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 = $dst3", []>;
297 def VLD3LNd8 : VLD3LN<0b0010, "vld3.8">;
298 def VLD3LNd16 : VLD3LN<0b0110, "vld3.16">;
299 def VLD3LNd32 : VLD3LN<0b1010, "vld3.32">;
301 // vld3 to double-spaced even registers.
302 def VLD3LNq16a: VLD3LN<0b0110, "vld3.16">;
303 def VLD3LNq32a: VLD3LN<0b1010, "vld3.32">;
305 // vld3 to double-spaced odd registers.
306 def VLD3LNq16b: VLD3LN<0b0110, "vld3.16">;
307 def VLD3LNq32b: VLD3LN<0b1010, "vld3.32">;
309 // VLD4LN : Vector Load (single 4-element structure to one lane)
310 class VLD4LN<bits<4> op11_8, string OpcodeStr>
311 : NLdSt<1,0b10,op11_8,0b0000,
312 (outs DPR:$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 = $dst4", []>;
319 def VLD4LNd8 : VLD4LN<0b0011, "vld4.8">;
320 def VLD4LNd16 : VLD4LN<0b0111, "vld4.16">;
321 def VLD4LNd32 : VLD4LN<0b1011, "vld4.32">;
323 // vld4 to double-spaced even registers.
324 def VLD4LNq16a: VLD4LN<0b0111, "vld4.16">;
325 def VLD4LNq32a: VLD4LN<0b1011, "vld4.32">;
327 // vld4 to double-spaced odd registers.
328 def VLD4LNq16b: VLD4LN<0b0111, "vld4.16">;
329 def VLD4LNq32b: VLD4LN<0b1011, "vld4.32">;
331 // VLD1DUP : Vector Load (single element to all lanes)
332 // VLD2DUP : Vector Load (single 2-element structure to all lanes)
333 // VLD3DUP : Vector Load (single 3-element structure to all lanes)
334 // VLD4DUP : Vector Load (single 4-element structure to all lanes)
335 // FIXME: Not yet implemented.
336 } // mayLoad = 1, hasExtraDefRegAllocReq = 1
338 // VST1 : Vector Store (multiple single elements)
339 class VST1D<bits<4> op7_4, string OpcodeStr, ValueType Ty, Intrinsic IntOp>
340 : NLdSt<0,0b00,0b0111,op7_4, (outs), (ins addrmode6:$addr, DPR:$src), IIC_VST,
341 !strconcat(OpcodeStr, "\t\\{$src\\}, $addr"), "",
342 [(IntOp addrmode6:$addr, (Ty DPR:$src))]>;
343 class VST1Q<bits<4> op7_4, string OpcodeStr, ValueType Ty, Intrinsic IntOp>
344 : NLdSt<0,0b00,0b1010,op7_4, (outs), (ins addrmode6:$addr, QPR:$src), IIC_VST,
345 !strconcat(OpcodeStr, "\t${src:dregpair}, $addr"), "",
346 [(IntOp addrmode6:$addr, (Ty QPR:$src))]>;
348 let hasExtraSrcRegAllocReq = 1 in {
349 def VST1d8 : VST1D<0b0000, "vst1.8", v8i8, int_arm_neon_vst1>;
350 def VST1d16 : VST1D<0b0100, "vst1.16", v4i16, int_arm_neon_vst1>;
351 def VST1d32 : VST1D<0b1000, "vst1.32", v2i32, int_arm_neon_vst1>;
352 def VST1df : VST1D<0b1000, "vst1.32", v2f32, int_arm_neon_vst1>;
353 def VST1d64 : VST1D<0b1100, "vst1.64", v1i64, int_arm_neon_vst1>;
355 def VST1q8 : VST1Q<0b0000, "vst1.8", v16i8, int_arm_neon_vst1>;
356 def VST1q16 : VST1Q<0b0100, "vst1.16", v8i16, int_arm_neon_vst1>;
357 def VST1q32 : VST1Q<0b1000, "vst1.32", v4i32, int_arm_neon_vst1>;
358 def VST1qf : VST1Q<0b1000, "vst1.32", v4f32, int_arm_neon_vst1>;
359 def VST1q64 : VST1Q<0b1100, "vst1.64", v2i64, int_arm_neon_vst1>;
360 } // hasExtraSrcRegAllocReq
362 let mayStore = 1, hasExtraSrcRegAllocReq = 1 in {
364 // VST2 : Vector Store (multiple 2-element structures)
365 class VST2D<bits<4> op7_4, string OpcodeStr>
366 : NLdSt<0,0b00,0b1000,op7_4, (outs),
367 (ins addrmode6:$addr, DPR:$src1, DPR:$src2), IIC_VST,
368 !strconcat(OpcodeStr, "\t\\{$src1,$src2\\}, $addr"), "", []>;
369 class VST2Q<bits<4> op7_4, string OpcodeStr>
370 : NLdSt<0,0b00,0b0011,op7_4, (outs),
371 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
373 !strconcat(OpcodeStr, "\t\\{$src1,$src2,$src3,$src4\\}, $addr"),
376 def VST2d8 : VST2D<0b0000, "vst2.8">;
377 def VST2d16 : VST2D<0b0100, "vst2.16">;
378 def VST2d32 : VST2D<0b1000, "vst2.32">;
379 def VST2d64 : NLdSt<0,0b00,0b1010,0b1100, (outs),
380 (ins addrmode6:$addr, DPR:$src1, DPR:$src2), IIC_VST,
381 "vst1.64\t\\{$src1,$src2\\}, $addr", "", []>;
383 def VST2q8 : VST2Q<0b0000, "vst2.8">;
384 def VST2q16 : VST2Q<0b0100, "vst2.16">;
385 def VST2q32 : VST2Q<0b1000, "vst2.32">;
387 // VST3 : Vector Store (multiple 3-element structures)
388 class VST3D<bits<4> op7_4, string OpcodeStr>
389 : NLdSt<0,0b00,0b0100,op7_4, (outs),
390 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3), IIC_VST,
391 !strconcat(OpcodeStr, "\t\\{$src1,$src2,$src3\\}, $addr"), "", []>;
392 class VST3WB<bits<4> op7_4, string OpcodeStr>
393 : NLdSt<0,0b00,0b0101,op7_4, (outs GPR:$wb),
394 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3), IIC_VST,
395 !strconcat(OpcodeStr, "\t\\{$src1,$src2,$src3\\}, $addr"),
396 "$addr.addr = $wb", []>;
398 def VST3d8 : VST3D<0b0000, "vst3.8">;
399 def VST3d16 : VST3D<0b0100, "vst3.16">;
400 def VST3d32 : VST3D<0b1000, "vst3.32">;
401 def VST3d64 : NLdSt<0,0b00,0b0110,0b1100, (outs),
402 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3),
404 "vst1.64\t\\{$src1,$src2,$src3\\}, $addr", "", []>;
406 // vst3 to double-spaced even registers.
407 def VST3q8a : VST3WB<0b0000, "vst3.8">;
408 def VST3q16a : VST3WB<0b0100, "vst3.16">;
409 def VST3q32a : VST3WB<0b1000, "vst3.32">;
411 // vst3 to double-spaced odd registers.
412 def VST3q8b : VST3WB<0b0000, "vst3.8">;
413 def VST3q16b : VST3WB<0b0100, "vst3.16">;
414 def VST3q32b : VST3WB<0b1000, "vst3.32">;
416 // VST4 : Vector Store (multiple 4-element structures)
417 class VST4D<bits<4> op7_4, string OpcodeStr>
418 : NLdSt<0,0b00,0b0000,op7_4, (outs),
419 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
421 !strconcat(OpcodeStr, "\t\\{$src1,$src2,$src3,$src4\\}, $addr"),
423 class VST4WB<bits<4> op7_4, string OpcodeStr>
424 : NLdSt<0,0b00,0b0001,op7_4, (outs GPR:$wb),
425 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
427 !strconcat(OpcodeStr, "\t\\{$src1,$src2,$src3,$src4\\}, $addr"),
428 "$addr.addr = $wb", []>;
430 def VST4d8 : VST4D<0b0000, "vst4.8">;
431 def VST4d16 : VST4D<0b0100, "vst4.16">;
432 def VST4d32 : VST4D<0b1000, "vst4.32">;
433 def VST4d64 : NLdSt<0,0b00,0b0010,0b1100, (outs),
434 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3,
436 "vst1.64\t\\{$src1,$src2,$src3,$src4\\}, $addr", "", []>;
438 // vst4 to double-spaced even registers.
439 def VST4q8a : VST4WB<0b0000, "vst4.8">;
440 def VST4q16a : VST4WB<0b0100, "vst4.16">;
441 def VST4q32a : VST4WB<0b1000, "vst4.32">;
443 // vst4 to double-spaced odd registers.
444 def VST4q8b : VST4WB<0b0000, "vst4.8">;
445 def VST4q16b : VST4WB<0b0100, "vst4.16">;
446 def VST4q32b : VST4WB<0b1000, "vst4.32">;
448 // VST1LN : Vector Store (single element from one lane)
449 // FIXME: Not yet implemented.
451 // VST2LN : Vector Store (single 2-element structure from one lane)
452 class VST2LN<bits<4> op11_8, string OpcodeStr>
453 : NLdSt<1,0b00,op11_8,0b0000, (outs),
454 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, nohash_imm:$lane),
456 !strconcat(OpcodeStr, "\t\\{$src1[$lane],$src2[$lane]\\}, $addr"),
459 def VST2LNd8 : VST2LN<0b0000, "vst2.8">;
460 def VST2LNd16 : VST2LN<0b0100, "vst2.16">;
461 def VST2LNd32 : VST2LN<0b1000, "vst2.32">;
463 // vst2 to double-spaced even registers.
464 def VST2LNq16a: VST2LN<0b0100, "vst2.16">;
465 def VST2LNq32a: VST2LN<0b1000, "vst2.32">;
467 // vst2 to double-spaced odd registers.
468 def VST2LNq16b: VST2LN<0b0100, "vst2.16">;
469 def VST2LNq32b: VST2LN<0b1000, "vst2.32">;
471 // VST3LN : Vector Store (single 3-element structure from one lane)
472 class VST3LND<bits<4> op11_8, string OpcodeStr>
473 : NLdSt<1,0b00,op11_8,0b0000, (outs),
474 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3,
475 nohash_imm:$lane), IIC_VST,
476 !strconcat(OpcodeStr,
477 "\t\\{$src1[$lane],$src2[$lane],$src3[$lane]\\}, $addr"), "", []>;
479 def VST3LNd8 : VST3LND<0b0010, "vst3.8">;
480 def VST3LNd16 : VST3LND<0b0110, "vst3.16">;
481 def VST3LNd32 : VST3LND<0b1010, "vst3.32">;
483 // VST4LN : Vector Store (single 4-element structure from one lane)
484 class VST4LND<bits<4> op11_8, string OpcodeStr>
485 : NLdSt<1,0b00,op11_8,0b0000, (outs),
486 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
487 nohash_imm:$lane), IIC_VST,
488 !strconcat(OpcodeStr,
489 "\t\\{$src1[$lane],$src2[$lane],$src3[$lane],$src4[$lane]\\}, $addr"),
492 def VST4LNd8 : VST4LND<0b0011, "vst4.8">;
493 def VST4LNd16 : VST4LND<0b0111, "vst4.16">;
494 def VST4LNd32 : VST4LND<0b1011, "vst4.32">;
495 } // mayStore = 1, hasExtraSrcRegAllocReq = 1
498 //===----------------------------------------------------------------------===//
499 // NEON pattern fragments
500 //===----------------------------------------------------------------------===//
502 // Extract D sub-registers of Q registers.
503 // (arm_dsubreg_0 is 5; arm_dsubreg_1 is 6)
504 def DSubReg_i8_reg : SDNodeXForm<imm, [{
505 return CurDAG->getTargetConstant(5 + N->getZExtValue() / 8, MVT::i32);
507 def DSubReg_i16_reg : SDNodeXForm<imm, [{
508 return CurDAG->getTargetConstant(5 + N->getZExtValue() / 4, MVT::i32);
510 def DSubReg_i32_reg : SDNodeXForm<imm, [{
511 return CurDAG->getTargetConstant(5 + N->getZExtValue() / 2, MVT::i32);
513 def DSubReg_f64_reg : SDNodeXForm<imm, [{
514 return CurDAG->getTargetConstant(5 + N->getZExtValue(), MVT::i32);
516 def DSubReg_f64_other_reg : SDNodeXForm<imm, [{
517 return CurDAG->getTargetConstant(5 + (1 - N->getZExtValue()), MVT::i32);
520 // Extract S sub-registers of Q/D registers.
521 // (arm_ssubreg_0 is 1; arm_ssubreg_1 is 2; etc.)
522 def SSubReg_f32_reg : SDNodeXForm<imm, [{
523 return CurDAG->getTargetConstant(1 + N->getZExtValue(), MVT::i32);
526 // Translate lane numbers from Q registers to D subregs.
527 def SubReg_i8_lane : SDNodeXForm<imm, [{
528 return CurDAG->getTargetConstant(N->getZExtValue() & 7, MVT::i32);
530 def SubReg_i16_lane : SDNodeXForm<imm, [{
531 return CurDAG->getTargetConstant(N->getZExtValue() & 3, MVT::i32);
533 def SubReg_i32_lane : SDNodeXForm<imm, [{
534 return CurDAG->getTargetConstant(N->getZExtValue() & 1, MVT::i32);
537 //===----------------------------------------------------------------------===//
538 // Instruction Classes
539 //===----------------------------------------------------------------------===//
541 // Basic 2-register operations, both double- and quad-register.
542 class N2VD<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
543 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
544 ValueType ResTy, ValueType OpTy, SDNode OpNode>
545 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
546 (ins DPR:$src), IIC_VUNAD, !strconcat(OpcodeStr, "\t$dst, $src"), "",
547 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src))))]>;
548 class N2VQ<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
549 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
550 ValueType ResTy, ValueType OpTy, SDNode OpNode>
551 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
552 (ins QPR:$src), IIC_VUNAQ, !strconcat(OpcodeStr, "\t$dst, $src"), "",
553 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src))))]>;
555 // Basic 2-register operations, scalar single-precision.
556 class N2VDs<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
557 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
558 ValueType ResTy, ValueType OpTy, SDNode OpNode>
559 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
560 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src),
561 IIC_VUNAD, !strconcat(OpcodeStr, "\t$dst, $src"), "", []>;
563 class N2VDsPat<SDNode OpNode, ValueType ResTy, ValueType OpTy, NeonI Inst>
564 : NEONFPPat<(ResTy (OpNode SPR:$a)),
566 (Inst (INSERT_SUBREG (OpTy (IMPLICIT_DEF)), SPR:$a, arm_ssubreg_0)),
569 // Basic 2-register intrinsics, both double- and quad-register.
570 class N2VDInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
571 bits<2> op17_16, bits<5> op11_7, bit op4,
572 InstrItinClass itin, string OpcodeStr,
573 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
574 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
575 (ins DPR:$src), itin, !strconcat(OpcodeStr, "\t$dst, $src"), "",
576 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
577 class N2VQInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
578 bits<2> op17_16, bits<5> op11_7, bit op4,
579 InstrItinClass itin, string OpcodeStr,
580 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
581 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
582 (ins QPR:$src), itin, !strconcat(OpcodeStr, "\t$dst, $src"), "",
583 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
585 // Basic 2-register intrinsics, scalar single-precision
586 class N2VDInts<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
587 bits<2> op17_16, bits<5> op11_7, bit op4,
588 InstrItinClass itin, string OpcodeStr,
589 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
590 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
591 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), itin,
592 !strconcat(OpcodeStr, "\t$dst, $src"), "", []>;
594 class N2VDIntsPat<SDNode OpNode, NeonI Inst>
595 : NEONFPPat<(f32 (OpNode SPR:$a)),
597 (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$a, arm_ssubreg_0)),
600 // Narrow 2-register intrinsics.
601 class N2VNInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
602 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
603 InstrItinClass itin, string OpcodeStr,
604 ValueType TyD, ValueType TyQ, Intrinsic IntOp>
605 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$dst),
606 (ins QPR:$src), itin, !strconcat(OpcodeStr, "\t$dst, $src"), "",
607 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src))))]>;
609 // Long 2-register intrinsics. (This is currently only used for VMOVL and is
610 // derived from N2VImm instead of N2V because of the way the size is encoded.)
611 class N2VLInt<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
612 bit op6, bit op4, InstrItinClass itin, string OpcodeStr,
613 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
614 : N2VImm<op24, op23, op21_16, op11_8, op7, op6, op4, (outs QPR:$dst),
615 (ins DPR:$src), itin, !strconcat(OpcodeStr, "\t$dst, $src"), "",
616 [(set QPR:$dst, (TyQ (IntOp (TyD DPR:$src))))]>;
618 // 2-register shuffles (VTRN/VZIP/VUZP), both double- and quad-register.
619 class N2VDShuffle<bits<2> op19_18, bits<5> op11_7, string OpcodeStr>
620 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 0, 0, (outs DPR:$dst1, DPR:$dst2),
621 (ins DPR:$src1, DPR:$src2), IIC_VPERMD,
622 !strconcat(OpcodeStr, "\t$dst1, $dst2"),
623 "$src1 = $dst1, $src2 = $dst2", []>;
624 class N2VQShuffle<bits<2> op19_18, bits<5> op11_7,
625 InstrItinClass itin, string OpcodeStr>
626 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 1, 0, (outs QPR:$dst1, QPR:$dst2),
627 (ins QPR:$src1, QPR:$src2), itin,
628 !strconcat(OpcodeStr, "\t$dst1, $dst2"),
629 "$src1 = $dst1, $src2 = $dst2", []>;
631 // Basic 3-register operations, both double- and quad-register.
632 class N3VD<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
633 InstrItinClass itin, string OpcodeStr, ValueType ResTy, ValueType OpTy,
634 SDNode OpNode, bit Commutable>
635 : N3V<op24, op23, op21_20, op11_8, 0, op4,
636 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), itin,
637 !strconcat(OpcodeStr, "\t$dst, $src1, $src2"), "",
638 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src1), (OpTy DPR:$src2))))]> {
639 let isCommutable = Commutable;
641 class N3VDSL<bits<2> op21_20, bits<4> op11_8,
642 InstrItinClass itin, string OpcodeStr, ValueType Ty, SDNode ShOp>
643 : N3V<0, 1, op21_20, op11_8, 1, 0,
644 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
645 itin, !strconcat(OpcodeStr, "\t$dst, $src1, $src2[$lane]"), "",
647 (Ty (ShOp (Ty DPR:$src1),
648 (Ty (NEONvduplane (Ty DPR_VFP2:$src2),
650 let isCommutable = 0;
652 class N3VDSL16<bits<2> op21_20, bits<4> op11_8,
653 string OpcodeStr, ValueType Ty, SDNode ShOp>
654 : N3V<0, 1, op21_20, op11_8, 1, 0,
655 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
657 !strconcat(OpcodeStr, "\t$dst, $src1, $src2[$lane]"), "",
659 (Ty (ShOp (Ty DPR:$src1),
660 (Ty (NEONvduplane (Ty DPR_8:$src2),
662 let isCommutable = 0;
665 class N3VQ<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
666 InstrItinClass itin, string OpcodeStr, ValueType ResTy, ValueType OpTy,
667 SDNode OpNode, bit Commutable>
668 : N3V<op24, op23, op21_20, op11_8, 1, op4,
669 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), itin,
670 !strconcat(OpcodeStr, "\t$dst, $src1, $src2"), "",
671 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src1), (OpTy QPR:$src2))))]> {
672 let isCommutable = Commutable;
674 class N3VQSL<bits<2> op21_20, bits<4> op11_8,
675 InstrItinClass itin, string OpcodeStr,
676 ValueType ResTy, ValueType OpTy, SDNode ShOp>
677 : N3V<1, 1, op21_20, op11_8, 1, 0,
678 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
679 itin, !strconcat(OpcodeStr, "\t$dst, $src1, $src2[$lane]"), "",
680 [(set (ResTy QPR:$dst),
681 (ResTy (ShOp (ResTy QPR:$src1),
682 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
684 let isCommutable = 0;
686 class N3VQSL16<bits<2> op21_20, bits<4> op11_8,
687 string OpcodeStr, ValueType ResTy, ValueType OpTy, SDNode ShOp>
688 : N3V<1, 1, op21_20, op11_8, 1, 0,
689 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
691 !strconcat(OpcodeStr, "\t$dst, $src1, $src2[$lane]"), "",
692 [(set (ResTy QPR:$dst),
693 (ResTy (ShOp (ResTy QPR:$src1),
694 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
696 let isCommutable = 0;
699 // Basic 3-register operations, scalar single-precision
700 class N3VDs<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
701 string OpcodeStr, ValueType ResTy, ValueType OpTy,
702 SDNode OpNode, bit Commutable>
703 : N3V<op24, op23, op21_20, op11_8, 0, op4,
704 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src1, DPR_VFP2:$src2), IIC_VBIND,
705 !strconcat(OpcodeStr, "\t$dst, $src1, $src2"), "", []> {
706 let isCommutable = Commutable;
708 class N3VDsPat<SDNode OpNode, NeonI Inst>
709 : NEONFPPat<(f32 (OpNode SPR:$a, SPR:$b)),
711 (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$a, arm_ssubreg_0),
712 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$b, arm_ssubreg_0)),
715 // Basic 3-register intrinsics, both double- and quad-register.
