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\\}, $addr", "", []>;
615 def VST2q8 : VST2Q<0b0000, "8">;
616 def VST2q16 : VST2Q<0b0100, "16">;
617 def VST2q32 : VST2Q<0b1000, "32">;
619 // ...with address register writeback:
620 class VST2DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
621 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
622 (ins addrmode6:$addr, DPR:$src1, DPR:$src2),
623 IIC_VST, "vst2", Dt, "\\{$src1, $src2\\}, $addr",
624 "$addr.addr = $wb", []>;
625 class VST2QWB<bits<4> op7_4, string Dt>
626 : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
627 (ins addrmode6:$addr,
628 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
629 IIC_VST, "vst2", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr",
630 "$addr.addr = $wb", []>;
632 def VST2d8_UPD : VST2DWB<0b1000, 0b0000, "8">;
633 def VST2d16_UPD : VST2DWB<0b1000, 0b0100, "16">;
634 def VST2d32_UPD : VST2DWB<0b1000, 0b1000, "32">;
635 def VST2d64_UPD : NLdSt<0,0b00,0b1010,0b1100, (outs GPR:$wb),
636 (ins addrmode6:$addr,
637 DPR:$src1, DPR:$src2), IIC_VST,
638 "vst1", "64", "\\{$src1, $src2\\}, $addr",
639 "$addr.addr = $wb", []>;
641 def VST2q8_UPD : VST2QWB<0b0000, "8">;
642 def VST2q16_UPD : VST2QWB<0b0100, "16">;
643 def VST2q32_UPD : VST2QWB<0b1000, "32">;
645 // ...with double-spaced registers (for disassembly only):
646 def VST2b8 : VST2D<0b1001, 0b0000, "8">;
647 def VST2b16 : VST2D<0b1001, 0b0100, "16">;
648 def VST2b32 : VST2D<0b1001, 0b1000, "32">;
649 def VST2b8_UPD : VST2DWB<0b1001, 0b0000, "8">;
650 def VST2b16_UPD : VST2DWB<0b1001, 0b0100, "16">;
651 def VST2b32_UPD : VST2DWB<0b1001, 0b1000, "32">;
653 // VST3 : Vector Store (multiple 3-element structures)
654 class VST3D<bits<4> op11_8, bits<4> op7_4, string Dt>
655 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
656 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3), IIC_VST,
657 "vst3", Dt, "\\{$src1, $src2, $src3\\}, $addr", "", []>;
659 def VST3d8 : VST3D<0b0100, 0b0000, "8">;
660 def VST3d16 : VST3D<0b0100, 0b0100, "16">;
661 def VST3d32 : VST3D<0b0100, 0b1000, "32">;
662 def VST3d64 : NLdSt<0,0b00,0b0110,0b1100, (outs),
663 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3),
665 "vst1", "64", "\\{$src1, $src2, $src3\\}, $addr", "", []>;
667 // ...with address register writeback:
668 class VST3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
669 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
670 (ins addrmode6:$addr,
671 DPR:$src1, DPR:$src2, DPR:$src3), IIC_VST,
672 "vst3", Dt, "\\{$src1, $src2, $src3\\}, $addr",
673 "$addr.addr = $wb", []>;
675 def VST3d8_UPD : VST3DWB<0b0100, 0b0000, "8">;
676 def VST3d16_UPD : VST3DWB<0b0100, 0b0100, "16">;
677 def VST3d32_UPD : VST3DWB<0b0100, 0b1000, "32">;
678 def VST3d64_UPD : NLdSt<0,0b00,0b0110,0b1100, (outs GPR:$wb),
679 (ins addrmode6:$addr,
680 DPR:$src1, DPR:$src2, DPR:$src3), IIC_VST,
681 "vst1", "64", "\\{$src1, $src2, $src3\\}, $addr",
682 "$addr.addr = $wb", []>;
684 // ...with double-spaced registers (non-updating versions for disassembly only):
685 def VST3q8 : VST3D<0b0101, 0b0000, "8">;
686 def VST3q16 : VST3D<0b0101, 0b0100, "16">;
687 def VST3q32 : VST3D<0b0101, 0b1000, "32">;
688 def VST3q8_UPD : VST3DWB<0b0101, 0b0000, "8">;
689 def VST3q16_UPD : VST3DWB<0b0101, 0b0100, "16">;
690 def VST3q32_UPD : VST3DWB<0b0101, 0b1000, "32">;
692 // ...alternate versions to be allocated odd register numbers:
693 def VST3q8odd_UPD : VST3DWB<0b0101, 0b0000, "8">;
694 def VST3q16odd_UPD : VST3DWB<0b0101, 0b0100, "16">;
695 def VST3q32odd_UPD : VST3DWB<0b0101, 0b1000, "32">;
697 // VST4 : Vector Store (multiple 4-element structures)
698 class VST4D<bits<4> op11_8, bits<4> op7_4, string Dt>
699 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
700 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
701 IIC_VST, "vst4", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr",
704 def VST4d8 : VST4D<0b0000, 0b0000, "8">;
705 def VST4d16 : VST4D<0b0000, 0b0100, "16">;
706 def VST4d32 : VST4D<0b0000, 0b1000, "32">;
707 def VST4d64 : NLdSt<0,0b00,0b0010,0b1100, (outs),
708 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3,
710 "vst1", "64", "\\{$src1, $src2, $src3, $src4\\}, $addr",
713 // ...with address register writeback:
714 class VST4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
715 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
716 (ins addrmode6:$addr,
717 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST,
718 "vst4", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr",
719 "$addr.addr = $wb", []>;
721 def VST4d8_UPD : VST4DWB<0b0000, 0b0000, "8">;
722 def VST4d16_UPD : VST4DWB<0b0000, 0b0100, "16">;
723 def VST4d32_UPD : VST4DWB<0b0000, 0b1000, "32">;
724 def VST4d64_UPD : NLdSt<0,0b00,0b0010,0b1100, (outs GPR:$wb),
725 (ins addrmode6:$addr,
726 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST,
728 "\\{$src1, $src2, $src3, $src4\\}, $addr",
729 "$addr.addr = $wb", []>;
731 // ...with double-spaced registers (non-updating versions for disassembly only):
732 def VST4q8 : VST4D<0b0001, 0b0000, "8">;
733 def VST4q16 : VST4D<0b0001, 0b0100, "16">;
734 def VST4q32 : VST4D<0b0001, 0b1000, "32">;
735 def VST4q8_UPD : VST4DWB<0b0001, 0b0000, "8">;
736 def VST4q16_UPD : VST4DWB<0b0001, 0b0100, "16">;
737 def VST4q32_UPD : VST4DWB<0b0001, 0b1000, "32">;
739 // ...alternate versions to be allocated odd register numbers:
740 def VST4q8odd_UPD : VST4DWB<0b0001, 0b0000, "8">;
741 def VST4q16odd_UPD : VST4DWB<0b0001, 0b0100, "16">;
742 def VST4q32odd_UPD : VST4DWB<0b0001, 0b1000, "32">;
744 // VST1LN : Vector Store (single element from one lane)
745 // FIXME: Not yet implemented.
747 // VST2LN : Vector Store (single 2-element structure from one lane)
748 class VST2LN<bits<4> op11_8, string Dt>
749 : NLdSt<1, 0b00, op11_8, {?,?,?,?}, (outs),
750 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, nohash_imm:$lane),
751 IIC_VST, "vst2", Dt, "\\{$src1[$lane], $src2[$lane]\\}, $addr",
754 def VST2LNd8 : VST2LN<0b0001, "8">;
755 def VST2LNd16 : VST2LN<0b0101, "16"> { let Inst{5} = 0; }
756 def VST2LNd32 : VST2LN<0b1001, "32"> { let Inst{6} = 0; }
758 // ...with double-spaced registers:
759 def VST2LNq16 : VST2LN<0b0101, "16"> { let Inst{5} = 1; }
760 def VST2LNq32 : VST2LN<0b1001, "32"> { let Inst{6} = 1; }
762 // ...alternate versions to be allocated odd register numbers:
763 def VST2LNq16odd : VST2LN<0b0101, "16"> { let Inst{5} = 1; }
764 def VST2LNq32odd : VST2LN<0b1001, "32"> { let Inst{6} = 1; }
766 // VST3LN : Vector Store (single 3-element structure from one lane)
767 class VST3LN<bits<4> op11_8, string Dt>
768 : NLdSt<1, 0b00, op11_8, {?,?,?,?}, (outs),
769 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3,
770 nohash_imm:$lane), IIC_VST, "vst3", Dt,
771 "\\{$src1[$lane], $src2[$lane], $src3[$lane]\\}, $addr", "", []>;
773 def VST3LNd8 : VST3LN<0b0010, "8"> { let Inst{4} = 0; }
774 def VST3LNd16 : VST3LN<0b0110, "16"> { let Inst{5-4} = 0b00; }
775 def VST3LNd32 : VST3LN<0b1010, "32"> { let Inst{6-4} = 0b000; }
777 // ...with double-spaced registers:
778 def VST3LNq16 : VST3LN<0b0110, "16"> { let Inst{5-4} = 0b10; }
779 def VST3LNq32 : VST3LN<0b1010, "32"> { let Inst{6-4} = 0b100; }
781 // ...alternate versions to be allocated odd register numbers:
782 def VST3LNq16odd : VST3LN<0b0110, "16"> { let Inst{5-4} = 0b10; }
783 def VST3LNq32odd : VST3LN<0b1010, "32"> { let Inst{6-4} = 0b100; }
785 // VST4LN : Vector Store (single 4-element structure from one lane)
786 class VST4LN<bits<4> op11_8, string Dt>
787 : NLdSt<1, 0b00, op11_8, {?,?,?,?}, (outs),
788 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
789 nohash_imm:$lane), IIC_VST, "vst4", Dt,
790 "\\{$src1[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $addr",
793 def VST4LNd8 : VST4LN<0b0011, "8">;
794 def VST4LNd16 : VST4LN<0b0111, "16"> { let Inst{5} = 0; }
795 def VST4LNd32 : VST4LN<0b1011, "32"> { let Inst{6} = 0; }
797 // ...with double-spaced registers:
798 def VST4LNq16 : VST4LN<0b0111, "16"> { let Inst{5} = 1; }
799 def VST4LNq32 : VST4LN<0b1011, "32"> { let Inst{6} = 1; }
801 // ...alternate versions to be allocated odd register numbers:
802 def VST4LNq16odd : VST4LN<0b0111, "16"> { let Inst{5} = 1; }
803 def VST4LNq32odd : VST4LN<0b1011, "32"> { let Inst{6} = 1; }
805 } // mayStore = 1, hasExtraSrcRegAllocReq = 1
808 //===----------------------------------------------------------------------===//
809 // NEON pattern fragments
810 //===----------------------------------------------------------------------===//
812 // Extract D sub-registers of Q registers.
813 // (arm_dsubreg_0 is 5; arm_dsubreg_1 is 6)
814 def DSubReg_i8_reg : SDNodeXForm<imm, [{
815 return CurDAG->getTargetConstant(5 + N->getZExtValue() / 8, MVT::i32);
817 def DSubReg_i16_reg : SDNodeXForm<imm, [{
818 return CurDAG->getTargetConstant(5 + N->getZExtValue() / 4, MVT::i32);
820 def DSubReg_i32_reg : SDNodeXForm<imm, [{
821 return CurDAG->getTargetConstant(5 + N->getZExtValue() / 2, MVT::i32);
823 def DSubReg_f64_reg : SDNodeXForm<imm, [{
824 return CurDAG->getTargetConstant(5 + N->getZExtValue(), MVT::i32);
826 def DSubReg_f64_other_reg : SDNodeXForm<imm, [{
827 return CurDAG->getTargetConstant(5 + (1 - N->getZExtValue()), MVT::i32);
830 // Extract S sub-registers of Q/D registers.
831 // (arm_ssubreg_0 is 1; arm_ssubreg_1 is 2; etc.)
832 def SSubReg_f32_reg : SDNodeXForm<imm, [{
833 return CurDAG->getTargetConstant(1 + N->getZExtValue(), MVT::i32);
836 // Translate lane numbers from Q registers to D subregs.
837 def SubReg_i8_lane : SDNodeXForm<imm, [{
838 return CurDAG->getTargetConstant(N->getZExtValue() & 7, MVT::i32);
840 def SubReg_i16_lane : SDNodeXForm<imm, [{
841 return CurDAG->getTargetConstant(N->getZExtValue() & 3, MVT::i32);
843 def SubReg_i32_lane : SDNodeXForm<imm, [{
844 return CurDAG->getTargetConstant(N->getZExtValue() & 1, MVT::i32);
847 //===----------------------------------------------------------------------===//
848 // Instruction Classes
849 //===----------------------------------------------------------------------===//
851 // Basic 2-register operations: single-, double- and quad-register.
852 class N2VS<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
853 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
854 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
855 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
856 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src),
857 IIC_VUNAD, OpcodeStr, Dt, "$dst, $src", "", []>;
858 class N2VD<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
859 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
860 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
861 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
862 (ins DPR:$src), IIC_VUNAD, OpcodeStr, Dt, "$dst, $src", "",
863 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src))))]>;
864 class N2VQ<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
865 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
866 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
867 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
868 (ins QPR:$src), IIC_VUNAQ, OpcodeStr, Dt, "$dst, $src", "",
869 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src))))]>;
871 // Basic 2-register intrinsics, both double- and quad-register.
872 class N2VDInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
873 bits<2> op17_16, bits<5> op11_7, bit op4,
874 InstrItinClass itin, string OpcodeStr, string Dt,
875 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
876 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
877 (ins DPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
878 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
879 class N2VQInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
880 bits<2> op17_16, bits<5> op11_7, bit op4,
881 InstrItinClass itin, string OpcodeStr, string Dt,
882 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
883 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
884 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
885 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
887 // Narrow 2-register intrinsics.
888 class N2VNInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
889 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
890 InstrItinClass itin, string OpcodeStr, string Dt,
891 ValueType TyD, ValueType TyQ, Intrinsic IntOp>
892 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$dst),
893 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
894 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src))))]>;
896 // Long 2-register intrinsics (currently only used for VMOVL).
897 class N2VLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
898 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
899 InstrItinClass itin, string OpcodeStr, string Dt,
900 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
901 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$dst),
902 (ins DPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
903 [(set QPR:$dst, (TyQ (IntOp (TyD DPR:$src))))]>;
905 // 2-register shuffles (VTRN/VZIP/VUZP), both double- and quad-register.
906 class N2VDShuffle<bits<2> op19_18, bits<5> op11_7, string OpcodeStr, string Dt>
907 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 0, 0, (outs DPR:$dst1, DPR:$dst2),
908 (ins DPR:$src1, DPR:$src2), IIC_VPERMD,
909 OpcodeStr, Dt, "$dst1, $dst2",
910 "$src1 = $dst1, $src2 = $dst2", []>;
911 class N2VQShuffle<bits<2> op19_18, bits<5> op11_7,
912 InstrItinClass itin, string OpcodeStr, string Dt>
913 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 1, 0, (outs QPR:$dst1, QPR:$dst2),
914 (ins QPR:$src1, QPR:$src2), itin, OpcodeStr, Dt, "$dst1, $dst2",
915 "$src1 = $dst1, $src2 = $dst2", []>;
917 // Basic 3-register operations: single-, double- and quad-register.
918 class N3VS<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
919 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
920 SDNode OpNode, bit Commutable>
921 : N3V<op24, op23, op21_20, op11_8, 0, op4,
922 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src1, DPR_VFP2:$src2), IIC_VBIND,
923 OpcodeStr, Dt, "$dst, $src1, $src2", "", []> {
924 let isCommutable = Commutable;
927 class N3VD<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
928 InstrItinClass itin, string OpcodeStr, string Dt,
929 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
930 : N3V<op24, op23, op21_20, op11_8, 0, op4,
931 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), itin,
932 OpcodeStr, Dt, "$dst, $src1, $src2", "",
933 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src1), (OpTy DPR:$src2))))]> {
934 let isCommutable = Commutable;
936 // Same as N3VD but no data type.
937 class N3VDX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
938 InstrItinClass itin, string OpcodeStr,
939 ValueType ResTy, ValueType OpTy,
940 SDNode OpNode, bit Commutable>
941 : N3VX<op24, op23, op21_20, op11_8, 0, op4,
942 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), itin,
943 OpcodeStr, "$dst, $src1, $src2", "",
944 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src1), (OpTy DPR:$src2))))]>{
945 let isCommutable = Commutable;
947 class N3VDSL<bits<2> op21_20, bits<4> op11_8,
948 InstrItinClass itin, string OpcodeStr, string Dt,
949 ValueType Ty, SDNode ShOp>
950 : N3V<0, 1, op21_20, op11_8, 1, 0,
951 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
952 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
954 (Ty (ShOp (Ty DPR:$src1),
955 (Ty (NEONvduplane (Ty DPR_VFP2:$src2), imm:$lane)))))]>{
956 let isCommutable = 0;
958 class N3VDSL16<bits<2> op21_20, bits<4> op11_8,
959 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
960 : N3V<0, 1, op21_20, op11_8, 1, 0,
961 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
962 IIC_VMULi16D, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
964 (Ty (ShOp (Ty DPR:$src1),
965 (Ty (NEONvduplane (Ty DPR_8:$src2), imm:$lane)))))]> {
966 let isCommutable = 0;
969 class N3VQ<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
970 InstrItinClass itin, string OpcodeStr, string Dt,
971 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
972 : N3V<op24, op23, op21_20, op11_8, 1, op4,
973 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), itin,
974 OpcodeStr, Dt, "$dst, $src1, $src2", "",
975 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src1), (OpTy QPR:$src2))))]> {
976 let isCommutable = Commutable;
978 class N3VQX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
979 InstrItinClass itin, string OpcodeStr,
980 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
981 : N3VX<op24, op23, op21_20, op11_8, 1, op4,
982 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), itin,
983 OpcodeStr, "$dst, $src1, $src2", "",
984 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src1), (OpTy QPR:$src2))))]>{
985 let isCommutable = Commutable;
987 class N3VQSL<bits<2> op21_20, bits<4> op11_8,
988 InstrItinClass itin, string OpcodeStr, string Dt,
989 ValueType ResTy, ValueType OpTy, SDNode ShOp>
990 : N3V<1, 1, op21_20, op11_8, 1, 0,
991 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
992 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
993 [(set (ResTy QPR:$dst),
994 (ResTy (ShOp (ResTy QPR:$src1),
995 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
997 let isCommutable = 0;
999 class N3VQSL16<bits<2> op21_20, bits<4> op11_8, string OpcodeStr, string Dt,
1000 ValueType ResTy, ValueType OpTy, SDNode ShOp>
1001 : N3V<1, 1, op21_20, op11_8, 1, 0,
1002 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1003 IIC_VMULi16Q, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1004 [(set (ResTy QPR:$dst),
1005 (ResTy (ShOp (ResTy QPR:$src1),
1006 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
1008 let isCommutable = 0;
1011 // Basic 3-register intrinsics, both double- and quad-register.
