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\\}, $addr", "", []>;
255 // vld4 to double-spaced even registers.
256 def VLD4q8a : VLD4WB<0b0000, "vld4.8">;
257 def VLD4q16a : VLD4WB<0b0100, "vld4.16">;
258 def VLD4q32a : VLD4WB<0b1000, "vld4.32">;
260 // vld4 to double-spaced odd registers.
261 def VLD4q8b : VLD4WB<0b0000, "vld4.8">;
262 def VLD4q16b : VLD4WB<0b0100, "vld4.16">;
263 def VLD4q32b : VLD4WB<0b1000, "vld4.32">;
265 // VLD1LN : Vector Load (single element to one lane)
266 // FIXME: Not yet implemented.
268 // VLD2LN : Vector Load (single 2-element structure to one lane)
269 class VLD2LND<bits<4> op11_8, string OpcodeStr>
270 : NLdSt<1,0b10,op11_8,0b0000, (outs DPR:$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 = $dst2", []>;
276 def VLD2LNd8 : VLD2LND<0b0001, "vld2.8">;
277 def VLD2LNd16 : VLD2LND<0b0101, "vld2.16">;
278 def VLD2LNd32 : VLD2LND<0b1001, "vld2.32">;
280 // VLD3LN : Vector Load (single 3-element structure to one lane)
281 class VLD3LND<bits<4> op11_8, string OpcodeStr>
282 : NLdSt<1,0b10,op11_8,0b0000, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3),
283 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3,
284 nohash_imm:$lane), IIC_VLD3,
285 !strconcat(OpcodeStr,
286 "\t\\{$dst1[$lane],$dst2[$lane],$dst3[$lane]\\}, $addr"),
287 "$src1 = $dst1, $src2 = $dst2, $src3 = $dst3", []>;
289 def VLD3LNd8 : VLD3LND<0b0010, "vld3.8">;
290 def VLD3LNd16 : VLD3LND<0b0110, "vld3.16">;
291 def VLD3LNd32 : VLD3LND<0b1010, "vld3.32">;
293 // VLD4LN : Vector Load (single 4-element structure to one lane)
294 class VLD4LND<bits<4> op11_8, string OpcodeStr>
295 : NLdSt<1,0b10,op11_8,0b0000,
296 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
297 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
298 nohash_imm:$lane), IIC_VLD4,
299 !strconcat(OpcodeStr,
300 "\t\\{$dst1[$lane],$dst2[$lane],$dst3[$lane],$dst4[$lane]\\}, $addr"),
301 "$src1 = $dst1, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4", []>;
303 def VLD4LNd8 : VLD4LND<0b0011, "vld4.8">;
304 def VLD4LNd16 : VLD4LND<0b0111, "vld4.16">;
305 def VLD4LNd32 : VLD4LND<0b1011, "vld4.32">;
307 // VLD1DUP : Vector Load (single element to all lanes)
308 // VLD2DUP : Vector Load (single 2-element structure to all lanes)
309 // VLD3DUP : Vector Load (single 3-element structure to all lanes)
310 // VLD4DUP : Vector Load (single 4-element structure to all lanes)
311 // FIXME: Not yet implemented.
312 } // mayLoad = 1, hasExtraDefRegAllocReq = 1
314 // VST1 : Vector Store (multiple single elements)
315 class VST1D<bits<4> op7_4, string OpcodeStr, ValueType Ty, Intrinsic IntOp>
316 : NLdSt<0,0b00,0b0111,op7_4, (outs), (ins addrmode6:$addr, DPR:$src), IIC_VST,
317 !strconcat(OpcodeStr, "\t\\{$src\\}, $addr"), "",
318 [(IntOp addrmode6:$addr, (Ty DPR:$src))]>;
319 class VST1Q<bits<4> op7_4, string OpcodeStr, ValueType Ty, Intrinsic IntOp>
320 : NLdSt<0,0b00,0b1010,op7_4, (outs), (ins addrmode6:$addr, QPR:$src), IIC_VST,
321 !strconcat(OpcodeStr, "\t${src:dregpair}, $addr"), "",
322 [(IntOp addrmode6:$addr, (Ty QPR:$src))]>;
324 let hasExtraSrcRegAllocReq = 1 in {
325 def VST1d8 : VST1D<0b0000, "vst1.8", v8i8, int_arm_neon_vst1>;
326 def VST1d16 : VST1D<0b0100, "vst1.16", v4i16, int_arm_neon_vst1>;
327 def VST1d32 : VST1D<0b1000, "vst1.32", v2i32, int_arm_neon_vst1>;
328 def VST1df : VST1D<0b1000, "vst1.32", v2f32, int_arm_neon_vst1>;
329 def VST1d64 : VST1D<0b1100, "vst1.64", v1i64, int_arm_neon_vst1>;
331 def VST1q8 : VST1Q<0b0000, "vst1.8", v16i8, int_arm_neon_vst1>;
332 def VST1q16 : VST1Q<0b0100, "vst1.16", v8i16, int_arm_neon_vst1>;
333 def VST1q32 : VST1Q<0b1000, "vst1.32", v4i32, int_arm_neon_vst1>;
334 def VST1qf : VST1Q<0b1000, "vst1.32", v4f32, int_arm_neon_vst1>;
335 def VST1q64 : VST1Q<0b1100, "vst1.64", v2i64, int_arm_neon_vst1>;
336 } // hasExtraSrcRegAllocReq
338 let mayStore = 1, hasExtraSrcRegAllocReq = 1 in {
340 // VST2 : Vector Store (multiple 2-element structures)
341 class VST2D<bits<4> op7_4, string OpcodeStr>
342 : NLdSt<0,0b00,0b1000,op7_4, (outs),
343 (ins addrmode6:$addr, DPR:$src1, DPR:$src2), IIC_VST,
344 !strconcat(OpcodeStr, "\t\\{$src1,$src2\\}, $addr"), "", []>;
345 class VST2Q<bits<4> op7_4, string OpcodeStr>
346 : NLdSt<0,0b00,0b0011,op7_4, (outs),
347 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
349 !strconcat(OpcodeStr, "\t\\{$src1,$src2,$src3,$src4\\}, $addr"),
352 def VST2d8 : VST2D<0b0000, "vst2.8">;
353 def VST2d16 : VST2D<0b0100, "vst2.16">;
354 def VST2d32 : VST2D<0b1000, "vst2.32">;
356 def VST2q8 : VST2Q<0b0000, "vst2.8">;
357 def VST2q16 : VST2Q<0b0100, "vst2.16">;
358 def VST2q32 : VST2Q<0b1000, "vst2.32">;
360 // VST3 : Vector Store (multiple 3-element structures)
361 class VST3D<bits<4> op7_4, string OpcodeStr>
362 : NLdSt<0,0b00,0b0100,op7_4, (outs),
363 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3), IIC_VST,
364 !strconcat(OpcodeStr, "\t\\{$src1,$src2,$src3\\}, $addr"), "", []>;
365 class VST3WB<bits<4> op7_4, string OpcodeStr>
366 : NLdSt<0,0b00,0b0101,op7_4, (outs GPR:$wb),
367 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3), IIC_VST,
368 !strconcat(OpcodeStr, "\t\\{$src1,$src2,$src3\\}, $addr"),
369 "$addr.addr = $wb", []>;
371 def VST3d8 : VST3D<0b0000, "vst3.8">;
372 def VST3d16 : VST3D<0b0100, "vst3.16">;
373 def VST3d32 : VST3D<0b1000, "vst3.32">;
375 // vst3 to double-spaced even registers.
376 def VST3q8a : VST3WB<0b0000, "vst3.8">;
377 def VST3q16a : VST3WB<0b0100, "vst3.16">;
378 def VST3q32a : VST3WB<0b1000, "vst3.32">;
380 // vst3 to double-spaced odd registers.
381 def VST3q8b : VST3WB<0b0000, "vst3.8">;
382 def VST3q16b : VST3WB<0b0100, "vst3.16">;
383 def VST3q32b : VST3WB<0b1000, "vst3.32">;
385 // VST4 : Vector Store (multiple 4-element structures)
386 class VST4D<bits<4> op7_4, string OpcodeStr>
387 : NLdSt<0,0b00,0b0000,op7_4, (outs),
388 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
390 !strconcat(OpcodeStr, "\t\\{$src1,$src2,$src3,$src4\\}, $addr"),
392 class VST4WB<bits<4> op7_4, string OpcodeStr>
393 : NLdSt<0,0b00,0b0001,op7_4, (outs GPR:$wb),
394 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
396 !strconcat(OpcodeStr, "\t\\{$src1,$src2,$src3,$src4\\}, $addr"),
397 "$addr.addr = $wb", []>;
399 def VST4d8 : VST4D<0b0000, "vst4.8">;
400 def VST4d16 : VST4D<0b0100, "vst4.16">;
401 def VST4d32 : VST4D<0b1000, "vst4.32">;
403 // vst4 to double-spaced even registers.
404 def VST4q8a : VST4WB<0b0000, "vst4.8">;
405 def VST4q16a : VST4WB<0b0100, "vst4.16">;
406 def VST4q32a : VST4WB<0b1000, "vst4.32">;
408 // vst4 to double-spaced odd registers.
409 def VST4q8b : VST4WB<0b0000, "vst4.8">;
410 def VST4q16b : VST4WB<0b0100, "vst4.16">;
411 def VST4q32b : VST4WB<0b1000, "vst4.32">;
413 // VST1LN : Vector Store (single element from one lane)
414 // FIXME: Not yet implemented.
416 // VST2LN : Vector Store (single 2-element structure from one lane)
417 class VST2LND<bits<4> op11_8, string OpcodeStr>
418 : NLdSt<1,0b00,op11_8,0b0000, (outs),
419 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, nohash_imm:$lane),
421 !strconcat(OpcodeStr, "\t\\{$src1[$lane],$src2[$lane]\\}, $addr"),
424 def VST2LNd8 : VST2LND<0b0000, "vst2.8">;
425 def VST2LNd16 : VST2LND<0b0100, "vst2.16">;
426 def VST2LNd32 : VST2LND<0b1000, "vst2.32">;
428 // VST3LN : Vector Store (single 3-element structure from one lane)
429 class VST3LND<bits<4> op11_8, string OpcodeStr>
430 : NLdSt<1,0b00,op11_8,0b0000, (outs),
431 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3,
432 nohash_imm:$lane), IIC_VST,
433 !strconcat(OpcodeStr,
434 "\t\\{$src1[$lane],$src2[$lane],$src3[$lane]\\}, $addr"), "", []>;
436 def VST3LNd8 : VST3LND<0b0010, "vst3.8">;
437 def VST3LNd16 : VST3LND<0b0110, "vst3.16">;
438 def VST3LNd32 : VST3LND<0b1010, "vst3.32">;
440 // VST4LN : Vector Store (single 4-element structure from one lane)
441 class VST4LND<bits<4> op11_8, string OpcodeStr>
442 : NLdSt<1,0b00,op11_8,0b0000, (outs),
443 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
444 nohash_imm:$lane), IIC_VST,
445 !strconcat(OpcodeStr,
446 "\t\\{$src1[$lane],$src2[$lane],$src3[$lane],$src4[$lane]\\}, $addr"),
449 def VST4LNd8 : VST4LND<0b0011, "vst4.8">;
450 def VST4LNd16 : VST4LND<0b0111, "vst4.16">;
451 def VST4LNd32 : VST4LND<0b1011, "vst4.32">;
452 } // mayStore = 1, hasExtraSrcRegAllocReq = 1
455 //===----------------------------------------------------------------------===//
456 // NEON pattern fragments
457 //===----------------------------------------------------------------------===//
459 // Extract D sub-registers of Q registers.
460 // (arm_dsubreg_0 is 5; arm_dsubreg_1 is 6)
461 def DSubReg_i8_reg : SDNodeXForm<imm, [{
462 return CurDAG->getTargetConstant(5 + N->getZExtValue() / 8, MVT::i32);
464 def DSubReg_i16_reg : SDNodeXForm<imm, [{
465 return CurDAG->getTargetConstant(5 + N->getZExtValue() / 4, MVT::i32);
467 def DSubReg_i32_reg : SDNodeXForm<imm, [{
468 return CurDAG->getTargetConstant(5 + N->getZExtValue() / 2, MVT::i32);
470 def DSubReg_f64_reg : SDNodeXForm<imm, [{
471 return CurDAG->getTargetConstant(5 + N->getZExtValue(), MVT::i32);
473 def DSubReg_f64_other_reg : SDNodeXForm<imm, [{
474 return CurDAG->getTargetConstant(5 + (1 - N->getZExtValue()), MVT::i32);
477 // Extract S sub-registers of Q/D registers.
478 // (arm_ssubreg_0 is 1; arm_ssubreg_1 is 2; etc.)
479 def SSubReg_f32_reg : SDNodeXForm<imm, [{
480 return CurDAG->getTargetConstant(1 + N->getZExtValue(), MVT::i32);
483 // Translate lane numbers from Q registers to D subregs.
484 def SubReg_i8_lane : SDNodeXForm<imm, [{
485 return CurDAG->getTargetConstant(N->getZExtValue() & 7, MVT::i32);
487 def SubReg_i16_lane : SDNodeXForm<imm, [{
488 return CurDAG->getTargetConstant(N->getZExtValue() & 3, MVT::i32);
490 def SubReg_i32_lane : SDNodeXForm<imm, [{
491 return CurDAG->getTargetConstant(N->getZExtValue() & 1, MVT::i32);
494 //===----------------------------------------------------------------------===//
495 // Instruction Classes
496 //===----------------------------------------------------------------------===//
498 // Basic 2-register operations, both double- and quad-register.
499 class N2VD<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
500 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
501 ValueType ResTy, ValueType OpTy, SDNode OpNode>
502 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
503 (ins DPR:$src), IIC_VUNAD, !strconcat(OpcodeStr, "\t$dst, $src"), "",
504 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src))))]>;
505 class N2VQ<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
506 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
507 ValueType ResTy, ValueType OpTy, SDNode OpNode>
508 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
509 (ins QPR:$src), IIC_VUNAQ, !strconcat(OpcodeStr, "\t$dst, $src"), "",
510 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src))))]>;
512 // Basic 2-register operations, scalar single-precision.
513 class N2VDs<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
514 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
515 ValueType ResTy, ValueType OpTy, SDNode OpNode>
516 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
517 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src),
518 IIC_VUNAD, !strconcat(OpcodeStr, "\t$dst, $src"), "", []>;
520 class N2VDsPat<SDNode OpNode, ValueType ResTy, ValueType OpTy, NeonI Inst>
521 : NEONFPPat<(ResTy (OpNode SPR:$a)),
523 (Inst (INSERT_SUBREG (OpTy (IMPLICIT_DEF)), SPR:$a, arm_ssubreg_0)),
526 // Basic 2-register intrinsics, both double- and quad-register.
527 class N2VDInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
528 bits<2> op17_16, bits<5> op11_7, bit op4,
529 InstrItinClass itin, string OpcodeStr,
530 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
531 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
532 (ins DPR:$src), itin, !strconcat(OpcodeStr, "\t$dst, $src"), "",
533 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
534 class N2VQInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
535 bits<2> op17_16, bits<5> op11_7, bit op4,
536 InstrItinClass itin, string OpcodeStr,
537 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
538 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
539 (ins QPR:$src), itin, !strconcat(OpcodeStr, "\t$dst, $src"), "",
540 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
542 // Basic 2-register intrinsics, scalar single-precision
543 class N2VDInts<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
544 bits<2> op17_16, bits<5> op11_7, bit op4,
545 InstrItinClass itin, string OpcodeStr,
546 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
547 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
548 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), itin,
549 !strconcat(OpcodeStr, "\t$dst, $src"), "", []>;
551 class N2VDIntsPat<SDNode OpNode, NeonI Inst>
552 : NEONFPPat<(f32 (OpNode SPR:$a)),
554 (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$a, arm_ssubreg_0)),
557 // Narrow 2-register intrinsics.
558 class N2VNInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
559 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
560 InstrItinClass itin, string OpcodeStr,
561 ValueType TyD, ValueType TyQ, Intrinsic IntOp>
562 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$dst),
563 (ins QPR:$src), itin, !strconcat(OpcodeStr, "\t$dst, $src"), "",
564 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src))))]>;
566 // Long 2-register intrinsics. (This is currently only used for VMOVL and is
567 // derived from N2VImm instead of N2V because of the way the size is encoded.)
568 class N2VLInt<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
569 bit op6, bit op4, InstrItinClass itin, string OpcodeStr,
570 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
571 : N2VImm<op24, op23, op21_16, op11_8, op7, op6, op4, (outs QPR:$dst),
572 (ins DPR:$src), itin, !strconcat(OpcodeStr, "\t$dst, $src"), "",
573 [(set QPR:$dst, (TyQ (IntOp (TyD DPR:$src))))]>;
575 // 2-register shuffles (VTRN/VZIP/VUZP), both double- and quad-register.
576 class N2VDShuffle<bits<2> op19_18, bits<5> op11_7, string OpcodeStr>
577 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 0, 0, (outs DPR:$dst1, DPR:$dst2),
578 (ins DPR:$src1, DPR:$src2), IIC_VPERMD,
579 !strconcat(OpcodeStr, "\t$dst1, $dst2"),
580 "$src1 = $dst1, $src2 = $dst2", []>;
581 class N2VQShuffle<bits<2> op19_18, bits<5> op11_7,
582 InstrItinClass itin, string OpcodeStr>
583 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 1, 0, (outs QPR:$dst1, QPR:$dst2),
584 (ins QPR:$src1, QPR:$src2), itin,
585 !strconcat(OpcodeStr, "\t$dst1, $dst2"),
586 "$src1 = $dst1, $src2 = $dst2", []>;
588 // Basic 3-register operations, both double- and quad-register.
589 class N3VD<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
590 InstrItinClass itin, string OpcodeStr, ValueType ResTy, ValueType OpTy,
591 SDNode OpNode, bit Commutable>
592 : N3V<op24, op23, op21_20, op11_8, 0, op4,
593 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), itin,
594 !strconcat(OpcodeStr, "\t$dst, $src1, $src2"), "",
595 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src1), (OpTy DPR:$src2))))]> {
596 let isCommutable = Commutable;
598 class N3VDSL<bits<2> op21_20, bits<4> op11_8,
599 InstrItinClass itin, string OpcodeStr, ValueType Ty, SDNode ShOp>
600 : N3V<0, 1, op21_20, op11_8, 1, 0,
601 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
602 itin, !strconcat(OpcodeStr, "\t$dst, $src1, $src2[$lane]"), "",
604 (Ty (ShOp (Ty DPR:$src1),
605 (Ty (NEONvduplane (Ty DPR_VFP2:$src2),
607 let isCommutable = 0;
609 class N3VDSL16<bits<2> op21_20, bits<4> op11_8,
610 string OpcodeStr, ValueType Ty, SDNode ShOp>
611 : N3V<0, 1, op21_20, op11_8, 1, 0,
612 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
614 !strconcat(OpcodeStr, "\t$dst, $src1, $src2[$lane]"), "",
616 (Ty (ShOp (Ty DPR:$src1),
617 (Ty (NEONvduplane (Ty DPR_8:$src2),
619 let isCommutable = 0;
622 class N3VQ<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
623 InstrItinClass itin, string OpcodeStr, ValueType ResTy, ValueType OpTy,
624 SDNode OpNode, bit Commutable>
625 : N3V<op24, op23, op21_20, op11_8, 1, op4,
626 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), itin,
627 !strconcat(OpcodeStr, "\t$dst, $src1, $src2"), "",
628 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src1), (OpTy QPR:$src2))))]> {
629 let isCommutable = Commutable;
631 class N3VQSL<bits<2> op21_20, bits<4> op11_8,
632 InstrItinClass itin, string OpcodeStr,
633 ValueType ResTy, ValueType OpTy, SDNode ShOp>
634 : N3V<1, 1, op21_20, op11_8, 1, 0,
635 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
636 itin, !strconcat(OpcodeStr, "\t$dst, $src1, $src2[$lane]"), "",
637 [(set (ResTy QPR:$dst),
638 (ResTy (ShOp (ResTy QPR:$src1),
639 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
641 let isCommutable = 0;
643 class N3VQSL16<bits<2> op21_20, bits<4> op11_8,
644 string OpcodeStr, ValueType ResTy, ValueType OpTy, SDNode ShOp>
645 : N3V<1, 1, op21_20, op11_8, 1, 0,
646 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
648 !strconcat(OpcodeStr, "\t$dst, $src1, $src2[$lane]"), "",
649 [(set (ResTy QPR:$dst),
650 (ResTy (ShOp (ResTy QPR:$src1),
651 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
653 let isCommutable = 0;
656 // Basic 3-register operations, scalar single-precision
657 class N3VDs<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
658 string OpcodeStr, ValueType ResTy, ValueType OpTy,
659 SDNode OpNode, bit Commutable>
660 : N3V<op24, op23, op21_20, op11_8, 0, op4,
661 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src1, DPR_VFP2:$src2), IIC_VBIND,
662 !strconcat(OpcodeStr, "\t$dst, $src1, $src2"), "", []> {
663 let isCommutable = Commutable;
665 class N3VDsPat<SDNode OpNode, NeonI Inst>
666 : NEONFPPat<(f32 (OpNode SPR:$a, SPR:$b)),
668 (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$a, arm_ssubreg_0),
669 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$b, arm_ssubreg_0)),
672 // Basic 3-register intrinsics, both double- and quad-register.