716 class N3VDInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
717 InstrItinClass itin, string OpcodeStr, ValueType ResTy, ValueType OpTy,
718 Intrinsic IntOp, bit Commutable>
719 : N3V<op24, op23, op21_20, op11_8, 0, op4,
720 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), itin,
721 !strconcat(OpcodeStr, "\t$dst, $src1, $src2"), "",
722 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src1), (OpTy DPR:$src2))))]> {
723 let isCommutable = Commutable;
725 class N3VDIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
726 string OpcodeStr, ValueType Ty, Intrinsic IntOp>
727 : N3V<0, 1, op21_20, op11_8, 1, 0,
728 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
729 itin, !strconcat(OpcodeStr, "\t$dst, $src1, $src2[$lane]"), "",
731 (Ty (IntOp (Ty DPR:$src1),
732 (Ty (NEONvduplane (Ty DPR_VFP2:$src2),
734 let isCommutable = 0;
736 class N3VDIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
737 string OpcodeStr, ValueType Ty, Intrinsic IntOp>
738 : N3V<0, 1, op21_20, op11_8, 1, 0,
739 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
740 itin, !strconcat(OpcodeStr, "\t$dst, $src1, $src2[$lane]"), "",
742 (Ty (IntOp (Ty DPR:$src1),
743 (Ty (NEONvduplane (Ty DPR_8:$src2),
745 let isCommutable = 0;
748 class N3VQInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
749 InstrItinClass itin, string OpcodeStr, ValueType ResTy, ValueType OpTy,
750 Intrinsic IntOp, bit Commutable>
751 : N3V<op24, op23, op21_20, op11_8, 1, op4,
752 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), itin,
753 !strconcat(OpcodeStr, "\t$dst, $src1, $src2"), "",
754 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src1), (OpTy QPR:$src2))))]> {
755 let isCommutable = Commutable;
757 class N3VQIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
758 string OpcodeStr, ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
759 : N3V<1, 1, op21_20, op11_8, 1, 0,
760 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
761 itin, !strconcat(OpcodeStr, "\t$dst, $src1, $src2[$lane]"), "",
762 [(set (ResTy QPR:$dst),
763 (ResTy (IntOp (ResTy QPR:$src1),
764 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
766 let isCommutable = 0;
768 class N3VQIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
769 string OpcodeStr, ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
770 : N3V<1, 1, op21_20, op11_8, 1, 0,
771 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
772 itin, !strconcat(OpcodeStr, "\t$dst, $src1, $src2[$lane]"), "",
773 [(set (ResTy QPR:$dst),
774 (ResTy (IntOp (ResTy QPR:$src1),
775 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
777 let isCommutable = 0;
780 // Multiply-Add/Sub operations, both double- and quad-register.
781 class N3VDMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
782 InstrItinClass itin, string OpcodeStr,
783 ValueType Ty, SDNode MulOp, SDNode OpNode>
784 : N3V<op24, op23, op21_20, op11_8, 0, op4,
785 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3), itin,
786 !strconcat(OpcodeStr, "\t$dst, $src2, $src3"), "$src1 = $dst",
787 [(set DPR:$dst, (Ty (OpNode DPR:$src1,
788 (Ty (MulOp DPR:$src2, DPR:$src3)))))]>;
789 class N3VDMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
790 string OpcodeStr, ValueType Ty, SDNode MulOp, SDNode ShOp>
791 : N3V<0, 1, op21_20, op11_8, 1, 0,
793 (ins DPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane), itin,
794 !strconcat(OpcodeStr, "\t$dst, $src2, $src3[$lane]"), "$src1 = $dst",
796 (Ty (ShOp (Ty DPR:$src1),
797 (Ty (MulOp DPR:$src2,
798 (Ty (NEONvduplane (Ty DPR_VFP2:$src3),
800 class N3VDMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
801 string OpcodeStr, ValueType Ty, SDNode MulOp, SDNode ShOp>
802 : N3V<0, 1, op21_20, op11_8, 1, 0,
804 (ins DPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane), itin,
805 !strconcat(OpcodeStr, "\t$dst, $src2, $src3[$lane]"), "$src1 = $dst",
807 (Ty (ShOp (Ty DPR:$src1),
808 (Ty (MulOp DPR:$src2,
809 (Ty (NEONvduplane (Ty DPR_8:$src3),
812 class N3VQMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
813 InstrItinClass itin, string OpcodeStr, ValueType Ty,
814 SDNode MulOp, SDNode OpNode>
815 : N3V<op24, op23, op21_20, op11_8, 1, op4,
816 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3), itin,
817 !strconcat(OpcodeStr, "\t$dst, $src2, $src3"), "$src1 = $dst",
818 [(set QPR:$dst, (Ty (OpNode QPR:$src1,
819 (Ty (MulOp QPR:$src2, QPR:$src3)))))]>;
820 class N3VQMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
821 string OpcodeStr, ValueType ResTy, ValueType OpTy,
822 SDNode MulOp, SDNode ShOp>
823 : N3V<1, 1, op21_20, op11_8, 1, 0,
825 (ins QPR:$src1, QPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane), itin,
826 !strconcat(OpcodeStr, "\t$dst, $src2, $src3[$lane]"), "$src1 = $dst",
827 [(set (ResTy QPR:$dst),
828 (ResTy (ShOp (ResTy QPR:$src1),
829 (ResTy (MulOp QPR:$src2,
830 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src3),
832 class N3VQMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
833 string OpcodeStr, ValueType ResTy, ValueType OpTy,
834 SDNode MulOp, SDNode ShOp>
835 : N3V<1, 1, op21_20, op11_8, 1, 0,
837 (ins QPR:$src1, QPR:$src2, DPR_8:$src3, nohash_imm:$lane), itin,
838 !strconcat(OpcodeStr, "\t$dst, $src2, $src3[$lane]"), "$src1 = $dst",
839 [(set (ResTy QPR:$dst),
840 (ResTy (ShOp (ResTy QPR:$src1),
841 (ResTy (MulOp QPR:$src2,
842 (ResTy (NEONvduplane (OpTy DPR_8:$src3),
845 // Multiply-Add/Sub operations, scalar single-precision
846 class N3VDMulOps<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
847 InstrItinClass itin, string OpcodeStr,
848 ValueType Ty, SDNode MulOp, SDNode OpNode>
849 : N3V<op24, op23, op21_20, op11_8, 0, op4,
850 (outs DPR_VFP2:$dst),
851 (ins DPR_VFP2:$src1, DPR_VFP2:$src2, DPR_VFP2:$src3), itin,
852 !strconcat(OpcodeStr, "\t$dst, $src2, $src3"), "$src1 = $dst", []>;
854 class N3VDMulOpsPat<SDNode MulNode, SDNode OpNode, NeonI Inst>
855 : NEONFPPat<(f32 (OpNode SPR:$acc, (f32 (MulNode SPR:$a, SPR:$b)))),
857 (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$acc, arm_ssubreg_0),
858 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$a, arm_ssubreg_0),
859 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$b, arm_ssubreg_0)),
862 // Neon 3-argument intrinsics, both double- and quad-register.
863 // The destination register is also used as the first source operand register.
864 class N3VDInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
865 InstrItinClass itin, string OpcodeStr,
866 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
867 : N3V<op24, op23, op21_20, op11_8, 0, op4,
868 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3), itin,
869 !strconcat(OpcodeStr, "\t$dst, $src2, $src3"), "$src1 = $dst",
870 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src1),
871 (OpTy DPR:$src2), (OpTy DPR:$src3))))]>;
872 class N3VQInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
873 InstrItinClass itin, string OpcodeStr,
874 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
875 : N3V<op24, op23, op21_20, op11_8, 1, op4,
876 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3), itin,
877 !strconcat(OpcodeStr, "\t$dst, $src2, $src3"), "$src1 = $dst",
878 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src1),
879 (OpTy QPR:$src2), (OpTy QPR:$src3))))]>;
881 // Neon Long 3-argument intrinsic. The destination register is
882 // a quad-register and is also used as the first source operand register.
883 class N3VLInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
884 InstrItinClass itin, string OpcodeStr,
885 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
886 : N3V<op24, op23, op21_20, op11_8, 0, op4,
887 (outs QPR:$dst), (ins QPR:$src1, DPR:$src2, DPR:$src3), itin,
888 !strconcat(OpcodeStr, "\t$dst, $src2, $src3"), "$src1 = $dst",
890 (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$src2), (TyD DPR:$src3))))]>;
891 class N3VLInt3SL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
892 string OpcodeStr, ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
893 : N3V<op24, 1, op21_20, op11_8, 1, 0,
895 (ins QPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane), itin,
896 !strconcat(OpcodeStr, "\t$dst, $src2, $src3[$lane]"), "$src1 = $dst",
897 [(set (ResTy QPR:$dst),
898 (ResTy (IntOp (ResTy QPR:$src1),
900 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src3),
902 class N3VLInt3SL16<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
903 string OpcodeStr, ValueType ResTy, ValueType OpTy,
905 : N3V<op24, 1, op21_20, op11_8, 1, 0,
907 (ins QPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane), itin,
908 !strconcat(OpcodeStr, "\t$dst, $src2, $src3[$lane]"), "$src1 = $dst",
909 [(set (ResTy QPR:$dst),
910 (ResTy (IntOp (ResTy QPR:$src1),
912 (OpTy (NEONvduplane (OpTy DPR_8:$src3),
916 // Narrowing 3-register intrinsics.
917 class N3VNInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
918 string OpcodeStr, ValueType TyD, ValueType TyQ,
919 Intrinsic IntOp, bit Commutable>
920 : N3V<op24, op23, op21_20, op11_8, 0, op4,
921 (outs DPR:$dst), (ins QPR:$src1, QPR:$src2), IIC_VBINi4D,
922 !strconcat(OpcodeStr, "\t$dst, $src1, $src2"), "",
923 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src1), (TyQ QPR:$src2))))]> {
924 let isCommutable = Commutable;
927 // Long 3-register intrinsics.
928 class N3VLInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
929 InstrItinClass itin, string OpcodeStr, ValueType TyQ, ValueType TyD,
930 Intrinsic IntOp, bit Commutable>
931 : N3V<op24, op23, op21_20, op11_8, 0, op4,
932 (outs QPR:$dst), (ins DPR:$src1, DPR:$src2), itin,
933 !strconcat(OpcodeStr, "\t$dst, $src1, $src2"), "",
934 [(set QPR:$dst, (TyQ (IntOp (TyD DPR:$src1), (TyD DPR:$src2))))]> {
935 let isCommutable = Commutable;
937 class N3VLIntSL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
938 string OpcodeStr, ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
939 : N3V<op24, 1, op21_20, op11_8, 1, 0,
940 (outs QPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
941 itin, !strconcat(OpcodeStr, "\t$dst, $src1, $src2[$lane]"), "",
942 [(set (ResTy QPR:$dst),
943 (ResTy (IntOp (OpTy DPR:$src1),
944 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src2),
946 class N3VLIntSL16<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
947 string OpcodeStr, ValueType ResTy, ValueType OpTy,
949 : N3V<op24, 1, op21_20, op11_8, 1, 0,
950 (outs QPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
951 itin, !strconcat(OpcodeStr, "\t$dst, $src1, $src2[$lane]"), "",
952 [(set (ResTy QPR:$dst),
953 (ResTy (IntOp (OpTy DPR:$src1),
954 (OpTy (NEONvduplane (OpTy DPR_8:$src2),
957 // Wide 3-register intrinsics.
958 class N3VWInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
959 string OpcodeStr, ValueType TyQ, ValueType TyD,
960 Intrinsic IntOp, bit Commutable>
961 : N3V<op24, op23, op21_20, op11_8, 0, op4,
962 (outs QPR:$dst), (ins QPR:$src1, DPR:$src2), IIC_VSUBiD,
963 !strconcat(OpcodeStr, "\t$dst, $src1, $src2"), "",
964 [(set QPR:$dst, (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$src2))))]> {
965 let isCommutable = Commutable;
968 // Pairwise long 2-register intrinsics, both double- and quad-register.
969 class N2VDPLInt<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, (outs DPR:$dst),
973 (ins DPR:$src), IIC_VSHLiD, !strconcat(OpcodeStr, "\t$dst, $src"), "",
974 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
975 class N2VQPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
976 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
977 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
978 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
979 (ins QPR:$src), IIC_VSHLiD, !strconcat(OpcodeStr, "\t$dst, $src"), "",
980 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
982 // Pairwise long 2-register accumulate intrinsics,
983 // both double- and quad-register.
984 // The destination register is also used as the first source operand register.
985 class N2VDPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
986 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
987 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
988 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
989 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), IIC_VPALiD,
990 !strconcat(OpcodeStr, "\t$dst, $src2"), "$src1 = $dst",
991 [(set DPR:$dst, (ResTy (IntOp (ResTy DPR:$src1), (OpTy DPR:$src2))))]>;
992 class N2VQPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
993 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
994 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
995 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4,
996 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), IIC_VPALiQ,
997 !strconcat(OpcodeStr, "\t$dst, $src2"), "$src1 = $dst",
998 [(set QPR:$dst, (ResTy (IntOp (ResTy QPR:$src1), (OpTy QPR:$src2))))]>;
1000 // Shift by immediate,
1001 // both double- and quad-register.
1002 class N2VDSh<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
1003 bit op4, InstrItinClass itin, string OpcodeStr,
1004 ValueType Ty, SDNode OpNode>
1005 : N2VImm<op24, op23, op21_16, op11_8, op7, 0, op4,
1006 (outs DPR:$dst), (ins DPR:$src, i32imm:$SIMM), itin,
1007 !strconcat(OpcodeStr, "\t$dst, $src, $SIMM"), "",
1008 [(set DPR:$dst, (Ty (OpNode (Ty DPR:$src), (i32 imm:$SIMM))))]>;
1009 class N2VQSh<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
1010 bit op4, InstrItinClass itin, string OpcodeStr,
1011 ValueType Ty, SDNode OpNode>
1012 : N2VImm<op24, op23, op21_16, op11_8, op7, 1, op4,
1013 (outs QPR:$dst), (ins QPR:$src, i32imm:$SIMM), itin,
1014 !strconcat(OpcodeStr, "\t$dst, $src, $SIMM"), "",
1015 [(set QPR:$dst, (Ty (OpNode (Ty QPR:$src), (i32 imm:$SIMM))))]>;
1017 // Long shift by immediate.
1018 class N2VLSh<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
1019 bit op6, bit op4, string OpcodeStr, ValueType ResTy,
1020 ValueType OpTy, SDNode OpNode>
1021 : N2VImm<op24, op23, op21_16, op11_8, op7, op6, op4,
1022 (outs QPR:$dst), (ins DPR:$src, i32imm:$SIMM), IIC_VSHLiD,
1023 !strconcat(OpcodeStr, "\t$dst, $src, $SIMM"), "",
1024 [(set QPR:$dst, (ResTy (OpNode (OpTy DPR:$src),
1025 (i32 imm:$SIMM))))]>;
1027 // Narrow shift by immediate.
1028 class N2VNSh<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
1029 bit op6, bit op4, InstrItinClass itin, string OpcodeStr,
1030 ValueType ResTy, ValueType OpTy, SDNode OpNode>
1031 : N2VImm<op24, op23, op21_16, op11_8, op7, op6, op4,
1032 (outs DPR:$dst), (ins QPR:$src, i32imm:$SIMM), itin,
1033 !strconcat(OpcodeStr, "\t$dst, $src, $SIMM"), "",
1034 [(set DPR:$dst, (ResTy (OpNode (OpTy QPR:$src),
1035 (i32 imm:$SIMM))))]>;
1037 // Shift right by immediate and accumulate,
1038 // both double- and quad-register.
1039 class N2VDShAdd<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
1040 bit op4, string OpcodeStr, ValueType Ty, SDNode ShOp>
1041 : N2VImm<op24, op23, op21_16, op11_8, op7, 0, op4,
1042 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, i32imm:$SIMM),
1044 !strconcat(OpcodeStr, "\t$dst, $src2, $SIMM"), "$src1 = $dst",
1045 [(set DPR:$dst, (Ty (add DPR:$src1,
1046 (Ty (ShOp DPR:$src2, (i32 imm:$SIMM))))))]>;
1047 class N2VQShAdd<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
1048 bit op4, string OpcodeStr, ValueType Ty, SDNode ShOp>
1049 : N2VImm<op24, op23, op21_16, op11_8, op7, 1, op4,
1050 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, i32imm:$SIMM),
1052 !strconcat(OpcodeStr, "\t$dst, $src2, $SIMM"), "$src1 = $dst",
1053 [(set QPR:$dst, (Ty (add QPR:$src1,
1054 (Ty (ShOp QPR:$src2, (i32 imm:$SIMM))))))]>;
1056 // Shift by immediate and insert,
1057 // both double- and quad-register.
1058 class N2VDShIns<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
1059 bit op4, string OpcodeStr, ValueType Ty, SDNode ShOp>
1060 : N2VImm<op24, op23, op21_16, op11_8, op7, 0, op4,
1061 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, i32imm:$SIMM),
1063 !strconcat(OpcodeStr, "\t$dst, $src2, $SIMM"), "$src1 = $dst",
1064 [(set DPR:$dst, (Ty (ShOp DPR:$src1, DPR:$src2, (i32 imm:$SIMM))))]>;
1065 class N2VQShIns<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
1066 bit op4, string OpcodeStr, ValueType Ty, SDNode ShOp>
1067 : N2VImm<op24, op23, op21_16, op11_8, op7, 1, op4,
1068 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, i32imm:$SIMM),
1070 !strconcat(OpcodeStr, "\t$dst, $src2, $SIMM"), "$src1 = $dst",
1071 [(set QPR:$dst, (Ty (ShOp QPR:$src1, QPR:$src2, (i32 imm:$SIMM))))]>;
1073 // Convert, with fractional bits immediate,
1074 // both double- and quad-register.
1075 class N2VCvtD<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
1076 bit op4, string OpcodeStr, ValueType ResTy, ValueType OpTy,
1078 : N2VImm<op24, op23, op21_16, op11_8, op7, 0, op4,
1079 (outs DPR:$dst), (ins DPR:$src, i32imm:$SIMM), IIC_VUNAD,
1080 !strconcat(OpcodeStr, "\t$dst, $src, $SIMM"), "",
1081 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src), (i32 imm:$SIMM))))]>;
1082 class N2VCvtQ<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
1083 bit op4, string OpcodeStr, ValueType ResTy, ValueType OpTy,
1085 : N2VImm<op24, op23, op21_16, op11_8, op7, 1, op4,
1086 (outs QPR:$dst), (ins QPR:$src, i32imm:$SIMM), IIC_VUNAQ,
1087 !strconcat(OpcodeStr, "\t$dst, $src, $SIMM"), "",
1088 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src), (i32 imm:$SIMM))))]>;
1090 //===----------------------------------------------------------------------===//
1092 //===----------------------------------------------------------------------===//
1094 // Abbreviations used in multiclass suffixes:
1095 // Q = quarter int (8 bit) elements
1096 // H = half int (16 bit) elements
1097 // S = single int (32 bit) elements
1098 // D = double int (64 bit) elements
1100 // Neon 3-register vector operations.
1102 // First with only element sizes of 8, 16 and 32 bits:
1103 multiclass N3V_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1104 InstrItinClass itinD16, InstrItinClass itinD32,
1105 InstrItinClass itinQ16, InstrItinClass itinQ32,
1106 string OpcodeStr, SDNode OpNode, bit Commutable = 0> {
1107 // 64-bit vector types.
1108 def v8i8 : N3VD<op24, op23, 0b00, op11_8, op4, itinD16,
1109 !strconcat(OpcodeStr, "8"), v8i8, v8i8, OpNode, Commutable>;
1110 def v4i16 : N3VD<op24, op23, 0b01, op11_8, op4, itinD16,
1111 !strconcat(OpcodeStr, "16"), v4i16, v4i16, OpNode, Commutable>;
1112 def v2i32 : N3VD<op24, op23, 0b10, op11_8, op4, itinD32,
1113 !strconcat(OpcodeStr, "32"), v2i32, v2i32, OpNode, Commutable>;
1115 // 128-bit vector types.
