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:$src))))]>;
589 // Basic 2-register intrinsics: single-, double- and quad-register.
590 class N2VSInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
591 bits<2> op17_16, bits<5> op11_7, bit op4,
592 InstrItinClass itin, string OpcodeStr, string Dt,
593 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
594 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
595 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), itin,
596 OpcodeStr, Dt, "$dst, $src", "", []>;
597 class N2VDInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
598 bits<2> op17_16, bits<5> op11_7, bit op4,
599 InstrItinClass itin, string OpcodeStr, string Dt,
600 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
601 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
602 (ins DPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
603 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
604 class N2VQInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
605 bits<2> op17_16, bits<5> op11_7, bit op4,
606 InstrItinClass itin, string OpcodeStr, string Dt,
607 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
608 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
609 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
610 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
612 // Narrow 2-register intrinsics.
613 class N2VNInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
614 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
615 InstrItinClass itin, string OpcodeStr, string Dt,
616 ValueType TyD, ValueType TyQ, Intrinsic IntOp>
617 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$dst),
618 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
619 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src))))]>;
621 // Long 2-register intrinsics (currently only used for VMOVL).
622 class N2VLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
623 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
624 InstrItinClass itin, string OpcodeStr, string Dt,
625 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
626 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$dst),
627 (ins DPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
628 [(set QPR:$dst, (TyQ (IntOp (TyD DPR:$src))))]>;
630 // 2-register shuffles (VTRN/VZIP/VUZP), both double- and quad-register.
631 class N2VDShuffle<bits<2> op19_18, bits<5> op11_7, string OpcodeStr, string Dt>
632 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 0, 0, (outs DPR:$dst1, DPR:$dst2),
633 (ins DPR:$src1, DPR:$src2), IIC_VPERMD,
634 OpcodeStr, Dt, "$dst1, $dst2",
635 "$src1 = $dst1, $src2 = $dst2", []>;
636 class N2VQShuffle<bits<2> op19_18, bits<5> op11_7,
637 InstrItinClass itin, string OpcodeStr, string Dt>
638 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 1, 0, (outs QPR:$dst1, QPR:$dst2),
639 (ins QPR:$src1, QPR:$src2), itin, OpcodeStr, Dt, "$dst1, $dst2",
640 "$src1 = $dst1, $src2 = $dst2", []>;
642 // Basic 3-register operations: single-, double- and quad-register.
643 class N3VS<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
644 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
645 SDNode OpNode, bit Commutable>
646 : N3V<op24, op23, op21_20, op11_8, 0, op4,
647 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src1, DPR_VFP2:$src2), IIC_VBIND,
648 OpcodeStr, Dt, "$dst, $src1, $src2", "", []> {
649 let isCommutable = Commutable;
652 class N3VD<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
653 InstrItinClass itin, string OpcodeStr, string Dt,
654 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
655 : N3V<op24, op23, op21_20, op11_8, 0, op4,
656 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), itin,
657 OpcodeStr, Dt, "$dst, $src1, $src2", "",
658 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src1), (OpTy DPR:$src2))))]> {
659 let isCommutable = Commutable;
661 // Same as N3VD but no data type.
662 class N3VDX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
663 InstrItinClass itin, string OpcodeStr,
664 ValueType ResTy, ValueType OpTy,
665 SDNode OpNode, bit Commutable>
666 : N3VX<op24, op23, op21_20, op11_8, 0, op4,
667 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), itin,
668 OpcodeStr, "$dst, $src1, $src2", "",
669 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src1), (OpTy DPR:$src2))))]>{
670 let isCommutable = Commutable;
672 class N3VDSL<bits<2> op21_20, bits<4> op11_8,
673 InstrItinClass itin, string OpcodeStr, string Dt,
674 ValueType Ty, SDNode ShOp>
675 : N3V<0, 1, op21_20, op11_8, 1, 0,
676 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
677 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
679 (Ty (ShOp (Ty DPR:$src1),
680 (Ty (NEONvduplane (Ty DPR_VFP2:$src2), imm:$lane)))))]>{
681 let isCommutable = 0;
683 class N3VDSL16<bits<2> op21_20, bits<4> op11_8,
684 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
685 : N3V<0, 1, op21_20, op11_8, 1, 0,
686 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
687 IIC_VMULi16D, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
689 (Ty (ShOp (Ty DPR:$src1),
690 (Ty (NEONvduplane (Ty DPR_8:$src2), imm:$lane)))))]> {
691 let isCommutable = 0;
694 class N3VQ<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
695 InstrItinClass itin, string OpcodeStr, string Dt,
696 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
697 : N3V<op24, op23, op21_20, op11_8, 1, op4,
698 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), itin,
699 OpcodeStr, Dt, "$dst, $src1, $src2", "",
700 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src1), (OpTy QPR:$src2))))]> {
701 let isCommutable = Commutable;
703 class N3VQX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
704 InstrItinClass itin, string OpcodeStr,
705 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
706 : N3VX<op24, op23, op21_20, op11_8, 1, op4,
707 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), itin,
708 OpcodeStr, "$dst, $src1, $src2", "",
709 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src1), (OpTy QPR:$src2))))]>{
710 let isCommutable = Commutable;
712 class N3VQSL<bits<2> op21_20, bits<4> op11_8,
713 InstrItinClass itin, string OpcodeStr, string Dt,
714 ValueType ResTy, ValueType OpTy, SDNode ShOp>
715 : N3V<1, 1, op21_20, op11_8, 1, 0,
716 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
717 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
718 [(set (ResTy QPR:$dst),
719 (ResTy (ShOp (ResTy QPR:$src1),
720 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
722 let isCommutable = 0;
724 class N3VQSL16<bits<2> op21_20, bits<4> op11_8, string OpcodeStr, string Dt,
725 ValueType ResTy, ValueType OpTy, SDNode ShOp>
726 : N3V<1, 1, op21_20, op11_8, 1, 0,
727 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
728 IIC_VMULi16Q, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
729 [(set (ResTy QPR:$dst),
730 (ResTy (ShOp (ResTy QPR:$src1),
731 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
733 let isCommutable = 0;
736 // Basic 3-register intrinsics, both double- and quad-register.
737 class N3VDInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
738 InstrItinClass itin, string OpcodeStr, string Dt,
739 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
740 : N3V<op24, op23, op21_20, op11_8, 0, op4,
741 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), itin,
742 OpcodeStr, Dt, "$dst, $src1, $src2", "",
743 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src1), (OpTy DPR:$src2))))]> {
744 let isCommutable = Commutable;
746 class N3VDIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
747 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
748 : N3V<0, 1, op21_20, op11_8, 1, 0,
749 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
750 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
752 (Ty (IntOp (Ty DPR:$src1),
753 (Ty (NEONvduplane (Ty DPR_VFP2:$src2),
755 let isCommutable = 0;
757 class N3VDIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
758 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
759 : N3V<0, 1, op21_20, op11_8, 1, 0,
760 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
761 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
763 (Ty (IntOp (Ty DPR:$src1),
764 (Ty (NEONvduplane (Ty DPR_8:$src2),
766 let isCommutable = 0;
769 class N3VQInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
770 InstrItinClass itin, string OpcodeStr, string Dt,
771 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
772 : N3V<op24, op23, op21_20, op11_8, 1, op4,
773 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), itin,
774 OpcodeStr, Dt, "$dst, $src1, $src2", "",
775 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src1), (OpTy QPR:$src2))))]> {
776 let isCommutable = Commutable;
778 class N3VQIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
779 string OpcodeStr, string Dt,
780 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
781 : N3V<1, 1, op21_20, op11_8, 1, 0,
782 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
783 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
784 [(set (ResTy QPR:$dst),
785 (ResTy (IntOp (ResTy QPR:$src1),
786 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
788 let isCommutable = 0;
790 class N3VQIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
791 string OpcodeStr, string Dt,
792 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
793 : N3V<1, 1, op21_20, op11_8, 1, 0,
794 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
795 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
796 [(set (ResTy QPR:$dst),
797 (ResTy (IntOp (ResTy QPR:$src1),
798 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
800 let isCommutable = 0;
803 // Multiply-Add/Sub operations: single-, double- and quad-register.
804 class N3VSMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
805 InstrItinClass itin, string OpcodeStr, string Dt,
806 ValueType Ty, SDNode MulOp, SDNode OpNode>
807 : N3V<op24, op23, op21_20, op11_8, 0, op4,
808 (outs DPR_VFP2:$dst),
809 (ins DPR_VFP2:$src1, DPR_VFP2:$src2, DPR_VFP2:$src3), itin,
810 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst", []>;
812 class N3VDMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
813 InstrItinClass itin, string OpcodeStr, string Dt,
814 ValueType Ty, SDNode MulOp, SDNode OpNode>
815 : N3V<op24, op23, op21_20, op11_8, 0, op4,
816 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3), itin,
817 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
818 [(set DPR:$dst, (Ty (OpNode DPR:$src1,
819 (Ty (MulOp DPR:$src2, DPR:$src3)))))]>;
820 class N3VDMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
821 string OpcodeStr, string Dt,
822 ValueType Ty, SDNode MulOp, SDNode ShOp>
823 : N3V<0, 1, op21_20, op11_8, 1, 0,
825 (ins DPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane), itin,
826 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
828 (Ty (ShOp (Ty DPR:$src1),
829 (Ty (MulOp DPR:$src2,
830 (Ty (NEONvduplane (Ty DPR_VFP2:$src3),
832 class N3VDMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
833 string OpcodeStr, string Dt,
834 ValueType Ty, SDNode MulOp, SDNode ShOp>
835 : N3V<0, 1, op21_20, op11_8, 1, 0,
837 (ins DPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane), itin,
838 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
840 (Ty (ShOp (Ty DPR:$src1),
841 (Ty (MulOp DPR:$src2,
842 (Ty (NEONvduplane (Ty DPR_8:$src3),
845 class N3VQMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
846 InstrItinClass itin, string OpcodeStr, string Dt, ValueType Ty,
847 SDNode MulOp, SDNode OpNode>
848 : N3V<op24, op23, op21_20, op11_8, 1, op4,
849 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3), itin,
850 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
851 [(set QPR:$dst, (Ty (OpNode QPR:$src1,
852 (Ty (MulOp QPR:$src2, QPR:$src3)))))]>;
853 class N3VQMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
854 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
855 SDNode MulOp, SDNode ShOp>
856 : N3V<1, 1, op21_20, op11_8, 1, 0,
858 (ins QPR:$src1, QPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane), itin,
859 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
860 [(set (ResTy QPR:$dst),
861 (ResTy (ShOp (ResTy QPR:$src1),
862 (ResTy (MulOp QPR:$src2,
863 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src3),
865 class N3VQMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
866 string OpcodeStr, string Dt,
867 ValueType ResTy, ValueType OpTy,
868 SDNode MulOp, SDNode ShOp>
869 : N3V<1, 1, op21_20, op11_8, 1, 0,
871 (ins QPR:$src1, QPR:$src2, DPR_8:$src3, nohash_imm:$lane), itin,
872 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
873 [(set (ResTy QPR:$dst),
874 (ResTy (ShOp (ResTy QPR:$src1),
875 (ResTy (MulOp QPR:$src2,
876 (ResTy (NEONvduplane (OpTy DPR_8:$src3),
879 // Neon 3-argument intrinsics, both double- and quad-register.
