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\\}, $addr", "",
199 [/* For disassembly only; pattern left blank */]>;
201 def VLD1d8T : VLD1D3<0b0000, "8">;
202 def VLD1d16T : VLD1D3<0b0100, "16">;
203 def VLD1d32T : VLD1D3<0b1000, "32">;
204 //def VLD1d64T : VLD1D3<0b1100, "64">;
206 def VLD1d8Q : VLD1D4<0b0000, "8">;
207 def VLD1d16Q : VLD1D4<0b0100, "16">;
208 def VLD1d32Q : VLD1D4<0b1000, "32">;
209 //def VLD1d64Q : VLD1D4<0b1100, "64">;
211 // ...with address register writeback:
212 class VLD1D3WB<bits<4> op7_4, string Dt>
213 : NLdSt<0,0b10,0b0110,op7_4, (outs DPR:$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 = $wb",
222 [/* For disassembly only; pattern left blank */]>;
224 def VLD1d8T_UPD : VLD1D3WB<0b0000, "8">;
225 def VLD1d16T_UPD : VLD1D3WB<0b0100, "16">;
226 def VLD1d32T_UPD : VLD1D3WB<0b1000, "32">;
227 // VLD1d64T_UPD : implemented as VLD3d64_UPD
229 def VLD1d8Q_UPD : VLD1D4WB<0b0000, "8">;
230 def VLD1d16Q_UPD : VLD1D4WB<0b0100, "16">;
231 def VLD1d32Q_UPD : VLD1D4WB<0b1000, "32">;
232 // VLD1d64Q_UPD : implemented as VLD4d64_UPD
234 // VLD2 : Vector Load (multiple 2-element structures)
235 class VLD2D<bits<4> op7_4, string Dt>
236 : NLdSt<0,0b10,0b1000,op7_4, (outs DPR:$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\\}, $addr", "", []>;
245 def VLD2d8 : VLD2D<0b0000, "8">;
246 def VLD2d16 : VLD2D<0b0100, "16">;
247 def VLD2d32 : VLD2D<0b1000, "32">;
248 def VLD2d64 : NLdSt<0,0b10,0b1010,0b1100, (outs DPR:$dst1, DPR:$dst2),
249 (ins addrmode6:$addr), IIC_VLD1,
250 "vld1", "64", "\\{$dst1, $dst2\\}, $addr", "", []>;
252 def VLD2q8 : VLD2Q<0b0000, "8">;
253 def VLD2q16 : VLD2Q<0b0100, "16">;
254 def VLD2q32 : VLD2Q<0b1000, "32">;
256 // These (double-spaced dreg pair) are for disassembly only.
257 class VLD2Ddbl<bits<4> op7_4, string Dt>
258 : NLdSt<0,0b10,0b1001,op7_4, (outs DPR:$dst1, DPR:$dst2),
259 (ins addrmode6:$addr), IIC_VLD2,
260 "vld2", Dt, "\\{$dst1, $dst2\\}, $addr", "", []>;
262 def VLD2d8D : VLD2Ddbl<0b0000, "8">;
263 def VLD2d16D : VLD2Ddbl<0b0100, "16">;
264 def VLD2d32D : VLD2Ddbl<0b1000, "32">;
266 // VLD3 : Vector Load (multiple 3-element structures)
267 class VLD3D<bits<4> op7_4, string Dt>
268 : NLdSt<0,0b10,0b0100,op7_4, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3),
269 (ins addrmode6:$addr), IIC_VLD3,
270 "vld3", Dt, "\\{$dst1, $dst2, $dst3\\}, $addr", "", []>;
271 class VLD3WB<bits<4> op7_4, string Dt>
272 : NLdSt<0,0b10,0b0101,op7_4, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, GPR:$wb),
273 (ins addrmode6:$addr), IIC_VLD3,
274 "vld3", Dt, "\\{$dst1, $dst2, $dst3\\}, $addr",
275 "$addr.addr = $wb", []>;
277 def VLD3d8 : VLD3D<0b0000, "8">;
278 def VLD3d16 : VLD3D<0b0100, "16">;
279 def VLD3d32 : VLD3D<0b1000, "32">;
280 def VLD3d64 : NLdSt<0,0b10,0b0110,0b1100,
281 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3),
282 (ins addrmode6:$addr), IIC_VLD1,
283 "vld1", "64", "\\{$dst1, $dst2, $dst3\\}, $addr", "", []>;
285 // vld3 to double-spaced even registers.
286 def VLD3q8a : VLD3WB<0b0000, "8">;
287 def VLD3q16a : VLD3WB<0b0100, "16">;
288 def VLD3q32a : VLD3WB<0b1000, "32">;
290 // vld3 to double-spaced odd registers.
291 def VLD3q8b : VLD3WB<0b0000, "8">;
292 def VLD3q16b : VLD3WB<0b0100, "16">;
293 def VLD3q32b : VLD3WB<0b1000, "32">;
295 // VLD4 : Vector Load (multiple 4-element structures)
296 class VLD4D<bits<4> op7_4, string Dt>
297 : NLdSt<0,0b10,0b0000,op7_4,
298 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
299 (ins addrmode6:$addr), IIC_VLD4,
300 "vld4", Dt, "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr", "", []>;
301 class VLD4WB<bits<4> op7_4, string Dt>
302 : NLdSt<0,0b10,0b0001,op7_4,
303 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
304 (ins addrmode6:$addr), IIC_VLD4,
305 "vld4", Dt, "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr",
306 "$addr.addr = $wb", []>;
308 def VLD4d8 : VLD4D<0b0000, "8">;
309 def VLD4d16 : VLD4D<0b0100, "16">;
310 def VLD4d32 : VLD4D<0b1000, "32">;
311 def VLD4d64 : NLdSt<0,0b10,0b0010,0b1100,
312 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
313 (ins addrmode6:$addr), IIC_VLD1,
314 "vld1", "64", "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr",
317 // vld4 to double-spaced even registers.
318 def VLD4q8a : VLD4WB<0b0000, "8">;
319 def VLD4q16a : VLD4WB<0b0100, "16">;
320 def VLD4q32a : VLD4WB<0b1000, "32">;
322 // vld4 to double-spaced odd registers.
323 def VLD4q8b : VLD4WB<0b0000, "8">;
324 def VLD4q16b : VLD4WB<0b0100, "16">;
325 def VLD4q32b : VLD4WB<0b1000, "32">;
327 // VLD1LN : Vector Load (single element to one lane)
328 // FIXME: Not yet implemented.
330 // VLD2LN : Vector Load (single 2-element structure to one lane)
331 class VLD2LN<bits<4> op11_8, string Dt>
332 : NLdSt<1,0b10,op11_8,{?,?,?,?}, (outs DPR:$dst1, DPR:$dst2),
333 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, nohash_imm:$lane),
334 IIC_VLD2, "vld2", Dt, "\\{$dst1[$lane], $dst2[$lane]\\}, $addr",
335 "$src1 = $dst1, $src2 = $dst2", []>;
337 // vld2 to single-spaced registers.
338 def VLD2LNd8 : VLD2LN<0b0001, "8">;
339 def VLD2LNd16 : VLD2LN<0b0101, "16"> { let Inst{5} = 0; }
340 def VLD2LNd32 : VLD2LN<0b1001, "32"> { let Inst{6} = 0; }
342 // vld2 to double-spaced even registers.
343 def VLD2LNq16a: VLD2LN<0b0101, "16"> { let Inst{5} = 1; }
344 def VLD2LNq32a: VLD2LN<0b1001, "32"> { let Inst{6} = 1; }
346 // vld2 to double-spaced odd registers.
347 def VLD2LNq16b: VLD2LN<0b0101, "16"> { let Inst{5} = 1; }
348 def VLD2LNq32b: VLD2LN<0b1001, "32"> { let Inst{6} = 1; }
350 // VLD3LN : Vector Load (single 3-element structure to one lane)
351 class VLD3LN<bits<4> op11_8, string Dt>
352 : NLdSt<1,0b10,op11_8,{?,?,?,?}, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3),
353 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3,
354 nohash_imm:$lane), IIC_VLD3, "vld3", Dt,
355 "\\{$dst1[$lane], $dst2[$lane], $dst3[$lane]\\}, $addr",
356 "$src1 = $dst1, $src2 = $dst2, $src3 = $dst3", []>;
358 // vld3 to single-spaced registers.
359 def VLD3LNd8 : VLD3LN<0b0010, "8"> { let Inst{4} = 0; }
360 def VLD3LNd16 : VLD3LN<0b0110, "16"> { let Inst{5-4} = 0b00; }
361 def VLD3LNd32 : VLD3LN<0b1010, "32"> { let Inst{6-4} = 0b000; }
363 // vld3 to double-spaced even registers.
364 def VLD3LNq16a: VLD3LN<0b0110, "16"> { let Inst{5-4} = 0b10; }
365 def VLD3LNq32a: VLD3LN<0b1010, "32"> { let Inst{6-4} = 0b100; }
367 // vld3 to double-spaced odd registers.
368 def VLD3LNq16b: VLD3LN<0b0110, "16"> { let Inst{5-4} = 0b10; }
369 def VLD3LNq32b: VLD3LN<0b1010, "32"> { let Inst{6-4} = 0b100; }
371 // VLD4LN : Vector Load (single 4-element structure to one lane)
372 class VLD4LN<bits<4> op11_8, string Dt>
373 : NLdSt<1,0b10,op11_8,{?,?,?,?},
374 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
375 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
376 nohash_imm:$lane), IIC_VLD4, "vld4", Dt,
377 "\\{$dst1[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $addr",
378 "$src1 = $dst1, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4", []>;
380 // vld4 to single-spaced registers.
381 def VLD4LNd8 : VLD4LN<0b0011, "8">;
382 def VLD4LNd16 : VLD4LN<0b0111, "16"> { let Inst{5} = 0; }
383 def VLD4LNd32 : VLD4LN<0b1011, "32"> { let Inst{6} = 0; }
385 // vld4 to double-spaced even registers.
386 def VLD4LNq16a: VLD4LN<0b0111, "16"> { let Inst{5} = 1; }
387 def VLD4LNq32a: VLD4LN<0b1011, "32"> { let Inst{6} = 1; }
389 // vld4 to double-spaced odd registers.
390 def VLD4LNq16b: VLD4LN<0b0111, "16"> { let Inst{5} = 1; }
391 def VLD4LNq32b: VLD4LN<0b1011, "32"> { let Inst{6} = 1; }
393 // VLD1DUP : Vector Load (single element to all lanes)
394 // VLD2DUP : Vector Load (single 2-element structure to all lanes)
395 // VLD3DUP : Vector Load (single 3-element structure to all lanes)
396 // VLD4DUP : Vector Load (single 4-element structure to all lanes)
397 // FIXME: Not yet implemented.
398 } // mayLoad = 1, hasExtraDefRegAllocReq = 1
400 // VST1 : Vector Store (multiple single elements)
401 class VST1D<bits<4> op7_4, string Dt, ValueType Ty>
402 : NLdSt<0,0b00,0b0111,op7_4, (outs), (ins addrmode6:$addr, DPR:$src), IIC_VST,
403 "vst1", Dt, "\\{$src\\}, $addr", "",
404 [(int_arm_neon_vst1 addrmode6:$addr, (Ty DPR:$src))]>;
405 class VST1Q<bits<4> op7_4, string Dt, ValueType Ty>
406 : NLdSt<0,0b00,0b1010,op7_4, (outs), (ins addrmode6:$addr, QPR:$src), IIC_VST,
407 "vst1", Dt, "${src:dregpair}, $addr", "",
408 [(int_arm_neon_vst1 addrmode6:$addr, (Ty QPR:$src))]>;
410 let hasExtraSrcRegAllocReq = 1 in {
411 def VST1d8 : VST1D<0b0000, "8", v8i8>;
412 def VST1d16 : VST1D<0b0100, "16", v4i16>;
413 def VST1d32 : VST1D<0b1000, "32", v2i32>;
414 def VST1df : VST1D<0b1000, "32", v2f32>;
415 def VST1d64 : VST1D<0b1100, "64", v1i64>;
417 def VST1q8 : VST1Q<0b0000, "8", v16i8>;
418 def VST1q16 : VST1Q<0b0100, "16", v8i16>;
419 def VST1q32 : VST1Q<0b1000, "32", v4i32>;
420 def VST1qf : VST1Q<0b1000, "32", v4f32>;
421 def VST1q64 : VST1Q<0b1100, "64", v2i64>;
422 } // hasExtraSrcRegAllocReq
424 // These (dreg triple/quadruple) are for disassembly only.
425 class VST1D3<bits<4> op7_4, string Dt>
426 : NLdSt<0, 0b00, 0b0110, op7_4, (outs),
427 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3), IIC_VST,
428 "vst1", Dt, "\\{$src1, $src2, $src3\\}, $addr", "",
429 [/* For disassembly only; pattern left blank */]>;
430 class VST1D4<bits<4> op7_4, string Dt>
431 : NLdSt<0, 0b00, 0b0010, op7_4, (outs),
432 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
433 IIC_VST, "vst1", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr", "",
434 [/* For disassembly only; pattern left blank */]>;
436 def VST1d8T : VST1D3<0b0000, "8">;
437 def VST1d16T : VST1D3<0b0100, "16">;
438 def VST1d32T : VST1D3<0b1000, "32">;
439 //def VST1d64T : VST1D3<0b1100, "64">;
441 def VST1d8Q : VST1D4<0b0000, "8">;
442 def VST1d16Q : VST1D4<0b0100, "16">;
443 def VST1d32Q : VST1D4<0b1000, "32">;
444 //def VST1d64Q : VST1D4<0b1100, "64">;
447 let mayStore = 1, hasExtraSrcRegAllocReq = 1 in {
449 // VST2 : Vector Store (multiple 2-element structures)
450 class VST2D<bits<4> op7_4, string Dt>
451 : NLdSt<0,0b00,0b1000,op7_4, (outs),
452 (ins addrmode6:$addr, DPR:$src1, DPR:$src2), IIC_VST,
453 "vst2", Dt, "\\{$src1, $src2\\}, $addr", "", []>;
454 class VST2Q<bits<4> op7_4, string Dt>
455 : NLdSt<0,0b00,0b0011,op7_4, (outs),
456 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
457 IIC_VST, "vst2", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr",
460 def VST2d8 : VST2D<0b0000, "8">;
461 def VST2d16 : VST2D<0b0100, "16">;
462 def VST2d32 : VST2D<0b1000, "32">;
463 def VST2d64 : NLdSt<0,0b00,0b1010,0b1100, (outs),
464 (ins addrmode6:$addr, DPR:$src1, DPR:$src2), IIC_VST,
465 "vst1", "64", "\\{$src1, $src2\\}, $addr", "", []>;
467 def VST2q8 : VST2Q<0b0000, "8">;
468 def VST2q16 : VST2Q<0b0100, "16">;
469 def VST2q32 : VST2Q<0b1000, "32">;
471 // These (double-spaced dreg pair) are for disassembly only.
472 class VST2Ddbl<bits<4> op7_4, string Dt>
473 : NLdSt<0, 0b00, 0b1001, op7_4, (outs),
474 (ins addrmode6:$addr, DPR:$src1, DPR:$src2), IIC_VST,
475 "vst2", Dt, "\\{$src1, $src2\\}, $addr", "", []>;
477 def VST2d8D : VST2Ddbl<0b0000, "8">;
478 def VST2d16D : VST2Ddbl<0b0100, "16">;
479 def VST2d32D : VST2Ddbl<0b1000, "32">;
481 // VST3 : Vector Store (multiple 3-element structures)
482 class VST3D<bits<4> op7_4, string Dt>
483 : NLdSt<0,0b00,0b0100,op7_4, (outs),
484 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3), IIC_VST,
485 "vst3", Dt, "\\{$src1, $src2, $src3\\}, $addr", "", []>;
486 class VST3WB<bits<4> op7_4, string Dt>
487 : NLdSt<0,0b00,0b0101,op7_4, (outs GPR:$wb),
488 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3), IIC_VST,
489 "vst3", Dt, "\\{$src1, $src2, $src3\\}, $addr",
490 "$addr.addr = $wb", []>;
492 def VST3d8 : VST3D<0b0000, "8">;
493 def VST3d16 : VST3D<0b0100, "16">;
494 def VST3d32 : VST3D<0b1000, "32">;
495 def VST3d64 : NLdSt<0,0b00,0b0110,0b1100, (outs),
496 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3),
498 "vst1", "64", "\\{$src1, $src2, $src3\\}, $addr", "", []>;
500 // vst3 to double-spaced even registers.
501 def VST3q8a : VST3WB<0b0000, "8">;
502 def VST3q16a : VST3WB<0b0100, "16">;
503 def VST3q32a : VST3WB<0b1000, "32">;
505 // vst3 to double-spaced odd registers.
506 def VST3q8b : VST3WB<0b0000, "8">;
507 def VST3q16b : VST3WB<0b0100, "16">;
508 def VST3q32b : VST3WB<0b1000, "32">;
510 // VST4 : Vector Store (multiple 4-element structures)
511 class VST4D<bits<4> op7_4, string Dt>
512 : NLdSt<0,0b00,0b0000,op7_4, (outs),
513 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
514 IIC_VST, "vst4", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr",
516 class VST4WB<bits<4> op7_4, string Dt>
517 : NLdSt<0,0b00,0b0001,op7_4, (outs GPR:$wb),
518 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
519 IIC_VST, "vst4", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr",
520 "$addr.addr = $wb", []>;
522 def VST4d8 : VST4D<0b0000, "8">;
523 def VST4d16 : VST4D<0b0100, "16">;
524 def VST4d32 : VST4D<0b1000, "32">;
525 def VST4d64 : NLdSt<0,0b00,0b0010,0b1100, (outs),
526 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3,
528 "vst1", "64", "\\{$src1, $src2, $src3, $src4\\}, $addr",
531 // vst4 to double-spaced even registers.
532 def VST4q8a : VST4WB<0b0000, "8">;
533 def VST4q16a : VST4WB<0b0100, "16">;
534 def VST4q32a : VST4WB<0b1000, "32">;
536 // vst4 to double-spaced odd registers.
537 def VST4q8b : VST4WB<0b0000, "8">;
538 def VST4q16b : VST4WB<0b0100, "16">;
539 def VST4q32b : VST4WB<0b1000, "32">;
541 // VST1LN : Vector Store (single element from one lane)
542 // FIXME: Not yet implemented.
544 // VST2LN : Vector Store (single 2-element structure from one lane)
545 class VST2LN<bits<4> op11_8, string Dt>
546 : NLdSt<1,0b00,op11_8,{?,?,?,?}, (outs),
547 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, nohash_imm:$lane),
548 IIC_VST, "vst2", Dt, "\\{$src1[$lane], $src2[$lane]\\}, $addr",
551 // vst2 to single-spaced registers.
552 def VST2LNd8 : VST2LN<0b0001, "8">;
553 def VST2LNd16 : VST2LN<0b0101, "16"> { let Inst{5} = 0; }
554 def VST2LNd32 : VST2LN<0b1001, "32"> { let Inst{6} = 0; }
556 // vst2 to double-spaced even registers.
557 def VST2LNq16a: VST2LN<0b0101, "16"> { let Inst{5} = 1; }
558 def VST2LNq32a: VST2LN<0b1001, "32"> { let Inst{6} = 1; }
560 // vst2 to double-spaced odd registers.
561 def VST2LNq16b: VST2LN<0b0101, "16"> { let Inst{5} = 1; }
562 def VST2LNq32b: VST2LN<0b1001, "32"> { let Inst{6} = 1; }
564 // VST3LN : Vector Store (single 3-element structure from one lane)
565 class VST3LN<bits<4> op11_8, string Dt>
566 : NLdSt<1,0b00,op11_8,{?,?,?,?}, (outs),
567 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3,
568 nohash_imm:$lane), IIC_VST, "vst3", Dt,
569 "\\{$src1[$lane], $src2[$lane], $src3[$lane]\\}, $addr", "", []>;
571 // vst3 to single-spaced registers.
572 def VST3LNd8 : VST3LN<0b0010, "8"> { let Inst{4} = 0; }
573 def VST3LNd16 : VST3LN<0b0110, "16"> { let Inst{5-4} = 0b00; }
574 def VST3LNd32 : VST3LN<0b1010, "32"> { let Inst{6-4} = 0b000; }
576 // vst3 to double-spaced even registers.
577 def VST3LNq16a: VST3LN<0b0110, "16"> { let Inst{5-4} = 0b10; }
578 def VST3LNq32a: VST3LN<0b1010, "32"> { let Inst{6-4} = 0b100; }
580 // vst3 to double-spaced odd registers.
581 def VST3LNq16b: VST3LN<0b0110, "16"> { let Inst{5-4} = 0b10; }
582 def VST3LNq32b: VST3LN<0b1010, "32"> { let Inst{6-4} = 0b100; }
584 // VST4LN : Vector Store (single 4-element structure from one lane)
585 class VST4LN<bits<4> op11_8, string Dt>
586 : NLdSt<1,0b00,op11_8,{?,?,?,?}, (outs),
587 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
588 nohash_imm:$lane), IIC_VST, "vst4", Dt,
589 "\\{$src1[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $addr",
592 // vst4 to single-spaced registers.