1116 def v16i8 : N3VQ<op24, op23, 0b00, op11_8, op4, itinQ16,
1117 !strconcat(OpcodeStr, "8"), v16i8, v16i8, OpNode, Commutable>;
1118 def v8i16 : N3VQ<op24, op23, 0b01, op11_8, op4, itinQ16,
1119 !strconcat(OpcodeStr, "16"), v8i16, v8i16, OpNode, Commutable>;
1120 def v4i32 : N3VQ<op24, op23, 0b10, op11_8, op4, itinQ32,
1121 !strconcat(OpcodeStr, "32"), v4i32, v4i32, OpNode, Commutable>;
1124 multiclass N3VSL_HS<bits<4> op11_8, string OpcodeStr, SDNode ShOp> {
1125 def v4i16 : N3VDSL16<0b01, op11_8, !strconcat(OpcodeStr, "16"), v4i16, ShOp>;
1126 def v2i32 : N3VDSL<0b10, op11_8, IIC_VMULi32D, !strconcat(OpcodeStr, "32"), v2i32, ShOp>;
1127 def v8i16 : N3VQSL16<0b01, op11_8, !strconcat(OpcodeStr, "16"), v8i16, v4i16, ShOp>;
1128 def v4i32 : N3VQSL<0b10, op11_8, IIC_VMULi32Q, !strconcat(OpcodeStr, "32"), v4i32, v2i32, ShOp>;
1131 // ....then also with element size 64 bits:
1132 multiclass N3V_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1133 InstrItinClass itinD, InstrItinClass itinQ,
1134 string OpcodeStr, SDNode OpNode, bit Commutable = 0>
1135 : N3V_QHS<op24, op23, op11_8, op4, itinD, itinD, itinQ, itinQ,
1136 OpcodeStr, OpNode, Commutable> {
1137 def v1i64 : N3VD<op24, op23, 0b11, op11_8, op4, itinD,
1138 !strconcat(OpcodeStr, "64"), v1i64, v1i64, OpNode, Commutable>;
1139 def v2i64 : N3VQ<op24, op23, 0b11, op11_8, op4, itinQ,
1140 !strconcat(OpcodeStr, "64"), v2i64, v2i64, OpNode, Commutable>;
1144 // Neon Narrowing 2-register vector intrinsics,
1145 // source operand element sizes of 16, 32 and 64 bits:
1146 multiclass N2VNInt_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1147 bits<5> op11_7, bit op6, bit op4,
1148 InstrItinClass itin, string OpcodeStr,
1150 def v8i8 : N2VNInt<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
1151 itin, !strconcat(OpcodeStr, "16"), v8i8, v8i16, IntOp>;
1152 def v4i16 : N2VNInt<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
1153 itin, !strconcat(OpcodeStr, "32"), v4i16, v4i32, IntOp>;
1154 def v2i32 : N2VNInt<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
1155 itin, !strconcat(OpcodeStr, "64"), v2i32, v2i64, IntOp>;
1159 // Neon Lengthening 2-register vector intrinsic (currently specific to VMOVL).
1160 // source operand element sizes of 16, 32 and 64 bits:
1161 multiclass N2VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
1162 bit op4, string OpcodeStr, Intrinsic IntOp> {
1163 def v8i16 : N2VLInt<op24, op23, 0b001000, op11_8, op7, op6, op4,
1164 IIC_VQUNAiD, !strconcat(OpcodeStr, "8"), v8i16, v8i8, IntOp>;
1165 def v4i32 : N2VLInt<op24, op23, 0b010000, op11_8, op7, op6, op4,
1166 IIC_VQUNAiD, !strconcat(OpcodeStr, "16"), v4i32, v4i16, IntOp>;
1167 def v2i64 : N2VLInt<op24, op23, 0b100000, op11_8, op7, op6, op4,
1168 IIC_VQUNAiD, !strconcat(OpcodeStr, "32"), v2i64, v2i32, IntOp>;
1172 // Neon 3-register vector intrinsics.
1174 // First with only element sizes of 16 and 32 bits:
1175 multiclass N3VInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
1176 InstrItinClass itinD16, InstrItinClass itinD32,
1177 InstrItinClass itinQ16, InstrItinClass itinQ32,
1178 string OpcodeStr, Intrinsic IntOp, bit Commutable = 0> {
1179 // 64-bit vector types.
1180 def v4i16 : N3VDInt<op24, op23, 0b01, op11_8, op4, itinD16, !strconcat(OpcodeStr,"16"),
1181 v4i16, v4i16, IntOp, Commutable>;
1182 def v2i32 : N3VDInt<op24, op23, 0b10, op11_8, op4, itinD32, !strconcat(OpcodeStr,"32"),
1183 v2i32, v2i32, IntOp, Commutable>;
1185 // 128-bit vector types.
1186 def v8i16 : N3VQInt<op24, op23, 0b01, op11_8, op4, itinQ16, !strconcat(OpcodeStr,"16"),
1187 v8i16, v8i16, IntOp, Commutable>;
1188 def v4i32 : N3VQInt<op24, op23, 0b10, op11_8, op4, itinQ32, !strconcat(OpcodeStr,"32"),
1189 v4i32, v4i32, IntOp, Commutable>;
1192 multiclass N3VIntSL_HS<bits<4> op11_8,
1193 InstrItinClass itinD16, InstrItinClass itinD32,
1194 InstrItinClass itinQ16, InstrItinClass itinQ32,
1195 string OpcodeStr, Intrinsic IntOp> {
1196 def v4i16 : N3VDIntSL16<0b01, op11_8, itinD16, !strconcat(OpcodeStr, "16"), v4i16, IntOp>;
1197 def v2i32 : N3VDIntSL<0b10, op11_8, itinD32, !strconcat(OpcodeStr, "32"), v2i32, IntOp>;
1198 def v8i16 : N3VQIntSL16<0b01, op11_8, itinQ16, !strconcat(OpcodeStr, "16"), v8i16, v4i16, IntOp>;
1199 def v4i32 : N3VQIntSL<0b10, op11_8, itinQ32, !strconcat(OpcodeStr, "32"), v4i32, v2i32, IntOp>;
1202 // ....then also with element size of 8 bits:
1203 multiclass N3VInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1204 InstrItinClass itinD16, InstrItinClass itinD32,
1205 InstrItinClass itinQ16, InstrItinClass itinQ32,
1206 string OpcodeStr, Intrinsic IntOp, bit Commutable = 0>
1207 : N3VInt_HS<op24, op23, op11_8, op4, itinD16, itinD32, itinQ16, itinQ32,
1208 OpcodeStr, IntOp, Commutable> {
1209 def v8i8 : N3VDInt<op24, op23, 0b00, op11_8, op4, itinD16,
1210 !strconcat(OpcodeStr, "8"), v8i8, v8i8, IntOp, Commutable>;
1211 def v16i8 : N3VQInt<op24, op23, 0b00, op11_8, op4, itinQ16,
1212 !strconcat(OpcodeStr, "8"), v16i8, v16i8, IntOp, Commutable>;
1215 // ....then also with element size of 64 bits:
1216 multiclass N3VInt_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1217 InstrItinClass itinD16, InstrItinClass itinD32,
1218 InstrItinClass itinQ16, InstrItinClass itinQ32,
1219 string OpcodeStr, Intrinsic IntOp, bit Commutable = 0>
1220 : N3VInt_QHS<op24, op23, op11_8, op4, itinD16, itinD32, itinQ16, itinQ32,
1221 OpcodeStr, IntOp, Commutable> {
1222 def v1i64 : N3VDInt<op24, op23, 0b11, op11_8, op4, itinD32,
1223 !strconcat(OpcodeStr,"64"), v1i64, v1i64, IntOp, Commutable>;
1224 def v2i64 : N3VQInt<op24, op23, 0b11, op11_8, op4, itinQ32,
1225 !strconcat(OpcodeStr,"64"), v2i64, v2i64, IntOp, Commutable>;
1229 // Neon Narrowing 3-register vector intrinsics,
1230 // source operand element sizes of 16, 32 and 64 bits:
1231 multiclass N3VNInt_HSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1232 string OpcodeStr, Intrinsic IntOp, bit Commutable = 0> {
1233 def v8i8 : N3VNInt<op24, op23, 0b00, op11_8, op4, !strconcat(OpcodeStr,"16"),
1234 v8i8, v8i16, IntOp, Commutable>;
1235 def v4i16 : N3VNInt<op24, op23, 0b01, op11_8, op4, !strconcat(OpcodeStr,"32"),
1236 v4i16, v4i32, IntOp, Commutable>;
1237 def v2i32 : N3VNInt<op24, op23, 0b10, op11_8, op4, !strconcat(OpcodeStr,"64"),
1238 v2i32, v2i64, IntOp, Commutable>;
1242 // Neon Long 3-register vector intrinsics.
1244 // First with only element sizes of 16 and 32 bits:
1245 multiclass N3VLInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
1246 InstrItinClass itin, string OpcodeStr,
1247 Intrinsic IntOp, bit Commutable = 0> {
1248 def v4i32 : N3VLInt<op24, op23, 0b01, op11_8, op4, itin,
1249 !strconcat(OpcodeStr,"16"), v4i32, v4i16, IntOp, Commutable>;
1250 def v2i64 : N3VLInt<op24, op23, 0b10, op11_8, op4, itin,
1251 !strconcat(OpcodeStr,"32"), v2i64, v2i32, IntOp, Commutable>;
1254 multiclass N3VLIntSL_HS<bit op24, bits<4> op11_8,
1255 InstrItinClass itin, string OpcodeStr, Intrinsic IntOp> {
1256 def v4i16 : N3VLIntSL16<op24, 0b01, op11_8, itin,
1257 !strconcat(OpcodeStr, "16"), v4i32, v4i16, IntOp>;
1258 def v2i32 : N3VLIntSL<op24, 0b10, op11_8, itin,
1259 !strconcat(OpcodeStr, "32"), v2i64, v2i32, IntOp>;
1262 // ....then also with element size of 8 bits:
1263 multiclass N3VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1264 InstrItinClass itin, string OpcodeStr,
1265 Intrinsic IntOp, bit Commutable = 0>
1266 : N3VLInt_HS<op24, op23, op11_8, op4, itin, OpcodeStr, IntOp, Commutable> {
1267 def v8i16 : N3VLInt<op24, op23, 0b00, op11_8, op4, itin,
1268 !strconcat(OpcodeStr, "8"), v8i16, v8i8, IntOp, Commutable>;
1272 // Neon Wide 3-register vector intrinsics,
1273 // source operand element sizes of 8, 16 and 32 bits:
1274 multiclass N3VWInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1275 string OpcodeStr, Intrinsic IntOp, bit Commutable = 0> {
1276 def v8i16 : N3VWInt<op24, op23, 0b00, op11_8, op4, !strconcat(OpcodeStr, "8"),
1277 v8i16, v8i8, IntOp, Commutable>;
1278 def v4i32 : N3VWInt<op24, op23, 0b01, op11_8, op4, !strconcat(OpcodeStr,"16"),
1279 v4i32, v4i16, IntOp, Commutable>;
1280 def v2i64 : N3VWInt<op24, op23, 0b10, op11_8, op4, !strconcat(OpcodeStr,"32"),
1281 v2i64, v2i32, IntOp, Commutable>;
1285 // Neon Multiply-Op vector operations,
1286 // element sizes of 8, 16 and 32 bits:
1287 multiclass N3VMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1288 InstrItinClass itinD16, InstrItinClass itinD32,
1289 InstrItinClass itinQ16, InstrItinClass itinQ32,
1290 string OpcodeStr, SDNode OpNode> {
1291 // 64-bit vector types.
1292 def v8i8 : N3VDMulOp<op24, op23, 0b00, op11_8, op4, itinD16,
1293 !strconcat(OpcodeStr, "8"), v8i8, mul, OpNode>;
1294 def v4i16 : N3VDMulOp<op24, op23, 0b01, op11_8, op4, itinD16,
1295 !strconcat(OpcodeStr, "16"), v4i16, mul, OpNode>;
1296 def v2i32 : N3VDMulOp<op24, op23, 0b10, op11_8, op4, itinD32,
1297 !strconcat(OpcodeStr, "32"), v2i32, mul, OpNode>;
1299 // 128-bit vector types.
1300 def v16i8 : N3VQMulOp<op24, op23, 0b00, op11_8, op4, itinQ16,
1301 !strconcat(OpcodeStr, "8"), v16i8, mul, OpNode>;
1302 def v8i16 : N3VQMulOp<op24, op23, 0b01, op11_8, op4, itinQ16,
1303 !strconcat(OpcodeStr, "16"), v8i16, mul, OpNode>;
1304 def v4i32 : N3VQMulOp<op24, op23, 0b10, op11_8, op4, itinQ32,
1305 !strconcat(OpcodeStr, "32"), v4i32, mul, OpNode>;
1308 multiclass N3VMulOpSL_HS<bits<4> op11_8,
1309 InstrItinClass itinD16, InstrItinClass itinD32,
1310 InstrItinClass itinQ16, InstrItinClass itinQ32,
1311 string OpcodeStr, SDNode ShOp> {
1312 def v4i16 : N3VDMulOpSL16<0b01, op11_8, itinD16,
1313 !strconcat(OpcodeStr, "16"), v4i16, mul, ShOp>;
1314 def v2i32 : N3VDMulOpSL<0b10, op11_8, itinD32,
1315 !strconcat(OpcodeStr, "32"), v2i32, mul, ShOp>;
1316 def v8i16 : N3VQMulOpSL16<0b01, op11_8, itinQ16,
1317 !strconcat(OpcodeStr, "16"), v8i16, v4i16, mul, ShOp>;
1318 def v4i32 : N3VQMulOpSL<0b10, op11_8, itinQ32,
1319 !strconcat(OpcodeStr, "32"), v4i32, v2i32, mul, ShOp>;
1322 // Neon 3-argument intrinsics,
1323 // element sizes of 8, 16 and 32 bits:
1324 multiclass N3VInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1325 string OpcodeStr, Intrinsic IntOp> {
1326 // 64-bit vector types.
1327 def v8i8 : N3VDInt3<op24, op23, 0b00, op11_8, op4, IIC_VMACi16D,
1328 !strconcat(OpcodeStr, "8"), v8i8, v8i8, IntOp>;
1329 def v4i16 : N3VDInt3<op24, op23, 0b01, op11_8, op4, IIC_VMACi16D,
1330 !strconcat(OpcodeStr, "16"), v4i16, v4i16, IntOp>;
1331 def v2i32 : N3VDInt3<op24, op23, 0b10, op11_8, op4, IIC_VMACi32D,
1332 !strconcat(OpcodeStr, "32"), v2i32, v2i32, IntOp>;
1334 // 128-bit vector types.
1335 def v16i8 : N3VQInt3<op24, op23, 0b00, op11_8, op4, IIC_VMACi16Q,
1336 !strconcat(OpcodeStr, "8"), v16i8, v16i8, IntOp>;
1337 def v8i16 : N3VQInt3<op24, op23, 0b01, op11_8, op4, IIC_VMACi16Q,
1338 !strconcat(OpcodeStr, "16"), v8i16, v8i16, IntOp>;
1339 def v4i32 : N3VQInt3<op24, op23, 0b10, op11_8, op4, IIC_VMACi32Q,
1340 !strconcat(OpcodeStr, "32"), v4i32, v4i32, IntOp>;
1344 // Neon Long 3-argument intrinsics.
1346 // First with only element sizes of 16 and 32 bits:
1347 multiclass N3VLInt3_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
1348 string OpcodeStr, Intrinsic IntOp> {
1349 def v4i32 : N3VLInt3<op24, op23, 0b01, op11_8, op4, IIC_VMACi16D,
1350 !strconcat(OpcodeStr, "16"), v4i32, v4i16, IntOp>;
1351 def v2i64 : N3VLInt3<op24, op23, 0b10, op11_8, op4, IIC_VMACi16D,
1352 !strconcat(OpcodeStr, "32"), v2i64, v2i32, IntOp>;
1355 multiclass N3VLInt3SL_HS<bit op24, bits<4> op11_8,
1356 string OpcodeStr, Intrinsic IntOp> {
1357 def v4i16 : N3VLInt3SL16<op24, 0b01, op11_8, IIC_VMACi16D,
1358 !strconcat(OpcodeStr, "16"), v4i32, v4i16, IntOp>;
1359 def v2i32 : N3VLInt3SL<op24, 0b10, op11_8, IIC_VMACi32D,
1360 !strconcat(OpcodeStr, "32"), v2i64, v2i32, IntOp>;
1363 // ....then also with element size of 8 bits:
1364 multiclass N3VLInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1365 string OpcodeStr, Intrinsic IntOp>
1366 : N3VLInt3_HS<op24, op23, op11_8, op4, OpcodeStr, IntOp> {
1367 def v8i16 : N3VLInt3<op24, op23, 0b01, op11_8, op4, IIC_VMACi16D,
1368 !strconcat(OpcodeStr, "8"), v8i16, v8i8, IntOp>;
1372 // Neon 2-register vector intrinsics,
1373 // element sizes of 8, 16 and 32 bits:
1374 multiclass N2VInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1375 bits<5> op11_7, bit op4,
1376 InstrItinClass itinD, InstrItinClass itinQ,
1377 string OpcodeStr, Intrinsic IntOp> {
1378 // 64-bit vector types.
1379 def v8i8 : N2VDInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1380 itinD, !strconcat(OpcodeStr, "8"), v8i8, v8i8, IntOp>;
1381 def v4i16 : N2VDInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1382 itinD, !strconcat(OpcodeStr, "16"), v4i16, v4i16, IntOp>;
1383 def v2i32 : N2VDInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1384 itinD, !strconcat(OpcodeStr, "32"), v2i32, v2i32, IntOp>;
1386 // 128-bit vector types.
1387 def v16i8 : N2VQInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1388 itinQ, !strconcat(OpcodeStr, "8"), v16i8, v16i8, IntOp>;
1389 def v8i16 : N2VQInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1390 itinQ, !strconcat(OpcodeStr, "16"), v8i16, v8i16, IntOp>;
1391 def v4i32 : N2VQInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1392 itinQ, !strconcat(OpcodeStr, "32"), v4i32, v4i32, IntOp>;
1396 // Neon Pairwise long 2-register intrinsics,
1397 // element sizes of 8, 16 and 32 bits:
1398 multiclass N2VPLInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1399 bits<5> op11_7, bit op4,
1400 string OpcodeStr, Intrinsic IntOp> {
1401 // 64-bit vector types.
1402 def v8i8 : N2VDPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1403 !strconcat(OpcodeStr, "8"), v4i16, v8i8, IntOp>;
1404 def v4i16 : N2VDPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1405 !strconcat(OpcodeStr, "16"), v2i32, v4i16, IntOp>;
1406 def v2i32 : N2VDPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1407 !strconcat(OpcodeStr, "32"), v1i64, v2i32, IntOp>;
1409 // 128-bit vector types.
1410 def v16i8 : N2VQPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1411 !strconcat(OpcodeStr, "8"), v8i16, v16i8, IntOp>;
1412 def v8i16 : N2VQPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1413 !strconcat(OpcodeStr, "16"), v4i32, v8i16, IntOp>;
1414 def v4i32 : N2VQPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1415 !strconcat(OpcodeStr, "32"), v2i64, v4i32, IntOp>;
1419 // Neon Pairwise long 2-register accumulate intrinsics,
1420 // element sizes of 8, 16 and 32 bits:
1421 multiclass N2VPLInt2_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1422 bits<5> op11_7, bit op4,
1423 string OpcodeStr, Intrinsic IntOp> {
1424 // 64-bit vector types.
1425 def v8i8 : N2VDPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1426 !strconcat(OpcodeStr, "8"), v4i16, v8i8, IntOp>;
1427 def v4i16 : N2VDPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1428 !strconcat(OpcodeStr, "16"), v2i32, v4i16, IntOp>;
1429 def v2i32 : N2VDPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1430 !strconcat(OpcodeStr, "32"), v1i64, v2i32, IntOp>;
1432 // 128-bit vector types.
1433 def v16i8 : N2VQPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1434 !strconcat(OpcodeStr, "8"), v8i16, v16i8, IntOp>;
1435 def v8i16 : N2VQPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1436 !strconcat(OpcodeStr, "16"), v4i32, v8i16, IntOp>;
1437 def v4i32 : N2VQPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1438 !strconcat(OpcodeStr, "32"), v2i64, v4i32, IntOp>;
1442 // Neon 2-register vector shift by immediate,
1443 // element sizes of 8, 16, 32 and 64 bits:
1444 multiclass N2VSh_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1445 InstrItinClass itin, string OpcodeStr, SDNode OpNode> {
1446 // 64-bit vector types.
1447 def v8i8 : N2VDSh<op24, op23, 0b001000, op11_8, 0, op4, itin,
1448 !strconcat(OpcodeStr, "8"), v8i8, OpNode>;
1449 def v4i16 : N2VDSh<op24, op23, 0b010000, op11_8, 0, op4, itin,
1450 !strconcat(OpcodeStr, "16"), v4i16, OpNode>;
1451 def v2i32 : N2VDSh<op24, op23, 0b100000, op11_8, 0, op4, itin,
1452 !strconcat(OpcodeStr, "32"), v2i32, OpNode>;
1453 def v1i64 : N2VDSh<op24, op23, 0b000000, op11_8, 1, op4, itin,
1454 !strconcat(OpcodeStr, "64"), v1i64, OpNode>;
1456 // 128-bit vector types.