880 // The destination register is also used as the first source operand register.
881 class N3VDInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
882 InstrItinClass itin, string OpcodeStr, string Dt,
883 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
884 : N3V<op24, op23, op21_20, op11_8, 0, op4,
885 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3), itin,
886 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
887 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src1),
888 (OpTy DPR:$src2), (OpTy DPR:$src3))))]>;
889 class N3VQInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
890 InstrItinClass itin, string OpcodeStr, string Dt,
891 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
892 : N3V<op24, op23, op21_20, op11_8, 1, op4,
893 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3), itin,
894 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
895 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src1),
896 (OpTy QPR:$src2), (OpTy QPR:$src3))))]>;
898 // Neon Long 3-argument intrinsic. The destination register is
899 // a quad-register and is also used as the first source operand register.
900 class N3VLInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
901 InstrItinClass itin, string OpcodeStr, string Dt,
902 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
903 : N3V<op24, op23, op21_20, op11_8, 0, op4,
904 (outs QPR:$dst), (ins QPR:$src1, DPR:$src2, DPR:$src3), itin,
905 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
907 (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$src2), (TyD DPR:$src3))))]>;
908 class N3VLInt3SL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
909 string OpcodeStr, string Dt,
910 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
911 : N3V<op24, 1, op21_20, op11_8, 1, 0,
913 (ins QPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane), itin,
914 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
915 [(set (ResTy QPR:$dst),
916 (ResTy (IntOp (ResTy QPR:$src1),
918 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src3),
920 class N3VLInt3SL16<bit op24, bits<2> op21_20, bits<4> op11_8,
921 InstrItinClass itin, string OpcodeStr, string Dt,
922 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
923 : N3V<op24, 1, op21_20, op11_8, 1, 0,
925 (ins QPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane), itin,
926 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
927 [(set (ResTy QPR:$dst),
928 (ResTy (IntOp (ResTy QPR:$src1),
930 (OpTy (NEONvduplane (OpTy DPR_8:$src3),
933 // Narrowing 3-register intrinsics.
934 class N3VNInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
935 string OpcodeStr, string Dt, ValueType TyD, ValueType TyQ,
936 Intrinsic IntOp, bit Commutable>
937 : N3V<op24, op23, op21_20, op11_8, 0, op4,
938 (outs DPR:$dst), (ins QPR:$src1, QPR:$src2), IIC_VBINi4D,
939 OpcodeStr, Dt, "$dst, $src1, $src2", "",
940 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src1), (TyQ QPR:$src2))))]> {
941 let isCommutable = Commutable;
944 // Long 3-register intrinsics.
945 class N3VLInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
946 InstrItinClass itin, string OpcodeStr, string Dt,
947 ValueType TyQ, ValueType TyD, Intrinsic IntOp, bit Commutable>
948 : N3V<op24, op23, op21_20, op11_8, 0, op4,
949 (outs QPR:$dst), (ins DPR:$src1, DPR:$src2), itin,
950 OpcodeStr, Dt, "$dst, $src1, $src2", "",
951 [(set QPR:$dst, (TyQ (IntOp (TyD DPR:$src1), (TyD DPR:$src2))))]> {
952 let isCommutable = Commutable;
954 class N3VLIntSL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
955 string OpcodeStr, string Dt,
956 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
957 : N3V<op24, 1, op21_20, op11_8, 1, 0,
958 (outs QPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
959 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
960 [(set (ResTy QPR:$dst),
961 (ResTy (IntOp (OpTy DPR:$src1),
962 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src2),
964 class N3VLIntSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
965 InstrItinClass itin, string OpcodeStr, string Dt,
966 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
967 : N3V<op24, 1, op21_20, op11_8, 1, 0,
968 (outs QPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
969 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
970 [(set (ResTy QPR:$dst),
971 (ResTy (IntOp (OpTy DPR:$src1),
972 (OpTy (NEONvduplane (OpTy DPR_8:$src2),
975 // Wide 3-register intrinsics.
976 class N3VWInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
977 string OpcodeStr, string Dt, ValueType TyQ, ValueType TyD,
978 Intrinsic IntOp, bit Commutable>
979 : N3V<op24, op23, op21_20, op11_8, 0, op4,
980 (outs QPR:$dst), (ins QPR:$src1, DPR:$src2), IIC_VSUBiD,
981 OpcodeStr, Dt, "$dst, $src1, $src2", "",
982 [(set QPR:$dst, (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$src2))))]> {
983 let isCommutable = Commutable;
986 // Pairwise long 2-register intrinsics, both double- and quad-register.
987 class N2VDPLInt<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, 0, op4, (outs DPR:$dst),
992 (ins DPR:$src), IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
993 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
994 class N2VQPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
995 bits<2> op17_16, bits<5> op11_7, bit op4,
996 string OpcodeStr, string Dt,
997 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
998 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
999 (ins QPR:$src), IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
1000 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
1002 // Pairwise long 2-register accumulate intrinsics,
1003 // both double- and quad-register.
1004 // The destination register is also used as the first source operand register.
1005 class N2VDPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1006 bits<2> op17_16, bits<5> op11_7, bit op4,
1007 string OpcodeStr, string Dt,
1008 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1009 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
1010 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), IIC_VPALiD,
1011 OpcodeStr, Dt, "$dst, $src2", "$src1 = $dst",
1012 [(set DPR:$dst, (ResTy (IntOp (ResTy DPR:$src1), (OpTy DPR:$src2))))]>;
1013 class N2VQPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1014 bits<2> op17_16, bits<5> op11_7, bit op4,
1015 string OpcodeStr, string Dt,
1016 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1017 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4,
1018 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), IIC_VPALiQ,
1019 OpcodeStr, Dt, "$dst, $src2", "$src1 = $dst",
1020 [(set QPR:$dst, (ResTy (IntOp (ResTy QPR:$src1), (OpTy QPR:$src2))))]>;
1022 // Shift by immediate,
1023 // both double- and quad-register.
1024 class N2VDSh<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, 0, op4,
1028 (outs DPR:$dst), (ins DPR:$src, i32imm:$SIMM), itin,
1029 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1030 [(set DPR:$dst, (Ty (OpNode (Ty DPR:$src), (i32 imm:$SIMM))))]>;
1031 class N2VQSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1032 InstrItinClass itin, string OpcodeStr, string Dt,
1033 ValueType Ty, SDNode OpNode>
1034 : N2VImm<op24, op23, op11_8, op7, 1, op4,
1035 (outs QPR:$dst), (ins QPR:$src, i32imm:$SIMM), itin,
1036 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1037 [(set QPR:$dst, (Ty (OpNode (Ty QPR:$src), (i32 imm:$SIMM))))]>;
1039 // Long shift by immediate.
1040 class N2VLSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
1041 string OpcodeStr, string Dt,
1042 ValueType ResTy, ValueType OpTy, SDNode OpNode>
1043 : N2VImm<op24, op23, op11_8, op7, op6, op4,
1044 (outs QPR:$dst), (ins DPR:$src, i32imm:$SIMM), IIC_VSHLiD,
1045 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1046 [(set QPR:$dst, (ResTy (OpNode (OpTy DPR:$src),
1047 (i32 imm:$SIMM))))]>;
1049 // Narrow shift by immediate.
1050 class N2VNSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
1051 InstrItinClass itin, string OpcodeStr, string Dt,
1052 ValueType ResTy, ValueType OpTy, SDNode OpNode>
1053 : N2VImm<op24, op23, op11_8, op7, op6, op4,
1054 (outs DPR:$dst), (ins QPR:$src, i32imm:$SIMM), itin,
1055 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1056 [(set DPR:$dst, (ResTy (OpNode (OpTy QPR:$src),
1057 (i32 imm:$SIMM))))]>;
1059 // Shift right by immediate and accumulate,
1060 // both double- and quad-register.
1061 class N2VDShAdd<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, 0, op4, (outs DPR:$dst),
1064 (ins DPR:$src1, DPR:$src2, i32imm:$SIMM), IIC_VPALiD,
1065 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1066 [(set DPR:$dst, (Ty (add DPR:$src1,
1067 (Ty (ShOp DPR:$src2, (i32 imm:$SIMM))))))]>;
1068 class N2VQShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1069 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1070 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$dst),
1071 (ins QPR:$src1, QPR:$src2, i32imm:$SIMM), IIC_VPALiD,
1072 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1073 [(set QPR:$dst, (Ty (add QPR:$src1,
1074 (Ty (ShOp QPR:$src2, (i32 imm:$SIMM))))))]>;
1076 // Shift by immediate and insert,
1077 // both double- and quad-register.
