUF : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisSameAs<0, 1>]>;
81 def NEONvrev64 : SDNode<"ARMISD::VREV64", SDTARMVSHUF>;
82 def NEONvrev32 : SDNode<"ARMISD::VREV32", SDTARMVSHUF>;
83 def NEONvrev16 : SDNode<"ARMISD::VREV16", SDTARMVSHUF>;
85 def SDTARMVSHUF2 : SDTypeProfile<2, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
88 def NEONzip : SDNode<"ARMISD::VZIP", SDTARMVSHUF2>;
89 def NEONuzp : SDNode<"ARMISD::VUZP", SDTARMVSHUF2>;
90 def NEONtrn : SDNode<"ARMISD::VTRN", SDTARMVSHUF2>;
92 def SDTARMFMAX : SDTypeProfile<1, 2, [SDTCisVT<0, f32>, SDTCisSameAs<0, 1>,
94 def NEONfmax : SDNode<"ARMISD::FMAX", SDTARMFMAX>;
95 def NEONfmin : SDNode<"ARMISD::FMIN", SDTARMFMAX>;
97 //===----------------------------------------------------------------------===//
98 // NEON operand definitions
99 //===----------------------------------------------------------------------===//
101 def h8imm : Operand<i8> {
102 let PrintMethod = "printHex8ImmOperand";
104 def h16imm : Operand<i16> {
105 let PrintMethod = "printHex16ImmOperand";
107 def h32imm : Operand<i32> {
108 let PrintMethod = "printHex32ImmOperand";
110 def h64imm : Operand<i64> {
111 let PrintMethod = "printHex64ImmOperand";
114 //===----------------------------------------------------------------------===//
115 // NEON load / store instructions
116 //===----------------------------------------------------------------------===//
118 // Use vldmia to load a Q register as a D register pair.
119 def VLDRQ : NI4<(outs QPR:$dst), (ins addrmode4:$addr), IIC_fpLoadm,
120 "vldmia", "$addr, ${dst:dregpair}",
121 [(set QPR:$dst, (v2f64 (load addrmode4:$addr)))]> {
122 let Inst{27-25} = 0b110;
123 let Inst{24} = 0; // P bit
124 let Inst{23} = 1; // U bit
126 let Inst{11-8} = 0b1011;
129 // Use vstmia to store a Q register as a D register pair.
130 def VSTRQ : NI4<(outs), (ins QPR:$src, addrmode4:$addr), IIC_fpStorem,
131 "vstmia", "$addr, ${src:dregpair}",
132 [(store (v2f64 QPR:$src), addrmode4:$addr)]> {
133 let Inst{27-25} = 0b110;
134 let Inst{24} = 0; // P bit
135 let Inst{23} = 1; // U bit
137 let Inst{11-8} = 0b1011;
140 // VLD1 : Vector Load (multiple single elements)
141 class VLD1D<bits<4> op7_4, string Dt, ValueType Ty>
142 : NLdSt<0,0b10,0b0111,op7_4, (outs DPR:$dst), (ins addrmode6:$addr), IIC_VLD1,
143 "vld1", Dt, "\\{$dst\\}, $addr", "",
144 [(set DPR:$dst, (Ty (int_arm_neon_vld1 addrmode6:$addr)))]>;
145 class VLD1Q<bits<4> op7_4, string Dt, ValueType Ty>
146 : NLdSt<0,0b10,0b1010,op7_4, (outs QPR:$dst), (ins addrmode6:$addr), IIC_VLD1,
147 "vld1", Dt, "${dst:dregpair}, $addr", "",
148 [(set QPR:$dst, (Ty (int_arm_neon_vld1 addrmode6:$addr)))]>;
150 def VLD1d8 : VLD1D<0b0000, "8", v8i8>;
151 def VLD1d16 : VLD1D<0b0100, "16", v4i16>;
152 def VLD1d32 : VLD1D<0b1000, "32", v2i32>;
153 def VLD1df : VLD1D<0b1000, "32", v2f32>;
154 def VLD1d64 : VLD1D<0b1100, "64", v1i64>;
156 def VLD1q8 : VLD1Q<0b0000, "8", v16i8>;
157 def VLD1q16 : VLD1Q<0b0100, "16", v8i16>;
158 def VLD1q32 : VLD1Q<0b1000, "32", v4i32>;
159 def VLD1qf : VLD1Q<0b1000, "32", v4f32>;
160 def VLD1q64 : VLD1Q<0b1100, "64", v2i64>;
164 // ...with address register writeback:
165 class VLD1DWB<bits<4> op7_4, string Dt>
166 : NLdSt<0,0b10,0b0111,op7_4, (outs DPR:$dst, GPR:$wb),
167 (ins addrmode6:$addr), IIC_VLD1,
168 "vld1", Dt, "\\{$dst\\}, $addr",
169 "$addr.addr = $wb", []>;
170 class VLD1QWB<bits<4> op7_4, string Dt>
171 : NLdSt<0,0b10,0b1010,op7_4, (outs QPR:$dst, GPR:$wb),
172 (ins addrmode6:$addr), IIC_VLD1,
173 "vld1", Dt, "${dst:dregpair}, $addr",
174 "$addr.addr = $wb", []>;
176 def VLD1d8_UPD : VLD1DWB<0b0000, "8">;
177 def VLD1d16_UPD : VLD1DWB<0b0100, "16">;
178 def VLD1d32_UPD : VLD1DWB<0b1000, "32">;
179 def VLD1d64_UPD : VLD1DWB<0b1100, "64">;
181 def VLD1q8_UPD : VLD1QWB<0b0000, "8">;
182 def VLD1q16_UPD : VLD1QWB<0b0100, "16">;
183 def VLD1q32_UPD : VLD1QWB<0b1000, "32">;
184 def VLD1q64_UPD : VLD1QWB<0b1100, "64">;
187 let mayLoad = 1, hasExtraDefRegAllocReq = 1 in {
189 // These (dreg triple/quadruple) are for disassembly only.
190 class VLD1D3<bits<4> op7_4, string Dt>
191 : NLdSt<0,0b10,0b0110,op7_4, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3),
192 (ins addrmode6:$addr), IIC_VLD1, "vld1", Dt,
193 "\\{$dst1, $dst2, $dst3\\}, $addr", "",
194 [/* For disassembly only; pattern left blank */]>;
195 class VLD1D4<bits<4> op7_4, string Dt>
196 : NLdSt<0,0b10,0b0010,op7_4,(outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
197 (ins addrmode6:$addr), IIC_VLD1, "vld1", Dt,
198 "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr", "",
199 [/* For disassembly only; pattern left blank */]>;
201 def VLD1d8T : VLD1D3<0b0000, "8">;
202 def VLD1d16T : VLD1D3<0b0100, "16">;
203 def VLD1d32T : VLD1D3<0b1000, "32">;
204 // VLD1d64T : implemented as VLD3d64
206 def VLD1d8Q : VLD1D4<0b0000, "8">;
207 def VLD1d16Q : VLD1D4<0b0100, "16">;
208 def VLD1d32Q : VLD1D4<0b1000, "32">;
209 // VLD1d64Q : implemented as VLD4d64
211 // ...with address register writeback:
212 class VLD1D3WB<bits<4> op7_4, string Dt>
213 : NLdSt<0,0b10,0b0110,op7_4, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, GPR:$wb),
214 (ins addrmode6:$addr), IIC_VLD1, "vld1", Dt,
215 "\\{$dst1, $dst2, $dst3\\}, $addr", "$addr.addr = $wb",
216 [/* For disassembly only; pattern left blank */]>;
217 class VLD1D4WB<bits<4> op7_4, string Dt>
218 : NLdSt<0,0b10,0b0010,op7_4,
219 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
220 (ins addrmode6:$addr), IIC_VLD1, "vld1", Dt,
221 "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr", "$addr.addr = $wb",
222 [/* For disassembly only; pattern left blank */]>;
224 def VLD1d8T_UPD : VLD1D3WB<0b0000, "8">;
225 def VLD1d16T_UPD : VLD1D3WB<0b0100, "16">;
226 def VLD1d32T_UPD : VLD1D3WB<0b1000, "32">;
227 // VLD1d64T_UPD : implemented as VLD3d64_UPD
229 def VLD1d8Q_UPD : VLD1D4WB<0b0000, "8">;
230 def VLD1d16Q_UPD : VLD1D4WB<0b0100, "16">;
231 def VLD1d32Q_UPD : VLD1D4WB<0b1000, "32">;
232 // VLD1d64Q_UPD : implemented as VLD4d64_UPD
234 // VLD2 : Vector Load (multiple 2-element structures)
235 class VLD2D<bits<4> op11_8, bits<4> op7_4, string Dt>
236 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2),
237 (ins addrmode6:$addr), IIC_VLD2,
238 "vld2", Dt, "\\{$dst1, $dst2\\}, $addr", "", []>;
239 class VLD2Q<bits<4> op7_4, string Dt>
240 : NLdSt<0, 0b10, 0b0011, op7_4,
241 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
242 (ins addrmode6:$addr), IIC_VLD2,
243 "vld2", Dt, "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr", "", []>;
245 def VLD2d8 : VLD2D<0b1000, 0b0000, "8">;
246 def VLD2d16 : VLD2D<0b1000, 0b0100, "16">;
247 def VLD2d32 : VLD2D<0b1000, 0b1000, "32">;
248 def VLD2d64 : NLdSt<0,0b10,0b1010,0b1100, (outs DPR:$dst1, DPR:$dst2),
249 (ins addrmode6:$addr), IIC_VLD1,
250 "vld1", "64", "\\{$dst1, $dst2\\}, $addr", "", []>;
252 def VLD2q8 : VLD2Q<0b0000, "8">;
253 def VLD2q16 : VLD2Q<0b0100, "16">;
254 def VLD2q32 : VLD2Q<0b1000, "32">;
256 // ...with address register writeback:
257 class VLD2DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
258 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2, GPR:$wb),
259 (ins addrmode6:$addr), IIC_VLD2,
260 "vld2", Dt, "\\{$dst1, $dst2\\}, $addr",
261 "$addr.addr = $wb", []>;
262 class VLD2QWB<bits<4> op7_4, string Dt>
263 : NLdSt<0, 0b10, 0b0011, op7_4,
264 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
265 (ins addrmode6:$addr), IIC_VLD2,
266 "vld2", Dt, "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr",
267 "$addr.addr = $wb", []>;
269 def VLD2d8_UPD : VLD2DWB<0b1000, 0b0000, "8">;
270 def VLD2d16_UPD : VLD2DWB<0b1000, 0b0100, "16">;
271 def VLD2d32_UPD : VLD2DWB<0b1000, 0b1000, "32">;
272 def VLD2d64_UPD : NLdSt<0,0b10,0b1010,0b1100,
273 (outs DPR:$dst1, DPR:$dst2, GPR:$wb),
274 (ins addrmode6:$addr), IIC_VLD1,
275 "vld1", "64", "\\{$dst1, $dst2\\}, $addr",
276 "$addr.addr = $wb", []>;
278 def VLD2q8_UPD : VLD2QWB<0b0000, "8">;
279 def VLD2q16_UPD : VLD2QWB<0b0100, "16">;
280 def VLD2q32_UPD : VLD2QWB<0b1000, "32">;
282 // ...with double-spaced registers (for disassembly only):
283 def VLD2b8 : VLD2D<0b1001, 0b0000, "8">;
284 def VLD2b16 : VLD2D<0b1001, 0b0100, "16">;
285 def VLD2b32 : VLD2D<0b1001, 0b1000, "32">;
286 def VLD2b8_UPD : VLD2DWB<0b1001, 0b0000, "8">;
287 def VLD2b16_UPD : VLD2DWB<0b1001, 0b0100, "16">;
288 def VLD2b32_UPD : VLD2DWB<0b1001, 0b1000, "32">;
290 // VLD3 : Vector Load (multiple 3-element structures)
291 class VLD3D<bits<4> op11_8, bits<4> op7_4, string Dt>
292 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3),
293 (ins addrmode6:$addr), IIC_VLD3,
294 "vld3", Dt, "\\{$dst1, $dst2, $dst3\\}, $addr", "", []>;
296 def VLD3d8 : VLD3D<0b0100, 0b0000, "8">;
297 def VLD3d16 : VLD3D<0b0100, 0b0100, "16">;
298 def VLD3d32 : VLD3D<0b0100, 0b1000, "32">;
299 def VLD3d64 : NLdSt<0,0b10,0b0110,0b1100,
300 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3),
301 (ins addrmode6:$addr), IIC_VLD1,
302 "vld1", "64", "\\{$dst1, $dst2, $dst3\\}, $addr", "", []>;
304 // ...with address register writeback:
305 class VLD3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
306 : NLdSt<0, 0b10, op11_8, op7_4,
307 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, GPR:$wb),
308 (ins addrmode6:$addr), IIC_VLD3,
309 "vld3", Dt, "\\{$dst1, $dst2, $dst3\\}, $addr",
310 "$addr.addr = $wb", []>;
312 def VLD3d8_UPD : VLD3DWB<0b0100, 0b0000, "8">;
313 def VLD3d16_UPD : VLD3DWB<0b0100, 0b0100, "16">;
314 def VLD3d32_UPD : VLD3DWB<0b0100, 0b1000, "32">;
315 def VLD3d64_UPD : NLdSt<0,0b10,0b0110,0b1100,
316 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, GPR:$wb),
317 (ins addrmode6:$addr), IIC_VLD1,
318 "vld1", "64", "\\{$dst1, $dst2, $dst3\\}, $addr",
319 "$addr.addr = $wb", []>;
321 // ...with double-spaced registers (non-updating versions for disassembly only):
322 def VLD3q8 : VLD3D<0b0101, 0b0000, "8">;
323 def VLD3q16 : VLD3D<0b0101, 0b0100, "16">;
324 def VLD3q32 : VLD3D<0b0101, 0b1000, "32">;
325 def VLD3q8_UPD : VLD3DWB<0b0101, 0b0000, "8">;
326 def VLD3q16_UPD : VLD3DWB<0b0101, 0b0100, "16">;
327 def VLD3q32_UPD : VLD3DWB<0b0101, 0b1000, "32">;
329 // ...alternate versions to be allocated odd register numbers:
330 def VLD3q8odd_UPD : VLD3DWB<0b0101, 0b0000, "8">;
331 def VLD3q16odd_UPD : VLD3DWB<0b0101, 0b0100, "16">;
332 def VLD3q32odd_UPD : VLD3DWB<0b0101, 0b1000, "32">;
334 // VLD4 : Vector Load (multiple 4-element structures)
335 class VLD4D<bits<4> op11_8, bits<4> op7_4, string Dt>
336 : NLdSt<0, 0b10, op11_8, op7_4,
337 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
338 (ins addrmode6:$addr), IIC_VLD4,
339 "vld4", Dt, "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr", "", []>;
341 def VLD4d8 : VLD4D<0b0000, 0b0000, "8">;
342 def VLD4d16 : VLD4D<0b0000, 0b0100, "16">;
343 def VLD4d32 : VLD4D<0b0000, 0b1000, "32">;
344 def VLD4d64 : NLdSt<0,0b10,0b0010,0b1100,
345 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
346 (ins addrmode6:$addr), IIC_VLD1,
347 "vld1", "64", "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr",
350 // ...with address register writeback:
351 class VLD4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
352 : NLdSt<0, 0b10, op11_8, op7_4,
353 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
354 (ins addrmode6:$addr), IIC_VLD4,
355 "vld4", Dt, "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr",
356 "$addr.addr = $wb", []>;
358 def VLD4d8_UPD : VLD4DWB<0b0000, 0b0000, "8">;
359 def VLD4d16_UPD : VLD4DWB<0b0000, 0b0100, "16">;
360 def VLD4d32_UPD : VLD4DWB<0b0000, 0b1000, "32">;
361 def VLD4d64_UPD : NLdSt<0,0b10,0b0010,0b1100,
362 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4,
364 (ins addrmode6:$addr), IIC_VLD1,
366 "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr",
367 "$addr.addr = $wb", []>;
369 // ...with double-spaced registers (non-updating versions for disassembly only):
370 def VLD4q8 : VLD4D<0b0001, 0b0000, "8">;
371 def VLD4q16 : VLD4D<0b0001, 0b0100, "16">;
372 def VLD4q32 : VLD4D<0b0001, 0b1000, "32">;
373 def VLD4q8_UPD : VLD4DWB<0b0001, 0b0000, "8">;
374 def VLD4q16_UPD : VLD4DWB<0b0001, 0b0100, "16">;
375 def VLD4q32_UPD : VLD4DWB<0b0001, 0b1000, "32">;
377 // ...alternate versions to be allocated odd register numbers:
378 def VLD4q8odd_UPD : VLD4DWB<0b0001, 0b0000, "8">;
379 def VLD4q16odd_UPD : VLD4DWB<0b0001, 0b0100, "16">;
380 def VLD4q32odd_UPD : VLD4DWB<0b0001, 0b1000, "32">;
382 // VLD1LN : Vector Load (single element to one lane)
383 // FIXME: Not yet implemented.
385 // VLD2LN : Vector Load (single 2-element structure to one lane)
386 class VLD2LN<bits<4> op11_8, string Dt>
387 : NLdSt<1, 0b10, op11_8, {?,?,?,?}, (outs DPR:$dst1, DPR:$dst2),
388 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, nohash_imm:$lane),
389 IIC_VLD2, "vld2", Dt, "\\{$dst1[$lane], $dst2[$lane]\\}, $addr",
390 "$src1 = $dst1, $src2 = $dst2", []>;
392 def VLD2LNd8 : VLD2LN<0b0001, "8">;
393 def VLD2LNd16 : VLD2LN<0b0101, "16"> { let Inst{5} = 0; }
394 def VLD2LNd32 : VLD2LN<0b1001, "32"> { let Inst{6} = 0; }
396 // ...with double-spaced registers:
397 def VLD2LNq16 : VLD2LN<0b0101, "16"> { let Inst{5} = 1; }
398 def VLD2LNq32 : VLD2LN<0b1001, "32"> { let Inst{6} = 1; }
400 // ...alternate versions to be allocated odd register numbers:
401 def VLD2LNq16odd : VLD2LN<0b0101, "16"> { let Inst{5} = 1; }
402 def VLD2LNq32odd : VLD2LN<0b1001, "32"> { let Inst{6} = 1; }
404 // VLD3LN : Vector Load (single 3-element structure to one lane)
405 class VLD3LN<bits<4> op11_8, string Dt>
406 : NLdSt<1, 0b10, op11_8, {?,?,?,?}, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3),
407 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3,
408 nohash_imm:$lane), IIC_VLD3, "vld3", Dt,
409 "\\{$dst1[$lane], $dst2[$lane], $dst3[$lane]\\}, $addr",
410 "$src1 = $dst1, $src2 = $dst2, $src3 = $dst3", []>;
412 def VLD3LNd8 : VLD3LN<0b0010, "8"> { let Inst{4} = 0; }
413 def VLD3LNd16 : VLD3LN<0b0110, "16"> { let Inst{5-4} = 0b00; }
414 def VLD3LNd32 : VLD3LN<0b1010, "32"> { let Inst{6-4} = 0b000; }
416 // ...with double-spaced registers:
417 def VLD3LNq16 : VLD3LN<0b0110, "16"> { let Inst{5-4} = 0b10; }
418 def VLD3LNq32 : VLD3LN<0b1010, "32"> { let Inst{6-4} = 0b100; }
420 // ...alternate versions to be allocated odd register numbers:
421 def VLD3LNq16odd : VLD3LN<0b0110, "16"> { let Inst{5-4} = 0b10; }
422 def VLD3LNq32odd : VLD3LN<0b1010, "32"> { let Inst{6-4} = 0b100; }
424 // VLD4LN : Vector Load (single 4-element structure to one lane)
425 class VLD4LN<bits<4> op11_8, string Dt>
426 : NLdSt<1, 0b10, op11_8, {?,?,?,?},
427 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
428 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
429 nohash_imm:$lane), IIC_VLD4, "vld4", Dt,
430 "\\{$dst1[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $addr",
431 "$src1 = $dst1, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4", []>;
433 def VLD4LNd8 : VLD4LN<0b0011, "8">;
434 def VLD4LNd16 : VLD4LN<0b0111, "16"> { let Inst{5} = 0; }
435 def VLD4LNd32 : VLD4LN<0b1011, "32"> { let Inst{6} = 0; }
437 // ...with double-spaced registers:
438 def VLD4LNq16 : VLD4LN<0b0111, "16"> { let Inst{5} = 1; }
439 def VLD4LNq32 : VLD4LN<0b1011, "32"> { let Inst{6} = 1; }
441 // ...alternate versions to be allocated odd register numbers:
442 def VLD4LNq16odd : VLD4LN<0b0111, "16"> { let Inst{5} = 1; }
443 def VLD4LNq32odd : VLD4LN<0b1011, "32"> { let Inst{6} = 1; }
445 // VLD1DUP : Vector Load (single element to all lanes)
446 // VLD2DUP : Vector Load (single 2-element structure to all lanes)
447 // VLD3DUP : Vector Load (single 3-element structure to all lanes)
448 // VLD4DUP : Vector Load (single 4-element structure to all lanes)
449 // FIXME: Not yet implemented.
450 } // mayLoad = 1, hasExtraDefRegAllocReq = 1
452 // VST1 : Vector Store (multiple single elements)
453 class VST1D<bits<4> op7_4, string Dt, ValueType Ty>
454 : NLdSt<0,0b00,0b0111,op7_4, (outs), (ins addrmode6:$addr, DPR:$src), IIC_VST,
455 "vst1", Dt, "\\{$src\\}, $addr", "",
456 [(int_arm_neon_vst1 addrmode6:$addr, (Ty DPR:$src))]>;
457 class VST1Q<bits<4> op7_4, string Dt, ValueType Ty>
458 : NLdSt<0,0b00,0b1010,op7_4, (outs), (ins addrmode6:$addr, QPR:$src), IIC_VST,
459 "vst1", Dt, "${src:dregpair}, $addr", "",
460 [(int_arm_neon_vst1 addrmode6:$addr, (Ty QPR:$src))]>;
462 let hasExtraSrcRegAllocReq = 1 in {
463 def VST1d8 : VST1D<0b0000, "8", v8i8>;
464 def VST1d16 : VST1D<0b0100, "16", v4i16>;
465 def VST1d32 : VST1D<0b1000, "32", v2i32>;
466 def VST1df : VST1D<0b1000, "32", v2f32>;
467 def VST1d64 : VST1D<0b1100, "64", v1i64>;
469 def VST1q8 : VST1Q<0b0000, "8", v16i8>;
470 def VST1q16 : VST1Q<0b0100, "16", v8i16>;
471 def VST1q32 : VST1Q<0b1000, "32", v4i32>;
472 def VST1qf : VST1Q<0b1000, "32", v4f32>;
473 def VST1q64 : VST1Q<0b1100, "64", v2i64>;
474 } // hasExtraSrcRegAllocReq
476 // These (dreg triple/quadruple) are for disassembly only.
