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:$addr)]> {
133 let Inst{27-25} = 0b110;
134 let Inst{24} = 0; // P bit
135 let Inst{23} = 1; // U bit
137 let Inst{11-8} = 0b1011;
140 // VLD1 : Vector Load (multiple single elements)
141 class VLD1D<bits<4> op7_4, string OpcodeStr, string Dt,
142 ValueType Ty, Intrinsic IntOp>
143 : NLdSt<0,0b10,0b0111,op7_4, (outs DPR:$dst), (ins addrmode6:$addr), IIC_VLD1,
144 OpcodeStr, Dt, "\\{$dst\\}, $addr", "",
145 [(set DPR:$dst, (Ty (IntOp addrmode6:$addr)))]>;
146 class VLD1Q<bits<4> op7_4, string OpcodeStr, string Dt,
147 ValueType Ty, Intrinsic IntOp>
148 : NLdSt<0,0b10,0b1010,op7_4, (outs QPR:$dst), (ins addrmode6:$addr), IIC_VLD1,
149 OpcodeStr, Dt, "${dst:dregpair}, $addr", "",
150 [(set QPR:$dst, (Ty (IntOp addrmode6:$addr)))]>;
152 def VLD1d8 : VLD1D<0b0000, "vld1", "8", v8i8, int_arm_neon_vld1>;
153 def VLD1d16 : VLD1D<0b0100, "vld1", "16", v4i16, int_arm_neon_vld1>;
154 def VLD1d32 : VLD1D<0b1000, "vld1", "32", v2i32, int_arm_neon_vld1>;
155 def VLD1df : VLD1D<0b1000, "vld1", "32", v2f32, int_arm_neon_vld1>;
156 def VLD1d64 : VLD1D<0b1100, "vld1", "64", v1i64, int_arm_neon_vld1>;
158 def VLD1q8 : VLD1Q<0b0000, "vld1", "8", v16i8, int_arm_neon_vld1>;
159 def VLD1q16 : VLD1Q<0b0100, "vld1", "16", v8i16, int_arm_neon_vld1>;
160 def VLD1q32 : VLD1Q<0b1000, "vld1", "32", v4i32, int_arm_neon_vld1>;
161 def VLD1qf : VLD1Q<0b1000, "vld1", "32", v4f32, int_arm_neon_vld1>;
162 def VLD1q64 : VLD1Q<0b1100, "vld1", "64", v2i64, int_arm_neon_vld1>;
164 // These (dreg triple/quadruple) are for disassembly only.
165 class VLD1D3<bits<4> op7_4, string OpcodeStr, string Dt>
166 : NLdSt<0, 0b10, 0b0110, op7_4, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3),
167 (ins addrmode6:$addr), IIC_VLD1, OpcodeStr, Dt,
168 "\\{$dst1, $dst2, $dst3\\}, $addr", "",
169 [/* For disassembly only; pattern left blank */]>;
170 class VLD1D4<bits<4> op7_4, string OpcodeStr, string Dt>
171 : NLdSt<0,0b10,0b0010,op7_4,(outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
172 (ins addrmode6:$addr), IIC_VLD1, OpcodeStr, Dt,
173 "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr", "",
174 [/* For disassembly only; pattern left blank */]>;
176 def VLD1d8T : VLD1D3<0b0000, "vld1", "8">;
177 def VLD1d16T : VLD1D3<0b0100, "vld1", "16">;
178 def VLD1d32T : VLD1D3<0b1000, "vld1", "32">;
179 //def VLD1d64T : VLD1D3<0b1100, "vld1", "64">;
181 def VLD1d8Q : VLD1D4<0b0000, "vld1", "8">;
182 def VLD1d16Q : VLD1D4<0b0100, "vld1", "16">;
183 def VLD1d32Q : VLD1D4<0b1000, "vld1", "32">;
184 //def VLD1d64Q : VLD1D4<0b1100, "vld1", "64">;
187 let mayLoad = 1, hasExtraDefRegAllocReq = 1 in {
189 // VLD2 : Vector Load (multiple 2-element structures)
190 class VLD2D<bits<4> op7_4, string OpcodeStr, string Dt>
191 : NLdSt<0,0b10,0b1000,op7_4, (outs DPR:$dst1, DPR:$dst2),
192 (ins addrmode6:$addr), IIC_VLD2,
193 OpcodeStr, Dt, "\\{$dst1, $dst2\\}, $addr", "", []>;
194 class VLD2Q<bits<4> op7_4, string OpcodeStr, string Dt>
195 : NLdSt<0,0b10,0b0011,op7_4,
196 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
197 (ins addrmode6:$addr), IIC_VLD2,
198 OpcodeStr, Dt, "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr",
201 def VLD2d8 : VLD2D<0b0000, "vld2", "8">;
202 def VLD2d16 : VLD2D<0b0100, "vld2", "16">;
203 def VLD2d32 : VLD2D<0b1000, "vld2", "32">;
204 def VLD2d64 : NLdSt<0,0b10,0b1010,0b1100, (outs DPR:$dst1, DPR:$dst2),
205 (ins addrmode6:$addr), IIC_VLD1,
206 "vld1", "64", "\\{$dst1, $dst2\\}, $addr", "", []>;
208 def VLD2q8 : VLD2Q<0b0000, "vld2", "8">;
209 def VLD2q16 : VLD2Q<0b0100, "vld2", "16">;
210 def VLD2q32 : VLD2Q<0b1000, "vld2", "32">;
212 // These (double-spaced dreg pair) are for disassembly only.
213 class VLD2Ddbl<bits<4> op7_4, string OpcodeStr, string Dt>
214 : NLdSt<0,0b10,0b1001,op7_4, (outs DPR:$dst1, DPR:$dst2),
215 (ins addrmode6:$addr), IIC_VLD2,
216 OpcodeStr, Dt, "\\{$dst1, $dst2\\}, $addr", "", []>;
218 def VLD2d8D : VLD2Ddbl<0b0000, "vld2", "8">;
219 def VLD2d16D : VLD2Ddbl<0b0100, "vld2", "16">;
220 def VLD2d32D : VLD2Ddbl<0b1000, "vld2", "32">;
222 // VLD3 : Vector Load (multiple 3-element structures)
223 class VLD3D<bits<4> op7_4, string OpcodeStr, string Dt>
224 : NLdSt<0,0b10,0b0100,op7_4, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3),
225 (ins addrmode6:$addr), IIC_VLD3,
226 OpcodeStr, Dt, "\\{$dst1, $dst2, $dst3\\}, $addr", "", []>;
227 class VLD3WB<bits<4> op7_4, string OpcodeStr, string Dt>
228 : NLdSt<0,0b10,0b0101,op7_4, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, GPR:$wb),
229 (ins addrmode6:$addr), IIC_VLD3,
230 OpcodeStr, Dt, "\\{$dst1, $dst2, $dst3\\}, $addr!",
231 "$addr.addr = $wb", []>;
233 def VLD3d8 : VLD3D<0b0000, "vld3", "8">;
234 def VLD3d16 : VLD3D<0b0100, "vld3", "16">;
235 def VLD3d32 : VLD3D<0b1000, "vld3", "32">;
236 def VLD3d64 : NLdSt<0,0b10,0b0110,0b1100,
237 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3),
238 (ins addrmode6:$addr), IIC_VLD1,
239 "vld1", "64", "\\{$dst1, $dst2, $dst3\\}, $addr", "", []>;
241 // vld3 to double-spaced even registers.
242 def VLD3q8a : VLD3WB<0b0000, "vld3", "8">;
243 def VLD3q16a : VLD3WB<0b0100, "vld3", "16">;
244 def VLD3q32a : VLD3WB<0b1000, "vld3", "32">;
246 // vld3 to double-spaced odd registers.
247 def VLD3q8b : VLD3WB<0b0000, "vld3", "8">;
248 def VLD3q16b : VLD3WB<0b0100, "vld3", "16">;
249 def VLD3q32b : VLD3WB<0b1000, "vld3", "32">;
251 // VLD4 : Vector Load (multiple 4-element structures)
252 class VLD4D<bits<4> op7_4, string OpcodeStr, string Dt>
253 : NLdSt<0,0b10,0b0000,op7_4,
254 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
255 (ins addrmode6:$addr), IIC_VLD4,
256 OpcodeStr, Dt, "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr",
258 class VLD4WB<bits<4> op7_4, string OpcodeStr, string Dt>
259 : NLdSt<0,0b10,0b0001,op7_4,
260 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
261 (ins addrmode6:$addr), IIC_VLD4,
262 OpcodeStr, Dt, "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr!",
263 "$addr.addr = $wb", []>;
265 def VLD4d8 : VLD4D<0b0000, "vld4", "8">;
266 def VLD4d16 : VLD4D<0b0100, "vld4", "16">;
267 def VLD4d32 : VLD4D<0b1000, "vld4", "32">;
268 def VLD4d64 : NLdSt<0,0b10,0b0010,0b1100,
269 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
270 (ins addrmode6:$addr), IIC_VLD1,
271 "vld1", "64", "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr",
274 // vld4 to double-spaced even registers.
275 def VLD4q8a : VLD4WB<0b0000, "vld4", "8">;
276 def VLD4q16a : VLD4WB<0b0100, "vld4", "16">;
277 def VLD4q32a : VLD4WB<0b1000, "vld4", "32">;
279 // vld4 to double-spaced odd registers.
280 def VLD4q8b : VLD4WB<0b0000, "vld4", "8">;
281 def VLD4q16b : VLD4WB<0b0100, "vld4", "16">;
282 def VLD4q32b : VLD4WB<0b1000, "vld4", "32">;
284 // VLD1LN : Vector Load (single element to one lane)
285 // FIXME: Not yet implemented.
287 // VLD2LN : Vector Load (single 2-element structure to one lane)
288 class VLD2LN<bits<4> op11_8, string OpcodeStr, string Dt>
289 : NLdSt<1,0b10,op11_8,{?,?,?,?}, (outs DPR:$dst1, DPR:$dst2),
290 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, nohash_imm:$lane),
291 IIC_VLD2, OpcodeStr, Dt, "\\{$dst1[$lane], $dst2[$lane]\\}, $addr",
292 "$src1 = $dst1, $src2 = $dst2", []>;
294 // vld2 to single-spaced registers.
295 def VLD2LNd8 : VLD2LN<0b0001, "vld2", "8">;
296 def VLD2LNd16 : VLD2LN<0b0101, "vld2", "16"> { let Inst{5} = 0; }
297 def VLD2LNd32 : VLD2LN<0b1001, "vld2", "32"> { let Inst{6} = 0; }
299 // vld2 to double-spaced even registers.
300 def VLD2LNq16a: VLD2LN<0b0101, "vld2", "16"> { let Inst{5} = 1; }
301 def VLD2LNq32a: VLD2LN<0b1001, "vld2", "32"> { let Inst{6} = 1; }
303 // vld2 to double-spaced odd registers.
304 def VLD2LNq16b: VLD2LN<0b0101, "vld2", "16"> { let Inst{5} = 1; }
305 def VLD2LNq32b: VLD2LN<0b1001, "vld2", "32"> { let Inst{6} = 1; }
307 // VLD3LN : Vector Load (single 3-element structure to one lane)
308 class VLD3LN<bits<4> op11_8, string OpcodeStr, string Dt>
309 : NLdSt<1,0b10,op11_8,{?,?,?,?}, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3),
310 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3,
311 nohash_imm:$lane), IIC_VLD3, OpcodeStr, Dt,
312 "\\{$dst1[$lane], $dst2[$lane], $dst3[$lane]\\}, $addr",
313 "$src1 = $dst1, $src2 = $dst2, $src3 = $dst3", []>;
315 // vld3 to single-spaced registers.
316 def VLD3LNd8 : VLD3LN<0b0010, "vld3", "8"> { let Inst{4} = 0; }
317 def VLD3LNd16 : VLD3LN<0b0110, "vld3", "16"> { let Inst{5-4} = 0b00; }
318 def VLD3LNd32 : VLD3LN<0b1010, "vld3", "32"> { let Inst{6-4} = 0b000; }
320 // vld3 to double-spaced even registers.
321 def VLD3LNq16a: VLD3LN<0b0110, "vld3", "16"> { let Inst{5-4} = 0b10; }
322 def VLD3LNq32a: VLD3LN<0b1010, "vld3", "32"> { let Inst{6-4} = 0b100; }
324 // vld3 to double-spaced odd registers.
325 def VLD3LNq16b: VLD3LN<0b0110, "vld3", "16"> { let Inst{5-4} = 0b10; }
326 def VLD3LNq32b: VLD3LN<0b1010, "vld3", "32"> { let Inst{6-4} = 0b100; }
328 // VLD4LN : Vector Load (single 4-element structure to one lane)
329 class VLD4LN<bits<4> op11_8, string OpcodeStr, string Dt>
330 : NLdSt<1,0b10,op11_8,{?,?,?,?},
331 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
332 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
333 nohash_imm:$lane), IIC_VLD4, OpcodeStr, Dt,
334 "\\{$dst1[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $addr",
335 "$src1 = $dst1, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4", []>;
337 // vld4 to single-spaced registers.
338 def VLD4LNd8 : VLD4LN<0b0011, "vld4", "8">;
339 def VLD4LNd16 : VLD4LN<0b0111, "vld4", "16"> { let Inst{5} = 0; }
340 def VLD4LNd32 : VLD4LN<0b1011, "vld4", "32"> { let Inst{6} = 0; }
342 // vld4 to double-spaced even registers.
343 def VLD4LNq16a: VLD4LN<0b0111, "vld4", "16"> { let Inst{5} = 1; }
344 def VLD4LNq32a: VLD4LN<0b1011, "vld4", "32"> { let Inst{6} = 1; }
346 // vld4 to double-spaced odd registers.
347 def VLD4LNq16b: VLD4LN<0b0111, "vld4", "16"> { let Inst{5} = 1; }
348 def VLD4LNq32b: VLD4LN<0b1011, "vld4", "32"> { let Inst{6} = 1; }
350 // VLD1DUP : Vector Load (single element to all lanes)
351 // VLD2DUP : Vector Load (single 2-element structure to all lanes)
352 // VLD3DUP : Vector Load (single 3-element structure to all lanes)
353 // VLD4DUP : Vector Load (single 4-element structure to all lanes)
354 // FIXME: Not yet implemented.
355 } // mayLoad = 1, hasExtraDefRegAllocReq = 1
357 // VST1 : Vector Store (multiple single elements)
358 class VST1D<bits<4> op7_4, string OpcodeStr, string Dt,
359 ValueType Ty, Intrinsic IntOp>
360 : NLdSt<0,0b00,0b0111,op7_4, (outs), (ins addrmode6:$addr, DPR:$src), IIC_VST,
361 OpcodeStr, Dt, "\\{$src\\}, $addr", "",
362 [(IntOp addrmode6:$addr, (Ty DPR:$src))]>;
363 class VST1Q<bits<4> op7_4, string OpcodeStr, string Dt,
364 ValueType Ty, Intrinsic IntOp>
365 : NLdSt<0,0b00,0b1010,op7_4, (outs), (ins addrmode6:$addr, QPR:$src), IIC_VST,
366 OpcodeStr, Dt, "${src:dregpair}, $addr", "",
367 [(IntOp addrmode6:$addr, (Ty QPR:$src))]>;
369 let hasExtraSrcRegAllocReq = 1 in {
370 def VST1d8 : VST1D<0b0000, "vst1", "8", v8i8, int_arm_neon_vst1>;
371 def VST1d16 : VST1D<0b0100, "vst1", "16", v4i16, int_arm_neon_vst1>;
372 def VST1d32 : VST1D<0b1000, "vst1", "32", v2i32, int_arm_neon_vst1>;
373 def VST1df : VST1D<0b1000, "vst1", "32", v2f32, int_arm_neon_vst1>;
374 def VST1d64 : VST1D<0b1100, "vst1", "64", v1i64, int_arm_neon_vst1>;
376 def VST1q8 : VST1Q<0b0000, "vst1", "8", v16i8, int_arm_neon_vst1>;
377 def VST1q16 : VST1Q<0b0100, "vst1", "16", v8i16, int_arm_neon_vst1>;
378 def VST1q32 : VST1Q<0b1000, "vst1", "32", v4i32, int_arm_neon_vst1>;
379 def VST1qf : VST1Q<0b1000, "vst1", "32", v4f32, int_arm_neon_vst1>;
380 def VST1q64 : VST1Q<0b1100, "vst1", "64", v2i64, int_arm_neon_vst1>;
381 } // hasExtraSrcRegAllocReq
383 // These (dreg triple/quadruple) are for disassembly only.
384 class VST1D3<bits<4> op7_4, string OpcodeStr, string Dt>
385 : NLdSt<0, 0b00, 0b0110, op7_4, (outs),
386 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3), IIC_VST,
388 "\\{$src1, $src2, $src3\\}, $addr", "",
389 [/* For disassembly only; pattern left blank */]>;
390 class VST1D4<bits<4> op7_4, string OpcodeStr, string Dt>
391 : NLdSt<0, 0b00, 0b0010, op7_4, (outs),
392 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
393 IIC_VST, OpcodeStr, Dt,
394 "\\{$src1, $src2, $src3, $src4\\}, $addr", "",
395 [/* For disassembly only; pattern left blank */]>;
397 def VST1d8T : VST1D3<0b0000, "vst1", "8">;
398 def VST1d16T : VST1D3<0b0100, "vst1", "16">;
399 def VST1d32T : VST1D3<0b1000, "vst1", "32">;
400 //def VST1d64T : VST1D3<0b1100, "vst1", "64">;
402 def VST1d8Q : VST1D4<0b0000, "vst1", "8">;
403 def VST1d16Q : VST1D4<0b0100, "vst1", "16">;
404 def VST1d32Q : VST1D4<0b1000, "vst1", "32">;
405 //def VST1d64Q : VST1D4<0b1100, "vst1", "64">;
408 let mayStore = 1, hasExtraSrcRegAllocReq = 1 in {
410 // VST2 : Vector Store (multiple 2-element structures)
411 class VST2D<bits<4> op7_4, string OpcodeStr, string Dt>
412 : NLdSt<0,0b00,0b1000,op7_4, (outs),
413 (ins addrmode6:$addr, DPR:$src1, DPR:$src2), IIC_VST,
414 OpcodeStr, Dt, "\\{$src1, $src2\\}, $addr", "", []>;
415 class VST2Q<bits<4> op7_4, string OpcodeStr, string Dt>
416 : NLdSt<0,0b00,0b0011,op7_4, (outs),
417 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
418 IIC_VST, OpcodeStr, Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr",
421 def VST2d8 : VST2D<0b0000, "vst2", "8">;
422 def VST2d16 : VST2D<0b0100, "vst2", "16">;
423 def VST2d32 : VST2D<0b1000, "vst2", "32">;
424 def VST2d64 : NLdSt<0,0b00,0b1010,0b1100, (outs),
425 (ins addrmode6:$addr, DPR:$src1, DPR:$src2), IIC_VST,
426 "vst1", "64", "\\{$src1, $src2\\}, $addr", "", []>;
428 def VST2q8 : VST2Q<0b0000, "vst2", "8">;
429 def VST2q16 : VST2Q<0b0100, "vst2", "16">;
430 def VST2q32 : VST2Q<0b1000, "vst2", "32">;
432 // These (double-spaced dreg pair) are for disassembly only.
433 class VST2Ddbl<bits<4> op7_4, string OpcodeStr, string Dt>
434 : NLdSt<0, 0b00, 0b1001, op7_4, (outs),
435 (ins addrmode6:$addr, DPR:$src1, DPR:$src2), IIC_VST,
436 OpcodeStr, Dt, "\\{$src1, $src2\\}, $addr", "", []>;
438 def VST2d8D : VST2Ddbl<0b0000, "vst2", "8">;
439 def VST2d16D : VST2Ddbl<0b0100, "vst2", "16">;
440 def VST2d32D : VST2Ddbl<0b1000, "vst2", "32">;
442 // VST3 : Vector Store (multiple 3-element structures)
443 class VST3D<bits<4> op7_4, string OpcodeStr, string Dt>
444 : NLdSt<0,0b00,0b0100,op7_4, (outs),
445 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3), IIC_VST,
446 OpcodeStr, Dt, "\\{$src1, $src2, $src3\\}, $addr", "", []>;
447 class VST3WB<bits<4> op7_4, string OpcodeStr, string Dt>
448 : NLdSt<0,0b00,0b0101,op7_4, (outs GPR:$wb),
449 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3), IIC_VST,
450 OpcodeStr, Dt, "\\{$src1, $src2, $src3\\}, $addr!",
451 "$addr.addr = $wb", []>;
453 def VST3d8 : VST3D<0b0000, "vst3", "8">;
454 def VST3d16 : VST3D<0b0100, "vst3", "16">;
455 def VST3d32 : VST3D<0b1000, "vst3", "32">;
456 def VST3d64 : NLdSt<0,0b00,0b0110,0b1100, (outs),
457 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3),
459 "vst1", "64", "\\{$src1, $src2, $src3\\}, $addr", "", []>;
461 // vst3 to double-spaced even registers.
462 def VST3q8a : VST3WB<0b0000, "vst3", "8">;
463 def VST3q16a : VST3WB<0b0100, "vst3", "16">;
464 def VST3q32a : VST3WB<0b1000, "vst3", "32">;
466 // vst3 to double-spaced odd registers.