1078 class N2VDShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1079 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1080 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$dst),
1081 (ins DPR:$src1, DPR:$src2, i32imm:$SIMM), IIC_VSHLiD,
1082 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1083 [(set DPR:$dst, (Ty (ShOp DPR:$src1, DPR:$src2, (i32 imm:$SIMM))))]>;
1084 class N2VQShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1085 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1086 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$dst),
1087 (ins QPR:$src1, QPR:$src2, i32imm:$SIMM), IIC_VSHLiQ,
1088 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1089 [(set QPR:$dst, (Ty (ShOp QPR:$src1, QPR:$src2, (i32 imm:$SIMM))))]>;
1091 // Convert, with fractional bits immediate,
1092 // both double- and quad-register.
1093 class N2VCvtD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1094 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1096 : N2VImm<op24, op23, op11_8, op7, 0, op4,
1097 (outs DPR:$dst), (ins DPR:$src, i32imm:$SIMM), IIC_VUNAD,
1098 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1099 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src), (i32 imm:$SIMM))))]>;
1100 class N2VCvtQ<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1101 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1103 : N2VImm<op24, op23, op11_8, op7, 1, op4,
1104 (outs QPR:$dst), (ins QPR:$src, i32imm:$SIMM), IIC_VUNAQ,
1105 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1106 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src), (i32 imm:$SIMM))))]>;
1108 //===----------------------------------------------------------------------===//
1110 //===----------------------------------------------------------------------===//
1112 // Abbreviations used in multiclass suffixes:
1113 // Q = quarter int (8 bit) elements
1114 // H = half int (16 bit) elements
1115 // S = single int (32 bit) elements
1116 // D = double int (64 bit) elements
1118 // Neon 3-register vector operations.
1120 // First with only element sizes of 8, 16 and 32 bits:
1121 multiclass N3V_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1122 InstrItinClass itinD16, InstrItinClass itinD32,
1123 InstrItinClass itinQ16, InstrItinClass itinQ32,
1124 string OpcodeStr, string Dt,
1125 SDNode OpNode, bit Commutable = 0> {
1126 // 64-bit vector types.
1127 def v8i8 : N3VD<op24, op23, 0b00, op11_8, op4, itinD16,
1128 OpcodeStr, !strconcat(Dt, "8"),
1129 v8i8, v8i8, OpNode, Commutable>;
1130 def v4i16 : N3VD<op24, op23, 0b01, op11_8, op4, itinD16,
1131 OpcodeStr, !strconcat(Dt, "16"),
1132 v4i16, v4i16, OpNode, Commutable>;
1133 def v2i32 : N3VD<op24, op23, 0b10, op11_8, op4, itinD32,
1134 OpcodeStr, !strconcat(Dt, "32"),
1135 v2i32, v2i32, OpNode, Commutable>;
1137 // 128-bit vector types.
1138 def v16i8 : N3VQ<op24, op23, 0b00, op11_8, op4, itinQ16,
1139 OpcodeStr, !strconcat(Dt, "8"),
1140 v16i8, v16i8, OpNode, Commutable>;
1141 def v8i16 : N3VQ<op24, op23, 0b01, op11_8, op4, itinQ16,
1142 OpcodeStr, !strconcat(Dt, "16"),
1143 v8i16, v8i16, OpNode, Commutable>;
1144 def v4i32 : N3VQ<op24, op23, 0b10, op11_8, op4, itinQ32,
1145 OpcodeStr, !strconcat(Dt, "32"),
1146 v4i32, v4i32, OpNode, Commutable>;
1149 multiclass N3VSL_HS<bits<4> op11_8, string OpcodeStr, string Dt, SDNode ShOp> {
1150 def v4i16 : N3VDSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
1152 def v2i32 : N3VDSL<0b10, op11_8, IIC_VMULi32D, OpcodeStr, !strconcat(Dt,"32"),
1154 def v8i16 : N3VQSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
1155 v8i16, v4i16, ShOp>;
1156 def v4i32 : N3VQSL<0b10, op11_8, IIC_VMULi32Q, OpcodeStr, !strconcat(Dt,"32"),
1157 v4i32, v2i32, ShOp>;
1160 // ....then also with element size 64 bits:
1161 multiclass N3V_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1162 InstrItinClass itinD, InstrItinClass itinQ,
1163 string OpcodeStr, string Dt,
1164 SDNode OpNode, bit Commutable = 0>
1165 : N3V_QHS<op24, op23, op11_8, op4, itinD, itinD, itinQ, itinQ,
1166 OpcodeStr, Dt, OpNode, Commutable> {
1167 def v1i64 : N3VD<op24, op23, 0b11, op11_8, op4, itinD,
1168 OpcodeStr, !strconcat(Dt, "64"),
1169 v1i64, v1i64, OpNode, Commutable>;
1170 def v2i64 : N3VQ<op24, op23, 0b11, op11_8, op4, itinQ,
1171 OpcodeStr, !strconcat(Dt, "64"),
1172 v2i64, v2i64, OpNode, Commutable>;
1176 // Neon Narrowing 2-register vector intrinsics,
1177 // source operand element sizes of 16, 32 and 64 bits:
1178 multiclass N2VNInt_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1179 bits<5> op11_7, bit op6, bit op4,
1180 InstrItinClass itin, string OpcodeStr, string Dt,
1182 def v8i8 : N2VNInt<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
1183 itin, OpcodeStr, !strconcat(Dt, "16"),
1184 v8i8, v8i16, IntOp>;
1185 def v4i16 : N2VNInt<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
1186 itin, OpcodeStr, !strconcat(Dt, "32"),
1187 v4i16, v4i32, IntOp>;
1188 def v2i32 : N2VNInt<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
1189 itin, OpcodeStr, !strconcat(Dt, "64"),
1190 v2i32, v2i64, IntOp>;
1194 // Neon Lengthening 2-register vector intrinsic (currently specific to VMOVL).
1195 // source operand element sizes of 16, 32 and 64 bits:
1196 multiclass N2VLInt_QHS<bits<2> op24_23, bits<5> op11_7, bit op6, bit op4,
1197 string OpcodeStr, string Dt, Intrinsic IntOp> {
1198 def v8i16 : N2VLInt<op24_23, 0b00, 0b10, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
1199 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
1200 def v4i32 : N2VLInt<op24_23, 0b01, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
1201 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
1202 def v2i64 : N2VLInt<op24_23, 0b10, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
1203 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1207 // Neon 3-register vector intrinsics.
1209 // First with only element sizes of 16 and 32 bits:
1210 multiclass N3VInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
1211 InstrItinClass itinD16, InstrItinClass itinD32,
1212 InstrItinClass itinQ16, InstrItinClass itinQ32,
1213 string OpcodeStr, string Dt,
1214 Intrinsic IntOp, bit Commutable = 0> {
1215 // 64-bit vector types.
1216 def v4i16 : N3VDInt<op24, op23, 0b01, op11_8, op4, itinD16,
1217 OpcodeStr, !strconcat(Dt, "16"),
1218 v4i16, v4i16, IntOp, Commutable>;
1219 def v2i32 : N3VDInt<op24, op23, 0b10, op11_8, op4, itinD32,
1220 OpcodeStr, !strconcat(Dt, "32"),
1221 v2i32, v2i32, IntOp, Commutable>;
1223 // 128-bit vector types.
1224 def v8i16 : N3VQInt<op24, op23, 0b01, op11_8, op4, itinQ16,
1225 OpcodeStr, !strconcat(Dt, "16"),
1226 v8i16, v8i16, IntOp, Commutable>;
1227 def v4i32 : N3VQInt<op24, op23, 0b10, op11_8, op4, itinQ32,
1228 OpcodeStr, !strconcat(Dt, "32"),
1229 v4i32, v4i32, IntOp, Commutable>;
1232 multiclass N3VIntSL_HS<bits<4> op11_8,
1233 InstrItinClass itinD16, InstrItinClass itinD32,
1234 InstrItinClass itinQ16, InstrItinClass itinQ32,
1235 string OpcodeStr, string Dt, Intrinsic IntOp> {
1236 def v4i16 : N3VDIntSL16<0b01, op11_8, itinD16,
1237 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp>;
1238 def v2i32 : N3VDIntSL<0b10, op11_8, itinD32,
1239 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp>;
1240 def v8i16 : N3VQIntSL16<0b01, op11_8, itinQ16,
1241 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16, IntOp>;
1242 def v4i32 : N3VQIntSL<0b10, op11_8, itinQ32,
1243 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32, IntOp>;
1246 // ....then also with element size of 8 bits:
1247 multiclass N3VInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1248 InstrItinClass itinD16, InstrItinClass itinD32,
1249 InstrItinClass itinQ16, InstrItinClass itinQ32,
1250 string OpcodeStr, string Dt,
1251 Intrinsic IntOp, bit Commutable = 0>
1252 : N3VInt_HS<op24, op23, op11_8, op4, itinD16, itinD32, itinQ16, itinQ32,
1253 OpcodeStr, Dt, IntOp, Commutable> {
1254 def v8i8 : N3VDInt<op24, op23, 0b00, op11_8, op4, itinD16,
1255 OpcodeStr, !strconcat(Dt, "8"),
1256 v8i8, v8i8, IntOp, Commutable>;
1257 def v16i8 : N3VQInt<op24, op23, 0b00, op11_8, op4, itinQ16,
1258 OpcodeStr, !strconcat(Dt, "8"),
1259 v16i8, v16i8, IntOp, Commutable>;
1262 // ....then also with element size of 64 bits:
1263 multiclass N3VInt_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1264 InstrItinClass itinD16, InstrItinClass itinD32,
1265 InstrItinClass itinQ16, InstrItinClass itinQ32,
1266 string OpcodeStr, string Dt,
1267 Intrinsic IntOp, bit Commutable = 0>
1268 : N3VInt_QHS<op24, op23, op11_8, op4, itinD16, itinD32, itinQ16, itinQ32,
1269 OpcodeStr, Dt, IntOp, Commutable> {
1270 def v1i64 : N3VDInt<op24, op23, 0b11, op11_8, op4, itinD32,
1271 OpcodeStr, !strconcat(Dt, "64"),
1272 v1i64, v1i64, IntOp, Commutable>;
1273 def v2i64 : N3VQInt<op24, op23, 0b11, op11_8, op4, itinQ32,
1274 OpcodeStr, !strconcat(Dt, "64"),
1275 v2i64, v2i64, IntOp, Commutable>;
1279 // Neon Narrowing 3-register vector intrinsics,
1280 // source operand element sizes of 16, 32 and 64 bits:
1281 multiclass N3VNInt_HSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1282 string OpcodeStr, string Dt,
1283 Intrinsic IntOp, bit Commutable = 0> {
1284 def v8i8 : N3VNInt<op24, op23, 0b00, op11_8, op4,
1285 OpcodeStr, !strconcat(Dt, "16"),
1286 v8i8, v8i16, IntOp, Commutable>;
1287 def v4i16 : N3VNInt<op24, op23, 0b01, op11_8, op4,
1288 OpcodeStr, !strconcat(Dt, "32"),
1289 v4i16, v4i32, IntOp, Commutable>;
1290 def v2i32 : N3VNInt<op24, op23, 0b10, op11_8, op4,
1291 OpcodeStr, !strconcat(Dt, "64"),
1292 v2i32, v2i64, IntOp, Commutable>;
1296 // Neon Long 3-register vector intrinsics.