477 class VST1D3<bits<4> op7_4, string Dt>
478 : NLdSt<0, 0b00, 0b0110, op7_4, (outs),
479 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3),
480 IIC_VST, "vst1", Dt, "\\{$src1, $src2, $src3\\}, $addr", "",
481 [/* For disassembly only; pattern left blank */]>;
482 class VST1D4<bits<4> op7_4, string Dt>
483 : NLdSt<0, 0b00, 0b0010, op7_4, (outs),
484 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
485 IIC_VST, "vst1", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr", "",
486 [/* For disassembly only; pattern left blank */]>;
488 def VST1d8T : VST1D3<0b0000, "8">;
489 def VST1d16T : VST1D3<0b0100, "16">;
490 def VST1d32T : VST1D3<0b1000, "32">;
491 // VST1d64T : implemented as VST3d64
493 def VST1d8Q : VST1D4<0b0000, "8">;
494 def VST1d16Q : VST1D4<0b0100, "16">;
495 def VST1d32Q : VST1D4<0b1000, "32">;
496 // VST1d64Q : implemented as VST4d64
498 let mayStore = 1, hasExtraSrcRegAllocReq = 1 in {
500 // VST2 : Vector Store (multiple 2-element structures)
501 class VST2D<bits<4> op7_4, string Dt>
502 : NLdSt<0,0b00,0b1000,op7_4, (outs),
503 (ins addrmode6:$addr, DPR:$src1, DPR:$src2), IIC_VST,
504 "vst2", Dt, "\\{$src1, $src2\\}, $addr", "", []>;
505 class VST2Q<bits<4> op7_4, string Dt>
506 : NLdSt<0,0b00,0b0011,op7_4, (outs),
507 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
508 IIC_VST, "vst2", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr",
511 def VST2d8 : VST2D<0b0000, "8">;
512 def VST2d16 : VST2D<0b0100, "16">;
513 def VST2d32 : VST2D<0b1000, "32">;
514 def VST2d64 : NLdSt<0,0b00,0b1010,0b1100, (outs),
515 (ins addrmode6:$addr, DPR:$src1, DPR:$src2), IIC_VST,
516 "vst1", "64", "\\{$src1, $src2\\}, $addr", "", []>;
518 def VST2q8 : VST2Q<0b0000, "8">;
519 def VST2q16 : VST2Q<0b0100, "16">;
520 def VST2q32 : VST2Q<0b1000, "32">;
522 // These (double-spaced dreg pair) are for disassembly only.
523 class VST2Ddbl<bits<4> op7_4, string Dt>
524 : NLdSt<0, 0b00, 0b1001, op7_4, (outs),
525 (ins addrmode6:$addr, DPR:$src1, DPR:$src2), IIC_VST,
526 "vst2", Dt, "\\{$src1, $src2\\}, $addr", "", []>;
528 def VST2d8D : VST2Ddbl<0b0000, "8">;
529 def VST2d16D : VST2Ddbl<0b0100, "16">;
530 def VST2d32D : VST2Ddbl<0b1000, "32">;
532 // VST3 : Vector Store (multiple 3-element structures)
533 class VST3D<bits<4> op7_4, string Dt>
534 : NLdSt<0,0b00,0b0100,op7_4, (outs),
535 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3), IIC_VST,
536 "vst3", Dt, "\\{$src1, $src2, $src3\\}, $addr", "", []>;
537 class VST3WB<bits<4> op7_4, string Dt>
538 : NLdSt<0,0b00,0b0101,op7_4, (outs GPR:$wb),
539 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3), IIC_VST,
540 "vst3", Dt, "\\{$src1, $src2, $src3\\}, $addr",
541 "$addr.addr = $wb", []>;
543 def VST3d8 : VST3D<0b0000, "8">;
544 def VST3d16 : VST3D<0b0100, "16">;
545 def VST3d32 : VST3D<0b1000, "32">;
546 def VST3d64 : NLdSt<0,0b00,0b0110,0b1100, (outs),
547 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3),
549 "vst1", "64", "\\{$src1, $src2, $src3\\}, $addr", "", []>;
551 // vst3 to double-spaced even registers.
552 def VST3q8_UPD : VST3WB<0b0000, "8">;
553 def VST3q16_UPD : VST3WB<0b0100, "16">;
554 def VST3q32_UPD : VST3WB<0b1000, "32">;
556 // vst3 to double-spaced odd registers.
557 def VST3q8odd_UPD : VST3WB<0b0000, "8">;
558 def VST3q16odd_UPD : VST3WB<0b0100, "16">;
559 def VST3q32odd_UPD : VST3WB<0b1000, "32">;
561 // VST4 : Vector Store (multiple 4-element structures)
562 class VST4D<bits<4> op7_4, string Dt>
563 : NLdSt<0,0b00,0b0000,op7_4, (outs),
564 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
565 IIC_VST, "vst4", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr",
567 class VST4WB<bits<4> op7_4, string Dt>
568 : NLdSt<0,0b00,0b0001,op7_4, (outs GPR:$wb),
569 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
570 IIC_VST, "vst4", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr",
571 "$addr.addr = $wb", []>;
573 def VST4d8 : VST4D<0b0000, "8">;
574 def VST4d16 : VST4D<0b0100, "16">;
575 def VST4d32 : VST4D<0b1000, "32">;
576 def VST4d64 : NLdSt<0,0b00,0b0010,0b1100, (outs),
577 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3,
579 "vst1", "64", "\\{$src1, $src2, $src3, $src4\\}, $addr",
582 // vst4 to double-spaced even registers.
583 def VST4q8_UPD : VST4WB<0b0000, "8">;
584 def VST4q16_UPD : VST4WB<0b0100, "16">;
585 def VST4q32_UPD : VST4WB<0b1000, "32">;
587 // vst4 to double-spaced odd registers.
588 def VST4q8odd_UPD : VST4WB<0b0000, "8">;
589 def VST4q16odd_UPD : VST4WB<0b0100, "16">;
590 def VST4q32odd_UPD : VST4WB<0b1000, "32">;
592 // VST1LN : Vector Store (single element from one lane)
593 // FIXME: Not yet implemented.
595 // VST2LN : Vector Store (single 2-element structure from one lane)
596 class VST2LN<bits<4> op11_8, string Dt>
597 : NLdSt<1, 0b00, op11_8, {?,?,?,?}, (outs),
598 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, nohash_imm:$lane),
599 IIC_VST, "vst2", Dt, "\\{$src1[$lane], $src2[$lane]\\}, $addr",
602 def VST2LNd8 : VST2LN<0b0001, "8">;
603 def VST2LNd16 : VST2LN<0b0101, "16"> { let Inst{5} = 0; }
604 def VST2LNd32 : VST2LN<0b1001, "32"> { let Inst{6} = 0; }
606 // ...with double-spaced registers:
607 def VST2LNq16 : VST2LN<0b0101, "16"> { let Inst{5} = 1; }
608 def VST2LNq32 : VST2LN<0b1001, "32"> { let Inst{6} = 1; }
610 // ...alternate versions to be allocated odd register numbers:
611 def VST2LNq16odd : VST2LN<0b0101, "16"> { let Inst{5} = 1; }
612 def VST2LNq32odd : VST2LN<0b1001, "32"> { let Inst{6} = 1; }
614 // VST3LN : Vector Store (single 3-element structure from one lane)
615 class VST3LN<bits<4> op11_8, string Dt>
616 : NLdSt<1, 0b00, op11_8, {?,?,?,?}, (outs),
617 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3,
618 nohash_imm:$lane), IIC_VST, "vst3", Dt,
619 "\\{$src1[$lane], $src2[$lane], $src3[$lane]\\}, $addr", "", []>;
621 def VST3LNd8 : VST3LN<0b0010, "8"> { let Inst{4} = 0; }
622 def VST3LNd16 : VST3LN<0b0110, "16"> { let Inst{5-4} = 0b00; }
623 def VST3LNd32 : VST3LN<0b1010, "32"> { let Inst{6-4} = 0b000; }
625 // ...with double-spaced registers:
626 def VST3LNq16 : VST3LN<0b0110, "16"> { let Inst{5-4} = 0b10; }
627 def VST3LNq32 : VST3LN<0b1010, "32"> { let Inst{6-4} = 0b100; }
629 // ...alternate versions to be allocated odd register numbers:
630 def VST3LNq16odd : VST3LN<0b0110, "16"> { let Inst{5-4} = 0b10; }
631 def VST3LNq32odd : VST3LN<0b1010, "32"> { let Inst{6-4} = 0b100; }
633 // VST4LN : Vector Store (single 4-element structure from one lane)
634 class VST4LN<bits<4> op11_8, string Dt>
635 : NLdSt<1, 0b00, op11_8, {?,?,?,?}, (outs),
636 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
637 nohash_imm:$lane), IIC_VST, "vst4", Dt,
638 "\\{$src1[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $addr",
641 def VST4LNd8 : VST4LN<0b0011, "8">;
642 def VST4LNd16 : VST4LN<0b0111, "16"> { let Inst{5} = 0; }
643 def VST4LNd32 : VST4LN<0b1011, "32"> { let Inst{6} = 0; }
645 // ...with double-spaced registers:
646 def VST4LNq16 : VST4LN<0b0111, "16"> { let Inst{5} = 1; }
647 def VST4LNq32 : VST4LN<0b1011, "32"> { let Inst{6} = 1; }
649 // ...alternate versions to be allocated odd register numbers:
650 def VST4LNq16odd : VST4LN<0b0111, "16"> { let Inst{5} = 1; }
651 def VST4LNq32odd : VST4LN<0b1011, "32"> { let Inst{6} = 1; }
653 } // mayStore = 1, hasExtraSrcRegAllocReq = 1
656 //===----------------------------------------------------------------------===//
657 // NEON pattern fragments
658 //===----------------------------------------------------------------------===//
660 // Extract D sub-registers of Q registers.
661 // (arm_dsubreg_0 is 5; arm_dsubreg_1 is 6)
662 def DSubReg_i8_reg : SDNodeXForm<imm, [{
663 return CurDAG->getTargetConstant(5 + N->getZExtValue() / 8, MVT::i32);
665 def DSubReg_i16_reg : SDNodeXForm<imm, [{
666 return CurDAG->getTargetConstant(5 + N->getZExtValue() / 4, MVT::i32);
668 def DSubReg_i32_reg : SDNodeXForm<imm, [{
669 return CurDAG->getTargetConstant(5 + N->getZExtValue() / 2, MVT::i32);
671 def DSubReg_f64_reg : SDNodeXForm<imm, [{
672 return CurDAG->getTargetConstant(5 + N->getZExtValue(), MVT::i32);
674 def DSubReg_f64_other_reg : SDNodeXForm<imm, [{
675 return CurDAG->getTargetConstant(5 + (1 - N->getZExtValue()), MVT::i32);
678 // Extract S sub-registers of Q/D registers.
679 // (arm_ssubreg_0 is 1; arm_ssubreg_1 is 2; etc.)
680 def SSubReg_f32_reg : SDNodeXForm<imm, [{
681 return CurDAG->getTargetConstant(1 + N->getZExtValue(), MVT::i32);
684 // Translate lane numbers from Q registers to D subregs.
685 def SubReg_i8_lane : SDNodeXForm<imm, [{
686 return CurDAG->getTargetConstant(N->getZExtValue() & 7, MVT::i32);
688 def SubReg_i16_lane : SDNodeXForm<imm, [{
689 return CurDAG->getTargetConstant(N->getZExtValue() & 3, MVT::i32);
691 def SubReg_i32_lane : SDNodeXForm<imm, [{
692 return CurDAG->getTargetConstant(N->getZExtValue() & 1, MVT::i32);
695 //===----------------------------------------------------------------------===//
696 // Instruction Classes
697 //===----------------------------------------------------------------------===//
699 // Basic 2-register operations: single-, double- and quad-register.
700 class N2VS<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
701 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
702 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
703 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
704 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src),
705 IIC_VUNAD, OpcodeStr, Dt, "$dst, $src", "", []>;
706 class N2VD<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
707 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
708 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
709 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
710 (ins DPR:$src), IIC_VUNAD, OpcodeStr, Dt, "$dst, $src", "",
711 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src))))]>;
712 class N2VQ<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
713 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
714 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
715 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
716 (ins QPR:$src), IIC_VUNAQ, OpcodeStr, Dt, "$dst, $src", "",
717 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src))))]>;
719 // Basic 2-register intrinsics, both double- and quad-register.
720 class N2VDInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
721 bits<2> op17_16, bits<5> op11_7, bit op4,
722 InstrItinClass itin, string OpcodeStr, string Dt,
723 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
724 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
725 (ins DPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
726 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
727 class N2VQInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
728 bits<2> op17_16, bits<5> op11_7, bit op4,
729 InstrItinClass itin, string OpcodeStr, string Dt,
730 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
731 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
732 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
733 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
735 // Narrow 2-register intrinsics.
736 class N2VNInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
737 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
738 InstrItinClass itin, string OpcodeStr, string Dt,
739 ValueType TyD, ValueType TyQ, Intrinsic IntOp>
740 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$dst),
741 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
742 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src))))]>;
744 // Long 2-register intrinsics (currently only used for VMOVL).
745 class N2VLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
746 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
747 InstrItinClass itin, string OpcodeStr, string Dt,
748 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
749 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$dst),
750 (ins DPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
751 [(set QPR:$dst, (TyQ (IntOp (TyD DPR:$src))))]>;
753 // 2-register shuffles (VTRN/VZIP/VUZP), both double- and quad-register.
754 class N2VDShuffle<bits<2> op19_18, bits<5> op11_7, string OpcodeStr, string Dt>
755 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 0, 0, (outs DPR:$dst1, DPR:$dst2),
756 (ins DPR:$src1, DPR:$src2), IIC_VPERMD,
757 OpcodeStr, Dt, "$dst1, $dst2",
758 "$src1 = $dst1, $src2 = $dst2", []>;
759 class N2VQShuffle<bits<2> op19_18, bits<5> op11_7,
760 InstrItinClass itin, string OpcodeStr, string Dt>
761 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 1, 0, (outs QPR:$dst1, QPR:$dst2),
762 (ins QPR:$src1, QPR:$src2), itin, OpcodeStr, Dt, "$dst1, $dst2",
763 "$src1 = $dst1, $src2 = $dst2", []>;
765 // Basic 3-register operations: single-, double- and quad-register.
766 class N3VS<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
767 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
768 SDNode OpNode, bit Commutable>
769 : N3V<op24, op23, op21_20, op11_8, 0, op4,
770 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src1, DPR_VFP2:$src2), IIC_VBIND,
771 OpcodeStr, Dt, "$dst, $src1, $src2", "", []> {
772 let isCommutable = Commutable;
775 class N3VD<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
776 InstrItinClass itin, string OpcodeStr, string Dt,
777 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
778 : N3V<op24, op23, op21_20, op11_8, 0, op4,
779 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), itin,
780 OpcodeStr, Dt, "$dst, $src1, $src2", "",
781 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src1), (OpTy DPR:$src2))))]> {
782 let isCommutable = Commutable;
784 // Same as N3VD but no data type.
785 class N3VDX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
786 InstrItinClass itin, string OpcodeStr,
787 ValueType ResTy, ValueType OpTy,
788 SDNode OpNode, bit Commutable>
789 : N3VX<op24, op23, op21_20, op11_8, 0, op4,
790 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), itin,
791 OpcodeStr, "$dst, $src1, $src2", "",
792 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src1), (OpTy DPR:$src2))))]>{
793 let isCommutable = Commutable;
795 class N3VDSL<bits<2> op21_20, bits<4> op11_8,
796 InstrItinClass itin, string OpcodeStr, string Dt,
797 ValueType Ty, SDNode ShOp>
798 : N3V<0, 1, op21_20, op11_8, 1, 0,
799 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
800 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
802 (Ty (ShOp (Ty DPR:$src1),
803 (Ty (NEONvduplane (Ty DPR_VFP2:$src2), imm:$lane)))))]>{
804 let isCommutable = 0;
806 class N3VDSL16<bits<2> op21_20, bits<4> op11_8,
807 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
808 : N3V<0, 1, op21_20, op11_8, 1, 0,
809 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
810 IIC_VMULi16D, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
812 (Ty (ShOp (Ty DPR:$src1),
813 (Ty (NEONvduplane (Ty DPR_8:$src2), imm:$lane)))))]> {
814 let isCommutable = 0;
817 class N3VQ<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
818 InstrItinClass itin, string OpcodeStr, string Dt,
819 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
820 : N3V<op24, op23, op21_20, op11_8, 1, op4,
821 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), itin,
822 OpcodeStr, Dt, "$dst, $src1, $src2", "",
823 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src1), (OpTy QPR:$src2))))]> {
824 let isCommutable = Commutable;
826 class N3VQX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
827 InstrItinClass itin, string OpcodeStr,
828 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
829 : N3VX<op24, op23, op21_20, op11_8, 1, op4,
830 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), itin,
831 OpcodeStr, "$dst, $src1, $src2", "",
832 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src1), (OpTy QPR:$src2))))]>{
833 let isCommutable = Commutable;
835 class N3VQSL<bits<2> op21_20, bits<4> op11_8,
836 InstrItinClass itin, string OpcodeStr, string Dt,
837 ValueType ResTy, ValueType OpTy, SDNode ShOp>
838 : N3V<1, 1, op21_20, op11_8, 1, 0,
839 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
840 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
841 [(set (ResTy QPR:$dst),
842 (ResTy (ShOp (ResTy QPR:$src1),
843 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
845 let isCommutable = 0;
847 class N3VQSL16<bits<2> op21_20, bits<4> op11_8, string OpcodeStr, string Dt,
848 ValueType ResTy, ValueType OpTy, SDNode ShOp>
849 : N3V<1, 1, op21_20, op11_8, 1, 0,
850 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
851 IIC_VMULi16Q, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
852 [(set (ResTy QPR:$dst),
853 (ResTy (ShOp (ResTy QPR:$src1),
854 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
856 let isCommutable = 0;
859 // Basic 3-register intrinsics, both double- and quad-register.
860 class N3VDInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
861 InstrItinClass itin, string OpcodeStr, string Dt,
862 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
863 : N3V<op24, op23, op21_20, op11_8, 0, op4,
864 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), itin,
865 OpcodeStr, Dt, "$dst, $src1, $src2", "",
866 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src1), (OpTy DPR:$src2))))]> {
867 let isCommutable = Commutable;
869 class N3VDIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
870 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
871 : N3V<0, 1, op21_20, op11_8, 1, 0,
872 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
873 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
875 (Ty (IntOp (Ty DPR:$src1),
876 (Ty (NEONvduplane (Ty DPR_VFP2:$src2),
878 let isCommutable = 0;
880 class N3VDIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
881 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
882 : N3V<0, 1, op21_20, op11_8, 1, 0,
883 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
884 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
886 (Ty (IntOp (Ty DPR:$src1),
887 (Ty (NEONvduplane (Ty DPR_8:$src2),
889 let isCommutable = 0;
892 class N3VQInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
893 InstrItinClass itin, string OpcodeStr, string Dt,
894 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
895 : N3V<op24, op23, op21_20, op11_8, 1, op4,
896 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), itin,
897 OpcodeStr, Dt, "$dst, $src1, $src2", "",
898 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src1), (OpTy QPR:$src2))))]> {
899 let isCommutable = Commutable;
901 class N3VQIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
902 string OpcodeStr, string Dt,
903 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
904 : N3V<1, 1, op21_20, op11_8, 1, 0,
905 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
906 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
907 [(set (ResTy QPR:$dst),
908 (ResTy (IntOp (ResTy QPR:$src1),
909 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
911 let isCommutable = 0;
913 class N3VQIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
914 string OpcodeStr, string Dt,
915 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
916 : N3V<1, 1, op21_20, op11_8, 1, 0,
917 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
918 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
919 [(set (ResTy QPR:$dst),
920 (ResTy (IntOp (ResTy QPR:$src1),
921 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
923 let isCommutable = 0;
926 // Multiply-Add/Sub operations: single-, double- and quad-register.
927 class N3VSMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
928 InstrItinClass itin, string OpcodeStr, string Dt,
929 ValueType Ty, SDNode MulOp, SDNode OpNode>
930 : N3V<op24, op23, op21_20, op11_8, 0, op4,
931 (outs DPR_VFP2:$dst),
932 (ins DPR_VFP2:$src1, DPR_VFP2:$src2, DPR_VFP2:$src3), itin,
933 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst", []>;
935 class N3VDMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
936 InstrItinClass itin, string OpcodeStr, string Dt,
937 ValueType Ty, SDNode MulOp, SDNode OpNode>
938 : N3V<op24, op23, op21_20, op11_8, 0, op4,
939 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3), itin,
940 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
941 [(set DPR:$dst, (Ty (OpNode DPR:$src1,
942 (Ty (MulOp DPR:$src2, DPR:$src3)))))]>;
943 class N3VDMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
944 string OpcodeStr, string Dt,
945 ValueType Ty, SDNode MulOp, SDNode ShOp>
946 : N3V<0, 1, op21_20, op11_8, 1, 0,
948 (ins DPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane), itin,
949 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
951 (Ty (ShOp (Ty DPR:$src1),
952 (Ty (MulOp DPR:$src2,
953 (Ty (NEONvduplane (Ty DPR_VFP2:$src3),
955 class N3VDMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
956 string OpcodeStr, string Dt,
957 ValueType Ty, SDNode MulOp, SDNode ShOp>
958 : N3V<0, 1, op21_20, op11_8, 1, 0,
960 (ins DPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane), itin,
961 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
963 (Ty (ShOp (Ty DPR:$src1),
964 (Ty (MulOp DPR:$src2,
965 (Ty (NEONvduplane (Ty DPR_8:$src3),
968 class N3VQMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
969 InstrItinClass itin, string OpcodeStr, string Dt, ValueType Ty,
970 SDNode MulOp, SDNode OpNode>
971 : N3V<op24, op23, op21_20, op11_8, 1, op4,
972 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3), itin,
973 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
974 [(set QPR:$dst, (Ty (OpNode QPR:$src1,
975 (Ty (MulOp QPR:$src2, QPR:$src3)))))]>;
976 class N3VQMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
977 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
978 SDNode MulOp, SDNode ShOp>
979 : N3V<1, 1, op21_20, op11_8, 1, 0,
981 (ins QPR:$src1, QPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane), itin,
982 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
983 [(set (ResTy QPR:$dst),
984 (ResTy (ShOp (ResTy QPR:$src1),
985 (ResTy (MulOp QPR:$src2,
986 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src3),
988 class N3VQMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
989 string OpcodeStr, string Dt,
990 ValueType ResTy, ValueType OpTy,
991 SDNode MulOp, SDNode ShOp>
992 : N3V<1, 1, op21_20, op11_8, 1, 0,
994 (ins QPR:$src1, QPR:$src2, DPR_8:$src3, nohash_imm:$lane), itin,
995 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
996 [(set (ResTy QPR:$dst),
997 (ResTy (ShOp (ResTy QPR:$src1),
998 (ResTy (MulOp QPR:$src2,
999 (ResTy (NEONvduplane (OpTy DPR_8:$src3),
1002 // Neon 3-argument intrinsics, both double- and quad-register.
