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>,
86 SDTCisSameAs<0, 2>, SDTCisSameAs<0, 3>]>;
87 def NEONzip : SDNode<"ARMISD::VZIP", SDTARMVSHUF2>;
88 def NEONuzp : SDNode<"ARMISD::VUZP", SDTARMVSHUF2>;
89 def NEONtrn : SDNode<"ARMISD::VTRN", SDTARMVSHUF2>;
91 //===----------------------------------------------------------------------===//
92 // NEON operand definitions
93 //===----------------------------------------------------------------------===//
95 // addrmode_neonldstm := reg
97 /* TODO: Take advantage of vldm.
98 def addrmode_neonldstm : Operand<i32>,
99 ComplexPattern<i32, 2, "SelectAddrModeNeonLdStM", []> {
100 let PrintMethod = "printAddrNeonLdStMOperand";
101 let MIOperandInfo = (ops GPR, i32imm);
105 //===----------------------------------------------------------------------===//
106 // NEON load / store instructions
107 //===----------------------------------------------------------------------===//
109 /* TODO: Take advantage of vldm.
110 let mayLoad = 1, hasExtraDefRegAllocReq = 1 in {
111 def VLDMD : NI<(outs),
112 (ins addrmode_neonldstm:$addr, reglist:$dst1, variable_ops),
114 "vldm${addr:submode} ${addr:base}, $dst1",
116 let Inst{27-25} = 0b110;
118 let Inst{11-9} = 0b101;
121 def VLDMS : NI<(outs),
122 (ins addrmode_neonldstm:$addr, reglist:$dst1, variable_ops),
124 "vldm${addr:submode} ${addr:base}, $dst1",
126 let Inst{27-25} = 0b110;
128 let Inst{11-9} = 0b101;
133 // Use vldmia to load a Q register as a D register pair.
134 def VLDRQ : NI4<(outs QPR:$dst), (ins addrmode4:$addr),
136 "vldmia $addr, ${dst:dregpair}",
137 [(set QPR:$dst, (v2f64 (load addrmode4:$addr)))]> {
138 let Inst{27-25} = 0b110;
139 let Inst{24} = 0; // P bit
140 let Inst{23} = 1; // U bit
142 let Inst{11-9} = 0b101;
145 // Use vstmia to store a Q register as a D register pair.
146 def VSTRQ : NI4<(outs), (ins QPR:$src, addrmode4:$addr),
148 "vstmia $addr, ${src:dregpair}",
149 [(store (v2f64 QPR:$src), addrmode4:$addr)]> {
150 let Inst{27-25} = 0b110;
151 let Inst{24} = 0; // P bit
152 let Inst{23} = 1; // U bit
154 let Inst{11-9} = 0b101;
157 // VLD1 : Vector Load (multiple single elements)
158 class VLD1D<bits<4> op7_4, string OpcodeStr, ValueType Ty, Intrinsic IntOp>
159 : NLdSt<0,0b10,0b0111,op7_4, (outs DPR:$dst), (ins addrmode6:$addr), IIC_VLD1,
160 !strconcat(OpcodeStr, "\t\\{$dst\\}, $addr"), "",
161 [(set DPR:$dst, (Ty (IntOp addrmode6:$addr)))]>;
162 class VLD1Q<bits<4> op7_4, string OpcodeStr, ValueType Ty, Intrinsic IntOp>
163 : NLdSt<0,0b10,0b1010,op7_4, (outs QPR:$dst), (ins addrmode6:$addr), IIC_VLD1,
164 !strconcat(OpcodeStr, "\t${dst:dregpair}, $addr"), "",
165 [(set QPR:$dst, (Ty (IntOp addrmode6:$addr)))]>;
167 def VLD1d8 : VLD1D<0b0000, "vld1.8", v8i8, int_arm_neon_vld1>;
168 def VLD1d16 : VLD1D<0b0100, "vld1.16", v4i16, int_arm_neon_vld1>;
169 def VLD1d32 : VLD1D<0b1000, "vld1.32", v2i32, int_arm_neon_vld1>;
170 def VLD1df : VLD1D<0b1000, "vld1.32", v2f32, int_arm_neon_vld1>;
171 def VLD1d64 : VLD1D<0b1100, "vld1.64", v1i64, int_arm_neon_vld1>;
173 def VLD1q8 : VLD1Q<0b0000, "vld1.8", v16i8, int_arm_neon_vld1>;
174 def VLD1q16 : VLD1Q<0b0100, "vld1.16", v8i16, int_arm_neon_vld1>;
175 def VLD1q32 : VLD1Q<0b1000, "vld1.32", v4i32, int_arm_neon_vld1>;
176 def VLD1qf : VLD1Q<0b1000, "vld1.32", v4f32, int_arm_neon_vld1>;
177 def VLD1q64 : VLD1Q<0b1100, "vld1.64", v2i64, int_arm_neon_vld1>;
179 let mayLoad = 1, hasExtraDefRegAllocReq = 1 in {
181 // VLD2 : Vector Load (multiple 2-element structures)
182 class VLD2D<bits<4> op7_4, string OpcodeStr>
183 : NLdSt<0,0b10,0b1000,op7_4, (outs DPR:$dst1, DPR:$dst2),
184 (ins addrmode6:$addr), IIC_VLD2,
185 !strconcat(OpcodeStr, "\t\\{$dst1,$dst2\\}, $addr"), "", []>;
186 class VLD2Q<bits<4> op7_4, string OpcodeStr>
187 : NLdSt<0,0b10,0b0011,op7_4,
188 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
189 (ins addrmode6:$addr), IIC_VLD2,
190 !strconcat(OpcodeStr, "\t\\{$dst1,$dst2,$dst3,$dst4\\}, $addr"),
193 def VLD2d8 : VLD2D<0b0000, "vld2.8">;
194 def VLD2d16 : VLD2D<0b0100, "vld2.16">;
195 def VLD2d32 : VLD2D<0b1000, "vld2.32">;
196 def VLD2d64 : NLdSt<0,0b10,0b1010,0b1100, (outs DPR:$dst1, DPR:$dst2),
197 (ins addrmode6:$addr), IIC_VLD1,
198 "vld1.64\t\\{$dst1,$dst2\\}, $addr", "", []>;
200 def VLD2q8 : VLD2Q<0b0000, "vld2.8">;
201 def VLD2q16 : VLD2Q<0b0100, "vld2.16">;
202 def VLD2q32 : VLD2Q<0b1000, "vld2.32">;
204 // VLD3 : Vector Load (multiple 3-element structures)
205 class VLD3D<bits<4> op7_4, string OpcodeStr>
206 : NLdSt<0,0b10,0b0100,op7_4, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3),
207 (ins addrmode6:$addr), IIC_VLD3,
208 !strconcat(OpcodeStr, "\t\\{$dst1,$dst2,$dst3\\}, $addr"), "", []>;
209 class VLD3WB<bits<4> op7_4, string OpcodeStr>
210 : NLdSt<0,0b10,0b0101,op7_4, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, GPR:$wb),
211 (ins addrmode6:$addr), IIC_VLD3,
212 !strconcat(OpcodeStr, "\t\\{$dst1,$dst2,$dst3\\}, $addr"),
213 "$addr.addr = $wb", []>;
215 def VLD3d8 : VLD3D<0b0000, "vld3.8">;
216 def VLD3d16 : VLD3D<0b0100, "vld3.16">;
217 def VLD3d32 : VLD3D<0b1000, "vld3.32">;
218 def VLD3d64 : NLdSt<0,0b10,0b0110,0b1100,
219 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3),
220 (ins addrmode6:$addr), IIC_VLD1,
221 "vld1.64\t\\{$dst1,$dst2,$dst3\\}, $addr", "", []>;
223 // vld3 to double-spaced even registers.
224 def VLD3q8a : VLD3WB<0b0000, "vld3.8">;
225 def VLD3q16a : VLD3WB<0b0100, "vld3.16">;
226 def VLD3q32a : VLD3WB<0b1000, "vld3.32">;
228 // vld3 to double-spaced odd registers.
229 def VLD3q8b : VLD3WB<0b0000, "vld3.8">;
230 def VLD3q16b : VLD3WB<0b0100, "vld3.16">;
231 def VLD3q32b : VLD3WB<0b1000, "vld3.32">;
233 // VLD4 : Vector Load (multiple 4-element structures)
234 class VLD4D<bits<4> op7_4, string OpcodeStr>
235 : NLdSt<0,0b10,0b0000,op7_4,
236 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
237 (ins addrmode6:$addr), IIC_VLD4,
238 !strconcat(OpcodeStr, "\t\\{$dst1,$dst2,$dst3,$dst4\\}, $addr"),
240 class VLD4WB<bits<4> op7_4, string OpcodeStr>
241 : NLdSt<0,0b10,0b0001,op7_4,
242 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
243 (ins addrmode6:$addr), IIC_VLD4,
244 !strconcat(OpcodeStr, "\t\\{$dst1,$dst2,$dst3,$dst4\\}, $addr"),
245 "$addr.addr = $wb", []>;
247 def VLD4d8 : VLD4D<0b0000, "vld4.8">;
248 def VLD4d16 : VLD4D<0b0100, "vld4.16">;
249 def VLD4d32 : VLD4D<0b1000, "vld4.32">;
250 def VLD4d64 : NLdSt<0,0b10,0b0010,0b1100,
251 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
252 (ins addrmode6:$addr), IIC_VLD1,
253 "vld1.64\t\\{$dst1,$dst2,$dst3,$dst4\\}, $addr", "", []>;
255 // vld4 to double-spaced even registers.
256 def VLD4q8a : VLD4WB<0b0000, "vld4.8">;
257 def VLD4q16a : VLD4WB<0b0100, "vld4.16">;
258 def VLD4q32a : VLD4WB<0b1000, "vld4.32">;
260 // vld4 to double-spaced odd registers.
261 def VLD4q8b : VLD4WB<0b0000, "vld4.8">;
262 def VLD4q16b : VLD4WB<0b0100, "vld4.16">;
263 def VLD4q32b : VLD4WB<0b1000, "vld4.32">;
265 // VLD1LN : Vector Load (single element to one lane)
266 // FIXME: Not yet implemented.
268 // VLD2LN : Vector Load (single 2-element structure to one lane)
269 class VLD2LN<bits<4> op11_8, string OpcodeStr>
270 : NLdSt<1,0b10,op11_8,0b0000, (outs DPR:$dst1, DPR:$dst2),
271 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, nohash_imm:$lane),
273 !strconcat(OpcodeStr, "\t\\{$dst1[$lane],$dst2[$lane]\\}, $addr"),
274 "$src1 = $dst1, $src2 = $dst2", []>;
276 def VLD2LNd8 : VLD2LN<0b0001, "vld2.8">;
277 def VLD2LNd16 : VLD2LN<0b0101, "vld2.16">;
278 def VLD2LNd32 : VLD2LN<0b1001, "vld2.32">;
280 // vld2 to double-spaced even registers.
281 def VLD2LNq16a: VLD2LN<0b0101, "vld2.16">;
282 def VLD2LNq32a: VLD2LN<0b1001, "vld2.32">;
284 // vld2 to double-spaced odd registers.
285 def VLD2LNq16b: VLD2LN<0b0101, "vld2.16">;
286 def VLD2LNq32b: VLD2LN<0b1001, "vld2.32">;
288 // VLD3LN : Vector Load (single 3-element structure to one lane)
289 class VLD3LN<bits<4> op11_8, string OpcodeStr>
290 : NLdSt<1,0b10,op11_8,0b0000, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3),
291 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3,
292 nohash_imm:$lane), IIC_VLD3,
293 !strconcat(OpcodeStr,
294 "\t\\{$dst1[$lane],$dst2[$lane],$dst3[$lane]\\}, $addr"),
295 "$src1 = $dst1, $src2 = $dst2, $src3 = $dst3", []>;
297 def VLD3LNd8 : VLD3LN<0b0010, "vld3.8">;
298 def VLD3LNd16 : VLD3LN<0b0110, "vld3.16">;
299 def VLD3LNd32 : VLD3LN<0b1010, "vld3.32">;
301 // vld3 to double-spaced even registers.
302 def VLD3LNq16a: VLD3LN<0b0101, "vld3.16">;
303 def VLD3LNq32a: VLD3LN<0b1001, "vld3.32">;
305 // vld3 to double-spaced odd registers.
306 def VLD3LNq16b: VLD3LN<0b0101, "vld3.16">;
307 def VLD3LNq32b: VLD3LN<0b1001, "vld3.32">;
309 // VLD4LN : Vector Load (single 4-element structure to one lane)
310 class VLD4LND<bits<4> op11_8, string OpcodeStr>
311 : NLdSt<1,0b10,op11_8,0b0000,
312 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
313 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
314 nohash_imm:$lane), IIC_VLD4,
315 !strconcat(OpcodeStr,
316 "\t\\{$dst1[$lane],$dst2[$lane],$dst3[$lane],$dst4[$lane]\\}, $addr"),
317 "$src1 = $dst1, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4", []>;
319 def VLD4LNd8 : VLD4LND<0b0011, "vld4.8">;
320 def VLD4LNd16 : VLD4LND<0b0111, "vld4.16">;
321 def VLD4LNd32 : VLD4LND<0b1011, "vld4.32">;
323 // VLD1DUP : Vector Load (single element to all lanes)
324 // VLD2DUP : Vector Load (single 2-element structure to all lanes)
325 // VLD3DUP : Vector Load (single 3-element structure to all lanes)
326 // VLD4DUP : Vector Load (single 4-element structure to all lanes)
327 // FIXME: Not yet implemented.
328 } // mayLoad = 1, hasExtraDefRegAllocReq = 1
330 // VST1 : Vector Store (multiple single elements)
331 class VST1D<bits<4> op7_4, string OpcodeStr, ValueType Ty, Intrinsic IntOp>
332 : NLdSt<0,0b00,0b0111,op7_4, (outs), (ins addrmode6:$addr, DPR:$src), IIC_VST,
333 !strconcat(OpcodeStr, "\t\\{$src\\}, $addr"), "",
334 [(IntOp addrmode6:$addr, (Ty DPR:$src))]>;
335 class VST1Q<bits<4> op7_4, string OpcodeStr, ValueType Ty, Intrinsic IntOp>
336 : NLdSt<0,0b00,0b1010,op7_4, (outs), (ins addrmode6:$addr, QPR:$src), IIC_VST,
337 !strconcat(OpcodeStr, "\t${src:dregpair}, $addr"), "",
338 [(IntOp addrmode6:$addr, (Ty QPR:$src))]>;
340 let hasExtraSrcRegAllocReq = 1 in {
341 def VST1d8 : VST1D<0b0000, "vst1.8", v8i8, int_arm_neon_vst1>;
342 def VST1d16 : VST1D<0b0100, "vst1.16", v4i16, int_arm_neon_vst1>;
343 def VST1d32 : VST1D<0b1000, "vst1.32", v2i32, int_arm_neon_vst1>;
344 def VST1df : VST1D<0b1000, "vst1.32", v2f32, int_arm_neon_vst1>;
345 def VST1d64 : VST1D<0b1100, "vst1.64", v1i64, int_arm_neon_vst1>;
347 def VST1q8 : VST1Q<0b0000, "vst1.8", v16i8, int_arm_neon_vst1>;
348 def VST1q16 : VST1Q<0b0100, "vst1.16", v8i16, int_arm_neon_vst1>;
349 def VST1q32 : VST1Q<0b1000, "vst1.32", v4i32, int_arm_neon_vst1>;
350 def VST1qf : VST1Q<0b1000, "vst1.32", v4f32, int_arm_neon_vst1>;
351 def VST1q64 : VST1Q<0b1100, "vst1.64", v2i64, int_arm_neon_vst1>;
352 } // hasExtraSrcRegAllocReq
354 let mayStore = 1, hasExtraSrcRegAllocReq = 1 in {
356 // VST2 : Vector Store (multiple 2-element structures)
357 class VST2D<bits<4> op7_4, string OpcodeStr>
358 : NLdSt<0,0b00,0b1000,op7_4, (outs),
359 (ins addrmode6:$addr, DPR:$src1, DPR:$src2), IIC_VST,
360 !strconcat(OpcodeStr, "\t\\{$src1,$src2\\}, $addr"), "", []>;
361 class VST2Q<bits<4> op7_4, string OpcodeStr>
362 : NLdSt<0,0b00,0b0011,op7_4, (outs),
363 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
365 !strconcat(OpcodeStr, "\t\\{$src1,$src2,$src3,$src4\\}, $addr"),
368 def VST2d8 : VST2D<0b0000, "vst2.8">;
369 def VST2d16 : VST2D<0b0100, "vst2.16">;
370 def VST2d32 : VST2D<0b1000, "vst2.32">;
371 def VST2d64 : NLdSt<0,0b00,0b1010,0b1100, (outs),
372 (ins addrmode6:$addr, DPR:$src1, DPR:$src2), IIC_VST,
373 "vst1.64\t\\{$src1,$src2\\}, $addr", "", []>;
375 def VST2q8 : VST2Q<0b0000, "vst2.8">;
376 def VST2q16 : VST2Q<0b0100, "vst2.16">;
377 def VST2q32 : VST2Q<0b1000, "vst2.32">;
379 // VST3 : Vector Store (multiple 3-element structures)
380 class VST3D<bits<4> op7_4, string OpcodeStr>
381 : NLdSt<0,0b00,0b0100,op7_4, (outs),
382 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3), IIC_VST,
383 !strconcat(OpcodeStr, "\t\\{$src1,$src2,$src3\\}, $addr"), "", []>;
384 class VST3WB<bits<4> op7_4, string OpcodeStr>
385 : NLdSt<0,0b00,0b0101,op7_4, (outs GPR:$wb),
386 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3), IIC_VST,
387 !strconcat(OpcodeStr, "\t\\{$src1,$src2,$src3\\}, $addr"),
388 "$addr.addr = $wb", []>;
390 def VST3d8 : VST3D<0b0000, "vst3.8">;
391 def VST3d16 : VST3D<0b0100, "vst3.16">;
392 def VST3d32 : VST3D<0b1000, "vst3.32">;
393 def VST3d64 : NLdSt<0,0b00,0b0110,0b1100, (outs),
394 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3),
396 "vst1.64\t\\{$src1,$src2,$src3\\}, $addr", "", []>;
398 // vst3 to double-spaced even registers.
399 def VST3q8a : VST3WB<0b0000, "vst3.8">;
400 def VST3q16a : VST3WB<0b0100, "vst3.16">;
401 def VST3q32a : VST3WB<0b1000, "vst3.32">;
403 // vst3 to double-spaced odd registers.
404 def VST3q8b : VST3WB<0b0000, "vst3.8">;
405 def VST3q16b : VST3WB<0b0100, "vst3.16">;
406 def VST3q32b : VST3WB<0b1000, "vst3.32">;
408 // VST4 : Vector Store (multiple 4-element structures)
409 class VST4D<bits<4> op7_4, string OpcodeStr>
410 : NLdSt<0,0b00,0b0000,op7_4, (outs),
411 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
413 !strconcat(OpcodeStr, "\t\\{$src1,$src2,$src3,$src4\\}, $addr"),
415 class VST4WB<bits<4> op7_4, string OpcodeStr>
416 : NLdSt<0,0b00,0b0001,op7_4, (outs GPR:$wb),
417 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
419 !strconcat(OpcodeStr, "\t\\{$src1,$src2,$src3,$src4\\}, $addr"),
420 "$addr.addr = $wb", []>;
422 def VST4d8 : VST4D<0b0000, "vst4.8">;
423 def VST4d16 : VST4D<0b0100, "vst4.16">;
424 def VST4d32 : VST4D<0b1000, "vst4.32">;
425 def VST4d64 : NLdSt<0,0b00,0b0010,0b1100, (outs),
426 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3,
428 "vst1.64\t\\{$src1,$src2,$src3,$src4\\}, $addr", "", []>;
430 // vst4 to double-spaced even registers.
431 def VST4q8a : VST4WB<0b0000, "vst4.8">;
432 def VST4q16a : VST4WB<0b0100, "vst4.16">;
433 def VST4q32a : VST4WB<0b1000, "vst4.32">;
435 // vst4 to double-spaced odd registers.
436 def VST4q8b : VST4WB<0b0000, "vst4.8">;
437 def VST4q16b : VST4WB<0b0100, "vst4.16">;
438 def VST4q32b : VST4WB<0b1000, "vst4.32">;
440 // VST1LN : Vector Store (single element from one lane)
441 // FIXME: Not yet implemented.
443 // VST2LN : Vector Store (single 2-element structure from one lane)
444 class VST2LND<bits<4> op11_8, string OpcodeStr>
445 : NLdSt<1,0b00,op11_8,0b0000, (outs),
446 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, nohash_imm:$lane),
448 !strconcat(OpcodeStr, "\t\\{$src1[$lane],$src2[$lane]\\}, $addr"),
451 def VST2LNd8 : VST2LND<0b0000, "vst2.8">;
452 def VST2LNd16 : VST2LND<0b0100, "vst2.16">;
453 def VST2LNd32 : VST2LND<0b1000, "vst2.32">;
455 // VST3LN : Vector Store (single 3-element structure from one lane)
456 class VST3LND<bits<4> op11_8, string OpcodeStr>
457 : NLdSt<1,0b00,op11_8,0b0000, (outs),
458 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3,
459 nohash_imm:$lane), IIC_VST,
460 !strconcat(OpcodeStr,
461 "\t\\{$src1[$lane],$src2[$lane],$src3[$lane]\\}, $addr"), "", []>;
463 def VST3LNd8 : VST3LND<0b0010, "vst3.8">;
464 def VST3LNd16 : VST3LND<0b0110, "vst3.16">;
465 def VST3LNd32 : VST3LND<0b1010, "vst3.32">;
467 // VST4LN : Vector Store (single 4-element structure from one lane)
468 class VST4LND<bits<4> op11_8, string OpcodeStr>
469 : NLdSt<1,0b00,op11_8,0b0000, (outs),
470 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
471 nohash_imm:$lane), IIC_VST,
472 !strconcat(OpcodeStr,
473 "\t\\{$src1[$lane],$src2[$lane],$src3[$lane],$src4[$lane]\\}, $addr"),
476 def VST4LNd8 : VST4LND<0b0011, "vst4.8">;
477 def VST4LNd16 : VST4LND<0b0111, "vst4.16">;
478 def VST4LNd32 : VST4LND<0b1011, "vst4.32">;
479 } // mayStore = 1, hasExtraSrcRegAllocReq = 1
482 //===----------------------------------------------------------------------===//
483 // NEON pattern fragments
484 //===----------------------------------------------------------------------===//
486 // Extract D sub-registers of Q registers.
