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 = $dst2", []>;
392 def VLD2LNd8 : VLD2LN<0b0001, "8">;
393 def VLD2LNd16 : VLD2LN<0b0101, "16"> { let Inst{5} = 0; }
394 def VLD2LNd32 : VLD2LN<0b1001, "32"> { let Inst{6} = 0; }
396 // ...with double-spaced registers:
397 def VLD2LNq16 : VLD2LN<0b0101, "16"> { let Inst{5} = 1; }
398 def VLD2LNq32 : VLD2LN<0b1001, "32"> { let Inst{6} = 1; }
400 // ...alternate versions to be allocated odd register numbers:
401 def VLD2LNq16odd : VLD2LN<0b0101, "16"> { let Inst{5} = 1; }
402 def VLD2LNq32odd : VLD2LN<0b1001, "32"> { let Inst{6} = 1; }
404 // ...with address register writeback:
405 class VLD2LNWB<bits<4> op11_8, string Dt>
406 : NLdSt<1, 0b10, op11_8, {?,?,?,?}, (outs DPR:$dst1, DPR:$dst2, GPR:$wb),
407 (ins addrmode6:$addr,
408 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VLD2, "vld2", Dt,
409 "\\{$dst1[$lane], $dst2[$lane]\\}, $addr",
410 "$src1 = $dst1, $src2 = $dst2, $addr.addr = $wb", []>;
412 def VLD2LNd8_UPD : VLD2LNWB<0b0001, "8">;
413 def VLD2LNd16_UPD : VLD2LNWB<0b0101, "16"> { let Inst{5} = 0; }
414 def VLD2LNd32_UPD : VLD2LNWB<0b1001, "32"> { let Inst{6} = 0; }
416 def VLD2LNq16_UPD : VLD2LNWB<0b0101, "16"> { let Inst{5} = 1; }
417 def VLD2LNq32_UPD : VLD2LNWB<0b1001, "32"> { let Inst{6} = 1; }
419 // VLD3LN : Vector Load (single 3-element structure to one lane)
420 class VLD3LN<bits<4> op11_8, string Dt>
421 : NLdSt<1, 0b10, op11_8, {?,?,?,?}, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3),
422 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3,
423 nohash_imm:$lane), IIC_VLD3, "vld3", Dt,
424 "\\{$dst1[$lane], $dst2[$lane], $dst3[$lane]\\}, $addr",
425 "$src1 = $dst1, $src2 = $dst2, $src3 = $dst3", []>;
427 def VLD3LNd8 : VLD3LN<0b0010, "8"> { let Inst{4} = 0; }
428 def VLD3LNd16 : VLD3LN<0b0110, "16"> { let Inst{5-4} = 0b00; }
429 def VLD3LNd32 : VLD3LN<0b1010, "32"> { let Inst{6-4} = 0b000; }
431 // ...with double-spaced registers:
432 def VLD3LNq16 : VLD3LN<0b0110, "16"> { let Inst{5-4} = 0b10; }
433 def VLD3LNq32 : VLD3LN<0b1010, "32"> { let Inst{6-4} = 0b100; }
435 // ...alternate versions to be allocated odd register numbers:
436 def VLD3LNq16odd : VLD3LN<0b0110, "16"> { let Inst{5-4} = 0b10; }
437 def VLD3LNq32odd : VLD3LN<0b1010, "32"> { let Inst{6-4} = 0b100; }
439 // ...with address register writeback:
440 class VLD3LNWB<bits<4> op11_8, string Dt>
441 : NLdSt<1, 0b10, op11_8, {?,?,?,?},
442 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, GPR:$wb),
443 (ins addrmode6:$addr,
444 DPR:$src1, DPR:$src2, DPR:$src3, nohash_imm:$lane),
445 IIC_VLD3, "vld3", Dt,
446 "\\{$dst1[$lane], $dst2[$lane], $dst3[$lane]\\}, $addr",
447 "$src1 = $dst1, $src2 = $dst2, $src3 = $dst3, $addr.addr = $wb",
450 def VLD3LNd8_UPD : VLD3LNWB<0b0010, "8"> { let Inst{4} = 0; }
451 def VLD3LNd16_UPD : VLD3LNWB<0b0110, "16"> { let Inst{5-4} = 0b00; }
452 def VLD3LNd32_UPD : VLD3LNWB<0b1010, "32"> { let Inst{6-4} = 0b000; }
454 def VLD3LNq16_UPD : VLD3LNWB<0b0110, "16"> { let Inst{5-4} = 0b10; }
455 def VLD3LNq32_UPD : VLD3LNWB<0b1010, "32"> { let Inst{6-4} = 0b100; }
457 // VLD4LN : Vector Load (single 4-element structure to one lane)
458 class VLD4LN<bits<4> op11_8, string Dt>
459 : NLdSt<1, 0b10, op11_8, {?,?,?,?},
460 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
461 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
462 nohash_imm:$lane), IIC_VLD4, "vld4", Dt,
463 "\\{$dst1[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $addr",
464 "$src1 = $dst1, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4", []>;
466 def VLD4LNd8 : VLD4LN<0b0011, "8">;
467 def VLD4LNd16 : VLD4LN<0b0111, "16"> { let Inst{5} = 0; }
468 def VLD4LNd32 : VLD4LN<0b1011, "32"> { let Inst{6} = 0; }
470 // ...with double-spaced registers:
471 def VLD4LNq16 : VLD4LN<0b0111, "16"> { let Inst{5} = 1; }
472 def VLD4LNq32 : VLD4LN<0b1011, "32"> { let Inst{6} = 1; }
474 // ...alternate versions to be allocated odd register numbers:
475 def VLD4LNq16odd : VLD4LN<0b0111, "16"> { let Inst{5} = 1; }
476 def VLD4LNq32odd : VLD4LN<0b1011, "32"> { let Inst{6} = 1; }
478 // ...with address register writeback:
479 class VLD4LNWB<bits<4> op11_8, string Dt>
480 : NLdSt<1, 0b10, op11_8, {?,?,?,?},
481 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
482 (ins addrmode6:$addr,
483 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
484 IIC_VLD4, "vld4", Dt,
485 "\\{$dst1[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $addr",
486 "$src1 = $dst1, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4, $addr.addr = $wb",
489 def VLD4LNd8_UPD : VLD4LNWB<0b0011, "8">;
490 def VLD4LNd16_UPD : VLD4LNWB<0b0111, "16"> { let Inst{5} = 0; }
491 def VLD4LNd32_UPD : VLD4LNWB<0b1011, "32"> { let Inst{6} = 0; }
493 def VLD4LNq16_UPD : VLD4LNWB<0b0111, "16"> { let Inst{5} = 1; }
494 def VLD4LNq32_UPD : VLD4LNWB<0b1011, "32"> { let Inst{6} = 1; }
496 // VLD1DUP : Vector Load (single element to all lanes)
497 // VLD2DUP : Vector Load (single 2-element structure to all lanes)
498 // VLD3DUP : Vector Load (single 3-element structure to all lanes)
499 // VLD4DUP : Vector Load (single 4-element structure to all lanes)
500 // FIXME: Not yet implemented.
501 } // mayLoad = 1, hasExtraDefRegAllocReq = 1
503 // VST1 : Vector Store (multiple single elements)
504 class VST1D<bits<4> op7_4, string Dt, ValueType Ty>
505 : NLdSt<0,0b00,0b0111,op7_4, (outs), (ins addrmode6:$addr, DPR:$src), IIC_VST,
506 "vst1", Dt, "\\{$src\\}, $addr", "",
507 [(int_arm_neon_vst1 addrmode6:$addr, (Ty DPR:$src))]>;
508 class VST1Q<bits<4> op7_4, string Dt, ValueType Ty>
509 : NLdSt<0,0b00,0b1010,op7_4, (outs), (ins addrmode6:$addr, QPR:$src), IIC_VST,
510 "vst1", Dt, "${src:dregpair}, $addr", "",
511 [(int_arm_neon_vst1 addrmode6:$addr, (Ty QPR:$src))]>;
513 let hasExtraSrcRegAllocReq = 1 in {
514 def VST1d8 : VST1D<0b0000, "8", v8i8>;
515 def VST1d16 : VST1D<0b0100, "16", v4i16>;
516 def VST1d32 : VST1D<0b1000, "32", v2i32>;
517 def VST1df : VST1D<0b1000, "32", v2f32>;
518 def VST1d64 : VST1D<0b1100, "64", v1i64>;
520 def VST1q8 : VST1Q<0b0000, "8", v16i8>;
521 def VST1q16 : VST1Q<0b0100, "16", v8i16>;
522 def VST1q32 : VST1Q<0b1000, "32", v4i32>;
523 def VST1qf : VST1Q<0b1000, "32", v4f32>;
524 def VST1q64 : VST1Q<0b1100, "64", v2i64>;
525 } // hasExtraSrcRegAllocReq
527 let mayStore = 1, hasExtraSrcRegAllocReq = 1 in {
529 // ...with address register writeback:
530 class VST1DWB<bits<4> op7_4, string Dt>
531 : NLdSt<0, 0b00, 0b0111, op7_4, (outs GPR:$wb),
532 (ins addrmode6:$addr, DPR:$src), IIC_VST,
533 "vst1", Dt, "\\{$src\\}, $addr", "$addr.addr = $wb", []>;
534 class VST1QWB<bits<4> op7_4, string Dt>
535 : NLdSt<0, 0b00, 0b1010, op7_4, (outs GPR:$wb),
536 (ins addrmode6:$addr, QPR:$src), IIC_VST,
537 "vst1", Dt, "${src:dregpair}, $addr", "$addr.addr = $wb", []>;
539 def VST1d8_UPD : VST1DWB<0b0000, "8">;
540 def VST1d16_UPD : VST1DWB<0b0100, "16">;
541 def VST1d32_UPD : VST1DWB<0b1000, "32">;
542 def VST1d64_UPD : VST1DWB<0b1100, "64">;
544 def VST1q8_UPD : VST1QWB<0b0000, "8">;
545 def VST1q16_UPD : VST1QWB<0b0100, "16">;
546 def VST1q32_UPD : VST1QWB<0b1000, "32">;
547 def VST1q64_UPD : VST1QWB<0b1100, "64">;
549 // These (dreg triple/quadruple) are for disassembly only.
550 class VST1D3<bits<4> op7_4, string Dt>
551 : NLdSt<0, 0b00, 0b0110, op7_4, (outs),
552 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3),
553 IIC_VST, "vst1", Dt, "\\{$src1, $src2, $src3\\}, $addr", "",
554 [/* For disassembly only; pattern left blank */]>;
555 class VST1D4<bits<4> op7_4, string Dt>
556 : NLdSt<0, 0b00, 0b0010, op7_4, (outs),
557 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
558 IIC_VST, "vst1", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr", "",
559 [/* For disassembly only; pattern left blank */]>;
561 def VST1d8T : VST1D3<0b0000, "8">;
562 def VST1d16T : VST1D3<0b0100, "16">;
563 def VST1d32T : VST1D3<0b1000, "32">;
564 // VST1d64T : implemented as VST3d64
566 def VST1d8Q : VST1D4<0b0000, "8">;
567 def VST1d16Q : VST1D4<0b0100, "16">;
568 def VST1d32Q : VST1D4<0b1000, "32">;
569 // VST1d64Q : implemented as VST4d64
571 // ...with address register writeback:
572 class VST1D3WB<bits<4> op7_4, string Dt>
573 : NLdSt<0, 0b00, 0b0110, op7_4, (outs GPR:$wb),
574 (ins addrmode6:$addr,
575 DPR:$src1, DPR:$src2, DPR:$src3),
576 IIC_VST, "vst1", Dt, "\\{$src1, $src2, $src3\\}, $addr",
578 [/* For disassembly only; pattern left blank */]>;
579 class VST1D4WB<bits<4> op7_4, string Dt>
580 : NLdSt<0, 0b00, 0b0010, op7_4, (outs GPR:$wb),
581 (ins addrmode6:$addr,
582 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
583 IIC_VST, "vst1", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr",
585 [/* For disassembly only; pattern left blank */]>;
587 def VST1d8T_UPD : VST1D3WB<0b0000, "8">;
588 def VST1d16T_UPD : VST1D3WB<0b0100, "16">;
589 def VST1d32T_UPD : VST1D3WB<0b1000, "32">;
590 // VST1d64T_UPD : implemented as VST3d64_UPD
592 def VST1d8Q_UPD : VST1D4WB<0b0000, "8">;
593 def VST1d16Q_UPD : VST1D4WB<0b0100, "16">;
594 def VST1d32Q_UPD : VST1D4WB<0b1000, "32">;
595 // VST1d64Q_UPD : implemented as VST4d64_UPD
597 // VST2 : Vector Store (multiple 2-element structures)
598 class VST2D<bits<4> op11_8, bits<4> op7_4, string Dt>
599 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
600 (ins addrmode6:$addr, DPR:$src1, DPR:$src2),
601 IIC_VST, "vst2", Dt, "\\{$src1, $src2\\}, $addr", "", []>;
602 class VST2Q<bits<4> op7_4, string Dt>
603 : NLdSt<0, 0b00, 0b0011, op7_4, (outs),
604 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
605 IIC_VST, "vst2", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr",
608 def VST2d8 : VST2D<0b1000, 0b0000, "8">;
609 def VST2d16 : VST2D<0b1000, 0b0100, "16">;
610 def VST2d32 : VST2D<0b1000, 0b1000, "32">;
611 def VST2d64 : NLdSt<0,0b00,0b1010,0b1100, (outs),
612 (ins addrmode6:$addr, DPR:$src1, DPR:$src2), IIC_VST,
613 "vst1", "64", "\\{$src1, $src2\\}, $addr", "", []>;
615 def VST2q8 : VST2Q<0b0000, "8">;
616 def VST2q16 : VST2Q<0b0100, "16">;
617 def VST2q32 : VST2Q<0b1000, "32">;
619 // ...with double-spaced registers (for disassembly only):
620 def VST2b8 : VST2D<0b1001, 0b0000, "8">;
621 def VST2b16 : VST2D<0b1001, 0b0100, "16">;
622 def VST2b32 : VST2D<0b1001, 0b1000, "32">;
624 // VST3 : Vector Store (multiple 3-element structures)
625 class VST3D<bits<4> op11_8, bits<4> op7_4, string Dt>
626 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
627 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3), IIC_VST,
628 "vst3", Dt, "\\{$src1, $src2, $src3\\}, $addr", "", []>;
629 class VST3WB<bits<4> op7_4, string Dt>
630 : NLdSt<0,0b00,0b0101,op7_4, (outs GPR:$wb),
631 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3), IIC_VST,
632 "vst3", Dt, "\\{$src1, $src2, $src3\\}, $addr",
633 "$addr.addr = $wb", []>;
635 def VST3d8 : VST3D<0b0100, 0b0000, "8">;
636 def VST3d16 : VST3D<0b0100, 0b0100, "16">;
637 def VST3d32 : VST3D<0b0100, 0b1000, "32">;
638 def VST3d64 : NLdSt<0,0b00,0b0110,0b1100, (outs),
639 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3),
641 "vst1", "64", "\\{$src1, $src2, $src3\\}, $addr", "", []>;
643 // ...with double-spaced registers:
644 def VST3q8 : VST3D<0b0101, 0b0000, "8">;
645 def VST3q16 : VST3D<0b0101, 0b0100, "16">;
646 def VST3q32 : VST3D<0b0101, 0b1000, "32">;
648 // vst3 to double-spaced even registers.
649 def VST3q8_UPD : VST3WB<0b0000, "8">;
650 def VST3q16_UPD : VST3WB<0b0100, "16">;
651 def VST3q32_UPD : VST3WB<0b1000, "32">;
653 // vst3 to double-spaced odd registers.
654 def VST3q8odd_UPD : VST3WB<0b0000, "8">;
655 def VST3q16odd_UPD : VST3WB<0b0100, "16">;
656 def VST3q32odd_UPD : VST3WB<0b1000, "32">;
658 // VST4 : Vector Store (multiple 4-element structures)
659 class VST4D<bits<4> op11_8, bits<4> op7_4, string Dt>
660 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
661 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
662 IIC_VST, "vst4", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr",
664 class VST4WB<bits<4> op7_4, string Dt>
665 : NLdSt<0,0b00,0b0001,op7_4, (outs GPR:$wb),
666 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
667 IIC_VST, "vst4", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr",
668 "$addr.addr = $wb", []>;
670 def VST4d8 : VST4D<0b0000, 0b0000, "8">;
671 def VST4d16 : VST4D<0b0000, 0b0100, "16">;
672 def VST4d32 : VST4D<0b0000, 0b1000, "32">;
673 def VST4d64 : NLdSt<0,0b00,0b0010,0b1100, (outs),
674 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3,
676 "vst1", "64", "\\{$src1, $src2, $src3, $src4\\}, $addr",
679 // ...with double-spaced registers:
680 def VST4q8 : VST4D<0b0001, 0b0000, "8">;
681 def VST4q16 : VST4D<0b0001, 0b0100, "16">;
682 def VST4q32 : VST4D<0b0001, 0b1000, "32">;
684 // vst4 to double-spaced even registers.
685 def VST4q8_UPD : VST4WB<0b0000, "8">;
686 def VST4q16_UPD : VST4WB<0b0100, "16">;
687 def VST4q32_UPD : VST4WB<0b1000, "32">;
689 // vst4 to double-spaced odd registers.
690 def VST4q8odd_UPD : VST4WB<0b0000, "8">;
691 def VST4q16odd_UPD : VST4WB<0b0100, "16">;
692 def VST4q32odd_UPD : VST4WB<0b1000, "32">;
694 // VST1LN : Vector Store (single element from one lane)
695 // FIXME: Not yet implemented.
697 // VST2LN : Vector Store (single 2-element structure from one lane)
698 class VST2LN<bits<4> op11_8, string Dt>
699 : NLdSt<1, 0b00, op11_8, {?,?,?,?}, (outs),
700 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, nohash_imm:$lane),
701 IIC_VST, "vst2", Dt, "\\{$src1[$lane], $src2[$lane]\\}, $addr",
704 def VST2LNd8 : VST2LN<0b0001, "8">;
705 def VST2LNd16 : VST2LN<0b0101, "16"> { let Inst{5} = 0; }
706 def VST2LNd32 : VST2LN<0b1001, "32"> { let Inst{6} = 0; }
708 // ...with double-spaced registers:
709 def VST2LNq16 : VST2LN<0b0101, "16"> { let Inst{5} = 1; }
710 def VST2LNq32 : VST2LN<0b1001, "32"> { let Inst{6} = 1; }
712 // ...alternate versions to be allocated odd register numbers:
713 def VST2LNq16odd : VST2LN<0b0101, "16"> { let Inst{5} = 1; }
714 def VST2LNq32odd : VST2LN<0b1001, "32"> { let Inst{6} = 1; }
716 // VST3LN : Vector Store (single 3-element structure from one lane)
717 class VST3LN<bits<4> op11_8, string Dt>
718 : NLdSt<1, 0b00, op11_8, {?,?,?,?}, (outs),
719 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3,
720 nohash_imm:$lane), IIC_VST, "vst3", Dt,
721 "\\{$src1[$lane], $src2[$lane], $src3[$lane]\\}, $addr", "", []>;
723 def VST3LNd8 : VST3LN<0b0010, "8"> { let Inst{4} = 0; }
724 def VST3LNd16 : VST3LN<0b0110, "16"> { let Inst{5-4} = 0b00; }
725 def VST3LNd32 : VST3LN<0b1010, "32"> { let Inst{6-4} = 0b000; }
727 // ...with double-spaced registers:
728 def VST3LNq16 : VST3LN<0b0110, "16"> { let Inst{5-4} = 0b10; }
729 def VST3LNq32 : VST3LN<0b1010, "32"> { let Inst{6-4} = 0b100; }
731 // ...alternate versions to be allocated odd register numbers:
732 def VST3LNq16odd : VST3LN<0b0110, "16"> { let Inst{5-4} = 0b10; }
733 def VST3LNq32odd : VST3LN<0b1010, "32"> { let Inst{6-4} = 0b100; }
735 // VST4LN : Vector Store (single 4-element structure from one lane)
736 class VST4LN<bits<4> op11_8, string Dt>
737 : NLdSt<1, 0b00, op11_8, {?,?,?,?}, (outs),
738 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
739 nohash_imm:$lane), IIC_VST, "vst4", Dt,
740 "\\{$src1[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $addr",
743 def VST4LNd8 : VST4LN<0b0011, "8">;
744 def VST4LNd16 : VST4LN<0b0111, "16"> { let Inst{5} = 0; }
745 def VST4LNd32 : VST4LN<0b1011, "32"> { let Inst{6} = 0; }
747 // ...with double-spaced registers:
748 def VST4LNq16 : VST4LN<0b0111, "16"> { let Inst{5} = 1; }
749 def VST4LNq32 : VST4LN<0b1011, "32"> { let Inst{6} = 1; }
751 // ...alternate versions to be allocated odd register numbers:
752 def VST4LNq16odd : VST4LN<0b0111, "16"> { let Inst{5} = 1; }
753 def VST4LNq32odd : VST4LN<0b1011, "32"> { let Inst{6} = 1; }
755 } // mayStore = 1, hasExtraSrcRegAllocReq = 1
758 //===----------------------------------------------------------------------===//
759 // NEON pattern fragments
760 //===----------------------------------------------------------------------===//
762 // Extract D sub-registers of Q registers.