467 def VST3q8b : VST3WB<0b0000, "vst3", "8">;
468 def VST3q16b : VST3WB<0b0100, "vst3", "16">;
469 def VST3q32b : VST3WB<0b1000, "vst3", "32">;
471 // VST4 : Vector Store (multiple 4-element structures)
472 class VST4D<bits<4> op7_4, string OpcodeStr, string Dt>
473 : NLdSt<0,0b00,0b0000,op7_4, (outs),
474 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
475 IIC_VST, OpcodeStr, Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr",
477 class VST4WB<bits<4> op7_4, string OpcodeStr, string Dt>
478 : NLdSt<0,0b00,0b0001,op7_4, (outs GPR:$wb),
479 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
480 IIC_VST, OpcodeStr, Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr!",
481 "$addr.addr = $wb", []>;
483 def VST4d8 : VST4D<0b0000, "vst4", "8">;
484 def VST4d16 : VST4D<0b0100, "vst4", "16">;
485 def VST4d32 : VST4D<0b1000, "vst4", "32">;
486 def VST4d64 : NLdSt<0,0b00,0b0010,0b1100, (outs),
487 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3,
489 "vst1", "64", "\\{$src1, $src2, $src3, $src4\\}, $addr",
492 // vst4 to double-spaced even registers.
493 def VST4q8a : VST4WB<0b0000, "vst4", "8">;
494 def VST4q16a : VST4WB<0b0100, "vst4", "16">;
495 def VST4q32a : VST4WB<0b1000, "vst4", "32">;
497 // vst4 to double-spaced odd registers.
498 def VST4q8b : VST4WB<0b0000, "vst4", "8">;
499 def VST4q16b : VST4WB<0b0100, "vst4", "16">;
500 def VST4q32b : VST4WB<0b1000, "vst4", "32">;
502 // VST1LN : Vector Store (single element from one lane)
503 // FIXME: Not yet implemented.
505 // VST2LN : Vector Store (single 2-element structure from one lane)
506 class VST2LN<bits<4> op11_8, string OpcodeStr, string Dt>
507 : NLdSt<1,0b00,op11_8,{?,?,?,?}, (outs),
508 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, nohash_imm:$lane),
509 IIC_VST, OpcodeStr, Dt, "\\{$src1[$lane], $src2[$lane]\\}, $addr",
512 // vst2 to single-spaced registers.
513 def VST2LNd8 : VST2LN<0b0001, "vst2", "8">;
514 def VST2LNd16 : VST2LN<0b0101, "vst2", "16"> { let Inst{5} = 0; }
515 def VST2LNd32 : VST2LN<0b1001, "vst2", "32"> { let Inst{6} = 0; }
517 // vst2 to double-spaced even registers.
518 def VST2LNq16a: VST2LN<0b0101, "vst2", "16"> { let Inst{5} = 1; }
519 def VST2LNq32a: VST2LN<0b1001, "vst2", "32"> { let Inst{6} = 1; }
521 // vst2 to double-spaced odd registers.
522 def VST2LNq16b: VST2LN<0b0101, "vst2", "16"> { let Inst{5} = 1; }
523 def VST2LNq32b: VST2LN<0b1001, "vst2", "32"> { let Inst{6} = 1; }
525 // VST3LN : Vector Store (single 3-element structure from one lane)
526 class VST3LN<bits<4> op11_8, string OpcodeStr, string Dt>
527 : NLdSt<1,0b00,op11_8,{?,?,?,?}, (outs),
528 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3,
529 nohash_imm:$lane), IIC_VST, OpcodeStr, Dt,
530 "\\{$src1[$lane], $src2[$lane], $src3[$lane]\\}, $addr", "", []>;
532 // vst3 to single-spaced registers.
533 def VST3LNd8 : VST3LN<0b0010, "vst3", "8"> { let Inst{4} = 0; }
534 def VST3LNd16 : VST3LN<0b0110, "vst3", "16"> { let Inst{5-4} = 0b00; }
535 def VST3LNd32 : VST3LN<0b1010, "vst3", "32"> { let Inst{6-4} = 0b000; }
537 // vst3 to double-spaced even registers.
538 def VST3LNq16a: VST3LN<0b0110, "vst3", "16"> { let Inst{5-4} = 0b10; }
539 def VST3LNq32a: VST3LN<0b1010, "vst3", "32"> { let Inst{6-4} = 0b100; }
541 // vst3 to double-spaced odd registers.
542 def VST3LNq16b: VST3LN<0b0110, "vst3", "16"> { let Inst{5-4} = 0b10; }
543 def VST3LNq32b: VST3LN<0b1010, "vst3", "32"> { let Inst{6-4} = 0b100; }
545 // VST4LN : Vector Store (single 4-element structure from one lane)
546 class VST4LN<bits<4> op11_8, string OpcodeStr, string Dt>
547 : NLdSt<1,0b00,op11_8,{?,?,?,?}, (outs),
548 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
549 nohash_imm:$lane), IIC_VST, OpcodeStr, Dt,
550 "\\{$src1[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $addr",
553 // vst4 to single-spaced registers.
554 def VST4LNd8 : VST4LN<0b0011, "vst4", "8">;
555 def VST4LNd16 : VST4LN<0b0111, "vst4", "16"> { let Inst{5} = 0; }
556 def VST4LNd32 : VST4LN<0b1011, "vst4", "32"> { let Inst{6} = 0; }
558 // vst4 to double-spaced even registers.
559 def VST4LNq16a: VST4LN<0b0111, "vst4", "16"> { let Inst{5} = 1; }
560 def VST4LNq32a: VST4LN<0b1011, "vst4", "32"> { let Inst{6} = 1; }
562 // vst4 to double-spaced odd registers.
563 def VST4LNq16b: VST4LN<0b0111, "vst4", "16"> { let Inst{5} = 1; }
564 def VST4LNq32b: VST4LN<0b1011, "vst4", "32"> { let Inst{6} = 1; }
566 } // mayStore = 1, hasExtraSrcRegAllocReq = 1
569 //===----------------------------------------------------------------------===//
570 // NEON pattern fragments
571 //===----------------------------------------------------------------------===//
573 // Extract D sub-registers of Q registers.
574 // (arm_dsubreg_0 is 5; arm_dsubreg_1 is 6)
575 def DSubReg_i8_reg : SDNodeXForm<imm, [{
576 return CurDAG->getTargetConstant(5 + N->getZExtValue() / 8, MVT::i32);
578 def DSubReg_i16_reg : SDNodeXForm<imm, [{
579 return CurDAG->getTargetConstant(5 + N->getZExtValue() / 4, MVT::i32);
581 def DSubReg_i32_reg : SDNodeXForm<imm, [{
582 return CurDAG->getTargetConstant(5 + N->getZExtValue() / 2, MVT::i32);
584 def DSubReg_f64_reg : SDNodeXForm<imm, [{
585 return CurDAG->getTargetConstant(5 + N->getZExtValue(), MVT::i32);
587 def DSubReg_f64_other_reg : SDNodeXForm<imm, [{
588 return CurDAG->getTargetConstant(5 + (1 - N->getZExtValue()), MVT::i32);
591 // Extract S sub-registers of Q/D registers.
592 // (arm_ssubreg_0 is 1; arm_ssubreg_1 is 2; etc.)
593 def SSubReg_f32_reg : SDNodeXForm<imm, [{
594 return CurDAG->getTargetConstant(1 + N->getZExtValue(), MVT::i32);
597 // Translate lane numbers from Q registers to D subregs.
598 def SubReg_i8_lane : SDNodeXForm<imm, [{
599 return CurDAG->getTargetConstant(N->getZExtValue() & 7, MVT::i32);
601 def SubReg_i16_lane : SDNodeXForm<imm, [{
602 return CurDAG->getTargetConstant(N->getZExtValue() & 3, MVT::i32);
604 def SubReg_i32_lane : SDNodeXForm<imm, [{
605 return CurDAG->getTargetConstant(N->getZExtValue() & 1, MVT::i32);
608 //===----------------------------------------------------------------------===//
609 // Instruction Classes
610 //===----------------------------------------------------------------------===//
612 // Basic 2-register operations: single-, double- and quad-register.
613 class N2VS<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
614 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
615 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
616 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
617 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src),
618 IIC_VUNAD, OpcodeStr, Dt, "$dst, $src", "", []>;
619 class N2VD<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
620 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
621 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
622 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
623 (ins DPR:$src), IIC_VUNAD, OpcodeStr, Dt, "$dst, $src", "",
624 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src))))]>;
625 class N2VQ<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
626 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
627 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
628 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
629 (ins QPR:$src), IIC_VUNAQ, OpcodeStr, Dt, "$dst, $src", "",
630 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src))))]>;
632 // Basic 2-register intrinsics, both double- and quad-register.
633 class N2VDInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
634 bits<2> op17_16, bits<5> op11_7, bit op4,
635 InstrItinClass itin, string OpcodeStr, string Dt,
636 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
637 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
638 (ins DPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
639 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
640 class N2VQInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
641 bits<2> op17_16, bits<5> op11_7, bit op4,
642 InstrItinClass itin, string OpcodeStr, string Dt,
643 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
644 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
645 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
646 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
648 // Narrow 2-register intrinsics.
649 class N2VNInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
650 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
651 InstrItinClass itin, string OpcodeStr, string Dt,
652 ValueType TyD, ValueType TyQ, Intrinsic IntOp>
653 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$dst),
654 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
655 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src))))]>;
657 // Long 2-register intrinsics (currently only used for VMOVL).
658 class N2VLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
659 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
660 InstrItinClass itin, string OpcodeStr, string Dt,
661 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
662 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$dst),
663 (ins DPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
664 [(set QPR:$dst, (TyQ (IntOp (TyD DPR:$src))))]>;
666 // 2-register shuffles (VTRN/VZIP/VUZP), both double- and quad-register.
667 class N2VDShuffle<bits<2> op19_18, bits<5> op11_7, string OpcodeStr, string Dt>
668 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 0, 0, (outs DPR:$dst1, DPR:$dst2),
669 (ins DPR:$src1, DPR:$src2), IIC_VPERMD,
670 OpcodeStr, Dt, "$dst1, $dst2",
671 "$src1 = $dst1, $src2 = $dst2", []>;
672 class N2VQShuffle<bits<2> op19_18, bits<5> op11_7,
673 InstrItinClass itin, string OpcodeStr, string Dt>
674 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 1, 0, (outs QPR:$dst1, QPR:$dst2),
675 (ins QPR:$src1, QPR:$src2), itin, OpcodeStr, Dt, "$dst1, $dst2",
676 "$src1 = $dst1, $src2 = $dst2", []>;
678 // Basic 3-register operations: single-, double- and quad-register.
679 class N3VS<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
680 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
681 SDNode OpNode, bit Commutable>
682 : N3V<op24, op23, op21_20, op11_8, 0, op4,
683 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src1, DPR_VFP2:$src2), IIC_VBIND,
684 OpcodeStr, Dt, "$dst, $src1, $src2", "", []> {
685 let isCommutable = Commutable;
688 class N3VD<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
689 InstrItinClass itin, string OpcodeStr, string Dt,
690 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
691 : N3V<op24, op23, op21_20, op11_8, 0, op4,
692 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), itin,
693 OpcodeStr, Dt, "$dst, $src1, $src2", "",
694 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src1), (OpTy DPR:$src2))))]> {
695 let isCommutable = Commutable;
697 // Same as N3VD but no data type.
698 class N3VDX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
699 InstrItinClass itin, string OpcodeStr,
700 ValueType ResTy, ValueType OpTy,
701 SDNode OpNode, bit Commutable>
702 : N3VX<op24, op23, op21_20, op11_8, 0, op4,
703 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), itin,
704 OpcodeStr, "$dst, $src1, $src2", "",
705 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src1), (OpTy DPR:$src2))))]>{
706 let isCommutable = Commutable;
708 class N3VDSL<bits<2> op21_20, bits<4> op11_8,
709 InstrItinClass itin, string OpcodeStr, string Dt,
710 ValueType Ty, SDNode ShOp>
711 : N3V<0, 1, op21_20, op11_8, 1, 0,
712 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
713 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
715 (Ty (ShOp (Ty DPR:$src1),
716 (Ty (NEONvduplane (Ty DPR_VFP2:$src2), imm:$lane)))))]>{
717 let isCommutable = 0;
719 class N3VDSL16<bits<2> op21_20, bits<4> op11_8,
720 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
721 : N3V<0, 1, op21_20, op11_8, 1, 0,
722 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
723 IIC_VMULi16D, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
725 (Ty (ShOp (Ty DPR:$src1),
726 (Ty (NEONvduplane (Ty DPR_8:$src2), imm:$lane)))))]> {
727 let isCommutable = 0;
730 class N3VQ<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
731 InstrItinClass itin, string OpcodeStr, string Dt,
732 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
733 : N3V<op24, op23, op21_20, op11_8, 1, op4,
734 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), itin,
735 OpcodeStr, Dt, "$dst, $src1, $src2", "",
736 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src1), (OpTy QPR:$src2))))]> {
737 let isCommutable = Commutable;
739 class N3VQX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
740 InstrItinClass itin, string OpcodeStr,
741 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
742 : N3VX<op24, op23, op21_20, op11_8, 1, op4,
743 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), itin,
744 OpcodeStr, "$dst, $src1, $src2", "",
745 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src1), (OpTy QPR:$src2))))]>{
746 let isCommutable = Commutable;
748 class N3VQSL<bits<2> op21_20, bits<4> op11_8,
749 InstrItinClass itin, string OpcodeStr, string Dt,
750 ValueType ResTy, ValueType OpTy, SDNode ShOp>
751 : N3V<1, 1, op21_20, op11_8, 1, 0,
752 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
753 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
754 [(set (ResTy QPR:$dst),
755 (ResTy (ShOp (ResTy QPR:$src1),
756 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
758 let isCommutable = 0;
760 class N3VQSL16<bits<2> op21_20, bits<4> op11_8, string OpcodeStr, string Dt,
761 ValueType ResTy, ValueType OpTy, SDNode ShOp>
762 : N3V<1, 1, op21_20, op11_8, 1, 0,
763 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
764 IIC_VMULi16Q, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
765 [(set (ResTy QPR:$dst),
766 (ResTy (ShOp (ResTy QPR:$src1),
767 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
769 let isCommutable = 0;
772 // Basic 3-register intrinsics, both double- and quad-register.
773 class N3VDInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
774 InstrItinClass itin, string OpcodeStr, string Dt,
775 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
776 : N3V<op24, op23, op21_20, op11_8, 0, op4,
777 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), itin,
778 OpcodeStr, Dt, "$dst, $src1, $src2", "",
779 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src1), (OpTy DPR:$src2))))]> {
780 let isCommutable = Commutable;
782 class N3VDIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
783 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
784 : N3V<0, 1, op21_20, op11_8, 1, 0,
785 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
786 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
788 (Ty (IntOp (Ty DPR:$src1),
789 (Ty (NEONvduplane (Ty DPR_VFP2:$src2),
791 let isCommutable = 0;
793 class N3VDIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
794 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
795 : N3V<0, 1, op21_20, op11_8, 1, 0,
796 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
797 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
799 (Ty (IntOp (Ty DPR:$src1),
800 (Ty (NEONvduplane (Ty DPR_8:$src2),
802 let isCommutable = 0;
805 class N3VQInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
806 InstrItinClass itin, string OpcodeStr, string Dt,
807 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
808 : N3V<op24, op23, op21_20, op11_8, 1, op4,
809 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), itin,
810 OpcodeStr, Dt, "$dst, $src1, $src2", "",
811 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src1), (OpTy QPR:$src2))))]> {
812 let isCommutable = Commutable;
814 class N3VQIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
815 string OpcodeStr, string Dt,
816 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
817 : N3V<1, 1, op21_20, op11_8, 1, 0,
818 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
819 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
820 [(set (ResTy QPR:$dst),
821 (ResTy (IntOp (ResTy QPR:$src1),
822 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
824 let isCommutable = 0;
826 class N3VQIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
827 string OpcodeStr, string Dt,
828 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
829 : N3V<1, 1, op21_20, op11_8, 1, 0,
830 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
831 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
832 [(set (ResTy QPR:$dst),
833 (ResTy (IntOp (ResTy QPR:$src1),
834 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
836 let isCommutable = 0;
839 // Multiply-Add/Sub operations: single-, double- and quad-register.
840 class N3VSMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
841 InstrItinClass itin, string OpcodeStr, string Dt,
842 ValueType Ty, SDNode MulOp, SDNode OpNode>
843 : N3V<op24, op23, op21_20, op11_8, 0, op4,
844 (outs DPR_VFP2:$dst),
845 (ins DPR_VFP2:$src1, DPR_VFP2:$src2, DPR_VFP2:$src3), itin,
846 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst", []>;
848 class N3VDMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
849 InstrItinClass itin, string OpcodeStr, string Dt,
850 ValueType Ty, SDNode MulOp, SDNode OpNode>
851 : N3V<op24, op23, op21_20, op11_8, 0, op4,
852 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3), itin,
853 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
854 [(set DPR:$dst, (Ty (OpNode DPR:$src1,
855 (Ty (MulOp DPR:$src2, DPR:$src3)))))]>;
856 class N3VDMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
857 string OpcodeStr, string Dt,
858 ValueType Ty, SDNode MulOp, SDNode ShOp>
859 : N3V<0, 1, op21_20, op11_8, 1, 0,
861 (ins DPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane), itin,
862 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
864 (Ty (ShOp (Ty DPR:$src1),
865 (Ty (MulOp DPR:$src2,
866 (Ty (NEONvduplane (Ty DPR_VFP2:$src3),
868 class N3VDMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
869 string OpcodeStr, string Dt,
870 ValueType Ty, SDNode MulOp, SDNode ShOp>
871 : N3V<0, 1, op21_20, op11_8, 1, 0,
873 (ins DPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane), itin,
874 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
876 (Ty (ShOp (Ty DPR:$src1),
877 (Ty (MulOp DPR:$src2,
878 (Ty (NEONvduplane (Ty DPR_8:$src3),
881 class N3VQMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
882 InstrItinClass itin, string OpcodeStr, string Dt, ValueType Ty,
883 SDNode MulOp, SDNode OpNode>
884 : N3V<op24, op23, op21_20, op11_8, 1, op4,
885 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3), itin,
886 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
887 [(set QPR:$dst, (Ty (OpNode QPR:$src1,
888 (Ty (MulOp QPR:$src2, QPR:$src3)))))]>;
889 class N3VQMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
890 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
891 SDNode MulOp, SDNode ShOp>
892 : N3V<1, 1, op21_20, op11_8, 1, 0,
894 (ins QPR:$src1, QPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane), itin,
895 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
896 [(set (ResTy QPR:$dst),
897 (ResTy (ShOp (ResTy QPR:$src1),
898 (ResTy (MulOp QPR:$src2,
899 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src3),
901 class N3VQMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
902 string OpcodeStr, string Dt,
903 ValueType ResTy, ValueType OpTy,
904 SDNode MulOp, SDNode ShOp>
905 : N3V<1, 1, op21_20, op11_8, 1, 0,
907 (ins QPR:$src1, QPR:$src2, DPR_8:$src3, nohash_imm:$lane), itin,
908 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
909 [(set (ResTy QPR:$dst),
910 (ResTy (ShOp (ResTy QPR:$src1),
911 (ResTy (MulOp QPR:$src2,
912 (ResTy (NEONvduplane (OpTy DPR_8:$src3),
915 // Neon 3-argument intrinsics, both double- and quad-register.
916 // The destination register is also used as the first source operand register.
917 class N3VDInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
918 InstrItinClass itin, string OpcodeStr, string Dt,
919 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
920 : N3V<op24, op23, op21_20, op11_8, 0, op4,
921 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3), itin,
922 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
923 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src1),
924 (OpTy DPR:$src2), (OpTy DPR:$src3))))]>;
925 class N3VQInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
926 InstrItinClass itin, string OpcodeStr, string Dt,
927 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
928 : N3V<op24, op23, op21_20, op11_8, 1, op4,
929 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3), itin,
930 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
931 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src1),
932 (OpTy QPR:$src2), (OpTy QPR:$src3))))]>;
934 // Neon Long 3-argument intrinsic. The destination register is
935 // a quad-register and is also used as the first source operand register.
936 class N3VLInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
937 InstrItinClass itin, string OpcodeStr, string Dt,
938 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
939 : N3V<op24, op23, op21_20, op11_8, 0, op4,
940 (outs QPR:$dst), (ins QPR:$src1, DPR:$src2, DPR:$src3), itin,
941 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
943 (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$src2), (TyD DPR:$src3))))]>;
944 class N3VLInt3SL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
945 string OpcodeStr, string Dt,
946 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
947 : N3V<op24, 1, op21_20, op11_8, 1, 0,
949 (ins QPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane), itin,
950 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
951 [(set (ResTy QPR:$dst),
952 (ResTy (IntOp (ResTy QPR:$src1),
954 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src3),
956 class N3VLInt3SL16<bit op24, bits<2> op21_20, bits<4> op11_8,
957 InstrItinClass itin, string OpcodeStr, string Dt,
958 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
959 : N3V<op24, 1, op21_20, op11_8, 1, 0,
961 (ins QPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane), itin,
962 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
963 [(set (ResTy QPR:$dst),
964 (ResTy (IntOp (ResTy QPR:$src1),
966 (OpTy (NEONvduplane (OpTy DPR_8:$src3),
969 // Narrowing 3-register intrinsics.
970 class N3VNInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
971 string OpcodeStr, string Dt, ValueType TyD, ValueType TyQ,
972 Intrinsic IntOp, bit Commutable>
973 : N3V<op24, op23, op21_20, op11_8, 0, op4,
974 (outs DPR:$dst), (ins QPR:$src1, QPR:$src2), IIC_VBINi4D,
975 OpcodeStr, Dt, "$dst, $src1, $src2", "",
976 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src1), (TyQ QPR:$src2))))]> {
977 let isCommutable = Commutable;
980 // Long 3-register intrinsics.