1298 // First with only element sizes of 16 and 32 bits:
1299 multiclass N3VLInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
1300 InstrItinClass itin, string OpcodeStr, string Dt,
1301 Intrinsic IntOp, bit Commutable = 0> {
1302 def v4i32 : N3VLInt<op24, op23, 0b01, op11_8, op4, itin,
1303 OpcodeStr, !strconcat(Dt, "16"),
1304 v4i32, v4i16, IntOp, Commutable>;
1305 def v2i64 : N3VLInt<op24, op23, 0b10, op11_8, op4, itin,
1306 OpcodeStr, !strconcat(Dt, "32"),
1307 v2i64, v2i32, IntOp, Commutable>;
1310 multiclass N3VLIntSL_HS<bit op24, bits<4> op11_8,
1311 InstrItinClass itin, string OpcodeStr, string Dt,
1313 def v4i16 : N3VLIntSL16<op24, 0b01, op11_8, itin,
1314 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
1315 def v2i32 : N3VLIntSL<op24, 0b10, op11_8, itin,
1316 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1319 // ....then also with element size of 8 bits:
1320 multiclass N3VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1321 InstrItinClass itin, string OpcodeStr, string Dt,
1322 Intrinsic IntOp, bit Commutable = 0>
1323 : N3VLInt_HS<op24, op23, op11_8, op4, itin, OpcodeStr, Dt,
1324 IntOp, Commutable> {
1325 def v8i16 : N3VLInt<op24, op23, 0b00, op11_8, op4, itin,
1326 OpcodeStr, !strconcat(Dt, "8"),
1327 v8i16, v8i8, IntOp, Commutable>;
1331 // Neon Wide 3-register vector intrinsics,
1332 // source operand element sizes of 8, 16 and 32 bits:
1333 multiclass N3VWInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1334 string OpcodeStr, string Dt,
1335 Intrinsic IntOp, bit Commutable = 0> {
1336 def v8i16 : N3VWInt<op24, op23, 0b00, op11_8, op4,
1337 OpcodeStr, !strconcat(Dt, "8"),
1338 v8i16, v8i8, IntOp, Commutable>;
1339 def v4i32 : N3VWInt<op24, op23, 0b01, op11_8, op4,
1340 OpcodeStr, !strconcat(Dt, "16"),
1341 v4i32, v4i16, IntOp, Commutable>;
1342 def v2i64 : N3VWInt<op24, op23, 0b10, op11_8, op4,
1343 OpcodeStr, !strconcat(Dt, "32"),
1344 v2i64, v2i32, IntOp, Commutable>;
1348 // Neon Multiply-Op vector operations,
1349 // element sizes of 8, 16 and 32 bits:
1350 multiclass N3VMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1351 InstrItinClass itinD16, InstrItinClass itinD32,
1352 InstrItinClass itinQ16, InstrItinClass itinQ32,
1353 string OpcodeStr, string Dt, SDNode OpNode> {
1354 // 64-bit vector types.
1355 def v8i8 : N3VDMulOp<op24, op23, 0b00, op11_8, op4, itinD16,
1356 OpcodeStr, !strconcat(Dt, "8"), v8i8, mul, OpNode>;
1357 def v4i16 : N3VDMulOp<op24, op23, 0b01, op11_8, op4, itinD16,
1358 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, OpNode>;
1359 def v2i32 : N3VDMulOp<op24, op23, 0b10, op11_8, op4, itinD32,
1360 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, OpNode>;
1362 // 128-bit vector types.
1363 def v16i8 : N3VQMulOp<op24, op23, 0b00, op11_8, op4, itinQ16,
1364 OpcodeStr, !strconcat(Dt, "8"), v16i8, mul, OpNode>;
1365 def v8i16 : N3VQMulOp<op24, op23, 0b01, op11_8, op4, itinQ16,
1366 OpcodeStr, !strconcat(Dt, "16"), v8i16, mul, OpNode>;
1367 def v4i32 : N3VQMulOp<op24, op23, 0b10, op11_8, op4, itinQ32,
1368 OpcodeStr, !strconcat(Dt, "32"), v4i32, mul, OpNode>;
1371 multiclass N3VMulOpSL_HS<bits<4> op11_8,
1372 InstrItinClass itinD16, InstrItinClass itinD32,
1373 InstrItinClass itinQ16, InstrItinClass itinQ32,
1374 string OpcodeStr, string Dt, SDNode ShOp> {
1375 def v4i16 : N3VDMulOpSL16<0b01, op11_8, itinD16,
1376 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, ShOp>;
1377 def v2i32 : N3VDMulOpSL<0b10, op11_8, itinD32,
1378 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, ShOp>;
1379 def v8i16 : N3VQMulOpSL16<0b01, op11_8, itinQ16,
1380 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16,
1382 def v4i32 : N3VQMulOpSL<0b10, op11_8, itinQ32,
1383 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32,
1387 // Neon 3-argument intrinsics,
1388 // element sizes of 8, 16 and 32 bits:
1389 multiclass N3VInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1390 string OpcodeStr, string Dt, Intrinsic IntOp> {
1391 // 64-bit vector types.
1392 def v8i8 : N3VDInt3<op24, op23, 0b00, op11_8, op4, IIC_VMACi16D,
1393 OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
1394 def v4i16 : N3VDInt3<op24, op23, 0b01, op11_8, op4, IIC_VMACi16D,
1395 OpcodeStr, !strconcat(Dt, "16"), v4i16, v4i16, IntOp>;
1396 def v2i32 : N3VDInt3<op24, op23, 0b10, op11_8, op4, IIC_VMACi32D,
1397 OpcodeStr, !strconcat(Dt, "32"), v2i32, v2i32, IntOp>;
1399 // 128-bit vector types.
1400 def v16i8 : N3VQInt3<op24, op23, 0b00, op11_8, op4, IIC_VMACi16Q,
1401 OpcodeStr, !strconcat(Dt, "8"), v16i8, v16i8, IntOp>;
1402 def v8i16 : N3VQInt3<op24, op23, 0b01, op11_8, op4, IIC_VMACi16Q,
1403 OpcodeStr, !strconcat(Dt, "16"), v8i16, v8i16, IntOp>;
1404 def v4i32 : N3VQInt3<op24, op23, 0b10, op11_8, op4, IIC_VMACi32Q,
1405 OpcodeStr, !strconcat(Dt, "32"), v4i32, v4i32, IntOp>;
1409 // Neon Long 3-argument intrinsics.
1411 // First with only element sizes of 16 and 32 bits:
1412 multiclass N3VLInt3_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
1413 string OpcodeStr, string Dt, Intrinsic IntOp> {
1414 def v4i32 : N3VLInt3<op24, op23, 0b01, op11_8, op4, IIC_VMACi16D,
1415 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
1416 def v2i64 : N3VLInt3<op24, op23, 0b10, op11_8, op4, IIC_VMACi16D,
1417 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1420 multiclass N3VLInt3SL_HS<bit op24, bits<4> op11_8,
1421 string OpcodeStr, string Dt, Intrinsic IntOp> {
1422 def v4i16 : N3VLInt3SL16<op24, 0b01, op11_8, IIC_VMACi16D,
1423 OpcodeStr, !strconcat(Dt,"16"), v4i32, v4i16, IntOp>;
1424 def v2i32 : N3VLInt3SL<op24, 0b10, op11_8, IIC_VMACi32D,
1425 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1428 // ....then also with element size of 8 bits:
1429 multiclass N3VLInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1430 string OpcodeStr, string Dt, Intrinsic IntOp>
1431 : N3VLInt3_HS<op24, op23, op11_8, op4, OpcodeStr, Dt, IntOp> {
1432 def v8i16 : N3VLInt3<op24, op23, 0b00, op11_8, op4, IIC_VMACi16D,
1433 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
1437 // Neon 2-register vector intrinsics,
1438 // element sizes of 8, 16 and 32 bits:
1439 multiclass N2VInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1440 bits<5> op11_7, bit op4,
1441 InstrItinClass itinD, InstrItinClass itinQ,
1442 string OpcodeStr, string Dt, Intrinsic IntOp> {
1443 // 64-bit vector types.