487 // (arm_dsubreg_0 is 5; arm_dsubreg_1 is 6)
488 def DSubReg_i8_reg : SDNodeXForm<imm, [{
489 return CurDAG->getTargetConstant(5 + N->getZExtValue() / 8, MVT::i32);
491 def DSubReg_i16_reg : SDNodeXForm<imm, [{
492 return CurDAG->getTargetConstant(5 + N->getZExtValue() / 4, MVT::i32);
494 def DSubReg_i32_reg : SDNodeXForm<imm, [{
495 return CurDAG->getTargetConstant(5 + N->getZExtValue() / 2, MVT::i32);
497 def DSubReg_f64_reg : SDNodeXForm<imm, [{
498 return CurDAG->getTargetConstant(5 + N->getZExtValue(), MVT::i32);
500 def DSubReg_f64_other_reg : SDNodeXForm<imm, [{
501 return CurDAG->getTargetConstant(5 + (1 - N->getZExtValue()), MVT::i32);
504 // Extract S sub-registers of Q/D registers.
505 // (arm_ssubreg_0 is 1; arm_ssubreg_1 is 2; etc.)
506 def SSubReg_f32_reg : SDNodeXForm<imm, [{
507 return CurDAG->getTargetConstant(1 + N->getZExtValue(), MVT::i32);
510 // Translate lane numbers from Q registers to D subregs.
511 def SubReg_i8_lane : SDNodeXForm<imm, [{
512 return CurDAG->getTargetConstant(N->getZExtValue() & 7, MVT::i32);
514 def SubReg_i16_lane : SDNodeXForm<imm, [{
515 return CurDAG->getTargetConstant(N->getZExtValue() & 3, MVT::i32);
517 def SubReg_i32_lane : SDNodeXForm<imm, [{
518 return CurDAG->getTargetConstant(N->getZExtValue() & 1, MVT::i32);
521 //===----------------------------------------------------------------------===//
522 // Instruction Classes
523 //===----------------------------------------------------------------------===//
525 // Basic 2-register operations, both double- and quad-register.
526 class N2VD<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
527 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
528 ValueType ResTy, ValueType OpTy, SDNode OpNode>
529 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
530 (ins DPR:$src), IIC_VUNAD, !strconcat(OpcodeStr, "\t$dst, $src"), "",
531 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src))))]>;
532 class N2VQ<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
533 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
534 ValueType ResTy, ValueType OpTy, SDNode OpNode>
535 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
536 (ins QPR:$src), IIC_VUNAQ, !strconcat(OpcodeStr, "\t$dst, $src"), "",
537 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src))))]>;
539 // Basic 2-register operations, scalar single-precision.
540 class N2VDs<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
541 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
542 ValueType ResTy, ValueType OpTy, SDNode OpNode>
543 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
544 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src),
545 IIC_VUNAD, !strconcat(OpcodeStr, "\t$dst, $src"), "", []>;
547 class N2VDsPat<SDNode OpNode, ValueType ResTy, ValueType OpTy, NeonI Inst>
548 : NEONFPPat<(ResTy (OpNode SPR:$a)),
550 (Inst (INSERT_SUBREG (OpTy (IMPLICIT_DEF)), SPR:$a, arm_ssubreg_0)),
553 // Basic 2-register intrinsics, both double- and quad-register.
554 class N2VDInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
555 bits<2> op17_16, bits<5> op11_7, bit op4,
556 InstrItinClass itin, string OpcodeStr,
557 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
558 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
559 (ins DPR:$src), itin, !strconcat(OpcodeStr, "\t$dst, $src"), "",
560 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
561 class N2VQInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
562 bits<2> op17_16, bits<5> op11_7, bit op4,
563 InstrItinClass itin, string OpcodeStr,
564 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
565 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
566 (ins QPR:$src), itin, !strconcat(OpcodeStr, "\t$dst, $src"), "",
567 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
569 // Basic 2-register intrinsics, scalar single-precision
570 class N2VDInts<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
571 bits<2> op17_16, bits<5> op11_7, bit op4,
572 InstrItinClass itin, string OpcodeStr,
573 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
574 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
575 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), itin,
576 !strconcat(OpcodeStr, "\t$dst, $src"), "", []>;
578 class N2VDIntsPat<SDNode OpNode, NeonI Inst>
579 : NEONFPPat<(f32 (OpNode SPR:$a)),
581 (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$a, arm_ssubreg_0)),
584 // Narrow 2-register intrinsics.
585 class N2VNInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
586 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
587 InstrItinClass itin, string OpcodeStr,
588 ValueType TyD, ValueType TyQ, Intrinsic IntOp>
589 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$dst),
590 (ins QPR:$src), itin, !strconcat(OpcodeStr, "\t$dst, $src"), "",
591 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src))))]>;
593 // Long 2-register intrinsics. (This is currently only used for VMOVL and is
594 // derived from N2VImm instead of N2V because of the way the size is encoded.)
595 class N2VLInt<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
596 bit op6, bit op4, InstrItinClass itin, string OpcodeStr,
597 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
598 : N2VImm<op24, op23, op21_16, op11_8, op7, op6, op4, (outs QPR:$dst),
599 (ins DPR:$src), itin, !strconcat(OpcodeStr, "\t$dst, $src"), "",
600 [(set QPR:$dst, (TyQ (IntOp (TyD DPR:$src))))]>;
602 // 2-register shuffles (VTRN/VZIP/VUZP), both double- and quad-register.
603 class N2VDShuffle<bits<2> op19_18, bits<5> op11_7, string OpcodeStr>
604 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 0, 0, (outs DPR:$dst1, DPR:$dst2),
605 (ins DPR:$src1, DPR:$src2), IIC_VPERMD,
606 !strconcat(OpcodeStr, "\t$dst1, $dst2"),
607 "$src1 = $dst1, $src2 = $dst2", []>;
608 class N2VQShuffle<bits<2> op19_18, bits<5> op11_7,
609 InstrItinClass itin, string OpcodeStr>
610 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 1, 0, (outs QPR:$dst1, QPR:$dst2),
611 (ins QPR:$src1, QPR:$src2), itin,
612 !strconcat(OpcodeStr, "\t$dst1, $dst2"),
613 "$src1 = $dst1, $src2 = $dst2", []>;
615 // Basic 3-register operations, both double- and quad-register.
616 class N3VD<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
617 InstrItinClass itin, string OpcodeStr, ValueType ResTy, ValueType OpTy,
618 SDNode OpNode, bit Commutable>
619 : N3V<op24, op23, op21_20, op11_8, 0, op4,
620 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), itin,
621 !strconcat(OpcodeStr, "\t$dst, $src1, $src2"), "",
622 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src1), (OpTy DPR:$src2))))]> {
623 let isCommutable = Commutable;
625 class N3VDSL<bits<2> op21_20, bits<4> op11_8,
626 InstrItinClass itin, string OpcodeStr, ValueType Ty, SDNode ShOp>
627 : N3V<0, 1, op21_20, op11_8, 1, 0,
628 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
629 itin, !strconcat(OpcodeStr, "\t$dst, $src1, $src2[$lane]"), "",
631 (Ty (ShOp (Ty DPR:$src1),
632 (Ty (NEONvduplane (Ty DPR_VFP2:$src2),
634 let isCommutable = 0;
636 class N3VDSL16<bits<2> op21_20, bits<4> op11_8,
637 string OpcodeStr, ValueType Ty, SDNode ShOp>
638 : N3V<0, 1, op21_20, op11_8, 1, 0,
639 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
641 !strconcat(OpcodeStr, "\t$dst, $src1, $src2[$lane]"), "",
643 (Ty (ShOp (Ty DPR:$src1),
644 (Ty (NEONvduplane (Ty DPR_8:$src2),
646 let isCommutable = 0;
649 class N3VQ<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
650 InstrItinClass itin, string OpcodeStr, ValueType ResTy, ValueType OpTy,
651 SDNode OpNode, bit Commutable>
652 : N3V<op24, op23, op21_20, op11_8, 1, op4,
653 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), itin,
654 !strconcat(OpcodeStr, "\t$dst, $src1, $src2"), "",
655 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src1), (OpTy QPR:$src2))))]> {
656 let isCommutable = Commutable;
658 class N3VQSL<bits<2> op21_20, bits<4> op11_8,
659 InstrItinClass itin, string OpcodeStr,
660 ValueType ResTy, ValueType OpTy, SDNode ShOp>
661 : N3V<1, 1, op21_20, op11_8, 1, 0,
662 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
663 itin, !strconcat(OpcodeStr, "\t$dst, $src1, $src2[$lane]"), "",
664 [(set (ResTy QPR:$dst),
665 (ResTy (ShOp (ResTy QPR:$src1),
666 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
668 let isCommutable = 0;
670 class N3VQSL16<bits<2> op21_20, bits<4> op11_8,
671 string OpcodeStr, ValueType ResTy, ValueType OpTy, SDNode ShOp>
672 : N3V<1, 1, op21_20, op11_8, 1, 0,
673 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
675 !strconcat(OpcodeStr, "\t$dst, $src1, $src2[$lane]"), "",
676 [(set (ResTy QPR:$dst),
677 (ResTy (ShOp (ResTy QPR:$src1),
678 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
680 let isCommutable = 0;
683 // Basic 3-register operations, scalar single-precision
684 class N3VDs<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
685 string OpcodeStr, ValueType ResTy, ValueType OpTy,
686 SDNode OpNode, bit Commutable>
687 : N3V<op24, op23, op21_20, op11_8, 0, op4,
688 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src1, DPR_VFP2:$src2), IIC_VBIND,
689 !strconcat(OpcodeStr, "\t$dst, $src1, $src2"), "", []> {
690 let isCommutable = Commutable;
692 class N3VDsPat<SDNode OpNode, NeonI Inst>
693 : NEONFPPat<(f32 (OpNode SPR:$a, SPR:$b)),
695 (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$a, arm_ssubreg_0),
696 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$b, arm_ssubreg_0)),
699 // Basic 3-register intrinsics, both double- and quad-register.
700 class N3VDInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
701 InstrItinClass itin, string OpcodeStr, ValueType ResTy, ValueType OpTy,
702 Intrinsic IntOp, bit Commutable>
703 : N3V<op24, op23, op21_20, op11_8, 0, op4,
704 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), itin,
705 !strconcat(OpcodeStr, "\t$dst, $src1, $src2"), "",
706 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src1), (OpTy DPR:$src2))))]> {
707 let isCommutable = Commutable;
709 class N3VDIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
710 string OpcodeStr, ValueType Ty, Intrinsic IntOp>
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, !strconcat(OpcodeStr, "\t$dst, $src1, $src2[$lane]"), "",
715 (Ty (IntOp (Ty DPR:$src1),
716 (Ty (NEONvduplane (Ty DPR_VFP2:$src2),
718 let isCommutable = 0;
720 class N3VDIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
721 string OpcodeStr, ValueType Ty, Intrinsic IntOp>
722 : N3V<0, 1, op21_20, op11_8, 1, 0,
723 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
724 itin, !strconcat(OpcodeStr, "\t$dst, $src1, $src2[$lane]"), "",
726 (Ty (IntOp (Ty DPR:$src1),
727 (Ty (NEONvduplane (Ty DPR_8:$src2),
729 let isCommutable = 0;
732 class N3VQInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
733 InstrItinClass itin, string OpcodeStr, ValueType ResTy, ValueType OpTy,
734 Intrinsic IntOp, bit Commutable>
735 : N3V<op24, op23, op21_20, op11_8, 1, op4,
736 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), itin,
737 !strconcat(OpcodeStr, "\t$dst, $src1, $src2"), "",
738 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src1), (OpTy QPR:$src2))))]> {
739 let isCommutable = Commutable;
741 class N3VQIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
742 string OpcodeStr, ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
743 : N3V<1, 1, op21_20, op11_8, 1, 0,
744 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
745 itin, !strconcat(OpcodeStr, "\t$dst, $src1, $src2[$lane]"), "",
746 [(set (ResTy QPR:$dst),
747 (ResTy (IntOp (ResTy QPR:$src1),
748 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
750 let isCommutable = 0;
752 class N3VQIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
753 string OpcodeStr, ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
754 : N3V<1, 1, op21_20, op11_8, 1, 0,
755 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
756 itin, !strconcat(OpcodeStr, "\t$dst, $src1, $src2[$lane]"), "",
757 [(set (ResTy QPR:$dst),
758 (ResTy (IntOp (ResTy QPR:$src1),
759 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
761 let isCommutable = 0;
764 // Multiply-Add/Sub operations, both double- and quad-register.
765 class N3VDMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
766 InstrItinClass itin, string OpcodeStr,
767 ValueType Ty, SDNode MulOp, SDNode OpNode>
768 : N3V<op24, op23, op21_20, op11_8, 0, op4,
769 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3), itin,
770 !strconcat(OpcodeStr, "\t$dst, $src2, $src3"), "$src1 = $dst",
771 [(set DPR:$dst, (Ty (OpNode DPR:$src1,
772 (Ty (MulOp DPR:$src2, DPR:$src3)))))]>;
773 class N3VDMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
774 string OpcodeStr, ValueType Ty, SDNode MulOp, SDNode ShOp>
775 : N3V<0, 1, op21_20, op11_8, 1, 0,
777 (ins DPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane), itin,
778 !strconcat(OpcodeStr, "\t$dst, $src2, $src3[$lane]"), "$src1 = $dst",
780 (Ty (ShOp (Ty DPR:$src1),
781 (Ty (MulOp DPR:$src2,
782 (Ty (NEONvduplane (Ty DPR_VFP2:$src3),
784 class N3VDMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
785 string OpcodeStr, ValueType Ty, SDNode MulOp, SDNode ShOp>
786 : N3V<0, 1, op21_20, op11_8, 1, 0,
788 (ins DPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane), itin,
789 !strconcat(OpcodeStr, "\t$dst, $src2, $src3[$lane]"), "$src1 = $dst",
791 (Ty (ShOp (Ty DPR:$src1),
792 (Ty (MulOp DPR:$src2,
793 (Ty (NEONvduplane (Ty DPR_8:$src3),
796 class N3VQMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
797 InstrItinClass itin, string OpcodeStr, ValueType Ty,
798 SDNode MulOp, SDNode OpNode>
799 : N3V<op24, op23, op21_20, op11_8, 1, op4,
800 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3), itin,
801 !strconcat(OpcodeStr, "\t$dst, $src2, $src3"), "$src1 = $dst",
802 [(set QPR:$dst, (Ty (OpNode QPR:$src1,
803 (Ty (MulOp QPR:$src2, QPR:$src3)))))]>;
804 class N3VQMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
805 string OpcodeStr, ValueType ResTy, ValueType OpTy,
806 SDNode MulOp, SDNode ShOp>
807 : N3V<1, 1, op21_20, op11_8, 1, 0,
809 (ins QPR:$src1, QPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane), itin,
810 !strconcat(OpcodeStr, "\t$dst, $src2, $src3[$lane]"), "$src1 = $dst",
811 [(set (ResTy QPR:$dst),
812 (ResTy (ShOp (ResTy QPR:$src1),
813 (ResTy (MulOp QPR:$src2,
814 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src3),
816 class N3VQMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
817 string OpcodeStr, ValueType ResTy, ValueType OpTy,
818 SDNode MulOp, SDNode ShOp>
819 : N3V<1, 1, op21_20, op11_8, 1, 0,
821 (ins QPR:$src1, QPR:$src2, DPR_8:$src3, nohash_imm:$lane), itin,
822 !strconcat(OpcodeStr, "\t$dst, $src2, $src3[$lane]"), "$src1 = $dst",
823 [(set (ResTy QPR:$dst),
824 (ResTy (ShOp (ResTy QPR:$src1),
825 (ResTy (MulOp QPR:$src2,
826 (ResTy (NEONvduplane (OpTy DPR_8:$src3),
829 // Multiply-Add/Sub operations, scalar single-precision
830 class N3VDMulOps<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
831 InstrItinClass itin, string OpcodeStr,
832 ValueType Ty, SDNode MulOp, SDNode OpNode>
833 : N3V<op24, op23, op21_20, op11_8, 0, op4,
834 (outs DPR_VFP2:$dst),
835 (ins DPR_VFP2:$src1, DPR_VFP2:$src2, DPR_VFP2:$src3), itin,
836 !strconcat(OpcodeStr, "\t$dst, $src2, $src3"), "$src1 = $dst", []>;
838 class N3VDMulOpsPat<SDNode MulNode, SDNode OpNode, NeonI Inst>
839 : NEONFPPat<(f32 (OpNode SPR:$acc, (f32 (MulNode SPR:$a, SPR:$b)))),
841 (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$acc, arm_ssubreg_0),
842 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$a, arm_ssubreg_0),
843 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$b, arm_ssubreg_0)),
846 // Neon 3-argument intrinsics, both double- and quad-register.
847 // The destination register is also used as the first source operand register.
848 class N3VDInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
849 InstrItinClass itin, string OpcodeStr,
850 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
851 : N3V<op24, op23, op21_20, op11_8, 0, op4,
852 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3), itin,
853 !strconcat(OpcodeStr, "\t$dst, $src2, $src3"), "$src1 = $dst",
854 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src1),
855 (OpTy DPR:$src2), (OpTy DPR:$src3))))]>;
856 class N3VQInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
857 InstrItinClass itin, string OpcodeStr,
858 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
859 : N3V<op24, op23, op21_20, op11_8, 1, op4,
860 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3), itin,
861 !strconcat(OpcodeStr, "\t$dst, $src2, $src3"), "$src1 = $dst",
862 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src1),
863 (OpTy QPR:$src2), (OpTy QPR:$src3))))]>;
865 // Neon Long 3-argument intrinsic. The destination register is
866 // a quad-register and is also used as the first source operand register.
867 class N3VLInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
868 InstrItinClass itin, string OpcodeStr,
869 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
870 : N3V<op24, op23, op21_20, op11_8, 0, op4,
871 (outs QPR:$dst), (ins QPR:$src1, DPR:$src2, DPR:$src3), itin,
872 !strconcat(OpcodeStr, "\t$dst, $src2, $src3"), "$src1 = $dst",
874 (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$src2), (TyD DPR:$src3))))]>;
875 class N3VLInt3SL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
876 string OpcodeStr, ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
877 : N3V<op24, 1, op21_20, op11_8, 1, 0,
879 (ins QPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane), itin,
880 !strconcat(OpcodeStr, "\t$dst, $src2, $src3[$lane]"), "$src1 = $dst",
881 [(set (ResTy QPR:$dst),
882 (ResTy (IntOp (ResTy QPR:$src1),
884 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src3),
886 class N3VLInt3SL16<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
887 string OpcodeStr, ValueType ResTy, ValueType OpTy,
889 : N3V<op24, 1, op21_20, op11_8, 1, 0,
891 (ins QPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane), itin,
892 !strconcat(OpcodeStr, "\t$dst, $src2, $src3[$lane]"), "$src1 = $dst",
893 [(set (ResTy QPR:$dst),
894 (ResTy (IntOp (ResTy QPR:$src1),
896 (OpTy (NEONvduplane (OpTy DPR_8:$src3),
900 // Narrowing 3-register intrinsics.
901 class N3VNInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
902 string OpcodeStr, ValueType TyD, ValueType TyQ,
903 Intrinsic IntOp, bit Commutable>
904 : N3V<op24, op23, op21_20, op11_8, 0, op4,
905 (outs DPR:$dst), (ins QPR:$src1, QPR:$src2), IIC_VBINi4D,
906 !strconcat(OpcodeStr, "\t$dst, $src1, $src2"), "",
907 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src1), (TyQ QPR:$src2))))]> {
908 let isCommutable = Commutable;
911 // Long 3-register intrinsics.
912 class N3VLInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
913 InstrItinClass itin, string OpcodeStr, ValueType TyQ, ValueType TyD,
914 Intrinsic IntOp, bit Commutable>
915 : N3V<op24, op23, op21_20, op11_8, 0, op4,
916 (outs QPR:$dst), (ins DPR:$src1, DPR:$src2), itin,
917 !strconcat(OpcodeStr, "\t$dst, $src1, $src2"), "",
918 [(set QPR:$dst, (TyQ (IntOp (TyD DPR:$src1), (TyD DPR:$src2))))]> {
919 let isCommutable = Commutable;
921 class N3VLIntSL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
922 string OpcodeStr, ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
923 : N3V<op24, 1, op21_20, op11_8, 1, 0,
924 (outs QPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
925 itin, !strconcat(OpcodeStr, "\t$dst, $src1, $src2[$lane]"), "",
926 [(set (ResTy QPR:$dst),
927 (ResTy (IntOp (OpTy DPR:$src1),
928 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src2),
930 class N3VLIntSL16<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
931 string OpcodeStr, ValueType ResTy, ValueType OpTy,
933 : N3V<op24, 1, op21_20, op11_8, 1, 0,
934 (outs QPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
935 itin, !strconcat(OpcodeStr, "\t$dst, $src1, $src2[$lane]"), "",
936 [(set (ResTy QPR:$dst),
937 (ResTy (IntOp (OpTy DPR:$src1),
938 (OpTy (NEONvduplane (OpTy DPR_8:$src2),
941 // Wide 3-register intrinsics.
942 class N3VWInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
943 string OpcodeStr, ValueType TyQ, ValueType TyD,
944 Intrinsic IntOp, bit Commutable>
945 : N3V<op24, op23, op21_20, op11_8, 0, op4,
946 (outs QPR:$dst), (ins QPR:$src1, DPR:$src2), IIC_VSUBiD,
947 !strconcat(OpcodeStr, "\t$dst, $src1, $src2"), "",
948 [(set QPR:$dst, (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$src2))))]> {
949 let isCommutable = Commutable;
952 // Pairwise long 2-register intrinsics, both double- and quad-register.
953 class N2VDPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
954 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
955 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
956 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
957 (ins DPR:$src), IIC_VSHLiD, !strconcat(OpcodeStr, "\t$dst, $src"), "",
958 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
959 class N2VQPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
960 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
961 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
962 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
963 (ins QPR:$src), IIC_VSHLiD, !strconcat(OpcodeStr, "\t$dst, $src"), "",
964 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
966 // Pairwise long 2-register accumulate intrinsics,
967 // both double- and quad-register.