673 class N3VDInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
674 InstrItinClass itin, string OpcodeStr, ValueType ResTy, ValueType OpTy,
675 Intrinsic IntOp, bit Commutable>
676 : N3V<op24, op23, op21_20, op11_8, 0, op4,
677 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), itin,
678 !strconcat(OpcodeStr, "\t$dst, $src1, $src2"), "",
679 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src1), (OpTy DPR:$src2))))]> {
680 let isCommutable = Commutable;
682 class N3VDIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
683 string OpcodeStr, ValueType Ty, Intrinsic IntOp>
684 : N3V<0, 1, op21_20, op11_8, 1, 0,
685 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
686 itin, !strconcat(OpcodeStr, "\t$dst, $src1, $src2[$lane]"), "",
688 (Ty (IntOp (Ty DPR:$src1),
689 (Ty (NEONvduplane (Ty DPR_VFP2:$src2),
691 let isCommutable = 0;
693 class N3VDIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
694 string OpcodeStr, ValueType Ty, Intrinsic IntOp>
695 : N3V<0, 1, op21_20, op11_8, 1, 0,
696 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
697 itin, !strconcat(OpcodeStr, "\t$dst, $src1, $src2[$lane]"), "",
699 (Ty (IntOp (Ty DPR:$src1),
700 (Ty (NEONvduplane (Ty DPR_8:$src2),
702 let isCommutable = 0;
705 class N3VQInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
706 InstrItinClass itin, string OpcodeStr, ValueType ResTy, ValueType OpTy,
707 Intrinsic IntOp, bit Commutable>
708 : N3V<op24, op23, op21_20, op11_8, 1, op4,
709 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), itin,
710 !strconcat(OpcodeStr, "\t$dst, $src1, $src2"), "",
711 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src1), (OpTy QPR:$src2))))]> {
712 let isCommutable = Commutable;
714 class N3VQIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
715 string OpcodeStr, ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
716 : N3V<1, 1, op21_20, op11_8, 1, 0,
717 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
718 itin, !strconcat(OpcodeStr, "\t$dst, $src1, $src2[$lane]"), "",
719 [(set (ResTy QPR:$dst),
720 (ResTy (IntOp (ResTy QPR:$src1),
721 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
723 let isCommutable = 0;
725 class N3VQIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
726 string OpcodeStr, ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
727 : N3V<1, 1, op21_20, op11_8, 1, 0,
728 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
729 itin, !strconcat(OpcodeStr, "\t$dst, $src1, $src2[$lane]"), "",
730 [(set (ResTy QPR:$dst),
731 (ResTy (IntOp (ResTy QPR:$src1),
732 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
734 let isCommutable = 0;
737 // Multiply-Add/Sub operations, both double- and quad-register.
738 class N3VDMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
739 InstrItinClass itin, string OpcodeStr,
740 ValueType Ty, SDNode MulOp, SDNode OpNode>
741 : N3V<op24, op23, op21_20, op11_8, 0, op4,
742 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3), itin,
743 !strconcat(OpcodeStr, "\t$dst, $src2, $src3"), "$src1 = $dst",
744 [(set DPR:$dst, (Ty (OpNode DPR:$src1,
745 (Ty (MulOp DPR:$src2, DPR:$src3)))))]>;
746 class N3VDMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
747 string OpcodeStr, ValueType Ty, SDNode MulOp, SDNode ShOp>
748 : N3V<0, 1, op21_20, op11_8, 1, 0,
750 (ins DPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane), itin,
751 !strconcat(OpcodeStr, "\t$dst, $src2, $src3[$lane]"), "$src1 = $dst",
753 (Ty (ShOp (Ty DPR:$src1),
754 (Ty (MulOp DPR:$src2,
755 (Ty (NEONvduplane (Ty DPR_VFP2:$src3),
757 class N3VDMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
758 string OpcodeStr, ValueType Ty, SDNode MulOp, SDNode ShOp>
759 : N3V<0, 1, op21_20, op11_8, 1, 0,
761 (ins DPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane), itin,
762 !strconcat(OpcodeStr, "\t$dst, $src2, $src3[$lane]"), "$src1 = $dst",
764 (Ty (ShOp (Ty DPR:$src1),
765 (Ty (MulOp DPR:$src2,
766 (Ty (NEONvduplane (Ty DPR_8:$src3),
769 class N3VQMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
770 InstrItinClass itin, string OpcodeStr, ValueType Ty,
771 SDNode MulOp, SDNode OpNode>
772 : N3V<op24, op23, op21_20, op11_8, 1, op4,
773 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3), itin,
774 !strconcat(OpcodeStr, "\t$dst, $src2, $src3"), "$src1 = $dst",
775 [(set QPR:$dst, (Ty (OpNode QPR:$src1,
776 (Ty (MulOp QPR:$src2, QPR:$src3)))))]>;
777 class N3VQMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
778 string OpcodeStr, ValueType ResTy, ValueType OpTy,
779 SDNode MulOp, SDNode ShOp>
780 : N3V<1, 1, op21_20, op11_8, 1, 0,
782 (ins QPR:$src1, QPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane), itin,
783 !strconcat(OpcodeStr, "\t$dst, $src2, $src3[$lane]"), "$src1 = $dst",
784 [(set (ResTy QPR:$dst),
785 (ResTy (ShOp (ResTy QPR:$src1),
786 (ResTy (MulOp QPR:$src2,
787 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src3),
789 class N3VQMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
790 string OpcodeStr, ValueType ResTy, ValueType OpTy,
791 SDNode MulOp, SDNode ShOp>
792 : N3V<1, 1, op21_20, op11_8, 1, 0,
794 (ins QPR:$src1, QPR:$src2, DPR_8:$src3, nohash_imm:$lane), itin,
795 !strconcat(OpcodeStr, "\t$dst, $src2, $src3[$lane]"), "$src1 = $dst",
796 [(set (ResTy QPR:$dst),
797 (ResTy (ShOp (ResTy QPR:$src1),
798 (ResTy (MulOp QPR:$src2,
799 (ResTy (NEONvduplane (OpTy DPR_8:$src3),
802 // Multiply-Add/Sub operations, scalar single-precision
803 class N3VDMulOps<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
804 InstrItinClass itin, string OpcodeStr,
805 ValueType Ty, SDNode MulOp, SDNode OpNode>
806 : N3V<op24, op23, op21_20, op11_8, 0, op4,
807 (outs DPR_VFP2:$dst),
808 (ins DPR_VFP2:$src1, DPR_VFP2:$src2, DPR_VFP2:$src3), itin,
809 !strconcat(OpcodeStr, "\t$dst, $src2, $src3"), "$src1 = $dst", []>;
811 class N3VDMulOpsPat<SDNode MulNode, SDNode OpNode, NeonI Inst>
812 : NEONFPPat<(f32 (OpNode SPR:$acc, (f32 (MulNode SPR:$a, SPR:$b)))),
814 (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$acc, arm_ssubreg_0),
815 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$a, arm_ssubreg_0),
816 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$b, arm_ssubreg_0)),
819 // Neon 3-argument intrinsics, both double- and quad-register.
820 // The destination register is also used as the first source operand register.
821 class N3VDInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
822 InstrItinClass itin, string OpcodeStr,
823 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
824 : N3V<op24, op23, op21_20, op11_8, 0, op4,
825 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3), itin,
826 !strconcat(OpcodeStr, "\t$dst, $src2, $src3"), "$src1 = $dst",
827 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src1),
828 (OpTy DPR:$src2), (OpTy DPR:$src3))))]>;
829 class N3VQInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
830 InstrItinClass itin, string OpcodeStr,
831 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
832 : N3V<op24, op23, op21_20, op11_8, 1, op4,
833 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3), itin,
834 !strconcat(OpcodeStr, "\t$dst, $src2, $src3"), "$src1 = $dst",
835 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src1),
836 (OpTy QPR:$src2), (OpTy QPR:$src3))))]>;
838 // Neon Long 3-argument intrinsic. The destination register is
839 // a quad-register and is also used as the first source operand register.
840 class N3VLInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
841 InstrItinClass itin, string OpcodeStr,
842 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
843 : N3V<op24, op23, op21_20, op11_8, 0, op4,
844 (outs QPR:$dst), (ins QPR:$src1, DPR:$src2, DPR:$src3), itin,
845 !strconcat(OpcodeStr, "\t$dst, $src2, $src3"), "$src1 = $dst",
847 (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$src2), (TyD DPR:$src3))))]>;
848 class N3VLInt3SL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
849 string OpcodeStr, ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
850 : N3V<op24, 1, op21_20, op11_8, 1, 0,
852 (ins QPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane), itin,
853 !strconcat(OpcodeStr, "\t$dst, $src2, $src3[$lane]"), "$src1 = $dst",
854 [(set (ResTy QPR:$dst),
855 (ResTy (IntOp (ResTy QPR:$src1),
857 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src3),
859 class N3VLInt3SL16<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
860 string OpcodeStr, ValueType ResTy, ValueType OpTy,
862 : N3V<op24, 1, op21_20, op11_8, 1, 0,
864 (ins QPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane), itin,
865 !strconcat(OpcodeStr, "\t$dst, $src2, $src3[$lane]"), "$src1 = $dst",
866 [(set (ResTy QPR:$dst),
867 (ResTy (IntOp (ResTy QPR:$src1),
869 (OpTy (NEONvduplane (OpTy DPR_8:$src3),
873 // Narrowing 3-register intrinsics.
874 class N3VNInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
875 string OpcodeStr, ValueType TyD, ValueType TyQ,
876 Intrinsic IntOp, bit Commutable>
877 : N3V<op24, op23, op21_20, op11_8, 0, op4,
878 (outs DPR:$dst), (ins QPR:$src1, QPR:$src2), IIC_VBINi4D,
879 !strconcat(OpcodeStr, "\t$dst, $src1, $src2"), "",
880 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src1), (TyQ QPR:$src2))))]> {
881 let isCommutable = Commutable;
884 // Long 3-register intrinsics.
885 class N3VLInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
886 InstrItinClass itin, string OpcodeStr, ValueType TyQ, ValueType TyD,
887 Intrinsic IntOp, bit Commutable>
888 : N3V<op24, op23, op21_20, op11_8, 0, op4,
889 (outs QPR:$dst), (ins DPR:$src1, DPR:$src2), itin,
890 !strconcat(OpcodeStr, "\t$dst, $src1, $src2"), "",
891 [(set QPR:$dst, (TyQ (IntOp (TyD DPR:$src1), (TyD DPR:$src2))))]> {
892 let isCommutable = Commutable;
894 class N3VLIntSL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
895 string OpcodeStr, ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
896 : N3V<op24, 1, op21_20, op11_8, 1, 0,
897 (outs QPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
898 itin, !strconcat(OpcodeStr, "\t$dst, $src1, $src2[$lane]"), "",
899 [(set (ResTy QPR:$dst),
900 (ResTy (IntOp (OpTy DPR:$src1),
901 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src2),
903 class N3VLIntSL16<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
904 string OpcodeStr, ValueType ResTy, ValueType OpTy,
906 : N3V<op24, 1, op21_20, op11_8, 1, 0,
907 (outs QPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
908 itin, !strconcat(OpcodeStr, "\t$dst, $src1, $src2[$lane]"), "",
909 [(set (ResTy QPR:$dst),
910 (ResTy (IntOp (OpTy DPR:$src1),
911 (OpTy (NEONvduplane (OpTy DPR_8:$src2),
914 // Wide 3-register intrinsics.
915 class N3VWInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
916 string OpcodeStr, ValueType TyQ, ValueType TyD,
917 Intrinsic IntOp, bit Commutable>
918 : N3V<op24, op23, op21_20, op11_8, 0, op4,
919 (outs QPR:$dst), (ins QPR:$src1, DPR:$src2), IIC_VSUBiD,
920 !strconcat(OpcodeStr, "\t$dst, $src1, $src2"), "",
921 [(set QPR:$dst, (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$src2))))]> {
922 let isCommutable = Commutable;
925 // Pairwise long 2-register intrinsics, both double- and quad-register.
926 class N2VDPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
927 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
928 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
929 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
930 (ins DPR:$src), IIC_VSHLiD, !strconcat(OpcodeStr, "\t$dst, $src"), "",
931 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
932 class N2VQPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
933 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
934 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
935 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
936 (ins QPR:$src), IIC_VSHLiD, !strconcat(OpcodeStr, "\t$dst, $src"), "",
937 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
939 // Pairwise long 2-register accumulate intrinsics,
940 // both double- and quad-register.
941 // The destination register is also used as the first source operand register.
942 class N2VDPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
943 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
944 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
945 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
946 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), IIC_VPALiD,
947 !strconcat(OpcodeStr, "\t$dst, $src2"), "$src1 = $dst",
948 [(set DPR:$dst, (ResTy (IntOp (ResTy DPR:$src1), (OpTy DPR:$src2))))]>;
949 class N2VQPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
950 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
951 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
952 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4,
953 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), IIC_VPALiQ,
954 !strconcat(OpcodeStr, "\t$dst, $src2"), "$src1 = $dst",
955 [(set QPR:$dst, (ResTy (IntOp (ResTy QPR:$src1), (OpTy QPR:$src2))))]>;
957 // Shift by immediate,
958 // both double- and quad-register.
959 class N2VDSh<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
960 bit op4, InstrItinClass itin, string OpcodeStr,
961 ValueType Ty, SDNode OpNode>
962 : N2VImm<op24, op23, op21_16, op11_8, op7, 0, op4,
963 (outs DPR:$dst), (ins DPR:$src, i32imm:$SIMM), itin,
964 !strconcat(OpcodeStr, "\t$dst, $src, $SIMM"), "",
965 [(set DPR:$dst, (Ty (OpNode (Ty DPR:$src), (i32 imm:$SIMM))))]>;
966 class N2VQSh<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
967 bit op4, InstrItinClass itin, string OpcodeStr,
968 ValueType Ty, SDNode OpNode>
969 : N2VImm<op24, op23, op21_16, op11_8, op7, 1, op4,
970 (outs QPR:$dst), (ins QPR:$src, i32imm:$SIMM), itin,
971 !strconcat(OpcodeStr, "\t$dst, $src, $SIMM"), "",
972 [(set QPR:$dst, (Ty (OpNode (Ty QPR:$src), (i32 imm:$SIMM))))]>;
974 // Long shift by immediate.
975 class N2VLSh<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
976 bit op6, bit op4, string OpcodeStr, ValueType ResTy,
977 ValueType OpTy, SDNode OpNode>
978 : N2VImm<op24, op23, op21_16, op11_8, op7, op6, op4,
979 (outs QPR:$dst), (ins DPR:$src, i32imm:$SIMM), IIC_VSHLiD,
980 !strconcat(OpcodeStr, "\t$dst, $src, $SIMM"), "",
981 [(set QPR:$dst, (ResTy (OpNode (OpTy DPR:$src),
982 (i32 imm:$SIMM))))]>;
984 // Narrow shift by immediate.
985 class N2VNSh<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
986 bit op6, bit op4, InstrItinClass itin, string OpcodeStr,
987 ValueType ResTy, ValueType OpTy, SDNode OpNode>
988 : N2VImm<op24, op23, op21_16, op11_8, op7, op6, op4,
989 (outs DPR:$dst), (ins QPR:$src, i32imm:$SIMM), itin,
990 !strconcat(OpcodeStr, "\t$dst, $src, $SIMM"), "",
991 [(set DPR:$dst, (ResTy (OpNode (OpTy QPR:$src),
992 (i32 imm:$SIMM))))]>;
994 // Shift right by immediate and accumulate,
995 // both double- and quad-register.
996 class N2VDShAdd<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
997 bit op4, string OpcodeStr, ValueType Ty, SDNode ShOp>
998 : N2VImm<op24, op23, op21_16, op11_8, op7, 0, op4,
999 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, i32imm:$SIMM),
1001 !strconcat(OpcodeStr, "\t$dst, $src2, $SIMM"), "$src1 = $dst",
1002 [(set DPR:$dst, (Ty (add DPR:$src1,
1003 (Ty (ShOp DPR:$src2, (i32 imm:$SIMM))))))]>;
1004 class N2VQShAdd<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
1005 bit op4, string OpcodeStr, ValueType Ty, SDNode ShOp>
1006 : N2VImm<op24, op23, op21_16, op11_8, op7, 1, op4,
1007 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, i32imm:$SIMM),
1009 !strconcat(OpcodeStr, "\t$dst, $src2, $SIMM"), "$src1 = $dst",
1010 [(set QPR:$dst, (Ty (add QPR:$src1,
1011 (Ty (ShOp QPR:$src2, (i32 imm:$SIMM))))))]>;
1013 // Shift by immediate and insert,
1014 // both double- and quad-register.
1015 class N2VDShIns<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
1016 bit op4, string OpcodeStr, ValueType Ty, SDNode ShOp>
1017 : N2VImm<op24, op23, op21_16, op11_8, op7, 0, op4,
1018 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, i32imm:$SIMM),
1020 !strconcat(OpcodeStr, "\t$dst, $src2, $SIMM"), "$src1 = $dst",
1021 [(set DPR:$dst, (Ty (ShOp DPR:$src1, DPR:$src2, (i32 imm:$SIMM))))]>;
1022 class N2VQShIns<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
1023 bit op4, string OpcodeStr, ValueType Ty, SDNode ShOp>
1024 : N2VImm<op24, op23, op21_16, op11_8, op7, 1, op4,
1025 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, i32imm:$SIMM),
1027 !strconcat(OpcodeStr, "\t$dst, $src2, $SIMM"), "$src1 = $dst",
1028 [(set QPR:$dst, (Ty (ShOp QPR:$src1, QPR:$src2, (i32 imm:$SIMM))))]>;
1030 // Convert, with fractional bits immediate,
1031 // both double- and quad-register.
1032 class N2VCvtD<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
1033 bit op4, string OpcodeStr, ValueType ResTy, ValueType OpTy,
1035 : N2VImm<op24, op23, op21_16, op11_8, op7, 0, op4,
1036 (outs DPR:$dst), (ins DPR:$src, i32imm:$SIMM), IIC_VUNAD,
1037 !strconcat(OpcodeStr, "\t$dst, $src, $SIMM"), "",
1038 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src), (i32 imm:$SIMM))))]>;
1039 class N2VCvtQ<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
1040 bit op4, string OpcodeStr, ValueType ResTy, ValueType OpTy,
1042 : N2VImm<op24, op23, op21_16, op11_8, op7, 1, op4,
1043 (outs QPR:$dst), (ins QPR:$src, i32imm:$SIMM), IIC_VUNAQ,
1044 !strconcat(OpcodeStr, "\t$dst, $src, $SIMM"), "",
1045 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src), (i32 imm:$SIMM))))]>;
1047 //===----------------------------------------------------------------------===//
1049 //===----------------------------------------------------------------------===//
1051 // Abbreviations used in multiclass suffixes:
1052 // Q = quarter int (8 bit) elements
1053 // H = half int (16 bit) elements
1054 // S = single int (32 bit) elements
1055 // D = double int (64 bit) elements
1057 // Neon 3-register vector operations.
1059 // First with only element sizes of 8, 16 and 32 bits:
1060 multiclass N3V_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1061 InstrItinClass itinD16, InstrItinClass itinD32,
1062 InstrItinClass itinQ16, InstrItinClass itinQ32,
1063 string OpcodeStr, SDNode OpNode, bit Commutable = 0> {
1064 // 64-bit vector types.
1065 def v8i8 : N3VD<op24, op23, 0b00, op11_8, op4, itinD16,
1066 !strconcat(OpcodeStr, "8"), v8i8, v8i8, OpNode, Commutable>;
1067 def v4i16 : N3VD<op24, op23, 0b01, op11_8, op4, itinD16,
1068 !strconcat(OpcodeStr, "16"), v4i16, v4i16, OpNode, Commutable>;
1069 def v2i32 : N3VD<op24, op23, 0b10, op11_8, op4, itinD32,
1070 !strconcat(OpcodeStr, "32"), v2i32, v2i32, OpNode, Commutable>;
1072 // 128-bit vector types.
