1 //===- ARMInstrNEON.td - NEON support for ARM -----------------------------===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file describes the ARM NEON instruction set.
12 //===----------------------------------------------------------------------===//
14 //===----------------------------------------------------------------------===//
15 // NEON-specific DAG Nodes.
16 //===----------------------------------------------------------------------===//
18 def SDTARMVCMP : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisSameAs<1, 2>]>;
20 def NEONvceq : SDNode<"ARMISD::VCEQ", SDTARMVCMP>;
21 def NEONvcge : SDNode<"ARMISD::VCGE", SDTARMVCMP>;
22 def NEONvcgeu : SDNode<"ARMISD::VCGEU", SDTARMVCMP>;
23 def NEONvcgt : SDNode<"ARMISD::VCGT", SDTARMVCMP>;
24 def NEONvcgtu : SDNode<"ARMISD::VCGTU", SDTARMVCMP>;
25 def NEONvtst : SDNode<"ARMISD::VTST", SDTARMVCMP>;
27 // Types for vector shift by immediates. The "SHX" version is for long and
28 // narrow operations where the source and destination vectors have different
29 // types. The "SHINS" version is for shift and insert operations.
30 def SDTARMVSH : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
32 def SDTARMVSHX : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
34 def SDTARMVSHINS : SDTypeProfile<1, 3, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
35 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
37 def NEONvshl : SDNode<"ARMISD::VSHL", SDTARMVSH>;
38 def NEONvshrs : SDNode<"ARMISD::VSHRs", SDTARMVSH>;
39 def NEONvshru : SDNode<"ARMISD::VSHRu", SDTARMVSH>;
40 def NEONvshlls : SDNode<"ARMISD::VSHLLs", SDTARMVSHX>;
41 def NEONvshllu : SDNode<"ARMISD::VSHLLu", SDTARMVSHX>;
42 def NEONvshlli : SDNode<"ARMISD::VSHLLi", SDTARMVSHX>;
43 def NEONvshrn : SDNode<"ARMISD::VSHRN", SDTARMVSHX>;
45 def NEONvrshrs : SDNode<"ARMISD::VRSHRs", SDTARMVSH>;
46 def NEONvrshru : SDNode<"ARMISD::VRSHRu", SDTARMVSH>;
47 def NEONvrshrn : SDNode<"ARMISD::VRSHRN", SDTARMVSHX>;
49 def NEONvqshls : SDNode<"ARMISD::VQSHLs", SDTARMVSH>;
50 def NEONvqshlu : SDNode<"ARMISD::VQSHLu", SDTARMVSH>;
51 def NEONvqshlsu : SDNode<"ARMISD::VQSHLsu", SDTARMVSH>;
52 def NEONvqshrns : SDNode<"ARMISD::VQSHRNs", SDTARMVSHX>;
53 def NEONvqshrnu : SDNode<"ARMISD::VQSHRNu", SDTARMVSHX>;
54 def NEONvqshrnsu : SDNode<"ARMISD::VQSHRNsu", SDTARMVSHX>;
56 def NEONvqrshrns : SDNode<"ARMISD::VQRSHRNs", SDTARMVSHX>;
57 def NEONvqrshrnu : SDNode<"ARMISD::VQRSHRNu", SDTARMVSHX>;
58 def NEONvqrshrnsu : SDNode<"ARMISD::VQRSHRNsu", SDTARMVSHX>;
60 def NEONvsli : SDNode<"ARMISD::VSLI", SDTARMVSHINS>;
61 def NEONvsri : SDNode<"ARMISD::VSRI", SDTARMVSHINS>;
63 def SDTARMVGETLN : SDTypeProfile<1, 2, [SDTCisVT<0, i32>, SDTCisInt<1>,
65 def NEONvgetlaneu : SDNode<"ARMISD::VGETLANEu", SDTARMVGETLN>;
66 def NEONvgetlanes : SDNode<"ARMISD::VGETLANEs", SDTARMVGETLN>;
68 def NEONvdup : SDNode<"ARMISD::VDUP", SDTypeProfile<1, 1, [SDTCisVec<0>]>>;
70 // VDUPLANE can produce a quad-register result from a double-register source,
71 // so the result is not constrained to match the source.
72 def NEONvduplane : SDNode<"ARMISD::VDUPLANE",
73 SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisVec<1>,
76 def SDTARMVEXT : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
77 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
78 def NEONvext : SDNode<"ARMISD::VEXT", SDTARMVEXT>;
80 def SDTARMVSHUF : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisSameAs<0, 1>]>;
81 def NEONvrev64 : SDNode<"ARMISD::VREV64", SDTARMVSHUF>;
82 def NEONvrev32 : SDNode<"ARMISD::VREV32", SDTARMVSHUF>;
83 def NEONvrev16 : SDNode<"ARMISD::VREV16", SDTARMVSHUF>;
85 def SDTARMVSHUF2 : SDTypeProfile<2, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
88 def NEONzip : SDNode<"ARMISD::VZIP", SDTARMVSHUF2>;
89 def NEONuzp : SDNode<"ARMISD::VUZP", SDTARMVSHUF2>;
90 def NEONtrn : SDNode<"ARMISD::VTRN", SDTARMVSHUF2>;
92 //===----------------------------------------------------------------------===//
93 // NEON operand definitions
94 //===----------------------------------------------------------------------===//
96 // addrmode_neonldstm := reg
98 /* TODO: Take advantage of vldm.
99 def addrmode_neonldstm : Operand<i32>,
100 ComplexPattern<i32, 2, "SelectAddrModeNeonLdStM", []> {
101 let PrintMethod = "printAddrNeonLdStMOperand";
102 let MIOperandInfo = (ops GPR, i32imm);
106 def h8imm : Operand<i8> {
107 let PrintMethod = "printHex8ImmOperand";
109 def h16imm : Operand<i16> {
110 let PrintMethod = "printHex16ImmOperand";
112 def h32imm : Operand<i32> {
113 let PrintMethod = "printHex32ImmOperand";
115 def h64imm : Operand<i64> {
116 let PrintMethod = "printHex64ImmOperand";
119 //===----------------------------------------------------------------------===//
120 // NEON load / store instructions
121 //===----------------------------------------------------------------------===//
123 /* TODO: Take advantage of vldm.
124 let mayLoad = 1, hasExtraDefRegAllocReq = 1 in {
125 def VLDMD : NI<(outs),
126 (ins addrmode_neonldstm:$addr, reglist:$dst1, variable_ops),
127 IIC_fpLoadm, "vldm", "${addr:submode} ${addr:base}, $dst1", []> {
128 let Inst{27-25} = 0b110;
130 let Inst{11-9} = 0b101;
133 def VLDMS : NI<(outs),
134 (ins addrmode_neonldstm:$addr, reglist:$dst1, variable_ops),
135 IIC_fpLoadm, "vldm", "${addr:submode} ${addr:base}, $dst1", []> {
136 let Inst{27-25} = 0b110;
138 let Inst{11-9} = 0b101;
143 // Use vldmia to load a Q register as a D register pair.
144 def VLDRQ : NI4<(outs QPR:$dst), (ins addrmode4:$addr), IIC_fpLoadm,
145 "vldmia", "$addr, ${dst:dregpair}",
146 [(set QPR:$dst, (v2f64 (load addrmode4:$addr)))]> {
147 let Inst{27-25} = 0b110;
148 let Inst{24} = 0; // P bit
149 let Inst{23} = 1; // U bit
151 let Inst{11-8} = 0b1011;
154 // Use vstmia to store a Q register as a D register pair.
155 def VSTRQ : NI4<(outs), (ins QPR:$src, addrmode4:$addr), IIC_fpStorem,
156 "vstmia", "$addr, ${src:dregpair}",
157 [(store (v2f64 QPR:$src), addrmode4:$addr)]> {
158 let Inst{27-25} = 0b110;
159 let Inst{24} = 0; // P bit
160 let Inst{23} = 1; // U bit
162 let Inst{11-8} = 0b1011;
165 // VLD1 : Vector Load (multiple single elements)
166 class VLD1D<bits<4> op7_4, string OpcodeStr, string Dt,
167 ValueType Ty, Intrinsic IntOp>
168 : NLdSt<0,0b10,0b0111,op7_4, (outs DPR:$dst), (ins addrmode6:$addr), IIC_VLD1,
169 OpcodeStr, Dt, "\\{$dst\\}, $addr", "",
170 [(set DPR:$dst, (Ty (IntOp addrmode6:$addr)))]>;
171 class VLD1Q<bits<4> op7_4, string OpcodeStr, string Dt,
172 ValueType Ty, Intrinsic IntOp>
173 : NLdSt<0,0b10,0b1010,op7_4, (outs QPR:$dst), (ins addrmode6:$addr), IIC_VLD1,
174 OpcodeStr, Dt, "${dst:dregpair}, $addr", "",
175 [(set QPR:$dst, (Ty (IntOp addrmode6:$addr)))]>;
177 def VLD1d8 : VLD1D<0b0000, "vld1", "8", v8i8, int_arm_neon_vld1>;
178 def VLD1d16 : VLD1D<0b0100, "vld1", "16", v4i16, int_arm_neon_vld1>;
179 def VLD1d32 : VLD1D<0b1000, "vld1", "32", v2i32, int_arm_neon_vld1>;
180 def VLD1df : VLD1D<0b1000, "vld1", "32", v2f32, int_arm_neon_vld1>;
181 def VLD1d64 : VLD1D<0b1100, "vld1", "64", v1i64, int_arm_neon_vld1>;
183 def VLD1q8 : VLD1Q<0b0000, "vld1", "8", v16i8, int_arm_neon_vld1>;
184 def VLD1q16 : VLD1Q<0b0100, "vld1", "16", v8i16, int_arm_neon_vld1>;
185 def VLD1q32 : VLD1Q<0b1000, "vld1", "32", v4i32, int_arm_neon_vld1>;
186 def VLD1qf : VLD1Q<0b1000, "vld1", "32", v4f32, int_arm_neon_vld1>;
187 def VLD1q64 : VLD1Q<0b1100, "vld1", "64", v2i64, int_arm_neon_vld1>;
189 let mayLoad = 1, hasExtraDefRegAllocReq = 1 in {
191 // VLD2 : Vector Load (multiple 2-element structures)
192 class VLD2D<bits<4> op7_4, string OpcodeStr, string Dt>
193 : NLdSt<0,0b10,0b1000,op7_4, (outs DPR:$dst1, DPR:$dst2),
194 (ins addrmode6:$addr), IIC_VLD2,
195 OpcodeStr, Dt, "\\{$dst1, $dst2\\}, $addr", "", []>;
196 class VLD2Q<bits<4> op7_4, string OpcodeStr, string Dt>
197 : NLdSt<0,0b10,0b0011,op7_4,
198 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
199 (ins addrmode6:$addr), IIC_VLD2,
200 OpcodeStr, Dt, "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr",
203 def VLD2d8 : VLD2D<0b0000, "vld2", "8">;
204 def VLD2d16 : VLD2D<0b0100, "vld2", "16">;
205 def VLD2d32 : VLD2D<0b1000, "vld2", "32">;
206 def VLD2d64 : NLdSt<0,0b10,0b1010,0b1100, (outs DPR:$dst1, DPR:$dst2),
207 (ins addrmode6:$addr), IIC_VLD1,
208 "vld1", "64", "\\{$dst1, $dst2\\}, $addr", "", []>;
210 def VLD2q8 : VLD2Q<0b0000, "vld2", "8">;
211 def VLD2q16 : VLD2Q<0b0100, "vld2", "16">;
212 def VLD2q32 : VLD2Q<0b1000, "vld2", "32">;
214 // VLD3 : Vector Load (multiple 3-element structures)
215 class VLD3D<bits<4> op7_4, string OpcodeStr, string Dt>
216 : NLdSt<0,0b10,0b0100,op7_4, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3),
217 (ins addrmode6:$addr), IIC_VLD3,
218 OpcodeStr, Dt, "\\{$dst1, $dst2, $dst3\\}, $addr", "", []>;
219 class VLD3WB<bits<4> op7_4, string OpcodeStr, string Dt>
220 : NLdSt<0,0b10,0b0101,op7_4, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, GPR:$wb),
221 (ins addrmode6:$addr), IIC_VLD3,
222 OpcodeStr, Dt, "\\{$dst1, $dst2, $dst3\\}, $addr",
223 "$addr.addr = $wb", []>;
225 def VLD3d8 : VLD3D<0b0000, "vld3", "8">;
226 def VLD3d16 : VLD3D<0b0100, "vld3", "16">;
227 def VLD3d32 : VLD3D<0b1000, "vld3", "32">;
228 def VLD3d64 : NLdSt<0,0b10,0b0110,0b1100,
229 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3),
230 (ins addrmode6:$addr), IIC_VLD1,
231 "vld1", "64", "\\{$dst1, $dst2, $dst3\\}, $addr", "", []>;
233 // vld3 to double-spaced even registers.
234 def VLD3q8a : VLD3WB<0b0000, "vld3", "8">;
235 def VLD3q16a : VLD3WB<0b0100, "vld3", "16">;
236 def VLD3q32a : VLD3WB<0b1000, "vld3", "32">;
238 // vld3 to double-spaced odd registers.
239 def VLD3q8b : VLD3WB<0b0000, "vld3", "8">;
240 def VLD3q16b : VLD3WB<0b0100, "vld3", "16">;
241 def VLD3q32b : VLD3WB<0b1000, "vld3", "32">;
243 // VLD4 : Vector Load (multiple 4-element structures)
244 class VLD4D<bits<4> op7_4, string OpcodeStr, string Dt>
245 : NLdSt<0,0b10,0b0000,op7_4,
246 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
247 (ins addrmode6:$addr), IIC_VLD4,
248 OpcodeStr, Dt, "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr",
250 class VLD4WB<bits<4> op7_4, string OpcodeStr, string Dt>
251 : NLdSt<0,0b10,0b0001,op7_4,
252 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
253 (ins addrmode6:$addr), IIC_VLD4,
254 OpcodeStr, Dt, "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr",
255 "$addr.addr = $wb", []>;
257 def VLD4d8 : VLD4D<0b0000, "vld4", "8">;
258 def VLD4d16 : VLD4D<0b0100, "vld4", "16">;
259 def VLD4d32 : VLD4D<0b1000, "vld4", "32">;
260 def VLD4d64 : NLdSt<0,0b10,0b0010,0b1100,
261 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
262 (ins addrmode6:$addr), IIC_VLD1,
263 "vld1", "64", "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr",
266 // vld4 to double-spaced even registers.
267 def VLD4q8a : VLD4WB<0b0000, "vld4", "8">;
268 def VLD4q16a : VLD4WB<0b0100, "vld4", "16">;
269 def VLD4q32a : VLD4WB<0b1000, "vld4", "32">;
271 // vld4 to double-spaced odd registers.
272 def VLD4q8b : VLD4WB<0b0000, "vld4", "8">;
273 def VLD4q16b : VLD4WB<0b0100, "vld4", "16">;
274 def VLD4q32b : VLD4WB<0b1000, "vld4", "32">;
276 // VLD1LN : Vector Load (single element to one lane)
277 // FIXME: Not yet implemented.
279 // VLD2LN : Vector Load (single 2-element structure to one lane)
280 class VLD2LN<bits<4> op11_8, string OpcodeStr, string Dt>
281 : NLdSt<1,0b10,op11_8,{?,?,?,?}, (outs DPR:$dst1, DPR:$dst2),
282 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, nohash_imm:$lane),
283 IIC_VLD2, OpcodeStr, Dt, "\\{$dst1[$lane], $dst2[$lane]\\}, $addr",
284 "$src1 = $dst1, $src2 = $dst2", []>;
286 // vld2 to single-spaced registers.
287 def VLD2LNd8 : VLD2LN<0b0001, "vld2", "8">;
288 def VLD2LNd16 : VLD2LN<0b0101, "vld2", "16"> { let Inst{5} = 0; }
289 def VLD2LNd32 : VLD2LN<0b1001, "vld2", "32"> { let Inst{6} = 0; }
291 // vld2 to double-spaced even registers.
292 def VLD2LNq16a: VLD2LN<0b0101, "vld2", "16"> { let Inst{5} = 1; }
293 def VLD2LNq32a: VLD2LN<0b1001, "vld2", "32"> { let Inst{6} = 1; }
295 // vld2 to double-spaced odd registers.
296 def VLD2LNq16b: VLD2LN<0b0101, "vld2", "16"> { let Inst{5} = 1; }
297 def VLD2LNq32b: VLD2LN<0b1001, "vld2", "32"> { let Inst{6} = 1; }
299 // VLD3LN : Vector Load (single 3-element structure to one lane)
300 class VLD3LN<bits<4> op11_8, string OpcodeStr, string Dt>
301 : NLdSt<1,0b10,op11_8,{?,?,?,?}, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3),
302 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3,
303 nohash_imm:$lane), IIC_VLD3, OpcodeStr, Dt,
304 "\\{$dst1[$lane], $dst2[$lane], $dst3[$lane]\\}, $addr",
305 "$src1 = $dst1, $src2 = $dst2, $src3 = $dst3", []>;
307 // vld3 to single-spaced registers.
308 def VLD3LNd8 : VLD3LN<0b0010, "vld3", "8"> { let Inst{4} = 0; }
309 def VLD3LNd16 : VLD3LN<0b0110, "vld3", "16"> { let Inst{5-4} = 0b00; }
310 def VLD3LNd32 : VLD3LN<0b1010, "vld3", "32"> { let Inst{6-4} = 0b000; }
312 // vld3 to double-spaced even registers.
313 def VLD3LNq16a: VLD3LN<0b0110, "vld3", "16"> { let Inst{5-4} = 0b10; }
314 def VLD3LNq32a: VLD3LN<0b1010, "vld3", "32"> { let Inst{6-4} = 0b100; }
316 // vld3 to double-spaced odd registers.
317 def VLD3LNq16b: VLD3LN<0b0110, "vld3", "16"> { let Inst{5-4} = 0b10; }
318 def VLD3LNq32b: VLD3LN<0b1010, "vld3", "32"> { let Inst{6-4} = 0b100; }
320 // VLD4LN : Vector Load (single 4-element structure to one lane)
321 class VLD4LN<bits<4> op11_8, string OpcodeStr, string Dt>
322 : NLdSt<1,0b10,op11_8,{?,?,?,?},
323 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
324 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
325 nohash_imm:$lane), IIC_VLD4, OpcodeStr, Dt,
326 "\\{$dst1[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $addr",
327 "$src1 = $dst1, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4", []>;
329 // vld4 to single-spaced registers.
330 def VLD4LNd8 : VLD4LN<0b0011, "vld4", "8">;
331 def VLD4LNd16 : VLD4LN<0b0111, "vld4", "16"> { let Inst{5} = 0; }
332 def VLD4LNd32 : VLD4LN<0b1011, "vld4", "32"> { let Inst{6} = 0; }
334 // vld4 to double-spaced even registers.
335 def VLD4LNq16a: VLD4LN<0b0111, "vld4", "16"> { let Inst{5} = 1; }
336 def VLD4LNq32a: VLD4LN<0b1011, "vld4", "32"> { let Inst{6} = 1; }
338 // vld4 to double-spaced odd registers.
339 def VLD4LNq16b: VLD4LN<0b0111, "vld4", "16"> { let Inst{5} = 1; }
340 def VLD4LNq32b: VLD4LN<0b1011, "vld4", "32"> { let Inst{6} = 1; }
342 // VLD1DUP : Vector Load (single element to all lanes)
343 // VLD2DUP : Vector Load (single 2-element structure to all lanes)
344 // VLD3DUP : Vector Load (single 3-element structure to all lanes)
345 // VLD4DUP : Vector Load (single 4-element structure to all lanes)
346 // FIXME: Not yet implemented.