968 // The destination register is also used as the first source operand register.
969 class N2VDPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
970 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
971 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
972 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
973 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), IIC_VPALiD,
974 !strconcat(OpcodeStr, "\t$dst, $src2"), "$src1 = $dst",
975 [(set DPR:$dst, (ResTy (IntOp (ResTy DPR:$src1), (OpTy DPR:$src2))))]>;
976 class N2VQPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
977 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
978 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
979 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4,
980 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), IIC_VPALiQ,
981 !strconcat(OpcodeStr, "\t$dst, $src2"), "$src1 = $dst",
982 [(set QPR:$dst, (ResTy (IntOp (ResTy QPR:$src1), (OpTy QPR:$src2))))]>;
984 // Shift by immediate,
985 // both double- and quad-register.
986 class N2VDSh<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
987 bit op4, InstrItinClass itin, string OpcodeStr,
988 ValueType Ty, SDNode OpNode>
989 : N2VImm<op24, op23, op21_16, op11_8, op7, 0, op4,
990 (outs DPR:$dst), (ins DPR:$src, i32imm:$SIMM), itin,
991 !strconcat(OpcodeStr, "\t$dst, $src, $SIMM"), "",
992 [(set DPR:$dst, (Ty (OpNode (Ty DPR:$src), (i32 imm:$SIMM))))]>;
993 class N2VQSh<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
994 bit op4, InstrItinClass itin, string OpcodeStr,
995 ValueType Ty, SDNode OpNode>
996 : N2VImm<op24, op23, op21_16, op11_8, op7, 1, op4,
997 (outs QPR:$dst), (ins QPR:$src, i32imm:$SIMM), itin,
998 !strconcat(OpcodeStr, "\t$dst, $src, $SIMM"), "",
999 [(set QPR:$dst, (Ty (OpNode (Ty QPR:$src), (i32 imm:$SIMM))))]>;
1001 // Long shift by immediate.
1002 class N2VLSh<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
1003 bit op6, bit op4, string OpcodeStr, ValueType ResTy,
1004 ValueType OpTy, SDNode OpNode>
1005 : N2VImm<op24, op23, op21_16, op11_8, op7, op6, op4,
1006 (outs QPR:$dst), (ins DPR:$src, i32imm:$SIMM), IIC_VSHLiD,
1007 !strconcat(OpcodeStr, "\t$dst, $src, $SIMM"), "",
1008 [(set QPR:$dst, (ResTy (OpNode (OpTy DPR:$src),
1009 (i32 imm:$SIMM))))]>;
1011 // Narrow shift by immediate.
1012 class N2VNSh<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
1013 bit op6, bit op4, InstrItinClass itin, string OpcodeStr,
1014 ValueType ResTy, ValueType OpTy, SDNode OpNode>
1015 : N2VImm<op24, op23, op21_16, op11_8, op7, op6, op4,
1016 (outs DPR:$dst), (ins QPR:$src, i32imm:$SIMM), itin,
1017 !strconcat(OpcodeStr, "\t$dst, $src, $SIMM"), "",
1018 [(set DPR:$dst, (ResTy (OpNode (OpTy QPR:$src),
1019 (i32 imm:$SIMM))))]>;
1021 // Shift right by immediate and accumulate,
1022 // both double- and quad-register.
1023 class N2VDShAdd<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
1024 bit op4, string OpcodeStr, ValueType Ty, SDNode ShOp>
1025 : N2VImm<op24, op23, op21_16, op11_8, op7, 0, op4,
1026 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, i32imm:$SIMM),
1028 !strconcat(OpcodeStr, "\t$dst, $src2, $SIMM"), "$src1 = $dst",
1029 [(set DPR:$dst, (Ty (add DPR:$src1,
1030 (Ty (ShOp DPR:$src2, (i32 imm:$SIMM))))))]>;
1031 class N2VQShAdd<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
1032 bit op4, string OpcodeStr, ValueType Ty, SDNode ShOp>
1033 : N2VImm<op24, op23, op21_16, op11_8, op7, 1, op4,
1034 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, i32imm:$SIMM),
1036 !strconcat(OpcodeStr, "\t$dst, $src2, $SIMM"), "$src1 = $dst",
1037 [(set QPR:$dst, (Ty (add QPR:$src1,
1038 (Ty (ShOp QPR:$src2, (i32 imm:$SIMM))))))]>;
1040 // Shift by immediate and insert,
1041 // both double- and quad-register.
1042 class N2VDShIns<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
1043 bit op4, string OpcodeStr, ValueType Ty, SDNode ShOp>
1044 : N2VImm<op24, op23, op21_16, op11_8, op7, 0, op4,
1045 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, i32imm:$SIMM),
1047 !strconcat(OpcodeStr, "\t$dst, $src2, $SIMM"), "$src1 = $dst",
1048 [(set DPR:$dst, (Ty (ShOp DPR:$src1, DPR:$src2, (i32 imm:$SIMM))))]>;
1049 class N2VQShIns<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
1050 bit op4, string OpcodeStr, ValueType Ty, SDNode ShOp>
1051 : N2VImm<op24, op23, op21_16, op11_8, op7, 1, op4,
1052 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, i32imm:$SIMM),
1054 !strconcat(OpcodeStr, "\t$dst, $src2, $SIMM"), "$src1 = $dst",
1055 [(set QPR:$dst, (Ty (ShOp QPR:$src1, QPR:$src2, (i32 imm:$SIMM))))]>;
1057 // Convert, with fractional bits immediate,
1058 // both double- and quad-register.
1059 class N2VCvtD<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
1060 bit op4, string OpcodeStr, ValueType ResTy, ValueType OpTy,
1062 : N2VImm<op24, op23, op21_16, op11_8, op7, 0, op4,
1063 (outs DPR:$dst), (ins DPR:$src, i32imm:$SIMM), IIC_VUNAD,
1064 !strconcat(OpcodeStr, "\t$dst, $src, $SIMM"), "",
1065 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src), (i32 imm:$SIMM))))]>;
1066 class N2VCvtQ<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
1067 bit op4, string OpcodeStr, ValueType ResTy, ValueType OpTy,
1069 : N2VImm<op24, op23, op21_16, op11_8, op7, 1, op4,
1070 (outs QPR:$dst), (ins QPR:$src, i32imm:$SIMM), IIC_VUNAQ,
1071 !strconcat(OpcodeStr, "\t$dst, $src, $SIMM"), "",
1072 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src), (i32 imm:$SIMM))))]>;
1074 //===----------------------------------------------------------------------===//
1076 //===----------------------------------------------------------------------===//
1078 // Abbreviations used in multiclass suffixes:
1079 // Q = quarter int (8 bit) elements
1080 // H = half int (16 bit) elements
1081 // S = single int (32 bit) elements
1082 // D = double int (64 bit) elements
1084 // Neon 3-register vector operations.
1086 // First with only element sizes of 8, 16 and 32 bits:
1087 multiclass N3V_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1088 InstrItinClass itinD16, InstrItinClass itinD32,
1089 InstrItinClass itinQ16, InstrItinClass itinQ32,
1090 string OpcodeStr, SDNode OpNode, bit Commutable = 0> {
1091 // 64-bit vector types.
1092 def v8i8 : N3VD<op24, op23, 0b00, op11_8, op4, itinD16,
1093 !strconcat(OpcodeStr, "8"), v8i8, v8i8, OpNode, Commutable>;
1094 def v4i16 : N3VD<op24, op23, 0b01, op11_8, op4, itinD16,
1095 !strconcat(OpcodeStr, "16"), v4i16, v4i16, OpNode, Commutable>;
1096 def v2i32 : N3VD<op24, op23, 0b10, op11_8, op4, itinD32,
1097 !strconcat(OpcodeStr, "32"), v2i32, v2i32, OpNode, Commutable>;
1099 // 128-bit vector types.
1100 def v16i8 : N3VQ<op24, op23, 0b00, op11_8, op4, itinQ16,
1101 !strconcat(OpcodeStr, "8"), v16i8, v16i8, OpNode, Commutable>;
1102 def v8i16 : N3VQ<op24, op23, 0b01, op11_8, op4, itinQ16,
1103 !strconcat(OpcodeStr, "16"), v8i16, v8i16, OpNode, Commutable>;
1104 def v4i32 : N3VQ<op24, op23, 0b10, op11_8, op4, itinQ32,
1105 !strconcat(OpcodeStr, "32"), v4i32, v4i32, OpNode, Commutable>;
1108 multiclass N3VSL_HS<bits<4> op11_8, string OpcodeStr, SDNode ShOp> {
1109 def v4i16 : N3VDSL16<0b01, op11_8, !strconcat(OpcodeStr, "16"), v4i16, ShOp>;
1110 def v2i32 : N3VDSL<0b10, op11_8, IIC_VMULi32D, !strconcat(OpcodeStr, "32"), v2i32, ShOp>;
1111 def v8i16 : N3VQSL16<0b01, op11_8, !strconcat(OpcodeStr, "16"), v8i16, v4i16, ShOp>;
1112 def v4i32 : N3VQSL<0b10, op11_8, IIC_VMULi32Q, !strconcat(OpcodeStr, "32"), v4i32, v2i32, ShOp>;
1115 // ....then also with element size 64 bits:
1116 multiclass N3V_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1117 InstrItinClass itinD, InstrItinClass itinQ,
1118 string OpcodeStr, SDNode OpNode, bit Commutable = 0>
1119 : N3V_QHS<op24, op23, op11_8, op4, itinD, itinD, itinQ, itinQ,
1120 OpcodeStr, OpNode, Commutable> {
1121 def v1i64 : N3VD<op24, op23, 0b11, op11_8, op4, itinD,
1122 !strconcat(OpcodeStr, "64"), v1i64, v1i64, OpNode, Commutable>;
1123 def v2i64 : N3VQ<op24, op23, 0b11, op11_8, op4, itinQ,
1124 !strconcat(OpcodeStr, "64"), v2i64, v2i64, OpNode, Commutable>;
1128 // Neon Narrowing 2-register vector intrinsics,
1129 // source operand element sizes of 16, 32 and 64 bits:
1130 multiclass N2VNInt_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1131 bits<5> op11_7, bit op6, bit op4,
1132 InstrItinClass itin, string OpcodeStr,
1134 def v8i8 : N2VNInt<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
1135 itin, !strconcat(OpcodeStr, "16"), v8i8, v8i16, IntOp>;
1136 def v4i16 : N2VNInt<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
1137 itin, !strconcat(OpcodeStr, "32"), v4i16, v4i32, IntOp>;
1138 def v2i32 : N2VNInt<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
1139 itin, !strconcat(OpcodeStr, "64"), v2i32, v2i64, IntOp>;
1143 // Neon Lengthening 2-register vector intrinsic (currently specific to VMOVL).
1144 // source operand element sizes of 16, 32 and 64 bits:
1145 multiclass N2VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
1146 bit op4, string OpcodeStr, Intrinsic IntOp> {
1147 def v8i16 : N2VLInt<op24, op23, 0b001000, op11_8, op7, op6, op4,
1148 IIC_VQUNAiD, !strconcat(OpcodeStr, "8"), v8i16, v8i8, IntOp>;
1149 def v4i32 : N2VLInt<op24, op23, 0b010000, op11_8, op7, op6, op4,
1150 IIC_VQUNAiD, !strconcat(OpcodeStr, "16"), v4i32, v4i16, IntOp>;
1151 def v2i64 : N2VLInt<op24, op23, 0b100000, op11_8, op7, op6, op4,
1152 IIC_VQUNAiD, !strconcat(OpcodeStr, "32"), v2i64, v2i32, IntOp>;
1156 // Neon 3-register vector intrinsics.
1158 // First with only element sizes of 16 and 32 bits:
1159 multiclass N3VInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
1160 InstrItinClass itinD16, InstrItinClass itinD32,
1161 InstrItinClass itinQ16, InstrItinClass itinQ32,
1162 string OpcodeStr, Intrinsic IntOp, bit Commutable = 0> {
1163 // 64-bit vector types.
1164 def v4i16 : N3VDInt<op24, op23, 0b01, op11_8, op4, itinD16, !strconcat(OpcodeStr,"16"),
1165 v4i16, v4i16, IntOp, Commutable>;
1166 def v2i32 : N3VDInt<op24, op23, 0b10, op11_8, op4, itinD32, !strconcat(OpcodeStr,"32"),
1167 v2i32, v2i32, IntOp, Commutable>;
1169 // 128-bit vector types.
1170 def v8i16 : N3VQInt<op24, op23, 0b01, op11_8, op4, itinQ16, !strconcat(OpcodeStr,"16"),
1171 v8i16, v8i16, IntOp, Commutable>;
1172 def v4i32 : N3VQInt<op24, op23, 0b10, op11_8, op4, itinQ32, !strconcat(OpcodeStr,"32"),
1173 v4i32, v4i32, IntOp, Commutable>;
1176 multiclass N3VIntSL_HS<bits<4> op11_8,
1177 InstrItinClass itinD16, InstrItinClass itinD32,
1178 InstrItinClass itinQ16, InstrItinClass itinQ32,
1179 string OpcodeStr, Intrinsic IntOp> {
1180 def v4i16 : N3VDIntSL16<0b01, op11_8, itinD16, !strconcat(OpcodeStr, "16"), v4i16, IntOp>;
1181 def v2i32 : N3VDIntSL<0b10, op11_8, itinD32, !strconcat(OpcodeStr, "32"), v2i32, IntOp>;
1182 def v8i16 : N3VQIntSL16<0b01, op11_8, itinQ16, !strconcat(OpcodeStr, "16"), v8i16, v4i16, IntOp>;
1183 def v4i32 : N3VQIntSL<0b10, op11_8, itinQ32, !strconcat(OpcodeStr, "32"), v4i32, v2i32, IntOp>;
1186 // ....then also with element size of 8 bits:
1187 multiclass N3VInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1188 InstrItinClass itinD16, InstrItinClass itinD32,
1189 InstrItinClass itinQ16, InstrItinClass itinQ32,
1190 string OpcodeStr, Intrinsic IntOp, bit Commutable = 0>
1191 : N3VInt_HS<op24, op23, op11_8, op4, itinD16, itinD32, itinQ16, itinQ32,
1192 OpcodeStr, IntOp, Commutable> {
1193 def v8i8 : N3VDInt<op24, op23, 0b00, op11_8, op4, itinD16,
1194 !strconcat(OpcodeStr, "8"), v8i8, v8i8, IntOp, Commutable>;
1195 def v16i8 : N3VQInt<op24, op23, 0b00, op11_8, op4, itinQ16,
1196 !strconcat(OpcodeStr, "8"), v16i8, v16i8, IntOp, Commutable>;
1199 // ....then also with element size of 64 bits:
1200 multiclass N3VInt_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1201 InstrItinClass itinD16, InstrItinClass itinD32,
1202 InstrItinClass itinQ16, InstrItinClass itinQ32,
1203 string OpcodeStr, Intrinsic IntOp, bit Commutable = 0>
1204 : N3VInt_QHS<op24, op23, op11_8, op4, itinD16, itinD32, itinQ16, itinQ32,
1205 OpcodeStr, IntOp, Commutable> {
1206 def v1i64 : N3VDInt<op24, op23, 0b11, op11_8, op4, itinD32,
1207 !strconcat(OpcodeStr,"64"), v1i64, v1i64, IntOp, Commutable>;
1208 def v2i64 : N3VQInt<op24, op23, 0b11, op11_8, op4, itinQ32,
1209 !strconcat(OpcodeStr,"64"), v2i64, v2i64, IntOp, Commutable>;
1213 // Neon Narrowing 3-register vector intrinsics,
1214 // source operand element sizes of 16, 32 and 64 bits:
1215 multiclass N3VNInt_HSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1216 string OpcodeStr, Intrinsic IntOp, bit Commutable = 0> {
1217 def v8i8 : N3VNInt<op24, op23, 0b00, op11_8, op4, !strconcat(OpcodeStr,"16"),
1218 v8i8, v8i16, IntOp, Commutable>;
1219 def v4i16 : N3VNInt<op24, op23, 0b01, op11_8, op4, !strconcat(OpcodeStr,"32"),
1220 v4i16, v4i32, IntOp, Commutable>;
1221 def v2i32 : N3VNInt<op24, op23, 0b10, op11_8, op4, !strconcat(OpcodeStr,"64"),
1222 v2i32, v2i64, IntOp, Commutable>;
1226 // Neon Long 3-register vector intrinsics.
1228 // First with only element sizes of 16 and 32 bits:
1229 multiclass N3VLInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
1230 InstrItinClass itin, string OpcodeStr,
1231 Intrinsic IntOp, bit Commutable = 0> {
1232 def v4i32 : N3VLInt<op24, op23, 0b01, op11_8, op4, itin,
1233 !strconcat(OpcodeStr,"16"), v4i32, v4i16, IntOp, Commutable>;
1234 def v2i64 : N3VLInt<op24, op23, 0b10, op11_8, op4, itin,
1235 !strconcat(OpcodeStr,"32"), v2i64, v2i32, IntOp, Commutable>;
1238 multiclass N3VLIntSL_HS<bit op24, bits<4> op11_8,
1239 InstrItinClass itin, string OpcodeStr, Intrinsic IntOp> {
1240 def v4i16 : N3VLIntSL16<op24, 0b01, op11_8, itin,
1241 !strconcat(OpcodeStr, "16"), v4i32, v4i16, IntOp>;
1242 def v2i32 : N3VLIntSL<op24, 0b10, op11_8, itin,
1243 !strconcat(OpcodeStr, "32"), v2i64, v2i32, IntOp>;
1246 // ....then also with element size of 8 bits:
1247 multiclass N3VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1248 InstrItinClass itin, string OpcodeStr,
1249 Intrinsic IntOp, bit Commutable = 0>
1250 : N3VLInt_HS<op24, op23, op11_8, op4, itin, OpcodeStr, IntOp, Commutable> {
1251 def v8i16 : N3VLInt<op24, op23, 0b00, op11_8, op4, itin,
1252 !strconcat(OpcodeStr, "8"), v8i16, v8i8, IntOp, Commutable>;
1256 // Neon Wide 3-register vector intrinsics,
1257 // source operand element sizes of 8, 16 and 32 bits:
1258 multiclass N3VWInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1259 string OpcodeStr, Intrinsic IntOp, bit Commutable = 0> {
1260 def v8i16 : N3VWInt<op24, op23, 0b00, op11_8, op4, !strconcat(OpcodeStr, "8"),
1261 v8i16, v8i8, IntOp, Commutable>;
1262 def v4i32 : N3VWInt<op24, op23, 0b01, op11_8, op4, !strconcat(OpcodeStr,"16"),
1263 v4i32, v4i16, IntOp, Commutable>;
1264 def v2i64 : N3VWInt<op24, op23, 0b10, op11_8, op4, !strconcat(OpcodeStr,"32"),
1265 v2i64, v2i32, IntOp, Commutable>;
1269 // Neon Multiply-Op vector operations,
1270 // element sizes of 8, 16 and 32 bits:
1271 multiclass N3VMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1272 InstrItinClass itinD16, InstrItinClass itinD32,
1273 InstrItinClass itinQ16, InstrItinClass itinQ32,
1274 string OpcodeStr, SDNode OpNode> {
1275 // 64-bit vector types.
1276 def v8i8 : N3VDMulOp<op24, op23, 0b00, op11_8, op4, itinD16,
1277 !strconcat(OpcodeStr, "8"), v8i8, mul, OpNode>;
1278 def v4i16 : N3VDMulOp<op24, op23, 0b01, op11_8, op4, itinD16,
1279 !strconcat(OpcodeStr, "16"), v4i16, mul, OpNode>;
1280 def v2i32 : N3VDMulOp<op24, op23, 0b10, op11_8, op4, itinD32,
1281 !strconcat(OpcodeStr, "32"), v2i32, mul, OpNode>;
1283 // 128-bit vector types.
1284 def v16i8 : N3VQMulOp<op24, op23, 0b00, op11_8, op4, itinQ16,
1285 !strconcat(OpcodeStr, "8"), v16i8, mul, OpNode>;
1286 def v8i16 : N3VQMulOp<op24, op23, 0b01, op11_8, op4, itinQ16,
1287 !strconcat(OpcodeStr, "16"), v8i16, mul, OpNode>;
1288 def v4i32 : N3VQMulOp<op24, op23, 0b10, op11_8, op4, itinQ32,
1289 !strconcat(OpcodeStr, "32"), v4i32, mul, OpNode>;
1292 multiclass N3VMulOpSL_HS<bits<4> op11_8,
1293 InstrItinClass itinD16, InstrItinClass itinD32,
1294 InstrItinClass itinQ16, InstrItinClass itinQ32,
1295 string OpcodeStr, SDNode ShOp> {
1296 def v4i16 : N3VDMulOpSL16<0b01, op11_8, itinD16,
1297 !strconcat(OpcodeStr, "16"), v4i16, mul, ShOp>;
1298 def v2i32 : N3VDMulOpSL<0b10, op11_8, itinD32,
1299 !strconcat(OpcodeStr, "32"), v2i32, mul, ShOp>;
1300 def v8i16 : N3VQMulOpSL16<0b01, op11_8, itinQ16,
1301 !strconcat(OpcodeStr, "16"), v8i16, v4i16, mul, ShOp>;
1302 def v4i32 : N3VQMulOpSL<0b10, op11_8, itinQ32,
1303 !strconcat(OpcodeStr, "32"), v4i32, v2i32, mul, ShOp>;
1306 // Neon 3-argument intrinsics,
1307 // element sizes of 8, 16 and 32 bits:
1308 multiclass N3VInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1309 string OpcodeStr, Intrinsic IntOp> {
1310 // 64-bit vector types.
1311 def v8i8 : N3VDInt3<op24, op23, 0b00, op11_8, op4, IIC_VMACi16D,
1312 !strconcat(OpcodeStr, "8"), v8i8, v8i8, IntOp>;
1313 def v4i16 : N3VDInt3<op24, op23, 0b01, op11_8, op4, IIC_VMACi16D,
1314 !strconcat(OpcodeStr, "16"), v4i16, v4i16, IntOp>;
1315 def v2i32 : N3VDInt3<op24, op23, 0b10, op11_8, op4, IIC_VMACi32D,
1316 !strconcat(OpcodeStr, "32"), v2i32, v2i32, IntOp>;
1318 // 128-bit vector types.