1444 def v8i8 : N2VDInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1445 itinD, OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
1446 def v4i16 : N2VDInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1447 itinD, OpcodeStr, !strconcat(Dt, "16"),v4i16,v4i16,IntOp>;
1448 def v2i32 : N2VDInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1449 itinD, OpcodeStr, !strconcat(Dt, "32"),v2i32,v2i32,IntOp>;
1451 // 128-bit vector types.
1452 def v16i8 : N2VQInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1453 itinQ, OpcodeStr, !strconcat(Dt, "8"), v16i8,v16i8,IntOp>;
1454 def v8i16 : N2VQInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1455 itinQ, OpcodeStr, !strconcat(Dt, "16"),v8i16,v8i16,IntOp>;
1456 def v4i32 : N2VQInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1457 itinQ, OpcodeStr, !strconcat(Dt, "32"),v4i32,v4i32,IntOp>;
1461 // Neon Pairwise long 2-register intrinsics,
1462 // element sizes of 8, 16 and 32 bits:
1463 multiclass N2VPLInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1464 bits<5> op11_7, bit op4,
1465 string OpcodeStr, string Dt, Intrinsic IntOp> {
1466 // 64-bit vector types.
1467 def v8i8 : N2VDPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1468 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
1469 def v4i16 : N2VDPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1470 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
1471 def v2i32 : N2VDPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1472 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
1474 // 128-bit vector types.
1475 def v16i8 : N2VQPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1476 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
1477 def v8i16 : N2VQPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1478 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
1479 def v4i32 : N2VQPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1480 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
1484 // Neon Pairwise long 2-register accumulate intrinsics,
1485 // element sizes of 8, 16 and 32 bits:
1486 multiclass N2VPLInt2_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1487 bits<5> op11_7, bit op4,
1488 string OpcodeStr, string Dt, Intrinsic IntOp> {
1489 // 64-bit vector types.
1490 def v8i8 : N2VDPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1491 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
1492 def v4i16 : N2VDPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1493 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
1494 def v2i32 : N2VDPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1495 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
1497 // 128-bit vector types.
1498 def v16i8 : N2VQPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1499 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
1500 def v8i16 : N2VQPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1501 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
1502 def v4i32 : N2VQPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1503 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
1507 // Neon 2-register vector shift by immediate,
1508 // element sizes of 8, 16, 32 and 64 bits:
1509 multiclass N2VSh_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1510 InstrItinClass itin, string OpcodeStr, string Dt,
1512 // 64-bit vector types.
1513 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, itin,
1514 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
1515 let Inst{21-19} = 0b001; // imm6 = 001xxx
1517 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, itin,
1518 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
1519 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1521 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, itin,
1522 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
1523 let Inst{21} = 0b1; // imm6 = 1xxxxx
1525 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, itin,
1526 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
1529 // 128-bit vector types.
1530 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, itin,
1531 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
1532 let Inst{21-19} = 0b001; // imm6 = 001xxx
1534 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, itin,
1535 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
1536 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1538 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, itin,
1539 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
1540 let Inst{21} = 0b1; // imm6 = 1xxxxx
1542 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, itin,
1543 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
1548 // Neon Shift-Accumulate vector operations,
1549 // element sizes of 8, 16, 32 and 64 bits:
1550 multiclass N2VShAdd_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1551 string OpcodeStr, string Dt, SDNode ShOp> {
1552 // 64-bit vector types.
1553 def v8i8 : N2VDShAdd<op24, op23, op11_8, 0, op4,
1554 OpcodeStr, !strconcat(Dt, "8"), v8i8, ShOp> {
1555 let Inst{21-19} = 0b001; // imm6 = 001xxx
1557 def v4i16 : N2VDShAdd<op24, op23, op11_8, 0, op4,
1558 OpcodeStr, !strconcat(Dt, "16"), v4i16, ShOp> {
1559 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1561 def v2i32 : N2VDShAdd<op24, op23, op11_8, 0, op4,
1562 OpcodeStr, !strconcat(Dt, "32"), v2i32, ShOp> {
1563 let Inst{21} = 0b1; // imm6 = 1xxxxx
1565 def v1i64 : N2VDShAdd<op24, op23, op11_8, 1, op4,
1566 OpcodeStr, !strconcat(Dt, "64"), v1i64, ShOp>;
1569 // 128-bit vector types.
1570 def v16i8 : N2VQShAdd<op24, op23, op11_8, 0, op4,
1571 OpcodeStr, !strconcat(Dt, "8"), v16i8, ShOp> {
1572 let Inst{21-19} = 0b001; // imm6 = 001xxx
1574 def v8i16 : N2VQShAdd<op24, op23, op11_8, 0, op4,
1575 OpcodeStr, !strconcat(Dt, "16"), v8i16, ShOp> {
1576 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1578 def v4i32 : N2VQShAdd<op24, op23, op11_8, 0, op4,
1579 OpcodeStr, !strconcat(Dt, "32"), v4i32, ShOp> {
1580 let Inst{21} = 0b1; // imm6 = 1xxxxx
1582 def v2i64 : N2VQShAdd<op24, op23, op11_8, 1, op4,
1583 OpcodeStr, !strconcat(Dt, "64"), v2i64, ShOp>;
1588 // Neon Shift-Insert vector operations,
1589 // element sizes of 8, 16, 32 and 64 bits:
1590 multiclass N2VShIns_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1591 string OpcodeStr, SDNode ShOp> {
1592 // 64-bit vector types.
1593 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4,
1594 OpcodeStr, "8", v8i8, ShOp> {
1595 let Inst{21-19} = 0b001; // imm6 = 001xxx
1597 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4,
1598 OpcodeStr, "16", v4i16, ShOp> {
1599 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1601 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4,
1602 OpcodeStr, "32", v2i32, ShOp> {
1603 let Inst{21} = 0b1; // imm6 = 1xxxxx
1605 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4,
1606 OpcodeStr, "64", v1i64, ShOp>;
1609 // 128-bit vector types.
1610 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4,
1611 OpcodeStr, "8", v16i8, ShOp> {
1612 let Inst{21-19} = 0b001; // imm6 = 001xxx
1614 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4,
1615 OpcodeStr, "16", v8i16, ShOp> {
1616 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1618 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4,
1619 OpcodeStr, "32", v4i32, ShOp> {
1620 let Inst{21} = 0b1; // imm6 = 1xxxxx
1622 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4,
1623 OpcodeStr, "64", v2i64, ShOp>;
1627 // Neon Shift Long operations,
1628 // element sizes of 8, 16, 32 bits:
1629 multiclass N2VLSh_QHS<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
1630 bit op4, string OpcodeStr, string Dt, SDNode OpNode> {
1631 def v8i16 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
1632 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode> {
1633 let Inst{21-19} = 0b001; // imm6 = 001xxx
1635 def v4i32 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
1636 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode> {
1637 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1639 def v2i64 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
1640 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode> {
1641 let Inst{21} = 0b1; // imm6 = 1xxxxx
1645 // Neon Shift Narrow operations,
1646 // element sizes of 16, 32, 64 bits:
1647 multiclass N2VNSh_HSD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
1648 bit op4, InstrItinClass itin, string OpcodeStr, string Dt,
1650 def v8i8 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
1651 OpcodeStr, !strconcat(Dt, "16"), v8i8, v8i16, OpNode> {
1652 let Inst{21-19} = 0b001; // imm6 = 001xxx
1654 def v4i16 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
1655 OpcodeStr, !strconcat(Dt, "32"), v4i16, v4i32, OpNode> {
1656 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1658 def v2i32 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
1659 OpcodeStr, !strconcat(Dt, "64"), v2i32, v2i64, OpNode> {
1660 let Inst{21} = 0b1; // imm6 = 1xxxxx
1664 //===----------------------------------------------------------------------===//
1665 // Instruction Definitions.
1666 //===----------------------------------------------------------------------===//
1668 // Vector Add Operations.
1670 // VADD : Vector Add (integer and floating-point)
1671 defm VADD : N3V_QHSD<0, 0, 0b1000, 0, IIC_VBINiD, IIC_VBINiQ, "vadd", "i",
1673 def VADDfd : N3VD<0, 0, 0b00, 0b1101, 0, IIC_VBIND, "vadd", "f32",
1674 v2f32, v2f32, fadd, 1>;
1675 def VADDfq : N3VQ<0, 0, 0b00, 0b1101, 0, IIC_VBINQ, "vadd", "f32",
1676 v4f32, v4f32, fadd, 1>;
1677 // VADDL : Vector Add Long (Q = D + D)
1678 defm VADDLs : N3VLInt_QHS<0,1,0b0000,0, IIC_VSHLiD, "vaddl", "s",
1679 int_arm_neon_vaddls, 1>;
1680 defm VADDLu : N3VLInt_QHS<1,1,0b0000,0, IIC_VSHLiD, "vaddl", "u",
1681 int_arm_neon_vaddlu, 1>;
1682 // VADDW : Vector Add Wide (Q = Q + D)
1683 defm VADDWs : N3VWInt_QHS<0,1,0b0001,0, "vaddw", "s", int_arm_neon_vaddws, 0>;
1684 defm VADDWu : N3VWInt_QHS<1,1,0b0001,0, "vaddw", "u", int_arm_neon_vaddwu, 0>;
1685 // VHADD : Vector Halving Add
1686 defm VHADDs : N3VInt_QHS<0,0,0b0000,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1687 IIC_VBINi4Q, "vhadd", "s", int_arm_neon_vhadds, 1>;
1688 defm VHADDu : N3VInt_QHS<1,0,0b0000,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1689 IIC_VBINi4Q, "vhadd", "u", int_arm_neon_vhaddu, 1>;
1690 // VRHADD : Vector Rounding Halving Add
1691 defm VRHADDs : N3VInt_QHS<0,0,0b0001,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1692 IIC_VBINi4Q, "vrhadd", "s", int_arm_neon_vrhadds, 1>;
1693 defm VRHADDu : N3VInt_QHS<1,0,0b0001,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1694 IIC_VBINi4Q, "vrhadd", "u", int_arm_neon_vrhaddu, 1>;
1695 // VQADD : Vector Saturating Add
1696 defm VQADDs : N3VInt_QHSD<0,0,0b0000,1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1697 IIC_VBINi4Q, "vqadd", "s", int_arm_neon_vqadds, 1>;
1698 defm VQADDu : N3VInt_QHSD<1,0,0b0000,1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1699 IIC_VBINi4Q, "vqadd", "u", int_arm_neon_vqaddu, 1>;
1700 // VADDHN : Vector Add and Narrow Returning High Half (D = Q + Q)
1701 defm VADDHN : N3VNInt_HSD<0,1,0b0100,0, "vaddhn", "i",
1702 int_arm_neon_vaddhn, 1>;
1703 // VRADDHN : Vector Rounding Add and Narrow Returning High Half (D = Q + Q)
1704 defm VRADDHN : N3VNInt_HSD<1,1,0b0100,0, "vraddhn", "i",
1705 int_arm_neon_vraddhn, 1>;
1707 // Vector Multiply Operations.