347 } // mayLoad = 1, hasExtraDefRegAllocReq = 1
349 // VST1 : Vector Store (multiple single elements)
350 class VST1D<bits<4> op7_4, string OpcodeStr, string Dt,
351 ValueType Ty, Intrinsic IntOp>
352 : NLdSt<0,0b00,0b0111,op7_4, (outs), (ins addrmode6:$addr, DPR:$src), IIC_VST,
353 OpcodeStr, Dt, "\\{$src\\}, $addr", "",
354 [(IntOp addrmode6:$addr, (Ty DPR:$src))]>;
355 class VST1Q<bits<4> op7_4, string OpcodeStr, string Dt,
356 ValueType Ty, Intrinsic IntOp>
357 : NLdSt<0,0b00,0b1010,op7_4, (outs), (ins addrmode6:$addr, QPR:$src), IIC_VST,
358 OpcodeStr, Dt, "${src:dregpair}, $addr", "",
359 [(IntOp addrmode6:$addr, (Ty QPR:$src))]>;
361 let hasExtraSrcRegAllocReq = 1 in {
362 def VST1d8 : VST1D<0b0000, "vst1", "8", v8i8, int_arm_neon_vst1>;
363 def VST1d16 : VST1D<0b0100, "vst1", "16", v4i16, int_arm_neon_vst1>;
364 def VST1d32 : VST1D<0b1000, "vst1", "32", v2i32, int_arm_neon_vst1>;
365 def VST1df : VST1D<0b1000, "vst1", "32", v2f32, int_arm_neon_vst1>;
366 def VST1d64 : VST1D<0b1100, "vst1", "64", v1i64, int_arm_neon_vst1>;
368 def VST1q8 : VST1Q<0b0000, "vst1", "8", v16i8, int_arm_neon_vst1>;
369 def VST1q16 : VST1Q<0b0100, "vst1", "16", v8i16, int_arm_neon_vst1>;
370 def VST1q32 : VST1Q<0b1000, "vst1", "32", v4i32, int_arm_neon_vst1>;
371 def VST1qf : VST1Q<0b1000, "vst1", "32", v4f32, int_arm_neon_vst1>;
372 def VST1q64 : VST1Q<0b1100, "vst1", "64", v2i64, int_arm_neon_vst1>;
373 } // hasExtraSrcRegAllocReq
375 let mayStore = 1, hasExtraSrcRegAllocReq = 1 in {
377 // VST2 : Vector Store (multiple 2-element structures)
378 class VST2D<bits<4> op7_4, string OpcodeStr, string Dt>
379 : NLdSt<0,0b00,0b1000,op7_4, (outs),
380 (ins addrmode6:$addr, DPR:$src1, DPR:$src2), IIC_VST,
381 OpcodeStr, Dt, "\\{$src1, $src2\\}, $addr", "", []>;
382 class VST2Q<bits<4> op7_4, string OpcodeStr, string Dt>
383 : NLdSt<0,0b00,0b0011,op7_4, (outs),
384 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
385 IIC_VST, OpcodeStr, Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr",
388 def VST2d8 : VST2D<0b0000, "vst2", "8">;
389 def VST2d16 : VST2D<0b0100, "vst2", "16">;
390 def VST2d32 : VST2D<0b1000, "vst2", "32">;
391 def VST2d64 : NLdSt<0,0b00,0b1010,0b1100, (outs),
392 (ins addrmode6:$addr, DPR:$src1, DPR:$src2), IIC_VST,
393 "vst1", "64", "\\{$src1, $src2\\}, $addr", "", []>;
395 def VST2q8 : VST2Q<0b0000, "vst2", "8">;
396 def VST2q16 : VST2Q<0b0100, "vst2", "16">;
397 def VST2q32 : VST2Q<0b1000, "vst2", "32">;
399 // VST3 : Vector Store (multiple 3-element structures)
400 class VST3D<bits<4> op7_4, string OpcodeStr, string Dt>
401 : NLdSt<0,0b00,0b0100,op7_4, (outs),
402 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3), IIC_VST,
403 OpcodeStr, Dt, "\\{$src1, $src2, $src3\\}, $addr", "", []>;
404 class VST3WB<bits<4> op7_4, string OpcodeStr, string Dt>
405 : NLdSt<0,0b00,0b0101,op7_4, (outs GPR:$wb),
406 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3), IIC_VST,
407 OpcodeStr, Dt, "\\{$src1, $src2, $src3\\}, $addr",
408 "$addr.addr = $wb", []>;
410 def VST3d8 : VST3D<0b0000, "vst3", "8">;
411 def VST3d16 : VST3D<0b0100, "vst3", "16">;
412 def VST3d32 : VST3D<0b1000, "vst3", "32">;
413 def VST3d64 : NLdSt<0,0b00,0b0110,0b1100, (outs),
414 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3),
416 "vst1", "64", "\\{$src1, $src2, $src3\\}, $addr", "", []>;
418 // vst3 to double-spaced even registers.
419 def VST3q8a : VST3WB<0b0000, "vst3", "8">;
420 def VST3q16a : VST3WB<0b0100, "vst3", "16">;
421 def VST3q32a : VST3WB<0b1000, "vst3", "32">;
423 // vst3 to double-spaced odd registers.
424 def VST3q8b : VST3WB<0b0000, "vst3", "8">;
425 def VST3q16b : VST3WB<0b0100, "vst3", "16">;
426 def VST3q32b : VST3WB<0b1000, "vst3", "32">;
428 // VST4 : Vector Store (multiple 4-element structures)
429 class VST4D<bits<4> op7_4, string OpcodeStr, string Dt>
430 : NLdSt<0,0b00,0b0000,op7_4, (outs),
431 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
432 IIC_VST, OpcodeStr, Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr",
434 class VST4WB<bits<4> op7_4, string OpcodeStr, string Dt>
435 : NLdSt<0,0b00,0b0001,op7_4, (outs GPR:$wb),
436 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
437 IIC_VST, OpcodeStr, Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr",
438 "$addr.addr = $wb", []>;
440 def VST4d8 : VST4D<0b0000, "vst4", "8">;
441 def VST4d16 : VST4D<0b0100, "vst4", "16">;
442 def VST4d32 : VST4D<0b1000, "vst4", "32">;
443 def VST4d64 : NLdSt<0,0b00,0b0010,0b1100, (outs),
444 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3,
446 "vst1", "64", "\\{$src1, $src2, $src3, $src4\\}, $addr",
449 // vst4 to double-spaced even registers.
450 def VST4q8a : VST4WB<0b0000, "vst4", "8">;
451 def VST4q16a : VST4WB<0b0100, "vst4", "16">;
452 def VST4q32a : VST4WB<0b1000, "vst4", "32">;
454 // vst4 to double-spaced odd registers.
455 def VST4q8b : VST4WB<0b0000, "vst4", "8">;
456 def VST4q16b : VST4WB<0b0100, "vst4", "16">;
457 def VST4q32b : VST4WB<0b1000, "vst4", "32">;
459 // VST1LN : Vector Store (single element from one lane)
460 // FIXME: Not yet implemented.
462 // VST2LN : Vector Store (single 2-element structure from one lane)
463 class VST2LN<bits<4> op11_8, string OpcodeStr, string Dt>
464 : NLdSt<1,0b00,op11_8,{?,?,?,?}, (outs),
465 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, nohash_imm:$lane),
466 IIC_VST, OpcodeStr, Dt, "\\{$src1[$lane], $src2[$lane]\\}, $addr",
469 // vst2 to single-spaced registers.
470 def VST2LNd8 : VST2LN<0b0001, "vst2", "8">;
471 def VST2LNd16 : VST2LN<0b0101, "vst2", "16"> { let Inst{5} = 0; }
472 def VST2LNd32 : VST2LN<0b1001, "vst2", "32"> { let Inst{6} = 0; }
474 // vst2 to double-spaced even registers.
475 def VST2LNq16a: VST2LN<0b0101, "vst2", "16"> { let Inst{5} = 1; }
476 def VST2LNq32a: VST2LN<0b1001, "vst2", "32"> { let Inst{6} = 1; }
478 // vst2 to double-spaced odd registers.
479 def VST2LNq16b: VST2LN<0b0101, "vst2", "16"> { let Inst{5} = 1; }
480 def VST2LNq32b: VST2LN<0b1001, "vst2", "32"> { let Inst{6} = 1; }
482 // VST3LN : Vector Store (single 3-element structure from one lane)
483 class VST3LN<bits<4> op11_8, string OpcodeStr, string Dt>
484 : NLdSt<1,0b00,op11_8,{?,?,?,?}, (outs),
485 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3,
486 nohash_imm:$lane), IIC_VST, OpcodeStr, Dt,
487 "\\{$src1[$lane], $src2[$lane], $src3[$lane]\\}, $addr", "", []>;
489 // vst3 to single-spaced registers.
490 def VST3LNd8 : VST3LN<0b0010, "vst3", "8"> { let Inst{4} = 0; }
491 def VST3LNd16 : VST3LN<0b0110, "vst3", "16"> { let Inst{5-4} = 0b00; }
492 def VST3LNd32 : VST3LN<0b1010, "vst3", "32"> { let Inst{6-4} = 0b000; }
494 // vst3 to double-spaced even registers.
495 def VST3LNq16a: VST3LN<0b0110, "vst3", "16"> { let Inst{5-4} = 0b10; }
496 def VST3LNq32a: VST3LN<0b1010, "vst3", "32"> { let Inst{6-4} = 0b100; }
498 // vst3 to double-spaced odd registers.
499 def VST3LNq16b: VST3LN<0b0110, "vst3", "16"> { let Inst{5-4} = 0b10; }
500 def VST3LNq32b: VST3LN<0b1010, "vst3", "32"> { let Inst{6-4} = 0b100; }
502 // VST4LN : Vector Store (single 4-element structure from one lane)
503 class VST4LN<bits<4> op11_8, string OpcodeStr, string Dt>
504 : NLdSt<1,0b00,op11_8,{?,?,?,?}, (outs),
505 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
506 nohash_imm:$lane), IIC_VST, OpcodeStr, Dt,
507 "\\{$src1[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $addr",
510 // vst4 to single-spaced registers.
511 def VST4LNd8 : VST4LN<0b0011, "vst4", "8">;
512 def VST4LNd16 : VST4LN<0b0111, "vst4", "16"> { let Inst{5} = 0; }
513 def VST4LNd32 : VST4LN<0b1011, "vst4", "32"> { let Inst{6} = 0; }
515 // vst4 to double-spaced even registers.
516 def VST4LNq16a: VST4LN<0b0111, "vst4", "16"> { let Inst{5} = 1; }
517 def VST4LNq32a: VST4LN<0b1011, "vst4", "32"> { let Inst{6} = 1; }
519 // vst4 to double-spaced odd registers.
520 def VST4LNq16b: VST4LN<0b0111, "vst4", "16"> { let Inst{5} = 1; }
521 def VST4LNq32b: VST4LN<0b1011, "vst4", "32"> { let Inst{6} = 1; }
523 } // mayStore = 1, hasExtraSrcRegAllocReq = 1
526 //===----------------------------------------------------------------------===//
527 // NEON pattern fragments
528 //===----------------------------------------------------------------------===//
530 // Extract D sub-registers of Q registers.
531 // (arm_dsubreg_0 is 5; arm_dsubreg_1 is 6)
532 def DSubReg_i8_reg : SDNodeXForm<imm, [{
533 return CurDAG->getTargetConstant(5 + N->getZExtValue() / 8, MVT::i32);
535 def DSubReg_i16_reg : SDNodeXForm<imm, [{
536 return CurDAG->getTargetConstant(5 + N->getZExtValue() / 4, MVT::i32);
538 def DSubReg_i32_reg : SDNodeXForm<imm, [{
539 return CurDAG->getTargetConstant(5 + N->getZExtValue() / 2, MVT::i32);
541 def DSubReg_f64_reg : SDNodeXForm<imm, [{
542 return CurDAG->getTargetConstant(5 + N->getZExtValue(), MVT::i32);
544 def DSubReg_f64_other_reg : SDNodeXForm<imm, [{
545 return CurDAG->getTargetConstant(5 + (1 - N->getZExtValue()), MVT::i32);
548 // Extract S sub-registers of Q/D registers.
549 // (arm_ssubreg_0 is 1; arm_ssubreg_1 is 2; etc.)
550 def SSubReg_f32_reg : SDNodeXForm<imm, [{
551 return CurDAG->getTargetConstant(1 + N->getZExtValue(), MVT::i32);
554 // Translate lane numbers from Q registers to D subregs.
555 def SubReg_i8_lane : SDNodeXForm<imm, [{
556 return CurDAG->getTargetConstant(N->getZExtValue() & 7, MVT::i32);
558 def SubReg_i16_lane : SDNodeXForm<imm, [{
559 return CurDAG->getTargetConstant(N->getZExtValue() & 3, MVT::i32);
561 def SubReg_i32_lane : SDNodeXForm<imm, [{
562 return CurDAG->getTargetConstant(N->getZExtValue() & 1, MVT::i32);
565 //===----------------------------------------------------------------------===//
566 // Instruction Classes
567 //===----------------------------------------------------------------------===//
569 // Basic 2-register operations: single-, double- and quad-register.
570 class N2VS<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
571 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
572 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
573 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
574 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src),
575 IIC_VUNAD, OpcodeStr, Dt, "$dst, $src", "", []>;
576 class N2VD<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
577 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
578 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
579 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
580 (ins DPR:$src), IIC_VUNAD, OpcodeStr, Dt, "$dst, $src", "",
581 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src))))]>;
582 class N2VQ<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
583 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
584 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
585 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
586 (ins QPR:$src), IIC_VUNAQ, OpcodeStr, Dt, "$dst, $src", "",
587 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src))))]>;
589 // Basic 2-register intrinsics, both double- and quad-register.
590 class N2VDInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
591 bits<2> op17_16, bits<5> op11_7, bit op4,
592 InstrItinClass itin, string OpcodeStr, string Dt,
593 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
594 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
595 (ins DPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
596 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
597 class N2VQInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
598 bits<2> op17_16, bits<5> op11_7, bit op4,
599 InstrItinClass itin, string OpcodeStr, string Dt,
600 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
601 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
602 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
603 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
605 // Narrow 2-register intrinsics.
606 class N2VNInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
607 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
608 InstrItinClass itin, string OpcodeStr, string Dt,
609 ValueType TyD, ValueType TyQ, Intrinsic IntOp>
610 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$dst),
611 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
612 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src))))]>;
614 // Long 2-register intrinsics (currently only used for VMOVL).
615 class N2VLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
616 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
617 InstrItinClass itin, string OpcodeStr, string Dt,
618 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
619 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$dst),
620 (ins DPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
621 [(set QPR:$dst, (TyQ (IntOp (TyD DPR:$src))))]>;
623 // 2-register shuffles (VTRN/VZIP/VUZP), both double- and quad-register.
624 class N2VDShuffle<bits<2> op19_18, bits<5> op11_7, string OpcodeStr, string Dt>
625 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 0, 0, (outs DPR:$dst1, DPR:$dst2),
626 (ins DPR:$src1, DPR:$src2), IIC_VPERMD,
627 OpcodeStr, Dt, "$dst1, $dst2",
628 "$src1 = $dst1, $src2 = $dst2", []>;
629 class N2VQShuffle<bits<2> op19_18, bits<5> op11_7,
630 InstrItinClass itin, string OpcodeStr, string Dt>
631 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 1, 0, (outs QPR:$dst1, QPR:$dst2),
632 (ins QPR:$src1, QPR:$src2), itin, OpcodeStr, Dt, "$dst1, $dst2",
633 "$src1 = $dst1, $src2 = $dst2", []>;
635 // Basic 3-register operations: single-, double- and quad-register.
636 class N3VS<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
637 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
638 SDNode OpNode, bit Commutable>
639 : N3V<op24, op23, op21_20, op11_8, 0, op4,
640 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src1, DPR_VFP2:$src2), IIC_VBIND,
641 OpcodeStr, Dt, "$dst, $src1, $src2", "", []> {
642 let isCommutable = Commutable;
645 class N3VD<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
646 InstrItinClass itin, string OpcodeStr, string Dt,
647 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
648 : N3V<op24, op23, op21_20, op11_8, 0, op4,
649 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), itin,
650 OpcodeStr, Dt, "$dst, $src1, $src2", "",
651 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src1), (OpTy DPR:$src2))))]> {
652 let isCommutable = Commutable;
654 // Same as N3VD but no data type.
655 class N3VDX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
656 InstrItinClass itin, string OpcodeStr,
657 ValueType ResTy, ValueType OpTy,
658 SDNode OpNode, bit Commutable>
659 : N3VX<op24, op23, op21_20, op11_8, 0, op4,
660 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), itin,
661 OpcodeStr, "$dst, $src1, $src2", "",
662 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src1), (OpTy DPR:$src2))))]>{
663 let isCommutable = Commutable;
665 class N3VDSL<bits<2> op21_20, bits<4> op11_8,
666 InstrItinClass itin, string OpcodeStr, string Dt,
667 ValueType Ty, SDNode ShOp>
668 : N3V<0, 1, op21_20, op11_8, 1, 0,
669 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
670 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
672 (Ty (ShOp (Ty DPR:$src1),
673 (Ty (NEONvduplane (Ty DPR_VFP2:$src2), imm:$lane)))))]>{
674 let isCommutable = 0;
676 class N3VDSL16<bits<2> op21_20, bits<4> op11_8,
677 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
678 : N3V<0, 1, op21_20, op11_8, 1, 0,
679 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
680 IIC_VMULi16D, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
682 (Ty (ShOp (Ty DPR:$src1),
683 (Ty (NEONvduplane (Ty DPR_8:$src2), imm:$lane)))))]> {
684 let isCommutable = 0;
687 class N3VQ<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
688 InstrItinClass itin, string OpcodeStr, string Dt,
689 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
690 : N3V<op24, op23, op21_20, op11_8, 1, op4,
691 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), itin,
692 OpcodeStr, Dt, "$dst, $src1, $src2", "",
693 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src1), (OpTy QPR:$src2))))]> {
694 let isCommutable = Commutable;
696 class N3VQX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
697 InstrItinClass itin, string OpcodeStr,
698 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
699 : N3VX<op24, op23, op21_20, op11_8, 1, op4,
700 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), itin,
701 OpcodeStr, "$dst, $src1, $src2", "",
702 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src1), (OpTy QPR:$src2))))]>{
703 let isCommutable = Commutable;
705 class N3VQSL<bits<2> op21_20, bits<4> op11_8,
706 InstrItinClass itin, string OpcodeStr, string Dt,
707 ValueType ResTy, ValueType OpTy, SDNode ShOp>
708 : N3V<1, 1, op21_20, op11_8, 1, 0,
709 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
710 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
711 [(set (ResTy QPR:$dst),
712 (ResTy (ShOp (ResTy QPR:$src1),
713 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
715 let isCommutable = 0;
717 class N3VQSL16<bits<2> op21_20, bits<4> op11_8, string OpcodeStr, string Dt,
718 ValueType ResTy, ValueType OpTy, SDNode ShOp>
719 : N3V<1, 1, op21_20, op11_8, 1, 0,
720 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
721 IIC_VMULi16Q, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
722 [(set (ResTy QPR:$dst),
723 (ResTy (ShOp (ResTy QPR:$src1),
724 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
726 let isCommutable = 0;
729 // Basic 3-register intrinsics, both double- and quad-register.
730 class N3VDInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
731 InstrItinClass itin, string OpcodeStr, string Dt,
732 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
733 : N3V<op24, op23, op21_20, op11_8, 0, op4,
734 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), itin,
735 OpcodeStr, Dt, "$dst, $src1, $src2", "",
736 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src1), (OpTy DPR:$src2))))]> {
737 let isCommutable = Commutable;
739 class N3VDIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
740 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
741 : N3V<0, 1, op21_20, op11_8, 1, 0,
742 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
743 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
745 (Ty (IntOp (Ty DPR:$src1),
746 (Ty (NEONvduplane (Ty DPR_VFP2:$src2),
748 let isCommutable = 0;
750 class N3VDIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
751 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
752 : N3V<0, 1, op21_20, op11_8, 1, 0,
753 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
754 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
756 (Ty (IntOp (Ty DPR:$src1),
757 (Ty (NEONvduplane (Ty DPR_8:$src2),
759 let isCommutable = 0;
762 class N3VQInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
763 InstrItinClass itin, string OpcodeStr, string Dt,
764 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
765 : N3V<op24, op23, op21_20, op11_8, 1, op4,
766 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), itin,
767 OpcodeStr, Dt, "$dst, $src1, $src2", "",
768 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src1), (OpTy QPR:$src2))))]> {
769 let isCommutable = Commutable;
771 class N3VQIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
772 string OpcodeStr, string Dt,
773 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
774 : N3V<1, 1, op21_20, op11_8, 1, 0,
775 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
776 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
777 [(set (ResTy QPR:$dst),
778 (ResTy (IntOp (ResTy QPR:$src1),
779 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
781 let isCommutable = 0;
783 class N3VQIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
784 string OpcodeStr, string Dt,
785 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
786 : N3V<1, 1, op21_20, op11_8, 1, 0,
787 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
788 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
789 [(set (ResTy QPR:$dst),
790 (ResTy (IntOp (ResTy QPR:$src1),
791 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
793 let isCommutable = 0;
796 // Multiply-Add/Sub operations: single-, double- and quad-register.
797 class N3VSMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
798 InstrItinClass itin, string OpcodeStr, string Dt,
799 ValueType Ty, SDNode MulOp, SDNode OpNode>
800 : N3V<op24, op23, op21_20, op11_8, 0, op4,
801 (outs DPR_VFP2:$dst),
802 (ins DPR_VFP2:$src1, DPR_VFP2:$src2, DPR_VFP2:$src3), itin,
803 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst", []>;
805 class N3VDMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
806 InstrItinClass itin, string OpcodeStr, string Dt,
807 ValueType Ty, SDNode MulOp, SDNode OpNode>
808 : N3V<op24, op23, op21_20, op11_8, 0, op4,
809 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3), itin,
810 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
811 [(set DPR:$dst, (Ty (OpNode DPR:$src1,
812 (Ty (MulOp DPR:$src2, DPR:$src3)))))]>;
813 class N3VDMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
814 string OpcodeStr, string Dt,
815 ValueType Ty, SDNode MulOp, SDNode ShOp>
816 : N3V<0, 1, op21_20, op11_8, 1, 0,
818 (ins DPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane), itin,
819 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
821 (Ty (ShOp (Ty DPR:$src1),
822 (Ty (MulOp DPR:$src2,
823 (Ty (NEONvduplane (Ty DPR_VFP2:$src3),
825 class N3VDMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
826 string OpcodeStr, string Dt,
827 ValueType Ty, SDNode MulOp, SDNode ShOp>
828 : N3V<0, 1, op21_20, op11_8, 1, 0,
830 (ins DPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane), itin,
831 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
833 (Ty (ShOp (Ty DPR:$src1),
834 (Ty (MulOp DPR:$src2,
835 (Ty (NEONvduplane (Ty DPR_8:$src3),
838 class N3VQMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
839 InstrItinClass itin, string OpcodeStr, string Dt, ValueType Ty,
840 SDNode MulOp, SDNode OpNode>
841 : N3V<op24, op23, op21_20, op11_8, 1, op4,
842 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3), itin,
843 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
844 [(set QPR:$dst, (Ty (OpNode QPR:$src1,
845 (Ty (MulOp QPR:$src2, QPR:$src3)))))]>;
846 class N3VQMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
847 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
848 SDNode MulOp, SDNode ShOp>
849 : N3V<1, 1, op21_20, op11_8, 1, 0,
851 (ins QPR:$src1, QPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane), itin,
852 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
853 [(set (ResTy QPR:$dst),
854 (ResTy (ShOp (ResTy QPR:$src1),
855 (ResTy (MulOp QPR:$src2,
856 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src3),
858 class N3VQMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
859 string OpcodeStr, string Dt,
860 ValueType ResTy, ValueType OpTy,
861 SDNode MulOp, SDNode ShOp>
862 : N3V<1, 1, op21_20, op11_8, 1, 0,
864 (ins QPR:$src1, QPR:$src2, DPR_8:$src3, nohash_imm:$lane), itin,
865 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
866 [(set (ResTy QPR:$dst),
867 (ResTy (ShOp (ResTy QPR:$src1),
868 (ResTy (MulOp QPR:$src2,
869 (ResTy (NEONvduplane (OpTy DPR_8:$src3),
872 // Neon 3-argument intrinsics, both double- and quad-register.