1012 class N3VDInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1013 InstrItinClass itin, string OpcodeStr, string Dt,
1014 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
1015 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1016 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), itin,
1017 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1018 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src1), (OpTy DPR:$src2))))]> {
1019 let isCommutable = Commutable;
1021 class N3VDIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1022 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
1023 : N3V<0, 1, op21_20, op11_8, 1, 0,
1024 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1025 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1026 [(set (Ty DPR:$dst),
1027 (Ty (IntOp (Ty DPR:$src1),
1028 (Ty (NEONvduplane (Ty DPR_VFP2:$src2),
1030 let isCommutable = 0;
1032 class N3VDIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1033 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
1034 : N3V<0, 1, op21_20, op11_8, 1, 0,
1035 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1036 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1037 [(set (Ty DPR:$dst),
1038 (Ty (IntOp (Ty DPR:$src1),
1039 (Ty (NEONvduplane (Ty DPR_8:$src2),
1041 let isCommutable = 0;
1044 class N3VQInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1045 InstrItinClass itin, string OpcodeStr, string Dt,
1046 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
1047 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1048 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), itin,
1049 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1050 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src1), (OpTy QPR:$src2))))]> {
1051 let isCommutable = Commutable;
1053 class N3VQIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1054 string OpcodeStr, string Dt,
1055 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1056 : N3V<1, 1, op21_20, op11_8, 1, 0,
1057 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1058 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1059 [(set (ResTy QPR:$dst),
1060 (ResTy (IntOp (ResTy QPR:$src1),
1061 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1063 let isCommutable = 0;
1065 class N3VQIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1066 string OpcodeStr, string Dt,
1067 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1068 : N3V<1, 1, op21_20, op11_8, 1, 0,
1069 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1070 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1071 [(set (ResTy QPR:$dst),
1072 (ResTy (IntOp (ResTy QPR:$src1),
1073 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
1075 let isCommutable = 0;
1078 // Multiply-Add/Sub operations: single-, double- and quad-register.
1079 class N3VSMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1080 InstrItinClass itin, string OpcodeStr, string Dt,
1081 ValueType Ty, SDNode MulOp, SDNode OpNode>
1082 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1083 (outs DPR_VFP2:$dst),
1084 (ins DPR_VFP2:$src1, DPR_VFP2:$src2, DPR_VFP2:$src3), itin,
1085 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst", []>;
1087 class N3VDMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1088 InstrItinClass itin, string OpcodeStr, string Dt,
1089 ValueType Ty, SDNode MulOp, SDNode OpNode>
1090 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1091 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3), itin,
1092 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1093 [(set DPR:$dst, (Ty (OpNode DPR:$src1,
1094 (Ty (MulOp DPR:$src2, DPR:$src3)))))]>;
1095 class N3VDMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1096 string OpcodeStr, string Dt,
1097 ValueType Ty, SDNode MulOp, SDNode ShOp>
1098 : N3V<0, 1, op21_20, op11_8, 1, 0,
1100 (ins DPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane), itin,
1101 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1102 [(set (Ty DPR:$dst),
1103 (Ty (ShOp (Ty DPR:$src1),
1104 (Ty (MulOp DPR:$src2,
1105 (Ty (NEONvduplane (Ty DPR_VFP2:$src3),
1107 class N3VDMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1108 string OpcodeStr, string Dt,
1109 ValueType Ty, SDNode MulOp, SDNode ShOp>
1110 : N3V<0, 1, op21_20, op11_8, 1, 0,
1112 (ins DPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane), itin,
1113 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1114 [(set (Ty DPR:$dst),
1115 (Ty (ShOp (Ty DPR:$src1),
1116 (Ty (MulOp DPR:$src2,
1117 (Ty (NEONvduplane (Ty DPR_8:$src3),
1120 class N3VQMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1121 InstrItinClass itin, string OpcodeStr, string Dt, ValueType Ty,
1122 SDNode MulOp, SDNode OpNode>
1123 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1124 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3), itin,
1125 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1126 [(set QPR:$dst, (Ty (OpNode QPR:$src1,
1127 (Ty (MulOp QPR:$src2, QPR:$src3)))))]>;
1128 class N3VQMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1129 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1130 SDNode MulOp, SDNode ShOp>
1131 : N3V<1, 1, op21_20, op11_8, 1, 0,
1133 (ins QPR:$src1, QPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane), itin,
1134 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1135 [(set (ResTy QPR:$dst),
1136 (ResTy (ShOp (ResTy QPR:$src1),
1137 (ResTy (MulOp QPR:$src2,
1138 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src3),
1140 class N3VQMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1141 string OpcodeStr, string Dt,
1142 ValueType ResTy, ValueType OpTy,
1143 SDNode MulOp, SDNode ShOp>
1144 : N3V<1, 1, op21_20, op11_8, 1, 0,
1146 (ins QPR:$src1, QPR:$src2, DPR_8:$src3, nohash_imm:$lane), itin,
1147 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1148 [(set (ResTy QPR:$dst),
1149 (ResTy (ShOp (ResTy QPR:$src1),
1150 (ResTy (MulOp QPR:$src2,
1151 (ResTy (NEONvduplane (OpTy DPR_8:$src3),
1154 // Neon 3-argument intrinsics, both double- and quad-register.
1155 // The destination register is also used as the first source operand register.
1156 class N3VDInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1157 InstrItinClass itin, string OpcodeStr, string Dt,
1158 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1159 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1160 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3), itin,
1161 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1162 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src1),
1163 (OpTy DPR:$src2), (OpTy DPR:$src3))))]>;
1164 class N3VQInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1165 InstrItinClass itin, string OpcodeStr, string Dt,
1166 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1167 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1168 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3), itin,
1169 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1170 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src1),
1171 (OpTy QPR:$src2), (OpTy QPR:$src3))))]>;
1173 // Neon Long 3-argument intrinsic. The destination register is
1174 // a quad-register and is also used as the first source operand register.
1175 class N3VLInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1176 InstrItinClass itin, string OpcodeStr, string Dt,
1177 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
1178 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1179 (outs QPR:$dst), (ins QPR:$src1, DPR:$src2, DPR:$src3), itin,
1180 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1182 (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$src2), (TyD DPR:$src3))))]>;
1183 class N3VLInt3SL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1184 string OpcodeStr, string Dt,
1185 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1186 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1188 (ins QPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane), itin,
1189 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1190 [(set (ResTy QPR:$dst),
1191 (ResTy (IntOp (ResTy QPR:$src1),
1193 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src3),
1195 class N3VLInt3SL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1196 InstrItinClass itin, string OpcodeStr, string Dt,
1197 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1198 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1200 (ins QPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane), itin,
1201 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1202 [(set (ResTy QPR:$dst),
1203 (ResTy (IntOp (ResTy QPR:$src1),
1205 (OpTy (NEONvduplane (OpTy DPR_8:$src3),
1208 // Narrowing 3-register intrinsics.
1209 class N3VNInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1210 string OpcodeStr, string Dt, ValueType TyD, ValueType TyQ,
1211 Intrinsic IntOp, bit Commutable>
1212 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1213 (outs DPR:$dst), (ins QPR:$src1, QPR:$src2), IIC_VBINi4D,
1214 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1215 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src1), (TyQ QPR:$src2))))]> {
1216 let isCommutable = Commutable;
1219 // Long 3-register intrinsics.
1220 class N3VLInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1221 InstrItinClass itin, string OpcodeStr, string Dt,
1222 ValueType TyQ, ValueType TyD, Intrinsic IntOp, bit Commutable>
1223 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1224 (outs QPR:$dst), (ins DPR:$src1, DPR:$src2), itin,
1225 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1226 [(set QPR:$dst, (TyQ (IntOp (TyD DPR:$src1), (TyD DPR:$src2))))]> {
1227 let isCommutable = Commutable;
1229 class N3VLIntSL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1230 string OpcodeStr, string Dt,
1231 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1232 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1233 (outs QPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1234 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1235 [(set (ResTy QPR:$dst),
1236 (ResTy (IntOp (OpTy DPR:$src1),
1237 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1239 class N3VLIntSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1240 InstrItinClass itin, string OpcodeStr, string Dt,
1241 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1242 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1243 (outs QPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1244 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1245 [(set (ResTy QPR:$dst),
1246 (ResTy (IntOp (OpTy DPR:$src1),
1247 (OpTy (NEONvduplane (OpTy DPR_8:$src2),
1250 // Wide 3-register intrinsics.
1251 class N3VWInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1252 string OpcodeStr, string Dt, ValueType TyQ, ValueType TyD,
1253 Intrinsic IntOp, bit Commutable>
1254 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1255 (outs QPR:$dst), (ins QPR:$src1, DPR:$src2), IIC_VSUBiD,
1256 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1257 [(set QPR:$dst, (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$src2))))]> {
1258 let isCommutable = Commutable;
1261 // Pairwise long 2-register intrinsics, both double- and quad-register.
1262 class N2VDPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1263 bits<2> op17_16, bits<5> op11_7, bit op4,
1264 string OpcodeStr, string Dt,
1265 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1266 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
1267 (ins DPR:$src), IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
1268 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
1269 class N2VQPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1270 bits<2> op17_16, bits<5> op11_7, bit op4,
1271 string OpcodeStr, string Dt,
1272 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1273 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
1274 (ins QPR:$src), IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
1275 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
1277 // Pairwise long 2-register accumulate intrinsics,
1278 // both double- and quad-register.
1279 // The destination register is also used as the first source operand register.
1280 class N2VDPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1281 bits<2> op17_16, bits<5> op11_7, bit op4,
1282 string OpcodeStr, string Dt,
1283 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1284 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
1285 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), IIC_VPALiD,
1286 OpcodeStr, Dt, "$dst, $src2", "$src1 = $dst",
1287 [(set DPR:$dst, (ResTy (IntOp (ResTy DPR:$src1), (OpTy DPR:$src2))))]>;
1288 class N2VQPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1289 bits<2> op17_16, bits<5> op11_7, bit op4,
1290 string OpcodeStr, string Dt,
1291 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1292 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4,
1293 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), IIC_VPALiQ,
1294 OpcodeStr, Dt, "$dst, $src2", "$src1 = $dst",
1295 [(set QPR:$dst, (ResTy (IntOp (ResTy QPR:$src1), (OpTy QPR:$src2))))]>;
1297 // Shift by immediate,
1298 // both double- and quad-register.
1299 class N2VDSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1300 InstrItinClass itin, string OpcodeStr, string Dt,
1301 ValueType Ty, SDNode OpNode>
1302 : N2VImm<op24, op23, op11_8, op7, 0, op4,
1303 (outs DPR:$dst), (ins DPR:$src, i32imm:$SIMM), itin,
1304 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1305 [(set DPR:$dst, (Ty (OpNode (Ty DPR:$src), (i32 imm:$SIMM))))]>;
1306 class N2VQSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1307 InstrItinClass itin, string OpcodeStr, string Dt,
1308 ValueType Ty, SDNode OpNode>
1309 : N2VImm<op24, op23, op11_8, op7, 1, op4,
1310 (outs QPR:$dst), (ins QPR:$src, i32imm:$SIMM), itin,
1311 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1312 [(set QPR:$dst, (Ty (OpNode (Ty QPR:$src), (i32 imm:$SIMM))))]>;
1314 // Long shift by immediate.
1315 class N2VLSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
1316 string OpcodeStr, string Dt,
1317 ValueType ResTy, ValueType OpTy, SDNode OpNode>
1318 : N2VImm<op24, op23, op11_8, op7, op6, op4,
1319 (outs QPR:$dst), (ins DPR:$src, i32imm:$SIMM), IIC_VSHLiD,
1320 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1321 [(set QPR:$dst, (ResTy (OpNode (OpTy DPR:$src),
1322 (i32 imm:$SIMM))))]>;
1324 // Narrow shift by immediate.
1325 class N2VNSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
1326 InstrItinClass itin, string OpcodeStr, string Dt,
1327 ValueType ResTy, ValueType OpTy, SDNode OpNode>
1328 : N2VImm<op24, op23, op11_8, op7, op6, op4,
1329 (outs DPR:$dst), (ins QPR:$src, i32imm:$SIMM), itin,
1330 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1331 [(set DPR:$dst, (ResTy (OpNode (OpTy QPR:$src),
1332 (i32 imm:$SIMM))))]>;
1334 // Shift right by immediate and accumulate,
1335 // both double- and quad-register.
1336 class N2VDShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1337 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1338 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$dst),
1339 (ins DPR:$src1, DPR:$src2, i32imm:$SIMM), IIC_VPALiD,
1340 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1341 [(set DPR:$dst, (Ty (add DPR:$src1,
1342 (Ty (ShOp DPR:$src2, (i32 imm:$SIMM))))))]>;
1343 class N2VQShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1344 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1345 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$dst),
1346 (ins QPR:$src1, QPR:$src2, i32imm:$SIMM), IIC_VPALiD,
1347 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1348 [(set QPR:$dst, (Ty (add QPR:$src1,
1349 (Ty (ShOp QPR:$src2, (i32 imm:$SIMM))))))]>;
1351 // Shift by immediate and insert,
1352 // both double- and quad-register.
1353 class N2VDShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1354 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1355 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$dst),
1356 (ins DPR:$src1, DPR:$src2, i32imm:$SIMM), IIC_VSHLiD,
1357 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1358 [(set DPR:$dst, (Ty (ShOp DPR:$src1, DPR:$src2, (i32 imm:$SIMM))))]>;
1359 class N2VQShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1360 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1361 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$dst),
1362 (ins QPR:$src1, QPR:$src2, i32imm:$SIMM), IIC_VSHLiQ,
1363 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1364 [(set QPR:$dst, (Ty (ShOp QPR:$src1, QPR:$src2, (i32 imm:$SIMM))))]>;
1366 // Convert, with fractional bits immediate,
1367 // both double- and quad-register.
1368 class N2VCvtD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1369 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1371 : N2VImm<op24, op23, op11_8, op7, 0, op4,
1372 (outs DPR:$dst), (ins DPR:$src, i32imm:$SIMM), IIC_VUNAD,
1373 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1374 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src), (i32 imm:$SIMM))))]>;
1375 class N2VCvtQ<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1376 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1378 : N2VImm<op24, op23, op11_8, op7, 1, op4,
1379 (outs QPR:$dst), (ins QPR:$src, i32imm:$SIMM), IIC_VUNAQ,
1380 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1381 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src), (i32 imm:$SIMM))))]>;
1383 //===----------------------------------------------------------------------===//
1385 //===----------------------------------------------------------------------===//
1387 // Abbreviations used in multiclass suffixes:
1388 // Q = quarter int (8 bit) elements
1389 // H = half int (16 bit) elements
1390 // S = single int (32 bit) elements
1391 // D = double int (64 bit) elements
1393 // Neon 2-register vector operations -- for disassembly only.
1395 // First with only element sizes of 8, 16 and 32 bits:
1396 multiclass N2V_QHS_cmp<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1397 bits<5> op11_7, bit op4, string opc, string Dt,
1399 // 64-bit vector types.
1400 def v8i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 0, op4,
1401 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1402 opc, !strconcat(Dt, "8"), asm, "", []>;
1403 def v4i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 0, op4,
1404 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1405 opc, !strconcat(Dt, "16"), asm, "", []>;
1406 def v2i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
1407 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1408 opc, !strconcat(Dt, "32"), asm, "", []>;
1409 def v2f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
1410 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1411 opc, "f32", asm, "", []> {
1412 let Inst{10} = 1; // overwrite F = 1
1415 // 128-bit vector types.
1416 def v16i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 1, op4,
1417 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1418 opc, !strconcat(Dt, "8"), asm, "", []>;
1419 def v8i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 1, op4,
1420 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1421 opc, !strconcat(Dt, "16"), asm, "", []>;
1422 def v4i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
1423 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1424 opc, !strconcat(Dt, "32"), asm, "", []>;
1425 def v4f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
1426 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1427 opc, "f32", asm, "", []> {
1428 let Inst{10} = 1; // overwrite F = 1
1432 // Neon 3-register vector operations.
1434 // First with only element sizes of 8, 16 and 32 bits:
1435 multiclass N3V_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1436 InstrItinClass itinD16, InstrItinClass itinD32,
1437 InstrItinClass itinQ16, InstrItinClass itinQ32,
1438 string OpcodeStr, string Dt,
1439 SDNode OpNode, bit Commutable = 0> {
1440 // 64-bit vector types.
1441 def v8i8 : N3VD<op24, op23, 0b00, op11_8, op4, itinD16,
1442 OpcodeStr, !strconcat(Dt, "8"),
1443 v8i8, v8i8, OpNode, Commutable>;
1444 def v4i16 : N3VD<op24, op23, 0b01, op11_8, op4, itinD16,
1445 OpcodeStr, !strconcat(Dt, "16"),
1446 v4i16, v4i16, OpNode, Commutable>;
1447 def v2i32 : N3VD<op24, op23, 0b10, op11_8, op4, itinD32,
1448 OpcodeStr, !strconcat(Dt, "32"),
1449 v2i32, v2i32, OpNode, Commutable>;
1451 // 128-bit vector types.
1452 def v16i8 : N3VQ<op24, op23, 0b00, op11_8, op4, itinQ16,
1453 OpcodeStr, !strconcat(Dt, "8"),
1454 v16i8, v16i8, OpNode, Commutable>;
1455 def v8i16 : N3VQ<op24, op23, 0b01, op11_8, op4, itinQ16,
1456 OpcodeStr, !strconcat(Dt, "16"),
1457 v8i16, v8i16, OpNode, Commutable>;
1458 def v4i32 : N3VQ<op24, op23, 0b10, op11_8, op4, itinQ32,
1459 OpcodeStr, !strconcat(Dt, "32"),
1460 v4i32, v4i32, OpNode, Commutable>;
1463 multiclass N3VSL_HS<bits<4> op11_8, string OpcodeStr, string Dt, SDNode ShOp> {
1464 def v4i16 : N3VDSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
1466 def v2i32 : N3VDSL<0b10, op11_8, IIC_VMULi32D, OpcodeStr, !strconcat(Dt,"32"),
1468 def v8i16 : N3VQSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
1469 v8i16, v4i16, ShOp>;
1470 def v4i32 : N3VQSL<0b10, op11_8, IIC_VMULi32Q, OpcodeStr, !strconcat(Dt,"32"),
1471 v4i32, v2i32, ShOp>;
1474 // ....then also with element size 64 bits:
1475 multiclass N3V_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1476 InstrItinClass itinD, InstrItinClass itinQ,
1477 string OpcodeStr, string Dt,
1478 SDNode OpNode, bit Commutable = 0>
1479 : N3V_QHS<op24, op23, op11_8, op4, itinD, itinD, itinQ, itinQ,
1480 OpcodeStr, Dt, OpNode, Commutable> {
1481 def v1i64 : N3VD<op24, op23, 0b11, op11_8, op4, itinD,
1482 OpcodeStr, !strconcat(Dt, "64"),
1483 v1i64, v1i64, OpNode, Commutable>;
1484 def v2i64 : N3VQ<op24, op23, 0b11, op11_8, op4, itinQ,
1485 OpcodeStr, !strconcat(Dt, "64"),
1486 v2i64, v2i64, OpNode, Commutable>;
1490 // Neon Narrowing 2-register vector intrinsics,
1491 // source operand element sizes of 16, 32 and 64 bits:
1492 multiclass N2VNInt_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1493 bits<5> op11_7, bit op6, bit op4,
1494 InstrItinClass itin, string OpcodeStr, string Dt,
1496 def v8i8 : N2VNInt<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
1497 itin, OpcodeStr, !strconcat(Dt, "16"),
1498 v8i8, v8i16, IntOp>;
1499 def v4i16 : N2VNInt<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
1500 itin, OpcodeStr, !strconcat(Dt, "32"),
1501 v4i16, v4i32, IntOp>;
1502 def v2i32 : N2VNInt<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
1503 itin, OpcodeStr, !strconcat(Dt, "64"),
1504 v2i32, v2i64, IntOp>;
1508 // Neon Lengthening 2-register vector intrinsic (currently specific to VMOVL).
1509 // source operand element sizes of 16, 32 and 64 bits:
1510 multiclass N2VLInt_QHS<bits<2> op24_23, bits<5> op11_7, bit op6, bit op4,
1511 string OpcodeStr, string Dt, Intrinsic IntOp> {
1512 def v8i16 : N2VLInt<op24_23, 0b00, 0b10, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
1513 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
1514 def v4i32 : N2VLInt<op24_23, 0b01, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
1515 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
1516 def v2i64 : N2VLInt<op24_23, 0b10, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
1517 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1521 // Neon 3-register vector intrinsics.