1709 // VMUL : Vector Multiply (integer, polynomial and floating-point)
1710 defm VMUL : N3V_QHS<0, 0, 0b1001, 1, IIC_VMULi16D, IIC_VMULi32D,
1711 IIC_VMULi16Q, IIC_VMULi32Q, "vmul", "i", mul, 1>;
1712 def VMULpd : N3VDInt<1, 0, 0b00, 0b1001, 1, IIC_VMULi16D, "vmul", "p8",
1713 v8i8, v8i8, int_arm_neon_vmulp, 1>;
1714 def VMULpq : N3VQInt<1, 0, 0b00, 0b1001, 1, IIC_VMULi16Q, "vmul", "p8",
1715 v16i8, v16i8, int_arm_neon_vmulp, 1>;
1716 def VMULfd : N3VD<1, 0, 0b00, 0b1101, 1, IIC_VBIND, "vmul", "f32",
1717 v2f32, v2f32, fmul, 1>;
1718 def VMULfq : N3VQ<1, 0, 0b00, 0b1101, 1, IIC_VBINQ, "vmul", "f32",
1719 v4f32, v4f32, fmul, 1>;
1720 defm VMULsl : N3VSL_HS<0b1000, "vmul", "i", mul>;
1721 def VMULslfd : N3VDSL<0b10, 0b1001, IIC_VBIND, "vmul", "f32", v2f32, fmul>;
1722 def VMULslfq : N3VQSL<0b10, 0b1001, IIC_VBINQ, "vmul", "f32", v4f32,
1725 def : Pat<(v8i16 (mul (v8i16 QPR:$src1),
1726 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
1727 (v8i16 (VMULslv8i16 (v8i16 QPR:$src1),
1728 (v4i16 (EXTRACT_SUBREG QPR:$src2,
1729 (DSubReg_i16_reg imm:$lane))),
1730 (SubReg_i16_lane imm:$lane)))>;
1731 def : Pat<(v4i32 (mul (v4i32 QPR:$src1),
1732 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
1733 (v4i32 (VMULslv4i32 (v4i32 QPR:$src1),
1734 (v2i32 (EXTRACT_SUBREG QPR:$src2,
1735 (DSubReg_i32_reg imm:$lane))),
1736 (SubReg_i32_lane imm:$lane)))>;
1737 def : Pat<(v4f32 (fmul (v4f32 QPR:$src1),
1738 (v4f32 (NEONvduplane (v4f32 QPR:$src2), imm:$lane)))),
1739 (v4f32 (VMULslfq (v4f32 QPR:$src1),
1740 (v2f32 (EXTRACT_SUBREG QPR:$src2,
1741 (DSubReg_i32_reg imm:$lane))),
1742 (SubReg_i32_lane imm:$lane)))>;
1744 // VQDMULH : Vector Saturating Doubling Multiply Returning High Half
1745 defm VQDMULH : N3VInt_HS<0, 0, 0b1011, 0, IIC_VMULi16D, IIC_VMULi32D,
1746 IIC_VMULi16Q, IIC_VMULi32Q,
1747 "vqdmulh", "s", int_arm_neon_vqdmulh, 1>;
1748 defm VQDMULHsl: N3VIntSL_HS<0b1100, IIC_VMULi16D, IIC_VMULi32D,
1749 IIC_VMULi16Q, IIC_VMULi32Q,
1750 "vqdmulh", "s", int_arm_neon_vqdmulh>;
1751 def : Pat<(v8i16 (int_arm_neon_vqdmulh (v8i16 QPR:$src1),
1752 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
1754 (v8i16 (VQDMULHslv8i16 (v8i16 QPR:$src1),
1755 (v4i16 (EXTRACT_SUBREG QPR:$src2,
1756 (DSubReg_i16_reg imm:$lane))),
1757 (SubReg_i16_lane imm:$lane)))>;
1758 def : Pat<(v4i32 (int_arm_neon_vqdmulh (v4i32 QPR:$src1),
1759 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
1761 (v4i32 (VQDMULHslv4i32 (v4i32 QPR:$src1),
1762 (v2i32 (EXTRACT_SUBREG QPR:$src2,
1763 (DSubReg_i32_reg imm:$lane))),
1764 (SubReg_i32_lane imm:$lane)))>;
1766 // VQRDMULH : Vector Rounding Saturating Doubling Multiply Returning High Half
1767 defm VQRDMULH : N3VInt_HS<1, 0, 0b1011, 0, IIC_VMULi16D, IIC_VMULi32D,
1768 IIC_VMULi16Q, IIC_VMULi32Q,
1769 "vqrdmulh", "s", int_arm_neon_vqrdmulh, 1>;
1770 defm VQRDMULHsl : N3VIntSL_HS<0b1101, IIC_VMULi16D, IIC_VMULi32D,
1771 IIC_VMULi16Q, IIC_VMULi32Q,
1772 "vqrdmulh", "s", int_arm_neon_vqrdmulh>;
1773 def : Pat<(v8i16 (int_arm_neon_vqrdmulh (v8i16 QPR:$src1),
1774 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
1776 (v8i16 (VQRDMULHslv8i16 (v8i16 QPR:$src1),
1777 (v4i16 (EXTRACT_SUBREG QPR:$src2,
1778 (DSubReg_i16_reg imm:$lane))),
1779 (SubReg_i16_lane imm:$lane)))>;
1780 def : Pat<(v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$src1),
1781 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
1783 (v4i32 (VQRDMULHslv4i32 (v4i32 QPR:$src1),
1784 (v2i32 (EXTRACT_SUBREG QPR:$src2,
1785 (DSubReg_i32_reg imm:$lane))),
1786 (SubReg_i32_lane imm:$lane)))>;
1788 // VMULL : Vector Multiply Long (integer and polynomial) (Q = D * D)
1789 defm VMULLs : N3VLInt_QHS<0,1,0b1100,0, IIC_VMULi16D, "vmull", "s",
1790 int_arm_neon_vmulls, 1>;
1791 defm VMULLu : N3VLInt_QHS<1,1,0b1100,0, IIC_VMULi16D, "vmull", "u",
1792 int_arm_neon_vmullu, 1>;
1793 def VMULLp : N3VLInt<0, 1, 0b00, 0b1110, 0, IIC_VMULi16D, "vmull", "p8",
1794 v8i16, v8i8, int_arm_neon_vmullp, 1>;
1795 defm VMULLsls : N3VLIntSL_HS<0, 0b1010, IIC_VMULi16D, "vmull", "s",
1796 int_arm_neon_vmulls>;
1797 defm VMULLslu : N3VLIntSL_HS<1, 0b1010, IIC_VMULi16D, "vmull", "u",
1798 int_arm_neon_vmullu>;
1800 // VQDMULL : Vector Saturating Doubling Multiply Long (Q = D * D)
1801 defm VQDMULL : N3VLInt_HS<0,1,0b1101,0, IIC_VMULi16D, "vqdmull", "s",
1802 int_arm_neon_vqdmull, 1>;
1803 defm VQDMULLsl: N3VLIntSL_HS<0, 0b1011, IIC_VMULi16D, "vqdmull", "s",
1804 int_arm_neon_vqdmull>;
1806 // Vector Multiply-Accumulate and Multiply-Subtract Operations.