873 // The destination register is also used as the first source operand register.
874 class N3VDInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
875 InstrItinClass itin, string OpcodeStr, string Dt,
876 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
877 : N3V<op24, op23, op21_20, op11_8, 0, op4,
878 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3), itin,
879 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
880 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src1),
881 (OpTy DPR:$src2), (OpTy DPR:$src3))))]>;
882 class N3VQInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
883 InstrItinClass itin, string OpcodeStr, string Dt,
884 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
885 : N3V<op24, op23, op21_20, op11_8, 1, op4,
886 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3), itin,
887 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
888 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src1),
889 (OpTy QPR:$src2), (OpTy QPR:$src3))))]>;
891 // Neon Long 3-argument intrinsic. The destination register is
892 // a quad-register and is also used as the first source operand register.
893 class N3VLInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
894 InstrItinClass itin, string OpcodeStr, string Dt,
895 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
896 : N3V<op24, op23, op21_20, op11_8, 0, op4,
897 (outs QPR:$dst), (ins QPR:$src1, DPR:$src2, DPR:$src3), itin,
898 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
900 (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$src2), (TyD DPR:$src3))))]>;
901 class N3VLInt3SL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
902 string OpcodeStr, string Dt,
903 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
904 : N3V<op24, 1, op21_20, op11_8, 1, 0,
906 (ins QPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane), itin,
907 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
908 [(set (ResTy QPR:$dst),
909 (ResTy (IntOp (ResTy QPR:$src1),
911 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src3),
913 class N3VLInt3SL16<bit op24, bits<2> op21_20, bits<4> op11_8,
914 InstrItinClass itin, string OpcodeStr, string Dt,
915 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
916 : N3V<op24, 1, op21_20, op11_8, 1, 0,
918 (ins QPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane), itin,
919 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
920 [(set (ResTy QPR:$dst),
921 (ResTy (IntOp (ResTy QPR:$src1),
923 (OpTy (NEONvduplane (OpTy DPR_8:$src3),
926 // Narrowing 3-register intrinsics.
927 class N3VNInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
928 string OpcodeStr, string Dt, ValueType TyD, ValueType TyQ,
929 Intrinsic IntOp, bit Commutable>
930 : N3V<op24, op23, op21_20, op11_8, 0, op4,
931 (outs DPR:$dst), (ins QPR:$src1, QPR:$src2), IIC_VBINi4D,
932 OpcodeStr, Dt, "$dst, $src1, $src2", "",
933 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src1), (TyQ QPR:$src2))))]> {
934 let isCommutable = Commutable;
937 // Long 3-register intrinsics.
938 class N3VLInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
939 InstrItinClass itin, string OpcodeStr, string Dt,
940 ValueType TyQ, ValueType TyD, Intrinsic IntOp, bit Commutable>
941 : N3V<op24, op23, op21_20, op11_8, 0, op4,
942 (outs QPR:$dst), (ins DPR:$src1, DPR:$src2), itin,
943 OpcodeStr, Dt, "$dst, $src1, $src2", "",
944 [(set QPR:$dst, (TyQ (IntOp (TyD DPR:$src1), (TyD DPR:$src2))))]> {
945 let isCommutable = Commutable;
947 class N3VLIntSL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
948 string OpcodeStr, string Dt,
949 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
950 : N3V<op24, 1, op21_20, op11_8, 1, 0,
951 (outs QPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
952 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
953 [(set (ResTy QPR:$dst),
954 (ResTy (IntOp (OpTy DPR:$src1),
955 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src2),
957 class N3VLIntSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
958 InstrItinClass itin, string OpcodeStr, string Dt,
959 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
960 : N3V<op24, 1, op21_20, op11_8, 1, 0,
961 (outs QPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
962 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
963 [(set (ResTy QPR:$dst),
964 (ResTy (IntOp (OpTy DPR:$src1),
965 (OpTy (NEONvduplane (OpTy DPR_8:$src2),
968 // Wide 3-register intrinsics.
969 class N3VWInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
970 string OpcodeStr, string Dt, ValueType TyQ, ValueType TyD,
971 Intrinsic IntOp, bit Commutable>
972 : N3V<op24, op23, op21_20, op11_8, 0, op4,
973 (outs QPR:$dst), (ins QPR:$src1, DPR:$src2), IIC_VSUBiD,
974 OpcodeStr, Dt, "$dst, $src1, $src2", "",
975 [(set QPR:$dst, (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$src2))))]> {
976 let isCommutable = Commutable;
979 // Pairwise long 2-register intrinsics, both double- and quad-register.
980 class N2VDPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
981 bits<2> op17_16, bits<5> op11_7, bit op4,
982 string OpcodeStr, string Dt,
983 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
984 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
985 (ins DPR:$src), IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
986 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
987 class N2VQPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
988 bits<2> op17_16, bits<5> op11_7, bit op4,
989 string OpcodeStr, string Dt,
990 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
991 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
992 (ins QPR:$src), IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
993 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
995 // Pairwise long 2-register accumulate intrinsics,
996 // both double- and quad-register.
997 // The destination register is also used as the first source operand register.
998 class N2VDPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
999 bits<2> op17_16, bits<5> op11_7, bit op4,
1000 string OpcodeStr, string Dt,
1001 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1002 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
1003 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), IIC_VPALiD,
1004 OpcodeStr, Dt, "$dst, $src2", "$src1 = $dst",
1005 [(set DPR:$dst, (ResTy (IntOp (ResTy DPR:$src1), (OpTy DPR:$src2))))]>;
1006 class N2VQPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1007 bits<2> op17_16, bits<5> op11_7, bit op4,
1008 string OpcodeStr, string Dt,
1009 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1010 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4,
1011 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), IIC_VPALiQ,
1012 OpcodeStr, Dt, "$dst, $src2", "$src1 = $dst",
1013 [(set QPR:$dst, (ResTy (IntOp (ResTy QPR:$src1), (OpTy QPR:$src2))))]>;
1015 // Shift by immediate,
1016 // both double- and quad-register.
1017 class N2VDSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1018 InstrItinClass itin, string OpcodeStr, string Dt,
1019 ValueType Ty, SDNode OpNode>
1020 : N2VImm<op24, op23, op11_8, op7, 0, op4,
1021 (outs DPR:$dst), (ins DPR:$src, i32imm:$SIMM), itin,
1022 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1023 [(set DPR:$dst, (Ty (OpNode (Ty DPR:$src), (i32 imm:$SIMM))))]>;
1024 class N2VQSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1025 InstrItinClass itin, string OpcodeStr, string Dt,
1026 ValueType Ty, SDNode OpNode>
1027 : N2VImm<op24, op23, op11_8, op7, 1, op4,
1028 (outs QPR:$dst), (ins QPR:$src, i32imm:$SIMM), itin,
1029 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1030 [(set QPR:$dst, (Ty (OpNode (Ty QPR:$src), (i32 imm:$SIMM))))]>;
1032 // Long shift by immediate.
1033 class N2VLSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
1034 string OpcodeStr, string Dt,
1035 ValueType ResTy, ValueType OpTy, SDNode OpNode>
1036 : N2VImm<op24, op23, op11_8, op7, op6, op4,
1037 (outs QPR:$dst), (ins DPR:$src, i32imm:$SIMM), IIC_VSHLiD,
1038 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1039 [(set QPR:$dst, (ResTy (OpNode (OpTy DPR:$src),
1040 (i32 imm:$SIMM))))]>;
1042 // Narrow shift by immediate.
1043 class N2VNSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
1044 InstrItinClass itin, string OpcodeStr, string Dt,
1045 ValueType ResTy, ValueType OpTy, SDNode OpNode>
1046 : N2VImm<op24, op23, op11_8, op7, op6, op4,
1047 (outs DPR:$dst), (ins QPR:$src, i32imm:$SIMM), itin,
1048 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1049 [(set DPR:$dst, (ResTy (OpNode (OpTy QPR:$src),
1050 (i32 imm:$SIMM))))]>;
1052 // Shift right by immediate and accumulate,
1053 // both double- and quad-register.
1054 class N2VDShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1055 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1056 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$dst),
1057 (ins DPR:$src1, DPR:$src2, i32imm:$SIMM), IIC_VPALiD,
1058 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1059 [(set DPR:$dst, (Ty (add DPR:$src1,
1060 (Ty (ShOp DPR:$src2, (i32 imm:$SIMM))))))]>;
1061 class N2VQShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1062 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1063 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$dst),
1064 (ins QPR:$src1, QPR:$src2, i32imm:$SIMM), IIC_VPALiD,
1065 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1066 [(set QPR:$dst, (Ty (add QPR:$src1,
1067 (Ty (ShOp QPR:$src2, (i32 imm:$SIMM))))))]>;
1069 // Shift by immediate and insert,
1070 // both double- and quad-register.
1071 class N2VDShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1072 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1073 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$dst),
1074 (ins DPR:$src1, DPR:$src2, i32imm:$SIMM), IIC_VSHLiD,
1075 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1076 [(set DPR:$dst, (Ty (ShOp DPR:$src1, DPR:$src2, (i32 imm:$SIMM))))]>;
1077 class N2VQShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1078 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1079 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$dst),
1080 (ins QPR:$src1, QPR:$src2, i32imm:$SIMM), IIC_VSHLiQ,
1081 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1082 [(set QPR:$dst, (Ty (ShOp QPR:$src1, QPR:$src2, (i32 imm:$SIMM))))]>;
1084 // Convert, with fractional bits immediate,
1085 // both double- and quad-register.
1086 class N2VCvtD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1087 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1089 : N2VImm<op24, op23, op11_8, op7, 0, op4,
1090 (outs DPR:$dst), (ins DPR:$src, i32imm:$SIMM), IIC_VUNAD,
1091 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1092 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src), (i32 imm:$SIMM))))]>;
1093 class N2VCvtQ<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1094 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1096 : N2VImm<op24, op23, op11_8, op7, 1, op4,
1097 (outs QPR:$dst), (ins QPR:$src, i32imm:$SIMM), IIC_VUNAQ,
1098 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1099 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src), (i32 imm:$SIMM))))]>;
1101 //===----------------------------------------------------------------------===//
1103 //===----------------------------------------------------------------------===//
1105 // Abbreviations used in multiclass suffixes:
1106 // Q = quarter int (8 bit) elements
1107 // H = half int (16 bit) elements
1108 // S = single int (32 bit) elements
1109 // D = double int (64 bit) elements
1111 // Neon 3-register vector operations.
1113 // First with only element sizes of 8, 16 and 32 bits:
1114 multiclass N3V_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1115 InstrItinClass itinD16, InstrItinClass itinD32,
1116 InstrItinClass itinQ16, InstrItinClass itinQ32,
1117 string OpcodeStr, string Dt,
1118 SDNode OpNode, bit Commutable = 0> {
1119 // 64-bit vector types.
1120 def v8i8 : N3VD<op24, op23, 0b00, op11_8, op4, itinD16,
1121 OpcodeStr, !strconcat(Dt, "8"),
1122 v8i8, v8i8, OpNode, Commutable>;
1123 def v4i16 : N3VD<op24, op23, 0b01, op11_8, op4, itinD16,
1124 OpcodeStr, !strconcat(Dt, "16"),
1125 v4i16, v4i16, OpNode, Commutable>;
1126 def v2i32 : N3VD<op24, op23, 0b10, op11_8, op4, itinD32,
1127 OpcodeStr, !strconcat(Dt, "32"),
1128 v2i32, v2i32, OpNode, Commutable>;
1130 // 128-bit vector types.
1131 def v16i8 : N3VQ<op24, op23, 0b00, op11_8, op4, itinQ16,
1132 OpcodeStr, !strconcat(Dt, "8"),
1133 v16i8, v16i8, OpNode, Commutable>;
1134 def v8i16 : N3VQ<op24, op23, 0b01, op11_8, op4, itinQ16,
1135 OpcodeStr, !strconcat(Dt, "16"),
1136 v8i16, v8i16, OpNode, Commutable>;
1137 def v4i32 : N3VQ<op24, op23, 0b10, op11_8, op4, itinQ32,
1138 OpcodeStr, !strconcat(Dt, "32"),
1139 v4i32, v4i32, OpNode, Commutable>;
1142 multiclass N3VSL_HS<bits<4> op11_8, string OpcodeStr, string Dt, SDNode ShOp> {
1143 def v4i16 : N3VDSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
1145 def v2i32 : N3VDSL<0b10, op11_8, IIC_VMULi32D, OpcodeStr, !strconcat(Dt,"32"),
1147 def v8i16 : N3VQSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
1148 v8i16, v4i16, ShOp>;
1149 def v4i32 : N3VQSL<0b10, op11_8, IIC_VMULi32Q, OpcodeStr, !strconcat(Dt,"32"),
1150 v4i32, v2i32, ShOp>;
1153 // ....then also with element size 64 bits:
1154 multiclass N3V_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1155 InstrItinClass itinD, InstrItinClass itinQ,
1156 string OpcodeStr, string Dt,
1157 SDNode OpNode, bit Commutable = 0>
1158 : N3V_QHS<op24, op23, op11_8, op4, itinD, itinD, itinQ, itinQ,
1159 OpcodeStr, Dt, OpNode, Commutable> {
1160 def v1i64 : N3VD<op24, op23, 0b11, op11_8, op4, itinD,
1161 OpcodeStr, !strconcat(Dt, "64"),
1162 v1i64, v1i64, OpNode, Commutable>;
1163 def v2i64 : N3VQ<op24, op23, 0b11, op11_8, op4, itinQ,
1164 OpcodeStr, !strconcat(Dt, "64"),
1165 v2i64, v2i64, OpNode, Commutable>;
1169 // Neon Narrowing 2-register vector intrinsics,
1170 // source operand element sizes of 16, 32 and 64 bits:
1171 multiclass N2VNInt_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1172 bits<5> op11_7, bit op6, bit op4,
1173 InstrItinClass itin, string OpcodeStr, string Dt,
1175 def v8i8 : N2VNInt<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
1176 itin, OpcodeStr, !strconcat(Dt, "16"),
1177 v8i8, v8i16, IntOp>;
1178 def v4i16 : N2VNInt<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
1179 itin, OpcodeStr, !strconcat(Dt, "32"),
1180 v4i16, v4i32, IntOp>;
1181 def v2i32 : N2VNInt<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
1182 itin, OpcodeStr, !strconcat(Dt, "64"),
1183 v2i32, v2i64, IntOp>;
1187 // Neon Lengthening 2-register vector intrinsic (currently specific to VMOVL).
1188 // source operand element sizes of 16, 32 and 64 bits:
1189 multiclass N2VLInt_QHS<bits<2> op24_23, bits<5> op11_7, bit op6, bit op4,
1190 string OpcodeStr, string Dt, Intrinsic IntOp> {
1191 def v8i16 : N2VLInt<op24_23, 0b00, 0b10, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
1192 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
1193 def v4i32 : N2VLInt<op24_23, 0b01, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
1194 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
1195 def v2i64 : N2VLInt<op24_23, 0b10, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
1196 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1200 // Neon 3-register vector intrinsics.
1202 // First with only element sizes of 16 and 32 bits:
1203 multiclass N3VInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
1204 InstrItinClass itinD16, InstrItinClass itinD32,
1205 InstrItinClass itinQ16, InstrItinClass itinQ32,
1206 string OpcodeStr, string Dt,
1207 Intrinsic IntOp, bit Commutable = 0> {
1208 // 64-bit vector types.
1209 def v4i16 : N3VDInt<op24, op23, 0b01, op11_8, op4, itinD16,
1210 OpcodeStr, !strconcat(Dt, "16"),
1211 v4i16, v4i16, IntOp, Commutable>;
1212 def v2i32 : N3VDInt<op24, op23, 0b10, op11_8, op4, itinD32,
1213 OpcodeStr, !strconcat(Dt, "32"),
1214 v2i32, v2i32, IntOp, Commutable>;
1216 // 128-bit vector types.
1217 def v8i16 : N3VQInt<op24, op23, 0b01, op11_8, op4, itinQ16,
1218 OpcodeStr, !strconcat(Dt, "16"),
1219 v8i16, v8i16, IntOp, Commutable>;
1220 def v4i32 : N3VQInt<op24, op23, 0b10, op11_8, op4, itinQ32,
1221 OpcodeStr, !strconcat(Dt, "32"),
1222 v4i32, v4i32, IntOp, Commutable>;
1225 multiclass N3VIntSL_HS<bits<4> op11_8,
1226 InstrItinClass itinD16, InstrItinClass itinD32,
1227 InstrItinClass itinQ16, InstrItinClass itinQ32,
1228 string OpcodeStr, string Dt, Intrinsic IntOp> {
1229 def v4i16 : N3VDIntSL16<0b01, op11_8, itinD16,
1230 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp>;
1231 def v2i32 : N3VDIntSL<0b10, op11_8, itinD32,
1232 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp>;
1233 def v8i16 : N3VQIntSL16<0b01, op11_8, itinQ16,
1234 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16, IntOp>;
1235 def v4i32 : N3VQIntSL<0b10, op11_8, itinQ32,
1236 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32, IntOp>;
1239 // ....then also with element size of 8 bits:
1240 multiclass N3VInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1241 InstrItinClass itinD16, InstrItinClass itinD32,
1242 InstrItinClass itinQ16, InstrItinClass itinQ32,
1243 string OpcodeStr, string Dt,
1244 Intrinsic IntOp, bit Commutable = 0>
1245 : N3VInt_HS<op24, op23, op11_8, op4, itinD16, itinD32, itinQ16, itinQ32,
1246 OpcodeStr, Dt, IntOp, Commutable> {
1247 def v8i8 : N3VDInt<op24, op23, 0b00, op11_8, op4, itinD16,
1248 OpcodeStr, !strconcat(Dt, "8"),
1249 v8i8, v8i8, IntOp, Commutable>;
1250 def v16i8 : N3VQInt<op24, op23, 0b00, op11_8, op4, itinQ16,
1251 OpcodeStr, !strconcat(Dt, "8"),
1252 v16i8, v16i8, IntOp, Commutable>;
1255 // ....then also with element size of 64 bits:
1256 multiclass N3VInt_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1257 InstrItinClass itinD16, InstrItinClass itinD32,
1258 InstrItinClass itinQ16, InstrItinClass itinQ32,
1259 string OpcodeStr, string Dt,
1260 Intrinsic IntOp, bit Commutable = 0>
1261 : N3VInt_QHS<op24, op23, op11_8, op4, itinD16, itinD32, itinQ16, itinQ32,
1262 OpcodeStr, Dt, IntOp, Commutable> {
1263 def v1i64 : N3VDInt<op24, op23, 0b11, op11_8, op4, itinD32,
1264 OpcodeStr, !strconcat(Dt, "64"),
1265 v1i64, v1i64, IntOp, Commutable>;
1266 def v2i64 : N3VQInt<op24, op23, 0b11, op11_8, op4, itinQ32,
1267 OpcodeStr, !strconcat(Dt, "64"),
1268 v2i64, v2i64, IntOp, Commutable>;
1272 // Neon Narrowing 3-register vector intrinsics,
1273 // source operand element sizes of 16, 32 and 64 bits:
1274 multiclass N3VNInt_HSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1275 string OpcodeStr, string Dt,
1276 Intrinsic IntOp, bit Commutable = 0> {
1277 def v8i8 : N3VNInt<op24, op23, 0b00, op11_8, op4,
1278 OpcodeStr, !strconcat(Dt, "16"),
1279 v8i8, v8i16, IntOp, Commutable>;
1280 def v4i16 : N3VNInt<op24, op23, 0b01, op11_8, op4,
1281 OpcodeStr, !strconcat(Dt, "32"),
1282 v4i16, v4i32, IntOp, Commutable>;
1283 def v2i32 : N3VNInt<op24, op23, 0b10, op11_8, op4,
1284 OpcodeStr, !strconcat(Dt, "64"),
1285 v2i32, v2i64, IntOp, Commutable>;
1289 // Neon Long 3-register vector intrinsics.
1291 // First with only element sizes of 16 and 32 bits:
1292 multiclass N3VLInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
1293 InstrItinClass itin, string OpcodeStr, string Dt,
1294 Intrinsic IntOp, bit Commutable = 0> {
1295 def v4i32 : N3VLInt<op24, op23, 0b01, op11_8, op4, itin,
1296 OpcodeStr, !strconcat(Dt, "16"),
1297 v4i32, v4i16, IntOp, Commutable>;
1298 def v2i64 : N3VLInt<op24, op23, 0b10, op11_8, op4, itin,
1299 OpcodeStr, !strconcat(Dt, "32"),
1300 v2i64, v2i32, IntOp, Commutable>;
1303 multiclass N3VLIntSL_HS<bit op24, bits<4> op11_8,
1304 InstrItinClass itin, string OpcodeStr, string Dt,
1306 def v4i16 : N3VLIntSL16<op24, 0b01, op11_8, itin,
1307 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
1308 def v2i32 : N3VLIntSL<op24, 0b10, op11_8, itin,
1309 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1312 // ....then also with element size of 8 bits:
1313 multiclass N3VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1314 InstrItinClass itin, string OpcodeStr, string Dt,
1315 Intrinsic IntOp, bit Commutable = 0>
1316 : N3VLInt_HS<op24, op23, op11_8, op4, itin, OpcodeStr, Dt,
1317 IntOp, Commutable> {
1318 def v8i16 : N3VLInt<op24, op23, 0b00, op11_8, op4, itin,
1319 OpcodeStr, !strconcat(Dt, "8"),
1320 v8i16, v8i8, IntOp, Commutable>;
1324 // Neon Wide 3-register vector intrinsics,
1325 // source operand element sizes of 8, 16 and 32 bits:
1326 multiclass N3VWInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1327 string OpcodeStr, string Dt,
1328 Intrinsic IntOp, bit Commutable = 0> {
1329 def v8i16 : N3VWInt<op24, op23, 0b00, op11_8, op4,
1330 OpcodeStr, !strconcat(Dt, "8"),
1331 v8i16, v8i8, IntOp, Commutable>;
1332 def v4i32 : N3VWInt<op24, op23, 0b01, op11_8, op4,
1333 OpcodeStr, !strconcat(Dt, "16"),
1334 v4i32, v4i16, IntOp, Commutable>;
1335 def v2i64 : N3VWInt<op24, op23, 0b10, op11_8, op4,
1336 OpcodeStr, !strconcat(Dt, "32"),
1337 v2i64, v2i32, IntOp, Commutable>;
1341 // Neon Multiply-Op vector operations,
1342 // element sizes of 8, 16 and 32 bits:
1343 multiclass N3VMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1344 InstrItinClass itinD16, InstrItinClass itinD32,
1345 InstrItinClass itinQ16, InstrItinClass itinQ32,
1346 string OpcodeStr, string Dt, SDNode OpNode> {
1347 // 64-bit vector types.