981 class N3VLInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
982 InstrItinClass itin, string OpcodeStr, string Dt,
983 ValueType TyQ, ValueType TyD, Intrinsic IntOp, bit Commutable>
984 : N3V<op24, op23, op21_20, op11_8, 0, op4,
985 (outs QPR:$dst), (ins DPR:$src1, DPR:$src2), itin,
986 OpcodeStr, Dt, "$dst, $src1, $src2", "",
987 [(set QPR:$dst, (TyQ (IntOp (TyD DPR:$src1), (TyD DPR:$src2))))]> {
988 let isCommutable = Commutable;
990 class N3VLIntSL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
991 string OpcodeStr, string Dt,
992 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
993 : N3V<op24, 1, op21_20, op11_8, 1, 0,
994 (outs QPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
995 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
996 [(set (ResTy QPR:$dst),
997 (ResTy (IntOp (OpTy DPR:$src1),
998 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1000 class N3VLIntSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1001 InstrItinClass itin, string OpcodeStr, string Dt,
1002 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1003 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1004 (outs QPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1005 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1006 [(set (ResTy QPR:$dst),
1007 (ResTy (IntOp (OpTy DPR:$src1),
1008 (OpTy (NEONvduplane (OpTy DPR_8:$src2),
1011 // Wide 3-register intrinsics.
1012 class N3VWInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1013 string OpcodeStr, string Dt, ValueType TyQ, ValueType TyD,
1014 Intrinsic IntOp, bit Commutable>
1015 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1016 (outs QPR:$dst), (ins QPR:$src1, DPR:$src2), IIC_VSUBiD,
1017 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1018 [(set QPR:$dst, (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$src2))))]> {
1019 let isCommutable = Commutable;
1022 // Pairwise long 2-register intrinsics, both double- and quad-register.
1023 class N2VDPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1024 bits<2> op17_16, bits<5> op11_7, bit op4,
1025 string OpcodeStr, string Dt,
1026 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1027 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
1028 (ins DPR:$src), IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
1029 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
1030 class N2VQPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1031 bits<2> op17_16, bits<5> op11_7, bit op4,
1032 string OpcodeStr, string Dt,
1033 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1034 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
1035 (ins QPR:$src), IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
1036 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
1038 // Pairwise long 2-register accumulate intrinsics,
1039 // both double- and quad-register.
1040 // The destination register is also used as the first source operand register.
1041 class N2VDPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1042 bits<2> op17_16, bits<5> op11_7, bit op4,
1043 string OpcodeStr, string Dt,
1044 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1045 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
1046 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), IIC_VPALiD,
1047 OpcodeStr, Dt, "$dst, $src2", "$src1 = $dst",
1048 [(set DPR:$dst, (ResTy (IntOp (ResTy DPR:$src1), (OpTy DPR:$src2))))]>;
1049 class N2VQPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1050 bits<2> op17_16, bits<5> op11_7, bit op4,
1051 string OpcodeStr, string Dt,
1052 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1053 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4,
1054 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), IIC_VPALiQ,
1055 OpcodeStr, Dt, "$dst, $src2", "$src1 = $dst",
1056 [(set QPR:$dst, (ResTy (IntOp (ResTy QPR:$src1), (OpTy QPR:$src2))))]>;
1058 // Shift by immediate,
1059 // both double- and quad-register.
1060 class N2VDSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1061 InstrItinClass itin, string OpcodeStr, string Dt,
1062 ValueType Ty, SDNode OpNode>
1063 : N2VImm<op24, op23, op11_8, op7, 0, op4,
1064 (outs DPR:$dst), (ins DPR:$src, i32imm:$SIMM), itin,
1065 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1066 [(set DPR:$dst, (Ty (OpNode (Ty DPR:$src), (i32 imm:$SIMM))))]>;
1067 class N2VQSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1068 InstrItinClass itin, string OpcodeStr, string Dt,
1069 ValueType Ty, SDNode OpNode>
1070 : N2VImm<op24, op23, op11_8, op7, 1, op4,
1071 (outs QPR:$dst), (ins QPR:$src, i32imm:$SIMM), itin,
1072 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1073 [(set QPR:$dst, (Ty (OpNode (Ty QPR:$src), (i32 imm:$SIMM))))]>;
1075 // Long shift by immediate.
1076 class N2VLSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
1077 string OpcodeStr, string Dt,
1078 ValueType ResTy, ValueType OpTy, SDNode OpNode>
1079 : N2VImm<op24, op23, op11_8, op7, op6, op4,
1080 (outs QPR:$dst), (ins DPR:$src, i32imm:$SIMM), IIC_VSHLiD,
1081 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1082 [(set QPR:$dst, (ResTy (OpNode (OpTy DPR:$src),
1083 (i32 imm:$SIMM))))]>;
1085 // Narrow shift by immediate.
1086 class N2VNSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
1087 InstrItinClass itin, string OpcodeStr, string Dt,
1088 ValueType ResTy, ValueType OpTy, SDNode OpNode>
1089 : N2VImm<op24, op23, op11_8, op7, op6, op4,
1090 (outs DPR:$dst), (ins QPR:$src, i32imm:$SIMM), itin,
1091 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1092 [(set DPR:$dst, (ResTy (OpNode (OpTy QPR:$src),
1093 (i32 imm:$SIMM))))]>;
1095 // Shift right by immediate and accumulate,
1096 // both double- and quad-register.
1097 class N2VDShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1098 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1099 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$dst),
1100 (ins DPR:$src1, DPR:$src2, i32imm:$SIMM), IIC_VPALiD,
1101 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1102 [(set DPR:$dst, (Ty (add DPR:$src1,
1103 (Ty (ShOp DPR:$src2, (i32 imm:$SIMM))))))]>;
1104 class N2VQShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1105 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1106 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$dst),
1107 (ins QPR:$src1, QPR:$src2, i32imm:$SIMM), IIC_VPALiD,
1108 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1109 [(set QPR:$dst, (Ty (add QPR:$src1,
1110 (Ty (ShOp QPR:$src2, (i32 imm:$SIMM))))))]>;
1112 // Shift by immediate and insert,
1113 // both double- and quad-register.
1114 class N2VDShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1115 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1116 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$dst),
1117 (ins DPR:$src1, DPR:$src2, i32imm:$SIMM), IIC_VSHLiD,
1118 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1119 [(set DPR:$dst, (Ty (ShOp DPR:$src1, DPR:$src2, (i32 imm:$SIMM))))]>;
1120 class N2VQShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1121 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1122 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$dst),
1123 (ins QPR:$src1, QPR:$src2, i32imm:$SIMM), IIC_VSHLiQ,
1124 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1125 [(set QPR:$dst, (Ty (ShOp QPR:$src1, QPR:$src2, (i32 imm:$SIMM))))]>;
1127 // Convert, with fractional bits immediate,
1128 // both double- and quad-register.
1129 class N2VCvtD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1130 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1132 : N2VImm<op24, op23, op11_8, op7, 0, op4,
1133 (outs DPR:$dst), (ins DPR:$src, i32imm:$SIMM), IIC_VUNAD,
1134 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1135 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src), (i32 imm:$SIMM))))]>;
1136 class N2VCvtQ<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1137 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1139 : N2VImm<op24, op23, op11_8, op7, 1, op4,
1140 (outs QPR:$dst), (ins QPR:$src, i32imm:$SIMM), IIC_VUNAQ,
1141 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1142 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src), (i32 imm:$SIMM))))]>;
1144 //===----------------------------------------------------------------------===//
1146 //===----------------------------------------------------------------------===//
1148 // Abbreviations used in multiclass suffixes:
1149 // Q = quarter int (8 bit) elements
1150 // H = half int (16 bit) elements
1151 // S = single int (32 bit) elements
1152 // D = double int (64 bit) elements
1154 // Neon 2-register vector operations -- for disassembly only.
1156 // First with only element sizes of 8, 16 and 32 bits:
1157 multiclass N2V_QHS_cmp<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1158 bits<5> op11_7, bit op4, string opc, string Dt,
1160 // 64-bit vector types.
1161 def v8i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 0, op4,
1162 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1163 opc, !strconcat(Dt, "8"), asm, "", []>;
1164 def v4i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 0, op4,
1165 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1166 opc, !strconcat(Dt, "16"), asm, "", []>;
1167 def v2i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
1168 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1169 opc, !strconcat(Dt, "32"), asm, "", []>;
1170 def v2f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
1171 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1172 opc, "f32", asm, "", []> {
1173 let Inst{10} = 1; // overwrite F = 1
1176 // 128-bit vector types.
1177 def v16i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 1, op4,
1178 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1179 opc, !strconcat(Dt, "8"), asm, "", []>;
1180 def v8i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 1, op4,
1181 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1182 opc, !strconcat(Dt, "16"), asm, "", []>;
1183 def v4i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
1184 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1185 opc, !strconcat(Dt, "32"), asm, "", []>;
1186 def v4f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
1187 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1188 opc, "f32", asm, "", []> {
1189 let Inst{10} = 1; // overwrite F = 1
1193 // Neon 3-register vector operations.
1195 // First with only element sizes of 8, 16 and 32 bits:
1196 multiclass N3V_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1197 InstrItinClass itinD16, InstrItinClass itinD32,
1198 InstrItinClass itinQ16, InstrItinClass itinQ32,
1199 string OpcodeStr, string Dt,
1200 SDNode OpNode, bit Commutable = 0> {
1201 // 64-bit vector types.
1202 def v8i8 : N3VD<op24, op23, 0b00, op11_8, op4, itinD16,
1203 OpcodeStr, !strconcat(Dt, "8"),
1204 v8i8, v8i8, OpNode, Commutable>;
1205 def v4i16 : N3VD<op24, op23, 0b01, op11_8, op4, itinD16,
1206 OpcodeStr, !strconcat(Dt, "16"),
1207 v4i16, v4i16, OpNode, Commutable>;
1208 def v2i32 : N3VD<op24, op23, 0b10, op11_8, op4, itinD32,
1209 OpcodeStr, !strconcat(Dt, "32"),
1210 v2i32, v2i32, OpNode, Commutable>;
1212 // 128-bit vector types.
1213 def v16i8 : N3VQ<op24, op23, 0b00, op11_8, op4, itinQ16,
1214 OpcodeStr, !strconcat(Dt, "8"),
1215 v16i8, v16i8, OpNode, Commutable>;
1216 def v8i16 : N3VQ<op24, op23, 0b01, op11_8, op4, itinQ16,
1217 OpcodeStr, !strconcat(Dt, "16"),
1218 v8i16, v8i16, OpNode, Commutable>;
1219 def v4i32 : N3VQ<op24, op23, 0b10, op11_8, op4, itinQ32,
1220 OpcodeStr, !strconcat(Dt, "32"),
1221 v4i32, v4i32, OpNode, Commutable>;
1224 multiclass N3VSL_HS<bits<4> op11_8, string OpcodeStr, string Dt, SDNode ShOp> {
1225 def v4i16 : N3VDSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
1227 def v2i32 : N3VDSL<0b10, op11_8, IIC_VMULi32D, OpcodeStr, !strconcat(Dt,"32"),
1229 def v8i16 : N3VQSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
1230 v8i16, v4i16, ShOp>;
1231 def v4i32 : N3VQSL<0b10, op11_8, IIC_VMULi32Q, OpcodeStr, !strconcat(Dt,"32"),
1232 v4i32, v2i32, ShOp>;
1235 // ....then also with element size 64 bits:
1236 multiclass N3V_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1237 InstrItinClass itinD, InstrItinClass itinQ,
1238 string OpcodeStr, string Dt,
1239 SDNode OpNode, bit Commutable = 0>
1240 : N3V_QHS<op24, op23, op11_8, op4, itinD, itinD, itinQ, itinQ,
1241 OpcodeStr, Dt, OpNode, Commutable> {
1242 def v1i64 : N3VD<op24, op23, 0b11, op11_8, op4, itinD,
1243 OpcodeStr, !strconcat(Dt, "64"),
1244 v1i64, v1i64, OpNode, Commutable>;
1245 def v2i64 : N3VQ<op24, op23, 0b11, op11_8, op4, itinQ,
1246 OpcodeStr, !strconcat(Dt, "64"),
1247 v2i64, v2i64, OpNode, Commutable>;
1251 // Neon Narrowing 2-register vector intrinsics,
1252 // source operand element sizes of 16, 32 and 64 bits:
1253 multiclass N2VNInt_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1254 bits<5> op11_7, bit op6, bit op4,
1255 InstrItinClass itin, string OpcodeStr, string Dt,
1257 def v8i8 : N2VNInt<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
1258 itin, OpcodeStr, !strconcat(Dt, "16"),
1259 v8i8, v8i16, IntOp>;
1260 def v4i16 : N2VNInt<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
1261 itin, OpcodeStr, !strconcat(Dt, "32"),
1262 v4i16, v4i32, IntOp>;
1263 def v2i32 : N2VNInt<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
1264 itin, OpcodeStr, !strconcat(Dt, "64"),
1265 v2i32, v2i64, IntOp>;
1269 // Neon Lengthening 2-register vector intrinsic (currently specific to VMOVL).
1270 // source operand element sizes of 16, 32 and 64 bits:
1271 multiclass N2VLInt_QHS<bits<2> op24_23, bits<5> op11_7, bit op6, bit op4,
1272 string OpcodeStr, string Dt, Intrinsic IntOp> {
1273 def v8i16 : N2VLInt<op24_23, 0b00, 0b10, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
1274 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
1275 def v4i32 : N2VLInt<op24_23, 0b01, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
1276 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
1277 def v2i64 : N2VLInt<op24_23, 0b10, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
1278 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1282 // Neon 3-register vector intrinsics.
1284 // First with only element sizes of 16 and 32 bits:
1285 multiclass N3VInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
1286 InstrItinClass itinD16, InstrItinClass itinD32,
1287 InstrItinClass itinQ16, InstrItinClass itinQ32,
1288 string OpcodeStr, string Dt,
1289 Intrinsic IntOp, bit Commutable = 0> {
1290 // 64-bit vector types.
1291 def v4i16 : N3VDInt<op24, op23, 0b01, op11_8, op4, itinD16,
1292 OpcodeStr, !strconcat(Dt, "16"),
1293 v4i16, v4i16, IntOp, Commutable>;
1294 def v2i32 : N3VDInt<op24, op23, 0b10, op11_8, op4, itinD32,
1295 OpcodeStr, !strconcat(Dt, "32"),
1296 v2i32, v2i32, IntOp, Commutable>;
1298 // 128-bit vector types.
1299 def v8i16 : N3VQInt<op24, op23, 0b01, op11_8, op4, itinQ16,
1300 OpcodeStr, !strconcat(Dt, "16"),
1301 v8i16, v8i16, IntOp, Commutable>;
1302 def v4i32 : N3VQInt<op24, op23, 0b10, op11_8, op4, itinQ32,
1303 OpcodeStr, !strconcat(Dt, "32"),
1304 v4i32, v4i32, IntOp, Commutable>;
1307 multiclass N3VIntSL_HS<bits<4> op11_8,
1308 InstrItinClass itinD16, InstrItinClass itinD32,
1309 InstrItinClass itinQ16, InstrItinClass itinQ32,
1310 string OpcodeStr, string Dt, Intrinsic IntOp> {
1311 def v4i16 : N3VDIntSL16<0b01, op11_8, itinD16,
1312 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp>;
1313 def v2i32 : N3VDIntSL<0b10, op11_8, itinD32,
1314 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp>;
1315 def v8i16 : N3VQIntSL16<0b01, op11_8, itinQ16,
1316 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16, IntOp>;
1317 def v4i32 : N3VQIntSL<0b10, op11_8, itinQ32,
1318 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32, IntOp>;
1321 // ....then also with element size of 8 bits:
1322 multiclass N3VInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1323 InstrItinClass itinD16, InstrItinClass itinD32,
1324 InstrItinClass itinQ16, InstrItinClass itinQ32,
1325 string OpcodeStr, string Dt,
1326 Intrinsic IntOp, bit Commutable = 0>
1327 : N3VInt_HS<op24, op23, op11_8, op4, itinD16, itinD32, itinQ16, itinQ32,
1328 OpcodeStr, Dt, IntOp, Commutable> {
1329 def v8i8 : N3VDInt<op24, op23, 0b00, op11_8, op4, itinD16,
1330 OpcodeStr, !strconcat(Dt, "8"),
1331 v8i8, v8i8, IntOp, Commutable>;
1332 def v16i8 : N3VQInt<op24, op23, 0b00, op11_8, op4, itinQ16,
1333 OpcodeStr, !strconcat(Dt, "8"),
1334 v16i8, v16i8, IntOp, Commutable>;
1337 // ....then also with element size of 64 bits:
1338 multiclass N3VInt_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1339 InstrItinClass itinD16, InstrItinClass itinD32,
1340 InstrItinClass itinQ16, InstrItinClass itinQ32,
1341 string OpcodeStr, string Dt,
1342 Intrinsic IntOp, bit Commutable = 0>
1343 : N3VInt_QHS<op24, op23, op11_8, op4, itinD16, itinD32, itinQ16, itinQ32,
1344 OpcodeStr, Dt, IntOp, Commutable> {
1345 def v1i64 : N3VDInt<op24, op23, 0b11, op11_8, op4, itinD32,
1346 OpcodeStr, !strconcat(Dt, "64"),
1347 v1i64, v1i64, IntOp, Commutable>;
1348 def v2i64 : N3VQInt<op24, op23, 0b11, op11_8, op4, itinQ32,
1349 OpcodeStr, !strconcat(Dt, "64"),
1350 v2i64, v2i64, IntOp, Commutable>;
1354 // Neon Narrowing 3-register vector intrinsics,
1355 // source operand element sizes of 16, 32 and 64 bits:
1356 multiclass N3VNInt_HSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1357 string OpcodeStr, string Dt,
1358 Intrinsic IntOp, bit Commutable = 0> {
1359 def v8i8 : N3VNInt<op24, op23, 0b00, op11_8, op4,
1360 OpcodeStr, !strconcat(Dt, "16"),
1361 v8i8, v8i16, IntOp, Commutable>;
1362 def v4i16 : N3VNInt<op24, op23, 0b01, op11_8, op4,
1363 OpcodeStr, !strconcat(Dt, "32"),
1364 v4i16, v4i32, IntOp, Commutable>;
1365 def v2i32 : N3VNInt<op24, op23, 0b10, op11_8, op4,
1366 OpcodeStr, !strconcat(Dt, "64"),
1367 v2i32, v2i64, IntOp, Commutable>;
1371 // Neon Long 3-register vector intrinsics.
1373 // First with only element sizes of 16 and 32 bits:
1374 multiclass N3VLInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
1375 InstrItinClass itin, string OpcodeStr, string Dt,
1376 Intrinsic IntOp, bit Commutable = 0> {
1377 def v4i32 : N3VLInt<op24, op23, 0b01, op11_8, op4, itin,
1378 OpcodeStr, !strconcat(Dt, "16"),
1379 v4i32, v4i16, IntOp, Commutable>;
1380 def v2i64 : N3VLInt<op24, op23, 0b10, op11_8, op4, itin,
1381 OpcodeStr, !strconcat(Dt, "32"),
1382 v2i64, v2i32, IntOp, Commutable>;
1385 multiclass N3VLIntSL_HS<bit op24, bits<4> op11_8,
1386 InstrItinClass itin, string OpcodeStr, string Dt,
1388 def v4i16 : N3VLIntSL16<op24, 0b01, op11_8, itin,
1389 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
1390 def v2i32 : N3VLIntSL<op24, 0b10, op11_8, itin,
1391 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1394 // ....then also with element size of 8 bits:
1395 multiclass N3VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1396 InstrItinClass itin, string OpcodeStr, string Dt,
1397 Intrinsic IntOp, bit Commutable = 0>
1398 : N3VLInt_HS<op24, op23, op11_8, op4, itin, OpcodeStr, Dt,
1399 IntOp, Commutable> {
1400 def v8i16 : N3VLInt<op24, op23, 0b00, op11_8, op4, itin,
1401 OpcodeStr, !strconcat(Dt, "8"),
1402 v8i16, v8i8, IntOp, Commutable>;
1406 // Neon Wide 3-register vector intrinsics,
1407 // source operand element sizes of 8, 16 and 32 bits:
1408 multiclass N3VWInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1409 string OpcodeStr, string Dt,
1410 Intrinsic IntOp, bit Commutable = 0> {
1411 def v8i16 : N3VWInt<op24, op23, 0b00, op11_8, op4,
1412 OpcodeStr, !strconcat(Dt, "8"),
1413 v8i16, v8i8, IntOp, Commutable>;
1414 def v4i32 : N3VWInt<op24, op23, 0b01, op11_8, op4,
1415 OpcodeStr, !strconcat(Dt, "16"),
1416 v4i32, v4i16, IntOp, Commutable>;
1417 def v2i64 : N3VWInt<op24, op23, 0b10, op11_8, op4,
1418 OpcodeStr, !strconcat(Dt, "32"),
1419 v2i64, v2i32, IntOp, Commutable>;
1423 // Neon Multiply-Op vector operations,
1424 // element sizes of 8, 16 and 32 bits:
1425 multiclass N3VMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1426 InstrItinClass itinD16, InstrItinClass itinD32,
1427 InstrItinClass itinQ16, InstrItinClass itinQ32,
1428 string OpcodeStr, string Dt, SDNode OpNode> {
1429 // 64-bit vector types.
1430 def v8i8 : N3VDMulOp<op24, op23, 0b00, op11_8, op4, itinD16,
1431 OpcodeStr, !strconcat(Dt, "8"), v8i8, mul, OpNode>;
1432 def v4i16 : N3VDMulOp<op24, op23, 0b01, op11_8, op4, itinD16,
1433 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, OpNode>;
1434 def v2i32 : N3VDMulOp<op24, op23, 0b10, op11_8, op4, itinD32,
1435 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, OpNode>;
1437 // 128-bit vector types.