1003 // The destination register is also used as the first source operand register.
1004 class N3VDInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1005 InstrItinClass itin, string OpcodeStr, string Dt,
1006 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1007 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1008 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3), itin,
1009 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1010 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src1),
1011 (OpTy DPR:$src2), (OpTy DPR:$src3))))]>;
1012 class N3VQInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1013 InstrItinClass itin, string OpcodeStr, string Dt,
1014 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1015 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1016 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3), itin,
1017 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1018 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src1),
1019 (OpTy QPR:$src2), (OpTy QPR:$src3))))]>;
1021 // Neon Long 3-argument intrinsic. The destination register is
1022 // a quad-register and is also used as the first source operand register.
1023 class N3VLInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1024 InstrItinClass itin, string OpcodeStr, string Dt,
1025 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
1026 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1027 (outs QPR:$dst), (ins QPR:$src1, DPR:$src2, DPR:$src3), itin,
1028 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1030 (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$src2), (TyD DPR:$src3))))]>;
1031 class N3VLInt3SL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1032 string OpcodeStr, string Dt,
1033 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1034 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1036 (ins QPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane), itin,
1037 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1038 [(set (ResTy QPR:$dst),
1039 (ResTy (IntOp (ResTy QPR:$src1),
1041 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src3),
1043 class N3VLInt3SL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1044 InstrItinClass itin, string OpcodeStr, string Dt,
1045 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1046 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1048 (ins QPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane), itin,
1049 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1050 [(set (ResTy QPR:$dst),
1051 (ResTy (IntOp (ResTy QPR:$src1),
1053 (OpTy (NEONvduplane (OpTy DPR_8:$src3),
1056 // Narrowing 3-register intrinsics.
1057 class N3VNInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1058 string OpcodeStr, string Dt, ValueType TyD, ValueType TyQ,
1059 Intrinsic IntOp, bit Commutable>
1060 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1061 (outs DPR:$dst), (ins QPR:$src1, QPR:$src2), IIC_VBINi4D,
1062 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1063 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src1), (TyQ QPR:$src2))))]> {
1064 let isCommutable = Commutable;
1067 // Long 3-register intrinsics.
1068 class N3VLInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1069 InstrItinClass itin, string OpcodeStr, string Dt,
1070 ValueType TyQ, ValueType TyD, Intrinsic IntOp, bit Commutable>
1071 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1072 (outs QPR:$dst), (ins DPR:$src1, DPR:$src2), itin,
1073 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1074 [(set QPR:$dst, (TyQ (IntOp (TyD DPR:$src1), (TyD DPR:$src2))))]> {
1075 let isCommutable = Commutable;
1077 class N3VLIntSL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1078 string OpcodeStr, string Dt,
1079 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1080 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1081 (outs QPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1082 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1083 [(set (ResTy QPR:$dst),
1084 (ResTy (IntOp (OpTy DPR:$src1),
1085 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1087 class N3VLIntSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1088 InstrItinClass itin, string OpcodeStr, string Dt,
1089 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1090 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1091 (outs QPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1092 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1093 [(set (ResTy QPR:$dst),
1094 (ResTy (IntOp (OpTy DPR:$src1),
1095 (OpTy (NEONvduplane (OpTy DPR_8:$src2),
1098 // Wide 3-register intrinsics.
1099 class N3VWInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1100 string OpcodeStr, string Dt, ValueType TyQ, ValueType TyD,
1101 Intrinsic IntOp, bit Commutable>
1102 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1103 (outs QPR:$dst), (ins QPR:$src1, DPR:$src2), IIC_VSUBiD,
1104 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1105 [(set QPR:$dst, (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$src2))))]> {
1106 let isCommutable = Commutable;
1109 // Pairwise long 2-register intrinsics, both double- and quad-register.
1110 class N2VDPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1111 bits<2> op17_16, bits<5> op11_7, bit op4,
1112 string OpcodeStr, string Dt,
1113 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1114 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
1115 (ins DPR:$src), IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
1116 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
1117 class N2VQPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1118 bits<2> op17_16, bits<5> op11_7, bit op4,
1119 string OpcodeStr, string Dt,
1120 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1121 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
1122 (ins QPR:$src), IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
1123 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
1125 // Pairwise long 2-register accumulate intrinsics,
1126 // both double- and quad-register.
1127 // The destination register is also used as the first source operand register.
1128 class N2VDPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1129 bits<2> op17_16, bits<5> op11_7, bit op4,
1130 string OpcodeStr, string Dt,
1131 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1132 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
1133 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), IIC_VPALiD,
1134 OpcodeStr, Dt, "$dst, $src2", "$src1 = $dst",
1135 [(set DPR:$dst, (ResTy (IntOp (ResTy DPR:$src1), (OpTy DPR:$src2))))]>;
1136 class N2VQPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1137 bits<2> op17_16, bits<5> op11_7, bit op4,
1138 string OpcodeStr, string Dt,
1139 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1140 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4,
1141 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), IIC_VPALiQ,
1142 OpcodeStr, Dt, "$dst, $src2", "$src1 = $dst",
1143 [(set QPR:$dst, (ResTy (IntOp (ResTy QPR:$src1), (OpTy QPR:$src2))))]>;
1145 // Shift by immediate,
1146 // both double- and quad-register.
1147 class N2VDSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1148 InstrItinClass itin, string OpcodeStr, string Dt,
1149 ValueType Ty, SDNode OpNode>
1150 : N2VImm<op24, op23, op11_8, op7, 0, op4,
1151 (outs DPR:$dst), (ins DPR:$src, i32imm:$SIMM), itin,
1152 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1153 [(set DPR:$dst, (Ty (OpNode (Ty DPR:$src), (i32 imm:$SIMM))))]>;
1154 class N2VQSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1155 InstrItinClass itin, string OpcodeStr, string Dt,
1156 ValueType Ty, SDNode OpNode>
1157 : N2VImm<op24, op23, op11_8, op7, 1, op4,
1158 (outs QPR:$dst), (ins QPR:$src, i32imm:$SIMM), itin,
1159 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1160 [(set QPR:$dst, (Ty (OpNode (Ty QPR:$src), (i32 imm:$SIMM))))]>;
1162 // Long shift by immediate.
1163 class N2VLSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
1164 string OpcodeStr, string Dt,
1165 ValueType ResTy, ValueType OpTy, SDNode OpNode>
1166 : N2VImm<op24, op23, op11_8, op7, op6, op4,
1167 (outs QPR:$dst), (ins DPR:$src, i32imm:$SIMM), IIC_VSHLiD,
1168 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1169 [(set QPR:$dst, (ResTy (OpNode (OpTy DPR:$src),
1170 (i32 imm:$SIMM))))]>;
1172 // Narrow shift by immediate.
1173 class N2VNSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
1174 InstrItinClass itin, string OpcodeStr, string Dt,
1175 ValueType ResTy, ValueType OpTy, SDNode OpNode>
1176 : N2VImm<op24, op23, op11_8, op7, op6, op4,
1177 (outs DPR:$dst), (ins QPR:$src, i32imm:$SIMM), itin,
1178 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1179 [(set DPR:$dst, (ResTy (OpNode (OpTy QPR:$src),
1180 (i32 imm:$SIMM))))]>;
1182 // Shift right by immediate and accumulate,
1183 // both double- and quad-register.
1184 class N2VDShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1185 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1186 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$dst),
1187 (ins DPR:$src1, DPR:$src2, i32imm:$SIMM), IIC_VPALiD,
1188 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1189 [(set DPR:$dst, (Ty (add DPR:$src1,
1190 (Ty (ShOp DPR:$src2, (i32 imm:$SIMM))))))]>;
1191 class N2VQShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1192 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1193 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$dst),
1194 (ins QPR:$src1, QPR:$src2, i32imm:$SIMM), IIC_VPALiD,
1195 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1196 [(set QPR:$dst, (Ty (add QPR:$src1,
1197 (Ty (ShOp QPR:$src2, (i32 imm:$SIMM))))))]>;
1199 // Shift by immediate and insert,
1200 // both double- and quad-register.
1201 class N2VDShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1202 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1203 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$dst),
1204 (ins DPR:$src1, DPR:$src2, i32imm:$SIMM), IIC_VSHLiD,
1205 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1206 [(set DPR:$dst, (Ty (ShOp DPR:$src1, DPR:$src2, (i32 imm:$SIMM))))]>;
1207 class N2VQShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1208 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1209 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$dst),
1210 (ins QPR:$src1, QPR:$src2, i32imm:$SIMM), IIC_VSHLiQ,
1211 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1212 [(set QPR:$dst, (Ty (ShOp QPR:$src1, QPR:$src2, (i32 imm:$SIMM))))]>;
1214 // Convert, with fractional bits immediate,
1215 // both double- and quad-register.
1216 class N2VCvtD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1217 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1219 : N2VImm<op24, op23, op11_8, op7, 0, op4,
1220 (outs DPR:$dst), (ins DPR:$src, i32imm:$SIMM), IIC_VUNAD,
1221 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1222 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src), (i32 imm:$SIMM))))]>;
1223 class N2VCvtQ<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1224 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1226 : N2VImm<op24, op23, op11_8, op7, 1, op4,
1227 (outs QPR:$dst), (ins QPR:$src, i32imm:$SIMM), IIC_VUNAQ,
1228 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1229 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src), (i32 imm:$SIMM))))]>;
1231 //===----------------------------------------------------------------------===//
1233 //===----------------------------------------------------------------------===//
1235 // Abbreviations used in multiclass suffixes:
1236 // Q = quarter int (8 bit) elements
1237 // H = half int (16 bit) elements
1238 // S = single int (32 bit) elements
1239 // D = double int (64 bit) elements
1241 // Neon 2-register vector operations -- for disassembly only.
1243 // First with only element sizes of 8, 16 and 32 bits:
1244 multiclass N2V_QHS_cmp<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1245 bits<5> op11_7, bit op4, string opc, string Dt,
1247 // 64-bit vector types.
1248 def v8i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 0, op4,
1249 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1250 opc, !strconcat(Dt, "8"), asm, "", []>;
1251 def v4i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 0, op4,
1252 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1253 opc, !strconcat(Dt, "16"), asm, "", []>;
1254 def v2i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
1255 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1256 opc, !strconcat(Dt, "32"), asm, "", []>;
1257 def v2f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
1258 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1259 opc, "f32", asm, "", []> {
1260 let Inst{10} = 1; // overwrite F = 1
1263 // 128-bit vector types.
1264 def v16i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 1, op4,
1265 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1266 opc, !strconcat(Dt, "8"), asm, "", []>;
1267 def v8i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 1, op4,
1268 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1269 opc, !strconcat(Dt, "16"), asm, "", []>;
1270 def v4i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
1271 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1272 opc, !strconcat(Dt, "32"), asm, "", []>;
1273 def v4f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
1274 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1275 opc, "f32", asm, "", []> {
1276 let Inst{10} = 1; // overwrite F = 1
1280 // Neon 3-register vector operations.
1282 // First with only element sizes of 8, 16 and 32 bits:
1283 multiclass N3V_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1284 InstrItinClass itinD16, InstrItinClass itinD32,
1285 InstrItinClass itinQ16, InstrItinClass itinQ32,
1286 string OpcodeStr, string Dt,
1287 SDNode OpNode, bit Commutable = 0> {
1288 // 64-bit vector types.
1289 def v8i8 : N3VD<op24, op23, 0b00, op11_8, op4, itinD16,
1290 OpcodeStr, !strconcat(Dt, "8"),
1291 v8i8, v8i8, OpNode, Commutable>;
1292 def v4i16 : N3VD<op24, op23, 0b01, op11_8, op4, itinD16,
1293 OpcodeStr, !strconcat(Dt, "16"),
1294 v4i16, v4i16, OpNode, Commutable>;
1295 def v2i32 : N3VD<op24, op23, 0b10, op11_8, op4, itinD32,
1296 OpcodeStr, !strconcat(Dt, "32"),
1297 v2i32, v2i32, OpNode, Commutable>;
1299 // 128-bit vector types.
1300 def v16i8 : N3VQ<op24, op23, 0b00, op11_8, op4, itinQ16,
1301 OpcodeStr, !strconcat(Dt, "8"),
1302 v16i8, v16i8, OpNode, Commutable>;
1303 def v8i16 : N3VQ<op24, op23, 0b01, op11_8, op4, itinQ16,
1304 OpcodeStr, !strconcat(Dt, "16"),
1305 v8i16, v8i16, OpNode, Commutable>;
1306 def v4i32 : N3VQ<op24, op23, 0b10, op11_8, op4, itinQ32,
1307 OpcodeStr, !strconcat(Dt, "32"),
1308 v4i32, v4i32, OpNode, Commutable>;
1311 multiclass N3VSL_HS<bits<4> op11_8, string OpcodeStr, string Dt, SDNode ShOp> {
1312 def v4i16 : N3VDSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
1314 def v2i32 : N3VDSL<0b10, op11_8, IIC_VMULi32D, OpcodeStr, !strconcat(Dt,"32"),
1316 def v8i16 : N3VQSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
1317 v8i16, v4i16, ShOp>;
1318 def v4i32 : N3VQSL<0b10, op11_8, IIC_VMULi32Q, OpcodeStr, !strconcat(Dt,"32"),
1319 v4i32, v2i32, ShOp>;
1322 // ....then also with element size 64 bits:
1323 multiclass N3V_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1324 InstrItinClass itinD, InstrItinClass itinQ,
1325 string OpcodeStr, string Dt,
1326 SDNode OpNode, bit Commutable = 0>
1327 : N3V_QHS<op24, op23, op11_8, op4, itinD, itinD, itinQ, itinQ,
1328 OpcodeStr, Dt, OpNode, Commutable> {
1329 def v1i64 : N3VD<op24, op23, 0b11, op11_8, op4, itinD,
1330 OpcodeStr, !strconcat(Dt, "64"),
1331 v1i64, v1i64, OpNode, Commutable>;
1332 def v2i64 : N3VQ<op24, op23, 0b11, op11_8, op4, itinQ,
1333 OpcodeStr, !strconcat(Dt, "64"),
1334 v2i64, v2i64, OpNode, Commutable>;
1338 // Neon Narrowing 2-register vector intrinsics,
1339 // source operand element sizes of 16, 32 and 64 bits:
1340 multiclass N2VNInt_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1341 bits<5> op11_7, bit op6, bit op4,
1342 InstrItinClass itin, string OpcodeStr, string Dt,
1344 def v8i8 : N2VNInt<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
1345 itin, OpcodeStr, !strconcat(Dt, "16"),
1346 v8i8, v8i16, IntOp>;
1347 def v4i16 : N2VNInt<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
1348 itin, OpcodeStr, !strconcat(Dt, "32"),
1349 v4i16, v4i32, IntOp>;
1350 def v2i32 : N2VNInt<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
1351 itin, OpcodeStr, !strconcat(Dt, "64"),
1352 v2i32, v2i64, IntOp>;
1356 // Neon Lengthening 2-register vector intrinsic (currently specific to VMOVL).
1357 // source operand element sizes of 16, 32 and 64 bits:
1358 multiclass N2VLInt_QHS<bits<2> op24_23, bits<5> op11_7, bit op6, bit op4,
1359 string OpcodeStr, string Dt, Intrinsic IntOp> {
1360 def v8i16 : N2VLInt<op24_23, 0b00, 0b10, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
1361 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
1362 def v4i32 : N2VLInt<op24_23, 0b01, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
1363 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
1364 def v2i64 : N2VLInt<op24_23, 0b10, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
1365 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1369 // Neon 3-register vector intrinsics.
1371 // First with only element sizes of 16 and 32 bits:
1372 multiclass N3VInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
1373 InstrItinClass itinD16, InstrItinClass itinD32,
1374 InstrItinClass itinQ16, InstrItinClass itinQ32,
1375 string OpcodeStr, string Dt,
1376 Intrinsic IntOp, bit Commutable = 0> {
1377 // 64-bit vector types.
1378 def v4i16 : N3VDInt<op24, op23, 0b01, op11_8, op4, itinD16,
1379 OpcodeStr, !strconcat(Dt, "16"),
1380 v4i16, v4i16, IntOp, Commutable>;
1381 def v2i32 : N3VDInt<op24, op23, 0b10, op11_8, op4, itinD32,
1382 OpcodeStr, !strconcat(Dt, "32"),
1383 v2i32, v2i32, IntOp, Commutable>;
1385 // 128-bit vector types.
1386 def v8i16 : N3VQInt<op24, op23, 0b01, op11_8, op4, itinQ16,
1387 OpcodeStr, !strconcat(Dt, "16"),
1388 v8i16, v8i16, IntOp, Commutable>;
1389 def v4i32 : N3VQInt<op24, op23, 0b10, op11_8, op4, itinQ32,
1390 OpcodeStr, !strconcat(Dt, "32"),
1391 v4i32, v4i32, IntOp, Commutable>;
1394 multiclass N3VIntSL_HS<bits<4> op11_8,
1395 InstrItinClass itinD16, InstrItinClass itinD32,
1396 InstrItinClass itinQ16, InstrItinClass itinQ32,
1397 string OpcodeStr, string Dt, Intrinsic IntOp> {
1398 def v4i16 : N3VDIntSL16<0b01, op11_8, itinD16,
1399 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp>;
1400 def v2i32 : N3VDIntSL<0b10, op11_8, itinD32,
1401 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp>;
1402 def v8i16 : N3VQIntSL16<0b01, op11_8, itinQ16,
1403 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16, IntOp>;
1404 def v4i32 : N3VQIntSL<0b10, op11_8, itinQ32,
1405 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32, IntOp>;
1408 // ....then also with element size of 8 bits:
1409 multiclass N3VInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1410 InstrItinClass itinD16, InstrItinClass itinD32,
1411 InstrItinClass itinQ16, InstrItinClass itinQ32,
1412 string OpcodeStr, string Dt,
1413 Intrinsic IntOp, bit Commutable = 0>
1414 : N3VInt_HS<op24, op23, op11_8, op4, itinD16, itinD32, itinQ16, itinQ32,
1415 OpcodeStr, Dt, IntOp, Commutable> {
1416 def v8i8 : N3VDInt<op24, op23, 0b00, op11_8, op4, itinD16,
1417 OpcodeStr, !strconcat(Dt, "8"),
1418 v8i8, v8i8, IntOp, Commutable>;
1419 def v16i8 : N3VQInt<op24, op23, 0b00, op11_8, op4, itinQ16,
1420 OpcodeStr, !strconcat(Dt, "8"),
1421 v16i8, v16i8, IntOp, Commutable>;
1424 // ....then also with element size of 64 bits:
1425 multiclass N3VInt_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1426 InstrItinClass itinD16, InstrItinClass itinD32,
1427 InstrItinClass itinQ16, InstrItinClass itinQ32,
1428 string OpcodeStr, string Dt,
1429 Intrinsic IntOp, bit Commutable = 0>
1430 : N3VInt_QHS<op24, op23, op11_8, op4, itinD16, itinD32, itinQ16, itinQ32,
1431 OpcodeStr, Dt, IntOp, Commutable> {
1432 def v1i64 : N3VDInt<op24, op23, 0b11, op11_8, op4, itinD32,
1433 OpcodeStr, !strconcat(Dt, "64"),
1434 v1i64, v1i64, IntOp, Commutable>;
1435 def v2i64 : N3VQInt<op24, op23, 0b11, op11_8, op4, itinQ32,
1436 OpcodeStr, !strconcat(Dt, "64"),
1437 v2i64, v2i64, IntOp, Commutable>;
1441 // Neon Narrowing 3-register vector intrinsics,
1442 // source operand element sizes of 16, 32 and 64 bits:
1443 multiclass N3VNInt_HSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1444 string OpcodeStr, string Dt,
1445 Intrinsic IntOp, bit Commutable = 0> {
1446 def v8i8 : N3VNInt<op24, op23, 0b00, op11_8, op4,
1447 OpcodeStr, !strconcat(Dt, "16"),
1448 v8i8, v8i16, IntOp, Commutable>;
1449 def v4i16 : N3VNInt<op24, op23, 0b01, op11_8, op4,
1450 OpcodeStr, !strconcat(Dt, "32"),
1451 v4i16, v4i32, IntOp, Commutable>;
1452 def v2i32 : N3VNInt<op24, op23, 0b10, op11_8, op4,
1453 OpcodeStr, !strconcat(Dt, "64"),
1454 v2i32, v2i64, IntOp, Commutable>;
1458 // Neon Long 3-register vector intrinsics.