1348 def v8i8 : N3VDMulOp<op24, op23, 0b00, op11_8, op4, itinD16,
1349 OpcodeStr, !strconcat(Dt, "8"), v8i8, mul, OpNode>;
1350 def v4i16 : N3VDMulOp<op24, op23, 0b01, op11_8, op4, itinD16,
1351 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, OpNode>;
1352 def v2i32 : N3VDMulOp<op24, op23, 0b10, op11_8, op4, itinD32,
1353 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, OpNode>;
1355 // 128-bit vector types.
1356 def v16i8 : N3VQMulOp<op24, op23, 0b00, op11_8, op4, itinQ16,
1357 OpcodeStr, !strconcat(Dt, "8"), v16i8, mul, OpNode>;
1358 def v8i16 : N3VQMulOp<op24, op23, 0b01, op11_8, op4, itinQ16,
1359 OpcodeStr, !strconcat(Dt, "16"), v8i16, mul, OpNode>;
1360 def v4i32 : N3VQMulOp<op24, op23, 0b10, op11_8, op4, itinQ32,
1361 OpcodeStr, !strconcat(Dt, "32"), v4i32, mul, OpNode>;
1364 multiclass N3VMulOpSL_HS<bits<4> op11_8,
1365 InstrItinClass itinD16, InstrItinClass itinD32,
1366 InstrItinClass itinQ16, InstrItinClass itinQ32,
1367 string OpcodeStr, string Dt, SDNode ShOp> {
1368 def v4i16 : N3VDMulOpSL16<0b01, op11_8, itinD16,
1369 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, ShOp>;
1370 def v2i32 : N3VDMulOpSL<0b10, op11_8, itinD32,
1371 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, ShOp>;
1372 def v8i16 : N3VQMulOpSL16<0b01, op11_8, itinQ16,
1373 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16,
1375 def v4i32 : N3VQMulOpSL<0b10, op11_8, itinQ32,
1376 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32,
1380 // Neon 3-argument intrinsics,
1381 // element sizes of 8, 16 and 32 bits:
1382 multiclass N3VInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1383 string OpcodeStr, string Dt, Intrinsic IntOp> {
1384 // 64-bit vector types.
1385 def v8i8 : N3VDInt3<op24, op23, 0b00, op11_8, op4, IIC_VMACi16D,
1386 OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
1387 def v4i16 : N3VDInt3<op24, op23, 0b01, op11_8, op4, IIC_VMACi16D,
1388 OpcodeStr, !strconcat(Dt, "16"), v4i16, v4i16, IntOp>;
1389 def v2i32 : N3VDInt3<op24, op23, 0b10, op11_8, op4, IIC_VMACi32D,
1390 OpcodeStr, !strconcat(Dt, "32"), v2i32, v2i32, IntOp>;
1392 // 128-bit vector types.
1393 def v16i8 : N3VQInt3<op24, op23, 0b00, op11_8, op4, IIC_VMACi16Q,
1394 OpcodeStr, !strconcat(Dt, "8"), v16i8, v16i8, IntOp>;
1395 def v8i16 : N3VQInt3<op24, op23, 0b01, op11_8, op4, IIC_VMACi16Q,
1396 OpcodeStr, !strconcat(Dt, "16"), v8i16, v8i16, IntOp>;
1397 def v4i32 : N3VQInt3<op24, op23, 0b10, op11_8, op4, IIC_VMACi32Q,
1398 OpcodeStr, !strconcat(Dt, "32"), v4i32, v4i32, IntOp>;
1402 // Neon Long 3-argument intrinsics.
1404 // First with only element sizes of 16 and 32 bits:
1405 multiclass N3VLInt3_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
1406 string OpcodeStr, string Dt, Intrinsic IntOp> {
1407 def v4i32 : N3VLInt3<op24, op23, 0b01, op11_8, op4, IIC_VMACi16D,
1408 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
1409 def v2i64 : N3VLInt3<op24, op23, 0b10, op11_8, op4, IIC_VMACi16D,
1410 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1413 multiclass N3VLInt3SL_HS<bit op24, bits<4> op11_8,
1414 string OpcodeStr, string Dt, Intrinsic IntOp> {
1415 def v4i16 : N3VLInt3SL16<op24, 0b01, op11_8, IIC_VMACi16D,
1416 OpcodeStr, !strconcat(Dt,"16"), v4i32, v4i16, IntOp>;
1417 def v2i32 : N3VLInt3SL<op24, 0b10, op11_8, IIC_VMACi32D,
1418 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1421 // ....then also with element size of 8 bits:
1422 multiclass N3VLInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1423 string OpcodeStr, string Dt, Intrinsic IntOp>
1424 : N3VLInt3_HS<op24, op23, op11_8, op4, OpcodeStr, Dt, IntOp> {
1425 def v8i16 : N3VLInt3<op24, op23, 0b00, op11_8, op4, IIC_VMACi16D,
1426 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
1430 // Neon 2-register vector intrinsics,
1431 // element sizes of 8, 16 and 32 bits:
1432 multiclass N2VInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1433 bits<5> op11_7, bit op4,
1434 InstrItinClass itinD, InstrItinClass itinQ,
1435 string OpcodeStr, string Dt, Intrinsic IntOp> {
1436 // 64-bit vector types.
1437 def v8i8 : N2VDInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1438 itinD, OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
1439 def v4i16 : N2VDInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1440 itinD, OpcodeStr, !strconcat(Dt, "16"),v4i16,v4i16,IntOp>;
1441 def v2i32 : N2VDInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1442 itinD, OpcodeStr, !strconcat(Dt, "32"),v2i32,v2i32,IntOp>;
1444 // 128-bit vector types.
1445 def v16i8 : N2VQInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1446 itinQ, OpcodeStr, !strconcat(Dt, "8"), v16i8,v16i8,IntOp>;
1447 def v8i16 : N2VQInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1448 itinQ, OpcodeStr, !strconcat(Dt, "16"),v8i16,v8i16,IntOp>;
1449 def v4i32 : N2VQInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1450 itinQ, OpcodeStr, !strconcat(Dt, "32"),v4i32,v4i32,IntOp>;
1454 // Neon Pairwise long 2-register intrinsics,
1455 // element sizes of 8, 16 and 32 bits:
1456 multiclass N2VPLInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1457 bits<5> op11_7, bit op4,
1458 string OpcodeStr, string Dt, Intrinsic IntOp> {
1459 // 64-bit vector types.
1460 def v8i8 : N2VDPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1461 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
1462 def v4i16 : N2VDPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1463 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
1464 def v2i32 : N2VDPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1465 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
1467 // 128-bit vector types.
1468 def v16i8 : N2VQPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1469 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
1470 def v8i16 : N2VQPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1471 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
1472 def v4i32 : N2VQPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1473 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
1477 // Neon Pairwise long 2-register accumulate intrinsics,
1478 // element sizes of 8, 16 and 32 bits:
1479 multiclass N2VPLInt2_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1480 bits<5> op11_7, bit op4,
1481 string OpcodeStr, string Dt, Intrinsic IntOp> {
1482 // 64-bit vector types.
1483 def v8i8 : N2VDPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1484 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
1485 def v4i16 : N2VDPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1486 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
1487 def v2i32 : N2VDPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1488 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
1490 // 128-bit vector types.
1491 def v16i8 : N2VQPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1492 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
1493 def v8i16 : N2VQPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1494 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
1495 def v4i32 : N2VQPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1496 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
1500 // Neon 2-register vector shift by immediate,
1501 // element sizes of 8, 16, 32 and 64 bits:
1502 multiclass N2VSh_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1503 InstrItinClass itin, string OpcodeStr, string Dt,
1505 // 64-bit vector types.
1506 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, itin,
1507 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
1508 let Inst{21-19} = 0b001; // imm6 = 001xxx
1510 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, itin,
1511 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
1512 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1514 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, itin,
1515 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
1516 let Inst{21} = 0b1; // imm6 = 1xxxxx
1518 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, itin,
1519 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
1522 // 128-bit vector types.
1523 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, itin,
1524 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
1525 let Inst{21-19} = 0b001; // imm6 = 001xxx
1527 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, itin,
1528 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
1529 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1531 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, itin,
1532 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
1533 let Inst{21} = 0b1; // imm6 = 1xxxxx
1535 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, itin,
1536 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
1541 // Neon Shift-Accumulate vector operations,
1542 // element sizes of 8, 16, 32 and 64 bits:
1543 multiclass N2VShAdd_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1544 string OpcodeStr, string Dt, SDNode ShOp> {
1545 // 64-bit vector types.
1546 def v8i8 : N2VDShAdd<op24, op23, op11_8, 0, op4,
1547 OpcodeStr, !strconcat(Dt, "8"), v8i8, ShOp> {
1548 let Inst{21-19} = 0b001; // imm6 = 001xxx
1550 def v4i16 : N2VDShAdd<op24, op23, op11_8, 0, op4,
1551 OpcodeStr, !strconcat(Dt, "16"), v4i16, ShOp> {
1552 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1554 def v2i32 : N2VDShAdd<op24, op23, op11_8, 0, op4,
1555 OpcodeStr, !strconcat(Dt, "32"), v2i32, ShOp> {
1556 let Inst{21} = 0b1; // imm6 = 1xxxxx
1558 def v1i64 : N2VDShAdd<op24, op23, op11_8, 1, op4,
1559 OpcodeStr, !strconcat(Dt, "64"), v1i64, ShOp>;
1562 // 128-bit vector types.
1563 def v16i8 : N2VQShAdd<op24, op23, op11_8, 0, op4,
1564 OpcodeStr, !strconcat(Dt, "8"), v16i8, ShOp> {
1565 let Inst{21-19} = 0b001; // imm6 = 001xxx
1567 def v8i16 : N2VQShAdd<op24, op23, op11_8, 0, op4,
1568 OpcodeStr, !strconcat(Dt, "16"), v8i16, ShOp> {
1569 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1571 def v4i32 : N2VQShAdd<op24, op23, op11_8, 0, op4,
1572 OpcodeStr, !strconcat(Dt, "32"), v4i32, ShOp> {
1573 let Inst{21} = 0b1; // imm6 = 1xxxxx
1575 def v2i64 : N2VQShAdd<op24, op23, op11_8, 1, op4,
1576 OpcodeStr, !strconcat(Dt, "64"), v2i64, ShOp>;
1581 // Neon Shift-Insert vector operations,
1582 // element sizes of 8, 16, 32 and 64 bits:
1583 multiclass N2VShIns_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1584 string OpcodeStr, SDNode ShOp> {
1585 // 64-bit vector types.
1586 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4,
1587 OpcodeStr, "8", v8i8, ShOp> {
1588 let Inst{21-19} = 0b001; // imm6 = 001xxx
1590 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4,
1591 OpcodeStr, "16", v4i16, ShOp> {
1592 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1594 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4,
1595 OpcodeStr, "32", v2i32, ShOp> {
1596 let Inst{21} = 0b1; // imm6 = 1xxxxx
1598 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4,
1599 OpcodeStr, "64", v1i64, ShOp>;
1602 // 128-bit vector types.
1603 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4,
1604 OpcodeStr, "8", v16i8, ShOp> {
1605 let Inst{21-19} = 0b001; // imm6 = 001xxx
1607 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4,
1608 OpcodeStr, "16", v8i16, ShOp> {
1609 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1611 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4,
1612 OpcodeStr, "32", v4i32, ShOp> {
1613 let Inst{21} = 0b1; // imm6 = 1xxxxx
1615 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4,
1616 OpcodeStr, "64", v2i64, ShOp>;
1620 // Neon Shift Long operations,
1621 // element sizes of 8, 16, 32 bits:
1622 multiclass N2VLSh_QHS<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
1623 bit op4, string OpcodeStr, string Dt, SDNode OpNode> {
1624 def v8i16 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
1625 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode> {
1626 let Inst{21-19} = 0b001; // imm6 = 001xxx
1628 def v4i32 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
1629 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode> {
1630 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1632 def v2i64 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
1633 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode> {
1634 let Inst{21} = 0b1; // imm6 = 1xxxxx
1638 // Neon Shift Narrow operations,
1639 // element sizes of 16, 32, 64 bits:
1640 multiclass N2VNSh_HSD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
1641 bit op4, InstrItinClass itin, string OpcodeStr, string Dt,
1643 def v8i8 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
1644 OpcodeStr, !strconcat(Dt, "16"), v8i8, v8i16, OpNode> {
1645 let Inst{21-19} = 0b001; // imm6 = 001xxx
1647 def v4i16 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
1648 OpcodeStr, !strconcat(Dt, "32"), v4i16, v4i32, OpNode> {
1649 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1651 def v2i32 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
1652 OpcodeStr, !strconcat(Dt, "64"), v2i32, v2i64, OpNode> {
1653 let Inst{21} = 0b1; // imm6 = 1xxxxx
1657 //===----------------------------------------------------------------------===//
1658 // Instruction Definitions.
1659 //===----------------------------------------------------------------------===//
1661 // Vector Add Operations.
1663 // VADD : Vector Add (integer and floating-point)
1664 defm VADD : N3V_QHSD<0, 0, 0b1000, 0, IIC_VBINiD, IIC_VBINiQ, "vadd", "i",
1666 def VADDfd : N3VD<0, 0, 0b00, 0b1101, 0, IIC_VBIND, "vadd", "f32",
1667 v2f32, v2f32, fadd, 1>;
1668 def VADDfq : N3VQ<0, 0, 0b00, 0b1101, 0, IIC_VBINQ, "vadd", "f32",
1669 v4f32, v4f32, fadd, 1>;
1670 // VADDL : Vector Add Long (Q = D + D)
1671 defm VADDLs : N3VLInt_QHS<0,1,0b0000,0, IIC_VSHLiD, "vaddl", "s",
1672 int_arm_neon_vaddls, 1>;
1673 defm VADDLu : N3VLInt_QHS<1,1,0b0000,0, IIC_VSHLiD, "vaddl", "u",
1674 int_arm_neon_vaddlu, 1>;
1675 // VADDW : Vector Add Wide (Q = Q + D)
1676 defm VADDWs : N3VWInt_QHS<0,1,0b0001,0, "vaddw", "s", int_arm_neon_vaddws, 0>;
1677 defm VADDWu : N3VWInt_QHS<1,1,0b0001,0, "vaddw", "u", int_arm_neon_vaddwu, 0>;
1678 // VHADD : Vector Halving Add
1679 defm VHADDs : N3VInt_QHS<0,0,0b0000,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1680 IIC_VBINi4Q, "vhadd", "s", int_arm_neon_vhadds, 1>;
1681 defm VHADDu : N3VInt_QHS<1,0,0b0000,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1682 IIC_VBINi4Q, "vhadd", "u", int_arm_neon_vhaddu, 1>;
1683 // VRHADD : Vector Rounding Halving Add
1684 defm VRHADDs : N3VInt_QHS<0,0,0b0001,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1685 IIC_VBINi4Q, "vrhadd", "s", int_arm_neon_vrhadds, 1>;
1686 defm VRHADDu : N3VInt_QHS<1,0,0b0001,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1687 IIC_VBINi4Q, "vrhadd", "u", int_arm_neon_vrhaddu, 1>;
1688 // VQADD : Vector Saturating Add
1689 defm VQADDs : N3VInt_QHSD<0,0,0b0000,1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1690 IIC_VBINi4Q, "vqadd", "s", int_arm_neon_vqadds, 1>;
1691 defm VQADDu : N3VInt_QHSD<1,0,0b0000,1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1692 IIC_VBINi4Q, "vqadd", "u", int_arm_neon_vqaddu, 1>;
1693 // VADDHN : Vector Add and Narrow Returning High Half (D = Q + Q)
1694 defm VADDHN : N3VNInt_HSD<0,1,0b0100,0, "vaddhn", "i",
1695 int_arm_neon_vaddhn, 1>;
1696 // VRADDHN : Vector Rounding Add and Narrow Returning High Half (D = Q + Q)
1697 defm VRADDHN : N3VNInt_HSD<1,1,0b0100,0, "vraddhn", "i",
1698 int_arm_neon_vraddhn, 1>;
1700 // Vector Multiply Operations.
1702 // VMUL : Vector Multiply (integer, polynomial and floating-point)
1703 defm VMUL : N3V_QHS<0, 0, 0b1001, 1, IIC_VMULi16D, IIC_VMULi32D,
1704 IIC_VMULi16Q, IIC_VMULi32Q, "vmul", "i", mul, 1>;
1705 def VMULpd : N3VDInt<1, 0, 0b00, 0b1001, 1, IIC_VMULi16D, "vmul", "p8",
1706 v8i8, v8i8, int_arm_neon_vmulp, 1>;
1707 def VMULpq : N3VQInt<1, 0, 0b00, 0b1001, 1, IIC_VMULi16Q, "vmul", "p8",
1708 v16i8, v16i8, int_arm_neon_vmulp, 1>;
1709 def VMULfd : N3VD<1, 0, 0b00, 0b1101, 1, IIC_VBIND, "vmul", "f32",
1710 v2f32, v2f32, fmul, 1>;
1711 def VMULfq : N3VQ<1, 0, 0b00, 0b1101, 1, IIC_VBINQ, "vmul", "f32",
1712 v4f32, v4f32, fmul, 1>;
1713 defm VMULsl : N3VSL_HS<0b1000, "vmul", "i", mul>;
1714 def VMULslfd : N3VDSL<0b10, 0b1001, IIC_VBIND, "vmul", "f32", v2f32, fmul>;
1715 def VMULslfq : N3VQSL<0b10, 0b1001, IIC_VBINQ, "vmul", "f32", v4f32,
1718 def : Pat<(v8i16 (mul (v8i16 QPR:$src1),
1719 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
1720 (v8i16 (VMULslv8i16 (v8i16 QPR:$src1),
1721 (v4i16 (EXTRACT_SUBREG QPR:$src2,
1722 (DSubReg_i16_reg imm:$lane))),
1723 (SubReg_i16_lane imm:$lane)))>;
1724 def : Pat<(v4i32 (mul (v4i32 QPR:$src1),
1725 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
1726 (v4i32 (VMULslv4i32 (v4i32 QPR:$src1),
1727 (v2i32 (EXTRACT_SUBREG QPR:$src2,
1728 (DSubReg_i32_reg imm:$lane))),
1729 (SubReg_i32_lane imm:$lane)))>;
1730 def : Pat<(v4f32 (fmul (v4f32 QPR:$src1),
1731 (v4f32 (NEONvduplane (v4f32 QPR:$src2), imm:$lane)))),
1732 (v4f32 (VMULslfq (v4f32 QPR:$src1),
1733 (v2f32 (EXTRACT_SUBREG QPR:$src2,
1734 (DSubReg_i32_reg imm:$lane))),
1735 (SubReg_i32_lane imm:$lane)))>;
1737 // VQDMULH : Vector Saturating Doubling Multiply Returning High Half
1738 defm VQDMULH : N3VInt_HS<0, 0, 0b1011, 0, IIC_VMULi16D, IIC_VMULi32D,
1739 IIC_VMULi16Q, IIC_VMULi32Q,
1740 "vqdmulh", "s", int_arm_neon_vqdmulh, 1>;
1741 defm VQDMULHsl: N3VIntSL_HS<0b1100, IIC_VMULi16D, IIC_VMULi32D,
1742 IIC_VMULi16Q, IIC_VMULi32Q,
1743 "vqdmulh", "s", int_arm_neon_vqdmulh>;
1744 def : Pat<(v8i16 (int_arm_neon_vqdmulh (v8i16 QPR:$src1),
1745 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
1747 (v8i16 (VQDMULHslv8i16 (v8i16 QPR:$src1),
1748 (v4i16 (EXTRACT_SUBREG QPR:$src2,
1749 (DSubReg_i16_reg imm:$lane))),
1750 (SubReg_i16_lane imm:$lane)))>;
1751 def : Pat<(v4i32 (int_arm_neon_vqdmulh (v4i32 QPR:$src1),
1752 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
1754 (v4i32 (VQDMULHslv4i32 (v4i32 QPR:$src1),
1755 (v2i32 (EXTRACT_SUBREG QPR:$src2,
1756 (DSubReg_i32_reg imm:$lane))),
1757 (SubReg_i32_lane imm:$lane)))>;
1759 // VQRDMULH : Vector Rounding Saturating Doubling Multiply Returning High Half
1760 defm VQRDMULH : N3VInt_HS<1, 0, 0b1011, 0, IIC_VMULi16D, IIC_VMULi32D,
1761 IIC_VMULi16Q, IIC_VMULi32Q,
1762 "vqrdmulh", "s", int_arm_neon_vqrdmulh, 1>;
1763 defm VQRDMULHsl : N3VIntSL_HS<0b1101, IIC_VMULi16D, IIC_VMULi32D,
1764 IIC_VMULi16Q, IIC_VMULi32Q,
1765 "vqrdmulh", "s", int_arm_neon_vqrdmulh>;
1766 def : Pat<(v8i16 (int_arm_neon_vqrdmulh (v8i16 QPR:$src1),
1767 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
1769 (v8i16 (VQRDMULHslv8i16 (v8i16 QPR:$src1),
1770 (v4i16 (EXTRACT_SUBREG QPR:$src2,
1771 (DSubReg_i16_reg imm:$lane))),
1772 (SubReg_i16_lane imm:$lane)))>;
1773 def : Pat<(v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$src1),
1774 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
1776 (v4i32 (VQRDMULHslv4i32 (v4i32 QPR:$src1),
1777 (v2i32 (EXTRACT_SUBREG QPR:$src2,
1778 (DSubReg_i32_reg imm:$lane))),
1779 (SubReg_i32_lane imm:$lane)))>;
1781 // VMULL : Vector Multiply Long (integer and polynomial) (Q = D * D)
1782 defm VMULLs : N3VLInt_QHS<0,1,0b1100,0, IIC_VMULi16D, "vmull", "s",
1783 int_arm_neon_vmulls, 1>;
1784 defm VMULLu : N3VLInt_QHS<1,1,0b1100,0, IIC_VMULi16D, "vmull", "u",
1785 int_arm_neon_vmullu, 1>;
1786 def VMULLp : N3VLInt<0, 1, 0b00, 0b1110, 0, IIC_VMULi16D, "vmull", "p8",
1787 v8i16, v8i8, int_arm_neon_vmullp, 1>;
1788 defm VMULLsls : N3VLIntSL_HS<0, 0b1010, IIC_VMULi16D, "vmull", "s",
1789 int_arm_neon_vmulls>;
1790 defm VMULLslu : N3VLIntSL_HS<1, 0b1010, IIC_VMULi16D, "vmull", "u",
1791 int_arm_neon_vmullu>;
1793 // VQDMULL : Vector Saturating Doubling Multiply Long (Q = D * D)
1794 defm VQDMULL : N3VLInt_HS<0,1,0b1101,0, IIC_VMULi16D, "vqdmull", "s",
1795 int_arm_neon_vqdmull, 1>;
1796 defm VQDMULLsl: N3VLIntSL_HS<0, 0b1011, IIC_VMULi16D, "vqdmull", "s",
1797 int_arm_neon_vqdmull>;
1799 // Vector Multiply-Accumulate and Multiply-Subtract Operations.