1523 // First with only element sizes of 16 and 32 bits:
1524 multiclass N3VInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
1525 InstrItinClass itinD16, InstrItinClass itinD32,
1526 InstrItinClass itinQ16, InstrItinClass itinQ32,
1527 string OpcodeStr, string Dt,
1528 Intrinsic IntOp, bit Commutable = 0> {
1529 // 64-bit vector types.
1530 def v4i16 : N3VDInt<op24, op23, 0b01, op11_8, op4, itinD16,
1531 OpcodeStr, !strconcat(Dt, "16"),
1532 v4i16, v4i16, IntOp, Commutable>;
1533 def v2i32 : N3VDInt<op24, op23, 0b10, op11_8, op4, itinD32,
1534 OpcodeStr, !strconcat(Dt, "32"),
1535 v2i32, v2i32, IntOp, Commutable>;
1537 // 128-bit vector types.
1538 def v8i16 : N3VQInt<op24, op23, 0b01, op11_8, op4, itinQ16,
1539 OpcodeStr, !strconcat(Dt, "16"),
1540 v8i16, v8i16, IntOp, Commutable>;
1541 def v4i32 : N3VQInt<op24, op23, 0b10, op11_8, op4, itinQ32,
1542 OpcodeStr, !strconcat(Dt, "32"),
1543 v4i32, v4i32, IntOp, Commutable>;
1546 multiclass N3VIntSL_HS<bits<4> op11_8,
1547 InstrItinClass itinD16, InstrItinClass itinD32,
1548 InstrItinClass itinQ16, InstrItinClass itinQ32,
1549 string OpcodeStr, string Dt, Intrinsic IntOp> {
1550 def v4i16 : N3VDIntSL16<0b01, op11_8, itinD16,
1551 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp>;
1552 def v2i32 : N3VDIntSL<0b10, op11_8, itinD32,
1553 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp>;
1554 def v8i16 : N3VQIntSL16<0b01, op11_8, itinQ16,
1555 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16, IntOp>;
1556 def v4i32 : N3VQIntSL<0b10, op11_8, itinQ32,
1557 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32, IntOp>;
1560 // ....then also with element size of 8 bits:
1561 multiclass N3VInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1562 InstrItinClass itinD16, InstrItinClass itinD32,
1563 InstrItinClass itinQ16, InstrItinClass itinQ32,
1564 string OpcodeStr, string Dt,
1565 Intrinsic IntOp, bit Commutable = 0>
1566 : N3VInt_HS<op24, op23, op11_8, op4, itinD16, itinD32, itinQ16, itinQ32,
1567 OpcodeStr, Dt, IntOp, Commutable> {
1568 def v8i8 : N3VDInt<op24, op23, 0b00, op11_8, op4, itinD16,
1569 OpcodeStr, !strconcat(Dt, "8"),
1570 v8i8, v8i8, IntOp, Commutable>;
1571 def v16i8 : N3VQInt<op24, op23, 0b00, op11_8, op4, itinQ16,
1572 OpcodeStr, !strconcat(Dt, "8"),
1573 v16i8, v16i8, IntOp, Commutable>;
1576 // ....then also with element size of 64 bits:
1577 multiclass N3VInt_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1578 InstrItinClass itinD16, InstrItinClass itinD32,
1579 InstrItinClass itinQ16, InstrItinClass itinQ32,
1580 string OpcodeStr, string Dt,
1581 Intrinsic IntOp, bit Commutable = 0>
1582 : N3VInt_QHS<op24, op23, op11_8, op4, itinD16, itinD32, itinQ16, itinQ32,
1583 OpcodeStr, Dt, IntOp, Commutable> {
1584 def v1i64 : N3VDInt<op24, op23, 0b11, op11_8, op4, itinD32,
1585 OpcodeStr, !strconcat(Dt, "64"),
1586 v1i64, v1i64, IntOp, Commutable>;
1587 def v2i64 : N3VQInt<op24, op23, 0b11, op11_8, op4, itinQ32,
1588 OpcodeStr, !strconcat(Dt, "64"),
1589 v2i64, v2i64, IntOp, Commutable>;
1593 // Neon Narrowing 3-register vector intrinsics,
1594 // source operand element sizes of 16, 32 and 64 bits:
1595 multiclass N3VNInt_HSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1596 string OpcodeStr, string Dt,
1597 Intrinsic IntOp, bit Commutable = 0> {
1598 def v8i8 : N3VNInt<op24, op23, 0b00, op11_8, op4,
1599 OpcodeStr, !strconcat(Dt, "16"),
1600 v8i8, v8i16, IntOp, Commutable>;
1601 def v4i16 : N3VNInt<op24, op23, 0b01, op11_8, op4,
1602 OpcodeStr, !strconcat(Dt, "32"),
1603 v4i16, v4i32, IntOp, Commutable>;
1604 def v2i32 : N3VNInt<op24, op23, 0b10, op11_8, op4,
1605 OpcodeStr, !strconcat(Dt, "64"),
1606 v2i32, v2i64, IntOp, Commutable>;
1610 // Neon Long 3-register vector intrinsics.
1612 // First with only element sizes of 16 and 32 bits:
1613 multiclass N3VLInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
1614 InstrItinClass itin, string OpcodeStr, string Dt,
1615 Intrinsic IntOp, bit Commutable = 0> {
1616 def v4i32 : N3VLInt<op24, op23, 0b01, op11_8, op4, itin,
1617 OpcodeStr, !strconcat(Dt, "16"),
1618 v4i32, v4i16, IntOp, Commutable>;
1619 def v2i64 : N3VLInt<op24, op23, 0b10, op11_8, op4, itin,
1620 OpcodeStr, !strconcat(Dt, "32"),
1621 v2i64, v2i32, IntOp, Commutable>;
1624 multiclass N3VLIntSL_HS<bit op24, bits<4> op11_8,
1625 InstrItinClass itin, string OpcodeStr, string Dt,
1627 def v4i16 : N3VLIntSL16<op24, 0b01, op11_8, itin,
1628 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
1629 def v2i32 : N3VLIntSL<op24, 0b10, op11_8, itin,
1630 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1633 // ....then also with element size of 8 bits:
1634 multiclass N3VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1635 InstrItinClass itin, string OpcodeStr, string Dt,
1636 Intrinsic IntOp, bit Commutable = 0>
1637 : N3VLInt_HS<op24, op23, op11_8, op4, itin, OpcodeStr, Dt,
1638 IntOp, Commutable> {
1639 def v8i16 : N3VLInt<op24, op23, 0b00, op11_8, op4, itin,
1640 OpcodeStr, !strconcat(Dt, "8"),
1641 v8i16, v8i8, IntOp, Commutable>;
1645 // Neon Wide 3-register vector intrinsics,
1646 // source operand element sizes of 8, 16 and 32 bits:
1647 multiclass N3VWInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1648 string OpcodeStr, string Dt,
1649 Intrinsic IntOp, bit Commutable = 0> {
1650 def v8i16 : N3VWInt<op24, op23, 0b00, op11_8, op4,
1651 OpcodeStr, !strconcat(Dt, "8"),
1652 v8i16, v8i8, IntOp, Commutable>;
1653 def v4i32 : N3VWInt<op24, op23, 0b01, op11_8, op4,
1654 OpcodeStr, !strconcat(Dt, "16"),
1655 v4i32, v4i16, IntOp, Commutable>;
1656 def v2i64 : N3VWInt<op24, op23, 0b10, op11_8, op4,
1657 OpcodeStr, !strconcat(Dt, "32"),
1658 v2i64, v2i32, IntOp, Commutable>;
1662 // Neon Multiply-Op vector operations,
1663 // element sizes of 8, 16 and 32 bits:
1664 multiclass N3VMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1665 InstrItinClass itinD16, InstrItinClass itinD32,
1666 InstrItinClass itinQ16, InstrItinClass itinQ32,
1667 string OpcodeStr, string Dt, SDNode OpNode> {
1668 // 64-bit vector types.
1669 def v8i8 : N3VDMulOp<op24, op23, 0b00, op11_8, op4, itinD16,
1670 OpcodeStr, !strconcat(Dt, "8"), v8i8, mul, OpNode>;
1671 def v4i16 : N3VDMulOp<op24, op23, 0b01, op11_8, op4, itinD16,
1672 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, OpNode>;
1673 def v2i32 : N3VDMulOp<op24, op23, 0b10, op11_8, op4, itinD32,
1674 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, OpNode>;
1676 // 128-bit vector types.
1677 def v16i8 : N3VQMulOp<op24, op23, 0b00, op11_8, op4, itinQ16,
1678 OpcodeStr, !strconcat(Dt, "8"), v16i8, mul, OpNode>;
1679 def v8i16 : N3VQMulOp<op24, op23, 0b01, op11_8, op4, itinQ16,
1680 OpcodeStr, !strconcat(Dt, "16"), v8i16, mul, OpNode>;
1681 def v4i32 : N3VQMulOp<op24, op23, 0b10, op11_8, op4, itinQ32,
1682 OpcodeStr, !strconcat(Dt, "32"), v4i32, mul, OpNode>;
1685 multiclass N3VMulOpSL_HS<bits<4> op11_8,
1686 InstrItinClass itinD16, InstrItinClass itinD32,
1687 InstrItinClass itinQ16, InstrItinClass itinQ32,
1688 string OpcodeStr, string Dt, SDNode ShOp> {
1689 def v4i16 : N3VDMulOpSL16<0b01, op11_8, itinD16,
1690 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, ShOp>;
1691 def v2i32 : N3VDMulOpSL<0b10, op11_8, itinD32,
1692 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, ShOp>;
1693 def v8i16 : N3VQMulOpSL16<0b01, op11_8, itinQ16,
1694 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16,
1696 def v4i32 : N3VQMulOpSL<0b10, op11_8, itinQ32,
1697 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32,
1701 // Neon 3-argument intrinsics,
1702 // element sizes of 8, 16 and 32 bits:
1703 multiclass N3VInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1704 string OpcodeStr, string Dt, Intrinsic IntOp> {
1705 // 64-bit vector types.
1706 def v8i8 : N3VDInt3<op24, op23, 0b00, op11_8, op4, IIC_VMACi16D,
1707 OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
1708 def v4i16 : N3VDInt3<op24, op23, 0b01, op11_8, op4, IIC_VMACi16D,
1709 OpcodeStr, !strconcat(Dt, "16"), v4i16, v4i16, IntOp>;
1710 def v2i32 : N3VDInt3<op24, op23, 0b10, op11_8, op4, IIC_VMACi32D,
1711 OpcodeStr, !strconcat(Dt, "32"), v2i32, v2i32, IntOp>;
1713 // 128-bit vector types.
1714 def v16i8 : N3VQInt3<op24, op23, 0b00, op11_8, op4, IIC_VMACi16Q,
1715 OpcodeStr, !strconcat(Dt, "8"), v16i8, v16i8, IntOp>;
1716 def v8i16 : N3VQInt3<op24, op23, 0b01, op11_8, op4, IIC_VMACi16Q,
1717 OpcodeStr, !strconcat(Dt, "16"), v8i16, v8i16, IntOp>;
1718 def v4i32 : N3VQInt3<op24, op23, 0b10, op11_8, op4, IIC_VMACi32Q,
1719 OpcodeStr, !strconcat(Dt, "32"), v4i32, v4i32, IntOp>;
1723 // Neon Long 3-argument intrinsics.
1725 // First with only element sizes of 16 and 32 bits:
1726 multiclass N3VLInt3_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
1727 string OpcodeStr, string Dt, Intrinsic IntOp> {
1728 def v4i32 : N3VLInt3<op24, op23, 0b01, op11_8, op4, IIC_VMACi16D,
1729 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
1730 def v2i64 : N3VLInt3<op24, op23, 0b10, op11_8, op4, IIC_VMACi16D,
1731 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1734 multiclass N3VLInt3SL_HS<bit op24, bits<4> op11_8,
1735 string OpcodeStr, string Dt, Intrinsic IntOp> {
1736 def v4i16 : N3VLInt3SL16<op24, 0b01, op11_8, IIC_VMACi16D,
1737 OpcodeStr, !strconcat(Dt,"16"), v4i32, v4i16, IntOp>;
1738 def v2i32 : N3VLInt3SL<op24, 0b10, op11_8, IIC_VMACi32D,
1739 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1742 // ....then also with element size of 8 bits:
1743 multiclass N3VLInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1744 string OpcodeStr, string Dt, Intrinsic IntOp>
1745 : N3VLInt3_HS<op24, op23, op11_8, op4, OpcodeStr, Dt, IntOp> {
1746 def v8i16 : N3VLInt3<op24, op23, 0b00, op11_8, op4, IIC_VMACi16D,
1747 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
1751 // Neon 2-register vector intrinsics,
1752 // element sizes of 8, 16 and 32 bits:
1753 multiclass N2VInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1754 bits<5> op11_7, bit op4,
1755 InstrItinClass itinD, InstrItinClass itinQ,
1756 string OpcodeStr, string Dt, Intrinsic IntOp> {
1757 // 64-bit vector types.
1758 def v8i8 : N2VDInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1759 itinD, OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
1760 def v4i16 : N2VDInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1761 itinD, OpcodeStr, !strconcat(Dt, "16"),v4i16,v4i16,IntOp>;
1762 def v2i32 : N2VDInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1763 itinD, OpcodeStr, !strconcat(Dt, "32"),v2i32,v2i32,IntOp>;
1765 // 128-bit vector types.
1766 def v16i8 : N2VQInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1767 itinQ, OpcodeStr, !strconcat(Dt, "8"), v16i8,v16i8,IntOp>;
1768 def v8i16 : N2VQInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1769 itinQ, OpcodeStr, !strconcat(Dt, "16"),v8i16,v8i16,IntOp>;
1770 def v4i32 : N2VQInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1771 itinQ, OpcodeStr, !strconcat(Dt, "32"),v4i32,v4i32,IntOp>;
1775 // Neon Pairwise long 2-register intrinsics,
1776 // element sizes of 8, 16 and 32 bits:
1777 multiclass N2VPLInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1778 bits<5> op11_7, bit op4,
1779 string OpcodeStr, string Dt, Intrinsic IntOp> {
1780 // 64-bit vector types.
1781 def v8i8 : N2VDPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1782 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
1783 def v4i16 : N2VDPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1784 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
1785 def v2i32 : N2VDPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1786 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
1788 // 128-bit vector types.
1789 def v16i8 : N2VQPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1790 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
1791 def v8i16 : N2VQPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1792 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
1793 def v4i32 : N2VQPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1794 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
1798 // Neon Pairwise long 2-register accumulate intrinsics,
1799 // element sizes of 8, 16 and 32 bits:
1800 multiclass N2VPLInt2_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1801 bits<5> op11_7, bit op4,
1802 string OpcodeStr, string Dt, Intrinsic IntOp> {
1803 // 64-bit vector types.
1804 def v8i8 : N2VDPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1805 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
1806 def v4i16 : N2VDPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1807 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
1808 def v2i32 : N2VDPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1809 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
1811 // 128-bit vector types.
1812 def v16i8 : N2VQPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1813 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
1814 def v8i16 : N2VQPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1815 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
1816 def v4i32 : N2VQPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1817 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
1821 // Neon 2-register vector shift by immediate,
1822 // element sizes of 8, 16, 32 and 64 bits:
1823 multiclass N2VSh_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1824 InstrItinClass itin, string OpcodeStr, string Dt,
1826 // 64-bit vector types.
1827 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, itin,
1828 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
1829 let Inst{21-19} = 0b001; // imm6 = 001xxx
1831 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, itin,
1832 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
1833 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1835 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, itin,
1836 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
1837 let Inst{21} = 0b1; // imm6 = 1xxxxx
1839 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, itin,
1840 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
1843 // 128-bit vector types.
1844 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, itin,
1845 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
1846 let Inst{21-19} = 0b001; // imm6 = 001xxx
1848 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, itin,
1849 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
1850 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1852 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, itin,
1853 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
1854 let Inst{21} = 0b1; // imm6 = 1xxxxx
1856 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, itin,
1857 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
1862 // Neon Shift-Accumulate vector operations,
1863 // element sizes of 8, 16, 32 and 64 bits:
1864 multiclass N2VShAdd_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1865 string OpcodeStr, string Dt, SDNode ShOp> {
1866 // 64-bit vector types.
1867 def v8i8 : N2VDShAdd<op24, op23, op11_8, 0, op4,
1868 OpcodeStr, !strconcat(Dt, "8"), v8i8, ShOp> {
1869 let Inst{21-19} = 0b001; // imm6 = 001xxx
1871 def v4i16 : N2VDShAdd<op24, op23, op11_8, 0, op4,
1872 OpcodeStr, !strconcat(Dt, "16"), v4i16, ShOp> {
1873 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1875 def v2i32 : N2VDShAdd<op24, op23, op11_8, 0, op4,
1876 OpcodeStr, !strconcat(Dt, "32"), v2i32, ShOp> {
1877 let Inst{21} = 0b1; // imm6 = 1xxxxx
1879 def v1i64 : N2VDShAdd<op24, op23, op11_8, 1, op4,
1880 OpcodeStr, !strconcat(Dt, "64"), v1i64, ShOp>;
1883 // 128-bit vector types.
1884 def v16i8 : N2VQShAdd<op24, op23, op11_8, 0, op4,
1885 OpcodeStr, !strconcat(Dt, "8"), v16i8, ShOp> {
1886 let Inst{21-19} = 0b001; // imm6 = 001xxx
1888 def v8i16 : N2VQShAdd<op24, op23, op11_8, 0, op4,
1889 OpcodeStr, !strconcat(Dt, "16"), v8i16, ShOp> {
1890 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1892 def v4i32 : N2VQShAdd<op24, op23, op11_8, 0, op4,
1893 OpcodeStr, !strconcat(Dt, "32"), v4i32, ShOp> {
1894 let Inst{21} = 0b1; // imm6 = 1xxxxx
1896 def v2i64 : N2VQShAdd<op24, op23, op11_8, 1, op4,
1897 OpcodeStr, !strconcat(Dt, "64"), v2i64, ShOp>;
1902 // Neon Shift-Insert vector operations,
1903 // element sizes of 8, 16, 32 and 64 bits:
1904 multiclass N2VShIns_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1905 string OpcodeStr, SDNode ShOp> {
1906 // 64-bit vector types.
1907 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4,
1908 OpcodeStr, "8", v8i8, ShOp> {
1909 let Inst{21-19} = 0b001; // imm6 = 001xxx
1911 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4,
1912 OpcodeStr, "16", v4i16, ShOp> {
1913 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1915 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4,
1916 OpcodeStr, "32", v2i32, ShOp> {
1917 let Inst{21} = 0b1; // imm6 = 1xxxxx
1919 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4,
1920 OpcodeStr, "64", v1i64, ShOp>;
1923 // 128-bit vector types.
1924 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4,
1925 OpcodeStr, "8", v16i8, ShOp> {
1926 let Inst{21-19} = 0b001; // imm6 = 001xxx
1928 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4,
1929 OpcodeStr, "16", v8i16, ShOp> {
1930 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1932 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4,
1933 OpcodeStr, "32", v4i32, ShOp> {
1934 let Inst{21} = 0b1; // imm6 = 1xxxxx
1936 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4,
1937 OpcodeStr, "64", v2i64, ShOp>;
1941 // Neon Shift Long operations,
1942 // element sizes of 8, 16, 32 bits:
1943 multiclass N2VLSh_QHS<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
1944 bit op4, string OpcodeStr, string Dt, SDNode OpNode> {
1945 def v8i16 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
1946 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode> {
1947 let Inst{21-19} = 0b001; // imm6 = 001xxx
1949 def v4i32 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
1950 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode> {
1951 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1953 def v2i64 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
1954 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode> {
1955 let Inst{21} = 0b1; // imm6 = 1xxxxx
1959 // Neon Shift Narrow operations,
1960 // element sizes of 16, 32, 64 bits:
1961 multiclass N2VNSh_HSD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
1962 bit op4, InstrItinClass itin, string OpcodeStr, string Dt,
1964 def v8i8 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
1965 OpcodeStr, !strconcat(Dt, "16"), v8i8, v8i16, OpNode> {
1966 let Inst{21-19} = 0b001; // imm6 = 001xxx
1968 def v4i16 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
1969 OpcodeStr, !strconcat(Dt, "32"), v4i16, v4i32, OpNode> {
1970 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1972 def v2i32 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
1973 OpcodeStr, !strconcat(Dt, "64"), v2i32, v2i64, OpNode> {
1974 let Inst{21} = 0b1; // imm6 = 1xxxxx
1978 //===----------------------------------------------------------------------===//
1979 // Instruction Definitions.