763 // (arm_dsubreg_0 is 5; arm_dsubreg_1 is 6)
764 def DSubReg_i8_reg : SDNodeXForm<imm, [{
765 return CurDAG->getTargetConstant(5 + N->getZExtValue() / 8, MVT::i32);
767 def DSubReg_i16_reg : SDNodeXForm<imm, [{
768 return CurDAG->getTargetConstant(5 + N->getZExtValue() / 4, MVT::i32);
770 def DSubReg_i32_reg : SDNodeXForm<imm, [{
771 return CurDAG->getTargetConstant(5 + N->getZExtValue() / 2, MVT::i32);
773 def DSubReg_f64_reg : SDNodeXForm<imm, [{
774 return CurDAG->getTargetConstant(5 + N->getZExtValue(), MVT::i32);
776 def DSubReg_f64_other_reg : SDNodeXForm<imm, [{
777 return CurDAG->getTargetConstant(5 + (1 - N->getZExtValue()), MVT::i32);
780 // Extract S sub-registers of Q/D registers.
781 // (arm_ssubreg_0 is 1; arm_ssubreg_1 is 2; etc.)
782 def SSubReg_f32_reg : SDNodeXForm<imm, [{
783 return CurDAG->getTargetConstant(1 + N->getZExtValue(), MVT::i32);
786 // Translate lane numbers from Q registers to D subregs.
787 def SubReg_i8_lane : SDNodeXForm<imm, [{
788 return CurDAG->getTargetConstant(N->getZExtValue() & 7, MVT::i32);
790 def SubReg_i16_lane : SDNodeXForm<imm, [{
791 return CurDAG->getTargetConstant(N->getZExtValue() & 3, MVT::i32);
793 def SubReg_i32_lane : SDNodeXForm<imm, [{
794 return CurDAG->getTargetConstant(N->getZExtValue() & 1, MVT::i32);
797 //===----------------------------------------------------------------------===//
798 // Instruction Classes
799 //===----------------------------------------------------------------------===//
801 // Basic 2-register operations: single-, double- and quad-register.
802 class N2VS<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
803 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
804 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
805 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
806 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src),
807 IIC_VUNAD, OpcodeStr, Dt, "$dst, $src", "", []>;
808 class N2VD<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
809 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
810 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
811 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
812 (ins DPR:$src), IIC_VUNAD, OpcodeStr, Dt, "$dst, $src", "",
813 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src))))]>;
814 class N2VQ<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
815 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
816 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
817 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
818 (ins QPR:$src), IIC_VUNAQ, OpcodeStr, Dt, "$dst, $src", "",
819 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src))))]>;
821 // Basic 2-register intrinsics, both double- and quad-register.
822 class N2VDInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
823 bits<2> op17_16, bits<5> op11_7, bit op4,
824 InstrItinClass itin, string OpcodeStr, string Dt,
825 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
826 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
827 (ins DPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
828 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
829 class N2VQInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
830 bits<2> op17_16, bits<5> op11_7, bit op4,
831 InstrItinClass itin, string OpcodeStr, string Dt,
832 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
833 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
834 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
835 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
837 // Narrow 2-register intrinsics.
838 class N2VNInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
839 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
840 InstrItinClass itin, string OpcodeStr, string Dt,
841 ValueType TyD, ValueType TyQ, Intrinsic IntOp>
842 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$dst),
843 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
844 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src))))]>;
846 // Long 2-register intrinsics (currently only used for VMOVL).
847 class N2VLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
848 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
849 InstrItinClass itin, string OpcodeStr, string Dt,
850 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
851 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$dst),
852 (ins DPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
853 [(set QPR:$dst, (TyQ (IntOp (TyD DPR:$src))))]>;
855 // 2-register shuffles (VTRN/VZIP/VUZP), both double- and quad-register.
856 class N2VDShuffle<bits<2> op19_18, bits<5> op11_7, string OpcodeStr, string Dt>
857 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 0, 0, (outs DPR:$dst1, DPR:$dst2),
858 (ins DPR:$src1, DPR:$src2), IIC_VPERMD,
859 OpcodeStr, Dt, "$dst1, $dst2",
860 "$src1 = $dst1, $src2 = $dst2", []>;
861 class N2VQShuffle<bits<2> op19_18, bits<5> op11_7,
862 InstrItinClass itin, string OpcodeStr, string Dt>
863 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 1, 0, (outs QPR:$dst1, QPR:$dst2),
864 (ins QPR:$src1, QPR:$src2), itin, OpcodeStr, Dt, "$dst1, $dst2",
865 "$src1 = $dst1, $src2 = $dst2", []>;
867 // Basic 3-register operations: single-, double- and quad-register.
868 class N3VS<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
869 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
870 SDNode OpNode, bit Commutable>
871 : N3V<op24, op23, op21_20, op11_8, 0, op4,
872 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src1, DPR_VFP2:$src2), IIC_VBIND,
873 OpcodeStr, Dt, "$dst, $src1, $src2", "", []> {
874 let isCommutable = Commutable;
877 class N3VD<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
878 InstrItinClass itin, string OpcodeStr, string Dt,
879 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
880 : N3V<op24, op23, op21_20, op11_8, 0, op4,
881 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), itin,
882 OpcodeStr, Dt, "$dst, $src1, $src2", "",
883 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src1), (OpTy DPR:$src2))))]> {
884 let isCommutable = Commutable;
886 // Same as N3VD but no data type.
887 class N3VDX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
888 InstrItinClass itin, string OpcodeStr,
889 ValueType ResTy, ValueType OpTy,
890 SDNode OpNode, bit Commutable>
891 : N3VX<op24, op23, op21_20, op11_8, 0, op4,
892 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), itin,
893 OpcodeStr, "$dst, $src1, $src2", "",
894 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src1), (OpTy DPR:$src2))))]>{
895 let isCommutable = Commutable;
897 class N3VDSL<bits<2> op21_20, bits<4> op11_8,
898 InstrItinClass itin, string OpcodeStr, string Dt,
899 ValueType Ty, SDNode ShOp>
900 : N3V<0, 1, op21_20, op11_8, 1, 0,
901 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
902 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
904 (Ty (ShOp (Ty DPR:$src1),
905 (Ty (NEONvduplane (Ty DPR_VFP2:$src2), imm:$lane)))))]>{
906 let isCommutable = 0;
908 class N3VDSL16<bits<2> op21_20, bits<4> op11_8,
909 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
910 : N3V<0, 1, op21_20, op11_8, 1, 0,
911 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
912 IIC_VMULi16D, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
914 (Ty (ShOp (Ty DPR:$src1),
915 (Ty (NEONvduplane (Ty DPR_8:$src2), imm:$lane)))))]> {
916 let isCommutable = 0;
919 class N3VQ<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
920 InstrItinClass itin, string OpcodeStr, string Dt,
921 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
922 : N3V<op24, op23, op21_20, op11_8, 1, op4,
923 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), itin,
924 OpcodeStr, Dt, "$dst, $src1, $src2", "",
925 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src1), (OpTy QPR:$src2))))]> {
926 let isCommutable = Commutable;
928 class N3VQX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
929 InstrItinClass itin, string OpcodeStr,
930 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
931 : N3VX<op24, op23, op21_20, op11_8, 1, op4,
932 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), itin,
933 OpcodeStr, "$dst, $src1, $src2", "",
934 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src1), (OpTy QPR:$src2))))]>{
935 let isCommutable = Commutable;
937 class N3VQSL<bits<2> op21_20, bits<4> op11_8,
938 InstrItinClass itin, string OpcodeStr, string Dt,
939 ValueType ResTy, ValueType OpTy, SDNode ShOp>
940 : N3V<1, 1, op21_20, op11_8, 1, 0,
941 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
942 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
943 [(set (ResTy QPR:$dst),
944 (ResTy (ShOp (ResTy QPR:$src1),
945 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
947 let isCommutable = 0;
949 class N3VQSL16<bits<2> op21_20, bits<4> op11_8, string OpcodeStr, string Dt,
950 ValueType ResTy, ValueType OpTy, SDNode ShOp>
951 : N3V<1, 1, op21_20, op11_8, 1, 0,
952 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
953 IIC_VMULi16Q, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
954 [(set (ResTy QPR:$dst),
955 (ResTy (ShOp (ResTy QPR:$src1),
956 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
958 let isCommutable = 0;
961 // Basic 3-register intrinsics, both double- and quad-register.
962 class N3VDInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
963 InstrItinClass itin, string OpcodeStr, string Dt,
964 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
965 : N3V<op24, op23, op21_20, op11_8, 0, op4,
966 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), itin,
967 OpcodeStr, Dt, "$dst, $src1, $src2", "",
968 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src1), (OpTy DPR:$src2))))]> {
969 let isCommutable = Commutable;
971 class N3VDIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
972 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
973 : N3V<0, 1, op21_20, op11_8, 1, 0,
974 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
975 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
977 (Ty (IntOp (Ty DPR:$src1),
978 (Ty (NEONvduplane (Ty DPR_VFP2:$src2),
980 let isCommutable = 0;
982 class N3VDIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
983 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
984 : N3V<0, 1, op21_20, op11_8, 1, 0,
985 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
986 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
988 (Ty (IntOp (Ty DPR:$src1),
989 (Ty (NEONvduplane (Ty DPR_8:$src2),
991 let isCommutable = 0;
994 class N3VQInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
995 InstrItinClass itin, string OpcodeStr, string Dt,
996 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
997 : N3V<op24, op23, op21_20, op11_8, 1, op4,
998 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), itin,
999 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1000 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src1), (OpTy QPR:$src2))))]> {
1001 let isCommutable = Commutable;
1003 class N3VQIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1004 string OpcodeStr, string Dt,
1005 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1006 : N3V<1, 1, op21_20, op11_8, 1, 0,
1007 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1008 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1009 [(set (ResTy QPR:$dst),
1010 (ResTy (IntOp (ResTy QPR:$src1),
1011 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1013 let isCommutable = 0;
1015 class N3VQIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1016 string OpcodeStr, string Dt,
1017 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1018 : N3V<1, 1, op21_20, op11_8, 1, 0,
1019 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1020 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1021 [(set (ResTy QPR:$dst),
1022 (ResTy (IntOp (ResTy QPR:$src1),
1023 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
1025 let isCommutable = 0;
1028 // Multiply-Add/Sub operations: single-, double- and quad-register.
1029 class N3VSMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1030 InstrItinClass itin, string OpcodeStr, string Dt,
1031 ValueType Ty, SDNode MulOp, SDNode OpNode>
1032 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1033 (outs DPR_VFP2:$dst),
1034 (ins DPR_VFP2:$src1, DPR_VFP2:$src2, DPR_VFP2:$src3), itin,
1035 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst", []>;
1037 class N3VDMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1038 InstrItinClass itin, string OpcodeStr, string Dt,
1039 ValueType Ty, SDNode MulOp, SDNode OpNode>
1040 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1041 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3), itin,
1042 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1043 [(set DPR:$dst, (Ty (OpNode DPR:$src1,
1044 (Ty (MulOp DPR:$src2, DPR:$src3)))))]>;
1045 class N3VDMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1046 string OpcodeStr, string Dt,
1047 ValueType Ty, SDNode MulOp, SDNode ShOp>
1048 : N3V<0, 1, op21_20, op11_8, 1, 0,
1050 (ins DPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane), itin,
1051 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1052 [(set (Ty DPR:$dst),
1053 (Ty (ShOp (Ty DPR:$src1),
1054 (Ty (MulOp DPR:$src2,
1055 (Ty (NEONvduplane (Ty DPR_VFP2:$src3),
1057 class N3VDMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1058 string OpcodeStr, string Dt,
1059 ValueType Ty, SDNode MulOp, SDNode ShOp>
1060 : N3V<0, 1, op21_20, op11_8, 1, 0,
1062 (ins DPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane), itin,
1063 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1064 [(set (Ty DPR:$dst),
1065 (Ty (ShOp (Ty DPR:$src1),
1066 (Ty (MulOp DPR:$src2,
1067 (Ty (NEONvduplane (Ty DPR_8:$src3),
1070 class N3VQMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1071 InstrItinClass itin, string OpcodeStr, string Dt, ValueType Ty,
1072 SDNode MulOp, SDNode OpNode>
1073 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1074 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3), itin,
1075 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1076 [(set QPR:$dst, (Ty (OpNode QPR:$src1,
1077 (Ty (MulOp QPR:$src2, QPR:$src3)))))]>;
1078 class N3VQMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1079 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1080 SDNode MulOp, SDNode ShOp>
1081 : N3V<1, 1, op21_20, op11_8, 1, 0,
1083 (ins QPR:$src1, QPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane), itin,
1084 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1085 [(set (ResTy QPR:$dst),
1086 (ResTy (ShOp (ResTy QPR:$src1),
1087 (ResTy (MulOp QPR:$src2,
1088 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src3),
1090 class N3VQMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1091 string OpcodeStr, string Dt,
1092 ValueType ResTy, ValueType OpTy,
1093 SDNode MulOp, SDNode ShOp>
1094 : N3V<1, 1, op21_20, op11_8, 1, 0,
1096 (ins QPR:$src1, QPR:$src2, DPR_8:$src3, nohash_imm:$lane), itin,
1097 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1098 [(set (ResTy QPR:$dst),
1099 (ResTy (ShOp (ResTy QPR:$src1),
1100 (ResTy (MulOp QPR:$src2,
1101 (ResTy (NEONvduplane (OpTy DPR_8:$src3),
1104 // Neon 3-argument intrinsics, both double- and quad-register.
1105 // The destination register is also used as the first source operand register.
1106 class N3VDInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1107 InstrItinClass itin, string OpcodeStr, string Dt,
1108 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1109 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1110 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3), itin,
1111 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1112 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src1),
1113 (OpTy DPR:$src2), (OpTy DPR:$src3))))]>;
1114 class N3VQInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1115 InstrItinClass itin, string OpcodeStr, string Dt,
1116 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1117 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1118 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3), itin,
1119 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1120 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src1),
1121 (OpTy QPR:$src2), (OpTy QPR:$src3))))]>;
1123 // Neon Long 3-argument intrinsic. The destination register is
1124 // a quad-register and is also used as the first source operand register.
1125 class N3VLInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1126 InstrItinClass itin, string OpcodeStr, string Dt,
1127 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
1128 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1129 (outs QPR:$dst), (ins QPR:$src1, DPR:$src2, DPR:$src3), itin,
1130 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1132 (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$src2), (TyD DPR:$src3))))]>;
1133 class N3VLInt3SL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1134 string OpcodeStr, string Dt,
1135 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1136 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1138 (ins QPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane), itin,
1139 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1140 [(set (ResTy QPR:$dst),
1141 (ResTy (IntOp (ResTy QPR:$src1),
1143 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src3),
1145 class N3VLInt3SL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1146 InstrItinClass itin, string OpcodeStr, string Dt,
1147 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1148 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1150 (ins QPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane), itin,
1151 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1152 [(set (ResTy QPR:$dst),
1153 (ResTy (IntOp (ResTy QPR:$src1),
1155 (OpTy (NEONvduplane (OpTy DPR_8:$src3),
1158 // Narrowing 3-register intrinsics.
1159 class N3VNInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1160 string OpcodeStr, string Dt, ValueType TyD, ValueType TyQ,
1161 Intrinsic IntOp, bit Commutable>
1162 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1163 (outs DPR:$dst), (ins QPR:$src1, QPR:$src2), IIC_VBINi4D,
1164 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1165 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src1), (TyQ QPR:$src2))))]> {
1166 let isCommutable = Commutable;
1169 // Long 3-register intrinsics.
1170 class N3VLInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1171 InstrItinClass itin, string OpcodeStr, string Dt,
1172 ValueType TyQ, ValueType TyD, Intrinsic IntOp, bit Commutable>
1173 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1174 (outs QPR:$dst), (ins DPR:$src1, DPR:$src2), itin,
1175 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1176 [(set QPR:$dst, (TyQ (IntOp (TyD DPR:$src1), (TyD DPR:$src2))))]> {
1177 let isCommutable = Commutable;
1179 class N3VLIntSL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1180 string OpcodeStr, string Dt,
1181 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1182 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1183 (outs QPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1184 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1185 [(set (ResTy QPR:$dst),
1186 (ResTy (IntOp (OpTy DPR:$src1),
1187 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1189 class N3VLIntSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1190 InstrItinClass itin, string OpcodeStr, string Dt,
1191 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1192 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1193 (outs QPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1194 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1195 [(set (ResTy QPR:$dst),
1196 (ResTy (IntOp (OpTy DPR:$src1),
1197 (OpTy (NEONvduplane (OpTy DPR_8:$src2),
1200 // Wide 3-register intrinsics.
1201 class N3VWInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1202 string OpcodeStr, string Dt, ValueType TyQ, ValueType TyD,
1203 Intrinsic IntOp, bit Commutable>
1204 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1205 (outs QPR:$dst), (ins QPR:$src1, DPR:$src2), IIC_VSUBiD,
1206 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1207 [(set QPR:$dst, (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$src2))))]> {
1208 let isCommutable = Commutable;
1211 // Pairwise long 2-register intrinsics, both double- and quad-register.
1212 class N2VDPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1213 bits<2> op17_16, bits<5> op11_7, bit op4,
1214 string OpcodeStr, string Dt,
1215 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1216 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
1217 (ins DPR:$src), IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
1218 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
1219 class N2VQPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1220 bits<2> op17_16, bits<5> op11_7, bit op4,
1221 string OpcodeStr, string Dt,
1222 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1223 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
1224 (ins QPR:$src), IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
1225 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
1227 // Pairwise long 2-register accumulate intrinsics,
1228 // both double- and quad-register.
1229 // The destination register is also used as the first source operand register.
1230 class N2VDPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1231 bits<2> op17_16, bits<5> op11_7, bit op4,
1232 string OpcodeStr, string Dt,
1233 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1234 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
1235 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), IIC_VPALiD,
1236 OpcodeStr, Dt, "$dst, $src2", "$src1 = $dst",
1237 [(set DPR:$dst, (ResTy (IntOp (ResTy DPR:$src1), (OpTy DPR:$src2))))]>;
1238 class N2VQPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1239 bits<2> op17_16, bits<5> op11_7, bit op4,
1240 string OpcodeStr, string Dt,
1241 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1242 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4,
1243 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), IIC_VPALiQ,
1244 OpcodeStr, Dt, "$dst, $src2", "$src1 = $dst",
1245 [(set QPR:$dst, (ResTy (IntOp (ResTy QPR:$src1), (OpTy QPR:$src2))))]>;
1247 // Shift by immediate,
1248 // both double- and quad-register.
1249 class N2VDSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1250 InstrItinClass itin, string OpcodeStr, string Dt,
1251 ValueType Ty, SDNode OpNode>
1252 : N2VImm<op24, op23, op11_8, op7, 0, op4,
1253 (outs DPR:$dst), (ins DPR:$src, i32imm:$SIMM), itin,
1254 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1255 [(set DPR:$dst, (Ty (OpNode (Ty DPR:$src), (i32 imm:$SIMM))))]>;
1256 class N2VQSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1257 InstrItinClass itin, string OpcodeStr, string Dt,
1258 ValueType Ty, SDNode OpNode>
1259 : N2VImm<op24, op23, op11_8, op7, 1, op4,
1260 (outs QPR:$dst), (ins QPR:$src, i32imm:$SIMM), itin,
1261 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1262 [(set QPR:$dst, (Ty (OpNode (Ty QPR:$src), (i32 imm:$SIMM))))]>;
1264 // Long shift by immediate.
1265 class N2VLSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
1266 string OpcodeStr, string Dt,
1267 ValueType ResTy, ValueType OpTy, SDNode OpNode>
1268 : N2VImm<op24, op23, op11_8, op7, op6, op4,
1269 (outs QPR:$dst), (ins DPR:$src, i32imm:$SIMM), IIC_VSHLiD,
1270 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1271 [(set QPR:$dst, (ResTy (OpNode (OpTy DPR:$src),
1272 (i32 imm:$SIMM))))]>;
1274 // Narrow shift by immediate.
1275 class N2VNSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
1276 InstrItinClass itin, string OpcodeStr, string Dt,
1277 ValueType ResTy, ValueType OpTy, SDNode OpNode>
1278 : N2VImm<op24, op23, op11_8, op7, op6, op4,
1279 (outs DPR:$dst), (ins QPR:$src, i32imm:$SIMM), itin,
1280 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1281 [(set DPR:$dst, (ResTy (OpNode (OpTy QPR:$src),
1282 (i32 imm:$SIMM))))]>;
1284 // Shift right by immediate and accumulate,
1285 // both double- and quad-register.