1073 def v16i8 : N3VQ<op24, op23, 0b00, op11_8, op4, itinQ16,
1074 !strconcat(OpcodeStr, "8"), v16i8, v16i8, OpNode, Commutable>;
1075 def v8i16 : N3VQ<op24, op23, 0b01, op11_8, op4, itinQ16,
1076 !strconcat(OpcodeStr, "16"), v8i16, v8i16, OpNode, Commutable>;
1077 def v4i32 : N3VQ<op24, op23, 0b10, op11_8, op4, itinQ32,
1078 !strconcat(OpcodeStr, "32"), v4i32, v4i32, OpNode, Commutable>;
1081 multiclass N3VSL_HS<bits<4> op11_8, string OpcodeStr, SDNode ShOp> {
1082 def v4i16 : N3VDSL16<0b01, op11_8, !strconcat(OpcodeStr, "16"), v4i16, ShOp>;
1083 def v2i32 : N3VDSL<0b10, op11_8, IIC_VMULi32D, !strconcat(OpcodeStr, "32"), v2i32, ShOp>;
1084 def v8i16 : N3VQSL16<0b01, op11_8, !strconcat(OpcodeStr, "16"), v8i16, v4i16, ShOp>;
1085 def v4i32 : N3VQSL<0b10, op11_8, IIC_VMULi32Q, !strconcat(OpcodeStr, "32"), v4i32, v2i32, ShOp>;
1088 // ....then also with element size 64 bits:
1089 multiclass N3V_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1090 InstrItinClass itinD, InstrItinClass itinQ,
1091 string OpcodeStr, SDNode OpNode, bit Commutable = 0>
1092 : N3V_QHS<op24, op23, op11_8, op4, itinD, itinD, itinQ, itinQ,
1093 OpcodeStr, OpNode, Commutable> {
1094 def v1i64 : N3VD<op24, op23, 0b11, op11_8, op4, itinD,
1095 !strconcat(OpcodeStr, "64"), v1i64, v1i64, OpNode, Commutable>;
1096 def v2i64 : N3VQ<op24, op23, 0b11, op11_8, op4, itinQ,
1097 !strconcat(OpcodeStr, "64"), v2i64, v2i64, OpNode, Commutable>;
1101 // Neon Narrowing 2-register vector intrinsics,
1102 // source operand element sizes of 16, 32 and 64 bits:
1103 multiclass N2VNInt_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1104 bits<5> op11_7, bit op6, bit op4,
1105 InstrItinClass itin, string OpcodeStr,
1107 def v8i8 : N2VNInt<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
1108 itin, !strconcat(OpcodeStr, "16"), v8i8, v8i16, IntOp>;
1109 def v4i16 : N2VNInt<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
1110 itin, !strconcat(OpcodeStr, "32"), v4i16, v4i32, IntOp>;
1111 def v2i32 : N2VNInt<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
1112 itin, !strconcat(OpcodeStr, "64"), v2i32, v2i64, IntOp>;
1116 // Neon Lengthening 2-register vector intrinsic (currently specific to VMOVL).
1117 // source operand element sizes of 16, 32 and 64 bits:
1118 multiclass N2VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
1119 bit op4, string OpcodeStr, Intrinsic IntOp> {
1120 def v8i16 : N2VLInt<op24, op23, 0b001000, op11_8, op7, op6, op4,
1121 IIC_VQUNAiD, !strconcat(OpcodeStr, "8"), v8i16, v8i8, IntOp>;
1122 def v4i32 : N2VLInt<op24, op23, 0b010000, op11_8, op7, op6, op4,
1123 IIC_VQUNAiD, !strconcat(OpcodeStr, "16"), v4i32, v4i16, IntOp>;
1124 def v2i64 : N2VLInt<op24, op23, 0b100000, op11_8, op7, op6, op4,
1125 IIC_VQUNAiD, !strconcat(OpcodeStr, "32"), v2i64, v2i32, IntOp>;
1129 // Neon 3-register vector intrinsics.
1131 // First with only element sizes of 16 and 32 bits:
1132 multiclass N3VInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
1133 InstrItinClass itinD16, InstrItinClass itinD32,
1134 InstrItinClass itinQ16, InstrItinClass itinQ32,
1135 string OpcodeStr, Intrinsic IntOp, bit Commutable = 0> {
1136 // 64-bit vector types.
1137 def v4i16 : N3VDInt<op24, op23, 0b01, op11_8, op4, itinD16, !strconcat(OpcodeStr,"16"),
1138 v4i16, v4i16, IntOp, Commutable>;
1139 def v2i32 : N3VDInt<op24, op23, 0b10, op11_8, op4, itinD32, !strconcat(OpcodeStr,"32"),
1140 v2i32, v2i32, IntOp, Commutable>;
1142 // 128-bit vector types.
1143 def v8i16 : N3VQInt<op24, op23, 0b01, op11_8, op4, itinQ16, !strconcat(OpcodeStr,"16"),
1144 v8i16, v8i16, IntOp, Commutable>;
1145 def v4i32 : N3VQInt<op24, op23, 0b10, op11_8, op4, itinQ32, !strconcat(OpcodeStr,"32"),
1146 v4i32, v4i32, IntOp, Commutable>;
1149 multiclass N3VIntSL_HS<bits<4> op11_8,
1150 InstrItinClass itinD16, InstrItinClass itinD32,
1151 InstrItinClass itinQ16, InstrItinClass itinQ32,
1152 string OpcodeStr, Intrinsic IntOp> {
1153 def v4i16 : N3VDIntSL16<0b01, op11_8, itinD16, !strconcat(OpcodeStr, "16"), v4i16, IntOp>;
1154 def v2i32 : N3VDIntSL<0b10, op11_8, itinD32, !strconcat(OpcodeStr, "32"), v2i32, IntOp>;
1155 def v8i16 : N3VQIntSL16<0b01, op11_8, itinQ16, !strconcat(OpcodeStr, "16"), v8i16, v4i16, IntOp>;
1156 def v4i32 : N3VQIntSL<0b10, op11_8, itinQ32, !strconcat(OpcodeStr, "32"), v4i32, v2i32, IntOp>;
1159 // ....then also with element size of 8 bits:
1160 multiclass N3VInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1161 InstrItinClass itinD16, InstrItinClass itinD32,
1162 InstrItinClass itinQ16, InstrItinClass itinQ32,
1163 string OpcodeStr, Intrinsic IntOp, bit Commutable = 0>
1164 : N3VInt_HS<op24, op23, op11_8, op4, itinD16, itinD32, itinQ16, itinQ32,
1165 OpcodeStr, IntOp, Commutable> {
1166 def v8i8 : N3VDInt<op24, op23, 0b00, op11_8, op4, itinD16,
1167 !strconcat(OpcodeStr, "8"), v8i8, v8i8, IntOp, Commutable>;
1168 def v16i8 : N3VQInt<op24, op23, 0b00, op11_8, op4, itinQ16,
1169 !strconcat(OpcodeStr, "8"), v16i8, v16i8, IntOp, Commutable>;
1172 // ....then also with element size of 64 bits:
1173 multiclass N3VInt_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1174 InstrItinClass itinD16, InstrItinClass itinD32,
1175 InstrItinClass itinQ16, InstrItinClass itinQ32,
1176 string OpcodeStr, Intrinsic IntOp, bit Commutable = 0>
1177 : N3VInt_QHS<op24, op23, op11_8, op4, itinD16, itinD32, itinQ16, itinQ32,
1178 OpcodeStr, IntOp, Commutable> {
1179 def v1i64 : N3VDInt<op24, op23, 0b11, op11_8, op4, itinD32,
1180 !strconcat(OpcodeStr,"64"), v1i64, v1i64, IntOp, Commutable>;
1181 def v2i64 : N3VQInt<op24, op23, 0b11, op11_8, op4, itinQ32,
1182 !strconcat(OpcodeStr,"64"), v2i64, v2i64, IntOp, Commutable>;
1186 // Neon Narrowing 3-register vector intrinsics,
1187 // source operand element sizes of 16, 32 and 64 bits:
1188 multiclass N3VNInt_HSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1189 string OpcodeStr, Intrinsic IntOp, bit Commutable = 0> {
1190 def v8i8 : N3VNInt<op24, op23, 0b00, op11_8, op4, !strconcat(OpcodeStr,"16"),
1191 v8i8, v8i16, IntOp, Commutable>;
1192 def v4i16 : N3VNInt<op24, op23, 0b01, op11_8, op4, !strconcat(OpcodeStr,"32"),
1193 v4i16, v4i32, IntOp, Commutable>;
1194 def v2i32 : N3VNInt<op24, op23, 0b10, op11_8, op4, !strconcat(OpcodeStr,"64"),
1195 v2i32, v2i64, IntOp, Commutable>;
1199 // Neon Long 3-register vector intrinsics.
1201 // First with only element sizes of 16 and 32 bits:
1202 multiclass N3VLInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
1203 InstrItinClass itin, string OpcodeStr,
1204 Intrinsic IntOp, bit Commutable = 0> {
1205 def v4i32 : N3VLInt<op24, op23, 0b01, op11_8, op4, itin,
1206 !strconcat(OpcodeStr,"16"), v4i32, v4i16, IntOp, Commutable>;
1207 def v2i64 : N3VLInt<op24, op23, 0b10, op11_8, op4, itin,
1208 !strconcat(OpcodeStr,"32"), v2i64, v2i32, IntOp, Commutable>;
1211 multiclass N3VLIntSL_HS<bit op24, bits<4> op11_8,
1212 InstrItinClass itin, string OpcodeStr, Intrinsic IntOp> {
1213 def v4i16 : N3VLIntSL16<op24, 0b01, op11_8, itin,
1214 !strconcat(OpcodeStr, "16"), v4i32, v4i16, IntOp>;
1215 def v2i32 : N3VLIntSL<op24, 0b10, op11_8, itin,
1216 !strconcat(OpcodeStr, "32"), v2i64, v2i32, IntOp>;
1219 // ....then also with element size of 8 bits:
1220 multiclass N3VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1221 InstrItinClass itin, string OpcodeStr,
1222 Intrinsic IntOp, bit Commutable = 0>
1223 : N3VLInt_HS<op24, op23, op11_8, op4, itin, OpcodeStr, IntOp, Commutable> {
1224 def v8i16 : N3VLInt<op24, op23, 0b00, op11_8, op4, itin,
1225 !strconcat(OpcodeStr, "8"), v8i16, v8i8, IntOp, Commutable>;
1229 // Neon Wide 3-register vector intrinsics,
1230 // source operand element sizes of 8, 16 and 32 bits:
1231 multiclass N3VWInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1232 string OpcodeStr, Intrinsic IntOp, bit Commutable = 0> {
1233 def v8i16 : N3VWInt<op24, op23, 0b00, op11_8, op4, !strconcat(OpcodeStr, "8"),
1234 v8i16, v8i8, IntOp, Commutable>;
1235 def v4i32 : N3VWInt<op24, op23, 0b01, op11_8, op4, !strconcat(OpcodeStr,"16"),
1236 v4i32, v4i16, IntOp, Commutable>;
1237 def v2i64 : N3VWInt<op24, op23, 0b10, op11_8, op4, !strconcat(OpcodeStr,"32"),
1238 v2i64, v2i32, IntOp, Commutable>;
1242 // Neon Multiply-Op vector operations,
1243 // element sizes of 8, 16 and 32 bits:
1244 multiclass N3VMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1245 InstrItinClass itinD16, InstrItinClass itinD32,
1246 InstrItinClass itinQ16, InstrItinClass itinQ32,
1247 string OpcodeStr, SDNode OpNode> {
1248 // 64-bit vector types.
1249 def v8i8 : N3VDMulOp<op24, op23, 0b00, op11_8, op4, itinD16,
1250 !strconcat(OpcodeStr, "8"), v8i8, mul, OpNode>;
1251 def v4i16 : N3VDMulOp<op24, op23, 0b01, op11_8, op4, itinD16,
1252 !strconcat(OpcodeStr, "16"), v4i16, mul, OpNode>;
1253 def v2i32 : N3VDMulOp<op24, op23, 0b10, op11_8, op4, itinD32,
1254 !strconcat(OpcodeStr, "32"), v2i32, mul, OpNode>;
1256 // 128-bit vector types.
1257 def v16i8 : N3VQMulOp<op24, op23, 0b00, op11_8, op4, itinQ16,
1258 !strconcat(OpcodeStr, "8"), v16i8, mul, OpNode>;
1259 def v8i16 : N3VQMulOp<op24, op23, 0b01, op11_8, op4, itinQ16,
1260 !strconcat(OpcodeStr, "16"), v8i16, mul, OpNode>;
1261 def v4i32 : N3VQMulOp<op24, op23, 0b10, op11_8, op4, itinQ32,
1262 !strconcat(OpcodeStr, "32"), v4i32, mul, OpNode>;
1265 multiclass N3VMulOpSL_HS<bits<4> op11_8,
1266 InstrItinClass itinD16, InstrItinClass itinD32,
1267 InstrItinClass itinQ16, InstrItinClass itinQ32,
1268 string OpcodeStr, SDNode ShOp> {
1269 def v4i16 : N3VDMulOpSL16<0b01, op11_8, itinD16,
1270 !strconcat(OpcodeStr, "16"), v4i16, mul, ShOp>;
1271 def v2i32 : N3VDMulOpSL<0b10, op11_8, itinD32,
1272 !strconcat(OpcodeStr, "32"), v2i32, mul, ShOp>;
1273 def v8i16 : N3VQMulOpSL16<0b01, op11_8, itinQ16,
1274 !strconcat(OpcodeStr, "16"), v8i16, v4i16, mul, ShOp>;
1275 def v4i32 : N3VQMulOpSL<0b10, op11_8, itinQ32,
1276 !strconcat(OpcodeStr, "32"), v4i32, v2i32, mul, ShOp>;
1279 // Neon 3-argument intrinsics,
1280 // element sizes of 8, 16 and 32 bits:
1281 multiclass N3VInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1282 string OpcodeStr, Intrinsic IntOp> {
1283 // 64-bit vector types.
1284 def v8i8 : N3VDInt3<op24, op23, 0b00, op11_8, op4, IIC_VMACi16D,
1285 !strconcat(OpcodeStr, "8"), v8i8, v8i8, IntOp>;
1286 def v4i16 : N3VDInt3<op24, op23, 0b01, op11_8, op4, IIC_VMACi16D,
1287 !strconcat(OpcodeStr, "16"), v4i16, v4i16, IntOp>;
1288 def v2i32 : N3VDInt3<op24, op23, 0b10, op11_8, op4, IIC_VMACi32D,
1289 !strconcat(OpcodeStr, "32"), v2i32, v2i32, IntOp>;
1291 // 128-bit vector types.
1292 def v16i8 : N3VQInt3<op24, op23, 0b00, op11_8, op4, IIC_VMACi16Q,
1293 !strconcat(OpcodeStr, "8"), v16i8, v16i8, IntOp>;
1294 def v8i16 : N3VQInt3<op24, op23, 0b01, op11_8, op4, IIC_VMACi16Q,
1295 !strconcat(OpcodeStr, "16"), v8i16, v8i16, IntOp>;
1296 def v4i32 : N3VQInt3<op24, op23, 0b10, op11_8, op4, IIC_VMACi32Q,
1297 !strconcat(OpcodeStr, "32"), v4i32, v4i32, IntOp>;
1301 // Neon Long 3-argument intrinsics.
1303 // First with only element sizes of 16 and 32 bits:
1304 multiclass N3VLInt3_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
1305 string OpcodeStr, Intrinsic IntOp> {
1306 def v4i32 : N3VLInt3<op24, op23, 0b01, op11_8, op4, IIC_VMACi16D,
1307 !strconcat(OpcodeStr, "16"), v4i32, v4i16, IntOp>;
1308 def v2i64 : N3VLInt3<op24, op23, 0b10, op11_8, op4, IIC_VMACi16D,
1309 !strconcat(OpcodeStr, "32"), v2i64, v2i32, IntOp>;
1312 multiclass N3VLInt3SL_HS<bit op24, bits<4> op11_8,
1313 string OpcodeStr, Intrinsic IntOp> {
1314 def v4i16 : N3VLInt3SL16<op24, 0b01, op11_8, IIC_VMACi16D,
1315 !strconcat(OpcodeStr, "16"), v4i32, v4i16, IntOp>;
1316 def v2i32 : N3VLInt3SL<op24, 0b10, op11_8, IIC_VMACi32D,
1317 !strconcat(OpcodeStr, "32"), v2i64, v2i32, IntOp>;
1320 // ....then also with element size of 8 bits:
1321 multiclass N3VLInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1322 string OpcodeStr, Intrinsic IntOp>
1323 : N3VLInt3_HS<op24, op23, op11_8, op4, OpcodeStr, IntOp> {
1324 def v8i16 : N3VLInt3<op24, op23, 0b01, op11_8, op4, IIC_VMACi16D,
1325 !strconcat(OpcodeStr, "8"), v8i16, v8i8, IntOp>;
1329 // Neon 2-register vector intrinsics,
1330 // element sizes of 8, 16 and 32 bits:
1331 multiclass N2VInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1332 bits<5> op11_7, bit op4,
1333 InstrItinClass itinD, InstrItinClass itinQ,
1334 string OpcodeStr, Intrinsic IntOp> {
1335 // 64-bit vector types.
1336 def v8i8 : N2VDInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1337 itinD, !strconcat(OpcodeStr, "8"), v8i8, v8i8, IntOp>;
1338 def v4i16 : N2VDInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1339 itinD, !strconcat(OpcodeStr, "16"), v4i16, v4i16, IntOp>;
1340 def v2i32 : N2VDInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1341 itinD, !strconcat(OpcodeStr, "32"), v2i32, v2i32, IntOp>;
1343 // 128-bit vector types.
1344 def v16i8 : N2VQInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1345 itinQ, !strconcat(OpcodeStr, "8"), v16i8, v16i8, IntOp>;
1346 def v8i16 : N2VQInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1347 itinQ, !strconcat(OpcodeStr, "16"), v8i16, v8i16, IntOp>;
1348 def v4i32 : N2VQInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1349 itinQ, !strconcat(OpcodeStr, "32"), v4i32, v4i32, IntOp>;
1353 // Neon Pairwise long 2-register intrinsics,
1354 // element sizes of 8, 16 and 32 bits:
1355 multiclass N2VPLInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1356 bits<5> op11_7, bit op4,
1357 string OpcodeStr, Intrinsic IntOp> {
1358 // 64-bit vector types.
1359 def v8i8 : N2VDPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1360 !strconcat(OpcodeStr, "8"), v4i16, v8i8, IntOp>;
1361 def v4i16 : N2VDPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1362 !strconcat(OpcodeStr, "16"), v2i32, v4i16, IntOp>;
1363 def v2i32 : N2VDPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1364 !strconcat(OpcodeStr, "32"), v1i64, v2i32, IntOp>;
1366 // 128-bit vector types.
1367 def v16i8 : N2VQPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1368 !strconcat(OpcodeStr, "8"), v8i16, v16i8, IntOp>;
1369 def v8i16 : N2VQPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1370 !strconcat(OpcodeStr, "16"), v4i32, v8i16, IntOp>;
1371 def v4i32 : N2VQPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1372 !strconcat(OpcodeStr, "32"), v2i64, v4i32, IntOp>;
1376 // Neon Pairwise long 2-register accumulate intrinsics,
1377 // element sizes of 8, 16 and 32 bits:
1378 multiclass N2VPLInt2_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1379 bits<5> op11_7, bit op4,
1380 string OpcodeStr, Intrinsic IntOp> {
1381 // 64-bit vector types.
1382 def v8i8 : N2VDPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1383 !strconcat(OpcodeStr, "8"), v4i16, v8i8, IntOp>;
1384 def v4i16 : N2VDPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1385 !strconcat(OpcodeStr, "16"), v2i32, v4i16, IntOp>;
1386 def v2i32 : N2VDPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1387 !strconcat(OpcodeStr, "32"), v1i64, v2i32, IntOp>;
1389 // 128-bit vector types.
1390 def v16i8 : N2VQPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1391 !strconcat(OpcodeStr, "8"), v8i16, v16i8, IntOp>;
1392 def v8i16 : N2VQPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1393 !strconcat(OpcodeStr, "16"), v4i32, v8i16, IntOp>;
1394 def v4i32 : N2VQPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1395 !strconcat(OpcodeStr, "32"), v2i64, v4i32, IntOp>;
1399 // Neon 2-register vector shift by immediate,
1400 // element sizes of 8, 16, 32 and 64 bits:
1401 multiclass N2VSh_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1402 InstrItinClass itin, string OpcodeStr, SDNode OpNode> {
1403 // 64-bit vector types.
1404 def v8i8 : N2VDSh<op24, op23, 0b001000, op11_8, 0, op4, itin,
1405 !strconcat(OpcodeStr, "8"), v8i8, OpNode>;
1406 def v4i16 : N2VDSh<op24, op23, 0b010000, op11_8, 0, op4, itin,
1407 !strconcat(OpcodeStr, "16"), v4i16, OpNode>;
1408 def v2i32 : N2VDSh<op24, op23, 0b100000, op11_8, 0, op4, itin,
1409 !strconcat(OpcodeStr, "32"), v2i32, OpNode>;
1410 def v1i64 : N2VDSh<op24, op23, 0b000000, op11_8, 1, op4, itin,
1411 !strconcat(OpcodeStr, "64"), v1i64, OpNode>;
1413 // 128-bit vector types.