1801 // VMLA : Vector Multiply Accumulate (integer and floating-point)
1802 defm VMLA : N3VMulOp_QHS<0, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
1803 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
1804 def VMLAfd : N3VDMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACD, "vmla", "f32",
1806 def VMLAfq : N3VQMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACQ, "vmla", "f32",
1808 defm VMLAsl : N3VMulOpSL_HS<0b0000, IIC_VMACi16D, IIC_VMACi32D,
1809 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
1810 def VMLAslfd : N3VDMulOpSL<0b10, 0b0001, IIC_VMACD, "vmla", "f32",
1812 def VMLAslfq : N3VQMulOpSL<0b10, 0b0001, IIC_VMACQ, "vmla", "f32",
1813 v4f32, v2f32, fmul, fadd>;
1815 def : Pat<(v8i16 (add (v8i16 QPR:$src1),
1816 (mul (v8i16 QPR:$src2),
1817 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
1818 (v8i16 (VMLAslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
1819 (v4i16 (EXTRACT_SUBREG QPR:$src3,
1820 (DSubReg_i16_reg imm:$lane))),
1821 (SubReg_i16_lane imm:$lane)))>;
1823 def : Pat<(v4i32 (add (v4i32 QPR:$src1),
1824 (mul (v4i32 QPR:$src2),
1825 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
1826 (v4i32 (VMLAslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
1827 (v2i32 (EXTRACT_SUBREG QPR:$src3,
1828 (DSubReg_i32_reg imm:$lane))),
1829 (SubReg_i32_lane imm:$lane)))>;
1831 def : Pat<(v4f32 (fadd (v4f32 QPR:$src1),
1832 (fmul (v4f32 QPR:$src2),
1833 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
1834 (v4f32 (VMLAslfq (v4f32 QPR:$src1),
1836 (v2f32 (EXTRACT_SUBREG QPR:$src3,
1837 (DSubReg_i32_reg imm:$lane))),
1838 (SubReg_i32_lane imm:$lane)))>;
1840 // VMLAL : Vector Multiply Accumulate Long (Q += D * D)
1841 defm VMLALs : N3VLInt3_QHS<0,1,0b1000,0, "vmlal", "s", int_arm_neon_vmlals>;
1842 defm VMLALu : N3VLInt3_QHS<1,1,0b1000,0, "vmlal", "u", int_arm_neon_vmlalu>;
1844 defm VMLALsls : N3VLInt3SL_HS<0, 0b0010, "vmlal", "s", int_arm_neon_vmlals>;
1845 defm VMLALslu : N3VLInt3SL_HS<1, 0b0010, "vmlal", "u", int_arm_neon_vmlalu>;
1847 // VQDMLAL : Vector Saturating Doubling Multiply Accumulate Long (Q += D * D)
1848 defm VQDMLAL : N3VLInt3_HS<0, 1, 0b1001, 0, "vqdmlal", "s",
1849 int_arm_neon_vqdmlal>;
1850 defm VQDMLALsl: N3VLInt3SL_HS<0, 0b0011, "vqdmlal", "s", int_arm_neon_vqdmlal>;
1852 // VMLS : Vector Multiply Subtract (integer and floating-point)
1853 defm VMLS : N3VMulOp_QHS<1, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
1854 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
1855 def VMLSfd : N3VDMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACD, "vmls", "f32",
1857 def VMLSfq : N3VQMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACQ, "vmls", "f32",
1859 defm VMLSsl : N3VMulOpSL_HS<0b0100, IIC_VMACi16D, IIC_VMACi32D,
1860 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
1861 def VMLSslfd : N3VDMulOpSL<0b10, 0b0101, IIC_VMACD, "vmls", "f32",
1863 def VMLSslfq : N3VQMulOpSL<0b10, 0b0101, IIC_VMACQ, "vmls", "f32",
1864 v4f32, v2f32, fmul, fsub>;
1866 def : Pat<(v8i16 (sub (v8i16 QPR:$src1),
1867 (mul (v8i16 QPR:$src2),
1868 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
1869 (v8i16 (VMLSslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
1870 (v4i16 (EXTRACT_SUBREG QPR:$src3,
1871 (DSubReg_i16_reg imm:$lane))),
1872 (SubReg_i16_lane imm:$lane)))>;
1874 def : Pat<(v4i32 (sub (v4i32 QPR:$src1),
1875 (mul (v4i32 QPR:$src2),
1876 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
1877 (v4i32 (VMLSslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
1878 (v2i32 (EXTRACT_SUBREG QPR:$src3,
1879 (DSubReg_i32_reg imm:$lane))),
1880 (SubReg_i32_lane imm:$lane)))>;
1882 def : Pat<(v4f32 (fsub (v4f32 QPR:$src1),
1883 (fmul (v4f32 QPR:$src2),
1884 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
1885 (v4f32 (VMLSslfq (v4f32 QPR:$src1), (v4f32 QPR:$src2),
1886 (v2f32 (EXTRACT_SUBREG QPR:$src3,
1887 (DSubReg_i32_reg imm:$lane))),
1888 (SubReg_i32_lane imm:$lane)))>;
1890 // VMLSL : Vector Multiply Subtract Long (Q -= D * D)
1891 defm VMLSLs : N3VLInt3_QHS<0,1,0b1010,0, "vmlsl", "s", int_arm_neon_vmlsls>;
1892 defm VMLSLu : N3VLInt3_QHS<1,1,0b1010,0, "vmlsl", "u", int_arm_neon_vmlslu>;
1894 defm VMLSLsls : N3VLInt3SL_HS<0, 0b0110, "vmlsl", "s", int_arm_neon_vmlsls>;
1895 defm VMLSLslu : N3VLInt3SL_HS<1, 0b0110, "vmlsl", "u", int_arm_neon_vmlslu>;
1897 // VQDMLSL : Vector Saturating Doubling Multiply Subtract Long (Q -= D * D)
1898 defm VQDMLSL : N3VLInt3_HS<0, 1, 0b1011, 0, "vqdmlsl", "s",
1899 int_arm_neon_vqdmlsl>;
1900 defm VQDMLSLsl: N3VLInt3SL_HS<0, 0b111, "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
1902 // Vector Subtract Operations.
1904 // VSUB : Vector Subtract (integer and floating-point)
1905 defm VSUB : N3V_QHSD<1, 0, 0b1000, 0, IIC_VSUBiD, IIC_VSUBiQ,
1906 "vsub", "i", sub, 0>;
1907 def VSUBfd : N3VD<0, 0, 0b10, 0b1101, 0, IIC_VBIND, "vsub", "f32",
1908 v2f32, v2f32, fsub, 0>;
1909 def VSUBfq : N3VQ<0, 0, 0b10, 0b1101, 0, IIC_VBINQ, "vsub", "f32",
1910 v4f32, v4f32, fsub, 0>;
1911 // VSUBL : Vector Subtract Long (Q = D - D)
1912 defm VSUBLs : N3VLInt_QHS<0,1,0b0010,0, IIC_VSHLiD, "vsubl", "s",
1913 int_arm_neon_vsubls, 1>;
1914 defm VSUBLu : N3VLInt_QHS<1,1,0b0010,0, IIC_VSHLiD, "vsubl", "u",
1915 int_arm_neon_vsublu, 1>;
1916 // VSUBW : Vector Subtract Wide (Q = Q - D)
1917 defm VSUBWs : N3VWInt_QHS<0,1,0b0011,0, "vsubw", "s", int_arm_neon_vsubws, 0>;
1918 defm VSUBWu : N3VWInt_QHS<1,1,0b0011,0, "vsubw", "u", int_arm_neon_vsubwu, 0>;
1919 // VHSUB : Vector Halving Subtract
1920 defm VHSUBs : N3VInt_QHS<0, 0, 0b0010, 0, IIC_VBINi4D, IIC_VBINi4D,
1921 IIC_VBINi4Q, IIC_VBINi4Q,
1922 "vhsub", "s", int_arm_neon_vhsubs, 0>;
1923 defm VHSUBu : N3VInt_QHS<1, 0, 0b0010, 0, IIC_VBINi4D, IIC_VBINi4D,
1924 IIC_VBINi4Q, IIC_VBINi4Q,
1925 "vhsub", "u", int_arm_neon_vhsubu, 0>;
1926 // VQSUB : Vector Saturing Subtract
1927 defm VQSUBs : N3VInt_QHSD<0, 0, 0b0010, 1, IIC_VBINi4D, IIC_VBINi4D,
1928 IIC_VBINi4Q, IIC_VBINi4Q,
1929 "vqsub", "s", int_arm_neon_vqsubs, 0>;
1930 defm VQSUBu : N3VInt_QHSD<1, 0, 0b0010, 1, IIC_VBINi4D, IIC_VBINi4D,
1931 IIC_VBINi4Q, IIC_VBINi4Q,
1932 "vqsub", "u", int_arm_neon_vqsubu, 0>;
1933 // VSUBHN : Vector Subtract and Narrow Returning High Half (D = Q - Q)
1934 defm VSUBHN : N3VNInt_HSD<0,1,0b0110,0, "vsubhn", "i",
1935 int_arm_neon_vsubhn, 0>;
1936 // VRSUBHN : Vector Rounding Subtract and Narrow Returning High Half (D=Q-Q)
1937 defm VRSUBHN : N3VNInt_HSD<1,1,0b0110,0, "vrsubhn", "i",
1938 int_arm_neon_vrsubhn, 0>;
1940 // Vector Comparisons.
1942 // VCEQ : Vector Compare Equal
1943 defm VCEQ : N3V_QHS<1, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1944 IIC_VBINi4Q, "vceq", "i", NEONvceq, 1>;
1945 def VCEQfd : N3VD<0,0,0b00,0b1110,0, IIC_VBIND, "vceq", "f32", v2i32, v2f32,
1947 def VCEQfq : N3VQ<0,0,0b00,0b1110,0, IIC_VBINQ, "vceq", "f32", v4i32, v4f32,
1949 // VCGE : Vector Compare Greater Than or Equal
1950 defm VCGEs : N3V_QHS<0, 0, 0b0011, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1951 IIC_VBINi4Q, "vcge", "s", NEONvcge, 0>;
1952 defm VCGEu : N3V_QHS<1, 0, 0b0011, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1953 IIC_VBINi4Q, "vcge", "u", NEONvcgeu, 0>;
1954 def VCGEfd : N3VD<1,0,0b00,0b1110,0, IIC_VBIND, "vcge", "f32",
1955 v2i32, v2f32, NEONvcge, 0>;
1956 def VCGEfq : N3VQ<1,0,0b00,0b1110,0, IIC_VBINQ, "vcge", "f32", v4i32, v4f32,
1958 // VCGT : Vector Compare Greater Than
1959 defm VCGTs : N3V_QHS<0, 0, 0b0011, 0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1960 IIC_VBINi4Q, "vcgt", "s", NEONvcgt, 0>;
1961 defm VCGTu : N3V_QHS<1, 0, 0b0011, 0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1962 IIC_VBINi4Q, "vcgt", "u", NEONvcgtu, 0>;
1963 def VCGTfd : N3VD<1,0,0b10,0b1110,0, IIC_VBIND, "vcgt", "f32", v2i32, v2f32,
1965 def VCGTfq : N3VQ<1,0,0b10,0b1110,0, IIC_VBINQ, "vcgt", "f32", v4i32, v4f32,
1967 // VACGE : Vector Absolute Compare Greater Than or Equal (aka VCAGE)
1968 def VACGEd : N3VDInt<1, 0, 0b00, 0b1110, 1, IIC_VBIND, "vacge", "f32",
1969 v2i32, v2f32, int_arm_neon_vacged, 0>;
1970 def VACGEq : N3VQInt<1, 0, 0b00, 0b1110, 1, IIC_VBINQ, "vacge", "f32",
1971 v4i32, v4f32, int_arm_neon_vacgeq, 0>;
1972 // VACGT : Vector Absolute Compare Greater Than (aka VCAGT)
1973 def VACGTd : N3VDInt<1, 0, 0b10, 0b1110, 1, IIC_VBIND, "vacgt", "f32",
1974 v2i32, v2f32, int_arm_neon_vacgtd, 0>;
1975 def VACGTq : N3VQInt<1, 0, 0b10, 0b1110, 1, IIC_VBINQ, "vacgt", "f32",
1976 v4i32, v4f32, int_arm_neon_vacgtq, 0>;
1977 // VTST : Vector Test Bits
1978 defm VTST : N3V_QHS<0, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1979 IIC_VBINi4Q, "vtst", "", NEONvtst, 1>;
1981 // Vector Bitwise Operations.
1983 // VAND : Vector Bitwise AND
1984 def VANDd : N3VDX<0, 0, 0b00, 0b0001, 1, IIC_VBINiD, "vand",
1985 v2i32, v2i32, and, 1>;
1986 def VANDq : N3VQX<0, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "vand",
1987 v4i32, v4i32, and, 1>;
1989 // VEOR : Vector Bitwise Exclusive OR
1990 def VEORd : N3VDX<1, 0, 0b00, 0b0001, 1, IIC_VBINiD, "veor",
1991 v2i32, v2i32, xor, 1>;
1992 def VEORq : N3VQX<1, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "veor",
1993 v4i32, v4i32, xor, 1>;
1995 // VORR : Vector Bitwise OR
1996 def VORRd : N3VDX<0, 0, 0b10, 0b0001, 1, IIC_VBINiD, "vorr",
1997 v2i32, v2i32, or, 1>;
1998 def VORRq : N3VQX<0, 0, 0b10, 0b0001, 1, IIC_VBINiQ, "vorr",
1999 v4i32, v4i32, or, 1>;
2001 // VBIC : Vector Bitwise Bit Clear (AND NOT)
2002 def VBICd : N3VX<0, 0, 0b01, 0b0001, 0, 1, (outs DPR:$dst),
2003 (ins DPR:$src1, DPR:$src2), IIC_VBINiD,
2004 "vbic", "$dst, $src1, $src2", "",
2005 [(set DPR:$dst, (v2i32 (and DPR:$src1,
2006 (vnot_conv DPR:$src2))))]>;
2007 def VBICq : N3VX<0, 0, 0b01, 0b0001, 1, 1, (outs QPR:$dst),
2008 (ins QPR:$src1, QPR:$src2), IIC_VBINiQ,
2009 "vbic", "$dst, $src1, $src2", "",
2010 [(set QPR:$dst, (v4i32 (and QPR:$src1,
2011 (vnot_conv QPR:$src2))))]>;
2013 // VORN : Vector Bitwise OR NOT
2014 def VORNd : N3VX<0, 0, 0b11, 0b0001, 0, 1, (outs DPR:$dst),
2015 (ins DPR:$src1, DPR:$src2), IIC_VBINiD,
2016 "vorn", "$dst, $src1, $src2", "",
2017 [(set DPR:$dst, (v2i32 (or DPR:$src1,
2018 (vnot_conv DPR:$src2))))]>;
2019 def VORNq : N3VX<0, 0, 0b11, 0b0001, 1, 1, (outs QPR:$dst),
2020 (ins QPR:$src1, QPR:$src2), IIC_VBINiQ,
2021 "vorn", "$dst, $src1, $src2", "",
2022 [(set QPR:$dst, (v4i32 (or QPR:$src1,
2023 (vnot_conv QPR:$src2))))]>;
2025 // VMVN : Vector Bitwise NOT
2026 def VMVNd : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 0, 0,
2027 (outs DPR:$dst), (ins DPR:$src), IIC_VSHLiD,
2028 "vmvn", "$dst, $src", "",
2029 [(set DPR:$dst, (v2i32 (vnot DPR:$src)))]>;
2030 def VMVNq : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 1, 0,
2031 (outs QPR:$dst), (ins QPR:$src), IIC_VSHLiD,
2032 "vmvn", "$dst, $src", "",
2033 [(set QPR:$dst, (v4i32 (vnot QPR:$src)))]>;
2034 def : Pat<(v2i32 (vnot_conv DPR:$src)), (VMVNd DPR:$src)>;
2035 def : Pat<(v4i32 (vnot_conv QPR:$src)), (VMVNq QPR:$src)>;
2037 // VBSL : Vector Bitwise Select
2038 def VBSLd : N3VX<1, 0, 0b01, 0b0001, 0, 1, (outs DPR:$dst),
2039 (ins DPR:$src1, DPR:$src2, DPR:$src3), IIC_VCNTiD,
2040 "vbsl", "$dst, $src2, $src3", "$src1 = $dst",
2042 (v2i32 (or (and DPR:$src2, DPR:$src1),
2043 (and DPR:$src3, (vnot_conv DPR:$src1)))))]>;
2044 def VBSLq : N3VX<1, 0, 0b01, 0b0001, 1, 1, (outs QPR:$dst),
2045 (ins QPR:$src1, QPR:$src2, QPR:$src3), IIC_VCNTiQ,
2046 "vbsl", "$dst, $src2, $src3", "$src1 = $dst",
2048 (v4i32 (or (and QPR:$src2, QPR:$src1),
2049 (and QPR:$src3, (vnot_conv QPR:$src1)))))]>;
2051 // VBIF : Vector Bitwise Insert if False
2052 // like VBSL but with: "vbif $dst, $src3, $src1", "$src2 = $dst",
2053 def VBIFd : N3VX<1, 0, 0b11, 0b0001, 0, 1,
2054 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3),
2055 IIC_VBINiD, "vbif", "$dst, $src2, $src3", "$src1 = $dst",
2056 [/* For disassembly only; pattern left blank */]>;
2057 def VBIFq : N3VX<1, 0, 0b11, 0b0001, 1, 1,
2058 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3),
2059 IIC_VBINiQ, "vbif", "$dst, $src2, $src3", "$src1 = $dst",
2060 [/* For disassembly only; pattern left blank */]>;
2062 // VBIT : Vector Bitwise Insert if True
2063 // like VBSL but with: "vbit $dst, $src2, $src1", "$src3 = $dst",
2064 def VBITd : N3VX<1, 0, 0b10, 0b0001, 0, 1,
2065 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3),
2066 IIC_VBINiD, "vbit", "$dst, $src2, $src3", "$src1 = $dst",
2067 [/* For disassembly only; pattern left blank */]>;
2068 def VBITq : N3VX<1, 0, 0b10, 0b0001, 1, 1,
2069 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3),
2070 IIC_VBINiQ, "vbit", "$dst, $src2, $src3", "$src1 = $dst",
2071 [/* For disassembly only; pattern left blank */]>;
2073 // VBIT/VBIF are not yet implemented. The TwoAddress pass will not go looking
2074 // for equivalent operations with different register constraints; it just
2077 // Vector Absolute Differences.
2079 // VABD : Vector Absolute Difference
2080 defm VABDs : N3VInt_QHS<0, 0, 0b0111, 0, IIC_VBINi4D, IIC_VBINi4D,
2081 IIC_VBINi4Q, IIC_VBINi4Q,
2082 "vabd", "s", int_arm_neon_vabds, 0>;
2083 defm VABDu : N3VInt_QHS<1, 0, 0b0111, 0, IIC_VBINi4D, IIC_VBINi4D,
2084 IIC_VBINi4Q, IIC_VBINi4Q,
2085 "vabd", "u", int_arm_neon_vabdu, 0>;
2086 def VABDfd : N3VDInt<1, 0, 0b10, 0b1101, 0, IIC_VBIND,
2087 "vabd", "f32", v2f32, v2f32, int_arm_neon_vabds, 0>;
2088 def VABDfq : N3VQInt<1, 0, 0b10, 0b1101, 0, IIC_VBINQ,
2089 "vabd", "f32", v4f32, v4f32, int_arm_neon_vabds, 0>;
2091 // VABDL : Vector Absolute Difference Long (Q = | D - D |)
2092 defm VABDLs : N3VLInt_QHS<0,1,0b0111,0, IIC_VBINi4Q,
2093 "vabdl", "s", int_arm_neon_vabdls, 0>;
2094 defm VABDLu : N3VLInt_QHS<1,1,0b0111,0, IIC_VBINi4Q,
2095 "vabdl", "u", int_arm_neon_vabdlu, 0>;
2097 // VABA : Vector Absolute Difference and Accumulate
2098 defm VABAs : N3VInt3_QHS<0,0,0b0111,1, "vaba", "s", int_arm_neon_vabas>;
2099 defm VABAu : N3VInt3_QHS<1,0,0b0111,1, "vaba", "u", int_arm_neon_vabau>;
2101 // VABAL : Vector Absolute Difference and Accumulate Long (Q += | D - D |)
2102 defm VABALs : N3VLInt3_QHS<0,1,0b0101,0, "vabal", "s", int_arm_neon_vabals>;
2103 defm VABALu : N3VLInt3_QHS<1,1,0b0101,0, "vabal", "u", int_arm_neon_vabalu>;
2105 // Vector Maximum and Minimum.