1319 def v16i8 : N3VQInt3<op24, op23, 0b00, op11_8, op4, IIC_VMACi16Q,
1320 !strconcat(OpcodeStr, "8"), v16i8, v16i8, IntOp>;
1321 def v8i16 : N3VQInt3<op24, op23, 0b01, op11_8, op4, IIC_VMACi16Q,
1322 !strconcat(OpcodeStr, "16"), v8i16, v8i16, IntOp>;
1323 def v4i32 : N3VQInt3<op24, op23, 0b10, op11_8, op4, IIC_VMACi32Q,
1324 !strconcat(OpcodeStr, "32"), v4i32, v4i32, IntOp>;
1328 // Neon Long 3-argument intrinsics.
1330 // First with only element sizes of 16 and 32 bits:
1331 multiclass N3VLInt3_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
1332 string OpcodeStr, Intrinsic IntOp> {
1333 def v4i32 : N3VLInt3<op24, op23, 0b01, op11_8, op4, IIC_VMACi16D,
1334 !strconcat(OpcodeStr, "16"), v4i32, v4i16, IntOp>;
1335 def v2i64 : N3VLInt3<op24, op23, 0b10, op11_8, op4, IIC_VMACi16D,
1336 !strconcat(OpcodeStr, "32"), v2i64, v2i32, IntOp>;
1339 multiclass N3VLInt3SL_HS<bit op24, bits<4> op11_8,
1340 string OpcodeStr, Intrinsic IntOp> {
1341 def v4i16 : N3VLInt3SL16<op24, 0b01, op11_8, IIC_VMACi16D,
1342 !strconcat(OpcodeStr, "16"), v4i32, v4i16, IntOp>;
1343 def v2i32 : N3VLInt3SL<op24, 0b10, op11_8, IIC_VMACi32D,
1344 !strconcat(OpcodeStr, "32"), v2i64, v2i32, IntOp>;
1347 // ....then also with element size of 8 bits:
1348 multiclass N3VLInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1349 string OpcodeStr, Intrinsic IntOp>
1350 : N3VLInt3_HS<op24, op23, op11_8, op4, OpcodeStr, IntOp> {
1351 def v8i16 : N3VLInt3<op24, op23, 0b01, op11_8, op4, IIC_VMACi16D,
1352 !strconcat(OpcodeStr, "8"), v8i16, v8i8, IntOp>;
1356 // Neon 2-register vector intrinsics,
1357 // element sizes of 8, 16 and 32 bits:
1358 multiclass N2VInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1359 bits<5> op11_7, bit op4,
1360 InstrItinClass itinD, InstrItinClass itinQ,
1361 string OpcodeStr, Intrinsic IntOp> {
1362 // 64-bit vector types.
1363 def v8i8 : N2VDInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1364 itinD, !strconcat(OpcodeStr, "8"), v8i8, v8i8, IntOp>;
1365 def v4i16 : N2VDInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1366 itinD, !strconcat(OpcodeStr, "16"), v4i16, v4i16, IntOp>;
1367 def v2i32 : N2VDInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1368 itinD, !strconcat(OpcodeStr, "32"), v2i32, v2i32, IntOp>;
1370 // 128-bit vector types.
1371 def v16i8 : N2VQInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1372 itinQ, !strconcat(OpcodeStr, "8"), v16i8, v16i8, IntOp>;
1373 def v8i16 : N2VQInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1374 itinQ, !strconcat(OpcodeStr, "16"), v8i16, v8i16, IntOp>;
1375 def v4i32 : N2VQInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1376 itinQ, !strconcat(OpcodeStr, "32"), v4i32, v4i32, IntOp>;
1380 // Neon Pairwise long 2-register intrinsics,
1381 // element sizes of 8, 16 and 32 bits:
1382 multiclass N2VPLInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1383 bits<5> op11_7, bit op4,
1384 string OpcodeStr, Intrinsic IntOp> {
1385 // 64-bit vector types.
1386 def v8i8 : N2VDPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1387 !strconcat(OpcodeStr, "8"), v4i16, v8i8, IntOp>;
1388 def v4i16 : N2VDPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1389 !strconcat(OpcodeStr, "16"), v2i32, v4i16, IntOp>;
1390 def v2i32 : N2VDPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1391 !strconcat(OpcodeStr, "32"), v1i64, v2i32, IntOp>;
1393 // 128-bit vector types.
1394 def v16i8 : N2VQPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1395 !strconcat(OpcodeStr, "8"), v8i16, v16i8, IntOp>;
1396 def v8i16 : N2VQPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1397 !strconcat(OpcodeStr, "16"), v4i32, v8i16, IntOp>;
1398 def v4i32 : N2VQPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1399 !strconcat(OpcodeStr, "32"), v2i64, v4i32, IntOp>;
1403 // Neon Pairwise long 2-register accumulate intrinsics,
1404 // element sizes of 8, 16 and 32 bits:
1405 multiclass N2VPLInt2_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1406 bits<5> op11_7, bit op4,
1407 string OpcodeStr, Intrinsic IntOp> {
1408 // 64-bit vector types.
1409 def v8i8 : N2VDPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1410 !strconcat(OpcodeStr, "8"), v4i16, v8i8, IntOp>;
1411 def v4i16 : N2VDPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1412 !strconcat(OpcodeStr, "16"), v2i32, v4i16, IntOp>;
1413 def v2i32 : N2VDPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1414 !strconcat(OpcodeStr, "32"), v1i64, v2i32, IntOp>;
1416 // 128-bit vector types.
1417 def v16i8 : N2VQPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1418 !strconcat(OpcodeStr, "8"), v8i16, v16i8, IntOp>;
1419 def v8i16 : N2VQPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1420 !strconcat(OpcodeStr, "16"), v4i32, v8i16, IntOp>;
1421 def v4i32 : N2VQPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1422 !strconcat(OpcodeStr, "32"), v2i64, v4i32, IntOp>;
1426 // Neon 2-register vector shift by immediate,
1427 // element sizes of 8, 16, 32 and 64 bits:
1428 multiclass N2VSh_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1429 InstrItinClass itin, string OpcodeStr, SDNode OpNode> {
1430 // 64-bit vector types.
1431 def v8i8 : N2VDSh<op24, op23, 0b001000, op11_8, 0, op4, itin,
1432 !strconcat(OpcodeStr, "8"), v8i8, OpNode>;
1433 def v4i16 : N2VDSh<op24, op23, 0b010000, op11_8, 0, op4, itin,
1434 !strconcat(OpcodeStr, "16"), v4i16, OpNode>;
1435 def v2i32 : N2VDSh<op24, op23, 0b100000, op11_8, 0, op4, itin,
1436 !strconcat(OpcodeStr, "32"), v2i32, OpNode>;
1437 def v1i64 : N2VDSh<op24, op23, 0b000000, op11_8, 1, op4, itin,
1438 !strconcat(OpcodeStr, "64"), v1i64, OpNode>;
1440 // 128-bit vector types.
1441 def v16i8 : N2VQSh<op24, op23, 0b001000, op11_8, 0, op4, itin,
1442 !strconcat(OpcodeStr, "8"), v16i8, OpNode>;
1443 def v8i16 : N2VQSh<op24, op23, 0b010000, op11_8, 0, op4, itin,
1444 !strconcat(OpcodeStr, "16"), v8i16, OpNode>;
1445 def v4i32 : N2VQSh<op24, op23, 0b100000, op11_8, 0, op4, itin,
1446 !strconcat(OpcodeStr, "32"), v4i32, OpNode>;
1447 def v2i64 : N2VQSh<op24, op23, 0b000000, op11_8, 1, op4, itin,
1448 !strconcat(OpcodeStr, "64"), v2i64, OpNode>;
1452 // Neon Shift-Accumulate vector operations,
1453 // element sizes of 8, 16, 32 and 64 bits:
1454 multiclass N2VShAdd_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1455 string OpcodeStr, SDNode ShOp> {
1456 // 64-bit vector types.
1457 def v8i8 : N2VDShAdd<op24, op23, 0b001000, op11_8, 0, op4,
1458 !strconcat(OpcodeStr, "8"), v8i8, ShOp>;
1459 def v4i16 : N2VDShAdd<op24, op23, 0b010000, op11_8, 0, op4,
1460 !strconcat(OpcodeStr, "16"), v4i16, ShOp>;
1461 def v2i32 : N2VDShAdd<op24, op23, 0b100000, op11_8, 0, op4,
1462 !strconcat(OpcodeStr, "32"), v2i32, ShOp>;
1463 def v1i64 : N2VDShAdd<op24, op23, 0b000000, op11_8, 1, op4,
1464 !strconcat(OpcodeStr, "64"), v1i64, ShOp>;
1466 // 128-bit vector types.
1467 def v16i8 : N2VQShAdd<op24, op23, 0b001000, op11_8, 0, op4,
1468 !strconcat(OpcodeStr, "8"), v16i8, ShOp>;
1469 def v8i16 : N2VQShAdd<op24, op23, 0b010000, op11_8, 0, op4,
1470 !strconcat(OpcodeStr, "16"), v8i16, ShOp>;
1471 def v4i32 : N2VQShAdd<op24, op23, 0b100000, op11_8, 0, op4,
1472 !strconcat(OpcodeStr, "32"), v4i32, ShOp>;
1473 def v2i64 : N2VQShAdd<op24, op23, 0b000000, op11_8, 1, op4,
1474 !strconcat(OpcodeStr, "64"), v2i64, ShOp>;
1478 // Neon Shift-Insert vector operations,
1479 // element sizes of 8, 16, 32 and 64 bits:
1480 multiclass N2VShIns_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1481 string OpcodeStr, SDNode ShOp> {
1482 // 64-bit vector types.
1483 def v8i8 : N2VDShIns<op24, op23, 0b001000, op11_8, 0, op4,
1484 !strconcat(OpcodeStr, "8"), v8i8, ShOp>;
1485 def v4i16 : N2VDShIns<op24, op23, 0b010000, op11_8, 0, op4,
1486 !strconcat(OpcodeStr, "16"), v4i16, ShOp>;
1487 def v2i32 : N2VDShIns<op24, op23, 0b100000, op11_8, 0, op4,
1488 !strconcat(OpcodeStr, "32"), v2i32, ShOp>;
1489 def v1i64 : N2VDShIns<op24, op23, 0b000000, op11_8, 1, op4,
1490 !strconcat(OpcodeStr, "64"), v1i64, ShOp>;
1492 // 128-bit vector types.
1493 def v16i8 : N2VQShIns<op24, op23, 0b001000, op11_8, 0, op4,
1494 !strconcat(OpcodeStr, "8"), v16i8, ShOp>;
1495 def v8i16 : N2VQShIns<op24, op23, 0b010000, op11_8, 0, op4,
1496 !strconcat(OpcodeStr, "16"), v8i16, ShOp>;
1497 def v4i32 : N2VQShIns<op24, op23, 0b100000, op11_8, 0, op4,
1498 !strconcat(OpcodeStr, "32"), v4i32, ShOp>;
1499 def v2i64 : N2VQShIns<op24, op23, 0b000000, op11_8, 1, op4,
1500 !strconcat(OpcodeStr, "64"), v2i64, ShOp>;
1503 //===----------------------------------------------------------------------===//
1504 // Instruction Definitions.
1505 //===----------------------------------------------------------------------===//
1507 // Vector Add Operations.
1509 // VADD : Vector Add (integer and floating-point)
1510 defm VADD : N3V_QHSD<0, 0, 0b1000, 0, IIC_VBINiD, IIC_VBINiQ, "vadd.i", add, 1>;
1511 def VADDfd : N3VD<0, 0, 0b00, 0b1101, 0, IIC_VBIND, "vadd.f32", v2f32, v2f32, fadd, 1>;
1512 def VADDfq : N3VQ<0, 0, 0b00, 0b1101, 0, IIC_VBINQ, "vadd.f32", v4f32, v4f32, fadd, 1>;
1513 // VADDL : Vector Add Long (Q = D + D)
1514 defm VADDLs : N3VLInt_QHS<0,1,0b0000,0, IIC_VSHLiD, "vaddl.s", int_arm_neon_vaddls, 1>;
1515 defm VADDLu : N3VLInt_QHS<1,1,0b0000,0, IIC_VSHLiD, "vaddl.u", int_arm_neon_vaddlu, 1>;
1516 // VADDW : Vector Add Wide (Q = Q + D)
1517 defm VADDWs : N3VWInt_QHS<0,1,0b0001,0, "vaddw.s", int_arm_neon_vaddws, 0>;
1518 defm VADDWu : N3VWInt_QHS<1,1,0b0001,0, "vaddw.u", int_arm_neon_vaddwu, 0>;
1519 // VHADD : Vector Halving Add
1520 defm VHADDs : N3VInt_QHS<0,0,0b0000,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1521 IIC_VBINi4Q, "vhadd.s", int_arm_neon_vhadds, 1>;
1522 defm VHADDu : N3VInt_QHS<1,0,0b0000,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1523 IIC_VBINi4Q, "vhadd.u", int_arm_neon_vhaddu, 1>;
1524 // VRHADD : Vector Rounding Halving Add
1525 defm VRHADDs : N3VInt_QHS<0,0,0b0001,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1526 IIC_VBINi4Q, "vrhadd.s", int_arm_neon_vrhadds, 1>;
1527 defm VRHADDu : N3VInt_QHS<1,0,0b0001,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1528 IIC_VBINi4Q, "vrhadd.u", int_arm_neon_vrhaddu, 1>;
1529 // VQADD : Vector Saturating Add
1530 defm VQADDs : N3VInt_QHSD<0,0,0b0000,1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1531 IIC_VBINi4Q, "vqadd.s", int_arm_neon_vqadds, 1>;
1532 defm VQADDu : N3VInt_QHSD<1,0,0b0000,1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1533 IIC_VBINi4Q, "vqadd.u", int_arm_neon_vqaddu, 1>;
1534 // VADDHN : Vector Add and Narrow Returning High Half (D = Q + Q)
1535 defm VADDHN : N3VNInt_HSD<0,1,0b0100,0, "vaddhn.i", int_arm_neon_vaddhn, 1>;
1536 // VRADDHN : Vector Rounding Add and Narrow Returning High Half (D = Q + Q)
1537 defm VRADDHN : N3VNInt_HSD<1,1,0b0100,0, "vraddhn.i", int_arm_neon_vraddhn, 1>;
1539 // Vector Multiply Operations.
1541 // VMUL : Vector Multiply (integer, polynomial and floating-point)
1542 defm VMUL : N3V_QHS<0, 0, 0b1001, 1, IIC_VMULi16D, IIC_VMULi32D, IIC_VMULi16Q,
1543 IIC_VMULi32Q, "vmul.i", mul, 1>;
1544 def VMULpd : N3VDInt<1, 0, 0b00, 0b1001, 1, IIC_VMULi16D, "vmul.p8", v8i8, v8i8,
1545 int_arm_neon_vmulp, 1>;
1546 def VMULpq : N3VQInt<1, 0, 0b00, 0b1001, 1, IIC_VMULi16Q, "vmul.p8", v16i8, v16i8,
1547 int_arm_neon_vmulp, 1>;
1548 def VMULfd : N3VD<1, 0, 0b00, 0b1101, 1, IIC_VBIND, "vmul.f32", v2f32, v2f32, fmul, 1>;
1549 def VMULfq : N3VQ<1, 0, 0b00, 0b1101, 1, IIC_VBINQ, "vmul.f32", v4f32, v4f32, fmul, 1>;
1550 defm VMULsl : N3VSL_HS<0b1000, "vmul.i", mul>;
1551 def VMULslfd : N3VDSL<0b10, 0b1001, IIC_VBIND, "vmul.f32", v2f32, fmul>;
1552 def VMULslfq : N3VQSL<0b10, 0b1001, IIC_VBINQ, "vmul.f32", v4f32, v2f32, fmul>;
1553 def : Pat<(v8i16 (mul (v8i16 QPR:$src1),
1554 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
1555 (v8i16 (VMULslv8i16 (v8i16 QPR:$src1),
1556 (v4i16 (EXTRACT_SUBREG QPR:$src2,
1557 (DSubReg_i16_reg imm:$lane))),
1558 (SubReg_i16_lane imm:$lane)))>;
1559 def : Pat<(v4i32 (mul (v4i32 QPR:$src1),
1560 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
1561 (v4i32 (VMULslv4i32 (v4i32 QPR:$src1),
1562 (v2i32 (EXTRACT_SUBREG QPR:$src2,
1563 (DSubReg_i32_reg imm:$lane))),
1564 (SubReg_i32_lane imm:$lane)))>;
1565 def : Pat<(v4f32 (fmul (v4f32 QPR:$src1),
1566 (v4f32 (NEONvduplane (v4f32 QPR:$src2), imm:$lane)))),
1567 (v4f32 (VMULslfq (v4f32 QPR:$src1),
1568 (v2f32 (EXTRACT_SUBREG QPR:$src2,
1569 (DSubReg_i32_reg imm:$lane))),
1570 (SubReg_i32_lane imm:$lane)))>;
1572 // VQDMULH : Vector Saturating Doubling Multiply Returning High Half
1573 defm VQDMULH : N3VInt_HS<0, 0, 0b1011, 0, IIC_VMULi16D, IIC_VMULi32D,
1574 IIC_VMULi16Q, IIC_VMULi32Q,
1575 "vqdmulh.s", int_arm_neon_vqdmulh, 1>;
1576 defm VQDMULHsl: N3VIntSL_HS<0b1100, IIC_VMULi16D, IIC_VMULi32D,
1577 IIC_VMULi16Q, IIC_VMULi32Q,
1578 "vqdmulh.s", int_arm_neon_vqdmulh>;
1579 def : Pat<(v8i16 (int_arm_neon_vqdmulh (v8i16 QPR:$src1),
1580 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
1581 (v8i16 (VQDMULHslv8i16 (v8i16 QPR:$src1),
1582 (v4i16 (EXTRACT_SUBREG QPR:$src2,
1583 (DSubReg_i16_reg imm:$lane))),
1584 (SubReg_i16_lane imm:$lane)))>;
1585 def : Pat<(v4i32 (int_arm_neon_vqdmulh (v4i32 QPR:$src1),
1586 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
1587 (v4i32 (VQDMULHslv4i32 (v4i32 QPR:$src1),
1588 (v2i32 (EXTRACT_SUBREG QPR:$src2,
1589 (DSubReg_i32_reg imm:$lane))),
1590 (SubReg_i32_lane imm:$lane)))>;
1592 // VQRDMULH : Vector Rounding Saturating Doubling Multiply Returning High Half
1593 defm VQRDMULH : N3VInt_HS<1, 0, 0b1011, 0, IIC_VMULi16D, IIC_VMULi32D,
1594 IIC_VMULi16Q, IIC_VMULi32Q,
1595 "vqrdmulh.s", int_arm_neon_vqrdmulh, 1>;
1596 defm VQRDMULHsl : N3VIntSL_HS<0b1101, IIC_VMULi16D, IIC_VMULi32D,
1597 IIC_VMULi16Q, IIC_VMULi32Q,
1598 "vqrdmulh.s", int_arm_neon_vqrdmulh>;
1599 def : Pat<(v8i16 (int_arm_neon_vqrdmulh (v8i16 QPR:$src1),
1600 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
1601 (v8i16 (VQRDMULHslv8i16 (v8i16 QPR:$src1),
1602 (v4i16 (EXTRACT_SUBREG QPR:$src2,
1603 (DSubReg_i16_reg imm:$lane))),
1604 (SubReg_i16_lane imm:$lane)))>;
1605 def : Pat<(v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$src1),
1606 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
1607 (v4i32 (VQRDMULHslv4i32 (v4i32 QPR:$src1),
1608 (v2i32 (EXTRACT_SUBREG QPR:$src2,
1609 (DSubReg_i32_reg imm:$lane))),
1610 (SubReg_i32_lane imm:$lane)))>;
1612 // VMULL : Vector Multiply Long (integer and polynomial) (Q = D * D)
1613 defm VMULLs : N3VLInt_QHS<0,1,0b1100,0, IIC_VMULi16D, "vmull.s", int_arm_neon_vmulls, 1>;
1614 defm VMULLu : N3VLInt_QHS<1,1,0b1100,0, IIC_VMULi16D, "vmull.u", int_arm_neon_vmullu, 1>;
1615 def VMULLp : N3VLInt<0, 1, 0b00, 0b1110, 0, IIC_VMULi16D, "vmull.p8", v8i16, v8i8,
1616 int_arm_neon_vmullp, 1>;
1617 defm VMULLsls : N3VLIntSL_HS<0, 0b1010, IIC_VMULi16D, "vmull.s", int_arm_neon_vmulls>;
1618 defm VMULLslu : N3VLIntSL_HS<1, 0b1010, IIC_VMULi16D, "vmull.u", int_arm_neon_vmullu>;
1620 // VQDMULL : Vector Saturating Doubling Multiply Long (Q = D * D)
1621 defm VQDMULL : N3VLInt_HS<0,1,0b1101,0, IIC_VMULi16D, "vqdmull.s", int_arm_neon_vqdmull, 1>;
1622 defm VQDMULLsl: N3VLIntSL_HS<0, 0b1011, IIC_VMULi16D, "vqdmull.s", int_arm_neon_vqdmull>;
1624 // Vector Multiply-Accumulate and Multiply-Subtract Operations.