1414 def v16i8 : N2VQSh<op24, op23, 0b001000, op11_8, 0, op4, itin,
1415 !strconcat(OpcodeStr, "8"), v16i8, OpNode>;
1416 def v8i16 : N2VQSh<op24, op23, 0b010000, op11_8, 0, op4, itin,
1417 !strconcat(OpcodeStr, "16"), v8i16, OpNode>;
1418 def v4i32 : N2VQSh<op24, op23, 0b100000, op11_8, 0, op4, itin,
1419 !strconcat(OpcodeStr, "32"), v4i32, OpNode>;
1420 def v2i64 : N2VQSh<op24, op23, 0b000000, op11_8, 1, op4, itin,
1421 !strconcat(OpcodeStr, "64"), v2i64, OpNode>;
1425 // Neon Shift-Accumulate vector operations,
1426 // element sizes of 8, 16, 32 and 64 bits:
1427 multiclass N2VShAdd_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1428 string OpcodeStr, SDNode ShOp> {
1429 // 64-bit vector types.
1430 def v8i8 : N2VDShAdd<op24, op23, 0b001000, op11_8, 0, op4,
1431 !strconcat(OpcodeStr, "8"), v8i8, ShOp>;
1432 def v4i16 : N2VDShAdd<op24, op23, 0b010000, op11_8, 0, op4,
1433 !strconcat(OpcodeStr, "16"), v4i16, ShOp>;
1434 def v2i32 : N2VDShAdd<op24, op23, 0b100000, op11_8, 0, op4,
1435 !strconcat(OpcodeStr, "32"), v2i32, ShOp>;
1436 def v1i64 : N2VDShAdd<op24, op23, 0b000000, op11_8, 1, op4,
1437 !strconcat(OpcodeStr, "64"), v1i64, ShOp>;
1439 // 128-bit vector types.
1440 def v16i8 : N2VQShAdd<op24, op23, 0b001000, op11_8, 0, op4,
1441 !strconcat(OpcodeStr, "8"), v16i8, ShOp>;
1442 def v8i16 : N2VQShAdd<op24, op23, 0b010000, op11_8, 0, op4,
1443 !strconcat(OpcodeStr, "16"), v8i16, ShOp>;
1444 def v4i32 : N2VQShAdd<op24, op23, 0b100000, op11_8, 0, op4,
1445 !strconcat(OpcodeStr, "32"), v4i32, ShOp>;
1446 def v2i64 : N2VQShAdd<op24, op23, 0b000000, op11_8, 1, op4,
1447 !strconcat(OpcodeStr, "64"), v2i64, ShOp>;
1451 // Neon Shift-Insert vector operations,
1452 // element sizes of 8, 16, 32 and 64 bits:
1453 multiclass N2VShIns_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1454 string OpcodeStr, SDNode ShOp> {
1455 // 64-bit vector types.
1456 def v8i8 : N2VDShIns<op24, op23, 0b001000, op11_8, 0, op4,
1457 !strconcat(OpcodeStr, "8"), v8i8, ShOp>;
1458 def v4i16 : N2VDShIns<op24, op23, 0b010000, op11_8, 0, op4,
1459 !strconcat(OpcodeStr, "16"), v4i16, ShOp>;
1460 def v2i32 : N2VDShIns<op24, op23, 0b100000, op11_8, 0, op4,
1461 !strconcat(OpcodeStr, "32"), v2i32, ShOp>;
1462 def v1i64 : N2VDShIns<op24, op23, 0b000000, op11_8, 1, op4,
1463 !strconcat(OpcodeStr, "64"), v1i64, ShOp>;
1465 // 128-bit vector types.
1466 def v16i8 : N2VQShIns<op24, op23, 0b001000, op11_8, 0, op4,
1467 !strconcat(OpcodeStr, "8"), v16i8, ShOp>;
1468 def v8i16 : N2VQShIns<op24, op23, 0b010000, op11_8, 0, op4,
1469 !strconcat(OpcodeStr, "16"), v8i16, ShOp>;
1470 def v4i32 : N2VQShIns<op24, op23, 0b100000, op11_8, 0, op4,
1471 !strconcat(OpcodeStr, "32"), v4i32, ShOp>;
1472 def v2i64 : N2VQShIns<op24, op23, 0b000000, op11_8, 1, op4,
1473 !strconcat(OpcodeStr, "64"), v2i64, ShOp>;
1476 //===----------------------------------------------------------------------===//
1477 // Instruction Definitions.
1478 //===----------------------------------------------------------------------===//
1480 // Vector Add Operations.
1482 // VADD : Vector Add (integer and floating-point)
1483 defm VADD : N3V_QHSD<0, 0, 0b1000, 0, IIC_VBINiD, IIC_VBINiQ, "vadd.i", add, 1>;
1484 def VADDfd : N3VD<0, 0, 0b00, 0b1101, 0, IIC_VBIND, "vadd.f32", v2f32, v2f32, fadd, 1>;
1485 def VADDfq : N3VQ<0, 0, 0b00, 0b1101, 0, IIC_VBINQ, "vadd.f32", v4f32, v4f32, fadd, 1>;
1486 // VADDL : Vector Add Long (Q = D + D)
1487 defm VADDLs : N3VLInt_QHS<0,1,0b0000,0, IIC_VSHLiD, "vaddl.s", int_arm_neon_vaddls, 1>;
1488 defm VADDLu : N3VLInt_QHS<1,1,0b0000,0, IIC_VSHLiD, "vaddl.u", int_arm_neon_vaddlu, 1>;
1489 // VADDW : Vector Add Wide (Q = Q + D)
1490 defm VADDWs : N3VWInt_QHS<0,1,0b0001,0, "vaddw.s", int_arm_neon_vaddws, 0>;
1491 defm VADDWu : N3VWInt_QHS<1,1,0b0001,0, "vaddw.u", int_arm_neon_vaddwu, 0>;
1492 // VHADD : Vector Halving Add
1493 defm VHADDs : N3VInt_QHS<0,0,0b0000,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1494 IIC_VBINi4Q, "vhadd.s", int_arm_neon_vhadds, 1>;
1495 defm VHADDu : N3VInt_QHS<1,0,0b0000,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1496 IIC_VBINi4Q, "vhadd.u", int_arm_neon_vhaddu, 1>;
1497 // VRHADD : Vector Rounding Halving Add
1498 defm VRHADDs : N3VInt_QHS<0,0,0b0001,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1499 IIC_VBINi4Q, "vrhadd.s", int_arm_neon_vrhadds, 1>;
1500 defm VRHADDu : N3VInt_QHS<1,0,0b0001,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1501 IIC_VBINi4Q, "vrhadd.u", int_arm_neon_vrhaddu, 1>;
1502 // VQADD : Vector Saturating Add
1503 defm VQADDs : N3VInt_QHSD<0,0,0b0000,1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1504 IIC_VBINi4Q, "vqadd.s", int_arm_neon_vqadds, 1>;
1505 defm VQADDu : N3VInt_QHSD<1,0,0b0000,1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1506 IIC_VBINi4Q, "vqadd.u", int_arm_neon_vqaddu, 1>;
1507 // VADDHN : Vector Add and Narrow Returning High Half (D = Q + Q)
1508 defm VADDHN : N3VNInt_HSD<0,1,0b0100,0, "vaddhn.i", int_arm_neon_vaddhn, 1>;
1509 // VRADDHN : Vector Rounding Add and Narrow Returning High Half (D = Q + Q)
1510 defm VRADDHN : N3VNInt_HSD<1,1,0b0100,0, "vraddhn.i", int_arm_neon_vraddhn, 1>;
1512 // Vector Multiply Operations.
1514 // VMUL : Vector Multiply (integer, polynomial and floating-point)
1515 defm VMUL : N3V_QHS<0, 0, 0b1001, 1, IIC_VMULi16D, IIC_VMULi32D, IIC_VMULi16Q,
1516 IIC_VMULi32Q, "vmul.i", mul, 1>;
1517 def VMULpd : N3VDInt<1, 0, 0b00, 0b1001, 1, IIC_VMULi16D, "vmul.p8", v8i8, v8i8,
1518 int_arm_neon_vmulp, 1>;
1519 def VMULpq : N3VQInt<1, 0, 0b00, 0b1001, 1, IIC_VMULi16Q, "vmul.p8", v16i8, v16i8,
1520 int_arm_neon_vmulp, 1>;
1521 def VMULfd : N3VD<1, 0, 0b00, 0b1101, 1, IIC_VBIND, "vmul.f32", v2f32, v2f32, fmul, 1>;
1522 def VMULfq : N3VQ<1, 0, 0b00, 0b1101, 1, IIC_VBINQ, "vmul.f32", v4f32, v4f32, fmul, 1>;
1523 defm VMULsl : N3VSL_HS<0b1000, "vmul.i", mul>;
1524 def VMULslfd : N3VDSL<0b10, 0b1001, IIC_VBIND, "vmul.f32", v2f32, fmul>;
1525 def VMULslfq : N3VQSL<0b10, 0b1001, IIC_VBINQ, "vmul.f32", v4f32, v2f32, fmul>;
1526 def : Pat<(v8i16 (mul (v8i16 QPR:$src1),
1527 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
1528 (v8i16 (VMULslv8i16 (v8i16 QPR:$src1),
1529 (v4i16 (EXTRACT_SUBREG QPR:$src2,
1530 (DSubReg_i16_reg imm:$lane))),
1531 (SubReg_i16_lane imm:$lane)))>;
1532 def : Pat<(v4i32 (mul (v4i32 QPR:$src1),
1533 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
1534 (v4i32 (VMULslv4i32 (v4i32 QPR:$src1),
1535 (v2i32 (EXTRACT_SUBREG QPR:$src2,
1536 (DSubReg_i32_reg imm:$lane))),
1537 (SubReg_i32_lane imm:$lane)))>;
1538 def : Pat<(v4f32 (fmul (v4f32 QPR:$src1),
1539 (v4f32 (NEONvduplane (v4f32 QPR:$src2), imm:$lane)))),
1540 (v4f32 (VMULslfq (v4f32 QPR:$src1),
1541 (v2f32 (EXTRACT_SUBREG QPR:$src2,
1542 (DSubReg_i32_reg imm:$lane))),
1543 (SubReg_i32_lane imm:$lane)))>;
1545 // VQDMULH : Vector Saturating Doubling Multiply Returning High Half
1546 defm VQDMULH : N3VInt_HS<0, 0, 0b1011, 0, IIC_VMULi16D, IIC_VMULi32D,
1547 IIC_VMULi16Q, IIC_VMULi32Q,
1548 "vqdmulh.s", int_arm_neon_vqdmulh, 1>;
1549 defm VQDMULHsl: N3VIntSL_HS<0b1100, IIC_VMULi16D, IIC_VMULi32D,
1550 IIC_VMULi16Q, IIC_VMULi32Q,
1551 "vqdmulh.s", int_arm_neon_vqdmulh>;
1552 def : Pat<(v8i16 (int_arm_neon_vqdmulh (v8i16 QPR:$src1),
1553 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
1554 (v8i16 (VQDMULHslv8i16 (v8i16 QPR:$src1),
1555 (v4i16 (EXTRACT_SUBREG QPR:$src2,
1556 (DSubReg_i16_reg imm:$lane))),
1557 (SubReg_i16_lane imm:$lane)))>;
1558 def : Pat<(v4i32 (int_arm_neon_vqdmulh (v4i32 QPR:$src1),
1559 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
1560 (v4i32 (VQDMULHslv4i32 (v4i32 QPR:$src1),
1561 (v2i32 (EXTRACT_SUBREG QPR:$src2,
1562 (DSubReg_i32_reg imm:$lane))),
1563 (SubReg_i32_lane imm:$lane)))>;
1565 // VQRDMULH : Vector Rounding Saturating Doubling Multiply Returning High Half
1566 defm VQRDMULH : N3VInt_HS<1, 0, 0b1011, 0, IIC_VMULi16D, IIC_VMULi32D,
1567 IIC_VMULi16Q, IIC_VMULi32Q,
1568 "vqrdmulh.s", int_arm_neon_vqrdmulh, 1>;
1569 defm VQRDMULHsl : N3VIntSL_HS<0b1101, IIC_VMULi16D, IIC_VMULi32D,
1570 IIC_VMULi16Q, IIC_VMULi32Q,
1571 "vqrdmulh.s", int_arm_neon_vqrdmulh>;
1572 def : Pat<(v8i16 (int_arm_neon_vqrdmulh (v8i16 QPR:$src1),
1573 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
1574 (v8i16 (VQRDMULHslv8i16 (v8i16 QPR:$src1),
1575 (v4i16 (EXTRACT_SUBREG QPR:$src2,
1576 (DSubReg_i16_reg imm:$lane))),
1577 (SubReg_i16_lane imm:$lane)))>;
1578 def : Pat<(v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$src1),
1579 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
1580 (v4i32 (VQRDMULHslv4i32 (v4i32 QPR:$src1),
1581 (v2i32 (EXTRACT_SUBREG QPR:$src2,
1582 (DSubReg_i32_reg imm:$lane))),
1583 (SubReg_i32_lane imm:$lane)))>;
1585 // VMULL : Vector Multiply Long (integer and polynomial) (Q = D * D)
1586 defm VMULLs : N3VLInt_QHS<0,1,0b1100,0, IIC_VMULi16D, "vmull.s", int_arm_neon_vmulls, 1>;
1587 defm VMULLu : N3VLInt_QHS<1,1,0b1100,0, IIC_VMULi16D, "vmull.u", int_arm_neon_vmullu, 1>;
1588 def VMULLp : N3VLInt<0, 1, 0b00, 0b1110, 0, IIC_VMULi16D, "vmull.p8", v8i16, v8i8,
1589 int_arm_neon_vmullp, 1>;
1590 defm VMULLsls : N3VLIntSL_HS<0, 0b1010, IIC_VMULi16D, "vmull.s", int_arm_neon_vmulls>;
1591 defm VMULLslu : N3VLIntSL_HS<1, 0b1010, IIC_VMULi16D, "vmull.u", int_arm_neon_vmullu>;
1593 // VQDMULL : Vector Saturating Doubling Multiply Long (Q = D * D)
1594 defm VQDMULL : N3VLInt_HS<0,1,0b1101,0, IIC_VMULi16D, "vqdmull.s", int_arm_neon_vqdmull, 1>;
1595 defm VQDMULLsl: N3VLIntSL_HS<0, 0b1011, IIC_VMULi16D, "vqdmull.s", int_arm_neon_vqdmull>;
1597 // Vector Multiply-Accumulate and Multiply-Subtract Operations.
1599 // VMLA : Vector Multiply Accumulate (integer and floating-point)
1600 defm VMLA : N3VMulOp_QHS<0, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
1601 IIC_VMACi16Q, IIC_VMACi32Q, "vmla.i", add>;
1602 def VMLAfd : N3VDMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACD, "vmla.f32", v2f32, fmul, fadd>;
1603 def VMLAfq : N3VQMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACQ, "vmla.f32", v4f32, fmul, fadd>;
1604 defm VMLAsl : N3VMulOpSL_HS<0b0000, IIC_VMACi16D, IIC_VMACi32D,
1605 IIC_VMACi16Q, IIC_VMACi32Q, "vmla.i", add>;
1606 def VMLAslfd : N3VDMulOpSL<0b10, 0b0001, IIC_VMACD, "vmla.f32", v2f32, fmul, fadd>;
1607 def VMLAslfq : N3VQMulOpSL<0b10, 0b0001, IIC_VMACQ, "vmla.f32", v4f32, v2f32, fmul, fadd>;
1609 def : Pat<(v8i16 (add (v8i16 QPR:$src1),
1610 (mul (v8i16 QPR:$src2),
1611 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
1612 (v8i16 (VMLAslv8i16 (v8i16 QPR:$src1),
1614 (v4i16 (EXTRACT_SUBREG QPR:$src3,
1615 (DSubReg_i16_reg imm:$lane))),
1616 (SubReg_i16_lane imm:$lane)))>;
1618 def : Pat<(v4i32 (add (v4i32 QPR:$src1),
1619 (mul (v4i32 QPR:$src2),
1620 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
1621 (v4i32 (VMLAslv4i32 (v4i32 QPR:$src1),
1623 (v2i32 (EXTRACT_SUBREG QPR:$src3,
1624 (DSubReg_i32_reg imm:$lane))),
1625 (SubReg_i32_lane imm:$lane)))>;
1627 def : Pat<(v4f32 (fadd (v4f32 QPR:$src1),
1628 (fmul (v4f32 QPR:$src2),
1629 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
1630 (v4f32 (VMLAslfq (v4f32 QPR:$src1),
1632 (v2f32 (EXTRACT_SUBREG QPR:$src3,
1633 (DSubReg_i32_reg imm:$lane))),
1634 (SubReg_i32_lane imm:$lane)))>;
1636 // VMLAL : Vector Multiply Accumulate Long (Q += D * D)
1637 defm VMLALs : N3VLInt3_QHS<0,1,0b1000,0, "vmlal.s", int_arm_neon_vmlals>;
1638 defm VMLALu : N3VLInt3_QHS<1,1,0b1000,0, "vmlal.u", int_arm_neon_vmlalu>;
1640 defm VMLALsls : N3VLInt3SL_HS<0, 0b0010, "vmlal.s", int_arm_neon_vmlals>;
1641 defm VMLALslu : N3VLInt3SL_HS<1, 0b0010, "vmlal.u", int_arm_neon_vmlalu>;
1643 // VQDMLAL : Vector Saturating Doubling Multiply Accumulate Long (Q += D * D)
1644 defm VQDMLAL : N3VLInt3_HS<0, 1, 0b1001, 0, "vqdmlal.s", int_arm_neon_vqdmlal>;
1645 defm VQDMLALsl: N3VLInt3SL_HS<0, 0b0011, "vqdmlal.s", int_arm_neon_vqdmlal>;
1647 // VMLS : Vector Multiply Subtract (integer and floating-point)
1648 defm VMLS : N3VMulOp_QHS<1, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
1649 IIC_VMACi16Q, IIC_VMACi32Q, "vmls.i", sub>;
1650 def VMLSfd : N3VDMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACD, "vmls.f32", v2f32, fmul, fsub>;
1651 def VMLSfq : N3VQMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACQ, "vmls.f32", v4f32, fmul, fsub>;
1652 defm VMLSsl : N3VMulOpSL_HS<0b0100, IIC_VMACi16D, IIC_VMACi32D,
1653 IIC_VMACi16Q, IIC_VMACi32Q, "vmls.i", sub>;
1654 def VMLSslfd : N3VDMulOpSL<0b10, 0b0101, IIC_VMACD, "vmls.f32", v2f32, fmul, fsub>;
1655 def VMLSslfq : N3VQMulOpSL<0b10, 0b0101, IIC_VMACQ, "vmls.f32", v4f32, v2f32, fmul, fsub>;
1657 def : Pat<(v8i16 (sub (v8i16 QPR:$src1),
1658 (mul (v8i16 QPR:$src2),
1659 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
1660 (v8i16 (VMLSslv8i16 (v8i16 QPR:$src1),
1662 (v4i16 (EXTRACT_SUBREG QPR:$src3,
1663 (DSubReg_i16_reg imm:$lane))),
1664 (SubReg_i16_lane imm:$lane)))>;
1666 def : Pat<(v4i32 (sub (v4i32 QPR:$src1),
1667 (mul (v4i32 QPR:$src2),
1668 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
1669 (v4i32 (VMLSslv4i32 (v4i32 QPR:$src1),
1671 (v2i32 (EXTRACT_SUBREG QPR:$src3,
1672 (DSubReg_i32_reg imm:$lane))),
1673 (SubReg_i32_lane imm:$lane)))>;
1675 def : Pat<(v4f32 (fsub (v4f32 QPR:$src1),
1676 (fmul (v4f32 QPR:$src2),
1677 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
1678 (v4f32 (VMLSslfq (v4f32 QPR:$src1),
1680 (v2f32 (EXTRACT_SUBREG QPR:$src3,
1681 (DSubReg_i32_reg imm:$lane))),
1682 (SubReg_i32_lane imm:$lane)))>;
1684 // VMLSL : Vector Multiply Subtract Long (Q -= D * D)
1685 defm VMLSLs : N3VLInt3_QHS<0,1,0b1010,0, "vmlsl.s", int_arm_neon_vmlsls>;
1686 defm VMLSLu : N3VLInt3_QHS<1,1,0b1010,0, "vmlsl.u", int_arm_neon_vmlslu>;
1688 defm VMLSLsls : N3VLInt3SL_HS<0, 0b0110, "vmlsl.s", int_arm_neon_vmlsls>;
1689 defm VMLSLslu : N3VLInt3SL_HS<1, 0b0110, "vmlsl.u", int_arm_neon_vmlslu>;
1691 // VQDMLSL : Vector Saturating Doubling Multiply Subtract Long (Q -= D * D)
1692 defm VQDMLSL : N3VLInt3_HS<0, 1, 0b1011, 0, "vqdmlsl.s", int_arm_neon_vqdmlsl>;
1693 defm VQDMLSLsl: N3VLInt3SL_HS<0, 0b111, "vqdmlsl.s", int_arm_neon_vqdmlsl>;
1695 // Vector Subtract Operations.