1460 // First with only element sizes of 16 and 32 bits:
1461 multiclass N3VLInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
1462 InstrItinClass itin, string OpcodeStr, string Dt,
1463 Intrinsic IntOp, bit Commutable = 0> {
1464 def v4i32 : N3VLInt<op24, op23, 0b01, op11_8, op4, itin,
1465 OpcodeStr, !strconcat(Dt, "16"),
1466 v4i32, v4i16, IntOp, Commutable>;
1467 def v2i64 : N3VLInt<op24, op23, 0b10, op11_8, op4, itin,
1468 OpcodeStr, !strconcat(Dt, "32"),
1469 v2i64, v2i32, IntOp, Commutable>;
1472 multiclass N3VLIntSL_HS<bit op24, bits<4> op11_8,
1473 InstrItinClass itin, string OpcodeStr, string Dt,
1475 def v4i16 : N3VLIntSL16<op24, 0b01, op11_8, itin,
1476 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
1477 def v2i32 : N3VLIntSL<op24, 0b10, op11_8, itin,
1478 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1481 // ....then also with element size of 8 bits:
1482 multiclass N3VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1483 InstrItinClass itin, string OpcodeStr, string Dt,
1484 Intrinsic IntOp, bit Commutable = 0>
1485 : N3VLInt_HS<op24, op23, op11_8, op4, itin, OpcodeStr, Dt,
1486 IntOp, Commutable> {
1487 def v8i16 : N3VLInt<op24, op23, 0b00, op11_8, op4, itin,
1488 OpcodeStr, !strconcat(Dt, "8"),
1489 v8i16, v8i8, IntOp, Commutable>;
1493 // Neon Wide 3-register vector intrinsics,
1494 // source operand element sizes of 8, 16 and 32 bits:
1495 multiclass N3VWInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1496 string OpcodeStr, string Dt,
1497 Intrinsic IntOp, bit Commutable = 0> {
1498 def v8i16 : N3VWInt<op24, op23, 0b00, op11_8, op4,
1499 OpcodeStr, !strconcat(Dt, "8"),
1500 v8i16, v8i8, IntOp, Commutable>;
1501 def v4i32 : N3VWInt<op24, op23, 0b01, op11_8, op4,
1502 OpcodeStr, !strconcat(Dt, "16"),
1503 v4i32, v4i16, IntOp, Commutable>;
1504 def v2i64 : N3VWInt<op24, op23, 0b10, op11_8, op4,
1505 OpcodeStr, !strconcat(Dt, "32"),
1506 v2i64, v2i32, IntOp, Commutable>;
1510 // Neon Multiply-Op vector operations,
1511 // element sizes of 8, 16 and 32 bits:
1512 multiclass N3VMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1513 InstrItinClass itinD16, InstrItinClass itinD32,
1514 InstrItinClass itinQ16, InstrItinClass itinQ32,
1515 string OpcodeStr, string Dt, SDNode OpNode> {
1516 // 64-bit vector types.
1517 def v8i8 : N3VDMulOp<op24, op23, 0b00, op11_8, op4, itinD16,
1518 OpcodeStr, !strconcat(Dt, "8"), v8i8, mul, OpNode>;
1519 def v4i16 : N3VDMulOp<op24, op23, 0b01, op11_8, op4, itinD16,
1520 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, OpNode>;
1521 def v2i32 : N3VDMulOp<op24, op23, 0b10, op11_8, op4, itinD32,
1522 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, OpNode>;
1524 // 128-bit vector types.
1525 def v16i8 : N3VQMulOp<op24, op23, 0b00, op11_8, op4, itinQ16,
1526 OpcodeStr, !strconcat(Dt, "8"), v16i8, mul, OpNode>;
1527 def v8i16 : N3VQMulOp<op24, op23, 0b01, op11_8, op4, itinQ16,
1528 OpcodeStr, !strconcat(Dt, "16"), v8i16, mul, OpNode>;
1529 def v4i32 : N3VQMulOp<op24, op23, 0b10, op11_8, op4, itinQ32,
1530 OpcodeStr, !strconcat(Dt, "32"), v4i32, mul, OpNode>;
1533 multiclass N3VMulOpSL_HS<bits<4> op11_8,
1534 InstrItinClass itinD16, InstrItinClass itinD32,
1535 InstrItinClass itinQ16, InstrItinClass itinQ32,
1536 string OpcodeStr, string Dt, SDNode ShOp> {
1537 def v4i16 : N3VDMulOpSL16<0b01, op11_8, itinD16,
1538 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, ShOp>;
1539 def v2i32 : N3VDMulOpSL<0b10, op11_8, itinD32,
1540 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, ShOp>;
1541 def v8i16 : N3VQMulOpSL16<0b01, op11_8, itinQ16,
1542 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16,
1544 def v4i32 : N3VQMulOpSL<0b10, op11_8, itinQ32,
1545 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32,
1549 // Neon 3-argument intrinsics,
1550 // element sizes of 8, 16 and 32 bits:
1551 multiclass N3VInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1552 string OpcodeStr, string Dt, Intrinsic IntOp> {
1553 // 64-bit vector types.
1554 def v8i8 : N3VDInt3<op24, op23, 0b00, op11_8, op4, IIC_VMACi16D,
1555 OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
1556 def v4i16 : N3VDInt3<op24, op23, 0b01, op11_8, op4, IIC_VMACi16D,
1557 OpcodeStr, !strconcat(Dt, "16"), v4i16, v4i16, IntOp>;
1558 def v2i32 : N3VDInt3<op24, op23, 0b10, op11_8, op4, IIC_VMACi32D,
1559 OpcodeStr, !strconcat(Dt, "32"), v2i32, v2i32, IntOp>;
1561 // 128-bit vector types.
1562 def v16i8 : N3VQInt3<op24, op23, 0b00, op11_8, op4, IIC_VMACi16Q,
1563 OpcodeStr, !strconcat(Dt, "8"), v16i8, v16i8, IntOp>;
1564 def v8i16 : N3VQInt3<op24, op23, 0b01, op11_8, op4, IIC_VMACi16Q,
1565 OpcodeStr, !strconcat(Dt, "16"), v8i16, v8i16, IntOp>;
1566 def v4i32 : N3VQInt3<op24, op23, 0b10, op11_8, op4, IIC_VMACi32Q,
1567 OpcodeStr, !strconcat(Dt, "32"), v4i32, v4i32, IntOp>;
1571 // Neon Long 3-argument intrinsics.
1573 // First with only element sizes of 16 and 32 bits:
1574 multiclass N3VLInt3_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
1575 string OpcodeStr, string Dt, Intrinsic IntOp> {
1576 def v4i32 : N3VLInt3<op24, op23, 0b01, op11_8, op4, IIC_VMACi16D,
1577 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
1578 def v2i64 : N3VLInt3<op24, op23, 0b10, op11_8, op4, IIC_VMACi16D,
1579 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1582 multiclass N3VLInt3SL_HS<bit op24, bits<4> op11_8,
1583 string OpcodeStr, string Dt, Intrinsic IntOp> {
1584 def v4i16 : N3VLInt3SL16<op24, 0b01, op11_8, IIC_VMACi16D,
1585 OpcodeStr, !strconcat(Dt,"16"), v4i32, v4i16, IntOp>;
1586 def v2i32 : N3VLInt3SL<op24, 0b10, op11_8, IIC_VMACi32D,
1587 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1590 // ....then also with element size of 8 bits:
1591 multiclass N3VLInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1592 string OpcodeStr, string Dt, Intrinsic IntOp>
1593 : N3VLInt3_HS<op24, op23, op11_8, op4, OpcodeStr, Dt, IntOp> {
1594 def v8i16 : N3VLInt3<op24, op23, 0b00, op11_8, op4, IIC_VMACi16D,
1595 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
1599 // Neon 2-register vector intrinsics,
1600 // element sizes of 8, 16 and 32 bits:
1601 multiclass N2VInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1602 bits<5> op11_7, bit op4,
1603 InstrItinClass itinD, InstrItinClass itinQ,
1604 string OpcodeStr, string Dt, Intrinsic IntOp> {
1605 // 64-bit vector types.
1606 def v8i8 : N2VDInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1607 itinD, OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
1608 def v4i16 : N2VDInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1609 itinD, OpcodeStr, !strconcat(Dt, "16"),v4i16,v4i16,IntOp>;
1610 def v2i32 : N2VDInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1611 itinD, OpcodeStr, !strconcat(Dt, "32"),v2i32,v2i32,IntOp>;
1613 // 128-bit vector types.
1614 def v16i8 : N2VQInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1615 itinQ, OpcodeStr, !strconcat(Dt, "8"), v16i8,v16i8,IntOp>;
1616 def v8i16 : N2VQInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1617 itinQ, OpcodeStr, !strconcat(Dt, "16"),v8i16,v8i16,IntOp>;
1618 def v4i32 : N2VQInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1619 itinQ, OpcodeStr, !strconcat(Dt, "32"),v4i32,v4i32,IntOp>;
1623 // Neon Pairwise long 2-register intrinsics,
1624 // element sizes of 8, 16 and 32 bits:
1625 multiclass N2VPLInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1626 bits<5> op11_7, bit op4,
1627 string OpcodeStr, string Dt, Intrinsic IntOp> {
1628 // 64-bit vector types.
1629 def v8i8 : N2VDPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1630 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
1631 def v4i16 : N2VDPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1632 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
1633 def v2i32 : N2VDPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1634 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
1636 // 128-bit vector types.
1637 def v16i8 : N2VQPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1638 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
1639 def v8i16 : N2VQPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1640 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
1641 def v4i32 : N2VQPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1642 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
1646 // Neon Pairwise long 2-register accumulate intrinsics,
1647 // element sizes of 8, 16 and 32 bits:
1648 multiclass N2VPLInt2_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1649 bits<5> op11_7, bit op4,
1650 string OpcodeStr, string Dt, Intrinsic IntOp> {
1651 // 64-bit vector types.
1652 def v8i8 : N2VDPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1653 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
1654 def v4i16 : N2VDPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1655 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
1656 def v2i32 : N2VDPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1657 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
1659 // 128-bit vector types.
1660 def v16i8 : N2VQPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1661 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
1662 def v8i16 : N2VQPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1663 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
1664 def v4i32 : N2VQPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1665 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
1669 // Neon 2-register vector shift by immediate,
1670 // element sizes of 8, 16, 32 and 64 bits:
1671 multiclass N2VSh_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1672 InstrItinClass itin, string OpcodeStr, string Dt,
1674 // 64-bit vector types.
1675 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, itin,
1676 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
1677 let Inst{21-19} = 0b001; // imm6 = 001xxx
1679 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, itin,
1680 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
1681 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1683 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, itin,
1684 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
1685 let Inst{21} = 0b1; // imm6 = 1xxxxx
1687 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, itin,
1688 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
1691 // 128-bit vector types.
1692 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, itin,
1693 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
1694 let Inst{21-19} = 0b001; // imm6 = 001xxx
1696 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, itin,
1697 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
1698 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1700 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, itin,
1701 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
1702 let Inst{21} = 0b1; // imm6 = 1xxxxx
1704 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, itin,
1705 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
1710 // Neon Shift-Accumulate vector operations,
1711 // element sizes of 8, 16, 32 and 64 bits:
1712 multiclass N2VShAdd_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1713 string OpcodeStr, string Dt, SDNode ShOp> {
1714 // 64-bit vector types.
1715 def v8i8 : N2VDShAdd<op24, op23, op11_8, 0, op4,
1716 OpcodeStr, !strconcat(Dt, "8"), v8i8, ShOp> {
1717 let Inst{21-19} = 0b001; // imm6 = 001xxx
1719 def v4i16 : N2VDShAdd<op24, op23, op11_8, 0, op4,
1720 OpcodeStr, !strconcat(Dt, "16"), v4i16, ShOp> {
1721 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1723 def v2i32 : N2VDShAdd<op24, op23, op11_8, 0, op4,
1724 OpcodeStr, !strconcat(Dt, "32"), v2i32, ShOp> {
1725 let Inst{21} = 0b1; // imm6 = 1xxxxx
1727 def v1i64 : N2VDShAdd<op24, op23, op11_8, 1, op4,
1728 OpcodeStr, !strconcat(Dt, "64"), v1i64, ShOp>;
1731 // 128-bit vector types.
1732 def v16i8 : N2VQShAdd<op24, op23, op11_8, 0, op4,
1733 OpcodeStr, !strconcat(Dt, "8"), v16i8, ShOp> {
1734 let Inst{21-19} = 0b001; // imm6 = 001xxx
1736 def v8i16 : N2VQShAdd<op24, op23, op11_8, 0, op4,
1737 OpcodeStr, !strconcat(Dt, "16"), v8i16, ShOp> {
1738 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1740 def v4i32 : N2VQShAdd<op24, op23, op11_8, 0, op4,
1741 OpcodeStr, !strconcat(Dt, "32"), v4i32, ShOp> {
1742 let Inst{21} = 0b1; // imm6 = 1xxxxx
1744 def v2i64 : N2VQShAdd<op24, op23, op11_8, 1, op4,
1745 OpcodeStr, !strconcat(Dt, "64"), v2i64, ShOp>;
1750 // Neon Shift-Insert vector operations,
1751 // element sizes of 8, 16, 32 and 64 bits:
1752 multiclass N2VShIns_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1753 string OpcodeStr, SDNode ShOp> {
1754 // 64-bit vector types.
1755 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4,
1756 OpcodeStr, "8", v8i8, ShOp> {
1757 let Inst{21-19} = 0b001; // imm6 = 001xxx
1759 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4,
1760 OpcodeStr, "16", v4i16, ShOp> {
1761 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1763 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4,
1764 OpcodeStr, "32", v2i32, ShOp> {
1765 let Inst{21} = 0b1; // imm6 = 1xxxxx
1767 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4,
1768 OpcodeStr, "64", v1i64, ShOp>;
1771 // 128-bit vector types.
1772 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4,
1773 OpcodeStr, "8", v16i8, ShOp> {
1774 let Inst{21-19} = 0b001; // imm6 = 001xxx
1776 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4,
1777 OpcodeStr, "16", v8i16, ShOp> {
1778 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1780 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4,
1781 OpcodeStr, "32", v4i32, ShOp> {
1782 let Inst{21} = 0b1; // imm6 = 1xxxxx
1784 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4,
1785 OpcodeStr, "64", v2i64, ShOp>;
1789 // Neon Shift Long operations,
1790 // element sizes of 8, 16, 32 bits:
1791 multiclass N2VLSh_QHS<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
1792 bit op4, string OpcodeStr, string Dt, SDNode OpNode> {
1793 def v8i16 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
1794 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode> {
1795 let Inst{21-19} = 0b001; // imm6 = 001xxx
1797 def v4i32 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
1798 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode> {
1799 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1801 def v2i64 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
1802 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode> {
1803 let Inst{21} = 0b1; // imm6 = 1xxxxx
1807 // Neon Shift Narrow operations,
1808 // element sizes of 16, 32, 64 bits:
1809 multiclass N2VNSh_HSD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
1810 bit op4, InstrItinClass itin, string OpcodeStr, string Dt,
1812 def v8i8 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
1813 OpcodeStr, !strconcat(Dt, "16"), v8i8, v8i16, OpNode> {
1814 let Inst{21-19} = 0b001; // imm6 = 001xxx
1816 def v4i16 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
1817 OpcodeStr, !strconcat(Dt, "32"), v4i16, v4i32, OpNode> {
1818 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1820 def v2i32 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
1821 OpcodeStr, !strconcat(Dt, "64"), v2i32, v2i64, OpNode> {
1822 let Inst{21} = 0b1; // imm6 = 1xxxxx
1826 //===----------------------------------------------------------------------===//
1827 // Instruction Definitions.
1828 //===----------------------------------------------------------------------===//
1830 // Vector Add Operations.
1832 // VADD : Vector Add (integer and floating-point)
1833 defm VADD : N3V_QHSD<0, 0, 0b1000, 0, IIC_VBINiD, IIC_VBINiQ, "vadd", "i",
1835 def VADDfd : N3VD<0, 0, 0b00, 0b1101, 0, IIC_VBIND, "vadd", "f32",
1836 v2f32, v2f32, fadd, 1>;
1837 def VADDfq : N3VQ<0, 0, 0b00, 0b1101, 0, IIC_VBINQ, "vadd", "f32",
1838 v4f32, v4f32, fadd, 1>;
1839 // VADDL : Vector Add Long (Q = D + D)
1840 defm VADDLs : N3VLInt_QHS<0,1,0b0000,0, IIC_VSHLiD, "vaddl", "s",
1841 int_arm_neon_vaddls, 1>;
1842 defm VADDLu : N3VLInt_QHS<1,1,0b0000,0, IIC_VSHLiD, "vaddl", "u",
1843 int_arm_neon_vaddlu, 1>;
1844 // VADDW : Vector Add Wide (Q = Q + D)
1845 defm VADDWs : N3VWInt_QHS<0,1,0b0001,0, "vaddw", "s", int_arm_neon_vaddws, 0>;
1846 defm VADDWu : N3VWInt_QHS<1,1,0b0001,0, "vaddw", "u", int_arm_neon_vaddwu, 0>;
1847 // VHADD : Vector Halving Add
1848 defm VHADDs : N3VInt_QHS<0,0,0b0000,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1849 IIC_VBINi4Q, "vhadd", "s", int_arm_neon_vhadds, 1>;
1850 defm VHADDu : N3VInt_QHS<1,0,0b0000,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1851 IIC_VBINi4Q, "vhadd", "u", int_arm_neon_vhaddu, 1>;
1852 // VRHADD : Vector Rounding Halving Add
1853 defm VRHADDs : N3VInt_QHS<0,0,0b0001,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1854 IIC_VBINi4Q, "vrhadd", "s", int_arm_neon_vrhadds, 1>;
1855 defm VRHADDu : N3VInt_QHS<1,0,0b0001,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1856 IIC_VBINi4Q, "vrhadd", "u", int_arm_neon_vrhaddu, 1>;
1857 // VQADD : Vector Saturating Add
1858 defm VQADDs : N3VInt_QHSD<0,0,0b0000,1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1859 IIC_VBINi4Q, "vqadd", "s", int_arm_neon_vqadds, 1>;
1860 defm VQADDu : N3VInt_QHSD<1,0,0b0000,1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1861 IIC_VBINi4Q, "vqadd", "u", int_arm_neon_vqaddu, 1>;
1862 // VADDHN : Vector Add and Narrow Returning High Half (D = Q + Q)
1863 defm VADDHN : N3VNInt_HSD<0,1,0b0100,0, "vaddhn", "i",
1864 int_arm_neon_vaddhn, 1>;
1865 // VRADDHN : Vector Rounding Add and Narrow Returning High Half (D = Q + Q)
1866 defm VRADDHN : N3VNInt_HSD<1,1,0b0100,0, "vraddhn", "i",
1867 int_arm_neon_vraddhn, 1>;
1869 // Vector Multiply Operations.
1871 // VMUL : Vector Multiply (integer, polynomial and floating-point)
1872 defm VMUL : N3V_QHS<0, 0, 0b1001, 1, IIC_VMULi16D, IIC_VMULi32D,
1873 IIC_VMULi16Q, IIC_VMULi32Q, "vmul", "i", mul, 1>;
1874 def VMULpd : N3VDInt<1, 0, 0b00, 0b1001, 1, IIC_VMULi16D, "vmul", "p8",
1875 v8i8, v8i8, int_arm_neon_vmulp, 1>;
1876 def VMULpq : N3VQInt<1, 0, 0b00, 0b1001, 1, IIC_VMULi16Q, "vmul", "p8",
1877 v16i8, v16i8, int_arm_neon_vmulp, 1>;
1878 def VMULfd : N3VD<1, 0, 0b00, 0b1101, 1, IIC_VBIND, "vmul", "f32",
1879 v2f32, v2f32, fmul, 1>;
1880 def VMULfq : N3VQ<1, 0, 0b00, 0b1101, 1, IIC_VBINQ, "vmul", "f32",
1881 v4f32, v4f32, fmul, 1>;
1882 defm VMULsl : N3VSL_HS<0b1000, "vmul", "i", mul>;
1883 def VMULslfd : N3VDSL<0b10, 0b1001, IIC_VBIND, "vmul", "f32", v2f32, fmul>;
1884 def VMULslfq : N3VQSL<0b10, 0b1001, IIC_VBINQ, "vmul", "f32", v4f32,
1887 def : Pat<(v8i16 (mul (v8i16 QPR:$src1),
1888 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
1889 (v8i16 (VMULslv8i16 (v8i16 QPR:$src1),
1890 (v4i16 (EXTRACT_SUBREG QPR:$src2,
1891 (DSubReg_i16_reg imm:$lane))),
1892 (SubReg_i16_lane imm:$lane)))>;
1893 def : Pat<(v4i32 (mul (v4i32 QPR:$src1),
1894 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
1895 (v4i32 (VMULslv4i32 (v4i32 QPR:$src1),
1896 (v2i32 (EXTRACT_SUBREG QPR:$src2,
1897 (DSubReg_i32_reg imm:$lane))),
1898 (SubReg_i32_lane imm:$lane)))>;
1899 def : Pat<(v4f32 (fmul (v4f32 QPR:$src1),
1900 (v4f32 (NEONvduplane (v4f32 QPR:$src2), imm:$lane)))),
1901 (v4f32 (VMULslfq (v4f32 QPR:$src1),
1902 (v2f32 (EXTRACT_SUBREG QPR:$src2,
1903 (DSubReg_i32_reg imm:$lane))),
1904 (SubReg_i32_lane imm:$lane)))>;
1906 // VQDMULH : Vector Saturating Doubling Multiply Returning High Half
1907 defm VQDMULH : N3VInt_HS<0, 0, 0b1011, 0, IIC_VMULi16D, IIC_VMULi32D,
1908 IIC_VMULi16Q, IIC_VMULi32Q,
1909 "vqdmulh", "s", int_arm_neon_vqdmulh, 1>;
1910 defm VQDMULHsl: N3VIntSL_HS<0b1100, IIC_VMULi16D, IIC_VMULi32D,
1911 IIC_VMULi16Q, IIC_VMULi32Q,
1912 "vqdmulh", "s", int_arm_neon_vqdmulh>;
1913 def : Pat<(v8i16 (int_arm_neon_vqdmulh (v8i16 QPR:$src1),
1914 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
1916 (v8i16 (VQDMULHslv8i16 (v8i16 QPR:$src1),
1917 (v4i16 (EXTRACT_SUBREG QPR:$src2,
1918 (DSubReg_i16_reg imm:$lane))),
1919 (SubReg_i16_lane imm:$lane)))>;
1920 def : Pat<(v4i32 (int_arm_neon_vqdmulh (v4i32 QPR:$src1),
1921 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
1923 (v4i32 (VQDMULHslv4i32 (v4i32 QPR:$src1),
1924 (v2i32 (EXTRACT_SUBREG QPR:$src2,
1925 (DSubReg_i32_reg imm:$lane))),
1926 (SubReg_i32_lane imm:$lane)))>;
1928 // VQRDMULH : Vector Rounding Saturating Doubling Multiply Returning High Half
1929 defm VQRDMULH : N3VInt_HS<1, 0, 0b1011, 0, IIC_VMULi16D, IIC_VMULi32D,
1930 IIC_VMULi16Q, IIC_VMULi32Q,
1931 "vqrdmulh", "s", int_arm_neon_vqrdmulh, 1>;
1932 defm VQRDMULHsl : N3VIntSL_HS<0b1101, IIC_VMULi16D, IIC_VMULi32D,
1933 IIC_VMULi16Q, IIC_VMULi32Q,
1934 "vqrdmulh", "s", int_arm_neon_vqrdmulh>;
1935 def : Pat<(v8i16 (int_arm_neon_vqrdmulh (v8i16 QPR:$src1),
1936 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
1938 (v8i16 (VQRDMULHslv8i16 (v8i16 QPR:$src1),
1939 (v4i16 (EXTRACT_SUBREG QPR:$src2,
1940 (DSubReg_i16_reg imm:$lane))),
1941 (SubReg_i16_lane imm:$lane)))>;
1942 def : Pat<(v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$src1),
1943 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
1945 (v4i32 (VQRDMULHslv4i32 (v4i32 QPR:$src1),
1946 (v2i32 (EXTRACT_SUBREG QPR:$src2,
1947 (DSubReg_i32_reg imm:$lane))),
1948 (SubReg_i32_lane imm:$lane)))>;
1950 // VMULL : Vector Multiply Long (integer and polynomial) (Q = D * D)
1951 defm VMULLs : N3VLInt_QHS<0,1,0b1100,0, IIC_VMULi16D, "vmull", "s",
1952 int_arm_neon_vmulls, 1>;
1953 defm VMULLu : N3VLInt_QHS<1,1,0b1100,0, IIC_VMULi16D, "vmull", "u",
1954 int_arm_neon_vmullu, 1>;
1955 def VMULLp : N3VLInt<0, 1, 0b00, 0b1110, 0, IIC_VMULi16D, "vmull", "p8",
1956 v8i16, v8i8, int_arm_neon_vmullp, 1>;
1957 defm VMULLsls : N3VLIntSL_HS<0, 0b1010, IIC_VMULi16D, "vmull", "s",
1958 int_arm_neon_vmulls>;
1959 defm VMULLslu : N3VLIntSL_HS<1, 0b1010, IIC_VMULi16D, "vmull", "u",
1960 int_arm_neon_vmullu>;
1962 // VQDMULL : Vector Saturating Doubling Multiply Long (Q = D * D)
1963 defm VQDMULL : N3VLInt_HS<0,1,0b1101,0, IIC_VMULi16D, "vqdmull", "s",
1964 int_arm_neon_vqdmull, 1>;
1965 defm VQDMULLsl: N3VLIntSL_HS<0, 0b1011, IIC_VMULi16D, "vqdmull", "s",
1966 int_arm_neon_vqdmull>;
1968 // Vector Multiply-Accumulate and Multiply-Subtract Operations.