1457 def v16i8 : N2VQSh<op24, op23, 0b001000, op11_8, 0, op4, itin,
1458 !strconcat(OpcodeStr, "8"), v16i8, OpNode>;
1459 def v8i16 : N2VQSh<op24, op23, 0b010000, op11_8, 0, op4, itin,
1460 !strconcat(OpcodeStr, "16"), v8i16, OpNode>;
1461 def v4i32 : N2VQSh<op24, op23, 0b100000, op11_8, 0, op4, itin,
1462 !strconcat(OpcodeStr, "32"), v4i32, OpNode>;
1463 def v2i64 : N2VQSh<op24, op23, 0b000000, op11_8, 1, op4, itin,
1464 !strconcat(OpcodeStr, "64"), v2i64, OpNode>;
1468 // Neon Shift-Accumulate vector operations,
1469 // element sizes of 8, 16, 32 and 64 bits:
1470 multiclass N2VShAdd_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1471 string OpcodeStr, SDNode ShOp> {
1472 // 64-bit vector types.
1473 def v8i8 : N2VDShAdd<op24, op23, 0b001000, op11_8, 0, op4,
1474 !strconcat(OpcodeStr, "8"), v8i8, ShOp>;
1475 def v4i16 : N2VDShAdd<op24, op23, 0b010000, op11_8, 0, op4,
1476 !strconcat(OpcodeStr, "16"), v4i16, ShOp>;
1477 def v2i32 : N2VDShAdd<op24, op23, 0b100000, op11_8, 0, op4,
1478 !strconcat(OpcodeStr, "32"), v2i32, ShOp>;
1479 def v1i64 : N2VDShAdd<op24, op23, 0b000000, op11_8, 1, op4,
1480 !strconcat(OpcodeStr, "64"), v1i64, ShOp>;
1482 // 128-bit vector types.
1483 def v16i8 : N2VQShAdd<op24, op23, 0b001000, op11_8, 0, op4,
1484 !strconcat(OpcodeStr, "8"), v16i8, ShOp>;
1485 def v8i16 : N2VQShAdd<op24, op23, 0b010000, op11_8, 0, op4,
1486 !strconcat(OpcodeStr, "16"), v8i16, ShOp>;
1487 def v4i32 : N2VQShAdd<op24, op23, 0b100000, op11_8, 0, op4,
1488 !strconcat(OpcodeStr, "32"), v4i32, ShOp>;
1489 def v2i64 : N2VQShAdd<op24, op23, 0b000000, op11_8, 1, op4,
1490 !strconcat(OpcodeStr, "64"), v2i64, ShOp>;
1494 // Neon Shift-Insert vector operations,
1495 // element sizes of 8, 16, 32 and 64 bits:
1496 multiclass N2VShIns_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1497 string OpcodeStr, SDNode ShOp> {
1498 // 64-bit vector types.
1499 def v8i8 : N2VDShIns<op24, op23, 0b001000, op11_8, 0, op4,
1500 !strconcat(OpcodeStr, "8"), v8i8, ShOp>;
1501 def v4i16 : N2VDShIns<op24, op23, 0b010000, op11_8, 0, op4,
1502 !strconcat(OpcodeStr, "16"), v4i16, ShOp>;
1503 def v2i32 : N2VDShIns<op24, op23, 0b100000, op11_8, 0, op4,
1504 !strconcat(OpcodeStr, "32"), v2i32, ShOp>;
1505 def v1i64 : N2VDShIns<op24, op23, 0b000000, op11_8, 1, op4,
1506 !strconcat(OpcodeStr, "64"), v1i64, ShOp>;
1508 // 128-bit vector types.
1509 def v16i8 : N2VQShIns<op24, op23, 0b001000, op11_8, 0, op4,
1510 !strconcat(OpcodeStr, "8"), v16i8, ShOp>;
1511 def v8i16 : N2VQShIns<op24, op23, 0b010000, op11_8, 0, op4,
1512 !strconcat(OpcodeStr, "16"), v8i16, ShOp>;
1513 def v4i32 : N2VQShIns<op24, op23, 0b100000, op11_8, 0, op4,
1514 !strconcat(OpcodeStr, "32"), v4i32, ShOp>;
1515 def v2i64 : N2VQShIns<op24, op23, 0b000000, op11_8, 1, op4,
1516 !strconcat(OpcodeStr, "64"), v2i64, ShOp>;
1519 //===----------------------------------------------------------------------===//
1520 // Instruction Definitions.
1521 //===----------------------------------------------------------------------===//
1523 // Vector Add Operations.
1525 // VADD : Vector Add (integer and floating-point)
1526 defm VADD : N3V_QHSD<0, 0, 0b1000, 0, IIC_VBINiD, IIC_VBINiQ, "vadd.i", add, 1>;
1527 def VADDfd : N3VD<0, 0, 0b00, 0b1101, 0, IIC_VBIND, "vadd.f32", v2f32, v2f32, fadd, 1>;
1528 def VADDfq : N3VQ<0, 0, 0b00, 0b1101, 0, IIC_VBINQ, "vadd.f32", v4f32, v4f32, fadd, 1>;
1529 // VADDL : Vector Add Long (Q = D + D)
1530 defm VADDLs : N3VLInt_QHS<0,1,0b0000,0, IIC_VSHLiD, "vaddl.s", int_arm_neon_vaddls, 1>;
1531 defm VADDLu : N3VLInt_QHS<1,1,0b0000,0, IIC_VSHLiD, "vaddl.u", int_arm_neon_vaddlu, 1>;
1532 // VADDW : Vector Add Wide (Q = Q + D)
1533 defm VADDWs : N3VWInt_QHS<0,1,0b0001,0, "vaddw.s", int_arm_neon_vaddws, 0>;
1534 defm VADDWu : N3VWInt_QHS<1,1,0b0001,0, "vaddw.u", int_arm_neon_vaddwu, 0>;
1535 // VHADD : Vector Halving Add
1536 defm VHADDs : N3VInt_QHS<0,0,0b0000,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1537 IIC_VBINi4Q, "vhadd.s", int_arm_neon_vhadds, 1>;
1538 defm VHADDu : N3VInt_QHS<1,0,0b0000,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1539 IIC_VBINi4Q, "vhadd.u", int_arm_neon_vhaddu, 1>;
1540 // VRHADD : Vector Rounding Halving Add
1541 defm VRHADDs : N3VInt_QHS<0,0,0b0001,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1542 IIC_VBINi4Q, "vrhadd.s", int_arm_neon_vrhadds, 1>;
1543 defm VRHADDu : N3VInt_QHS<1,0,0b0001,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1544 IIC_VBINi4Q, "vrhadd.u", int_arm_neon_vrhaddu, 1>;
1545 // VQADD : Vector Saturating Add
1546 defm VQADDs : N3VInt_QHSD<0,0,0b0000,1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1547 IIC_VBINi4Q, "vqadd.s", int_arm_neon_vqadds, 1>;
1548 defm VQADDu : N3VInt_QHSD<1,0,0b0000,1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1549 IIC_VBINi4Q, "vqadd.u", int_arm_neon_vqaddu, 1>;
1550 // VADDHN : Vector Add and Narrow Returning High Half (D = Q + Q)
1551 defm VADDHN : N3VNInt_HSD<0,1,0b0100,0, "vaddhn.i", int_arm_neon_vaddhn, 1>;
1552 // VRADDHN : Vector Rounding Add and Narrow Returning High Half (D = Q + Q)
1553 defm VRADDHN : N3VNInt_HSD<1,1,0b0100,0, "vraddhn.i", int_arm_neon_vraddhn, 1>;
1555 // Vector Multiply Operations.
1557 // VMUL : Vector Multiply (integer, polynomial and floating-point)
1558 defm VMUL : N3V_QHS<0, 0, 0b1001, 1, IIC_VMULi16D, IIC_VMULi32D, IIC_VMULi16Q,
1559 IIC_VMULi32Q, "vmul.i", mul, 1>;
1560 def VMULpd : N3VDInt<1, 0, 0b00, 0b1001, 1, IIC_VMULi16D, "vmul.p8", v8i8, v8i8,
1561 int_arm_neon_vmulp, 1>;
1562 def VMULpq : N3VQInt<1, 0, 0b00, 0b1001, 1, IIC_VMULi16Q, "vmul.p8", v16i8, v16i8,
1563 int_arm_neon_vmulp, 1>;
1564 def VMULfd : N3VD<1, 0, 0b00, 0b1101, 1, IIC_VBIND, "vmul.f32", v2f32, v2f32, fmul, 1>;
1565 def VMULfq : N3VQ<1, 0, 0b00, 0b1101, 1, IIC_VBINQ, "vmul.f32", v4f32, v4f32, fmul, 1>;
1566 defm VMULsl : N3VSL_HS<0b1000, "vmul.i", mul>;
1567 def VMULslfd : N3VDSL<0b10, 0b1001, IIC_VBIND, "vmul.f32", v2f32, fmul>;
1568 def VMULslfq : N3VQSL<0b10, 0b1001, IIC_VBINQ, "vmul.f32", v4f32, v2f32, fmul>;
1569 def : Pat<(v8i16 (mul (v8i16 QPR:$src1),
1570 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
1571 (v8i16 (VMULslv8i16 (v8i16 QPR:$src1),
1572 (v4i16 (EXTRACT_SUBREG QPR:$src2,
1573 (DSubReg_i16_reg imm:$lane))),
1574 (SubReg_i16_lane imm:$lane)))>;
1575 def : Pat<(v4i32 (mul (v4i32 QPR:$src1),
1576 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
1577 (v4i32 (VMULslv4i32 (v4i32 QPR:$src1),
1578 (v2i32 (EXTRACT_SUBREG QPR:$src2,
1579 (DSubReg_i32_reg imm:$lane))),
1580 (SubReg_i32_lane imm:$lane)))>;
1581 def : Pat<(v4f32 (fmul (v4f32 QPR:$src1),
1582 (v4f32 (NEONvduplane (v4f32 QPR:$src2), imm:$lane)))),
1583 (v4f32 (VMULslfq (v4f32 QPR:$src1),
1584 (v2f32 (EXTRACT_SUBREG QPR:$src2,
1585 (DSubReg_i32_reg imm:$lane))),
1586 (SubReg_i32_lane imm:$lane)))>;
1588 // VQDMULH : Vector Saturating Doubling Multiply Returning High Half
1589 defm VQDMULH : N3VInt_HS<0, 0, 0b1011, 0, IIC_VMULi16D, IIC_VMULi32D,
1590 IIC_VMULi16Q, IIC_VMULi32Q,
1591 "vqdmulh.s", int_arm_neon_vqdmulh, 1>;
1592 defm VQDMULHsl: N3VIntSL_HS<0b1100, IIC_VMULi16D, IIC_VMULi32D,
1593 IIC_VMULi16Q, IIC_VMULi32Q,
1594 "vqdmulh.s", int_arm_neon_vqdmulh>;
1595 def : Pat<(v8i16 (int_arm_neon_vqdmulh (v8i16 QPR:$src1),
1596 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
1597 (v8i16 (VQDMULHslv8i16 (v8i16 QPR:$src1),
1598 (v4i16 (EXTRACT_SUBREG QPR:$src2,
1599 (DSubReg_i16_reg imm:$lane))),
1600 (SubReg_i16_lane imm:$lane)))>;
1601 def : Pat<(v4i32 (int_arm_neon_vqdmulh (v4i32 QPR:$src1),
1602 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
1603 (v4i32 (VQDMULHslv4i32 (v4i32 QPR:$src1),
1604 (v2i32 (EXTRACT_SUBREG QPR:$src2,
1605 (DSubReg_i32_reg imm:$lane))),
1606 (SubReg_i32_lane imm:$lane)))>;
1608 // VQRDMULH : Vector Rounding Saturating Doubling Multiply Returning High Half
1609 defm VQRDMULH : N3VInt_HS<1, 0, 0b1011, 0, IIC_VMULi16D, IIC_VMULi32D,
1610 IIC_VMULi16Q, IIC_VMULi32Q,
1611 "vqrdmulh.s", int_arm_neon_vqrdmulh, 1>;
1612 defm VQRDMULHsl : N3VIntSL_HS<0b1101, IIC_VMULi16D, IIC_VMULi32D,
1613 IIC_VMULi16Q, IIC_VMULi32Q,
1614 "vqrdmulh.s", int_arm_neon_vqrdmulh>;
1615 def : Pat<(v8i16 (int_arm_neon_vqrdmulh (v8i16 QPR:$src1),
1616 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
1617 (v8i16 (VQRDMULHslv8i16 (v8i16 QPR:$src1),
1618 (v4i16 (EXTRACT_SUBREG QPR:$src2,
1619 (DSubReg_i16_reg imm:$lane))),
1620 (SubReg_i16_lane imm:$lane)))>;
1621 def : Pat<(v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$src1),
1622 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
1623 (v4i32 (VQRDMULHslv4i32 (v4i32 QPR:$src1),
1624 (v2i32 (EXTRACT_SUBREG QPR:$src2,
1625 (DSubReg_i32_reg imm:$lane))),
1626 (SubReg_i32_lane imm:$lane)))>;
1628 // VMULL : Vector Multiply Long (integer and polynomial) (Q = D * D)
1629 defm VMULLs : N3VLInt_QHS<0,1,0b1100,0, IIC_VMULi16D, "vmull.s", int_arm_neon_vmulls, 1>;
1630 defm VMULLu : N3VLInt_QHS<1,1,0b1100,0, IIC_VMULi16D, "vmull.u", int_arm_neon_vmullu, 1>;
1631 def VMULLp : N3VLInt<0, 1, 0b00, 0b1110, 0, IIC_VMULi16D, "vmull.p8", v8i16, v8i8,
1632 int_arm_neon_vmullp, 1>;
1633 defm VMULLsls : N3VLIntSL_HS<0, 0b1010, IIC_VMULi16D, "vmull.s", int_arm_neon_vmulls>;
1634 defm VMULLslu : N3VLIntSL_HS<1, 0b1010, IIC_VMULi16D, "vmull.u", int_arm_neon_vmullu>;
1636 // VQDMULL : Vector Saturating Doubling Multiply Long (Q = D * D)
1637 defm VQDMULL : N3VLInt_HS<0,1,0b1101,0, IIC_VMULi16D, "vqdmull.s", int_arm_neon_vqdmull, 1>;
1638 defm VQDMULLsl: N3VLIntSL_HS<0, 0b1011, IIC_VMULi16D, "vqdmull.s", int_arm_neon_vqdmull>;
1640 // Vector Multiply-Accumulate and Multiply-Subtract Operations.
1642 // VMLA : Vector Multiply Accumulate (integer and floating-point)
1643 defm VMLA : N3VMulOp_QHS<0, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
1644 IIC_VMACi16Q, IIC_VMACi32Q, "vmla.i", add>;
1645 def VMLAfd : N3VDMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACD, "vmla.f32", v2f32, fmul, fadd>;
1646 def VMLAfq : N3VQMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACQ, "vmla.f32", v4f32, fmul, fadd>;
1647 defm VMLAsl : N3VMulOpSL_HS<0b0000, IIC_VMACi16D, IIC_VMACi32D,
1648 IIC_VMACi16Q, IIC_VMACi32Q, "vmla.i", add>;
1649 def VMLAslfd : N3VDMulOpSL<0b10, 0b0001, IIC_VMACD, "vmla.f32", v2f32, fmul, fadd>;
1650 def VMLAslfq : N3VQMulOpSL<0b10, 0b0001, IIC_VMACQ, "vmla.f32", v4f32, v2f32, fmul, fadd>;
1652 def : Pat<(v8i16 (add (v8i16 QPR:$src1),
1653 (mul (v8i16 QPR:$src2),
1654 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
1655 (v8i16 (VMLAslv8i16 (v8i16 QPR:$src1),
1657 (v4i16 (EXTRACT_SUBREG QPR:$src3,
1658 (DSubReg_i16_reg imm:$lane))),
1659 (SubReg_i16_lane imm:$lane)))>;
1661 def : Pat<(v4i32 (add (v4i32 QPR:$src1),
1662 (mul (v4i32 QPR:$src2),
1663 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
1664 (v4i32 (VMLAslv4i32 (v4i32 QPR:$src1),
1666 (v2i32 (EXTRACT_SUBREG QPR:$src3,
1667 (DSubReg_i32_reg imm:$lane))),
1668 (SubReg_i32_lane imm:$lane)))>;
1670 def : Pat<(v4f32 (fadd (v4f32 QPR:$src1),
1671 (fmul (v4f32 QPR:$src2),
1672 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
1673 (v4f32 (VMLAslfq (v4f32 QPR:$src1),
1675 (v2f32 (EXTRACT_SUBREG QPR:$src3,
1676 (DSubReg_i32_reg imm:$lane))),
1677 (SubReg_i32_lane imm:$lane)))>;
1679 // VMLAL : Vector Multiply Accumulate Long (Q += D * D)
1680 defm VMLALs : N3VLInt3_QHS<0,1,0b1000,0, "vmlal.s", int_arm_neon_vmlals>;
1681 defm VMLALu : N3VLInt3_QHS<1,1,0b1000,0, "vmlal.u", int_arm_neon_vmlalu>;
1683 defm VMLALsls : N3VLInt3SL_HS<0, 0b0010, "vmlal.s", int_arm_neon_vmlals>;
1684 defm VMLALslu : N3VLInt3SL_HS<1, 0b0010, "vmlal.u", int_arm_neon_vmlalu>;
1686 // VQDMLAL : Vector Saturating Doubling Multiply Accumulate Long (Q += D * D)
1687 defm VQDMLAL : N3VLInt3_HS<0, 1, 0b1001, 0, "vqdmlal.s", int_arm_neon_vqdmlal>;
1688 defm VQDMLALsl: N3VLInt3SL_HS<0, 0b0011, "vqdmlal.s", int_arm_neon_vqdmlal>;
1690 // VMLS : Vector Multiply Subtract (integer and floating-point)
1691 defm VMLS : N3VMulOp_QHS<1, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
1692 IIC_VMACi16Q, IIC_VMACi32Q, "vmls.i", sub>;
1693 def VMLSfd : N3VDMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACD, "vmls.f32", v2f32, fmul, fsub>;
1694 def VMLSfq : N3VQMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACQ, "vmls.f32", v4f32, fmul, fsub>;
1695 defm VMLSsl : N3VMulOpSL_HS<0b0100, IIC_VMACi16D, IIC_VMACi32D,
1696 IIC_VMACi16Q, IIC_VMACi32Q, "vmls.i", sub>;
1697 def VMLSslfd : N3VDMulOpSL<0b10, 0b0101, IIC_VMACD, "vmls.f32", v2f32, fmul, fsub>;
1698 def VMLSslfq : N3VQMulOpSL<0b10, 0b0101, IIC_VMACQ, "vmls.f32", v4f32, v2f32, fmul, fsub>;
1700 def : Pat<(v8i16 (sub (v8i16 QPR:$src1),
1701 (mul (v8i16 QPR:$src2),
1702 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
1703 (v8i16 (VMLSslv8i16 (v8i16 QPR:$src1),
1705 (v4i16 (EXTRACT_SUBREG QPR:$src3,
1706 (DSubReg_i16_reg imm:$lane))),
1707 (SubReg_i16_lane imm:$lane)))>;
1709 def : Pat<(v4i32 (sub (v4i32 QPR:$src1),
1710 (mul (v4i32 QPR:$src2),
1711 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
1712 (v4i32 (VMLSslv4i32 (v4i32 QPR:$src1),
1714 (v2i32 (EXTRACT_SUBREG QPR:$src3,
1715 (DSubReg_i32_reg imm:$lane))),
1716 (SubReg_i32_lane imm:$lane)))>;
1718 def : Pat<(v4f32 (fsub (v4f32 QPR:$src1),
1719 (fmul (v4f32 QPR:$src2),
1720 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
1721 (v4f32 (VMLSslfq (v4f32 QPR:$src1),
1723 (v2f32 (EXTRACT_SUBREG QPR:$src3,
1724 (DSubReg_i32_reg imm:$lane))),
1725 (SubReg_i32_lane imm:$lane)))>;
1727 // VMLSL : Vector Multiply Subtract Long (Q -= D * D)
1728 defm VMLSLs : N3VLInt3_QHS<0,1,0b1010,0, "vmlsl.s", int_arm_neon_vmlsls>;
1729 defm VMLSLu : N3VLInt3_QHS<1,1,0b1010,0, "vmlsl.u", int_arm_neon_vmlslu>;
1731 defm VMLSLsls : N3VLInt3SL_HS<0, 0b0110, "vmlsl.s", int_arm_neon_vmlsls>;
1732 defm VMLSLslu : N3VLInt3SL_HS<1, 0b0110, "vmlsl.u", int_arm_neon_vmlslu>;
1734 // VQDMLSL : Vector Saturating Doubling Multiply Subtract Long (Q -= D * D)
1735 defm VQDMLSL : N3VLInt3_HS<0, 1, 0b1011, 0, "vqdmlsl.s", int_arm_neon_vqdmlsl>;
1736 defm VQDMLSLsl: N3VLInt3SL_HS<0, 0b111, "vqdmlsl.s", int_arm_neon_vqdmlsl>;
1738 // Vector Subtract Operations.