1286 class N2VDShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1287 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1288 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$dst),
1289 (ins DPR:$src1, DPR:$src2, i32imm:$SIMM), IIC_VPALiD,
1290 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1291 [(set DPR:$dst, (Ty (add DPR:$src1,
1292 (Ty (ShOp DPR:$src2, (i32 imm:$SIMM))))))]>;
1293 class N2VQShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1294 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1295 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$dst),
1296 (ins QPR:$src1, QPR:$src2, i32imm:$SIMM), IIC_VPALiD,
1297 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1298 [(set QPR:$dst, (Ty (add QPR:$src1,
1299 (Ty (ShOp QPR:$src2, (i32 imm:$SIMM))))))]>;
1301 // Shift by immediate and insert,
1302 // both double- and quad-register.
1303 class N2VDShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1304 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1305 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$dst),
1306 (ins DPR:$src1, DPR:$src2, i32imm:$SIMM), IIC_VSHLiD,
1307 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1308 [(set DPR:$dst, (Ty (ShOp DPR:$src1, DPR:$src2, (i32 imm:$SIMM))))]>;
1309 class N2VQShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1310 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1311 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$dst),
1312 (ins QPR:$src1, QPR:$src2, i32imm:$SIMM), IIC_VSHLiQ,
1313 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1314 [(set QPR:$dst, (Ty (ShOp QPR:$src1, QPR:$src2, (i32 imm:$SIMM))))]>;
1316 // Convert, with fractional bits immediate,
1317 // both double- and quad-register.
1318 class N2VCvtD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1319 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1321 : N2VImm<op24, op23, op11_8, op7, 0, op4,
1322 (outs DPR:$dst), (ins DPR:$src, i32imm:$SIMM), IIC_VUNAD,
1323 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1324 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src), (i32 imm:$SIMM))))]>;
1325 class N2VCvtQ<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1326 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1328 : N2VImm<op24, op23, op11_8, op7, 1, op4,
1329 (outs QPR:$dst), (ins QPR:$src, i32imm:$SIMM), IIC_VUNAQ,
1330 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1331 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src), (i32 imm:$SIMM))))]>;
1333 //===----------------------------------------------------------------------===//
1335 //===----------------------------------------------------------------------===//
1337 // Abbreviations used in multiclass suffixes:
1338 // Q = quarter int (8 bit) elements
1339 // H = half int (16 bit) elements
1340 // S = single int (32 bit) elements
1341 // D = double int (64 bit) elements
1343 // Neon 2-register vector operations -- for disassembly only.
1345 // First with only element sizes of 8, 16 and 32 bits:
1346 multiclass N2V_QHS_cmp<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1347 bits<5> op11_7, bit op4, string opc, string Dt,
1349 // 64-bit vector types.
1350 def v8i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 0, op4,
1351 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1352 opc, !strconcat(Dt, "8"), asm, "", []>;
1353 def v4i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 0, op4,
1354 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1355 opc, !strconcat(Dt, "16"), asm, "", []>;
1356 def v2i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
1357 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1358 opc, !strconcat(Dt, "32"), asm, "", []>;
1359 def v2f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
1360 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1361 opc, "f32", asm, "", []> {
1362 let Inst{10} = 1; // overwrite F = 1
1365 // 128-bit vector types.
1366 def v16i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 1, op4,
1367 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1368 opc, !strconcat(Dt, "8"), asm, "", []>;
1369 def v8i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 1, op4,
1370 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1371 opc, !strconcat(Dt, "16"), asm, "", []>;
1372 def v4i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
1373 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1374 opc, !strconcat(Dt, "32"), asm, "", []>;
1375 def v4f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
1376 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1377 opc, "f32", asm, "", []> {
1378 let Inst{10} = 1; // overwrite F = 1
1382 // Neon 3-register vector operations.
1384 // First with only element sizes of 8, 16 and 32 bits:
1385 multiclass N3V_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1386 InstrItinClass itinD16, InstrItinClass itinD32,
1387 InstrItinClass itinQ16, InstrItinClass itinQ32,
1388 string OpcodeStr, string Dt,
1389 SDNode OpNode, bit Commutable = 0> {
1390 // 64-bit vector types.
1391 def v8i8 : N3VD<op24, op23, 0b00, op11_8, op4, itinD16,
1392 OpcodeStr, !strconcat(Dt, "8"),
1393 v8i8, v8i8, OpNode, Commutable>;
1394 def v4i16 : N3VD<op24, op23, 0b01, op11_8, op4, itinD16,
1395 OpcodeStr, !strconcat(Dt, "16"),
1396 v4i16, v4i16, OpNode, Commutable>;
1397 def v2i32 : N3VD<op24, op23, 0b10, op11_8, op4, itinD32,
1398 OpcodeStr, !strconcat(Dt, "32"),
1399 v2i32, v2i32, OpNode, Commutable>;
1401 // 128-bit vector types.
1402 def v16i8 : N3VQ<op24, op23, 0b00, op11_8, op4, itinQ16,
1403 OpcodeStr, !strconcat(Dt, "8"),
1404 v16i8, v16i8, OpNode, Commutable>;
1405 def v8i16 : N3VQ<op24, op23, 0b01, op11_8, op4, itinQ16,
1406 OpcodeStr, !strconcat(Dt, "16"),
1407 v8i16, v8i16, OpNode, Commutable>;
1408 def v4i32 : N3VQ<op24, op23, 0b10, op11_8, op4, itinQ32,
1409 OpcodeStr, !strconcat(Dt, "32"),
1410 v4i32, v4i32, OpNode, Commutable>;
1413 multiclass N3VSL_HS<bits<4> op11_8, string OpcodeStr, string Dt, SDNode ShOp> {
1414 def v4i16 : N3VDSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
1416 def v2i32 : N3VDSL<0b10, op11_8, IIC_VMULi32D, OpcodeStr, !strconcat(Dt,"32"),
1418 def v8i16 : N3VQSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
1419 v8i16, v4i16, ShOp>;
1420 def v4i32 : N3VQSL<0b10, op11_8, IIC_VMULi32Q, OpcodeStr, !strconcat(Dt,"32"),
1421 v4i32, v2i32, ShOp>;
1424 // ....then also with element size 64 bits:
1425 multiclass N3V_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1426 InstrItinClass itinD, InstrItinClass itinQ,
1427 string OpcodeStr, string Dt,
1428 SDNode OpNode, bit Commutable = 0>
1429 : N3V_QHS<op24, op23, op11_8, op4, itinD, itinD, itinQ, itinQ,
1430 OpcodeStr, Dt, OpNode, Commutable> {
1431 def v1i64 : N3VD<op24, op23, 0b11, op11_8, op4, itinD,
1432 OpcodeStr, !strconcat(Dt, "64"),
1433 v1i64, v1i64, OpNode, Commutable>;
1434 def v2i64 : N3VQ<op24, op23, 0b11, op11_8, op4, itinQ,
1435 OpcodeStr, !strconcat(Dt, "64"),
1436 v2i64, v2i64, OpNode, Commutable>;
1440 // Neon Narrowing 2-register vector intrinsics,
1441 // source operand element sizes of 16, 32 and 64 bits:
1442 multiclass N2VNInt_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1443 bits<5> op11_7, bit op6, bit op4,
1444 InstrItinClass itin, string OpcodeStr, string Dt,
1446 def v8i8 : N2VNInt<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
1447 itin, OpcodeStr, !strconcat(Dt, "16"),
1448 v8i8, v8i16, IntOp>;
1449 def v4i16 : N2VNInt<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
1450 itin, OpcodeStr, !strconcat(Dt, "32"),
1451 v4i16, v4i32, IntOp>;
1452 def v2i32 : N2VNInt<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
1453 itin, OpcodeStr, !strconcat(Dt, "64"),
1454 v2i32, v2i64, IntOp>;
1458 // Neon Lengthening 2-register vector intrinsic (currently specific to VMOVL).
1459 // source operand element sizes of 16, 32 and 64 bits:
1460 multiclass N2VLInt_QHS<bits<2> op24_23, bits<5> op11_7, bit op6, bit op4,
1461 string OpcodeStr, string Dt, Intrinsic IntOp> {
1462 def v8i16 : N2VLInt<op24_23, 0b00, 0b10, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
1463 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
1464 def v4i32 : N2VLInt<op24_23, 0b01, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
1465 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
1466 def v2i64 : N2VLInt<op24_23, 0b10, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
1467 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1471 // Neon 3-register vector intrinsics.
1473 // First with only element sizes of 16 and 32 bits:
1474 multiclass N3VInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
1475 InstrItinClass itinD16, InstrItinClass itinD32,
1476 InstrItinClass itinQ16, InstrItinClass itinQ32,
1477 string OpcodeStr, string Dt,
1478 Intrinsic IntOp, bit Commutable = 0> {
1479 // 64-bit vector types.
1480 def v4i16 : N3VDInt<op24, op23, 0b01, op11_8, op4, itinD16,
1481 OpcodeStr, !strconcat(Dt, "16"),
1482 v4i16, v4i16, IntOp, Commutable>;
1483 def v2i32 : N3VDInt<op24, op23, 0b10, op11_8, op4, itinD32,
1484 OpcodeStr, !strconcat(Dt, "32"),
1485 v2i32, v2i32, IntOp, Commutable>;
1487 // 128-bit vector types.
1488 def v8i16 : N3VQInt<op24, op23, 0b01, op11_8, op4, itinQ16,
1489 OpcodeStr, !strconcat(Dt, "16"),
1490 v8i16, v8i16, IntOp, Commutable>;
1491 def v4i32 : N3VQInt<op24, op23, 0b10, op11_8, op4, itinQ32,
1492 OpcodeStr, !strconcat(Dt, "32"),
1493 v4i32, v4i32, IntOp, Commutable>;
1496 multiclass N3VIntSL_HS<bits<4> op11_8,
1497 InstrItinClass itinD16, InstrItinClass itinD32,
1498 InstrItinClass itinQ16, InstrItinClass itinQ32,
1499 string OpcodeStr, string Dt, Intrinsic IntOp> {
1500 def v4i16 : N3VDIntSL16<0b01, op11_8, itinD16,
1501 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp>;
1502 def v2i32 : N3VDIntSL<0b10, op11_8, itinD32,
1503 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp>;
1504 def v8i16 : N3VQIntSL16<0b01, op11_8, itinQ16,
1505 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16, IntOp>;
1506 def v4i32 : N3VQIntSL<0b10, op11_8, itinQ32,
1507 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32, IntOp>;
1510 // ....then also with element size of 8 bits:
1511 multiclass N3VInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1512 InstrItinClass itinD16, InstrItinClass itinD32,
1513 InstrItinClass itinQ16, InstrItinClass itinQ32,
1514 string OpcodeStr, string Dt,
1515 Intrinsic IntOp, bit Commutable = 0>
1516 : N3VInt_HS<op24, op23, op11_8, op4, itinD16, itinD32, itinQ16, itinQ32,
1517 OpcodeStr, Dt, IntOp, Commutable> {
1518 def v8i8 : N3VDInt<op24, op23, 0b00, op11_8, op4, itinD16,
1519 OpcodeStr, !strconcat(Dt, "8"),
1520 v8i8, v8i8, IntOp, Commutable>;
1521 def v16i8 : N3VQInt<op24, op23, 0b00, op11_8, op4, itinQ16,
1522 OpcodeStr, !strconcat(Dt, "8"),
1523 v16i8, v16i8, IntOp, Commutable>;
1526 // ....then also with element size of 64 bits:
1527 multiclass N3VInt_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1528 InstrItinClass itinD16, InstrItinClass itinD32,
1529 InstrItinClass itinQ16, InstrItinClass itinQ32,
1530 string OpcodeStr, string Dt,
1531 Intrinsic IntOp, bit Commutable = 0>
1532 : N3VInt_QHS<op24, op23, op11_8, op4, itinD16, itinD32, itinQ16, itinQ32,
1533 OpcodeStr, Dt, IntOp, Commutable> {
1534 def v1i64 : N3VDInt<op24, op23, 0b11, op11_8, op4, itinD32,
1535 OpcodeStr, !strconcat(Dt, "64"),
1536 v1i64, v1i64, IntOp, Commutable>;
1537 def v2i64 : N3VQInt<op24, op23, 0b11, op11_8, op4, itinQ32,
1538 OpcodeStr, !strconcat(Dt, "64"),
1539 v2i64, v2i64, IntOp, Commutable>;
1543 // Neon Narrowing 3-register vector intrinsics,
1544 // source operand element sizes of 16, 32 and 64 bits:
1545 multiclass N3VNInt_HSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1546 string OpcodeStr, string Dt,
1547 Intrinsic IntOp, bit Commutable = 0> {
1548 def v8i8 : N3VNInt<op24, op23, 0b00, op11_8, op4,
1549 OpcodeStr, !strconcat(Dt, "16"),
1550 v8i8, v8i16, IntOp, Commutable>;
1551 def v4i16 : N3VNInt<op24, op23, 0b01, op11_8, op4,
1552 OpcodeStr, !strconcat(Dt, "32"),
1553 v4i16, v4i32, IntOp, Commutable>;
1554 def v2i32 : N3VNInt<op24, op23, 0b10, op11_8, op4,
1555 OpcodeStr, !strconcat(Dt, "64"),
1556 v2i32, v2i64, IntOp, Commutable>;
1560 // Neon Long 3-register vector intrinsics.
1562 // First with only element sizes of 16 and 32 bits:
1563 multiclass N3VLInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
1564 InstrItinClass itin, string OpcodeStr, string Dt,
1565 Intrinsic IntOp, bit Commutable = 0> {
1566 def v4i32 : N3VLInt<op24, op23, 0b01, op11_8, op4, itin,
1567 OpcodeStr, !strconcat(Dt, "16"),
1568 v4i32, v4i16, IntOp, Commutable>;
1569 def v2i64 : N3VLInt<op24, op23, 0b10, op11_8, op4, itin,
1570 OpcodeStr, !strconcat(Dt, "32"),
1571 v2i64, v2i32, IntOp, Commutable>;
1574 multiclass N3VLIntSL_HS<bit op24, bits<4> op11_8,
1575 InstrItinClass itin, string OpcodeStr, string Dt,
1577 def v4i16 : N3VLIntSL16<op24, 0b01, op11_8, itin,
1578 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
1579 def v2i32 : N3VLIntSL<op24, 0b10, op11_8, itin,
1580 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1583 // ....then also with element size of 8 bits:
1584 multiclass N3VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1585 InstrItinClass itin, string OpcodeStr, string Dt,
1586 Intrinsic IntOp, bit Commutable = 0>
1587 : N3VLInt_HS<op24, op23, op11_8, op4, itin, OpcodeStr, Dt,
1588 IntOp, Commutable> {
1589 def v8i16 : N3VLInt<op24, op23, 0b00, op11_8, op4, itin,
1590 OpcodeStr, !strconcat(Dt, "8"),
1591 v8i16, v8i8, IntOp, Commutable>;
1595 // Neon Wide 3-register vector intrinsics,
1596 // source operand element sizes of 8, 16 and 32 bits:
1597 multiclass N3VWInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1598 string OpcodeStr, string Dt,
1599 Intrinsic IntOp, bit Commutable = 0> {
1600 def v8i16 : N3VWInt<op24, op23, 0b00, op11_8, op4,
1601 OpcodeStr, !strconcat(Dt, "8"),
1602 v8i16, v8i8, IntOp, Commutable>;
1603 def v4i32 : N3VWInt<op24, op23, 0b01, op11_8, op4,
1604 OpcodeStr, !strconcat(Dt, "16"),
1605 v4i32, v4i16, IntOp, Commutable>;
1606 def v2i64 : N3VWInt<op24, op23, 0b10, op11_8, op4,
1607 OpcodeStr, !strconcat(Dt, "32"),
1608 v2i64, v2i32, IntOp, Commutable>;
1612 // Neon Multiply-Op vector operations,
1613 // element sizes of 8, 16 and 32 bits:
1614 multiclass N3VMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1615 InstrItinClass itinD16, InstrItinClass itinD32,
1616 InstrItinClass itinQ16, InstrItinClass itinQ32,
1617 string OpcodeStr, string Dt, SDNode OpNode> {
1618 // 64-bit vector types.
1619 def v8i8 : N3VDMulOp<op24, op23, 0b00, op11_8, op4, itinD16,
1620 OpcodeStr, !strconcat(Dt, "8"), v8i8, mul, OpNode>;
1621 def v4i16 : N3VDMulOp<op24, op23, 0b01, op11_8, op4, itinD16,
1622 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, OpNode>;
1623 def v2i32 : N3VDMulOp<op24, op23, 0b10, op11_8, op4, itinD32,
1624 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, OpNode>;
1626 // 128-bit vector types.
1627 def v16i8 : N3VQMulOp<op24, op23, 0b00, op11_8, op4, itinQ16,
1628 OpcodeStr, !strconcat(Dt, "8"), v16i8, mul, OpNode>;
1629 def v8i16 : N3VQMulOp<op24, op23, 0b01, op11_8, op4, itinQ16,
1630 OpcodeStr, !strconcat(Dt, "16"), v8i16, mul, OpNode>;
1631 def v4i32 : N3VQMulOp<op24, op23, 0b10, op11_8, op4, itinQ32,
1632 OpcodeStr, !strconcat(Dt, "32"), v4i32, mul, OpNode>;
1635 multiclass N3VMulOpSL_HS<bits<4> op11_8,
1636 InstrItinClass itinD16, InstrItinClass itinD32,
1637 InstrItinClass itinQ16, InstrItinClass itinQ32,
1638 string OpcodeStr, string Dt, SDNode ShOp> {
1639 def v4i16 : N3VDMulOpSL16<0b01, op11_8, itinD16,
1640 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, ShOp>;
1641 def v2i32 : N3VDMulOpSL<0b10, op11_8, itinD32,
1642 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, ShOp>;
1643 def v8i16 : N3VQMulOpSL16<0b01, op11_8, itinQ16,
1644 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16,
1646 def v4i32 : N3VQMulOpSL<0b10, op11_8, itinQ32,
1647 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32,
1651 // Neon 3-argument intrinsics,
1652 // element sizes of 8, 16 and 32 bits:
1653 multiclass N3VInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1654 string OpcodeStr, string Dt, Intrinsic IntOp> {
1655 // 64-bit vector types.
1656 def v8i8 : N3VDInt3<op24, op23, 0b00, op11_8, op4, IIC_VMACi16D,
1657 OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
1658 def v4i16 : N3VDInt3<op24, op23, 0b01, op11_8, op4, IIC_VMACi16D,
1659 OpcodeStr, !strconcat(Dt, "16"), v4i16, v4i16, IntOp>;
1660 def v2i32 : N3VDInt3<op24, op23, 0b10, op11_8, op4, IIC_VMACi32D,
1661 OpcodeStr, !strconcat(Dt, "32"), v2i32, v2i32, IntOp>;
1663 // 128-bit vector types.
1664 def v16i8 : N3VQInt3<op24, op23, 0b00, op11_8, op4, IIC_VMACi16Q,
1665 OpcodeStr, !strconcat(Dt, "8"), v16i8, v16i8, IntOp>;
1666 def v8i16 : N3VQInt3<op24, op23, 0b01, op11_8, op4, IIC_VMACi16Q,
1667 OpcodeStr, !strconcat(Dt, "16"), v8i16, v8i16, IntOp>;
1668 def v4i32 : N3VQInt3<op24, op23, 0b10, op11_8, op4, IIC_VMACi32Q,
1669 OpcodeStr, !strconcat(Dt, "32"), v4i32, v4i32, IntOp>;
1673 // Neon Long 3-argument intrinsics.
1675 // First with only element sizes of 16 and 32 bits:
1676 multiclass N3VLInt3_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
1677 string OpcodeStr, string Dt, Intrinsic IntOp> {
1678 def v4i32 : N3VLInt3<op24, op23, 0b01, op11_8, op4, IIC_VMACi16D,
1679 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
1680 def v2i64 : N3VLInt3<op24, op23, 0b10, op11_8, op4, IIC_VMACi16D,
1681 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1684 multiclass N3VLInt3SL_HS<bit op24, bits<4> op11_8,
1685 string OpcodeStr, string Dt, Intrinsic IntOp> {
1686 def v4i16 : N3VLInt3SL16<op24, 0b01, op11_8, IIC_VMACi16D,
1687 OpcodeStr, !strconcat(Dt,"16"), v4i32, v4i16, IntOp>;
1688 def v2i32 : N3VLInt3SL<op24, 0b10, op11_8, IIC_VMACi32D,
1689 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1692 // ....then also with element size of 8 bits:
1693 multiclass N3VLInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1694 string OpcodeStr, string Dt, Intrinsic IntOp>
1695 : N3VLInt3_HS<op24, op23, op11_8, op4, OpcodeStr, Dt, IntOp> {
1696 def v8i16 : N3VLInt3<op24, op23, 0b00, op11_8, op4, IIC_VMACi16D,
1697 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
1701 // Neon 2-register vector intrinsics,
1702 // element sizes of 8, 16 and 32 bits:
1703 multiclass N2VInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1704 bits<5> op11_7, bit op4,
1705 InstrItinClass itinD, InstrItinClass itinQ,
1706 string OpcodeStr, string Dt, Intrinsic IntOp> {
1707 // 64-bit vector types.