1626 // VMLA : Vector Multiply Accumulate (integer and floating-point)
1627 defm VMLA : N3VMulOp_QHS<0, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
1628 IIC_VMACi16Q, IIC_VMACi32Q, "vmla.i", add>;
1629 def VMLAfd : N3VDMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACD, "vmla.f32", v2f32, fmul, fadd>;
1630 def VMLAfq : N3VQMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACQ, "vmla.f32", v4f32, fmul, fadd>;
1631 defm VMLAsl : N3VMulOpSL_HS<0b0000, IIC_VMACi16D, IIC_VMACi32D,
1632 IIC_VMACi16Q, IIC_VMACi32Q, "vmla.i", add>;
1633 def VMLAslfd : N3VDMulOpSL<0b10, 0b0001, IIC_VMACD, "vmla.f32", v2f32, fmul, fadd>;
1634 def VMLAslfq : N3VQMulOpSL<0b10, 0b0001, IIC_VMACQ, "vmla.f32", v4f32, v2f32, fmul, fadd>;
1636 def : Pat<(v8i16 (add (v8i16 QPR:$src1),
1637 (mul (v8i16 QPR:$src2),
1638 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
1639 (v8i16 (VMLAslv8i16 (v8i16 QPR:$src1),
1641 (v4i16 (EXTRACT_SUBREG QPR:$src3,
1642 (DSubReg_i16_reg imm:$lane))),
1643 (SubReg_i16_lane imm:$lane)))>;
1645 def : Pat<(v4i32 (add (v4i32 QPR:$src1),
1646 (mul (v4i32 QPR:$src2),
1647 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
1648 (v4i32 (VMLAslv4i32 (v4i32 QPR:$src1),
1650 (v2i32 (EXTRACT_SUBREG QPR:$src3,
1651 (DSubReg_i32_reg imm:$lane))),
1652 (SubReg_i32_lane imm:$lane)))>;
1654 def : Pat<(v4f32 (fadd (v4f32 QPR:$src1),
1655 (fmul (v4f32 QPR:$src2),
1656 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
1657 (v4f32 (VMLAslfq (v4f32 QPR:$src1),
1659 (v2f32 (EXTRACT_SUBREG QPR:$src3,
1660 (DSubReg_i32_reg imm:$lane))),
1661 (SubReg_i32_lane imm:$lane)))>;
1663 // VMLAL : Vector Multiply Accumulate Long (Q += D * D)
1664 defm VMLALs : N3VLInt3_QHS<0,1,0b1000,0, "vmlal.s", int_arm_neon_vmlals>;
1665 defm VMLALu : N3VLInt3_QHS<1,1,0b1000,0, "vmlal.u", int_arm_neon_vmlalu>;
1667 defm VMLALsls : N3VLInt3SL_HS<0, 0b0010, "vmlal.s", int_arm_neon_vmlals>;
1668 defm VMLALslu : N3VLInt3SL_HS<1, 0b0010, "vmlal.u", int_arm_neon_vmlalu>;
1670 // VQDMLAL : Vector Saturating Doubling Multiply Accumulate Long (Q += D * D)
1671 defm VQDMLAL : N3VLInt3_HS<0, 1, 0b1001, 0, "vqdmlal.s", int_arm_neon_vqdmlal>;
1672 defm VQDMLALsl: N3VLInt3SL_HS<0, 0b0011, "vqdmlal.s", int_arm_neon_vqdmlal>;
1674 // VMLS : Vector Multiply Subtract (integer and floating-point)
1675 defm VMLS : N3VMulOp_QHS<1, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
1676 IIC_VMACi16Q, IIC_VMACi32Q, "vmls.i", sub>;
1677 def VMLSfd : N3VDMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACD, "vmls.f32", v2f32, fmul, fsub>;
1678 def VMLSfq : N3VQMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACQ, "vmls.f32", v4f32, fmul, fsub>;
1679 defm VMLSsl : N3VMulOpSL_HS<0b0100, IIC_VMACi16D, IIC_VMACi32D,
1680 IIC_VMACi16Q, IIC_VMACi32Q, "vmls.i", sub>;
1681 def VMLSslfd : N3VDMulOpSL<0b10, 0b0101, IIC_VMACD, "vmls.f32", v2f32, fmul, fsub>;
1682 def VMLSslfq : N3VQMulOpSL<0b10, 0b0101, IIC_VMACQ, "vmls.f32", v4f32, v2f32, fmul, fsub>;
1684 def : Pat<(v8i16 (sub (v8i16 QPR:$src1),
1685 (mul (v8i16 QPR:$src2),
1686 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
1687 (v8i16 (VMLSslv8i16 (v8i16 QPR:$src1),
1689 (v4i16 (EXTRACT_SUBREG QPR:$src3,
1690 (DSubReg_i16_reg imm:$lane))),
1691 (SubReg_i16_lane imm:$lane)))>;
1693 def : Pat<(v4i32 (sub (v4i32 QPR:$src1),
1694 (mul (v4i32 QPR:$src2),
1695 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
1696 (v4i32 (VMLSslv4i32 (v4i32 QPR:$src1),
1698 (v2i32 (EXTRACT_SUBREG QPR:$src3,
1699 (DSubReg_i32_reg imm:$lane))),
1700 (SubReg_i32_lane imm:$lane)))>;
1702 def : Pat<(v4f32 (fsub (v4f32 QPR:$src1),
1703 (fmul (v4f32 QPR:$src2),
1704 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
1705 (v4f32 (VMLSslfq (v4f32 QPR:$src1),
1707 (v2f32 (EXTRACT_SUBREG QPR:$src3,
1708 (DSubReg_i32_reg imm:$lane))),
1709 (SubReg_i32_lane imm:$lane)))>;
1711 // VMLSL : Vector Multiply Subtract Long (Q -= D * D)
1712 defm VMLSLs : N3VLInt3_QHS<0,1,0b1010,0, "vmlsl.s", int_arm_neon_vmlsls>;
1713 defm VMLSLu : N3VLInt3_QHS<1,1,0b1010,0, "vmlsl.u", int_arm_neon_vmlslu>;
1715 defm VMLSLsls : N3VLInt3SL_HS<0, 0b0110, "vmlsl.s", int_arm_neon_vmlsls>;
1716 defm VMLSLslu : N3VLInt3SL_HS<1, 0b0110, "vmlsl.u", int_arm_neon_vmlslu>;
1718 // VQDMLSL : Vector Saturating Doubling Multiply Subtract Long (Q -= D * D)
1719 defm VQDMLSL : N3VLInt3_HS<0, 1, 0b1011, 0, "vqdmlsl.s", int_arm_neon_vqdmlsl>;
1720 defm VQDMLSLsl: N3VLInt3SL_HS<0, 0b111, "vqdmlsl.s", int_arm_neon_vqdmlsl>;
1722 // Vector Subtract Operations.
1724 // VSUB : Vector Subtract (integer and floating-point)
1725 defm VSUB : N3V_QHSD<1, 0, 0b1000, 0, IIC_VSUBiD, IIC_VSUBiQ, "vsub.i", sub, 0>;
1726 def VSUBfd : N3VD<0, 0, 0b10, 0b1101, 0, IIC_VBIND, "vsub.f32", v2f32, v2f32, fsub, 0>;
1727 def VSUBfq : N3VQ<0, 0, 0b10, 0b1101, 0, IIC_VBINQ, "vsub.f32", v4f32, v4f32, fsub, 0>;
1728 // VSUBL : Vector Subtract Long (Q = D - D)
1729 defm VSUBLs : N3VLInt_QHS<0,1,0b0010,0, IIC_VSHLiD, "vsubl.s", int_arm_neon_vsubls, 1>;
1730 defm VSUBLu : N3VLInt_QHS<1,1,0b0010,0, IIC_VSHLiD, "vsubl.u", int_arm_neon_vsublu, 1>;
1731 // VSUBW : Vector Subtract Wide (Q = Q - D)
1732 defm VSUBWs : N3VWInt_QHS<0,1,0b0011,0, "vsubw.s", int_arm_neon_vsubws, 0>;
1733 defm VSUBWu : N3VWInt_QHS<1,1,0b0011,0, "vsubw.u", int_arm_neon_vsubwu, 0>;
1734 // VHSUB : Vector Halving Subtract
1735 defm VHSUBs : N3VInt_QHS<0, 0, 0b0010, 0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1736 IIC_VBINi4Q, "vhsub.s", int_arm_neon_vhsubs, 0>;
1737 defm VHSUBu : N3VInt_QHS<1, 0, 0b0010, 0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1738 IIC_VBINi4Q, "vhsub.u", int_arm_neon_vhsubu, 0>;
1739 // VQSUB : Vector Saturing Subtract
1740 defm VQSUBs : N3VInt_QHSD<0, 0, 0b0010, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1741 IIC_VBINi4Q, "vqsub.s", int_arm_neon_vqsubs, 0>;
1742 defm VQSUBu : N3VInt_QHSD<1, 0, 0b0010, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1743 IIC_VBINi4Q, "vqsub.u", int_arm_neon_vqsubu, 0>;
1744 // VSUBHN : Vector Subtract and Narrow Returning High Half (D = Q - Q)
1745 defm VSUBHN : N3VNInt_HSD<0,1,0b0110,0, "vsubhn.i", int_arm_neon_vsubhn, 0>;
1746 // VRSUBHN : Vector Rounding Subtract and Narrow Returning High Half (D=Q-Q)
1747 defm VRSUBHN : N3VNInt_HSD<1,1,0b0110,0, "vrsubhn.i", int_arm_neon_vrsubhn, 0>;
1749 // Vector Comparisons.
1751 // VCEQ : Vector Compare Equal
1752 defm VCEQ : N3V_QHS<1, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1753 IIC_VBINi4Q, "vceq.i", NEONvceq, 1>;
1754 def VCEQfd : N3VD<0,0,0b00,0b1110,0, IIC_VBIND, "vceq.f32", v2i32, v2f32, NEONvceq, 1>;
1755 def VCEQfq : N3VQ<0,0,0b00,0b1110,0, IIC_VBINQ, "vceq.f32", v4i32, v4f32, NEONvceq, 1>;
1756 // VCGE : Vector Compare Greater Than or Equal
1757 defm VCGEs : N3V_QHS<0, 0, 0b0011, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1758 IIC_VBINi4Q, "vcge.s", NEONvcge, 0>;
1759 defm VCGEu : N3V_QHS<1, 0, 0b0011, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1760 IIC_VBINi4Q, "vcge.u", NEONvcgeu, 0>;
1761 def VCGEfd : N3VD<1,0,0b00,0b1110,0, IIC_VBIND, "vcge.f32", v2i32, v2f32, NEONvcge, 0>;
1762 def VCGEfq : N3VQ<1,0,0b00,0b1110,0, IIC_VBINQ, "vcge.f32", v4i32, v4f32, NEONvcge, 0>;
1763 // VCGT : Vector Compare Greater Than
1764 defm VCGTs : N3V_QHS<0, 0, 0b0011, 0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1765 IIC_VBINi4Q, "vcgt.s", NEONvcgt, 0>;
1766 defm VCGTu : N3V_QHS<1, 0, 0b0011, 0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1767 IIC_VBINi4Q, "vcgt.u", NEONvcgtu, 0>;
1768 def VCGTfd : N3VD<1,0,0b10,0b1110,0, IIC_VBIND, "vcgt.f32", v2i32, v2f32, NEONvcgt, 0>;
1769 def VCGTfq : N3VQ<1,0,0b10,0b1110,0, IIC_VBINQ, "vcgt.f32", v4i32, v4f32, NEONvcgt, 0>;
1770 // VACGE : Vector Absolute Compare Greater Than or Equal (aka VCAGE)
1771 def VACGEd : N3VDInt<1, 0, 0b00, 0b1110, 1, IIC_VBIND, "vacge.f32", v2i32, v2f32,
1772 int_arm_neon_vacged, 0>;
1773 def VACGEq : N3VQInt<1, 0, 0b00, 0b1110, 1, IIC_VBINQ, "vacge.f32", v4i32, v4f32,
1774 int_arm_neon_vacgeq, 0>;
1775 // VACGT : Vector Absolute Compare Greater Than (aka VCAGT)
1776 def VACGTd : N3VDInt<1, 0, 0b10, 0b1110, 1, IIC_VBIND, "vacgt.f32", v2i32, v2f32,
1777 int_arm_neon_vacgtd, 0>;
1778 def VACGTq : N3VQInt<1, 0, 0b10, 0b1110, 1, IIC_VBINQ, "vacgt.f32", v4i32, v4f32,
1779 int_arm_neon_vacgtq, 0>;
1780 // VTST : Vector Test Bits
1781 defm VTST : N3V_QHS<0, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1782 IIC_VBINi4Q, "vtst.i", NEONvtst, 1>;
1784 // Vector Bitwise Operations.
1786 // VAND : Vector Bitwise AND
1787 def VANDd : N3VD<0, 0, 0b00, 0b0001, 1, IIC_VBINiD, "vand", v2i32, v2i32, and, 1>;
1788 def VANDq : N3VQ<0, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "vand", v4i32, v4i32, and, 1>;
1790 // VEOR : Vector Bitwise Exclusive OR
1791 def VEORd : N3VD<1, 0, 0b00, 0b0001, 1, IIC_VBINiD, "veor", v2i32, v2i32, xor, 1>;
1792 def VEORq : N3VQ<1, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "veor", v4i32, v4i32, xor, 1>;
1794 // VORR : Vector Bitwise OR
1795 def VORRd : N3VD<0, 0, 0b10, 0b0001, 1, IIC_VBINiD, "vorr", v2i32, v2i32, or, 1>;
1796 def VORRq : N3VQ<0, 0, 0b10, 0b0001, 1, IIC_VBINiQ, "vorr", v4i32, v4i32, or, 1>;
1798 // VBIC : Vector Bitwise Bit Clear (AND NOT)
1799 def VBICd : N3V<0, 0, 0b01, 0b0001, 0, 1, (outs DPR:$dst),
1800 (ins DPR:$src1, DPR:$src2), IIC_VBINiD,
1801 "vbic\t$dst, $src1, $src2", "",
1802 [(set DPR:$dst, (v2i32 (and DPR:$src1,
1803 (vnot_conv DPR:$src2))))]>;
1804 def VBICq : N3V<0, 0, 0b01, 0b0001, 1, 1, (outs QPR:$dst),
1805 (ins QPR:$src1, QPR:$src2), IIC_VBINiQ,
1806 "vbic\t$dst, $src1, $src2", "",
1807 [(set QPR:$dst, (v4i32 (and QPR:$src1,
1808 (vnot_conv QPR:$src2))))]>;
1810 // VORN : Vector Bitwise OR NOT
1811 def VORNd : N3V<0, 0, 0b11, 0b0001, 0, 1, (outs DPR:$dst),
1812 (ins DPR:$src1, DPR:$src2), IIC_VBINiD,
1813 "vorn\t$dst, $src1, $src2", "",
1814 [(set DPR:$dst, (v2i32 (or DPR:$src1,
1815 (vnot_conv DPR:$src2))))]>;
1816 def VORNq : N3V<0, 0, 0b11, 0b0001, 1, 1, (outs QPR:$dst),
1817 (ins QPR:$src1, QPR:$src2), IIC_VBINiQ,
1818 "vorn\t$dst, $src1, $src2", "",
1819 [(set QPR:$dst, (v4i32 (or QPR:$src1,
1820 (vnot_conv QPR:$src2))))]>;
1822 // VMVN : Vector Bitwise NOT
1823 def VMVNd : N2V<0b11, 0b11, 0b00, 0b00, 0b01011, 0, 0,
1824 (outs DPR:$dst), (ins DPR:$src), IIC_VSHLiD,
1825 "vmvn\t$dst, $src", "",
1826 [(set DPR:$dst, (v2i32 (vnot DPR:$src)))]>;
1827 def VMVNq : N2V<0b11, 0b11, 0b00, 0b00, 0b01011, 1, 0,
1828 (outs QPR:$dst), (ins QPR:$src), IIC_VSHLiD,
1829 "vmvn\t$dst, $src", "",
1830 [(set QPR:$dst, (v4i32 (vnot QPR:$src)))]>;
1831 def : Pat<(v2i32 (vnot_conv DPR:$src)), (VMVNd DPR:$src)>;
1832 def : Pat<(v4i32 (vnot_conv QPR:$src)), (VMVNq QPR:$src)>;
1834 // VBSL : Vector Bitwise Select
1835 def VBSLd : N3V<1, 0, 0b01, 0b0001, 0, 1, (outs DPR:$dst),
1836 (ins DPR:$src1, DPR:$src2, DPR:$src3), IIC_VCNTiD,
1837 "vbsl\t$dst, $src2, $src3", "$src1 = $dst",
1839 (v2i32 (or (and DPR:$src2, DPR:$src1),
1840 (and DPR:$src3, (vnot_conv DPR:$src1)))))]>;
1841 def VBSLq : N3V<1, 0, 0b01, 0b0001, 1, 1, (outs QPR:$dst),
1842 (ins QPR:$src1, QPR:$src2, QPR:$src3), IIC_VCNTiQ,
1843 "vbsl\t$dst, $src2, $src3", "$src1 = $dst",
1845 (v4i32 (or (and QPR:$src2, QPR:$src1),
1846 (and QPR:$src3, (vnot_conv QPR:$src1)))))]>;
1848 // VBIF : Vector Bitwise Insert if False
1849 // like VBSL but with: "vbif\t$dst, $src3, $src1", "$src2 = $dst",
1850 // VBIT : Vector Bitwise Insert if True
1851 // like VBSL but with: "vbit\t$dst, $src2, $src1", "$src3 = $dst",
1852 // These are not yet implemented. The TwoAddress pass will not go looking
1853 // for equivalent operations with different register constraints; it just
1856 // Vector Absolute Differences.
1858 // VABD : Vector Absolute Difference
1859 defm VABDs : N3VInt_QHS<0, 0, 0b0111, 0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1860 IIC_VBINi4Q, "vabd.s", int_arm_neon_vabds, 0>;
1861 defm VABDu : N3VInt_QHS<1, 0, 0b0111, 0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1862 IIC_VBINi4Q, "vabd.u", int_arm_neon_vabdu, 0>;
1863 def VABDfd : N3VDInt<1, 0, 0b10, 0b1101, 0, IIC_VBIND, "vabd.f32", v2f32, v2f32,
1864 int_arm_neon_vabds, 0>;
1865 def VABDfq : N3VQInt<1, 0, 0b10, 0b1101, 0, IIC_VBINQ, "vabd.f32", v4f32, v4f32,
1866 int_arm_neon_vabds, 0>;
1868 // VABDL : Vector Absolute Difference Long (Q = | D - D |)
1869 defm VABDLs : N3VLInt_QHS<0,1,0b0111,0, IIC_VBINi4Q, "vabdl.s", int_arm_neon_vabdls, 0>;
1870 defm VABDLu : N3VLInt_QHS<1,1,0b0111,0, IIC_VBINi4Q, "vabdl.u", int_arm_neon_vabdlu, 0>;
1872 // VABA : Vector Absolute Difference and Accumulate
1873 defm VABAs : N3VInt3_QHS<0,1,0b0101,0, "vaba.s", int_arm_neon_vabas>;
1874 defm VABAu : N3VInt3_QHS<1,1,0b0101,0, "vaba.u", int_arm_neon_vabau>;
1876 // VABAL : Vector Absolute Difference and Accumulate Long (Q += | D - D |)
1877 defm VABALs : N3VLInt3_QHS<0,1,0b0101,0, "vabal.s", int_arm_neon_vabals>;
1878 defm VABALu : N3VLInt3_QHS<1,1,0b0101,0, "vabal.u", int_arm_neon_vabalu>;
1880 // Vector Maximum and Minimum.
1882 // VMAX : Vector Maximum
1883 defm VMAXs : N3VInt_QHS<0, 0, 0b0110, 0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1884 IIC_VBINi4Q, "vmax.s", int_arm_neon_vmaxs, 1>;
1885 defm VMAXu : N3VInt_QHS<1, 0, 0b0110, 0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1886 IIC_VBINi4Q, "vmax.u", int_arm_neon_vmaxu, 1>;
1887 def VMAXfd : N3VDInt<0, 0, 0b00, 0b1111, 0, IIC_VBIND, "vmax.f32", v2f32, v2f32,
1888 int_arm_neon_vmaxs, 1>;
1889 def VMAXfq : N3VQInt<0, 0, 0b00, 0b1111, 0, IIC_VBINQ, "vmax.f32", v4f32, v4f32,
1890 int_arm_neon_vmaxs, 1>;
1892 // VMIN : Vector Minimum
1893 defm VMINs : N3VInt_QHS<0, 0, 0b0110, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1894 IIC_VBINi4Q, "vmin.s", int_arm_neon_vmins, 1>;
1895 defm VMINu : N3VInt_QHS<1, 0, 0b0110, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1896 IIC_VBINi4Q, "vmin.u", int_arm_neon_vminu, 1>;
1897 def VMINfd : N3VDInt<0, 0, 0b10, 0b1111, 0, IIC_VBIND, "vmin.f32", v2f32, v2f32,
1898 int_arm_neon_vmins, 1>;
1899 def VMINfq : N3VQInt<0, 0, 0b10, 0b1111, 0, IIC_VBINQ, "vmin.f32", v4f32, v4f32,
1900 int_arm_neon_vmins, 1>;
1902 // Vector Pairwise Operations.