1980 //===----------------------------------------------------------------------===//
1982 // Vector Add Operations.
1984 // VADD : Vector Add (integer and floating-point)
1985 defm VADD : N3V_QHSD<0, 0, 0b1000, 0, IIC_VBINiD, IIC_VBINiQ, "vadd", "i",
1987 def VADDfd : N3VD<0, 0, 0b00, 0b1101, 0, IIC_VBIND, "vadd", "f32",
1988 v2f32, v2f32, fadd, 1>;
1989 def VADDfq : N3VQ<0, 0, 0b00, 0b1101, 0, IIC_VBINQ, "vadd", "f32",
1990 v4f32, v4f32, fadd, 1>;
1991 // VADDL : Vector Add Long (Q = D + D)
1992 defm VADDLs : N3VLInt_QHS<0,1,0b0000,0, IIC_VSHLiD, "vaddl", "s",
1993 int_arm_neon_vaddls, 1>;
1994 defm VADDLu : N3VLInt_QHS<1,1,0b0000,0, IIC_VSHLiD, "vaddl", "u",
1995 int_arm_neon_vaddlu, 1>;
1996 // VADDW : Vector Add Wide (Q = Q + D)
1997 defm VADDWs : N3VWInt_QHS<0,1,0b0001,0, "vaddw", "s", int_arm_neon_vaddws, 0>;
1998 defm VADDWu : N3VWInt_QHS<1,1,0b0001,0, "vaddw", "u", int_arm_neon_vaddwu, 0>;
1999 // VHADD : Vector Halving Add
2000 defm VHADDs : N3VInt_QHS<0,0,0b0000,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2001 IIC_VBINi4Q, "vhadd", "s", int_arm_neon_vhadds, 1>;
2002 defm VHADDu : N3VInt_QHS<1,0,0b0000,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2003 IIC_VBINi4Q, "vhadd", "u", int_arm_neon_vhaddu, 1>;
2004 // VRHADD : Vector Rounding Halving Add
2005 defm VRHADDs : N3VInt_QHS<0,0,0b0001,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2006 IIC_VBINi4Q, "vrhadd", "s", int_arm_neon_vrhadds, 1>;
2007 defm VRHADDu : N3VInt_QHS<1,0,0b0001,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2008 IIC_VBINi4Q, "vrhadd", "u", int_arm_neon_vrhaddu, 1>;
2009 // VQADD : Vector Saturating Add
2010 defm VQADDs : N3VInt_QHSD<0,0,0b0000,1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2011 IIC_VBINi4Q, "vqadd", "s", int_arm_neon_vqadds, 1>;
2012 defm VQADDu : N3VInt_QHSD<1,0,0b0000,1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2013 IIC_VBINi4Q, "vqadd", "u", int_arm_neon_vqaddu, 1>;
2014 // VADDHN : Vector Add and Narrow Returning High Half (D = Q + Q)
2015 defm VADDHN : N3VNInt_HSD<0,1,0b0100,0, "vaddhn", "i",
2016 int_arm_neon_vaddhn, 1>;
2017 // VRADDHN : Vector Rounding Add and Narrow Returning High Half (D = Q + Q)
2018 defm VRADDHN : N3VNInt_HSD<1,1,0b0100,0, "vraddhn", "i",
2019 int_arm_neon_vraddhn, 1>;
2021 // Vector Multiply Operations.
2023 // VMUL : Vector Multiply (integer, polynomial and floating-point)
2024 defm VMUL : N3V_QHS<0, 0, 0b1001, 1, IIC_VMULi16D, IIC_VMULi32D,
2025 IIC_VMULi16Q, IIC_VMULi32Q, "vmul", "i", mul, 1>;
2026 def VMULpd : N3VDInt<1, 0, 0b00, 0b1001, 1, IIC_VMULi16D, "vmul", "p8",
2027 v8i8, v8i8, int_arm_neon_vmulp, 1>;
2028 def VMULpq : N3VQInt<1, 0, 0b00, 0b1001, 1, IIC_VMULi16Q, "vmul", "p8",
2029 v16i8, v16i8, int_arm_neon_vmulp, 1>;
2030 def VMULfd : N3VD<1, 0, 0b00, 0b1101, 1, IIC_VBIND, "vmul", "f32",
2031 v2f32, v2f32, fmul, 1>;
2032 def VMULfq : N3VQ<1, 0, 0b00, 0b1101, 1, IIC_VBINQ, "vmul", "f32",
2033 v4f32, v4f32, fmul, 1>;
2034 defm VMULsl : N3VSL_HS<0b1000, "vmul", "i", mul>;
2035 def VMULslfd : N3VDSL<0b10, 0b1001, IIC_VBIND, "vmul", "f32", v2f32, fmul>;
2036 def VMULslfq : N3VQSL<0b10, 0b1001, IIC_VBINQ, "vmul", "f32", v4f32,
2039 def : Pat<(v8i16 (mul (v8i16 QPR:$src1),
2040 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
2041 (v8i16 (VMULslv8i16 (v8i16 QPR:$src1),
2042 (v4i16 (EXTRACT_SUBREG QPR:$src2,
2043 (DSubReg_i16_reg imm:$lane))),
2044 (SubReg_i16_lane imm:$lane)))>;
2045 def : Pat<(v4i32 (mul (v4i32 QPR:$src1),
2046 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
2047 (v4i32 (VMULslv4i32 (v4i32 QPR:$src1),
2048 (v2i32 (EXTRACT_SUBREG QPR:$src2,
2049 (DSubReg_i32_reg imm:$lane))),
2050 (SubReg_i32_lane imm:$lane)))>;
2051 def : Pat<(v4f32 (fmul (v4f32 QPR:$src1),
2052 (v4f32 (NEONvduplane (v4f32 QPR:$src2), imm:$lane)))),
2053 (v4f32 (VMULslfq (v4f32 QPR:$src1),
2054 (v2f32 (EXTRACT_SUBREG QPR:$src2,
2055 (DSubReg_i32_reg imm:$lane))),
2056 (SubReg_i32_lane imm:$lane)))>;
2058 // VQDMULH : Vector Saturating Doubling Multiply Returning High Half
2059 defm VQDMULH : N3VInt_HS<0, 0, 0b1011, 0, IIC_VMULi16D, IIC_VMULi32D,
2060 IIC_VMULi16Q, IIC_VMULi32Q,
2061 "vqdmulh", "s", int_arm_neon_vqdmulh, 1>;
2062 defm VQDMULHsl: N3VIntSL_HS<0b1100, IIC_VMULi16D, IIC_VMULi32D,
2063 IIC_VMULi16Q, IIC_VMULi32Q,
2064 "vqdmulh", "s", int_arm_neon_vqdmulh>;
2065 def : Pat<(v8i16 (int_arm_neon_vqdmulh (v8i16 QPR:$src1),
2066 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
2068 (v8i16 (VQDMULHslv8i16 (v8i16 QPR:$src1),
2069 (v4i16 (EXTRACT_SUBREG QPR:$src2,
2070 (DSubReg_i16_reg imm:$lane))),
2071 (SubReg_i16_lane imm:$lane)))>;
2072 def : Pat<(v4i32 (int_arm_neon_vqdmulh (v4i32 QPR:$src1),
2073 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
2075 (v4i32 (VQDMULHslv4i32 (v4i32 QPR:$src1),
2076 (v2i32 (EXTRACT_SUBREG QPR:$src2,
2077 (DSubReg_i32_reg imm:$lane))),
2078 (SubReg_i32_lane imm:$lane)))>;
2080 // VQRDMULH : Vector Rounding Saturating Doubling Multiply Returning High Half
2081 defm VQRDMULH : N3VInt_HS<1, 0, 0b1011, 0, IIC_VMULi16D, IIC_VMULi32D,
2082 IIC_VMULi16Q, IIC_VMULi32Q,
2083 "vqrdmulh", "s", int_arm_neon_vqrdmulh, 1>;
2084 defm VQRDMULHsl : N3VIntSL_HS<0b1101, IIC_VMULi16D, IIC_VMULi32D,
2085 IIC_VMULi16Q, IIC_VMULi32Q,
2086 "vqrdmulh", "s", int_arm_neon_vqrdmulh>;
2087 def : Pat<(v8i16 (int_arm_neon_vqrdmulh (v8i16 QPR:$src1),
2088 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
2090 (v8i16 (VQRDMULHslv8i16 (v8i16 QPR:$src1),
2091 (v4i16 (EXTRACT_SUBREG QPR:$src2,
2092 (DSubReg_i16_reg imm:$lane))),
2093 (SubReg_i16_lane imm:$lane)))>;
2094 def : Pat<(v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$src1),
2095 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
2097 (v4i32 (VQRDMULHslv4i32 (v4i32 QPR:$src1),
2098 (v2i32 (EXTRACT_SUBREG QPR:$src2,
2099 (DSubReg_i32_reg imm:$lane))),
2100 (SubReg_i32_lane imm:$lane)))>;
2102 // VMULL : Vector Multiply Long (integer and polynomial) (Q = D * D)
2103 defm VMULLs : N3VLInt_QHS<0,1,0b1100,0, IIC_VMULi16D, "vmull", "s",
2104 int_arm_neon_vmulls, 1>;
2105 defm VMULLu : N3VLInt_QHS<1,1,0b1100,0, IIC_VMULi16D, "vmull", "u",
2106 int_arm_neon_vmullu, 1>;
2107 def VMULLp : N3VLInt<0, 1, 0b00, 0b1110, 0, IIC_VMULi16D, "vmull", "p8",
2108 v8i16, v8i8, int_arm_neon_vmullp, 1>;
2109 defm VMULLsls : N3VLIntSL_HS<0, 0b1010, IIC_VMULi16D, "vmull", "s",
2110 int_arm_neon_vmulls>;
2111 defm VMULLslu : N3VLIntSL_HS<1, 0b1010, IIC_VMULi16D, "vmull", "u",
2112 int_arm_neon_vmullu>;
2114 // VQDMULL : Vector Saturating Doubling Multiply Long (Q = D * D)
2115 defm VQDMULL : N3VLInt_HS<0,1,0b1101,0, IIC_VMULi16D, "vqdmull", "s",
2116 int_arm_neon_vqdmull, 1>;
2117 defm VQDMULLsl: N3VLIntSL_HS<0, 0b1011, IIC_VMULi16D, "vqdmull", "s",
2118 int_arm_neon_vqdmull>;
2120 // Vector Multiply-Accumulate and Multiply-Subtract Operations.
2122 // VMLA : Vector Multiply Accumulate (integer and floating-point)
2123 defm VMLA : N3VMulOp_QHS<0, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
2124 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
2125 def VMLAfd : N3VDMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACD, "vmla", "f32",
2127 def VMLAfq : N3VQMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACQ, "vmla", "f32",
2129 defm VMLAsl : N3VMulOpSL_HS<0b0000, IIC_VMACi16D, IIC_VMACi32D,
2130 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
2131 def VMLAslfd : N3VDMulOpSL<0b10, 0b0001, IIC_VMACD, "vmla", "f32",
2133 def VMLAslfq : N3VQMulOpSL<0b10, 0b0001, IIC_VMACQ, "vmla", "f32",
2134 v4f32, v2f32, fmul, fadd>;
2136 def : Pat<(v8i16 (add (v8i16 QPR:$src1),
2137 (mul (v8i16 QPR:$src2),
2138 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
2139 (v8i16 (VMLAslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
2140 (v4i16 (EXTRACT_SUBREG QPR:$src3,
2141 (DSubReg_i16_reg imm:$lane))),
2142 (SubReg_i16_lane imm:$lane)))>;
2144 def : Pat<(v4i32 (add (v4i32 QPR:$src1),
2145 (mul (v4i32 QPR:$src2),
2146 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
2147 (v4i32 (VMLAslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
2148 (v2i32 (EXTRACT_SUBREG QPR:$src3,
2149 (DSubReg_i32_reg imm:$lane))),
2150 (SubReg_i32_lane imm:$lane)))>;
2152 def : Pat<(v4f32 (fadd (v4f32 QPR:$src1),
2153 (fmul (v4f32 QPR:$src2),
2154 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
2155 (v4f32 (VMLAslfq (v4f32 QPR:$src1),
2157 (v2f32 (EXTRACT_SUBREG QPR:$src3,
2158 (DSubReg_i32_reg imm:$lane))),
2159 (SubReg_i32_lane imm:$lane)))>;
2161 // VMLAL : Vector Multiply Accumulate Long (Q += D * D)
2162 defm VMLALs : N3VLInt3_QHS<0,1,0b1000,0, "vmlal", "s", int_arm_neon_vmlals>;
2163 defm VMLALu : N3VLInt3_QHS<1,1,0b1000,0, "vmlal", "u", int_arm_neon_vmlalu>;
2165 defm VMLALsls : N3VLInt3SL_HS<0, 0b0010, "vmlal", "s", int_arm_neon_vmlals>;
2166 defm VMLALslu : N3VLInt3SL_HS<1, 0b0010, "vmlal", "u", int_arm_neon_vmlalu>;
2168 // VQDMLAL : Vector Saturating Doubling Multiply Accumulate Long (Q += D * D)
2169 defm VQDMLAL : N3VLInt3_HS<0, 1, 0b1001, 0, "vqdmlal", "s",
2170 int_arm_neon_vqdmlal>;
2171 defm VQDMLALsl: N3VLInt3SL_HS<0, 0b0011, "vqdmlal", "s", int_arm_neon_vqdmlal>;
2173 // VMLS : Vector Multiply Subtract (integer and floating-point)
2174 defm VMLS : N3VMulOp_QHS<1, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
2175 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
2176 def VMLSfd : N3VDMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACD, "vmls", "f32",
2178 def VMLSfq : N3VQMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACQ, "vmls", "f32",
2180 defm VMLSsl : N3VMulOpSL_HS<0b0100, IIC_VMACi16D, IIC_VMACi32D,
2181 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
2182 def VMLSslfd : N3VDMulOpSL<0b10, 0b0101, IIC_VMACD, "vmls", "f32",
2184 def VMLSslfq : N3VQMulOpSL<0b10, 0b0101, IIC_VMACQ, "vmls", "f32",
2185 v4f32, v2f32, fmul, fsub>;
2187 def : Pat<(v8i16 (sub (v8i16 QPR:$src1),
2188 (mul (v8i16 QPR:$src2),
2189 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
2190 (v8i16 (VMLSslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
2191 (v4i16 (EXTRACT_SUBREG QPR:$src3,
2192 (DSubReg_i16_reg imm:$lane))),
2193 (SubReg_i16_lane imm:$lane)))>;
2195 def : Pat<(v4i32 (sub (v4i32 QPR:$src1),
2196 (mul (v4i32 QPR:$src2),
2197 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
2198 (v4i32 (VMLSslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
2199 (v2i32 (EXTRACT_SUBREG QPR:$src3,
2200 (DSubReg_i32_reg imm:$lane))),
2201 (SubReg_i32_lane imm:$lane)))>;
2203 def : Pat<(v4f32 (fsub (v4f32 QPR:$src1),
2204 (fmul (v4f32 QPR:$src2),
2205 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
2206 (v4f32 (VMLSslfq (v4f32 QPR:$src1), (v4f32 QPR:$src2),
2207 (v2f32 (EXTRACT_SUBREG QPR:$src3,
2208 (DSubReg_i32_reg imm:$lane))),
2209 (SubReg_i32_lane imm:$lane)))>;
2211 // VMLSL : Vector Multiply Subtract Long (Q -= D * D)
2212 defm VMLSLs : N3VLInt3_QHS<0,1,0b1010,0, "vmlsl", "s", int_arm_neon_vmlsls>;
2213 defm VMLSLu : N3VLInt3_QHS<1,1,0b1010,0, "vmlsl", "u", int_arm_neon_vmlslu>;
2215 defm VMLSLsls : N3VLInt3SL_HS<0, 0b0110, "vmlsl", "s", int_arm_neon_vmlsls>;
2216 defm VMLSLslu : N3VLInt3SL_HS<1, 0b0110, "vmlsl", "u", int_arm_neon_vmlslu>;
2218 // VQDMLSL : Vector Saturating Doubling Multiply Subtract Long (Q -= D * D)
2219 defm VQDMLSL : N3VLInt3_HS<0, 1, 0b1011, 0, "vqdmlsl", "s",
2220 int_arm_neon_vqdmlsl>;
2221 defm VQDMLSLsl: N3VLInt3SL_HS<0, 0b111, "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
2223 // Vector Subtract Operations.
2225 // VSUB : Vector Subtract (integer and floating-point)
2226 defm VSUB : N3V_QHSD<1, 0, 0b1000, 0, IIC_VSUBiD, IIC_VSUBiQ,
2227 "vsub", "i", sub, 0>;
2228 def VSUBfd : N3VD<0, 0, 0b10, 0b1101, 0, IIC_VBIND, "vsub", "f32",
2229 v2f32, v2f32, fsub, 0>;
2230 def VSUBfq : N3VQ<0, 0, 0b10, 0b1101, 0, IIC_VBINQ, "vsub", "f32",
2231 v4f32, v4f32, fsub, 0>;
2232 // VSUBL : Vector Subtract Long (Q = D - D)
2233 defm VSUBLs : N3VLInt_QHS<0,1,0b0010,0, IIC_VSHLiD, "vsubl", "s",
2234 int_arm_neon_vsubls, 1>;
2235 defm VSUBLu : N3VLInt_QHS<1,1,0b0010,0, IIC_VSHLiD, "vsubl", "u",
2236 int_arm_neon_vsublu, 1>;
2237 // VSUBW : Vector Subtract Wide (Q = Q - D)
2238 defm VSUBWs : N3VWInt_QHS<0,1,0b0011,0, "vsubw", "s", int_arm_neon_vsubws, 0>;
2239 defm VSUBWu : N3VWInt_QHS<1,1,0b0011,0, "vsubw", "u", int_arm_neon_vsubwu, 0>;
2240 // VHSUB : Vector Halving Subtract
2241 defm VHSUBs : N3VInt_QHS<0, 0, 0b0010, 0, IIC_VBINi4D, IIC_VBINi4D,
2242 IIC_VBINi4Q, IIC_VBINi4Q,
2243 "vhsub", "s", int_arm_neon_vhsubs, 0>;
2244 defm VHSUBu : N3VInt_QHS<1, 0, 0b0010, 0, IIC_VBINi4D, IIC_VBINi4D,
2245 IIC_VBINi4Q, IIC_VBINi4Q,
2246 "vhsub", "u", int_arm_neon_vhsubu, 0>;
2247 // VQSUB : Vector Saturing Subtract
2248 defm VQSUBs : N3VInt_QHSD<0, 0, 0b0010, 1, IIC_VBINi4D, IIC_VBINi4D,
2249 IIC_VBINi4Q, IIC_VBINi4Q,
2250 "vqsub", "s", int_arm_neon_vqsubs, 0>;
2251 defm VQSUBu : N3VInt_QHSD<1, 0, 0b0010, 1, IIC_VBINi4D, IIC_VBINi4D,
2252 IIC_VBINi4Q, IIC_VBINi4Q,
2253 "vqsub", "u", int_arm_neon_vqsubu, 0>;
2254 // VSUBHN : Vector Subtract and Narrow Returning High Half (D = Q - Q)
2255 defm VSUBHN : N3VNInt_HSD<0,1,0b0110,0, "vsubhn", "i",
2256 int_arm_neon_vsubhn, 0>;
2257 // VRSUBHN : Vector Rounding Subtract and Narrow Returning High Half (D=Q-Q)
2258 defm VRSUBHN : N3VNInt_HSD<1,1,0b0110,0, "vrsubhn", "i",
2259 int_arm_neon_vrsubhn, 0>;
2261 // Vector Comparisons.