1970 // VMLA : Vector Multiply Accumulate (integer and floating-point)
1971 defm VMLA : N3VMulOp_QHS<0, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
1972 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
1973 def VMLAfd : N3VDMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACD, "vmla", "f32",
1975 def VMLAfq : N3VQMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACQ, "vmla", "f32",
1977 defm VMLAsl : N3VMulOpSL_HS<0b0000, IIC_VMACi16D, IIC_VMACi32D,
1978 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
1979 def VMLAslfd : N3VDMulOpSL<0b10, 0b0001, IIC_VMACD, "vmla", "f32",
1981 def VMLAslfq : N3VQMulOpSL<0b10, 0b0001, IIC_VMACQ, "vmla", "f32",
1982 v4f32, v2f32, fmul, fadd>;
1984 def : Pat<(v8i16 (add (v8i16 QPR:$src1),
1985 (mul (v8i16 QPR:$src2),
1986 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
1987 (v8i16 (VMLAslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
1988 (v4i16 (EXTRACT_SUBREG QPR:$src3,
1989 (DSubReg_i16_reg imm:$lane))),
1990 (SubReg_i16_lane imm:$lane)))>;
1992 def : Pat<(v4i32 (add (v4i32 QPR:$src1),
1993 (mul (v4i32 QPR:$src2),
1994 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
1995 (v4i32 (VMLAslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
1996 (v2i32 (EXTRACT_SUBREG QPR:$src3,
1997 (DSubReg_i32_reg imm:$lane))),
1998 (SubReg_i32_lane imm:$lane)))>;
2000 def : Pat<(v4f32 (fadd (v4f32 QPR:$src1),
2001 (fmul (v4f32 QPR:$src2),
2002 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
2003 (v4f32 (VMLAslfq (v4f32 QPR:$src1),
2005 (v2f32 (EXTRACT_SUBREG QPR:$src3,
2006 (DSubReg_i32_reg imm:$lane))),
2007 (SubReg_i32_lane imm:$lane)))>;
2009 // VMLAL : Vector Multiply Accumulate Long (Q += D * D)
2010 defm VMLALs : N3VLInt3_QHS<0,1,0b1000,0, "vmlal", "s", int_arm_neon_vmlals>;
2011 defm VMLALu : N3VLInt3_QHS<1,1,0b1000,0, "vmlal", "u", int_arm_neon_vmlalu>;
2013 defm VMLALsls : N3VLInt3SL_HS<0, 0b0010, "vmlal", "s", int_arm_neon_vmlals>;
2014 defm VMLALslu : N3VLInt3SL_HS<1, 0b0010, "vmlal", "u", int_arm_neon_vmlalu>;
2016 // VQDMLAL : Vector Saturating Doubling Multiply Accumulate Long (Q += D * D)
2017 defm VQDMLAL : N3VLInt3_HS<0, 1, 0b1001, 0, "vqdmlal", "s",
2018 int_arm_neon_vqdmlal>;
2019 defm VQDMLALsl: N3VLInt3SL_HS<0, 0b0011, "vqdmlal", "s", int_arm_neon_vqdmlal>;
2021 // VMLS : Vector Multiply Subtract (integer and floating-point)
2022 defm VMLS : N3VMulOp_QHS<1, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
2023 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
2024 def VMLSfd : N3VDMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACD, "vmls", "f32",
2026 def VMLSfq : N3VQMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACQ, "vmls", "f32",
2028 defm VMLSsl : N3VMulOpSL_HS<0b0100, IIC_VMACi16D, IIC_VMACi32D,
2029 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
2030 def VMLSslfd : N3VDMulOpSL<0b10, 0b0101, IIC_VMACD, "vmls", "f32",
2032 def VMLSslfq : N3VQMulOpSL<0b10, 0b0101, IIC_VMACQ, "vmls", "f32",
2033 v4f32, v2f32, fmul, fsub>;
2035 def : Pat<(v8i16 (sub (v8i16 QPR:$src1),
2036 (mul (v8i16 QPR:$src2),
2037 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
2038 (v8i16 (VMLSslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
2039 (v4i16 (EXTRACT_SUBREG QPR:$src3,
2040 (DSubReg_i16_reg imm:$lane))),
2041 (SubReg_i16_lane imm:$lane)))>;
2043 def : Pat<(v4i32 (sub (v4i32 QPR:$src1),
2044 (mul (v4i32 QPR:$src2),
2045 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
2046 (v4i32 (VMLSslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
2047 (v2i32 (EXTRACT_SUBREG QPR:$src3,
2048 (DSubReg_i32_reg imm:$lane))),
2049 (SubReg_i32_lane imm:$lane)))>;
2051 def : Pat<(v4f32 (fsub (v4f32 QPR:$src1),
2052 (fmul (v4f32 QPR:$src2),
2053 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
2054 (v4f32 (VMLSslfq (v4f32 QPR:$src1), (v4f32 QPR:$src2),
2055 (v2f32 (EXTRACT_SUBREG QPR:$src3,
2056 (DSubReg_i32_reg imm:$lane))),
2057 (SubReg_i32_lane imm:$lane)))>;
2059 // VMLSL : Vector Multiply Subtract Long (Q -= D * D)
2060 defm VMLSLs : N3VLInt3_QHS<0,1,0b1010,0, "vmlsl", "s", int_arm_neon_vmlsls>;
2061 defm VMLSLu : N3VLInt3_QHS<1,1,0b1010,0, "vmlsl", "u", int_arm_neon_vmlslu>;
2063 defm VMLSLsls : N3VLInt3SL_HS<0, 0b0110, "vmlsl", "s", int_arm_neon_vmlsls>;
2064 defm VMLSLslu : N3VLInt3SL_HS<1, 0b0110, "vmlsl", "u", int_arm_neon_vmlslu>;
2066 // VQDMLSL : Vector Saturating Doubling Multiply Subtract Long (Q -= D * D)
2067 defm VQDMLSL : N3VLInt3_HS<0, 1, 0b1011, 0, "vqdmlsl", "s",
2068 int_arm_neon_vqdmlsl>;
2069 defm VQDMLSLsl: N3VLInt3SL_HS<0, 0b111, "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
2071 // Vector Subtract Operations.
2073 // VSUB : Vector Subtract (integer and floating-point)
2074 defm VSUB : N3V_QHSD<1, 0, 0b1000, 0, IIC_VSUBiD, IIC_VSUBiQ,
2075 "vsub", "i", sub, 0>;
2076 def VSUBfd : N3VD<0, 0, 0b10, 0b1101, 0, IIC_VBIND, "vsub", "f32",
2077 v2f32, v2f32, fsub, 0>;
2078 def VSUBfq : N3VQ<0, 0, 0b10, 0b1101, 0, IIC_VBINQ, "vsub", "f32",
2079 v4f32, v4f32, fsub, 0>;
2080 // VSUBL : Vector Subtract Long (Q = D - D)
2081 defm VSUBLs : N3VLInt_QHS<0,1,0b0010,0, IIC_VSHLiD, "vsubl", "s",
2082 int_arm_neon_vsubls, 1>;
2083 defm VSUBLu : N3VLInt_QHS<1,1,0b0010,0, IIC_VSHLiD, "vsubl", "u",
2084 int_arm_neon_vsublu, 1>;
2085 // VSUBW : Vector Subtract Wide (Q = Q - D)
2086 defm VSUBWs : N3VWInt_QHS<0,1,0b0011,0, "vsubw", "s", int_arm_neon_vsubws, 0>;
2087 defm VSUBWu : N3VWInt_QHS<1,1,0b0011,0, "vsubw", "u", int_arm_neon_vsubwu, 0>;
2088 // VHSUB : Vector Halving Subtract
2089 defm VHSUBs : N3VInt_QHS<0, 0, 0b0010, 0, IIC_VBINi4D, IIC_VBINi4D,
2090 IIC_VBINi4Q, IIC_VBINi4Q,
2091 "vhsub", "s", int_arm_neon_vhsubs, 0>;
2092 defm VHSUBu : N3VInt_QHS<1, 0, 0b0010, 0, IIC_VBINi4D, IIC_VBINi4D,
2093 IIC_VBINi4Q, IIC_VBINi4Q,
2094 "vhsub", "u", int_arm_neon_vhsubu, 0>;
2095 // VQSUB : Vector Saturing Subtract
2096 defm VQSUBs : N3VInt_QHSD<0, 0, 0b0010, 1, IIC_VBINi4D, IIC_VBINi4D,
2097 IIC_VBINi4Q, IIC_VBINi4Q,
2098 "vqsub", "s", int_arm_neon_vqsubs, 0>;
2099 defm VQSUBu : N3VInt_QHSD<1, 0, 0b0010, 1, IIC_VBINi4D, IIC_VBINi4D,
2100 IIC_VBINi4Q, IIC_VBINi4Q,
2101 "vqsub", "u", int_arm_neon_vqsubu, 0>;
2102 // VSUBHN : Vector Subtract and Narrow Returning High Half (D = Q - Q)
2103 defm VSUBHN : N3VNInt_HSD<0,1,0b0110,0, "vsubhn", "i",
2104 int_arm_neon_vsubhn, 0>;
2105 // VRSUBHN : Vector Rounding Subtract and Narrow Returning High Half (D=Q-Q)
2106 defm VRSUBHN : N3VNInt_HSD<1,1,0b0110,0, "vrsubhn", "i",
2107 int_arm_neon_vrsubhn, 0>;
2109 // Vector Comparisons.
2111 // VCEQ : Vector Compare Equal
2112 defm VCEQ : N3V_QHS<1, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2113 IIC_VBINi4Q, "vceq", "i", NEONvceq, 1>;
2114 def VCEQfd : N3VD<0,0,0b00,0b1110,0, IIC_VBIND, "vceq", "f32", v2i32, v2f32,
2116 def VCEQfq : N3VQ<0,0,0b00,0b1110,0, IIC_VBINQ, "vceq", "f32", v4i32, v4f32,
2118 // For disassembly only.
2119 defm VCEQz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00010, 0, "vceq", "i",
2122 // VCGE : Vector Compare Greater Than or Equal
2123 defm VCGEs : N3V_QHS<0, 0, 0b0011, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2124 IIC_VBINi4Q, "vcge", "s", NEONvcge, 0>;
2125 defm VCGEu : N3V_QHS<1, 0, 0b0011, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2126 IIC_VBINi4Q, "vcge", "u", NEONvcgeu, 0>;
2127 def VCGEfd : N3VD<1,0,0b00,0b1110,0, IIC_VBIND, "vcge", "f32",
2128 v2i32, v2f32, NEONvcge, 0>;
2129 def VCGEfq : N3VQ<1,0,0b00,0b1110,0, IIC_VBINQ, "vcge", "f32", v4i32, v4f32,
2131 // For disassembly only.
2132 defm VCGEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00001, 0, "vcge", "s",
2134 // For disassembly only.
2135 defm VCLEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00011, 0, "vcle", "s",
2138 // VCGT : Vector Compare Greater Than
2139 defm VCGTs : N3V_QHS<0, 0, 0b0011, 0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2140 IIC_VBINi4Q, "vcgt", "s", NEONvcgt, 0>;
2141 defm VCGTu : N3V_QHS<1, 0, 0b0011, 0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2142 IIC_VBINi4Q, "vcgt", "u", NEONvcgtu, 0>;
2143 def VCGTfd : N3VD<1,0,0b10,0b1110,0, IIC_VBIND, "vcgt", "f32", v2i32, v2f32,
2145 def VCGTfq : N3VQ<1,0,0b10,0b1110,0, IIC_VBINQ, "vcgt", "f32", v4i32, v4f32,
2147 // For disassembly only.
2148 defm VCGTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00000, 0, "vcgt", "s",
2150 // For disassembly only.
2151 defm VCLTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00100, 0, "vclt", "s",
2154 // VACGE : Vector Absolute Compare Greater Than or Equal (aka VCAGE)
2155 def VACGEd : N3VDInt<1, 0, 0b00, 0b1110, 1, IIC_VBIND, "vacge", "f32",
2156 v2i32, v2f32, int_arm_neon_vacged, 0>;
2157 def VACGEq : N3VQInt<1, 0, 0b00, 0b1110, 1, IIC_VBINQ, "vacge", "f32",
2158 v4i32, v4f32, int_arm_neon_vacgeq, 0>;
2159 // VACGT : Vector Absolute Compare Greater Than (aka VCAGT)
2160 def VACGTd : N3VDInt<1, 0, 0b10, 0b1110, 1, IIC_VBIND, "vacgt", "f32",
2161 v2i32, v2f32, int_arm_neon_vacgtd, 0>;
2162 def VACGTq : N3VQInt<1, 0, 0b10, 0b1110, 1, IIC_VBINQ, "vacgt", "f32",
2163 v4i32, v4f32, int_arm_neon_vacgtq, 0>;
2164 // VTST : Vector Test Bits
2165 defm VTST : N3V_QHS<0, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2166 IIC_VBINi4Q, "vtst", "", NEONvtst, 1>;
2168 // Vector Bitwise Operations.
2170 // VAND : Vector Bitwise AND
2171 def VANDd : N3VDX<0, 0, 0b00, 0b0001, 1, IIC_VBINiD, "vand",
2172 v2i32, v2i32, and, 1>;
2173 def VANDq : N3VQX<0, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "vand",
2174 v4i32, v4i32, and, 1>;
2176 // VEOR : Vector Bitwise Exclusive OR
2177 def VEORd : N3VDX<1, 0, 0b00, 0b0001, 1, IIC_VBINiD, "veor",
2178 v2i32, v2i32, xor, 1>;
2179 def VEORq : N3VQX<1, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "veor",
2180 v4i32, v4i32, xor, 1>;
2182 // VORR : Vector Bitwise OR
2183 def VORRd : N3VDX<0, 0, 0b10, 0b0001, 1, IIC_VBINiD, "vorr",
2184 v2i32, v2i32, or, 1>;
2185 def VORRq : N3VQX<0, 0, 0b10, 0b0001, 1, IIC_VBINiQ, "vorr",
2186 v4i32, v4i32, or, 1>;
2188 // VBIC : Vector Bitwise Bit Clear (AND NOT)
2189 def VBICd : N3VX<0, 0, 0b01, 0b0001, 0, 1, (outs DPR:$dst),
2190 (ins DPR:$src1, DPR:$src2), IIC_VBINiD,
2191 "vbic", "$dst, $src1, $src2", "",
2192 [(set DPR:$dst, (v2i32 (and DPR:$src1,
2193 (vnot_conv DPR:$src2))))]>;
2194 def VBICq : N3VX<0, 0, 0b01, 0b0001, 1, 1, (outs QPR:$dst),
2195 (ins QPR:$src1, QPR:$src2), IIC_VBINiQ,
2196 "vbic", "$dst, $src1, $src2", "",
2197 [(set QPR:$dst, (v4i32 (and QPR:$src1,
2198 (vnot_conv QPR:$src2))))]>;
2200 // VORN : Vector Bitwise OR NOT
2201 def VORNd : N3VX<0, 0, 0b11, 0b0001, 0, 1, (outs DPR:$dst),
2202 (ins DPR:$src1, DPR:$src2), IIC_VBINiD,
2203 "vorn", "$dst, $src1, $src2", "",
2204 [(set DPR:$dst, (v2i32 (or DPR:$src1,
2205 (vnot_conv DPR:$src2))))]>;
2206 def VORNq : N3VX<0, 0, 0b11, 0b0001, 1, 1, (outs QPR:$dst),
2207 (ins QPR:$src1, QPR:$src2), IIC_VBINiQ,
2208 "vorn", "$dst, $src1, $src2", "",
2209 [(set QPR:$dst, (v4i32 (or QPR:$src1,
2210 (vnot_conv QPR:$src2))))]>;
2212 // VMVN : Vector Bitwise NOT
2213 def VMVNd : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 0, 0,
2214 (outs DPR:$dst), (ins DPR:$src), IIC_VSHLiD,
2215 "vmvn", "$dst, $src", "",
2216 [(set DPR:$dst, (v2i32 (vnot DPR:$src)))]>;
2217 def VMVNq : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 1, 0,
2218 (outs QPR:$dst), (ins QPR:$src), IIC_VSHLiD,
2219 "vmvn", "$dst, $src", "",
2220 [(set QPR:$dst, (v4i32 (vnot QPR:$src)))]>;
2221 def : Pat<(v2i32 (vnot_conv DPR:$src)), (VMVNd DPR:$src)>;
2222 def : Pat<(v4i32 (vnot_conv QPR:$src)), (VMVNq QPR:$src)>;
2224 // VBSL : Vector Bitwise Select
2225 def VBSLd : N3VX<1, 0, 0b01, 0b0001, 0, 1, (outs DPR:$dst),
2226 (ins DPR:$src1, DPR:$src2, DPR:$src3), IIC_VCNTiD,
2227 "vbsl", "$dst, $src2, $src3", "$src1 = $dst",
2229 (v2i32 (or (and DPR:$src2, DPR:$src1),
2230 (and DPR:$src3, (vnot_conv DPR:$src1)))))]>;
2231 def VBSLq : N3VX<1, 0, 0b01, 0b0001, 1, 1, (outs QPR:$dst),
2232 (ins QPR:$src1, QPR:$src2, QPR:$src3), IIC_VCNTiQ,
2233 "vbsl", "$dst, $src2, $src3", "$src1 = $dst",
2235 (v4i32 (or (and QPR:$src2, QPR:$src1),
2236 (and QPR:$src3, (vnot_conv QPR:$src1)))))]>;
2238 // VBIF : Vector Bitwise Insert if False
2239 // like VBSL but with: "vbif $dst, $src3, $src1", "$src2 = $dst",
2240 def VBIFd : N3VX<1, 0, 0b11, 0b0001, 0, 1,
2241 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3),
2242 IIC_VBINiD, "vbif", "$dst, $src2, $src3", "$src1 = $dst",
2243 [/* For disassembly only; pattern left blank */]>;
2244 def VBIFq : N3VX<1, 0, 0b11, 0b0001, 1, 1,
2245 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3),
2246 IIC_VBINiQ, "vbif", "$dst, $src2, $src3", "$src1 = $dst",
2247 [/* For disassembly only; pattern left blank */]>;
2249 // VBIT : Vector Bitwise Insert if True
2250 // like VBSL but with: "vbit $dst, $src2, $src1", "$src3 = $dst",
2251 def VBITd : N3VX<1, 0, 0b10, 0b0001, 0, 1,
2252 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3),
2253 IIC_VBINiD, "vbit", "$dst, $src2, $src3", "$src1 = $dst",
2254 [/* For disassembly only; pattern left blank */]>;
2255 def VBITq : N3VX<1, 0, 0b10, 0b0001, 1, 1,
2256 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3),
2257 IIC_VBINiQ, "vbit", "$dst, $src2, $src3", "$src1 = $dst",
2258 [/* For disassembly only; pattern left blank */]>;
2260 // VBIT/VBIF are not yet implemented. The TwoAddress pass will not go looking
2261 // for equivalent operations with different register constraints; it just
2264 // Vector Absolute Differences.