2107 // VMAX : Vector Maximum
2108 defm VMAXs : N3VInt_QHS<0,0,0b0110,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2109 IIC_VBINi4Q, "vmax", "s", int_arm_neon_vmaxs, 1>;
2110 defm VMAXu : N3VInt_QHS<1,0,0b0110,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2111 IIC_VBINi4Q, "vmax", "u", int_arm_neon_vmaxu, 1>;
2112 def VMAXfd : N3VDInt<0, 0, 0b00, 0b1111, 0, IIC_VBIND, "vmax", "f32",
2113 v2f32, v2f32, int_arm_neon_vmaxs, 1>;
2114 def VMAXfq : N3VQInt<0, 0, 0b00, 0b1111, 0, IIC_VBINQ, "vmax", "f32",
2115 v4f32, v4f32, int_arm_neon_vmaxs, 1>;
2117 // VMIN : Vector Minimum
2118 defm VMINs : N3VInt_QHS<0,0,0b0110,1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2119 IIC_VBINi4Q, "vmin", "s", int_arm_neon_vmins, 1>;
2120 defm VMINu : N3VInt_QHS<1,0,0b0110,1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2121 IIC_VBINi4Q, "vmin", "u", int_arm_neon_vminu, 1>;
2122 def VMINfd : N3VDInt<0, 0, 0b10, 0b1111, 0, IIC_VBIND, "vmin", "f32",
2123 v2f32, v2f32, int_arm_neon_vmins, 1>;
2124 def VMINfq : N3VQInt<0, 0, 0b10, 0b1111, 0, IIC_VBINQ, "vmin", "f32",
2125 v4f32, v4f32, int_arm_neon_vmins, 1>;
2127 // Vector Pairwise Operations.
2129 // VPADD : Vector Pairwise Add
2130 def VPADDi8 : N3VDInt<0, 0, 0b00, 0b1011, 1, IIC_VBINiD, "vpadd", "i8",
2131 v8i8, v8i8, int_arm_neon_vpadd, 0>;
2132 def VPADDi16 : N3VDInt<0, 0, 0b01, 0b1011, 1, IIC_VBINiD, "vpadd", "i16",
2133 v4i16, v4i16, int_arm_neon_vpadd, 0>;
2134 def VPADDi32 : N3VDInt<0, 0, 0b10, 0b1011, 1, IIC_VBINiD, "vpadd", "i32",
2135 v2i32, v2i32, int_arm_neon_vpadd, 0>;
2136 def VPADDf : N3VDInt<1, 0, 0b00, 0b1101, 0, IIC_VBIND, "vpadd", "f32",
2137 v2f32, v2f32, int_arm_neon_vpadd, 0>;
2139 // VPADDL : Vector Pairwise Add Long
2140 defm VPADDLs : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00100, 0, "vpaddl", "s",
2141 int_arm_neon_vpaddls>;
2142 defm VPADDLu : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00101, 0, "vpaddl", "u",
2143 int_arm_neon_vpaddlu>;
2145 // VPADAL : Vector Pairwise Add and Accumulate Long
2146 defm VPADALs : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01100, 0, "vpadal", "s",
2147 int_arm_neon_vpadals>;
2148 defm VPADALu : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01101, 0, "vpadal", "u",
2149 int_arm_neon_vpadalu>;
2151 // VPMAX : Vector Pairwise Maximum
2152 def VPMAXs8 : N3VDInt<0, 0, 0b00, 0b1010, 0, IIC_VBINi4D, "vpmax", "s8",
2153 v8i8, v8i8, int_arm_neon_vpmaxs, 0>;
2154 def VPMAXs16 : N3VDInt<0, 0, 0b01, 0b1010, 0, IIC_VBINi4D, "vpmax", "s16",
2155 v4i16, v4i16, int_arm_neon_vpmaxs, 0>;
2156 def VPMAXs32 : N3VDInt<0, 0, 0b10, 0b1010, 0, IIC_VBINi4D, "vpmax", "s32",
2157 v2i32, v2i32, int_arm_neon_vpmaxs, 0>;
2158 def VPMAXu8 : N3VDInt<1, 0, 0b00, 0b1010, 0, IIC_VBINi4D, "vpmax", "u8",
2159 v8i8, v8i8, int_arm_neon_vpmaxu, 0>;
2160 def VPMAXu16 : N3VDInt<1, 0, 0b01, 0b1010, 0, IIC_VBINi4D, "vpmax", "u16",
2161 v4i16, v4i16, int_arm_neon_vpmaxu, 0>;
2162 def VPMAXu32 : N3VDInt<1, 0, 0b10, 0b1010, 0, IIC_VBINi4D, "vpmax", "u32",
2163 v2i32, v2i32, int_arm_neon_vpmaxu, 0>;
2164 def VPMAXf : N3VDInt<1, 0, 0b00, 0b1111, 0, IIC_VBINi4D, "vpmax", "f32",
2165 v2f32, v2f32, int_arm_neon_vpmaxs, 0>;
2167 // VPMIN : Vector Pairwise Minimum
2168 def VPMINs8 : N3VDInt<0, 0, 0b00, 0b1010, 1, IIC_VBINi4D, "vpmin", "s8",
2169 v8i8, v8i8, int_arm_neon_vpmins, 0>;
2170 def VPMINs16 : N3VDInt<0, 0, 0b01, 0b1010, 1, IIC_VBINi4D, "vpmin", "s16",
2171 v4i16, v4i16, int_arm_neon_vpmins, 0>;
2172 def VPMINs32 : N3VDInt<0, 0, 0b10, 0b1010, 1, IIC_VBINi4D, "vpmin", "s32",
2173 v2i32, v2i32, int_arm_neon_vpmins, 0>;
2174 def VPMINu8 : N3VDInt<1, 0, 0b00, 0b1010, 1, IIC_VBINi4D, "vpmin", "u8",
2175 v8i8, v8i8, int_arm_neon_vpminu, 0>;
2176 def VPMINu16 : N3VDInt<1, 0, 0b01, 0b1010, 1, IIC_VBINi4D, "vpmin", "u16",
2177 v4i16, v4i16, int_arm_neon_vpminu, 0>;
2178 def VPMINu32 : N3VDInt<1, 0, 0b10, 0b1010, 1, IIC_VBINi4D, "vpmin", "u32",
2179 v2i32, v2i32, int_arm_neon_vpminu, 0>;
2180 def VPMINf : N3VDInt<1, 0, 0b10, 0b1111, 0, IIC_VBINi4D, "vpmin", "f32",
2181 v2f32, v2f32, int_arm_neon_vpmins, 0>;
2183 // Vector Reciprocal and Reciprocal Square Root Estimate and Step.
2185 // VRECPE : Vector Reciprocal Estimate
2186 def VRECPEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
2187 IIC_VUNAD, "vrecpe", "u32",
2188 v2i32, v2i32, int_arm_neon_vrecpe>;
2189 def VRECPEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
2190 IIC_VUNAQ, "vrecpe", "u32",
2191 v4i32, v4i32, int_arm_neon_vrecpe>;
2192 def VRECPEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
2193 IIC_VUNAD, "vrecpe", "f32",
2194 v2f32, v2f32, int_arm_neon_vrecpe>;
2195 def VRECPEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
2196 IIC_VUNAQ, "vrecpe", "f32",
2197 v4f32, v4f32, int_arm_neon_vrecpe>;
2199 // VRECPS : Vector Reciprocal Step
2200 def VRECPSfd : N3VDInt<0, 0, 0b00, 0b1111, 1,
2201 IIC_VRECSD, "vrecps", "f32",
2202 v2f32, v2f32, int_arm_neon_vrecps, 1>;
2203 def VRECPSfq : N3VQInt<0, 0, 0b00, 0b1111, 1,
2204 IIC_VRECSQ, "vrecps", "f32",
2205 v4f32, v4f32, int_arm_neon_vrecps, 1>;
2207 // VRSQRTE : Vector Reciprocal Square Root Estimate
2208 def VRSQRTEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
2209 IIC_VUNAD, "vrsqrte", "u32",
2210 v2i32, v2i32, int_arm_neon_vrsqrte>;
2211 def VRSQRTEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
2212 IIC_VUNAQ, "vrsqrte", "u32",
2213 v4i32, v4i32, int_arm_neon_vrsqrte>;
2214 def VRSQRTEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
2215 IIC_VUNAD, "vrsqrte", "f32",
2216 v2f32, v2f32, int_arm_neon_vrsqrte>;
2217 def VRSQRTEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
2218 IIC_VUNAQ, "vrsqrte", "f32",
2219 v4f32, v4f32, int_arm_neon_vrsqrte>;
2221 // VRSQRTS : Vector Reciprocal Square Root Step
2222 def VRSQRTSfd : N3VDInt<0, 0, 0b10, 0b1111, 1,
2223 IIC_VRECSD, "vrsqrts", "f32",
2224 v2f32, v2f32, int_arm_neon_vrsqrts, 1>;
2225 def VRSQRTSfq : N3VQInt<0, 0, 0b10, 0b1111, 1,
2226 IIC_VRECSQ, "vrsqrts", "f32",
2227 v4f32, v4f32, int_arm_neon_vrsqrts, 1>;
2231 // VSHL : Vector Shift
2232 defm VSHLs : N3VInt_QHSD<0, 0, 0b0100, 0, IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ,
2233 IIC_VSHLiQ, "vshl", "s", int_arm_neon_vshifts, 0>;
2234 defm VSHLu : N3VInt_QHSD<1, 0, 0b0100, 0, IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ,
2235 IIC_VSHLiQ, "vshl", "u", int_arm_neon_vshiftu, 0>;
2236 // VSHL : Vector Shift Left (Immediate)
2237 defm VSHLi : N2VSh_QHSD<0, 1, 0b0101, 1, IIC_VSHLiD, "vshl", "i", NEONvshl>;
2238 // VSHR : Vector Shift Right (Immediate)
2239 defm VSHRs : N2VSh_QHSD<0, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "s", NEONvshrs>;
2240 defm VSHRu : N2VSh_QHSD<1, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "u", NEONvshru>;
2242 // VSHLL : Vector Shift Left Long
2243 defm VSHLLs : N2VLSh_QHS<0, 1, 0b1010, 0, 0, 1, "vshll", "s", NEONvshlls>;
2244 defm VSHLLu : N2VLSh_QHS<1, 1, 0b1010, 0, 0, 1, "vshll", "u", NEONvshllu>;
2246 // VSHLL : Vector Shift Left Long (with maximum shift count)
2247 class N2VLShMax<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
2248 bit op6, bit op4, string OpcodeStr, string Dt, ValueType ResTy,
2249 ValueType OpTy, SDNode OpNode>
2250 : N2VLSh<op24, op23, op11_8, op7, op6, op4, OpcodeStr, Dt,
2251 ResTy, OpTy, OpNode> {
2252 let Inst{21-16} = op21_16;
2254 def VSHLLi8 : N2VLShMax<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll", "i8",
2255 v8i16, v8i8, NEONvshlli>;
2256 def VSHLLi16 : N2VLShMax<1, 1, 0b110110, 0b0011, 0, 0, 0, "vshll", "i16",
2257 v4i32, v4i16, NEONvshlli>;
2258 def VSHLLi32 : N2VLShMax<1, 1, 0b111010, 0b0011, 0, 0, 0, "vshll", "i32",
2259 v2i64, v2i32, NEONvshlli>;
2261 // VSHRN : Vector Shift Right and Narrow
2262 defm VSHRN : N2VNSh_HSD<0,1,0b1000,0,0,1, IIC_VSHLiD, "vshrn", "i",
2265 // VRSHL : Vector Rounding Shift
2266 defm VRSHLs : N3VInt_QHSD<0,0,0b0101,0, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2267 IIC_VSHLi4Q, "vrshl", "s", int_arm_neon_vrshifts,0>;
2268 defm VRSHLu : N3VInt_QHSD<1,0,0b0101,0, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2269 IIC_VSHLi4Q, "vrshl", "u", int_arm_neon_vrshiftu,0>;
2270 // VRSHR : Vector Rounding Shift Right
2271 defm VRSHRs : N2VSh_QHSD<0,1,0b0010,1, IIC_VSHLi4D, "vrshr", "s", NEONvrshrs>;
2272 defm VRSHRu : N2VSh_QHSD<1,1,0b0010,1, IIC_VSHLi4D, "vrshr", "u", NEONvrshru>;
2274 // VRSHRN : Vector Rounding Shift Right and Narrow
2275 defm VRSHRN : N2VNSh_HSD<0, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vrshrn", "i",
2278 // VQSHL : Vector Saturating Shift
2279 defm VQSHLs : N3VInt_QHSD<0,0,0b0100,1, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2280 IIC_VSHLi4Q, "vqshl", "s", int_arm_neon_vqshifts,0>;
2281 defm VQSHLu : N3VInt_QHSD<1,0,0b0100,1, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2282 IIC_VSHLi4Q, "vqshl", "u", int_arm_neon_vqshiftu,0>;
2283 // VQSHL : Vector Saturating Shift Left (Immediate)
2284 defm VQSHLsi : N2VSh_QHSD<0,1,0b0111,1, IIC_VSHLi4D, "vqshl", "s", NEONvqshls>;
2285 defm VQSHLui : N2VSh_QHSD<1,1,0b0111,1, IIC_VSHLi4D, "vqshl", "u", NEONvqshlu>;
2286 // VQSHLU : Vector Saturating Shift Left (Immediate, Unsigned)
2287 defm VQSHLsu : N2VSh_QHSD<1,1,0b0110,1, IIC_VSHLi4D, "vqshlu","s",NEONvqshlsu>;
2289 // VQSHRN : Vector Saturating Shift Right and Narrow
2290 defm VQSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "s",
2292 defm VQSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "u",
2295 // VQSHRUN : Vector Saturating Shift Right and Narrow (Unsigned)
2296 defm VQSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 0, 1, IIC_VSHLi4D, "vqshrun", "s",
2299 // VQRSHL : Vector Saturating Rounding Shift
2300 defm VQRSHLs : N3VInt_QHSD<0,0,0b0101,1, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2301 IIC_VSHLi4Q, "vqrshl", "s",
2302 int_arm_neon_vqrshifts, 0>;
2303 defm VQRSHLu : N3VInt_QHSD<1,0,0b0101,1, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2304 IIC_VSHLi4Q, "vqrshl", "u",
2305 int_arm_neon_vqrshiftu, 0>;
2307 // VQRSHRN : Vector Saturating Rounding Shift Right and Narrow
2308 defm VQRSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "s",
2310 defm VQRSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "u",
2313 // VQRSHRUN : Vector Saturating Rounding Shift Right and Narrow (Unsigned)
2314 defm VQRSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vqrshrun", "s",
2317 // VSRA : Vector Shift Right and Accumulate
2318 defm VSRAs : N2VShAdd_QHSD<0, 1, 0b0001, 1, "vsra", "s", NEONvshrs>;
2319 defm VSRAu : N2VShAdd_QHSD<1, 1, 0b0001, 1, "vsra", "u", NEONvshru>;
2320 // VRSRA : Vector Rounding Shift Right and Accumulate
2321 defm VRSRAs : N2VShAdd_QHSD<0, 1, 0b0011, 1, "vrsra", "s", NEONvrshrs>;
2322 defm VRSRAu : N2VShAdd_QHSD<1, 1, 0b0011, 1, "vrsra", "u", NEONvrshru>;
2324 // VSLI : Vector Shift Left and Insert
2325 defm VSLI : N2VShIns_QHSD<1, 1, 0b0101, 1, "vsli", NEONvsli>;
2326 // VSRI : Vector Shift Right and Insert
2327 defm VSRI : N2VShIns_QHSD<1, 1, 0b0100, 1, "vsri", NEONvsri>;
2329 // Vector Absolute and Saturating Absolute.
2331 // VABS : Vector Absolute Value
2332 defm VABS : N2VInt_QHS<0b11, 0b11, 0b01, 0b00110, 0,
2333 IIC_VUNAiD, IIC_VUNAiQ, "vabs", "s",
2335 def VABSfd : N2VDInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
2336 IIC_VUNAD, "vabs", "f32",
2337 v2f32, v2f32, int_arm_neon_vabs>;
2338 def VABSfq : N2VQInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
2339 IIC_VUNAQ, "vabs", "f32",
2340 v4f32, v4f32, int_arm_neon_vabs>;
2342 // VQABS : Vector Saturating Absolute Value
2343 defm VQABS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01110, 0,
2344 IIC_VQUNAiD, IIC_VQUNAiQ, "vqabs", "s",
2345 int_arm_neon_vqabs>;
2349 def vneg : PatFrag<(ops node:$in), (sub immAllZerosV, node:$in)>;
2350 def vneg_conv : PatFrag<(ops node:$in), (sub immAllZerosV_bc, node:$in)>;
2352 class VNEGD<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
2353 : N2V<0b11, 0b11, size, 0b01, 0b00111, 0, 0, (outs DPR:$dst), (ins DPR:$src),
2354 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
2355 [(set DPR:$dst, (Ty (vneg DPR:$src)))]>;
2356 class VNEGQ<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
2357 : N2V<0b11, 0b11, size, 0b01, 0b00111, 1, 0, (outs QPR:$dst), (ins QPR:$src),
2358 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
2359 [(set QPR:$dst, (Ty (vneg QPR:$src)))]>;
2361 // VNEG : Vector Negate
2362 def VNEGs8d : VNEGD<0b00, "vneg", "s8", v8i8>;
2363 def VNEGs16d : VNEGD<0b01, "vneg", "s16", v4i16>;
2364 def VNEGs32d : VNEGD<0b10, "vneg", "s32", v2i32>;
2365 def VNEGs8q : VNEGQ<0b00, "vneg", "s8", v16i8>;
2366 def VNEGs16q : VNEGQ<0b01, "vneg", "s16", v8i16>;
2367 def VNEGs32q : VNEGQ<0b10, "vneg", "s32", v4i32>;
2369 // VNEG : Vector Negate (floating-point)
2370 def VNEGfd : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
2371 (outs DPR:$dst), (ins DPR:$src), IIC_VUNAD,
2372 "vneg", "f32", "$dst, $src", "",
2373 [(set DPR:$dst, (v2f32 (fneg DPR:$src)))]>;
2374 def VNEGf32q : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 1, 0,
2375 (outs QPR:$dst), (ins QPR:$src), IIC_VUNAQ,
2376 "vneg", "f32", "$dst, $src", "",
2377 [(set QPR:$dst, (v4f32 (fneg QPR:$src)))]>;
2379 def : Pat<(v8i8 (vneg_conv DPR:$src)), (VNEGs8d DPR:$src)>;
2380 def : Pat<(v4i16 (vneg_conv DPR:$src)), (VNEGs16d DPR:$src)>;
2381 def : Pat<(v2i32 (vneg_conv DPR:$src)), (VNEGs32d DPR:$src)>;
2382 def : Pat<(v16i8 (vneg_conv QPR:$src)), (VNEGs8q QPR:$src)>;
2383 def : Pat<(v8i16 (vneg_conv QPR:$src)), (VNEGs16q QPR:$src)>;
2384 def : Pat<(v4i32 (vneg_conv QPR:$src)), (VNEGs32q QPR:$src)>;
2386 // VQNEG : Vector Saturating Negate
2387 defm VQNEG : N2VInt_QHS<0b11, 0b11, 0b00, 0b01111, 0,
2388 IIC_VQUNAiD, IIC_VQUNAiQ, "vqneg", "s",
2389 int_arm_neon_vqneg>;
2391 // Vector Bit Counting Operations.
2393 // VCLS : Vector Count Leading Sign Bits
2394 defm VCLS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01000, 0,
2395 IIC_VCNTiD, IIC_VCNTiQ, "vcls", "s",
2397 // VCLZ : Vector Count Leading Zeros
2398 defm VCLZ : N2VInt_QHS<0b11, 0b11, 0b00, 0b01001, 0,
2399 IIC_VCNTiD, IIC_VCNTiQ, "vclz", "i",
2401 // VCNT : Vector Count One Bits
2402 def VCNTd : N2VDInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
2403 IIC_VCNTiD, "vcnt", "8",
2404 v8i8, v8i8, int_arm_neon_vcnt>;
2405 def VCNTq : N2VQInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
2406 IIC_VCNTiQ, "vcnt", "8",
2407 v16i8, v16i8, int_arm_neon_vcnt>;
2409 // Vector Move Operations.
2411 // VMOV : Vector Move (Register)
2413 def VMOVDneon: N3VX<0, 0, 0b10, 0b0001, 0, 1, (outs DPR:$dst), (ins DPR:$src),
2414 IIC_VMOVD, "vmov", "$dst, $src", "", []>;
2415 def VMOVQ : N3VX<0, 0, 0b10, 0b0001, 1, 1, (outs QPR:$dst), (ins QPR:$src),
2416 IIC_VMOVD, "vmov", "$dst, $src", "", []>;
2418 // VMOV : Vector Move (Immediate)
2420 // VMOV_get_imm8 xform function: convert build_vector to VMOV.i8 imm.