593 def VST4LNd8 : VST4LN<0b0011, "8">;
594 def VST4LNd16 : VST4LN<0b0111, "16"> { let Inst{5} = 0; }
595 def VST4LNd32 : VST4LN<0b1011, "32"> { let Inst{6} = 0; }
597 // vst4 to double-spaced even registers.
598 def VST4LNq16a: VST4LN<0b0111, "16"> { let Inst{5} = 1; }
599 def VST4LNq32a: VST4LN<0b1011, "32"> { let Inst{6} = 1; }
601 // vst4 to double-spaced odd registers.
602 def VST4LNq16b: VST4LN<0b0111, "16"> { let Inst{5} = 1; }
603 def VST4LNq32b: VST4LN<0b1011, "32"> { let Inst{6} = 1; }
605 } // mayStore = 1, hasExtraSrcRegAllocReq = 1
608 //===----------------------------------------------------------------------===//
609 // NEON pattern fragments
610 //===----------------------------------------------------------------------===//
612 // Extract D sub-registers of Q registers.
613 // (arm_dsubreg_0 is 5; arm_dsubreg_1 is 6)
614 def DSubReg_i8_reg : SDNodeXForm<imm, [{
615 return CurDAG->getTargetConstant(5 + N->getZExtValue() / 8, MVT::i32);
617 def DSubReg_i16_reg : SDNodeXForm<imm, [{
618 return CurDAG->getTargetConstant(5 + N->getZExtValue() / 4, MVT::i32);
620 def DSubReg_i32_reg : SDNodeXForm<imm, [{
621 return CurDAG->getTargetConstant(5 + N->getZExtValue() / 2, MVT::i32);
623 def DSubReg_f64_reg : SDNodeXForm<imm, [{
624 return CurDAG->getTargetConstant(5 + N->getZExtValue(), MVT::i32);
626 def DSubReg_f64_other_reg : SDNodeXForm<imm, [{
627 return CurDAG->getTargetConstant(5 + (1 - N->getZExtValue()), MVT::i32);
630 // Extract S sub-registers of Q/D registers.
631 // (arm_ssubreg_0 is 1; arm_ssubreg_1 is 2; etc.)
632 def SSubReg_f32_reg : SDNodeXForm<imm, [{
633 return CurDAG->getTargetConstant(1 + N->getZExtValue(), MVT::i32);
636 // Translate lane numbers from Q registers to D subregs.
637 def SubReg_i8_lane : SDNodeXForm<imm, [{
638 return CurDAG->getTargetConstant(N->getZExtValue() & 7, MVT::i32);
640 def SubReg_i16_lane : SDNodeXForm<imm, [{
641 return CurDAG->getTargetConstant(N->getZExtValue() & 3, MVT::i32);
643 def SubReg_i32_lane : SDNodeXForm<imm, [{
644 return CurDAG->getTargetConstant(N->getZExtValue() & 1, MVT::i32);
647 //===----------------------------------------------------------------------===//
648 // Instruction Classes
649 //===----------------------------------------------------------------------===//
651 // Basic 2-register operations: single-, double- and quad-register.
652 class N2VS<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
653 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
654 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
655 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
656 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src),
657 IIC_VUNAD, OpcodeStr, Dt, "$dst, $src", "", []>;
658 class N2VD<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
659 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
660 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
661 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
662 (ins DPR:$src), IIC_VUNAD, OpcodeStr, Dt, "$dst, $src", "",
663 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src))))]>;
664 class N2VQ<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
665 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
666 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
667 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
668 (ins QPR:$src), IIC_VUNAQ, OpcodeStr, Dt, "$dst, $src", "",
669 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src))))]>;
671 // Basic 2-register intrinsics, both double- and quad-register.
672 class N2VDInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
673 bits<2> op17_16, bits<5> op11_7, bit op4,
674 InstrItinClass itin, string OpcodeStr, string Dt,
675 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
676 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
677 (ins DPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
678 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
679 class N2VQInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
680 bits<2> op17_16, bits<5> op11_7, bit op4,
681 InstrItinClass itin, string OpcodeStr, string Dt,
682 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
683 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
684 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
685 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
687 // Narrow 2-register intrinsics.
688 class N2VNInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
689 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
690 InstrItinClass itin, string OpcodeStr, string Dt,
691 ValueType TyD, ValueType TyQ, Intrinsic IntOp>
692 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$dst),
693 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
694 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src))))]>;
696 // Long 2-register intrinsics (currently only used for VMOVL).
697 class N2VLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
698 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
699 InstrItinClass itin, string OpcodeStr, string Dt,
700 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
701 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$dst),
702 (ins DPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
703 [(set QPR:$dst, (TyQ (IntOp (TyD DPR:$src))))]>;
705 // 2-register shuffles (VTRN/VZIP/VUZP), both double- and quad-register.
706 class N2VDShuffle<bits<2> op19_18, bits<5> op11_7, string OpcodeStr, string Dt>
707 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 0, 0, (outs DPR:$dst1, DPR:$dst2),
708 (ins DPR:$src1, DPR:$src2), IIC_VPERMD,
709 OpcodeStr, Dt, "$dst1, $dst2",
710 "$src1 = $dst1, $src2 = $dst2", []>;
711 class N2VQShuffle<bits<2> op19_18, bits<5> op11_7,
712 InstrItinClass itin, string OpcodeStr, string Dt>
713 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 1, 0, (outs QPR:$dst1, QPR:$dst2),
714 (ins QPR:$src1, QPR:$src2), itin, OpcodeStr, Dt, "$dst1, $dst2",
715 "$src1 = $dst1, $src2 = $dst2", []>;
717 // Basic 3-register operations: single-, double- and quad-register.
718 class N3VS<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
719 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
720 SDNode OpNode, bit Commutable>
721 : N3V<op24, op23, op21_20, op11_8, 0, op4,
722 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src1, DPR_VFP2:$src2), IIC_VBIND,
723 OpcodeStr, Dt, "$dst, $src1, $src2", "", []> {
724 let isCommutable = Commutable;
727 class N3VD<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
728 InstrItinClass itin, string OpcodeStr, string Dt,
729 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
730 : N3V<op24, op23, op21_20, op11_8, 0, op4,
731 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), itin,
732 OpcodeStr, Dt, "$dst, $src1, $src2", "",
733 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src1), (OpTy DPR:$src2))))]> {
734 let isCommutable = Commutable;
736 // Same as N3VD but no data type.
737 class N3VDX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
738 InstrItinClass itin, string OpcodeStr,
739 ValueType ResTy, ValueType OpTy,
740 SDNode OpNode, bit Commutable>
741 : N3VX<op24, op23, op21_20, op11_8, 0, op4,
742 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), itin,
743 OpcodeStr, "$dst, $src1, $src2", "",
744 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src1), (OpTy DPR:$src2))))]>{
745 let isCommutable = Commutable;
747 class N3VDSL<bits<2> op21_20, bits<4> op11_8,
748 InstrItinClass itin, string OpcodeStr, string Dt,
749 ValueType Ty, SDNode ShOp>
750 : N3V<0, 1, op21_20, op11_8, 1, 0,
751 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
752 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
754 (Ty (ShOp (Ty DPR:$src1),
755 (Ty (NEONvduplane (Ty DPR_VFP2:$src2), imm:$lane)))))]>{
756 let isCommutable = 0;
758 class N3VDSL16<bits<2> op21_20, bits<4> op11_8,
759 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
760 : N3V<0, 1, op21_20, op11_8, 1, 0,
761 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
762 IIC_VMULi16D, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
764 (Ty (ShOp (Ty DPR:$src1),
765 (Ty (NEONvduplane (Ty DPR_8:$src2), imm:$lane)))))]> {
766 let isCommutable = 0;
769 class N3VQ<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
770 InstrItinClass itin, string OpcodeStr, string Dt,
771 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
772 : N3V<op24, op23, op21_20, op11_8, 1, op4,
773 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), itin,
774 OpcodeStr, Dt, "$dst, $src1, $src2", "",
775 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src1), (OpTy QPR:$src2))))]> {
776 let isCommutable = Commutable;
778 class N3VQX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
779 InstrItinClass itin, string OpcodeStr,
780 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
781 : N3VX<op24, op23, op21_20, op11_8, 1, op4,
782 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), itin,
783 OpcodeStr, "$dst, $src1, $src2", "",
784 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src1), (OpTy QPR:$src2))))]>{
785 let isCommutable = Commutable;
787 class N3VQSL<bits<2> op21_20, bits<4> op11_8,
788 InstrItinClass itin, string OpcodeStr, string Dt,
789 ValueType ResTy, ValueType OpTy, SDNode ShOp>
790 : N3V<1, 1, op21_20, op11_8, 1, 0,
791 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
792 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
793 [(set (ResTy QPR:$dst),
794 (ResTy (ShOp (ResTy QPR:$src1),
795 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
797 let isCommutable = 0;
799 class N3VQSL16<bits<2> op21_20, bits<4> op11_8, string OpcodeStr, string Dt,
800 ValueType ResTy, ValueType OpTy, SDNode ShOp>
801 : N3V<1, 1, op21_20, op11_8, 1, 0,
802 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
803 IIC_VMULi16Q, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
804 [(set (ResTy QPR:$dst),
805 (ResTy (ShOp (ResTy QPR:$src1),
806 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
808 let isCommutable = 0;
811 // Basic 3-register intrinsics, both double- and quad-register.
812 class N3VDInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
813 InstrItinClass itin, string OpcodeStr, string Dt,
814 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
815 : N3V<op24, op23, op21_20, op11_8, 0, op4,
816 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), itin,
817 OpcodeStr, Dt, "$dst, $src1, $src2", "",
818 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src1), (OpTy DPR:$src2))))]> {
819 let isCommutable = Commutable;
821 class N3VDIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
822 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
823 : N3V<0, 1, op21_20, op11_8, 1, 0,
824 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
825 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
827 (Ty (IntOp (Ty DPR:$src1),
828 (Ty (NEONvduplane (Ty DPR_VFP2:$src2),
830 let isCommutable = 0;
832 class N3VDIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
833 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
834 : N3V<0, 1, op21_20, op11_8, 1, 0,
835 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
836 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
838 (Ty (IntOp (Ty DPR:$src1),
839 (Ty (NEONvduplane (Ty DPR_8:$src2),
841 let isCommutable = 0;
844 class N3VQInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
845 InstrItinClass itin, string OpcodeStr, string Dt,
846 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
847 : N3V<op24, op23, op21_20, op11_8, 1, op4,
848 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), itin,
849 OpcodeStr, Dt, "$dst, $src1, $src2", "",
850 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src1), (OpTy QPR:$src2))))]> {
851 let isCommutable = Commutable;
853 class N3VQIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
854 string OpcodeStr, string Dt,
855 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
856 : N3V<1, 1, op21_20, op11_8, 1, 0,
857 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
858 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
859 [(set (ResTy QPR:$dst),
860 (ResTy (IntOp (ResTy QPR:$src1),
861 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
863 let isCommutable = 0;
865 class N3VQIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
866 string OpcodeStr, string Dt,
867 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
868 : N3V<1, 1, op21_20, op11_8, 1, 0,
869 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
870 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
871 [(set (ResTy QPR:$dst),
872 (ResTy (IntOp (ResTy QPR:$src1),
873 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
875 let isCommutable = 0;
878 // Multiply-Add/Sub operations: single-, double- and quad-register.
879 class N3VSMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
880 InstrItinClass itin, string OpcodeStr, string Dt,
881 ValueType Ty, SDNode MulOp, SDNode OpNode>
882 : N3V<op24, op23, op21_20, op11_8, 0, op4,
883 (outs DPR_VFP2:$dst),
884 (ins DPR_VFP2:$src1, DPR_VFP2:$src2, DPR_VFP2:$src3), itin,
885 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst", []>;
887 class N3VDMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
888 InstrItinClass itin, string OpcodeStr, string Dt,
889 ValueType Ty, SDNode MulOp, SDNode OpNode>
890 : N3V<op24, op23, op21_20, op11_8, 0, op4,
891 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3), itin,
892 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
893 [(set DPR:$dst, (Ty (OpNode DPR:$src1,
894 (Ty (MulOp DPR:$src2, DPR:$src3)))))]>;
895 class N3VDMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
896 string OpcodeStr, string Dt,
897 ValueType Ty, SDNode MulOp, SDNode ShOp>
898 : N3V<0, 1, op21_20, op11_8, 1, 0,
900 (ins DPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane), itin,
901 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
903 (Ty (ShOp (Ty DPR:$src1),
904 (Ty (MulOp DPR:$src2,
905 (Ty (NEONvduplane (Ty DPR_VFP2:$src3),
907 class N3VDMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
908 string OpcodeStr, string Dt,
909 ValueType Ty, SDNode MulOp, SDNode ShOp>
910 : N3V<0, 1, op21_20, op11_8, 1, 0,
912 (ins DPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane), itin,
913 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
915 (Ty (ShOp (Ty DPR:$src1),
916 (Ty (MulOp DPR:$src2,
917 (Ty (NEONvduplane (Ty DPR_8:$src3),
920 class N3VQMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
921 InstrItinClass itin, string OpcodeStr, string Dt, ValueType Ty,
922 SDNode MulOp, SDNode OpNode>
923 : N3V<op24, op23, op21_20, op11_8, 1, op4,
924 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3), itin,
925 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
926 [(set QPR:$dst, (Ty (OpNode QPR:$src1,
927 (Ty (MulOp QPR:$src2, QPR:$src3)))))]>;
928 class N3VQMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
929 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
930 SDNode MulOp, SDNode ShOp>
931 : N3V<1, 1, op21_20, op11_8, 1, 0,
933 (ins QPR:$src1, QPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane), itin,
934 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
935 [(set (ResTy QPR:$dst),
936 (ResTy (ShOp (ResTy QPR:$src1),
937 (ResTy (MulOp QPR:$src2,
938 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src3),
940 class N3VQMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
941 string OpcodeStr, string Dt,
942 ValueType ResTy, ValueType OpTy,
943 SDNode MulOp, SDNode ShOp>
944 : N3V<1, 1, op21_20, op11_8, 1, 0,
946 (ins QPR:$src1, QPR:$src2, DPR_8:$src3, nohash_imm:$lane), itin,
947 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
948 [(set (ResTy QPR:$dst),
949 (ResTy (ShOp (ResTy QPR:$src1),
950 (ResTy (MulOp QPR:$src2,
951 (ResTy (NEONvduplane (OpTy DPR_8:$src3),
954 // Neon 3-argument intrinsics, both double- and quad-register.
955 // The destination register is also used as the first source operand register.
956 class N3VDInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
957 InstrItinClass itin, string OpcodeStr, string Dt,
958 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
959 : N3V<op24, op23, op21_20, op11_8, 0, op4,
960 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3), itin,
961 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
962 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src1),
963 (OpTy DPR:$src2), (OpTy DPR:$src3))))]>;
964 class N3VQInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
965 InstrItinClass itin, string OpcodeStr, string Dt,
966 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
967 : N3V<op24, op23, op21_20, op11_8, 1, op4,
968 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3), itin,
969 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
970 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src1),
971 (OpTy QPR:$src2), (OpTy QPR:$src3))))]>;
973 // Neon Long 3-argument intrinsic. The destination register is
974 // a quad-register and is also used as the first source operand register.
975 class N3VLInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
976 InstrItinClass itin, string OpcodeStr, string Dt,
977 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
978 : N3V<op24, op23, op21_20, op11_8, 0, op4,
979 (outs QPR:$dst), (ins QPR:$src1, DPR:$src2, DPR:$src3), itin,
980 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
982 (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$src2), (TyD DPR:$src3))))]>;
983 class N3VLInt3SL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
984 string OpcodeStr, string Dt,
985 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
986 : N3V<op24, 1, op21_20, op11_8, 1, 0,
988 (ins QPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane), itin,
989 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
990 [(set (ResTy QPR:$dst),
991 (ResTy (IntOp (ResTy QPR:$src1),
993 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src3),
995 class N3VLInt3SL16<bit op24, bits<2> op21_20, bits<4> op11_8,
996 InstrItinClass itin, string OpcodeStr, string Dt,
997 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
998 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1000 (ins QPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane), itin,
1001 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1002 [(set (ResTy QPR:$dst),
1003 (ResTy (IntOp (ResTy QPR:$src1),
1005 (OpTy (NEONvduplane (OpTy DPR_8:$src3),
1008 // Narrowing 3-register intrinsics.
1009 class N3VNInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1010 string OpcodeStr, string Dt, ValueType TyD, ValueType TyQ,
1011 Intrinsic IntOp, bit Commutable>
1012 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1013 (outs DPR:$dst), (ins QPR:$src1, QPR:$src2), IIC_VBINi4D,
1014 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1015 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src1), (TyQ QPR:$src2))))]> {
1016 let isCommutable = Commutable;
1019 // Long 3-register intrinsics.
1020 class N3VLInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1021 InstrItinClass itin, string OpcodeStr, string Dt,
1022 ValueType TyQ, ValueType TyD, Intrinsic IntOp, bit Commutable>
1023 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1024 (outs QPR:$dst), (ins DPR:$src1, DPR:$src2), itin,
1025 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1026 [(set QPR:$dst, (TyQ (IntOp (TyD DPR:$src1), (TyD DPR:$src2))))]> {
1027 let isCommutable = Commutable;
1029 class N3VLIntSL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1030 string OpcodeStr, string Dt,
1031 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1032 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1033 (outs QPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1034 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1035 [(set (ResTy QPR:$dst),
1036 (ResTy (IntOp (OpTy DPR:$src1),
1037 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1039 class N3VLIntSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1040 InstrItinClass itin, string OpcodeStr, string Dt,
1041 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1042 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1043 (outs QPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1044 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1045 [(set (ResTy QPR:$dst),
1046 (ResTy (IntOp (OpTy DPR:$src1),
1047 (OpTy (NEONvduplane (OpTy DPR_8:$src2),
1050 // Wide 3-register intrinsics.
1051 class N3VWInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1052 string OpcodeStr, string Dt, ValueType TyQ, ValueType TyD,
1053 Intrinsic IntOp, bit Commutable>
1054 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1055 (outs QPR:$dst), (ins QPR:$src1, DPR:$src2), IIC_VSUBiD,
1056 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1057 [(set QPR:$dst, (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$src2))))]> {
1058 let isCommutable = Commutable;
1061 // Pairwise long 2-register intrinsics, both double- and quad-register.
1062 class N2VDPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1063 bits<2> op17_16, bits<5> op11_7, bit op4,
1064 string OpcodeStr, string Dt,
1065 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1066 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
1067 (ins DPR:$src), IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
1068 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
1069 class N2VQPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1070 bits<2> op17_16, bits<5> op11_7, bit op4,
1071 string OpcodeStr, string Dt,
1072 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1073 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
1074 (ins QPR:$src), IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
1075 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
1077 // Pairwise long 2-register accumulate intrinsics,
1078 // both double- and quad-register.
1079 // The destination register is also used as the first source operand register.
1080 class N2VDPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1081 bits<2> op17_16, bits<5> op11_7, bit op4,
1082 string OpcodeStr, string Dt,
1083 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1084 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
1085 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), IIC_VPALiD,
1086 OpcodeStr, Dt, "$dst, $src2", "$src1 = $dst",
1087 [(set DPR:$dst, (ResTy (IntOp (ResTy DPR:$src1), (OpTy DPR:$src2))))]>;
1088 class N2VQPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1089 bits<2> op17_16, bits<5> op11_7, bit op4,
1090 string OpcodeStr, string Dt,
1091 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1092 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4,
1093 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), IIC_VPALiQ,
1094 OpcodeStr, Dt, "$dst, $src2", "$src1 = $dst",
1095 [(set QPR:$dst, (ResTy (IntOp (ResTy QPR:$src1), (OpTy QPR:$src2))))]>;
1097 // Shift by immediate,
1098 // both double- and quad-register.
1099 class N2VDSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1100 InstrItinClass itin, string OpcodeStr, string Dt,
1101 ValueType Ty, SDNode OpNode>
1102 : N2VImm<op24, op23, op11_8, op7, 0, op4,
1103 (outs DPR:$dst), (ins DPR:$src, i32imm:$SIMM), itin,
1104 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1105 [(set DPR:$dst, (Ty (OpNode (Ty DPR:$src), (i32 imm:$SIMM))))]>;
1106 class N2VQSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1107 InstrItinClass itin, string OpcodeStr, string Dt,
1108 ValueType Ty, SDNode OpNode>
1109 : N2VImm<op24, op23, op11_8, op7, 1, op4,
1110 (outs QPR:$dst), (ins QPR:$src, i32imm:$SIMM), itin,
1111 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1112 [(set QPR:$dst, (Ty (OpNode (Ty QPR:$src), (i32 imm:$SIMM))))]>;
1114 // Long shift by immediate.
1115 class N2VLSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
1116 string OpcodeStr, string Dt,
1117 ValueType ResTy, ValueType OpTy, SDNode OpNode>
1118 : N2VImm<op24, op23, op11_8, op7, op6, op4,
1119 (outs QPR:$dst), (ins DPR:$src, i32imm:$SIMM), IIC_VSHLiD,
1120 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1121 [(set QPR:$dst, (ResTy (OpNode (OpTy DPR:$src),
1122 (i32 imm:$SIMM))))]>;
1124 // Narrow shift by immediate.