1808 // VMLA : Vector Multiply Accumulate (integer and floating-point)
1809 defm VMLA : N3VMulOp_QHS<0, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
1810 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
1811 def VMLAfd : N3VDMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACD, "vmla", "f32",
1813 def VMLAfq : N3VQMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACQ, "vmla", "f32",
1815 defm VMLAsl : N3VMulOpSL_HS<0b0000, IIC_VMACi16D, IIC_VMACi32D,
1816 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
1817 def VMLAslfd : N3VDMulOpSL<0b10, 0b0001, IIC_VMACD, "vmla", "f32",
1819 def VMLAslfq : N3VQMulOpSL<0b10, 0b0001, IIC_VMACQ, "vmla", "f32",
1820 v4f32, v2f32, fmul, fadd>;
1822 def : Pat<(v8i16 (add (v8i16 QPR:$src1),
1823 (mul (v8i16 QPR:$src2),
1824 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
1825 (v8i16 (VMLAslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
1826 (v4i16 (EXTRACT_SUBREG QPR:$src3,
1827 (DSubReg_i16_reg imm:$lane))),
1828 (SubReg_i16_lane imm:$lane)))>;
1830 def : Pat<(v4i32 (add (v4i32 QPR:$src1),
1831 (mul (v4i32 QPR:$src2),
1832 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
1833 (v4i32 (VMLAslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
1834 (v2i32 (EXTRACT_SUBREG QPR:$src3,
1835 (DSubReg_i32_reg imm:$lane))),
1836 (SubReg_i32_lane imm:$lane)))>;
1838 def : Pat<(v4f32 (fadd (v4f32 QPR:$src1),
1839 (fmul (v4f32 QPR:$src2),
1840 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
1841 (v4f32 (VMLAslfq (v4f32 QPR:$src1),
1843 (v2f32 (EXTRACT_SUBREG QPR:$src3,
1844 (DSubReg_i32_reg imm:$lane))),
1845 (SubReg_i32_lane imm:$lane)))>;
1847 // VMLAL : Vector Multiply Accumulate Long (Q += D * D)
1848 defm VMLALs : N3VLInt3_QHS<0,1,0b1000,0, "vmlal", "s", int_arm_neon_vmlals>;
1849 defm VMLALu : N3VLInt3_QHS<1,1,0b1000,0, "vmlal", "u", int_arm_neon_vmlalu>;
1851 defm VMLALsls : N3VLInt3SL_HS<0, 0b0010, "vmlal", "s", int_arm_neon_vmlals>;
1852 defm VMLALslu : N3VLInt3SL_HS<1, 0b0010, "vmlal", "u", int_arm_neon_vmlalu>;
1854 // VQDMLAL : Vector Saturating Doubling Multiply Accumulate Long (Q += D * D)
1855 defm VQDMLAL : N3VLInt3_HS<0, 1, 0b1001, 0, "vqdmlal", "s",
1856 int_arm_neon_vqdmlal>;
1857 defm VQDMLALsl: N3VLInt3SL_HS<0, 0b0011, "vqdmlal", "s", int_arm_neon_vqdmlal>;
1859 // VMLS : Vector Multiply Subtract (integer and floating-point)
1860 defm VMLS : N3VMulOp_QHS<1, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
1861 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
1862 def VMLSfd : N3VDMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACD, "vmls", "f32",
1864 def VMLSfq : N3VQMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACQ, "vmls", "f32",
1866 defm VMLSsl : N3VMulOpSL_HS<0b0100, IIC_VMACi16D, IIC_VMACi32D,
1867 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
1868 def VMLSslfd : N3VDMulOpSL<0b10, 0b0101, IIC_VMACD, "vmls", "f32",
1870 def VMLSslfq : N3VQMulOpSL<0b10, 0b0101, IIC_VMACQ, "vmls", "f32",
1871 v4f32, v2f32, fmul, fsub>;
1873 def : Pat<(v8i16 (sub (v8i16 QPR:$src1),
1874 (mul (v8i16 QPR:$src2),
1875 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
1876 (v8i16 (VMLSslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
1877 (v4i16 (EXTRACT_SUBREG QPR:$src3,
1878 (DSubReg_i16_reg imm:$lane))),
1879 (SubReg_i16_lane imm:$lane)))>;
1881 def : Pat<(v4i32 (sub (v4i32 QPR:$src1),
1882 (mul (v4i32 QPR:$src2),
1883 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
1884 (v4i32 (VMLSslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
1885 (v2i32 (EXTRACT_SUBREG QPR:$src3,
1886 (DSubReg_i32_reg imm:$lane))),
1887 (SubReg_i32_lane imm:$lane)))>;
1889 def : Pat<(v4f32 (fsub (v4f32 QPR:$src1),
1890 (fmul (v4f32 QPR:$src2),
1891 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
1892 (v4f32 (VMLSslfq (v4f32 QPR:$src1), (v4f32 QPR:$src2),
1893 (v2f32 (EXTRACT_SUBREG QPR:$src3,
1894 (DSubReg_i32_reg imm:$lane))),
1895 (SubReg_i32_lane imm:$lane)))>;
1897 // VMLSL : Vector Multiply Subtract Long (Q -= D * D)
1898 defm VMLSLs : N3VLInt3_QHS<0,1,0b1010,0, "vmlsl", "s", int_arm_neon_vmlsls>;
1899 defm VMLSLu : N3VLInt3_QHS<1,1,0b1010,0, "vmlsl", "u", int_arm_neon_vmlslu>;
1901 defm VMLSLsls : N3VLInt3SL_HS<0, 0b0110, "vmlsl", "s", int_arm_neon_vmlsls>;
1902 defm VMLSLslu : N3VLInt3SL_HS<1, 0b0110, "vmlsl", "u", int_arm_neon_vmlslu>;
1904 // VQDMLSL : Vector Saturating Doubling Multiply Subtract Long (Q -= D * D)
1905 defm VQDMLSL : N3VLInt3_HS<0, 1, 0b1011, 0, "vqdmlsl", "s",
1906 int_arm_neon_vqdmlsl>;
1907 defm VQDMLSLsl: N3VLInt3SL_HS<0, 0b111, "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
1909 // Vector Subtract Operations.
1911 // VSUB : Vector Subtract (integer and floating-point)
1912 defm VSUB : N3V_QHSD<1, 0, 0b1000, 0, IIC_VSUBiD, IIC_VSUBiQ,
1913 "vsub", "i", sub, 0>;
1914 def VSUBfd : N3VD<0, 0, 0b10, 0b1101, 0, IIC_VBIND, "vsub", "f32",
1915 v2f32, v2f32, fsub, 0>;
1916 def VSUBfq : N3VQ<0, 0, 0b10, 0b1101, 0, IIC_VBINQ, "vsub", "f32",
1917 v4f32, v4f32, fsub, 0>;
1918 // VSUBL : Vector Subtract Long (Q = D - D)
1919 defm VSUBLs : N3VLInt_QHS<0,1,0b0010,0, IIC_VSHLiD, "vsubl", "s",
1920 int_arm_neon_vsubls, 1>;
1921 defm VSUBLu : N3VLInt_QHS<1,1,0b0010,0, IIC_VSHLiD, "vsubl", "u",
1922 int_arm_neon_vsublu, 1>;
1923 // VSUBW : Vector Subtract Wide (Q = Q - D)
1924 defm VSUBWs : N3VWInt_QHS<0,1,0b0011,0, "vsubw", "s", int_arm_neon_vsubws, 0>;
1925 defm VSUBWu : N3VWInt_QHS<1,1,0b0011,0, "vsubw", "u", int_arm_neon_vsubwu, 0>;
1926 // VHSUB : Vector Halving Subtract
1927 defm VHSUBs : N3VInt_QHS<0, 0, 0b0010, 0, IIC_VBINi4D, IIC_VBINi4D,
1928 IIC_VBINi4Q, IIC_VBINi4Q,
1929 "vhsub", "s", int_arm_neon_vhsubs, 0>;
1930 defm VHSUBu : N3VInt_QHS<1, 0, 0b0010, 0, IIC_VBINi4D, IIC_VBINi4D,
1931 IIC_VBINi4Q, IIC_VBINi4Q,
1932 "vhsub", "u", int_arm_neon_vhsubu, 0>;
1933 // VQSUB : Vector Saturing Subtract
1934 defm VQSUBs : N3VInt_QHSD<0, 0, 0b0010, 1, IIC_VBINi4D, IIC_VBINi4D,
1935 IIC_VBINi4Q, IIC_VBINi4Q,
1936 "vqsub", "s", int_arm_neon_vqsubs, 0>;
1937 defm VQSUBu : N3VInt_QHSD<1, 0, 0b0010, 1, IIC_VBINi4D, IIC_VBINi4D,
1938 IIC_VBINi4Q, IIC_VBINi4Q,
1939 "vqsub", "u", int_arm_neon_vqsubu, 0>;
1940 // VSUBHN : Vector Subtract and Narrow Returning High Half (D = Q - Q)
1941 defm VSUBHN : N3VNInt_HSD<0,1,0b0110,0, "vsubhn", "i",
1942 int_arm_neon_vsubhn, 0>;
1943 // VRSUBHN : Vector Rounding Subtract and Narrow Returning High Half (D=Q-Q)
1944 defm VRSUBHN : N3VNInt_HSD<1,1,0b0110,0, "vrsubhn", "i",
1945 int_arm_neon_vrsubhn, 0>;
1947 // Vector Comparisons.
1949 // VCEQ : Vector Compare Equal
1950 defm VCEQ : N3V_QHS<1, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1951 IIC_VBINi4Q, "vceq", "i", NEONvceq, 1>;
1952 def VCEQfd : N3VD<0,0,0b00,0b1110,0, IIC_VBIND, "vceq", "f32", v2i32, v2f32,
1954 def VCEQfq : N3VQ<0,0,0b00,0b1110,0, IIC_VBINQ, "vceq", "f32", v4i32, v4f32,
1956 // VCGE : Vector Compare Greater Than or Equal
1957 defm VCGEs : N3V_QHS<0, 0, 0b0011, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1958 IIC_VBINi4Q, "vcge", "s", NEONvcge, 0>;
1959 defm VCGEu : N3V_QHS<1, 0, 0b0011, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1960 IIC_VBINi4Q, "vcge", "u", NEONvcgeu, 0>;
1961 def VCGEfd : N3VD<1,0,0b00,0b1110,0, IIC_VBIND, "vcge", "f32",
1962 v2i32, v2f32, NEONvcge, 0>;
1963 def VCGEfq : N3VQ<1,0,0b00,0b1110,0, IIC_VBINQ, "vcge", "f32", v4i32, v4f32,
1965 // VCGT : Vector Compare Greater Than
1966 defm VCGTs : N3V_QHS<0, 0, 0b0011, 0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1967 IIC_VBINi4Q, "vcgt", "s", NEONvcgt, 0>;
1968 defm VCGTu : N3V_QHS<1, 0, 0b0011, 0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1969 IIC_VBINi4Q, "vcgt", "u", NEONvcgtu, 0>;
1970 def VCGTfd : N3VD<1,0,0b10,0b1110,0, IIC_VBIND, "vcgt", "f32", v2i32, v2f32,
1972 def VCGTfq : N3VQ<1,0,0b10,0b1110,0, IIC_VBINQ, "vcgt", "f32", v4i32, v4f32,
1974 // VACGE : Vector Absolute Compare Greater Than or Equal (aka VCAGE)
1975 def VACGEd : N3VDInt<1, 0, 0b00, 0b1110, 1, IIC_VBIND, "vacge", "f32",
1976 v2i32, v2f32, int_arm_neon_vacged, 0>;
1977 def VACGEq : N3VQInt<1, 0, 0b00, 0b1110, 1, IIC_VBINQ, "vacge", "f32",
1978 v4i32, v4f32, int_arm_neon_vacgeq, 0>;
1979 // VACGT : Vector Absolute Compare Greater Than (aka VCAGT)
1980 def VACGTd : N3VDInt<1, 0, 0b10, 0b1110, 1, IIC_VBIND, "vacgt", "f32",
1981 v2i32, v2f32, int_arm_neon_vacgtd, 0>;
1982 def VACGTq : N3VQInt<1, 0, 0b10, 0b1110, 1, IIC_VBINQ, "vacgt", "f32",
1983 v4i32, v4f32, int_arm_neon_vacgtq, 0>;
1984 // VTST : Vector Test Bits
1985 defm VTST : N3V_QHS<0, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
1986 IIC_VBINi4Q, "vtst", "", NEONvtst, 1>;
1988 // Vector Bitwise Operations.