1740 // VSUB : Vector Subtract (integer and floating-point)
1741 defm VSUB : N3V_QHSD<1, 0, 0b1000, 0, IIC_VSUBiD, IIC_VSUBiQ, "vsub.i", sub, 0>;
1742 def VSUBfd : N3VD<0, 0, 0b10, 0b1101, 0, IIC_VBIND, "vsub.f32", v2f32, v2f32, fsub, 0>;
1743 def VSUBfq : N3VQ<0, 0, 0b10, 0b1101, 0, IIC_VBINQ, "vsub.f32", v4f32, v4f32, fsub, 0>;
1744 // VSUBL : Vector Subtract Long (Q = D - D)
1745 defm VSUBLs : N3VLInt_QHS<0,1,0b0010,0, IIC_VSHLiD, "vsubl.s", int_arm_neon_vsubls, 1>;
1746 defm VSUBLu : N3VLInt_QHS<1,1,0b0010,0, IIC_VSHLiD, "vsubl.u", int_arm_neon_vsublu, 1>;
1747 // VSUBW : Vector Subtract Wide (Q = Q - D)
1748 defm VSUBWs : N3VWInt_QHS<0,1,0b0011,0, "vsubw.s", int_arm_neon_vsubws, 0>;
1749 defm VSUBWu : N3VWInt_QHS<1,1,0b0011,0, "vsubw.u", int_arm_neon_vsubwu, 0>;
1750 // VHSUB : Vector Halving Subtract
1751 defm VHSUBs : N3VInt_QHS<0, 0, 0b0010, 0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1752 IIC_VBINi4Q, "vhsub.s", int_arm_neon_vhsubs, 0>;
1753 defm VHSUBu : N3VInt_QHS<1, 0, 0b0010, 0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1754 IIC_VBINi4Q, "vhsub.u", int_arm_neon_vhsubu, 0>;
1755 // VQSUB : Vector Saturing Subtract
1756 defm VQSUBs : N3VInt_QHSD<0, 0, 0b0010, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1757 IIC_VBINi4Q, "vqsub.s", int_arm_neon_vqsubs, 0>;
1758 defm VQSUBu : N3VInt_QHSD<1, 0, 0b0010, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1759 IIC_VBINi4Q, "vqsub.u", int_arm_neon_vqsubu, 0>;
1760 // VSUBHN : Vector Subtract and Narrow Returning High Half (D = Q - Q)
1761 defm VSUBHN : N3VNInt_HSD<0,1,0b0110,0, "vsubhn.i", int_arm_neon_vsubhn, 0>;
1762 // VRSUBHN : Vector Rounding Subtract and Narrow Returning High Half (D=Q-Q)
1763 defm VRSUBHN : N3VNInt_HSD<1,1,0b0110,0, "vrsubhn.i", int_arm_neon_vrsubhn, 0>;
1765 // Vector Comparisons.
1767 // VCEQ : Vector Compare Equal
1768 defm VCEQ : N3V_QHS<1, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1769 IIC_VBINi4Q, "vceq.i", NEONvceq, 1>;
1770 def VCEQfd : N3VD<0,0,0b00,0b1110,0, IIC_VBIND, "vceq.f32", v2i32, v2f32, NEONvceq, 1>;
1771 def VCEQfq : N3VQ<0,0,0b00,0b1110,0, IIC_VBINQ, "vceq.f32", v4i32, v4f32, NEONvceq, 1>;
1772 // VCGE : Vector Compare Greater Than or Equal
1773 defm VCGEs : N3V_QHS<0, 0, 0b0011, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1774 IIC_VBINi4Q, "vcge.s", NEONvcge, 0>;
1775 defm VCGEu : N3V_QHS<1, 0, 0b0011, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1776 IIC_VBINi4Q, "vcge.u", NEONvcgeu, 0>;
1777 def VCGEfd : N3VD<1,0,0b00,0b1110,0, IIC_VBIND, "vcge.f32", v2i32, v2f32, NEONvcge, 0>;
1778 def VCGEfq : N3VQ<1,0,0b00,0b1110,0, IIC_VBINQ, "vcge.f32", v4i32, v4f32, NEONvcge, 0>;
1779 // VCGT : Vector Compare Greater Than
1780 defm VCGTs : N3V_QHS<0, 0, 0b0011, 0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1781 IIC_VBINi4Q, "vcgt.s", NEONvcgt, 0>;
1782 defm VCGTu : N3V_QHS<1, 0, 0b0011, 0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1783 IIC_VBINi4Q, "vcgt.u", NEONvcgtu, 0>;
1784 def VCGTfd : N3VD<1,0,0b10,0b1110,0, IIC_VBIND, "vcgt.f32", v2i32, v2f32, NEONvcgt, 0>;
1785 def VCGTfq : N3VQ<1,0,0b10,0b1110,0, IIC_VBINQ, "vcgt.f32", v4i32, v4f32, NEONvcgt, 0>;
1786 // VACGE : Vector Absolute Compare Greater Than or Equal (aka VCAGE)
1787 def VACGEd : N3VDInt<1, 0, 0b00, 0b1110, 1, IIC_VBIND, "vacge.f32", v2i32, v2f32,
1788 int_arm_neon_vacged, 0>;
1789 def VACGEq : N3VQInt<1, 0, 0b00, 0b1110, 1, IIC_VBINQ, "vacge.f32", v4i32, v4f32,
1790 int_arm_neon_vacgeq, 0>;
1791 // VACGT : Vector Absolute Compare Greater Than (aka VCAGT)
1792 def VACGTd : N3VDInt<1, 0, 0b10, 0b1110, 1, IIC_VBIND, "vacgt.f32", v2i32, v2f32,
1793 int_arm_neon_vacgtd, 0>;
1794 def VACGTq : N3VQInt<1, 0, 0b10, 0b1110, 1, IIC_VBINQ, "vacgt.f32", v4i32, v4f32,
1795 int_arm_neon_vacgtq, 0>;
1796 // VTST : Vector Test Bits
1797 defm VTST : N3V_QHS<0, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1798 IIC_VBINi4Q, "vtst.i", NEONvtst, 1>;
1800 // Vector Bitwise Operations.
1802 // VAND : Vector Bitwise AND
1803 def VANDd : N3VD<0, 0, 0b00, 0b0001, 1, IIC_VBINiD, "vand", v2i32, v2i32, and, 1>;
1804 def VANDq : N3VQ<0, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "vand", v4i32, v4i32, and, 1>;
1806 // VEOR : Vector Bitwise Exclusive OR
1807 def VEORd : N3VD<1, 0, 0b00, 0b0001, 1, IIC_VBINiD, "veor", v2i32, v2i32, xor, 1>;
1808 def VEORq : N3VQ<1, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "veor", v4i32, v4i32, xor, 1>;
1810 // VORR : Vector Bitwise OR
1811 def VORRd : N3VD<0, 0, 0b10, 0b0001, 1, IIC_VBINiD, "vorr", v2i32, v2i32, or, 1>;
1812 def VORRq : N3VQ<0, 0, 0b10, 0b0001, 1, IIC_VBINiQ, "vorr", v4i32, v4i32, or, 1>;
1814 // VBIC : Vector Bitwise Bit Clear (AND NOT)
1815 def VBICd : N3V<0, 0, 0b01, 0b0001, 0, 1, (outs DPR:$dst),
1816 (ins DPR:$src1, DPR:$src2), IIC_VBINiD,
1817 "vbic\t$dst, $src1, $src2", "",
1818 [(set DPR:$dst, (v2i32 (and DPR:$src1,
1819 (vnot_conv DPR:$src2))))]>;
1820 def VBICq : N3V<0, 0, 0b01, 0b0001, 1, 1, (outs QPR:$dst),
1821 (ins QPR:$src1, QPR:$src2), IIC_VBINiQ,
1822 "vbic\t$dst, $src1, $src2", "",
1823 [(set QPR:$dst, (v4i32 (and QPR:$src1,
1824 (vnot_conv QPR:$src2))))]>;
1826 // VORN : Vector Bitwise OR NOT
1827 def VORNd : N3V<0, 0, 0b11, 0b0001, 0, 1, (outs DPR:$dst),
1828 (ins DPR:$src1, DPR:$src2), IIC_VBINiD,
1829 "vorn\t$dst, $src1, $src2", "",
1830 [(set DPR:$dst, (v2i32 (or DPR:$src1,
1831 (vnot_conv DPR:$src2))))]>;
1832 def VORNq : N3V<0, 0, 0b11, 0b0001, 1, 1, (outs QPR:$dst),
1833 (ins QPR:$src1, QPR:$src2), IIC_VBINiQ,
1834 "vorn\t$dst, $src1, $src2", "",
1835 [(set QPR:$dst, (v4i32 (or QPR:$src1,
1836 (vnot_conv QPR:$src2))))]>;
1838 // VMVN : Vector Bitwise NOT
1839 def VMVNd : N2V<0b11, 0b11, 0b00, 0b00, 0b01011, 0, 0,
1840 (outs DPR:$dst), (ins DPR:$src), IIC_VSHLiD,
1841 "vmvn\t$dst, $src", "",
1842 [(set DPR:$dst, (v2i32 (vnot DPR:$src)))]>;
1843 def VMVNq : N2V<0b11, 0b11, 0b00, 0b00, 0b01011, 1, 0,
1844 (outs QPR:$dst), (ins QPR:$src), IIC_VSHLiD,
1845 "vmvn\t$dst, $src", "",
1846 [(set QPR:$dst, (v4i32 (vnot QPR:$src)))]>;
1847 def : Pat<(v2i32 (vnot_conv DPR:$src)), (VMVNd DPR:$src)>;
1848 def : Pat<(v4i32 (vnot_conv QPR:$src)), (VMVNq QPR:$src)>;
1850 // VBSL : Vector Bitwise Select
1851 def VBSLd : N3V<1, 0, 0b01, 0b0001, 0, 1, (outs DPR:$dst),
1852 (ins DPR:$src1, DPR:$src2, DPR:$src3), IIC_VCNTiD,
1853 "vbsl\t$dst, $src2, $src3", "$src1 = $dst",
1855 (v2i32 (or (and DPR:$src2, DPR:$src1),
1856 (and DPR:$src3, (vnot_conv DPR:$src1)))))]>;
1857 def VBSLq : N3V<1, 0, 0b01, 0b0001, 1, 1, (outs QPR:$dst),
1858 (ins QPR:$src1, QPR:$src2, QPR:$src3), IIC_VCNTiQ,
1859 "vbsl\t$dst, $src2, $src3", "$src1 = $dst",
1861 (v4i32 (or (and QPR:$src2, QPR:$src1),
1862 (and QPR:$src3, (vnot_conv QPR:$src1)))))]>;
1864 // VBIF : Vector Bitwise Insert if False
1865 // like VBSL but with: "vbif\t$dst, $src3, $src1", "$src2 = $dst",
1866 // VBIT : Vector Bitwise Insert if True
1867 // like VBSL but with: "vbit\t$dst, $src2, $src1", "$src3 = $dst",
1868 // These are not yet implemented. The TwoAddress pass will not go looking
1869 // for equivalent operations with different register constraints; it just
1872 // Vector Absolute Differences.
1874 // VABD : Vector Absolute Difference
1875 defm VABDs : N3VInt_QHS<0, 0, 0b0111, 0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1876 IIC_VBINi4Q, "vabd.s", int_arm_neon_vabds, 0>;
1877 defm VABDu : N3VInt_QHS<1, 0, 0b0111, 0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1878 IIC_VBINi4Q, "vabd.u", int_arm_neon_vabdu, 0>;
1879 def VABDfd : N3VDInt<1, 0, 0b10, 0b1101, 0, IIC_VBIND, "vabd.f32", v2f32, v2f32,
1880 int_arm_neon_vabds, 0>;
1881 def VABDfq : N3VQInt<1, 0, 0b10, 0b1101, 0, IIC_VBINQ, "vabd.f32", v4f32, v4f32,
1882 int_arm_neon_vabds, 0>;
1884 // VABDL : Vector Absolute Difference Long (Q = | D - D |)
1885 defm VABDLs : N3VLInt_QHS<0,1,0b0111,0, IIC_VBINi4Q, "vabdl.s", int_arm_neon_vabdls, 0>;
1886 defm VABDLu : N3VLInt_QHS<1,1,0b0111,0, IIC_VBINi4Q, "vabdl.u", int_arm_neon_vabdlu, 0>;
1888 // VABA : Vector Absolute Difference and Accumulate
1889 defm VABAs : N3VInt3_QHS<0,1,0b0101,0, "vaba.s", int_arm_neon_vabas>;
1890 defm VABAu : N3VInt3_QHS<1,1,0b0101,0, "vaba.u", int_arm_neon_vabau>;
1892 // VABAL : Vector Absolute Difference and Accumulate Long (Q += | D - D |)
1893 defm VABALs : N3VLInt3_QHS<0,1,0b0101,0, "vabal.s", int_arm_neon_vabals>;
1894 defm VABALu : N3VLInt3_QHS<1,1,0b0101,0, "vabal.u", int_arm_neon_vabalu>;
1896 // Vector Maximum and Minimum.
1898 // VMAX : Vector Maximum
1899 defm VMAXs : N3VInt_QHS<0, 0, 0b0110, 0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1900 IIC_VBINi4Q, "vmax.s", int_arm_neon_vmaxs, 1>;
1901 defm VMAXu : N3VInt_QHS<1, 0, 0b0110, 0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1902 IIC_VBINi4Q, "vmax.u", int_arm_neon_vmaxu, 1>;
1903 def VMAXfd : N3VDInt<0, 0, 0b00, 0b1111, 0, IIC_VBIND, "vmax.f32", v2f32, v2f32,
1904 int_arm_neon_vmaxs, 1>;
1905 def VMAXfq : N3VQInt<0, 0, 0b00, 0b1111, 0, IIC_VBINQ, "vmax.f32", v4f32, v4f32,
1906 int_arm_neon_vmaxs, 1>;
1908 // VMIN : Vector Minimum
1909 defm VMINs : N3VInt_QHS<0, 0, 0b0110, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1910 IIC_VBINi4Q, "vmin.s", int_arm_neon_vmins, 1>;
1911 defm VMINu : N3VInt_QHS<1, 0, 0b0110, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1912 IIC_VBINi4Q, "vmin.u", int_arm_neon_vminu, 1>;
1913 def VMINfd : N3VDInt<0, 0, 0b10, 0b1111, 0, IIC_VBIND, "vmin.f32", v2f32, v2f32,
1914 int_arm_neon_vmins, 1>;
1915 def VMINfq : N3VQInt<0, 0, 0b10, 0b1111, 0, IIC_VBINQ, "vmin.f32", v4f32, v4f32,
1916 int_arm_neon_vmins, 1>;
1918 // Vector Pairwise Operations.
1920 // VPADD : Vector Pairwise Add
1921 def VPADDi8 : N3VDInt<0, 0, 0b00, 0b1011, 1, IIC_VBINiD, "vpadd.i8", v8i8, v8i8,
1922 int_arm_neon_vpadd, 0>;
1923 def VPADDi16 : N3VDInt<0, 0, 0b01, 0b1011, 1, IIC_VBINiD, "vpadd.i16", v4i16, v4i16,
1924 int_arm_neon_vpadd, 0>;
1925 def VPADDi32 : N3VDInt<0, 0, 0b10, 0b1011, 1, IIC_VBINiD, "vpadd.i32", v2i32, v2i32,
1926 int_arm_neon_vpadd, 0>;
1927 def VPADDf : N3VDInt<1, 0, 0b00, 0b1101, 0, IIC_VBIND, "vpadd.f32", v2f32, v2f32,
1928 int_arm_neon_vpadd, 0>;
1930 // VPADDL : Vector Pairwise Add Long
1931 defm VPADDLs : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00100, 0, "vpaddl.s",
1932 int_arm_neon_vpaddls>;
1933 defm VPADDLu : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00101, 0, "vpaddl.u",
1934 int_arm_neon_vpaddlu>;
1936 // VPADAL : Vector Pairwise Add and Accumulate Long
1937 defm VPADALs : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b00100, 0, "vpadal.s",
1938 int_arm_neon_vpadals>;
1939 defm VPADALu : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b00101, 0, "vpadal.u",
1940 int_arm_neon_vpadalu>;
1942 // VPMAX : Vector Pairwise Maximum
1943 def VPMAXs8 : N3VDInt<0, 0, 0b00, 0b1010, 0, IIC_VBINi4D, "vpmax.s8", v8i8, v8i8,
1944 int_arm_neon_vpmaxs, 0>;
1945 def VPMAXs16 : N3VDInt<0, 0, 0b01, 0b1010, 0, IIC_VBINi4D, "vpmax.s16", v4i16, v4i16,
1946 int_arm_neon_vpmaxs, 0>;
1947 def VPMAXs32 : N3VDInt<0, 0, 0b10, 0b1010, 0, IIC_VBINi4D, "vpmax.s32", v2i32, v2i32,
1948 int_arm_neon_vpmaxs, 0>;
1949 def VPMAXu8 : N3VDInt<1, 0, 0b00, 0b1010, 0, IIC_VBINi4D, "vpmax.u8", v8i8, v8i8,
1950 int_arm_neon_vpmaxu, 0>;
1951 def VPMAXu16 : N3VDInt<1, 0, 0b01, 0b1010, 0, IIC_VBINi4D, "vpmax.u16", v4i16, v4i16,
1952 int_arm_neon_vpmaxu, 0>;
1953 def VPMAXu32 : N3VDInt<1, 0, 0b10, 0b1010, 0, IIC_VBINi4D, "vpmax.u32", v2i32, v2i32,
1954 int_arm_neon_vpmaxu, 0>;
1955 def VPMAXf : N3VDInt<1, 0, 0b00, 0b1111, 0, IIC_VBINi4D, "vpmax.f32", v2f32, v2f32,
1956 int_arm_neon_vpmaxs, 0>;
1958 // VPMIN : Vector Pairwise Minimum
1959 def VPMINs8 : N3VDInt<0, 0, 0b00, 0b1010, 1, IIC_VBINi4D, "vpmin.s8", v8i8, v8i8,
1960 int_arm_neon_vpmins, 0>;
1961 def VPMINs16 : N3VDInt<0, 0, 0b01, 0b1010, 1, IIC_VBINi4D, "vpmin.s16", v4i16, v4i16,
1962 int_arm_neon_vpmins, 0>;
1963 def VPMINs32 : N3VDInt<0, 0, 0b10, 0b1010, 1, IIC_VBINi4D, "vpmin.s32", v2i32, v2i32,
1964 int_arm_neon_vpmins, 0>;
1965 def VPMINu8 : N3VDInt<1, 0, 0b00, 0b1010, 1, IIC_VBINi4D, "vpmin.u8", v8i8, v8i8,
1966 int_arm_neon_vpminu, 0>;
1967 def VPMINu16 : N3VDInt<1, 0, 0b01, 0b1010, 1, IIC_VBINi4D, "vpmin.u16", v4i16, v4i16,
1968 int_arm_neon_vpminu, 0>;
1969 def VPMINu32 : N3VDInt<1, 0, 0b10, 0b1010, 1, IIC_VBINi4D, "vpmin.u32", v2i32, v2i32,
1970 int_arm_neon_vpminu, 0>;
1971 def VPMINf : N3VDInt<1, 0, 0b10, 0b1111, 0, IIC_VBINi4D, "vpmin.f32", v2f32, v2f32,
1972 int_arm_neon_vpmins, 0>;
1974 // Vector Reciprocal and Reciprocal Square Root Estimate and Step.