1125 class N2VNSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
1126 InstrItinClass itin, string OpcodeStr, string Dt,
1127 ValueType ResTy, ValueType OpTy, SDNode OpNode>
1128 : N2VImm<op24, op23, op11_8, op7, op6, op4,
1129 (outs DPR:$dst), (ins QPR:$src, i32imm:$SIMM), itin,
1130 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1131 [(set DPR:$dst, (ResTy (OpNode (OpTy QPR:$src),
1132 (i32 imm:$SIMM))))]>;
1134 // Shift right by immediate and accumulate,
1135 // both double- and quad-register.
1136 class N2VDShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1137 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1138 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$dst),
1139 (ins DPR:$src1, DPR:$src2, i32imm:$SIMM), IIC_VPALiD,
1140 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1141 [(set DPR:$dst, (Ty (add DPR:$src1,
1142 (Ty (ShOp DPR:$src2, (i32 imm:$SIMM))))))]>;
1143 class N2VQShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1144 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1145 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$dst),
1146 (ins QPR:$src1, QPR:$src2, i32imm:$SIMM), IIC_VPALiD,
1147 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1148 [(set QPR:$dst, (Ty (add QPR:$src1,
1149 (Ty (ShOp QPR:$src2, (i32 imm:$SIMM))))))]>;
1151 // Shift by immediate and insert,
1152 // both double- and quad-register.
1153 class N2VDShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1154 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1155 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$dst),
1156 (ins DPR:$src1, DPR:$src2, i32imm:$SIMM), IIC_VSHLiD,
1157 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1158 [(set DPR:$dst, (Ty (ShOp DPR:$src1, DPR:$src2, (i32 imm:$SIMM))))]>;
1159 class N2VQShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1160 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1161 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$dst),
1162 (ins QPR:$src1, QPR:$src2, i32imm:$SIMM), IIC_VSHLiQ,
1163 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1164 [(set QPR:$dst, (Ty (ShOp QPR:$src1, QPR:$src2, (i32 imm:$SIMM))))]>;
1166 // Convert, with fractional bits immediate,
1167 // both double- and quad-register.
1168 class N2VCvtD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1169 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1171 : N2VImm<op24, op23, op11_8, op7, 0, op4,
1172 (outs DPR:$dst), (ins DPR:$src, i32imm:$SIMM), IIC_VUNAD,
1173 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1174 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src), (i32 imm:$SIMM))))]>;
1175 class N2VCvtQ<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1176 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1178 : N2VImm<op24, op23, op11_8, op7, 1, op4,
1179 (outs QPR:$dst), (ins QPR:$src, i32imm:$SIMM), IIC_VUNAQ,
1180 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1181 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src), (i32 imm:$SIMM))))]>;
1183 //===----------------------------------------------------------------------===//
1185 //===----------------------------------------------------------------------===//
1187 // Abbreviations used in multiclass suffixes:
1188 // Q = quarter int (8 bit) elements
1189 // H = half int (16 bit) elements
1190 // S = single int (32 bit) elements
1191 // D = double int (64 bit) elements
1193 // Neon 2-register vector operations -- for disassembly only.
1195 // First with only element sizes of 8, 16 and 32 bits:
1196 multiclass N2V_QHS_cmp<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1197 bits<5> op11_7, bit op4, string opc, string Dt,
1199 // 64-bit vector types.
1200 def v8i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 0, op4,
1201 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1202 opc, !strconcat(Dt, "8"), asm, "", []>;
1203 def v4i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 0, op4,
1204 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1205 opc, !strconcat(Dt, "16"), asm, "", []>;
1206 def v2i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
1207 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1208 opc, !strconcat(Dt, "32"), asm, "", []>;
1209 def v2f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
1210 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1211 opc, "f32", asm, "", []> {
1212 let Inst{10} = 1; // overwrite F = 1
1215 // 128-bit vector types.
1216 def v16i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 1, op4,
1217 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1218 opc, !strconcat(Dt, "8"), asm, "", []>;
1219 def v8i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 1, op4,
1220 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1221 opc, !strconcat(Dt, "16"), asm, "", []>;
1222 def v4i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
1223 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1224 opc, !strconcat(Dt, "32"), asm, "", []>;
1225 def v4f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
1226 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1227 opc, "f32", asm, "", []> {
1228 let Inst{10} = 1; // overwrite F = 1
1232 // Neon 3-register vector operations.
1234 // First with only element sizes of 8, 16 and 32 bits:
1235 multiclass N3V_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1236 InstrItinClass itinD16, InstrItinClass itinD32,
1237 InstrItinClass itinQ16, InstrItinClass itinQ32,
1238 string OpcodeStr, string Dt,
1239 SDNode OpNode, bit Commutable = 0> {
1240 // 64-bit vector types.
1241 def v8i8 : N3VD<op24, op23, 0b00, op11_8, op4, itinD16,
1242 OpcodeStr, !strconcat(Dt, "8"),
1243 v8i8, v8i8, OpNode, Commutable>;
1244 def v4i16 : N3VD<op24, op23, 0b01, op11_8, op4, itinD16,
1245 OpcodeStr, !strconcat(Dt, "16"),
1246 v4i16, v4i16, OpNode, Commutable>;
1247 def v2i32 : N3VD<op24, op23, 0b10, op11_8, op4, itinD32,
1248 OpcodeStr, !strconcat(Dt, "32"),
1249 v2i32, v2i32, OpNode, Commutable>;
1251 // 128-bit vector types.
1252 def v16i8 : N3VQ<op24, op23, 0b00, op11_8, op4, itinQ16,
1253 OpcodeStr, !strconcat(Dt, "8"),
1254 v16i8, v16i8, OpNode, Commutable>;
1255 def v8i16 : N3VQ<op24, op23, 0b01, op11_8, op4, itinQ16,
1256 OpcodeStr, !strconcat(Dt, "16"),
1257 v8i16, v8i16, OpNode, Commutable>;
1258 def v4i32 : N3VQ<op24, op23, 0b10, op11_8, op4, itinQ32,
1259 OpcodeStr, !strconcat(Dt, "32"),
1260 v4i32, v4i32, OpNode, Commutable>;
1263 multiclass N3VSL_HS<bits<4> op11_8, string OpcodeStr, string Dt, SDNode ShOp> {
1264 def v4i16 : N3VDSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
1266 def v2i32 : N3VDSL<0b10, op11_8, IIC_VMULi32D, OpcodeStr, !strconcat(Dt,"32"),
1268 def v8i16 : N3VQSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
1269 v8i16, v4i16, ShOp>;
1270 def v4i32 : N3VQSL<0b10, op11_8, IIC_VMULi32Q, OpcodeStr, !strconcat(Dt,"32"),
1271 v4i32, v2i32, ShOp>;
1274 // ....then also with element size 64 bits:
1275 multiclass N3V_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1276 InstrItinClass itinD, InstrItinClass itinQ,
1277 string OpcodeStr, string Dt,
1278 SDNode OpNode, bit Commutable = 0>
1279 : N3V_QHS<op24, op23, op11_8, op4, itinD, itinD, itinQ, itinQ,
1280 OpcodeStr, Dt, OpNode, Commutable> {
1281 def v1i64 : N3VD<op24, op23, 0b11, op11_8, op4, itinD,
1282 OpcodeStr, !strconcat(Dt, "64"),
1283 v1i64, v1i64, OpNode, Commutable>;
1284 def v2i64 : N3VQ<op24, op23, 0b11, op11_8, op4, itinQ,
1285 OpcodeStr, !strconcat(Dt, "64"),
1286 v2i64, v2i64, OpNode, Commutable>;
1290 // Neon Narrowing 2-register vector intrinsics,
1291 // source operand element sizes of 16, 32 and 64 bits:
1292 multiclass N2VNInt_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1293 bits<5> op11_7, bit op6, bit op4,
1294 InstrItinClass itin, string OpcodeStr, string Dt,
1296 def v8i8 : N2VNInt<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
1297 itin, OpcodeStr, !strconcat(Dt, "16"),
1298 v8i8, v8i16, IntOp>;
1299 def v4i16 : N2VNInt<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
1300 itin, OpcodeStr, !strconcat(Dt, "32"),
1301 v4i16, v4i32, IntOp>;
1302 def v2i32 : N2VNInt<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
1303 itin, OpcodeStr, !strconcat(Dt, "64"),
1304 v2i32, v2i64, IntOp>;
1308 // Neon Lengthening 2-register vector intrinsic (currently specific to VMOVL).
1309 // source operand element sizes of 16, 32 and 64 bits:
1310 multiclass N2VLInt_QHS<bits<2> op24_23, bits<5> op11_7, bit op6, bit op4,
1311 string OpcodeStr, string Dt, Intrinsic IntOp> {
1312 def v8i16 : N2VLInt<op24_23, 0b00, 0b10, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
1313 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
1314 def v4i32 : N2VLInt<op24_23, 0b01, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
1315 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
1316 def v2i64 : N2VLInt<op24_23, 0b10, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
1317 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1321 // Neon 3-register vector intrinsics.
1323 // First with only element sizes of 16 and 32 bits:
1324 multiclass N3VInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
1325 InstrItinClass itinD16, InstrItinClass itinD32,
1326 InstrItinClass itinQ16, InstrItinClass itinQ32,
1327 string OpcodeStr, string Dt,
1328 Intrinsic IntOp, bit Commutable = 0> {
1329 // 64-bit vector types.
1330 def v4i16 : N3VDInt<op24, op23, 0b01, op11_8, op4, itinD16,
1331 OpcodeStr, !strconcat(Dt, "16"),
1332 v4i16, v4i16, IntOp, Commutable>;
1333 def v2i32 : N3VDInt<op24, op23, 0b10, op11_8, op4, itinD32,
1334 OpcodeStr, !strconcat(Dt, "32"),
1335 v2i32, v2i32, IntOp, Commutable>;
1337 // 128-bit vector types.
1338 def v8i16 : N3VQInt<op24, op23, 0b01, op11_8, op4, itinQ16,
1339 OpcodeStr, !strconcat(Dt, "16"),
1340 v8i16, v8i16, IntOp, Commutable>;
1341 def v4i32 : N3VQInt<op24, op23, 0b10, op11_8, op4, itinQ32,
1342 OpcodeStr, !strconcat(Dt, "32"),
1343 v4i32, v4i32, IntOp, Commutable>;
1346 multiclass N3VIntSL_HS<bits<4> op11_8,
1347 InstrItinClass itinD16, InstrItinClass itinD32,
1348 InstrItinClass itinQ16, InstrItinClass itinQ32,
1349 string OpcodeStr, string Dt, Intrinsic IntOp> {
1350 def v4i16 : N3VDIntSL16<0b01, op11_8, itinD16,
1351 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp>;
1352 def v2i32 : N3VDIntSL<0b10, op11_8, itinD32,
1353 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp>;
1354 def v8i16 : N3VQIntSL16<0b01, op11_8, itinQ16,
1355 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16, IntOp>;
1356 def v4i32 : N3VQIntSL<0b10, op11_8, itinQ32,
1357 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32, IntOp>;
1360 // ....then also with element size of 8 bits:
1361 multiclass N3VInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1362 InstrItinClass itinD16, InstrItinClass itinD32,
1363 InstrItinClass itinQ16, InstrItinClass itinQ32,
1364 string OpcodeStr, string Dt,
1365 Intrinsic IntOp, bit Commutable = 0>
1366 : N3VInt_HS<op24, op23, op11_8, op4, itinD16, itinD32, itinQ16, itinQ32,
1367 OpcodeStr, Dt, IntOp, Commutable> {
1368 def v8i8 : N3VDInt<op24, op23, 0b00, op11_8, op4, itinD16,
1369 OpcodeStr, !strconcat(Dt, "8"),
1370 v8i8, v8i8, IntOp, Commutable>;
1371 def v16i8 : N3VQInt<op24, op23, 0b00, op11_8, op4, itinQ16,
1372 OpcodeStr, !strconcat(Dt, "8"),
1373 v16i8, v16i8, IntOp, Commutable>;
1376 // ....then also with element size of 64 bits:
1377 multiclass N3VInt_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1378 InstrItinClass itinD16, InstrItinClass itinD32,
1379 InstrItinClass itinQ16, InstrItinClass itinQ32,
1380 string OpcodeStr, string Dt,
1381 Intrinsic IntOp, bit Commutable = 0>
1382 : N3VInt_QHS<op24, op23, op11_8, op4, itinD16, itinD32, itinQ16, itinQ32,
1383 OpcodeStr, Dt, IntOp, Commutable> {
1384 def v1i64 : N3VDInt<op24, op23, 0b11, op11_8, op4, itinD32,
1385 OpcodeStr, !strconcat(Dt, "64"),
1386 v1i64, v1i64, IntOp, Commutable>;
1387 def v2i64 : N3VQInt<op24, op23, 0b11, op11_8, op4, itinQ32,
1388 OpcodeStr, !strconcat(Dt, "64"),
1389 v2i64, v2i64, IntOp, Commutable>;
1393 // Neon Narrowing 3-register vector intrinsics,
1394 // source operand element sizes of 16, 32 and 64 bits:
1395 multiclass N3VNInt_HSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1396 string OpcodeStr, string Dt,
1397 Intrinsic IntOp, bit Commutable = 0> {
1398 def v8i8 : N3VNInt<op24, op23, 0b00, op11_8, op4,
1399 OpcodeStr, !strconcat(Dt, "16"),
1400 v8i8, v8i16, IntOp, Commutable>;
1401 def v4i16 : N3VNInt<op24, op23, 0b01, op11_8, op4,
1402 OpcodeStr, !strconcat(Dt, "32"),
1403 v4i16, v4i32, IntOp, Commutable>;
1404 def v2i32 : N3VNInt<op24, op23, 0b10, op11_8, op4,
1405 OpcodeStr, !strconcat(Dt, "64"),
1406 v2i32, v2i64, IntOp, Commutable>;
1410 // Neon Long 3-register vector intrinsics.
1412 // First with only element sizes of 16 and 32 bits:
1413 multiclass N3VLInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
1414 InstrItinClass itin, string OpcodeStr, string Dt,
1415 Intrinsic IntOp, bit Commutable = 0> {
1416 def v4i32 : N3VLInt<op24, op23, 0b01, op11_8, op4, itin,
1417 OpcodeStr, !strconcat(Dt, "16"),
1418 v4i32, v4i16, IntOp, Commutable>;
1419 def v2i64 : N3VLInt<op24, op23, 0b10, op11_8, op4, itin,
1420 OpcodeStr, !strconcat(Dt, "32"),
1421 v2i64, v2i32, IntOp, Commutable>;
1424 multiclass N3VLIntSL_HS<bit op24, bits<4> op11_8,
1425 InstrItinClass itin, string OpcodeStr, string Dt,
1427 def v4i16 : N3VLIntSL16<op24, 0b01, op11_8, itin,
1428 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
1429 def v2i32 : N3VLIntSL<op24, 0b10, op11_8, itin,
1430 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1433 // ....then also with element size of 8 bits:
1434 multiclass N3VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1435 InstrItinClass itin, string OpcodeStr, string Dt,
1436 Intrinsic IntOp, bit Commutable = 0>
1437 : N3VLInt_HS<op24, op23, op11_8, op4, itin, OpcodeStr, Dt,
1438 IntOp, Commutable> {
1439 def v8i16 : N3VLInt<op24, op23, 0b00, op11_8, op4, itin,
1440 OpcodeStr, !strconcat(Dt, "8"),
1441 v8i16, v8i8, IntOp, Commutable>;
1445 // Neon Wide 3-register vector intrinsics,
1446 // source operand element sizes of 8, 16 and 32 bits:
1447 multiclass N3VWInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1448 string OpcodeStr, string Dt,
1449 Intrinsic IntOp, bit Commutable = 0> {
1450 def v8i16 : N3VWInt<op24, op23, 0b00, op11_8, op4,
1451 OpcodeStr, !strconcat(Dt, "8"),
1452 v8i16, v8i8, IntOp, Commutable>;
1453 def v4i32 : N3VWInt<op24, op23, 0b01, op11_8, op4,
1454 OpcodeStr, !strconcat(Dt, "16"),
1455 v4i32, v4i16, IntOp, Commutable>;
1456 def v2i64 : N3VWInt<op24, op23, 0b10, op11_8, op4,
1457 OpcodeStr, !strconcat(Dt, "32"),
1458 v2i64, v2i32, IntOp, Commutable>;
1462 // Neon Multiply-Op vector operations,
1463 // element sizes of 8, 16 and 32 bits:
1464 multiclass N3VMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1465 InstrItinClass itinD16, InstrItinClass itinD32,
1466 InstrItinClass itinQ16, InstrItinClass itinQ32,
1467 string OpcodeStr, string Dt, SDNode OpNode> {
1468 // 64-bit vector types.
1469 def v8i8 : N3VDMulOp<op24, op23, 0b00, op11_8, op4, itinD16,
1470 OpcodeStr, !strconcat(Dt, "8"), v8i8, mul, OpNode>;
1471 def v4i16 : N3VDMulOp<op24, op23, 0b01, op11_8, op4, itinD16,
1472 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, OpNode>;
1473 def v2i32 : N3VDMulOp<op24, op23, 0b10, op11_8, op4, itinD32,
1474 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, OpNode>;
1476 // 128-bit vector types.
1477 def v16i8 : N3VQMulOp<op24, op23, 0b00, op11_8, op4, itinQ16,
1478 OpcodeStr, !strconcat(Dt, "8"), v16i8, mul, OpNode>;
1479 def v8i16 : N3VQMulOp<op24, op23, 0b01, op11_8, op4, itinQ16,
1480 OpcodeStr, !strconcat(Dt, "16"), v8i16, mul, OpNode>;
1481 def v4i32 : N3VQMulOp<op24, op23, 0b10, op11_8, op4, itinQ32,
1482 OpcodeStr, !strconcat(Dt, "32"), v4i32, mul, OpNode>;
1485 multiclass N3VMulOpSL_HS<bits<4> op11_8,
1486 InstrItinClass itinD16, InstrItinClass itinD32,
1487 InstrItinClass itinQ16, InstrItinClass itinQ32,
1488 string OpcodeStr, string Dt, SDNode ShOp> {
1489 def v4i16 : N3VDMulOpSL16<0b01, op11_8, itinD16,
1490 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, ShOp>;
1491 def v2i32 : N3VDMulOpSL<0b10, op11_8, itinD32,
1492 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, ShOp>;
1493 def v8i16 : N3VQMulOpSL16<0b01, op11_8, itinQ16,
1494 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16,
1496 def v4i32 : N3VQMulOpSL<0b10, op11_8, itinQ32,
1497 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32,
1501 // Neon 3-argument intrinsics,
1502 // element sizes of 8, 16 and 32 bits:
1503 multiclass N3VInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1504 string OpcodeStr, string Dt, Intrinsic IntOp> {
1505 // 64-bit vector types.
1506 def v8i8 : N3VDInt3<op24, op23, 0b00, op11_8, op4, IIC_VMACi16D,
1507 OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
1508 def v4i16 : N3VDInt3<op24, op23, 0b01, op11_8, op4, IIC_VMACi16D,
1509 OpcodeStr, !strconcat(Dt, "16"), v4i16, v4i16, IntOp>;
1510 def v2i32 : N3VDInt3<op24, op23, 0b10, op11_8, op4, IIC_VMACi32D,
1511 OpcodeStr, !strconcat(Dt, "32"), v2i32, v2i32, IntOp>;
1513 // 128-bit vector types.
1514 def v16i8 : N3VQInt3<op24, op23, 0b00, op11_8, op4, IIC_VMACi16Q,
1515 OpcodeStr, !strconcat(Dt, "8"), v16i8, v16i8, IntOp>;
1516 def v8i16 : N3VQInt3<op24, op23, 0b01, op11_8, op4, IIC_VMACi16Q,
1517 OpcodeStr, !strconcat(Dt, "16"), v8i16, v8i16, IntOp>;
1518 def v4i32 : N3VQInt3<op24, op23, 0b10, op11_8, op4, IIC_VMACi32Q,
1519 OpcodeStr, !strconcat(Dt, "32"), v4i32, v4i32, IntOp>;
1523 // Neon Long 3-argument intrinsics.
1525 // First with only element sizes of 16 and 32 bits:
1526 multiclass N3VLInt3_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
1527 string OpcodeStr, string Dt, Intrinsic IntOp> {
1528 def v4i32 : N3VLInt3<op24, op23, 0b01, op11_8, op4, IIC_VMACi16D,
1529 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
1530 def v2i64 : N3VLInt3<op24, op23, 0b10, op11_8, op4, IIC_VMACi16D,
1531 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1534 multiclass N3VLInt3SL_HS<bit op24, bits<4> op11_8,
1535 string OpcodeStr, string Dt, Intrinsic IntOp> {
1536 def v4i16 : N3VLInt3SL16<op24, 0b01, op11_8, IIC_VMACi16D,
1537 OpcodeStr, !strconcat(Dt,"16"), v4i32, v4i16, IntOp>;
1538 def v2i32 : N3VLInt3SL<op24, 0b10, op11_8, IIC_VMACi32D,
1539 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1542 // ....then also with element size of 8 bits:
1543 multiclass N3VLInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1544 string OpcodeStr, string Dt, Intrinsic IntOp>
1545 : N3VLInt3_HS<op24, op23, op11_8, op4, OpcodeStr, Dt, IntOp> {
1546 def v8i16 : N3VLInt3<op24, op23, 0b00, op11_8, op4, IIC_VMACi16D,
1547 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
1551 // Neon 2-register vector intrinsics,
1552 // element sizes of 8, 16 and 32 bits:
1553 multiclass N2VInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1554 bits<5> op11_7, bit op4,
1555 InstrItinClass itinD, InstrItinClass itinQ,
1556 string OpcodeStr, string Dt, Intrinsic IntOp> {
1557 // 64-bit vector types.