2421 def VMOV_get_imm8 : SDNodeXForm<build_vector, [{
2422 return ARM::getVMOVImm(N, 1, *CurDAG);
2424 def vmovImm8 : PatLeaf<(build_vector), [{
2425 return ARM::getVMOVImm(N, 1, *CurDAG).getNode() != 0;
2428 // VMOV_get_imm16 xform function: convert build_vector to VMOV.i16 imm.
2429 def VMOV_get_imm16 : SDNodeXForm<build_vector, [{
2430 return ARM::getVMOVImm(N, 2, *CurDAG);
2432 def vmovImm16 : PatLeaf<(build_vector), [{
2433 return ARM::getVMOVImm(N, 2, *CurDAG).getNode() != 0;
2434 }], VMOV_get_imm16>;
2436 // VMOV_get_imm32 xform function: convert build_vector to VMOV.i32 imm.
2437 def VMOV_get_imm32 : SDNodeXForm<build_vector, [{
2438 return ARM::getVMOVImm(N, 4, *CurDAG);
2440 def vmovImm32 : PatLeaf<(build_vector), [{
2441 return ARM::getVMOVImm(N, 4, *CurDAG).getNode() != 0;
2442 }], VMOV_get_imm32>;
2444 // VMOV_get_imm64 xform function: convert build_vector to VMOV.i64 imm.
2445 def VMOV_get_imm64 : SDNodeXForm<build_vector, [{
2446 return ARM::getVMOVImm(N, 8, *CurDAG);
2448 def vmovImm64 : PatLeaf<(build_vector), [{
2449 return ARM::getVMOVImm(N, 8, *CurDAG).getNode() != 0;
2450 }], VMOV_get_imm64>;
2452 // Note: Some of the cmode bits in the following VMOV instructions need to
2453 // be encoded based on the immed values.
2455 def VMOVv8i8 : N1ModImm<1, 0b000, 0b1110, 0, 0, 0, 1, (outs DPR:$dst),
2456 (ins h8imm:$SIMM), IIC_VMOVImm,
2457 "vmov", "i8", "$dst, $SIMM", "",
2458 [(set DPR:$dst, (v8i8 vmovImm8:$SIMM))]>;
2459 def VMOVv16i8 : N1ModImm<1, 0b000, 0b1110, 0, 1, 0, 1, (outs QPR:$dst),
2460 (ins h8imm:$SIMM), IIC_VMOVImm,
2461 "vmov", "i8", "$dst, $SIMM", "",
2462 [(set QPR:$dst, (v16i8 vmovImm8:$SIMM))]>;
2464 def VMOVv4i16 : N1ModImm<1, 0b000, {1,0,?,?}, 0, 0, {?}, 1, (outs DPR:$dst),
2465 (ins h16imm:$SIMM), IIC_VMOVImm,
2466 "vmov", "i16", "$dst, $SIMM", "",
2467 [(set DPR:$dst, (v4i16 vmovImm16:$SIMM))]>;
2468 def VMOVv8i16 : N1ModImm<1, 0b000, {1,0,?,?}, 0, 1, {?}, 1, (outs QPR:$dst),
2469 (ins h16imm:$SIMM), IIC_VMOVImm,
2470 "vmov", "i16", "$dst, $SIMM", "",
2471 [(set QPR:$dst, (v8i16 vmovImm16:$SIMM))]>;
2473 def VMOVv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, {?}, 1, (outs DPR:$dst),
2474 (ins h32imm:$SIMM), IIC_VMOVImm,
2475 "vmov", "i32", "$dst, $SIMM", "",
2476 [(set DPR:$dst, (v2i32 vmovImm32:$SIMM))]>;
2477 def VMOVv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, {?}, 1, (outs QPR:$dst),
2478 (ins h32imm:$SIMM), IIC_VMOVImm,
2479 "vmov", "i32", "$dst, $SIMM", "",
2480 [(set QPR:$dst, (v4i32 vmovImm32:$SIMM))]>;
2482 def VMOVv1i64 : N1ModImm<1, 0b000, 0b1110, 0, 0, 1, 1, (outs DPR:$dst),
2483 (ins h64imm:$SIMM), IIC_VMOVImm,
2484 "vmov", "i64", "$dst, $SIMM", "",
2485 [(set DPR:$dst, (v1i64 vmovImm64:$SIMM))]>;
2486 def VMOVv2i64 : N1ModImm<1, 0b000, 0b1110, 0, 1, 1, 1, (outs QPR:$dst),
2487 (ins h64imm:$SIMM), IIC_VMOVImm,
2488 "vmov", "i64", "$dst, $SIMM", "",
2489 [(set QPR:$dst, (v2i64 vmovImm64:$SIMM))]>;
2491 // VMOV : Vector Get Lane (move scalar to ARM core register)
2493 def VGETLNs8 : NVGetLane<{1,1,1,0,0,1,?,1}, 0b1011, {?,?},
2494 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2495 IIC_VMOVSI, "vmov", "s8", "$dst, $src[$lane]",
2496 [(set GPR:$dst, (NEONvgetlanes (v8i8 DPR:$src),
2498 def VGETLNs16 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, {?,1},
2499 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2500 IIC_VMOVSI, "vmov", "s16", "$dst, $src[$lane]",
2501 [(set GPR:$dst, (NEONvgetlanes (v4i16 DPR:$src),
2503 def VGETLNu8 : NVGetLane<{1,1,1,0,1,1,?,1}, 0b1011, {?,?},
2504 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2505 IIC_VMOVSI, "vmov", "u8", "$dst, $src[$lane]",
2506 [(set GPR:$dst, (NEONvgetlaneu (v8i8 DPR:$src),
2508 def VGETLNu16 : NVGetLane<{1,1,1,0,1,0,?,1}, 0b1011, {?,1},
2509 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2510 IIC_VMOVSI, "vmov", "u16", "$dst, $src[$lane]",
2511 [(set GPR:$dst, (NEONvgetlaneu (v4i16 DPR:$src),
2513 def VGETLNi32 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, 0b00,
2514 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2515 IIC_VMOVSI, "vmov", "32", "$dst, $src[$lane]",
2516 [(set GPR:$dst, (extractelt (v2i32 DPR:$src),
2518 // def VGETLNf32: see FMRDH and FMRDL in ARMInstrVFP.td
2519 def : Pat<(NEONvgetlanes (v16i8 QPR:$src), imm:$lane),
2520 (VGETLNs8 (v8i8 (EXTRACT_SUBREG QPR:$src,
2521 (DSubReg_i8_reg imm:$lane))),
2522 (SubReg_i8_lane imm:$lane))>;
2523 def : Pat<(NEONvgetlanes (v8i16 QPR:$src), imm:$lane),
2524 (VGETLNs16 (v4i16 (EXTRACT_SUBREG QPR:$src,
2525 (DSubReg_i16_reg imm:$lane))),
2526 (SubReg_i16_lane imm:$lane))>;
2527 def : Pat<(NEONvgetlaneu (v16i8 QPR:$src), imm:$lane),
2528 (VGETLNu8 (v8i8 (EXTRACT_SUBREG QPR:$src,
2529 (DSubReg_i8_reg imm:$lane))),
2530 (SubReg_i8_lane imm:$lane))>;
2531 def : Pat<(NEONvgetlaneu (v8i16 QPR:$src), imm:$lane),
2532 (VGETLNu16 (v4i16 (EXTRACT_SUBREG QPR:$src,
2533 (DSubReg_i16_reg imm:$lane))),
2534 (SubReg_i16_lane imm:$lane))>;
2535 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
2536 (VGETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src,
2537 (DSubReg_i32_reg imm:$lane))),
2538 (SubReg_i32_lane imm:$lane))>;
2539 def : Pat<(extractelt (v2f32 DPR:$src1), imm:$src2),
2540 (EXTRACT_SUBREG (v2f32 (COPY_TO_REGCLASS (v2f32 DPR:$src1),DPR_VFP2)),
2541 (SSubReg_f32_reg imm:$src2))>;
2542 def : Pat<(extractelt (v4f32 QPR:$src1), imm:$src2),
2543 (EXTRACT_SUBREG (v4f32 (COPY_TO_REGCLASS (v4f32 QPR:$src1),QPR_VFP2)),
2544 (SSubReg_f32_reg imm:$src2))>;
2545 //def : Pat<(extractelt (v2i64 QPR:$src1), imm:$src2),
2546 // (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
2547 def : Pat<(extractelt (v2f64 QPR:$src1), imm:$src2),
2548 (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
2551 // VMOV : Vector Set Lane (move ARM core register to scalar)
2553 let Constraints = "$src1 = $dst" in {
2554 def VSETLNi8 : NVSetLane<{1,1,1,0,0,1,?,0}, 0b1011, {?,?}, (outs DPR:$dst),
2555 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
2556 IIC_VMOVISL, "vmov", "8", "$dst[$lane], $src2",
2557 [(set DPR:$dst, (vector_insert (v8i8 DPR:$src1),
2558 GPR:$src2, imm:$lane))]>;
2559 def VSETLNi16 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, {?,1}, (outs DPR:$dst),
2560 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
2561 IIC_VMOVISL, "vmov", "16", "$dst[$lane], $src2",
2562 [(set DPR:$dst, (vector_insert (v4i16 DPR:$src1),
2563 GPR:$src2, imm:$lane))]>;
2564 def VSETLNi32 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, 0b00, (outs DPR:$dst),
2565 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
2566 IIC_VMOVISL, "vmov", "32", "$dst[$lane], $src2",
2567 [(set DPR:$dst, (insertelt (v2i32 DPR:$src1),
2568 GPR:$src2, imm:$lane))]>;
2570 def : Pat<(vector_insert (v16i8 QPR:$src1), GPR:$src2, imm:$lane),
2571 (v16i8 (INSERT_SUBREG QPR:$src1,
2572 (VSETLNi8 (v8i8 (EXTRACT_SUBREG QPR:$src1,
2573 (DSubReg_i8_reg imm:$lane))),
2574 GPR:$src2, (SubReg_i8_lane imm:$lane)),
2575 (DSubReg_i8_reg imm:$lane)))>;
2576 def : Pat<(vector_insert (v8i16 QPR:$src1), GPR:$src2, imm:$lane),
2577 (v8i16 (INSERT_SUBREG QPR:$src1,
2578 (VSETLNi16 (v4i16 (EXTRACT_SUBREG QPR:$src1,
2579 (DSubReg_i16_reg imm:$lane))),
2580 GPR:$src2, (SubReg_i16_lane imm:$lane)),
2581 (DSubReg_i16_reg imm:$lane)))>;
2582 def : Pat<(insertelt (v4i32 QPR:$src1), GPR:$src2, imm:$lane),
2583 (v4i32 (INSERT_SUBREG QPR:$src1,
2584 (VSETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src1,
2585 (DSubReg_i32_reg imm:$lane))),
2586 GPR:$src2, (SubReg_i32_lane imm:$lane)),
2587 (DSubReg_i32_reg imm:$lane)))>;
2589 def : Pat<(v2f32 (insertelt DPR:$src1, SPR:$src2, imm:$src3)),
2590 (INSERT_SUBREG (v2f32 (COPY_TO_REGCLASS DPR:$src1, DPR_VFP2)),
2591 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
2592 def : Pat<(v4f32 (insertelt QPR:$src1, SPR:$src2, imm:$src3)),
2593 (INSERT_SUBREG (v4f32 (COPY_TO_REGCLASS QPR:$src1, QPR_VFP2)),
2594 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
2596 //def : Pat<(v2i64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
2597 // (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
2598 def : Pat<(v2f64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
2599 (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
2601 def : Pat<(v2f32 (scalar_to_vector SPR:$src)),
2602 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$src, arm_ssubreg_0)>;
2603 def : Pat<(v2f64 (scalar_to_vector DPR:$src)),
2604 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), DPR:$src, arm_dsubreg_0)>;
2605 def : Pat<(v4f32 (scalar_to_vector SPR:$src)),
2606 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), SPR:$src, arm_ssubreg_0)>;
2608 def : Pat<(v8i8 (scalar_to_vector GPR:$src)),
2609 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
2610 def : Pat<(v4i16 (scalar_to_vector GPR:$src)),
2611 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
2612 def : Pat<(v2i32 (scalar_to_vector GPR:$src)),
2613 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
2615 def : Pat<(v16i8 (scalar_to_vector GPR:$src)),
2616 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
2617 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
2619 def : Pat<(v8i16 (scalar_to_vector GPR:$src)),
2620 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
2621 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
2623 def : Pat<(v4i32 (scalar_to_vector GPR:$src)),
2624 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
2625 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
2628 // VDUP : Vector Duplicate (from ARM core register to all elements)
2630 class VDUPD<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
2631 : NVDup<opcod1, 0b1011, opcod3, (outs DPR:$dst), (ins GPR:$src),
2632 IIC_VMOVIS, "vdup", Dt, "$dst, $src",
2633 [(set DPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
2634 class VDUPQ<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
2635 : NVDup<opcod1, 0b1011, opcod3, (outs QPR:$dst), (ins GPR:$src),
2636 IIC_VMOVIS, "vdup", Dt, "$dst, $src",
2637 [(set QPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
2639 def VDUP8d : VDUPD<0b11101100, 0b00, "8", v8i8>;
2640 def VDUP16d : VDUPD<0b11101000, 0b01, "16", v4i16>;
2641 def VDUP32d : VDUPD<0b11101000, 0b00, "32", v2i32>;
2642 def VDUP8q : VDUPQ<0b11101110, 0b00, "8", v16i8>;
2643 def VDUP16q : VDUPQ<0b11101010, 0b01, "16", v8i16>;
2644 def VDUP32q : VDUPQ<0b11101010, 0b00, "32", v4i32>;
2646 def VDUPfd : NVDup<0b11101000, 0b1011, 0b00, (outs DPR:$dst), (ins GPR:$src),
2647 IIC_VMOVIS, "vdup", "32", "$dst, $src",
2648 [(set DPR:$dst, (v2f32 (NEONvdup
2649 (f32 (bitconvert GPR:$src)))))]>;
2650 def VDUPfq : NVDup<0b11101010, 0b1011, 0b00, (outs QPR:$dst), (ins GPR:$src),
2651 IIC_VMOVIS, "vdup", "32", "$dst, $src",
2652 [(set QPR:$dst, (v4f32 (NEONvdup
2653 (f32 (bitconvert GPR:$src)))))]>;
2655 // VDUP : Vector Duplicate Lane (from scalar to all elements)
2657 class VDUPLND<bits<2> op19_18, bits<2> op17_16,
2658 string OpcodeStr, string Dt, ValueType Ty>
2659 : N2V<0b11, 0b11, op19_18, op17_16, 0b11000, 0, 0,
2660 (outs DPR:$dst), (ins DPR:$src, nohash_imm:$lane), IIC_VMOVD,
2661 OpcodeStr, Dt, "$dst, $src[$lane]", "",
2662 [(set DPR:$dst, (Ty (NEONvduplane (Ty DPR:$src), imm:$lane)))]>;
2664 class VDUPLNQ<bits<2> op19_18, bits<2> op17_16, string OpcodeStr, string Dt,
2665 ValueType ResTy, ValueType OpTy>
2666 : N2V<0b11, 0b11, op19_18, op17_16, 0b11000, 1, 0,
2667 (outs QPR:$dst), (ins DPR:$src, nohash_imm:$lane), IIC_VMOVD,
2668 OpcodeStr, Dt, "$dst, $src[$lane]", "",
2669 [(set QPR:$dst, (ResTy (NEONvduplane (OpTy DPR:$src), imm:$lane)))]>;
2671 // Inst{19-16} is partially specified depending on the element size.
2673 def VDUPLN8d : VDUPLND<{?,?}, {?,1}, "vdup", "8", v8i8>;
2674 def VDUPLN16d : VDUPLND<{?,?}, {1,0}, "vdup", "16", v4i16>;
2675 def VDUPLN32d : VDUPLND<{?,1}, {0,0}, "vdup", "32", v2i32>;
2676 def VDUPLNfd : VDUPLND<{?,1}, {0,0}, "vdup", "32", v2f32>;
2677 def VDUPLN8q : VDUPLNQ<{?,?}, {?,1}, "vdup", "8", v16i8, v8i8>;
2678 def VDUPLN16q : VDUPLNQ<{?,?}, {1,0}, "vdup", "16", v8i16, v4i16>;
2679 def VDUPLN32q : VDUPLNQ<{?,1}, {0,0}, "vdup", "32", v4i32, v2i32>;
2680 def VDUPLNfq : VDUPLNQ<{?,1}, {0,0}, "vdup", "32", v4f32, v2f32>;
2682 def : Pat<(v16i8 (NEONvduplane (v16i8 QPR:$src), imm:$lane)),
2683 (v16i8 (VDUPLN8q (v8i8 (EXTRACT_SUBREG QPR:$src,
2684 (DSubReg_i8_reg imm:$lane))),
2685 (SubReg_i8_lane imm:$lane)))>;
2686 def : Pat<(v8i16 (NEONvduplane (v8i16 QPR:$src), imm:$lane)),
2687 (v8i16 (VDUPLN16q (v4i16 (EXTRACT_SUBREG QPR:$src,
2688 (DSubReg_i16_reg imm:$lane))),
2689 (SubReg_i16_lane imm:$lane)))>;
2690 def : Pat<(v4i32 (NEONvduplane (v4i32 QPR:$src), imm:$lane)),
2691 (v4i32 (VDUPLN32q (v2i32 (EXTRACT_SUBREG QPR:$src,
2692 (DSubReg_i32_reg imm:$lane))),
2693 (SubReg_i32_lane imm:$lane)))>;
2694 def : Pat<(v4f32 (NEONvduplane (v4f32 QPR:$src), imm:$lane)),
2695 (v4f32 (VDUPLNfq (v2f32 (EXTRACT_SUBREG QPR:$src,
2696 (DSubReg_i32_reg imm:$lane))),
2697 (SubReg_i32_lane imm:$lane)))>;
2699 def VDUPfdf : N2V<0b11, 0b11, {?,1}, {0,0}, 0b11000, 0, 0,
2700 (outs DPR:$dst), (ins SPR:$src),
2701 IIC_VMOVD, "vdup", "32", "$dst, ${src:lane}", "",
2702 [(set DPR:$dst, (v2f32 (NEONvdup (f32 SPR:$src))))]>;
2704 def VDUPfqf : N2V<0b11, 0b11, {?,1}, {0,0}, 0b11000, 1, 0,
2705 (outs QPR:$dst), (ins SPR:$src),
2706 IIC_VMOVD, "vdup", "32", "$dst, ${src:lane}", "",
2707 [(set QPR:$dst, (v4f32 (NEONvdup (f32 SPR:$src))))]>;
2709 def : Pat<(v2i64 (NEONvduplane (v2i64 QPR:$src), imm:$lane)),
2710 (INSERT_SUBREG QPR:$src,
2711 (i64 (EXTRACT_SUBREG QPR:$src,
2712 (DSubReg_f64_reg imm:$lane))),
2713 (DSubReg_f64_other_reg imm:$lane))>;
2714 def : Pat<(v2f64 (NEONvduplane (v2f64 QPR:$src), imm:$lane)),
2715 (INSERT_SUBREG QPR:$src,
2716 (f64 (EXTRACT_SUBREG QPR:$src,
2717 (DSubReg_f64_reg imm:$lane))),
2718 (DSubReg_f64_other_reg imm:$lane))>;
2720 // VMOVN : Vector Narrowing Move
2721 defm VMOVN : N2VNInt_HSD<0b11,0b11,0b10,0b00100,0,0, IIC_VMOVD,
2722 "vmovn", "i", int_arm_neon_vmovn>;
2723 // VQMOVN : Vector Saturating Narrowing Move
2724 defm VQMOVNs : N2VNInt_HSD<0b11,0b11,0b10,0b00101,0,0, IIC_VQUNAiD,
2725 "vqmovn", "s", int_arm_neon_vqmovns>;
2726 defm VQMOVNu : N2VNInt_HSD<0b11,0b11,0b10,0b00101,1,0, IIC_VQUNAiD,
2727 "vqmovn", "u", int_arm_neon_vqmovnu>;
2728 defm VQMOVNsu : N2VNInt_HSD<0b11,0b11,0b10,0b00100,1,0, IIC_VQUNAiD,
2729 "vqmovun", "s", int_arm_neon_vqmovnsu>;
2730 // VMOVL : Vector Lengthening Move
2731 defm VMOVLs : N2VLInt_QHS<0b01,0b10100,0,1, "vmovl", "s",
2732 int_arm_neon_vmovls>;
2733 defm VMOVLu : N2VLInt_QHS<0b11,0b10100,0,1, "vmovl", "u",
2734 int_arm_neon_vmovlu>;
2736 // Vector Conversions.
2738 // VCVT : Vector Convert Between Floating-Point and Integers
2739 def VCVTf2sd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
2740 v2i32, v2f32, fp_to_sint>;
2741 def VCVTf2ud : N2VD<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
2742 v2i32, v2f32, fp_to_uint>;
2743 def VCVTs2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
2744 v2f32, v2i32, sint_to_fp>;
2745 def VCVTu2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
2746 v2f32, v2i32, uint_to_fp>;
2748 def VCVTf2sq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
2749 v4i32, v4f32, fp_to_sint>;
2750 def VCVTf2uq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
2751 v4i32, v4f32, fp_to_uint>;
2752 def VCVTs2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
2753 v4f32, v4i32, sint_to_fp>;
2754 def VCVTu2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
2755 v4f32, v4i32, uint_to_fp>;
2757 // VCVT : Vector Convert Between Floating-Point and Fixed-Point.