1976 // VRECPE : Vector Reciprocal Estimate
1977 def VRECPEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
1978 IIC_VUNAD, "vrecpe.u32",
1979 v2i32, v2i32, int_arm_neon_vrecpe>;
1980 def VRECPEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
1981 IIC_VUNAQ, "vrecpe.u32",
1982 v4i32, v4i32, int_arm_neon_vrecpe>;
1983 def VRECPEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
1984 IIC_VUNAD, "vrecpe.f32",
1985 v2f32, v2f32, int_arm_neon_vrecpe>;
1986 def VRECPEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
1987 IIC_VUNAQ, "vrecpe.f32",
1988 v4f32, v4f32, int_arm_neon_vrecpe>;
1990 // VRECPS : Vector Reciprocal Step
1991 def VRECPSfd : N3VDInt<0, 0, 0b00, 0b1111, 1, IIC_VRECSD, "vrecps.f32", v2f32, v2f32,
1992 int_arm_neon_vrecps, 1>;
1993 def VRECPSfq : N3VQInt<0, 0, 0b00, 0b1111, 1, IIC_VRECSQ, "vrecps.f32", v4f32, v4f32,
1994 int_arm_neon_vrecps, 1>;
1996 // VRSQRTE : Vector Reciprocal Square Root Estimate
1997 def VRSQRTEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
1998 IIC_VUNAD, "vrsqrte.u32",
1999 v2i32, v2i32, int_arm_neon_vrsqrte>;
2000 def VRSQRTEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
2001 IIC_VUNAQ, "vrsqrte.u32",
2002 v4i32, v4i32, int_arm_neon_vrsqrte>;
2003 def VRSQRTEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
2004 IIC_VUNAD, "vrsqrte.f32",
2005 v2f32, v2f32, int_arm_neon_vrsqrte>;
2006 def VRSQRTEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
2007 IIC_VUNAQ, "vrsqrte.f32",
2008 v4f32, v4f32, int_arm_neon_vrsqrte>;
2010 // VRSQRTS : Vector Reciprocal Square Root Step
2011 def VRSQRTSfd : N3VDInt<0, 0, 0b10, 0b1111, 1, IIC_VRECSD, "vrsqrts.f32", v2f32, v2f32,
2012 int_arm_neon_vrsqrts, 1>;
2013 def VRSQRTSfq : N3VQInt<0, 0, 0b10, 0b1111, 1, IIC_VRECSQ, "vrsqrts.f32", v4f32, v4f32,
2014 int_arm_neon_vrsqrts, 1>;
2018 // VSHL : Vector Shift
2019 defm VSHLs : N3VInt_QHSD<0, 0, 0b0100, 0, IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ,
2020 IIC_VSHLiQ, "vshl.s", int_arm_neon_vshifts, 0>;
2021 defm VSHLu : N3VInt_QHSD<1, 0, 0b0100, 0, IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ,
2022 IIC_VSHLiQ, "vshl.u", int_arm_neon_vshiftu, 0>;
2023 // VSHL : Vector Shift Left (Immediate)
2024 defm VSHLi : N2VSh_QHSD<0, 1, 0b0111, 1, IIC_VSHLiD, "vshl.i", NEONvshl>;
2025 // VSHR : Vector Shift Right (Immediate)
2026 defm VSHRs : N2VSh_QHSD<0, 1, 0b0000, 1, IIC_VSHLiD, "vshr.s", NEONvshrs>;
2027 defm VSHRu : N2VSh_QHSD<1, 1, 0b0000, 1, IIC_VSHLiD, "vshr.u", NEONvshru>;
2029 // VSHLL : Vector Shift Left Long
2030 def VSHLLs8 : N2VLSh<0, 1, 0b001000, 0b1010, 0, 0, 1, "vshll.s8",
2031 v8i16, v8i8, NEONvshlls>;
2032 def VSHLLs16 : N2VLSh<0, 1, 0b010000, 0b1010, 0, 0, 1, "vshll.s16",
2033 v4i32, v4i16, NEONvshlls>;
2034 def VSHLLs32 : N2VLSh<0, 1, 0b100000, 0b1010, 0, 0, 1, "vshll.s32",
2035 v2i64, v2i32, NEONvshlls>;
2036 def VSHLLu8 : N2VLSh<1, 1, 0b001000, 0b1010, 0, 0, 1, "vshll.u8",
2037 v8i16, v8i8, NEONvshllu>;
2038 def VSHLLu16 : N2VLSh<1, 1, 0b010000, 0b1010, 0, 0, 1, "vshll.u16",
2039 v4i32, v4i16, NEONvshllu>;
2040 def VSHLLu32 : N2VLSh<1, 1, 0b100000, 0b1010, 0, 0, 1, "vshll.u32",
2041 v2i64, v2i32, NEONvshllu>;
2043 // VSHLL : Vector Shift Left Long (with maximum shift count)
2044 def VSHLLi8 : N2VLSh<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll.i8",
2045 v8i16, v8i8, NEONvshlli>;
2046 def VSHLLi16 : N2VLSh<1, 1, 0b110110, 0b0011, 0, 0, 0, "vshll.i16",
2047 v4i32, v4i16, NEONvshlli>;
2048 def VSHLLi32 : N2VLSh<1, 1, 0b111010, 0b0011, 0, 0, 0, "vshll.i32",
2049 v2i64, v2i32, NEONvshlli>;
2051 // VSHRN : Vector Shift Right and Narrow
2052 def VSHRN16 : N2VNSh<0, 1, 0b001000, 0b1000, 0, 0, 1,
2053 IIC_VSHLiD, "vshrn.i16", v8i8, v8i16, NEONvshrn>;
2054 def VSHRN32 : N2VNSh<0, 1, 0b010000, 0b1000, 0, 0, 1,
2055 IIC_VSHLiD, "vshrn.i32", v4i16, v4i32, NEONvshrn>;
2056 def VSHRN64 : N2VNSh<0, 1, 0b100000, 0b1000, 0, 0, 1,
2057 IIC_VSHLiD, "vshrn.i64", v2i32, v2i64, NEONvshrn>;
2059 // VRSHL : Vector Rounding Shift
2060 defm VRSHLs : N3VInt_QHSD<0,0,0b0101,0, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2061 IIC_VSHLi4Q, "vrshl.s", int_arm_neon_vrshifts, 0>;
2062 defm VRSHLu : N3VInt_QHSD<1,0,0b0101,0, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2063 IIC_VSHLi4Q, "vrshl.u", int_arm_neon_vrshiftu, 0>;
2064 // VRSHR : Vector Rounding Shift Right
2065 defm VRSHRs : N2VSh_QHSD<0, 1, 0b0010, 1, IIC_VSHLi4D, "vrshr.s", NEONvrshrs>;
2066 defm VRSHRu : N2VSh_QHSD<1, 1, 0b0010, 1, IIC_VSHLi4D, "vrshr.u", NEONvrshru>;
2068 // VRSHRN : Vector Rounding Shift Right and Narrow
2069 def VRSHRN16 : N2VNSh<0, 1, 0b001000, 0b1000, 0, 1, 1,
2070 IIC_VSHLi4D, "vrshrn.i16", v8i8, v8i16, NEONvrshrn>;
2071 def VRSHRN32 : N2VNSh<0, 1, 0b010000, 0b1000, 0, 1, 1,
2072 IIC_VSHLi4D, "vrshrn.i32", v4i16, v4i32, NEONvrshrn>;
2073 def VRSHRN64 : N2VNSh<0, 1, 0b100000, 0b1000, 0, 1, 1,
2074 IIC_VSHLi4D, "vrshrn.i64", v2i32, v2i64, NEONvrshrn>;
2076 // VQSHL : Vector Saturating Shift
2077 defm VQSHLs : N3VInt_QHSD<0,0,0b0100,1, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2078 IIC_VSHLi4Q, "vqshl.s", int_arm_neon_vqshifts, 0>;
2079 defm VQSHLu : N3VInt_QHSD<1,0,0b0100,1, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2080 IIC_VSHLi4Q, "vqshl.u", int_arm_neon_vqshiftu, 0>;
2081 // VQSHL : Vector Saturating Shift Left (Immediate)
2082 defm VQSHLsi : N2VSh_QHSD<0, 1, 0b0111, 1, IIC_VSHLi4D, "vqshl.s", NEONvqshls>;
2083 defm VQSHLui : N2VSh_QHSD<1, 1, 0b0111, 1, IIC_VSHLi4D, "vqshl.u", NEONvqshlu>;
2084 // VQSHLU : Vector Saturating Shift Left (Immediate, Unsigned)
2085 defm VQSHLsu : N2VSh_QHSD<1, 1, 0b0110, 1, IIC_VSHLi4D, "vqshlu.s", NEONvqshlsu>;
2087 // VQSHRN : Vector Saturating Shift Right and Narrow
2088 def VQSHRNs16 : N2VNSh<0, 1, 0b001000, 0b1001, 0, 0, 1,
2089 IIC_VSHLi4D, "vqshrn.s16", v8i8, v8i16, NEONvqshrns>;
2090 def VQSHRNs32 : N2VNSh<0, 1, 0b010000, 0b1001, 0, 0, 1,
2091 IIC_VSHLi4D, "vqshrn.s32", v4i16, v4i32, NEONvqshrns>;
2092 def VQSHRNs64 : N2VNSh<0, 1, 0b100000, 0b1001, 0, 0, 1,
2093 IIC_VSHLi4D, "vqshrn.s64", v2i32, v2i64, NEONvqshrns>;
2094 def VQSHRNu16 : N2VNSh<1, 1, 0b001000, 0b1001, 0, 0, 1,
2095 IIC_VSHLi4D, "vqshrn.u16", v8i8, v8i16, NEONvqshrnu>;
2096 def VQSHRNu32 : N2VNSh<1, 1, 0b010000, 0b1001, 0, 0, 1,
2097 IIC_VSHLi4D, "vqshrn.u32", v4i16, v4i32, NEONvqshrnu>;
2098 def VQSHRNu64 : N2VNSh<1, 1, 0b100000, 0b1001, 0, 0, 1,
2099 IIC_VSHLi4D, "vqshrn.u64", v2i32, v2i64, NEONvqshrnu>;
2101 // VQSHRUN : Vector Saturating Shift Right and Narrow (Unsigned)
2102 def VQSHRUN16 : N2VNSh<1, 1, 0b001000, 0b1000, 0, 0, 1,
2103 IIC_VSHLi4D, "vqshrun.s16", v8i8, v8i16, NEONvqshrnsu>;
2104 def VQSHRUN32 : N2VNSh<1, 1, 0b010000, 0b1000, 0, 0, 1,
2105 IIC_VSHLi4D, "vqshrun.s32", v4i16, v4i32, NEONvqshrnsu>;
2106 def VQSHRUN64 : N2VNSh<1, 1, 0b100000, 0b1000, 0, 0, 1,
2107 IIC_VSHLi4D, "vqshrun.s64", v2i32, v2i64, NEONvqshrnsu>;
2109 // VQRSHL : Vector Saturating Rounding Shift
2110 defm VQRSHLs : N3VInt_QHSD<0, 0, 0b0101, 1, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2111 IIC_VSHLi4Q, "vqrshl.s", int_arm_neon_vqrshifts, 0>;
2112 defm VQRSHLu : N3VInt_QHSD<1, 0, 0b0101, 1, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2113 IIC_VSHLi4Q, "vqrshl.u", int_arm_neon_vqrshiftu, 0>;
2115 // VQRSHRN : Vector Saturating Rounding Shift Right and Narrow
2116 def VQRSHRNs16: N2VNSh<0, 1, 0b001000, 0b1001, 0, 1, 1,
2117 IIC_VSHLi4D, "vqrshrn.s16", v8i8, v8i16, NEONvqrshrns>;
2118 def VQRSHRNs32: N2VNSh<0, 1, 0b010000, 0b1001, 0, 1, 1,
2119 IIC_VSHLi4D, "vqrshrn.s32", v4i16, v4i32, NEONvqrshrns>;
2120 def VQRSHRNs64: N2VNSh<0, 1, 0b100000, 0b1001, 0, 1, 1,
2121 IIC_VSHLi4D, "vqrshrn.s64", v2i32, v2i64, NEONvqrshrns>;
2122 def VQRSHRNu16: N2VNSh<1, 1, 0b001000, 0b1001, 0, 1, 1,
2123 IIC_VSHLi4D, "vqrshrn.u16", v8i8, v8i16, NEONvqrshrnu>;
2124 def VQRSHRNu32: N2VNSh<1, 1, 0b010000, 0b1001, 0, 1, 1,
2125 IIC_VSHLi4D, "vqrshrn.u32", v4i16, v4i32, NEONvqrshrnu>;
2126 def VQRSHRNu64: N2VNSh<1, 1, 0b100000, 0b1001, 0, 1, 1,
2127 IIC_VSHLi4D, "vqrshrn.u64", v2i32, v2i64, NEONvqrshrnu>;
2129 // VQRSHRUN : Vector Saturating Rounding Shift Right and Narrow (Unsigned)
2130 def VQRSHRUN16: N2VNSh<1, 1, 0b001000, 0b1000, 0, 1, 1,
2131 IIC_VSHLi4D, "vqrshrun.s16", v8i8, v8i16, NEONvqrshrnsu>;
2132 def VQRSHRUN32: N2VNSh<1, 1, 0b010000, 0b1000, 0, 1, 1,
2133 IIC_VSHLi4D, "vqrshrun.s32", v4i16, v4i32, NEONvqrshrnsu>;
2134 def VQRSHRUN64: N2VNSh<1, 1, 0b100000, 0b1000, 0, 1, 1,
2135 IIC_VSHLi4D, "vqrshrun.s64", v2i32, v2i64, NEONvqrshrnsu>;
2137 // VSRA : Vector Shift Right and Accumulate
2138 defm VSRAs : N2VShAdd_QHSD<0, 1, 0b0001, 1, "vsra.s", NEONvshrs>;
2139 defm VSRAu : N2VShAdd_QHSD<1, 1, 0b0001, 1, "vsra.u", NEONvshru>;
2140 // VRSRA : Vector Rounding Shift Right and Accumulate
2141 defm VRSRAs : N2VShAdd_QHSD<0, 1, 0b0011, 1, "vrsra.s", NEONvrshrs>;
2142 defm VRSRAu : N2VShAdd_QHSD<1, 1, 0b0011, 1, "vrsra.u", NEONvrshru>;
2144 // VSLI : Vector Shift Left and Insert
2145 defm VSLI : N2VShIns_QHSD<1, 1, 0b0101, 1, "vsli.", NEONvsli>;
2146 // VSRI : Vector Shift Right and Insert
2147 defm VSRI : N2VShIns_QHSD<1, 1, 0b0100, 1, "vsri.", NEONvsri>;
2149 // Vector Absolute and Saturating Absolute.
2151 // VABS : Vector Absolute Value
2152 defm VABS : N2VInt_QHS<0b11, 0b11, 0b01, 0b00110, 0,
2153 IIC_VUNAiD, IIC_VUNAiQ, "vabs.s",
2155 def VABSfd : N2VDInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
2156 IIC_VUNAD, "vabs.f32",
2157 v2f32, v2f32, int_arm_neon_vabs>;
2158 def VABSfq : N2VQInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
2159 IIC_VUNAQ, "vabs.f32",
2160 v4f32, v4f32, int_arm_neon_vabs>;
2162 // VQABS : Vector Saturating Absolute Value
2163 defm VQABS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01110, 0,
2164 IIC_VQUNAiD, IIC_VQUNAiQ, "vqabs.s",
2165 int_arm_neon_vqabs>;
2169 def vneg : PatFrag<(ops node:$in), (sub immAllZerosV, node:$in)>;
2170 def vneg_conv : PatFrag<(ops node:$in), (sub immAllZerosV_bc, node:$in)>;
2172 class VNEGD<bits<2> size, string OpcodeStr, ValueType Ty>
2173 : N2V<0b11, 0b11, size, 0b01, 0b00111, 0, 0, (outs DPR:$dst), (ins DPR:$src),
2174 IIC_VSHLiD, !strconcat(OpcodeStr, "\t$dst, $src"), "",
2175 [(set DPR:$dst, (Ty (vneg DPR:$src)))]>;
2176 class VNEGQ<bits<2> size, string OpcodeStr, ValueType Ty>
2177 : N2V<0b11, 0b11, size, 0b01, 0b00111, 1, 0, (outs QPR:$dst), (ins QPR:$src),
2178 IIC_VSHLiD, !strconcat(OpcodeStr, "\t$dst, $src"), "",
2179 [(set QPR:$dst, (Ty (vneg QPR:$src)))]>;
2181 // VNEG : Vector Negate
2182 def VNEGs8d : VNEGD<0b00, "vneg.s8", v8i8>;
2183 def VNEGs16d : VNEGD<0b01, "vneg.s16", v4i16>;
2184 def VNEGs32d : VNEGD<0b10, "vneg.s32", v2i32>;
2185 def VNEGs8q : VNEGQ<0b00, "vneg.s8", v16i8>;
2186 def VNEGs16q : VNEGQ<0b01, "vneg.s16", v8i16>;
2187 def VNEGs32q : VNEGQ<0b10, "vneg.s32", v4i32>;
2189 // VNEG : Vector Negate (floating-point)
2190 def VNEGf32d : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
2191 (outs DPR:$dst), (ins DPR:$src), IIC_VUNAD,
2192 "vneg.f32\t$dst, $src", "",
2193 [(set DPR:$dst, (v2f32 (fneg DPR:$src)))]>;
2194 def VNEGf32q : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 1, 0,
2195 (outs QPR:$dst), (ins QPR:$src), IIC_VUNAQ,
2196 "vneg.f32\t$dst, $src", "",
2197 [(set QPR:$dst, (v4f32 (fneg QPR:$src)))]>;
2199 def : Pat<(v8i8 (vneg_conv DPR:$src)), (VNEGs8d DPR:$src)>;
2200 def : Pat<(v4i16 (vneg_conv DPR:$src)), (VNEGs16d DPR:$src)>;
2201 def : Pat<(v2i32 (vneg_conv DPR:$src)), (VNEGs32d DPR:$src)>;
2202 def : Pat<(v16i8 (vneg_conv QPR:$src)), (VNEGs8q QPR:$src)>;
2203 def : Pat<(v8i16 (vneg_conv QPR:$src)), (VNEGs16q QPR:$src)>;
2204 def : Pat<(v4i32 (vneg_conv QPR:$src)), (VNEGs32q QPR:$src)>;
2206 // VQNEG : Vector Saturating Negate
2207 defm VQNEG : N2VInt_QHS<0b11, 0b11, 0b00, 0b01111, 0,
2208 IIC_VQUNAiD, IIC_VQUNAiQ, "vqneg.s",
2209 int_arm_neon_vqneg>;
2211 // Vector Bit Counting Operations.
2213 // VCLS : Vector Count Leading Sign Bits
2214 defm VCLS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01000, 0,
2215 IIC_VCNTiD, IIC_VCNTiQ, "vcls.s",
2217 // VCLZ : Vector Count Leading Zeros
2218 defm VCLZ : N2VInt_QHS<0b11, 0b11, 0b00, 0b01001, 0,
2219 IIC_VCNTiD, IIC_VCNTiQ, "vclz.i",
2221 // VCNT : Vector Count One Bits
2222 def VCNTd : N2VDInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
2223 IIC_VCNTiD, "vcnt.8",
2224 v8i8, v8i8, int_arm_neon_vcnt>;
2225 def VCNTq : N2VQInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
2226 IIC_VCNTiQ, "vcnt.8",
2227 v16i8, v16i8, int_arm_neon_vcnt>;
2229 // Vector Move Operations.
2231 // VMOV : Vector Move (Register)
2233 def VMOVD : N3V<0, 0, 0b10, 0b0001, 0, 1, (outs DPR:$dst), (ins DPR:$src),
2234 IIC_VMOVD, "vmov\t$dst, $src", "", []>;
2235 def VMOVQ : N3V<0, 0, 0b10, 0b0001, 1, 1, (outs QPR:$dst), (ins QPR:$src),
2236 IIC_VMOVD, "vmov\t$dst, $src", "", []>;
2238 // VMOV : Vector Move (Immediate)
2240 // VMOV_get_imm8 xform function: convert build_vector to VMOV.i8 imm.
2241 def VMOV_get_imm8 : SDNodeXForm<build_vector, [{
2242 return ARM::getVMOVImm(N, 1, *CurDAG);
2244 def vmovImm8 : PatLeaf<(build_vector), [{
2245 return ARM::getVMOVImm(N, 1, *CurDAG).getNode() != 0;
2248 // VMOV_get_imm16 xform function: convert build_vector to VMOV.i16 imm.
2249 def VMOV_get_imm16 : SDNodeXForm<build_vector, [{
2250 return ARM::getVMOVImm(N, 2, *CurDAG);
2252 def vmovImm16 : PatLeaf<(build_vector), [{
2253 return ARM::getVMOVImm(N, 2, *CurDAG).getNode() != 0;
2254 }], VMOV_get_imm16>;
2256 // VMOV_get_imm32 xform function: convert build_vector to VMOV.i32 imm.
2257 def VMOV_get_imm32 : SDNodeXForm<build_vector, [{
2258 return ARM::getVMOVImm(N, 4, *CurDAG);
2260 def vmovImm32 : PatLeaf<(build_vector), [{
2261 return ARM::getVMOVImm(N, 4, *CurDAG).getNode() != 0;
2262 }], VMOV_get_imm32>;
2264 // VMOV_get_imm64 xform function: convert build_vector to VMOV.i64 imm.
2265 def VMOV_get_imm64 : SDNodeXForm<build_vector, [{
2266 return ARM::getVMOVImm(N, 8, *CurDAG);
2268 def vmovImm64 : PatLeaf<(build_vector), [{
2269 return ARM::getVMOVImm(N, 8, *CurDAG).getNode() != 0;
2270 }], VMOV_get_imm64>;
2272 // Note: Some of the cmode bits in the following VMOV instructions need to
2273 // be encoded based on the immed values.