1558 def v8i8 : N2VDInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1559 itinD, OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
1560 def v4i16 : N2VDInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1561 itinD, OpcodeStr, !strconcat(Dt, "16"),v4i16,v4i16,IntOp>;
1562 def v2i32 : N2VDInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1563 itinD, OpcodeStr, !strconcat(Dt, "32"),v2i32,v2i32,IntOp>;
1565 // 128-bit vector types.
1566 def v16i8 : N2VQInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1567 itinQ, OpcodeStr, !strconcat(Dt, "8"), v16i8,v16i8,IntOp>;
1568 def v8i16 : N2VQInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1569 itinQ, OpcodeStr, !strconcat(Dt, "16"),v8i16,v8i16,IntOp>;
1570 def v4i32 : N2VQInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1571 itinQ, OpcodeStr, !strconcat(Dt, "32"),v4i32,v4i32,IntOp>;
1575 // Neon Pairwise long 2-register intrinsics,
1576 // element sizes of 8, 16 and 32 bits:
1577 multiclass N2VPLInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1578 bits<5> op11_7, bit op4,
1579 string OpcodeStr, string Dt, Intrinsic IntOp> {
1580 // 64-bit vector types.
1581 def v8i8 : N2VDPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1582 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
1583 def v4i16 : N2VDPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1584 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
1585 def v2i32 : N2VDPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1586 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
1588 // 128-bit vector types.
1589 def v16i8 : N2VQPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1590 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
1591 def v8i16 : N2VQPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1592 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
1593 def v4i32 : N2VQPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1594 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
1598 // Neon Pairwise long 2-register accumulate intrinsics,
1599 // element sizes of 8, 16 and 32 bits:
1600 multiclass N2VPLInt2_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1601 bits<5> op11_7, bit op4,
1602 string OpcodeStr, string Dt, Intrinsic IntOp> {
1603 // 64-bit vector types.
1604 def v8i8 : N2VDPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1605 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
1606 def v4i16 : N2VDPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1607 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
1608 def v2i32 : N2VDPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1609 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
1611 // 128-bit vector types.
1612 def v16i8 : N2VQPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1613 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
1614 def v8i16 : N2VQPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1615 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
1616 def v4i32 : N2VQPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1617 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
1621 // Neon 2-register vector shift by immediate,
1622 // element sizes of 8, 16, 32 and 64 bits:
1623 multiclass N2VSh_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1624 InstrItinClass itin, string OpcodeStr, string Dt,
1626 // 64-bit vector types.
1627 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, itin,
1628 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
1629 let Inst{21-19} = 0b001; // imm6 = 001xxx
1631 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, itin,
1632 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
1633 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1635 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, itin,
1636 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
1637 let Inst{21} = 0b1; // imm6 = 1xxxxx
1639 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, itin,
1640 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
1643 // 128-bit vector types.
1644 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, itin,
1645 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
1646 let Inst{21-19} = 0b001; // imm6 = 001xxx
1648 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, itin,
1649 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
1650 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1652 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, itin,
1653 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
1654 let Inst{21} = 0b1; // imm6 = 1xxxxx
1656 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, itin,
1657 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
1662 // Neon Shift-Accumulate vector operations,
1663 // element sizes of 8, 16, 32 and 64 bits:
1664 multiclass N2VShAdd_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1665 string OpcodeStr, string Dt, SDNode ShOp> {
1666 // 64-bit vector types.
1667 def v8i8 : N2VDShAdd<op24, op23, op11_8, 0, op4,
1668 OpcodeStr, !strconcat(Dt, "8"), v8i8, ShOp> {
1669 let Inst{21-19} = 0b001; // imm6 = 001xxx
1671 def v4i16 : N2VDShAdd<op24, op23, op11_8, 0, op4,
1672 OpcodeStr, !strconcat(Dt, "16"), v4i16, ShOp> {
1673 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1675 def v2i32 : N2VDShAdd<op24, op23, op11_8, 0, op4,
1676 OpcodeStr, !strconcat(Dt, "32"), v2i32, ShOp> {
1677 let Inst{21} = 0b1; // imm6 = 1xxxxx
1679 def v1i64 : N2VDShAdd<op24, op23, op11_8, 1, op4,
1680 OpcodeStr, !strconcat(Dt, "64"), v1i64, ShOp>;
1683 // 128-bit vector types.
1684 def v16i8 : N2VQShAdd<op24, op23, op11_8, 0, op4,
1685 OpcodeStr, !strconcat(Dt, "8"), v16i8, ShOp> {
1686 let Inst{21-19} = 0b001; // imm6 = 001xxx
1688 def v8i16 : N2VQShAdd<op24, op23, op11_8, 0, op4,
1689 OpcodeStr, !strconcat(Dt, "16"), v8i16, ShOp> {
1690 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1692 def v4i32 : N2VQShAdd<op24, op23, op11_8, 0, op4,
1693 OpcodeStr, !strconcat(Dt, "32"), v4i32, ShOp> {
1694 let Inst{21} = 0b1; // imm6 = 1xxxxx
1696 def v2i64 : N2VQShAdd<op24, op23, op11_8, 1, op4,
1697 OpcodeStr, !strconcat(Dt, "64"), v2i64, ShOp>;
1702 // Neon Shift-Insert vector operations,
1703 // element sizes of 8, 16, 32 and 64 bits:
1704 multiclass N2VShIns_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1705 string OpcodeStr, SDNode ShOp> {
1706 // 64-bit vector types.
1707 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4,
1708 OpcodeStr, "8", v8i8, ShOp> {
1709 let Inst{21-19} = 0b001; // imm6 = 001xxx
1711 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4,
1712 OpcodeStr, "16", v4i16, ShOp> {
1713 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1715 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4,
1716 OpcodeStr, "32", v2i32, ShOp> {
1717 let Inst{21} = 0b1; // imm6 = 1xxxxx
1719 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4,
1720 OpcodeStr, "64", v1i64, ShOp>;
1723 // 128-bit vector types.
1724 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4,
1725 OpcodeStr, "8", v16i8, ShOp> {
1726 let Inst{21-19} = 0b001; // imm6 = 001xxx
1728 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4,
1729 OpcodeStr, "16", v8i16, ShOp> {
1730 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1732 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4,
1733 OpcodeStr, "32", v4i32, ShOp> {
1734 let Inst{21} = 0b1; // imm6 = 1xxxxx
1736 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4,
1737 OpcodeStr, "64", v2i64, ShOp>;
1741 // Neon Shift Long operations,
1742 // element sizes of 8, 16, 32 bits:
1743 multiclass N2VLSh_QHS<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
1744 bit op4, string OpcodeStr, string Dt, SDNode OpNode> {
1745 def v8i16 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
1746 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode> {
1747 let Inst{21-19} = 0b001; // imm6 = 001xxx
1749 def v4i32 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
1750 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode> {
1751 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1753 def v2i64 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
1754 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode> {
1755 let Inst{21} = 0b1; // imm6 = 1xxxxx
1759 // Neon Shift Narrow operations,
1760 // element sizes of 16, 32, 64 bits:
1761 multiclass N2VNSh_HSD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
1762 bit op4, InstrItinClass itin, string OpcodeStr, string Dt,
1764 def v8i8 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
1765 OpcodeStr, !strconcat(Dt, "16"), v8i8, v8i16, OpNode> {
1766 let Inst{21-19} = 0b001; // imm6 = 001xxx
1768 def v4i16 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
1769 OpcodeStr, !strconcat(Dt, "32"), v4i16, v4i32, OpNode> {
1770 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1772 def v2i32 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
1773 OpcodeStr, !strconcat(Dt, "64"), v2i32, v2i64, OpNode> {
1774 let Inst{21} = 0b1; // imm6 = 1xxxxx
1778 //===----------------------------------------------------------------------===//
1779 // Instruction Definitions.
1780 //===----------------------------------------------------------------------===//
1782 // Vector Add Operations.
1784 // VADD : Vector Add (integer and floating-point)
1785 defm VADD : N3V_QHSD<0, 0, 0b1000, 0, IIC_VBINiD, IIC_VBINiQ, "vadd", "i",
1787 def VADDfd : N3VD<0, 0, 0b00, 0b1101, 0, IIC_VBIND, "vadd", "f32",
1788 v2f32, v2f32, fadd, 1>;
1789 def VADDfq : N3VQ<0, 0, 0b00, 0b1101, 0, IIC_VBINQ, "vadd", "f32",
1790 v4f32, v4f32, fadd, 1>;
1791 // VADDL : Vector Add Long (Q = D + D)
1792 defm VADDLs : N3VLInt_QHS<0,1,0b0000,0, IIC_VSHLiD, "vaddl", "s",
1793 int_arm_neon_vaddls, 1>;
1794 defm VADDLu : N3VLInt_QHS<1,1,0b0000,0, IIC_VSHLiD, "vaddl", "u",
1795 int_arm_neon_vaddlu, 1>;
1796 // VADDW : Vector Add Wide (Q = Q + D)
1797 defm VADDWs : N3VWInt_QHS<0,1,0b0001,0, "vaddw", "s", int_arm_neon_vaddws, 0>;
1798 defm VADDWu : N3VWInt_QHS<1,1,0b0001,0, "vaddw", "u", int_arm_neon_vaddwu, 0>;
1799 // VHADD : Vector Halving Add
1800 defm VHADDs : N3VInt_QHS<0,0,0b0000,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1801 IIC_VBINi4Q, "vhadd", "s", int_arm_neon_vhadds, 1>;
1802 defm VHADDu : N3VInt_QHS<1,0,0b0000,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1803 IIC_VBINi4Q, "vhadd", "u", int_arm_neon_vhaddu, 1>;
1804 // VRHADD : Vector Rounding Halving Add
1805 defm VRHADDs : N3VInt_QHS<0,0,0b0001,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1806 IIC_VBINi4Q, "vrhadd", "s", int_arm_neon_vrhadds, 1>;
1807 defm VRHADDu : N3VInt_QHS<1,0,0b0001,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1808 IIC_VBINi4Q, "vrhadd", "u", int_arm_neon_vrhaddu, 1>;
1809 // VQADD : Vector Saturating Add
1810 defm VQADDs : N3VInt_QHSD<0,0,0b0000,1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1811 IIC_VBINi4Q, "vqadd", "s", int_arm_neon_vqadds, 1>;
1812 defm VQADDu : N3VInt_QHSD<1,0,0b0000,1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1813 IIC_VBINi4Q, "vqadd", "u", int_arm_neon_vqaddu, 1>;
1814 // VADDHN : Vector Add and Narrow Returning High Half (D = Q + Q)
1815 defm VADDHN : N3VNInt_HSD<0,1,0b0100,0, "vaddhn", "i",
1816 int_arm_neon_vaddhn, 1>;
1817 // VRADDHN : Vector Rounding Add and Narrow Returning High Half (D = Q + Q)
1818 defm VRADDHN : N3VNInt_HSD<1,1,0b0100,0, "vraddhn", "i",
1819 int_arm_neon_vraddhn, 1>;
1821 // Vector Multiply Operations.
1823 // VMUL : Vector Multiply (integer, polynomial and floating-point)
1824 defm VMUL : N3V_QHS<0, 0, 0b1001, 1, IIC_VMULi16D, IIC_VMULi32D,
1825 IIC_VMULi16Q, IIC_VMULi32Q, "vmul", "i", mul, 1>;
1826 def VMULpd : N3VDInt<1, 0, 0b00, 0b1001, 1, IIC_VMULi16D, "vmul", "p8",
1827 v8i8, v8i8, int_arm_neon_vmulp, 1>;
1828 def VMULpq : N3VQInt<1, 0, 0b00, 0b1001, 1, IIC_VMULi16Q, "vmul", "p8",
1829 v16i8, v16i8, int_arm_neon_vmulp, 1>;
1830 def VMULfd : N3VD<1, 0, 0b00, 0b1101, 1, IIC_VBIND, "vmul", "f32",
1831 v2f32, v2f32, fmul, 1>;
1832 def VMULfq : N3VQ<1, 0, 0b00, 0b1101, 1, IIC_VBINQ, "vmul", "f32",
1833 v4f32, v4f32, fmul, 1>;
1834 defm VMULsl : N3VSL_HS<0b1000, "vmul", "i", mul>;
1835 def VMULslfd : N3VDSL<0b10, 0b1001, IIC_VBIND, "vmul", "f32", v2f32, fmul>;
1836 def VMULslfq : N3VQSL<0b10, 0b1001, IIC_VBINQ, "vmul", "f32", v4f32,
1839 def : Pat<(v8i16 (mul (v8i16 QPR:$src1),
1840 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
1841 (v8i16 (VMULslv8i16 (v8i16 QPR:$src1),
1842 (v4i16 (EXTRACT_SUBREG QPR:$src2,
1843 (DSubReg_i16_reg imm:$lane))),
1844 (SubReg_i16_lane imm:$lane)))>;
1845 def : Pat<(v4i32 (mul (v4i32 QPR:$src1),
1846 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
1847 (v4i32 (VMULslv4i32 (v4i32 QPR:$src1),
1848 (v2i32 (EXTRACT_SUBREG QPR:$src2,
1849 (DSubReg_i32_reg imm:$lane))),
1850 (SubReg_i32_lane imm:$lane)))>;
1851 def : Pat<(v4f32 (fmul (v4f32 QPR:$src1),
1852 (v4f32 (NEONvduplane (v4f32 QPR:$src2), imm:$lane)))),
1853 (v4f32 (VMULslfq (v4f32 QPR:$src1),
1854 (v2f32 (EXTRACT_SUBREG QPR:$src2,
1855 (DSubReg_i32_reg imm:$lane))),
1856 (SubReg_i32_lane imm:$lane)))>;
1858 // VQDMULH : Vector Saturating Doubling Multiply Returning High Half
1859 defm VQDMULH : N3VInt_HS<0, 0, 0b1011, 0, IIC_VMULi16D, IIC_VMULi32D,
1860 IIC_VMULi16Q, IIC_VMULi32Q,
1861 "vqdmulh", "s", int_arm_neon_vqdmulh, 1>;
1862 defm VQDMULHsl: N3VIntSL_HS<0b1100, IIC_VMULi16D, IIC_VMULi32D,
1863 IIC_VMULi16Q, IIC_VMULi32Q,
1864 "vqdmulh", "s", int_arm_neon_vqdmulh>;
1865 def : Pat<(v8i16 (int_arm_neon_vqdmulh (v8i16 QPR:$src1),
1866 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
1868 (v8i16 (VQDMULHslv8i16 (v8i16 QPR:$src1),
1869 (v4i16 (EXTRACT_SUBREG QPR:$src2,
1870 (DSubReg_i16_reg imm:$lane))),
1871 (SubReg_i16_lane imm:$lane)))>;
1872 def : Pat<(v4i32 (int_arm_neon_vqdmulh (v4i32 QPR:$src1),
1873 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
1875 (v4i32 (VQDMULHslv4i32 (v4i32 QPR:$src1),
1876 (v2i32 (EXTRACT_SUBREG QPR:$src2,
1877 (DSubReg_i32_reg imm:$lane))),
1878 (SubReg_i32_lane imm:$lane)))>;
1880 // VQRDMULH : Vector Rounding Saturating Doubling Multiply Returning High Half
1881 defm VQRDMULH : N3VInt_HS<1, 0, 0b1011, 0, IIC_VMULi16D, IIC_VMULi32D,
1882 IIC_VMULi16Q, IIC_VMULi32Q,
1883 "vqrdmulh", "s", int_arm_neon_vqrdmulh, 1>;
1884 defm VQRDMULHsl : N3VIntSL_HS<0b1101, IIC_VMULi16D, IIC_VMULi32D,
1885 IIC_VMULi16Q, IIC_VMULi32Q,
1886 "vqrdmulh", "s", int_arm_neon_vqrdmulh>;
1887 def : Pat<(v8i16 (int_arm_neon_vqrdmulh (v8i16 QPR:$src1),
1888 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
1890 (v8i16 (VQRDMULHslv8i16 (v8i16 QPR:$src1),
1891 (v4i16 (EXTRACT_SUBREG QPR:$src2,
1892 (DSubReg_i16_reg imm:$lane))),
1893 (SubReg_i16_lane imm:$lane)))>;
1894 def : Pat<(v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$src1),
1895 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
1897 (v4i32 (VQRDMULHslv4i32 (v4i32 QPR:$src1),
1898 (v2i32 (EXTRACT_SUBREG QPR:$src2,
1899 (DSubReg_i32_reg imm:$lane))),
1900 (SubReg_i32_lane imm:$lane)))>;
1902 // VMULL : Vector Multiply Long (integer and polynomial) (Q = D * D)
1903 defm VMULLs : N3VLInt_QHS<0,1,0b1100,0, IIC_VMULi16D, "vmull", "s",
1904 int_arm_neon_vmulls, 1>;
1905 defm VMULLu : N3VLInt_QHS<1,1,0b1100,0, IIC_VMULi16D, "vmull", "u",
1906 int_arm_neon_vmullu, 1>;
1907 def VMULLp : N3VLInt<0, 1, 0b00, 0b1110, 0, IIC_VMULi16D, "vmull", "p8",
1908 v8i16, v8i8, int_arm_neon_vmullp, 1>;
1909 defm VMULLsls : N3VLIntSL_HS<0, 0b1010, IIC_VMULi16D, "vmull", "s",
1910 int_arm_neon_vmulls>;
1911 defm VMULLslu : N3VLIntSL_HS<1, 0b1010, IIC_VMULi16D, "vmull", "u",
1912 int_arm_neon_vmullu>;
1914 // VQDMULL : Vector Saturating Doubling Multiply Long (Q = D * D)
1915 defm VQDMULL : N3VLInt_HS<0,1,0b1101,0, IIC_VMULi16D, "vqdmull", "s",
1916 int_arm_neon_vqdmull, 1>;
1917 defm VQDMULLsl: N3VLIntSL_HS<0, 0b1011, IIC_VMULi16D, "vqdmull", "s",
1918 int_arm_neon_vqdmull>;
1920 // Vector Multiply-Accumulate and Multiply-Subtract Operations.