1438 def v16i8 : N3VQMulOp<op24, op23, 0b00, op11_8, op4, itinQ16,
1439 OpcodeStr, !strconcat(Dt, "8"), v16i8, mul, OpNode>;
1440 def v8i16 : N3VQMulOp<op24, op23, 0b01, op11_8, op4, itinQ16,
1441 OpcodeStr, !strconcat(Dt, "16"), v8i16, mul, OpNode>;
1442 def v4i32 : N3VQMulOp<op24, op23, 0b10, op11_8, op4, itinQ32,
1443 OpcodeStr, !strconcat(Dt, "32"), v4i32, mul, OpNode>;
1446 multiclass N3VMulOpSL_HS<bits<4> op11_8,
1447 InstrItinClass itinD16, InstrItinClass itinD32,
1448 InstrItinClass itinQ16, InstrItinClass itinQ32,
1449 string OpcodeStr, string Dt, SDNode ShOp> {
1450 def v4i16 : N3VDMulOpSL16<0b01, op11_8, itinD16,
1451 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, ShOp>;
1452 def v2i32 : N3VDMulOpSL<0b10, op11_8, itinD32,
1453 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, ShOp>;
1454 def v8i16 : N3VQMulOpSL16<0b01, op11_8, itinQ16,
1455 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16,
1457 def v4i32 : N3VQMulOpSL<0b10, op11_8, itinQ32,
1458 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32,
1462 // Neon 3-argument intrinsics,
1463 // element sizes of 8, 16 and 32 bits:
1464 multiclass N3VInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1465 string OpcodeStr, string Dt, Intrinsic IntOp> {
1466 // 64-bit vector types.
1467 def v8i8 : N3VDInt3<op24, op23, 0b00, op11_8, op4, IIC_VMACi16D,
1468 OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
1469 def v4i16 : N3VDInt3<op24, op23, 0b01, op11_8, op4, IIC_VMACi16D,
1470 OpcodeStr, !strconcat(Dt, "16"), v4i16, v4i16, IntOp>;
1471 def v2i32 : N3VDInt3<op24, op23, 0b10, op11_8, op4, IIC_VMACi32D,
1472 OpcodeStr, !strconcat(Dt, "32"), v2i32, v2i32, IntOp>;
1474 // 128-bit vector types.
1475 def v16i8 : N3VQInt3<op24, op23, 0b00, op11_8, op4, IIC_VMACi16Q,
1476 OpcodeStr, !strconcat(Dt, "8"), v16i8, v16i8, IntOp>;
1477 def v8i16 : N3VQInt3<op24, op23, 0b01, op11_8, op4, IIC_VMACi16Q,
1478 OpcodeStr, !strconcat(Dt, "16"), v8i16, v8i16, IntOp>;
1479 def v4i32 : N3VQInt3<op24, op23, 0b10, op11_8, op4, IIC_VMACi32Q,
1480 OpcodeStr, !strconcat(Dt, "32"), v4i32, v4i32, IntOp>;
1484 // Neon Long 3-argument intrinsics.
1486 // First with only element sizes of 16 and 32 bits:
1487 multiclass N3VLInt3_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
1488 string OpcodeStr, string Dt, Intrinsic IntOp> {
1489 def v4i32 : N3VLInt3<op24, op23, 0b01, op11_8, op4, IIC_VMACi16D,
1490 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
1491 def v2i64 : N3VLInt3<op24, op23, 0b10, op11_8, op4, IIC_VMACi16D,
1492 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1495 multiclass N3VLInt3SL_HS<bit op24, bits<4> op11_8,
1496 string OpcodeStr, string Dt, Intrinsic IntOp> {
1497 def v4i16 : N3VLInt3SL16<op24, 0b01, op11_8, IIC_VMACi16D,
1498 OpcodeStr, !strconcat(Dt,"16"), v4i32, v4i16, IntOp>;
1499 def v2i32 : N3VLInt3SL<op24, 0b10, op11_8, IIC_VMACi32D,
1500 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1503 // ....then also with element size of 8 bits:
1504 multiclass N3VLInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1505 string OpcodeStr, string Dt, Intrinsic IntOp>
1506 : N3VLInt3_HS<op24, op23, op11_8, op4, OpcodeStr, Dt, IntOp> {
1507 def v8i16 : N3VLInt3<op24, op23, 0b00, op11_8, op4, IIC_VMACi16D,
1508 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
1512 // Neon 2-register vector intrinsics,
1513 // element sizes of 8, 16 and 32 bits:
1514 multiclass N2VInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1515 bits<5> op11_7, bit op4,
1516 InstrItinClass itinD, InstrItinClass itinQ,
1517 string OpcodeStr, string Dt, Intrinsic IntOp> {
1518 // 64-bit vector types.
1519 def v8i8 : N2VDInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1520 itinD, OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
1521 def v4i16 : N2VDInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1522 itinD, OpcodeStr, !strconcat(Dt, "16"),v4i16,v4i16,IntOp>;
1523 def v2i32 : N2VDInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1524 itinD, OpcodeStr, !strconcat(Dt, "32"),v2i32,v2i32,IntOp>;
1526 // 128-bit vector types.
1527 def v16i8 : N2VQInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1528 itinQ, OpcodeStr, !strconcat(Dt, "8"), v16i8,v16i8,IntOp>;
1529 def v8i16 : N2VQInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1530 itinQ, OpcodeStr, !strconcat(Dt, "16"),v8i16,v8i16,IntOp>;
1531 def v4i32 : N2VQInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1532 itinQ, OpcodeStr, !strconcat(Dt, "32"),v4i32,v4i32,IntOp>;
1536 // Neon Pairwise long 2-register intrinsics,
1537 // element sizes of 8, 16 and 32 bits:
1538 multiclass N2VPLInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1539 bits<5> op11_7, bit op4,
1540 string OpcodeStr, string Dt, Intrinsic IntOp> {
1541 // 64-bit vector types.
1542 def v8i8 : N2VDPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1543 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
1544 def v4i16 : N2VDPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1545 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
1546 def v2i32 : N2VDPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1547 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
1549 // 128-bit vector types.
1550 def v16i8 : N2VQPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1551 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
1552 def v8i16 : N2VQPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1553 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
1554 def v4i32 : N2VQPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1555 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
1559 // Neon Pairwise long 2-register accumulate intrinsics,
1560 // element sizes of 8, 16 and 32 bits:
1561 multiclass N2VPLInt2_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1562 bits<5> op11_7, bit op4,
1563 string OpcodeStr, string Dt, Intrinsic IntOp> {
1564 // 64-bit vector types.
1565 def v8i8 : N2VDPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1566 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
1567 def v4i16 : N2VDPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1568 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
1569 def v2i32 : N2VDPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1570 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
1572 // 128-bit vector types.
1573 def v16i8 : N2VQPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1574 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
1575 def v8i16 : N2VQPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1576 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
1577 def v4i32 : N2VQPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1578 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
1582 // Neon 2-register vector shift by immediate,
1583 // element sizes of 8, 16, 32 and 64 bits:
1584 multiclass N2VSh_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1585 InstrItinClass itin, string OpcodeStr, string Dt,
1587 // 64-bit vector types.
1588 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, itin,
1589 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
1590 let Inst{21-19} = 0b001; // imm6 = 001xxx
1592 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, itin,
1593 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
1594 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1596 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, itin,
1597 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
1598 let Inst{21} = 0b1; // imm6 = 1xxxxx
1600 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, itin,
1601 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
1604 // 128-bit vector types.
1605 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, itin,
1606 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
1607 let Inst{21-19} = 0b001; // imm6 = 001xxx
1609 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, itin,
1610 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
1611 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1613 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, itin,
1614 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
1615 let Inst{21} = 0b1; // imm6 = 1xxxxx
1617 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, itin,
1618 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
1623 // Neon Shift-Accumulate vector operations,
1624 // element sizes of 8, 16, 32 and 64 bits:
1625 multiclass N2VShAdd_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1626 string OpcodeStr, string Dt, SDNode ShOp> {
1627 // 64-bit vector types.
1628 def v8i8 : N2VDShAdd<op24, op23, op11_8, 0, op4,
1629 OpcodeStr, !strconcat(Dt, "8"), v8i8, ShOp> {
1630 let Inst{21-19} = 0b001; // imm6 = 001xxx
1632 def v4i16 : N2VDShAdd<op24, op23, op11_8, 0, op4,
1633 OpcodeStr, !strconcat(Dt, "16"), v4i16, ShOp> {
1634 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1636 def v2i32 : N2VDShAdd<op24, op23, op11_8, 0, op4,
1637 OpcodeStr, !strconcat(Dt, "32"), v2i32, ShOp> {
1638 let Inst{21} = 0b1; // imm6 = 1xxxxx
1640 def v1i64 : N2VDShAdd<op24, op23, op11_8, 1, op4,
1641 OpcodeStr, !strconcat(Dt, "64"), v1i64, ShOp>;
1644 // 128-bit vector types.
1645 def v16i8 : N2VQShAdd<op24, op23, op11_8, 0, op4,
1646 OpcodeStr, !strconcat(Dt, "8"), v16i8, ShOp> {
1647 let Inst{21-19} = 0b001; // imm6 = 001xxx
1649 def v8i16 : N2VQShAdd<op24, op23, op11_8, 0, op4,
1650 OpcodeStr, !strconcat(Dt, "16"), v8i16, ShOp> {
1651 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1653 def v4i32 : N2VQShAdd<op24, op23, op11_8, 0, op4,
1654 OpcodeStr, !strconcat(Dt, "32"), v4i32, ShOp> {
1655 let Inst{21} = 0b1; // imm6 = 1xxxxx
1657 def v2i64 : N2VQShAdd<op24, op23, op11_8, 1, op4,
1658 OpcodeStr, !strconcat(Dt, "64"), v2i64, ShOp>;
1663 // Neon Shift-Insert vector operations,
1664 // element sizes of 8, 16, 32 and 64 bits:
1665 multiclass N2VShIns_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1666 string OpcodeStr, SDNode ShOp> {
1667 // 64-bit vector types.
1668 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4,
1669 OpcodeStr, "8", v8i8, ShOp> {
1670 let Inst{21-19} = 0b001; // imm6 = 001xxx
1672 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4,
1673 OpcodeStr, "16", v4i16, ShOp> {
1674 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1676 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4,
1677 OpcodeStr, "32", v2i32, ShOp> {
1678 let Inst{21} = 0b1; // imm6 = 1xxxxx
1680 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4,
1681 OpcodeStr, "64", v1i64, ShOp>;
1684 // 128-bit vector types.
1685 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4,
1686 OpcodeStr, "8", v16i8, ShOp> {
1687 let Inst{21-19} = 0b001; // imm6 = 001xxx
1689 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4,
1690 OpcodeStr, "16", v8i16, ShOp> {
1691 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1693 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4,
1694 OpcodeStr, "32", v4i32, ShOp> {
1695 let Inst{21} = 0b1; // imm6 = 1xxxxx
1697 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4,
1698 OpcodeStr, "64", v2i64, ShOp>;
1702 // Neon Shift Long operations,
1703 // element sizes of 8, 16, 32 bits:
1704 multiclass N2VLSh_QHS<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
1705 bit op4, string OpcodeStr, string Dt, SDNode OpNode> {
1706 def v8i16 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
1707 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode> {
1708 let Inst{21-19} = 0b001; // imm6 = 001xxx
1710 def v4i32 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
1711 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode> {
1712 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1714 def v2i64 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
1715 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode> {
1716 let Inst{21} = 0b1; // imm6 = 1xxxxx
1720 // Neon Shift Narrow operations,
1721 // element sizes of 16, 32, 64 bits:
1722 multiclass N2VNSh_HSD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
1723 bit op4, InstrItinClass itin, string OpcodeStr, string Dt,
1725 def v8i8 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
1726 OpcodeStr, !strconcat(Dt, "16"), v8i8, v8i16, OpNode> {
1727 let Inst{21-19} = 0b001; // imm6 = 001xxx
1729 def v4i16 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
1730 OpcodeStr, !strconcat(Dt, "32"), v4i16, v4i32, OpNode> {
1731 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1733 def v2i32 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
1734 OpcodeStr, !strconcat(Dt, "64"), v2i32, v2i64, OpNode> {
1735 let Inst{21} = 0b1; // imm6 = 1xxxxx
1739 //===----------------------------------------------------------------------===//
1740 // Instruction Definitions.
1741 //===----------------------------------------------------------------------===//
1743 // Vector Add Operations.
1745 // VADD : Vector Add (integer and floating-point)
1746 defm VADD : N3V_QHSD<0, 0, 0b1000, 0, IIC_VBINiD, IIC_VBINiQ, "vadd", "i",
1748 def VADDfd : N3VD<0, 0, 0b00, 0b1101, 0, IIC_VBIND, "vadd", "f32",
1749 v2f32, v2f32, fadd, 1>;
1750 def VADDfq : N3VQ<0, 0, 0b00, 0b1101, 0, IIC_VBINQ, "vadd", "f32",
1751 v4f32, v4f32, fadd, 1>;
1752 // VADDL : Vector Add Long (Q = D + D)
1753 defm VADDLs : N3VLInt_QHS<0,1,0b0000,0, IIC_VSHLiD, "vaddl", "s",
1754 int_arm_neon_vaddls, 1>;
1755 defm VADDLu : N3VLInt_QHS<1,1,0b0000,0, IIC_VSHLiD, "vaddl", "u",
1756 int_arm_neon_vaddlu, 1>;
1757 // VADDW : Vector Add Wide (Q = Q + D)
1758 defm VADDWs : N3VWInt_QHS<0,1,0b0001,0, "vaddw", "s", int_arm_neon_vaddws, 0>;
1759 defm VADDWu : N3VWInt_QHS<1,1,0b0001,0, "vaddw", "u", int_arm_neon_vaddwu, 0>;
1760 // VHADD : Vector Halving Add
1761 defm VHADDs : N3VInt_QHS<0,0,0b0000,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1762 IIC_VBINi4Q, "vhadd", "s", int_arm_neon_vhadds, 1>;
1763 defm VHADDu : N3VInt_QHS<1,0,0b0000,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1764 IIC_VBINi4Q, "vhadd", "u", int_arm_neon_vhaddu, 1>;
1765 // VRHADD : Vector Rounding Halving Add
1766 defm VRHADDs : N3VInt_QHS<0,0,0b0001,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1767 IIC_VBINi4Q, "vrhadd", "s", int_arm_neon_vrhadds, 1>;
1768 defm VRHADDu : N3VInt_QHS<1,0,0b0001,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1769 IIC_VBINi4Q, "vrhadd", "u", int_arm_neon_vrhaddu, 1>;
1770 // VQADD : Vector Saturating Add
1771 defm VQADDs : N3VInt_QHSD<0,0,0b0000,1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1772 IIC_VBINi4Q, "vqadd", "s", int_arm_neon_vqadds, 1>;
1773 defm VQADDu : N3VInt_QHSD<1,0,0b0000,1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1774 IIC_VBINi4Q, "vqadd", "u", int_arm_neon_vqaddu, 1>;
1775 // VADDHN : Vector Add and Narrow Returning High Half (D = Q + Q)
1776 defm VADDHN : N3VNInt_HSD<0,1,0b0100,0, "vaddhn", "i",
1777 int_arm_neon_vaddhn, 1>;
1778 // VRADDHN : Vector Rounding Add and Narrow Returning High Half (D = Q + Q)
1779 defm VRADDHN : N3VNInt_HSD<1,1,0b0100,0, "vraddhn", "i",
1780 int_arm_neon_vraddhn, 1>;
1782 // Vector Multiply Operations.
1784 // VMUL : Vector Multiply (integer, polynomial and floating-point)
1785 defm VMUL : N3V_QHS<0, 0, 0b1001, 1, IIC_VMULi16D, IIC_VMULi32D,
1786 IIC_VMULi16Q, IIC_VMULi32Q, "vmul", "i", mul, 1>;
1787 def VMULpd : N3VDInt<1, 0, 0b00, 0b1001, 1, IIC_VMULi16D, "vmul", "p8",
1788 v8i8, v8i8, int_arm_neon_vmulp, 1>;
1789 def VMULpq : N3VQInt<1, 0, 0b00, 0b1001, 1, IIC_VMULi16Q, "vmul", "p8",
1790 v16i8, v16i8, int_arm_neon_vmulp, 1>;
1791 def VMULfd : N3VD<1, 0, 0b00, 0b1101, 1, IIC_VBIND, "vmul", "f32",
1792 v2f32, v2f32, fmul, 1>;
1793 def VMULfq : N3VQ<1, 0, 0b00, 0b1101, 1, IIC_VBINQ, "vmul", "f32",
1794 v4f32, v4f32, fmul, 1>;
1795 defm VMULsl : N3VSL_HS<0b1000, "vmul", "i", mul>;
1796 def VMULslfd : N3VDSL<0b10, 0b1001, IIC_VBIND, "vmul", "f32", v2f32, fmul>;
1797 def VMULslfq : N3VQSL<0b10, 0b1001, IIC_VBINQ, "vmul", "f32", v4f32,
1800 def : Pat<(v8i16 (mul (v8i16 QPR:$src1),
1801 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
1802 (v8i16 (VMULslv8i16 (v8i16 QPR:$src1),
1803 (v4i16 (EXTRACT_SUBREG QPR:$src2,
1804 (DSubReg_i16_reg imm:$lane))),
1805 (SubReg_i16_lane imm:$lane)))>;
1806 def : Pat<(v4i32 (mul (v4i32 QPR:$src1),
1807 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
1808 (v4i32 (VMULslv4i32 (v4i32 QPR:$src1),
1809 (v2i32 (EXTRACT_SUBREG QPR:$src2,
1810 (DSubReg_i32_reg imm:$lane))),
1811 (SubReg_i32_lane imm:$lane)))>;
1812 def : Pat<(v4f32 (fmul (v4f32 QPR:$src1),
1813 (v4f32 (NEONvduplane (v4f32 QPR:$src2), imm:$lane)))),
1814 (v4f32 (VMULslfq (v4f32 QPR:$src1),
1815 (v2f32 (EXTRACT_SUBREG QPR:$src2,
1816 (DSubReg_i32_reg imm:$lane))),
1817 (SubReg_i32_lane imm:$lane)))>;
1819 // VQDMULH : Vector Saturating Doubling Multiply Returning High Half
1820 defm VQDMULH : N3VInt_HS<0, 0, 0b1011, 0, IIC_VMULi16D, IIC_VMULi32D,
1821 IIC_VMULi16Q, IIC_VMULi32Q,
1822 "vqdmulh", "s", int_arm_neon_vqdmulh, 1>;
1823 defm VQDMULHsl: N3VIntSL_HS<0b1100, IIC_VMULi16D, IIC_VMULi32D,
1824 IIC_VMULi16Q, IIC_VMULi32Q,
1825 "vqdmulh", "s", int_arm_neon_vqdmulh>;
1826 def : Pat<(v8i16 (int_arm_neon_vqdmulh (v8i16 QPR:$src1),
1827 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
1829 (v8i16 (VQDMULHslv8i16 (v8i16 QPR:$src1),
1830 (v4i16 (EXTRACT_SUBREG QPR:$src2,
1831 (DSubReg_i16_reg imm:$lane))),
1832 (SubReg_i16_lane imm:$lane)))>;
1833 def : Pat<(v4i32 (int_arm_neon_vqdmulh (v4i32 QPR:$src1),
1834 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
1836 (v4i32 (VQDMULHslv4i32 (v4i32 QPR:$src1),
1837 (v2i32 (EXTRACT_SUBREG QPR:$src2,
1838 (DSubReg_i32_reg imm:$lane))),
1839 (SubReg_i32_lane imm:$lane)))>;
1841 // VQRDMULH : Vector Rounding Saturating Doubling Multiply Returning High Half
1842 defm VQRDMULH : N3VInt_HS<1, 0, 0b1011, 0, IIC_VMULi16D, IIC_VMULi32D,
1843 IIC_VMULi16Q, IIC_VMULi32Q,
1844 "vqrdmulh", "s", int_arm_neon_vqrdmulh, 1>;
1845 defm VQRDMULHsl : N3VIntSL_HS<0b1101, IIC_VMULi16D, IIC_VMULi32D,
1846 IIC_VMULi16Q, IIC_VMULi32Q,
1847 "vqrdmulh", "s", int_arm_neon_vqrdmulh>;
1848 def : Pat<(v8i16 (int_arm_neon_vqrdmulh (v8i16 QPR:$src1),
1849 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
1851 (v8i16 (VQRDMULHslv8i16 (v8i16 QPR:$src1),
1852 (v4i16 (EXTRACT_SUBREG QPR:$src2,
1853 (DSubReg_i16_reg imm:$lane))),
1854 (SubReg_i16_lane imm:$lane)))>;
1855 def : Pat<(v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$src1),
1856 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
1858 (v4i32 (VQRDMULHslv4i32 (v4i32 QPR:$src1),
1859 (v2i32 (EXTRACT_SUBREG QPR:$src2,
1860 (DSubReg_i32_reg imm:$lane))),
1861 (SubReg_i32_lane imm:$lane)))>;
1863 // VMULL : Vector Multiply Long (integer and polynomial) (Q = D * D)
1864 defm VMULLs : N3VLInt_QHS<0,1,0b1100,0, IIC_VMULi16D, "vmull", "s",
1865 int_arm_neon_vmulls, 1>;
1866 defm VMULLu : N3VLInt_QHS<1,1,0b1100,0, IIC_VMULi16D, "vmull", "u",
1867 int_arm_neon_vmullu, 1>;
1868 def VMULLp : N3VLInt<0, 1, 0b00, 0b1110, 0, IIC_VMULi16D, "vmull", "p8",
1869 v8i16, v8i8, int_arm_neon_vmullp, 1>;
1870 defm VMULLsls : N3VLIntSL_HS<0, 0b1010, IIC_VMULi16D, "vmull", "s",
1871 int_arm_neon_vmulls>;
1872 defm VMULLslu : N3VLIntSL_HS<1, 0b1010, IIC_VMULi16D, "vmull", "u",
1873 int_arm_neon_vmullu>;
1875 // VQDMULL : Vector Saturating Doubling Multiply Long (Q = D * D)
1876 defm VQDMULL : N3VLInt_HS<0,1,0b1101,0, IIC_VMULi16D, "vqdmull", "s",
1877 int_arm_neon_vqdmull, 1>;
1878 defm VQDMULLsl: N3VLIntSL_HS<0, 0b1011, IIC_VMULi16D, "vqdmull", "s",
1879 int_arm_neon_vqdmull>;
1881 // Vector Multiply-Accumulate and Multiply-Subtract Operations.