1697 // VSUB : Vector Subtract (integer and floating-point)
1698 defm VSUB : N3V_QHSD<1, 0, 0b1000, 0, IIC_VSUBiD, IIC_VSUBiQ, "vsub.i", sub, 0>;
1699 def VSUBfd : N3VD<0, 0, 0b10, 0b1101, 0, IIC_VBIND, "vsub.f32", v2f32, v2f32, fsub, 0>;
1700 def VSUBfq : N3VQ<0, 0, 0b10, 0b1101, 0, IIC_VBINQ, "vsub.f32", v4f32, v4f32, fsub, 0>;
1701 // VSUBL : Vector Subtract Long (Q = D - D)
1702 defm VSUBLs : N3VLInt_QHS<0,1,0b0010,0, IIC_VSHLiD, "vsubl.s", int_arm_neon_vsubls, 1>;
1703 defm VSUBLu : N3VLInt_QHS<1,1,0b0010,0, IIC_VSHLiD, "vsubl.u", int_arm_neon_vsublu, 1>;
1704 // VSUBW : Vector Subtract Wide (Q = Q - D)
1705 defm VSUBWs : N3VWInt_QHS<0,1,0b0011,0, "vsubw.s", int_arm_neon_vsubws, 0>;
1706 defm VSUBWu : N3VWInt_QHS<1,1,0b0011,0, "vsubw.u", int_arm_neon_vsubwu, 0>;
1707 // VHSUB : Vector Halving Subtract
1708 defm VHSUBs : N3VInt_QHS<0, 0, 0b0010, 0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1709 IIC_VBINi4Q, "vhsub.s", int_arm_neon_vhsubs, 0>;
1710 defm VHSUBu : N3VInt_QHS<1, 0, 0b0010, 0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1711 IIC_VBINi4Q, "vhsub.u", int_arm_neon_vhsubu, 0>;
1712 // VQSUB : Vector Saturing Subtract
1713 defm VQSUBs : N3VInt_QHSD<0, 0, 0b0010, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1714 IIC_VBINi4Q, "vqsub.s", int_arm_neon_vqsubs, 0>;
1715 defm VQSUBu : N3VInt_QHSD<1, 0, 0b0010, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1716 IIC_VBINi4Q, "vqsub.u", int_arm_neon_vqsubu, 0>;
1717 // VSUBHN : Vector Subtract and Narrow Returning High Half (D = Q - Q)
1718 defm VSUBHN : N3VNInt_HSD<0,1,0b0110,0, "vsubhn.i", int_arm_neon_vsubhn, 0>;
1719 // VRSUBHN : Vector Rounding Subtract and Narrow Returning High Half (D=Q-Q)
1720 defm VRSUBHN : N3VNInt_HSD<1,1,0b0110,0, "vrsubhn.i", int_arm_neon_vrsubhn, 0>;
1722 // Vector Comparisons.
1724 // VCEQ : Vector Compare Equal
1725 defm VCEQ : N3V_QHS<1, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1726 IIC_VBINi4Q, "vceq.i", NEONvceq, 1>;
1727 def VCEQfd : N3VD<0,0,0b00,0b1110,0, IIC_VBIND, "vceq.f32", v2i32, v2f32, NEONvceq, 1>;
1728 def VCEQfq : N3VQ<0,0,0b00,0b1110,0, IIC_VBINQ, "vceq.f32", v4i32, v4f32, NEONvceq, 1>;
1729 // VCGE : Vector Compare Greater Than or Equal
1730 defm VCGEs : N3V_QHS<0, 0, 0b0011, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1731 IIC_VBINi4Q, "vcge.s", NEONvcge, 0>;
1732 defm VCGEu : N3V_QHS<1, 0, 0b0011, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1733 IIC_VBINi4Q, "vcge.u", NEONvcgeu, 0>;
1734 def VCGEfd : N3VD<1,0,0b00,0b1110,0, IIC_VBIND, "vcge.f32", v2i32, v2f32, NEONvcge, 0>;
1735 def VCGEfq : N3VQ<1,0,0b00,0b1110,0, IIC_VBINQ, "vcge.f32", v4i32, v4f32, NEONvcge, 0>;
1736 // VCGT : Vector Compare Greater Than
1737 defm VCGTs : N3V_QHS<0, 0, 0b0011, 0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1738 IIC_VBINi4Q, "vcgt.s", NEONvcgt, 0>;
1739 defm VCGTu : N3V_QHS<1, 0, 0b0011, 0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1740 IIC_VBINi4Q, "vcgt.u", NEONvcgtu, 0>;
1741 def VCGTfd : N3VD<1,0,0b10,0b1110,0, IIC_VBIND, "vcgt.f32", v2i32, v2f32, NEONvcgt, 0>;
1742 def VCGTfq : N3VQ<1,0,0b10,0b1110,0, IIC_VBINQ, "vcgt.f32", v4i32, v4f32, NEONvcgt, 0>;
1743 // VACGE : Vector Absolute Compare Greater Than or Equal (aka VCAGE)
1744 def VACGEd : N3VDInt<1, 0, 0b00, 0b1110, 1, IIC_VBIND, "vacge.f32", v2i32, v2f32,
1745 int_arm_neon_vacged, 0>;
1746 def VACGEq : N3VQInt<1, 0, 0b00, 0b1110, 1, IIC_VBINQ, "vacge.f32", v4i32, v4f32,
1747 int_arm_neon_vacgeq, 0>;
1748 // VACGT : Vector Absolute Compare Greater Than (aka VCAGT)
1749 def VACGTd : N3VDInt<1, 0, 0b10, 0b1110, 1, IIC_VBIND, "vacgt.f32", v2i32, v2f32,
1750 int_arm_neon_vacgtd, 0>;
1751 def VACGTq : N3VQInt<1, 0, 0b10, 0b1110, 1, IIC_VBINQ, "vacgt.f32", v4i32, v4f32,
1752 int_arm_neon_vacgtq, 0>;
1753 // VTST : Vector Test Bits
1754 defm VTST : N3V_QHS<0, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1755 IIC_VBINi4Q, "vtst.i", NEONvtst, 1>;
1757 // Vector Bitwise Operations.
1759 // VAND : Vector Bitwise AND
1760 def VANDd : N3VD<0, 0, 0b00, 0b0001, 1, IIC_VBINiD, "vand", v2i32, v2i32, and, 1>;
1761 def VANDq : N3VQ<0, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "vand", v4i32, v4i32, and, 1>;
1763 // VEOR : Vector Bitwise Exclusive OR
1764 def VEORd : N3VD<1, 0, 0b00, 0b0001, 1, IIC_VBINiD, "veor", v2i32, v2i32, xor, 1>;
1765 def VEORq : N3VQ<1, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "veor", v4i32, v4i32, xor, 1>;
1767 // VORR : Vector Bitwise OR
1768 def VORRd : N3VD<0, 0, 0b10, 0b0001, 1, IIC_VBINiD, "vorr", v2i32, v2i32, or, 1>;
1769 def VORRq : N3VQ<0, 0, 0b10, 0b0001, 1, IIC_VBINiQ, "vorr", v4i32, v4i32, or, 1>;
1771 // VBIC : Vector Bitwise Bit Clear (AND NOT)
1772 def VBICd : N3V<0, 0, 0b01, 0b0001, 0, 1, (outs DPR:$dst),
1773 (ins DPR:$src1, DPR:$src2), IIC_VBINiD,
1774 "vbic\t$dst, $src1, $src2", "",
1775 [(set DPR:$dst, (v2i32 (and DPR:$src1,
1776 (vnot_conv DPR:$src2))))]>;
1777 def VBICq : N3V<0, 0, 0b01, 0b0001, 1, 1, (outs QPR:$dst),
1778 (ins QPR:$src1, QPR:$src2), IIC_VBINiQ,
1779 "vbic\t$dst, $src1, $src2", "",
1780 [(set QPR:$dst, (v4i32 (and QPR:$src1,
1781 (vnot_conv QPR:$src2))))]>;
1783 // VORN : Vector Bitwise OR NOT
1784 def VORNd : N3V<0, 0, 0b11, 0b0001, 0, 1, (outs DPR:$dst),
1785 (ins DPR:$src1, DPR:$src2), IIC_VBINiD,
1786 "vorn\t$dst, $src1, $src2", "",
1787 [(set DPR:$dst, (v2i32 (or DPR:$src1,
1788 (vnot_conv DPR:$src2))))]>;
1789 def VORNq : N3V<0, 0, 0b11, 0b0001, 1, 1, (outs QPR:$dst),
1790 (ins QPR:$src1, QPR:$src2), IIC_VBINiQ,
1791 "vorn\t$dst, $src1, $src2", "",
1792 [(set QPR:$dst, (v4i32 (or QPR:$src1,
1793 (vnot_conv QPR:$src2))))]>;
1795 // VMVN : Vector Bitwise NOT
1796 def VMVNd : N2V<0b11, 0b11, 0b00, 0b00, 0b01011, 0, 0,
1797 (outs DPR:$dst), (ins DPR:$src), IIC_VSHLiD,
1798 "vmvn\t$dst, $src", "",
1799 [(set DPR:$dst, (v2i32 (vnot DPR:$src)))]>;
1800 def VMVNq : N2V<0b11, 0b11, 0b00, 0b00, 0b01011, 1, 0,
1801 (outs QPR:$dst), (ins QPR:$src), IIC_VSHLiD,
1802 "vmvn\t$dst, $src", "",
1803 [(set QPR:$dst, (v4i32 (vnot QPR:$src)))]>;
1804 def : Pat<(v2i32 (vnot_conv DPR:$src)), (VMVNd DPR:$src)>;
1805 def : Pat<(v4i32 (vnot_conv QPR:$src)), (VMVNq QPR:$src)>;
1807 // VBSL : Vector Bitwise Select
1808 def VBSLd : N3V<1, 0, 0b01, 0b0001, 0, 1, (outs DPR:$dst),
1809 (ins DPR:$src1, DPR:$src2, DPR:$src3), IIC_VCNTiD,
1810 "vbsl\t$dst, $src2, $src3", "$src1 = $dst",
1812 (v2i32 (or (and DPR:$src2, DPR:$src1),
1813 (and DPR:$src3, (vnot_conv DPR:$src1)))))]>;
1814 def VBSLq : N3V<1, 0, 0b01, 0b0001, 1, 1, (outs QPR:$dst),
1815 (ins QPR:$src1, QPR:$src2, QPR:$src3), IIC_VCNTiQ,
1816 "vbsl\t$dst, $src2, $src3", "$src1 = $dst",
1818 (v4i32 (or (and QPR:$src2, QPR:$src1),
1819 (and QPR:$src3, (vnot_conv QPR:$src1)))))]>;
1821 // VBIF : Vector Bitwise Insert if False
1822 // like VBSL but with: "vbif\t$dst, $src3, $src1", "$src2 = $dst",
1823 // VBIT : Vector Bitwise Insert if True
1824 // like VBSL but with: "vbit\t$dst, $src2, $src1", "$src3 = $dst",
1825 // These are not yet implemented. The TwoAddress pass will not go looking
1826 // for equivalent operations with different register constraints; it just
1829 // Vector Absolute Differences.
1831 // VABD : Vector Absolute Difference
1832 defm VABDs : N3VInt_QHS<0, 0, 0b0111, 0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1833 IIC_VBINi4Q, "vabd.s", int_arm_neon_vabds, 0>;
1834 defm VABDu : N3VInt_QHS<1, 0, 0b0111, 0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1835 IIC_VBINi4Q, "vabd.u", int_arm_neon_vabdu, 0>;
1836 def VABDfd : N3VDInt<1, 0, 0b10, 0b1101, 0, IIC_VBIND, "vabd.f32", v2f32, v2f32,
1837 int_arm_neon_vabds, 0>;
1838 def VABDfq : N3VQInt<1, 0, 0b10, 0b1101, 0, IIC_VBINQ, "vabd.f32", v4f32, v4f32,
1839 int_arm_neon_vabds, 0>;
1841 // VABDL : Vector Absolute Difference Long (Q = | D - D |)
1842 defm VABDLs : N3VLInt_QHS<0,1,0b0111,0, IIC_VBINi4Q, "vabdl.s", int_arm_neon_vabdls, 0>;
1843 defm VABDLu : N3VLInt_QHS<1,1,0b0111,0, IIC_VBINi4Q, "vabdl.u", int_arm_neon_vabdlu, 0>;
1845 // VABA : Vector Absolute Difference and Accumulate
1846 defm VABAs : N3VInt3_QHS<0,1,0b0101,0, "vaba.s", int_arm_neon_vabas>;
1847 defm VABAu : N3VInt3_QHS<1,1,0b0101,0, "vaba.u", int_arm_neon_vabau>;
1849 // VABAL : Vector Absolute Difference and Accumulate Long (Q += | D - D |)
1850 defm VABALs : N3VLInt3_QHS<0,1,0b0101,0, "vabal.s", int_arm_neon_vabals>;
1851 defm VABALu : N3VLInt3_QHS<1,1,0b0101,0, "vabal.u", int_arm_neon_vabalu>;
1853 // Vector Maximum and Minimum.
1855 // VMAX : Vector Maximum
1856 defm VMAXs : N3VInt_QHS<0, 0, 0b0110, 0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1857 IIC_VBINi4Q, "vmax.s", int_arm_neon_vmaxs, 1>;
1858 defm VMAXu : N3VInt_QHS<1, 0, 0b0110, 0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1859 IIC_VBINi4Q, "vmax.u", int_arm_neon_vmaxu, 1>;
1860 def VMAXfd : N3VDInt<0, 0, 0b00, 0b1111, 0, IIC_VBIND, "vmax.f32", v2f32, v2f32,
1861 int_arm_neon_vmaxs, 1>;
1862 def VMAXfq : N3VQInt<0, 0, 0b00, 0b1111, 0, IIC_VBINQ, "vmax.f32", v4f32, v4f32,
1863 int_arm_neon_vmaxs, 1>;
1865 // VMIN : Vector Minimum
1866 defm VMINs : N3VInt_QHS<0, 0, 0b0110, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1867 IIC_VBINi4Q, "vmin.s", int_arm_neon_vmins, 1>;
1868 defm VMINu : N3VInt_QHS<1, 0, 0b0110, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1869 IIC_VBINi4Q, "vmin.u", int_arm_neon_vminu, 1>;
1870 def VMINfd : N3VDInt<0, 0, 0b10, 0b1111, 0, IIC_VBIND, "vmin.f32", v2f32, v2f32,
1871 int_arm_neon_vmins, 1>;
1872 def VMINfq : N3VQInt<0, 0, 0b10, 0b1111, 0, IIC_VBINQ, "vmin.f32", v4f32, v4f32,
1873 int_arm_neon_vmins, 1>;
1875 // Vector Pairwise Operations.
1877 // VPADD : Vector Pairwise Add
1878 def VPADDi8 : N3VDInt<0, 0, 0b00, 0b1011, 1, IIC_VBINiD, "vpadd.i8", v8i8, v8i8,
1879 int_arm_neon_vpadd, 0>;
1880 def VPADDi16 : N3VDInt<0, 0, 0b01, 0b1011, 1, IIC_VBINiD, "vpadd.i16", v4i16, v4i16,
1881 int_arm_neon_vpadd, 0>;
1882 def VPADDi32 : N3VDInt<0, 0, 0b10, 0b1011, 1, IIC_VBINiD, "vpadd.i32", v2i32, v2i32,
1883 int_arm_neon_vpadd, 0>;
1884 def VPADDf : N3VDInt<1, 0, 0b00, 0b1101, 0, IIC_VBIND, "vpadd.f32", v2f32, v2f32,
1885 int_arm_neon_vpadd, 0>;
1887 // VPADDL : Vector Pairwise Add Long
1888 defm VPADDLs : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00100, 0, "vpaddl.s",
1889 int_arm_neon_vpaddls>;
1890 defm VPADDLu : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00101, 0, "vpaddl.u",
1891 int_arm_neon_vpaddlu>;
1893 // VPADAL : Vector Pairwise Add and Accumulate Long
1894 defm VPADALs : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b00100, 0, "vpadal.s",
1895 int_arm_neon_vpadals>;
1896 defm VPADALu : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b00101, 0, "vpadal.u",
1897 int_arm_neon_vpadalu>;
1899 // VPMAX : Vector Pairwise Maximum
1900 def VPMAXs8 : N3VDInt<0, 0, 0b00, 0b1010, 0, IIC_VBINi4D, "vpmax.s8", v8i8, v8i8,
1901 int_arm_neon_vpmaxs, 0>;
1902 def VPMAXs16 : N3VDInt<0, 0, 0b01, 0b1010, 0, IIC_VBINi4D, "vpmax.s16", v4i16, v4i16,
1903 int_arm_neon_vpmaxs, 0>;
1904 def VPMAXs32 : N3VDInt<0, 0, 0b10, 0b1010, 0, IIC_VBINi4D, "vpmax.s32", v2i32, v2i32,
1905 int_arm_neon_vpmaxs, 0>;
1906 def VPMAXu8 : N3VDInt<1, 0, 0b00, 0b1010, 0, IIC_VBINi4D, "vpmax.u8", v8i8, v8i8,
1907 int_arm_neon_vpmaxu, 0>;
1908 def VPMAXu16 : N3VDInt<1, 0, 0b01, 0b1010, 0, IIC_VBINi4D, "vpmax.u16", v4i16, v4i16,
1909 int_arm_neon_vpmaxu, 0>;
1910 def VPMAXu32 : N3VDInt<1, 0, 0b10, 0b1010, 0, IIC_VBINi4D, "vpmax.u32", v2i32, v2i32,
1911 int_arm_neon_vpmaxu, 0>;
1912 def VPMAXf : N3VDInt<1, 0, 0b00, 0b1111, 0, IIC_VBINi4D, "vpmax.f32", v2f32, v2f32,
1913 int_arm_neon_vpmaxs, 0>;
1915 // VPMIN : Vector Pairwise Minimum
1916 def VPMINs8 : N3VDInt<0, 0, 0b00, 0b1010, 1, IIC_VBINi4D, "vpmin.s8", v8i8, v8i8,
1917 int_arm_neon_vpmins, 0>;
1918 def VPMINs16 : N3VDInt<0, 0, 0b01, 0b1010, 1, IIC_VBINi4D, "vpmin.s16", v4i16, v4i16,
1919 int_arm_neon_vpmins, 0>;
1920 def VPMINs32 : N3VDInt<0, 0, 0b10, 0b1010, 1, IIC_VBINi4D, "vpmin.s32", v2i32, v2i32,
1921 int_arm_neon_vpmins, 0>;
1922 def VPMINu8 : N3VDInt<1, 0, 0b00, 0b1010, 1, IIC_VBINi4D, "vpmin.u8", v8i8, v8i8,
1923 int_arm_neon_vpminu, 0>;
1924 def VPMINu16 : N3VDInt<1, 0, 0b01, 0b1010, 1, IIC_VBINi4D, "vpmin.u16", v4i16, v4i16,
1925 int_arm_neon_vpminu, 0>;
1926 def VPMINu32 : N3VDInt<1, 0, 0b10, 0b1010, 1, IIC_VBINi4D, "vpmin.u32", v2i32, v2i32,
1927 int_arm_neon_vpminu, 0>;
1928 def VPMINf : N3VDInt<1, 0, 0b10, 0b1111, 0, IIC_VBINi4D, "vpmin.f32", v2f32, v2f32,
1929 int_arm_neon_vpmins, 0>;
1931 // Vector Reciprocal and Reciprocal Square Root Estimate and Step.