1922 // VMLA : Vector Multiply Accumulate (integer and floating-point)
1923 defm VMLA : N3VMulOp_QHS<0, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
1924 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
1925 def VMLAfd : N3VDMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACD, "vmla", "f32",
1927 def VMLAfq : N3VQMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACQ, "vmla", "f32",
1929 defm VMLAsl : N3VMulOpSL_HS<0b0000, IIC_VMACi16D, IIC_VMACi32D,
1930 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
1931 def VMLAslfd : N3VDMulOpSL<0b10, 0b0001, IIC_VMACD, "vmla", "f32",
1933 def VMLAslfq : N3VQMulOpSL<0b10, 0b0001, IIC_VMACQ, "vmla", "f32",
1934 v4f32, v2f32, fmul, fadd>;
1936 def : Pat<(v8i16 (add (v8i16 QPR:$src1),
1937 (mul (v8i16 QPR:$src2),
1938 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
1939 (v8i16 (VMLAslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
1940 (v4i16 (EXTRACT_SUBREG QPR:$src3,
1941 (DSubReg_i16_reg imm:$lane))),
1942 (SubReg_i16_lane imm:$lane)))>;
1944 def : Pat<(v4i32 (add (v4i32 QPR:$src1),
1945 (mul (v4i32 QPR:$src2),
1946 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
1947 (v4i32 (VMLAslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
1948 (v2i32 (EXTRACT_SUBREG QPR:$src3,
1949 (DSubReg_i32_reg imm:$lane))),
1950 (SubReg_i32_lane imm:$lane)))>;
1952 def : Pat<(v4f32 (fadd (v4f32 QPR:$src1),
1953 (fmul (v4f32 QPR:$src2),
1954 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
1955 (v4f32 (VMLAslfq (v4f32 QPR:$src1),
1957 (v2f32 (EXTRACT_SUBREG QPR:$src3,
1958 (DSubReg_i32_reg imm:$lane))),
1959 (SubReg_i32_lane imm:$lane)))>;
1961 // VMLAL : Vector Multiply Accumulate Long (Q += D * D)
1962 defm VMLALs : N3VLInt3_QHS<0,1,0b1000,0, "vmlal", "s", int_arm_neon_vmlals>;
1963 defm VMLALu : N3VLInt3_QHS<1,1,0b1000,0, "vmlal", "u", int_arm_neon_vmlalu>;
1965 defm VMLALsls : N3VLInt3SL_HS<0, 0b0010, "vmlal", "s", int_arm_neon_vmlals>;
1966 defm VMLALslu : N3VLInt3SL_HS<1, 0b0010, "vmlal", "u", int_arm_neon_vmlalu>;
1968 // VQDMLAL : Vector Saturating Doubling Multiply Accumulate Long (Q += D * D)
1969 defm VQDMLAL : N3VLInt3_HS<0, 1, 0b1001, 0, "vqdmlal", "s",
1970 int_arm_neon_vqdmlal>;
1971 defm VQDMLALsl: N3VLInt3SL_HS<0, 0b0011, "vqdmlal", "s", int_arm_neon_vqdmlal>;
1973 // VMLS : Vector Multiply Subtract (integer and floating-point)
1974 defm VMLS : N3VMulOp_QHS<1, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
1975 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
1976 def VMLSfd : N3VDMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACD, "vmls", "f32",
1978 def VMLSfq : N3VQMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACQ, "vmls", "f32",
1980 defm VMLSsl : N3VMulOpSL_HS<0b0100, IIC_VMACi16D, IIC_VMACi32D,
1981 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
1982 def VMLSslfd : N3VDMulOpSL<0b10, 0b0101, IIC_VMACD, "vmls", "f32",
1984 def VMLSslfq : N3VQMulOpSL<0b10, 0b0101, IIC_VMACQ, "vmls", "f32",
1985 v4f32, v2f32, fmul, fsub>;
1987 def : Pat<(v8i16 (sub (v8i16 QPR:$src1),
1988 (mul (v8i16 QPR:$src2),
1989 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
1990 (v8i16 (VMLSslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
1991 (v4i16 (EXTRACT_SUBREG QPR:$src3,
1992 (DSubReg_i16_reg imm:$lane))),
1993 (SubReg_i16_lane imm:$lane)))>;
1995 def : Pat<(v4i32 (sub (v4i32 QPR:$src1),
1996 (mul (v4i32 QPR:$src2),
1997 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
1998 (v4i32 (VMLSslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
1999 (v2i32 (EXTRACT_SUBREG QPR:$src3,
2000 (DSubReg_i32_reg imm:$lane))),
2001 (SubReg_i32_lane imm:$lane)))>;
2003 def : Pat<(v4f32 (fsub (v4f32 QPR:$src1),
2004 (fmul (v4f32 QPR:$src2),
2005 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
2006 (v4f32 (VMLSslfq (v4f32 QPR:$src1), (v4f32 QPR:$src2),
2007 (v2f32 (EXTRACT_SUBREG QPR:$src3,
2008 (DSubReg_i32_reg imm:$lane))),
2009 (SubReg_i32_lane imm:$lane)))>;
2011 // VMLSL : Vector Multiply Subtract Long (Q -= D * D)
2012 defm VMLSLs : N3VLInt3_QHS<0,1,0b1010,0, "vmlsl", "s", int_arm_neon_vmlsls>;
2013 defm VMLSLu : N3VLInt3_QHS<1,1,0b1010,0, "vmlsl", "u", int_arm_neon_vmlslu>;
2015 defm VMLSLsls : N3VLInt3SL_HS<0, 0b0110, "vmlsl", "s", int_arm_neon_vmlsls>;
2016 defm VMLSLslu : N3VLInt3SL_HS<1, 0b0110, "vmlsl", "u", int_arm_neon_vmlslu>;
2018 // VQDMLSL : Vector Saturating Doubling Multiply Subtract Long (Q -= D * D)
2019 defm VQDMLSL : N3VLInt3_HS<0, 1, 0b1011, 0, "vqdmlsl", "s",
2020 int_arm_neon_vqdmlsl>;
2021 defm VQDMLSLsl: N3VLInt3SL_HS<0, 0b111, "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
2023 // Vector Subtract Operations.
2025 // VSUB : Vector Subtract (integer and floating-point)
2026 defm VSUB : N3V_QHSD<1, 0, 0b1000, 0, IIC_VSUBiD, IIC_VSUBiQ,
2027 "vsub", "i", sub, 0>;
2028 def VSUBfd : N3VD<0, 0, 0b10, 0b1101, 0, IIC_VBIND, "vsub", "f32",
2029 v2f32, v2f32, fsub, 0>;
2030 def VSUBfq : N3VQ<0, 0, 0b10, 0b1101, 0, IIC_VBINQ, "vsub", "f32",
2031 v4f32, v4f32, fsub, 0>;
2032 // VSUBL : Vector Subtract Long (Q = D - D)
2033 defm VSUBLs : N3VLInt_QHS<0,1,0b0010,0, IIC_VSHLiD, "vsubl", "s",
2034 int_arm_neon_vsubls, 1>;
2035 defm VSUBLu : N3VLInt_QHS<1,1,0b0010,0, IIC_VSHLiD, "vsubl", "u",
2036 int_arm_neon_vsublu, 1>;
2037 // VSUBW : Vector Subtract Wide (Q = Q - D)
2038 defm VSUBWs : N3VWInt_QHS<0,1,0b0011,0, "vsubw", "s", int_arm_neon_vsubws, 0>;
2039 defm VSUBWu : N3VWInt_QHS<1,1,0b0011,0, "vsubw", "u", int_arm_neon_vsubwu, 0>;
2040 // VHSUB : Vector Halving Subtract
2041 defm VHSUBs : N3VInt_QHS<0, 0, 0b0010, 0, IIC_VBINi4D, IIC_VBINi4D,
2042 IIC_VBINi4Q, IIC_VBINi4Q,
2043 "vhsub", "s", int_arm_neon_vhsubs, 0>;
2044 defm VHSUBu : N3VInt_QHS<1, 0, 0b0010, 0, IIC_VBINi4D, IIC_VBINi4D,
2045 IIC_VBINi4Q, IIC_VBINi4Q,
2046 "vhsub", "u", int_arm_neon_vhsubu, 0>;
2047 // VQSUB : Vector Saturing Subtract
2048 defm VQSUBs : N3VInt_QHSD<0, 0, 0b0010, 1, IIC_VBINi4D, IIC_VBINi4D,
2049 IIC_VBINi4Q, IIC_VBINi4Q,
2050 "vqsub", "s", int_arm_neon_vqsubs, 0>;
2051 defm VQSUBu : N3VInt_QHSD<1, 0, 0b0010, 1, IIC_VBINi4D, IIC_VBINi4D,
2052 IIC_VBINi4Q, IIC_VBINi4Q,
2053 "vqsub", "u", int_arm_neon_vqsubu, 0>;
2054 // VSUBHN : Vector Subtract and Narrow Returning High Half (D = Q - Q)
2055 defm VSUBHN : N3VNInt_HSD<0,1,0b0110,0, "vsubhn", "i",
2056 int_arm_neon_vsubhn, 0>;
2057 // VRSUBHN : Vector Rounding Subtract and Narrow Returning High Half (D=Q-Q)
2058 defm VRSUBHN : N3VNInt_HSD<1,1,0b0110,0, "vrsubhn", "i",
2059 int_arm_neon_vrsubhn, 0>;
2061 // Vector Comparisons.
2063 // VCEQ : Vector Compare Equal
2064 defm VCEQ : N3V_QHS<1, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2065 IIC_VBINi4Q, "vceq", "i", NEONvceq, 1>;
2066 def VCEQfd : N3VD<0,0,0b00,0b1110,0, IIC_VBIND, "vceq", "f32", v2i32, v2f32,
2068 def VCEQfq : N3VQ<0,0,0b00,0b1110,0, IIC_VBINQ, "vceq", "f32", v4i32, v4f32,
2070 // For disassembly only.
2071 defm VCEQz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00010, 0, "vceq", "i",
2074 // VCGE : Vector Compare Greater Than or Equal
2075 defm VCGEs : N3V_QHS<0, 0, 0b0011, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2076 IIC_VBINi4Q, "vcge", "s", NEONvcge, 0>;
2077 defm VCGEu : N3V_QHS<1, 0, 0b0011, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2078 IIC_VBINi4Q, "vcge", "u", NEONvcgeu, 0>;
2079 def VCGEfd : N3VD<1,0,0b00,0b1110,0, IIC_VBIND, "vcge", "f32",
2080 v2i32, v2f32, NEONvcge, 0>;
2081 def VCGEfq : N3VQ<1,0,0b00,0b1110,0, IIC_VBINQ, "vcge", "f32", v4i32, v4f32,
2083 // For disassembly only.
2084 defm VCGEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00001, 0, "vcge", "s",
2086 // For disassembly only.
2087 defm VCLEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00011, 0, "vcle", "s",
2090 // VCGT : Vector Compare Greater Than
2091 defm VCGTs : N3V_QHS<0, 0, 0b0011, 0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2092 IIC_VBINi4Q, "vcgt", "s", NEONvcgt, 0>;
2093 defm VCGTu : N3V_QHS<1, 0, 0b0011, 0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2094 IIC_VBINi4Q, "vcgt", "u", NEONvcgtu, 0>;
2095 def VCGTfd : N3VD<1,0,0b10,0b1110,0, IIC_VBIND, "vcgt", "f32", v2i32, v2f32,
2097 def VCGTfq : N3VQ<1,0,0b10,0b1110,0, IIC_VBINQ, "vcgt", "f32", v4i32, v4f32,
2099 // For disassembly only.
2100 defm VCGTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00000, 0, "vcgt", "s",
2102 // For disassembly only.
2103 defm VCLTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00100, 0, "vclt", "s",
2106 // VACGE : Vector Absolute Compare Greater Than or Equal (aka VCAGE)
2107 def VACGEd : N3VDInt<1, 0, 0b00, 0b1110, 1, IIC_VBIND, "vacge", "f32",
2108 v2i32, v2f32, int_arm_neon_vacged, 0>;
2109 def VACGEq : N3VQInt<1, 0, 0b00, 0b1110, 1, IIC_VBINQ, "vacge", "f32",
2110 v4i32, v4f32, int_arm_neon_vacgeq, 0>;
2111 // VACGT : Vector Absolute Compare Greater Than (aka VCAGT)
2112 def VACGTd : N3VDInt<1, 0, 0b10, 0b1110, 1, IIC_VBIND, "vacgt", "f32",
2113 v2i32, v2f32, int_arm_neon_vacgtd, 0>;
2114 def VACGTq : N3VQInt<1, 0, 0b10, 0b1110, 1, IIC_VBINQ, "vacgt", "f32",
2115 v4i32, v4f32, int_arm_neon_vacgtq, 0>;
2116 // VTST : Vector Test Bits
2117 defm VTST : N3V_QHS<0, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2118 IIC_VBINi4Q, "vtst", "", NEONvtst, 1>;
2120 // Vector Bitwise Operations.
2122 // VAND : Vector Bitwise AND
2123 def VANDd : N3VDX<0, 0, 0b00, 0b0001, 1, IIC_VBINiD, "vand",
2124 v2i32, v2i32, and, 1>;
2125 def VANDq : N3VQX<0, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "vand",
2126 v4i32, v4i32, and, 1>;
2128 // VEOR : Vector Bitwise Exclusive OR
2129 def VEORd : N3VDX<1, 0, 0b00, 0b0001, 1, IIC_VBINiD, "veor",
2130 v2i32, v2i32, xor, 1>;
2131 def VEORq : N3VQX<1, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "veor",
2132 v4i32, v4i32, xor, 1>;
2134 // VORR : Vector Bitwise OR
2135 def VORRd : N3VDX<0, 0, 0b10, 0b0001, 1, IIC_VBINiD, "vorr",
2136 v2i32, v2i32, or, 1>;
2137 def VORRq : N3VQX<0, 0, 0b10, 0b0001, 1, IIC_VBINiQ, "vorr",
2138 v4i32, v4i32, or, 1>;
2140 // VBIC : Vector Bitwise Bit Clear (AND NOT)
2141 def VBICd : N3VX<0, 0, 0b01, 0b0001, 0, 1, (outs DPR:$dst),
2142 (ins DPR:$src1, DPR:$src2), IIC_VBINiD,
2143 "vbic", "$dst, $src1, $src2", "",
2144 [(set DPR:$dst, (v2i32 (and DPR:$src1,
2145 (vnot_conv DPR:$src2))))]>;
2146 def VBICq : N3VX<0, 0, 0b01, 0b0001, 1, 1, (outs QPR:$dst),
2147 (ins QPR:$src1, QPR:$src2), IIC_VBINiQ,
2148 "vbic", "$dst, $src1, $src2", "",
2149 [(set QPR:$dst, (v4i32 (and QPR:$src1,
2150 (vnot_conv QPR:$src2))))]>;
2152 // VORN : Vector Bitwise OR NOT
2153 def VORNd : N3VX<0, 0, 0b11, 0b0001, 0, 1, (outs DPR:$dst),
2154 (ins DPR:$src1, DPR:$src2), IIC_VBINiD,
2155 "vorn", "$dst, $src1, $src2", "",
2156 [(set DPR:$dst, (v2i32 (or DPR:$src1,
2157 (vnot_conv DPR:$src2))))]>;
2158 def VORNq : N3VX<0, 0, 0b11, 0b0001, 1, 1, (outs QPR:$dst),
2159 (ins QPR:$src1, QPR:$src2), IIC_VBINiQ,
2160 "vorn", "$dst, $src1, $src2", "",
2161 [(set QPR:$dst, (v4i32 (or QPR:$src1,
2162 (vnot_conv QPR:$src2))))]>;
2164 // VMVN : Vector Bitwise NOT
2165 def VMVNd : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 0, 0,
2166 (outs DPR:$dst), (ins DPR:$src), IIC_VSHLiD,
2167 "vmvn", "$dst, $src", "",
2168 [(set DPR:$dst, (v2i32 (vnot DPR:$src)))]>;
2169 def VMVNq : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 1, 0,
2170 (outs QPR:$dst), (ins QPR:$src), IIC_VSHLiD,
2171 "vmvn", "$dst, $src", "",
2172 [(set QPR:$dst, (v4i32 (vnot QPR:$src)))]>;
2173 def : Pat<(v2i32 (vnot_conv DPR:$src)), (VMVNd DPR:$src)>;
2174 def : Pat<(v4i32 (vnot_conv QPR:$src)), (VMVNq QPR:$src)>;
2176 // VBSL : Vector Bitwise Select
2177 def VBSLd : N3VX<1, 0, 0b01, 0b0001, 0, 1, (outs DPR:$dst),
2178 (ins DPR:$src1, DPR:$src2, DPR:$src3), IIC_VCNTiD,
2179 "vbsl", "$dst, $src2, $src3", "$src1 = $dst",
2181 (v2i32 (or (and DPR:$src2, DPR:$src1),
2182 (and DPR:$src3, (vnot_conv DPR:$src1)))))]>;
2183 def VBSLq : N3VX<1, 0, 0b01, 0b0001, 1, 1, (outs QPR:$dst),
2184 (ins QPR:$src1, QPR:$src2, QPR:$src3), IIC_VCNTiQ,
2185 "vbsl", "$dst, $src2, $src3", "$src1 = $dst",
2187 (v4i32 (or (and QPR:$src2, QPR:$src1),
2188 (and QPR:$src3, (vnot_conv QPR:$src1)))))]>;
2190 // VBIF : Vector Bitwise Insert if False
2191 // like VBSL but with: "vbif $dst, $src3, $src1", "$src2 = $dst",
2192 def VBIFd : N3VX<1, 0, 0b11, 0b0001, 0, 1,
2193 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3),
2194 IIC_VBINiD, "vbif", "$dst, $src2, $src3", "$src1 = $dst",
2195 [/* For disassembly only; pattern left blank */]>;
2196 def VBIFq : N3VX<1, 0, 0b11, 0b0001, 1, 1,
2197 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3),
2198 IIC_VBINiQ, "vbif", "$dst, $src2, $src3", "$src1 = $dst",
2199 [/* For disassembly only; pattern left blank */]>;
2201 // VBIT : Vector Bitwise Insert if True
2202 // like VBSL but with: "vbit $dst, $src2, $src1", "$src3 = $dst",
2203 def VBITd : N3VX<1, 0, 0b10, 0b0001, 0, 1,
2204 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3),
2205 IIC_VBINiD, "vbit", "$dst, $src2, $src3", "$src1 = $dst",
2206 [/* For disassembly only; pattern left blank */]>;
2207 def VBITq : N3VX<1, 0, 0b10, 0b0001, 1, 1,
2208 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3),
2209 IIC_VBINiQ, "vbit", "$dst, $src2, $src3", "$src1 = $dst",
2210 [/* For disassembly only; pattern left blank */]>;
2212 // VBIT/VBIF are not yet implemented. The TwoAddress pass will not go looking
2213 // for equivalent operations with different register constraints; it just
2216 // Vector Absolute Differences.
2218 // VABD : Vector Absolute Difference
2219 defm VABDs : N3VInt_QHS<0, 0, 0b0111, 0, IIC_VBINi4D, IIC_VBINi4D,
2220 IIC_VBINi4Q, IIC_VBINi4Q,
2221 "vabd", "s", int_arm_neon_vabds, 0>;
2222 defm VABDu : N3VInt_QHS<1, 0, 0b0111, 0, IIC_VBINi4D, IIC_VBINi4D,
2223 IIC_VBINi4Q, IIC_VBINi4Q,
2224 "vabd", "u", int_arm_neon_vabdu, 0>;
2225 def VABDfd : N3VDInt<1, 0, 0b10, 0b1101, 0, IIC_VBIND,
2226 "vabd", "f32", v2f32, v2f32, int_arm_neon_vabds, 0>;
2227 def VABDfq : N3VQInt<1, 0, 0b10, 0b1101, 0, IIC_VBINQ,
2228 "vabd", "f32", v4f32, v4f32, int_arm_neon_vabds, 0>;
2230 // VABDL : Vector Absolute Difference Long (Q = | D - D |)
2231 defm VABDLs : N3VLInt_QHS<0,1,0b0111,0, IIC_VBINi4Q,
2232 "vabdl", "s", int_arm_neon_vabdls, 0>;
2233 defm VABDLu : N3VLInt_QHS<1,1,0b0111,0, IIC_VBINi4Q,
2234 "vabdl", "u", int_arm_neon_vabdlu, 0>;
2236 // VABA : Vector Absolute Difference and Accumulate
2237 defm VABAs : N3VInt3_QHS<0,0,0b0111,1, "vaba", "s", int_arm_neon_vabas>;
2238 defm VABAu : N3VInt3_QHS<1,0,0b0111,1, "vaba", "u", int_arm_neon_vabau>;
2240 // VABAL : Vector Absolute Difference and Accumulate Long (Q += | D - D |)
2241 defm VABALs : N3VLInt3_QHS<0,1,0b0101,0, "vabal", "s", int_arm_neon_vabals>;
2242 defm VABALu : N3VLInt3_QHS<1,1,0b0101,0, "vabal", "u", int_arm_neon_vabalu>;
2244 // Vector Maximum and Minimum.
2246 // VMAX : Vector Maximum
2247 defm VMAXs : N3VInt_QHS<0,0,0b0110,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2248 IIC_VBINi4Q, "vmax", "s", int_arm_neon_vmaxs, 1>;
2249 defm VMAXu : N3VInt_QHS<1,0,0b0110,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2250 IIC_VBINi4Q, "vmax", "u", int_arm_neon_vmaxu, 1>;
2251 def VMAXfd : N3VDInt<0, 0, 0b00, 0b1111, 0, IIC_VBIND, "vmax", "f32",
2252 v2f32, v2f32, int_arm_neon_vmaxs, 1>;
2253 def VMAXfq : N3VQInt<0, 0, 0b00, 0b1111, 0, IIC_VBINQ, "vmax", "f32",
2254 v4f32, v4f32, int_arm_neon_vmaxs, 1>;
2256 // VMIN : Vector Minimum
2257 defm VMINs : N3VInt_QHS<0,0,0b0110,1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2258 IIC_VBINi4Q, "vmin", "s", int_arm_neon_vmins, 1>;
2259 defm VMINu : N3VInt_QHS<1,0,0b0110,1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2260 IIC_VBINi4Q, "vmin", "u", int_arm_neon_vminu, 1>;
2261 def VMINfd : N3VDInt<0, 0, 0b10, 0b1111, 0, IIC_VBIND, "vmin", "f32",
2262 v2f32, v2f32, int_arm_neon_vmins, 1>;
2263 def VMINfq : N3VQInt<0, 0, 0b10, 0b1111, 0, IIC_VBINQ, "vmin", "f32",
2264 v4f32, v4f32, int_arm_neon_vmins, 1>;
2266 // Vector Pairwise Operations.