1904 // VPADD : Vector Pairwise Add
1905 def VPADDi8 : N3VDInt<0, 0, 0b00, 0b1011, 1, IIC_VBINiD, "vpadd.i8", v8i8, v8i8,
1906 int_arm_neon_vpadd, 0>;
1907 def VPADDi16 : N3VDInt<0, 0, 0b01, 0b1011, 1, IIC_VBINiD, "vpadd.i16", v4i16, v4i16,
1908 int_arm_neon_vpadd, 0>;
1909 def VPADDi32 : N3VDInt<0, 0, 0b10, 0b1011, 1, IIC_VBINiD, "vpadd.i32", v2i32, v2i32,
1910 int_arm_neon_vpadd, 0>;
1911 def VPADDf : N3VDInt<1, 0, 0b00, 0b1101, 0, IIC_VBIND, "vpadd.f32", v2f32, v2f32,
1912 int_arm_neon_vpadd, 0>;
1914 // VPADDL : Vector Pairwise Add Long
1915 defm VPADDLs : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00100, 0, "vpaddl.s",
1916 int_arm_neon_vpaddls>;
1917 defm VPADDLu : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00101, 0, "vpaddl.u",
1918 int_arm_neon_vpaddlu>;
1920 // VPADAL : Vector Pairwise Add and Accumulate Long
1921 defm VPADALs : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b00100, 0, "vpadal.s",
1922 int_arm_neon_vpadals>;
1923 defm VPADALu : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b00101, 0, "vpadal.u",
1924 int_arm_neon_vpadalu>;
1926 // VPMAX : Vector Pairwise Maximum
1927 def VPMAXs8 : N3VDInt<0, 0, 0b00, 0b1010, 0, IIC_VBINi4D, "vpmax.s8", v8i8, v8i8,
1928 int_arm_neon_vpmaxs, 0>;
1929 def VPMAXs16 : N3VDInt<0, 0, 0b01, 0b1010, 0, IIC_VBINi4D, "vpmax.s16", v4i16, v4i16,
1930 int_arm_neon_vpmaxs, 0>;
1931 def VPMAXs32 : N3VDInt<0, 0, 0b10, 0b1010, 0, IIC_VBINi4D, "vpmax.s32", v2i32, v2i32,
1932 int_arm_neon_vpmaxs, 0>;
1933 def VPMAXu8 : N3VDInt<1, 0, 0b00, 0b1010, 0, IIC_VBINi4D, "vpmax.u8", v8i8, v8i8,
1934 int_arm_neon_vpmaxu, 0>;
1935 def VPMAXu16 : N3VDInt<1, 0, 0b01, 0b1010, 0, IIC_VBINi4D, "vpmax.u16", v4i16, v4i16,
1936 int_arm_neon_vpmaxu, 0>;
1937 def VPMAXu32 : N3VDInt<1, 0, 0b10, 0b1010, 0, IIC_VBINi4D, "vpmax.u32", v2i32, v2i32,
1938 int_arm_neon_vpmaxu, 0>;
1939 def VPMAXf : N3VDInt<1, 0, 0b00, 0b1111, 0, IIC_VBINi4D, "vpmax.f32", v2f32, v2f32,
1940 int_arm_neon_vpmaxs, 0>;
1942 // VPMIN : Vector Pairwise Minimum
1943 def VPMINs8 : N3VDInt<0, 0, 0b00, 0b1010, 1, IIC_VBINi4D, "vpmin.s8", v8i8, v8i8,
1944 int_arm_neon_vpmins, 0>;
1945 def VPMINs16 : N3VDInt<0, 0, 0b01, 0b1010, 1, IIC_VBINi4D, "vpmin.s16", v4i16, v4i16,
1946 int_arm_neon_vpmins, 0>;
1947 def VPMINs32 : N3VDInt<0, 0, 0b10, 0b1010, 1, IIC_VBINi4D, "vpmin.s32", v2i32, v2i32,
1948 int_arm_neon_vpmins, 0>;
1949 def VPMINu8 : N3VDInt<1, 0, 0b00, 0b1010, 1, IIC_VBINi4D, "vpmin.u8", v8i8, v8i8,
1950 int_arm_neon_vpminu, 0>;
1951 def VPMINu16 : N3VDInt<1, 0, 0b01, 0b1010, 1, IIC_VBINi4D, "vpmin.u16", v4i16, v4i16,
1952 int_arm_neon_vpminu, 0>;
1953 def VPMINu32 : N3VDInt<1, 0, 0b10, 0b1010, 1, IIC_VBINi4D, "vpmin.u32", v2i32, v2i32,
1954 int_arm_neon_vpminu, 0>;
1955 def VPMINf : N3VDInt<1, 0, 0b10, 0b1111, 0, IIC_VBINi4D, "vpmin.f32", v2f32, v2f32,
1956 int_arm_neon_vpmins, 0>;
1958 // Vector Reciprocal and Reciprocal Square Root Estimate and Step.
1960 // VRECPE : Vector Reciprocal Estimate
1961 def VRECPEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
1962 IIC_VUNAD, "vrecpe.u32",
1963 v2i32, v2i32, int_arm_neon_vrecpe>;
1964 def VRECPEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
1965 IIC_VUNAQ, "vrecpe.u32",
1966 v4i32, v4i32, int_arm_neon_vrecpe>;
1967 def VRECPEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
1968 IIC_VUNAD, "vrecpe.f32",
1969 v2f32, v2f32, int_arm_neon_vrecpe>;
1970 def VRECPEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
1971 IIC_VUNAQ, "vrecpe.f32",
1972 v4f32, v4f32, int_arm_neon_vrecpe>;
1974 // VRECPS : Vector Reciprocal Step
1975 def VRECPSfd : N3VDInt<0, 0, 0b00, 0b1111, 1, IIC_VRECSD, "vrecps.f32", v2f32, v2f32,
1976 int_arm_neon_vrecps, 1>;
1977 def VRECPSfq : N3VQInt<0, 0, 0b00, 0b1111, 1, IIC_VRECSQ, "vrecps.f32", v4f32, v4f32,
1978 int_arm_neon_vrecps, 1>;
1980 // VRSQRTE : Vector Reciprocal Square Root Estimate
1981 def VRSQRTEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
1982 IIC_VUNAD, "vrsqrte.u32",
1983 v2i32, v2i32, int_arm_neon_vrsqrte>;
1984 def VRSQRTEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
1985 IIC_VUNAQ, "vrsqrte.u32",
1986 v4i32, v4i32, int_arm_neon_vrsqrte>;
1987 def VRSQRTEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
1988 IIC_VUNAD, "vrsqrte.f32",
1989 v2f32, v2f32, int_arm_neon_vrsqrte>;
1990 def VRSQRTEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
1991 IIC_VUNAQ, "vrsqrte.f32",
1992 v4f32, v4f32, int_arm_neon_vrsqrte>;
1994 // VRSQRTS : Vector Reciprocal Square Root Step
1995 def VRSQRTSfd : N3VDInt<0, 0, 0b10, 0b1111, 1, IIC_VRECSD, "vrsqrts.f32", v2f32, v2f32,
1996 int_arm_neon_vrsqrts, 1>;
1997 def VRSQRTSfq : N3VQInt<0, 0, 0b10, 0b1111, 1, IIC_VRECSQ, "vrsqrts.f32", v4f32, v4f32,
1998 int_arm_neon_vrsqrts, 1>;
2002 // VSHL : Vector Shift
2003 defm VSHLs : N3VInt_QHSD<0, 0, 0b0100, 0, IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ,
2004 IIC_VSHLiQ, "vshl.s", int_arm_neon_vshifts, 0>;
2005 defm VSHLu : N3VInt_QHSD<1, 0, 0b0100, 0, IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ,
2006 IIC_VSHLiQ, "vshl.u", int_arm_neon_vshiftu, 0>;
2007 // VSHL : Vector Shift Left (Immediate)
2008 defm VSHLi : N2VSh_QHSD<0, 1, 0b0111, 1, IIC_VSHLiD, "vshl.i", NEONvshl>;
2009 // VSHR : Vector Shift Right (Immediate)
2010 defm VSHRs : N2VSh_QHSD<0, 1, 0b0000, 1, IIC_VSHLiD, "vshr.s", NEONvshrs>;
2011 defm VSHRu : N2VSh_QHSD<1, 1, 0b0000, 1, IIC_VSHLiD, "vshr.u", NEONvshru>;
2013 // VSHLL : Vector Shift Left Long
2014 def VSHLLs8 : N2VLSh<0, 1, 0b001000, 0b1010, 0, 0, 1, "vshll.s8",
2015 v8i16, v8i8, NEONvshlls>;
2016 def VSHLLs16 : N2VLSh<0, 1, 0b010000, 0b1010, 0, 0, 1, "vshll.s16",
2017 v4i32, v4i16, NEONvshlls>;
2018 def VSHLLs32 : N2VLSh<0, 1, 0b100000, 0b1010, 0, 0, 1, "vshll.s32",
2019 v2i64, v2i32, NEONvshlls>;
2020 def VSHLLu8 : N2VLSh<1, 1, 0b001000, 0b1010, 0, 0, 1, "vshll.u8",
2021 v8i16, v8i8, NEONvshllu>;
2022 def VSHLLu16 : N2VLSh<1, 1, 0b010000, 0b1010, 0, 0, 1, "vshll.u16",
2023 v4i32, v4i16, NEONvshllu>;
2024 def VSHLLu32 : N2VLSh<1, 1, 0b100000, 0b1010, 0, 0, 1, "vshll.u32",
2025 v2i64, v2i32, NEONvshllu>;
2027 // VSHLL : Vector Shift Left Long (with maximum shift count)
2028 def VSHLLi8 : N2VLSh<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll.i8",
2029 v8i16, v8i8, NEONvshlli>;
2030 def VSHLLi16 : N2VLSh<1, 1, 0b110110, 0b0011, 0, 0, 0, "vshll.i16",
2031 v4i32, v4i16, NEONvshlli>;
2032 def VSHLLi32 : N2VLSh<1, 1, 0b111010, 0b0011, 0, 0, 0, "vshll.i32",
2033 v2i64, v2i32, NEONvshlli>;
2035 // VSHRN : Vector Shift Right and Narrow
2036 def VSHRN16 : N2VNSh<0, 1, 0b001000, 0b1000, 0, 0, 1,
2037 IIC_VSHLiD, "vshrn.i16", v8i8, v8i16, NEONvshrn>;
2038 def VSHRN32 : N2VNSh<0, 1, 0b010000, 0b1000, 0, 0, 1,
2039 IIC_VSHLiD, "vshrn.i32", v4i16, v4i32, NEONvshrn>;
2040 def VSHRN64 : N2VNSh<0, 1, 0b100000, 0b1000, 0, 0, 1,
2041 IIC_VSHLiD, "vshrn.i64", v2i32, v2i64, NEONvshrn>;
2043 // VRSHL : Vector Rounding Shift
2044 defm VRSHLs : N3VInt_QHSD<0,0,0b0101,0, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2045 IIC_VSHLi4Q, "vrshl.s", int_arm_neon_vrshifts, 0>;
2046 defm VRSHLu : N3VInt_QHSD<1,0,0b0101,0, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2047 IIC_VSHLi4Q, "vrshl.u", int_arm_neon_vrshiftu, 0>;
2048 // VRSHR : Vector Rounding Shift Right
2049 defm VRSHRs : N2VSh_QHSD<0, 1, 0b0010, 1, IIC_VSHLi4D, "vrshr.s", NEONvrshrs>;
2050 defm VRSHRu : N2VSh_QHSD<1, 1, 0b0010, 1, IIC_VSHLi4D, "vrshr.u", NEONvrshru>;
2052 // VRSHRN : Vector Rounding Shift Right and Narrow
2053 def VRSHRN16 : N2VNSh<0, 1, 0b001000, 0b1000, 0, 1, 1,
2054 IIC_VSHLi4D, "vrshrn.i16", v8i8, v8i16, NEONvrshrn>;
2055 def VRSHRN32 : N2VNSh<0, 1, 0b010000, 0b1000, 0, 1, 1,
2056 IIC_VSHLi4D, "vrshrn.i32", v4i16, v4i32, NEONvrshrn>;
2057 def VRSHRN64 : N2VNSh<0, 1, 0b100000, 0b1000, 0, 1, 1,
2058 IIC_VSHLi4D, "vrshrn.i64", v2i32, v2i64, NEONvrshrn>;
2060 // VQSHL : Vector Saturating Shift
2061 defm VQSHLs : N3VInt_QHSD<0,0,0b0100,1, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2062 IIC_VSHLi4Q, "vqshl.s", int_arm_neon_vqshifts, 0>;
2063 defm VQSHLu : N3VInt_QHSD<1,0,0b0100,1, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2064 IIC_VSHLi4Q, "vqshl.u", int_arm_neon_vqshiftu, 0>;
2065 // VQSHL : Vector Saturating Shift Left (Immediate)
2066 defm VQSHLsi : N2VSh_QHSD<0, 1, 0b0111, 1, IIC_VSHLi4D, "vqshl.s", NEONvqshls>;
2067 defm VQSHLui : N2VSh_QHSD<1, 1, 0b0111, 1, IIC_VSHLi4D, "vqshl.u", NEONvqshlu>;
2068 // VQSHLU : Vector Saturating Shift Left (Immediate, Unsigned)
2069 defm VQSHLsu : N2VSh_QHSD<1, 1, 0b0110, 1, IIC_VSHLi4D, "vqshlu.s", NEONvqshlsu>;
2071 // VQSHRN : Vector Saturating Shift Right and Narrow
2072 def VQSHRNs16 : N2VNSh<0, 1, 0b001000, 0b1001, 0, 0, 1,
2073 IIC_VSHLi4D, "vqshrn.s16", v8i8, v8i16, NEONvqshrns>;
2074 def VQSHRNs32 : N2VNSh<0, 1, 0b010000, 0b1001, 0, 0, 1,
2075 IIC_VSHLi4D, "vqshrn.s32", v4i16, v4i32, NEONvqshrns>;
2076 def VQSHRNs64 : N2VNSh<0, 1, 0b100000, 0b1001, 0, 0, 1,
2077 IIC_VSHLi4D, "vqshrn.s64", v2i32, v2i64, NEONvqshrns>;
2078 def VQSHRNu16 : N2VNSh<1, 1, 0b001000, 0b1001, 0, 0, 1,
2079 IIC_VSHLi4D, "vqshrn.u16", v8i8, v8i16, NEONvqshrnu>;
2080 def VQSHRNu32 : N2VNSh<1, 1, 0b010000, 0b1001, 0, 0, 1,
2081 IIC_VSHLi4D, "vqshrn.u32", v4i16, v4i32, NEONvqshrnu>;
2082 def VQSHRNu64 : N2VNSh<1, 1, 0b100000, 0b1001, 0, 0, 1,
2083 IIC_VSHLi4D, "vqshrn.u64", v2i32, v2i64, NEONvqshrnu>;
2085 // VQSHRUN : Vector Saturating Shift Right and Narrow (Unsigned)
2086 def VQSHRUN16 : N2VNSh<1, 1, 0b001000, 0b1000, 0, 0, 1,
2087 IIC_VSHLi4D, "vqshrun.s16", v8i8, v8i16, NEONvqshrnsu>;
2088 def VQSHRUN32 : N2VNSh<1, 1, 0b010000, 0b1000, 0, 0, 1,
2089 IIC_VSHLi4D, "vqshrun.s32", v4i16, v4i32, NEONvqshrnsu>;
2090 def VQSHRUN64 : N2VNSh<1, 1, 0b100000, 0b1000, 0, 0, 1,
2091 IIC_VSHLi4D, "vqshrun.s64", v2i32, v2i64, NEONvqshrnsu>;
2093 // VQRSHL : Vector Saturating Rounding Shift
2094 defm VQRSHLs : N3VInt_QHSD<0, 0, 0b0101, 1, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2095 IIC_VSHLi4Q, "vqrshl.s", int_arm_neon_vqrshifts, 0>;
2096 defm VQRSHLu : N3VInt_QHSD<1, 0, 0b0101, 1, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2097 IIC_VSHLi4Q, "vqrshl.u", int_arm_neon_vqrshiftu, 0>;
2099 // VQRSHRN : Vector Saturating Rounding Shift Right and Narrow
2100 def VQRSHRNs16: N2VNSh<0, 1, 0b001000, 0b1001, 0, 1, 1,
2101 IIC_VSHLi4D, "vqrshrn.s16", v8i8, v8i16, NEONvqrshrns>;
2102 def VQRSHRNs32: N2VNSh<0, 1, 0b010000, 0b1001, 0, 1, 1,
2103 IIC_VSHLi4D, "vqrshrn.s32", v4i16, v4i32, NEONvqrshrns>;
2104 def VQRSHRNs64: N2VNSh<0, 1, 0b100000, 0b1001, 0, 1, 1,
2105 IIC_VSHLi4D, "vqrshrn.s64", v2i32, v2i64, NEONvqrshrns>;
2106 def VQRSHRNu16: N2VNSh<1, 1, 0b001000, 0b1001, 0, 1, 1,
2107 IIC_VSHLi4D, "vqrshrn.u16", v8i8, v8i16, NEONvqrshrnu>;
2108 def VQRSHRNu32: N2VNSh<1, 1, 0b010000, 0b1001, 0, 1, 1,
2109 IIC_VSHLi4D, "vqrshrn.u32", v4i16, v4i32, NEONvqrshrnu>;
2110 def VQRSHRNu64: N2VNSh<1, 1, 0b100000, 0b1001, 0, 1, 1,
2111 IIC_VSHLi4D, "vqrshrn.u64", v2i32, v2i64, NEONvqrshrnu>;
2113 // VQRSHRUN : Vector Saturating Rounding Shift Right and Narrow (Unsigned)
2114 def VQRSHRUN16: N2VNSh<1, 1, 0b001000, 0b1000, 0, 1, 1,
2115 IIC_VSHLi4D, "vqrshrun.s16", v8i8, v8i16, NEONvqrshrnsu>;
2116 def VQRSHRUN32: N2VNSh<1, 1, 0b010000, 0b1000, 0, 1, 1,
2117 IIC_VSHLi4D, "vqrshrun.s32", v4i16, v4i32, NEONvqrshrnsu>;
2118 def VQRSHRUN64: N2VNSh<1, 1, 0b100000, 0b1000, 0, 1, 1,
2119 IIC_VSHLi4D, "vqrshrun.s64", v2i32, v2i64, NEONvqrshrnsu>;
2121 // VSRA : Vector Shift Right and Accumulate
2122 defm VSRAs : N2VShAdd_QHSD<0, 1, 0b0001, 1, "vsra.s", NEONvshrs>;
2123 defm VSRAu : N2VShAdd_QHSD<1, 1, 0b0001, 1, "vsra.u", NEONvshru>;
2124 // VRSRA : Vector Rounding Shift Right and Accumulate
2125 defm VRSRAs : N2VShAdd_QHSD<0, 1, 0b0011, 1, "vrsra.s", NEONvrshrs>;
2126 defm VRSRAu : N2VShAdd_QHSD<1, 1, 0b0011, 1, "vrsra.u", NEONvrshru>;
2128 // VSLI : Vector Shift Left and Insert
2129 defm VSLI : N2VShIns_QHSD<1, 1, 0b0101, 1, "vsli.", NEONvsli>;
2130 // VSRI : Vector Shift Right and Insert
2131 defm VSRI : N2VShIns_QHSD<1, 1, 0b0100, 1, "vsri.", NEONvsri>;
2133 // Vector Absolute and Saturating Absolute.
2135 // VABS : Vector Absolute Value
2136 defm VABS : N2VInt_QHS<0b11, 0b11, 0b01, 0b00110, 0,
2137 IIC_VUNAiD, IIC_VUNAiQ, "vabs.s",
2139 def VABSfd : N2VDInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
2140 IIC_VUNAD, "vabs.f32",
2141 v2f32, v2f32, int_arm_neon_vabs>;
2142 def VABSfq : N2VQInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
2143 IIC_VUNAQ, "vabs.f32",
2144 v4f32, v4f32, int_arm_neon_vabs>;
2146 // VQABS : Vector Saturating Absolute Value
2147 defm VQABS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01110, 0,
2148 IIC_VQUNAiD, IIC_VQUNAiQ, "vqabs.s",
2149 int_arm_neon_vqabs>;
2153 def vneg : PatFrag<(ops node:$in), (sub immAllZerosV, node:$in)>;
2154 def vneg_conv : PatFrag<(ops node:$in), (sub immAllZerosV_bc, node:$in)>;
2156 class VNEGD<bits<2> size, string OpcodeStr, ValueType Ty>
2157 : N2V<0b11, 0b11, size, 0b01, 0b00111, 0, 0, (outs DPR:$dst), (ins DPR:$src),
2158 IIC_VSHLiD, !strconcat(OpcodeStr, "\t$dst, $src"), "",
2159 [(set DPR:$dst, (Ty (vneg DPR:$src)))]>;
2160 class VNEGQ<bits<2> size, string OpcodeStr, ValueType Ty>
2161 : N2V<0b11, 0b11, size, 0b01, 0b00111, 1, 0, (outs QPR:$dst), (ins QPR:$src),
2162 IIC_VSHLiD, !strconcat(OpcodeStr, "\t$dst, $src"), "",
2163 [(set QPR:$dst, (Ty (vneg QPR:$src)))]>;
2165 // VNEG : Vector Negate
2166 def VNEGs8d : VNEGD<0b00, "vneg.s8", v8i8>;
2167 def VNEGs16d : VNEGD<0b01, "vneg.s16", v4i16>;
2168 def VNEGs32d : VNEGD<0b10, "vneg.s32", v2i32>;
2169 def VNEGs8q : VNEGQ<0b00, "vneg.s8", v16i8>;
2170 def VNEGs16q : VNEGQ<0b01, "vneg.s16", v8i16>;
2171 def VNEGs32q : VNEGQ<0b10, "vneg.s32", v4i32>;
2173 // VNEG : Vector Negate (floating-point)
2174 def VNEGf32d : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
2175 (outs DPR:$dst), (ins DPR:$src), IIC_VUNAD,
2176 "vneg.f32\t$dst, $src", "",
2177 [(set DPR:$dst, (v2f32 (fneg DPR:$src)))]>;
2178 def VNEGf32q : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 1, 0,
2179 (outs QPR:$dst), (ins QPR:$src), IIC_VUNAQ,
2180 "vneg.f32\t$dst, $src", "",
2181 [(set QPR:$dst, (v4f32 (fneg QPR:$src)))]>;
2183 def : Pat<(v8i8 (vneg_conv DPR:$src)), (VNEGs8d DPR:$src)>;
2184 def : Pat<(v4i16 (vneg_conv DPR:$src)), (VNEGs16d DPR:$src)>;
2185 def : Pat<(v2i32 (vneg_conv DPR:$src)), (VNEGs32d DPR:$src)>;
2186 def : Pat<(v16i8 (vneg_conv QPR:$src)), (VNEGs8q QPR:$src)>;
2187 def : Pat<(v8i16 (vneg_conv QPR:$src)), (VNEGs16q QPR:$src)>;
2188 def : Pat<(v4i32 (vneg_conv QPR:$src)), (VNEGs32q QPR:$src)>;
2190 // VQNEG : Vector Saturating Negate
2191 defm VQNEG : N2VInt_QHS<0b11, 0b11, 0b00, 0b01111, 0,
2192 IIC_VQUNAiD, IIC_VQUNAiQ, "vqneg.s",
2193 int_arm_neon_vqneg>;
2195 // Vector Bit Counting Operations.
2197 // VCLS : Vector Count Leading Sign Bits
2198 defm VCLS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01000, 0,
2199 IIC_VCNTiD, IIC_VCNTiQ, "vcls.s",
2201 // VCLZ : Vector Count Leading Zeros
2202 defm VCLZ : N2VInt_QHS<0b11, 0b11, 0b00, 0b01001, 0,
2203 IIC_VCNTiD, IIC_VCNTiQ, "vclz.i",
2205 // VCNT : Vector Count One Bits
2206 def VCNTd : N2VDInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
2207 IIC_VCNTiD, "vcnt.8",
2208 v8i8, v8i8, int_arm_neon_vcnt>;
2209 def VCNTq : N2VQInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
2210 IIC_VCNTiQ, "vcnt.8",
2211 v16i8, v16i8, int_arm_neon_vcnt>;
2213 // Vector Move Operations.