1933 // VRECPE : Vector Reciprocal Estimate
1934 def VRECPEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
1935 IIC_VUNAD, "vrecpe.u32",
1936 v2i32, v2i32, int_arm_neon_vrecpe>;
1937 def VRECPEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
1938 IIC_VUNAQ, "vrecpe.u32",
1939 v4i32, v4i32, int_arm_neon_vrecpe>;
1940 def VRECPEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
1941 IIC_VUNAD, "vrecpe.f32",
1942 v2f32, v2f32, int_arm_neon_vrecpe>;
1943 def VRECPEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
1944 IIC_VUNAQ, "vrecpe.f32",
1945 v4f32, v4f32, int_arm_neon_vrecpe>;
1947 // VRECPS : Vector Reciprocal Step
1948 def VRECPSfd : N3VDInt<0, 0, 0b00, 0b1111, 1, IIC_VRECSD, "vrecps.f32", v2f32, v2f32,
1949 int_arm_neon_vrecps, 1>;
1950 def VRECPSfq : N3VQInt<0, 0, 0b00, 0b1111, 1, IIC_VRECSQ, "vrecps.f32", v4f32, v4f32,
1951 int_arm_neon_vrecps, 1>;
1953 // VRSQRTE : Vector Reciprocal Square Root Estimate
1954 def VRSQRTEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
1955 IIC_VUNAD, "vrsqrte.u32",
1956 v2i32, v2i32, int_arm_neon_vrsqrte>;
1957 def VRSQRTEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
1958 IIC_VUNAQ, "vrsqrte.u32",
1959 v4i32, v4i32, int_arm_neon_vrsqrte>;
1960 def VRSQRTEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
1961 IIC_VUNAD, "vrsqrte.f32",
1962 v2f32, v2f32, int_arm_neon_vrsqrte>;
1963 def VRSQRTEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
1964 IIC_VUNAQ, "vrsqrte.f32",
1965 v4f32, v4f32, int_arm_neon_vrsqrte>;
1967 // VRSQRTS : Vector Reciprocal Square Root Step
1968 def VRSQRTSfd : N3VDInt<0, 0, 0b10, 0b1111, 1, IIC_VRECSD, "vrsqrts.f32", v2f32, v2f32,
1969 int_arm_neon_vrsqrts, 1>;
1970 def VRSQRTSfq : N3VQInt<0, 0, 0b10, 0b1111, 1, IIC_VRECSQ, "vrsqrts.f32", v4f32, v4f32,
1971 int_arm_neon_vrsqrts, 1>;
1975 // VSHL : Vector Shift
1976 defm VSHLs : N3VInt_QHSD<0, 0, 0b0100, 0, IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ,
1977 IIC_VSHLiQ, "vshl.s", int_arm_neon_vshifts, 0>;
1978 defm VSHLu : N3VInt_QHSD<1, 0, 0b0100, 0, IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ,
1979 IIC_VSHLiQ, "vshl.u", int_arm_neon_vshiftu, 0>;
1980 // VSHL : Vector Shift Left (Immediate)
1981 defm VSHLi : N2VSh_QHSD<0, 1, 0b0111, 1, IIC_VSHLiD, "vshl.i", NEONvshl>;
1982 // VSHR : Vector Shift Right (Immediate)
1983 defm VSHRs : N2VSh_QHSD<0, 1, 0b0000, 1, IIC_VSHLiD, "vshr.s", NEONvshrs>;
1984 defm VSHRu : N2VSh_QHSD<1, 1, 0b0000, 1, IIC_VSHLiD, "vshr.u", NEONvshru>;
1986 // VSHLL : Vector Shift Left Long
1987 def VSHLLs8 : N2VLSh<0, 1, 0b001000, 0b1010, 0, 0, 1, "vshll.s8",
1988 v8i16, v8i8, NEONvshlls>;
1989 def VSHLLs16 : N2VLSh<0, 1, 0b010000, 0b1010, 0, 0, 1, "vshll.s16",
1990 v4i32, v4i16, NEONvshlls>;
1991 def VSHLLs32 : N2VLSh<0, 1, 0b100000, 0b1010, 0, 0, 1, "vshll.s32",
1992 v2i64, v2i32, NEONvshlls>;
1993 def VSHLLu8 : N2VLSh<1, 1, 0b001000, 0b1010, 0, 0, 1, "vshll.u8",
1994 v8i16, v8i8, NEONvshllu>;
1995 def VSHLLu16 : N2VLSh<1, 1, 0b010000, 0b1010, 0, 0, 1, "vshll.u16",
1996 v4i32, v4i16, NEONvshllu>;
1997 def VSHLLu32 : N2VLSh<1, 1, 0b100000, 0b1010, 0, 0, 1, "vshll.u32",
1998 v2i64, v2i32, NEONvshllu>;
2000 // VSHLL : Vector Shift Left Long (with maximum shift count)
2001 def VSHLLi8 : N2VLSh<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll.i8",
2002 v8i16, v8i8, NEONvshlli>;
2003 def VSHLLi16 : N2VLSh<1, 1, 0b110110, 0b0011, 0, 0, 0, "vshll.i16",
2004 v4i32, v4i16, NEONvshlli>;
2005 def VSHLLi32 : N2VLSh<1, 1, 0b111010, 0b0011, 0, 0, 0, "vshll.i32",
2006 v2i64, v2i32, NEONvshlli>;
2008 // VSHRN : Vector Shift Right and Narrow
2009 def VSHRN16 : N2VNSh<0, 1, 0b001000, 0b1000, 0, 0, 1,
2010 IIC_VSHLiD, "vshrn.i16", v8i8, v8i16, NEONvshrn>;
2011 def VSHRN32 : N2VNSh<0, 1, 0b010000, 0b1000, 0, 0, 1,
2012 IIC_VSHLiD, "vshrn.i32", v4i16, v4i32, NEONvshrn>;
2013 def VSHRN64 : N2VNSh<0, 1, 0b100000, 0b1000, 0, 0, 1,
2014 IIC_VSHLiD, "vshrn.i64", v2i32, v2i64, NEONvshrn>;
2016 // VRSHL : Vector Rounding Shift
2017 defm VRSHLs : N3VInt_QHSD<0,0,0b0101,0, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2018 IIC_VSHLi4Q, "vrshl.s", int_arm_neon_vrshifts, 0>;
2019 defm VRSHLu : N3VInt_QHSD<1,0,0b0101,0, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2020 IIC_VSHLi4Q, "vrshl.u", int_arm_neon_vrshiftu, 0>;
2021 // VRSHR : Vector Rounding Shift Right
2022 defm VRSHRs : N2VSh_QHSD<0, 1, 0b0010, 1, IIC_VSHLi4D, "vrshr.s", NEONvrshrs>;
2023 defm VRSHRu : N2VSh_QHSD<1, 1, 0b0010, 1, IIC_VSHLi4D, "vrshr.u", NEONvrshru>;
2025 // VRSHRN : Vector Rounding Shift Right and Narrow
2026 def VRSHRN16 : N2VNSh<0, 1, 0b001000, 0b1000, 0, 1, 1,
2027 IIC_VSHLi4D, "vrshrn.i16", v8i8, v8i16, NEONvrshrn>;
2028 def VRSHRN32 : N2VNSh<0, 1, 0b010000, 0b1000, 0, 1, 1,
2029 IIC_VSHLi4D, "vrshrn.i32", v4i16, v4i32, NEONvrshrn>;
2030 def VRSHRN64 : N2VNSh<0, 1, 0b100000, 0b1000, 0, 1, 1,
2031 IIC_VSHLi4D, "vrshrn.i64", v2i32, v2i64, NEONvrshrn>;
2033 // VQSHL : Vector Saturating Shift
2034 defm VQSHLs : N3VInt_QHSD<0,0,0b0100,1, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2035 IIC_VSHLi4Q, "vqshl.s", int_arm_neon_vqshifts, 0>;
2036 defm VQSHLu : N3VInt_QHSD<1,0,0b0100,1, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2037 IIC_VSHLi4Q, "vqshl.u", int_arm_neon_vqshiftu, 0>;
2038 // VQSHL : Vector Saturating Shift Left (Immediate)
2039 defm VQSHLsi : N2VSh_QHSD<0, 1, 0b0111, 1, IIC_VSHLi4D, "vqshl.s", NEONvqshls>;
2040 defm VQSHLui : N2VSh_QHSD<1, 1, 0b0111, 1, IIC_VSHLi4D, "vqshl.u", NEONvqshlu>;
2041 // VQSHLU : Vector Saturating Shift Left (Immediate, Unsigned)
2042 defm VQSHLsu : N2VSh_QHSD<1, 1, 0b0110, 1, IIC_VSHLi4D, "vqshlu.s", NEONvqshlsu>;
2044 // VQSHRN : Vector Saturating Shift Right and Narrow
2045 def VQSHRNs16 : N2VNSh<0, 1, 0b001000, 0b1001, 0, 0, 1,
2046 IIC_VSHLi4D, "vqshrn.s16", v8i8, v8i16, NEONvqshrns>;
2047 def VQSHRNs32 : N2VNSh<0, 1, 0b010000, 0b1001, 0, 0, 1,
2048 IIC_VSHLi4D, "vqshrn.s32", v4i16, v4i32, NEONvqshrns>;
2049 def VQSHRNs64 : N2VNSh<0, 1, 0b100000, 0b1001, 0, 0, 1,
2050 IIC_VSHLi4D, "vqshrn.s64", v2i32, v2i64, NEONvqshrns>;
2051 def VQSHRNu16 : N2VNSh<1, 1, 0b001000, 0b1001, 0, 0, 1,
2052 IIC_VSHLi4D, "vqshrn.u16", v8i8, v8i16, NEONvqshrnu>;
2053 def VQSHRNu32 : N2VNSh<1, 1, 0b010000, 0b1001, 0, 0, 1,
2054 IIC_VSHLi4D, "vqshrn.u32", v4i16, v4i32, NEONvqshrnu>;
2055 def VQSHRNu64 : N2VNSh<1, 1, 0b100000, 0b1001, 0, 0, 1,
2056 IIC_VSHLi4D, "vqshrn.u64", v2i32, v2i64, NEONvqshrnu>;
2058 // VQSHRUN : Vector Saturating Shift Right and Narrow (Unsigned)
2059 def VQSHRUN16 : N2VNSh<1, 1, 0b001000, 0b1000, 0, 0, 1,
2060 IIC_VSHLi4D, "vqshrun.s16", v8i8, v8i16, NEONvqshrnsu>;
2061 def VQSHRUN32 : N2VNSh<1, 1, 0b010000, 0b1000, 0, 0, 1,
2062 IIC_VSHLi4D, "vqshrun.s32", v4i16, v4i32, NEONvqshrnsu>;
2063 def VQSHRUN64 : N2VNSh<1, 1, 0b100000, 0b1000, 0, 0, 1,
2064 IIC_VSHLi4D, "vqshrun.s64", v2i32, v2i64, NEONvqshrnsu>;
2066 // VQRSHL : Vector Saturating Rounding Shift
2067 defm VQRSHLs : N3VInt_QHSD<0, 0, 0b0101, 1, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2068 IIC_VSHLi4Q, "vqrshl.s", int_arm_neon_vqrshifts, 0>;
2069 defm VQRSHLu : N3VInt_QHSD<1, 0, 0b0101, 1, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2070 IIC_VSHLi4Q, "vqrshl.u", int_arm_neon_vqrshiftu, 0>;
2072 // VQRSHRN : Vector Saturating Rounding Shift Right and Narrow
2073 def VQRSHRNs16: N2VNSh<0, 1, 0b001000, 0b1001, 0, 1, 1,
2074 IIC_VSHLi4D, "vqrshrn.s16", v8i8, v8i16, NEONvqrshrns>;
2075 def VQRSHRNs32: N2VNSh<0, 1, 0b010000, 0b1001, 0, 1, 1,
2076 IIC_VSHLi4D, "vqrshrn.s32", v4i16, v4i32, NEONvqrshrns>;
2077 def VQRSHRNs64: N2VNSh<0, 1, 0b100000, 0b1001, 0, 1, 1,
2078 IIC_VSHLi4D, "vqrshrn.s64", v2i32, v2i64, NEONvqrshrns>;
2079 def VQRSHRNu16: N2VNSh<1, 1, 0b001000, 0b1001, 0, 1, 1,
2080 IIC_VSHLi4D, "vqrshrn.u16", v8i8, v8i16, NEONvqrshrnu>;
2081 def VQRSHRNu32: N2VNSh<1, 1, 0b010000, 0b1001, 0, 1, 1,
2082 IIC_VSHLi4D, "vqrshrn.u32", v4i16, v4i32, NEONvqrshrnu>;
2083 def VQRSHRNu64: N2VNSh<1, 1, 0b100000, 0b1001, 0, 1, 1,
2084 IIC_VSHLi4D, "vqrshrn.u64", v2i32, v2i64, NEONvqrshrnu>;
2086 // VQRSHRUN : Vector Saturating Rounding Shift Right and Narrow (Unsigned)
2087 def VQRSHRUN16: N2VNSh<1, 1, 0b001000, 0b1000, 0, 1, 1,
2088 IIC_VSHLi4D, "vqrshrun.s16", v8i8, v8i16, NEONvqrshrnsu>;
2089 def VQRSHRUN32: N2VNSh<1, 1, 0b010000, 0b1000, 0, 1, 1,
2090 IIC_VSHLi4D, "vqrshrun.s32", v4i16, v4i32, NEONvqrshrnsu>;
2091 def VQRSHRUN64: N2VNSh<1, 1, 0b100000, 0b1000, 0, 1, 1,
2092 IIC_VSHLi4D, "vqrshrun.s64", v2i32, v2i64, NEONvqrshrnsu>;
2094 // VSRA : Vector Shift Right and Accumulate
2095 defm VSRAs : N2VShAdd_QHSD<0, 1, 0b0001, 1, "vsra.s", NEONvshrs>;
2096 defm VSRAu : N2VShAdd_QHSD<1, 1, 0b0001, 1, "vsra.u", NEONvshru>;
2097 // VRSRA : Vector Rounding Shift Right and Accumulate
2098 defm VRSRAs : N2VShAdd_QHSD<0, 1, 0b0011, 1, "vrsra.s", NEONvrshrs>;
2099 defm VRSRAu : N2VShAdd_QHSD<1, 1, 0b0011, 1, "vrsra.u", NEONvrshru>;
2101 // VSLI : Vector Shift Left and Insert
2102 defm VSLI : N2VShIns_QHSD<1, 1, 0b0101, 1, "vsli.", NEONvsli>;
2103 // VSRI : Vector Shift Right and Insert
2104 defm VSRI : N2VShIns_QHSD<1, 1, 0b0100, 1, "vsri.", NEONvsri>;
2106 // Vector Absolute and Saturating Absolute.
2108 // VABS : Vector Absolute Value
2109 defm VABS : N2VInt_QHS<0b11, 0b11, 0b01, 0b00110, 0,
2110 IIC_VUNAiD, IIC_VUNAiQ, "vabs.s",
2112 def VABSfd : N2VDInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
2113 IIC_VUNAD, "vabs.f32",
2114 v2f32, v2f32, int_arm_neon_vabs>;
2115 def VABSfq : N2VQInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
2116 IIC_VUNAQ, "vabs.f32",
2117 v4f32, v4f32, int_arm_neon_vabs>;
2119 // VQABS : Vector Saturating Absolute Value
2120 defm VQABS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01110, 0,
2121 IIC_VQUNAiD, IIC_VQUNAiQ, "vqabs.s",
2122 int_arm_neon_vqabs>;
2126 def vneg : PatFrag<(ops node:$in), (sub immAllZerosV, node:$in)>;
2127 def vneg_conv : PatFrag<(ops node:$in), (sub immAllZerosV_bc, node:$in)>;
2129 class VNEGD<bits<2> size, string OpcodeStr, ValueType Ty>
2130 : N2V<0b11, 0b11, size, 0b01, 0b00111, 0, 0, (outs DPR:$dst), (ins DPR:$src),
2131 IIC_VSHLiD, !strconcat(OpcodeStr, "\t$dst, $src"), "",
2132 [(set DPR:$dst, (Ty (vneg DPR:$src)))]>;
2133 class VNEGQ<bits<2> size, string OpcodeStr, ValueType Ty>
2134 : N2V<0b11, 0b11, size, 0b01, 0b00111, 1, 0, (outs QPR:$dst), (ins QPR:$src),
2135 IIC_VSHLiD, !strconcat(OpcodeStr, "\t$dst, $src"), "",
2136 [(set QPR:$dst, (Ty (vneg QPR:$src)))]>;
2138 // VNEG : Vector Negate
2139 def VNEGs8d : VNEGD<0b00, "vneg.s8", v8i8>;
2140 def VNEGs16d : VNEGD<0b01, "vneg.s16", v4i16>;
2141 def VNEGs32d : VNEGD<0b10, "vneg.s32", v2i32>;
2142 def VNEGs8q : VNEGQ<0b00, "vneg.s8", v16i8>;
2143 def VNEGs16q : VNEGQ<0b01, "vneg.s16", v8i16>;
2144 def VNEGs32q : VNEGQ<0b10, "vneg.s32", v4i32>;
2146 // VNEG : Vector Negate (floating-point)
2147 def VNEGf32d : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
2148 (outs DPR:$dst), (ins DPR:$src), IIC_VUNAD,
2149 "vneg.f32\t$dst, $src", "",
2150 [(set DPR:$dst, (v2f32 (fneg DPR:$src)))]>;
2151 def VNEGf32q : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 1, 0,
2152 (outs QPR:$dst), (ins QPR:$src), IIC_VUNAQ,
2153 "vneg.f32\t$dst, $src", "",
2154 [(set QPR:$dst, (v4f32 (fneg QPR:$src)))]>;
2156 def : Pat<(v8i8 (vneg_conv DPR:$src)), (VNEGs8d DPR:$src)>;
2157 def : Pat<(v4i16 (vneg_conv DPR:$src)), (VNEGs16d DPR:$src)>;
2158 def : Pat<(v2i32 (vneg_conv DPR:$src)), (VNEGs32d DPR:$src)>;
2159 def : Pat<(v16i8 (vneg_conv QPR:$src)), (VNEGs8q QPR:$src)>;
2160 def : Pat<(v8i16 (vneg_conv QPR:$src)), (VNEGs16q QPR:$src)>;
2161 def : Pat<(v4i32 (vneg_conv QPR:$src)), (VNEGs32q QPR:$src)>;
2163 // VQNEG : Vector Saturating Negate
2164 defm VQNEG : N2VInt_QHS<0b11, 0b11, 0b00, 0b01111, 0,
2165 IIC_VQUNAiD, IIC_VQUNAiQ, "vqneg.s",
2166 int_arm_neon_vqneg>;
2168 // Vector Bit Counting Operations.
2170 // VCLS : Vector Count Leading Sign Bits
2171 defm VCLS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01000, 0,
2172 IIC_VCNTiD, IIC_VCNTiQ, "vcls.s",
2174 // VCLZ : Vector Count Leading Zeros
2175 defm VCLZ : N2VInt_QHS<0b11, 0b11, 0b00, 0b01001, 0,
2176 IIC_VCNTiD, IIC_VCNTiQ, "vclz.i",
2178 // VCNT : Vector Count One Bits
2179 def VCNTd : N2VDInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
2180 IIC_VCNTiD, "vcnt.8",
2181 v8i8, v8i8, int_arm_neon_vcnt>;
2182 def VCNTq : N2VQInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
2183 IIC_VCNTiQ, "vcnt.8",
2184 v16i8, v16i8, int_arm_neon_vcnt>;
2186 // Vector Move Operations.
2188 // VMOV : Vector Move (Register)
2190 def VMOVD : N3V<0, 0, 0b10, 0b0001, 0, 1, (outs DPR:$dst), (ins DPR:$src),
2191 IIC_VMOVD, "vmov\t$dst, $src", "", []>;
2192 def VMOVQ : N3V<0, 0, 0b10, 0b0001, 1, 1, (outs QPR:$dst), (ins QPR:$src),
2193 IIC_VMOVD, "vmov\t$dst, $src", "", []>;
2195 // VMOV : Vector Move (Immediate)
2197 // VMOV_get_imm8 xform function: convert build_vector to VMOV.i8 imm.
2198 def VMOV_get_imm8 : SDNodeXForm<build_vector, [{
2199 return ARM::getVMOVImm(N, 1, *CurDAG);
2201 def vmovImm8 : PatLeaf<(build_vector), [{
2202 return ARM::getVMOVImm(N, 1, *CurDAG).getNode() != 0;
2205 // VMOV_get_imm16 xform function: convert build_vector to VMOV.i16 imm.
2206 def VMOV_get_imm16 : SDNodeXForm<build_vector, [{
2207 return ARM::getVMOVImm(N, 2, *CurDAG);
2209 def vmovImm16 : PatLeaf<(build_vector), [{
2210 return ARM::getVMOVImm(N, 2, *CurDAG).getNode() != 0;
2211 }], VMOV_get_imm16>;
2213 // VMOV_get_imm32 xform function: convert build_vector to VMOV.i32 imm.
2214 def VMOV_get_imm32 : SDNodeXForm<build_vector, [{
2215 return ARM::getVMOVImm(N, 4, *CurDAG);
2217 def vmovImm32 : PatLeaf<(build_vector), [{
2218 return ARM::getVMOVImm(N, 4, *CurDAG).getNode() != 0;
2219 }], VMOV_get_imm32>;
2221 // VMOV_get_imm64 xform function: convert build_vector to VMOV.i64 imm.
2222 def VMOV_get_imm64 : SDNodeXForm<build_vector, [{
2223 return ARM::getVMOVImm(N, 8, *CurDAG);
2225 def vmovImm64 : PatLeaf<(build_vector), [{
2226 return ARM::getVMOVImm(N, 8, *CurDAG).getNode() != 0;
2227 }], VMOV_get_imm64>;
2229 // Note: Some of the cmode bits in the following VMOV instructions need to
2230 // be encoded based on the immed values.