2266 // VABD : Vector Absolute Difference
2267 defm VABDs : N3VInt_QHS<0, 0, 0b0111, 0, IIC_VBINi4D, IIC_VBINi4D,
2268 IIC_VBINi4Q, IIC_VBINi4Q,
2269 "vabd", "s", int_arm_neon_vabds, 0>;
2270 defm VABDu : N3VInt_QHS<1, 0, 0b0111, 0, IIC_VBINi4D, IIC_VBINi4D,
2271 IIC_VBINi4Q, IIC_VBINi4Q,
2272 "vabd", "u", int_arm_neon_vabdu, 0>;
2273 def VABDfd : N3VDInt<1, 0, 0b10, 0b1101, 0, IIC_VBIND,
2274 "vabd", "f32", v2f32, v2f32, int_arm_neon_vabds, 0>;
2275 def VABDfq : N3VQInt<1, 0, 0b10, 0b1101, 0, IIC_VBINQ,
2276 "vabd", "f32", v4f32, v4f32, int_arm_neon_vabds, 0>;
2278 // VABDL : Vector Absolute Difference Long (Q = | D - D |)
2279 defm VABDLs : N3VLInt_QHS<0,1,0b0111,0, IIC_VBINi4Q,
2280 "vabdl", "s", int_arm_neon_vabdls, 0>;
2281 defm VABDLu : N3VLInt_QHS<1,1,0b0111,0, IIC_VBINi4Q,
2282 "vabdl", "u", int_arm_neon_vabdlu, 0>;
2284 // VABA : Vector Absolute Difference and Accumulate
2285 defm VABAs : N3VInt3_QHS<0,0,0b0111,1, "vaba", "s", int_arm_neon_vabas>;
2286 defm VABAu : N3VInt3_QHS<1,0,0b0111,1, "vaba", "u", int_arm_neon_vabau>;
2288 // VABAL : Vector Absolute Difference and Accumulate Long (Q += | D - D |)
2289 defm VABALs : N3VLInt3_QHS<0,1,0b0101,0, "vabal", "s", int_arm_neon_vabals>;
2290 defm VABALu : N3VLInt3_QHS<1,1,0b0101,0, "vabal", "u", int_arm_neon_vabalu>;
2292 // Vector Maximum and Minimum.
2294 // VMAX : Vector Maximum
2295 defm VMAXs : N3VInt_QHS<0,0,0b0110,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2296 IIC_VBINi4Q, "vmax", "s", int_arm_neon_vmaxs, 1>;
2297 defm VMAXu : N3VInt_QHS<1,0,0b0110,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2298 IIC_VBINi4Q, "vmax", "u", int_arm_neon_vmaxu, 1>;
2299 def VMAXfd : N3VDInt<0, 0, 0b00, 0b1111, 0, IIC_VBIND, "vmax", "f32",
2300 v2f32, v2f32, int_arm_neon_vmaxs, 1>;
2301 def VMAXfq : N3VQInt<0, 0, 0b00, 0b1111, 0, IIC_VBINQ, "vmax", "f32",
2302 v4f32, v4f32, int_arm_neon_vmaxs, 1>;
2304 // VMIN : Vector Minimum
2305 defm VMINs : N3VInt_QHS<0,0,0b0110,1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2306 IIC_VBINi4Q, "vmin", "s", int_arm_neon_vmins, 1>;
2307 defm VMINu : N3VInt_QHS<1,0,0b0110,1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2308 IIC_VBINi4Q, "vmin", "u", int_arm_neon_vminu, 1>;
2309 def VMINfd : N3VDInt<0, 0, 0b10, 0b1111, 0, IIC_VBIND, "vmin", "f32",
2310 v2f32, v2f32, int_arm_neon_vmins, 1>;
2311 def VMINfq : N3VQInt<0, 0, 0b10, 0b1111, 0, IIC_VBINQ, "vmin", "f32",
2312 v4f32, v4f32, int_arm_neon_vmins, 1>;
2314 // Vector Pairwise Operations.
2316 // VPADD : Vector Pairwise Add
2317 def VPADDi8 : N3VDInt<0, 0, 0b00, 0b1011, 1, IIC_VBINiD, "vpadd", "i8",
2318 v8i8, v8i8, int_arm_neon_vpadd, 0>;
2319 def VPADDi16 : N3VDInt<0, 0, 0b01, 0b1011, 1, IIC_VBINiD, "vpadd", "i16",
2320 v4i16, v4i16, int_arm_neon_vpadd, 0>;
2321 def VPADDi32 : N3VDInt<0, 0, 0b10, 0b1011, 1, IIC_VBINiD, "vpadd", "i32",
2322 v2i32, v2i32, int_arm_neon_vpadd, 0>;
2323 def VPADDf : N3VDInt<1, 0, 0b00, 0b1101, 0, IIC_VBIND, "vpadd", "f32",
2324 v2f32, v2f32, int_arm_neon_vpadd, 0>;
2326 // VPADDL : Vector Pairwise Add Long
2327 defm VPADDLs : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00100, 0, "vpaddl", "s",
2328 int_arm_neon_vpaddls>;
2329 defm VPADDLu : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00101, 0, "vpaddl", "u",
2330 int_arm_neon_vpaddlu>;
2332 // VPADAL : Vector Pairwise Add and Accumulate Long
2333 defm VPADALs : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01100, 0, "vpadal", "s",
2334 int_arm_neon_vpadals>;
2335 defm VPADALu : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01101, 0, "vpadal", "u",
2336 int_arm_neon_vpadalu>;
2338 // VPMAX : Vector Pairwise Maximum
2339 def VPMAXs8 : N3VDInt<0, 0, 0b00, 0b1010, 0, IIC_VBINi4D, "vpmax", "s8",
2340 v8i8, v8i8, int_arm_neon_vpmaxs, 0>;
2341 def VPMAXs16 : N3VDInt<0, 0, 0b01, 0b1010, 0, IIC_VBINi4D, "vpmax", "s16",
2342 v4i16, v4i16, int_arm_neon_vpmaxs, 0>;
2343 def VPMAXs32 : N3VDInt<0, 0, 0b10, 0b1010, 0, IIC_VBINi4D, "vpmax", "s32",
2344 v2i32, v2i32, int_arm_neon_vpmaxs, 0>;
2345 def VPMAXu8 : N3VDInt<1, 0, 0b00, 0b1010, 0, IIC_VBINi4D, "vpmax", "u8",
2346 v8i8, v8i8, int_arm_neon_vpmaxu, 0>;
2347 def VPMAXu16 : N3VDInt<1, 0, 0b01, 0b1010, 0, IIC_VBINi4D, "vpmax", "u16",
2348 v4i16, v4i16, int_arm_neon_vpmaxu, 0>;
2349 def VPMAXu32 : N3VDInt<1, 0, 0b10, 0b1010, 0, IIC_VBINi4D, "vpmax", "u32",
2350 v2i32, v2i32, int_arm_neon_vpmaxu, 0>;
2351 def VPMAXf : N3VDInt<1, 0, 0b00, 0b1111, 0, IIC_VBINi4D, "vpmax", "f32",
2352 v2f32, v2f32, int_arm_neon_vpmaxs, 0>;
2354 // VPMIN : Vector Pairwise Minimum
2355 def VPMINs8 : N3VDInt<0, 0, 0b00, 0b1010, 1, IIC_VBINi4D, "vpmin", "s8",
2356 v8i8, v8i8, int_arm_neon_vpmins, 0>;
2357 def VPMINs16 : N3VDInt<0, 0, 0b01, 0b1010, 1, IIC_VBINi4D, "vpmin", "s16",
2358 v4i16, v4i16, int_arm_neon_vpmins, 0>;
2359 def VPMINs32 : N3VDInt<0, 0, 0b10, 0b1010, 1, IIC_VBINi4D, "vpmin", "s32",
2360 v2i32, v2i32, int_arm_neon_vpmins, 0>;
2361 def VPMINu8 : N3VDInt<1, 0, 0b00, 0b1010, 1, IIC_VBINi4D, "vpmin", "u8",
2362 v8i8, v8i8, int_arm_neon_vpminu, 0>;
2363 def VPMINu16 : N3VDInt<1, 0, 0b01, 0b1010, 1, IIC_VBINi4D, "vpmin", "u16",
2364 v4i16, v4i16, int_arm_neon_vpminu, 0>;
2365 def VPMINu32 : N3VDInt<1, 0, 0b10, 0b1010, 1, IIC_VBINi4D, "vpmin", "u32",
2366 v2i32, v2i32, int_arm_neon_vpminu, 0>;
2367 def VPMINf : N3VDInt<1, 0, 0b10, 0b1111, 0, IIC_VBINi4D, "vpmin", "f32",
2368 v2f32, v2f32, int_arm_neon_vpmins, 0>;
2370 // Vector Reciprocal and Reciprocal Square Root Estimate and Step.
2372 // VRECPE : Vector Reciprocal Estimate
2373 def VRECPEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
2374 IIC_VUNAD, "vrecpe", "u32",
2375 v2i32, v2i32, int_arm_neon_vrecpe>;
2376 def VRECPEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
2377 IIC_VUNAQ, "vrecpe", "u32",
2378 v4i32, v4i32, int_arm_neon_vrecpe>;
2379 def VRECPEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
2380 IIC_VUNAD, "vrecpe", "f32",
2381 v2f32, v2f32, int_arm_neon_vrecpe>;
2382 def VRECPEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
2383 IIC_VUNAQ, "vrecpe", "f32",
2384 v4f32, v4f32, int_arm_neon_vrecpe>;
2386 // VRECPS : Vector Reciprocal Step
2387 def VRECPSfd : N3VDInt<0, 0, 0b00, 0b1111, 1,
2388 IIC_VRECSD, "vrecps", "f32",
2389 v2f32, v2f32, int_arm_neon_vrecps, 1>;
2390 def VRECPSfq : N3VQInt<0, 0, 0b00, 0b1111, 1,
2391 IIC_VRECSQ, "vrecps", "f32",
2392 v4f32, v4f32, int_arm_neon_vrecps, 1>;
2394 // VRSQRTE : Vector Reciprocal Square Root Estimate
2395 def VRSQRTEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
2396 IIC_VUNAD, "vrsqrte", "u32",
2397 v2i32, v2i32, int_arm_neon_vrsqrte>;
2398 def VRSQRTEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
2399 IIC_VUNAQ, "vrsqrte", "u32",
2400 v4i32, v4i32, int_arm_neon_vrsqrte>;
2401 def VRSQRTEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
2402 IIC_VUNAD, "vrsqrte", "f32",
2403 v2f32, v2f32, int_arm_neon_vrsqrte>;
2404 def VRSQRTEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
2405 IIC_VUNAQ, "vrsqrte", "f32",
2406 v4f32, v4f32, int_arm_neon_vrsqrte>;
2408 // VRSQRTS : Vector Reciprocal Square Root Step
2409 def VRSQRTSfd : N3VDInt<0, 0, 0b10, 0b1111, 1,
2410 IIC_VRECSD, "vrsqrts", "f32",
2411 v2f32, v2f32, int_arm_neon_vrsqrts, 1>;
2412 def VRSQRTSfq : N3VQInt<0, 0, 0b10, 0b1111, 1,
2413 IIC_VRECSQ, "vrsqrts", "f32",
2414 v4f32, v4f32, int_arm_neon_vrsqrts, 1>;
2418 // VSHL : Vector Shift
2419 defm VSHLs : N3VInt_QHSD<0, 0, 0b0100, 0, IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ,
2420 IIC_VSHLiQ, "vshl", "s", int_arm_neon_vshifts, 0>;
2421 defm VSHLu : N3VInt_QHSD<1, 0, 0b0100, 0, IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ,
2422 IIC_VSHLiQ, "vshl", "u", int_arm_neon_vshiftu, 0>;
2423 // VSHL : Vector Shift Left (Immediate)
2424 defm VSHLi : N2VSh_QHSD<0, 1, 0b0101, 1, IIC_VSHLiD, "vshl", "i", NEONvshl>;
2425 // VSHR : Vector Shift Right (Immediate)
2426 defm VSHRs : N2VSh_QHSD<0, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "s", NEONvshrs>;
2427 defm VSHRu : N2VSh_QHSD<1, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "u", NEONvshru>;
2429 // VSHLL : Vector Shift Left Long
2430 defm VSHLLs : N2VLSh_QHS<0, 1, 0b1010, 0, 0, 1, "vshll", "s", NEONvshlls>;
2431 defm VSHLLu : N2VLSh_QHS<1, 1, 0b1010, 0, 0, 1, "vshll", "u", NEONvshllu>;
2433 // VSHLL : Vector Shift Left Long (with maximum shift count)
2434 class N2VLShMax<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
2435 bit op6, bit op4, string OpcodeStr, string Dt, ValueType ResTy,
2436 ValueType OpTy, SDNode OpNode>
2437 : N2VLSh<op24, op23, op11_8, op7, op6, op4, OpcodeStr, Dt,
2438 ResTy, OpTy, OpNode> {
2439 let Inst{21-16} = op21_16;
2441 def VSHLLi8 : N2VLShMax<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll", "i8",
2442 v8i16, v8i8, NEONvshlli>;
2443 def VSHLLi16 : N2VLShMax<1, 1, 0b110110, 0b0011, 0, 0, 0, "vshll", "i16",
2444 v4i32, v4i16, NEONvshlli>;
2445 def VSHLLi32 : N2VLShMax<1, 1, 0b111010, 0b0011, 0, 0, 0, "vshll", "i32",
2446 v2i64, v2i32, NEONvshlli>;
2448 // VSHRN : Vector Shift Right and Narrow
2449 defm VSHRN : N2VNSh_HSD<0,1,0b1000,0,0,1, IIC_VSHLiD, "vshrn", "i",
2452 // VRSHL : Vector Rounding Shift
2453 defm VRSHLs : N3VInt_QHSD<0,0,0b0101,0, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2454 IIC_VSHLi4Q, "vrshl", "s", int_arm_neon_vrshifts,0>;
2455 defm VRSHLu : N3VInt_QHSD<1,0,0b0101,0, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2456 IIC_VSHLi4Q, "vrshl", "u", int_arm_neon_vrshiftu,0>;
2457 // VRSHR : Vector Rounding Shift Right
2458 defm VRSHRs : N2VSh_QHSD<0,1,0b0010,1, IIC_VSHLi4D, "vrshr", "s", NEONvrshrs>;
2459 defm VRSHRu : N2VSh_QHSD<1,1,0b0010,1, IIC_VSHLi4D, "vrshr", "u", NEONvrshru>;
2461 // VRSHRN : Vector Rounding Shift Right and Narrow
2462 defm VRSHRN : N2VNSh_HSD<0, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vrshrn", "i",
2465 // VQSHL : Vector Saturating Shift
2466 defm VQSHLs : N3VInt_QHSD<0,0,0b0100,1, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2467 IIC_VSHLi4Q, "vqshl", "s", int_arm_neon_vqshifts,0>;
2468 defm VQSHLu : N3VInt_QHSD<1,0,0b0100,1, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2469 IIC_VSHLi4Q, "vqshl", "u", int_arm_neon_vqshiftu,0>;
2470 // VQSHL : Vector Saturating Shift Left (Immediate)
2471 defm VQSHLsi : N2VSh_QHSD<0,1,0b0111,1, IIC_VSHLi4D, "vqshl", "s", NEONvqshls>;
2472 defm VQSHLui : N2VSh_QHSD<1,1,0b0111,1, IIC_VSHLi4D, "vqshl", "u", NEONvqshlu>;
2473 // VQSHLU : Vector Saturating Shift Left (Immediate, Unsigned)
2474 defm VQSHLsu : N2VSh_QHSD<1,1,0b0110,1, IIC_VSHLi4D, "vqshlu","s",NEONvqshlsu>;
2476 // VQSHRN : Vector Saturating Shift Right and Narrow
2477 defm VQSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "s",
2479 defm VQSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "u",
2482 // VQSHRUN : Vector Saturating Shift Right and Narrow (Unsigned)
2483 defm VQSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 0, 1, IIC_VSHLi4D, "vqshrun", "s",
2486 // VQRSHL : Vector Saturating Rounding Shift
2487 defm VQRSHLs : N3VInt_QHSD<0,0,0b0101,1, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2488 IIC_VSHLi4Q, "vqrshl", "s",
2489 int_arm_neon_vqrshifts, 0>;
2490 defm VQRSHLu : N3VInt_QHSD<1,0,0b0101,1, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2491 IIC_VSHLi4Q, "vqrshl", "u",
2492 int_arm_neon_vqrshiftu, 0>;
2494 // VQRSHRN : Vector Saturating Rounding Shift Right and Narrow
2495 defm VQRSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "s",
2497 defm VQRSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "u",
2500 // VQRSHRUN : Vector Saturating Rounding Shift Right and Narrow (Unsigned)
2501 defm VQRSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vqrshrun", "s",
2504 // VSRA : Vector Shift Right and Accumulate
2505 defm VSRAs : N2VShAdd_QHSD<0, 1, 0b0001, 1, "vsra", "s", NEONvshrs>;
2506 defm VSRAu : N2VShAdd_QHSD<1, 1, 0b0001, 1, "vsra", "u", NEONvshru>;
2507 // VRSRA : Vector Rounding Shift Right and Accumulate
2508 defm VRSRAs : N2VShAdd_QHSD<0, 1, 0b0011, 1, "vrsra", "s", NEONvrshrs>;
2509 defm VRSRAu : N2VShAdd_QHSD<1, 1, 0b0011, 1, "vrsra", "u", NEONvrshru>;
2511 // VSLI : Vector Shift Left and Insert
2512 defm VSLI : N2VShIns_QHSD<1, 1, 0b0101, 1, "vsli", NEONvsli>;
2513 // VSRI : Vector Shift Right and Insert
2514 defm VSRI : N2VShIns_QHSD<1, 1, 0b0100, 1, "vsri", NEONvsri>;
2516 // Vector Absolute and Saturating Absolute.
2518 // VABS : Vector Absolute Value
2519 defm VABS : N2VInt_QHS<0b11, 0b11, 0b01, 0b00110, 0,
2520 IIC_VUNAiD, IIC_VUNAiQ, "vabs", "s",
2522 def VABSfd : N2VDInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
2523 IIC_VUNAD, "vabs", "f32",
2524 v2f32, v2f32, int_arm_neon_vabs>;
2525 def VABSfq : N2VQInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
2526 IIC_VUNAQ, "vabs", "f32",
2527 v4f32, v4f32, int_arm_neon_vabs>;
2529 // VQABS : Vector Saturating Absolute Value
2530 defm VQABS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01110, 0,
2531 IIC_VQUNAiD, IIC_VQUNAiQ, "vqabs", "s",
2532 int_arm_neon_vqabs>;
2536 def vneg : PatFrag<(ops node:$in), (sub immAllZerosV, node:$in)>;
2537 def vneg_conv : PatFrag<(ops node:$in), (sub immAllZerosV_bc, node:$in)>;
2539 class VNEGD<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
2540 : N2V<0b11, 0b11, size, 0b01, 0b00111, 0, 0, (outs DPR:$dst), (ins DPR:$src),
2541 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
2542 [(set DPR:$dst, (Ty (vneg DPR:$src)))]>;
2543 class VNEGQ<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
2544 : N2V<0b11, 0b11, size, 0b01, 0b00111, 1, 0, (outs QPR:$dst), (ins QPR:$src),
2545 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
2546 [(set QPR:$dst, (Ty (vneg QPR:$src)))]>;
2548 // VNEG : Vector Negate
2549 def VNEGs8d : VNEGD<0b00, "vneg", "s8", v8i8>;
2550 def VNEGs16d : VNEGD<0b01, "vneg", "s16", v4i16>;
2551 def VNEGs32d : VNEGD<0b10, "vneg", "s32", v2i32>;
2552 def VNEGs8q : VNEGQ<0b00, "vneg", "s8", v16i8>;
2553 def VNEGs16q : VNEGQ<0b01, "vneg", "s16", v8i16>;
2554 def VNEGs32q : VNEGQ<0b10, "vneg", "s32", v4i32>;
2556 // VNEG : Vector Negate (floating-point)
2557 def VNEGfd : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
2558 (outs DPR:$dst), (ins DPR:$src), IIC_VUNAD,
2559 "vneg", "f32", "$dst, $src", "",
2560 [(set DPR:$dst, (v2f32 (fneg DPR:$src)))]>;
2561 def VNEGf32q : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 1, 0,
2562 (outs QPR:$dst), (ins QPR:$src), IIC_VUNAQ,
2563 "vneg", "f32", "$dst, $src", "",
2564 [(set QPR:$dst, (v4f32 (fneg QPR:$src)))]>;
2566 def : Pat<(v8i8 (vneg_conv DPR:$src)), (VNEGs8d DPR:$src)>;
2567 def : Pat<(v4i16 (vneg_conv DPR:$src)), (VNEGs16d DPR:$src)>;
2568 def : Pat<(v2i32 (vneg_conv DPR:$src)), (VNEGs32d DPR:$src)>;
2569 def : Pat<(v16i8 (vneg_conv QPR:$src)), (VNEGs8q QPR:$src)>;
2570 def : Pat<(v8i16 (vneg_conv QPR:$src)), (VNEGs16q QPR:$src)>;
2571 def : Pat<(v4i32 (vneg_conv QPR:$src)), (VNEGs32q QPR:$src)>;
2573 // VQNEG : Vector Saturating Negate
2574 defm VQNEG : N2VInt_QHS<0b11, 0b11, 0b00, 0b01111, 0,
2575 IIC_VQUNAiD, IIC_VQUNAiQ, "vqneg", "s",
2576 int_arm_neon_vqneg>;
2578 // Vector Bit Counting Operations.
2580 // VCLS : Vector Count Leading Sign Bits
2581 defm VCLS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01000, 0,
2582 IIC_VCNTiD, IIC_VCNTiQ, "vcls", "s",
2584 // VCLZ : Vector Count Leading Zeros
2585 defm VCLZ : N2VInt_QHS<0b11, 0b11, 0b00, 0b01001, 0,
2586 IIC_VCNTiD, IIC_VCNTiQ, "vclz", "i",
2588 // VCNT : Vector Count One Bits
2589 def VCNTd : N2VDInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
2590 IIC_VCNTiD, "vcnt", "8",
2591 v8i8, v8i8, int_arm_neon_vcnt>;
2592 def VCNTq : N2VQInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
2593 IIC_VCNTiQ, "vcnt", "8",
2594 v16i8, v16i8, int_arm_neon_vcnt>;
2596 // Vector Swap -- for disassembly only.
2597 def VSWPd : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 0, 0,
2598 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
2599 "vswp", "$dst, $src", "", []>;
2600 def VSWPq : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 1, 0,
2601 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
2602 "vswp", "$dst, $src", "", []>;
2604 // Vector Move Operations.
2606 // VMOV : Vector Move (Register)
2608 def VMOVDneon: N3VX<0, 0, 0b10, 0b0001, 0, 1, (outs DPR:$dst), (ins DPR:$src),
2609 IIC_VMOVD, "vmov", "$dst, $src", "", []>;
2610 def VMOVQ : N3VX<0, 0, 0b10, 0b0001, 1, 1, (outs QPR:$dst), (ins QPR:$src),
2611 IIC_VMOVD, "vmov", "$dst, $src", "", []>;
2613 // VMOV : Vector Move (Immediate)
2615 // VMOV_get_imm8 xform function: convert build_vector to VMOV.i8 imm.