1708 def v8i8 : N2VDInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1709 itinD, OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
1710 def v4i16 : N2VDInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1711 itinD, OpcodeStr, !strconcat(Dt, "16"),v4i16,v4i16,IntOp>;
1712 def v2i32 : N2VDInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1713 itinD, OpcodeStr, !strconcat(Dt, "32"),v2i32,v2i32,IntOp>;
1715 // 128-bit vector types.
1716 def v16i8 : N2VQInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1717 itinQ, OpcodeStr, !strconcat(Dt, "8"), v16i8,v16i8,IntOp>;
1718 def v8i16 : N2VQInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1719 itinQ, OpcodeStr, !strconcat(Dt, "16"),v8i16,v8i16,IntOp>;
1720 def v4i32 : N2VQInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1721 itinQ, OpcodeStr, !strconcat(Dt, "32"),v4i32,v4i32,IntOp>;
1725 // Neon Pairwise long 2-register intrinsics,
1726 // element sizes of 8, 16 and 32 bits:
1727 multiclass N2VPLInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1728 bits<5> op11_7, bit op4,
1729 string OpcodeStr, string Dt, Intrinsic IntOp> {
1730 // 64-bit vector types.
1731 def v8i8 : N2VDPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1732 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
1733 def v4i16 : N2VDPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1734 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
1735 def v2i32 : N2VDPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1736 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
1738 // 128-bit vector types.
1739 def v16i8 : N2VQPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1740 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
1741 def v8i16 : N2VQPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1742 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
1743 def v4i32 : N2VQPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1744 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
1748 // Neon Pairwise long 2-register accumulate intrinsics,
1749 // element sizes of 8, 16 and 32 bits:
1750 multiclass N2VPLInt2_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1751 bits<5> op11_7, bit op4,
1752 string OpcodeStr, string Dt, Intrinsic IntOp> {
1753 // 64-bit vector types.
1754 def v8i8 : N2VDPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1755 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
1756 def v4i16 : N2VDPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1757 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
1758 def v2i32 : N2VDPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1759 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
1761 // 128-bit vector types.
1762 def v16i8 : N2VQPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1763 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
1764 def v8i16 : N2VQPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1765 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
1766 def v4i32 : N2VQPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1767 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
1771 // Neon 2-register vector shift by immediate,
1772 // element sizes of 8, 16, 32 and 64 bits:
1773 multiclass N2VSh_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1774 InstrItinClass itin, string OpcodeStr, string Dt,
1776 // 64-bit vector types.
1777 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, itin,
1778 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
1779 let Inst{21-19} = 0b001; // imm6 = 001xxx
1781 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, itin,
1782 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
1783 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1785 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, itin,
1786 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
1787 let Inst{21} = 0b1; // imm6 = 1xxxxx
1789 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, itin,
1790 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
1793 // 128-bit vector types.
1794 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, itin,
1795 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
1796 let Inst{21-19} = 0b001; // imm6 = 001xxx
1798 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, itin,
1799 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
1800 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1802 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, itin,
1803 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
1804 let Inst{21} = 0b1; // imm6 = 1xxxxx
1806 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, itin,
1807 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
1812 // Neon Shift-Accumulate vector operations,
1813 // element sizes of 8, 16, 32 and 64 bits:
1814 multiclass N2VShAdd_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1815 string OpcodeStr, string Dt, SDNode ShOp> {
1816 // 64-bit vector types.
1817 def v8i8 : N2VDShAdd<op24, op23, op11_8, 0, op4,
1818 OpcodeStr, !strconcat(Dt, "8"), v8i8, ShOp> {
1819 let Inst{21-19} = 0b001; // imm6 = 001xxx
1821 def v4i16 : N2VDShAdd<op24, op23, op11_8, 0, op4,
1822 OpcodeStr, !strconcat(Dt, "16"), v4i16, ShOp> {
1823 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1825 def v2i32 : N2VDShAdd<op24, op23, op11_8, 0, op4,
1826 OpcodeStr, !strconcat(Dt, "32"), v2i32, ShOp> {
1827 let Inst{21} = 0b1; // imm6 = 1xxxxx
1829 def v1i64 : N2VDShAdd<op24, op23, op11_8, 1, op4,
1830 OpcodeStr, !strconcat(Dt, "64"), v1i64, ShOp>;
1833 // 128-bit vector types.
1834 def v16i8 : N2VQShAdd<op24, op23, op11_8, 0, op4,
1835 OpcodeStr, !strconcat(Dt, "8"), v16i8, ShOp> {
1836 let Inst{21-19} = 0b001; // imm6 = 001xxx
1838 def v8i16 : N2VQShAdd<op24, op23, op11_8, 0, op4,
1839 OpcodeStr, !strconcat(Dt, "16"), v8i16, ShOp> {
1840 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1842 def v4i32 : N2VQShAdd<op24, op23, op11_8, 0, op4,
1843 OpcodeStr, !strconcat(Dt, "32"), v4i32, ShOp> {
1844 let Inst{21} = 0b1; // imm6 = 1xxxxx
1846 def v2i64 : N2VQShAdd<op24, op23, op11_8, 1, op4,
1847 OpcodeStr, !strconcat(Dt, "64"), v2i64, ShOp>;
1852 // Neon Shift-Insert vector operations,
1853 // element sizes of 8, 16, 32 and 64 bits:
1854 multiclass N2VShIns_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1855 string OpcodeStr, SDNode ShOp> {
1856 // 64-bit vector types.
1857 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4,
1858 OpcodeStr, "8", v8i8, ShOp> {
1859 let Inst{21-19} = 0b001; // imm6 = 001xxx
1861 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4,
1862 OpcodeStr, "16", v4i16, ShOp> {
1863 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1865 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4,
1866 OpcodeStr, "32", v2i32, ShOp> {
1867 let Inst{21} = 0b1; // imm6 = 1xxxxx
1869 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4,
1870 OpcodeStr, "64", v1i64, ShOp>;
1873 // 128-bit vector types.
1874 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4,
1875 OpcodeStr, "8", v16i8, ShOp> {
1876 let Inst{21-19} = 0b001; // imm6 = 001xxx
1878 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4,
1879 OpcodeStr, "16", v8i16, ShOp> {
1880 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1882 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4,
1883 OpcodeStr, "32", v4i32, ShOp> {
1884 let Inst{21} = 0b1; // imm6 = 1xxxxx
1886 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4,
1887 OpcodeStr, "64", v2i64, ShOp>;
1891 // Neon Shift Long operations,
1892 // element sizes of 8, 16, 32 bits:
1893 multiclass N2VLSh_QHS<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
1894 bit op4, string OpcodeStr, string Dt, SDNode OpNode> {
1895 def v8i16 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
1896 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode> {
1897 let Inst{21-19} = 0b001; // imm6 = 001xxx
1899 def v4i32 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
1900 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode> {
1901 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1903 def v2i64 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
1904 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode> {
1905 let Inst{21} = 0b1; // imm6 = 1xxxxx
1909 // Neon Shift Narrow operations,
1910 // element sizes of 16, 32, 64 bits:
1911 multiclass N2VNSh_HSD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
1912 bit op4, InstrItinClass itin, string OpcodeStr, string Dt,
1914 def v8i8 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
1915 OpcodeStr, !strconcat(Dt, "16"), v8i8, v8i16, OpNode> {
1916 let Inst{21-19} = 0b001; // imm6 = 001xxx
1918 def v4i16 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
1919 OpcodeStr, !strconcat(Dt, "32"), v4i16, v4i32, OpNode> {
1920 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1922 def v2i32 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
1923 OpcodeStr, !strconcat(Dt, "64"), v2i32, v2i64, OpNode> {
1924 let Inst{21} = 0b1; // imm6 = 1xxxxx
1928 //===----------------------------------------------------------------------===//
1929 // Instruction Definitions.
1930 //===----------------------------------------------------------------------===//
1932 // Vector Add Operations.
1934 // VADD : Vector Add (integer and floating-point)
1935 defm VADD : N3V_QHSD<0, 0, 0b1000, 0, IIC_VBINiD, IIC_VBINiQ, "vadd", "i",
1937 def VADDfd : N3VD<0, 0, 0b00, 0b1101, 0, IIC_VBIND, "vadd", "f32",
1938 v2f32, v2f32, fadd, 1>;
1939 def VADDfq : N3VQ<0, 0, 0b00, 0b1101, 0, IIC_VBINQ, "vadd", "f32",
1940 v4f32, v4f32, fadd, 1>;
1941 // VADDL : Vector Add Long (Q = D + D)
1942 defm VADDLs : N3VLInt_QHS<0,1,0b0000,0, IIC_VSHLiD, "vaddl", "s",
1943 int_arm_neon_vaddls, 1>;
1944 defm VADDLu : N3VLInt_QHS<1,1,0b0000,0, IIC_VSHLiD, "vaddl", "u",
1945 int_arm_neon_vaddlu, 1>;
1946 // VADDW : Vector Add Wide (Q = Q + D)
1947 defm VADDWs : N3VWInt_QHS<0,1,0b0001,0, "vaddw", "s", int_arm_neon_vaddws, 0>;
1948 defm VADDWu : N3VWInt_QHS<1,1,0b0001,0, "vaddw", "u", int_arm_neon_vaddwu, 0>;
1949 // VHADD : Vector Halving Add
1950 defm VHADDs : N3VInt_QHS<0,0,0b0000,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1951 IIC_VBINi4Q, "vhadd", "s", int_arm_neon_vhadds, 1>;
1952 defm VHADDu : N3VInt_QHS<1,0,0b0000,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1953 IIC_VBINi4Q, "vhadd", "u", int_arm_neon_vhaddu, 1>;
1954 // VRHADD : Vector Rounding Halving Add
1955 defm VRHADDs : N3VInt_QHS<0,0,0b0001,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1956 IIC_VBINi4Q, "vrhadd", "s", int_arm_neon_vrhadds, 1>;
1957 defm VRHADDu : N3VInt_QHS<1,0,0b0001,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1958 IIC_VBINi4Q, "vrhadd", "u", int_arm_neon_vrhaddu, 1>;
1959 // VQADD : Vector Saturating Add
1960 defm VQADDs : N3VInt_QHSD<0,0,0b0000,1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1961 IIC_VBINi4Q, "vqadd", "s", int_arm_neon_vqadds, 1>;
1962 defm VQADDu : N3VInt_QHSD<1,0,0b0000,1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1963 IIC_VBINi4Q, "vqadd", "u", int_arm_neon_vqaddu, 1>;
1964 // VADDHN : Vector Add and Narrow Returning High Half (D = Q + Q)
1965 defm VADDHN : N3VNInt_HSD<0,1,0b0100,0, "vaddhn", "i",
1966 int_arm_neon_vaddhn, 1>;
1967 // VRADDHN : Vector Rounding Add and Narrow Returning High Half (D = Q + Q)
1968 defm VRADDHN : N3VNInt_HSD<1,1,0b0100,0, "vraddhn", "i",
1969 int_arm_neon_vraddhn, 1>;
1971 // Vector Multiply Operations.
1973 // VMUL : Vector Multiply (integer, polynomial and floating-point)
1974 defm VMUL : N3V_QHS<0, 0, 0b1001, 1, IIC_VMULi16D, IIC_VMULi32D,
1975 IIC_VMULi16Q, IIC_VMULi32Q, "vmul", "i", mul, 1>;
1976 def VMULpd : N3VDInt<1, 0, 0b00, 0b1001, 1, IIC_VMULi16D, "vmul", "p8",
1977 v8i8, v8i8, int_arm_neon_vmulp, 1>;
1978 def VMULpq : N3VQInt<1, 0, 0b00, 0b1001, 1, IIC_VMULi16Q, "vmul", "p8",
1979 v16i8, v16i8, int_arm_neon_vmulp, 1>;
1980 def VMULfd : N3VD<1, 0, 0b00, 0b1101, 1, IIC_VBIND, "vmul", "f32",
1981 v2f32, v2f32, fmul, 1>;
1982 def VMULfq : N3VQ<1, 0, 0b00, 0b1101, 1, IIC_VBINQ, "vmul", "f32",
1983 v4f32, v4f32, fmul, 1>;
1984 defm VMULsl : N3VSL_HS<0b1000, "vmul", "i", mul>;
1985 def VMULslfd : N3VDSL<0b10, 0b1001, IIC_VBIND, "vmul", "f32", v2f32, fmul>;
1986 def VMULslfq : N3VQSL<0b10, 0b1001, IIC_VBINQ, "vmul", "f32", v4f32,
1989 def : Pat<(v8i16 (mul (v8i16 QPR:$src1),
1990 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
1991 (v8i16 (VMULslv8i16 (v8i16 QPR:$src1),
1992 (v4i16 (EXTRACT_SUBREG QPR:$src2,
1993 (DSubReg_i16_reg imm:$lane))),
1994 (SubReg_i16_lane imm:$lane)))>;
1995 def : Pat<(v4i32 (mul (v4i32 QPR:$src1),
1996 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
1997 (v4i32 (VMULslv4i32 (v4i32 QPR:$src1),
1998 (v2i32 (EXTRACT_SUBREG QPR:$src2,
1999 (DSubReg_i32_reg imm:$lane))),
2000 (SubReg_i32_lane imm:$lane)))>;
2001 def : Pat<(v4f32 (fmul (v4f32 QPR:$src1),
2002 (v4f32 (NEONvduplane (v4f32 QPR:$src2), imm:$lane)))),
2003 (v4f32 (VMULslfq (v4f32 QPR:$src1),
2004 (v2f32 (EXTRACT_SUBREG QPR:$src2,
2005 (DSubReg_i32_reg imm:$lane))),
2006 (SubReg_i32_lane imm:$lane)))>;
2008 // VQDMULH : Vector Saturating Doubling Multiply Returning High Half
2009 defm VQDMULH : N3VInt_HS<0, 0, 0b1011, 0, IIC_VMULi16D, IIC_VMULi32D,
2010 IIC_VMULi16Q, IIC_VMULi32Q,
2011 "vqdmulh", "s", int_arm_neon_vqdmulh, 1>;
2012 defm VQDMULHsl: N3VIntSL_HS<0b1100, IIC_VMULi16D, IIC_VMULi32D,
2013 IIC_VMULi16Q, IIC_VMULi32Q,
2014 "vqdmulh", "s", int_arm_neon_vqdmulh>;
2015 def : Pat<(v8i16 (int_arm_neon_vqdmulh (v8i16 QPR:$src1),
2016 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
2018 (v8i16 (VQDMULHslv8i16 (v8i16 QPR:$src1),
2019 (v4i16 (EXTRACT_SUBREG QPR:$src2,
2020 (DSubReg_i16_reg imm:$lane))),
2021 (SubReg_i16_lane imm:$lane)))>;
2022 def : Pat<(v4i32 (int_arm_neon_vqdmulh (v4i32 QPR:$src1),
2023 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
2025 (v4i32 (VQDMULHslv4i32 (v4i32 QPR:$src1),
2026 (v2i32 (EXTRACT_SUBREG QPR:$src2,
2027 (DSubReg_i32_reg imm:$lane))),
2028 (SubReg_i32_lane imm:$lane)))>;
2030 // VQRDMULH : Vector Rounding Saturating Doubling Multiply Returning High Half
2031 defm VQRDMULH : N3VInt_HS<1, 0, 0b1011, 0, IIC_VMULi16D, IIC_VMULi32D,
2032 IIC_VMULi16Q, IIC_VMULi32Q,
2033 "vqrdmulh", "s", int_arm_neon_vqrdmulh, 1>;
2034 defm VQRDMULHsl : N3VIntSL_HS<0b1101, IIC_VMULi16D, IIC_VMULi32D,
2035 IIC_VMULi16Q, IIC_VMULi32Q,
2036 "vqrdmulh", "s", int_arm_neon_vqrdmulh>;
2037 def : Pat<(v8i16 (int_arm_neon_vqrdmulh (v8i16 QPR:$src1),
2038 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
2040 (v8i16 (VQRDMULHslv8i16 (v8i16 QPR:$src1),
2041 (v4i16 (EXTRACT_SUBREG QPR:$src2,
2042 (DSubReg_i16_reg imm:$lane))),
2043 (SubReg_i16_lane imm:$lane)))>;
2044 def : Pat<(v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$src1),
2045 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
2047 (v4i32 (VQRDMULHslv4i32 (v4i32 QPR:$src1),
2048 (v2i32 (EXTRACT_SUBREG QPR:$src2,
2049 (DSubReg_i32_reg imm:$lane))),
2050 (SubReg_i32_lane imm:$lane)))>;
2052 // VMULL : Vector Multiply Long (integer and polynomial) (Q = D * D)
2053 defm VMULLs : N3VLInt_QHS<0,1,0b1100,0, IIC_VMULi16D, "vmull", "s",
2054 int_arm_neon_vmulls, 1>;
2055 defm VMULLu : N3VLInt_QHS<1,1,0b1100,0, IIC_VMULi16D, "vmull", "u",
2056 int_arm_neon_vmullu, 1>;
2057 def VMULLp : N3VLInt<0, 1, 0b00, 0b1110, 0, IIC_VMULi16D, "vmull", "p8",
2058 v8i16, v8i8, int_arm_neon_vmullp, 1>;
2059 defm VMULLsls : N3VLIntSL_HS<0, 0b1010, IIC_VMULi16D, "vmull", "s",
2060 int_arm_neon_vmulls>;
2061 defm VMULLslu : N3VLIntSL_HS<1, 0b1010, IIC_VMULi16D, "vmull", "u",
2062 int_arm_neon_vmullu>;
2064 // VQDMULL : Vector Saturating Doubling Multiply Long (Q = D * D)
2065 defm VQDMULL : N3VLInt_HS<0,1,0b1101,0, IIC_VMULi16D, "vqdmull", "s",
2066 int_arm_neon_vqdmull, 1>;
2067 defm VQDMULLsl: N3VLIntSL_HS<0, 0b1011, IIC_VMULi16D, "vqdmull", "s",
2068 int_arm_neon_vqdmull>;
2070 // Vector Multiply-Accumulate and Multiply-Subtract Operations.
2072 // VMLA : Vector Multiply Accumulate (integer and floating-point)
2073 defm VMLA : N3VMulOp_QHS<0, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
2074 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
2075 def VMLAfd : N3VDMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACD, "vmla", "f32",
2077 def VMLAfq : N3VQMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACQ, "vmla", "f32",
2079 defm VMLAsl : N3VMulOpSL_HS<0b0000, IIC_VMACi16D, IIC_VMACi32D,
2080 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
2081 def VMLAslfd : N3VDMulOpSL<0b10, 0b0001, IIC_VMACD, "vmla", "f32",
2083 def VMLAslfq : N3VQMulOpSL<0b10, 0b0001, IIC_VMACQ, "vmla", "f32",
2084 v4f32, v2f32, fmul, fadd>;
2086 def : Pat<(v8i16 (add (v8i16 QPR:$src1),
2087 (mul (v8i16 QPR:$src2),
2088 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
2089 (v8i16 (VMLAslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
2090 (v4i16 (EXTRACT_SUBREG QPR:$src3,
2091 (DSubReg_i16_reg imm:$lane))),
2092 (SubReg_i16_lane imm:$lane)))>;
2094 def : Pat<(v4i32 (add (v4i32 QPR:$src1),
2095 (mul (v4i32 QPR:$src2),
2096 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
2097 (v4i32 (VMLAslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
2098 (v2i32 (EXTRACT_SUBREG QPR:$src3,
2099 (DSubReg_i32_reg imm:$lane))),
2100 (SubReg_i32_lane imm:$lane)))>;
2102 def : Pat<(v4f32 (fadd (v4f32 QPR:$src1),
2103 (fmul (v4f32 QPR:$src2),
2104 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
2105 (v4f32 (VMLAslfq (v4f32 QPR:$src1),
2107 (v2f32 (EXTRACT_SUBREG QPR:$src3,
2108 (DSubReg_i32_reg imm:$lane))),
2109 (SubReg_i32_lane imm:$lane)))>;
2111 // VMLAL : Vector Multiply Accumulate Long (Q += D * D)
2112 defm VMLALs : N3VLInt3_QHS<0,1,0b1000,0, "vmlal", "s", int_arm_neon_vmlals>;
2113 defm VMLALu : N3VLInt3_QHS<1,1,0b1000,0, "vmlal", "u", int_arm_neon_vmlalu>;
2115 defm VMLALsls : N3VLInt3SL_HS<0, 0b0010, "vmlal", "s", int_arm_neon_vmlals>;
2116 defm VMLALslu : N3VLInt3SL_HS<1, 0b0010, "vmlal", "u", int_arm_neon_vmlalu>;
2118 // VQDMLAL : Vector Saturating Doubling Multiply Accumulate Long (Q += D * D)
2119 defm VQDMLAL : N3VLInt3_HS<0, 1, 0b1001, 0, "vqdmlal", "s",
2120 int_arm_neon_vqdmlal>;
2121 defm VQDMLALsl: N3VLInt3SL_HS<0, 0b0011, "vqdmlal", "s", int_arm_neon_vqdmlal>;
2123 // VMLS : Vector Multiply Subtract (integer and floating-point)
2124 defm VMLS : N3VMulOp_QHS<1, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
2125 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
2126 def VMLSfd : N3VDMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACD, "vmls", "f32",
2128 def VMLSfq : N3VQMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACQ, "vmls", "f32",
2130 defm VMLSsl : N3VMulOpSL_HS<0b0100, IIC_VMACi16D, IIC_VMACi32D,
2131 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
2132 def VMLSslfd : N3VDMulOpSL<0b10, 0b0101, IIC_VMACD, "vmls", "f32",
2134 def VMLSslfq : N3VQMulOpSL<0b10, 0b0101, IIC_VMACQ, "vmls", "f32",
2135 v4f32, v2f32, fmul, fsub>;
2137 def : Pat<(v8i16 (sub (v8i16 QPR:$src1),
2138 (mul (v8i16 QPR:$src2),
2139 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
2140 (v8i16 (VMLSslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
2141 (v4i16 (EXTRACT_SUBREG QPR:$src3,
2142 (DSubReg_i16_reg imm:$lane))),
2143 (SubReg_i16_lane imm:$lane)))>;
2145 def : Pat<(v4i32 (sub (v4i32 QPR:$src1),
2146 (mul (v4i32 QPR:$src2),
2147 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
2148 (v4i32 (VMLSslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
2149 (v2i32 (EXTRACT_SUBREG QPR:$src3,
2150 (DSubReg_i32_reg imm:$lane))),
2151 (SubReg_i32_lane imm:$lane)))>;
2153 def : Pat<(v4f32 (fsub (v4f32 QPR:$src1),
2154 (fmul (v4f32 QPR:$src2),
2155 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
2156 (v4f32 (VMLSslfq (v4f32 QPR:$src1), (v4f32 QPR:$src2),
2157 (v2f32 (EXTRACT_SUBREG QPR:$src3,
2158 (DSubReg_i32_reg imm:$lane))),
2159 (SubReg_i32_lane imm:$lane)))>;
2161 // VMLSL : Vector Multiply Subtract Long (Q -= D * D)
2162 defm VMLSLs : N3VLInt3_QHS<0,1,0b1010,0, "vmlsl", "s", int_arm_neon_vmlsls>;
2163 defm VMLSLu : N3VLInt3_QHS<1,1,0b1010,0, "vmlsl", "u", int_arm_neon_vmlslu>;
2165 defm VMLSLsls : N3VLInt3SL_HS<0, 0b0110, "vmlsl", "s", int_arm_neon_vmlsls>;
2166 defm VMLSLslu : N3VLInt3SL_HS<1, 0b0110, "vmlsl", "u", int_arm_neon_vmlslu>;
2168 // VQDMLSL : Vector Saturating Doubling Multiply Subtract Long (Q -= D * D)
2169 defm VQDMLSL : N3VLInt3_HS<0, 1, 0b1011, 0, "vqdmlsl", "s",
2170 int_arm_neon_vqdmlsl>;
2171 defm VQDMLSLsl: N3VLInt3SL_HS<0, 0b111, "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
2173 // Vector Subtract Operations.