2263 // VCEQ : Vector Compare Equal
2264 defm VCEQ : N3V_QHS<1, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2265 IIC_VBINi4Q, "vceq", "i", NEONvceq, 1>;
2266 def VCEQfd : N3VD<0,0,0b00,0b1110,0, IIC_VBIND, "vceq", "f32", v2i32, v2f32,
2268 def VCEQfq : N3VQ<0,0,0b00,0b1110,0, IIC_VBINQ, "vceq", "f32", v4i32, v4f32,
2270 // For disassembly only.
2271 defm VCEQz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00010, 0, "vceq", "i",
2274 // VCGE : Vector Compare Greater Than or Equal
2275 defm VCGEs : N3V_QHS<0, 0, 0b0011, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2276 IIC_VBINi4Q, "vcge", "s", NEONvcge, 0>;
2277 defm VCGEu : N3V_QHS<1, 0, 0b0011, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2278 IIC_VBINi4Q, "vcge", "u", NEONvcgeu, 0>;
2279 def VCGEfd : N3VD<1,0,0b00,0b1110,0, IIC_VBIND, "vcge", "f32",
2280 v2i32, v2f32, NEONvcge, 0>;
2281 def VCGEfq : N3VQ<1,0,0b00,0b1110,0, IIC_VBINQ, "vcge", "f32", v4i32, v4f32,
2283 // For disassembly only.
2284 defm VCGEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00001, 0, "vcge", "s",
2286 // For disassembly only.
2287 defm VCLEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00011, 0, "vcle", "s",
2290 // VCGT : Vector Compare Greater Than
2291 defm VCGTs : N3V_QHS<0, 0, 0b0011, 0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2292 IIC_VBINi4Q, "vcgt", "s", NEONvcgt, 0>;
2293 defm VCGTu : N3V_QHS<1, 0, 0b0011, 0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2294 IIC_VBINi4Q, "vcgt", "u", NEONvcgtu, 0>;
2295 def VCGTfd : N3VD<1,0,0b10,0b1110,0, IIC_VBIND, "vcgt", "f32", v2i32, v2f32,
2297 def VCGTfq : N3VQ<1,0,0b10,0b1110,0, IIC_VBINQ, "vcgt", "f32", v4i32, v4f32,
2299 // For disassembly only.
2300 defm VCGTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00000, 0, "vcgt", "s",
2302 // For disassembly only.
2303 defm VCLTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00100, 0, "vclt", "s",
2306 // VACGE : Vector Absolute Compare Greater Than or Equal (aka VCAGE)
2307 def VACGEd : N3VDInt<1, 0, 0b00, 0b1110, 1, IIC_VBIND, "vacge", "f32",
2308 v2i32, v2f32, int_arm_neon_vacged, 0>;
2309 def VACGEq : N3VQInt<1, 0, 0b00, 0b1110, 1, IIC_VBINQ, "vacge", "f32",
2310 v4i32, v4f32, int_arm_neon_vacgeq, 0>;
2311 // VACGT : Vector Absolute Compare Greater Than (aka VCAGT)
2312 def VACGTd : N3VDInt<1, 0, 0b10, 0b1110, 1, IIC_VBIND, "vacgt", "f32",
2313 v2i32, v2f32, int_arm_neon_vacgtd, 0>;
2314 def VACGTq : N3VQInt<1, 0, 0b10, 0b1110, 1, IIC_VBINQ, "vacgt", "f32",
2315 v4i32, v4f32, int_arm_neon_vacgtq, 0>;
2316 // VTST : Vector Test Bits
2317 defm VTST : N3V_QHS<0, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2318 IIC_VBINi4Q, "vtst", "", NEONvtst, 1>;
2320 // Vector Bitwise Operations.
2322 // VAND : Vector Bitwise AND
2323 def VANDd : N3VDX<0, 0, 0b00, 0b0001, 1, IIC_VBINiD, "vand",
2324 v2i32, v2i32, and, 1>;
2325 def VANDq : N3VQX<0, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "vand",
2326 v4i32, v4i32, and, 1>;
2328 // VEOR : Vector Bitwise Exclusive OR
2329 def VEORd : N3VDX<1, 0, 0b00, 0b0001, 1, IIC_VBINiD, "veor",
2330 v2i32, v2i32, xor, 1>;
2331 def VEORq : N3VQX<1, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "veor",
2332 v4i32, v4i32, xor, 1>;
2334 // VORR : Vector Bitwise OR
2335 def VORRd : N3VDX<0, 0, 0b10, 0b0001, 1, IIC_VBINiD, "vorr",
2336 v2i32, v2i32, or, 1>;
2337 def VORRq : N3VQX<0, 0, 0b10, 0b0001, 1, IIC_VBINiQ, "vorr",
2338 v4i32, v4i32, or, 1>;
2340 // VBIC : Vector Bitwise Bit Clear (AND NOT)
2341 def VBICd : N3VX<0, 0, 0b01, 0b0001, 0, 1, (outs DPR:$dst),
2342 (ins DPR:$src1, DPR:$src2), IIC_VBINiD,
2343 "vbic", "$dst, $src1, $src2", "",
2344 [(set DPR:$dst, (v2i32 (and DPR:$src1,
2345 (vnot_conv DPR:$src2))))]>;
2346 def VBICq : N3VX<0, 0, 0b01, 0b0001, 1, 1, (outs QPR:$dst),
2347 (ins QPR:$src1, QPR:$src2), IIC_VBINiQ,
2348 "vbic", "$dst, $src1, $src2", "",
2349 [(set QPR:$dst, (v4i32 (and QPR:$src1,
2350 (vnot_conv QPR:$src2))))]>;
2352 // VORN : Vector Bitwise OR NOT
2353 def VORNd : N3VX<0, 0, 0b11, 0b0001, 0, 1, (outs DPR:$dst),
2354 (ins DPR:$src1, DPR:$src2), IIC_VBINiD,
2355 "vorn", "$dst, $src1, $src2", "",
2356 [(set DPR:$dst, (v2i32 (or DPR:$src1,
2357 (vnot_conv DPR:$src2))))]>;
2358 def VORNq : N3VX<0, 0, 0b11, 0b0001, 1, 1, (outs QPR:$dst),
2359 (ins QPR:$src1, QPR:$src2), IIC_VBINiQ,
2360 "vorn", "$dst, $src1, $src2", "",
2361 [(set QPR:$dst, (v4i32 (or QPR:$src1,
2362 (vnot_conv QPR:$src2))))]>;
2364 // VMVN : Vector Bitwise NOT
2365 def VMVNd : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 0, 0,
2366 (outs DPR:$dst), (ins DPR:$src), IIC_VSHLiD,
2367 "vmvn", "$dst, $src", "",
2368 [(set DPR:$dst, (v2i32 (vnot DPR:$src)))]>;
2369 def VMVNq : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 1, 0,
2370 (outs QPR:$dst), (ins QPR:$src), IIC_VSHLiD,
2371 "vmvn", "$dst, $src", "",
2372 [(set QPR:$dst, (v4i32 (vnot QPR:$src)))]>;
2373 def : Pat<(v2i32 (vnot_conv DPR:$src)), (VMVNd DPR:$src)>;
2374 def : Pat<(v4i32 (vnot_conv QPR:$src)), (VMVNq QPR:$src)>;
2376 // VBSL : Vector Bitwise Select
2377 def VBSLd : N3VX<1, 0, 0b01, 0b0001, 0, 1, (outs DPR:$dst),
2378 (ins DPR:$src1, DPR:$src2, DPR:$src3), IIC_VCNTiD,
2379 "vbsl", "$dst, $src2, $src3", "$src1 = $dst",
2381 (v2i32 (or (and DPR:$src2, DPR:$src1),
2382 (and DPR:$src3, (vnot_conv DPR:$src1)))))]>;
2383 def VBSLq : N3VX<1, 0, 0b01, 0b0001, 1, 1, (outs QPR:$dst),
2384 (ins QPR:$src1, QPR:$src2, QPR:$src3), IIC_VCNTiQ,
2385 "vbsl", "$dst, $src2, $src3", "$src1 = $dst",
2387 (v4i32 (or (and QPR:$src2, QPR:$src1),
2388 (and QPR:$src3, (vnot_conv QPR:$src1)))))]>;
2390 // VBIF : Vector Bitwise Insert if False
2391 // like VBSL but with: "vbif $dst, $src3, $src1", "$src2 = $dst",
2392 def VBIFd : N3VX<1, 0, 0b11, 0b0001, 0, 1,
2393 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3),
2394 IIC_VBINiD, "vbif", "$dst, $src2, $src3", "$src1 = $dst",
2395 [/* For disassembly only; pattern left blank */]>;
2396 def VBIFq : N3VX<1, 0, 0b11, 0b0001, 1, 1,
2397 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3),
2398 IIC_VBINiQ, "vbif", "$dst, $src2, $src3", "$src1 = $dst",
2399 [/* For disassembly only; pattern left blank */]>;
2401 // VBIT : Vector Bitwise Insert if True
2402 // like VBSL but with: "vbit $dst, $src2, $src1", "$src3 = $dst",
2403 def VBITd : N3VX<1, 0, 0b10, 0b0001, 0, 1,
2404 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3),
2405 IIC_VBINiD, "vbit", "$dst, $src2, $src3", "$src1 = $dst",
2406 [/* For disassembly only; pattern left blank */]>;
2407 def VBITq : N3VX<1, 0, 0b10, 0b0001, 1, 1,
2408 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3),
2409 IIC_VBINiQ, "vbit", "$dst, $src2, $src3", "$src1 = $dst",
2410 [/* For disassembly only; pattern left blank */]>;
2412 // VBIT/VBIF are not yet implemented. The TwoAddress pass will not go looking
2413 // for equivalent operations with different register constraints; it just
2416 // Vector Absolute Differences.
2418 // VABD : Vector Absolute Difference
2419 defm VABDs : N3VInt_QHS<0, 0, 0b0111, 0, IIC_VBINi4D, IIC_VBINi4D,
2420 IIC_VBINi4Q, IIC_VBINi4Q,
2421 "vabd", "s", int_arm_neon_vabds, 0>;
2422 defm VABDu : N3VInt_QHS<1, 0, 0b0111, 0, IIC_VBINi4D, IIC_VBINi4D,
2423 IIC_VBINi4Q, IIC_VBINi4Q,
2424 "vabd", "u", int_arm_neon_vabdu, 0>;
2425 def VABDfd : N3VDInt<1, 0, 0b10, 0b1101, 0, IIC_VBIND,
2426 "vabd", "f32", v2f32, v2f32, int_arm_neon_vabds, 0>;
2427 def VABDfq : N3VQInt<1, 0, 0b10, 0b1101, 0, IIC_VBINQ,
2428 "vabd", "f32", v4f32, v4f32, int_arm_neon_vabds, 0>;
2430 // VABDL : Vector Absolute Difference Long (Q = | D - D |)
2431 defm VABDLs : N3VLInt_QHS<0,1,0b0111,0, IIC_VBINi4Q,
2432 "vabdl", "s", int_arm_neon_vabdls, 0>;
2433 defm VABDLu : N3VLInt_QHS<1,1,0b0111,0, IIC_VBINi4Q,
2434 "vabdl", "u", int_arm_neon_vabdlu, 0>;
2436 // VABA : Vector Absolute Difference and Accumulate
2437 defm VABAs : N3VInt3_QHS<0,0,0b0111,1, "vaba", "s", int_arm_neon_vabas>;
2438 defm VABAu : N3VInt3_QHS<1,0,0b0111,1, "vaba", "u", int_arm_neon_vabau>;
2440 // VABAL : Vector Absolute Difference and Accumulate Long (Q += | D - D |)
2441 defm VABALs : N3VLInt3_QHS<0,1,0b0101,0, "vabal", "s", int_arm_neon_vabals>;
2442 defm VABALu : N3VLInt3_QHS<1,1,0b0101,0, "vabal", "u", int_arm_neon_vabalu>;
2444 // Vector Maximum and Minimum.
2446 // VMAX : Vector Maximum
2447 defm VMAXs : N3VInt_QHS<0,0,0b0110,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2448 IIC_VBINi4Q, "vmax", "s", int_arm_neon_vmaxs, 1>;
2449 defm VMAXu : N3VInt_QHS<1,0,0b0110,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2450 IIC_VBINi4Q, "vmax", "u", int_arm_neon_vmaxu, 1>;
2451 def VMAXfd : N3VDInt<0, 0, 0b00, 0b1111, 0, IIC_VBIND, "vmax", "f32",
2452 v2f32, v2f32, int_arm_neon_vmaxs, 1>;
2453 def VMAXfq : N3VQInt<0, 0, 0b00, 0b1111, 0, IIC_VBINQ, "vmax", "f32",
2454 v4f32, v4f32, int_arm_neon_vmaxs, 1>;
2456 // VMIN : Vector Minimum
2457 defm VMINs : N3VInt_QHS<0,0,0b0110,1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2458 IIC_VBINi4Q, "vmin", "s", int_arm_neon_vmins, 1>;
2459 defm VMINu : N3VInt_QHS<1,0,0b0110,1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2460 IIC_VBINi4Q, "vmin", "u", int_arm_neon_vminu, 1>;
2461 def VMINfd : N3VDInt<0, 0, 0b10, 0b1111, 0, IIC_VBIND, "vmin", "f32",
2462 v2f32, v2f32, int_arm_neon_vmins, 1>;
2463 def VMINfq : N3VQInt<0, 0, 0b10, 0b1111, 0, IIC_VBINQ, "vmin", "f32",
2464 v4f32, v4f32, int_arm_neon_vmins, 1>;
2466 // Vector Pairwise Operations.
2468 // VPADD : Vector Pairwise Add
2469 def VPADDi8 : N3VDInt<0, 0, 0b00, 0b1011, 1, IIC_VBINiD, "vpadd", "i8",
2470 v8i8, v8i8, int_arm_neon_vpadd, 0>;
2471 def VPADDi16 : N3VDInt<0, 0, 0b01, 0b1011, 1, IIC_VBINiD, "vpadd", "i16",
2472 v4i16, v4i16, int_arm_neon_vpadd, 0>;
2473 def VPADDi32 : N3VDInt<0, 0, 0b10, 0b1011, 1, IIC_VBINiD, "vpadd", "i32",
2474 v2i32, v2i32, int_arm_neon_vpadd, 0>;
2475 def VPADDf : N3VDInt<1, 0, 0b00, 0b1101, 0, IIC_VBIND, "vpadd", "f32",
2476 v2f32, v2f32, int_arm_neon_vpadd, 0>;
2478 // VPADDL : Vector Pairwise Add Long
2479 defm VPADDLs : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00100, 0, "vpaddl", "s",
2480 int_arm_neon_vpaddls>;
2481 defm VPADDLu : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00101, 0, "vpaddl", "u",
2482 int_arm_neon_vpaddlu>;
2484 // VPADAL : Vector Pairwise Add and Accumulate Long
2485 defm VPADALs : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01100, 0, "vpadal", "s",
2486 int_arm_neon_vpadals>;
2487 defm VPADALu : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01101, 0, "vpadal", "u",
2488 int_arm_neon_vpadalu>;
2490 // VPMAX : Vector Pairwise Maximum
2491 def VPMAXs8 : N3VDInt<0, 0, 0b00, 0b1010, 0, IIC_VBINi4D, "vpmax", "s8",
2492 v8i8, v8i8, int_arm_neon_vpmaxs, 0>;
2493 def VPMAXs16 : N3VDInt<0, 0, 0b01, 0b1010, 0, IIC_VBINi4D, "vpmax", "s16",
2494 v4i16, v4i16, int_arm_neon_vpmaxs, 0>;
2495 def VPMAXs32 : N3VDInt<0, 0, 0b10, 0b1010, 0, IIC_VBINi4D, "vpmax", "s32",
2496 v2i32, v2i32, int_arm_neon_vpmaxs, 0>;
2497 def VPMAXu8 : N3VDInt<1, 0, 0b00, 0b1010, 0, IIC_VBINi4D, "vpmax", "u8",
2498 v8i8, v8i8, int_arm_neon_vpmaxu, 0>;
2499 def VPMAXu16 : N3VDInt<1, 0, 0b01, 0b1010, 0, IIC_VBINi4D, "vpmax", "u16",
2500 v4i16, v4i16, int_arm_neon_vpmaxu, 0>;
2501 def VPMAXu32 : N3VDInt<1, 0, 0b10, 0b1010, 0, IIC_VBINi4D, "vpmax", "u32",
2502 v2i32, v2i32, int_arm_neon_vpmaxu, 0>;
2503 def VPMAXf : N3VDInt<1, 0, 0b00, 0b1111, 0, IIC_VBINi4D, "vpmax", "f32",
2504 v2f32, v2f32, int_arm_neon_vpmaxs, 0>;
2506 // VPMIN : Vector Pairwise Minimum
2507 def VPMINs8 : N3VDInt<0, 0, 0b00, 0b1010, 1, IIC_VBINi4D, "vpmin", "s8",
2508 v8i8, v8i8, int_arm_neon_vpmins, 0>;
2509 def VPMINs16 : N3VDInt<0, 0, 0b01, 0b1010, 1, IIC_VBINi4D, "vpmin", "s16",
2510 v4i16, v4i16, int_arm_neon_vpmins, 0>;
2511 def VPMINs32 : N3VDInt<0, 0, 0b10, 0b1010, 1, IIC_VBINi4D, "vpmin", "s32",
2512 v2i32, v2i32, int_arm_neon_vpmins, 0>;
2513 def VPMINu8 : N3VDInt<1, 0, 0b00, 0b1010, 1, IIC_VBINi4D, "vpmin", "u8",
2514 v8i8, v8i8, int_arm_neon_vpminu, 0>;
2515 def VPMINu16 : N3VDInt<1, 0, 0b01, 0b1010, 1, IIC_VBINi4D, "vpmin", "u16",
2516 v4i16, v4i16, int_arm_neon_vpminu, 0>;
2517 def VPMINu32 : N3VDInt<1, 0, 0b10, 0b1010, 1, IIC_VBINi4D, "vpmin", "u32",
2518 v2i32, v2i32, int_arm_neon_vpminu, 0>;
2519 def VPMINf : N3VDInt<1, 0, 0b10, 0b1111, 0, IIC_VBINi4D, "vpmin", "f32",
2520 v2f32, v2f32, int_arm_neon_vpmins, 0>;
2522 // Vector Reciprocal and Reciprocal Square Root Estimate and Step.