1883 // VMLA : Vector Multiply Accumulate (integer and floating-point)
1884 defm VMLA : N3VMulOp_QHS<0, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
1885 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
1886 def VMLAfd : N3VDMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACD, "vmla", "f32",
1888 def VMLAfq : N3VQMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACQ, "vmla", "f32",
1890 defm VMLAsl : N3VMulOpSL_HS<0b0000, IIC_VMACi16D, IIC_VMACi32D,
1891 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
1892 def VMLAslfd : N3VDMulOpSL<0b10, 0b0001, IIC_VMACD, "vmla", "f32",
1894 def VMLAslfq : N3VQMulOpSL<0b10, 0b0001, IIC_VMACQ, "vmla", "f32",
1895 v4f32, v2f32, fmul, fadd>;
1897 def : Pat<(v8i16 (add (v8i16 QPR:$src1),
1898 (mul (v8i16 QPR:$src2),
1899 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
1900 (v8i16 (VMLAslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
1901 (v4i16 (EXTRACT_SUBREG QPR:$src3,
1902 (DSubReg_i16_reg imm:$lane))),
1903 (SubReg_i16_lane imm:$lane)))>;
1905 def : Pat<(v4i32 (add (v4i32 QPR:$src1),
1906 (mul (v4i32 QPR:$src2),
1907 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
1908 (v4i32 (VMLAslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
1909 (v2i32 (EXTRACT_SUBREG QPR:$src3,
1910 (DSubReg_i32_reg imm:$lane))),
1911 (SubReg_i32_lane imm:$lane)))>;
1913 def : Pat<(v4f32 (fadd (v4f32 QPR:$src1),
1914 (fmul (v4f32 QPR:$src2),
1915 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
1916 (v4f32 (VMLAslfq (v4f32 QPR:$src1),
1918 (v2f32 (EXTRACT_SUBREG QPR:$src3,
1919 (DSubReg_i32_reg imm:$lane))),
1920 (SubReg_i32_lane imm:$lane)))>;
1922 // VMLAL : Vector Multiply Accumulate Long (Q += D * D)
1923 defm VMLALs : N3VLInt3_QHS<0,1,0b1000,0, "vmlal", "s", int_arm_neon_vmlals>;
1924 defm VMLALu : N3VLInt3_QHS<1,1,0b1000,0, "vmlal", "u", int_arm_neon_vmlalu>;
1926 defm VMLALsls : N3VLInt3SL_HS<0, 0b0010, "vmlal", "s", int_arm_neon_vmlals>;
1927 defm VMLALslu : N3VLInt3SL_HS<1, 0b0010, "vmlal", "u", int_arm_neon_vmlalu>;
1929 // VQDMLAL : Vector Saturating Doubling Multiply Accumulate Long (Q += D * D)
1930 defm VQDMLAL : N3VLInt3_HS<0, 1, 0b1001, 0, "vqdmlal", "s",
1931 int_arm_neon_vqdmlal>;
1932 defm VQDMLALsl: N3VLInt3SL_HS<0, 0b0011, "vqdmlal", "s", int_arm_neon_vqdmlal>;
1934 // VMLS : Vector Multiply Subtract (integer and floating-point)
1935 defm VMLS : N3VMulOp_QHS<1, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
1936 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
1937 def VMLSfd : N3VDMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACD, "vmls", "f32",
1939 def VMLSfq : N3VQMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACQ, "vmls", "f32",
1941 defm VMLSsl : N3VMulOpSL_HS<0b0100, IIC_VMACi16D, IIC_VMACi32D,
1942 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
1943 def VMLSslfd : N3VDMulOpSL<0b10, 0b0101, IIC_VMACD, "vmls", "f32",
1945 def VMLSslfq : N3VQMulOpSL<0b10, 0b0101, IIC_VMACQ, "vmls", "f32",
1946 v4f32, v2f32, fmul, fsub>;
1948 def : Pat<(v8i16 (sub (v8i16 QPR:$src1),
1949 (mul (v8i16 QPR:$src2),
1950 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
1951 (v8i16 (VMLSslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
1952 (v4i16 (EXTRACT_SUBREG QPR:$src3,
1953 (DSubReg_i16_reg imm:$lane))),
1954 (SubReg_i16_lane imm:$lane)))>;
1956 def : Pat<(v4i32 (sub (v4i32 QPR:$src1),
1957 (mul (v4i32 QPR:$src2),
1958 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
1959 (v4i32 (VMLSslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
1960 (v2i32 (EXTRACT_SUBREG QPR:$src3,
1961 (DSubReg_i32_reg imm:$lane))),
1962 (SubReg_i32_lane imm:$lane)))>;
1964 def : Pat<(v4f32 (fsub (v4f32 QPR:$src1),
1965 (fmul (v4f32 QPR:$src2),
1966 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
1967 (v4f32 (VMLSslfq (v4f32 QPR:$src1), (v4f32 QPR:$src2),
1968 (v2f32 (EXTRACT_SUBREG QPR:$src3,
1969 (DSubReg_i32_reg imm:$lane))),
1970 (SubReg_i32_lane imm:$lane)))>;
1972 // VMLSL : Vector Multiply Subtract Long (Q -= D * D)
1973 defm VMLSLs : N3VLInt3_QHS<0,1,0b1010,0, "vmlsl", "s", int_arm_neon_vmlsls>;
1974 defm VMLSLu : N3VLInt3_QHS<1,1,0b1010,0, "vmlsl", "u", int_arm_neon_vmlslu>;
1976 defm VMLSLsls : N3VLInt3SL_HS<0, 0b0110, "vmlsl", "s", int_arm_neon_vmlsls>;
1977 defm VMLSLslu : N3VLInt3SL_HS<1, 0b0110, "vmlsl", "u", int_arm_neon_vmlslu>;
1979 // VQDMLSL : Vector Saturating Doubling Multiply Subtract Long (Q -= D * D)
1980 defm VQDMLSL : N3VLInt3_HS<0, 1, 0b1011, 0, "vqdmlsl", "s",
1981 int_arm_neon_vqdmlsl>;
1982 defm VQDMLSLsl: N3VLInt3SL_HS<0, 0b111, "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
1984 // Vector Subtract Operations.
1986 // VSUB : Vector Subtract (integer and floating-point)
1987 defm VSUB : N3V_QHSD<1, 0, 0b1000, 0, IIC_VSUBiD, IIC_VSUBiQ,
1988 "vsub", "i", sub, 0>;
1989 def VSUBfd : N3VD<0, 0, 0b10, 0b1101, 0, IIC_VBIND, "vsub", "f32",
1990 v2f32, v2f32, fsub, 0>;
1991 def VSUBfq : N3VQ<0, 0, 0b10, 0b1101, 0, IIC_VBINQ, "vsub", "f32",
1992 v4f32, v4f32, fsub, 0>;
1993 // VSUBL : Vector Subtract Long (Q = D - D)
1994 defm VSUBLs : N3VLInt_QHS<0,1,0b0010,0, IIC_VSHLiD, "vsubl", "s",
1995 int_arm_neon_vsubls, 1>;
1996 defm VSUBLu : N3VLInt_QHS<1,1,0b0010,0, IIC_VSHLiD, "vsubl", "u",
1997 int_arm_neon_vsublu, 1>;
1998 // VSUBW : Vector Subtract Wide (Q = Q - D)
1999 defm VSUBWs : N3VWInt_QHS<0,1,0b0011,0, "vsubw", "s", int_arm_neon_vsubws, 0>;
2000 defm VSUBWu : N3VWInt_QHS<1,1,0b0011,0, "vsubw", "u", int_arm_neon_vsubwu, 0>;
2001 // VHSUB : Vector Halving Subtract
2002 defm VHSUBs : N3VInt_QHS<0, 0, 0b0010, 0, IIC_VBINi4D, IIC_VBINi4D,
2003 IIC_VBINi4Q, IIC_VBINi4Q,
2004 "vhsub", "s", int_arm_neon_vhsubs, 0>;
2005 defm VHSUBu : N3VInt_QHS<1, 0, 0b0010, 0, IIC_VBINi4D, IIC_VBINi4D,
2006 IIC_VBINi4Q, IIC_VBINi4Q,
2007 "vhsub", "u", int_arm_neon_vhsubu, 0>;
2008 // VQSUB : Vector Saturing Subtract
2009 defm VQSUBs : N3VInt_QHSD<0, 0, 0b0010, 1, IIC_VBINi4D, IIC_VBINi4D,
2010 IIC_VBINi4Q, IIC_VBINi4Q,
2011 "vqsub", "s", int_arm_neon_vqsubs, 0>;
2012 defm VQSUBu : N3VInt_QHSD<1, 0, 0b0010, 1, IIC_VBINi4D, IIC_VBINi4D,
2013 IIC_VBINi4Q, IIC_VBINi4Q,
2014 "vqsub", "u", int_arm_neon_vqsubu, 0>;
2015 // VSUBHN : Vector Subtract and Narrow Returning High Half (D = Q - Q)
2016 defm VSUBHN : N3VNInt_HSD<0,1,0b0110,0, "vsubhn", "i",
2017 int_arm_neon_vsubhn, 0>;
2018 // VRSUBHN : Vector Rounding Subtract and Narrow Returning High Half (D=Q-Q)
2019 defm VRSUBHN : N3VNInt_HSD<1,1,0b0110,0, "vrsubhn", "i",
2020 int_arm_neon_vrsubhn, 0>;
2022 // Vector Comparisons.
2024 // VCEQ : Vector Compare Equal
2025 defm VCEQ : N3V_QHS<1, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2026 IIC_VBINi4Q, "vceq", "i", NEONvceq, 1>;
2027 def VCEQfd : N3VD<0,0,0b00,0b1110,0, IIC_VBIND, "vceq", "f32", v2i32, v2f32,
2029 def VCEQfq : N3VQ<0,0,0b00,0b1110,0, IIC_VBINQ, "vceq", "f32", v4i32, v4f32,
2031 // For disassembly only.
2032 defm VCEQz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00010, 0, "vceq", "i",
2035 // VCGE : Vector Compare Greater Than or Equal
2036 defm VCGEs : N3V_QHS<0, 0, 0b0011, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2037 IIC_VBINi4Q, "vcge", "s", NEONvcge, 0>;
2038 defm VCGEu : N3V_QHS<1, 0, 0b0011, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2039 IIC_VBINi4Q, "vcge", "u", NEONvcgeu, 0>;
2040 def VCGEfd : N3VD<1,0,0b00,0b1110,0, IIC_VBIND, "vcge", "f32",
2041 v2i32, v2f32, NEONvcge, 0>;
2042 def VCGEfq : N3VQ<1,0,0b00,0b1110,0, IIC_VBINQ, "vcge", "f32", v4i32, v4f32,
2044 // For disassembly only.
2045 defm VCGEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00001, 0, "vcge", "s",
2047 // For disassembly only.
2048 defm VCLEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00011, 0, "vcle", "s",
2051 // VCGT : Vector Compare Greater Than
2052 defm VCGTs : N3V_QHS<0, 0, 0b0011, 0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2053 IIC_VBINi4Q, "vcgt", "s", NEONvcgt, 0>;
2054 defm VCGTu : N3V_QHS<1, 0, 0b0011, 0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2055 IIC_VBINi4Q, "vcgt", "u", NEONvcgtu, 0>;
2056 def VCGTfd : N3VD<1,0,0b10,0b1110,0, IIC_VBIND, "vcgt", "f32", v2i32, v2f32,
2058 def VCGTfq : N3VQ<1,0,0b10,0b1110,0, IIC_VBINQ, "vcgt", "f32", v4i32, v4f32,
2060 // For disassembly only.
2061 defm VCGTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00000, 0, "vcgt", "s",
2063 // For disassembly only.
2064 defm VCLTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00100, 0, "vclt", "s",
2067 // VACGE : Vector Absolute Compare Greater Than or Equal (aka VCAGE)
2068 def VACGEd : N3VDInt<1, 0, 0b00, 0b1110, 1, IIC_VBIND, "vacge", "f32",
2069 v2i32, v2f32, int_arm_neon_vacged, 0>;
2070 def VACGEq : N3VQInt<1, 0, 0b00, 0b1110, 1, IIC_VBINQ, "vacge", "f32",
2071 v4i32, v4f32, int_arm_neon_vacgeq, 0>;
2072 // VACGT : Vector Absolute Compare Greater Than (aka VCAGT)
2073 def VACGTd : N3VDInt<1, 0, 0b10, 0b1110, 1, IIC_VBIND, "vacgt", "f32",
2074 v2i32, v2f32, int_arm_neon_vacgtd, 0>;
2075 def VACGTq : N3VQInt<1, 0, 0b10, 0b1110, 1, IIC_VBINQ, "vacgt", "f32",
2076 v4i32, v4f32, int_arm_neon_vacgtq, 0>;
2077 // VTST : Vector Test Bits
2078 defm VTST : N3V_QHS<0, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2079 IIC_VBINi4Q, "vtst", "", NEONvtst, 1>;
2081 // Vector Bitwise Operations.
2083 // VAND : Vector Bitwise AND
2084 def VANDd : N3VDX<0, 0, 0b00, 0b0001, 1, IIC_VBINiD, "vand",
2085 v2i32, v2i32, and, 1>;
2086 def VANDq : N3VQX<0, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "vand",
2087 v4i32, v4i32, and, 1>;
2089 // VEOR : Vector Bitwise Exclusive OR
2090 def VEORd : N3VDX<1, 0, 0b00, 0b0001, 1, IIC_VBINiD, "veor",
2091 v2i32, v2i32, xor, 1>;
2092 def VEORq : N3VQX<1, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "veor",
2093 v4i32, v4i32, xor, 1>;
2095 // VORR : Vector Bitwise OR
2096 def VORRd : N3VDX<0, 0, 0b10, 0b0001, 1, IIC_VBINiD, "vorr",
2097 v2i32, v2i32, or, 1>;
2098 def VORRq : N3VQX<0, 0, 0b10, 0b0001, 1, IIC_VBINiQ, "vorr",
2099 v4i32, v4i32, or, 1>;
2101 // VBIC : Vector Bitwise Bit Clear (AND NOT)
2102 def VBICd : N3VX<0, 0, 0b01, 0b0001, 0, 1, (outs DPR:$dst),
2103 (ins DPR:$src1, DPR:$src2), IIC_VBINiD,
2104 "vbic", "$dst, $src1, $src2", "",
2105 [(set DPR:$dst, (v2i32 (and DPR:$src1,
2106 (vnot_conv DPR:$src2))))]>;
2107 def VBICq : N3VX<0, 0, 0b01, 0b0001, 1, 1, (outs QPR:$dst),
2108 (ins QPR:$src1, QPR:$src2), IIC_VBINiQ,
2109 "vbic", "$dst, $src1, $src2", "",
2110 [(set QPR:$dst, (v4i32 (and QPR:$src1,
2111 (vnot_conv QPR:$src2))))]>;
2113 // VORN : Vector Bitwise OR NOT
2114 def VORNd : N3VX<0, 0, 0b11, 0b0001, 0, 1, (outs DPR:$dst),
2115 (ins DPR:$src1, DPR:$src2), IIC_VBINiD,
2116 "vorn", "$dst, $src1, $src2", "",
2117 [(set DPR:$dst, (v2i32 (or DPR:$src1,
2118 (vnot_conv DPR:$src2))))]>;
2119 def VORNq : N3VX<0, 0, 0b11, 0b0001, 1, 1, (outs QPR:$dst),
2120 (ins QPR:$src1, QPR:$src2), IIC_VBINiQ,
2121 "vorn", "$dst, $src1, $src2", "",
2122 [(set QPR:$dst, (v4i32 (or QPR:$src1,
2123 (vnot_conv QPR:$src2))))]>;
2125 // VMVN : Vector Bitwise NOT
2126 def VMVNd : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 0, 0,
2127 (outs DPR:$dst), (ins DPR:$src), IIC_VSHLiD,
2128 "vmvn", "$dst, $src", "",
2129 [(set DPR:$dst, (v2i32 (vnot DPR:$src)))]>;
2130 def VMVNq : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 1, 0,
2131 (outs QPR:$dst), (ins QPR:$src), IIC_VSHLiD,
2132 "vmvn", "$dst, $src", "",
2133 [(set QPR:$dst, (v4i32 (vnot QPR:$src)))]>;
2134 def : Pat<(v2i32 (vnot_conv DPR:$src)), (VMVNd DPR:$src)>;
2135 def : Pat<(v4i32 (vnot_conv QPR:$src)), (VMVNq QPR:$src)>;
2137 // VBSL : Vector Bitwise Select
2138 def VBSLd : N3VX<1, 0, 0b01, 0b0001, 0, 1, (outs DPR:$dst),
2139 (ins DPR:$src1, DPR:$src2, DPR:$src3), IIC_VCNTiD,
2140 "vbsl", "$dst, $src2, $src3", "$src1 = $dst",
2142 (v2i32 (or (and DPR:$src2, DPR:$src1),
2143 (and DPR:$src3, (vnot_conv DPR:$src1)))))]>;
2144 def VBSLq : N3VX<1, 0, 0b01, 0b0001, 1, 1, (outs QPR:$dst),
2145 (ins QPR:$src1, QPR:$src2, QPR:$src3), IIC_VCNTiQ,
2146 "vbsl", "$dst, $src2, $src3", "$src1 = $dst",
2148 (v4i32 (or (and QPR:$src2, QPR:$src1),
2149 (and QPR:$src3, (vnot_conv QPR:$src1)))))]>;
2151 // VBIF : Vector Bitwise Insert if False
2152 // like VBSL but with: "vbif $dst, $src3, $src1", "$src2 = $dst",
2153 def VBIFd : N3VX<1, 0, 0b11, 0b0001, 0, 1,
2154 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3),
2155 IIC_VBINiD, "vbif", "$dst, $src2, $src3", "$src1 = $dst",
2156 [/* For disassembly only; pattern left blank */]>;
2157 def VBIFq : N3VX<1, 0, 0b11, 0b0001, 1, 1,
2158 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3),
2159 IIC_VBINiQ, "vbif", "$dst, $src2, $src3", "$src1 = $dst",
2160 [/* For disassembly only; pattern left blank */]>;
2162 // VBIT : Vector Bitwise Insert if True
2163 // like VBSL but with: "vbit $dst, $src2, $src1", "$src3 = $dst",
2164 def VBITd : N3VX<1, 0, 0b10, 0b0001, 0, 1,
2165 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3),
2166 IIC_VBINiD, "vbit", "$dst, $src2, $src3", "$src1 = $dst",
2167 [/* For disassembly only; pattern left blank */]>;
2168 def VBITq : N3VX<1, 0, 0b10, 0b0001, 1, 1,
2169 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3),
2170 IIC_VBINiQ, "vbit", "$dst, $src2, $src3", "$src1 = $dst",
2171 [/* For disassembly only; pattern left blank */]>;
2173 // VBIT/VBIF are not yet implemented. The TwoAddress pass will not go looking
2174 // for equivalent operations with different register constraints; it just
2177 // Vector Absolute Differences.
2179 // VABD : Vector Absolute Difference
2180 defm VABDs : N3VInt_QHS<0, 0, 0b0111, 0, IIC_VBINi4D, IIC_VBINi4D,
2181 IIC_VBINi4Q, IIC_VBINi4Q,
2182 "vabd", "s", int_arm_neon_vabds, 0>;
2183 defm VABDu : N3VInt_QHS<1, 0, 0b0111, 0, IIC_VBINi4D, IIC_VBINi4D,
2184 IIC_VBINi4Q, IIC_VBINi4Q,
2185 "vabd", "u", int_arm_neon_vabdu, 0>;
2186 def VABDfd : N3VDInt<1, 0, 0b10, 0b1101, 0, IIC_VBIND,
2187 "vabd", "f32", v2f32, v2f32, int_arm_neon_vabds, 0>;
2188 def VABDfq : N3VQInt<1, 0, 0b10, 0b1101, 0, IIC_VBINQ,
2189 "vabd", "f32", v4f32, v4f32, int_arm_neon_vabds, 0>;
2191 // VABDL : Vector Absolute Difference Long (Q = | D - D |)
2192 defm VABDLs : N3VLInt_QHS<0,1,0b0111,0, IIC_VBINi4Q,
2193 "vabdl", "s", int_arm_neon_vabdls, 0>;
2194 defm VABDLu : N3VLInt_QHS<1,1,0b0111,0, IIC_VBINi4Q,
2195 "vabdl", "u", int_arm_neon_vabdlu, 0>;
2197 // VABA : Vector Absolute Difference and Accumulate
2198 defm VABAs : N3VInt3_QHS<0,0,0b0111,1, "vaba", "s", int_arm_neon_vabas>;
2199 defm VABAu : N3VInt3_QHS<1,0,0b0111,1, "vaba", "u", int_arm_neon_vabau>;
2201 // VABAL : Vector Absolute Difference and Accumulate Long (Q += | D - D |)
2202 defm VABALs : N3VLInt3_QHS<0,1,0b0101,0, "vabal", "s", int_arm_neon_vabals>;
2203 defm VABALu : N3VLInt3_QHS<1,1,0b0101,0, "vabal", "u", int_arm_neon_vabalu>;
2205 // Vector Maximum and Minimum.