2175 // VSUB : Vector Subtract (integer and floating-point)
2176 defm VSUB : N3V_QHSD<1, 0, 0b1000, 0, IIC_VSUBiD, IIC_VSUBiQ,
2177 "vsub", "i", sub, 0>;
2178 def VSUBfd : N3VD<0, 0, 0b10, 0b1101, 0, IIC_VBIND, "vsub", "f32",
2179 v2f32, v2f32, fsub, 0>;
2180 def VSUBfq : N3VQ<0, 0, 0b10, 0b1101, 0, IIC_VBINQ, "vsub", "f32",
2181 v4f32, v4f32, fsub, 0>;
2182 // VSUBL : Vector Subtract Long (Q = D - D)
2183 defm VSUBLs : N3VLInt_QHS<0,1,0b0010,0, IIC_VSHLiD, "vsubl", "s",
2184 int_arm_neon_vsubls, 1>;
2185 defm VSUBLu : N3VLInt_QHS<1,1,0b0010,0, IIC_VSHLiD, "vsubl", "u",
2186 int_arm_neon_vsublu, 1>;
2187 // VSUBW : Vector Subtract Wide (Q = Q - D)
2188 defm VSUBWs : N3VWInt_QHS<0,1,0b0011,0, "vsubw", "s", int_arm_neon_vsubws, 0>;
2189 defm VSUBWu : N3VWInt_QHS<1,1,0b0011,0, "vsubw", "u", int_arm_neon_vsubwu, 0>;
2190 // VHSUB : Vector Halving Subtract
2191 defm VHSUBs : N3VInt_QHS<0, 0, 0b0010, 0, IIC_VBINi4D, IIC_VBINi4D,
2192 IIC_VBINi4Q, IIC_VBINi4Q,
2193 "vhsub", "s", int_arm_neon_vhsubs, 0>;
2194 defm VHSUBu : N3VInt_QHS<1, 0, 0b0010, 0, IIC_VBINi4D, IIC_VBINi4D,
2195 IIC_VBINi4Q, IIC_VBINi4Q,
2196 "vhsub", "u", int_arm_neon_vhsubu, 0>;
2197 // VQSUB : Vector Saturing Subtract
2198 defm VQSUBs : N3VInt_QHSD<0, 0, 0b0010, 1, IIC_VBINi4D, IIC_VBINi4D,
2199 IIC_VBINi4Q, IIC_VBINi4Q,
2200 "vqsub", "s", int_arm_neon_vqsubs, 0>;
2201 defm VQSUBu : N3VInt_QHSD<1, 0, 0b0010, 1, IIC_VBINi4D, IIC_VBINi4D,
2202 IIC_VBINi4Q, IIC_VBINi4Q,
2203 "vqsub", "u", int_arm_neon_vqsubu, 0>;
2204 // VSUBHN : Vector Subtract and Narrow Returning High Half (D = Q - Q)
2205 defm VSUBHN : N3VNInt_HSD<0,1,0b0110,0, "vsubhn", "i",
2206 int_arm_neon_vsubhn, 0>;
2207 // VRSUBHN : Vector Rounding Subtract and Narrow Returning High Half (D=Q-Q)
2208 defm VRSUBHN : N3VNInt_HSD<1,1,0b0110,0, "vrsubhn", "i",
2209 int_arm_neon_vrsubhn, 0>;
2211 // Vector Comparisons.
2213 // VCEQ : Vector Compare Equal
2214 defm VCEQ : N3V_QHS<1, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2215 IIC_VBINi4Q, "vceq", "i", NEONvceq, 1>;
2216 def VCEQfd : N3VD<0,0,0b00,0b1110,0, IIC_VBIND, "vceq", "f32", v2i32, v2f32,
2218 def VCEQfq : N3VQ<0,0,0b00,0b1110,0, IIC_VBINQ, "vceq", "f32", v4i32, v4f32,
2220 // For disassembly only.
2221 defm VCEQz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00010, 0, "vceq", "i",
2224 // VCGE : Vector Compare Greater Than or Equal
2225 defm VCGEs : N3V_QHS<0, 0, 0b0011, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2226 IIC_VBINi4Q, "vcge", "s", NEONvcge, 0>;
2227 defm VCGEu : N3V_QHS<1, 0, 0b0011, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2228 IIC_VBINi4Q, "vcge", "u", NEONvcgeu, 0>;
2229 def VCGEfd : N3VD<1,0,0b00,0b1110,0, IIC_VBIND, "vcge", "f32",
2230 v2i32, v2f32, NEONvcge, 0>;
2231 def VCGEfq : N3VQ<1,0,0b00,0b1110,0, IIC_VBINQ, "vcge", "f32", v4i32, v4f32,
2233 // For disassembly only.
2234 defm VCGEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00001, 0, "vcge", "s",
2236 // For disassembly only.
2237 defm VCLEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00011, 0, "vcle", "s",
2240 // VCGT : Vector Compare Greater Than
2241 defm VCGTs : N3V_QHS<0, 0, 0b0011, 0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2242 IIC_VBINi4Q, "vcgt", "s", NEONvcgt, 0>;
2243 defm VCGTu : N3V_QHS<1, 0, 0b0011, 0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2244 IIC_VBINi4Q, "vcgt", "u", NEONvcgtu, 0>;
2245 def VCGTfd : N3VD<1,0,0b10,0b1110,0, IIC_VBIND, "vcgt", "f32", v2i32, v2f32,
2247 def VCGTfq : N3VQ<1,0,0b10,0b1110,0, IIC_VBINQ, "vcgt", "f32", v4i32, v4f32,
2249 // For disassembly only.
2250 defm VCGTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00000, 0, "vcgt", "s",
2252 // For disassembly only.
2253 defm VCLTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00100, 0, "vclt", "s",
2256 // VACGE : Vector Absolute Compare Greater Than or Equal (aka VCAGE)
2257 def VACGEd : N3VDInt<1, 0, 0b00, 0b1110, 1, IIC_VBIND, "vacge", "f32",
2258 v2i32, v2f32, int_arm_neon_vacged, 0>;
2259 def VACGEq : N3VQInt<1, 0, 0b00, 0b1110, 1, IIC_VBINQ, "vacge", "f32",
2260 v4i32, v4f32, int_arm_neon_vacgeq, 0>;
2261 // VACGT : Vector Absolute Compare Greater Than (aka VCAGT)
2262 def VACGTd : N3VDInt<1, 0, 0b10, 0b1110, 1, IIC_VBIND, "vacgt", "f32",
2263 v2i32, v2f32, int_arm_neon_vacgtd, 0>;
2264 def VACGTq : N3VQInt<1, 0, 0b10, 0b1110, 1, IIC_VBINQ, "vacgt", "f32",
2265 v4i32, v4f32, int_arm_neon_vacgtq, 0>;
2266 // VTST : Vector Test Bits
2267 defm VTST : N3V_QHS<0, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2268 IIC_VBINi4Q, "vtst", "", NEONvtst, 1>;
2270 // Vector Bitwise Operations.
2272 // VAND : Vector Bitwise AND
2273 def VANDd : N3VDX<0, 0, 0b00, 0b0001, 1, IIC_VBINiD, "vand",
2274 v2i32, v2i32, and, 1>;
2275 def VANDq : N3VQX<0, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "vand",
2276 v4i32, v4i32, and, 1>;
2278 // VEOR : Vector Bitwise Exclusive OR
2279 def VEORd : N3VDX<1, 0, 0b00, 0b0001, 1, IIC_VBINiD, "veor",
2280 v2i32, v2i32, xor, 1>;
2281 def VEORq : N3VQX<1, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "veor",
2282 v4i32, v4i32, xor, 1>;
2284 // VORR : Vector Bitwise OR
2285 def VORRd : N3VDX<0, 0, 0b10, 0b0001, 1, IIC_VBINiD, "vorr",
2286 v2i32, v2i32, or, 1>;
2287 def VORRq : N3VQX<0, 0, 0b10, 0b0001, 1, IIC_VBINiQ, "vorr",
2288 v4i32, v4i32, or, 1>;
2290 // VBIC : Vector Bitwise Bit Clear (AND NOT)
2291 def VBICd : N3VX<0, 0, 0b01, 0b0001, 0, 1, (outs DPR:$dst),
2292 (ins DPR:$src1, DPR:$src2), IIC_VBINiD,
2293 "vbic", "$dst, $src1, $src2", "",
2294 [(set DPR:$dst, (v2i32 (and DPR:$src1,
2295 (vnot_conv DPR:$src2))))]>;
2296 def VBICq : N3VX<0, 0, 0b01, 0b0001, 1, 1, (outs QPR:$dst),
2297 (ins QPR:$src1, QPR:$src2), IIC_VBINiQ,
2298 "vbic", "$dst, $src1, $src2", "",
2299 [(set QPR:$dst, (v4i32 (and QPR:$src1,
2300 (vnot_conv QPR:$src2))))]>;
2302 // VORN : Vector Bitwise OR NOT
2303 def VORNd : N3VX<0, 0, 0b11, 0b0001, 0, 1, (outs DPR:$dst),
2304 (ins DPR:$src1, DPR:$src2), IIC_VBINiD,
2305 "vorn", "$dst, $src1, $src2", "",
2306 [(set DPR:$dst, (v2i32 (or DPR:$src1,
2307 (vnot_conv DPR:$src2))))]>;
2308 def VORNq : N3VX<0, 0, 0b11, 0b0001, 1, 1, (outs QPR:$dst),
2309 (ins QPR:$src1, QPR:$src2), IIC_VBINiQ,
2310 "vorn", "$dst, $src1, $src2", "",
2311 [(set QPR:$dst, (v4i32 (or QPR:$src1,
2312 (vnot_conv QPR:$src2))))]>;
2314 // VMVN : Vector Bitwise NOT
2315 def VMVNd : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 0, 0,
2316 (outs DPR:$dst), (ins DPR:$src), IIC_VSHLiD,
2317 "vmvn", "$dst, $src", "",
2318 [(set DPR:$dst, (v2i32 (vnot DPR:$src)))]>;
2319 def VMVNq : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 1, 0,
2320 (outs QPR:$dst), (ins QPR:$src), IIC_VSHLiD,
2321 "vmvn", "$dst, $src", "",
2322 [(set QPR:$dst, (v4i32 (vnot QPR:$src)))]>;
2323 def : Pat<(v2i32 (vnot_conv DPR:$src)), (VMVNd DPR:$src)>;
2324 def : Pat<(v4i32 (vnot_conv QPR:$src)), (VMVNq QPR:$src)>;
2326 // VBSL : Vector Bitwise Select
2327 def VBSLd : N3VX<1, 0, 0b01, 0b0001, 0, 1, (outs DPR:$dst),
2328 (ins DPR:$src1, DPR:$src2, DPR:$src3), IIC_VCNTiD,
2329 "vbsl", "$dst, $src2, $src3", "$src1 = $dst",
2331 (v2i32 (or (and DPR:$src2, DPR:$src1),
2332 (and DPR:$src3, (vnot_conv DPR:$src1)))))]>;
2333 def VBSLq : N3VX<1, 0, 0b01, 0b0001, 1, 1, (outs QPR:$dst),
2334 (ins QPR:$src1, QPR:$src2, QPR:$src3), IIC_VCNTiQ,
2335 "vbsl", "$dst, $src2, $src3", "$src1 = $dst",
2337 (v4i32 (or (and QPR:$src2, QPR:$src1),
2338 (and QPR:$src3, (vnot_conv QPR:$src1)))))]>;
2340 // VBIF : Vector Bitwise Insert if False
2341 // like VBSL but with: "vbif $dst, $src3, $src1", "$src2 = $dst",
2342 def VBIFd : N3VX<1, 0, 0b11, 0b0001, 0, 1,
2343 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3),
2344 IIC_VBINiD, "vbif", "$dst, $src2, $src3", "$src1 = $dst",
2345 [/* For disassembly only; pattern left blank */]>;
2346 def VBIFq : N3VX<1, 0, 0b11, 0b0001, 1, 1,
2347 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3),
2348 IIC_VBINiQ, "vbif", "$dst, $src2, $src3", "$src1 = $dst",
2349 [/* For disassembly only; pattern left blank */]>;
2351 // VBIT : Vector Bitwise Insert if True
2352 // like VBSL but with: "vbit $dst, $src2, $src1", "$src3 = $dst",
2353 def VBITd : N3VX<1, 0, 0b10, 0b0001, 0, 1,
2354 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3),
2355 IIC_VBINiD, "vbit", "$dst, $src2, $src3", "$src1 = $dst",
2356 [/* For disassembly only; pattern left blank */]>;
2357 def VBITq : N3VX<1, 0, 0b10, 0b0001, 1, 1,
2358 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3),
2359 IIC_VBINiQ, "vbit", "$dst, $src2, $src3", "$src1 = $dst",
2360 [/* For disassembly only; pattern left blank */]>;
2362 // VBIT/VBIF are not yet implemented. The TwoAddress pass will not go looking
2363 // for equivalent operations with different register constraints; it just
2366 // Vector Absolute Differences.
2368 // VABD : Vector Absolute Difference
2369 defm VABDs : N3VInt_QHS<0, 0, 0b0111, 0, IIC_VBINi4D, IIC_VBINi4D,
2370 IIC_VBINi4Q, IIC_VBINi4Q,
2371 "vabd", "s", int_arm_neon_vabds, 0>;
2372 defm VABDu : N3VInt_QHS<1, 0, 0b0111, 0, IIC_VBINi4D, IIC_VBINi4D,
2373 IIC_VBINi4Q, IIC_VBINi4Q,
2374 "vabd", "u", int_arm_neon_vabdu, 0>;
2375 def VABDfd : N3VDInt<1, 0, 0b10, 0b1101, 0, IIC_VBIND,
2376 "vabd", "f32", v2f32, v2f32, int_arm_neon_vabds, 0>;
2377 def VABDfq : N3VQInt<1, 0, 0b10, 0b1101, 0, IIC_VBINQ,
2378 "vabd", "f32", v4f32, v4f32, int_arm_neon_vabds, 0>;
2380 // VABDL : Vector Absolute Difference Long (Q = | D - D |)
2381 defm VABDLs : N3VLInt_QHS<0,1,0b0111,0, IIC_VBINi4Q,
2382 "vabdl", "s", int_arm_neon_vabdls, 0>;
2383 defm VABDLu : N3VLInt_QHS<1,1,0b0111,0, IIC_VBINi4Q,
2384 "vabdl", "u", int_arm_neon_vabdlu, 0>;
2386 // VABA : Vector Absolute Difference and Accumulate
2387 defm VABAs : N3VInt3_QHS<0,0,0b0111,1, "vaba", "s", int_arm_neon_vabas>;
2388 defm VABAu : N3VInt3_QHS<1,0,0b0111,1, "vaba", "u", int_arm_neon_vabau>;
2390 // VABAL : Vector Absolute Difference and Accumulate Long (Q += | D - D |)
2391 defm VABALs : N3VLInt3_QHS<0,1,0b0101,0, "vabal", "s", int_arm_neon_vabals>;
2392 defm VABALu : N3VLInt3_QHS<1,1,0b0101,0, "vabal", "u", int_arm_neon_vabalu>;
2394 // Vector Maximum and Minimum.
2396 // VMAX : Vector Maximum
2397 defm VMAXs : N3VInt_QHS<0,0,0b0110,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2398 IIC_VBINi4Q, "vmax", "s", int_arm_neon_vmaxs, 1>;
2399 defm VMAXu : N3VInt_QHS<1,0,0b0110,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2400 IIC_VBINi4Q, "vmax", "u", int_arm_neon_vmaxu, 1>;
2401 def VMAXfd : N3VDInt<0, 0, 0b00, 0b1111, 0, IIC_VBIND, "vmax", "f32",
2402 v2f32, v2f32, int_arm_neon_vmaxs, 1>;
2403 def VMAXfq : N3VQInt<0, 0, 0b00, 0b1111, 0, IIC_VBINQ, "vmax", "f32",
2404 v4f32, v4f32, int_arm_neon_vmaxs, 1>;
2406 // VMIN : Vector Minimum
2407 defm VMINs : N3VInt_QHS<0,0,0b0110,1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2408 IIC_VBINi4Q, "vmin", "s", int_arm_neon_vmins, 1>;
2409 defm VMINu : N3VInt_QHS<1,0,0b0110,1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2410 IIC_VBINi4Q, "vmin", "u", int_arm_neon_vminu, 1>;
2411 def VMINfd : N3VDInt<0, 0, 0b10, 0b1111, 0, IIC_VBIND, "vmin", "f32",
2412 v2f32, v2f32, int_arm_neon_vmins, 1>;
2413 def VMINfq : N3VQInt<0, 0, 0b10, 0b1111, 0, IIC_VBINQ, "vmin", "f32",
2414 v4f32, v4f32, int_arm_neon_vmins, 1>;
2416 // Vector Pairwise Operations.