2215 // VMOV : Vector Move (Register)
2217 def VMOVD : N3V<0, 0, 0b10, 0b0001, 0, 1, (outs DPR:$dst), (ins DPR:$src),
2218 IIC_VMOVD, "vmov\t$dst, $src", "", []>;
2219 def VMOVQ : N3V<0, 0, 0b10, 0b0001, 1, 1, (outs QPR:$dst), (ins QPR:$src),
2220 IIC_VMOVD, "vmov\t$dst, $src", "", []>;
2222 // VMOV : Vector Move (Immediate)
2224 // VMOV_get_imm8 xform function: convert build_vector to VMOV.i8 imm.
2225 def VMOV_get_imm8 : SDNodeXForm<build_vector, [{
2226 return ARM::getVMOVImm(N, 1, *CurDAG);
2228 def vmovImm8 : PatLeaf<(build_vector), [{
2229 return ARM::getVMOVImm(N, 1, *CurDAG).getNode() != 0;
2232 // VMOV_get_imm16 xform function: convert build_vector to VMOV.i16 imm.
2233 def VMOV_get_imm16 : SDNodeXForm<build_vector, [{
2234 return ARM::getVMOVImm(N, 2, *CurDAG);
2236 def vmovImm16 : PatLeaf<(build_vector), [{
2237 return ARM::getVMOVImm(N, 2, *CurDAG).getNode() != 0;
2238 }], VMOV_get_imm16>;
2240 // VMOV_get_imm32 xform function: convert build_vector to VMOV.i32 imm.
2241 def VMOV_get_imm32 : SDNodeXForm<build_vector, [{
2242 return ARM::getVMOVImm(N, 4, *CurDAG);
2244 def vmovImm32 : PatLeaf<(build_vector), [{
2245 return ARM::getVMOVImm(N, 4, *CurDAG).getNode() != 0;
2246 }], VMOV_get_imm32>;
2248 // VMOV_get_imm64 xform function: convert build_vector to VMOV.i64 imm.
2249 def VMOV_get_imm64 : SDNodeXForm<build_vector, [{
2250 return ARM::getVMOVImm(N, 8, *CurDAG);
2252 def vmovImm64 : PatLeaf<(build_vector), [{
2253 return ARM::getVMOVImm(N, 8, *CurDAG).getNode() != 0;
2254 }], VMOV_get_imm64>;
2256 // Note: Some of the cmode bits in the following VMOV instructions need to
2257 // be encoded based on the immed values.
2259 def VMOVv8i8 : N1ModImm<1, 0b000, 0b1110, 0, 0, 0, 1, (outs DPR:$dst),
2260 (ins i8imm:$SIMM), IIC_VMOVImm,
2261 "vmov.i8\t$dst, $SIMM", "",
2262 [(set DPR:$dst, (v8i8 vmovImm8:$SIMM))]>;
2263 def VMOVv16i8 : N1ModImm<1, 0b000, 0b1110, 0, 1, 0, 1, (outs QPR:$dst),
2264 (ins i8imm:$SIMM), IIC_VMOVImm,
2265 "vmov.i8\t$dst, $SIMM", "",
2266 [(set QPR:$dst, (v16i8 vmovImm8:$SIMM))]>;
2268 def VMOVv4i16 : N1ModImm<1, 0b000, 0b1000, 0, 0, 0, 1, (outs DPR:$dst),
2269 (ins i16imm:$SIMM), IIC_VMOVImm,
2270 "vmov.i16\t$dst, $SIMM", "",
2271 [(set DPR:$dst, (v4i16 vmovImm16:$SIMM))]>;
2272 def VMOVv8i16 : N1ModImm<1, 0b000, 0b1000, 0, 1, 0, 1, (outs QPR:$dst),
2273 (ins i16imm:$SIMM), IIC_VMOVImm,
2274 "vmov.i16\t$dst, $SIMM", "",
2275 [(set QPR:$dst, (v8i16 vmovImm16:$SIMM))]>;
2277 def VMOVv2i32 : N1ModImm<1, 0b000, 0b0000, 0, 0, 0, 1, (outs DPR:$dst),
2278 (ins i32imm:$SIMM), IIC_VMOVImm,
2279 "vmov.i32\t$dst, $SIMM", "",
2280 [(set DPR:$dst, (v2i32 vmovImm32:$SIMM))]>;
2281 def VMOVv4i32 : N1ModImm<1, 0b000, 0b0000, 0, 1, 0, 1, (outs QPR:$dst),
2282 (ins i32imm:$SIMM), IIC_VMOVImm,
2283 "vmov.i32\t$dst, $SIMM", "",
2284 [(set QPR:$dst, (v4i32 vmovImm32:$SIMM))]>;
2286 def VMOVv1i64 : N1ModImm<1, 0b000, 0b1110, 0, 0, 1, 1, (outs DPR:$dst),
2287 (ins i64imm:$SIMM), IIC_VMOVImm,
2288 "vmov.i64\t$dst, $SIMM", "",
2289 [(set DPR:$dst, (v1i64 vmovImm64:$SIMM))]>;
2290 def VMOVv2i64 : N1ModImm<1, 0b000, 0b1110, 0, 1, 1, 1, (outs QPR:$dst),
2291 (ins i64imm:$SIMM), IIC_VMOVImm,
2292 "vmov.i64\t$dst, $SIMM", "",
2293 [(set QPR:$dst, (v2i64 vmovImm64:$SIMM))]>;
2295 // VMOV : Vector Get Lane (move scalar to ARM core register)
2297 def VGETLNs8 : NVGetLane<0b11100101, 0b1011, 0b00,
2298 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2299 IIC_VMOVSI, "vmov", ".s8\t$dst, $src[$lane]",
2300 [(set GPR:$dst, (NEONvgetlanes (v8i8 DPR:$src),
2302 def VGETLNs16 : NVGetLane<0b11100001, 0b1011, 0b01,
2303 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2304 IIC_VMOVSI, "vmov", ".s16\t$dst, $src[$lane]",
2305 [(set GPR:$dst, (NEONvgetlanes (v4i16 DPR:$src),
2307 def VGETLNu8 : NVGetLane<0b11101101, 0b1011, 0b00,
2308 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2309 IIC_VMOVSI, "vmov", ".u8\t$dst, $src[$lane]",
2310 [(set GPR:$dst, (NEONvgetlaneu (v8i8 DPR:$src),
2312 def VGETLNu16 : NVGetLane<0b11101001, 0b1011, 0b01,
2313 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2314 IIC_VMOVSI, "vmov", ".u16\t$dst, $src[$lane]",
2315 [(set GPR:$dst, (NEONvgetlaneu (v4i16 DPR:$src),
2317 def VGETLNi32 : NVGetLane<0b11100001, 0b1011, 0b00,
2318 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2319 IIC_VMOVSI, "vmov", ".32\t$dst, $src[$lane]",
2320 [(set GPR:$dst, (extractelt (v2i32 DPR:$src),
2322 // def VGETLNf32: see FMRDH and FMRDL in ARMInstrVFP.td
2323 def : Pat<(NEONvgetlanes (v16i8 QPR:$src), imm:$lane),
2324 (VGETLNs8 (v8i8 (EXTRACT_SUBREG QPR:$src,
2325 (DSubReg_i8_reg imm:$lane))),
2326 (SubReg_i8_lane imm:$lane))>;
2327 def : Pat<(NEONvgetlanes (v8i16 QPR:$src), imm:$lane),
2328 (VGETLNs16 (v4i16 (EXTRACT_SUBREG QPR:$src,
2329 (DSubReg_i16_reg imm:$lane))),
2330 (SubReg_i16_lane imm:$lane))>;
2331 def : Pat<(NEONvgetlaneu (v16i8 QPR:$src), imm:$lane),
2332 (VGETLNu8 (v8i8 (EXTRACT_SUBREG QPR:$src,
2333 (DSubReg_i8_reg imm:$lane))),
2334 (SubReg_i8_lane imm:$lane))>;
2335 def : Pat<(NEONvgetlaneu (v8i16 QPR:$src), imm:$lane),
2336 (VGETLNu16 (v4i16 (EXTRACT_SUBREG QPR:$src,
2337 (DSubReg_i16_reg imm:$lane))),
2338 (SubReg_i16_lane imm:$lane))>;
2339 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
2340 (VGETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src,
2341 (DSubReg_i32_reg imm:$lane))),
2342 (SubReg_i32_lane imm:$lane))>;
2343 def : Pat<(extractelt (v2f32 DPR:$src1), imm:$src2),
2344 (EXTRACT_SUBREG (COPY_TO_REGCLASS DPR:$src1, DPR_VFP2),
2345 (SSubReg_f32_reg imm:$src2))>;
2346 def : Pat<(extractelt (v4f32 QPR:$src1), imm:$src2),
2347 (EXTRACT_SUBREG (COPY_TO_REGCLASS QPR:$src1, QPR_VFP2),
2348 (SSubReg_f32_reg imm:$src2))>;
2349 //def : Pat<(extractelt (v2i64 QPR:$src1), imm:$src2),
2350 // (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
2351 def : Pat<(extractelt (v2f64 QPR:$src1), imm:$src2),
2352 (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
2355 // VMOV : Vector Set Lane (move ARM core register to scalar)
2357 let Constraints = "$src1 = $dst" in {
2358 def VSETLNi8 : NVSetLane<0b11100100, 0b1011, 0b00, (outs DPR:$dst),
2359 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
2360 IIC_VMOVISL, "vmov", ".8\t$dst[$lane], $src2",
2361 [(set DPR:$dst, (vector_insert (v8i8 DPR:$src1),
2362 GPR:$src2, imm:$lane))]>;
2363 def VSETLNi16 : NVSetLane<0b11100000, 0b1011, 0b01, (outs DPR:$dst),
2364 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
2365 IIC_VMOVISL, "vmov", ".16\t$dst[$lane], $src2",
2366 [(set DPR:$dst, (vector_insert (v4i16 DPR:$src1),
2367 GPR:$src2, imm:$lane))]>;
2368 def VSETLNi32 : NVSetLane<0b11100000, 0b1011, 0b00, (outs DPR:$dst),
2369 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
2370 IIC_VMOVISL, "vmov", ".32\t$dst[$lane], $src2",
2371 [(set DPR:$dst, (insertelt (v2i32 DPR:$src1),
2372 GPR:$src2, imm:$lane))]>;
2374 def : Pat<(vector_insert (v16i8 QPR:$src1), GPR:$src2, imm:$lane),
2375 (v16i8 (INSERT_SUBREG QPR:$src1,
2376 (VSETLNi8 (v8i8 (EXTRACT_SUBREG QPR:$src1,
2377 (DSubReg_i8_reg imm:$lane))),
2378 GPR:$src2, (SubReg_i8_lane imm:$lane)),
2379 (DSubReg_i8_reg imm:$lane)))>;
2380 def : Pat<(vector_insert (v8i16 QPR:$src1), GPR:$src2, imm:$lane),
2381 (v8i16 (INSERT_SUBREG QPR:$src1,
2382 (VSETLNi16 (v4i16 (EXTRACT_SUBREG QPR:$src1,
2383 (DSubReg_i16_reg imm:$lane))),
2384 GPR:$src2, (SubReg_i16_lane imm:$lane)),
2385 (DSubReg_i16_reg imm:$lane)))>;
2386 def : Pat<(insertelt (v4i32 QPR:$src1), GPR:$src2, imm:$lane),
2387 (v4i32 (INSERT_SUBREG QPR:$src1,
2388 (VSETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src1,
2389 (DSubReg_i32_reg imm:$lane))),
2390 GPR:$src2, (SubReg_i32_lane imm:$lane)),
2391 (DSubReg_i32_reg imm:$lane)))>;
2393 def : Pat<(v2f32 (insertelt DPR:$src1, SPR:$src2, imm:$src3)),
2394 (INSERT_SUBREG (COPY_TO_REGCLASS DPR:$src1, DPR_VFP2),
2395 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
2396 def : Pat<(v4f32 (insertelt QPR:$src1, SPR:$src2, imm:$src3)),
2397 (INSERT_SUBREG (COPY_TO_REGCLASS QPR:$src1, QPR_VFP2),
2398 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
2400 //def : Pat<(v2i64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
2401 // (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
2402 def : Pat<(v2f64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
2403 (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
2405 def : Pat<(v2f32 (scalar_to_vector SPR:$src)),
2406 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$src, arm_ssubreg_0)>;
2407 def : Pat<(v2f64 (scalar_to_vector DPR:$src)),
2408 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), DPR:$src, arm_dsubreg_0)>;
2409 def : Pat<(v4f32 (scalar_to_vector SPR:$src)),
2410 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), SPR:$src, arm_ssubreg_0)>;
2412 def : Pat<(v8i8 (scalar_to_vector GPR:$src)),
2413 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
2414 def : Pat<(v4i16 (scalar_to_vector GPR:$src)),
2415 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
2416 def : Pat<(v2i32 (scalar_to_vector GPR:$src)),
2417 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
2419 def : Pat<(v16i8 (scalar_to_vector GPR:$src)),
2420 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
2421 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
2423 def : Pat<(v8i16 (scalar_to_vector GPR:$src)),
2424 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
2425 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
2427 def : Pat<(v4i32 (scalar_to_vector GPR:$src)),
2428 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
2429 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
2432 // VDUP : Vector Duplicate (from ARM core register to all elements)
2434 class VDUPD<bits<8> opcod1, bits<2> opcod3, string asmSize, ValueType Ty>
2435 : NVDup<opcod1, 0b1011, opcod3, (outs DPR:$dst), (ins GPR:$src),
2436 IIC_VMOVIS, "vdup", !strconcat(asmSize, "\t$dst, $src"),
2437 [(set DPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
2438 class VDUPQ<bits<8> opcod1, bits<2> opcod3, string asmSize, ValueType Ty>
2439 : NVDup<opcod1, 0b1011, opcod3, (outs QPR:$dst), (ins GPR:$src),
2440 IIC_VMOVIS, "vdup", !strconcat(asmSize, "\t$dst, $src"),
2441 [(set QPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
2443 def VDUP8d : VDUPD<0b11101100, 0b00, ".8", v8i8>;
2444 def VDUP16d : VDUPD<0b11101000, 0b01, ".16", v4i16>;
2445 def VDUP32d : VDUPD<0b11101000, 0b00, ".32", v2i32>;
2446 def VDUP8q : VDUPQ<0b11101110, 0b00, ".8", v16i8>;
2447 def VDUP16q : VDUPQ<0b11101010, 0b01, ".16", v8i16>;
2448 def VDUP32q : VDUPQ<0b11101010, 0b00, ".32", v4i32>;
2450 def VDUPfd : NVDup<0b11101000, 0b1011, 0b00, (outs DPR:$dst), (ins GPR:$src),
2451 IIC_VMOVIS, "vdup", ".32\t$dst, $src",
2452 [(set DPR:$dst, (v2f32 (NEONvdup
2453 (f32 (bitconvert GPR:$src)))))]>;
2454 def VDUPfq : NVDup<0b11101010, 0b1011, 0b00, (outs QPR:$dst), (ins GPR:$src),
2455 IIC_VMOVIS, "vdup", ".32\t$dst, $src",
2456 [(set QPR:$dst, (v4f32 (NEONvdup
2457 (f32 (bitconvert GPR:$src)))))]>;
2459 // VDUP : Vector Duplicate Lane (from scalar to all elements)
2461 class VDUPLND<bits<2> op19_18, bits<2> op17_16, string OpcodeStr, ValueType Ty>
2462 : N2V<0b11, 0b11, op19_18, op17_16, 0b11000, 0, 0,
2463 (outs DPR:$dst), (ins DPR:$src, nohash_imm:$lane), IIC_VMOVD,
2464 !strconcat(OpcodeStr, "\t$dst, $src[$lane]"), "",
2465 [(set DPR:$dst, (Ty (NEONvduplane (Ty DPR:$src), imm:$lane)))]>;
2467 class VDUPLNQ<bits<2> op19_18, bits<2> op17_16, string OpcodeStr,
2468 ValueType ResTy, ValueType OpTy>
2469 : N2V<0b11, 0b11, op19_18, op17_16, 0b11000, 1, 0,
2470 (outs QPR:$dst), (ins DPR:$src, nohash_imm:$lane), IIC_VMOVD,
2471 !strconcat(OpcodeStr, "\t$dst, $src[$lane]"), "",
2472 [(set QPR:$dst, (ResTy (NEONvduplane (OpTy DPR:$src), imm:$lane)))]>;
2474 def VDUPLN8d : VDUPLND<0b00, 0b01, "vdup.8", v8i8>;
2475 def VDUPLN16d : VDUPLND<0b00, 0b10, "vdup.16", v4i16>;
2476 def VDUPLN32d : VDUPLND<0b01, 0b00, "vdup.32", v2i32>;
2477 def VDUPLNfd : VDUPLND<0b01, 0b00, "vdup.32", v2f32>;
2478 def VDUPLN8q : VDUPLNQ<0b00, 0b01, "vdup.8", v16i8, v8i8>;
2479 def VDUPLN16q : VDUPLNQ<0b00, 0b10, "vdup.16", v8i16, v4i16>;
2480 def VDUPLN32q : VDUPLNQ<0b01, 0b00, "vdup.32", v4i32, v2i32>;
2481 def VDUPLNfq : VDUPLNQ<0b01, 0b00, "vdup.32", v4f32, v2f32>;
2483 def : Pat<(v16i8 (NEONvduplane (v16i8 QPR:$src), imm:$lane)),
2484 (v16i8 (VDUPLN8q (v8i8 (EXTRACT_SUBREG QPR:$src,
2485 (DSubReg_i8_reg imm:$lane))),
2486 (SubReg_i8_lane imm:$lane)))>;
2487 def : Pat<(v8i16 (NEONvduplane (v8i16 QPR:$src), imm:$lane)),
2488 (v8i16 (VDUPLN16q (v4i16 (EXTRACT_SUBREG QPR:$src,
2489 (DSubReg_i16_reg imm:$lane))),
2490 (SubReg_i16_lane imm:$lane)))>;
2491 def : Pat<(v4i32 (NEONvduplane (v4i32 QPR:$src), imm:$lane)),
2492 (v4i32 (VDUPLN32q (v2i32 (EXTRACT_SUBREG QPR:$src,
2493 (DSubReg_i32_reg imm:$lane))),
2494 (SubReg_i32_lane imm:$lane)))>;
2495 def : Pat<(v4f32 (NEONvduplane (v4f32 QPR:$src), imm:$lane)),
2496 (v4f32 (VDUPLNfq (v2f32 (EXTRACT_SUBREG QPR:$src,
2497 (DSubReg_i32_reg imm:$lane))),
2498 (SubReg_i32_lane imm:$lane)))>;
2500 def VDUPfdf : N2V<0b11, 0b11, 0b01, 0b00, 0b11000, 0, 0,
2501 (outs DPR:$dst), (ins SPR:$src),
2502 IIC_VMOVD, "vdup.32\t$dst, ${src:lane}", "",
2503 [(set DPR:$dst, (v2f32 (NEONvdup (f32 SPR:$src))))]>;
2505 def VDUPfqf : N2V<0b11, 0b11, 0b01, 0b00, 0b11000, 1, 0,
2506 (outs QPR:$dst), (ins SPR:$src),
2507 IIC_VMOVD, "vdup.32\t$dst, ${src:lane}", "",
2508 [(set QPR:$dst, (v4f32 (NEONvdup (f32 SPR:$src))))]>;
2510 def : Pat<(v2i64 (NEONvduplane (v2i64 QPR:$src), imm:$lane)),
2511 (INSERT_SUBREG QPR:$src,
2512 (i64 (EXTRACT_SUBREG QPR:$src, (DSubReg_f64_reg imm:$lane))),
2513 (DSubReg_f64_other_reg imm:$lane))>;
2514 def : Pat<(v2f64 (NEONvduplane (v2f64 QPR:$src), imm:$lane)),
2515 (INSERT_SUBREG QPR:$src,
2516 (f64 (EXTRACT_SUBREG QPR:$src, (DSubReg_f64_reg imm:$lane))),
2517 (DSubReg_f64_other_reg imm:$lane))>;
2519 // VMOVN : Vector Narrowing Move
2520 defm VMOVN : N2VNInt_HSD<0b11,0b11,0b10,0b00100,0,0, IIC_VMOVD, "vmovn.i",
2521 int_arm_neon_vmovn>;
2522 // VQMOVN : Vector Saturating Narrowing Move
2523 defm VQMOVNs : N2VNInt_HSD<0b11,0b11,0b10,0b00101,0,0, IIC_VQUNAiD, "vqmovn.s",
2524 int_arm_neon_vqmovns>;
2525 defm VQMOVNu : N2VNInt_HSD<0b11,0b11,0b10,0b00101,1,0, IIC_VQUNAiD, "vqmovn.u",
2526 int_arm_neon_vqmovnu>;
2527 defm VQMOVNsu : N2VNInt_HSD<0b11,0b11,0b10,0b00100,1,0, IIC_VQUNAiD, "vqmovun.s",
2528 int_arm_neon_vqmovnsu>;
2529 // VMOVL : Vector Lengthening Move
2530 defm VMOVLs : N2VLInt_QHS<0,1,0b1010,0,0,1, "vmovl.s", int_arm_neon_vmovls>;
2531 defm VMOVLu : N2VLInt_QHS<1,1,0b1010,0,0,1, "vmovl.u", int_arm_neon_vmovlu>;
2533 // Vector Conversions.
2535 // VCVT : Vector Convert Between Floating-Point and Integers
2536 def VCVTf2sd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt.s32.f32",
2537 v2i32, v2f32, fp_to_sint>;
2538 def VCVTf2ud : N2VD<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt.u32.f32",
2539 v2i32, v2f32, fp_to_uint>;
2540 def VCVTs2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt.f32.s32",
2541 v2f32, v2i32, sint_to_fp>;
2542 def VCVTu2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt.f32.u32",
2543 v2f32, v2i32, uint_to_fp>;
2545 def VCVTf2sq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt.s32.f32",
2546 v4i32, v4f32, fp_to_sint>;
2547 def VCVTf2uq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt.u32.f32",
2548 v4i32, v4f32, fp_to_uint>;
2549 def VCVTs2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt.f32.s32",
2550 v4f32, v4i32, sint_to_fp>;
2551 def VCVTu2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt.f32.u32",
2552 v4f32, v4i32, uint_to_fp>;
2554 // VCVT : Vector Convert Between Floating-Point and Fixed-Point.
2555 // Note: Some of the opcode bits in the following VCVT instructions need to
2556 // be encoded based on the immed values.