2207 // VMAX : Vector Maximum
2208 defm VMAXs : N3VInt_QHS<0,0,0b0110,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2209 IIC_VBINi4Q, "vmax", "s", int_arm_neon_vmaxs, 1>;
2210 defm VMAXu : N3VInt_QHS<1,0,0b0110,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2211 IIC_VBINi4Q, "vmax", "u", int_arm_neon_vmaxu, 1>;
2212 def VMAXfd : N3VDInt<0, 0, 0b00, 0b1111, 0, IIC_VBIND, "vmax", "f32",
2213 v2f32, v2f32, int_arm_neon_vmaxs, 1>;
2214 def VMAXfq : N3VQInt<0, 0, 0b00, 0b1111, 0, IIC_VBINQ, "vmax", "f32",
2215 v4f32, v4f32, int_arm_neon_vmaxs, 1>;
2217 // VMIN : Vector Minimum
2218 defm VMINs : N3VInt_QHS<0,0,0b0110,1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2219 IIC_VBINi4Q, "vmin", "s", int_arm_neon_vmins, 1>;
2220 defm VMINu : N3VInt_QHS<1,0,0b0110,1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2221 IIC_VBINi4Q, "vmin", "u", int_arm_neon_vminu, 1>;
2222 def VMINfd : N3VDInt<0, 0, 0b10, 0b1111, 0, IIC_VBIND, "vmin", "f32",
2223 v2f32, v2f32, int_arm_neon_vmins, 1>;
2224 def VMINfq : N3VQInt<0, 0, 0b10, 0b1111, 0, IIC_VBINQ, "vmin", "f32",
2225 v4f32, v4f32, int_arm_neon_vmins, 1>;
2227 // Vector Pairwise Operations.
2229 // VPADD : Vector Pairwise Add
2230 def VPADDi8 : N3VDInt<0, 0, 0b00, 0b1011, 1, IIC_VBINiD, "vpadd", "i8",
2231 v8i8, v8i8, int_arm_neon_vpadd, 0>;
2232 def VPADDi16 : N3VDInt<0, 0, 0b01, 0b1011, 1, IIC_VBINiD, "vpadd", "i16",
2233 v4i16, v4i16, int_arm_neon_vpadd, 0>;
2234 def VPADDi32 : N3VDInt<0, 0, 0b10, 0b1011, 1, IIC_VBINiD, "vpadd", "i32",
2235 v2i32, v2i32, int_arm_neon_vpadd, 0>;
2236 def VPADDf : N3VDInt<1, 0, 0b00, 0b1101, 0, IIC_VBIND, "vpadd", "f32",
2237 v2f32, v2f32, int_arm_neon_vpadd, 0>;
2239 // VPADDL : Vector Pairwise Add Long
2240 defm VPADDLs : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00100, 0, "vpaddl", "s",
2241 int_arm_neon_vpaddls>;
2242 defm VPADDLu : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00101, 0, "vpaddl", "u",
2243 int_arm_neon_vpaddlu>;
2245 // VPADAL : Vector Pairwise Add and Accumulate Long
2246 defm VPADALs : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01100, 0, "vpadal", "s",
2247 int_arm_neon_vpadals>;
2248 defm VPADALu : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01101, 0, "vpadal", "u",
2249 int_arm_neon_vpadalu>;
2251 // VPMAX : Vector Pairwise Maximum
2252 def VPMAXs8 : N3VDInt<0, 0, 0b00, 0b1010, 0, IIC_VBINi4D, "vpmax", "s8",
2253 v8i8, v8i8, int_arm_neon_vpmaxs, 0>;
2254 def VPMAXs16 : N3VDInt<0, 0, 0b01, 0b1010, 0, IIC_VBINi4D, "vpmax", "s16",
2255 v4i16, v4i16, int_arm_neon_vpmaxs, 0>;
2256 def VPMAXs32 : N3VDInt<0, 0, 0b10, 0b1010, 0, IIC_VBINi4D, "vpmax", "s32",
2257 v2i32, v2i32, int_arm_neon_vpmaxs, 0>;
2258 def VPMAXu8 : N3VDInt<1, 0, 0b00, 0b1010, 0, IIC_VBINi4D, "vpmax", "u8",
2259 v8i8, v8i8, int_arm_neon_vpmaxu, 0>;
2260 def VPMAXu16 : N3VDInt<1, 0, 0b01, 0b1010, 0, IIC_VBINi4D, "vpmax", "u16",
2261 v4i16, v4i16, int_arm_neon_vpmaxu, 0>;
2262 def VPMAXu32 : N3VDInt<1, 0, 0b10, 0b1010, 0, IIC_VBINi4D, "vpmax", "u32",
2263 v2i32, v2i32, int_arm_neon_vpmaxu, 0>;
2264 def VPMAXf : N3VDInt<1, 0, 0b00, 0b1111, 0, IIC_VBINi4D, "vpmax", "f32",
2265 v2f32, v2f32, int_arm_neon_vpmaxs, 0>;
2267 // VPMIN : Vector Pairwise Minimum
2268 def VPMINs8 : N3VDInt<0, 0, 0b00, 0b1010, 1, IIC_VBINi4D, "vpmin", "s8",
2269 v8i8, v8i8, int_arm_neon_vpmins, 0>;
2270 def VPMINs16 : N3VDInt<0, 0, 0b01, 0b1010, 1, IIC_VBINi4D, "vpmin", "s16",
2271 v4i16, v4i16, int_arm_neon_vpmins, 0>;
2272 def VPMINs32 : N3VDInt<0, 0, 0b10, 0b1010, 1, IIC_VBINi4D, "vpmin", "s32",
2273 v2i32, v2i32, int_arm_neon_vpmins, 0>;
2274 def VPMINu8 : N3VDInt<1, 0, 0b00, 0b1010, 1, IIC_VBINi4D, "vpmin", "u8",
2275 v8i8, v8i8, int_arm_neon_vpminu, 0>;
2276 def VPMINu16 : N3VDInt<1, 0, 0b01, 0b1010, 1, IIC_VBINi4D, "vpmin", "u16",
2277 v4i16, v4i16, int_arm_neon_vpminu, 0>;
2278 def VPMINu32 : N3VDInt<1, 0, 0b10, 0b1010, 1, IIC_VBINi4D, "vpmin", "u32",
2279 v2i32, v2i32, int_arm_neon_vpminu, 0>;
2280 def VPMINf : N3VDInt<1, 0, 0b10, 0b1111, 0, IIC_VBINi4D, "vpmin", "f32",
2281 v2f32, v2f32, int_arm_neon_vpmins, 0>;
2283 // Vector Reciprocal and Reciprocal Square Root Estimate and Step.
2285 // VRECPE : Vector Reciprocal Estimate
2286 def VRECPEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
2287 IIC_VUNAD, "vrecpe", "u32",
2288 v2i32, v2i32, int_arm_neon_vrecpe>;
2289 def VRECPEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
2290 IIC_VUNAQ, "vrecpe", "u32",
2291 v4i32, v4i32, int_arm_neon_vrecpe>;
2292 def VRECPEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
2293 IIC_VUNAD, "vrecpe", "f32",
2294 v2f32, v2f32, int_arm_neon_vrecpe>;
2295 def VRECPEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
2296 IIC_VUNAQ, "vrecpe", "f32",
2297 v4f32, v4f32, int_arm_neon_vrecpe>;
2299 // VRECPS : Vector Reciprocal Step
2300 def VRECPSfd : N3VDInt<0, 0, 0b00, 0b1111, 1,
2301 IIC_VRECSD, "vrecps", "f32",
2302 v2f32, v2f32, int_arm_neon_vrecps, 1>;
2303 def VRECPSfq : N3VQInt<0, 0, 0b00, 0b1111, 1,
2304 IIC_VRECSQ, "vrecps", "f32",
2305 v4f32, v4f32, int_arm_neon_vrecps, 1>;
2307 // VRSQRTE : Vector Reciprocal Square Root Estimate
2308 def VRSQRTEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
2309 IIC_VUNAD, "vrsqrte", "u32",
2310 v2i32, v2i32, int_arm_neon_vrsqrte>;
2311 def VRSQRTEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
2312 IIC_VUNAQ, "vrsqrte", "u32",
2313 v4i32, v4i32, int_arm_neon_vrsqrte>;
2314 def VRSQRTEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
2315 IIC_VUNAD, "vrsqrte", "f32",
2316 v2f32, v2f32, int_arm_neon_vrsqrte>;
2317 def VRSQRTEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
2318 IIC_VUNAQ, "vrsqrte", "f32",
2319 v4f32, v4f32, int_arm_neon_vrsqrte>;
2321 // VRSQRTS : Vector Reciprocal Square Root Step
2322 def VRSQRTSfd : N3VDInt<0, 0, 0b10, 0b1111, 1,
2323 IIC_VRECSD, "vrsqrts", "f32",
2324 v2f32, v2f32, int_arm_neon_vrsqrts, 1>;
2325 def VRSQRTSfq : N3VQInt<0, 0, 0b10, 0b1111, 1,
2326 IIC_VRECSQ, "vrsqrts", "f32",
2327 v4f32, v4f32, int_arm_neon_vrsqrts, 1>;
2331 // VSHL : Vector Shift
2332 defm VSHLs : N3VInt_QHSD<0, 0, 0b0100, 0, IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ,
2333 IIC_VSHLiQ, "vshl", "s", int_arm_neon_vshifts, 0>;
2334 defm VSHLu : N3VInt_QHSD<1, 0, 0b0100, 0, IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ,
2335 IIC_VSHLiQ, "vshl", "u", int_arm_neon_vshiftu, 0>;
2336 // VSHL : Vector Shift Left (Immediate)
2337 defm VSHLi : N2VSh_QHSD<0, 1, 0b0101, 1, IIC_VSHLiD, "vshl", "i", NEONvshl>;
2338 // VSHR : Vector Shift Right (Immediate)
2339 defm VSHRs : N2VSh_QHSD<0, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "s", NEONvshrs>;
2340 defm VSHRu : N2VSh_QHSD<1, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "u", NEONvshru>;
2342 // VSHLL : Vector Shift Left Long
2343 defm VSHLLs : N2VLSh_QHS<0, 1, 0b1010, 0, 0, 1, "vshll", "s", NEONvshlls>;
2344 defm VSHLLu : N2VLSh_QHS<1, 1, 0b1010, 0, 0, 1, "vshll", "u", NEONvshllu>;
2346 // VSHLL : Vector Shift Left Long (with maximum shift count)
2347 class N2VLShMax<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
2348 bit op6, bit op4, string OpcodeStr, string Dt, ValueType ResTy,
2349 ValueType OpTy, SDNode OpNode>
2350 : N2VLSh<op24, op23, op11_8, op7, op6, op4, OpcodeStr, Dt,
2351 ResTy, OpTy, OpNode> {
2352 let Inst{21-16} = op21_16;
2354 def VSHLLi8 : N2VLShMax<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll", "i8",
2355 v8i16, v8i8, NEONvshlli>;
2356 def VSHLLi16 : N2VLShMax<1, 1, 0b110110, 0b0011, 0, 0, 0, "vshll", "i16",
2357 v4i32, v4i16, NEONvshlli>;
2358 def VSHLLi32 : N2VLShMax<1, 1, 0b111010, 0b0011, 0, 0, 0, "vshll", "i32",
2359 v2i64, v2i32, NEONvshlli>;
2361 // VSHRN : Vector Shift Right and Narrow
2362 defm VSHRN : N2VNSh_HSD<0,1,0b1000,0,0,1, IIC_VSHLiD, "vshrn", "i",
2365 // VRSHL : Vector Rounding Shift
2366 defm VRSHLs : N3VInt_QHSD<0,0,0b0101,0, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2367 IIC_VSHLi4Q, "vrshl", "s", int_arm_neon_vrshifts,0>;
2368 defm VRSHLu : N3VInt_QHSD<1,0,0b0101,0, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2369 IIC_VSHLi4Q, "vrshl", "u", int_arm_neon_vrshiftu,0>;
2370 // VRSHR : Vector Rounding Shift Right
2371 defm VRSHRs : N2VSh_QHSD<0,1,0b0010,1, IIC_VSHLi4D, "vrshr", "s", NEONvrshrs>;
2372 defm VRSHRu : N2VSh_QHSD<1,1,0b0010,1, IIC_VSHLi4D, "vrshr", "u", NEONvrshru>;
2374 // VRSHRN : Vector Rounding Shift Right and Narrow
2375 defm VRSHRN : N2VNSh_HSD<0, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vrshrn", "i",
2378 // VQSHL : Vector Saturating Shift
2379 defm VQSHLs : N3VInt_QHSD<0,0,0b0100,1, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2380 IIC_VSHLi4Q, "vqshl", "s", int_arm_neon_vqshifts,0>;
2381 defm VQSHLu : N3VInt_QHSD<1,0,0b0100,1, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2382 IIC_VSHLi4Q, "vqshl", "u", int_arm_neon_vqshiftu,0>;
2383 // VQSHL : Vector Saturating Shift Left (Immediate)
2384 defm VQSHLsi : N2VSh_QHSD<0,1,0b0111,1, IIC_VSHLi4D, "vqshl", "s", NEONvqshls>;
2385 defm VQSHLui : N2VSh_QHSD<1,1,0b0111,1, IIC_VSHLi4D, "vqshl", "u", NEONvqshlu>;
2386 // VQSHLU : Vector Saturating Shift Left (Immediate, Unsigned)
2387 defm VQSHLsu : N2VSh_QHSD<1,1,0b0110,1, IIC_VSHLi4D, "vqshlu","s",NEONvqshlsu>;
2389 // VQSHRN : Vector Saturating Shift Right and Narrow
2390 defm VQSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "s",
2392 defm VQSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "u",
2395 // VQSHRUN : Vector Saturating Shift Right and Narrow (Unsigned)
2396 defm VQSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 0, 1, IIC_VSHLi4D, "vqshrun", "s",
2399 // VQRSHL : Vector Saturating Rounding Shift
2400 defm VQRSHLs : N3VInt_QHSD<0,0,0b0101,1, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2401 IIC_VSHLi4Q, "vqrshl", "s",
2402 int_arm_neon_vqrshifts, 0>;
2403 defm VQRSHLu : N3VInt_QHSD<1,0,0b0101,1, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2404 IIC_VSHLi4Q, "vqrshl", "u",
2405 int_arm_neon_vqrshiftu, 0>;
2407 // VQRSHRN : Vector Saturating Rounding Shift Right and Narrow
2408 defm VQRSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "s",
2410 defm VQRSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "u",
2413 // VQRSHRUN : Vector Saturating Rounding Shift Right and Narrow (Unsigned)
2414 defm VQRSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vqrshrun", "s",
2417 // VSRA : Vector Shift Right and Accumulate
2418 defm VSRAs : N2VShAdd_QHSD<0, 1, 0b0001, 1, "vsra", "s", NEONvshrs>;
2419 defm VSRAu : N2VShAdd_QHSD<1, 1, 0b0001, 1, "vsra", "u", NEONvshru>;
2420 // VRSRA : Vector Rounding Shift Right and Accumulate
2421 defm VRSRAs : N2VShAdd_QHSD<0, 1, 0b0011, 1, "vrsra", "s", NEONvrshrs>;
2422 defm VRSRAu : N2VShAdd_QHSD<1, 1, 0b0011, 1, "vrsra", "u", NEONvrshru>;
2424 // VSLI : Vector Shift Left and Insert
2425 defm VSLI : N2VShIns_QHSD<1, 1, 0b0101, 1, "vsli", NEONvsli>;
2426 // VSRI : Vector Shift Right and Insert
2427 defm VSRI : N2VShIns_QHSD<1, 1, 0b0100, 1, "vsri", NEONvsri>;
2429 // Vector Absolute and Saturating Absolute.
2431 // VABS : Vector Absolute Value
2432 defm VABS : N2VInt_QHS<0b11, 0b11, 0b01, 0b00110, 0,
2433 IIC_VUNAiD, IIC_VUNAiQ, "vabs", "s",
2435 def VABSfd : N2VDInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
2436 IIC_VUNAD, "vabs", "f32",
2437 v2f32, v2f32, int_arm_neon_vabs>;
2438 def VABSfq : N2VQInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
2439 IIC_VUNAQ, "vabs", "f32",
2440 v4f32, v4f32, int_arm_neon_vabs>;
2442 // VQABS : Vector Saturating Absolute Value
2443 defm VQABS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01110, 0,
2444 IIC_VQUNAiD, IIC_VQUNAiQ, "vqabs", "s",
2445 int_arm_neon_vqabs>;
2449 def vneg : PatFrag<(ops node:$in), (sub immAllZerosV, node:$in)>;
2450 def vneg_conv : PatFrag<(ops node:$in), (sub immAllZerosV_bc, node:$in)>;
2452 class VNEGD<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
2453 : N2V<0b11, 0b11, size, 0b01, 0b00111, 0, 0, (outs DPR:$dst), (ins DPR:$src),
2454 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
2455 [(set DPR:$dst, (Ty (vneg DPR:$src)))]>;
2456 class VNEGQ<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
2457 : N2V<0b11, 0b11, size, 0b01, 0b00111, 1, 0, (outs QPR:$dst), (ins QPR:$src),
2458 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
2459 [(set QPR:$dst, (Ty (vneg QPR:$src)))]>;
2461 // VNEG : Vector Negate
2462 def VNEGs8d : VNEGD<0b00, "vneg", "s8", v8i8>;
2463 def VNEGs16d : VNEGD<0b01, "vneg", "s16", v4i16>;
2464 def VNEGs32d : VNEGD<0b10, "vneg", "s32", v2i32>;
2465 def VNEGs8q : VNEGQ<0b00, "vneg", "s8", v16i8>;
2466 def VNEGs16q : VNEGQ<0b01, "vneg", "s16", v8i16>;
2467 def VNEGs32q : VNEGQ<0b10, "vneg", "s32", v4i32>;
2469 // VNEG : Vector Negate (floating-point)
2470 def VNEGfd : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
2471 (outs DPR:$dst), (ins DPR:$src), IIC_VUNAD,
2472 "vneg", "f32", "$dst, $src", "",
2473 [(set DPR:$dst, (v2f32 (fneg DPR:$src)))]>;
2474 def VNEGf32q : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 1, 0,
2475 (outs QPR:$dst), (ins QPR:$src), IIC_VUNAQ,
2476 "vneg", "f32", "$dst, $src", "",
2477 [(set QPR:$dst, (v4f32 (fneg QPR:$src)))]>;
2479 def : Pat<(v8i8 (vneg_conv DPR:$src)), (VNEGs8d DPR:$src)>;
2480 def : Pat<(v4i16 (vneg_conv DPR:$src)), (VNEGs16d DPR:$src)>;
2481 def : Pat<(v2i32 (vneg_conv DPR:$src)), (VNEGs32d DPR:$src)>;
2482 def : Pat<(v16i8 (vneg_conv QPR:$src)), (VNEGs8q QPR:$src)>;
2483 def : Pat<(v8i16 (vneg_conv QPR:$src)), (VNEGs16q QPR:$src)>;
2484 def : Pat<(v4i32 (vneg_conv QPR:$src)), (VNEGs32q QPR:$src)>;
2486 // VQNEG : Vector Saturating Negate
2487 defm VQNEG : N2VInt_QHS<0b11, 0b11, 0b00, 0b01111, 0,
2488 IIC_VQUNAiD, IIC_VQUNAiQ, "vqneg", "s",
2489 int_arm_neon_vqneg>;
2491 // Vector Bit Counting Operations.
2493 // VCLS : Vector Count Leading Sign Bits
2494 defm VCLS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01000, 0,
2495 IIC_VCNTiD, IIC_VCNTiQ, "vcls", "s",
2497 // VCLZ : Vector Count Leading Zeros
2498 defm VCLZ : N2VInt_QHS<0b11, 0b11, 0b00, 0b01001, 0,
2499 IIC_VCNTiD, IIC_VCNTiQ, "vclz", "i",
2501 // VCNT : Vector Count One Bits
2502 def VCNTd : N2VDInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
2503 IIC_VCNTiD, "vcnt", "8",
2504 v8i8, v8i8, int_arm_neon_vcnt>;
2505 def VCNTq : N2VQInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
2506 IIC_VCNTiQ, "vcnt", "8",
2507 v16i8, v16i8, int_arm_neon_vcnt>;
2509 // Vector Swap -- for disassembly only.
2510 def VSWPd : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 0, 0,
2511 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
2512 "vswp", "$dst, $src", "", []>;
2513 def VSWPq : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 1, 0,
2514 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
2515 "vswp", "$dst, $src", "", []>;
2517 // Vector Move Operations.
2519 // VMOV : Vector Move (Register)
2521 def VMOVDneon: N3VX<0, 0, 0b10, 0b0001, 0, 1, (outs DPR:$dst), (ins DPR:$src),
2522 IIC_VMOVD, "vmov", "$dst, $src", "", []>;
2523 def VMOVQ : N3VX<0, 0, 0b10, 0b0001, 1, 1, (outs QPR:$dst), (ins QPR:$src),
2524 IIC_VMOVD, "vmov", "$dst, $src", "", []>;
2526 // VMOV : Vector Move (Immediate)
2528 // VMOV_get_imm8 xform function: convert build_vector to VMOV.i8 imm.
2529 def VMOV_get_imm8 : SDNodeXForm<build_vector, [{
2530 return ARM::getVMOVImm(N, 1, *CurDAG);
2532 def vmovImm8 : PatLeaf<(build_vector), [{
2533 return ARM::getVMOVImm(N, 1, *CurDAG).getNode() != 0;
2536 // VMOV_get_imm16 xform function: convert build_vector to VMOV.i16 imm.
2537 def VMOV_get_imm16 : SDNodeXForm<build_vector, [{
2538 return ARM::getVMOVImm(N, 2, *CurDAG);
2540 def vmovImm16 : PatLeaf<(build_vector), [{
2541 return ARM::getVMOVImm(N, 2, *CurDAG).getNode() != 0;
2542 }], VMOV_get_imm16>;
2544 // VMOV_get_imm32 xform function: convert build_vector to VMOV.i32 imm.
2545 def VMOV_get_imm32 : SDNodeXForm<build_vector, [{
2546 return ARM::getVMOVImm(N, 4, *CurDAG);
2548 def vmovImm32 : PatLeaf<(build_vector), [{
2549 return ARM::getVMOVImm(N, 4, *CurDAG).getNode() != 0;
2550 }], VMOV_get_imm32>;
2552 // VMOV_get_imm64 xform function: convert build_vector to VMOV.i64 imm.