2232 def VMOVv8i8 : N1ModImm<1, 0b000, 0b1110, 0, 0, 0, 1, (outs DPR:$dst),
2233 (ins i8imm:$SIMM), IIC_VMOVImm,
2234 "vmov.i8\t$dst, $SIMM", "",
2235 [(set DPR:$dst, (v8i8 vmovImm8:$SIMM))]>;
2236 def VMOVv16i8 : N1ModImm<1, 0b000, 0b1110, 0, 1, 0, 1, (outs QPR:$dst),
2237 (ins i8imm:$SIMM), IIC_VMOVImm,
2238 "vmov.i8\t$dst, $SIMM", "",
2239 [(set QPR:$dst, (v16i8 vmovImm8:$SIMM))]>;
2241 def VMOVv4i16 : N1ModImm<1, 0b000, 0b1000, 0, 0, 0, 1, (outs DPR:$dst),
2242 (ins i16imm:$SIMM), IIC_VMOVImm,
2243 "vmov.i16\t$dst, $SIMM", "",
2244 [(set DPR:$dst, (v4i16 vmovImm16:$SIMM))]>;
2245 def VMOVv8i16 : N1ModImm<1, 0b000, 0b1000, 0, 1, 0, 1, (outs QPR:$dst),
2246 (ins i16imm:$SIMM), IIC_VMOVImm,
2247 "vmov.i16\t$dst, $SIMM", "",
2248 [(set QPR:$dst, (v8i16 vmovImm16:$SIMM))]>;
2250 def VMOVv2i32 : N1ModImm<1, 0b000, 0b0000, 0, 0, 0, 1, (outs DPR:$dst),
2251 (ins i32imm:$SIMM), IIC_VMOVImm,
2252 "vmov.i32\t$dst, $SIMM", "",
2253 [(set DPR:$dst, (v2i32 vmovImm32:$SIMM))]>;
2254 def VMOVv4i32 : N1ModImm<1, 0b000, 0b0000, 0, 1, 0, 1, (outs QPR:$dst),
2255 (ins i32imm:$SIMM), IIC_VMOVImm,
2256 "vmov.i32\t$dst, $SIMM", "",
2257 [(set QPR:$dst, (v4i32 vmovImm32:$SIMM))]>;
2259 def VMOVv1i64 : N1ModImm<1, 0b000, 0b1110, 0, 0, 1, 1, (outs DPR:$dst),
2260 (ins i64imm:$SIMM), IIC_VMOVImm,
2261 "vmov.i64\t$dst, $SIMM", "",
2262 [(set DPR:$dst, (v1i64 vmovImm64:$SIMM))]>;
2263 def VMOVv2i64 : N1ModImm<1, 0b000, 0b1110, 0, 1, 1, 1, (outs QPR:$dst),
2264 (ins i64imm:$SIMM), IIC_VMOVImm,
2265 "vmov.i64\t$dst, $SIMM", "",
2266 [(set QPR:$dst, (v2i64 vmovImm64:$SIMM))]>;
2268 // VMOV : Vector Get Lane (move scalar to ARM core register)
2270 def VGETLNs8 : NVGetLane<0b11100101, 0b1011, 0b00,
2271 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2272 IIC_VMOVSI, "vmov", ".s8\t$dst, $src[$lane]",
2273 [(set GPR:$dst, (NEONvgetlanes (v8i8 DPR:$src),
2275 def VGETLNs16 : NVGetLane<0b11100001, 0b1011, 0b01,
2276 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2277 IIC_VMOVSI, "vmov", ".s16\t$dst, $src[$lane]",
2278 [(set GPR:$dst, (NEONvgetlanes (v4i16 DPR:$src),
2280 def VGETLNu8 : NVGetLane<0b11101101, 0b1011, 0b00,
2281 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2282 IIC_VMOVSI, "vmov", ".u8\t$dst, $src[$lane]",
2283 [(set GPR:$dst, (NEONvgetlaneu (v8i8 DPR:$src),
2285 def VGETLNu16 : NVGetLane<0b11101001, 0b1011, 0b01,
2286 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2287 IIC_VMOVSI, "vmov", ".u16\t$dst, $src[$lane]",
2288 [(set GPR:$dst, (NEONvgetlaneu (v4i16 DPR:$src),
2290 def VGETLNi32 : NVGetLane<0b11100001, 0b1011, 0b00,
2291 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2292 IIC_VMOVSI, "vmov", ".32\t$dst, $src[$lane]",
2293 [(set GPR:$dst, (extractelt (v2i32 DPR:$src),
2295 // def VGETLNf32: see FMRDH and FMRDL in ARMInstrVFP.td
2296 def : Pat<(NEONvgetlanes (v16i8 QPR:$src), imm:$lane),
2297 (VGETLNs8 (v8i8 (EXTRACT_SUBREG QPR:$src,
2298 (DSubReg_i8_reg imm:$lane))),
2299 (SubReg_i8_lane imm:$lane))>;
2300 def : Pat<(NEONvgetlanes (v8i16 QPR:$src), imm:$lane),
2301 (VGETLNs16 (v4i16 (EXTRACT_SUBREG QPR:$src,
2302 (DSubReg_i16_reg imm:$lane))),
2303 (SubReg_i16_lane imm:$lane))>;
2304 def : Pat<(NEONvgetlaneu (v16i8 QPR:$src), imm:$lane),
2305 (VGETLNu8 (v8i8 (EXTRACT_SUBREG QPR:$src,
2306 (DSubReg_i8_reg imm:$lane))),
2307 (SubReg_i8_lane imm:$lane))>;
2308 def : Pat<(NEONvgetlaneu (v8i16 QPR:$src), imm:$lane),
2309 (VGETLNu16 (v4i16 (EXTRACT_SUBREG QPR:$src,
2310 (DSubReg_i16_reg imm:$lane))),
2311 (SubReg_i16_lane imm:$lane))>;
2312 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
2313 (VGETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src,
2314 (DSubReg_i32_reg imm:$lane))),
2315 (SubReg_i32_lane imm:$lane))>;
2316 def : Pat<(extractelt (v2f32 DPR:$src1), imm:$src2),
2317 (EXTRACT_SUBREG (COPY_TO_REGCLASS DPR:$src1, DPR_VFP2),
2318 (SSubReg_f32_reg imm:$src2))>;
2319 def : Pat<(extractelt (v4f32 QPR:$src1), imm:$src2),
2320 (EXTRACT_SUBREG (COPY_TO_REGCLASS QPR:$src1, QPR_VFP2),
2321 (SSubReg_f32_reg imm:$src2))>;
2322 //def : Pat<(extractelt (v2i64 QPR:$src1), imm:$src2),
2323 // (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
2324 def : Pat<(extractelt (v2f64 QPR:$src1), imm:$src2),
2325 (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
2328 // VMOV : Vector Set Lane (move ARM core register to scalar)
2330 let Constraints = "$src1 = $dst" in {
2331 def VSETLNi8 : NVSetLane<0b11100100, 0b1011, 0b00, (outs DPR:$dst),
2332 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
2333 IIC_VMOVISL, "vmov", ".8\t$dst[$lane], $src2",
2334 [(set DPR:$dst, (vector_insert (v8i8 DPR:$src1),
2335 GPR:$src2, imm:$lane))]>;
2336 def VSETLNi16 : NVSetLane<0b11100000, 0b1011, 0b01, (outs DPR:$dst),
2337 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
2338 IIC_VMOVISL, "vmov", ".16\t$dst[$lane], $src2",
2339 [(set DPR:$dst, (vector_insert (v4i16 DPR:$src1),
2340 GPR:$src2, imm:$lane))]>;
2341 def VSETLNi32 : NVSetLane<0b11100000, 0b1011, 0b00, (outs DPR:$dst),
2342 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
2343 IIC_VMOVISL, "vmov", ".32\t$dst[$lane], $src2",
2344 [(set DPR:$dst, (insertelt (v2i32 DPR:$src1),
2345 GPR:$src2, imm:$lane))]>;
2347 def : Pat<(vector_insert (v16i8 QPR:$src1), GPR:$src2, imm:$lane),
2348 (v16i8 (INSERT_SUBREG QPR:$src1,
2349 (VSETLNi8 (v8i8 (EXTRACT_SUBREG QPR:$src1,
2350 (DSubReg_i8_reg imm:$lane))),
2351 GPR:$src2, (SubReg_i8_lane imm:$lane)),
2352 (DSubReg_i8_reg imm:$lane)))>;
2353 def : Pat<(vector_insert (v8i16 QPR:$src1), GPR:$src2, imm:$lane),
2354 (v8i16 (INSERT_SUBREG QPR:$src1,
2355 (VSETLNi16 (v4i16 (EXTRACT_SUBREG QPR:$src1,
2356 (DSubReg_i16_reg imm:$lane))),
2357 GPR:$src2, (SubReg_i16_lane imm:$lane)),
2358 (DSubReg_i16_reg imm:$lane)))>;
2359 def : Pat<(insertelt (v4i32 QPR:$src1), GPR:$src2, imm:$lane),
2360 (v4i32 (INSERT_SUBREG QPR:$src1,
2361 (VSETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src1,
2362 (DSubReg_i32_reg imm:$lane))),
2363 GPR:$src2, (SubReg_i32_lane imm:$lane)),
2364 (DSubReg_i32_reg imm:$lane)))>;
2366 def : Pat<(v2f32 (insertelt DPR:$src1, SPR:$src2, imm:$src3)),
2367 (INSERT_SUBREG (COPY_TO_REGCLASS DPR:$src1, DPR_VFP2),
2368 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
2369 def : Pat<(v4f32 (insertelt QPR:$src1, SPR:$src2, imm:$src3)),
2370 (INSERT_SUBREG (COPY_TO_REGCLASS QPR:$src1, QPR_VFP2),
2371 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
2373 //def : Pat<(v2i64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
2374 // (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
2375 def : Pat<(v2f64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
2376 (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
2378 def : Pat<(v2f32 (scalar_to_vector SPR:$src)),
2379 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$src, arm_ssubreg_0)>;
2380 def : Pat<(v2f64 (scalar_to_vector DPR:$src)),
2381 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), DPR:$src, arm_dsubreg_0)>;
2382 def : Pat<(v4f32 (scalar_to_vector SPR:$src)),
2383 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), SPR:$src, arm_ssubreg_0)>;
2385 def : Pat<(v8i8 (scalar_to_vector GPR:$src)),
2386 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
2387 def : Pat<(v4i16 (scalar_to_vector GPR:$src)),
2388 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
2389 def : Pat<(v2i32 (scalar_to_vector GPR:$src)),
2390 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
2392 def : Pat<(v16i8 (scalar_to_vector GPR:$src)),
2393 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
2394 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
2396 def : Pat<(v8i16 (scalar_to_vector GPR:$src)),
2397 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
2398 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
2400 def : Pat<(v4i32 (scalar_to_vector GPR:$src)),
2401 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
2402 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
2405 // VDUP : Vector Duplicate (from ARM core register to all elements)
2407 class VDUPD<bits<8> opcod1, bits<2> opcod3, string asmSize, ValueType Ty>
2408 : NVDup<opcod1, 0b1011, opcod3, (outs DPR:$dst), (ins GPR:$src),
2409 IIC_VMOVIS, "vdup", !strconcat(asmSize, "\t$dst, $src"),
2410 [(set DPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
2411 class VDUPQ<bits<8> opcod1, bits<2> opcod3, string asmSize, ValueType Ty>
2412 : NVDup<opcod1, 0b1011, opcod3, (outs QPR:$dst), (ins GPR:$src),
2413 IIC_VMOVIS, "vdup", !strconcat(asmSize, "\t$dst, $src"),
2414 [(set QPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
2416 def VDUP8d : VDUPD<0b11101100, 0b00, ".8", v8i8>;
2417 def VDUP16d : VDUPD<0b11101000, 0b01, ".16", v4i16>;
2418 def VDUP32d : VDUPD<0b11101000, 0b00, ".32", v2i32>;
2419 def VDUP8q : VDUPQ<0b11101110, 0b00, ".8", v16i8>;
2420 def VDUP16q : VDUPQ<0b11101010, 0b01, ".16", v8i16>;
2421 def VDUP32q : VDUPQ<0b11101010, 0b00, ".32", v4i32>;
2423 def VDUPfd : NVDup<0b11101000, 0b1011, 0b00, (outs DPR:$dst), (ins GPR:$src),
2424 IIC_VMOVIS, "vdup", ".32\t$dst, $src",
2425 [(set DPR:$dst, (v2f32 (NEONvdup
2426 (f32 (bitconvert GPR:$src)))))]>;
2427 def VDUPfq : NVDup<0b11101010, 0b1011, 0b00, (outs QPR:$dst), (ins GPR:$src),
2428 IIC_VMOVIS, "vdup", ".32\t$dst, $src",
2429 [(set QPR:$dst, (v4f32 (NEONvdup
2430 (f32 (bitconvert GPR:$src)))))]>;
2432 // VDUP : Vector Duplicate Lane (from scalar to all elements)
2434 class VDUPLND<bits<2> op19_18, bits<2> op17_16, string OpcodeStr, ValueType Ty>
2435 : N2V<0b11, 0b11, op19_18, op17_16, 0b11000, 0, 0,
2436 (outs DPR:$dst), (ins DPR:$src, nohash_imm:$lane), IIC_VMOVD,
2437 !strconcat(OpcodeStr, "\t$dst, $src[$lane]"), "",
2438 [(set DPR:$dst, (Ty (NEONvduplane (Ty DPR:$src), imm:$lane)))]>;
2440 class VDUPLNQ<bits<2> op19_18, bits<2> op17_16, string OpcodeStr,
2441 ValueType ResTy, ValueType OpTy>
2442 : N2V<0b11, 0b11, op19_18, op17_16, 0b11000, 1, 0,
2443 (outs QPR:$dst), (ins DPR:$src, nohash_imm:$lane), IIC_VMOVD,
2444 !strconcat(OpcodeStr, "\t$dst, $src[$lane]"), "",
2445 [(set QPR:$dst, (ResTy (NEONvduplane (OpTy DPR:$src), imm:$lane)))]>;
2447 def VDUPLN8d : VDUPLND<0b00, 0b01, "vdup.8", v8i8>;
2448 def VDUPLN16d : VDUPLND<0b00, 0b10, "vdup.16", v4i16>;
2449 def VDUPLN32d : VDUPLND<0b01, 0b00, "vdup.32", v2i32>;
2450 def VDUPLNfd : VDUPLND<0b01, 0b00, "vdup.32", v2f32>;
2451 def VDUPLN8q : VDUPLNQ<0b00, 0b01, "vdup.8", v16i8, v8i8>;
2452 def VDUPLN16q : VDUPLNQ<0b00, 0b10, "vdup.16", v8i16, v4i16>;
2453 def VDUPLN32q : VDUPLNQ<0b01, 0b00, "vdup.32", v4i32, v2i32>;
2454 def VDUPLNfq : VDUPLNQ<0b01, 0b00, "vdup.32", v4f32, v2f32>;
2456 def : Pat<(v16i8 (NEONvduplane (v16i8 QPR:$src), imm:$lane)),
2457 (v16i8 (VDUPLN8q (v8i8 (EXTRACT_SUBREG QPR:$src,
2458 (DSubReg_i8_reg imm:$lane))),
2459 (SubReg_i8_lane imm:$lane)))>;
2460 def : Pat<(v8i16 (NEONvduplane (v8i16 QPR:$src), imm:$lane)),
2461 (v8i16 (VDUPLN16q (v4i16 (EXTRACT_SUBREG QPR:$src,
2462 (DSubReg_i16_reg imm:$lane))),
2463 (SubReg_i16_lane imm:$lane)))>;
2464 def : Pat<(v4i32 (NEONvduplane (v4i32 QPR:$src), imm:$lane)),
2465 (v4i32 (VDUPLN32q (v2i32 (EXTRACT_SUBREG QPR:$src,
2466 (DSubReg_i32_reg imm:$lane))),
2467 (SubReg_i32_lane imm:$lane)))>;
2468 def : Pat<(v4f32 (NEONvduplane (v4f32 QPR:$src), imm:$lane)),
2469 (v4f32 (VDUPLNfq (v2f32 (EXTRACT_SUBREG QPR:$src,
2470 (DSubReg_i32_reg imm:$lane))),
2471 (SubReg_i32_lane imm:$lane)))>;
2473 def VDUPfdf : N2V<0b11, 0b11, 0b01, 0b00, 0b11000, 0, 0,
2474 (outs DPR:$dst), (ins SPR:$src),
2475 IIC_VMOVD, "vdup.32\t$dst, ${src:lane}", "",
2476 [(set DPR:$dst, (v2f32 (NEONvdup (f32 SPR:$src))))]>;
2478 def VDUPfqf : N2V<0b11, 0b11, 0b01, 0b00, 0b11000, 1, 0,
2479 (outs QPR:$dst), (ins SPR:$src),
2480 IIC_VMOVD, "vdup.32\t$dst, ${src:lane}", "",
2481 [(set QPR:$dst, (v4f32 (NEONvdup (f32 SPR:$src))))]>;
2483 def : Pat<(v2i64 (NEONvduplane (v2i64 QPR:$src), imm:$lane)),
2484 (INSERT_SUBREG QPR:$src,
2485 (i64 (EXTRACT_SUBREG QPR:$src, (DSubReg_f64_reg imm:$lane))),
2486 (DSubReg_f64_other_reg imm:$lane))>;
2487 def : Pat<(v2f64 (NEONvduplane (v2f64 QPR:$src), imm:$lane)),
2488 (INSERT_SUBREG QPR:$src,
2489 (f64 (EXTRACT_SUBREG QPR:$src, (DSubReg_f64_reg imm:$lane))),
2490 (DSubReg_f64_other_reg imm:$lane))>;
2492 // VMOVN : Vector Narrowing Move
2493 defm VMOVN : N2VNInt_HSD<0b11,0b11,0b10,0b00100,0,0, IIC_VMOVD, "vmovn.i",
2494 int_arm_neon_vmovn>;
2495 // VQMOVN : Vector Saturating Narrowing Move
2496 defm VQMOVNs : N2VNInt_HSD<0b11,0b11,0b10,0b00101,0,0, IIC_VQUNAiD, "vqmovn.s",
2497 int_arm_neon_vqmovns>;
2498 defm VQMOVNu : N2VNInt_HSD<0b11,0b11,0b10,0b00101,1,0, IIC_VQUNAiD, "vqmovn.u",
2499 int_arm_neon_vqmovnu>;
2500 defm VQMOVNsu : N2VNInt_HSD<0b11,0b11,0b10,0b00100,1,0, IIC_VQUNAiD, "vqmovun.s",
2501 int_arm_neon_vqmovnsu>;
2502 // VMOVL : Vector Lengthening Move
2503 defm VMOVLs : N2VLInt_QHS<0,1,0b1010,0,0,1, "vmovl.s", int_arm_neon_vmovls>;
2504 defm VMOVLu : N2VLInt_QHS<1,1,0b1010,0,0,1, "vmovl.u", int_arm_neon_vmovlu>;
2506 // Vector Conversions.
2508 // VCVT : Vector Convert Between Floating-Point and Integers
2509 def VCVTf2sd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt.s32.f32",
2510 v2i32, v2f32, fp_to_sint>;
2511 def VCVTf2ud : N2VD<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt.u32.f32",
2512 v2i32, v2f32, fp_to_uint>;
2513 def VCVTs2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt.f32.s32",
2514 v2f32, v2i32, sint_to_fp>;
2515 def VCVTu2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt.f32.u32",
2516 v2f32, v2i32, uint_to_fp>;
2518 def VCVTf2sq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt.s32.f32",
2519 v4i32, v4f32, fp_to_sint>;
2520 def VCVTf2uq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt.u32.f32",
2521 v4i32, v4f32, fp_to_uint>;
2522 def VCVTs2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt.f32.s32",
2523 v4f32, v4i32, sint_to_fp>;
2524 def VCVTu2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt.f32.u32",
2525 v4f32, v4i32, uint_to_fp>;
2527 // VCVT : Vector Convert Between Floating-Point and Fixed-Point.
2528 // Note: Some of the opcode bits in the following VCVT instructions need to
2529 // be encoded based on the immed values.