2268 // VPADD : Vector Pairwise Add
2269 def VPADDi8 : N3VDInt<0, 0, 0b00, 0b1011, 1, IIC_VBINiD, "vpadd", "i8",
2270 v8i8, v8i8, int_arm_neon_vpadd, 0>;
2271 def VPADDi16 : N3VDInt<0, 0, 0b01, 0b1011, 1, IIC_VBINiD, "vpadd", "i16",
2272 v4i16, v4i16, int_arm_neon_vpadd, 0>;
2273 def VPADDi32 : N3VDInt<0, 0, 0b10, 0b1011, 1, IIC_VBINiD, "vpadd", "i32",
2274 v2i32, v2i32, int_arm_neon_vpadd, 0>;
2275 def VPADDf : N3VDInt<1, 0, 0b00, 0b1101, 0, IIC_VBIND, "vpadd", "f32",
2276 v2f32, v2f32, int_arm_neon_vpadd, 0>;
2278 // VPADDL : Vector Pairwise Add Long
2279 defm VPADDLs : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00100, 0, "vpaddl", "s",
2280 int_arm_neon_vpaddls>;
2281 defm VPADDLu : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00101, 0, "vpaddl", "u",
2282 int_arm_neon_vpaddlu>;
2284 // VPADAL : Vector Pairwise Add and Accumulate Long
2285 defm VPADALs : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01100, 0, "vpadal", "s",
2286 int_arm_neon_vpadals>;
2287 defm VPADALu : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01101, 0, "vpadal", "u",
2288 int_arm_neon_vpadalu>;
2290 // VPMAX : Vector Pairwise Maximum
2291 def VPMAXs8 : N3VDInt<0, 0, 0b00, 0b1010, 0, IIC_VBINi4D, "vpmax", "s8",
2292 v8i8, v8i8, int_arm_neon_vpmaxs, 0>;
2293 def VPMAXs16 : N3VDInt<0, 0, 0b01, 0b1010, 0, IIC_VBINi4D, "vpmax", "s16",
2294 v4i16, v4i16, int_arm_neon_vpmaxs, 0>;
2295 def VPMAXs32 : N3VDInt<0, 0, 0b10, 0b1010, 0, IIC_VBINi4D, "vpmax", "s32",
2296 v2i32, v2i32, int_arm_neon_vpmaxs, 0>;
2297 def VPMAXu8 : N3VDInt<1, 0, 0b00, 0b1010, 0, IIC_VBINi4D, "vpmax", "u8",
2298 v8i8, v8i8, int_arm_neon_vpmaxu, 0>;
2299 def VPMAXu16 : N3VDInt<1, 0, 0b01, 0b1010, 0, IIC_VBINi4D, "vpmax", "u16",
2300 v4i16, v4i16, int_arm_neon_vpmaxu, 0>;
2301 def VPMAXu32 : N3VDInt<1, 0, 0b10, 0b1010, 0, IIC_VBINi4D, "vpmax", "u32",
2302 v2i32, v2i32, int_arm_neon_vpmaxu, 0>;
2303 def VPMAXf : N3VDInt<1, 0, 0b00, 0b1111, 0, IIC_VBINi4D, "vpmax", "f32",
2304 v2f32, v2f32, int_arm_neon_vpmaxs, 0>;
2306 // VPMIN : Vector Pairwise Minimum
2307 def VPMINs8 : N3VDInt<0, 0, 0b00, 0b1010, 1, IIC_VBINi4D, "vpmin", "s8",
2308 v8i8, v8i8, int_arm_neon_vpmins, 0>;
2309 def VPMINs16 : N3VDInt<0, 0, 0b01, 0b1010, 1, IIC_VBINi4D, "vpmin", "s16",
2310 v4i16, v4i16, int_arm_neon_vpmins, 0>;
2311 def VPMINs32 : N3VDInt<0, 0, 0b10, 0b1010, 1, IIC_VBINi4D, "vpmin", "s32",
2312 v2i32, v2i32, int_arm_neon_vpmins, 0>;
2313 def VPMINu8 : N3VDInt<1, 0, 0b00, 0b1010, 1, IIC_VBINi4D, "vpmin", "u8",
2314 v8i8, v8i8, int_arm_neon_vpminu, 0>;
2315 def VPMINu16 : N3VDInt<1, 0, 0b01, 0b1010, 1, IIC_VBINi4D, "vpmin", "u16",
2316 v4i16, v4i16, int_arm_neon_vpminu, 0>;
2317 def VPMINu32 : N3VDInt<1, 0, 0b10, 0b1010, 1, IIC_VBINi4D, "vpmin", "u32",
2318 v2i32, v2i32, int_arm_neon_vpminu, 0>;
2319 def VPMINf : N3VDInt<1, 0, 0b10, 0b1111, 0, IIC_VBINi4D, "vpmin", "f32",
2320 v2f32, v2f32, int_arm_neon_vpmins, 0>;
2322 // Vector Reciprocal and Reciprocal Square Root Estimate and Step.
2324 // VRECPE : Vector Reciprocal Estimate
2325 def VRECPEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
2326 IIC_VUNAD, "vrecpe", "u32",
2327 v2i32, v2i32, int_arm_neon_vrecpe>;
2328 def VRECPEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
2329 IIC_VUNAQ, "vrecpe", "u32",
2330 v4i32, v4i32, int_arm_neon_vrecpe>;
2331 def VRECPEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
2332 IIC_VUNAD, "vrecpe", "f32",
2333 v2f32, v2f32, int_arm_neon_vrecpe>;
2334 def VRECPEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
2335 IIC_VUNAQ, "vrecpe", "f32",
2336 v4f32, v4f32, int_arm_neon_vrecpe>;
2338 // VRECPS : Vector Reciprocal Step
2339 def VRECPSfd : N3VDInt<0, 0, 0b00, 0b1111, 1,
2340 IIC_VRECSD, "vrecps", "f32",
2341 v2f32, v2f32, int_arm_neon_vrecps, 1>;
2342 def VRECPSfq : N3VQInt<0, 0, 0b00, 0b1111, 1,
2343 IIC_VRECSQ, "vrecps", "f32",
2344 v4f32, v4f32, int_arm_neon_vrecps, 1>;
2346 // VRSQRTE : Vector Reciprocal Square Root Estimate
2347 def VRSQRTEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
2348 IIC_VUNAD, "vrsqrte", "u32",
2349 v2i32, v2i32, int_arm_neon_vrsqrte>;
2350 def VRSQRTEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
2351 IIC_VUNAQ, "vrsqrte", "u32",
2352 v4i32, v4i32, int_arm_neon_vrsqrte>;
2353 def VRSQRTEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
2354 IIC_VUNAD, "vrsqrte", "f32",
2355 v2f32, v2f32, int_arm_neon_vrsqrte>;
2356 def VRSQRTEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
2357 IIC_VUNAQ, "vrsqrte", "f32",
2358 v4f32, v4f32, int_arm_neon_vrsqrte>;
2360 // VRSQRTS : Vector Reciprocal Square Root Step
2361 def VRSQRTSfd : N3VDInt<0, 0, 0b10, 0b1111, 1,
2362 IIC_VRECSD, "vrsqrts", "f32",
2363 v2f32, v2f32, int_arm_neon_vrsqrts, 1>;
2364 def VRSQRTSfq : N3VQInt<0, 0, 0b10, 0b1111, 1,
2365 IIC_VRECSQ, "vrsqrts", "f32",
2366 v4f32, v4f32, int_arm_neon_vrsqrts, 1>;
2370 // VSHL : Vector Shift
2371 defm VSHLs : N3VInt_QHSD<0, 0, 0b0100, 0, IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ,
2372 IIC_VSHLiQ, "vshl", "s", int_arm_neon_vshifts, 0>;
2373 defm VSHLu : N3VInt_QHSD<1, 0, 0b0100, 0, IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ,
2374 IIC_VSHLiQ, "vshl", "u", int_arm_neon_vshiftu, 0>;
2375 // VSHL : Vector Shift Left (Immediate)
2376 defm VSHLi : N2VSh_QHSD<0, 1, 0b0101, 1, IIC_VSHLiD, "vshl", "i", NEONvshl>;
2377 // VSHR : Vector Shift Right (Immediate)
2378 defm VSHRs : N2VSh_QHSD<0, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "s", NEONvshrs>;
2379 defm VSHRu : N2VSh_QHSD<1, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "u", NEONvshru>;
2381 // VSHLL : Vector Shift Left Long
2382 defm VSHLLs : N2VLSh_QHS<0, 1, 0b1010, 0, 0, 1, "vshll", "s", NEONvshlls>;
2383 defm VSHLLu : N2VLSh_QHS<1, 1, 0b1010, 0, 0, 1, "vshll", "u", NEONvshllu>;
2385 // VSHLL : Vector Shift Left Long (with maximum shift count)
2386 class N2VLShMax<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
2387 bit op6, bit op4, string OpcodeStr, string Dt, ValueType ResTy,
2388 ValueType OpTy, SDNode OpNode>
2389 : N2VLSh<op24, op23, op11_8, op7, op6, op4, OpcodeStr, Dt,
2390 ResTy, OpTy, OpNode> {
2391 let Inst{21-16} = op21_16;
2393 def VSHLLi8 : N2VLShMax<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll", "i8",
2394 v8i16, v8i8, NEONvshlli>;
2395 def VSHLLi16 : N2VLShMax<1, 1, 0b110110, 0b0011, 0, 0, 0, "vshll", "i16",
2396 v4i32, v4i16, NEONvshlli>;
2397 def VSHLLi32 : N2VLShMax<1, 1, 0b111010, 0b0011, 0, 0, 0, "vshll", "i32",
2398 v2i64, v2i32, NEONvshlli>;
2400 // VSHRN : Vector Shift Right and Narrow
2401 defm VSHRN : N2VNSh_HSD<0,1,0b1000,0,0,1, IIC_VSHLiD, "vshrn", "i",
2404 // VRSHL : Vector Rounding Shift
2405 defm VRSHLs : N3VInt_QHSD<0,0,0b0101,0, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2406 IIC_VSHLi4Q, "vrshl", "s", int_arm_neon_vrshifts,0>;
2407 defm VRSHLu : N3VInt_QHSD<1,0,0b0101,0, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2408 IIC_VSHLi4Q, "vrshl", "u", int_arm_neon_vrshiftu,0>;
2409 // VRSHR : Vector Rounding Shift Right
2410 defm VRSHRs : N2VSh_QHSD<0,1,0b0010,1, IIC_VSHLi4D, "vrshr", "s", NEONvrshrs>;
2411 defm VRSHRu : N2VSh_QHSD<1,1,0b0010,1, IIC_VSHLi4D, "vrshr", "u", NEONvrshru>;
2413 // VRSHRN : Vector Rounding Shift Right and Narrow
2414 defm VRSHRN : N2VNSh_HSD<0, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vrshrn", "i",
2417 // VQSHL : Vector Saturating Shift
2418 defm VQSHLs : N3VInt_QHSD<0,0,0b0100,1, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2419 IIC_VSHLi4Q, "vqshl", "s", int_arm_neon_vqshifts,0>;
2420 defm VQSHLu : N3VInt_QHSD<1,0,0b0100,1, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2421 IIC_VSHLi4Q, "vqshl", "u", int_arm_neon_vqshiftu,0>;
2422 // VQSHL : Vector Saturating Shift Left (Immediate)
2423 defm VQSHLsi : N2VSh_QHSD<0,1,0b0111,1, IIC_VSHLi4D, "vqshl", "s", NEONvqshls>;
2424 defm VQSHLui : N2VSh_QHSD<1,1,0b0111,1, IIC_VSHLi4D, "vqshl", "u", NEONvqshlu>;
2425 // VQSHLU : Vector Saturating Shift Left (Immediate, Unsigned)
2426 defm VQSHLsu : N2VSh_QHSD<1,1,0b0110,1, IIC_VSHLi4D, "vqshlu","s",NEONvqshlsu>;
2428 // VQSHRN : Vector Saturating Shift Right and Narrow
2429 defm VQSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "s",
2431 defm VQSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "u",
2434 // VQSHRUN : Vector Saturating Shift Right and Narrow (Unsigned)
2435 defm VQSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 0, 1, IIC_VSHLi4D, "vqshrun", "s",
2438 // VQRSHL : Vector Saturating Rounding Shift
2439 defm VQRSHLs : N3VInt_QHSD<0,0,0b0101,1, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2440 IIC_VSHLi4Q, "vqrshl", "s",
2441 int_arm_neon_vqrshifts, 0>;
2442 defm VQRSHLu : N3VInt_QHSD<1,0,0b0101,1, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2443 IIC_VSHLi4Q, "vqrshl", "u",
2444 int_arm_neon_vqrshiftu, 0>;
2446 // VQRSHRN : Vector Saturating Rounding Shift Right and Narrow
2447 defm VQRSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "s",
2449 defm VQRSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "u",
2452 // VQRSHRUN : Vector Saturating Rounding Shift Right and Narrow (Unsigned)
2453 defm VQRSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vqrshrun", "s",
2456 // VSRA : Vector Shift Right and Accumulate
2457 defm VSRAs : N2VShAdd_QHSD<0, 1, 0b0001, 1, "vsra", "s", NEONvshrs>;
2458 defm VSRAu : N2VShAdd_QHSD<1, 1, 0b0001, 1, "vsra", "u", NEONvshru>;
2459 // VRSRA : Vector Rounding Shift Right and Accumulate
2460 defm VRSRAs : N2VShAdd_QHSD<0, 1, 0b0011, 1, "vrsra", "s", NEONvrshrs>;
2461 defm VRSRAu : N2VShAdd_QHSD<1, 1, 0b0011, 1, "vrsra", "u", NEONvrshru>;
2463 // VSLI : Vector Shift Left and Insert
2464 defm VSLI : N2VShIns_QHSD<1, 1, 0b0101, 1, "vsli", NEONvsli>;
2465 // VSRI : Vector Shift Right and Insert
2466 defm VSRI : N2VShIns_QHSD<1, 1, 0b0100, 1, "vsri", NEONvsri>;
2468 // Vector Absolute and Saturating Absolute.
2470 // VABS : Vector Absolute Value
2471 defm VABS : N2VInt_QHS<0b11, 0b11, 0b01, 0b00110, 0,
2472 IIC_VUNAiD, IIC_VUNAiQ, "vabs", "s",
2474 def VABSfd : N2VDInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
2475 IIC_VUNAD, "vabs", "f32",
2476 v2f32, v2f32, int_arm_neon_vabs>;
2477 def VABSfq : N2VQInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
2478 IIC_VUNAQ, "vabs", "f32",
2479 v4f32, v4f32, int_arm_neon_vabs>;
2481 // VQABS : Vector Saturating Absolute Value
2482 defm VQABS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01110, 0,
2483 IIC_VQUNAiD, IIC_VQUNAiQ, "vqabs", "s",
2484 int_arm_neon_vqabs>;
2488 def vneg : PatFrag<(ops node:$in), (sub immAllZerosV, node:$in)>;
2489 def vneg_conv : PatFrag<(ops node:$in), (sub immAllZerosV_bc, node:$in)>;
2491 class VNEGD<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
2492 : N2V<0b11, 0b11, size, 0b01, 0b00111, 0, 0, (outs DPR:$dst), (ins DPR:$src),
2493 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
2494 [(set DPR:$dst, (Ty (vneg DPR:$src)))]>;
2495 class VNEGQ<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
2496 : N2V<0b11, 0b11, size, 0b01, 0b00111, 1, 0, (outs QPR:$dst), (ins QPR:$src),
2497 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
2498 [(set QPR:$dst, (Ty (vneg QPR:$src)))]>;
2500 // VNEG : Vector Negate
2501 def VNEGs8d : VNEGD<0b00, "vneg", "s8", v8i8>;
2502 def VNEGs16d : VNEGD<0b01, "vneg", "s16", v4i16>;
2503 def VNEGs32d : VNEGD<0b10, "vneg", "s32", v2i32>;
2504 def VNEGs8q : VNEGQ<0b00, "vneg", "s8", v16i8>;
2505 def VNEGs16q : VNEGQ<0b01, "vneg", "s16", v8i16>;
2506 def VNEGs32q : VNEGQ<0b10, "vneg", "s32", v4i32>;
2508 // VNEG : Vector Negate (floating-point)
2509 def VNEGfd : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
2510 (outs DPR:$dst), (ins DPR:$src), IIC_VUNAD,
2511 "vneg", "f32", "$dst, $src", "",
2512 [(set DPR:$dst, (v2f32 (fneg DPR:$src)))]>;
2513 def VNEGf32q : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 1, 0,
2514 (outs QPR:$dst), (ins QPR:$src), IIC_VUNAQ,
2515 "vneg", "f32", "$dst, $src", "",
2516 [(set QPR:$dst, (v4f32 (fneg QPR:$src)))]>;
2518 def : Pat<(v8i8 (vneg_conv DPR:$src)), (VNEGs8d DPR:$src)>;
2519 def : Pat<(v4i16 (vneg_conv DPR:$src)), (VNEGs16d DPR:$src)>;
2520 def : Pat<(v2i32 (vneg_conv DPR:$src)), (VNEGs32d DPR:$src)>;
2521 def : Pat<(v16i8 (vneg_conv QPR:$src)), (VNEGs8q QPR:$src)>;
2522 def : Pat<(v8i16 (vneg_conv QPR:$src)), (VNEGs16q QPR:$src)>;
2523 def : Pat<(v4i32 (vneg_conv QPR:$src)), (VNEGs32q QPR:$src)>;
2525 // VQNEG : Vector Saturating Negate
2526 defm VQNEG : N2VInt_QHS<0b11, 0b11, 0b00, 0b01111, 0,
2527 IIC_VQUNAiD, IIC_VQUNAiQ, "vqneg", "s",
2528 int_arm_neon_vqneg>;
2530 // Vector Bit Counting Operations.
2532 // VCLS : Vector Count Leading Sign Bits
2533 defm VCLS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01000, 0,
2534 IIC_VCNTiD, IIC_VCNTiQ, "vcls", "s",
2536 // VCLZ : Vector Count Leading Zeros
2537 defm VCLZ : N2VInt_QHS<0b11, 0b11, 0b00, 0b01001, 0,
2538 IIC_VCNTiD, IIC_VCNTiQ, "vclz", "i",
2540 // VCNT : Vector Count One Bits
2541 def VCNTd : N2VDInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
2542 IIC_VCNTiD, "vcnt", "8",
2543 v8i8, v8i8, int_arm_neon_vcnt>;
2544 def VCNTq : N2VQInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
2545 IIC_VCNTiQ, "vcnt", "8",
2546 v16i8, v16i8, int_arm_neon_vcnt>;
2548 // Vector Swap -- for disassembly only.
2549 def VSWPd : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 0, 0,
2550 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
2551 "vswp", "$dst, $src", "", []>;
2552 def VSWPq : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 1, 0,
2553 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
2554 "vswp", "$dst, $src", "", []>;
2556 // Vector Move Operations.
2558 // VMOV : Vector Move (Register)
2560 def VMOVDneon: N3VX<0, 0, 0b10, 0b0001, 0, 1, (outs DPR:$dst), (ins DPR:$src),
2561 IIC_VMOVD, "vmov", "$dst, $src", "", []>;
2562 def VMOVQ : N3VX<0, 0, 0b10, 0b0001, 1, 1, (outs QPR:$dst), (ins QPR:$src),
2563 IIC_VMOVD, "vmov", "$dst, $src", "", []>;
2565 // VMOV : Vector Move (Immediate)
2567 // VMOV_get_imm8 xform function: convert build_vector to VMOV.i8 imm.
2568 def VMOV_get_imm8 : SDNodeXForm<build_vector, [{
2569 return ARM::getVMOVImm(N, 1, *CurDAG);
2571 def vmovImm8 : PatLeaf<(build_vector), [{
2572 return ARM::getVMOVImm(N, 1, *CurDAG).getNode() != 0;
2575 // VMOV_get_imm16 xform function: convert build_vector to VMOV.i16 imm.
2576 def VMOV_get_imm16 : SDNodeXForm<build_vector, [{
2577 return ARM::getVMOVImm(N, 2, *CurDAG);
2579 def vmovImm16 : PatLeaf<(build_vector), [{
2580 return ARM::getVMOVImm(N, 2, *CurDAG).getNode() != 0;
2581 }], VMOV_get_imm16>;
2583 // VMOV_get_imm32 xform function: convert build_vector to VMOV.i32 imm.
2584 def VMOV_get_imm32 : SDNodeXForm<build_vector, [{
2585 return ARM::getVMOVImm(N, 4, *CurDAG);
2587 def vmovImm32 : PatLeaf<(build_vector), [{
2588 return ARM::getVMOVImm(N, 4, *CurDAG).getNode() != 0;
2589 }], VMOV_get_imm32>;
2591 // VMOV_get_imm64 xform function: convert build_vector to VMOV.i64 imm.
2592 def VMOV_get_imm64 : SDNodeXForm<build_vector, [{
2593 return ARM::getVMOVImm(N, 8, *CurDAG);
2595 def vmovImm64 : PatLeaf<(build_vector), [{
2596 return ARM::getVMOVImm(N, 8, *CurDAG).getNode() != 0;
2597 }], VMOV_get_imm64>;
2599 // Note: Some of the cmode bits in the following VMOV instructions need to
2600 // be encoded based on the immed values.