2524 // VRECPE : Vector Reciprocal Estimate
2525 def VRECPEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
2526 IIC_VUNAD, "vrecpe", "u32",
2527 v2i32, v2i32, int_arm_neon_vrecpe>;
2528 def VRECPEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
2529 IIC_VUNAQ, "vrecpe", "u32",
2530 v4i32, v4i32, int_arm_neon_vrecpe>;
2531 def VRECPEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
2532 IIC_VUNAD, "vrecpe", "f32",
2533 v2f32, v2f32, int_arm_neon_vrecpe>;
2534 def VRECPEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
2535 IIC_VUNAQ, "vrecpe", "f32",
2536 v4f32, v4f32, int_arm_neon_vrecpe>;
2538 // VRECPS : Vector Reciprocal Step
2539 def VRECPSfd : N3VDInt<0, 0, 0b00, 0b1111, 1,
2540 IIC_VRECSD, "vrecps", "f32",
2541 v2f32, v2f32, int_arm_neon_vrecps, 1>;
2542 def VRECPSfq : N3VQInt<0, 0, 0b00, 0b1111, 1,
2543 IIC_VRECSQ, "vrecps", "f32",
2544 v4f32, v4f32, int_arm_neon_vrecps, 1>;
2546 // VRSQRTE : Vector Reciprocal Square Root Estimate
2547 def VRSQRTEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
2548 IIC_VUNAD, "vrsqrte", "u32",
2549 v2i32, v2i32, int_arm_neon_vrsqrte>;
2550 def VRSQRTEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
2551 IIC_VUNAQ, "vrsqrte", "u32",
2552 v4i32, v4i32, int_arm_neon_vrsqrte>;
2553 def VRSQRTEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
2554 IIC_VUNAD, "vrsqrte", "f32",
2555 v2f32, v2f32, int_arm_neon_vrsqrte>;
2556 def VRSQRTEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
2557 IIC_VUNAQ, "vrsqrte", "f32",
2558 v4f32, v4f32, int_arm_neon_vrsqrte>;
2560 // VRSQRTS : Vector Reciprocal Square Root Step
2561 def VRSQRTSfd : N3VDInt<0, 0, 0b10, 0b1111, 1,
2562 IIC_VRECSD, "vrsqrts", "f32",
2563 v2f32, v2f32, int_arm_neon_vrsqrts, 1>;
2564 def VRSQRTSfq : N3VQInt<0, 0, 0b10, 0b1111, 1,
2565 IIC_VRECSQ, "vrsqrts", "f32",
2566 v4f32, v4f32, int_arm_neon_vrsqrts, 1>;
2570 // VSHL : Vector Shift
2571 defm VSHLs : N3VInt_QHSD<0, 0, 0b0100, 0, IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ,
2572 IIC_VSHLiQ, "vshl", "s", int_arm_neon_vshifts, 0>;
2573 defm VSHLu : N3VInt_QHSD<1, 0, 0b0100, 0, IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ,
2574 IIC_VSHLiQ, "vshl", "u", int_arm_neon_vshiftu, 0>;
2575 // VSHL : Vector Shift Left (Immediate)
2576 defm VSHLi : N2VSh_QHSD<0, 1, 0b0101, 1, IIC_VSHLiD, "vshl", "i", NEONvshl>;
2577 // VSHR : Vector Shift Right (Immediate)
2578 defm VSHRs : N2VSh_QHSD<0, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "s", NEONvshrs>;
2579 defm VSHRu : N2VSh_QHSD<1, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "u", NEONvshru>;
2581 // VSHLL : Vector Shift Left Long
2582 defm VSHLLs : N2VLSh_QHS<0, 1, 0b1010, 0, 0, 1, "vshll", "s", NEONvshlls>;
2583 defm VSHLLu : N2VLSh_QHS<1, 1, 0b1010, 0, 0, 1, "vshll", "u", NEONvshllu>;
2585 // VSHLL : Vector Shift Left Long (with maximum shift count)
2586 class N2VLShMax<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
2587 bit op6, bit op4, string OpcodeStr, string Dt, ValueType ResTy,
2588 ValueType OpTy, SDNode OpNode>
2589 : N2VLSh<op24, op23, op11_8, op7, op6, op4, OpcodeStr, Dt,
2590 ResTy, OpTy, OpNode> {
2591 let Inst{21-16} = op21_16;
2593 def VSHLLi8 : N2VLShMax<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll", "i8",
2594 v8i16, v8i8, NEONvshlli>;
2595 def VSHLLi16 : N2VLShMax<1, 1, 0b110110, 0b0011, 0, 0, 0, "vshll", "i16",
2596 v4i32, v4i16, NEONvshlli>;
2597 def VSHLLi32 : N2VLShMax<1, 1, 0b111010, 0b0011, 0, 0, 0, "vshll", "i32",
2598 v2i64, v2i32, NEONvshlli>;
2600 // VSHRN : Vector Shift Right and Narrow
2601 defm VSHRN : N2VNSh_HSD<0,1,0b1000,0,0,1, IIC_VSHLiD, "vshrn", "i",
2604 // VRSHL : Vector Rounding Shift
2605 defm VRSHLs : N3VInt_QHSD<0,0,0b0101,0, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2606 IIC_VSHLi4Q, "vrshl", "s", int_arm_neon_vrshifts,0>;
2607 defm VRSHLu : N3VInt_QHSD<1,0,0b0101,0, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2608 IIC_VSHLi4Q, "vrshl", "u", int_arm_neon_vrshiftu,0>;
2609 // VRSHR : Vector Rounding Shift Right
2610 defm VRSHRs : N2VSh_QHSD<0,1,0b0010,1, IIC_VSHLi4D, "vrshr", "s", NEONvrshrs>;
2611 defm VRSHRu : N2VSh_QHSD<1,1,0b0010,1, IIC_VSHLi4D, "vrshr", "u", NEONvrshru>;
2613 // VRSHRN : Vector Rounding Shift Right and Narrow
2614 defm VRSHRN : N2VNSh_HSD<0, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vrshrn", "i",
2617 // VQSHL : Vector Saturating Shift
2618 defm VQSHLs : N3VInt_QHSD<0,0,0b0100,1, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2619 IIC_VSHLi4Q, "vqshl", "s", int_arm_neon_vqshifts,0>;
2620 defm VQSHLu : N3VInt_QHSD<1,0,0b0100,1, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2621 IIC_VSHLi4Q, "vqshl", "u", int_arm_neon_vqshiftu,0>;
2622 // VQSHL : Vector Saturating Shift Left (Immediate)
2623 defm VQSHLsi : N2VSh_QHSD<0,1,0b0111,1, IIC_VSHLi4D, "vqshl", "s", NEONvqshls>;
2624 defm VQSHLui : N2VSh_QHSD<1,1,0b0111,1, IIC_VSHLi4D, "vqshl", "u", NEONvqshlu>;
2625 // VQSHLU : Vector Saturating Shift Left (Immediate, Unsigned)
2626 defm VQSHLsu : N2VSh_QHSD<1,1,0b0110,1, IIC_VSHLi4D, "vqshlu","s",NEONvqshlsu>;
2628 // VQSHRN : Vector Saturating Shift Right and Narrow
2629 defm VQSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "s",
2631 defm VQSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "u",
2634 // VQSHRUN : Vector Saturating Shift Right and Narrow (Unsigned)
2635 defm VQSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 0, 1, IIC_VSHLi4D, "vqshrun", "s",
2638 // VQRSHL : Vector Saturating Rounding Shift
2639 defm VQRSHLs : N3VInt_QHSD<0,0,0b0101,1, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2640 IIC_VSHLi4Q, "vqrshl", "s",
2641 int_arm_neon_vqrshifts, 0>;
2642 defm VQRSHLu : N3VInt_QHSD<1,0,0b0101,1, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2643 IIC_VSHLi4Q, "vqrshl", "u",
2644 int_arm_neon_vqrshiftu, 0>;
2646 // VQRSHRN : Vector Saturating Rounding Shift Right and Narrow
2647 defm VQRSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "s",
2649 defm VQRSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "u",
2652 // VQRSHRUN : Vector Saturating Rounding Shift Right and Narrow (Unsigned)
2653 defm VQRSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vqrshrun", "s",
2656 // VSRA : Vector Shift Right and Accumulate
2657 defm VSRAs : N2VShAdd_QHSD<0, 1, 0b0001, 1, "vsra", "s", NEONvshrs>;
2658 defm VSRAu : N2VShAdd_QHSD<1, 1, 0b0001, 1, "vsra", "u", NEONvshru>;
2659 // VRSRA : Vector Rounding Shift Right and Accumulate
2660 defm VRSRAs : N2VShAdd_QHSD<0, 1, 0b0011, 1, "vrsra", "s", NEONvrshrs>;
2661 defm VRSRAu : N2VShAdd_QHSD<1, 1, 0b0011, 1, "vrsra", "u", NEONvrshru>;
2663 // VSLI : Vector Shift Left and Insert
2664 defm VSLI : N2VShIns_QHSD<1, 1, 0b0101, 1, "vsli", NEONvsli>;
2665 // VSRI : Vector Shift Right and Insert
2666 defm VSRI : N2VShIns_QHSD<1, 1, 0b0100, 1, "vsri", NEONvsri>;
2668 // Vector Absolute and Saturating Absolute.
2670 // VABS : Vector Absolute Value
2671 defm VABS : N2VInt_QHS<0b11, 0b11, 0b01, 0b00110, 0,
2672 IIC_VUNAiD, IIC_VUNAiQ, "vabs", "s",
2674 def VABSfd : N2VDInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
2675 IIC_VUNAD, "vabs", "f32",
2676 v2f32, v2f32, int_arm_neon_vabs>;
2677 def VABSfq : N2VQInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
2678 IIC_VUNAQ, "vabs", "f32",
2679 v4f32, v4f32, int_arm_neon_vabs>;
2681 // VQABS : Vector Saturating Absolute Value
2682 defm VQABS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01110, 0,
2683 IIC_VQUNAiD, IIC_VQUNAiQ, "vqabs", "s",
2684 int_arm_neon_vqabs>;
2688 def vneg : PatFrag<(ops node:$in), (sub immAllZerosV, node:$in)>;
2689 def vneg_conv : PatFrag<(ops node:$in), (sub immAllZerosV_bc, node:$in)>;
2691 class VNEGD<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
2692 : N2V<0b11, 0b11, size, 0b01, 0b00111, 0, 0, (outs DPR:$dst), (ins DPR:$src),
2693 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
2694 [(set DPR:$dst, (Ty (vneg DPR:$src)))]>;
2695 class VNEGQ<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
2696 : N2V<0b11, 0b11, size, 0b01, 0b00111, 1, 0, (outs QPR:$dst), (ins QPR:$src),
2697 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
2698 [(set QPR:$dst, (Ty (vneg QPR:$src)))]>;
2700 // VNEG : Vector Negate
2701 def VNEGs8d : VNEGD<0b00, "vneg", "s8", v8i8>;
2702 def VNEGs16d : VNEGD<0b01, "vneg", "s16", v4i16>;
2703 def VNEGs32d : VNEGD<0b10, "vneg", "s32", v2i32>;
2704 def VNEGs8q : VNEGQ<0b00, "vneg", "s8", v16i8>;
2705 def VNEGs16q : VNEGQ<0b01, "vneg", "s16", v8i16>;
2706 def VNEGs32q : VNEGQ<0b10, "vneg", "s32", v4i32>;
2708 // VNEG : Vector Negate (floating-point)
2709 def VNEGfd : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
2710 (outs DPR:$dst), (ins DPR:$src), IIC_VUNAD,
2711 "vneg", "f32", "$dst, $src", "",
2712 [(set DPR:$dst, (v2f32 (fneg DPR:$src)))]>;
2713 def VNEGf32q : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 1, 0,
2714 (outs QPR:$dst), (ins QPR:$src), IIC_VUNAQ,
2715 "vneg", "f32", "$dst, $src", "",
2716 [(set QPR:$dst, (v4f32 (fneg QPR:$src)))]>;
2718 def : Pat<(v8i8 (vneg_conv DPR:$src)), (VNEGs8d DPR:$src)>;
2719 def : Pat<(v4i16 (vneg_conv DPR:$src)), (VNEGs16d DPR:$src)>;
2720 def : Pat<(v2i32 (vneg_conv DPR:$src)), (VNEGs32d DPR:$src)>;
2721 def : Pat<(v16i8 (vneg_conv QPR:$src)), (VNEGs8q QPR:$src)>;
2722 def : Pat<(v8i16 (vneg_conv QPR:$src)), (VNEGs16q QPR:$src)>;
2723 def : Pat<(v4i32 (vneg_conv QPR:$src)), (VNEGs32q QPR:$src)>;
2725 // VQNEG : Vector Saturating Negate
2726 defm VQNEG : N2VInt_QHS<0b11, 0b11, 0b00, 0b01111, 0,
2727 IIC_VQUNAiD, IIC_VQUNAiQ, "vqneg", "s",
2728 int_arm_neon_vqneg>;
2730 // Vector Bit Counting Operations.
2732 // VCLS : Vector Count Leading Sign Bits
2733 defm VCLS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01000, 0,
2734 IIC_VCNTiD, IIC_VCNTiQ, "vcls", "s",
2736 // VCLZ : Vector Count Leading Zeros
2737 defm VCLZ : N2VInt_QHS<0b11, 0b11, 0b00, 0b01001, 0,
2738 IIC_VCNTiD, IIC_VCNTiQ, "vclz", "i",
2740 // VCNT : Vector Count One Bits
2741 def VCNTd : N2VDInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
2742 IIC_VCNTiD, "vcnt", "8",
2743 v8i8, v8i8, int_arm_neon_vcnt>;
2744 def VCNTq : N2VQInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
2745 IIC_VCNTiQ, "vcnt", "8",
2746 v16i8, v16i8, int_arm_neon_vcnt>;
2748 // Vector Swap -- for disassembly only.
2749 def VSWPd : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 0, 0,
2750 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
2751 "vswp", "$dst, $src", "", []>;
2752 def VSWPq : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 1, 0,
2753 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
2754 "vswp", "$dst, $src", "", []>;
2756 // Vector Move Operations.
2758 // VMOV : Vector Move (Register)
2760 def VMOVDneon: N3VX<0, 0, 0b10, 0b0001, 0, 1, (outs DPR:$dst), (ins DPR:$src),
2761 IIC_VMOVD, "vmov", "$dst, $src", "", []>;
2762 def VMOVQ : N3VX<0, 0, 0b10, 0b0001, 1, 1, (outs QPR:$dst), (ins QPR:$src),
2763 IIC_VMOVD, "vmov", "$dst, $src", "", []>;
2765 // VMOV : Vector Move (Immediate)
2767 // VMOV_get_imm8 xform function: convert build_vector to VMOV.i8 imm.
2768 def VMOV_get_imm8 : SDNodeXForm<build_vector, [{
2769 return ARM::getVMOVImm(N, 1, *CurDAG);
2771 def vmovImm8 : PatLeaf<(build_vector), [{
2772 return ARM::getVMOVImm(N, 1, *CurDAG).getNode() != 0;
2775 // VMOV_get_imm16 xform function: convert build_vector to VMOV.i16 imm.
2776 def VMOV_get_imm16 : SDNodeXForm<build_vector, [{
2777 return ARM::getVMOVImm(N, 2, *CurDAG);
2779 def vmovImm16 : PatLeaf<(build_vector), [{
2780 return ARM::getVMOVImm(N, 2, *CurDAG).getNode() != 0;
2781 }], VMOV_get_imm16>;
2783 // VMOV_get_imm32 xform function: convert build_vector to VMOV.i32 imm.
2784 def VMOV_get_imm32 : SDNodeXForm<build_vector, [{
2785 return ARM::getVMOVImm(N, 4, *CurDAG);
2787 def vmovImm32 : PatLeaf<(build_vector), [{
2788 return ARM::getVMOVImm(N, 4, *CurDAG).getNode() != 0;
2789 }], VMOV_get_imm32>;
2791 // VMOV_get_imm64 xform function: convert build_vector to VMOV.i64 imm.
2792 def VMOV_get_imm64 : SDNodeXForm<build_vector, [{
2793 return ARM::getVMOVImm(N, 8, *CurDAG);
2795 def vmovImm64 : PatLeaf<(build_vector), [{
2796 return ARM::getVMOVImm(N, 8, *CurDAG).getNode() != 0;
2797 }], VMOV_get_imm64>;
2799 // Note: Some of the cmode bits in the following VMOV instructions need to
2800 // be encoded based on the immed values.