2758 def VCVTf2xsd : N2VCvtD<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
2759 v2i32, v2f32, int_arm_neon_vcvtfp2fxs>;
2760 def VCVTf2xud : N2VCvtD<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
2761 v2i32, v2f32, int_arm_neon_vcvtfp2fxu>;
2762 def VCVTxs2fd : N2VCvtD<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
2763 v2f32, v2i32, int_arm_neon_vcvtfxs2fp>;
2764 def VCVTxu2fd : N2VCvtD<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
2765 v2f32, v2i32, int_arm_neon_vcvtfxu2fp>;
2767 def VCVTf2xsq : N2VCvtQ<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
2768 v4i32, v4f32, int_arm_neon_vcvtfp2fxs>;
2769 def VCVTf2xuq : N2VCvtQ<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
2770 v4i32, v4f32, int_arm_neon_vcvtfp2fxu>;
2771 def VCVTxs2fq : N2VCvtQ<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
2772 v4f32, v4i32, int_arm_neon_vcvtfxs2fp>;
2773 def VCVTxu2fq : N2VCvtQ<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
2774 v4f32, v4i32, int_arm_neon_vcvtfxu2fp>;
2778 // VREV64 : Vector Reverse elements within 64-bit doublewords
2780 class VREV64D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
2781 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 0, 0, (outs DPR:$dst),
2782 (ins DPR:$src), IIC_VMOVD,
2783 OpcodeStr, Dt, "$dst, $src", "",
2784 [(set DPR:$dst, (Ty (NEONvrev64 (Ty DPR:$src))))]>;
2785 class VREV64Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
2786 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 1, 0, (outs QPR:$dst),
2787 (ins QPR:$src), IIC_VMOVD,
2788 OpcodeStr, Dt, "$dst, $src", "",
2789 [(set QPR:$dst, (Ty (NEONvrev64 (Ty QPR:$src))))]>;
2791 def VREV64d8 : VREV64D<0b00, "vrev64", "8", v8i8>;
2792 def VREV64d16 : VREV64D<0b01, "vrev64", "16", v4i16>;
2793 def VREV64d32 : VREV64D<0b10, "vrev64", "32", v2i32>;
2794 def VREV64df : VREV64D<0b10, "vrev64", "32", v2f32>;
2796 def VREV64q8 : VREV64Q<0b00, "vrev64", "8", v16i8>;
2797 def VREV64q16 : VREV64Q<0b01, "vrev64", "16", v8i16>;
2798 def VREV64q32 : VREV64Q<0b10, "vrev64", "32", v4i32>;
2799 def VREV64qf : VREV64Q<0b10, "vrev64", "32", v4f32>;
2801 // VREV32 : Vector Reverse elements within 32-bit words
2803 class VREV32D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
2804 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 0, 0, (outs DPR:$dst),
2805 (ins DPR:$src), IIC_VMOVD,
2806 OpcodeStr, Dt, "$dst, $src", "",
2807 [(set DPR:$dst, (Ty (NEONvrev32 (Ty DPR:$src))))]>;
2808 class VREV32Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
2809 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 1, 0, (outs QPR:$dst),
2810 (ins QPR:$src), IIC_VMOVD,
2811 OpcodeStr, Dt, "$dst, $src", "",
2812 [(set QPR:$dst, (Ty (NEONvrev32 (Ty QPR:$src))))]>;
2814 def VREV32d8 : VREV32D<0b00, "vrev32", "8", v8i8>;
2815 def VREV32d16 : VREV32D<0b01, "vrev32", "16", v4i16>;
2817 def VREV32q8 : VREV32Q<0b00, "vrev32", "8", v16i8>;
2818 def VREV32q16 : VREV32Q<0b01, "vrev32", "16", v8i16>;
2820 // VREV16 : Vector Reverse elements within 16-bit halfwords
2822 class VREV16D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
2823 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 0, 0, (outs DPR:$dst),
2824 (ins DPR:$src), IIC_VMOVD,
2825 OpcodeStr, Dt, "$dst, $src", "",
2826 [(set DPR:$dst, (Ty (NEONvrev16 (Ty DPR:$src))))]>;
2827 class VREV16Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
2828 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 1, 0, (outs QPR:$dst),
2829 (ins QPR:$src), IIC_VMOVD,
2830 OpcodeStr, Dt, "$dst, $src", "",
2831 [(set QPR:$dst, (Ty (NEONvrev16 (Ty QPR:$src))))]>;
2833 def VREV16d8 : VREV16D<0b00, "vrev16", "8", v8i8>;
2834 def VREV16q8 : VREV16Q<0b00, "vrev16", "8", v16i8>;
2836 // Other Vector Shuffles.
2838 // VEXT : Vector Extract
2840 class VEXTd<string OpcodeStr, string Dt, ValueType Ty>
2841 : N3V<0,1,0b11,{?,?,?,?},0,0, (outs DPR:$dst),
2842 (ins DPR:$lhs, DPR:$rhs, i32imm:$index), IIC_VEXTD,
2843 OpcodeStr, Dt, "$dst, $lhs, $rhs, $index", "",
2844 [(set DPR:$dst, (Ty (NEONvext (Ty DPR:$lhs),
2845 (Ty DPR:$rhs), imm:$index)))]>;
2847 class VEXTq<string OpcodeStr, string Dt, ValueType Ty>
2848 : N3V<0,1,0b11,{?,?,?,?},1,0, (outs QPR:$dst),
2849 (ins QPR:$lhs, QPR:$rhs, i32imm:$index), IIC_VEXTQ,
2850 OpcodeStr, Dt, "$dst, $lhs, $rhs, $index", "",
2851 [(set QPR:$dst, (Ty (NEONvext (Ty QPR:$lhs),
2852 (Ty QPR:$rhs), imm:$index)))]>;
2854 def VEXTd8 : VEXTd<"vext", "8", v8i8>;
2855 def VEXTd16 : VEXTd<"vext", "16", v4i16>;
2856 def VEXTd32 : VEXTd<"vext", "32", v2i32>;
2857 def VEXTdf : VEXTd<"vext", "32", v2f32>;
2859 def VEXTq8 : VEXTq<"vext", "8", v16i8>;
2860 def VEXTq16 : VEXTq<"vext", "16", v8i16>;
2861 def VEXTq32 : VEXTq<"vext", "32", v4i32>;
2862 def VEXTqf : VEXTq<"vext", "32", v4f32>;
2864 // VTRN : Vector Transpose
2866 def VTRNd8 : N2VDShuffle<0b00, 0b00001, "vtrn", "8">;
2867 def VTRNd16 : N2VDShuffle<0b01, 0b00001, "vtrn", "16">;
2868 def VTRNd32 : N2VDShuffle<0b10, 0b00001, "vtrn", "32">;
2870 def VTRNq8 : N2VQShuffle<0b00, 0b00001, IIC_VPERMQ, "vtrn", "8">;
2871 def VTRNq16 : N2VQShuffle<0b01, 0b00001, IIC_VPERMQ, "vtrn", "16">;
2872 def VTRNq32 : N2VQShuffle<0b10, 0b00001, IIC_VPERMQ, "vtrn", "32">;
2874 // VUZP : Vector Unzip (Deinterleave)
2876 def VUZPd8 : N2VDShuffle<0b00, 0b00010, "vuzp", "8">;
2877 def VUZPd16 : N2VDShuffle<0b01, 0b00010, "vuzp", "16">;
2878 def VUZPd32 : N2VDShuffle<0b10, 0b00010, "vuzp", "32">;
2880 def VUZPq8 : N2VQShuffle<0b00, 0b00010, IIC_VPERMQ3, "vuzp", "8">;
2881 def VUZPq16 : N2VQShuffle<0b01, 0b00010, IIC_VPERMQ3, "vuzp", "16">;
2882 def VUZPq32 : N2VQShuffle<0b10, 0b00010, IIC_VPERMQ3, "vuzp", "32">;
2884 // VZIP : Vector Zip (Interleave)
2886 def VZIPd8 : N2VDShuffle<0b00, 0b00011, "vzip", "8">;
2887 def VZIPd16 : N2VDShuffle<0b01, 0b00011, "vzip", "16">;
2888 def VZIPd32 : N2VDShuffle<0b10, 0b00011, "vzip", "32">;
2890 def VZIPq8 : N2VQShuffle<0b00, 0b00011, IIC_VPERMQ3, "vzip", "8">;
2891 def VZIPq16 : N2VQShuffle<0b01, 0b00011, IIC_VPERMQ3, "vzip", "16">;
2892 def VZIPq32 : N2VQShuffle<0b10, 0b00011, IIC_VPERMQ3, "vzip", "32">;
2894 // Vector Table Lookup and Table Extension.
2896 // VTBL : Vector Table Lookup
2898 : N3V<1,1,0b11,0b1000,0,0, (outs DPR:$dst),
2899 (ins DPR:$tbl1, DPR:$src), IIC_VTB1,
2900 "vtbl", "8", "$dst, \\{$tbl1\\}, $src", "",
2901 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl1 DPR:$tbl1, DPR:$src)))]>;
2902 let hasExtraSrcRegAllocReq = 1 in {
2904 : N3V<1,1,0b11,0b1001,0,0, (outs DPR:$dst),
2905 (ins DPR:$tbl1, DPR:$tbl2, DPR:$src), IIC_VTB2,
2906 "vtbl", "8", "$dst, \\{$tbl1, $tbl2\\}, $src", "",
2907 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl2
2908 DPR:$tbl1, DPR:$tbl2, DPR:$src)))]>;
2910 : N3V<1,1,0b11,0b1010,0,0, (outs DPR:$dst),
2911 (ins DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$src), IIC_VTB3,
2912 "vtbl", "8", "$dst, \\{$tbl1, $tbl2, $tbl3\\}, $src", "",
2913 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl3
2914 DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$src)))]>;
2916 : N3V<1,1,0b11,0b1011,0,0, (outs DPR:$dst),
2917 (ins DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$src), IIC_VTB4,
2918 "vtbl", "8", "$dst, \\{$tbl1, $tbl2, $tbl3, $tbl4\\}, $src", "",
2919 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl4 DPR:$tbl1, DPR:$tbl2,
2920 DPR:$tbl3, DPR:$tbl4, DPR:$src)))]>;
2921 } // hasExtraSrcRegAllocReq = 1
2923 // VTBX : Vector Table Extension
2925 : N3V<1,1,0b11,0b1000,1,0, (outs DPR:$dst),
2926 (ins DPR:$orig, DPR:$tbl1, DPR:$src), IIC_VTBX1,
2927 "vtbx", "8", "$dst, \\{$tbl1\\}, $src", "$orig = $dst",
2928 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx1
2929 DPR:$orig, DPR:$tbl1, DPR:$src)))]>;
2930 let hasExtraSrcRegAllocReq = 1 in {
2932 : N3V<1,1,0b11,0b1001,1,0, (outs DPR:$dst),
2933 (ins DPR:$orig, DPR:$tbl1, DPR:$tbl2, DPR:$src), IIC_VTBX2,
2934 "vtbx", "8", "$dst, \\{$tbl1, $tbl2\\}, $src", "$orig = $dst",
2935 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx2
2936 DPR:$orig, DPR:$tbl1, DPR:$tbl2, DPR:$src)))]>;
2938 : N3V<1,1,0b11,0b1010,1,0, (outs DPR:$dst),
2939 (ins DPR:$orig, DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$src), IIC_VTBX3,
2940 "vtbx", "8", "$dst, \\{$tbl1, $tbl2, $tbl3\\}, $src", "$orig = $dst",
2941 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx3 DPR:$orig, DPR:$tbl1,
2942 DPR:$tbl2, DPR:$tbl3, DPR:$src)))]>;
2944 : N3V<1,1,0b11,0b1011,1,0, (outs DPR:$dst), (ins DPR:$orig, DPR:$tbl1,
2945 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$src), IIC_VTBX4,
2946 "vtbx", "8", "$dst, \\{$tbl1, $tbl2, $tbl3, $tbl4\\}, $src",
2948 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx4 DPR:$orig, DPR:$tbl1,
2949 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$src)))]>;
2950 } // hasExtraSrcRegAllocReq = 1
2952 //===----------------------------------------------------------------------===//
2953 // NEON instructions for single-precision FP math
2954 //===----------------------------------------------------------------------===//
2956 class N2VSPat<SDNode OpNode, ValueType ResTy, ValueType OpTy, NeonI Inst>
2957 : NEONFPPat<(ResTy (OpNode SPR:$a)),
2958 (EXTRACT_SUBREG (Inst (INSERT_SUBREG (OpTy (IMPLICIT_DEF)),
2959 SPR:$a, arm_ssubreg_0)),
2962 class N3VSPat<SDNode OpNode, NeonI Inst>
2963 : NEONFPPat<(f32 (OpNode SPR:$a, SPR:$b)),
2964 (EXTRACT_SUBREG (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
2965 SPR:$a, arm_ssubreg_0),
2966 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
2967 SPR:$b, arm_ssubreg_0)),
2970 class N3VSMulOpPat<SDNode MulNode, SDNode OpNode, NeonI Inst>
2971 : NEONFPPat<(f32 (OpNode SPR:$acc, (f32 (MulNode SPR:$a, SPR:$b)))),
2972 (EXTRACT_SUBREG (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
2973 SPR:$acc, arm_ssubreg_0),
2974 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
2975 SPR:$a, arm_ssubreg_0),
2976 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
2977 SPR:$b, arm_ssubreg_0)),
2980 // These need separate instructions because they must use DPR_VFP2 register
2981 // class which have SPR sub-registers.
2983 // Vector Add Operations used for single-precision FP
2984 let neverHasSideEffects = 1 in
2985 def VADDfd_sfp : N3VS<0,0,0b00,0b1101,0, "vadd", "f32", v2f32, v2f32, fadd, 1>;
2986 def : N3VSPat<fadd, VADDfd_sfp>;
2988 // Vector Sub Operations used for single-precision FP
2989 let neverHasSideEffects = 1 in
2990 def VSUBfd_sfp : N3VS<0,0,0b10,0b1101,0, "vsub", "f32", v2f32, v2f32, fsub, 0>;
2991 def : N3VSPat<fsub, VSUBfd_sfp>;
2993 // Vector Multiply Operations used for single-precision FP
2994 let neverHasSideEffects = 1 in
2995 def VMULfd_sfp : N3VS<1,0,0b00,0b1101,1, "vmul", "f32", v2f32, v2f32, fmul, 1>;
2996 def : N3VSPat<fmul, VMULfd_sfp>;
2998 // Vector Multiply-Accumulate/Subtract used for single-precision FP
2999 // vml[as].f32 can cause 4-8 cycle stalls in following ASIMD instructions, so
3000 // we want to avoid them for now. e.g., alternating vmla/vadd instructions.
3002 //let neverHasSideEffects = 1 in
3003 //def VMLAfd_sfp : N3VSMulOp<0,0,0b00,0b1101,1, IIC_VMACD, "vmla", "f32",
3004 // v2f32, fmul, fadd>;
3005 //def : N3VSMulOpPat<fmul, fadd, VMLAfd_sfp>;
3007 //let neverHasSideEffects = 1 in
3008 //def VMLSfd_sfp : N3VSMulOp<0,0,0b10,0b1101,1, IIC_VMACD, "vmls", "f32",
3009 // v2f32, fmul, fsub>;
3010 //def : N3VSMulOpPat<fmul, fsub, VMLSfd_sfp>;
3012 // Vector Absolute used for single-precision FP
3013 let neverHasSideEffects = 1 in
3014 def VABSfd_sfp : N2V<0b11, 0b11, 0b10, 0b01, 0b01110, 0, 0,
3015 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), IIC_VUNAD,
3016 "vabs", "f32", "$dst, $src", "", []>;
3017 def : N2VSPat<fabs, f32, v2f32, VABSfd_sfp>;
3019 // Vector Negate used for single-precision FP
3020 let neverHasSideEffects = 1 in
3021 def VNEGfd_sfp : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
3022 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), IIC_VUNAD,
3023 "vneg", "f32", "$dst, $src", "", []>;
3024 def : N2VSPat<fneg, f32, v2f32, VNEGfd_sfp>;
3026 // Vector Convert between single-precision FP and integer
3027 let neverHasSideEffects = 1 in
3028 def VCVTf2sd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
3029 v2i32, v2f32, fp_to_sint>;
3030 def : N2VSPat<arm_ftosi, f32, v2f32, VCVTf2sd_sfp>;
3032 let neverHasSideEffects = 1 in
3033 def VCVTf2ud_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
3034 v2i32, v2f32, fp_to_uint>;
3035 def : N2VSPat<arm_ftoui, f32, v2f32, VCVTf2ud_sfp>;
3037 let neverHasSideEffects = 1 in
3038 def VCVTs2fd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
3039 v2f32, v2i32, sint_to_fp>;
3040 def : N2VSPat<arm_sitof, f32, v2i32, VCVTs2fd_sfp>;
3042 let neverHasSideEffects = 1 in
3043 def VCVTu2fd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
3044 v2f32, v2i32, uint_to_fp>;
3045 def : N2VSPat<arm_uitof, f32, v2i32, VCVTu2fd_sfp>;
3047 //===----------------------------------------------------------------------===//
3048 // Non-Instruction Patterns
3049 //===----------------------------------------------------------------------===//
3052 def : Pat<(v1i64 (bitconvert (v2i32 DPR:$src))), (v1i64 DPR:$src)>;
3053 def : Pat<(v1i64 (bitconvert (v4i16 DPR:$src))), (v1i64 DPR:$src)>;
3054 def : Pat<(v1i64 (bitconvert (v8i8 DPR:$src))), (v1i64 DPR:$src)>;
3055 def : Pat<(v1i64 (bitconvert (f64 DPR:$src))), (v1i64 DPR:$src)>;
3056 def : Pat<(v1i64 (bitconvert (v2f32 DPR:$src))), (v1i64 DPR:$src)>;
3057 def : Pat<(v2i32 (bitconvert (v1i64 DPR:$src))), (v2i32 DPR:$src)>;
3058 def : Pat<(v2i32 (bitconvert (v4i16 DPR:$src))), (v2i32 DPR:$src)>;
3059 def : Pat<(v2i32 (bitconvert (v8i8 DPR:$src))), (v2i32 DPR:$src)>;
3060 def : Pat<(v2i32 (bitconvert (f64 DPR:$src))), (v2i32 DPR:$src)>;
3061 def : Pat<(v2i32 (bitconvert (v2f32 DPR:$src))), (v2i32 DPR:$src)>;
3062 def : Pat<(v4i16 (bitconvert (v1i64 DPR:$src))), (v4i16 DPR:$src)>;
3063 def : Pat<(v4i16 (bitconvert (v2i32 DPR:$src))), (v4i16 DPR:$src)>;
3064 def : Pat<(v4i16 (bitconvert (v8i8 DPR:$src))), (v4i16 DPR:$src)>;
3065 def : Pat<(v4i16 (bitconvert (f64 DPR:$src))), (v4i16 DPR:$src)>;
3066 def : Pat<(v4i16 (bitconvert (v2f32 DPR:$src))), (v4i16 DPR:$src)>;
3067 def : Pat<(v8i8 (bitconvert (v1i64 DPR:$src))), (v8i8 DPR:$src)>;
3068 def : Pat<(v8i8 (bitconvert (v2i32 DPR:$src))), (v8i8 DPR:$src)>;
3069 def : Pat<(v8i8 (bitconvert (v4i16 DPR:$src))), (v8i8 DPR:$src)>;
3070 def : Pat<(v8i8 (bitconvert (f64 DPR:$src))), (v8i8 DPR:$src)>;
3071 def : Pat<(v8i8 (bitconvert (v2f32 DPR:$src))), (v8i8 DPR:$src)>;
3072 def : Pat<(f64 (bitconvert (v1i64 DPR:$src))), (f64 DPR:$src)>;
3073 def : Pat<(f64 (bitconvert (v2i32 DPR:$src))), (f64 DPR:$src)>;
3074 def : Pat<(f64 (bitconvert (v4i16 DPR:$src))), (f64 DPR:$src)>;
3075 def : Pat<(f64 (bitconvert (v8i8 DPR:$src))), (f64 DPR:$src)>;
3076 def : Pat<(f64 (bitconvert (v2f32 DPR:$src))), (f64 DPR:$src)>;
3077 def : Pat<(v2f32 (bitconvert (f64 DPR:$src))), (v2f32 DPR:$src)>;
3078 def : Pat<(v2f32 (bitconvert (v1i64 DPR:$src))), (v2f32 DPR:$src)>;
3079 def : Pat<(v2f32 (bitconvert (v2i32 DPR:$src))), (v2f32 DPR:$src)>;
3080 def : Pat<(v2f32 (bitconvert (v4i16 DPR:$src))), (v2f32 DPR:$src)>;
3081 def : Pat<(v2f32 (bitconvert (v8i8 DPR:$src))), (v2f32 DPR:$src)>;
3083 def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (v2i64 QPR:$src)>;
3084 def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (v2i64 QPR:$src)>;
3085 def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (v2i64 QPR:$src)>;
3086 def : Pat<(v2i64 (bitconvert (v2f64 QPR:$src))), (v2i64 QPR:$src)>;
3087 def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (v2i64 QPR:$src)>;
3088 def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (v4i32 QPR:$src)>;
3089 def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (v4i32 QPR:$src)>;
3090 def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (v4i32 QPR:$src)>;
3091 def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (v4i32 QPR:$src)>;
3092 def : Pat<(v4i32 (bitconvert (v4f32 QPR:$src))), (v4i32 QPR:$src)>;
3093 def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (v8i16 QPR:$src)>;
3094 def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (v8i16 QPR:$src)>;
3095 def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (v8i16 QPR:$src)>;
3096 def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (v8i16 QPR:$src)>;
3097 def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (v8i16 QPR:$src)>;
3098 def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (v16i8 QPR:$src)>;
3099 def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (v16i8 QPR:$src)>;
3100 def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (v16i8 QPR:$src)>;
3101 def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (v16i8 QPR:$src)>;
3102 def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (v16i8 QPR:$src)>;
3103 def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (v4f32 QPR:$src)>;
3104 def : Pat<(v4f32 (bitconvert (v4i32 QPR:$src))), (v4f32 QPR:$src)>;
3105 def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (v4f32 QPR:$src)>;
3106 def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (v4f32 QPR:$src)>;
3107 def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (v4f32 QPR:$src)>;
3108 def : Pat<(v2f64 (bitconvert (v2i64 QPR:$src))), (v2f64 QPR:$src)>;
3109 def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (v2f64 QPR:$src)>;
3110 def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 QPR:$src)>;
3111 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 QPR:$src)>;
3112 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 QPR:$src)>;