2275 def VMOVv8i8 : N1ModImm<1, 0b000, 0b1110, 0, 0, 0, 1, (outs DPR:$dst),
2276 (ins i8imm:$SIMM), IIC_VMOVImm,
2277 "vmov.i8\t$dst, $SIMM", "",
2278 [(set DPR:$dst, (v8i8 vmovImm8:$SIMM))]>;
2279 def VMOVv16i8 : N1ModImm<1, 0b000, 0b1110, 0, 1, 0, 1, (outs QPR:$dst),
2280 (ins i8imm:$SIMM), IIC_VMOVImm,
2281 "vmov.i8\t$dst, $SIMM", "",
2282 [(set QPR:$dst, (v16i8 vmovImm8:$SIMM))]>;
2284 def VMOVv4i16 : N1ModImm<1, 0b000, 0b1000, 0, 0, 0, 1, (outs DPR:$dst),
2285 (ins i16imm:$SIMM), IIC_VMOVImm,
2286 "vmov.i16\t$dst, $SIMM", "",
2287 [(set DPR:$dst, (v4i16 vmovImm16:$SIMM))]>;
2288 def VMOVv8i16 : N1ModImm<1, 0b000, 0b1000, 0, 1, 0, 1, (outs QPR:$dst),
2289 (ins i16imm:$SIMM), IIC_VMOVImm,
2290 "vmov.i16\t$dst, $SIMM", "",
2291 [(set QPR:$dst, (v8i16 vmovImm16:$SIMM))]>;
2293 def VMOVv2i32 : N1ModImm<1, 0b000, 0b0000, 0, 0, 0, 1, (outs DPR:$dst),
2294 (ins i32imm:$SIMM), IIC_VMOVImm,
2295 "vmov.i32\t$dst, $SIMM", "",
2296 [(set DPR:$dst, (v2i32 vmovImm32:$SIMM))]>;
2297 def VMOVv4i32 : N1ModImm<1, 0b000, 0b0000, 0, 1, 0, 1, (outs QPR:$dst),
2298 (ins i32imm:$SIMM), IIC_VMOVImm,
2299 "vmov.i32\t$dst, $SIMM", "",
2300 [(set QPR:$dst, (v4i32 vmovImm32:$SIMM))]>;
2302 def VMOVv1i64 : N1ModImm<1, 0b000, 0b1110, 0, 0, 1, 1, (outs DPR:$dst),
2303 (ins i64imm:$SIMM), IIC_VMOVImm,
2304 "vmov.i64\t$dst, $SIMM", "",
2305 [(set DPR:$dst, (v1i64 vmovImm64:$SIMM))]>;
2306 def VMOVv2i64 : N1ModImm<1, 0b000, 0b1110, 0, 1, 1, 1, (outs QPR:$dst),
2307 (ins i64imm:$SIMM), IIC_VMOVImm,
2308 "vmov.i64\t$dst, $SIMM", "",
2309 [(set QPR:$dst, (v2i64 vmovImm64:$SIMM))]>;
2311 // VMOV : Vector Get Lane (move scalar to ARM core register)
2313 def VGETLNs8 : NVGetLane<0b11100101, 0b1011, 0b00,
2314 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2315 IIC_VMOVSI, "vmov", ".s8\t$dst, $src[$lane]",
2316 [(set GPR:$dst, (NEONvgetlanes (v8i8 DPR:$src),
2318 def VGETLNs16 : NVGetLane<0b11100001, 0b1011, 0b01,
2319 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2320 IIC_VMOVSI, "vmov", ".s16\t$dst, $src[$lane]",
2321 [(set GPR:$dst, (NEONvgetlanes (v4i16 DPR:$src),
2323 def VGETLNu8 : NVGetLane<0b11101101, 0b1011, 0b00,
2324 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2325 IIC_VMOVSI, "vmov", ".u8\t$dst, $src[$lane]",
2326 [(set GPR:$dst, (NEONvgetlaneu (v8i8 DPR:$src),
2328 def VGETLNu16 : NVGetLane<0b11101001, 0b1011, 0b01,
2329 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2330 IIC_VMOVSI, "vmov", ".u16\t$dst, $src[$lane]",
2331 [(set GPR:$dst, (NEONvgetlaneu (v4i16 DPR:$src),
2333 def VGETLNi32 : NVGetLane<0b11100001, 0b1011, 0b00,
2334 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2335 IIC_VMOVSI, "vmov", ".32\t$dst, $src[$lane]",
2336 [(set GPR:$dst, (extractelt (v2i32 DPR:$src),
2338 // def VGETLNf32: see FMRDH and FMRDL in ARMInstrVFP.td
2339 def : Pat<(NEONvgetlanes (v16i8 QPR:$src), imm:$lane),
2340 (VGETLNs8 (v8i8 (EXTRACT_SUBREG QPR:$src,
2341 (DSubReg_i8_reg imm:$lane))),
2342 (SubReg_i8_lane imm:$lane))>;
2343 def : Pat<(NEONvgetlanes (v8i16 QPR:$src), imm:$lane),
2344 (VGETLNs16 (v4i16 (EXTRACT_SUBREG QPR:$src,
2345 (DSubReg_i16_reg imm:$lane))),
2346 (SubReg_i16_lane imm:$lane))>;
2347 def : Pat<(NEONvgetlaneu (v16i8 QPR:$src), imm:$lane),
2348 (VGETLNu8 (v8i8 (EXTRACT_SUBREG QPR:$src,
2349 (DSubReg_i8_reg imm:$lane))),
2350 (SubReg_i8_lane imm:$lane))>;
2351 def : Pat<(NEONvgetlaneu (v8i16 QPR:$src), imm:$lane),
2352 (VGETLNu16 (v4i16 (EXTRACT_SUBREG QPR:$src,
2353 (DSubReg_i16_reg imm:$lane))),
2354 (SubReg_i16_lane imm:$lane))>;
2355 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
2356 (VGETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src,
2357 (DSubReg_i32_reg imm:$lane))),
2358 (SubReg_i32_lane imm:$lane))>;
2359 def : Pat<(extractelt (v2f32 DPR:$src1), imm:$src2),
2360 (EXTRACT_SUBREG (COPY_TO_REGCLASS DPR:$src1, DPR_VFP2),
2361 (SSubReg_f32_reg imm:$src2))>;
2362 def : Pat<(extractelt (v4f32 QPR:$src1), imm:$src2),
2363 (EXTRACT_SUBREG (COPY_TO_REGCLASS QPR:$src1, QPR_VFP2),
2364 (SSubReg_f32_reg imm:$src2))>;
2365 //def : Pat<(extractelt (v2i64 QPR:$src1), imm:$src2),
2366 // (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
2367 def : Pat<(extractelt (v2f64 QPR:$src1), imm:$src2),
2368 (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
2371 // VMOV : Vector Set Lane (move ARM core register to scalar)
2373 let Constraints = "$src1 = $dst" in {
2374 def VSETLNi8 : NVSetLane<0b11100100, 0b1011, 0b00, (outs DPR:$dst),
2375 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
2376 IIC_VMOVISL, "vmov", ".8\t$dst[$lane], $src2",
2377 [(set DPR:$dst, (vector_insert (v8i8 DPR:$src1),
2378 GPR:$src2, imm:$lane))]>;
2379 def VSETLNi16 : NVSetLane<0b11100000, 0b1011, 0b01, (outs DPR:$dst),
2380 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
2381 IIC_VMOVISL, "vmov", ".16\t$dst[$lane], $src2",
2382 [(set DPR:$dst, (vector_insert (v4i16 DPR:$src1),
2383 GPR:$src2, imm:$lane))]>;
2384 def VSETLNi32 : NVSetLane<0b11100000, 0b1011, 0b00, (outs DPR:$dst),
2385 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
2386 IIC_VMOVISL, "vmov", ".32\t$dst[$lane], $src2",
2387 [(set DPR:$dst, (insertelt (v2i32 DPR:$src1),
2388 GPR:$src2, imm:$lane))]>;
2390 def : Pat<(vector_insert (v16i8 QPR:$src1), GPR:$src2, imm:$lane),
2391 (v16i8 (INSERT_SUBREG QPR:$src1,
2392 (VSETLNi8 (v8i8 (EXTRACT_SUBREG QPR:$src1,
2393 (DSubReg_i8_reg imm:$lane))),
2394 GPR:$src2, (SubReg_i8_lane imm:$lane)),
2395 (DSubReg_i8_reg imm:$lane)))>;
2396 def : Pat<(vector_insert (v8i16 QPR:$src1), GPR:$src2, imm:$lane),
2397 (v8i16 (INSERT_SUBREG QPR:$src1,
2398 (VSETLNi16 (v4i16 (EXTRACT_SUBREG QPR:$src1,
2399 (DSubReg_i16_reg imm:$lane))),
2400 GPR:$src2, (SubReg_i16_lane imm:$lane)),
2401 (DSubReg_i16_reg imm:$lane)))>;
2402 def : Pat<(insertelt (v4i32 QPR:$src1), GPR:$src2, imm:$lane),
2403 (v4i32 (INSERT_SUBREG QPR:$src1,
2404 (VSETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src1,
2405 (DSubReg_i32_reg imm:$lane))),
2406 GPR:$src2, (SubReg_i32_lane imm:$lane)),
2407 (DSubReg_i32_reg imm:$lane)))>;
2409 def : Pat<(v2f32 (insertelt DPR:$src1, SPR:$src2, imm:$src3)),
2410 (INSERT_SUBREG (COPY_TO_REGCLASS DPR:$src1, DPR_VFP2),
2411 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
2412 def : Pat<(v4f32 (insertelt QPR:$src1, SPR:$src2, imm:$src3)),
2413 (INSERT_SUBREG (COPY_TO_REGCLASS QPR:$src1, QPR_VFP2),
2414 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
2416 //def : Pat<(v2i64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
2417 // (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
2418 def : Pat<(v2f64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
2419 (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
2421 def : Pat<(v2f32 (scalar_to_vector SPR:$src)),
2422 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$src, arm_ssubreg_0)>;
2423 def : Pat<(v2f64 (scalar_to_vector DPR:$src)),
2424 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), DPR:$src, arm_dsubreg_0)>;
2425 def : Pat<(v4f32 (scalar_to_vector SPR:$src)),
2426 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), SPR:$src, arm_ssubreg_0)>;
2428 def : Pat<(v8i8 (scalar_to_vector GPR:$src)),
2429 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
2430 def : Pat<(v4i16 (scalar_to_vector GPR:$src)),
2431 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
2432 def : Pat<(v2i32 (scalar_to_vector GPR:$src)),
2433 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
2435 def : Pat<(v16i8 (scalar_to_vector GPR:$src)),
2436 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
2437 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
2439 def : Pat<(v8i16 (scalar_to_vector GPR:$src)),
2440 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
2441 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
2443 def : Pat<(v4i32 (scalar_to_vector GPR:$src)),
2444 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
2445 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
2448 // VDUP : Vector Duplicate (from ARM core register to all elements)
2450 class VDUPD<bits<8> opcod1, bits<2> opcod3, string asmSize, ValueType Ty>
2451 : NVDup<opcod1, 0b1011, opcod3, (outs DPR:$dst), (ins GPR:$src),
2452 IIC_VMOVIS, "vdup", !strconcat(asmSize, "\t$dst, $src"),
2453 [(set DPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
2454 class VDUPQ<bits<8> opcod1, bits<2> opcod3, string asmSize, ValueType Ty>
2455 : NVDup<opcod1, 0b1011, opcod3, (outs QPR:$dst), (ins GPR:$src),
2456 IIC_VMOVIS, "vdup", !strconcat(asmSize, "\t$dst, $src"),
2457 [(set QPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
2459 def VDUP8d : VDUPD<0b11101100, 0b00, ".8", v8i8>;
2460 def VDUP16d : VDUPD<0b11101000, 0b01, ".16", v4i16>;
2461 def VDUP32d : VDUPD<0b11101000, 0b00, ".32", v2i32>;
2462 def VDUP8q : VDUPQ<0b11101110, 0b00, ".8", v16i8>;
2463 def VDUP16q : VDUPQ<0b11101010, 0b01, ".16", v8i16>;
2464 def VDUP32q : VDUPQ<0b11101010, 0b00, ".32", v4i32>;
2466 def VDUPfd : NVDup<0b11101000, 0b1011, 0b00, (outs DPR:$dst), (ins GPR:$src),
2467 IIC_VMOVIS, "vdup", ".32\t$dst, $src",
2468 [(set DPR:$dst, (v2f32 (NEONvdup
2469 (f32 (bitconvert GPR:$src)))))]>;
2470 def VDUPfq : NVDup<0b11101010, 0b1011, 0b00, (outs QPR:$dst), (ins GPR:$src),
2471 IIC_VMOVIS, "vdup", ".32\t$dst, $src",
2472 [(set QPR:$dst, (v4f32 (NEONvdup
2473 (f32 (bitconvert GPR:$src)))))]>;
2475 // VDUP : Vector Duplicate Lane (from scalar to all elements)
2477 class VDUPLND<bits<2> op19_18, bits<2> op17_16, string OpcodeStr, ValueType Ty>
2478 : N2V<0b11, 0b11, op19_18, op17_16, 0b11000, 0, 0,
2479 (outs DPR:$dst), (ins DPR:$src, nohash_imm:$lane), IIC_VMOVD,
2480 !strconcat(OpcodeStr, "\t$dst, $src[$lane]"), "",
2481 [(set DPR:$dst, (Ty (NEONvduplane (Ty DPR:$src), imm:$lane)))]>;
2483 class VDUPLNQ<bits<2> op19_18, bits<2> op17_16, string OpcodeStr,
2484 ValueType ResTy, ValueType OpTy>
2485 : N2V<0b11, 0b11, op19_18, op17_16, 0b11000, 1, 0,
2486 (outs QPR:$dst), (ins DPR:$src, nohash_imm:$lane), IIC_VMOVD,
2487 !strconcat(OpcodeStr, "\t$dst, $src[$lane]"), "",
2488 [(set QPR:$dst, (ResTy (NEONvduplane (OpTy DPR:$src), imm:$lane)))]>;
2490 def VDUPLN8d : VDUPLND<0b00, 0b01, "vdup.8", v8i8>;
2491 def VDUPLN16d : VDUPLND<0b00, 0b10, "vdup.16", v4i16>;
2492 def VDUPLN32d : VDUPLND<0b01, 0b00, "vdup.32", v2i32>;
2493 def VDUPLNfd : VDUPLND<0b01, 0b00, "vdup.32", v2f32>;
2494 def VDUPLN8q : VDUPLNQ<0b00, 0b01, "vdup.8", v16i8, v8i8>;
2495 def VDUPLN16q : VDUPLNQ<0b00, 0b10, "vdup.16", v8i16, v4i16>;
2496 def VDUPLN32q : VDUPLNQ<0b01, 0b00, "vdup.32", v4i32, v2i32>;
2497 def VDUPLNfq : VDUPLNQ<0b01, 0b00, "vdup.32", v4f32, v2f32>;
2499 def : Pat<(v16i8 (NEONvduplane (v16i8 QPR:$src), imm:$lane)),
2500 (v16i8 (VDUPLN8q (v8i8 (EXTRACT_SUBREG QPR:$src,
2501 (DSubReg_i8_reg imm:$lane))),
2502 (SubReg_i8_lane imm:$lane)))>;
2503 def : Pat<(v8i16 (NEONvduplane (v8i16 QPR:$src), imm:$lane)),
2504 (v8i16 (VDUPLN16q (v4i16 (EXTRACT_SUBREG QPR:$src,
2505 (DSubReg_i16_reg imm:$lane))),
2506 (SubReg_i16_lane imm:$lane)))>;
2507 def : Pat<(v4i32 (NEONvduplane (v4i32 QPR:$src), imm:$lane)),
2508 (v4i32 (VDUPLN32q (v2i32 (EXTRACT_SUBREG QPR:$src,
2509 (DSubReg_i32_reg imm:$lane))),
2510 (SubReg_i32_lane imm:$lane)))>;
2511 def : Pat<(v4f32 (NEONvduplane (v4f32 QPR:$src), imm:$lane)),
2512 (v4f32 (VDUPLNfq (v2f32 (EXTRACT_SUBREG QPR:$src,
2513 (DSubReg_i32_reg imm:$lane))),
2514 (SubReg_i32_lane imm:$lane)))>;
2516 def VDUPfdf : N2V<0b11, 0b11, 0b01, 0b00, 0b11000, 0, 0,
2517 (outs DPR:$dst), (ins SPR:$src),
2518 IIC_VMOVD, "vdup.32\t$dst, ${src:lane}", "",
2519 [(set DPR:$dst, (v2f32 (NEONvdup (f32 SPR:$src))))]>;
2521 def VDUPfqf : N2V<0b11, 0b11, 0b01, 0b00, 0b11000, 1, 0,
2522 (outs QPR:$dst), (ins SPR:$src),
2523 IIC_VMOVD, "vdup.32\t$dst, ${src:lane}", "",
2524 [(set QPR:$dst, (v4f32 (NEONvdup (f32 SPR:$src))))]>;
2526 def : Pat<(v2i64 (NEONvduplane (v2i64 QPR:$src), imm:$lane)),
2527 (INSERT_SUBREG QPR:$src,
2528 (i64 (EXTRACT_SUBREG QPR:$src, (DSubReg_f64_reg imm:$lane))),
2529 (DSubReg_f64_other_reg imm:$lane))>;
2530 def : Pat<(v2f64 (NEONvduplane (v2f64 QPR:$src), imm:$lane)),
2531 (INSERT_SUBREG QPR:$src,
2532 (f64 (EXTRACT_SUBREG QPR:$src, (DSubReg_f64_reg imm:$lane))),
2533 (DSubReg_f64_other_reg imm:$lane))>;
2535 // VMOVN : Vector Narrowing Move
2536 defm VMOVN : N2VNInt_HSD<0b11,0b11,0b10,0b00100,0,0, IIC_VMOVD, "vmovn.i",
2537 int_arm_neon_vmovn>;
2538 // VQMOVN : Vector Saturating Narrowing Move
2539 defm VQMOVNs : N2VNInt_HSD<0b11,0b11,0b10,0b00101,0,0, IIC_VQUNAiD, "vqmovn.s",
2540 int_arm_neon_vqmovns>;
2541 defm VQMOVNu : N2VNInt_HSD<0b11,0b11,0b10,0b00101,1,0, IIC_VQUNAiD, "vqmovn.u",
2542 int_arm_neon_vqmovnu>;
2543 defm VQMOVNsu : N2VNInt_HSD<0b11,0b11,0b10,0b00100,1,0, IIC_VQUNAiD, "vqmovun.s",
2544 int_arm_neon_vqmovnsu>;
2545 // VMOVL : Vector Lengthening Move
2546 defm VMOVLs : N2VLInt_QHS<0,1,0b1010,0,0,1, "vmovl.s", int_arm_neon_vmovls>;
2547 defm VMOVLu : N2VLInt_QHS<1,1,0b1010,0,0,1, "vmovl.u", int_arm_neon_vmovlu>;
2549 // Vector Conversions.
2551 // VCVT : Vector Convert Between Floating-Point and Integers
2552 def VCVTf2sd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt.s32.f32",
2553 v2i32, v2f32, fp_to_sint>;
2554 def VCVTf2ud : N2VD<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt.u32.f32",
2555 v2i32, v2f32, fp_to_uint>;
2556 def VCVTs2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt.f32.s32",
2557 v2f32, v2i32, sint_to_fp>;
2558 def VCVTu2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt.f32.u32",
2559 v2f32, v2i32, uint_to_fp>;
2561 def VCVTf2sq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt.s32.f32",
2562 v4i32, v4f32, fp_to_sint>;
2563 def VCVTf2uq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt.u32.f32",
2564 v4i32, v4f32, fp_to_uint>;
2565 def VCVTs2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt.f32.s32",
2566 v4f32, v4i32, sint_to_fp>;
2567 def VCVTu2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt.f32.u32",
2568 v4f32, v4i32, uint_to_fp>;
2570 // VCVT : Vector Convert Between Floating-Point and Fixed-Point.
2571 // Note: Some of the opcode bits in the following VCVT instructions need to
2572 // be encoded based on the immed values.