2602 def VMOVv8i8 : N1ModImm<1, 0b000, 0b1110, 0, 0, 0, 1, (outs DPR:$dst),
2603 (ins h8imm:$SIMM), IIC_VMOVImm,
2604 "vmov", "i8", "$dst, $SIMM", "",
2605 [(set DPR:$dst, (v8i8 vmovImm8:$SIMM))]>;
2606 def VMOVv16i8 : N1ModImm<1, 0b000, 0b1110, 0, 1, 0, 1, (outs QPR:$dst),
2607 (ins h8imm:$SIMM), IIC_VMOVImm,
2608 "vmov", "i8", "$dst, $SIMM", "",
2609 [(set QPR:$dst, (v16i8 vmovImm8:$SIMM))]>;
2611 def VMOVv4i16 : N1ModImm<1, 0b000, {1,0,?,?}, 0, 0, {?}, 1, (outs DPR:$dst),
2612 (ins h16imm:$SIMM), IIC_VMOVImm,
2613 "vmov", "i16", "$dst, $SIMM", "",
2614 [(set DPR:$dst, (v4i16 vmovImm16:$SIMM))]>;
2615 def VMOVv8i16 : N1ModImm<1, 0b000, {1,0,?,?}, 0, 1, {?}, 1, (outs QPR:$dst),
2616 (ins h16imm:$SIMM), IIC_VMOVImm,
2617 "vmov", "i16", "$dst, $SIMM", "",
2618 [(set QPR:$dst, (v8i16 vmovImm16:$SIMM))]>;
2620 def VMOVv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, {?}, 1, (outs DPR:$dst),
2621 (ins h32imm:$SIMM), IIC_VMOVImm,
2622 "vmov", "i32", "$dst, $SIMM", "",
2623 [(set DPR:$dst, (v2i32 vmovImm32:$SIMM))]>;
2624 def VMOVv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, {?}, 1, (outs QPR:$dst),
2625 (ins h32imm:$SIMM), IIC_VMOVImm,
2626 "vmov", "i32", "$dst, $SIMM", "",
2627 [(set QPR:$dst, (v4i32 vmovImm32:$SIMM))]>;
2629 def VMOVv1i64 : N1ModImm<1, 0b000, 0b1110, 0, 0, 1, 1, (outs DPR:$dst),
2630 (ins h64imm:$SIMM), IIC_VMOVImm,
2631 "vmov", "i64", "$dst, $SIMM", "",
2632 [(set DPR:$dst, (v1i64 vmovImm64:$SIMM))]>;
2633 def VMOVv2i64 : N1ModImm<1, 0b000, 0b1110, 0, 1, 1, 1, (outs QPR:$dst),
2634 (ins h64imm:$SIMM), IIC_VMOVImm,
2635 "vmov", "i64", "$dst, $SIMM", "",
2636 [(set QPR:$dst, (v2i64 vmovImm64:$SIMM))]>;
2638 // VMOV : Vector Get Lane (move scalar to ARM core register)
2640 def VGETLNs8 : NVGetLane<{1,1,1,0,0,1,?,1}, 0b1011, {?,?},
2641 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2642 IIC_VMOVSI, "vmov", "s8", "$dst, $src[$lane]",
2643 [(set GPR:$dst, (NEONvgetlanes (v8i8 DPR:$src),
2645 def VGETLNs16 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, {?,1},
2646 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2647 IIC_VMOVSI, "vmov", "s16", "$dst, $src[$lane]",
2648 [(set GPR:$dst, (NEONvgetlanes (v4i16 DPR:$src),
2650 def VGETLNu8 : NVGetLane<{1,1,1,0,1,1,?,1}, 0b1011, {?,?},
2651 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2652 IIC_VMOVSI, "vmov", "u8", "$dst, $src[$lane]",
2653 [(set GPR:$dst, (NEONvgetlaneu (v8i8 DPR:$src),
2655 def VGETLNu16 : NVGetLane<{1,1,1,0,1,0,?,1}, 0b1011, {?,1},
2656 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2657 IIC_VMOVSI, "vmov", "u16", "$dst, $src[$lane]",
2658 [(set GPR:$dst, (NEONvgetlaneu (v4i16 DPR:$src),
2660 def VGETLNi32 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, 0b00,
2661 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2662 IIC_VMOVSI, "vmov", "32", "$dst, $src[$lane]",
2663 [(set GPR:$dst, (extractelt (v2i32 DPR:$src),
2665 // def VGETLNf32: see FMRDH and FMRDL in ARMInstrVFP.td
2666 def : Pat<(NEONvgetlanes (v16i8 QPR:$src), imm:$lane),
2667 (VGETLNs8 (v8i8 (EXTRACT_SUBREG QPR:$src,
2668 (DSubReg_i8_reg imm:$lane))),
2669 (SubReg_i8_lane imm:$lane))>;
2670 def : Pat<(NEONvgetlanes (v8i16 QPR:$src), imm:$lane),
2671 (VGETLNs16 (v4i16 (EXTRACT_SUBREG QPR:$src,
2672 (DSubReg_i16_reg imm:$lane))),
2673 (SubReg_i16_lane imm:$lane))>;
2674 def : Pat<(NEONvgetlaneu (v16i8 QPR:$src), imm:$lane),
2675 (VGETLNu8 (v8i8 (EXTRACT_SUBREG QPR:$src,
2676 (DSubReg_i8_reg imm:$lane))),
2677 (SubReg_i8_lane imm:$lane))>;
2678 def : Pat<(NEONvgetlaneu (v8i16 QPR:$src), imm:$lane),
2679 (VGETLNu16 (v4i16 (EXTRACT_SUBREG QPR:$src,
2680 (DSubReg_i16_reg imm:$lane))),
2681 (SubReg_i16_lane imm:$lane))>;
2682 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
2683 (VGETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src,
2684 (DSubReg_i32_reg imm:$lane))),
2685 (SubReg_i32_lane imm:$lane))>;
2686 def : Pat<(extractelt (v2f32 DPR:$src1), imm:$src2),
2687 (EXTRACT_SUBREG (v2f32 (COPY_TO_REGCLASS (v2f32 DPR:$src1),DPR_VFP2)),
2688 (SSubReg_f32_reg imm:$src2))>;
2689 def : Pat<(extractelt (v4f32 QPR:$src1), imm:$src2),
2690 (EXTRACT_SUBREG (v4f32 (COPY_TO_REGCLASS (v4f32 QPR:$src1),QPR_VFP2)),
2691 (SSubReg_f32_reg imm:$src2))>;
2692 //def : Pat<(extractelt (v2i64 QPR:$src1), imm:$src2),
2693 // (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
2694 def : Pat<(extractelt (v2f64 QPR:$src1), imm:$src2),
2695 (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
2698 // VMOV : Vector Set Lane (move ARM core register to scalar)
2700 let Constraints = "$src1 = $dst" in {
2701 def VSETLNi8 : NVSetLane<{1,1,1,0,0,1,?,0}, 0b1011, {?,?}, (outs DPR:$dst),
2702 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
2703 IIC_VMOVISL, "vmov", "8", "$dst[$lane], $src2",
2704 [(set DPR:$dst, (vector_insert (v8i8 DPR:$src1),
2705 GPR:$src2, imm:$lane))]>;
2706 def VSETLNi16 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, {?,1}, (outs DPR:$dst),
2707 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
2708 IIC_VMOVISL, "vmov", "16", "$dst[$lane], $src2",
2709 [(set DPR:$dst, (vector_insert (v4i16 DPR:$src1),
2710 GPR:$src2, imm:$lane))]>;
2711 def VSETLNi32 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, 0b00, (outs DPR:$dst),
2712 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
2713 IIC_VMOVISL, "vmov", "32", "$dst[$lane], $src2",
2714 [(set DPR:$dst, (insertelt (v2i32 DPR:$src1),
2715 GPR:$src2, imm:$lane))]>;
2717 def : Pat<(vector_insert (v16i8 QPR:$src1), GPR:$src2, imm:$lane),
2718 (v16i8 (INSERT_SUBREG QPR:$src1,
2719 (v8i8 (VSETLNi8 (v8i8 (EXTRACT_SUBREG QPR:$src1,
2720 (DSubReg_i8_reg imm:$lane))),
2721 GPR:$src2, (SubReg_i8_lane imm:$lane))),
2722 (DSubReg_i8_reg imm:$lane)))>;
2723 def : Pat<(vector_insert (v8i16 QPR:$src1), GPR:$src2, imm:$lane),
2724 (v8i16 (INSERT_SUBREG QPR:$src1,
2725 (v4i16 (VSETLNi16 (v4i16 (EXTRACT_SUBREG QPR:$src1,
2726 (DSubReg_i16_reg imm:$lane))),
2727 GPR:$src2, (SubReg_i16_lane imm:$lane))),
2728 (DSubReg_i16_reg imm:$lane)))>;
2729 def : Pat<(insertelt (v4i32 QPR:$src1), GPR:$src2, imm:$lane),
2730 (v4i32 (INSERT_SUBREG QPR:$src1,
2731 (v2i32 (VSETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src1,
2732 (DSubReg_i32_reg imm:$lane))),
2733 GPR:$src2, (SubReg_i32_lane imm:$lane))),
2734 (DSubReg_i32_reg imm:$lane)))>;
2736 def : Pat<(v2f32 (insertelt DPR:$src1, SPR:$src2, imm:$src3)),
2737 (INSERT_SUBREG (v2f32 (COPY_TO_REGCLASS DPR:$src1, DPR_VFP2)),
2738 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
2739 def : Pat<(v4f32 (insertelt QPR:$src1, SPR:$src2, imm:$src3)),
2740 (INSERT_SUBREG (v4f32 (COPY_TO_REGCLASS QPR:$src1, QPR_VFP2)),
2741 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
2743 //def : Pat<(v2i64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
2744 // (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
2745 def : Pat<(v2f64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
2746 (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
2748 def : Pat<(v2f32 (scalar_to_vector SPR:$src)),
2749 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$src, arm_ssubreg_0)>;
2750 def : Pat<(v2f64 (scalar_to_vector (f64 DPR:$src))),
2751 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), DPR:$src, arm_dsubreg_0)>;
2752 def : Pat<(v4f32 (scalar_to_vector SPR:$src)),
2753 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), SPR:$src, arm_ssubreg_0)>;
2755 def : Pat<(v8i8 (scalar_to_vector GPR:$src)),
2756 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
2757 def : Pat<(v4i16 (scalar_to_vector GPR:$src)),
2758 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
2759 def : Pat<(v2i32 (scalar_to_vector GPR:$src)),
2760 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
2762 def : Pat<(v16i8 (scalar_to_vector GPR:$src)),
2763 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
2764 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
2766 def : Pat<(v8i16 (scalar_to_vector GPR:$src)),
2767 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
2768 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
2770 def : Pat<(v4i32 (scalar_to_vector GPR:$src)),
2771 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
2772 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
2775 // VDUP : Vector Duplicate (from ARM core register to all elements)
2777 class VDUPD<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
2778 : NVDup<opcod1, 0b1011, opcod3, (outs DPR:$dst), (ins GPR:$src),
2779 IIC_VMOVIS, "vdup", Dt, "$dst, $src",
2780 [(set DPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
2781 class VDUPQ<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
2782 : NVDup<opcod1, 0b1011, opcod3, (outs QPR:$dst), (ins GPR:$src),
2783 IIC_VMOVIS, "vdup", Dt, "$dst, $src",
2784 [(set QPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
2786 def VDUP8d : VDUPD<0b11101100, 0b00, "8", v8i8>;
2787 def VDUP16d : VDUPD<0b11101000, 0b01, "16", v4i16>;
2788 def VDUP32d : VDUPD<0b11101000, 0b00, "32", v2i32>;
2789 def VDUP8q : VDUPQ<0b11101110, 0b00, "8", v16i8>;
2790 def VDUP16q : VDUPQ<0b11101010, 0b01, "16", v8i16>;
2791 def VDUP32q : VDUPQ<0b11101010, 0b00, "32", v4i32>;
2793 def VDUPfd : NVDup<0b11101000, 0b1011, 0b00, (outs DPR:$dst), (ins GPR:$src),
2794 IIC_VMOVIS, "vdup", "32", "$dst, $src",
2795 [(set DPR:$dst, (v2f32 (NEONvdup
2796 (f32 (bitconvert GPR:$src)))))]>;
2797 def VDUPfq : NVDup<0b11101010, 0b1011, 0b00, (outs QPR:$dst), (ins GPR:$src),
2798 IIC_VMOVIS, "vdup", "32", "$dst, $src",
2799 [(set QPR:$dst, (v4f32 (NEONvdup
2800 (f32 (bitconvert GPR:$src)))))]>;
2802 // VDUP : Vector Duplicate Lane (from scalar to all elements)
2804 class VDUPLND<bits<2> op19_18, bits<2> op17_16,
2805 string OpcodeStr, string Dt, ValueType Ty>
2806 : N2V<0b11, 0b11, op19_18, op17_16, 0b11000, 0, 0,
2807 (outs DPR:$dst), (ins DPR:$src, nohash_imm:$lane), IIC_VMOVD,
2808 OpcodeStr, Dt, "$dst, $src[$lane]", "",
2809 [(set DPR:$dst, (Ty (NEONvduplane (Ty DPR:$src), imm:$lane)))]>;
2811 class VDUPLNQ<bits<2> op19_18, bits<2> op17_16, string OpcodeStr, string Dt,
2812 ValueType ResTy, ValueType OpTy>
2813 : N2V<0b11, 0b11, op19_18, op17_16, 0b11000, 1, 0,
2814 (outs QPR:$dst), (ins DPR:$src, nohash_imm:$lane), IIC_VMOVD,
2815 OpcodeStr, Dt, "$dst, $src[$lane]", "",
2816 [(set QPR:$dst, (ResTy (NEONvduplane (OpTy DPR:$src), imm:$lane)))]>;
2818 // Inst{19-16} is partially specified depending on the element size.
2820 def VDUPLN8d : VDUPLND<{?,?}, {?,1}, "vdup", "8", v8i8>;
2821 def VDUPLN16d : VDUPLND<{?,?}, {1,0}, "vdup", "16", v4i16>;
2822 def VDUPLN32d : VDUPLND<{?,1}, {0,0}, "vdup", "32", v2i32>;
2823 def VDUPLNfd : VDUPLND<{?,1}, {0,0}, "vdup", "32", v2f32>;
2824 def VDUPLN8q : VDUPLNQ<{?,?}, {?,1}, "vdup", "8", v16i8, v8i8>;
2825 def VDUPLN16q : VDUPLNQ<{?,?}, {1,0}, "vdup", "16", v8i16, v4i16>;
2826 def VDUPLN32q : VDUPLNQ<{?,1}, {0,0}, "vdup", "32", v4i32, v2i32>;
2827 def VDUPLNfq : VDUPLNQ<{?,1}, {0,0}, "vdup", "32", v4f32, v2f32>;
2829 def : Pat<(v16i8 (NEONvduplane (v16i8 QPR:$src), imm:$lane)),
2830 (v16i8 (VDUPLN8q (v8i8 (EXTRACT_SUBREG QPR:$src,
2831 (DSubReg_i8_reg imm:$lane))),
2832 (SubReg_i8_lane imm:$lane)))>;
2833 def : Pat<(v8i16 (NEONvduplane (v8i16 QPR:$src), imm:$lane)),
2834 (v8i16 (VDUPLN16q (v4i16 (EXTRACT_SUBREG QPR:$src,
2835 (DSubReg_i16_reg imm:$lane))),
2836 (SubReg_i16_lane imm:$lane)))>;
2837 def : Pat<(v4i32 (NEONvduplane (v4i32 QPR:$src), imm:$lane)),
2838 (v4i32 (VDUPLN32q (v2i32 (EXTRACT_SUBREG QPR:$src,
2839 (DSubReg_i32_reg imm:$lane))),
2840 (SubReg_i32_lane imm:$lane)))>;
2841 def : Pat<(v4f32 (NEONvduplane (v4f32 QPR:$src), imm:$lane)),
2842 (v4f32 (VDUPLNfq (v2f32 (EXTRACT_SUBREG QPR:$src,
2843 (DSubReg_i32_reg imm:$lane))),
2844 (SubReg_i32_lane imm:$lane)))>;
2846 def VDUPfdf : N2V<0b11, 0b11, {?,1}, {0,0}, 0b11000, 0, 0,
2847 (outs DPR:$dst), (ins SPR:$src),
2848 IIC_VMOVD, "vdup", "32", "$dst, ${src:lane}", "",
2849 [(set DPR:$dst, (v2f32 (NEONvdup (f32 SPR:$src))))]>;
2851 def VDUPfqf : N2V<0b11, 0b11, {?,1}, {0,0}, 0b11000, 1, 0,
2852 (outs QPR:$dst), (ins SPR:$src),
2853 IIC_VMOVD, "vdup", "32", "$dst, ${src:lane}", "",
2854 [(set QPR:$dst, (v4f32 (NEONvdup (f32 SPR:$src))))]>;
2856 def : Pat<(v2i64 (NEONvduplane (v2i64 QPR:$src), imm:$lane)),
2857 (INSERT_SUBREG QPR:$src,
2858 (i64 (EXTRACT_SUBREG QPR:$src,
2859 (DSubReg_f64_reg imm:$lane))),
2860 (DSubReg_f64_other_reg imm:$lane))>;
2861 def : Pat<(v2f64 (NEONvduplane (v2f64 QPR:$src), imm:$lane)),
2862 (INSERT_SUBREG QPR:$src,
2863 (f64 (EXTRACT_SUBREG QPR:$src,
2864 (DSubReg_f64_reg imm:$lane))),
2865 (DSubReg_f64_other_reg imm:$lane))>;
2867 // VMOVN : Vector Narrowing Move
2868 defm VMOVN : N2VNInt_HSD<0b11,0b11,0b10,0b00100,0,0, IIC_VMOVD,
2869 "vmovn", "i", int_arm_neon_vmovn>;
2870 // VQMOVN : Vector Saturating Narrowing Move
2871 defm VQMOVNs : N2VNInt_HSD<0b11,0b11,0b10,0b00101,0,0, IIC_VQUNAiD,
2872 "vqmovn", "s", int_arm_neon_vqmovns>;
2873 defm VQMOVNu : N2VNInt_HSD<0b11,0b11,0b10,0b00101,1,0, IIC_VQUNAiD,
2874 "vqmovn", "u", int_arm_neon_vqmovnu>;
2875 defm VQMOVNsu : N2VNInt_HSD<0b11,0b11,0b10,0b00100,1,0, IIC_VQUNAiD,
2876 "vqmovun", "s", int_arm_neon_vqmovnsu>;
2877 // VMOVL : Vector Lengthening Move
2878 defm VMOVLs : N2VLInt_QHS<0b01,0b10100,0,1, "vmovl", "s",
2879 int_arm_neon_vmovls>;
2880 defm VMOVLu : N2VLInt_QHS<0b11,0b10100,0,1, "vmovl", "u",
2881 int_arm_neon_vmovlu>;
2883 // Vector Conversions.
2885 // VCVT : Vector Convert Between Floating-Point and Integers
2886 def VCVTf2sd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
2887 v2i32, v2f32, fp_to_sint>;
2888 def VCVTf2ud : N2VD<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
2889 v2i32, v2f32, fp_to_uint>;
2890 def VCVTs2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
2891 v2f32, v2i32, sint_to_fp>;
2892 def VCVTu2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
2893 v2f32, v2i32, uint_to_fp>;
2895 def VCVTf2sq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
2896 v4i32, v4f32, fp_to_sint>;
2897 def VCVTf2uq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
2898 v4i32, v4f32, fp_to_uint>;
2899 def VCVTs2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
2900 v4f32, v4i32, sint_to_fp>;
2901 def VCVTu2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
2902 v4f32, v4i32, uint_to_fp>;
2904 // VCVT : Vector Convert Between Floating-Point and Fixed-Point.
2905 def VCVTf2xsd : N2VCvtD<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
2906 v2i32, v2f32, int_arm_neon_vcvtfp2fxs>;
2907 def VCVTf2xud : N2VCvtD<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
2908 v2i32, v2f32, int_arm_neon_vcvtfp2fxu>;
2909 def VCVTxs2fd : N2VCvtD<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
2910 v2f32, v2i32, int_arm_neon_vcvtfxs2fp>;
2911 def VCVTxu2fd : N2VCvtD<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
2912 v2f32, v2i32, int_arm_neon_vcvtfxu2fp>;
2914 def VCVTf2xsq : N2VCvtQ<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
2915 v4i32, v4f32, int_arm_neon_vcvtfp2fxs>;
2916 def VCVTf2xuq : N2VCvtQ<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
2917 v4i32, v4f32, int_arm_neon_vcvtfp2fxu>;
2918 def VCVTxs2fq : N2VCvtQ<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
2919 v4f32, v4i32, int_arm_neon_vcvtfxs2fp>;
2920 def VCVTxu2fq : N2VCvtQ<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
2921 v4f32, v4i32, int_arm_neon_vcvtfxu2fp>;
2925 // VREV64 : Vector Reverse elements within 64-bit doublewords
2927 class VREV64D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
2928 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 0, 0, (outs DPR:$dst),
2929 (ins DPR:$src), IIC_VMOVD,
2930 OpcodeStr, Dt, "$dst, $src", "",
2931 [(set DPR:$dst, (Ty (NEONvrev64 (Ty DPR:$src))))]>;
2932 class VREV64Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
2933 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 1, 0, (outs QPR:$dst),
2934 (ins QPR:$src), IIC_VMOVD,
2935 OpcodeStr, Dt, "$dst, $src", "",
2936 [(set QPR:$dst, (Ty (NEONvrev64 (Ty QPR:$src))))]>;
2938 def VREV64d8 : VREV64D<0b00, "vrev64", "8", v8i8>;
2939 def VREV64d16 : VREV64D<0b01, "vrev64", "16", v4i16>;
2940 def VREV64d32 : VREV64D<0b10, "vrev64", "32", v2i32>;
2941 def VREV64df : VREV64D<0b10, "vrev64", "32", v2f32>;
2943 def VREV64q8 : VREV64Q<0b00, "vrev64", "8", v16i8>;
2944 def VREV64q16 : VREV64Q<0b01, "vrev64", "16", v8i16>;
2945 def VREV64q32 : VREV64Q<0b10, "vrev64", "32", v4i32>;
2946 def VREV64qf : VREV64Q<0b10, "vrev64", "32", v4f32>;
2948 // VREV32 : Vector Reverse elements within 32-bit words
2950 class VREV32D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
2951 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 0, 0, (outs DPR:$dst),
2952 (ins DPR:$src), IIC_VMOVD,
2953 OpcodeStr, Dt, "$dst, $src", "",
2954 [(set DPR:$dst, (Ty (NEONvrev32 (Ty DPR:$src))))]>;
2955 class VREV32Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
2956 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 1, 0, (outs QPR:$dst),
2957 (ins QPR:$src), IIC_VMOVD,
2958 OpcodeStr, Dt, "$dst, $src", "",
2959 [(set QPR:$dst, (Ty (NEONvrev32 (Ty QPR:$src))))]>;
2961 def VREV32d8 : VREV32D<0b00, "vrev32", "8", v8i8>;
2962 def VREV32d16 : VREV32D<0b01, "vrev32", "16", v4i16>;
2964 def VREV32q8 : VREV32Q<0b00, "vrev32", "8", v16i8>;
2965 def VREV32q16 : VREV32Q<0b01, "vrev32", "16", v8i16>;
2967 // VREV16 : Vector Reverse elements within 16-bit halfwords
2969 class VREV16D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
2970 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 0, 0, (outs DPR:$dst),
2971 (ins DPR:$src), IIC_VMOVD,
2972 OpcodeStr, Dt, "$dst, $src", "",
2973 [(set DPR:$dst, (Ty (NEONvrev16 (Ty DPR:$src))))]>;
2974 class VREV16Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
2975 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 1, 0, (outs QPR:$dst),
2976 (ins QPR:$src), IIC_VMOVD,
2977 OpcodeStr, Dt, "$dst, $src", "",
2978 [(set QPR:$dst, (Ty (NEONvrev16 (Ty QPR:$src))))]>;
2980 def VREV16d8 : VREV16D<0b00, "vrev16", "8", v8i8>;
2981 def VREV16q8 : VREV16Q<0b00, "vrev16", "8", v16i8>;
2983 // Other Vector Shuffles.