2802 def VMOVv8i8 : N1ModImm<1, 0b000, 0b1110, 0, 0, 0, 1, (outs DPR:$dst),
2803 (ins h8imm:$SIMM), IIC_VMOVImm,
2804 "vmov", "i8", "$dst, $SIMM", "",
2805 [(set DPR:$dst, (v8i8 vmovImm8:$SIMM))]>;
2806 def VMOVv16i8 : N1ModImm<1, 0b000, 0b1110, 0, 1, 0, 1, (outs QPR:$dst),
2807 (ins h8imm:$SIMM), IIC_VMOVImm,
2808 "vmov", "i8", "$dst, $SIMM", "",
2809 [(set QPR:$dst, (v16i8 vmovImm8:$SIMM))]>;
2811 def VMOVv4i16 : N1ModImm<1, 0b000, {1,0,?,?}, 0, 0, {?}, 1, (outs DPR:$dst),
2812 (ins h16imm:$SIMM), IIC_VMOVImm,
2813 "vmov", "i16", "$dst, $SIMM", "",
2814 [(set DPR:$dst, (v4i16 vmovImm16:$SIMM))]>;
2815 def VMOVv8i16 : N1ModImm<1, 0b000, {1,0,?,?}, 0, 1, {?}, 1, (outs QPR:$dst),
2816 (ins h16imm:$SIMM), IIC_VMOVImm,
2817 "vmov", "i16", "$dst, $SIMM", "",
2818 [(set QPR:$dst, (v8i16 vmovImm16:$SIMM))]>;
2820 def VMOVv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, {?}, 1, (outs DPR:$dst),
2821 (ins h32imm:$SIMM), IIC_VMOVImm,
2822 "vmov", "i32", "$dst, $SIMM", "",
2823 [(set DPR:$dst, (v2i32 vmovImm32:$SIMM))]>;
2824 def VMOVv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, {?}, 1, (outs QPR:$dst),
2825 (ins h32imm:$SIMM), IIC_VMOVImm,
2826 "vmov", "i32", "$dst, $SIMM", "",
2827 [(set QPR:$dst, (v4i32 vmovImm32:$SIMM))]>;
2829 def VMOVv1i64 : N1ModImm<1, 0b000, 0b1110, 0, 0, 1, 1, (outs DPR:$dst),
2830 (ins h64imm:$SIMM), IIC_VMOVImm,
2831 "vmov", "i64", "$dst, $SIMM", "",
2832 [(set DPR:$dst, (v1i64 vmovImm64:$SIMM))]>;
2833 def VMOVv2i64 : N1ModImm<1, 0b000, 0b1110, 0, 1, 1, 1, (outs QPR:$dst),
2834 (ins h64imm:$SIMM), IIC_VMOVImm,
2835 "vmov", "i64", "$dst, $SIMM", "",
2836 [(set QPR:$dst, (v2i64 vmovImm64:$SIMM))]>;
2838 // VMOV : Vector Get Lane (move scalar to ARM core register)
2840 def VGETLNs8 : NVGetLane<{1,1,1,0,0,1,?,1}, 0b1011, {?,?},
2841 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2842 IIC_VMOVSI, "vmov", "s8", "$dst, $src[$lane]",
2843 [(set GPR:$dst, (NEONvgetlanes (v8i8 DPR:$src),
2845 def VGETLNs16 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, {?,1},
2846 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2847 IIC_VMOVSI, "vmov", "s16", "$dst, $src[$lane]",
2848 [(set GPR:$dst, (NEONvgetlanes (v4i16 DPR:$src),
2850 def VGETLNu8 : NVGetLane<{1,1,1,0,1,1,?,1}, 0b1011, {?,?},
2851 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2852 IIC_VMOVSI, "vmov", "u8", "$dst, $src[$lane]",
2853 [(set GPR:$dst, (NEONvgetlaneu (v8i8 DPR:$src),
2855 def VGETLNu16 : NVGetLane<{1,1,1,0,1,0,?,1}, 0b1011, {?,1},
2856 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2857 IIC_VMOVSI, "vmov", "u16", "$dst, $src[$lane]",
2858 [(set GPR:$dst, (NEONvgetlaneu (v4i16 DPR:$src),
2860 def VGETLNi32 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, 0b00,
2861 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2862 IIC_VMOVSI, "vmov", "32", "$dst, $src[$lane]",
2863 [(set GPR:$dst, (extractelt (v2i32 DPR:$src),
2865 // def VGETLNf32: see FMRDH and FMRDL in ARMInstrVFP.td
2866 def : Pat<(NEONvgetlanes (v16i8 QPR:$src), imm:$lane),
2867 (VGETLNs8 (v8i8 (EXTRACT_SUBREG QPR:$src,
2868 (DSubReg_i8_reg imm:$lane))),
2869 (SubReg_i8_lane imm:$lane))>;
2870 def : Pat<(NEONvgetlanes (v8i16 QPR:$src), imm:$lane),
2871 (VGETLNs16 (v4i16 (EXTRACT_SUBREG QPR:$src,
2872 (DSubReg_i16_reg imm:$lane))),
2873 (SubReg_i16_lane imm:$lane))>;
2874 def : Pat<(NEONvgetlaneu (v16i8 QPR:$src), imm:$lane),
2875 (VGETLNu8 (v8i8 (EXTRACT_SUBREG QPR:$src,
2876 (DSubReg_i8_reg imm:$lane))),
2877 (SubReg_i8_lane imm:$lane))>;
2878 def : Pat<(NEONvgetlaneu (v8i16 QPR:$src), imm:$lane),
2879 (VGETLNu16 (v4i16 (EXTRACT_SUBREG QPR:$src,
2880 (DSubReg_i16_reg imm:$lane))),
2881 (SubReg_i16_lane imm:$lane))>;
2882 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
2883 (VGETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src,
2884 (DSubReg_i32_reg imm:$lane))),
2885 (SubReg_i32_lane imm:$lane))>;
2886 def : Pat<(extractelt (v2f32 DPR:$src1), imm:$src2),
2887 (EXTRACT_SUBREG (v2f32 (COPY_TO_REGCLASS (v2f32 DPR:$src1),DPR_VFP2)),
2888 (SSubReg_f32_reg imm:$src2))>;
2889 def : Pat<(extractelt (v4f32 QPR:$src1), imm:$src2),
2890 (EXTRACT_SUBREG (v4f32 (COPY_TO_REGCLASS (v4f32 QPR:$src1),QPR_VFP2)),
2891 (SSubReg_f32_reg imm:$src2))>;
2892 //def : Pat<(extractelt (v2i64 QPR:$src1), imm:$src2),
2893 // (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
2894 def : Pat<(extractelt (v2f64 QPR:$src1), imm:$src2),
2895 (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
2898 // VMOV : Vector Set Lane (move ARM core register to scalar)
2900 let Constraints = "$src1 = $dst" in {
2901 def VSETLNi8 : NVSetLane<{1,1,1,0,0,1,?,0}, 0b1011, {?,?}, (outs DPR:$dst),
2902 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
2903 IIC_VMOVISL, "vmov", "8", "$dst[$lane], $src2",
2904 [(set DPR:$dst, (vector_insert (v8i8 DPR:$src1),
2905 GPR:$src2, imm:$lane))]>;
2906 def VSETLNi16 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, {?,1}, (outs DPR:$dst),
2907 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
2908 IIC_VMOVISL, "vmov", "16", "$dst[$lane], $src2",
2909 [(set DPR:$dst, (vector_insert (v4i16 DPR:$src1),
2910 GPR:$src2, imm:$lane))]>;
2911 def VSETLNi32 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, 0b00, (outs DPR:$dst),
2912 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
2913 IIC_VMOVISL, "vmov", "32", "$dst[$lane], $src2",
2914 [(set DPR:$dst, (insertelt (v2i32 DPR:$src1),
2915 GPR:$src2, imm:$lane))]>;
2917 def : Pat<(vector_insert (v16i8 QPR:$src1), GPR:$src2, imm:$lane),
2918 (v16i8 (INSERT_SUBREG QPR:$src1,
2919 (v8i8 (VSETLNi8 (v8i8 (EXTRACT_SUBREG QPR:$src1,
2920 (DSubReg_i8_reg imm:$lane))),
2921 GPR:$src2, (SubReg_i8_lane imm:$lane))),
2922 (DSubReg_i8_reg imm:$lane)))>;
2923 def : Pat<(vector_insert (v8i16 QPR:$src1), GPR:$src2, imm:$lane),
2924 (v8i16 (INSERT_SUBREG QPR:$src1,
2925 (v4i16 (VSETLNi16 (v4i16 (EXTRACT_SUBREG QPR:$src1,
2926 (DSubReg_i16_reg imm:$lane))),
2927 GPR:$src2, (SubReg_i16_lane imm:$lane))),
2928 (DSubReg_i16_reg imm:$lane)))>;
2929 def : Pat<(insertelt (v4i32 QPR:$src1), GPR:$src2, imm:$lane),
2930 (v4i32 (INSERT_SUBREG QPR:$src1,
2931 (v2i32 (VSETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src1,
2932 (DSubReg_i32_reg imm:$lane))),
2933 GPR:$src2, (SubReg_i32_lane imm:$lane))),
2934 (DSubReg_i32_reg imm:$lane)))>;
2936 def : Pat<(v2f32 (insertelt DPR:$src1, SPR:$src2, imm:$src3)),
2937 (INSERT_SUBREG (v2f32 (COPY_TO_REGCLASS DPR:$src1, DPR_VFP2)),
2938 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
2939 def : Pat<(v4f32 (insertelt QPR:$src1, SPR:$src2, imm:$src3)),
2940 (INSERT_SUBREG (v4f32 (COPY_TO_REGCLASS QPR:$src1, QPR_VFP2)),
2941 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
2943 //def : Pat<(v2i64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
2944 // (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
2945 def : Pat<(v2f64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
2946 (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
2948 def : Pat<(v2f32 (scalar_to_vector SPR:$src)),
2949 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$src, arm_ssubreg_0)>;
2950 def : Pat<(v2f64 (scalar_to_vector (f64 DPR:$src))),
2951 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), DPR:$src, arm_dsubreg_0)>;
2952 def : Pat<(v4f32 (scalar_to_vector SPR:$src)),
2953 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), SPR:$src, arm_ssubreg_0)>;
2955 def : Pat<(v8i8 (scalar_to_vector GPR:$src)),
2956 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
2957 def : Pat<(v4i16 (scalar_to_vector GPR:$src)),
2958 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
2959 def : Pat<(v2i32 (scalar_to_vector GPR:$src)),
2960 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
2962 def : Pat<(v16i8 (scalar_to_vector GPR:$src)),
2963 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
2964 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
2966 def : Pat<(v8i16 (scalar_to_vector GPR:$src)),
2967 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
2968 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
2970 def : Pat<(v4i32 (scalar_to_vector GPR:$src)),
2971 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
2972 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
2975 // VDUP : Vector Duplicate (from ARM core register to all elements)
2977 class VDUPD<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
2978 : NVDup<opcod1, 0b1011, opcod3, (outs DPR:$dst), (ins GPR:$src),
2979 IIC_VMOVIS, "vdup", Dt, "$dst, $src",
2980 [(set DPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
2981 class VDUPQ<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
2982 : NVDup<opcod1, 0b1011, opcod3, (outs QPR:$dst), (ins GPR:$src),
2983 IIC_VMOVIS, "vdup", Dt, "$dst, $src",
2984 [(set QPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
2986 def VDUP8d : VDUPD<0b11101100, 0b00, "8", v8i8>;
2987 def VDUP16d : VDUPD<0b11101000, 0b01, "16", v4i16>;
2988 def VDUP32d : VDUPD<0b11101000, 0b00, "32", v2i32>;
2989 def VDUP8q : VDUPQ<0b11101110, 0b00, "8", v16i8>;
2990 def VDUP16q : VDUPQ<0b11101010, 0b01, "16", v8i16>;
2991 def VDUP32q : VDUPQ<0b11101010, 0b00, "32", v4i32>;
2993 def VDUPfd : NVDup<0b11101000, 0b1011, 0b00, (outs DPR:$dst), (ins GPR:$src),
2994 IIC_VMOVIS, "vdup", "32", "$dst, $src",
2995 [(set DPR:$dst, (v2f32 (NEONvdup
2996 (f32 (bitconvert GPR:$src)))))]>;
2997 def VDUPfq : NVDup<0b11101010, 0b1011, 0b00, (outs QPR:$dst), (ins GPR:$src),
2998 IIC_VMOVIS, "vdup", "32", "$dst, $src",
2999 [(set QPR:$dst, (v4f32 (NEONvdup
3000 (f32 (bitconvert GPR:$src)))))]>;
3002 // VDUP : Vector Duplicate Lane (from scalar to all elements)
3004 class VDUPLND<bits<2> op19_18, bits<2> op17_16,
3005 string OpcodeStr, string Dt, ValueType Ty>
3006 : N2V<0b11, 0b11, op19_18, op17_16, 0b11000, 0, 0,
3007 (outs DPR:$dst), (ins DPR:$src, nohash_imm:$lane), IIC_VMOVD,
3008 OpcodeStr, Dt, "$dst, $src[$lane]", "",
3009 [(set DPR:$dst, (Ty (NEONvduplane (Ty DPR:$src), imm:$lane)))]>;
3011 class VDUPLNQ<bits<2> op19_18, bits<2> op17_16, string OpcodeStr, string Dt,
3012 ValueType ResTy, ValueType OpTy>
3013 : N2V<0b11, 0b11, op19_18, op17_16, 0b11000, 1, 0,
3014 (outs QPR:$dst), (ins DPR:$src, nohash_imm:$lane), IIC_VMOVD,
3015 OpcodeStr, Dt, "$dst, $src[$lane]", "",
3016 [(set QPR:$dst, (ResTy (NEONvduplane (OpTy DPR:$src), imm:$lane)))]>;
3018 // Inst{19-16} is partially specified depending on the element size.
3020 def VDUPLN8d : VDUPLND<{?,?}, {?,1}, "vdup", "8", v8i8>;
3021 def VDUPLN16d : VDUPLND<{?,?}, {1,0}, "vdup", "16", v4i16>;
3022 def VDUPLN32d : VDUPLND<{?,1}, {0,0}, "vdup", "32", v2i32>;
3023 def VDUPLNfd : VDUPLND<{?,1}, {0,0}, "vdup", "32", v2f32>;
3024 def VDUPLN8q : VDUPLNQ<{?,?}, {?,1}, "vdup", "8", v16i8, v8i8>;
3025 def VDUPLN16q : VDUPLNQ<{?,?}, {1,0}, "vdup", "16", v8i16, v4i16>;
3026 def VDUPLN32q : VDUPLNQ<{?,1}, {0,0}, "vdup", "32", v4i32, v2i32>;
3027 def VDUPLNfq : VDUPLNQ<{?,1}, {0,0}, "vdup", "32", v4f32, v2f32>;
3029 def : Pat<(v16i8 (NEONvduplane (v16i8 QPR:$src), imm:$lane)),
3030 (v16i8 (VDUPLN8q (v8i8 (EXTRACT_SUBREG QPR:$src,
3031 (DSubReg_i8_reg imm:$lane))),
3032 (SubReg_i8_lane imm:$lane)))>;
3033 def : Pat<(v8i16 (NEONvduplane (v8i16 QPR:$src), imm:$lane)),
3034 (v8i16 (VDUPLN16q (v4i16 (EXTRACT_SUBREG QPR:$src,
3035 (DSubReg_i16_reg imm:$lane))),
3036 (SubReg_i16_lane imm:$lane)))>;
3037 def : Pat<(v4i32 (NEONvduplane (v4i32 QPR:$src), imm:$lane)),
3038 (v4i32 (VDUPLN32q (v2i32 (EXTRACT_SUBREG QPR:$src,
3039 (DSubReg_i32_reg imm:$lane))),
3040 (SubReg_i32_lane imm:$lane)))>;
3041 def : Pat<(v4f32 (NEONvduplane (v4f32 QPR:$src), imm:$lane)),
3042 (v4f32 (VDUPLNfq (v2f32 (EXTRACT_SUBREG QPR:$src,
3043 (DSubReg_i32_reg imm:$lane))),
3044 (SubReg_i32_lane imm:$lane)))>;
3046 def VDUPfdf : N2V<0b11, 0b11, {?,1}, {0,0}, 0b11000, 0, 0,
3047 (outs DPR:$dst), (ins SPR:$src),
3048 IIC_VMOVD, "vdup", "32", "$dst, ${src:lane}", "",
3049 [(set DPR:$dst, (v2f32 (NEONvdup (f32 SPR:$src))))]>;
3051 def VDUPfqf : N2V<0b11, 0b11, {?,1}, {0,0}, 0b11000, 1, 0,
3052 (outs QPR:$dst), (ins SPR:$src),
3053 IIC_VMOVD, "vdup", "32", "$dst, ${src:lane}", "",
3054 [(set QPR:$dst, (v4f32 (NEONvdup (f32 SPR:$src))))]>;
3056 def : Pat<(v2i64 (NEONvduplane (v2i64 QPR:$src), imm:$lane)),
3057 (INSERT_SUBREG QPR:$src,
3058 (i64 (EXTRACT_SUBREG QPR:$src,
3059 (DSubReg_f64_reg imm:$lane))),
3060 (DSubReg_f64_other_reg imm:$lane))>;
3061 def : Pat<(v2f64 (NEONvduplane (v2f64 QPR:$src), imm:$lane)),
3062 (INSERT_SUBREG QPR:$src,
3063 (f64 (EXTRACT_SUBREG QPR:$src,
3064 (DSubReg_f64_reg imm:$lane))),
3065 (DSubReg_f64_other_reg imm:$lane))>;
3067 // VMOVN : Vector Narrowing Move
3068 defm VMOVN : N2VNInt_HSD<0b11,0b11,0b10,0b00100,0,0, IIC_VMOVD,
3069 "vmovn", "i", int_arm_neon_vmovn>;
3070 // VQMOVN : Vector Saturating Narrowing Move
3071 defm VQMOVNs : N2VNInt_HSD<0b11,0b11,0b10,0b00101,0,0, IIC_VQUNAiD,
3072 "vqmovn", "s", int_arm_neon_vqmovns>;
3073 defm VQMOVNu : N2VNInt_HSD<0b11,0b11,0b10,0b00101,1,0, IIC_VQUNAiD,
3074 "vqmovn", "u", int_arm_neon_vqmovnu>;
3075 defm VQMOVNsu : N2VNInt_HSD<0b11,0b11,0b10,0b00100,1,0, IIC_VQUNAiD,
3076 "vqmovun", "s", int_arm_neon_vqmovnsu>;
3077 // VMOVL : Vector Lengthening Move
3078 defm VMOVLs : N2VLInt_QHS<0b01,0b10100,0,1, "vmovl", "s",
3079 int_arm_neon_vmovls>;
3080 defm VMOVLu : N2VLInt_QHS<0b11,0b10100,0,1, "vmovl", "u",
3081 int_arm_neon_vmovlu>;
3083 // Vector Conversions.
3085 // VCVT : Vector Convert Between Floating-Point and Integers
3086 def VCVTf2sd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
3087 v2i32, v2f32, fp_to_sint>;
3088 def VCVTf2ud : N2VD<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
3089 v2i32, v2f32, fp_to_uint>;
3090 def VCVTs2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
3091 v2f32, v2i32, sint_to_fp>;
3092 def VCVTu2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
3093 v2f32, v2i32, uint_to_fp>;
3095 def VCVTf2sq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
3096 v4i32, v4f32, fp_to_sint>;
3097 def VCVTf2uq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
3098 v4i32, v4f32, fp_to_uint>;
3099 def VCVTs2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
3100 v4f32, v4i32, sint_to_fp>;
3101 def VCVTu2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
3102 v4f32, v4i32, uint_to_fp>;
3104 // VCVT : Vector Convert Between Floating-Point and Fixed-Point.
3105 def VCVTf2xsd : N2VCvtD<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
3106 v2i32, v2f32, int_arm_neon_vcvtfp2fxs>;
3107 def VCVTf2xud : N2VCvtD<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
3108 v2i32, v2f32, int_arm_neon_vcvtfp2fxu>;
3109 def VCVTxs2fd : N2VCvtD<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
3110 v2f32, v2i32, int_arm_neon_vcvtfxs2fp>;
3111 def VCVTxu2fd : N2VCvtD<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
3112 v2f32, v2i32, int_arm_neon_vcvtfxu2fp>;
3114 def VCVTf2xsq : N2VCvtQ<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
3115 v4i32, v4f32, int_arm_neon_vcvtfp2fxs>;
3116 def VCVTf2xuq : N2VCvtQ<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
3117 v4i32, v4f32, int_arm_neon_vcvtfp2fxu>;
3118 def VCVTxs2fq : N2VCvtQ<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
3119 v4f32, v4i32, int_arm_neon_vcvtfxs2fp>;
3120 def VCVTxu2fq : N2VCvtQ<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
3121 v4f32, v4i32, int_arm_neon_vcvtfxu2fp>;
3125 // VREV64 : Vector Reverse elements within 64-bit doublewords
3127 class VREV64D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3128 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 0, 0, (outs DPR:$dst),
3129 (ins DPR:$src), IIC_VMOVD,
3130 OpcodeStr, Dt, "$dst, $src", "",
3131 [(set DPR:$dst, (Ty (NEONvrev64 (Ty DPR:$src))))]>;
3132 class VREV64Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3133 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 1, 0, (outs QPR:$dst),
3134 (ins QPR:$src), IIC_VMOVD,
3135 OpcodeStr, Dt, "$dst, $src", "",
3136 [(set QPR:$dst, (Ty (NEONvrev64 (Ty QPR:$src))))]>;
3138 def VREV64d8 : VREV64D<0b00, "vrev64", "8", v8i8>;
3139 def VREV64d16 : VREV64D<0b01, "vrev64", "16", v4i16>;
3140 def VREV64d32 : VREV64D<0b10, "vrev64", "32", v2i32>;
3141 def VREV64df : VREV64D<0b10, "vrev64", "32", v2f32>;
3143 def VREV64q8 : VREV64Q<0b00, "vrev64", "8", v16i8>;
3144 def VREV64q16 : VREV64Q<0b01, "vrev64", "16", v8i16>;
3145 def VREV64q32 : VREV64Q<0b10, "vrev64", "32", v4i32>;
3146 def VREV64qf : VREV64Q<0b10, "vrev64", "32", v4f32>;
3148 // VREV32 : Vector Reverse elements within 32-bit words
3150 class VREV32D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3151 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 0, 0, (outs DPR:$dst),
3152 (ins DPR:$src), IIC_VMOVD,
3153 OpcodeStr, Dt, "$dst, $src", "",
3154 [(set DPR:$dst, (Ty (NEONvrev32 (Ty DPR:$src))))]>;
3155 class VREV32Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3156 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 1, 0, (outs QPR:$dst),
3157 (ins QPR:$src), IIC_VMOVD,
3158 OpcodeStr, Dt, "$dst, $src", "",
3159 [(set QPR:$dst, (Ty (NEONvrev32 (Ty QPR:$src))))]>;
3161 def VREV32d8 : VREV32D<0b00, "vrev32", "8", v8i8>;
3162 def VREV32d16 : VREV32D<0b01, "vrev32", "16", v4i16>;
3164 def VREV32q8 : VREV32Q<0b00, "vrev32", "8", v16i8>;
3165 def VREV32q16 : VREV32Q<0b01, "vrev32", "16", v8i16>;
3167 // VREV16 : Vector Reverse elements within 16-bit halfwords
3169 class VREV16D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3170 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 0, 0, (outs DPR:$dst),
3171 (ins DPR:$src), IIC_VMOVD,
3172 OpcodeStr, Dt, "$dst, $src", "",
3173 [(set DPR:$dst, (Ty (NEONvrev16 (Ty DPR:$src))))]>;
3174 class VREV16Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3175 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 1, 0, (outs QPR:$dst),
3176 (ins QPR:$src), IIC_VMOVD,
3177 OpcodeStr, Dt, "$dst, $src", "",
3178 [(set QPR:$dst, (Ty (NEONvrev16 (Ty QPR:$src))))]>;
3180 def VREV16d8 : VREV16D<0b00, "vrev16", "8", v8i8>;
3181 def VREV16q8 : VREV16Q<0b00, "vrev16", "8", v16i8>;
3183 // Other Vector Shuffles.