2557 def VCVTf2xsd : N2VCvtD<0, 1, 0b000000, 0b1111, 0, 1, "vcvt.s32.f32",
2558 v2i32, v2f32, int_arm_neon_vcvtfp2fxs>;
2559 def VCVTf2xud : N2VCvtD<1, 1, 0b000000, 0b1111, 0, 1, "vcvt.u32.f32",
2560 v2i32, v2f32, int_arm_neon_vcvtfp2fxu>;
2561 def VCVTxs2fd : N2VCvtD<0, 1, 0b000000, 0b1110, 0, 1, "vcvt.f32.s32",
2562 v2f32, v2i32, int_arm_neon_vcvtfxs2fp>;
2563 def VCVTxu2fd : N2VCvtD<1, 1, 0b000000, 0b1110, 0, 1, "vcvt.f32.u32",
2564 v2f32, v2i32, int_arm_neon_vcvtfxu2fp>;
2566 def VCVTf2xsq : N2VCvtQ<0, 1, 0b000000, 0b1111, 0, 1, "vcvt.s32.f32",
2567 v4i32, v4f32, int_arm_neon_vcvtfp2fxs>;
2568 def VCVTf2xuq : N2VCvtQ<1, 1, 0b000000, 0b1111, 0, 1, "vcvt.u32.f32",
2569 v4i32, v4f32, int_arm_neon_vcvtfp2fxu>;
2570 def VCVTxs2fq : N2VCvtQ<0, 1, 0b000000, 0b1110, 0, 1, "vcvt.f32.s32",
2571 v4f32, v4i32, int_arm_neon_vcvtfxs2fp>;
2572 def VCVTxu2fq : N2VCvtQ<1, 1, 0b000000, 0b1110, 0, 1, "vcvt.f32.u32",
2573 v4f32, v4i32, int_arm_neon_vcvtfxu2fp>;
2577 // VREV64 : Vector Reverse elements within 64-bit doublewords
2579 class VREV64D<bits<2> op19_18, string OpcodeStr, ValueType Ty>
2580 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 0, 0, (outs DPR:$dst),
2581 (ins DPR:$src), IIC_VMOVD,
2582 !strconcat(OpcodeStr, "\t$dst, $src"), "",
2583 [(set DPR:$dst, (Ty (NEONvrev64 (Ty DPR:$src))))]>;
2584 class VREV64Q<bits<2> op19_18, string OpcodeStr, ValueType Ty>
2585 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 1, 0, (outs QPR:$dst),
2586 (ins QPR:$src), IIC_VMOVD,
2587 !strconcat(OpcodeStr, "\t$dst, $src"), "",
2588 [(set QPR:$dst, (Ty (NEONvrev64 (Ty QPR:$src))))]>;
2590 def VREV64d8 : VREV64D<0b00, "vrev64.8", v8i8>;
2591 def VREV64d16 : VREV64D<0b01, "vrev64.16", v4i16>;
2592 def VREV64d32 : VREV64D<0b10, "vrev64.32", v2i32>;
2593 def VREV64df : VREV64D<0b10, "vrev64.32", v2f32>;
2595 def VREV64q8 : VREV64Q<0b00, "vrev64.8", v16i8>;
2596 def VREV64q16 : VREV64Q<0b01, "vrev64.16", v8i16>;
2597 def VREV64q32 : VREV64Q<0b10, "vrev64.32", v4i32>;
2598 def VREV64qf : VREV64Q<0b10, "vrev64.32", v4f32>;
2600 // VREV32 : Vector Reverse elements within 32-bit words
2602 class VREV32D<bits<2> op19_18, string OpcodeStr, ValueType Ty>
2603 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 0, 0, (outs DPR:$dst),
2604 (ins DPR:$src), IIC_VMOVD,
2605 !strconcat(OpcodeStr, "\t$dst, $src"), "",
2606 [(set DPR:$dst, (Ty (NEONvrev32 (Ty DPR:$src))))]>;
2607 class VREV32Q<bits<2> op19_18, string OpcodeStr, ValueType Ty>
2608 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 1, 0, (outs QPR:$dst),
2609 (ins QPR:$src), IIC_VMOVD,
2610 !strconcat(OpcodeStr, "\t$dst, $src"), "",
2611 [(set QPR:$dst, (Ty (NEONvrev32 (Ty QPR:$src))))]>;
2613 def VREV32d8 : VREV32D<0b00, "vrev32.8", v8i8>;
2614 def VREV32d16 : VREV32D<0b01, "vrev32.16", v4i16>;
2616 def VREV32q8 : VREV32Q<0b00, "vrev32.8", v16i8>;
2617 def VREV32q16 : VREV32Q<0b01, "vrev32.16", v8i16>;
2619 // VREV16 : Vector Reverse elements within 16-bit halfwords
2621 class VREV16D<bits<2> op19_18, string OpcodeStr, ValueType Ty>
2622 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 0, 0, (outs DPR:$dst),
2623 (ins DPR:$src), IIC_VMOVD,
2624 !strconcat(OpcodeStr, "\t$dst, $src"), "",
2625 [(set DPR:$dst, (Ty (NEONvrev16 (Ty DPR:$src))))]>;
2626 class VREV16Q<bits<2> op19_18, string OpcodeStr, ValueType Ty>
2627 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 1, 0, (outs QPR:$dst),
2628 (ins QPR:$src), IIC_VMOVD,
2629 !strconcat(OpcodeStr, "\t$dst, $src"), "",
2630 [(set QPR:$dst, (Ty (NEONvrev16 (Ty QPR:$src))))]>;
2632 def VREV16d8 : VREV16D<0b00, "vrev16.8", v8i8>;
2633 def VREV16q8 : VREV16Q<0b00, "vrev16.8", v16i8>;
2635 // Other Vector Shuffles.
2637 // VEXT : Vector Extract
2639 class VEXTd<string OpcodeStr, ValueType Ty>
2640 : N3V<0,1,0b11,0b0000,0,0, (outs DPR:$dst),
2641 (ins DPR:$lhs, DPR:$rhs, i32imm:$index), IIC_VEXTD,
2642 !strconcat(OpcodeStr, "\t$dst, $lhs, $rhs, $index"), "",
2643 [(set DPR:$dst, (Ty (NEONvext (Ty DPR:$lhs),
2644 (Ty DPR:$rhs), imm:$index)))]>;
2646 class VEXTq<string OpcodeStr, ValueType Ty>
2647 : N3V<0,1,0b11,0b0000,1,0, (outs QPR:$dst),
2648 (ins QPR:$lhs, QPR:$rhs, i32imm:$index), IIC_VEXTQ,
2649 !strconcat(OpcodeStr, "\t$dst, $lhs, $rhs, $index"), "",
2650 [(set QPR:$dst, (Ty (NEONvext (Ty QPR:$lhs),
2651 (Ty QPR:$rhs), imm:$index)))]>;
2653 def VEXTd8 : VEXTd<"vext.8", v8i8>;
2654 def VEXTd16 : VEXTd<"vext.16", v4i16>;
2655 def VEXTd32 : VEXTd<"vext.32", v2i32>;
2656 def VEXTdf : VEXTd<"vext.32", v2f32>;
2658 def VEXTq8 : VEXTq<"vext.8", v16i8>;
2659 def VEXTq16 : VEXTq<"vext.16", v8i16>;
2660 def VEXTq32 : VEXTq<"vext.32", v4i32>;
2661 def VEXTqf : VEXTq<"vext.32", v4f32>;
2663 // VTRN : Vector Transpose
2665 def VTRNd8 : N2VDShuffle<0b00, 0b00001, "vtrn.8">;
2666 def VTRNd16 : N2VDShuffle<0b01, 0b00001, "vtrn.16">;
2667 def VTRNd32 : N2VDShuffle<0b10, 0b00001, "vtrn.32">;
2669 def VTRNq8 : N2VQShuffle<0b00, 0b00001, IIC_VPERMQ, "vtrn.8">;
2670 def VTRNq16 : N2VQShuffle<0b01, 0b00001, IIC_VPERMQ, "vtrn.16">;
2671 def VTRNq32 : N2VQShuffle<0b10, 0b00001, IIC_VPERMQ, "vtrn.32">;
2673 // VUZP : Vector Unzip (Deinterleave)
2675 def VUZPd8 : N2VDShuffle<0b00, 0b00010, "vuzp.8">;
2676 def VUZPd16 : N2VDShuffle<0b01, 0b00010, "vuzp.16">;
2677 def VUZPd32 : N2VDShuffle<0b10, 0b00010, "vuzp.32">;
2679 def VUZPq8 : N2VQShuffle<0b00, 0b00010, IIC_VPERMQ3, "vuzp.8">;
2680 def VUZPq16 : N2VQShuffle<0b01, 0b00010, IIC_VPERMQ3, "vuzp.16">;
2681 def VUZPq32 : N2VQShuffle<0b10, 0b00010, IIC_VPERMQ3, "vuzp.32">;
2683 // VZIP : Vector Zip (Interleave)
2685 def VZIPd8 : N2VDShuffle<0b00, 0b00011, "vzip.8">;
2686 def VZIPd16 : N2VDShuffle<0b01, 0b00011, "vzip.16">;
2687 def VZIPd32 : N2VDShuffle<0b10, 0b00011, "vzip.32">;
2689 def VZIPq8 : N2VQShuffle<0b00, 0b00011, IIC_VPERMQ3, "vzip.8">;
2690 def VZIPq16 : N2VQShuffle<0b01, 0b00011, IIC_VPERMQ3, "vzip.16">;
2691 def VZIPq32 : N2VQShuffle<0b10, 0b00011, IIC_VPERMQ3, "vzip.32">;
2693 // Vector Table Lookup and Table Extension.
2695 // VTBL : Vector Table Lookup
2697 : N3V<1,1,0b11,0b1000,0,0, (outs DPR:$dst),
2698 (ins DPR:$tbl1, DPR:$src), IIC_VTB1,
2699 "vtbl.8\t$dst, \\{$tbl1\\}, $src", "",
2700 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl1 DPR:$tbl1, DPR:$src)))]>;
2701 let hasExtraSrcRegAllocReq = 1 in {
2703 : N3V<1,1,0b11,0b1001,0,0, (outs DPR:$dst),
2704 (ins DPR:$tbl1, DPR:$tbl2, DPR:$src), IIC_VTB2,
2705 "vtbl.8\t$dst, \\{$tbl1,$tbl2\\}, $src", "",
2706 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl2
2707 DPR:$tbl1, DPR:$tbl2, DPR:$src)))]>;
2709 : N3V<1,1,0b11,0b1010,0,0, (outs DPR:$dst),
2710 (ins DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$src), IIC_VTB3,
2711 "vtbl.8\t$dst, \\{$tbl1,$tbl2,$tbl3\\}, $src", "",
2712 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl3
2713 DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$src)))]>;
2715 : N3V<1,1,0b11,0b1011,0,0, (outs DPR:$dst),
2716 (ins DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$src), IIC_VTB4,
2717 "vtbl.8\t$dst, \\{$tbl1,$tbl2,$tbl3,$tbl4\\}, $src", "",
2718 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl4 DPR:$tbl1, DPR:$tbl2,
2719 DPR:$tbl3, DPR:$tbl4, DPR:$src)))]>;
2720 } // hasExtraSrcRegAllocReq = 1
2722 // VTBX : Vector Table Extension
2724 : N3V<1,1,0b11,0b1000,1,0, (outs DPR:$dst),
2725 (ins DPR:$orig, DPR:$tbl1, DPR:$src), IIC_VTBX1,
2726 "vtbx.8\t$dst, \\{$tbl1\\}, $src", "$orig = $dst",
2727 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx1
2728 DPR:$orig, DPR:$tbl1, DPR:$src)))]>;
2729 let hasExtraSrcRegAllocReq = 1 in {
2731 : N3V<1,1,0b11,0b1001,1,0, (outs DPR:$dst),
2732 (ins DPR:$orig, DPR:$tbl1, DPR:$tbl2, DPR:$src), IIC_VTBX2,
2733 "vtbx.8\t$dst, \\{$tbl1,$tbl2\\}, $src", "$orig = $dst",
2734 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx2
2735 DPR:$orig, DPR:$tbl1, DPR:$tbl2, DPR:$src)))]>;
2737 : N3V<1,1,0b11,0b1010,1,0, (outs DPR:$dst),
2738 (ins DPR:$orig, DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$src), IIC_VTBX3,
2739 "vtbx.8\t$dst, \\{$tbl1,$tbl2,$tbl3\\}, $src", "$orig = $dst",
2740 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx3 DPR:$orig, DPR:$tbl1,
2741 DPR:$tbl2, DPR:$tbl3, DPR:$src)))]>;
2743 : N3V<1,1,0b11,0b1011,1,0, (outs DPR:$dst), (ins DPR:$orig, DPR:$tbl1,
2744 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$src), IIC_VTBX4,
2745 "vtbx.8\t$dst, \\{$tbl1,$tbl2,$tbl3,$tbl4\\}, $src", "$orig = $dst",
2746 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx4 DPR:$orig, DPR:$tbl1,
2747 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$src)))]>;
2748 } // hasExtraSrcRegAllocReq = 1
2750 //===----------------------------------------------------------------------===//
2751 // NEON instructions for single-precision FP math
2752 //===----------------------------------------------------------------------===//
2754 // These need separate instructions because they must use DPR_VFP2 register
2755 // class which have SPR sub-registers.
2757 // Vector Add Operations used for single-precision FP
2758 let neverHasSideEffects = 1 in
2759 def VADDfd_sfp : N3VDs<0, 0, 0b00, 0b1101, 0, "vadd.f32", v2f32, v2f32, fadd,1>;
2760 def : N3VDsPat<fadd, VADDfd_sfp>;
2762 // Vector Sub Operations used for single-precision FP
2763 let neverHasSideEffects = 1 in
2764 def VSUBfd_sfp : N3VDs<0, 0, 0b10, 0b1101, 0, "vsub.f32", v2f32, v2f32, fsub,0>;
2765 def : N3VDsPat<fsub, VSUBfd_sfp>;
2767 // Vector Multiply Operations used for single-precision FP
2768 let neverHasSideEffects = 1 in
2769 def VMULfd_sfp : N3VDs<1, 0, 0b00, 0b1101, 1, "vmul.f32", v2f32, v2f32, fmul,1>;
2770 def : N3VDsPat<fmul, VMULfd_sfp>;
2772 // Vector Multiply-Accumulate/Subtract used for single-precision FP
2773 let neverHasSideEffects = 1 in
2774 def VMLAfd_sfp : N3VDMulOps<0, 0, 0b00, 0b1101, 1, IIC_VMACD, "vmla.f32", v2f32,fmul,fadd>;
2775 def : N3VDMulOpsPat<fmul, fadd, VMLAfd_sfp>;
2777 let neverHasSideEffects = 1 in
2778 def VMLSfd_sfp : N3VDMulOps<0, 0, 0b10, 0b1101, 1, IIC_VMACD, "vmls.f32", v2f32,fmul,fsub>;
2779 def : N3VDMulOpsPat<fmul, fsub, VMLSfd_sfp>;
2781 // Vector Absolute used for single-precision FP
2782 let neverHasSideEffects = 1 in
2783 def VABSfd_sfp : N2VDInts<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
2784 IIC_VUNAD, "vabs.f32",
2785 v2f32, v2f32, int_arm_neon_vabs>;
2786 def : N2VDIntsPat<fabs, VABSfd_sfp>;
2788 // Vector Negate used for single-precision FP
2789 let neverHasSideEffects = 1 in
2790 def VNEGf32d_sfp : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
2791 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), IIC_VUNAD,
2792 "vneg.f32\t$dst, $src", "", []>;
2793 def : N2VDIntsPat<fneg, VNEGf32d_sfp>;
2795 // Vector Convert between single-precision FP and integer
2796 let neverHasSideEffects = 1 in
2797 def VCVTf2sd_sfp : N2VDs<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt.s32.f32",
2798 v2i32, v2f32, fp_to_sint>;
2799 def : N2VDsPat<arm_ftosi, f32, v2f32, VCVTf2sd_sfp>;
2801 let neverHasSideEffects = 1 in
2802 def VCVTf2ud_sfp : N2VDs<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt.u32.f32",
2803 v2i32, v2f32, fp_to_uint>;
2804 def : N2VDsPat<arm_ftoui, f32, v2f32, VCVTf2ud_sfp>;
2806 let neverHasSideEffects = 1 in
2807 def VCVTs2fd_sfp : N2VDs<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt.f32.s32",
2808 v2f32, v2i32, sint_to_fp>;
2809 def : N2VDsPat<arm_sitof, f32, v2i32, VCVTs2fd_sfp>;
2811 let neverHasSideEffects = 1 in
2812 def VCVTu2fd_sfp : N2VDs<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt.f32.u32",
2813 v2f32, v2i32, uint_to_fp>;
2814 def : N2VDsPat<arm_uitof, f32, v2i32, VCVTu2fd_sfp>;
2816 //===----------------------------------------------------------------------===//
2817 // Non-Instruction Patterns
2818 //===----------------------------------------------------------------------===//
2821 def : Pat<(v1i64 (bitconvert (v2i32 DPR:$src))), (v1i64 DPR:$src)>;
2822 def : Pat<(v1i64 (bitconvert (v4i16 DPR:$src))), (v1i64 DPR:$src)>;
2823 def : Pat<(v1i64 (bitconvert (v8i8 DPR:$src))), (v1i64 DPR:$src)>;
2824 def : Pat<(v1i64 (bitconvert (f64 DPR:$src))), (v1i64 DPR:$src)>;
2825 def : Pat<(v1i64 (bitconvert (v2f32 DPR:$src))), (v1i64 DPR:$src)>;
2826 def : Pat<(v2i32 (bitconvert (v1i64 DPR:$src))), (v2i32 DPR:$src)>;
2827 def : Pat<(v2i32 (bitconvert (v4i16 DPR:$src))), (v2i32 DPR:$src)>;
2828 def : Pat<(v2i32 (bitconvert (v8i8 DPR:$src))), (v2i32 DPR:$src)>;
2829 def : Pat<(v2i32 (bitconvert (f64 DPR:$src))), (v2i32 DPR:$src)>;
2830 def : Pat<(v2i32 (bitconvert (v2f32 DPR:$src))), (v2i32 DPR:$src)>;
2831 def : Pat<(v4i16 (bitconvert (v1i64 DPR:$src))), (v4i16 DPR:$src)>;
2832 def : Pat<(v4i16 (bitconvert (v2i32 DPR:$src))), (v4i16 DPR:$src)>;
2833 def : Pat<(v4i16 (bitconvert (v8i8 DPR:$src))), (v4i16 DPR:$src)>;
2834 def : Pat<(v4i16 (bitconvert (f64 DPR:$src))), (v4i16 DPR:$src)>;
2835 def : Pat<(v4i16 (bitconvert (v2f32 DPR:$src))), (v4i16 DPR:$src)>;
2836 def : Pat<(v8i8 (bitconvert (v1i64 DPR:$src))), (v8i8 DPR:$src)>;
2837 def : Pat<(v8i8 (bitconvert (v2i32 DPR:$src))), (v8i8 DPR:$src)>;
2838 def : Pat<(v8i8 (bitconvert (v4i16 DPR:$src))), (v8i8 DPR:$src)>;
2839 def : Pat<(v8i8 (bitconvert (f64 DPR:$src))), (v8i8 DPR:$src)>;
2840 def : Pat<(v8i8 (bitconvert (v2f32 DPR:$src))), (v8i8 DPR:$src)>;
2841 def : Pat<(f64 (bitconvert (v1i64 DPR:$src))), (f64 DPR:$src)>;
2842 def : Pat<(f64 (bitconvert (v2i32 DPR:$src))), (f64 DPR:$src)>;
2843 def : Pat<(f64 (bitconvert (v4i16 DPR:$src))), (f64 DPR:$src)>;
2844 def : Pat<(f64 (bitconvert (v8i8 DPR:$src))), (f64 DPR:$src)>;
2845 def : Pat<(f64 (bitconvert (v2f32 DPR:$src))), (f64 DPR:$src)>;
2846 def : Pat<(v2f32 (bitconvert (f64 DPR:$src))), (v2f32 DPR:$src)>;
2847 def : Pat<(v2f32 (bitconvert (v1i64 DPR:$src))), (v2f32 DPR:$src)>;
2848 def : Pat<(v2f32 (bitconvert (v2i32 DPR:$src))), (v2f32 DPR:$src)>;
2849 def : Pat<(v2f32 (bitconvert (v4i16 DPR:$src))), (v2f32 DPR:$src)>;
2850 def : Pat<(v2f32 (bitconvert (v8i8 DPR:$src))), (v2f32 DPR:$src)>;
2852 def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (v2i64 QPR:$src)>;
2853 def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (v2i64 QPR:$src)>;
2854 def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (v2i64 QPR:$src)>;
2855 def : Pat<(v2i64 (bitconvert (v2f64 QPR:$src))), (v2i64 QPR:$src)>;
2856 def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (v2i64 QPR:$src)>;
2857 def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (v4i32 QPR:$src)>;
2858 def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (v4i32 QPR:$src)>;
2859 def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (v4i32 QPR:$src)>;
2860 def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (v4i32 QPR:$src)>;
2861 def : Pat<(v4i32 (bitconvert (v4f32 QPR:$src))), (v4i32 QPR:$src)>;
2862 def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (v8i16 QPR:$src)>;
2863 def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (v8i16 QPR:$src)>;
2864 def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (v8i16 QPR:$src)>;
2865 def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (v8i16 QPR:$src)>;
2866 def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (v8i16 QPR:$src)>;
2867 def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (v16i8 QPR:$src)>;
2868 def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (v16i8 QPR:$src)>;
2869 def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (v16i8 QPR:$src)>;
2870 def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (v16i8 QPR:$src)>;
2871 def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (v16i8 QPR:$src)>;
2872 def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (v4f32 QPR:$src)>;
2873 def : Pat<(v4f32 (bitconvert (v4i32 QPR:$src))), (v4f32 QPR:$src)>;
2874 def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (v4f32 QPR:$src)>;
2875 def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (v4f32 QPR:$src)>;
2876 def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (v4f32 QPR:$src)>;
2877 def : Pat<(v2f64 (bitconvert (v2i64 QPR:$src))), (v2f64 QPR:$src)>;
2878 def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (v2f64 QPR:$src)>;
2879 def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 QPR:$src)>;
2880 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 QPR:$src)>;
2881 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 QPR:$src)>;