2418 // VPADD : Vector Pairwise Add
2419 def VPADDi8 : N3VDInt<0, 0, 0b00, 0b1011, 1, IIC_VBINiD, "vpadd", "i8",
2420 v8i8, v8i8, int_arm_neon_vpadd, 0>;
2421 def VPADDi16 : N3VDInt<0, 0, 0b01, 0b1011, 1, IIC_VBINiD, "vpadd", "i16",
2422 v4i16, v4i16, int_arm_neon_vpadd, 0>;
2423 def VPADDi32 : N3VDInt<0, 0, 0b10, 0b1011, 1, IIC_VBINiD, "vpadd", "i32",
2424 v2i32, v2i32, int_arm_neon_vpadd, 0>;
2425 def VPADDf : N3VDInt<1, 0, 0b00, 0b1101, 0, IIC_VBIND, "vpadd", "f32",
2426 v2f32, v2f32, int_arm_neon_vpadd, 0>;
2428 // VPADDL : Vector Pairwise Add Long
2429 defm VPADDLs : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00100, 0, "vpaddl", "s",
2430 int_arm_neon_vpaddls>;
2431 defm VPADDLu : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00101, 0, "vpaddl", "u",
2432 int_arm_neon_vpaddlu>;
2434 // VPADAL : Vector Pairwise Add and Accumulate Long
2435 defm VPADALs : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01100, 0, "vpadal", "s",
2436 int_arm_neon_vpadals>;
2437 defm VPADALu : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01101, 0, "vpadal", "u",
2438 int_arm_neon_vpadalu>;
2440 // VPMAX : Vector Pairwise Maximum
2441 def VPMAXs8 : N3VDInt<0, 0, 0b00, 0b1010, 0, IIC_VBINi4D, "vpmax", "s8",
2442 v8i8, v8i8, int_arm_neon_vpmaxs, 0>;
2443 def VPMAXs16 : N3VDInt<0, 0, 0b01, 0b1010, 0, IIC_VBINi4D, "vpmax", "s16",
2444 v4i16, v4i16, int_arm_neon_vpmaxs, 0>;
2445 def VPMAXs32 : N3VDInt<0, 0, 0b10, 0b1010, 0, IIC_VBINi4D, "vpmax", "s32",
2446 v2i32, v2i32, int_arm_neon_vpmaxs, 0>;
2447 def VPMAXu8 : N3VDInt<1, 0, 0b00, 0b1010, 0, IIC_VBINi4D, "vpmax", "u8",
2448 v8i8, v8i8, int_arm_neon_vpmaxu, 0>;
2449 def VPMAXu16 : N3VDInt<1, 0, 0b01, 0b1010, 0, IIC_VBINi4D, "vpmax", "u16",
2450 v4i16, v4i16, int_arm_neon_vpmaxu, 0>;
2451 def VPMAXu32 : N3VDInt<1, 0, 0b10, 0b1010, 0, IIC_VBINi4D, "vpmax", "u32",
2452 v2i32, v2i32, int_arm_neon_vpmaxu, 0>;
2453 def VPMAXf : N3VDInt<1, 0, 0b00, 0b1111, 0, IIC_VBINi4D, "vpmax", "f32",
2454 v2f32, v2f32, int_arm_neon_vpmaxs, 0>;
2456 // VPMIN : Vector Pairwise Minimum
2457 def VPMINs8 : N3VDInt<0, 0, 0b00, 0b1010, 1, IIC_VBINi4D, "vpmin", "s8",
2458 v8i8, v8i8, int_arm_neon_vpmins, 0>;
2459 def VPMINs16 : N3VDInt<0, 0, 0b01, 0b1010, 1, IIC_VBINi4D, "vpmin", "s16",
2460 v4i16, v4i16, int_arm_neon_vpmins, 0>;
2461 def VPMINs32 : N3VDInt<0, 0, 0b10, 0b1010, 1, IIC_VBINi4D, "vpmin", "s32",
2462 v2i32, v2i32, int_arm_neon_vpmins, 0>;
2463 def VPMINu8 : N3VDInt<1, 0, 0b00, 0b1010, 1, IIC_VBINi4D, "vpmin", "u8",
2464 v8i8, v8i8, int_arm_neon_vpminu, 0>;
2465 def VPMINu16 : N3VDInt<1, 0, 0b01, 0b1010, 1, IIC_VBINi4D, "vpmin", "u16",
2466 v4i16, v4i16, int_arm_neon_vpminu, 0>;
2467 def VPMINu32 : N3VDInt<1, 0, 0b10, 0b1010, 1, IIC_VBINi4D, "vpmin", "u32",
2468 v2i32, v2i32, int_arm_neon_vpminu, 0>;
2469 def VPMINf : N3VDInt<1, 0, 0b10, 0b1111, 0, IIC_VBINi4D, "vpmin", "f32",
2470 v2f32, v2f32, int_arm_neon_vpmins, 0>;
2472 // Vector Reciprocal and Reciprocal Square Root Estimate and Step.
2474 // VRECPE : Vector Reciprocal Estimate
2475 def VRECPEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
2476 IIC_VUNAD, "vrecpe", "u32",
2477 v2i32, v2i32, int_arm_neon_vrecpe>;
2478 def VRECPEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
2479 IIC_VUNAQ, "vrecpe", "u32",
2480 v4i32, v4i32, int_arm_neon_vrecpe>;
2481 def VRECPEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
2482 IIC_VUNAD, "vrecpe", "f32",
2483 v2f32, v2f32, int_arm_neon_vrecpe>;
2484 def VRECPEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
2485 IIC_VUNAQ, "vrecpe", "f32",
2486 v4f32, v4f32, int_arm_neon_vrecpe>;
2488 // VRECPS : Vector Reciprocal Step
2489 def VRECPSfd : N3VDInt<0, 0, 0b00, 0b1111, 1,
2490 IIC_VRECSD, "vrecps", "f32",
2491 v2f32, v2f32, int_arm_neon_vrecps, 1>;
2492 def VRECPSfq : N3VQInt<0, 0, 0b00, 0b1111, 1,
2493 IIC_VRECSQ, "vrecps", "f32",
2494 v4f32, v4f32, int_arm_neon_vrecps, 1>;
2496 // VRSQRTE : Vector Reciprocal Square Root Estimate
2497 def VRSQRTEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
2498 IIC_VUNAD, "vrsqrte", "u32",
2499 v2i32, v2i32, int_arm_neon_vrsqrte>;
2500 def VRSQRTEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
2501 IIC_VUNAQ, "vrsqrte", "u32",
2502 v4i32, v4i32, int_arm_neon_vrsqrte>;
2503 def VRSQRTEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
2504 IIC_VUNAD, "vrsqrte", "f32",
2505 v2f32, v2f32, int_arm_neon_vrsqrte>;
2506 def VRSQRTEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
2507 IIC_VUNAQ, "vrsqrte", "f32",
2508 v4f32, v4f32, int_arm_neon_vrsqrte>;
2510 // VRSQRTS : Vector Reciprocal Square Root Step
2511 def VRSQRTSfd : N3VDInt<0, 0, 0b10, 0b1111, 1,
2512 IIC_VRECSD, "vrsqrts", "f32",
2513 v2f32, v2f32, int_arm_neon_vrsqrts, 1>;
2514 def VRSQRTSfq : N3VQInt<0, 0, 0b10, 0b1111, 1,
2515 IIC_VRECSQ, "vrsqrts", "f32",
2516 v4f32, v4f32, int_arm_neon_vrsqrts, 1>;
2520 // VSHL : Vector Shift
2521 defm VSHLs : N3VInt_QHSD<0, 0, 0b0100, 0, IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ,
2522 IIC_VSHLiQ, "vshl", "s", int_arm_neon_vshifts, 0>;
2523 defm VSHLu : N3VInt_QHSD<1, 0, 0b0100, 0, IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ,
2524 IIC_VSHLiQ, "vshl", "u", int_arm_neon_vshiftu, 0>;
2525 // VSHL : Vector Shift Left (Immediate)
2526 defm VSHLi : N2VSh_QHSD<0, 1, 0b0101, 1, IIC_VSHLiD, "vshl", "i", NEONvshl>;
2527 // VSHR : Vector Shift Right (Immediate)
2528 defm VSHRs : N2VSh_QHSD<0, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "s", NEONvshrs>;
2529 defm VSHRu : N2VSh_QHSD<1, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "u", NEONvshru>;
2531 // VSHLL : Vector Shift Left Long
2532 defm VSHLLs : N2VLSh_QHS<0, 1, 0b1010, 0, 0, 1, "vshll", "s", NEONvshlls>;
2533 defm VSHLLu : N2VLSh_QHS<1, 1, 0b1010, 0, 0, 1, "vshll", "u", NEONvshllu>;
2535 // VSHLL : Vector Shift Left Long (with maximum shift count)
2536 class N2VLShMax<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
2537 bit op6, bit op4, string OpcodeStr, string Dt, ValueType ResTy,
2538 ValueType OpTy, SDNode OpNode>
2539 : N2VLSh<op24, op23, op11_8, op7, op6, op4, OpcodeStr, Dt,
2540 ResTy, OpTy, OpNode> {
2541 let Inst{21-16} = op21_16;
2543 def VSHLLi8 : N2VLShMax<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll", "i8",
2544 v8i16, v8i8, NEONvshlli>;
2545 def VSHLLi16 : N2VLShMax<1, 1, 0b110110, 0b0011, 0, 0, 0, "vshll", "i16",
2546 v4i32, v4i16, NEONvshlli>;
2547 def VSHLLi32 : N2VLShMax<1, 1, 0b111010, 0b0011, 0, 0, 0, "vshll", "i32",
2548 v2i64, v2i32, NEONvshlli>;
2550 // VSHRN : Vector Shift Right and Narrow
2551 defm VSHRN : N2VNSh_HSD<0,1,0b1000,0,0,1, IIC_VSHLiD, "vshrn", "i",
2554 // VRSHL : Vector Rounding Shift
2555 defm VRSHLs : N3VInt_QHSD<0,0,0b0101,0, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2556 IIC_VSHLi4Q, "vrshl", "s", int_arm_neon_vrshifts,0>;
2557 defm VRSHLu : N3VInt_QHSD<1,0,0b0101,0, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2558 IIC_VSHLi4Q, "vrshl", "u", int_arm_neon_vrshiftu,0>;
2559 // VRSHR : Vector Rounding Shift Right
2560 defm VRSHRs : N2VSh_QHSD<0,1,0b0010,1, IIC_VSHLi4D, "vrshr", "s", NEONvrshrs>;
2561 defm VRSHRu : N2VSh_QHSD<1,1,0b0010,1, IIC_VSHLi4D, "vrshr", "u", NEONvrshru>;
2563 // VRSHRN : Vector Rounding Shift Right and Narrow
2564 defm VRSHRN : N2VNSh_HSD<0, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vrshrn", "i",
2567 // VQSHL : Vector Saturating Shift
2568 defm VQSHLs : N3VInt_QHSD<0,0,0b0100,1, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2569 IIC_VSHLi4Q, "vqshl", "s", int_arm_neon_vqshifts,0>;
2570 defm VQSHLu : N3VInt_QHSD<1,0,0b0100,1, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2571 IIC_VSHLi4Q, "vqshl", "u", int_arm_neon_vqshiftu,0>;
2572 // VQSHL : Vector Saturating Shift Left (Immediate)
2573 defm VQSHLsi : N2VSh_QHSD<0,1,0b0111,1, IIC_VSHLi4D, "vqshl", "s", NEONvqshls>;
2574 defm VQSHLui : N2VSh_QHSD<1,1,0b0111,1, IIC_VSHLi4D, "vqshl", "u", NEONvqshlu>;
2575 // VQSHLU : Vector Saturating Shift Left (Immediate, Unsigned)
2576 defm VQSHLsu : N2VSh_QHSD<1,1,0b0110,1, IIC_VSHLi4D, "vqshlu","s",NEONvqshlsu>;
2578 // VQSHRN : Vector Saturating Shift Right and Narrow
2579 defm VQSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "s",
2581 defm VQSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "u",
2584 // VQSHRUN : Vector Saturating Shift Right and Narrow (Unsigned)
2585 defm VQSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 0, 1, IIC_VSHLi4D, "vqshrun", "s",
2588 // VQRSHL : Vector Saturating Rounding Shift
2589 defm VQRSHLs : N3VInt_QHSD<0,0,0b0101,1, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2590 IIC_VSHLi4Q, "vqrshl", "s",
2591 int_arm_neon_vqrshifts, 0>;
2592 defm VQRSHLu : N3VInt_QHSD<1,0,0b0101,1, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2593 IIC_VSHLi4Q, "vqrshl", "u",
2594 int_arm_neon_vqrshiftu, 0>;
2596 // VQRSHRN : Vector Saturating Rounding Shift Right and Narrow
2597 defm VQRSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "s",
2599 defm VQRSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "u",
2602 // VQRSHRUN : Vector Saturating Rounding Shift Right and Narrow (Unsigned)
2603 defm VQRSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vqrshrun", "s",
2606 // VSRA : Vector Shift Right and Accumulate
2607 defm VSRAs : N2VShAdd_QHSD<0, 1, 0b0001, 1, "vsra", "s", NEONvshrs>;
2608 defm VSRAu : N2VShAdd_QHSD<1, 1, 0b0001, 1, "vsra", "u", NEONvshru>;
2609 // VRSRA : Vector Rounding Shift Right and Accumulate
2610 defm VRSRAs : N2VShAdd_QHSD<0, 1, 0b0011, 1, "vrsra", "s", NEONvrshrs>;
2611 defm VRSRAu : N2VShAdd_QHSD<1, 1, 0b0011, 1, "vrsra", "u", NEONvrshru>;
2613 // VSLI : Vector Shift Left and Insert
2614 defm VSLI : N2VShIns_QHSD<1, 1, 0b0101, 1, "vsli", NEONvsli>;
2615 // VSRI : Vector Shift Right and Insert
2616 defm VSRI : N2VShIns_QHSD<1, 1, 0b0100, 1, "vsri", NEONvsri>;
2618 // Vector Absolute and Saturating Absolute.
2620 // VABS : Vector Absolute Value
2621 defm VABS : N2VInt_QHS<0b11, 0b11, 0b01, 0b00110, 0,
2622 IIC_VUNAiD, IIC_VUNAiQ, "vabs", "s",
2624 def VABSfd : N2VDInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
2625 IIC_VUNAD, "vabs", "f32",
2626 v2f32, v2f32, int_arm_neon_vabs>;
2627 def VABSfq : N2VQInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
2628 IIC_VUNAQ, "vabs", "f32",
2629 v4f32, v4f32, int_arm_neon_vabs>;
2631 // VQABS : Vector Saturating Absolute Value
2632 defm VQABS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01110, 0,
2633 IIC_VQUNAiD, IIC_VQUNAiQ, "vqabs", "s",
2634 int_arm_neon_vqabs>;
2638 def vneg : PatFrag<(ops node:$in), (sub immAllZerosV, node:$in)>;
2639 def vneg_conv : PatFrag<(ops node:$in), (sub immAllZerosV_bc, node:$in)>;
2641 class VNEGD<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
2642 : N2V<0b11, 0b11, size, 0b01, 0b00111, 0, 0, (outs DPR:$dst), (ins DPR:$src),
2643 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
2644 [(set DPR:$dst, (Ty (vneg DPR:$src)))]>;
2645 class VNEGQ<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
2646 : N2V<0b11, 0b11, size, 0b01, 0b00111, 1, 0, (outs QPR:$dst), (ins QPR:$src),
2647 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
2648 [(set QPR:$dst, (Ty (vneg QPR:$src)))]>;
2650 // VNEG : Vector Negate
2651 def VNEGs8d : VNEGD<0b00, "vneg", "s8", v8i8>;
2652 def VNEGs16d : VNEGD<0b01, "vneg", "s16", v4i16>;
2653 def VNEGs32d : VNEGD<0b10, "vneg", "s32", v2i32>;
2654 def VNEGs8q : VNEGQ<0b00, "vneg", "s8", v16i8>;
2655 def VNEGs16q : VNEGQ<0b01, "vneg", "s16", v8i16>;
2656 def VNEGs32q : VNEGQ<0b10, "vneg", "s32", v4i32>;
2658 // VNEG : Vector Negate (floating-point)
2659 def VNEGfd : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
2660 (outs DPR:$dst), (ins DPR:$src), IIC_VUNAD,
2661 "vneg", "f32", "$dst, $src", "",
2662 [(set DPR:$dst, (v2f32 (fneg DPR:$src)))]>;
2663 def VNEGf32q : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 1, 0,
2664 (outs QPR:$dst), (ins QPR:$src), IIC_VUNAQ,
2665 "vneg", "f32", "$dst, $src", "",
2666 [(set QPR:$dst, (v4f32 (fneg QPR:$src)))]>;
2668 def : Pat<(v8i8 (vneg_conv DPR:$src)), (VNEGs8d DPR:$src)>;
2669 def : Pat<(v4i16 (vneg_conv DPR:$src)), (VNEGs16d DPR:$src)>;
2670 def : Pat<(v2i32 (vneg_conv DPR:$src)), (VNEGs32d DPR:$src)>;
2671 def : Pat<(v16i8 (vneg_conv QPR:$src)), (VNEGs8q QPR:$src)>;
2672 def : Pat<(v8i16 (vneg_conv QPR:$src)), (VNEGs16q QPR:$src)>;
2673 def : Pat<(v4i32 (vneg_conv QPR:$src)), (VNEGs32q QPR:$src)>;
2675 // VQNEG : Vector Saturating Negate
2676 defm VQNEG : N2VInt_QHS<0b11, 0b11, 0b00, 0b01111, 0,
2677 IIC_VQUNAiD, IIC_VQUNAiQ, "vqneg", "s",
2678 int_arm_neon_vqneg>;
2680 // Vector Bit Counting Operations.
2682 // VCLS : Vector Count Leading Sign Bits
2683 defm VCLS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01000, 0,
2684 IIC_VCNTiD, IIC_VCNTiQ, "vcls", "s",
2686 // VCLZ : Vector Count Leading Zeros
2687 defm VCLZ : N2VInt_QHS<0b11, 0b11, 0b00, 0b01001, 0,
2688 IIC_VCNTiD, IIC_VCNTiQ, "vclz", "i",
2690 // VCNT : Vector Count One Bits
2691 def VCNTd : N2VDInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
2692 IIC_VCNTiD, "vcnt", "8",
2693 v8i8, v8i8, int_arm_neon_vcnt>;
2694 def VCNTq : N2VQInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
2695 IIC_VCNTiQ, "vcnt", "8",
2696 v16i8, v16i8, int_arm_neon_vcnt>;
2698 // Vector Swap -- for disassembly only.
2699 def VSWPd : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 0, 0,
2700 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
2701 "vswp", "$dst, $src", "", []>;
2702 def VSWPq : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 1, 0,
2703 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
2704 "vswp", "$dst, $src", "", []>;
2706 // Vector Move Operations.
2708 // VMOV : Vector Move (Register)
2710 def VMOVDneon: N3VX<0, 0, 0b10, 0b0001, 0, 1, (outs DPR:$dst), (ins DPR:$src),
2711 IIC_VMOVD, "vmov", "$dst, $src", "", []>;
2712 def VMOVQ : N3VX<0, 0, 0b10, 0b0001, 1, 1, (outs QPR:$dst), (ins QPR:$src),
2713 IIC_VMOVD, "vmov", "$dst, $src", "", []>;
2715 // VMOV : Vector Move (Immediate)
2717 // VMOV_get_imm8 xform function: convert build_vector to VMOV.i8 imm.
2718 def VMOV_get_imm8 : SDNodeXForm<build_vector, [{
2719 return ARM::getVMOVImm(N, 1, *CurDAG);
2721 def vmovImm8 : PatLeaf<(build_vector), [{
2722 return ARM::getVMOVImm(N, 1, *CurDAG).getNode() != 0;
2725 // VMOV_get_imm16 xform function: convert build_vector to VMOV.i16 imm.
2726 def VMOV_get_imm16 : SDNodeXForm<build_vector, [{
2727 return ARM::getVMOVImm(N, 2, *CurDAG);
2729 def vmovImm16 : PatLeaf<(build_vector), [{
2730 return ARM::getVMOVImm(N, 2, *CurDAG).getNode() != 0;
2731 }], VMOV_get_imm16>;
2733 // VMOV_get_imm32 xform function: convert build_vector to VMOV.i32 imm.
2734 def VMOV_get_imm32 : SDNodeXForm<build_vector, [{
2735 return ARM::getVMOVImm(N, 4, *CurDAG);
2737 def vmovImm32 : PatLeaf<(build_vector), [{
2738 return ARM::getVMOVImm(N, 4, *CurDAG).getNode() != 0;
2739 }], VMOV_get_imm32>;
2741 // VMOV_get_imm64 xform function: convert build_vector to VMOV.i64 imm.
2742 def VMOV_get_imm64 : SDNodeXForm<build_vector, [{
2743 return ARM::getVMOVImm(N, 8, *CurDAG);
2745 def vmovImm64 : PatLeaf<(build_vector), [{
2746 return ARM::getVMOVImm(N, 8, *CurDAG).getNode() != 0;
2747 }], VMOV_get_imm64>;
2749 // Note: Some of the cmode bits in the following VMOV instructions need to
2750 // be encoded based on the immed values.