2985 // VEXT : Vector Extract
2987 class VEXTd<string OpcodeStr, string Dt, ValueType Ty>
2988 : N3V<0,1,0b11,{?,?,?,?},0,0, (outs DPR:$dst),
2989 (ins DPR:$lhs, DPR:$rhs, i32imm:$index), IIC_VEXTD,
2990 OpcodeStr, Dt, "$dst, $lhs, $rhs, $index", "",
2991 [(set DPR:$dst, (Ty (NEONvext (Ty DPR:$lhs),
2992 (Ty DPR:$rhs), imm:$index)))]>;
2994 class VEXTq<string OpcodeStr, string Dt, ValueType Ty>
2995 : N3V<0,1,0b11,{?,?,?,?},1,0, (outs QPR:$dst),
2996 (ins QPR:$lhs, QPR:$rhs, i32imm:$index), IIC_VEXTQ,
2997 OpcodeStr, Dt, "$dst, $lhs, $rhs, $index", "",
2998 [(set QPR:$dst, (Ty (NEONvext (Ty QPR:$lhs),
2999 (Ty QPR:$rhs), imm:$index)))]>;
3001 def VEXTd8 : VEXTd<"vext", "8", v8i8>;
3002 def VEXTd16 : VEXTd<"vext", "16", v4i16>;
3003 def VEXTd32 : VEXTd<"vext", "32", v2i32>;
3004 def VEXTdf : VEXTd<"vext", "32", v2f32>;
3006 def VEXTq8 : VEXTq<"vext", "8", v16i8>;
3007 def VEXTq16 : VEXTq<"vext", "16", v8i16>;
3008 def VEXTq32 : VEXTq<"vext", "32", v4i32>;
3009 def VEXTqf : VEXTq<"vext", "32", v4f32>;
3011 // VTRN : Vector Transpose
3013 def VTRNd8 : N2VDShuffle<0b00, 0b00001, "vtrn", "8">;
3014 def VTRNd16 : N2VDShuffle<0b01, 0b00001, "vtrn", "16">;
3015 def VTRNd32 : N2VDShuffle<0b10, 0b00001, "vtrn", "32">;
3017 def VTRNq8 : N2VQShuffle<0b00, 0b00001, IIC_VPERMQ, "vtrn", "8">;
3018 def VTRNq16 : N2VQShuffle<0b01, 0b00001, IIC_VPERMQ, "vtrn", "16">;
3019 def VTRNq32 : N2VQShuffle<0b10, 0b00001, IIC_VPERMQ, "vtrn", "32">;
3021 // VUZP : Vector Unzip (Deinterleave)
3023 def VUZPd8 : N2VDShuffle<0b00, 0b00010, "vuzp", "8">;
3024 def VUZPd16 : N2VDShuffle<0b01, 0b00010, "vuzp", "16">;
3025 def VUZPd32 : N2VDShuffle<0b10, 0b00010, "vuzp", "32">;
3027 def VUZPq8 : N2VQShuffle<0b00, 0b00010, IIC_VPERMQ3, "vuzp", "8">;
3028 def VUZPq16 : N2VQShuffle<0b01, 0b00010, IIC_VPERMQ3, "vuzp", "16">;
3029 def VUZPq32 : N2VQShuffle<0b10, 0b00010, IIC_VPERMQ3, "vuzp", "32">;
3031 // VZIP : Vector Zip (Interleave)
3033 def VZIPd8 : N2VDShuffle<0b00, 0b00011, "vzip", "8">;
3034 def VZIPd16 : N2VDShuffle<0b01, 0b00011, "vzip", "16">;
3035 def VZIPd32 : N2VDShuffle<0b10, 0b00011, "vzip", "32">;
3037 def VZIPq8 : N2VQShuffle<0b00, 0b00011, IIC_VPERMQ3, "vzip", "8">;
3038 def VZIPq16 : N2VQShuffle<0b01, 0b00011, IIC_VPERMQ3, "vzip", "16">;
3039 def VZIPq32 : N2VQShuffle<0b10, 0b00011, IIC_VPERMQ3, "vzip", "32">;
3041 // Vector Table Lookup and Table Extension.
3043 // VTBL : Vector Table Lookup
3045 : N3V<1,1,0b11,0b1000,0,0, (outs DPR:$dst),
3046 (ins DPR:$tbl1, DPR:$src), IIC_VTB1,
3047 "vtbl", "8", "$dst, \\{$tbl1\\}, $src", "",
3048 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl1 DPR:$tbl1, DPR:$src)))]>;
3049 let hasExtraSrcRegAllocReq = 1 in {
3051 : N3V<1,1,0b11,0b1001,0,0, (outs DPR:$dst),
3052 (ins DPR:$tbl1, DPR:$tbl2, DPR:$src), IIC_VTB2,
3053 "vtbl", "8", "$dst, \\{$tbl1, $tbl2\\}, $src", "",
3054 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl2
3055 DPR:$tbl1, DPR:$tbl2, DPR:$src)))]>;
3057 : N3V<1,1,0b11,0b1010,0,0, (outs DPR:$dst),
3058 (ins DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$src), IIC_VTB3,
3059 "vtbl", "8", "$dst, \\{$tbl1, $tbl2, $tbl3\\}, $src", "",
3060 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl3
3061 DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$src)))]>;
3063 : N3V<1,1,0b11,0b1011,0,0, (outs DPR:$dst),
3064 (ins DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$src), IIC_VTB4,
3065 "vtbl", "8", "$dst, \\{$tbl1, $tbl2, $tbl3, $tbl4\\}, $src", "",
3066 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl4 DPR:$tbl1, DPR:$tbl2,
3067 DPR:$tbl3, DPR:$tbl4, DPR:$src)))]>;
3068 } // hasExtraSrcRegAllocReq = 1
3070 // VTBX : Vector Table Extension
3072 : N3V<1,1,0b11,0b1000,1,0, (outs DPR:$dst),
3073 (ins DPR:$orig, DPR:$tbl1, DPR:$src), IIC_VTBX1,
3074 "vtbx", "8", "$dst, \\{$tbl1\\}, $src", "$orig = $dst",
3075 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx1
3076 DPR:$orig, DPR:$tbl1, DPR:$src)))]>;
3077 let hasExtraSrcRegAllocReq = 1 in {
3079 : N3V<1,1,0b11,0b1001,1,0, (outs DPR:$dst),
3080 (ins DPR:$orig, DPR:$tbl1, DPR:$tbl2, DPR:$src), IIC_VTBX2,
3081 "vtbx", "8", "$dst, \\{$tbl1, $tbl2\\}, $src", "$orig = $dst",
3082 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx2
3083 DPR:$orig, DPR:$tbl1, DPR:$tbl2, DPR:$src)))]>;
3085 : N3V<1,1,0b11,0b1010,1,0, (outs DPR:$dst),
3086 (ins DPR:$orig, DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$src), IIC_VTBX3,
3087 "vtbx", "8", "$dst, \\{$tbl1, $tbl2, $tbl3\\}, $src", "$orig = $dst",
3088 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx3 DPR:$orig, DPR:$tbl1,
3089 DPR:$tbl2, DPR:$tbl3, DPR:$src)))]>;
3091 : N3V<1,1,0b11,0b1011,1,0, (outs DPR:$dst), (ins DPR:$orig, DPR:$tbl1,
3092 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$src), IIC_VTBX4,
3093 "vtbx", "8", "$dst, \\{$tbl1, $tbl2, $tbl3, $tbl4\\}, $src",
3095 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx4 DPR:$orig, DPR:$tbl1,
3096 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$src)))]>;
3097 } // hasExtraSrcRegAllocReq = 1
3099 //===----------------------------------------------------------------------===//
3100 // NEON instructions for single-precision FP math
3101 //===----------------------------------------------------------------------===//
3103 class N2VSPat<SDNode OpNode, ValueType ResTy, ValueType OpTy, NeonI Inst>
3104 : NEONFPPat<(ResTy (OpNode SPR:$a)),
3105 (EXTRACT_SUBREG (OpTy (Inst (INSERT_SUBREG (OpTy (IMPLICIT_DEF)),
3106 SPR:$a, arm_ssubreg_0))),
3109 class N3VSPat<SDNode OpNode, NeonI Inst>
3110 : NEONFPPat<(f32 (OpNode SPR:$a, SPR:$b)),
3111 (EXTRACT_SUBREG (v2f32
3112 (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3113 SPR:$a, arm_ssubreg_0),
3114 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3115 SPR:$b, arm_ssubreg_0))),
3118 class N3VSMulOpPat<SDNode MulNode, SDNode OpNode, NeonI Inst>
3119 : NEONFPPat<(f32 (OpNode SPR:$acc, (f32 (MulNode SPR:$a, SPR:$b)))),
3120 (EXTRACT_SUBREG (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3121 SPR:$acc, arm_ssubreg_0),
3122 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3123 SPR:$a, arm_ssubreg_0),
3124 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3125 SPR:$b, arm_ssubreg_0)),
3128 // These need separate instructions because they must use DPR_VFP2 register
3129 // class which have SPR sub-registers.
3131 // Vector Add Operations used for single-precision FP
3132 let neverHasSideEffects = 1 in
3133 def VADDfd_sfp : N3VS<0,0,0b00,0b1101,0, "vadd", "f32", v2f32, v2f32, fadd, 1>;
3134 def : N3VSPat<fadd, VADDfd_sfp>;
3136 // Vector Sub Operations used for single-precision FP
3137 let neverHasSideEffects = 1 in
3138 def VSUBfd_sfp : N3VS<0,0,0b10,0b1101,0, "vsub", "f32", v2f32, v2f32, fsub, 0>;
3139 def : N3VSPat<fsub, VSUBfd_sfp>;
3141 // Vector Multiply Operations used for single-precision FP
3142 let neverHasSideEffects = 1 in
3143 def VMULfd_sfp : N3VS<1,0,0b00,0b1101,1, "vmul", "f32", v2f32, v2f32, fmul, 1>;
3144 def : N3VSPat<fmul, VMULfd_sfp>;
3146 // Vector Multiply-Accumulate/Subtract used for single-precision FP
3147 // vml[as].f32 can cause 4-8 cycle stalls in following ASIMD instructions, so
3148 // we want to avoid them for now. e.g., alternating vmla/vadd instructions.
3150 //let neverHasSideEffects = 1 in
3151 //def VMLAfd_sfp : N3VSMulOp<0,0,0b00,0b1101,1, IIC_VMACD, "vmla", "f32",
3152 // v2f32, fmul, fadd>;
3153 //def : N3VSMulOpPat<fmul, fadd, VMLAfd_sfp>;
3155 //let neverHasSideEffects = 1 in
3156 //def VMLSfd_sfp : N3VSMulOp<0,0,0b10,0b1101,1, IIC_VMACD, "vmls", "f32",
3157 // v2f32, fmul, fsub>;
3158 //def : N3VSMulOpPat<fmul, fsub, VMLSfd_sfp>;
3160 // Vector Absolute used for single-precision FP
3161 let neverHasSideEffects = 1 in
3162 def VABSfd_sfp : N2V<0b11, 0b11, 0b10, 0b01, 0b01110, 0, 0,
3163 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), IIC_VUNAD,
3164 "vabs", "f32", "$dst, $src", "", []>;
3165 def : N2VSPat<fabs, f32, v2f32, VABSfd_sfp>;
3167 // Vector Negate used for single-precision FP
3168 let neverHasSideEffects = 1 in
3169 def VNEGfd_sfp : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
3170 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), IIC_VUNAD,
3171 "vneg", "f32", "$dst, $src", "", []>;
3172 def : N2VSPat<fneg, f32, v2f32, VNEGfd_sfp>;
3174 // Vector Maximum used for single-precision FP
3175 let neverHasSideEffects = 1 in
3176 def VMAXfd_sfp : N3V<0, 0, 0b00, 0b1111, 0, 0, (outs DPR_VFP2:$dst),
3177 (ins DPR_VFP2:$src1, DPR_VFP2:$src2), IIC_VBIND,
3178 "vmax", "f32", "$dst, $src1, $src2", "", []>;
3179 def : N3VSPat<NEONfmax, VMAXfd_sfp>;
3181 // Vector Minimum used for single-precision FP
3182 let neverHasSideEffects = 1 in
3183 def VMINfd_sfp : N3V<0, 0, 0b00, 0b1111, 0, 0, (outs DPR_VFP2:$dst),
3184 (ins DPR_VFP2:$src1, DPR_VFP2:$src2), IIC_VBIND,
3185 "vmin", "f32", "$dst, $src1, $src2", "", []>;
3186 def : N3VSPat<NEONfmin, VMINfd_sfp>;
3188 // Vector Convert between single-precision FP and integer
3189 let neverHasSideEffects = 1 in
3190 def VCVTf2sd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
3191 v2i32, v2f32, fp_to_sint>;
3192 def : N2VSPat<arm_ftosi, f32, v2f32, VCVTf2sd_sfp>;
3194 let neverHasSideEffects = 1 in
3195 def VCVTf2ud_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
3196 v2i32, v2f32, fp_to_uint>;
3197 def : N2VSPat<arm_ftoui, f32, v2f32, VCVTf2ud_sfp>;
3199 let neverHasSideEffects = 1 in
3200 def VCVTs2fd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
3201 v2f32, v2i32, sint_to_fp>;
3202 def : N2VSPat<arm_sitof, f32, v2i32, VCVTs2fd_sfp>;
3204 let neverHasSideEffects = 1 in
3205 def VCVTu2fd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
3206 v2f32, v2i32, uint_to_fp>;
3207 def : N2VSPat<arm_uitof, f32, v2i32, VCVTu2fd_sfp>;
3209 //===----------------------------------------------------------------------===//
3210 // Non-Instruction Patterns
3211 //===----------------------------------------------------------------------===//
3214 def : Pat<(v1i64 (bitconvert (v2i32 DPR:$src))), (v1i64 DPR:$src)>;
3215 def : Pat<(v1i64 (bitconvert (v4i16 DPR:$src))), (v1i64 DPR:$src)>;
3216 def : Pat<(v1i64 (bitconvert (v8i8 DPR:$src))), (v1i64 DPR:$src)>;
3217 def : Pat<(v1i64 (bitconvert (f64 DPR:$src))), (v1i64 DPR:$src)>;
3218 def : Pat<(v1i64 (bitconvert (v2f32 DPR:$src))), (v1i64 DPR:$src)>;
3219 def : Pat<(v2i32 (bitconvert (v1i64 DPR:$src))), (v2i32 DPR:$src)>;
3220 def : Pat<(v2i32 (bitconvert (v4i16 DPR:$src))), (v2i32 DPR:$src)>;
3221 def : Pat<(v2i32 (bitconvert (v8i8 DPR:$src))), (v2i32 DPR:$src)>;
3222 def : Pat<(v2i32 (bitconvert (f64 DPR:$src))), (v2i32 DPR:$src)>;
3223 def : Pat<(v2i32 (bitconvert (v2f32 DPR:$src))), (v2i32 DPR:$src)>;
3224 def : Pat<(v4i16 (bitconvert (v1i64 DPR:$src))), (v4i16 DPR:$src)>;
3225 def : Pat<(v4i16 (bitconvert (v2i32 DPR:$src))), (v4i16 DPR:$src)>;
3226 def : Pat<(v4i16 (bitconvert (v8i8 DPR:$src))), (v4i16 DPR:$src)>;
3227 def : Pat<(v4i16 (bitconvert (f64 DPR:$src))), (v4i16 DPR:$src)>;
3228 def : Pat<(v4i16 (bitconvert (v2f32 DPR:$src))), (v4i16 DPR:$src)>;
3229 def : Pat<(v8i8 (bitconvert (v1i64 DPR:$src))), (v8i8 DPR:$src)>;
3230 def : Pat<(v8i8 (bitconvert (v2i32 DPR:$src))), (v8i8 DPR:$src)>;
3231 def : Pat<(v8i8 (bitconvert (v4i16 DPR:$src))), (v8i8 DPR:$src)>;
3232 def : Pat<(v8i8 (bitconvert (f64 DPR:$src))), (v8i8 DPR:$src)>;
3233 def : Pat<(v8i8 (bitconvert (v2f32 DPR:$src))), (v8i8 DPR:$src)>;
3234 def : Pat<(f64 (bitconvert (v1i64 DPR:$src))), (f64 DPR:$src)>;
3235 def : Pat<(f64 (bitconvert (v2i32 DPR:$src))), (f64 DPR:$src)>;
3236 def : Pat<(f64 (bitconvert (v4i16 DPR:$src))), (f64 DPR:$src)>;
3237 def : Pat<(f64 (bitconvert (v8i8 DPR:$src))), (f64 DPR:$src)>;
3238 def : Pat<(f64 (bitconvert (v2f32 DPR:$src))), (f64 DPR:$src)>;
3239 def : Pat<(v2f32 (bitconvert (f64 DPR:$src))), (v2f32 DPR:$src)>;
3240 def : Pat<(v2f32 (bitconvert (v1i64 DPR:$src))), (v2f32 DPR:$src)>;
3241 def : Pat<(v2f32 (bitconvert (v2i32 DPR:$src))), (v2f32 DPR:$src)>;
3242 def : Pat<(v2f32 (bitconvert (v4i16 DPR:$src))), (v2f32 DPR:$src)>;
3243 def : Pat<(v2f32 (bitconvert (v8i8 DPR:$src))), (v2f32 DPR:$src)>;
3245 def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (v2i64 QPR:$src)>;
3246 def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (v2i64 QPR:$src)>;
3247 def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (v2i64 QPR:$src)>;
3248 def : Pat<(v2i64 (bitconvert (v2f64 QPR:$src))), (v2i64 QPR:$src)>;
3249 def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (v2i64 QPR:$src)>;
3250 def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (v4i32 QPR:$src)>;
3251 def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (v4i32 QPR:$src)>;
3252 def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (v4i32 QPR:$src)>;
3253 def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (v4i32 QPR:$src)>;
3254 def : Pat<(v4i32 (bitconvert (v4f32 QPR:$src))), (v4i32 QPR:$src)>;
3255 def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (v8i16 QPR:$src)>;
3256 def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (v8i16 QPR:$src)>;
3257 def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (v8i16 QPR:$src)>;
3258 def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (v8i16 QPR:$src)>;
3259 def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (v8i16 QPR:$src)>;
3260 def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (v16i8 QPR:$src)>;
3261 def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (v16i8 QPR:$src)>;
3262 def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (v16i8 QPR:$src)>;
3263 def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (v16i8 QPR:$src)>;
3264 def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (v16i8 QPR:$src)>;
3265 def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (v4f32 QPR:$src)>;
3266 def : Pat<(v4f32 (bitconvert (v4i32 QPR:$src))), (v4f32 QPR:$src)>;
3267 def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (v4f32 QPR:$src)>;
3268 def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (v4f32 QPR:$src)>;
3269 def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (v4f32 QPR:$src)>;
3270 def : Pat<(v2f64 (bitconvert (v2i64 QPR:$src))), (v2f64 QPR:$src)>;
3271 def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (v2f64 QPR:$src)>;
3272 def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 QPR:$src)>;
3273 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 QPR:$src)>;
3274 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 QPR:$src)>;