2553 def VMOV_get_imm64 : SDNodeXForm<build_vector, [{
2554 return ARM::getVMOVImm(N, 8, *CurDAG);
2556 def vmovImm64 : PatLeaf<(build_vector), [{
2557 return ARM::getVMOVImm(N, 8, *CurDAG).getNode() != 0;
2558 }], VMOV_get_imm64>;
2560 // Note: Some of the cmode bits in the following VMOV instructions need to
2561 // be encoded based on the immed values.
2563 def VMOVv8i8 : N1ModImm<1, 0b000, 0b1110, 0, 0, 0, 1, (outs DPR:$dst),
2564 (ins h8imm:$SIMM), IIC_VMOVImm,
2565 "vmov", "i8", "$dst, $SIMM", "",
2566 [(set DPR:$dst, (v8i8 vmovImm8:$SIMM))]>;
2567 def VMOVv16i8 : N1ModImm<1, 0b000, 0b1110, 0, 1, 0, 1, (outs QPR:$dst),
2568 (ins h8imm:$SIMM), IIC_VMOVImm,
2569 "vmov", "i8", "$dst, $SIMM", "",
2570 [(set QPR:$dst, (v16i8 vmovImm8:$SIMM))]>;
2572 def VMOVv4i16 : N1ModImm<1, 0b000, {1,0,?,?}, 0, 0, {?}, 1, (outs DPR:$dst),
2573 (ins h16imm:$SIMM), IIC_VMOVImm,
2574 "vmov", "i16", "$dst, $SIMM", "",
2575 [(set DPR:$dst, (v4i16 vmovImm16:$SIMM))]>;
2576 def VMOVv8i16 : N1ModImm<1, 0b000, {1,0,?,?}, 0, 1, {?}, 1, (outs QPR:$dst),
2577 (ins h16imm:$SIMM), IIC_VMOVImm,
2578 "vmov", "i16", "$dst, $SIMM", "",
2579 [(set QPR:$dst, (v8i16 vmovImm16:$SIMM))]>;
2581 def VMOVv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, {?}, 1, (outs DPR:$dst),
2582 (ins h32imm:$SIMM), IIC_VMOVImm,
2583 "vmov", "i32", "$dst, $SIMM", "",
2584 [(set DPR:$dst, (v2i32 vmovImm32:$SIMM))]>;
2585 def VMOVv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, {?}, 1, (outs QPR:$dst),
2586 (ins h32imm:$SIMM), IIC_VMOVImm,
2587 "vmov", "i32", "$dst, $SIMM", "",
2588 [(set QPR:$dst, (v4i32 vmovImm32:$SIMM))]>;
2590 def VMOVv1i64 : N1ModImm<1, 0b000, 0b1110, 0, 0, 1, 1, (outs DPR:$dst),
2591 (ins h64imm:$SIMM), IIC_VMOVImm,
2592 "vmov", "i64", "$dst, $SIMM", "",
2593 [(set DPR:$dst, (v1i64 vmovImm64:$SIMM))]>;
2594 def VMOVv2i64 : N1ModImm<1, 0b000, 0b1110, 0, 1, 1, 1, (outs QPR:$dst),
2595 (ins h64imm:$SIMM), IIC_VMOVImm,
2596 "vmov", "i64", "$dst, $SIMM", "",
2597 [(set QPR:$dst, (v2i64 vmovImm64:$SIMM))]>;
2599 // VMOV : Vector Get Lane (move scalar to ARM core register)
2601 def VGETLNs8 : NVGetLane<{1,1,1,0,0,1,?,1}, 0b1011, {?,?},
2602 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2603 IIC_VMOVSI, "vmov", "s8", "$dst, $src[$lane]",
2604 [(set GPR:$dst, (NEONvgetlanes (v8i8 DPR:$src),
2606 def VGETLNs16 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, {?,1},
2607 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2608 IIC_VMOVSI, "vmov", "s16", "$dst, $src[$lane]",
2609 [(set GPR:$dst, (NEONvgetlanes (v4i16 DPR:$src),
2611 def VGETLNu8 : NVGetLane<{1,1,1,0,1,1,?,1}, 0b1011, {?,?},
2612 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2613 IIC_VMOVSI, "vmov", "u8", "$dst, $src[$lane]",
2614 [(set GPR:$dst, (NEONvgetlaneu (v8i8 DPR:$src),
2616 def VGETLNu16 : NVGetLane<{1,1,1,0,1,0,?,1}, 0b1011, {?,1},
2617 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2618 IIC_VMOVSI, "vmov", "u16", "$dst, $src[$lane]",
2619 [(set GPR:$dst, (NEONvgetlaneu (v4i16 DPR:$src),
2621 def VGETLNi32 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, 0b00,
2622 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2623 IIC_VMOVSI, "vmov", "32", "$dst, $src[$lane]",
2624 [(set GPR:$dst, (extractelt (v2i32 DPR:$src),
2626 // def VGETLNf32: see FMRDH and FMRDL in ARMInstrVFP.td
2627 def : Pat<(NEONvgetlanes (v16i8 QPR:$src), imm:$lane),
2628 (VGETLNs8 (v8i8 (EXTRACT_SUBREG QPR:$src,
2629 (DSubReg_i8_reg imm:$lane))),
2630 (SubReg_i8_lane imm:$lane))>;
2631 def : Pat<(NEONvgetlanes (v8i16 QPR:$src), imm:$lane),
2632 (VGETLNs16 (v4i16 (EXTRACT_SUBREG QPR:$src,
2633 (DSubReg_i16_reg imm:$lane))),
2634 (SubReg_i16_lane imm:$lane))>;
2635 def : Pat<(NEONvgetlaneu (v16i8 QPR:$src), imm:$lane),
2636 (VGETLNu8 (v8i8 (EXTRACT_SUBREG QPR:$src,
2637 (DSubReg_i8_reg imm:$lane))),
2638 (SubReg_i8_lane imm:$lane))>;
2639 def : Pat<(NEONvgetlaneu (v8i16 QPR:$src), imm:$lane),
2640 (VGETLNu16 (v4i16 (EXTRACT_SUBREG QPR:$src,
2641 (DSubReg_i16_reg imm:$lane))),
2642 (SubReg_i16_lane imm:$lane))>;
2643 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
2644 (VGETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src,
2645 (DSubReg_i32_reg imm:$lane))),
2646 (SubReg_i32_lane imm:$lane))>;
2647 def : Pat<(extractelt (v2f32 DPR:$src1), imm:$src2),
2648 (EXTRACT_SUBREG (v2f32 (COPY_TO_REGCLASS (v2f32 DPR:$src1),DPR_VFP2)),
2649 (SSubReg_f32_reg imm:$src2))>;
2650 def : Pat<(extractelt (v4f32 QPR:$src1), imm:$src2),
2651 (EXTRACT_SUBREG (v4f32 (COPY_TO_REGCLASS (v4f32 QPR:$src1),QPR_VFP2)),
2652 (SSubReg_f32_reg imm:$src2))>;
2653 //def : Pat<(extractelt (v2i64 QPR:$src1), imm:$src2),
2654 // (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
2655 def : Pat<(extractelt (v2f64 QPR:$src1), imm:$src2),
2656 (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
2659 // VMOV : Vector Set Lane (move ARM core register to scalar)
2661 let Constraints = "$src1 = $dst" in {
2662 def VSETLNi8 : NVSetLane<{1,1,1,0,0,1,?,0}, 0b1011, {?,?}, (outs DPR:$dst),
2663 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
2664 IIC_VMOVISL, "vmov", "8", "$dst[$lane], $src2",
2665 [(set DPR:$dst, (vector_insert (v8i8 DPR:$src1),
2666 GPR:$src2, imm:$lane))]>;
2667 def VSETLNi16 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, {?,1}, (outs DPR:$dst),
2668 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
2669 IIC_VMOVISL, "vmov", "16", "$dst[$lane], $src2",
2670 [(set DPR:$dst, (vector_insert (v4i16 DPR:$src1),
2671 GPR:$src2, imm:$lane))]>;
2672 def VSETLNi32 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, 0b00, (outs DPR:$dst),
2673 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
2674 IIC_VMOVISL, "vmov", "32", "$dst[$lane], $src2",
2675 [(set DPR:$dst, (insertelt (v2i32 DPR:$src1),
2676 GPR:$src2, imm:$lane))]>;
2678 def : Pat<(vector_insert (v16i8 QPR:$src1), GPR:$src2, imm:$lane),
2679 (v16i8 (INSERT_SUBREG QPR:$src1,
2680 (v8i8 (VSETLNi8 (v8i8 (EXTRACT_SUBREG QPR:$src1,
2681 (DSubReg_i8_reg imm:$lane))),
2682 GPR:$src2, (SubReg_i8_lane imm:$lane))),
2683 (DSubReg_i8_reg imm:$lane)))>;
2684 def : Pat<(vector_insert (v8i16 QPR:$src1), GPR:$src2, imm:$lane),
2685 (v8i16 (INSERT_SUBREG QPR:$src1,
2686 (v4i16 (VSETLNi16 (v4i16 (EXTRACT_SUBREG QPR:$src1,
2687 (DSubReg_i16_reg imm:$lane))),
2688 GPR:$src2, (SubReg_i16_lane imm:$lane))),
2689 (DSubReg_i16_reg imm:$lane)))>;
2690 def : Pat<(insertelt (v4i32 QPR:$src1), GPR:$src2, imm:$lane),
2691 (v4i32 (INSERT_SUBREG QPR:$src1,
2692 (v2i32 (VSETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src1,
2693 (DSubReg_i32_reg imm:$lane))),
2694 GPR:$src2, (SubReg_i32_lane imm:$lane))),
2695 (DSubReg_i32_reg imm:$lane)))>;
2697 def : Pat<(v2f32 (insertelt DPR:$src1, SPR:$src2, imm:$src3)),
2698 (INSERT_SUBREG (v2f32 (COPY_TO_REGCLASS DPR:$src1, DPR_VFP2)),
2699 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
2700 def : Pat<(v4f32 (insertelt QPR:$src1, SPR:$src2, imm:$src3)),
2701 (INSERT_SUBREG (v4f32 (COPY_TO_REGCLASS QPR:$src1, QPR_VFP2)),
2702 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
2704 //def : Pat<(v2i64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
2705 // (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
2706 def : Pat<(v2f64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
2707 (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
2709 def : Pat<(v2f32 (scalar_to_vector SPR:$src)),
2710 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$src, arm_ssubreg_0)>;
2711 def : Pat<(v2f64 (scalar_to_vector (f64 DPR:$src))),
2712 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), DPR:$src, arm_dsubreg_0)>;
2713 def : Pat<(v4f32 (scalar_to_vector SPR:$src)),
2714 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), SPR:$src, arm_ssubreg_0)>;
2716 def : Pat<(v8i8 (scalar_to_vector GPR:$src)),
2717 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
2718 def : Pat<(v4i16 (scalar_to_vector GPR:$src)),
2719 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
2720 def : Pat<(v2i32 (scalar_to_vector GPR:$src)),
2721 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
2723 def : Pat<(v16i8 (scalar_to_vector GPR:$src)),
2724 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
2725 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
2727 def : Pat<(v8i16 (scalar_to_vector GPR:$src)),
2728 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
2729 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
2731 def : Pat<(v4i32 (scalar_to_vector GPR:$src)),
2732 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
2733 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
2736 // VDUP : Vector Duplicate (from ARM core register to all elements)
2738 class VDUPD<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
2739 : NVDup<opcod1, 0b1011, opcod3, (outs DPR:$dst), (ins GPR:$src),
2740 IIC_VMOVIS, "vdup", Dt, "$dst, $src",
2741 [(set DPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
2742 class VDUPQ<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
2743 : NVDup<opcod1, 0b1011, opcod3, (outs QPR:$dst), (ins GPR:$src),
2744 IIC_VMOVIS, "vdup", Dt, "$dst, $src",
2745 [(set QPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
2747 def VDUP8d : VDUPD<0b11101100, 0b00, "8", v8i8>;
2748 def VDUP16d : VDUPD<0b11101000, 0b01, "16", v4i16>;
2749 def VDUP32d : VDUPD<0b11101000, 0b00, "32", v2i32>;
2750 def VDUP8q : VDUPQ<0b11101110, 0b00, "8", v16i8>;
2751 def VDUP16q : VDUPQ<0b11101010, 0b01, "16", v8i16>;
2752 def VDUP32q : VDUPQ<0b11101010, 0b00, "32", v4i32>;
2754 def VDUPfd : NVDup<0b11101000, 0b1011, 0b00, (outs DPR:$dst), (ins GPR:$src),
2755 IIC_VMOVIS, "vdup", "32", "$dst, $src",
2756 [(set DPR:$dst, (v2f32 (NEONvdup
2757 (f32 (bitconvert GPR:$src)))))]>;
2758 def VDUPfq : NVDup<0b11101010, 0b1011, 0b00, (outs QPR:$dst), (ins GPR:$src),
2759 IIC_VMOVIS, "vdup", "32", "$dst, $src",
2760 [(set QPR:$dst, (v4f32 (NEONvdup
2761 (f32 (bitconvert GPR:$src)))))]>;
2763 // VDUP : Vector Duplicate Lane (from scalar to all elements)
2765 class VDUPLND<bits<2> op19_18, bits<2> op17_16,
2766 string OpcodeStr, string Dt, ValueType Ty>
2767 : N2V<0b11, 0b11, op19_18, op17_16, 0b11000, 0, 0,
2768 (outs DPR:$dst), (ins DPR:$src, nohash_imm:$lane), IIC_VMOVD,
2769 OpcodeStr, Dt, "$dst, $src[$lane]", "",
2770 [(set DPR:$dst, (Ty (NEONvduplane (Ty DPR:$src), imm:$lane)))]>;
2772 class VDUPLNQ<bits<2> op19_18, bits<2> op17_16, string OpcodeStr, string Dt,
2773 ValueType ResTy, ValueType OpTy>
2774 : N2V<0b11, 0b11, op19_18, op17_16, 0b11000, 1, 0,
2775 (outs QPR:$dst), (ins DPR:$src, nohash_imm:$lane), IIC_VMOVD,
2776 OpcodeStr, Dt, "$dst, $src[$lane]", "",
2777 [(set QPR:$dst, (ResTy (NEONvduplane (OpTy DPR:$src), imm:$lane)))]>;
2779 // Inst{19-16} is partially specified depending on the element size.
2781 def VDUPLN8d : VDUPLND<{?,?}, {?,1}, "vdup", "8", v8i8>;
2782 def VDUPLN16d : VDUPLND<{?,?}, {1,0}, "vdup", "16", v4i16>;
2783 def VDUPLN32d : VDUPLND<{?,1}, {0,0}, "vdup", "32", v2i32>;
2784 def VDUPLNfd : VDUPLND<{?,1}, {0,0}, "vdup", "32", v2f32>;
2785 def VDUPLN8q : VDUPLNQ<{?,?}, {?,1}, "vdup", "8", v16i8, v8i8>;
2786 def VDUPLN16q : VDUPLNQ<{?,?}, {1,0}, "vdup", "16", v8i16, v4i16>;
2787 def VDUPLN32q : VDUPLNQ<{?,1}, {0,0}, "vdup", "32", v4i32, v2i32>;
2788 def VDUPLNfq : VDUPLNQ<{?,1}, {0,0}, "vdup", "32", v4f32, v2f32>;
2790 def : Pat<(v16i8 (NEONvduplane (v16i8 QPR:$src), imm:$lane)),
2791 (v16i8 (VDUPLN8q (v8i8 (EXTRACT_SUBREG QPR:$src,
2792 (DSubReg_i8_reg imm:$lane))),
2793 (SubReg_i8_lane imm:$lane)))>;
2794 def : Pat<(v8i16 (NEONvduplane (v8i16 QPR:$src), imm:$lane)),
2795 (v8i16 (VDUPLN16q (v4i16 (EXTRACT_SUBREG QPR:$src,
2796 (DSubReg_i16_reg imm:$lane))),
2797 (SubReg_i16_lane imm:$lane)))>;
2798 def : Pat<(v4i32 (NEONvduplane (v4i32 QPR:$src), imm:$lane)),
2799 (v4i32 (VDUPLN32q (v2i32 (EXTRACT_SUBREG QPR:$src,
2800 (DSubReg_i32_reg imm:$lane))),
2801 (SubReg_i32_lane imm:$lane)))>;
2802 def : Pat<(v4f32 (NEONvduplane (v4f32 QPR:$src), imm:$lane)),
2803 (v4f32 (VDUPLNfq (v2f32 (EXTRACT_SUBREG QPR:$src,
2804 (DSubReg_i32_reg imm:$lane))),
2805 (SubReg_i32_lane imm:$lane)))>;
2807 def VDUPfdf : N2V<0b11, 0b11, {?,1}, {0,0}, 0b11000, 0, 0,
2808 (outs DPR:$dst), (ins SPR:$src),
2809 IIC_VMOVD, "vdup", "32", "$dst, ${src:lane}", "",
2810 [(set DPR:$dst, (v2f32 (NEONvdup (f32 SPR:$src))))]>;
2812 def VDUPfqf : N2V<0b11, 0b11, {?,1}, {0,0}, 0b11000, 1, 0,
2813 (outs QPR:$dst), (ins SPR:$src),
2814 IIC_VMOVD, "vdup", "32", "$dst, ${src:lane}", "",
2815 [(set QPR:$dst, (v4f32 (NEONvdup (f32 SPR:$src))))]>;
2817 def : Pat<(v2i64 (NEONvduplane (v2i64 QPR:$src), imm:$lane)),
2818 (INSERT_SUBREG QPR:$src,
2819 (i64 (EXTRACT_SUBREG QPR:$src,
2820 (DSubReg_f64_reg imm:$lane))),
2821 (DSubReg_f64_other_reg imm:$lane))>;
2822 def : Pat<(v2f64 (NEONvduplane (v2f64 QPR:$src), imm:$lane)),
2823 (INSERT_SUBREG QPR:$src,
2824 (f64 (EXTRACT_SUBREG QPR:$src,
2825 (DSubReg_f64_reg imm:$lane))),
2826 (DSubReg_f64_other_reg imm:$lane))>;
2828 // VMOVN : Vector Narrowing Move
2829 defm VMOVN : N2VNInt_HSD<0b11,0b11,0b10,0b00100,0,0, IIC_VMOVD,
2830 "vmovn", "i", int_arm_neon_vmovn>;
2831 // VQMOVN : Vector Saturating Narrowing Move
2832 defm VQMOVNs : N2VNInt_HSD<0b11,0b11,0b10,0b00101,0,0, IIC_VQUNAiD,
2833 "vqmovn", "s", int_arm_neon_vqmovns>;
2834 defm VQMOVNu : N2VNInt_HSD<0b11,0b11,0b10,0b00101,1,0, IIC_VQUNAiD,
2835 "vqmovn", "u", int_arm_neon_vqmovnu>;
2836 defm VQMOVNsu : N2VNInt_HSD<0b11,0b11,0b10,0b00100,1,0, IIC_VQUNAiD,
2837 "vqmovun", "s", int_arm_neon_vqmovnsu>;
2838 // VMOVL : Vector Lengthening Move
2839 defm VMOVLs : N2VLInt_QHS<0b01,0b10100,0,1, "vmovl", "s",
2840 int_arm_neon_vmovls>;
2841 defm VMOVLu : N2VLInt_QHS<0b11,0b10100,0,1, "vmovl", "u",
2842 int_arm_neon_vmovlu>;
2844 // Vector Conversions.
2846 // VCVT : Vector Convert Between Floating-Point and Integers
2847 def VCVTf2sd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
2848 v2i32, v2f32, fp_to_sint>;
2849 def VCVTf2ud : N2VD<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
2850 v2i32, v2f32, fp_to_uint>;
2851 def VCVTs2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
2852 v2f32, v2i32, sint_to_fp>;
2853 def VCVTu2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
2854 v2f32, v2i32, uint_to_fp>;
2856 def VCVTf2sq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
2857 v4i32, v4f32, fp_to_sint>;
2858 def VCVTf2uq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
2859 v4i32, v4f32, fp_to_uint>;
2860 def VCVTs2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
2861 v4f32, v4i32, sint_to_fp>;
2862 def VCVTu2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
2863 v4f32, v4i32, uint_to_fp>;
2865 // VCVT : Vector Convert Between Floating-Point and Fixed-Point.