2752 def VMOVv8i8 : N1ModImm<1, 0b000, 0b1110, 0, 0, 0, 1, (outs DPR:$dst),
2753 (ins h8imm:$SIMM), IIC_VMOVImm,
2754 "vmov", "i8", "$dst, $SIMM", "",
2755 [(set DPR:$dst, (v8i8 vmovImm8:$SIMM))]>;
2756 def VMOVv16i8 : N1ModImm<1, 0b000, 0b1110, 0, 1, 0, 1, (outs QPR:$dst),
2757 (ins h8imm:$SIMM), IIC_VMOVImm,
2758 "vmov", "i8", "$dst, $SIMM", "",
2759 [(set QPR:$dst, (v16i8 vmovImm8:$SIMM))]>;
2761 def VMOVv4i16 : N1ModImm<1, 0b000, {1,0,?,?}, 0, 0, {?}, 1, (outs DPR:$dst),
2762 (ins h16imm:$SIMM), IIC_VMOVImm,
2763 "vmov", "i16", "$dst, $SIMM", "",
2764 [(set DPR:$dst, (v4i16 vmovImm16:$SIMM))]>;
2765 def VMOVv8i16 : N1ModImm<1, 0b000, {1,0,?,?}, 0, 1, {?}, 1, (outs QPR:$dst),
2766 (ins h16imm:$SIMM), IIC_VMOVImm,
2767 "vmov", "i16", "$dst, $SIMM", "",
2768 [(set QPR:$dst, (v8i16 vmovImm16:$SIMM))]>;
2770 def VMOVv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, {?}, 1, (outs DPR:$dst),
2771 (ins h32imm:$SIMM), IIC_VMOVImm,
2772 "vmov", "i32", "$dst, $SIMM", "",
2773 [(set DPR:$dst, (v2i32 vmovImm32:$SIMM))]>;
2774 def VMOVv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, {?}, 1, (outs QPR:$dst),
2775 (ins h32imm:$SIMM), IIC_VMOVImm,
2776 "vmov", "i32", "$dst, $SIMM", "",
2777 [(set QPR:$dst, (v4i32 vmovImm32:$SIMM))]>;
2779 def VMOVv1i64 : N1ModImm<1, 0b000, 0b1110, 0, 0, 1, 1, (outs DPR:$dst),
2780 (ins h64imm:$SIMM), IIC_VMOVImm,
2781 "vmov", "i64", "$dst, $SIMM", "",
2782 [(set DPR:$dst, (v1i64 vmovImm64:$SIMM))]>;
2783 def VMOVv2i64 : N1ModImm<1, 0b000, 0b1110, 0, 1, 1, 1, (outs QPR:$dst),
2784 (ins h64imm:$SIMM), IIC_VMOVImm,
2785 "vmov", "i64", "$dst, $SIMM", "",
2786 [(set QPR:$dst, (v2i64 vmovImm64:$SIMM))]>;
2788 // VMOV : Vector Get Lane (move scalar to ARM core register)
2790 def VGETLNs8 : NVGetLane<{1,1,1,0,0,1,?,1}, 0b1011, {?,?},
2791 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2792 IIC_VMOVSI, "vmov", "s8", "$dst, $src[$lane]",
2793 [(set GPR:$dst, (NEONvgetlanes (v8i8 DPR:$src),
2795 def VGETLNs16 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, {?,1},
2796 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2797 IIC_VMOVSI, "vmov", "s16", "$dst, $src[$lane]",
2798 [(set GPR:$dst, (NEONvgetlanes (v4i16 DPR:$src),
2800 def VGETLNu8 : NVGetLane<{1,1,1,0,1,1,?,1}, 0b1011, {?,?},
2801 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2802 IIC_VMOVSI, "vmov", "u8", "$dst, $src[$lane]",
2803 [(set GPR:$dst, (NEONvgetlaneu (v8i8 DPR:$src),
2805 def VGETLNu16 : NVGetLane<{1,1,1,0,1,0,?,1}, 0b1011, {?,1},
2806 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2807 IIC_VMOVSI, "vmov", "u16", "$dst, $src[$lane]",
2808 [(set GPR:$dst, (NEONvgetlaneu (v4i16 DPR:$src),
2810 def VGETLNi32 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, 0b00,
2811 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2812 IIC_VMOVSI, "vmov", "32", "$dst, $src[$lane]",
2813 [(set GPR:$dst, (extractelt (v2i32 DPR:$src),
2815 // def VGETLNf32: see FMRDH and FMRDL in ARMInstrVFP.td
2816 def : Pat<(NEONvgetlanes (v16i8 QPR:$src), imm:$lane),
2817 (VGETLNs8 (v8i8 (EXTRACT_SUBREG QPR:$src,
2818 (DSubReg_i8_reg imm:$lane))),
2819 (SubReg_i8_lane imm:$lane))>;
2820 def : Pat<(NEONvgetlanes (v8i16 QPR:$src), imm:$lane),
2821 (VGETLNs16 (v4i16 (EXTRACT_SUBREG QPR:$src,
2822 (DSubReg_i16_reg imm:$lane))),
2823 (SubReg_i16_lane imm:$lane))>;
2824 def : Pat<(NEONvgetlaneu (v16i8 QPR:$src), imm:$lane),
2825 (VGETLNu8 (v8i8 (EXTRACT_SUBREG QPR:$src,
2826 (DSubReg_i8_reg imm:$lane))),
2827 (SubReg_i8_lane imm:$lane))>;
2828 def : Pat<(NEONvgetlaneu (v8i16 QPR:$src), imm:$lane),
2829 (VGETLNu16 (v4i16 (EXTRACT_SUBREG QPR:$src,
2830 (DSubReg_i16_reg imm:$lane))),
2831 (SubReg_i16_lane imm:$lane))>;
2832 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
2833 (VGETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src,
2834 (DSubReg_i32_reg imm:$lane))),
2835 (SubReg_i32_lane imm:$lane))>;
2836 def : Pat<(extractelt (v2f32 DPR:$src1), imm:$src2),
2837 (EXTRACT_SUBREG (v2f32 (COPY_TO_REGCLASS (v2f32 DPR:$src1),DPR_VFP2)),
2838 (SSubReg_f32_reg imm:$src2))>;
2839 def : Pat<(extractelt (v4f32 QPR:$src1), imm:$src2),
2840 (EXTRACT_SUBREG (v4f32 (COPY_TO_REGCLASS (v4f32 QPR:$src1),QPR_VFP2)),
2841 (SSubReg_f32_reg imm:$src2))>;
2842 //def : Pat<(extractelt (v2i64 QPR:$src1), imm:$src2),
2843 // (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
2844 def : Pat<(extractelt (v2f64 QPR:$src1), imm:$src2),
2845 (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
2848 // VMOV : Vector Set Lane (move ARM core register to scalar)
2850 let Constraints = "$src1 = $dst" in {
2851 def VSETLNi8 : NVSetLane<{1,1,1,0,0,1,?,0}, 0b1011, {?,?}, (outs DPR:$dst),
2852 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
2853 IIC_VMOVISL, "vmov", "8", "$dst[$lane], $src2",
2854 [(set DPR:$dst, (vector_insert (v8i8 DPR:$src1),
2855 GPR:$src2, imm:$lane))]>;
2856 def VSETLNi16 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, {?,1}, (outs DPR:$dst),
2857 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
2858 IIC_VMOVISL, "vmov", "16", "$dst[$lane], $src2",
2859 [(set DPR:$dst, (vector_insert (v4i16 DPR:$src1),
2860 GPR:$src2, imm:$lane))]>;
2861 def VSETLNi32 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, 0b00, (outs DPR:$dst),
2862 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
2863 IIC_VMOVISL, "vmov", "32", "$dst[$lane], $src2",
2864 [(set DPR:$dst, (insertelt (v2i32 DPR:$src1),
2865 GPR:$src2, imm:$lane))]>;
2867 def : Pat<(vector_insert (v16i8 QPR:$src1), GPR:$src2, imm:$lane),
2868 (v16i8 (INSERT_SUBREG QPR:$src1,
2869 (v8i8 (VSETLNi8 (v8i8 (EXTRACT_SUBREG QPR:$src1,
2870 (DSubReg_i8_reg imm:$lane))),
2871 GPR:$src2, (SubReg_i8_lane imm:$lane))),
2872 (DSubReg_i8_reg imm:$lane)))>;
2873 def : Pat<(vector_insert (v8i16 QPR:$src1), GPR:$src2, imm:$lane),
2874 (v8i16 (INSERT_SUBREG QPR:$src1,
2875 (v4i16 (VSETLNi16 (v4i16 (EXTRACT_SUBREG QPR:$src1,
2876 (DSubReg_i16_reg imm:$lane))),
2877 GPR:$src2, (SubReg_i16_lane imm:$lane))),
2878 (DSubReg_i16_reg imm:$lane)))>;
2879 def : Pat<(insertelt (v4i32 QPR:$src1), GPR:$src2, imm:$lane),
2880 (v4i32 (INSERT_SUBREG QPR:$src1,
2881 (v2i32 (VSETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src1,
2882 (DSubReg_i32_reg imm:$lane))),
2883 GPR:$src2, (SubReg_i32_lane imm:$lane))),
2884 (DSubReg_i32_reg imm:$lane)))>;
2886 def : Pat<(v2f32 (insertelt DPR:$src1, SPR:$src2, imm:$src3)),
2887 (INSERT_SUBREG (v2f32 (COPY_TO_REGCLASS DPR:$src1, DPR_VFP2)),
2888 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
2889 def : Pat<(v4f32 (insertelt QPR:$src1, SPR:$src2, imm:$src3)),
2890 (INSERT_SUBREG (v4f32 (COPY_TO_REGCLASS QPR:$src1, QPR_VFP2)),
2891 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
2893 //def : Pat<(v2i64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
2894 // (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
2895 def : Pat<(v2f64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
2896 (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
2898 def : Pat<(v2f32 (scalar_to_vector SPR:$src)),
2899 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$src, arm_ssubreg_0)>;
2900 def : Pat<(v2f64 (scalar_to_vector (f64 DPR:$src))),
2901 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), DPR:$src, arm_dsubreg_0)>;
2902 def : Pat<(v4f32 (scalar_to_vector SPR:$src)),
2903 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), SPR:$src, arm_ssubreg_0)>;
2905 def : Pat<(v8i8 (scalar_to_vector GPR:$src)),
2906 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
2907 def : Pat<(v4i16 (scalar_to_vector GPR:$src)),
2908 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
2909 def : Pat<(v2i32 (scalar_to_vector GPR:$src)),
2910 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
2912 def : Pat<(v16i8 (scalar_to_vector GPR:$src)),
2913 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
2914 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
2916 def : Pat<(v8i16 (scalar_to_vector GPR:$src)),
2917 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
2918 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
2920 def : Pat<(v4i32 (scalar_to_vector GPR:$src)),
2921 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
2922 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
2925 // VDUP : Vector Duplicate (from ARM core register to all elements)
2927 class VDUPD<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
2928 : NVDup<opcod1, 0b1011, opcod3, (outs DPR:$dst), (ins GPR:$src),
2929 IIC_VMOVIS, "vdup", Dt, "$dst, $src",
2930 [(set DPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
2931 class VDUPQ<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
2932 : NVDup<opcod1, 0b1011, opcod3, (outs QPR:$dst), (ins GPR:$src),
2933 IIC_VMOVIS, "vdup", Dt, "$dst, $src",
2934 [(set QPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
2936 def VDUP8d : VDUPD<0b11101100, 0b00, "8", v8i8>;
2937 def VDUP16d : VDUPD<0b11101000, 0b01, "16", v4i16>;
2938 def VDUP32d : VDUPD<0b11101000, 0b00, "32", v2i32>;
2939 def VDUP8q : VDUPQ<0b11101110, 0b00, "8", v16i8>;
2940 def VDUP16q : VDUPQ<0b11101010, 0b01, "16", v8i16>;
2941 def VDUP32q : VDUPQ<0b11101010, 0b00, "32", v4i32>;
2943 def VDUPfd : NVDup<0b11101000, 0b1011, 0b00, (outs DPR:$dst), (ins GPR:$src),
2944 IIC_VMOVIS, "vdup", "32", "$dst, $src",
2945 [(set DPR:$dst, (v2f32 (NEONvdup
2946 (f32 (bitconvert GPR:$src)))))]>;
2947 def VDUPfq : NVDup<0b11101010, 0b1011, 0b00, (outs QPR:$dst), (ins GPR:$src),
2948 IIC_VMOVIS, "vdup", "32", "$dst, $src",
2949 [(set QPR:$dst, (v4f32 (NEONvdup
2950 (f32 (bitconvert GPR:$src)))))]>;
2952 // VDUP : Vector Duplicate Lane (from scalar to all elements)
2954 class VDUPLND<bits<2> op19_18, bits<2> op17_16,
2955 string OpcodeStr, string Dt, ValueType Ty>
2956 : N2V<0b11, 0b11, op19_18, op17_16, 0b11000, 0, 0,
2957 (outs DPR:$dst), (ins DPR:$src, nohash_imm:$lane), IIC_VMOVD,
2958 OpcodeStr, Dt, "$dst, $src[$lane]", "",
2959 [(set DPR:$dst, (Ty (NEONvduplane (Ty DPR:$src), imm:$lane)))]>;
2961 class VDUPLNQ<bits<2> op19_18, bits<2> op17_16, string OpcodeStr, string Dt,
2962 ValueType ResTy, ValueType OpTy>
2963 : N2V<0b11, 0b11, op19_18, op17_16, 0b11000, 1, 0,
2964 (outs QPR:$dst), (ins DPR:$src, nohash_imm:$lane), IIC_VMOVD,
2965 OpcodeStr, Dt, "$dst, $src[$lane]", "",
2966 [(set QPR:$dst, (ResTy (NEONvduplane (OpTy DPR:$src), imm:$lane)))]>;
2968 // Inst{19-16} is partially specified depending on the element size.
2970 def VDUPLN8d : VDUPLND<{?,?}, {?,1}, "vdup", "8", v8i8>;
2971 def VDUPLN16d : VDUPLND<{?,?}, {1,0}, "vdup", "16", v4i16>;
2972 def VDUPLN32d : VDUPLND<{?,1}, {0,0}, "vdup", "32", v2i32>;
2973 def VDUPLNfd : VDUPLND<{?,1}, {0,0}, "vdup", "32", v2f32>;
2974 def VDUPLN8q : VDUPLNQ<{?,?}, {?,1}, "vdup", "8", v16i8, v8i8>;
2975 def VDUPLN16q : VDUPLNQ<{?,?}, {1,0}, "vdup", "16", v8i16, v4i16>;
2976 def VDUPLN32q : VDUPLNQ<{?,1}, {0,0}, "vdup", "32", v4i32, v2i32>;
2977 def VDUPLNfq : VDUPLNQ<{?,1}, {0,0}, "vdup", "32", v4f32, v2f32>;
2979 def : Pat<(v16i8 (NEONvduplane (v16i8 QPR:$src), imm:$lane)),
2980 (v16i8 (VDUPLN8q (v8i8 (EXTRACT_SUBREG QPR:$src,
2981 (DSubReg_i8_reg imm:$lane))),
2982 (SubReg_i8_lane imm:$lane)))>;
2983 def : Pat<(v8i16 (NEONvduplane (v8i16 QPR:$src), imm:$lane)),
2984 (v8i16 (VDUPLN16q (v4i16 (EXTRACT_SUBREG QPR:$src,
2985 (DSubReg_i16_reg imm:$lane))),
2986 (SubReg_i16_lane imm:$lane)))>;
2987 def : Pat<(v4i32 (NEONvduplane (v4i32 QPR:$src), imm:$lane)),
2988 (v4i32 (VDUPLN32q (v2i32 (EXTRACT_SUBREG QPR:$src,
2989 (DSubReg_i32_reg imm:$lane))),
2990 (SubReg_i32_lane imm:$lane)))>;
2991 def : Pat<(v4f32 (NEONvduplane (v4f32 QPR:$src), imm:$lane)),
2992 (v4f32 (VDUPLNfq (v2f32 (EXTRACT_SUBREG QPR:$src,
2993 (DSubReg_i32_reg imm:$lane))),
2994 (SubReg_i32_lane imm:$lane)))>;
2996 def VDUPfdf : N2V<0b11, 0b11, {?,1}, {0,0}, 0b11000, 0, 0,
2997 (outs DPR:$dst), (ins SPR:$src),
2998 IIC_VMOVD, "vdup", "32", "$dst, ${src:lane}", "",
2999 [(set DPR:$dst, (v2f32 (NEONvdup (f32 SPR:$src))))]>;
3001 def VDUPfqf : N2V<0b11, 0b11, {?,1}, {0,0}, 0b11000, 1, 0,
3002 (outs QPR:$dst), (ins SPR:$src),
3003 IIC_VMOVD, "vdup", "32", "$dst, ${src:lane}", "",
3004 [(set QPR:$dst, (v4f32 (NEONvdup (f32 SPR:$src))))]>;
3006 def : Pat<(v2i64 (NEONvduplane (v2i64 QPR:$src), imm:$lane)),
3007 (INSERT_SUBREG QPR:$src,
3008 (i64 (EXTRACT_SUBREG QPR:$src,
3009 (DSubReg_f64_reg imm:$lane))),
3010 (DSubReg_f64_other_reg imm:$lane))>;
3011 def : Pat<(v2f64 (NEONvduplane (v2f64 QPR:$src), imm:$lane)),
3012 (INSERT_SUBREG QPR:$src,
3013 (f64 (EXTRACT_SUBREG QPR:$src,
3014 (DSubReg_f64_reg imm:$lane))),
3015 (DSubReg_f64_other_reg imm:$lane))>;
3017 // VMOVN : Vector Narrowing Move
3018 defm VMOVN : N2VNInt_HSD<0b11,0b11,0b10,0b00100,0,0, IIC_VMOVD,
3019 "vmovn", "i", int_arm_neon_vmovn>;
3020 // VQMOVN : Vector Saturating Narrowing Move
3021 defm VQMOVNs : N2VNInt_HSD<0b11,0b11,0b10,0b00101,0,0, IIC_VQUNAiD,
3022 "vqmovn", "s", int_arm_neon_vqmovns>;
3023 defm VQMOVNu : N2VNInt_HSD<0b11,0b11,0b10,0b00101,1,0, IIC_VQUNAiD,
3024 "vqmovn", "u", int_arm_neon_vqmovnu>;
3025 defm VQMOVNsu : N2VNInt_HSD<0b11,0b11,0b10,0b00100,1,0, IIC_VQUNAiD,
3026 "vqmovun", "s", int_arm_neon_vqmovnsu>;
3027 // VMOVL : Vector Lengthening Move
3028 defm VMOVLs : N2VLInt_QHS<0b01,0b10100,0,1, "vmovl", "s",
3029 int_arm_neon_vmovls>;
3030 defm VMOVLu : N2VLInt_QHS<0b11,0b10100,0,1, "vmovl", "u",
3031 int_arm_neon_vmovlu>;
3033 // Vector Conversions.
3035 // VCVT : Vector Convert Between Floating-Point and Integers
3036 def VCVTf2sd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
3037 v2i32, v2f32, fp_to_sint>;
3038 def VCVTf2ud : N2VD<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
3039 v2i32, v2f32, fp_to_uint>;
3040 def VCVTs2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
3041 v2f32, v2i32, sint_to_fp>;
3042 def VCVTu2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
3043 v2f32, v2i32, uint_to_fp>;
3045 def VCVTf2sq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
3046 v4i32, v4f32, fp_to_sint>;
3047 def VCVTf2uq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
3048 v4i32, v4f32, fp_to_uint>;
3049 def VCVTs2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
3050 v4f32, v4i32, sint_to_fp>;
3051 def VCVTu2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
3052 v4f32, v4i32, uint_to_fp>;
3054 // VCVT : Vector Convert Between Floating-Point and Fixed-Point.
3055 def VCVTf2xsd : N2VCvtD<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
3056 v2i32, v2f32, int_arm_neon_vcvtfp2fxs>;
3057 def VCVTf2xud : N2VCvtD<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
3058 v2i32, v2f32, int_arm_neon_vcvtfp2fxu>;
3059 def VCVTxs2fd : N2VCvtD<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
3060 v2f32, v2i32, int_arm_neon_vcvtfxs2fp>;
3061 def VCVTxu2fd : N2VCvtD<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
3062 v2f32, v2i32, int_arm_neon_vcvtfxu2fp>;
3064 def VCVTf2xsq : N2VCvtQ<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
3065 v4i32, v4f32, int_arm_neon_vcvtfp2fxs>;
3066 def VCVTf2xuq : N2VCvtQ<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
3067 v4i32, v4f32, int_arm_neon_vcvtfp2fxu>;
3068 def VCVTxs2fq : N2VCvtQ<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
3069 v4f32, v4i32, int_arm_neon_vcvtfxs2fp>;
3070 def VCVTxu2fq : N2VCvtQ<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
3071 v4f32, v4i32, int_arm_neon_vcvtfxu2fp>;
3075 // VREV64 : Vector Reverse elements within 64-bit doublewords
3077 class VREV64D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3078 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 0, 0, (outs DPR:$dst),
3079 (ins DPR:$src), IIC_VMOVD,
3080 OpcodeStr, Dt, "$dst, $src", "",
3081 [(set DPR:$dst, (Ty (NEONvrev64 (Ty DPR:$src))))]>;
3082 class VREV64Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3083 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 1, 0, (outs QPR:$dst),
3084 (ins QPR:$src), IIC_VMOVD,
3085 OpcodeStr, Dt, "$dst, $src", "",
3086 [(set QPR:$dst, (Ty (NEONvrev64 (Ty QPR:$src))))]>;
3088 def VREV64d8 : VREV64D<0b00, "vrev64", "8", v8i8>;
3089 def VREV64d16 : VREV64D<0b01, "vrev64", "16", v4i16>;
3090 def VREV64d32 : VREV64D<0b10, "vrev64", "32", v2i32>;
3091 def VREV64df : VREV64D<0b10, "vrev64", "32", v2f32>;
3093 def VREV64q8 : VREV64Q<0b00, "vrev64", "8", v16i8>;
3094 def VREV64q16 : VREV64Q<0b01, "vrev64", "16", v8i16>;
3095 def VREV64q32 : VREV64Q<0b10, "vrev64", "32", v4i32>;
3096 def VREV64qf : VREV64Q<0b10, "vrev64", "32", v4f32>;
3098 // VREV32 : Vector Reverse elements within 32-bit words
3100 class VREV32D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3101 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 0, 0, (outs DPR:$dst),
3102 (ins DPR:$src), IIC_VMOVD,
3103 OpcodeStr, Dt, "$dst, $src", "",
3104 [(set DPR:$dst, (Ty (NEONvrev32 (Ty DPR:$src))))]>;
3105 class VREV32Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3106 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 1, 0, (outs QPR:$dst),
3107 (ins QPR:$src), IIC_VMOVD,
3108 OpcodeStr, Dt, "$dst, $src", "",
3109 [(set QPR:$dst, (Ty (NEONvrev32 (Ty QPR:$src))))]>;
3111 def VREV32d8 : VREV32D<0b00, "vrev32", "8", v8i8>;
3112 def VREV32d16 : VREV32D<0b01, "vrev32", "16", v4i16>;
3114 def VREV32q8 : VREV32Q<0b00, "vrev32", "8", v16i8>;
3115 def VREV32q16 : VREV32Q<0b01, "vrev32", "16", v8i16>;
3117 // VREV16 : Vector Reverse elements within 16-bit halfwords
3119 class VREV16D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3120 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 0, 0, (outs DPR:$dst),
3121 (ins DPR:$src), IIC_VMOVD,
3122 OpcodeStr, Dt, "$dst, $src", "",
3123 [(set DPR:$dst, (Ty (NEONvrev16 (Ty DPR:$src))))]>;
3124 class VREV16Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3125 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 1, 0, (outs QPR:$dst),
3126 (ins QPR:$src), IIC_VMOVD,
3127 OpcodeStr, Dt, "$dst, $src", "",
3128 [(set QPR:$dst, (Ty (NEONvrev16 (Ty QPR:$src))))]>;
3130 def VREV16d8 : VREV16D<0b00, "vrev16", "8", v8i8>;
3131 def VREV16q8 : VREV16Q<0b00, "vrev16", "8", v16i8>;
3133 // Other Vector Shuffles.