3185 // VEXT : Vector Extract
3187 class VEXTd<string OpcodeStr, string Dt, ValueType Ty>
3188 : N3V<0,1,0b11,{?,?,?,?},0,0, (outs DPR:$dst),
3189 (ins DPR:$lhs, DPR:$rhs, i32imm:$index), IIC_VEXTD,
3190 OpcodeStr, Dt, "$dst, $lhs, $rhs, $index", "",
3191 [(set DPR:$dst, (Ty (NEONvext (Ty DPR:$lhs),
3192 (Ty DPR:$rhs), imm:$index)))]>;
3194 class VEXTq<string OpcodeStr, string Dt, ValueType Ty>
3195 : N3V<0,1,0b11,{?,?,?,?},1,0, (outs QPR:$dst),
3196 (ins QPR:$lhs, QPR:$rhs, i32imm:$index), IIC_VEXTQ,
3197 OpcodeStr, Dt, "$dst, $lhs, $rhs, $index", "",
3198 [(set QPR:$dst, (Ty (NEONvext (Ty QPR:$lhs),
3199 (Ty QPR:$rhs), imm:$index)))]>;
3201 def VEXTd8 : VEXTd<"vext", "8", v8i8>;
3202 def VEXTd16 : VEXTd<"vext", "16", v4i16>;
3203 def VEXTd32 : VEXTd<"vext", "32", v2i32>;
3204 def VEXTdf : VEXTd<"vext", "32", v2f32>;
3206 def VEXTq8 : VEXTq<"vext", "8", v16i8>;
3207 def VEXTq16 : VEXTq<"vext", "16", v8i16>;
3208 def VEXTq32 : VEXTq<"vext", "32", v4i32>;
3209 def VEXTqf : VEXTq<"vext", "32", v4f32>;
3211 // VTRN : Vector Transpose
3213 def VTRNd8 : N2VDShuffle<0b00, 0b00001, "vtrn", "8">;
3214 def VTRNd16 : N2VDShuffle<0b01, 0b00001, "vtrn", "16">;
3215 def VTRNd32 : N2VDShuffle<0b10, 0b00001, "vtrn", "32">;
3217 def VTRNq8 : N2VQShuffle<0b00, 0b00001, IIC_VPERMQ, "vtrn", "8">;
3218 def VTRNq16 : N2VQShuffle<0b01, 0b00001, IIC_VPERMQ, "vtrn", "16">;
3219 def VTRNq32 : N2VQShuffle<0b10, 0b00001, IIC_VPERMQ, "vtrn", "32">;
3221 // VUZP : Vector Unzip (Deinterleave)
3223 def VUZPd8 : N2VDShuffle<0b00, 0b00010, "vuzp", "8">;
3224 def VUZPd16 : N2VDShuffle<0b01, 0b00010, "vuzp", "16">;
3225 def VUZPd32 : N2VDShuffle<0b10, 0b00010, "vuzp", "32">;
3227 def VUZPq8 : N2VQShuffle<0b00, 0b00010, IIC_VPERMQ3, "vuzp", "8">;
3228 def VUZPq16 : N2VQShuffle<0b01, 0b00010, IIC_VPERMQ3, "vuzp", "16">;
3229 def VUZPq32 : N2VQShuffle<0b10, 0b00010, IIC_VPERMQ3, "vuzp", "32">;
3231 // VZIP : Vector Zip (Interleave)
3233 def VZIPd8 : N2VDShuffle<0b00, 0b00011, "vzip", "8">;
3234 def VZIPd16 : N2VDShuffle<0b01, 0b00011, "vzip", "16">;
3235 def VZIPd32 : N2VDShuffle<0b10, 0b00011, "vzip", "32">;
3237 def VZIPq8 : N2VQShuffle<0b00, 0b00011, IIC_VPERMQ3, "vzip", "8">;
3238 def VZIPq16 : N2VQShuffle<0b01, 0b00011, IIC_VPERMQ3, "vzip", "16">;
3239 def VZIPq32 : N2VQShuffle<0b10, 0b00011, IIC_VPERMQ3, "vzip", "32">;
3241 // Vector Table Lookup and Table Extension.
3243 // VTBL : Vector Table Lookup
3245 : N3V<1,1,0b11,0b1000,0,0, (outs DPR:$dst),
3246 (ins DPR:$tbl1, DPR:$src), IIC_VTB1,
3247 "vtbl", "8", "$dst, \\{$tbl1\\}, $src", "",
3248 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl1 DPR:$tbl1, DPR:$src)))]>;
3249 let hasExtraSrcRegAllocReq = 1 in {
3251 : N3V<1,1,0b11,0b1001,0,0, (outs DPR:$dst),
3252 (ins DPR:$tbl1, DPR:$tbl2, DPR:$src), IIC_VTB2,
3253 "vtbl", "8", "$dst, \\{$tbl1, $tbl2\\}, $src", "",
3254 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl2
3255 DPR:$tbl1, DPR:$tbl2, DPR:$src)))]>;
3257 : N3V<1,1,0b11,0b1010,0,0, (outs DPR:$dst),
3258 (ins DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$src), IIC_VTB3,
3259 "vtbl", "8", "$dst, \\{$tbl1, $tbl2, $tbl3\\}, $src", "",
3260 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl3
3261 DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$src)))]>;
3263 : N3V<1,1,0b11,0b1011,0,0, (outs DPR:$dst),
3264 (ins DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$src), IIC_VTB4,
3265 "vtbl", "8", "$dst, \\{$tbl1, $tbl2, $tbl3, $tbl4\\}, $src", "",
3266 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl4 DPR:$tbl1, DPR:$tbl2,
3267 DPR:$tbl3, DPR:$tbl4, DPR:$src)))]>;
3268 } // hasExtraSrcRegAllocReq = 1
3270 // VTBX : Vector Table Extension
3272 : N3V<1,1,0b11,0b1000,1,0, (outs DPR:$dst),
3273 (ins DPR:$orig, DPR:$tbl1, DPR:$src), IIC_VTBX1,
3274 "vtbx", "8", "$dst, \\{$tbl1\\}, $src", "$orig = $dst",
3275 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx1
3276 DPR:$orig, DPR:$tbl1, DPR:$src)))]>;
3277 let hasExtraSrcRegAllocReq = 1 in {
3279 : N3V<1,1,0b11,0b1001,1,0, (outs DPR:$dst),
3280 (ins DPR:$orig, DPR:$tbl1, DPR:$tbl2, DPR:$src), IIC_VTBX2,
3281 "vtbx", "8", "$dst, \\{$tbl1, $tbl2\\}, $src", "$orig = $dst",
3282 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx2
3283 DPR:$orig, DPR:$tbl1, DPR:$tbl2, DPR:$src)))]>;
3285 : N3V<1,1,0b11,0b1010,1,0, (outs DPR:$dst),
3286 (ins DPR:$orig, DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$src), IIC_VTBX3,
3287 "vtbx", "8", "$dst, \\{$tbl1, $tbl2, $tbl3\\}, $src", "$orig = $dst",
3288 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx3 DPR:$orig, DPR:$tbl1,
3289 DPR:$tbl2, DPR:$tbl3, DPR:$src)))]>;
3291 : N3V<1,1,0b11,0b1011,1,0, (outs DPR:$dst), (ins DPR:$orig, DPR:$tbl1,
3292 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$src), IIC_VTBX4,
3293 "vtbx", "8", "$dst, \\{$tbl1, $tbl2, $tbl3, $tbl4\\}, $src",
3295 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx4 DPR:$orig, DPR:$tbl1,
3296 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$src)))]>;
3297 } // hasExtraSrcRegAllocReq = 1
3299 //===----------------------------------------------------------------------===//
3300 // NEON instructions for single-precision FP math
3301 //===----------------------------------------------------------------------===//
3303 class N2VSPat<SDNode OpNode, ValueType ResTy, ValueType OpTy, NeonI Inst>
3304 : NEONFPPat<(ResTy (OpNode SPR:$a)),
3305 (EXTRACT_SUBREG (OpTy (Inst (INSERT_SUBREG (OpTy (IMPLICIT_DEF)),
3306 SPR:$a, arm_ssubreg_0))),
3309 class N3VSPat<SDNode OpNode, NeonI Inst>
3310 : NEONFPPat<(f32 (OpNode SPR:$a, SPR:$b)),
3311 (EXTRACT_SUBREG (v2f32
3312 (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3313 SPR:$a, arm_ssubreg_0),
3314 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3315 SPR:$b, arm_ssubreg_0))),
3318 class N3VSMulOpPat<SDNode MulNode, SDNode OpNode, NeonI Inst>
3319 : NEONFPPat<(f32 (OpNode SPR:$acc, (f32 (MulNode SPR:$a, SPR:$b)))),
3320 (EXTRACT_SUBREG (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3321 SPR:$acc, arm_ssubreg_0),
3322 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3323 SPR:$a, arm_ssubreg_0),
3324 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3325 SPR:$b, arm_ssubreg_0)),
3328 // These need separate instructions because they must use DPR_VFP2 register
3329 // class which have SPR sub-registers.
3331 // Vector Add Operations used for single-precision FP
3332 let neverHasSideEffects = 1 in
3333 def VADDfd_sfp : N3VS<0,0,0b00,0b1101,0, "vadd", "f32", v2f32, v2f32, fadd, 1>;
3334 def : N3VSPat<fadd, VADDfd_sfp>;
3336 // Vector Sub Operations used for single-precision FP
3337 let neverHasSideEffects = 1 in
3338 def VSUBfd_sfp : N3VS<0,0,0b10,0b1101,0, "vsub", "f32", v2f32, v2f32, fsub, 0>;
3339 def : N3VSPat<fsub, VSUBfd_sfp>;
3341 // Vector Multiply Operations used for single-precision FP
3342 let neverHasSideEffects = 1 in
3343 def VMULfd_sfp : N3VS<1,0,0b00,0b1101,1, "vmul", "f32", v2f32, v2f32, fmul, 1>;
3344 def : N3VSPat<fmul, VMULfd_sfp>;
3346 // Vector Multiply-Accumulate/Subtract used for single-precision FP
3347 // vml[as].f32 can cause 4-8 cycle stalls in following ASIMD instructions, so
3348 // we want to avoid them for now. e.g., alternating vmla/vadd instructions.
3350 //let neverHasSideEffects = 1 in
3351 //def VMLAfd_sfp : N3VSMulOp<0,0,0b00,0b1101,1, IIC_VMACD, "vmla", "f32",
3352 // v2f32, fmul, fadd>;
3353 //def : N3VSMulOpPat<fmul, fadd, VMLAfd_sfp>;
3355 //let neverHasSideEffects = 1 in
3356 //def VMLSfd_sfp : N3VSMulOp<0,0,0b10,0b1101,1, IIC_VMACD, "vmls", "f32",
3357 // v2f32, fmul, fsub>;
3358 //def : N3VSMulOpPat<fmul, fsub, VMLSfd_sfp>;
3360 // Vector Absolute used for single-precision FP
3361 let neverHasSideEffects = 1 in
3362 def VABSfd_sfp : N2V<0b11, 0b11, 0b10, 0b01, 0b01110, 0, 0,
3363 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), IIC_VUNAD,
3364 "vabs", "f32", "$dst, $src", "", []>;
3365 def : N2VSPat<fabs, f32, v2f32, VABSfd_sfp>;
3367 // Vector Negate used for single-precision FP
3368 let neverHasSideEffects = 1 in
3369 def VNEGfd_sfp : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
3370 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), IIC_VUNAD,
3371 "vneg", "f32", "$dst, $src", "", []>;
3372 def : N2VSPat<fneg, f32, v2f32, VNEGfd_sfp>;
3374 // Vector Maximum used for single-precision FP
3375 let neverHasSideEffects = 1 in
3376 def VMAXfd_sfp : N3V<0, 0, 0b00, 0b1111, 0, 0, (outs DPR_VFP2:$dst),
3377 (ins DPR_VFP2:$src1, DPR_VFP2:$src2), IIC_VBIND,
3378 "vmax", "f32", "$dst, $src1, $src2", "", []>;
3379 def : N3VSPat<NEONfmax, VMAXfd_sfp>;
3381 // Vector Minimum used for single-precision FP
3382 let neverHasSideEffects = 1 in
3383 def VMINfd_sfp : N3V<0, 0, 0b00, 0b1111, 0, 0, (outs DPR_VFP2:$dst),
3384 (ins DPR_VFP2:$src1, DPR_VFP2:$src2), IIC_VBIND,
3385 "vmin", "f32", "$dst, $src1, $src2", "", []>;
3386 def : N3VSPat<NEONfmin, VMINfd_sfp>;
3388 // Vector Convert between single-precision FP and integer
3389 let neverHasSideEffects = 1 in
3390 def VCVTf2sd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
3391 v2i32, v2f32, fp_to_sint>;
3392 def : N2VSPat<arm_ftosi, f32, v2f32, VCVTf2sd_sfp>;
3394 let neverHasSideEffects = 1 in
3395 def VCVTf2ud_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
3396 v2i32, v2f32, fp_to_uint>;
3397 def : N2VSPat<arm_ftoui, f32, v2f32, VCVTf2ud_sfp>;
3399 let neverHasSideEffects = 1 in
3400 def VCVTs2fd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
3401 v2f32, v2i32, sint_to_fp>;
3402 def : N2VSPat<arm_sitof, f32, v2i32, VCVTs2fd_sfp>;
3404 let neverHasSideEffects = 1 in
3405 def VCVTu2fd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
3406 v2f32, v2i32, uint_to_fp>;
3407 def : N2VSPat<arm_uitof, f32, v2i32, VCVTu2fd_sfp>;
3409 //===----------------------------------------------------------------------===//
3410 // Non-Instruction Patterns
3411 //===----------------------------------------------------------------------===//
3414 def : Pat<(v1i64 (bitconvert (v2i32 DPR:$src))), (v1i64 DPR:$src)>;
3415 def : Pat<(v1i64 (bitconvert (v4i16 DPR:$src))), (v1i64 DPR:$src)>;
3416 def : Pat<(v1i64 (bitconvert (v8i8 DPR:$src))), (v1i64 DPR:$src)>;
3417 def : Pat<(v1i64 (bitconvert (f64 DPR:$src))), (v1i64 DPR:$src)>;
3418 def : Pat<(v1i64 (bitconvert (v2f32 DPR:$src))), (v1i64 DPR:$src)>;
3419 def : Pat<(v2i32 (bitconvert (v1i64 DPR:$src))), (v2i32 DPR:$src)>;
3420 def : Pat<(v2i32 (bitconvert (v4i16 DPR:$src))), (v2i32 DPR:$src)>;
3421 def : Pat<(v2i32 (bitconvert (v8i8 DPR:$src))), (v2i32 DPR:$src)>;
3422 def : Pat<(v2i32 (bitconvert (f64 DPR:$src))), (v2i32 DPR:$src)>;
3423 def : Pat<(v2i32 (bitconvert (v2f32 DPR:$src))), (v2i32 DPR:$src)>;
3424 def : Pat<(v4i16 (bitconvert (v1i64 DPR:$src))), (v4i16 DPR:$src)>;
3425 def : Pat<(v4i16 (bitconvert (v2i32 DPR:$src))), (v4i16 DPR:$src)>;
3426 def : Pat<(v4i16 (bitconvert (v8i8 DPR:$src))), (v4i16 DPR:$src)>;
3427 def : Pat<(v4i16 (bitconvert (f64 DPR:$src))), (v4i16 DPR:$src)>;
3428 def : Pat<(v4i16 (bitconvert (v2f32 DPR:$src))), (v4i16 DPR:$src)>;
3429 def : Pat<(v8i8 (bitconvert (v1i64 DPR:$src))), (v8i8 DPR:$src)>;
3430 def : Pat<(v8i8 (bitconvert (v2i32 DPR:$src))), (v8i8 DPR:$src)>;
3431 def : Pat<(v8i8 (bitconvert (v4i16 DPR:$src))), (v8i8 DPR:$src)>;
3432 def : Pat<(v8i8 (bitconvert (f64 DPR:$src))), (v8i8 DPR:$src)>;
3433 def : Pat<(v8i8 (bitconvert (v2f32 DPR:$src))), (v8i8 DPR:$src)>;
3434 def : Pat<(f64 (bitconvert (v1i64 DPR:$src))), (f64 DPR:$src)>;
3435 def : Pat<(f64 (bitconvert (v2i32 DPR:$src))), (f64 DPR:$src)>;
3436 def : Pat<(f64 (bitconvert (v4i16 DPR:$src))), (f64 DPR:$src)>;
3437 def : Pat<(f64 (bitconvert (v8i8 DPR:$src))), (f64 DPR:$src)>;
3438 def : Pat<(f64 (bitconvert (v2f32 DPR:$src))), (f64 DPR:$src)>;
3439 def : Pat<(v2f32 (bitconvert (f64 DPR:$src))), (v2f32 DPR:$src)>;
3440 def : Pat<(v2f32 (bitconvert (v1i64 DPR:$src))), (v2f32 DPR:$src)>;
3441 def : Pat<(v2f32 (bitconvert (v2i32 DPR:$src))), (v2f32 DPR:$src)>;
3442 def : Pat<(v2f32 (bitconvert (v4i16 DPR:$src))), (v2f32 DPR:$src)>;
3443 def : Pat<(v2f32 (bitconvert (v8i8 DPR:$src))), (v2f32 DPR:$src)>;
3445 def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (v2i64 QPR:$src)>;
3446 def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (v2i64 QPR:$src)>;
3447 def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (v2i64 QPR:$src)>;
3448 def : Pat<(v2i64 (bitconvert (v2f64 QPR:$src))), (v2i64 QPR:$src)>;
3449 def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (v2i64 QPR:$src)>;
3450 def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (v4i32 QPR:$src)>;
3451 def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (v4i32 QPR:$src)>;
3452 def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (v4i32 QPR:$src)>;
3453 def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (v4i32 QPR:$src)>;
3454 def : Pat<(v4i32 (bitconvert (v4f32 QPR:$src))), (v4i32 QPR:$src)>;
3455 def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (v8i16 QPR:$src)>;
3456 def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (v8i16 QPR:$src)>;
3457 def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (v8i16 QPR:$src)>;
3458 def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (v8i16 QPR:$src)>;
3459 def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (v8i16 QPR:$src)>;
3460 def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (v16i8 QPR:$src)>;
3461 def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (v16i8 QPR:$src)>;
3462 def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (v16i8 QPR:$src)>;
3463 def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (v16i8 QPR:$src)>;
3464 def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (v16i8 QPR:$src)>;
3465 def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (v4f32 QPR:$src)>;
3466 def : Pat<(v4f32 (bitconvert (v4i32 QPR:$src))), (v4f32 QPR:$src)>;
3467 def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (v4f32 QPR:$src)>;
3468 def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (v4f32 QPR:$src)>;
3469 def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (v4f32 QPR:$src)>;
3470 def : Pat<(v2f64 (bitconvert (v2i64 QPR:$src))), (v2f64 QPR:$src)>;
3471 def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (v2f64 QPR:$src)>;
3472 def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 QPR:$src)>;
3473 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 QPR:$src)>;
3474 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 QPR:$src)>;