2866 def VCVTf2xsd : N2VCvtD<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
2867 v2i32, v2f32, int_arm_neon_vcvtfp2fxs>;
2868 def VCVTf2xud : N2VCvtD<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
2869 v2i32, v2f32, int_arm_neon_vcvtfp2fxu>;
2870 def VCVTxs2fd : N2VCvtD<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
2871 v2f32, v2i32, int_arm_neon_vcvtfxs2fp>;
2872 def VCVTxu2fd : N2VCvtD<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
2873 v2f32, v2i32, int_arm_neon_vcvtfxu2fp>;
2875 def VCVTf2xsq : N2VCvtQ<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
2876 v4i32, v4f32, int_arm_neon_vcvtfp2fxs>;
2877 def VCVTf2xuq : N2VCvtQ<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
2878 v4i32, v4f32, int_arm_neon_vcvtfp2fxu>;
2879 def VCVTxs2fq : N2VCvtQ<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
2880 v4f32, v4i32, int_arm_neon_vcvtfxs2fp>;
2881 def VCVTxu2fq : N2VCvtQ<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
2882 v4f32, v4i32, int_arm_neon_vcvtfxu2fp>;
2886 // VREV64 : Vector Reverse elements within 64-bit doublewords
2888 class VREV64D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
2889 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 0, 0, (outs DPR:$dst),
2890 (ins DPR:$src), IIC_VMOVD,
2891 OpcodeStr, Dt, "$dst, $src", "",
2892 [(set DPR:$dst, (Ty (NEONvrev64 (Ty DPR:$src))))]>;
2893 class VREV64Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
2894 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 1, 0, (outs QPR:$dst),
2895 (ins QPR:$src), IIC_VMOVD,
2896 OpcodeStr, Dt, "$dst, $src", "",
2897 [(set QPR:$dst, (Ty (NEONvrev64 (Ty QPR:$src))))]>;
2899 def VREV64d8 : VREV64D<0b00, "vrev64", "8", v8i8>;
2900 def VREV64d16 : VREV64D<0b01, "vrev64", "16", v4i16>;
2901 def VREV64d32 : VREV64D<0b10, "vrev64", "32", v2i32>;
2902 def VREV64df : VREV64D<0b10, "vrev64", "32", v2f32>;
2904 def VREV64q8 : VREV64Q<0b00, "vrev64", "8", v16i8>;
2905 def VREV64q16 : VREV64Q<0b01, "vrev64", "16", v8i16>;
2906 def VREV64q32 : VREV64Q<0b10, "vrev64", "32", v4i32>;
2907 def VREV64qf : VREV64Q<0b10, "vrev64", "32", v4f32>;
2909 // VREV32 : Vector Reverse elements within 32-bit words
2911 class VREV32D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
2912 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 0, 0, (outs DPR:$dst),
2913 (ins DPR:$src), IIC_VMOVD,
2914 OpcodeStr, Dt, "$dst, $src", "",
2915 [(set DPR:$dst, (Ty (NEONvrev32 (Ty DPR:$src))))]>;
2916 class VREV32Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
2917 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 1, 0, (outs QPR:$dst),
2918 (ins QPR:$src), IIC_VMOVD,
2919 OpcodeStr, Dt, "$dst, $src", "",
2920 [(set QPR:$dst, (Ty (NEONvrev32 (Ty QPR:$src))))]>;
2922 def VREV32d8 : VREV32D<0b00, "vrev32", "8", v8i8>;
2923 def VREV32d16 : VREV32D<0b01, "vrev32", "16", v4i16>;
2925 def VREV32q8 : VREV32Q<0b00, "vrev32", "8", v16i8>;
2926 def VREV32q16 : VREV32Q<0b01, "vrev32", "16", v8i16>;
2928 // VREV16 : Vector Reverse elements within 16-bit halfwords
2930 class VREV16D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
2931 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 0, 0, (outs DPR:$dst),
2932 (ins DPR:$src), IIC_VMOVD,
2933 OpcodeStr, Dt, "$dst, $src", "",
2934 [(set DPR:$dst, (Ty (NEONvrev16 (Ty DPR:$src))))]>;
2935 class VREV16Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
2936 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 1, 0, (outs QPR:$dst),
2937 (ins QPR:$src), IIC_VMOVD,
2938 OpcodeStr, Dt, "$dst, $src", "",
2939 [(set QPR:$dst, (Ty (NEONvrev16 (Ty QPR:$src))))]>;
2941 def VREV16d8 : VREV16D<0b00, "vrev16", "8", v8i8>;
2942 def VREV16q8 : VREV16Q<0b00, "vrev16", "8", v16i8>;
2944 // Other Vector Shuffles.
2946 // VEXT : Vector Extract
2948 class VEXTd<string OpcodeStr, string Dt, ValueType Ty>
2949 : N3V<0,1,0b11,{?,?,?,?},0,0, (outs DPR:$dst),
2950 (ins DPR:$lhs, DPR:$rhs, i32imm:$index), IIC_VEXTD,
2951 OpcodeStr, Dt, "$dst, $lhs, $rhs, $index", "",
2952 [(set DPR:$dst, (Ty (NEONvext (Ty DPR:$lhs),
2953 (Ty DPR:$rhs), imm:$index)))]>;
2955 class VEXTq<string OpcodeStr, string Dt, ValueType Ty>
2956 : N3V<0,1,0b11,{?,?,?,?},1,0, (outs QPR:$dst),
2957 (ins QPR:$lhs, QPR:$rhs, i32imm:$index), IIC_VEXTQ,
2958 OpcodeStr, Dt, "$dst, $lhs, $rhs, $index", "",
2959 [(set QPR:$dst, (Ty (NEONvext (Ty QPR:$lhs),
2960 (Ty QPR:$rhs), imm:$index)))]>;
2962 def VEXTd8 : VEXTd<"vext", "8", v8i8>;
2963 def VEXTd16 : VEXTd<"vext", "16", v4i16>;
2964 def VEXTd32 : VEXTd<"vext", "32", v2i32>;
2965 def VEXTdf : VEXTd<"vext", "32", v2f32>;
2967 def VEXTq8 : VEXTq<"vext", "8", v16i8>;
2968 def VEXTq16 : VEXTq<"vext", "16", v8i16>;
2969 def VEXTq32 : VEXTq<"vext", "32", v4i32>;
2970 def VEXTqf : VEXTq<"vext", "32", v4f32>;
2972 // VTRN : Vector Transpose
2974 def VTRNd8 : N2VDShuffle<0b00, 0b00001, "vtrn", "8">;
2975 def VTRNd16 : N2VDShuffle<0b01, 0b00001, "vtrn", "16">;
2976 def VTRNd32 : N2VDShuffle<0b10, 0b00001, "vtrn", "32">;
2978 def VTRNq8 : N2VQShuffle<0b00, 0b00001, IIC_VPERMQ, "vtrn", "8">;
2979 def VTRNq16 : N2VQShuffle<0b01, 0b00001, IIC_VPERMQ, "vtrn", "16">;
2980 def VTRNq32 : N2VQShuffle<0b10, 0b00001, IIC_VPERMQ, "vtrn", "32">;
2982 // VUZP : Vector Unzip (Deinterleave)
2984 def VUZPd8 : N2VDShuffle<0b00, 0b00010, "vuzp", "8">;
2985 def VUZPd16 : N2VDShuffle<0b01, 0b00010, "vuzp", "16">;
2986 def VUZPd32 : N2VDShuffle<0b10, 0b00010, "vuzp", "32">;
2988 def VUZPq8 : N2VQShuffle<0b00, 0b00010, IIC_VPERMQ3, "vuzp", "8">;
2989 def VUZPq16 : N2VQShuffle<0b01, 0b00010, IIC_VPERMQ3, "vuzp", "16">;
2990 def VUZPq32 : N2VQShuffle<0b10, 0b00010, IIC_VPERMQ3, "vuzp", "32">;
2992 // VZIP : Vector Zip (Interleave)
2994 def VZIPd8 : N2VDShuffle<0b00, 0b00011, "vzip", "8">;
2995 def VZIPd16 : N2VDShuffle<0b01, 0b00011, "vzip", "16">;
2996 def VZIPd32 : N2VDShuffle<0b10, 0b00011, "vzip", "32">;
2998 def VZIPq8 : N2VQShuffle<0b00, 0b00011, IIC_VPERMQ3, "vzip", "8">;
2999 def VZIPq16 : N2VQShuffle<0b01, 0b00011, IIC_VPERMQ3, "vzip", "16">;
3000 def VZIPq32 : N2VQShuffle<0b10, 0b00011, IIC_VPERMQ3, "vzip", "32">;
3002 // Vector Table Lookup and Table Extension.
3004 // VTBL : Vector Table Lookup
3006 : N3V<1,1,0b11,0b1000,0,0, (outs DPR:$dst),
3007 (ins DPR:$tbl1, DPR:$src), IIC_VTB1,
3008 "vtbl", "8", "$dst, \\{$tbl1\\}, $src", "",
3009 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl1 DPR:$tbl1, DPR:$src)))]>;
3010 let hasExtraSrcRegAllocReq = 1 in {
3012 : N3V<1,1,0b11,0b1001,0,0, (outs DPR:$dst),
3013 (ins DPR:$tbl1, DPR:$tbl2, DPR:$src), IIC_VTB2,
3014 "vtbl", "8", "$dst, \\{$tbl1, $tbl2\\}, $src", "",
3015 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl2
3016 DPR:$tbl1, DPR:$tbl2, DPR:$src)))]>;
3018 : N3V<1,1,0b11,0b1010,0,0, (outs DPR:$dst),
3019 (ins DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$src), IIC_VTB3,
3020 "vtbl", "8", "$dst, \\{$tbl1, $tbl2, $tbl3\\}, $src", "",
3021 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl3
3022 DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$src)))]>;
3024 : N3V<1,1,0b11,0b1011,0,0, (outs DPR:$dst),
3025 (ins DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$src), IIC_VTB4,
3026 "vtbl", "8", "$dst, \\{$tbl1, $tbl2, $tbl3, $tbl4\\}, $src", "",
3027 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl4 DPR:$tbl1, DPR:$tbl2,
3028 DPR:$tbl3, DPR:$tbl4, DPR:$src)))]>;
3029 } // hasExtraSrcRegAllocReq = 1
3031 // VTBX : Vector Table Extension
3033 : N3V<1,1,0b11,0b1000,1,0, (outs DPR:$dst),
3034 (ins DPR:$orig, DPR:$tbl1, DPR:$src), IIC_VTBX1,
3035 "vtbx", "8", "$dst, \\{$tbl1\\}, $src", "$orig = $dst",
3036 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx1
3037 DPR:$orig, DPR:$tbl1, DPR:$src)))]>;
3038 let hasExtraSrcRegAllocReq = 1 in {
3040 : N3V<1,1,0b11,0b1001,1,0, (outs DPR:$dst),
3041 (ins DPR:$orig, DPR:$tbl1, DPR:$tbl2, DPR:$src), IIC_VTBX2,
3042 "vtbx", "8", "$dst, \\{$tbl1, $tbl2\\}, $src", "$orig = $dst",
3043 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx2
3044 DPR:$orig, DPR:$tbl1, DPR:$tbl2, DPR:$src)))]>;
3046 : N3V<1,1,0b11,0b1010,1,0, (outs DPR:$dst),
3047 (ins DPR:$orig, DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$src), IIC_VTBX3,
3048 "vtbx", "8", "$dst, \\{$tbl1, $tbl2, $tbl3\\}, $src", "$orig = $dst",
3049 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx3 DPR:$orig, DPR:$tbl1,
3050 DPR:$tbl2, DPR:$tbl3, DPR:$src)))]>;
3052 : N3V<1,1,0b11,0b1011,1,0, (outs DPR:$dst), (ins DPR:$orig, DPR:$tbl1,
3053 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$src), IIC_VTBX4,
3054 "vtbx", "8", "$dst, \\{$tbl1, $tbl2, $tbl3, $tbl4\\}, $src",
3056 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx4 DPR:$orig, DPR:$tbl1,
3057 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$src)))]>;
3058 } // hasExtraSrcRegAllocReq = 1
3060 //===----------------------------------------------------------------------===//
3061 // NEON instructions for single-precision FP math
3062 //===----------------------------------------------------------------------===//
3064 class N2VSPat<SDNode OpNode, ValueType ResTy, ValueType OpTy, NeonI Inst>
3065 : NEONFPPat<(ResTy (OpNode SPR:$a)),
3066 (EXTRACT_SUBREG (OpTy (Inst (INSERT_SUBREG (OpTy (IMPLICIT_DEF)),
3067 SPR:$a, arm_ssubreg_0))),
3070 class N3VSPat<SDNode OpNode, NeonI Inst>
3071 : NEONFPPat<(f32 (OpNode SPR:$a, SPR:$b)),
3072 (EXTRACT_SUBREG (v2f32
3073 (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3074 SPR:$a, arm_ssubreg_0),
3075 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3076 SPR:$b, arm_ssubreg_0))),
3079 class N3VSMulOpPat<SDNode MulNode, SDNode OpNode, NeonI Inst>
3080 : NEONFPPat<(f32 (OpNode SPR:$acc, (f32 (MulNode SPR:$a, SPR:$b)))),
3081 (EXTRACT_SUBREG (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3082 SPR:$acc, arm_ssubreg_0),
3083 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3084 SPR:$a, arm_ssubreg_0),
3085 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3086 SPR:$b, arm_ssubreg_0)),
3089 // These need separate instructions because they must use DPR_VFP2 register
3090 // class which have SPR sub-registers.
3092 // Vector Add Operations used for single-precision FP
3093 let neverHasSideEffects = 1 in
3094 def VADDfd_sfp : N3VS<0,0,0b00,0b1101,0, "vadd", "f32", v2f32, v2f32, fadd, 1>;
3095 def : N3VSPat<fadd, VADDfd_sfp>;
3097 // Vector Sub Operations used for single-precision FP
3098 let neverHasSideEffects = 1 in
3099 def VSUBfd_sfp : N3VS<0,0,0b10,0b1101,0, "vsub", "f32", v2f32, v2f32, fsub, 0>;
3100 def : N3VSPat<fsub, VSUBfd_sfp>;
3102 // Vector Multiply Operations used for single-precision FP
3103 let neverHasSideEffects = 1 in
3104 def VMULfd_sfp : N3VS<1,0,0b00,0b1101,1, "vmul", "f32", v2f32, v2f32, fmul, 1>;
3105 def : N3VSPat<fmul, VMULfd_sfp>;
3107 // Vector Multiply-Accumulate/Subtract used for single-precision FP
3108 // vml[as].f32 can cause 4-8 cycle stalls in following ASIMD instructions, so
3109 // we want to avoid them for now. e.g., alternating vmla/vadd instructions.
3111 //let neverHasSideEffects = 1 in
3112 //def VMLAfd_sfp : N3VSMulOp<0,0,0b00,0b1101,1, IIC_VMACD, "vmla", "f32",
3113 // v2f32, fmul, fadd>;
3114 //def : N3VSMulOpPat<fmul, fadd, VMLAfd_sfp>;
3116 //let neverHasSideEffects = 1 in
3117 //def VMLSfd_sfp : N3VSMulOp<0,0,0b10,0b1101,1, IIC_VMACD, "vmls", "f32",
3118 // v2f32, fmul, fsub>;
3119 //def : N3VSMulOpPat<fmul, fsub, VMLSfd_sfp>;
3121 // Vector Absolute used for single-precision FP
3122 let neverHasSideEffects = 1 in
3123 def VABSfd_sfp : N2V<0b11, 0b11, 0b10, 0b01, 0b01110, 0, 0,
3124 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), IIC_VUNAD,
3125 "vabs", "f32", "$dst, $src", "", []>;
3126 def : N2VSPat<fabs, f32, v2f32, VABSfd_sfp>;
3128 // Vector Negate used for single-precision FP
3129 let neverHasSideEffects = 1 in
3130 def VNEGfd_sfp : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
3131 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), IIC_VUNAD,
3132 "vneg", "f32", "$dst, $src", "", []>;
3133 def : N2VSPat<fneg, f32, v2f32, VNEGfd_sfp>;
3135 // Vector Maximum used for single-precision FP
3136 let neverHasSideEffects = 1 in
3137 def VMAXfd_sfp : N3V<0, 0, 0b00, 0b1111, 0, 0, (outs DPR_VFP2:$dst),
3138 (ins DPR_VFP2:$src1, DPR_VFP2:$src2), IIC_VBIND,
3139 "vmax", "f32", "$dst, $src1, $src2", "", []>;
3140 def : N3VSPat<NEONfmax, VMAXfd_sfp>;
3142 // Vector Minimum used for single-precision FP
3143 let neverHasSideEffects = 1 in
3144 def VMINfd_sfp : N3V<0, 0, 0b00, 0b1111, 0, 0, (outs DPR_VFP2:$dst),
3145 (ins DPR_VFP2:$src1, DPR_VFP2:$src2), IIC_VBIND,
3146 "vmin", "f32", "$dst, $src1, $src2", "", []>;
3147 def : N3VSPat<NEONfmin, VMINfd_sfp>;
3149 // Vector Convert between single-precision FP and integer
3150 let neverHasSideEffects = 1 in
3151 def VCVTf2sd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
3152 v2i32, v2f32, fp_to_sint>;
3153 def : N2VSPat<arm_ftosi, f32, v2f32, VCVTf2sd_sfp>;
3155 let neverHasSideEffects = 1 in
3156 def VCVTf2ud_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
3157 v2i32, v2f32, fp_to_uint>;
3158 def : N2VSPat<arm_ftoui, f32, v2f32, VCVTf2ud_sfp>;
3160 let neverHasSideEffects = 1 in
3161 def VCVTs2fd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
3162 v2f32, v2i32, sint_to_fp>;
3163 def : N2VSPat<arm_sitof, f32, v2i32, VCVTs2fd_sfp>;
3165 let neverHasSideEffects = 1 in
3166 def VCVTu2fd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
3167 v2f32, v2i32, uint_to_fp>;
3168 def : N2VSPat<arm_uitof, f32, v2i32, VCVTu2fd_sfp>;
3170 //===----------------------------------------------------------------------===//
3171 // Non-Instruction Patterns
3172 //===----------------------------------------------------------------------===//
3175 def : Pat<(v1i64 (bitconvert (v2i32 DPR:$src))), (v1i64 DPR:$src)>;
3176 def : Pat<(v1i64 (bitconvert (v4i16 DPR:$src))), (v1i64 DPR:$src)>;
3177 def : Pat<(v1i64 (bitconvert (v8i8 DPR:$src))), (v1i64 DPR:$src)>;
3178 def : Pat<(v1i64 (bitconvert (f64 DPR:$src))), (v1i64 DPR:$src)>;
3179 def : Pat<(v1i64 (bitconvert (v2f32 DPR:$src))), (v1i64 DPR:$src)>;
3180 def : Pat<(v2i32 (bitconvert (v1i64 DPR:$src))), (v2i32 DPR:$src)>;
3181 def : Pat<(v2i32 (bitconvert (v4i16 DPR:$src))), (v2i32 DPR:$src)>;
3182 def : Pat<(v2i32 (bitconvert (v8i8 DPR:$src))), (v2i32 DPR:$src)>;
3183 def : Pat<(v2i32 (bitconvert (f64 DPR:$src))), (v2i32 DPR:$src)>;
3184 def : Pat<(v2i32 (bitconvert (v2f32 DPR:$src))), (v2i32 DPR:$src)>;
3185 def : Pat<(v4i16 (bitconvert (v1i64 DPR:$src))), (v4i16 DPR:$src)>;
3186 def : Pat<(v4i16 (bitconvert (v2i32 DPR:$src))), (v4i16 DPR:$src)>;
3187 def : Pat<(v4i16 (bitconvert (v8i8 DPR:$src))), (v4i16 DPR:$src)>;
3188 def : Pat<(v4i16 (bitconvert (f64 DPR:$src))), (v4i16 DPR:$src)>;
3189 def : Pat<(v4i16 (bitconvert (v2f32 DPR:$src))), (v4i16 DPR:$src)>;
3190 def : Pat<(v8i8 (bitconvert (v1i64 DPR:$src))), (v8i8 DPR:$src)>;
3191 def : Pat<(v8i8 (bitconvert (v2i32 DPR:$src))), (v8i8 DPR:$src)>;
3192 def : Pat<(v8i8 (bitconvert (v4i16 DPR:$src))), (v8i8 DPR:$src)>;
3193 def : Pat<(v8i8 (bitconvert (f64 DPR:$src))), (v8i8 DPR:$src)>;
3194 def : Pat<(v8i8 (bitconvert (v2f32 DPR:$src))), (v8i8 DPR:$src)>;
3195 def : Pat<(f64 (bitconvert (v1i64 DPR:$src))), (f64 DPR:$src)>;
3196 def : Pat<(f64 (bitconvert (v2i32 DPR:$src))), (f64 DPR:$src)>;
3197 def : Pat<(f64 (bitconvert (v4i16 DPR:$src))), (f64 DPR:$src)>;
3198 def : Pat<(f64 (bitconvert (v8i8 DPR:$src))), (f64 DPR:$src)>;
3199 def : Pat<(f64 (bitconvert (v2f32 DPR:$src))), (f64 DPR:$src)>;
3200 def : Pat<(v2f32 (bitconvert (f64 DPR:$src))), (v2f32 DPR:$src)>;
3201 def : Pat<(v2f32 (bitconvert (v1i64 DPR:$src))), (v2f32 DPR:$src)>;
3202 def : Pat<(v2f32 (bitconvert (v2i32 DPR:$src))), (v2f32 DPR:$src)>;
3203 def : Pat<(v2f32 (bitconvert (v4i16 DPR:$src))), (v2f32 DPR:$src)>;
3204 def : Pat<(v2f32 (bitconvert (v8i8 DPR:$src))), (v2f32 DPR:$src)>;
3206 def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (v2i64 QPR:$src)>;
3207 def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (v2i64 QPR:$src)>;
3208 def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (v2i64 QPR:$src)>;
3209 def : Pat<(v2i64 (bitconvert (v2f64 QPR:$src))), (v2i64 QPR:$src)>;
3210 def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (v2i64 QPR:$src)>;
3211 def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (v4i32 QPR:$src)>;
3212 def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (v4i32 QPR:$src)>;
3213 def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (v4i32 QPR:$src)>;
3214 def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (v4i32 QPR:$src)>;
3215 def : Pat<(v4i32 (bitconvert (v4f32 QPR:$src))), (v4i32 QPR:$src)>;
3216 def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (v8i16 QPR:$src)>;
3217 def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (v8i16 QPR:$src)>;
3218 def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (v8i16 QPR:$src)>;
3219 def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (v8i16 QPR:$src)>;
3220 def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (v8i16 QPR:$src)>;
3221 def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (v16i8 QPR:$src)>;
3222 def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (v16i8 QPR:$src)>;
3223 def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (v16i8 QPR:$src)>;
3224 def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (v16i8 QPR:$src)>;
3225 def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (v16i8 QPR:$src)>;
3226 def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (v4f32 QPR:$src)>;
3227 def : Pat<(v4f32 (bitconvert (v4i32 QPR:$src))), (v4f32 QPR:$src)>;
3228 def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (v4f32 QPR:$src)>;
3229 def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (v4f32 QPR:$src)>;
3230 def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (v4f32 QPR:$src)>;
3231 def : Pat<(v2f64 (bitconvert (v2i64 QPR:$src))), (v2f64 QPR:$src)>;
3232 def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (v2f64 QPR:$src)>;
3233 def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 QPR:$src)>;
3234 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 QPR:$src)>;
3235 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 QPR:$src)>;