3135 // VEXT : Vector Extract
3137 class VEXTd<string OpcodeStr, string Dt, ValueType Ty>
3138 : N3V<0,1,0b11,{?,?,?,?},0,0, (outs DPR:$dst),
3139 (ins DPR:$lhs, DPR:$rhs, i32imm:$index), IIC_VEXTD,
3140 OpcodeStr, Dt, "$dst, $lhs, $rhs, $index", "",
3141 [(set DPR:$dst, (Ty (NEONvext (Ty DPR:$lhs),
3142 (Ty DPR:$rhs), imm:$index)))]>;
3144 class VEXTq<string OpcodeStr, string Dt, ValueType Ty>
3145 : N3V<0,1,0b11,{?,?,?,?},1,0, (outs QPR:$dst),
3146 (ins QPR:$lhs, QPR:$rhs, i32imm:$index), IIC_VEXTQ,
3147 OpcodeStr, Dt, "$dst, $lhs, $rhs, $index", "",
3148 [(set QPR:$dst, (Ty (NEONvext (Ty QPR:$lhs),
3149 (Ty QPR:$rhs), imm:$index)))]>;
3151 def VEXTd8 : VEXTd<"vext", "8", v8i8>;
3152 def VEXTd16 : VEXTd<"vext", "16", v4i16>;
3153 def VEXTd32 : VEXTd<"vext", "32", v2i32>;
3154 def VEXTdf : VEXTd<"vext", "32", v2f32>;
3156 def VEXTq8 : VEXTq<"vext", "8", v16i8>;
3157 def VEXTq16 : VEXTq<"vext", "16", v8i16>;
3158 def VEXTq32 : VEXTq<"vext", "32", v4i32>;
3159 def VEXTqf : VEXTq<"vext", "32", v4f32>;
3161 // VTRN : Vector Transpose
3163 def VTRNd8 : N2VDShuffle<0b00, 0b00001, "vtrn", "8">;
3164 def VTRNd16 : N2VDShuffle<0b01, 0b00001, "vtrn", "16">;
3165 def VTRNd32 : N2VDShuffle<0b10, 0b00001, "vtrn", "32">;
3167 def VTRNq8 : N2VQShuffle<0b00, 0b00001, IIC_VPERMQ, "vtrn", "8">;
3168 def VTRNq16 : N2VQShuffle<0b01, 0b00001, IIC_VPERMQ, "vtrn", "16">;
3169 def VTRNq32 : N2VQShuffle<0b10, 0b00001, IIC_VPERMQ, "vtrn", "32">;
3171 // VUZP : Vector Unzip (Deinterleave)
3173 def VUZPd8 : N2VDShuffle<0b00, 0b00010, "vuzp", "8">;
3174 def VUZPd16 : N2VDShuffle<0b01, 0b00010, "vuzp", "16">;
3175 def VUZPd32 : N2VDShuffle<0b10, 0b00010, "vuzp", "32">;
3177 def VUZPq8 : N2VQShuffle<0b00, 0b00010, IIC_VPERMQ3, "vuzp", "8">;
3178 def VUZPq16 : N2VQShuffle<0b01, 0b00010, IIC_VPERMQ3, "vuzp", "16">;
3179 def VUZPq32 : N2VQShuffle<0b10, 0b00010, IIC_VPERMQ3, "vuzp", "32">;
3181 // VZIP : Vector Zip (Interleave)
3183 def VZIPd8 : N2VDShuffle<0b00, 0b00011, "vzip", "8">;
3184 def VZIPd16 : N2VDShuffle<0b01, 0b00011, "vzip", "16">;
3185 def VZIPd32 : N2VDShuffle<0b10, 0b00011, "vzip", "32">;
3187 def VZIPq8 : N2VQShuffle<0b00, 0b00011, IIC_VPERMQ3, "vzip", "8">;
3188 def VZIPq16 : N2VQShuffle<0b01, 0b00011, IIC_VPERMQ3, "vzip", "16">;
3189 def VZIPq32 : N2VQShuffle<0b10, 0b00011, IIC_VPERMQ3, "vzip", "32">;
3191 // Vector Table Lookup and Table Extension.
3193 // VTBL : Vector Table Lookup
3195 : N3V<1,1,0b11,0b1000,0,0, (outs DPR:$dst),
3196 (ins DPR:$tbl1, DPR:$src), IIC_VTB1,
3197 "vtbl", "8", "$dst, \\{$tbl1\\}, $src", "",
3198 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl1 DPR:$tbl1, DPR:$src)))]>;
3199 let hasExtraSrcRegAllocReq = 1 in {
3201 : N3V<1,1,0b11,0b1001,0,0, (outs DPR:$dst),
3202 (ins DPR:$tbl1, DPR:$tbl2, DPR:$src), IIC_VTB2,
3203 "vtbl", "8", "$dst, \\{$tbl1, $tbl2\\}, $src", "",
3204 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl2
3205 DPR:$tbl1, DPR:$tbl2, DPR:$src)))]>;
3207 : N3V<1,1,0b11,0b1010,0,0, (outs DPR:$dst),
3208 (ins DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$src), IIC_VTB3,
3209 "vtbl", "8", "$dst, \\{$tbl1, $tbl2, $tbl3\\}, $src", "",
3210 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl3
3211 DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$src)))]>;
3213 : N3V<1,1,0b11,0b1011,0,0, (outs DPR:$dst),
3214 (ins DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$src), IIC_VTB4,
3215 "vtbl", "8", "$dst, \\{$tbl1, $tbl2, $tbl3, $tbl4\\}, $src", "",
3216 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl4 DPR:$tbl1, DPR:$tbl2,
3217 DPR:$tbl3, DPR:$tbl4, DPR:$src)))]>;
3218 } // hasExtraSrcRegAllocReq = 1
3220 // VTBX : Vector Table Extension
3222 : N3V<1,1,0b11,0b1000,1,0, (outs DPR:$dst),
3223 (ins DPR:$orig, DPR:$tbl1, DPR:$src), IIC_VTBX1,
3224 "vtbx", "8", "$dst, \\{$tbl1\\}, $src", "$orig = $dst",
3225 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx1
3226 DPR:$orig, DPR:$tbl1, DPR:$src)))]>;
3227 let hasExtraSrcRegAllocReq = 1 in {
3229 : N3V<1,1,0b11,0b1001,1,0, (outs DPR:$dst),
3230 (ins DPR:$orig, DPR:$tbl1, DPR:$tbl2, DPR:$src), IIC_VTBX2,
3231 "vtbx", "8", "$dst, \\{$tbl1, $tbl2\\}, $src", "$orig = $dst",
3232 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx2
3233 DPR:$orig, DPR:$tbl1, DPR:$tbl2, DPR:$src)))]>;
3235 : N3V<1,1,0b11,0b1010,1,0, (outs DPR:$dst),
3236 (ins DPR:$orig, DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$src), IIC_VTBX3,
3237 "vtbx", "8", "$dst, \\{$tbl1, $tbl2, $tbl3\\}, $src", "$orig = $dst",
3238 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx3 DPR:$orig, DPR:$tbl1,
3239 DPR:$tbl2, DPR:$tbl3, DPR:$src)))]>;
3241 : N3V<1,1,0b11,0b1011,1,0, (outs DPR:$dst), (ins DPR:$orig, DPR:$tbl1,
3242 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$src), IIC_VTBX4,
3243 "vtbx", "8", "$dst, \\{$tbl1, $tbl2, $tbl3, $tbl4\\}, $src",
3245 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx4 DPR:$orig, DPR:$tbl1,
3246 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$src)))]>;
3247 } // hasExtraSrcRegAllocReq = 1
3249 //===----------------------------------------------------------------------===//
3250 // NEON instructions for single-precision FP math
3251 //===----------------------------------------------------------------------===//
3253 class N2VSPat<SDNode OpNode, ValueType ResTy, ValueType OpTy, NeonI Inst>
3254 : NEONFPPat<(ResTy (OpNode SPR:$a)),
3255 (EXTRACT_SUBREG (OpTy (Inst (INSERT_SUBREG (OpTy (IMPLICIT_DEF)),
3256 SPR:$a, arm_ssubreg_0))),
3259 class N3VSPat<SDNode OpNode, NeonI Inst>
3260 : NEONFPPat<(f32 (OpNode SPR:$a, SPR:$b)),
3261 (EXTRACT_SUBREG (v2f32
3262 (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3263 SPR:$a, arm_ssubreg_0),
3264 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3265 SPR:$b, arm_ssubreg_0))),
3268 class N3VSMulOpPat<SDNode MulNode, SDNode OpNode, NeonI Inst>
3269 : NEONFPPat<(f32 (OpNode SPR:$acc, (f32 (MulNode SPR:$a, SPR:$b)))),
3270 (EXTRACT_SUBREG (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3271 SPR:$acc, arm_ssubreg_0),
3272 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3273 SPR:$a, arm_ssubreg_0),
3274 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3275 SPR:$b, arm_ssubreg_0)),
3278 // These need separate instructions because they must use DPR_VFP2 register
3279 // class which have SPR sub-registers.
3281 // Vector Add Operations used for single-precision FP
3282 let neverHasSideEffects = 1 in
3283 def VADDfd_sfp : N3VS<0,0,0b00,0b1101,0, "vadd", "f32", v2f32, v2f32, fadd, 1>;
3284 def : N3VSPat<fadd, VADDfd_sfp>;
3286 // Vector Sub Operations used for single-precision FP
3287 let neverHasSideEffects = 1 in
3288 def VSUBfd_sfp : N3VS<0,0,0b10,0b1101,0, "vsub", "f32", v2f32, v2f32, fsub, 0>;
3289 def : N3VSPat<fsub, VSUBfd_sfp>;
3291 // Vector Multiply Operations used for single-precision FP
3292 let neverHasSideEffects = 1 in
3293 def VMULfd_sfp : N3VS<1,0,0b00,0b1101,1, "vmul", "f32", v2f32, v2f32, fmul, 1>;
3294 def : N3VSPat<fmul, VMULfd_sfp>;
3296 // Vector Multiply-Accumulate/Subtract used for single-precision FP
3297 // vml[as].f32 can cause 4-8 cycle stalls in following ASIMD instructions, so
3298 // we want to avoid them for now. e.g., alternating vmla/vadd instructions.
3300 //let neverHasSideEffects = 1 in
3301 //def VMLAfd_sfp : N3VSMulOp<0,0,0b00,0b1101,1, IIC_VMACD, "vmla", "f32",
3302 // v2f32, fmul, fadd>;
3303 //def : N3VSMulOpPat<fmul, fadd, VMLAfd_sfp>;
3305 //let neverHasSideEffects = 1 in
3306 //def VMLSfd_sfp : N3VSMulOp<0,0,0b10,0b1101,1, IIC_VMACD, "vmls", "f32",
3307 // v2f32, fmul, fsub>;
3308 //def : N3VSMulOpPat<fmul, fsub, VMLSfd_sfp>;
3310 // Vector Absolute used for single-precision FP
3311 let neverHasSideEffects = 1 in
3312 def VABSfd_sfp : N2V<0b11, 0b11, 0b10, 0b01, 0b01110, 0, 0,
3313 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), IIC_VUNAD,
3314 "vabs", "f32", "$dst, $src", "", []>;
3315 def : N2VSPat<fabs, f32, v2f32, VABSfd_sfp>;
3317 // Vector Negate used for single-precision FP
3318 let neverHasSideEffects = 1 in
3319 def VNEGfd_sfp : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
3320 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), IIC_VUNAD,
3321 "vneg", "f32", "$dst, $src", "", []>;
3322 def : N2VSPat<fneg, f32, v2f32, VNEGfd_sfp>;
3324 // Vector Maximum used for single-precision FP
3325 let neverHasSideEffects = 1 in
3326 def VMAXfd_sfp : N3V<0, 0, 0b00, 0b1111, 0, 0, (outs DPR_VFP2:$dst),
3327 (ins DPR_VFP2:$src1, DPR_VFP2:$src2), IIC_VBIND,
3328 "vmax", "f32", "$dst, $src1, $src2", "", []>;
3329 def : N3VSPat<NEONfmax, VMAXfd_sfp>;
3331 // Vector Minimum used for single-precision FP
3332 let neverHasSideEffects = 1 in
3333 def VMINfd_sfp : N3V<0, 0, 0b00, 0b1111, 0, 0, (outs DPR_VFP2:$dst),
3334 (ins DPR_VFP2:$src1, DPR_VFP2:$src2), IIC_VBIND,
3335 "vmin", "f32", "$dst, $src1, $src2", "", []>;
3336 def : N3VSPat<NEONfmin, VMINfd_sfp>;
3338 // Vector Convert between single-precision FP and integer
3339 let neverHasSideEffects = 1 in
3340 def VCVTf2sd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
3341 v2i32, v2f32, fp_to_sint>;
3342 def : N2VSPat<arm_ftosi, f32, v2f32, VCVTf2sd_sfp>;
3344 let neverHasSideEffects = 1 in
3345 def VCVTf2ud_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
3346 v2i32, v2f32, fp_to_uint>;
3347 def : N2VSPat<arm_ftoui, f32, v2f32, VCVTf2ud_sfp>;
3349 let neverHasSideEffects = 1 in
3350 def VCVTs2fd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
3351 v2f32, v2i32, sint_to_fp>;
3352 def : N2VSPat<arm_sitof, f32, v2i32, VCVTs2fd_sfp>;
3354 let neverHasSideEffects = 1 in
3355 def VCVTu2fd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
3356 v2f32, v2i32, uint_to_fp>;
3357 def : N2VSPat<arm_uitof, f32, v2i32, VCVTu2fd_sfp>;
3359 //===----------------------------------------------------------------------===//
3360 // Non-Instruction Patterns
3361 //===----------------------------------------------------------------------===//
3364 def : Pat<(v1i64 (bitconvert (v2i32 DPR:$src))), (v1i64 DPR:$src)>;
3365 def : Pat<(v1i64 (bitconvert (v4i16 DPR:$src))), (v1i64 DPR:$src)>;
3366 def : Pat<(v1i64 (bitconvert (v8i8 DPR:$src))), (v1i64 DPR:$src)>;
3367 def : Pat<(v1i64 (bitconvert (f64 DPR:$src))), (v1i64 DPR:$src)>;
3368 def : Pat<(v1i64 (bitconvert (v2f32 DPR:$src))), (v1i64 DPR:$src)>;
3369 def : Pat<(v2i32 (bitconvert (v1i64 DPR:$src))), (v2i32 DPR:$src)>;
3370 def : Pat<(v2i32 (bitconvert (v4i16 DPR:$src))), (v2i32 DPR:$src)>;
3371 def : Pat<(v2i32 (bitconvert (v8i8 DPR:$src))), (v2i32 DPR:$src)>;
3372 def : Pat<(v2i32 (bitconvert (f64 DPR:$src))), (v2i32 DPR:$src)>;
3373 def : Pat<(v2i32 (bitconvert (v2f32 DPR:$src))), (v2i32 DPR:$src)>;
3374 def : Pat<(v4i16 (bitconvert (v1i64 DPR:$src))), (v4i16 DPR:$src)>;
3375 def : Pat<(v4i16 (bitconvert (v2i32 DPR:$src))), (v4i16 DPR:$src)>;
3376 def : Pat<(v4i16 (bitconvert (v8i8 DPR:$src))), (v4i16 DPR:$src)>;
3377 def : Pat<(v4i16 (bitconvert (f64 DPR:$src))), (v4i16 DPR:$src)>;
3378 def : Pat<(v4i16 (bitconvert (v2f32 DPR:$src))), (v4i16 DPR:$src)>;
3379 def : Pat<(v8i8 (bitconvert (v1i64 DPR:$src))), (v8i8 DPR:$src)>;
3380 def : Pat<(v8i8 (bitconvert (v2i32 DPR:$src))), (v8i8 DPR:$src)>;
3381 def : Pat<(v8i8 (bitconvert (v4i16 DPR:$src))), (v8i8 DPR:$src)>;
3382 def : Pat<(v8i8 (bitconvert (f64 DPR:$src))), (v8i8 DPR:$src)>;
3383 def : Pat<(v8i8 (bitconvert (v2f32 DPR:$src))), (v8i8 DPR:$src)>;
3384 def : Pat<(f64 (bitconvert (v1i64 DPR:$src))), (f64 DPR:$src)>;
3385 def : Pat<(f64 (bitconvert (v2i32 DPR:$src))), (f64 DPR:$src)>;
3386 def : Pat<(f64 (bitconvert (v4i16 DPR:$src))), (f64 DPR:$src)>;
3387 def : Pat<(f64 (bitconvert (v8i8 DPR:$src))), (f64 DPR:$src)>;
3388 def : Pat<(f64 (bitconvert (v2f32 DPR:$src))), (f64 DPR:$src)>;
3389 def : Pat<(v2f32 (bitconvert (f64 DPR:$src))), (v2f32 DPR:$src)>;
3390 def : Pat<(v2f32 (bitconvert (v1i64 DPR:$src))), (v2f32 DPR:$src)>;
3391 def : Pat<(v2f32 (bitconvert (v2i32 DPR:$src))), (v2f32 DPR:$src)>;
3392 def : Pat<(v2f32 (bitconvert (v4i16 DPR:$src))), (v2f32 DPR:$src)>;
3393 def : Pat<(v2f32 (bitconvert (v8i8 DPR:$src))), (v2f32 DPR:$src)>;
3395 def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (v2i64 QPR:$src)>;
3396 def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (v2i64 QPR:$src)>;
3397 def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (v2i64 QPR:$src)>;
3398 def : Pat<(v2i64 (bitconvert (v2f64 QPR:$src))), (v2i64 QPR:$src)>;
3399 def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (v2i64 QPR:$src)>;
3400 def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (v4i32 QPR:$src)>;
3401 def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (v4i32 QPR:$src)>;
3402 def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (v4i32 QPR:$src)>;
3403 def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (v4i32 QPR:$src)>;
3404 def : Pat<(v4i32 (bitconvert (v4f32 QPR:$src))), (v4i32 QPR:$src)>;
3405 def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (v8i16 QPR:$src)>;
3406 def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (v8i16 QPR:$src)>;
3407 def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (v8i16 QPR:$src)>;
3408 def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (v8i16 QPR:$src)>;
3409 def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (v8i16 QPR:$src)>;
3410 def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (v16i8 QPR:$src)>;
3411 def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (v16i8 QPR:$src)>;
3412 def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (v16i8 QPR:$src)>;
3413 def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (v16i8 QPR:$src)>;
3414 def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (v16i8 QPR:$src)>;
3415 def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (v4f32 QPR:$src)>;
3416 def : Pat<(v4f32 (bitconvert (v4i32 QPR:$src))), (v4f32 QPR:$src)>;
3417 def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (v4f32 QPR:$src)>;
3418 def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (v4f32 QPR:$src)>;
3419 def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (v4f32 QPR:$src)>;
3420 def : Pat<(v2f64 (bitconvert (v2i64 QPR:$src))), (v2f64 QPR:$src)>;
3421 def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (v2f64 QPR:$src)>;
3422 def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 QPR:$src)>;
3423 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 QPR:$src)>;
3424 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 QPR:$src)>;