2573 def VCVTf2xsd : N2VCvtD<0, 1, 0b000000, 0b1111, 0, 1, "vcvt.s32.f32",
2574 v2i32, v2f32, int_arm_neon_vcvtfp2fxs>;
2575 def VCVTf2xud : N2VCvtD<1, 1, 0b000000, 0b1111, 0, 1, "vcvt.u32.f32",
2576 v2i32, v2f32, int_arm_neon_vcvtfp2fxu>;
2577 def VCVTxs2fd : N2VCvtD<0, 1, 0b000000, 0b1110, 0, 1, "vcvt.f32.s32",
2578 v2f32, v2i32, int_arm_neon_vcvtfxs2fp>;
2579 def VCVTxu2fd : N2VCvtD<1, 1, 0b000000, 0b1110, 0, 1, "vcvt.f32.u32",
2580 v2f32, v2i32, int_arm_neon_vcvtfxu2fp>;
2582 def VCVTf2xsq : N2VCvtQ<0, 1, 0b000000, 0b1111, 0, 1, "vcvt.s32.f32",
2583 v4i32, v4f32, int_arm_neon_vcvtfp2fxs>;
2584 def VCVTf2xuq : N2VCvtQ<1, 1, 0b000000, 0b1111, 0, 1, "vcvt.u32.f32",
2585 v4i32, v4f32, int_arm_neon_vcvtfp2fxu>;
2586 def VCVTxs2fq : N2VCvtQ<0, 1, 0b000000, 0b1110, 0, 1, "vcvt.f32.s32",
2587 v4f32, v4i32, int_arm_neon_vcvtfxs2fp>;
2588 def VCVTxu2fq : N2VCvtQ<1, 1, 0b000000, 0b1110, 0, 1, "vcvt.f32.u32",
2589 v4f32, v4i32, int_arm_neon_vcvtfxu2fp>;
2593 // VREV64 : Vector Reverse elements within 64-bit doublewords
2595 class VREV64D<bits<2> op19_18, string OpcodeStr, ValueType Ty>
2596 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 0, 0, (outs DPR:$dst),
2597 (ins DPR:$src), IIC_VMOVD,
2598 !strconcat(OpcodeStr, "\t$dst, $src"), "",
2599 [(set DPR:$dst, (Ty (NEONvrev64 (Ty DPR:$src))))]>;
2600 class VREV64Q<bits<2> op19_18, string OpcodeStr, ValueType Ty>
2601 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 1, 0, (outs QPR:$dst),
2602 (ins QPR:$src), IIC_VMOVD,
2603 !strconcat(OpcodeStr, "\t$dst, $src"), "",
2604 [(set QPR:$dst, (Ty (NEONvrev64 (Ty QPR:$src))))]>;
2606 def VREV64d8 : VREV64D<0b00, "vrev64.8", v8i8>;
2607 def VREV64d16 : VREV64D<0b01, "vrev64.16", v4i16>;
2608 def VREV64d32 : VREV64D<0b10, "vrev64.32", v2i32>;
2609 def VREV64df : VREV64D<0b10, "vrev64.32", v2f32>;
2611 def VREV64q8 : VREV64Q<0b00, "vrev64.8", v16i8>;
2612 def VREV64q16 : VREV64Q<0b01, "vrev64.16", v8i16>;
2613 def VREV64q32 : VREV64Q<0b10, "vrev64.32", v4i32>;
2614 def VREV64qf : VREV64Q<0b10, "vrev64.32", v4f32>;
2616 // VREV32 : Vector Reverse elements within 32-bit words
2618 class VREV32D<bits<2> op19_18, string OpcodeStr, ValueType Ty>
2619 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 0, 0, (outs DPR:$dst),
2620 (ins DPR:$src), IIC_VMOVD,
2621 !strconcat(OpcodeStr, "\t$dst, $src"), "",
2622 [(set DPR:$dst, (Ty (NEONvrev32 (Ty DPR:$src))))]>;
2623 class VREV32Q<bits<2> op19_18, string OpcodeStr, ValueType Ty>
2624 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 1, 0, (outs QPR:$dst),
2625 (ins QPR:$src), IIC_VMOVD,
2626 !strconcat(OpcodeStr, "\t$dst, $src"), "",
2627 [(set QPR:$dst, (Ty (NEONvrev32 (Ty QPR:$src))))]>;
2629 def VREV32d8 : VREV32D<0b00, "vrev32.8", v8i8>;
2630 def VREV32d16 : VREV32D<0b01, "vrev32.16", v4i16>;
2632 def VREV32q8 : VREV32Q<0b00, "vrev32.8", v16i8>;
2633 def VREV32q16 : VREV32Q<0b01, "vrev32.16", v8i16>;
2635 // VREV16 : Vector Reverse elements within 16-bit halfwords
2637 class VREV16D<bits<2> op19_18, string OpcodeStr, ValueType Ty>
2638 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 0, 0, (outs DPR:$dst),
2639 (ins DPR:$src), IIC_VMOVD,
2640 !strconcat(OpcodeStr, "\t$dst, $src"), "",
2641 [(set DPR:$dst, (Ty (NEONvrev16 (Ty DPR:$src))))]>;
2642 class VREV16Q<bits<2> op19_18, string OpcodeStr, ValueType Ty>
2643 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 1, 0, (outs QPR:$dst),
2644 (ins QPR:$src), IIC_VMOVD,
2645 !strconcat(OpcodeStr, "\t$dst, $src"), "",
2646 [(set QPR:$dst, (Ty (NEONvrev16 (Ty QPR:$src))))]>;
2648 def VREV16d8 : VREV16D<0b00, "vrev16.8", v8i8>;
2649 def VREV16q8 : VREV16Q<0b00, "vrev16.8", v16i8>;
2651 // Other Vector Shuffles.
2653 // VEXT : Vector Extract
2655 class VEXTd<string OpcodeStr, ValueType Ty>
2656 : N3V<0,1,0b11,0b0000,0,0, (outs DPR:$dst),
2657 (ins DPR:$lhs, DPR:$rhs, i32imm:$index), IIC_VEXTD,
2658 !strconcat(OpcodeStr, "\t$dst, $lhs, $rhs, $index"), "",
2659 [(set DPR:$dst, (Ty (NEONvext (Ty DPR:$lhs),
2660 (Ty DPR:$rhs), imm:$index)))]>;
2662 class VEXTq<string OpcodeStr, ValueType Ty>
2663 : N3V<0,1,0b11,0b0000,1,0, (outs QPR:$dst),
2664 (ins QPR:$lhs, QPR:$rhs, i32imm:$index), IIC_VEXTQ,
2665 !strconcat(OpcodeStr, "\t$dst, $lhs, $rhs, $index"), "",
2666 [(set QPR:$dst, (Ty (NEONvext (Ty QPR:$lhs),
2667 (Ty QPR:$rhs), imm:$index)))]>;
2669 def VEXTd8 : VEXTd<"vext.8", v8i8>;
2670 def VEXTd16 : VEXTd<"vext.16", v4i16>;
2671 def VEXTd32 : VEXTd<"vext.32", v2i32>;
2672 def VEXTdf : VEXTd<"vext.32", v2f32>;
2674 def VEXTq8 : VEXTq<"vext.8", v16i8>;
2675 def VEXTq16 : VEXTq<"vext.16", v8i16>;
2676 def VEXTq32 : VEXTq<"vext.32", v4i32>;
2677 def VEXTqf : VEXTq<"vext.32", v4f32>;
2679 // VTRN : Vector Transpose
2681 def VTRNd8 : N2VDShuffle<0b00, 0b00001, "vtrn.8">;
2682 def VTRNd16 : N2VDShuffle<0b01, 0b00001, "vtrn.16">;
2683 def VTRNd32 : N2VDShuffle<0b10, 0b00001, "vtrn.32">;
2685 def VTRNq8 : N2VQShuffle<0b00, 0b00001, IIC_VPERMQ, "vtrn.8">;
2686 def VTRNq16 : N2VQShuffle<0b01, 0b00001, IIC_VPERMQ, "vtrn.16">;
2687 def VTRNq32 : N2VQShuffle<0b10, 0b00001, IIC_VPERMQ, "vtrn.32">;
2689 // VUZP : Vector Unzip (Deinterleave)
2691 def VUZPd8 : N2VDShuffle<0b00, 0b00010, "vuzp.8">;
2692 def VUZPd16 : N2VDShuffle<0b01, 0b00010, "vuzp.16">;
2693 def VUZPd32 : N2VDShuffle<0b10, 0b00010, "vuzp.32">;
2695 def VUZPq8 : N2VQShuffle<0b00, 0b00010, IIC_VPERMQ3, "vuzp.8">;
2696 def VUZPq16 : N2VQShuffle<0b01, 0b00010, IIC_VPERMQ3, "vuzp.16">;
2697 def VUZPq32 : N2VQShuffle<0b10, 0b00010, IIC_VPERMQ3, "vuzp.32">;
2699 // VZIP : Vector Zip (Interleave)
2701 def VZIPd8 : N2VDShuffle<0b00, 0b00011, "vzip.8">;
2702 def VZIPd16 : N2VDShuffle<0b01, 0b00011, "vzip.16">;
2703 def VZIPd32 : N2VDShuffle<0b10, 0b00011, "vzip.32">;
2705 def VZIPq8 : N2VQShuffle<0b00, 0b00011, IIC_VPERMQ3, "vzip.8">;
2706 def VZIPq16 : N2VQShuffle<0b01, 0b00011, IIC_VPERMQ3, "vzip.16">;
2707 def VZIPq32 : N2VQShuffle<0b10, 0b00011, IIC_VPERMQ3, "vzip.32">;
2709 // Vector Table Lookup and Table Extension.
2711 // VTBL : Vector Table Lookup
2713 : N3V<1,1,0b11,0b1000,0,0, (outs DPR:$dst),
2714 (ins DPR:$tbl1, DPR:$src), IIC_VTB1,
2715 "vtbl.8\t$dst, \\{$tbl1\\}, $src", "",
2716 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl1 DPR:$tbl1, DPR:$src)))]>;
2717 let hasExtraSrcRegAllocReq = 1 in {
2719 : N3V<1,1,0b11,0b1001,0,0, (outs DPR:$dst),
2720 (ins DPR:$tbl1, DPR:$tbl2, DPR:$src), IIC_VTB2,
2721 "vtbl.8\t$dst, \\{$tbl1,$tbl2\\}, $src", "",
2722 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl2
2723 DPR:$tbl1, DPR:$tbl2, DPR:$src)))]>;
2725 : N3V<1,1,0b11,0b1010,0,0, (outs DPR:$dst),
2726 (ins DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$src), IIC_VTB3,
2727 "vtbl.8\t$dst, \\{$tbl1,$tbl2,$tbl3\\}, $src", "",
2728 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl3
2729 DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$src)))]>;
2731 : N3V<1,1,0b11,0b1011,0,0, (outs DPR:$dst),
2732 (ins DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$src), IIC_VTB4,
2733 "vtbl.8\t$dst, \\{$tbl1,$tbl2,$tbl3,$tbl4\\}, $src", "",
2734 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl4 DPR:$tbl1, DPR:$tbl2,
2735 DPR:$tbl3, DPR:$tbl4, DPR:$src)))]>;
2736 } // hasExtraSrcRegAllocReq = 1
2738 // VTBX : Vector Table Extension
2740 : N3V<1,1,0b11,0b1000,1,0, (outs DPR:$dst),
2741 (ins DPR:$orig, DPR:$tbl1, DPR:$src), IIC_VTBX1,
2742 "vtbx.8\t$dst, \\{$tbl1\\}, $src", "$orig = $dst",
2743 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx1
2744 DPR:$orig, DPR:$tbl1, DPR:$src)))]>;
2745 let hasExtraSrcRegAllocReq = 1 in {
2747 : N3V<1,1,0b11,0b1001,1,0, (outs DPR:$dst),
2748 (ins DPR:$orig, DPR:$tbl1, DPR:$tbl2, DPR:$src), IIC_VTBX2,
2749 "vtbx.8\t$dst, \\{$tbl1,$tbl2\\}, $src", "$orig = $dst",
2750 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx2
2751 DPR:$orig, DPR:$tbl1, DPR:$tbl2, DPR:$src)))]>;
2753 : N3V<1,1,0b11,0b1010,1,0, (outs DPR:$dst),
2754 (ins DPR:$orig, DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$src), IIC_VTBX3,
2755 "vtbx.8\t$dst, \\{$tbl1,$tbl2,$tbl3\\}, $src", "$orig = $dst",
2756 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx3 DPR:$orig, DPR:$tbl1,
2757 DPR:$tbl2, DPR:$tbl3, DPR:$src)))]>;
2759 : N3V<1,1,0b11,0b1011,1,0, (outs DPR:$dst), (ins DPR:$orig, DPR:$tbl1,
2760 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$src), IIC_VTBX4,
2761 "vtbx.8\t$dst, \\{$tbl1,$tbl2,$tbl3,$tbl4\\}, $src", "$orig = $dst",
2762 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx4 DPR:$orig, DPR:$tbl1,
2763 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$src)))]>;
2764 } // hasExtraSrcRegAllocReq = 1
2766 //===----------------------------------------------------------------------===//
2767 // NEON instructions for single-precision FP math
2768 //===----------------------------------------------------------------------===//
2770 // These need separate instructions because they must use DPR_VFP2 register
2771 // class which have SPR sub-registers.
2773 // Vector Add Operations used for single-precision FP
2774 let neverHasSideEffects = 1 in
2775 def VADDfd_sfp : N3VDs<0, 0, 0b00, 0b1101, 0, "vadd.f32", v2f32, v2f32, fadd,1>;
2776 def : N3VDsPat<fadd, VADDfd_sfp>;
2778 // Vector Sub Operations used for single-precision FP
2779 let neverHasSideEffects = 1 in
2780 def VSUBfd_sfp : N3VDs<0, 0, 0b10, 0b1101, 0, "vsub.f32", v2f32, v2f32, fsub,0>;
2781 def : N3VDsPat<fsub, VSUBfd_sfp>;
2783 // Vector Multiply Operations used for single-precision FP
2784 let neverHasSideEffects = 1 in
2785 def VMULfd_sfp : N3VDs<1, 0, 0b00, 0b1101, 1, "vmul.f32", v2f32, v2f32, fmul,1>;
2786 def : N3VDsPat<fmul, VMULfd_sfp>;
2788 // Vector Multiply-Accumulate/Subtract used for single-precision FP
2789 let neverHasSideEffects = 1 in
2790 def VMLAfd_sfp : N3VDMulOps<0, 0, 0b00, 0b1101, 1, IIC_VMACD, "vmla.f32", v2f32,fmul,fadd>;
2791 def : N3VDMulOpsPat<fmul, fadd, VMLAfd_sfp>;
2793 let neverHasSideEffects = 1 in
2794 def VMLSfd_sfp : N3VDMulOps<0, 0, 0b10, 0b1101, 1, IIC_VMACD, "vmls.f32", v2f32,fmul,fsub>;
2795 def : N3VDMulOpsPat<fmul, fsub, VMLSfd_sfp>;
2797 // Vector Absolute used for single-precision FP
2798 let neverHasSideEffects = 1 in
2799 def VABSfd_sfp : N2VDInts<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
2800 IIC_VUNAD, "vabs.f32",
2801 v2f32, v2f32, int_arm_neon_vabs>;
2802 def : N2VDIntsPat<fabs, VABSfd_sfp>;
2804 // Vector Negate used for single-precision FP
2805 let neverHasSideEffects = 1 in
2806 def VNEGf32d_sfp : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
2807 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), IIC_VUNAD,
2808 "vneg.f32\t$dst, $src", "", []>;
2809 def : N2VDIntsPat<fneg, VNEGf32d_sfp>;
2811 // Vector Convert between single-precision FP and integer
2812 let neverHasSideEffects = 1 in
2813 def VCVTf2sd_sfp : N2VDs<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt.s32.f32",
2814 v2i32, v2f32, fp_to_sint>;
2815 def : N2VDsPat<arm_ftosi, f32, v2f32, VCVTf2sd_sfp>;
2817 let neverHasSideEffects = 1 in
2818 def VCVTf2ud_sfp : N2VDs<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt.u32.f32",
2819 v2i32, v2f32, fp_to_uint>;
2820 def : N2VDsPat<arm_ftoui, f32, v2f32, VCVTf2ud_sfp>;
2822 let neverHasSideEffects = 1 in
2823 def VCVTs2fd_sfp : N2VDs<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt.f32.s32",
2824 v2f32, v2i32, sint_to_fp>;
2825 def : N2VDsPat<arm_sitof, f32, v2i32, VCVTs2fd_sfp>;
2827 let neverHasSideEffects = 1 in
2828 def VCVTu2fd_sfp : N2VDs<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt.f32.u32",
2829 v2f32, v2i32, uint_to_fp>;
2830 def : N2VDsPat<arm_uitof, f32, v2i32, VCVTu2fd_sfp>;
2832 //===----------------------------------------------------------------------===//
2833 // Non-Instruction Patterns
2834 //===----------------------------------------------------------------------===//
2837 def : Pat<(v1i64 (bitconvert (v2i32 DPR:$src))), (v1i64 DPR:$src)>;
2838 def : Pat<(v1i64 (bitconvert (v4i16 DPR:$src))), (v1i64 DPR:$src)>;
2839 def : Pat<(v1i64 (bitconvert (v8i8 DPR:$src))), (v1i64 DPR:$src)>;
2840 def : Pat<(v1i64 (bitconvert (f64 DPR:$src))), (v1i64 DPR:$src)>;
2841 def : Pat<(v1i64 (bitconvert (v2f32 DPR:$src))), (v1i64 DPR:$src)>;
2842 def : Pat<(v2i32 (bitconvert (v1i64 DPR:$src))), (v2i32 DPR:$src)>;
2843 def : Pat<(v2i32 (bitconvert (v4i16 DPR:$src))), (v2i32 DPR:$src)>;
2844 def : Pat<(v2i32 (bitconvert (v8i8 DPR:$src))), (v2i32 DPR:$src)>;
2845 def : Pat<(v2i32 (bitconvert (f64 DPR:$src))), (v2i32 DPR:$src)>;
2846 def : Pat<(v2i32 (bitconvert (v2f32 DPR:$src))), (v2i32 DPR:$src)>;
2847 def : Pat<(v4i16 (bitconvert (v1i64 DPR:$src))), (v4i16 DPR:$src)>;
2848 def : Pat<(v4i16 (bitconvert (v2i32 DPR:$src))), (v4i16 DPR:$src)>;
2849 def : Pat<(v4i16 (bitconvert (v8i8 DPR:$src))), (v4i16 DPR:$src)>;
2850 def : Pat<(v4i16 (bitconvert (f64 DPR:$src))), (v4i16 DPR:$src)>;
2851 def : Pat<(v4i16 (bitconvert (v2f32 DPR:$src))), (v4i16 DPR:$src)>;
2852 def : Pat<(v8i8 (bitconvert (v1i64 DPR:$src))), (v8i8 DPR:$src)>;
2853 def : Pat<(v8i8 (bitconvert (v2i32 DPR:$src))), (v8i8 DPR:$src)>;
2854 def : Pat<(v8i8 (bitconvert (v4i16 DPR:$src))), (v8i8 DPR:$src)>;
2855 def : Pat<(v8i8 (bitconvert (f64 DPR:$src))), (v8i8 DPR:$src)>;
2856 def : Pat<(v8i8 (bitconvert (v2f32 DPR:$src))), (v8i8 DPR:$src)>;
2857 def : Pat<(f64 (bitconvert (v1i64 DPR:$src))), (f64 DPR:$src)>;
2858 def : Pat<(f64 (bitconvert (v2i32 DPR:$src))), (f64 DPR:$src)>;
2859 def : Pat<(f64 (bitconvert (v4i16 DPR:$src))), (f64 DPR:$src)>;
2860 def : Pat<(f64 (bitconvert (v8i8 DPR:$src))), (f64 DPR:$src)>;
2861 def : Pat<(f64 (bitconvert (v2f32 DPR:$src))), (f64 DPR:$src)>;
2862 def : Pat<(v2f32 (bitconvert (f64 DPR:$src))), (v2f32 DPR:$src)>;
2863 def : Pat<(v2f32 (bitconvert (v1i64 DPR:$src))), (v2f32 DPR:$src)>;
2864 def : Pat<(v2f32 (bitconvert (v2i32 DPR:$src))), (v2f32 DPR:$src)>;
2865 def : Pat<(v2f32 (bitconvert (v4i16 DPR:$src))), (v2f32 DPR:$src)>;
2866 def : Pat<(v2f32 (bitconvert (v8i8 DPR:$src))), (v2f32 DPR:$src)>;
2868 def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (v2i64 QPR:$src)>;
2869 def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (v2i64 QPR:$src)>;
2870 def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (v2i64 QPR:$src)>;
2871 def : Pat<(v2i64 (bitconvert (v2f64 QPR:$src))), (v2i64 QPR:$src)>;
2872 def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (v2i64 QPR:$src)>;
2873 def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (v4i32 QPR:$src)>;
2874 def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (v4i32 QPR:$src)>;
2875 def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (v4i32 QPR:$src)>;
2876 def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (v4i32 QPR:$src)>;
2877 def : Pat<(v4i32 (bitconvert (v4f32 QPR:$src))), (v4i32 QPR:$src)>;
2878 def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (v8i16 QPR:$src)>;
2879 def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (v8i16 QPR:$src)>;
2880 def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (v8i16 QPR:$src)>;
2881 def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (v8i16 QPR:$src)>;
2882 def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (v8i16 QPR:$src)>;
2883 def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (v16i8 QPR:$src)>;
2884 def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (v16i8 QPR:$src)>;
2885 def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (v16i8 QPR:$src)>;
2886 def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (v16i8 QPR:$src)>;
2887 def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (v16i8 QPR:$src)>;
2888 def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (v4f32 QPR:$src)>;
2889 def : Pat<(v4f32 (bitconvert (v4i32 QPR:$src))), (v4f32 QPR:$src)>;
2890 def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (v4f32 QPR:$src)>;
2891 def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (v4f32 QPR:$src)>;
2892 def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (v4f32 QPR:$src)>;
2893 def : Pat<(v2f64 (bitconvert (v2i64 QPR:$src))), (v2f64 QPR:$src)>;
2894 def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (v2f64 QPR:$src)>;
2895 def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 QPR:$src)>;
2896 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 QPR:$src)>;
2897 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 QPR:$src)>;