2616 def VMOV_get_imm8 : SDNodeXForm<build_vector, [{
2617 return ARM::getVMOVImm(N, 1, *CurDAG);
2619 def vmovImm8 : PatLeaf<(build_vector), [{
2620 return ARM::getVMOVImm(N, 1, *CurDAG).getNode() != 0;
2623 // VMOV_get_imm16 xform function: convert build_vector to VMOV.i16 imm.
2624 def VMOV_get_imm16 : SDNodeXForm<build_vector, [{
2625 return ARM::getVMOVImm(N, 2, *CurDAG);
2627 def vmovImm16 : PatLeaf<(build_vector), [{
2628 return ARM::getVMOVImm(N, 2, *CurDAG).getNode() != 0;
2629 }], VMOV_get_imm16>;
2631 // VMOV_get_imm32 xform function: convert build_vector to VMOV.i32 imm.
2632 def VMOV_get_imm32 : SDNodeXForm<build_vector, [{
2633 return ARM::getVMOVImm(N, 4, *CurDAG);
2635 def vmovImm32 : PatLeaf<(build_vector), [{
2636 return ARM::getVMOVImm(N, 4, *CurDAG).getNode() != 0;
2637 }], VMOV_get_imm32>;
2639 // VMOV_get_imm64 xform function: convert build_vector to VMOV.i64 imm.
2640 def VMOV_get_imm64 : SDNodeXForm<build_vector, [{
2641 return ARM::getVMOVImm(N, 8, *CurDAG);
2643 def vmovImm64 : PatLeaf<(build_vector), [{
2644 return ARM::getVMOVImm(N, 8, *CurDAG).getNode() != 0;
2645 }], VMOV_get_imm64>;
2647 // Note: Some of the cmode bits in the following VMOV instructions need to
2648 // be encoded based on the immed values.
2650 def VMOVv8i8 : N1ModImm<1, 0b000, 0b1110, 0, 0, 0, 1, (outs DPR:$dst),
2651 (ins h8imm:$SIMM), IIC_VMOVImm,
2652 "vmov", "i8", "$dst, $SIMM", "",
2653 [(set DPR:$dst, (v8i8 vmovImm8:$SIMM))]>;
2654 def VMOVv16i8 : N1ModImm<1, 0b000, 0b1110, 0, 1, 0, 1, (outs QPR:$dst),
2655 (ins h8imm:$SIMM), IIC_VMOVImm,
2656 "vmov", "i8", "$dst, $SIMM", "",
2657 [(set QPR:$dst, (v16i8 vmovImm8:$SIMM))]>;
2659 def VMOVv4i16 : N1ModImm<1, 0b000, {1,0,?,?}, 0, 0, {?}, 1, (outs DPR:$dst),
2660 (ins h16imm:$SIMM), IIC_VMOVImm,
2661 "vmov", "i16", "$dst, $SIMM", "",
2662 [(set DPR:$dst, (v4i16 vmovImm16:$SIMM))]>;
2663 def VMOVv8i16 : N1ModImm<1, 0b000, {1,0,?,?}, 0, 1, {?}, 1, (outs QPR:$dst),
2664 (ins h16imm:$SIMM), IIC_VMOVImm,
2665 "vmov", "i16", "$dst, $SIMM", "",
2666 [(set QPR:$dst, (v8i16 vmovImm16:$SIMM))]>;
2668 def VMOVv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, {?}, 1, (outs DPR:$dst),
2669 (ins h32imm:$SIMM), IIC_VMOVImm,
2670 "vmov", "i32", "$dst, $SIMM", "",
2671 [(set DPR:$dst, (v2i32 vmovImm32:$SIMM))]>;
2672 def VMOVv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, {?}, 1, (outs QPR:$dst),
2673 (ins h32imm:$SIMM), IIC_VMOVImm,
2674 "vmov", "i32", "$dst, $SIMM", "",
2675 [(set QPR:$dst, (v4i32 vmovImm32:$SIMM))]>;
2677 def VMOVv1i64 : N1ModImm<1, 0b000, 0b1110, 0, 0, 1, 1, (outs DPR:$dst),
2678 (ins h64imm:$SIMM), IIC_VMOVImm,
2679 "vmov", "i64", "$dst, $SIMM", "",
2680 [(set DPR:$dst, (v1i64 vmovImm64:$SIMM))]>;
2681 def VMOVv2i64 : N1ModImm<1, 0b000, 0b1110, 0, 1, 1, 1, (outs QPR:$dst),
2682 (ins h64imm:$SIMM), IIC_VMOVImm,
2683 "vmov", "i64", "$dst, $SIMM", "",
2684 [(set QPR:$dst, (v2i64 vmovImm64:$SIMM))]>;
2686 // VMOV : Vector Get Lane (move scalar to ARM core register)
2688 def VGETLNs8 : NVGetLane<{1,1,1,0,0,1,?,1}, 0b1011, {?,?},
2689 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2690 IIC_VMOVSI, "vmov", "s8", "$dst, $src[$lane]",
2691 [(set GPR:$dst, (NEONvgetlanes (v8i8 DPR:$src),
2693 def VGETLNs16 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, {?,1},
2694 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2695 IIC_VMOVSI, "vmov", "s16", "$dst, $src[$lane]",
2696 [(set GPR:$dst, (NEONvgetlanes (v4i16 DPR:$src),
2698 def VGETLNu8 : NVGetLane<{1,1,1,0,1,1,?,1}, 0b1011, {?,?},
2699 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2700 IIC_VMOVSI, "vmov", "u8", "$dst, $src[$lane]",
2701 [(set GPR:$dst, (NEONvgetlaneu (v8i8 DPR:$src),
2703 def VGETLNu16 : NVGetLane<{1,1,1,0,1,0,?,1}, 0b1011, {?,1},
2704 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2705 IIC_VMOVSI, "vmov", "u16", "$dst, $src[$lane]",
2706 [(set GPR:$dst, (NEONvgetlaneu (v4i16 DPR:$src),
2708 def VGETLNi32 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, 0b00,
2709 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2710 IIC_VMOVSI, "vmov", "32", "$dst, $src[$lane]",
2711 [(set GPR:$dst, (extractelt (v2i32 DPR:$src),
2713 // def VGETLNf32: see FMRDH and FMRDL in ARMInstrVFP.td
2714 def : Pat<(NEONvgetlanes (v16i8 QPR:$src), imm:$lane),
2715 (VGETLNs8 (v8i8 (EXTRACT_SUBREG QPR:$src,
2716 (DSubReg_i8_reg imm:$lane))),
2717 (SubReg_i8_lane imm:$lane))>;
2718 def : Pat<(NEONvgetlanes (v8i16 QPR:$src), imm:$lane),
2719 (VGETLNs16 (v4i16 (EXTRACT_SUBREG QPR:$src,
2720 (DSubReg_i16_reg imm:$lane))),
2721 (SubReg_i16_lane imm:$lane))>;
2722 def : Pat<(NEONvgetlaneu (v16i8 QPR:$src), imm:$lane),
2723 (VGETLNu8 (v8i8 (EXTRACT_SUBREG QPR:$src,
2724 (DSubReg_i8_reg imm:$lane))),
2725 (SubReg_i8_lane imm:$lane))>;
2726 def : Pat<(NEONvgetlaneu (v8i16 QPR:$src), imm:$lane),
2727 (VGETLNu16 (v4i16 (EXTRACT_SUBREG QPR:$src,
2728 (DSubReg_i16_reg imm:$lane))),
2729 (SubReg_i16_lane imm:$lane))>;
2730 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
2731 (VGETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src,
2732 (DSubReg_i32_reg imm:$lane))),
2733 (SubReg_i32_lane imm:$lane))>;
2734 def : Pat<(extractelt (v2f32 DPR:$src1), imm:$src2),
2735 (EXTRACT_SUBREG (v2f32 (COPY_TO_REGCLASS (v2f32 DPR:$src1),DPR_VFP2)),
2736 (SSubReg_f32_reg imm:$src2))>;
2737 def : Pat<(extractelt (v4f32 QPR:$src1), imm:$src2),
2738 (EXTRACT_SUBREG (v4f32 (COPY_TO_REGCLASS (v4f32 QPR:$src1),QPR_VFP2)),
2739 (SSubReg_f32_reg imm:$src2))>;
2740 //def : Pat<(extractelt (v2i64 QPR:$src1), imm:$src2),
2741 // (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
2742 def : Pat<(extractelt (v2f64 QPR:$src1), imm:$src2),
2743 (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
2746 // VMOV : Vector Set Lane (move ARM core register to scalar)
2748 let Constraints = "$src1 = $dst" in {
2749 def VSETLNi8 : NVSetLane<{1,1,1,0,0,1,?,0}, 0b1011, {?,?}, (outs DPR:$dst),
2750 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
2751 IIC_VMOVISL, "vmov", "8", "$dst[$lane], $src2",
2752 [(set DPR:$dst, (vector_insert (v8i8 DPR:$src1),
2753 GPR:$src2, imm:$lane))]>;
2754 def VSETLNi16 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, {?,1}, (outs DPR:$dst),
2755 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
2756 IIC_VMOVISL, "vmov", "16", "$dst[$lane], $src2",
2757 [(set DPR:$dst, (vector_insert (v4i16 DPR:$src1),
2758 GPR:$src2, imm:$lane))]>;
2759 def VSETLNi32 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, 0b00, (outs DPR:$dst),
2760 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
2761 IIC_VMOVISL, "vmov", "32", "$dst[$lane], $src2",
2762 [(set DPR:$dst, (insertelt (v2i32 DPR:$src1),
2763 GPR:$src2, imm:$lane))]>;
2765 def : Pat<(vector_insert (v16i8 QPR:$src1), GPR:$src2, imm:$lane),
2766 (v16i8 (INSERT_SUBREG QPR:$src1,
2767 (v8i8 (VSETLNi8 (v8i8 (EXTRACT_SUBREG QPR:$src1,
2768 (DSubReg_i8_reg imm:$lane))),
2769 GPR:$src2, (SubReg_i8_lane imm:$lane))),
2770 (DSubReg_i8_reg imm:$lane)))>;
2771 def : Pat<(vector_insert (v8i16 QPR:$src1), GPR:$src2, imm:$lane),
2772 (v8i16 (INSERT_SUBREG QPR:$src1,
2773 (v4i16 (VSETLNi16 (v4i16 (EXTRACT_SUBREG QPR:$src1,
2774 (DSubReg_i16_reg imm:$lane))),
2775 GPR:$src2, (SubReg_i16_lane imm:$lane))),
2776 (DSubReg_i16_reg imm:$lane)))>;
2777 def : Pat<(insertelt (v4i32 QPR:$src1), GPR:$src2, imm:$lane),
2778 (v4i32 (INSERT_SUBREG QPR:$src1,
2779 (v2i32 (VSETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src1,
2780 (DSubReg_i32_reg imm:$lane))),
2781 GPR:$src2, (SubReg_i32_lane imm:$lane))),
2782 (DSubReg_i32_reg imm:$lane)))>;
2784 def : Pat<(v2f32 (insertelt DPR:$src1, SPR:$src2, imm:$src3)),
2785 (INSERT_SUBREG (v2f32 (COPY_TO_REGCLASS DPR:$src1, DPR_VFP2)),
2786 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
2787 def : Pat<(v4f32 (insertelt QPR:$src1, SPR:$src2, imm:$src3)),
2788 (INSERT_SUBREG (v4f32 (COPY_TO_REGCLASS QPR:$src1, QPR_VFP2)),
2789 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
2791 //def : Pat<(v2i64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
2792 // (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
2793 def : Pat<(v2f64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
2794 (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
2796 def : Pat<(v2f32 (scalar_to_vector SPR:$src)),
2797 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$src, arm_ssubreg_0)>;
2798 def : Pat<(v2f64 (scalar_to_vector (f64 DPR:$src))),
2799 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), DPR:$src, arm_dsubreg_0)>;
2800 def : Pat<(v4f32 (scalar_to_vector SPR:$src)),
2801 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), SPR:$src, arm_ssubreg_0)>;
2803 def : Pat<(v8i8 (scalar_to_vector GPR:$src)),
2804 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
2805 def : Pat<(v4i16 (scalar_to_vector GPR:$src)),
2806 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
2807 def : Pat<(v2i32 (scalar_to_vector GPR:$src)),
2808 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
2810 def : Pat<(v16i8 (scalar_to_vector GPR:$src)),
2811 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
2812 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
2814 def : Pat<(v8i16 (scalar_to_vector GPR:$src)),
2815 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
2816 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
2818 def : Pat<(v4i32 (scalar_to_vector GPR:$src)),
2819 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
2820 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
2823 // VDUP : Vector Duplicate (from ARM core register to all elements)
2825 class VDUPD<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
2826 : NVDup<opcod1, 0b1011, opcod3, (outs DPR:$dst), (ins GPR:$src),
2827 IIC_VMOVIS, "vdup", Dt, "$dst, $src",
2828 [(set DPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
2829 class VDUPQ<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
2830 : NVDup<opcod1, 0b1011, opcod3, (outs QPR:$dst), (ins GPR:$src),
2831 IIC_VMOVIS, "vdup", Dt, "$dst, $src",
2832 [(set QPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
2834 def VDUP8d : VDUPD<0b11101100, 0b00, "8", v8i8>;
2835 def VDUP16d : VDUPD<0b11101000, 0b01, "16", v4i16>;
2836 def VDUP32d : VDUPD<0b11101000, 0b00, "32", v2i32>;
2837 def VDUP8q : VDUPQ<0b11101110, 0b00, "8", v16i8>;
2838 def VDUP16q : VDUPQ<0b11101010, 0b01, "16", v8i16>;
2839 def VDUP32q : VDUPQ<0b11101010, 0b00, "32", v4i32>;
2841 def VDUPfd : NVDup<0b11101000, 0b1011, 0b00, (outs DPR:$dst), (ins GPR:$src),
2842 IIC_VMOVIS, "vdup", "32", "$dst, $src",
2843 [(set DPR:$dst, (v2f32 (NEONvdup
2844 (f32 (bitconvert GPR:$src)))))]>;
2845 def VDUPfq : NVDup<0b11101010, 0b1011, 0b00, (outs QPR:$dst), (ins GPR:$src),
2846 IIC_VMOVIS, "vdup", "32", "$dst, $src",
2847 [(set QPR:$dst, (v4f32 (NEONvdup
2848 (f32 (bitconvert GPR:$src)))))]>;
2850 // VDUP : Vector Duplicate Lane (from scalar to all elements)
2852 class VDUPLND<bits<2> op19_18, bits<2> op17_16,
2853 string OpcodeStr, string Dt, ValueType Ty>
2854 : N2V<0b11, 0b11, op19_18, op17_16, 0b11000, 0, 0,
2855 (outs DPR:$dst), (ins DPR:$src, nohash_imm:$lane), IIC_VMOVD,
2856 OpcodeStr, Dt, "$dst, $src[$lane]", "",
2857 [(set DPR:$dst, (Ty (NEONvduplane (Ty DPR:$src), imm:$lane)))]>;
2859 class VDUPLNQ<bits<2> op19_18, bits<2> op17_16, string OpcodeStr, string Dt,
2860 ValueType ResTy, ValueType OpTy>
2861 : N2V<0b11, 0b11, op19_18, op17_16, 0b11000, 1, 0,
2862 (outs QPR:$dst), (ins DPR:$src, nohash_imm:$lane), IIC_VMOVD,
2863 OpcodeStr, Dt, "$dst, $src[$lane]", "",
2864 [(set QPR:$dst, (ResTy (NEONvduplane (OpTy DPR:$src), imm:$lane)))]>;
2866 // Inst{19-16} is partially specified depending on the element size.
2868 def VDUPLN8d : VDUPLND<{?,?}, {?,1}, "vdup", "8", v8i8>;
2869 def VDUPLN16d : VDUPLND<{?,?}, {1,0}, "vdup", "16", v4i16>;
2870 def VDUPLN32d : VDUPLND<{?,1}, {0,0}, "vdup", "32", v2i32>;
2871 def VDUPLNfd : VDUPLND<{?,1}, {0,0}, "vdup", "32", v2f32>;
2872 def VDUPLN8q : VDUPLNQ<{?,?}, {?,1}, "vdup", "8", v16i8, v8i8>;
2873 def VDUPLN16q : VDUPLNQ<{?,?}, {1,0}, "vdup", "16", v8i16, v4i16>;
2874 def VDUPLN32q : VDUPLNQ<{?,1}, {0,0}, "vdup", "32", v4i32, v2i32>;
2875 def VDUPLNfq : VDUPLNQ<{?,1}, {0,0}, "vdup", "32", v4f32, v2f32>;
2877 def : Pat<(v16i8 (NEONvduplane (v16i8 QPR:$src), imm:$lane)),
2878 (v16i8 (VDUPLN8q (v8i8 (EXTRACT_SUBREG QPR:$src,
2879 (DSubReg_i8_reg imm:$lane))),
2880 (SubReg_i8_lane imm:$lane)))>;
2881 def : Pat<(v8i16 (NEONvduplane (v8i16 QPR:$src), imm:$lane)),
2882 (v8i16 (VDUPLN16q (v4i16 (EXTRACT_SUBREG QPR:$src,
2883 (DSubReg_i16_reg imm:$lane))),
2884 (SubReg_i16_lane imm:$lane)))>;
2885 def : Pat<(v4i32 (NEONvduplane (v4i32 QPR:$src), imm:$lane)),
2886 (v4i32 (VDUPLN32q (v2i32 (EXTRACT_SUBREG QPR:$src,
2887 (DSubReg_i32_reg imm:$lane))),
2888 (SubReg_i32_lane imm:$lane)))>;
2889 def : Pat<(v4f32 (NEONvduplane (v4f32 QPR:$src), imm:$lane)),
2890 (v4f32 (VDUPLNfq (v2f32 (EXTRACT_SUBREG QPR:$src,
2891 (DSubReg_i32_reg imm:$lane))),
2892 (SubReg_i32_lane imm:$lane)))>;
2894 def VDUPfdf : N2V<0b11, 0b11, {?,1}, {0,0}, 0b11000, 0, 0,
2895 (outs DPR:$dst), (ins SPR:$src),
2896 IIC_VMOVD, "vdup", "32", "$dst, ${src:lane}", "",
2897 [(set DPR:$dst, (v2f32 (NEONvdup (f32 SPR:$src))))]>;
2899 def VDUPfqf : N2V<0b11, 0b11, {?,1}, {0,0}, 0b11000, 1, 0,
2900 (outs QPR:$dst), (ins SPR:$src),
2901 IIC_VMOVD, "vdup", "32", "$dst, ${src:lane}", "",
2902 [(set QPR:$dst, (v4f32 (NEONvdup (f32 SPR:$src))))]>;
2904 def : Pat<(v2i64 (NEONvduplane (v2i64 QPR:$src), imm:$lane)),
2905 (INSERT_SUBREG QPR:$src,
2906 (i64 (EXTRACT_SUBREG QPR:$src,
2907 (DSubReg_f64_reg imm:$lane))),
2908 (DSubReg_f64_other_reg imm:$lane))>;
2909 def : Pat<(v2f64 (NEONvduplane (v2f64 QPR:$src), imm:$lane)),
2910 (INSERT_SUBREG QPR:$src,
2911 (f64 (EXTRACT_SUBREG QPR:$src,
2912 (DSubReg_f64_reg imm:$lane))),
2913 (DSubReg_f64_other_reg imm:$lane))>;
2915 // VMOVN : Vector Narrowing Move
2916 defm VMOVN : N2VNInt_HSD<0b11,0b11,0b10,0b00100,0,0, IIC_VMOVD,
2917 "vmovn", "i", int_arm_neon_vmovn>;
2918 // VQMOVN : Vector Saturating Narrowing Move
2919 defm VQMOVNs : N2VNInt_HSD<0b11,0b11,0b10,0b00101,0,0, IIC_VQUNAiD,
2920 "vqmovn", "s", int_arm_neon_vqmovns>;
2921 defm VQMOVNu : N2VNInt_HSD<0b11,0b11,0b10,0b00101,1,0, IIC_VQUNAiD,
2922 "vqmovn", "u", int_arm_neon_vqmovnu>;
2923 defm VQMOVNsu : N2VNInt_HSD<0b11,0b11,0b10,0b00100,1,0, IIC_VQUNAiD,
2924 "vqmovun", "s", int_arm_neon_vqmovnsu>;
2925 // VMOVL : Vector Lengthening Move
2926 defm VMOVLs : N2VLInt_QHS<0b01,0b10100,0,1, "vmovl", "s",
2927 int_arm_neon_vmovls>;
2928 defm VMOVLu : N2VLInt_QHS<0b11,0b10100,0,1, "vmovl", "u",
2929 int_arm_neon_vmovlu>;
2931 // Vector Conversions.
2933 // VCVT : Vector Convert Between Floating-Point and Integers
2934 def VCVTf2sd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
2935 v2i32, v2f32, fp_to_sint>;
2936 def VCVTf2ud : N2VD<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
2937 v2i32, v2f32, fp_to_uint>;
2938 def VCVTs2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
2939 v2f32, v2i32, sint_to_fp>;
2940 def VCVTu2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
2941 v2f32, v2i32, uint_to_fp>;
2943 def VCVTf2sq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
2944 v4i32, v4f32, fp_to_sint>;
2945 def VCVTf2uq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
2946 v4i32, v4f32, fp_to_uint>;
2947 def VCVTs2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
2948 v4f32, v4i32, sint_to_fp>;
2949 def VCVTu2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
2950 v4f32, v4i32, uint_to_fp>;
2952 // VCVT : Vector Convert Between Floating-Point and Fixed-Point.