2530 def VCVTf2xsd : N2VCvtD<0, 1, 0b000000, 0b1111, 0, 1, "vcvt.s32.f32",
2531 v2i32, v2f32, int_arm_neon_vcvtfp2fxs>;
2532 def VCVTf2xud : N2VCvtD<1, 1, 0b000000, 0b1111, 0, 1, "vcvt.u32.f32",
2533 v2i32, v2f32, int_arm_neon_vcvtfp2fxu>;
2534 def VCVTxs2fd : N2VCvtD<0, 1, 0b000000, 0b1110, 0, 1, "vcvt.f32.s32",
2535 v2f32, v2i32, int_arm_neon_vcvtfxs2fp>;
2536 def VCVTxu2fd : N2VCvtD<1, 1, 0b000000, 0b1110, 0, 1, "vcvt.f32.u32",
2537 v2f32, v2i32, int_arm_neon_vcvtfxu2fp>;
2539 def VCVTf2xsq : N2VCvtQ<0, 1, 0b000000, 0b1111, 0, 1, "vcvt.s32.f32",
2540 v4i32, v4f32, int_arm_neon_vcvtfp2fxs>;
2541 def VCVTf2xuq : N2VCvtQ<1, 1, 0b000000, 0b1111, 0, 1, "vcvt.u32.f32",
2542 v4i32, v4f32, int_arm_neon_vcvtfp2fxu>;
2543 def VCVTxs2fq : N2VCvtQ<0, 1, 0b000000, 0b1110, 0, 1, "vcvt.f32.s32",
2544 v4f32, v4i32, int_arm_neon_vcvtfxs2fp>;
2545 def VCVTxu2fq : N2VCvtQ<1, 1, 0b000000, 0b1110, 0, 1, "vcvt.f32.u32",
2546 v4f32, v4i32, int_arm_neon_vcvtfxu2fp>;
2550 // VREV64 : Vector Reverse elements within 64-bit doublewords
2552 class VREV64D<bits<2> op19_18, string OpcodeStr, ValueType Ty>
2553 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 0, 0, (outs DPR:$dst),
2554 (ins DPR:$src), IIC_VMOVD,
2555 !strconcat(OpcodeStr, "\t$dst, $src"), "",
2556 [(set DPR:$dst, (Ty (NEONvrev64 (Ty DPR:$src))))]>;
2557 class VREV64Q<bits<2> op19_18, string OpcodeStr, ValueType Ty>
2558 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 1, 0, (outs QPR:$dst),
2559 (ins QPR:$src), IIC_VMOVD,
2560 !strconcat(OpcodeStr, "\t$dst, $src"), "",
2561 [(set QPR:$dst, (Ty (NEONvrev64 (Ty QPR:$src))))]>;
2563 def VREV64d8 : VREV64D<0b00, "vrev64.8", v8i8>;
2564 def VREV64d16 : VREV64D<0b01, "vrev64.16", v4i16>;
2565 def VREV64d32 : VREV64D<0b10, "vrev64.32", v2i32>;
2566 def VREV64df : VREV64D<0b10, "vrev64.32", v2f32>;
2568 def VREV64q8 : VREV64Q<0b00, "vrev64.8", v16i8>;
2569 def VREV64q16 : VREV64Q<0b01, "vrev64.16", v8i16>;
2570 def VREV64q32 : VREV64Q<0b10, "vrev64.32", v4i32>;
2571 def VREV64qf : VREV64Q<0b10, "vrev64.32", v4f32>;
2573 // VREV32 : Vector Reverse elements within 32-bit words
2575 class VREV32D<bits<2> op19_18, string OpcodeStr, ValueType Ty>
2576 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 0, 0, (outs DPR:$dst),
2577 (ins DPR:$src), IIC_VMOVD,
2578 !strconcat(OpcodeStr, "\t$dst, $src"), "",
2579 [(set DPR:$dst, (Ty (NEONvrev32 (Ty DPR:$src))))]>;
2580 class VREV32Q<bits<2> op19_18, string OpcodeStr, ValueType Ty>
2581 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 1, 0, (outs QPR:$dst),
2582 (ins QPR:$src), IIC_VMOVD,
2583 !strconcat(OpcodeStr, "\t$dst, $src"), "",
2584 [(set QPR:$dst, (Ty (NEONvrev32 (Ty QPR:$src))))]>;
2586 def VREV32d8 : VREV32D<0b00, "vrev32.8", v8i8>;
2587 def VREV32d16 : VREV32D<0b01, "vrev32.16", v4i16>;
2589 def VREV32q8 : VREV32Q<0b00, "vrev32.8", v16i8>;
2590 def VREV32q16 : VREV32Q<0b01, "vrev32.16", v8i16>;
2592 // VREV16 : Vector Reverse elements within 16-bit halfwords
2594 class VREV16D<bits<2> op19_18, string OpcodeStr, ValueType Ty>
2595 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 0, 0, (outs DPR:$dst),
2596 (ins DPR:$src), IIC_VMOVD,
2597 !strconcat(OpcodeStr, "\t$dst, $src"), "",
2598 [(set DPR:$dst, (Ty (NEONvrev16 (Ty DPR:$src))))]>;
2599 class VREV16Q<bits<2> op19_18, string OpcodeStr, ValueType Ty>
2600 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 1, 0, (outs QPR:$dst),
2601 (ins QPR:$src), IIC_VMOVD,
2602 !strconcat(OpcodeStr, "\t$dst, $src"), "",
2603 [(set QPR:$dst, (Ty (NEONvrev16 (Ty QPR:$src))))]>;
2605 def VREV16d8 : VREV16D<0b00, "vrev16.8", v8i8>;
2606 def VREV16q8 : VREV16Q<0b00, "vrev16.8", v16i8>;
2608 // Other Vector Shuffles.
2610 // VEXT : Vector Extract
2612 class VEXTd<string OpcodeStr, ValueType Ty>
2613 : N3V<0,1,0b11,0b0000,0,0, (outs DPR:$dst),
2614 (ins DPR:$lhs, DPR:$rhs, i32imm:$index), IIC_VEXTD,
2615 !strconcat(OpcodeStr, "\t$dst, $lhs, $rhs, $index"), "",
2616 [(set DPR:$dst, (Ty (NEONvext (Ty DPR:$lhs),
2617 (Ty DPR:$rhs), imm:$index)))]>;
2619 class VEXTq<string OpcodeStr, ValueType Ty>
2620 : N3V<0,1,0b11,0b0000,1,0, (outs QPR:$dst),
2621 (ins QPR:$lhs, QPR:$rhs, i32imm:$index), IIC_VEXTQ,
2622 !strconcat(OpcodeStr, "\t$dst, $lhs, $rhs, $index"), "",
2623 [(set QPR:$dst, (Ty (NEONvext (Ty QPR:$lhs),
2624 (Ty QPR:$rhs), imm:$index)))]>;
2626 def VEXTd8 : VEXTd<"vext.8", v8i8>;
2627 def VEXTd16 : VEXTd<"vext.16", v4i16>;
2628 def VEXTd32 : VEXTd<"vext.32", v2i32>;
2629 def VEXTdf : VEXTd<"vext.32", v2f32>;
2631 def VEXTq8 : VEXTq<"vext.8", v16i8>;
2632 def VEXTq16 : VEXTq<"vext.16", v8i16>;
2633 def VEXTq32 : VEXTq<"vext.32", v4i32>;
2634 def VEXTqf : VEXTq<"vext.32", v4f32>;
2636 // VTRN : Vector Transpose
2638 def VTRNd8 : N2VDShuffle<0b00, 0b00001, "vtrn.8">;
2639 def VTRNd16 : N2VDShuffle<0b01, 0b00001, "vtrn.16">;
2640 def VTRNd32 : N2VDShuffle<0b10, 0b00001, "vtrn.32">;
2642 def VTRNq8 : N2VQShuffle<0b00, 0b00001, IIC_VPERMQ, "vtrn.8">;
2643 def VTRNq16 : N2VQShuffle<0b01, 0b00001, IIC_VPERMQ, "vtrn.16">;
2644 def VTRNq32 : N2VQShuffle<0b10, 0b00001, IIC_VPERMQ, "vtrn.32">;
2646 // VUZP : Vector Unzip (Deinterleave)
2648 def VUZPd8 : N2VDShuffle<0b00, 0b00010, "vuzp.8">;
2649 def VUZPd16 : N2VDShuffle<0b01, 0b00010, "vuzp.16">;
2650 def VUZPd32 : N2VDShuffle<0b10, 0b00010, "vuzp.32">;
2652 def VUZPq8 : N2VQShuffle<0b00, 0b00010, IIC_VPERMQ3, "vuzp.8">;
2653 def VUZPq16 : N2VQShuffle<0b01, 0b00010, IIC_VPERMQ3, "vuzp.16">;
2654 def VUZPq32 : N2VQShuffle<0b10, 0b00010, IIC_VPERMQ3, "vuzp.32">;
2656 // VZIP : Vector Zip (Interleave)
2658 def VZIPd8 : N2VDShuffle<0b00, 0b00011, "vzip.8">;
2659 def VZIPd16 : N2VDShuffle<0b01, 0b00011, "vzip.16">;
2660 def VZIPd32 : N2VDShuffle<0b10, 0b00011, "vzip.32">;
2662 def VZIPq8 : N2VQShuffle<0b00, 0b00011, IIC_VPERMQ3, "vzip.8">;
2663 def VZIPq16 : N2VQShuffle<0b01, 0b00011, IIC_VPERMQ3, "vzip.16">;
2664 def VZIPq32 : N2VQShuffle<0b10, 0b00011, IIC_VPERMQ3, "vzip.32">;
2666 // Vector Table Lookup and Table Extension.
2668 // VTBL : Vector Table Lookup
2670 : N3V<1,1,0b11,0b1000,0,0, (outs DPR:$dst),
2671 (ins DPR:$tbl1, DPR:$src), IIC_VTB1,
2672 "vtbl.8\t$dst, \\{$tbl1\\}, $src", "",
2673 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl1 DPR:$tbl1, DPR:$src)))]>;
2674 let hasExtraSrcRegAllocReq = 1 in {
2676 : N3V<1,1,0b11,0b1001,0,0, (outs DPR:$dst),
2677 (ins DPR:$tbl1, DPR:$tbl2, DPR:$src), IIC_VTB2,
2678 "vtbl.8\t$dst, \\{$tbl1,$tbl2\\}, $src", "",
2679 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl2
2680 DPR:$tbl1, DPR:$tbl2, DPR:$src)))]>;
2682 : N3V<1,1,0b11,0b1010,0,0, (outs DPR:$dst),
2683 (ins DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$src), IIC_VTB3,
2684 "vtbl.8\t$dst, \\{$tbl1,$tbl2,$tbl3\\}, $src", "",
2685 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl3
2686 DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$src)))]>;
2688 : N3V<1,1,0b11,0b1011,0,0, (outs DPR:$dst),
2689 (ins DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$src), IIC_VTB4,
2690 "vtbl.8\t$dst, \\{$tbl1,$tbl2,$tbl3,$tbl4\\}, $src", "",
2691 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl4 DPR:$tbl1, DPR:$tbl2,
2692 DPR:$tbl3, DPR:$tbl4, DPR:$src)))]>;
2693 } // hasExtraSrcRegAllocReq = 1
2695 // VTBX : Vector Table Extension
2697 : N3V<1,1,0b11,0b1000,1,0, (outs DPR:$dst),
2698 (ins DPR:$orig, DPR:$tbl1, DPR:$src), IIC_VTBX1,
2699 "vtbx.8\t$dst, \\{$tbl1\\}, $src", "$orig = $dst",
2700 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx1
2701 DPR:$orig, DPR:$tbl1, DPR:$src)))]>;
2702 let hasExtraSrcRegAllocReq = 1 in {
2704 : N3V<1,1,0b11,0b1001,1,0, (outs DPR:$dst),
2705 (ins DPR:$orig, DPR:$tbl1, DPR:$tbl2, DPR:$src), IIC_VTBX2,
2706 "vtbx.8\t$dst, \\{$tbl1,$tbl2\\}, $src", "$orig = $dst",
2707 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx2
2708 DPR:$orig, DPR:$tbl1, DPR:$tbl2, DPR:$src)))]>;
2710 : N3V<1,1,0b11,0b1010,1,0, (outs DPR:$dst),
2711 (ins DPR:$orig, DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$src), IIC_VTBX3,
2712 "vtbx.8\t$dst, \\{$tbl1,$tbl2,$tbl3\\}, $src", "$orig = $dst",
2713 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx3 DPR:$orig, DPR:$tbl1,
2714 DPR:$tbl2, DPR:$tbl3, DPR:$src)))]>;
2716 : N3V<1,1,0b11,0b1011,1,0, (outs DPR:$dst), (ins DPR:$orig, DPR:$tbl1,
2717 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$src), IIC_VTBX4,
2718 "vtbx.8\t$dst, \\{$tbl1,$tbl2,$tbl3,$tbl4\\}, $src", "$orig = $dst",
2719 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx4 DPR:$orig, DPR:$tbl1,
2720 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$src)))]>;
2721 } // hasExtraSrcRegAllocReq = 1
2723 //===----------------------------------------------------------------------===//
2724 // NEON instructions for single-precision FP math
2725 //===----------------------------------------------------------------------===//
2727 // These need separate instructions because they must use DPR_VFP2 register
2728 // class which have SPR sub-registers.
2730 // Vector Add Operations used for single-precision FP
2731 let neverHasSideEffects = 1 in
2732 def VADDfd_sfp : N3VDs<0, 0, 0b00, 0b1101, 0, "vadd.f32", v2f32, v2f32, fadd,1>;
2733 def : N3VDsPat<fadd, VADDfd_sfp>;
2735 // Vector Sub Operations used for single-precision FP
2736 let neverHasSideEffects = 1 in
2737 def VSUBfd_sfp : N3VDs<0, 0, 0b10, 0b1101, 0, "vsub.f32", v2f32, v2f32, fsub,0>;
2738 def : N3VDsPat<fsub, VSUBfd_sfp>;
2740 // Vector Multiply Operations used for single-precision FP
2741 let neverHasSideEffects = 1 in
2742 def VMULfd_sfp : N3VDs<1, 0, 0b00, 0b1101, 1, "vmul.f32", v2f32, v2f32, fmul,1>;
2743 def : N3VDsPat<fmul, VMULfd_sfp>;
2745 // Vector Multiply-Accumulate/Subtract used for single-precision FP
2746 let neverHasSideEffects = 1 in
2747 def VMLAfd_sfp : N3VDMulOps<0, 0, 0b00, 0b1101, 1, IIC_VMACD, "vmla.f32", v2f32,fmul,fadd>;
2748 def : N3VDMulOpsPat<fmul, fadd, VMLAfd_sfp>;
2750 let neverHasSideEffects = 1 in
2751 def VMLSfd_sfp : N3VDMulOps<0, 0, 0b10, 0b1101, 1, IIC_VMACD, "vmls.f32", v2f32,fmul,fsub>;
2752 def : N3VDMulOpsPat<fmul, fsub, VMLSfd_sfp>;
2754 // Vector Absolute used for single-precision FP
2755 let neverHasSideEffects = 1 in
2756 def VABSfd_sfp : N2VDInts<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
2757 IIC_VUNAD, "vabs.f32",
2758 v2f32, v2f32, int_arm_neon_vabs>;
2759 def : N2VDIntsPat<fabs, VABSfd_sfp>;
2761 // Vector Negate used for single-precision FP
2762 let neverHasSideEffects = 1 in
2763 def VNEGf32d_sfp : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
2764 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), IIC_VUNAD,
2765 "vneg.f32\t$dst, $src", "", []>;
2766 def : N2VDIntsPat<fneg, VNEGf32d_sfp>;
2768 // Vector Convert between single-precision FP and integer
2769 let neverHasSideEffects = 1 in
2770 def VCVTf2sd_sfp : N2VDs<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt.s32.f32",
2771 v2i32, v2f32, fp_to_sint>;
2772 def : N2VDsPat<arm_ftosi, f32, v2f32, VCVTf2sd_sfp>;
2774 let neverHasSideEffects = 1 in
2775 def VCVTf2ud_sfp : N2VDs<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt.u32.f32",
2776 v2i32, v2f32, fp_to_uint>;
2777 def : N2VDsPat<arm_ftoui, f32, v2f32, VCVTf2ud_sfp>;
2779 let neverHasSideEffects = 1 in
2780 def VCVTs2fd_sfp : N2VDs<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt.f32.s32",
2781 v2f32, v2i32, sint_to_fp>;
2782 def : N2VDsPat<arm_sitof, f32, v2i32, VCVTs2fd_sfp>;
2784 let neverHasSideEffects = 1 in
2785 def VCVTu2fd_sfp : N2VDs<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt.f32.u32",
2786 v2f32, v2i32, uint_to_fp>;
2787 def : N2VDsPat<arm_uitof, f32, v2i32, VCVTu2fd_sfp>;
2789 //===----------------------------------------------------------------------===//
2790 // Non-Instruction Patterns
2791 //===----------------------------------------------------------------------===//
2794 def : Pat<(v1i64 (bitconvert (v2i32 DPR:$src))), (v1i64 DPR:$src)>;
2795 def : Pat<(v1i64 (bitconvert (v4i16 DPR:$src))), (v1i64 DPR:$src)>;
2796 def : Pat<(v1i64 (bitconvert (v8i8 DPR:$src))), (v1i64 DPR:$src)>;
2797 def : Pat<(v1i64 (bitconvert (f64 DPR:$src))), (v1i64 DPR:$src)>;
2798 def : Pat<(v1i64 (bitconvert (v2f32 DPR:$src))), (v1i64 DPR:$src)>;
2799 def : Pat<(v2i32 (bitconvert (v1i64 DPR:$src))), (v2i32 DPR:$src)>;
2800 def : Pat<(v2i32 (bitconvert (v4i16 DPR:$src))), (v2i32 DPR:$src)>;
2801 def : Pat<(v2i32 (bitconvert (v8i8 DPR:$src))), (v2i32 DPR:$src)>;
2802 def : Pat<(v2i32 (bitconvert (f64 DPR:$src))), (v2i32 DPR:$src)>;
2803 def : Pat<(v2i32 (bitconvert (v2f32 DPR:$src))), (v2i32 DPR:$src)>;
2804 def : Pat<(v4i16 (bitconvert (v1i64 DPR:$src))), (v4i16 DPR:$src)>;
2805 def : Pat<(v4i16 (bitconvert (v2i32 DPR:$src))), (v4i16 DPR:$src)>;
2806 def : Pat<(v4i16 (bitconvert (v8i8 DPR:$src))), (v4i16 DPR:$src)>;
2807 def : Pat<(v4i16 (bitconvert (f64 DPR:$src))), (v4i16 DPR:$src)>;
2808 def : Pat<(v4i16 (bitconvert (v2f32 DPR:$src))), (v4i16 DPR:$src)>;
2809 def : Pat<(v8i8 (bitconvert (v1i64 DPR:$src))), (v8i8 DPR:$src)>;
2810 def : Pat<(v8i8 (bitconvert (v2i32 DPR:$src))), (v8i8 DPR:$src)>;
2811 def : Pat<(v8i8 (bitconvert (v4i16 DPR:$src))), (v8i8 DPR:$src)>;
2812 def : Pat<(v8i8 (bitconvert (f64 DPR:$src))), (v8i8 DPR:$src)>;
2813 def : Pat<(v8i8 (bitconvert (v2f32 DPR:$src))), (v8i8 DPR:$src)>;
2814 def : Pat<(f64 (bitconvert (v1i64 DPR:$src))), (f64 DPR:$src)>;
2815 def : Pat<(f64 (bitconvert (v2i32 DPR:$src))), (f64 DPR:$src)>;
2816 def : Pat<(f64 (bitconvert (v4i16 DPR:$src))), (f64 DPR:$src)>;
2817 def : Pat<(f64 (bitconvert (v8i8 DPR:$src))), (f64 DPR:$src)>;
2818 def : Pat<(f64 (bitconvert (v2f32 DPR:$src))), (f64 DPR:$src)>;
2819 def : Pat<(v2f32 (bitconvert (f64 DPR:$src))), (v2f32 DPR:$src)>;
2820 def : Pat<(v2f32 (bitconvert (v1i64 DPR:$src))), (v2f32 DPR:$src)>;
2821 def : Pat<(v2f32 (bitconvert (v2i32 DPR:$src))), (v2f32 DPR:$src)>;
2822 def : Pat<(v2f32 (bitconvert (v4i16 DPR:$src))), (v2f32 DPR:$src)>;
2823 def : Pat<(v2f32 (bitconvert (v8i8 DPR:$src))), (v2f32 DPR:$src)>;
2825 def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (v2i64 QPR:$src)>;
2826 def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (v2i64 QPR:$src)>;
2827 def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (v2i64 QPR:$src)>;
2828 def : Pat<(v2i64 (bitconvert (v2f64 QPR:$src))), (v2i64 QPR:$src)>;
2829 def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (v2i64 QPR:$src)>;
2830 def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (v4i32 QPR:$src)>;
2831 def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (v4i32 QPR:$src)>;
2832 def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (v4i32 QPR:$src)>;
2833 def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (v4i32 QPR:$src)>;
2834 def : Pat<(v4i32 (bitconvert (v4f32 QPR:$src))), (v4i32 QPR:$src)>;
2835 def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (v8i16 QPR:$src)>;
2836 def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (v8i16 QPR:$src)>;
2837 def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (v8i16 QPR:$src)>;
2838 def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (v8i16 QPR:$src)>;
2839 def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (v8i16 QPR:$src)>;
2840 def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (v16i8 QPR:$src)>;
2841 def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (v16i8 QPR:$src)>;
2842 def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (v16i8 QPR:$src)>;
2843 def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (v16i8 QPR:$src)>;
2844 def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (v16i8 QPR:$src)>;
2845 def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (v4f32 QPR:$src)>;
2846 def : Pat<(v4f32 (bitconvert (v4i32 QPR:$src))), (v4f32 QPR:$src)>;
2847 def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (v4f32 QPR:$src)>;
2848 def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (v4f32 QPR:$src)>;
2849 def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (v4f32 QPR:$src)>;
2850 def : Pat<(v2f64 (bitconvert (v2i64 QPR:$src))), (v2f64 QPR:$src)>;
2851 def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (v2f64 QPR:$src)>;
2852 def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 QPR:$src)>;
2853 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 QPR:$src)>;
2854 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 QPR:$src)>;