2953 def VCVTf2xsd : N2VCvtD<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
2954 v2i32, v2f32, int_arm_neon_vcvtfp2fxs>;
2955 def VCVTf2xud : N2VCvtD<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
2956 v2i32, v2f32, int_arm_neon_vcvtfp2fxu>;
2957 def VCVTxs2fd : N2VCvtD<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
2958 v2f32, v2i32, int_arm_neon_vcvtfxs2fp>;
2959 def VCVTxu2fd : N2VCvtD<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
2960 v2f32, v2i32, int_arm_neon_vcvtfxu2fp>;
2962 def VCVTf2xsq : N2VCvtQ<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
2963 v4i32, v4f32, int_arm_neon_vcvtfp2fxs>;
2964 def VCVTf2xuq : N2VCvtQ<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
2965 v4i32, v4f32, int_arm_neon_vcvtfp2fxu>;
2966 def VCVTxs2fq : N2VCvtQ<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
2967 v4f32, v4i32, int_arm_neon_vcvtfxs2fp>;
2968 def VCVTxu2fq : N2VCvtQ<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
2969 v4f32, v4i32, int_arm_neon_vcvtfxu2fp>;
2973 // VREV64 : Vector Reverse elements within 64-bit doublewords
2975 class VREV64D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
2976 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 0, 0, (outs DPR:$dst),
2977 (ins DPR:$src), IIC_VMOVD,
2978 OpcodeStr, Dt, "$dst, $src", "",
2979 [(set DPR:$dst, (Ty (NEONvrev64 (Ty DPR:$src))))]>;
2980 class VREV64Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
2981 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 1, 0, (outs QPR:$dst),
2982 (ins QPR:$src), IIC_VMOVD,
2983 OpcodeStr, Dt, "$dst, $src", "",
2984 [(set QPR:$dst, (Ty (NEONvrev64 (Ty QPR:$src))))]>;
2986 def VREV64d8 : VREV64D<0b00, "vrev64", "8", v8i8>;
2987 def VREV64d16 : VREV64D<0b01, "vrev64", "16", v4i16>;
2988 def VREV64d32 : VREV64D<0b10, "vrev64", "32", v2i32>;
2989 def VREV64df : VREV64D<0b10, "vrev64", "32", v2f32>;
2991 def VREV64q8 : VREV64Q<0b00, "vrev64", "8", v16i8>;
2992 def VREV64q16 : VREV64Q<0b01, "vrev64", "16", v8i16>;
2993 def VREV64q32 : VREV64Q<0b10, "vrev64", "32", v4i32>;
2994 def VREV64qf : VREV64Q<0b10, "vrev64", "32", v4f32>;
2996 // VREV32 : Vector Reverse elements within 32-bit words
2998 class VREV32D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
2999 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 0, 0, (outs DPR:$dst),
3000 (ins DPR:$src), IIC_VMOVD,
3001 OpcodeStr, Dt, "$dst, $src", "",
3002 [(set DPR:$dst, (Ty (NEONvrev32 (Ty DPR:$src))))]>;
3003 class VREV32Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3004 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 1, 0, (outs QPR:$dst),
3005 (ins QPR:$src), IIC_VMOVD,
3006 OpcodeStr, Dt, "$dst, $src", "",
3007 [(set QPR:$dst, (Ty (NEONvrev32 (Ty QPR:$src))))]>;
3009 def VREV32d8 : VREV32D<0b00, "vrev32", "8", v8i8>;
3010 def VREV32d16 : VREV32D<0b01, "vrev32", "16", v4i16>;
3012 def VREV32q8 : VREV32Q<0b00, "vrev32", "8", v16i8>;
3013 def VREV32q16 : VREV32Q<0b01, "vrev32", "16", v8i16>;
3015 // VREV16 : Vector Reverse elements within 16-bit halfwords
3017 class VREV16D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3018 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 0, 0, (outs DPR:$dst),
3019 (ins DPR:$src), IIC_VMOVD,
3020 OpcodeStr, Dt, "$dst, $src", "",
3021 [(set DPR:$dst, (Ty (NEONvrev16 (Ty DPR:$src))))]>;
3022 class VREV16Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3023 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 1, 0, (outs QPR:$dst),
3024 (ins QPR:$src), IIC_VMOVD,
3025 OpcodeStr, Dt, "$dst, $src", "",
3026 [(set QPR:$dst, (Ty (NEONvrev16 (Ty QPR:$src))))]>;
3028 def VREV16d8 : VREV16D<0b00, "vrev16", "8", v8i8>;
3029 def VREV16q8 : VREV16Q<0b00, "vrev16", "8", v16i8>;
3031 // Other Vector Shuffles.
3033 // VEXT : Vector Extract
3035 class VEXTd<string OpcodeStr, string Dt, ValueType Ty>
3036 : N3V<0,1,0b11,{?,?,?,?},0,0, (outs DPR:$dst),
3037 (ins DPR:$lhs, DPR:$rhs, i32imm:$index), IIC_VEXTD,
3038 OpcodeStr, Dt, "$dst, $lhs, $rhs, $index", "",
3039 [(set DPR:$dst, (Ty (NEONvext (Ty DPR:$lhs),
3040 (Ty DPR:$rhs), imm:$index)))]>;
3042 class VEXTq<string OpcodeStr, string Dt, ValueType Ty>
3043 : N3V<0,1,0b11,{?,?,?,?},1,0, (outs QPR:$dst),
3044 (ins QPR:$lhs, QPR:$rhs, i32imm:$index), IIC_VEXTQ,
3045 OpcodeStr, Dt, "$dst, $lhs, $rhs, $index", "",
3046 [(set QPR:$dst, (Ty (NEONvext (Ty QPR:$lhs),
3047 (Ty QPR:$rhs), imm:$index)))]>;
3049 def VEXTd8 : VEXTd<"vext", "8", v8i8>;
3050 def VEXTd16 : VEXTd<"vext", "16", v4i16>;
3051 def VEXTd32 : VEXTd<"vext", "32", v2i32>;
3052 def VEXTdf : VEXTd<"vext", "32", v2f32>;
3054 def VEXTq8 : VEXTq<"vext", "8", v16i8>;
3055 def VEXTq16 : VEXTq<"vext", "16", v8i16>;
3056 def VEXTq32 : VEXTq<"vext", "32", v4i32>;
3057 def VEXTqf : VEXTq<"vext", "32", v4f32>;
3059 // VTRN : Vector Transpose
3061 def VTRNd8 : N2VDShuffle<0b00, 0b00001, "vtrn", "8">;
3062 def VTRNd16 : N2VDShuffle<0b01, 0b00001, "vtrn", "16">;
3063 def VTRNd32 : N2VDShuffle<0b10, 0b00001, "vtrn", "32">;
3065 def VTRNq8 : N2VQShuffle<0b00, 0b00001, IIC_VPERMQ, "vtrn", "8">;
3066 def VTRNq16 : N2VQShuffle<0b01, 0b00001, IIC_VPERMQ, "vtrn", "16">;
3067 def VTRNq32 : N2VQShuffle<0b10, 0b00001, IIC_VPERMQ, "vtrn", "32">;
3069 // VUZP : Vector Unzip (Deinterleave)
3071 def VUZPd8 : N2VDShuffle<0b00, 0b00010, "vuzp", "8">;
3072 def VUZPd16 : N2VDShuffle<0b01, 0b00010, "vuzp", "16">;
3073 def VUZPd32 : N2VDShuffle<0b10, 0b00010, "vuzp", "32">;
3075 def VUZPq8 : N2VQShuffle<0b00, 0b00010, IIC_VPERMQ3, "vuzp", "8">;
3076 def VUZPq16 : N2VQShuffle<0b01, 0b00010, IIC_VPERMQ3, "vuzp", "16">;
3077 def VUZPq32 : N2VQShuffle<0b10, 0b00010, IIC_VPERMQ3, "vuzp", "32">;
3079 // VZIP : Vector Zip (Interleave)
3081 def VZIPd8 : N2VDShuffle<0b00, 0b00011, "vzip", "8">;
3082 def VZIPd16 : N2VDShuffle<0b01, 0b00011, "vzip", "16">;
3083 def VZIPd32 : N2VDShuffle<0b10, 0b00011, "vzip", "32">;
3085 def VZIPq8 : N2VQShuffle<0b00, 0b00011, IIC_VPERMQ3, "vzip", "8">;
3086 def VZIPq16 : N2VQShuffle<0b01, 0b00011, IIC_VPERMQ3, "vzip", "16">;
3087 def VZIPq32 : N2VQShuffle<0b10, 0b00011, IIC_VPERMQ3, "vzip", "32">;
3089 // Vector Table Lookup and Table Extension.
3091 // VTBL : Vector Table Lookup
3093 : N3V<1,1,0b11,0b1000,0,0, (outs DPR:$dst),
3094 (ins DPR:$tbl1, DPR:$src), IIC_VTB1,
3095 "vtbl", "8", "$dst, \\{$tbl1\\}, $src", "",
3096 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl1 DPR:$tbl1, DPR:$src)))]>;
3097 let hasExtraSrcRegAllocReq = 1 in {
3099 : N3V<1,1,0b11,0b1001,0,0, (outs DPR:$dst),
3100 (ins DPR:$tbl1, DPR:$tbl2, DPR:$src), IIC_VTB2,
3101 "vtbl", "8", "$dst, \\{$tbl1, $tbl2\\}, $src", "",
3102 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl2
3103 DPR:$tbl1, DPR:$tbl2, DPR:$src)))]>;
3105 : N3V<1,1,0b11,0b1010,0,0, (outs DPR:$dst),
3106 (ins DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$src), IIC_VTB3,
3107 "vtbl", "8", "$dst, \\{$tbl1, $tbl2, $tbl3\\}, $src", "",
3108 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl3
3109 DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$src)))]>;
3111 : N3V<1,1,0b11,0b1011,0,0, (outs DPR:$dst),
3112 (ins DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$src), IIC_VTB4,
3113 "vtbl", "8", "$dst, \\{$tbl1, $tbl2, $tbl3, $tbl4\\}, $src", "",
3114 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl4 DPR:$tbl1, DPR:$tbl2,
3115 DPR:$tbl3, DPR:$tbl4, DPR:$src)))]>;
3116 } // hasExtraSrcRegAllocReq = 1
3118 // VTBX : Vector Table Extension
3120 : N3V<1,1,0b11,0b1000,1,0, (outs DPR:$dst),
3121 (ins DPR:$orig, DPR:$tbl1, DPR:$src), IIC_VTBX1,
3122 "vtbx", "8", "$dst, \\{$tbl1\\}, $src", "$orig = $dst",
3123 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx1
3124 DPR:$orig, DPR:$tbl1, DPR:$src)))]>;
3125 let hasExtraSrcRegAllocReq = 1 in {
3127 : N3V<1,1,0b11,0b1001,1,0, (outs DPR:$dst),
3128 (ins DPR:$orig, DPR:$tbl1, DPR:$tbl2, DPR:$src), IIC_VTBX2,
3129 "vtbx", "8", "$dst, \\{$tbl1, $tbl2\\}, $src", "$orig = $dst",
3130 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx2
3131 DPR:$orig, DPR:$tbl1, DPR:$tbl2, DPR:$src)))]>;
3133 : N3V<1,1,0b11,0b1010,1,0, (outs DPR:$dst),
3134 (ins DPR:$orig, DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$src), IIC_VTBX3,
3135 "vtbx", "8", "$dst, \\{$tbl1, $tbl2, $tbl3\\}, $src", "$orig = $dst",
3136 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx3 DPR:$orig, DPR:$tbl1,
3137 DPR:$tbl2, DPR:$tbl3, DPR:$src)))]>;
3139 : N3V<1,1,0b11,0b1011,1,0, (outs DPR:$dst), (ins DPR:$orig, DPR:$tbl1,
3140 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$src), IIC_VTBX4,
3141 "vtbx", "8", "$dst, \\{$tbl1, $tbl2, $tbl3, $tbl4\\}, $src",
3143 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx4 DPR:$orig, DPR:$tbl1,
3144 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$src)))]>;
3145 } // hasExtraSrcRegAllocReq = 1
3147 //===----------------------------------------------------------------------===//
3148 // NEON instructions for single-precision FP math
3149 //===----------------------------------------------------------------------===//
3151 class N2VSPat<SDNode OpNode, ValueType ResTy, ValueType OpTy, NeonI Inst>
3152 : NEONFPPat<(ResTy (OpNode SPR:$a)),
3153 (EXTRACT_SUBREG (OpTy (Inst (INSERT_SUBREG (OpTy (IMPLICIT_DEF)),
3154 SPR:$a, arm_ssubreg_0))),
3157 class N3VSPat<SDNode OpNode, NeonI Inst>
3158 : NEONFPPat<(f32 (OpNode SPR:$a, SPR:$b)),
3159 (EXTRACT_SUBREG (v2f32
3160 (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3161 SPR:$a, arm_ssubreg_0),
3162 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3163 SPR:$b, arm_ssubreg_0))),
3166 class N3VSMulOpPat<SDNode MulNode, SDNode OpNode, NeonI Inst>
3167 : NEONFPPat<(f32 (OpNode SPR:$acc, (f32 (MulNode SPR:$a, SPR:$b)))),
3168 (EXTRACT_SUBREG (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3169 SPR:$acc, arm_ssubreg_0),
3170 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3171 SPR:$a, arm_ssubreg_0),
3172 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3173 SPR:$b, arm_ssubreg_0)),
3176 // These need separate instructions because they must use DPR_VFP2 register
3177 // class which have SPR sub-registers.
3179 // Vector Add Operations used for single-precision FP
3180 let neverHasSideEffects = 1 in
3181 def VADDfd_sfp : N3VS<0,0,0b00,0b1101,0, "vadd", "f32", v2f32, v2f32, fadd, 1>;
3182 def : N3VSPat<fadd, VADDfd_sfp>;
3184 // Vector Sub Operations used for single-precision FP
3185 let neverHasSideEffects = 1 in
3186 def VSUBfd_sfp : N3VS<0,0,0b10,0b1101,0, "vsub", "f32", v2f32, v2f32, fsub, 0>;
3187 def : N3VSPat<fsub, VSUBfd_sfp>;
3189 // Vector Multiply Operations used for single-precision FP
3190 let neverHasSideEffects = 1 in
3191 def VMULfd_sfp : N3VS<1,0,0b00,0b1101,1, "vmul", "f32", v2f32, v2f32, fmul, 1>;
3192 def : N3VSPat<fmul, VMULfd_sfp>;
3194 // Vector Multiply-Accumulate/Subtract used for single-precision FP
3195 // vml[as].f32 can cause 4-8 cycle stalls in following ASIMD instructions, so
3196 // we want to avoid them for now. e.g., alternating vmla/vadd instructions.
3198 //let neverHasSideEffects = 1 in
3199 //def VMLAfd_sfp : N3VSMulOp<0,0,0b00,0b1101,1, IIC_VMACD, "vmla", "f32",
3200 // v2f32, fmul, fadd>;
3201 //def : N3VSMulOpPat<fmul, fadd, VMLAfd_sfp>;
3203 //let neverHasSideEffects = 1 in
3204 //def VMLSfd_sfp : N3VSMulOp<0,0,0b10,0b1101,1, IIC_VMACD, "vmls", "f32",
3205 // v2f32, fmul, fsub>;
3206 //def : N3VSMulOpPat<fmul, fsub, VMLSfd_sfp>;
3208 // Vector Absolute used for single-precision FP
3209 let neverHasSideEffects = 1 in
3210 def VABSfd_sfp : N2V<0b11, 0b11, 0b10, 0b01, 0b01110, 0, 0,
3211 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), IIC_VUNAD,
3212 "vabs", "f32", "$dst, $src", "", []>;
3213 def : N2VSPat<fabs, f32, v2f32, VABSfd_sfp>;
3215 // Vector Negate used for single-precision FP
3216 let neverHasSideEffects = 1 in
3217 def VNEGfd_sfp : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
3218 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), IIC_VUNAD,
3219 "vneg", "f32", "$dst, $src", "", []>;
3220 def : N2VSPat<fneg, f32, v2f32, VNEGfd_sfp>;
3222 // Vector Maximum used for single-precision FP
3223 let neverHasSideEffects = 1 in
3224 def VMAXfd_sfp : N3V<0, 0, 0b00, 0b1111, 0, 0, (outs DPR_VFP2:$dst),
3225 (ins DPR_VFP2:$src1, DPR_VFP2:$src2), IIC_VBIND,
3226 "vmax", "f32", "$dst, $src1, $src2", "", []>;
3227 def : N3VSPat<NEONfmax, VMAXfd_sfp>;
3229 // Vector Minimum used for single-precision FP
3230 let neverHasSideEffects = 1 in
3231 def VMINfd_sfp : N3V<0, 0, 0b00, 0b1111, 0, 0, (outs DPR_VFP2:$dst),
3232 (ins DPR_VFP2:$src1, DPR_VFP2:$src2), IIC_VBIND,
3233 "vmin", "f32", "$dst, $src1, $src2", "", []>;
3234 def : N3VSPat<NEONfmin, VMINfd_sfp>;
3236 // Vector Convert between single-precision FP and integer
3237 let neverHasSideEffects = 1 in
3238 def VCVTf2sd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
3239 v2i32, v2f32, fp_to_sint>;
3240 def : N2VSPat<arm_ftosi, f32, v2f32, VCVTf2sd_sfp>;
3242 let neverHasSideEffects = 1 in
3243 def VCVTf2ud_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
3244 v2i32, v2f32, fp_to_uint>;
3245 def : N2VSPat<arm_ftoui, f32, v2f32, VCVTf2ud_sfp>;
3247 let neverHasSideEffects = 1 in
3248 def VCVTs2fd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
3249 v2f32, v2i32, sint_to_fp>;
3250 def : N2VSPat<arm_sitof, f32, v2i32, VCVTs2fd_sfp>;
3252 let neverHasSideEffects = 1 in
3253 def VCVTu2fd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
3254 v2f32, v2i32, uint_to_fp>;
3255 def : N2VSPat<arm_uitof, f32, v2i32, VCVTu2fd_sfp>;
3257 //===----------------------------------------------------------------------===//
3258 // Non-Instruction Patterns
3259 //===----------------------------------------------------------------------===//
3262 def : Pat<(v1i64 (bitconvert (v2i32 DPR:$src))), (v1i64 DPR:$src)>;
3263 def : Pat<(v1i64 (bitconvert (v4i16 DPR:$src))), (v1i64 DPR:$src)>;
3264 def : Pat<(v1i64 (bitconvert (v8i8 DPR:$src))), (v1i64 DPR:$src)>;
3265 def : Pat<(v1i64 (bitconvert (f64 DPR:$src))), (v1i64 DPR:$src)>;
3266 def : Pat<(v1i64 (bitconvert (v2f32 DPR:$src))), (v1i64 DPR:$src)>;
3267 def : Pat<(v2i32 (bitconvert (v1i64 DPR:$src))), (v2i32 DPR:$src)>;
3268 def : Pat<(v2i32 (bitconvert (v4i16 DPR:$src))), (v2i32 DPR:$src)>;
3269 def : Pat<(v2i32 (bitconvert (v8i8 DPR:$src))), (v2i32 DPR:$src)>;
3270 def : Pat<(v2i32 (bitconvert (f64 DPR:$src))), (v2i32 DPR:$src)>;
3271 def : Pat<(v2i32 (bitconvert (v2f32 DPR:$src))), (v2i32 DPR:$src)>;
3272 def : Pat<(v4i16 (bitconvert (v1i64 DPR:$src))), (v4i16 DPR:$src)>;
3273 def : Pat<(v4i16 (bitconvert (v2i32 DPR:$src))), (v4i16 DPR:$src)>;
3274 def : Pat<(v4i16 (bitconvert (v8i8 DPR:$src))), (v4i16 DPR:$src)>;
3275 def : Pat<(v4i16 (bitconvert (f64 DPR:$src))), (v4i16 DPR:$src)>;
3276 def : Pat<(v4i16 (bitconvert (v2f32 DPR:$src))), (v4i16 DPR:$src)>;
3277 def : Pat<(v8i8 (bitconvert (v1i64 DPR:$src))), (v8i8 DPR:$src)>;
3278 def : Pat<(v8i8 (bitconvert (v2i32 DPR:$src))), (v8i8 DPR:$src)>;
3279 def : Pat<(v8i8 (bitconvert (v4i16 DPR:$src))), (v8i8 DPR:$src)>;
3280 def : Pat<(v8i8 (bitconvert (f64 DPR:$src))), (v8i8 DPR:$src)>;
3281 def : Pat<(v8i8 (bitconvert (v2f32 DPR:$src))), (v8i8 DPR:$src)>;
3282 def : Pat<(f64 (bitconvert (v1i64 DPR:$src))), (f64 DPR:$src)>;
3283 def : Pat<(f64 (bitconvert (v2i32 DPR:$src))), (f64 DPR:$src)>;
3284 def : Pat<(f64 (bitconvert (v4i16 DPR:$src))), (f64 DPR:$src)>;
3285 def : Pat<(f64 (bitconvert (v8i8 DPR:$src))), (f64 DPR:$src)>;
3286 def : Pat<(f64 (bitconvert (v2f32 DPR:$src))), (f64 DPR:$src)>;
3287 def : Pat<(v2f32 (bitconvert (f64 DPR:$src))), (v2f32 DPR:$src)>;
3288 def : Pat<(v2f32 (bitconvert (v1i64 DPR:$src))), (v2f32 DPR:$src)>;
3289 def : Pat<(v2f32 (bitconvert (v2i32 DPR:$src))), (v2f32 DPR:$src)>;
3290 def : Pat<(v2f32 (bitconvert (v4i16 DPR:$src))), (v2f32 DPR:$src)>;
3291 def : Pat<(v2f32 (bitconvert (v8i8 DPR:$src))), (v2f32 DPR:$src)>;
3293 def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (v2i64 QPR:$src)>;
3294 def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (v2i64 QPR:$src)>;
3295 def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (v2i64 QPR:$src)>;
3296 def : Pat<(v2i64 (bitconvert (v2f64 QPR:$src))), (v2i64 QPR:$src)>;
3297 def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (v2i64 QPR:$src)>;
3298 def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (v4i32 QPR:$src)>;
3299 def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (v4i32 QPR:$src)>;
3300 def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (v4i32 QPR:$src)>;
3301 def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (v4i32 QPR:$src)>;
3302 def : Pat<(v4i32 (bitconvert (v4f32 QPR:$src))), (v4i32 QPR:$src)>;
3303 def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (v8i16 QPR:$src)>;
3304 def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (v8i16 QPR:$src)>;
3305 def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (v8i16 QPR:$src)>;
3306 def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (v8i16 QPR:$src)>;
3307 def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (v8i16 QPR:$src)>;
3308 def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (v16i8 QPR:$src)>;
3309 def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (v16i8 QPR:$src)>;
3310 def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (v16i8 QPR:$src)>;
3311 def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (v16i8 QPR:$src)>;
3312 def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (v16i8 QPR:$src)>;
3313 def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (v4f32 QPR:$src)>;
3314 def : Pat<(v4f32 (bitconvert (v4i32 QPR:$src))), (v4f32 QPR:$src)>;
3315 def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (v4f32 QPR:$src)>;
3316 def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (v4f32 QPR:$src)>;
3317 def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (v4f32 QPR:$src)>;
3318 def : Pat<(v2f64 (bitconvert (v2i64 QPR:$src))), (v2f64 QPR:$src)>;
3319 def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (v2f64 QPR:$src)>;
3320 def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 QPR:$src)>;
3321 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 QPR:$src)>;
3322 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 QPR:$src)>;