1990 // VAND : Vector Bitwise AND
1991 def VANDd : N3VDX<0, 0, 0b00, 0b0001, 1, IIC_VBINiD, "vand",
1992 v2i32, v2i32, and, 1>;
1993 def VANDq : N3VQX<0, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "vand",
1994 v4i32, v4i32, and, 1>;
1996 // VEOR : Vector Bitwise Exclusive OR
1997 def VEORd : N3VDX<1, 0, 0b00, 0b0001, 1, IIC_VBINiD, "veor",
1998 v2i32, v2i32, xor, 1>;
1999 def VEORq : N3VQX<1, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "veor",
2000 v4i32, v4i32, xor, 1>;
2002 // VORR : Vector Bitwise OR
2003 def VORRd : N3VDX<0, 0, 0b10, 0b0001, 1, IIC_VBINiD, "vorr",
2004 v2i32, v2i32, or, 1>;
2005 def VORRq : N3VQX<0, 0, 0b10, 0b0001, 1, IIC_VBINiQ, "vorr",
2006 v4i32, v4i32, or, 1>;
2008 // VBIC : Vector Bitwise Bit Clear (AND NOT)
2009 def VBICd : N3VX<0, 0, 0b01, 0b0001, 0, 1, (outs DPR:$dst),
2010 (ins DPR:$src1, DPR:$src2), IIC_VBINiD,
2011 "vbic", "$dst, $src1, $src2", "",
2012 [(set DPR:$dst, (v2i32 (and DPR:$src1,
2013 (vnot_conv DPR:$src2))))]>;
2014 def VBICq : N3VX<0, 0, 0b01, 0b0001, 1, 1, (outs QPR:$dst),
2015 (ins QPR:$src1, QPR:$src2), IIC_VBINiQ,
2016 "vbic", "$dst, $src1, $src2", "",
2017 [(set QPR:$dst, (v4i32 (and QPR:$src1,
2018 (vnot_conv QPR:$src2))))]>;
2020 // VORN : Vector Bitwise OR NOT
2021 def VORNd : N3VX<0, 0, 0b11, 0b0001, 0, 1, (outs DPR:$dst),
2022 (ins DPR:$src1, DPR:$src2), IIC_VBINiD,
2023 "vorn", "$dst, $src1, $src2", "",
2024 [(set DPR:$dst, (v2i32 (or DPR:$src1,
2025 (vnot_conv DPR:$src2))))]>;
2026 def VORNq : N3VX<0, 0, 0b11, 0b0001, 1, 1, (outs QPR:$dst),
2027 (ins QPR:$src1, QPR:$src2), IIC_VBINiQ,
2028 "vorn", "$dst, $src1, $src2", "",
2029 [(set QPR:$dst, (v4i32 (or QPR:$src1,
2030 (vnot_conv QPR:$src2))))]>;
2032 // VMVN : Vector Bitwise NOT
2033 def VMVNd : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 0, 0,
2034 (outs DPR:$dst), (ins DPR:$src), IIC_VSHLiD,
2035 "vmvn", "$dst, $src", "",
2036 [(set DPR:$dst, (v2i32 (vnot DPR:$src)))]>;
2037 def VMVNq : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 1, 0,
2038 (outs QPR:$dst), (ins QPR:$src), IIC_VSHLiD,
2039 "vmvn", "$dst, $src", "",
2040 [(set QPR:$dst, (v4i32 (vnot QPR:$src)))]>;
2041 def : Pat<(v2i32 (vnot_conv DPR:$src)), (VMVNd DPR:$src)>;
2042 def : Pat<(v4i32 (vnot_conv QPR:$src)), (VMVNq QPR:$src)>;
2044 // VBSL : Vector Bitwise Select
2045 def VBSLd : N3VX<1, 0, 0b01, 0b0001, 0, 1, (outs DPR:$dst),
2046 (ins DPR:$src1, DPR:$src2, DPR:$src3), IIC_VCNTiD,
2047 "vbsl", "$dst, $src2, $src3", "$src1 = $dst",
2049 (v2i32 (or (and DPR:$src2, DPR:$src1),
2050 (and DPR:$src3, (vnot_conv DPR:$src1)))))]>;
2051 def VBSLq : N3VX<1, 0, 0b01, 0b0001, 1, 1, (outs QPR:$dst),
2052 (ins QPR:$src1, QPR:$src2, QPR:$src3), IIC_VCNTiQ,
2053 "vbsl", "$dst, $src2, $src3", "$src1 = $dst",
2055 (v4i32 (or (and QPR:$src2, QPR:$src1),
2056 (and QPR:$src3, (vnot_conv QPR:$src1)))))]>;
2058 // VBIF : Vector Bitwise Insert if False
2059 // like VBSL but with: "vbif $dst, $src3, $src1", "$src2 = $dst",
2060 def VBIFd : N3VX<1, 0, 0b11, 0b0001, 0, 1,
2061 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3),
2062 IIC_VBINiD, "vbif", "$dst, $src2, $src3", "$src1 = $dst",
2063 [/* For disassembly only; pattern left blank */]>;
2064 def VBIFq : N3VX<1, 0, 0b11, 0b0001, 1, 1,
2065 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3),
2066 IIC_VBINiQ, "vbif", "$dst, $src2, $src3", "$src1 = $dst",
2067 [/* For disassembly only; pattern left blank */]>;
2069 // VBIT : Vector Bitwise Insert if True
2070 // like VBSL but with: "vbit $dst, $src2, $src1", "$src3 = $dst",
2071 def VBITd : N3VX<1, 0, 0b10, 0b0001, 0, 1,
2072 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3),
2073 IIC_VBINiD, "vbit", "$dst, $src2, $src3", "$src1 = $dst",
2074 [/* For disassembly only; pattern left blank */]>;
2075 def VBITq : N3VX<1, 0, 0b10, 0b0001, 1, 1,
2076 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3),
2077 IIC_VBINiQ, "vbit", "$dst, $src2, $src3", "$src1 = $dst",
2078 [/* For disassembly only; pattern left blank */]>;
2080 // VBIT/VBIF are not yet implemented. The TwoAddress pass will not go looking
2081 // for equivalent operations with different register constraints; it just
2084 // Vector Absolute Differences.
2086 // VABD : Vector Absolute Difference
2087 defm VABDs : N3VInt_QHS<0, 0, 0b0111, 0, IIC_VBINi4D, IIC_VBINi4D,
2088 IIC_VBINi4Q, IIC_VBINi4Q,
2089 "vabd", "s", int_arm_neon_vabds, 0>;
2090 defm VABDu : N3VInt_QHS<1, 0, 0b0111, 0, IIC_VBINi4D, IIC_VBINi4D,
2091 IIC_VBINi4Q, IIC_VBINi4Q,
2092 "vabd", "u", int_arm_neon_vabdu, 0>;
2093 def VABDfd : N3VDInt<1, 0, 0b10, 0b1101, 0, IIC_VBIND,
2094 "vabd", "f32", v2f32, v2f32, int_arm_neon_vabds, 0>;
2095 def VABDfq : N3VQInt<1, 0, 0b10, 0b1101, 0, IIC_VBINQ,
2096 "vabd", "f32", v4f32, v4f32, int_arm_neon_vabds, 0>;
2098 // VABDL : Vector Absolute Difference Long (Q = | D - D |)
2099 defm VABDLs : N3VLInt_QHS<0,1,0b0111,0, IIC_VBINi4Q,
2100 "vabdl", "s", int_arm_neon_vabdls, 0>;
2101 defm VABDLu : N3VLInt_QHS<1,1,0b0111,0, IIC_VBINi4Q,
2102 "vabdl", "u", int_arm_neon_vabdlu, 0>;
2104 // VABA : Vector Absolute Difference and Accumulate
2105 defm VABAs : N3VInt3_QHS<0,0,0b0111,1, "vaba", "s", int_arm_neon_vabas>;
2106 defm VABAu : N3VInt3_QHS<1,0,0b0111,1, "vaba", "u", int_arm_neon_vabau>;
2108 // VABAL : Vector Absolute Difference and Accumulate Long (Q += | D - D |)
2109 defm VABALs : N3VLInt3_QHS<0,1,0b0101,0, "vabal", "s", int_arm_neon_vabals>;
2110 defm VABALu : N3VLInt3_QHS<1,1,0b0101,0, "vabal", "u", int_arm_neon_vabalu>;
2112 // Vector Maximum and Minimum.
2114 // VMAX : Vector Maximum
2115 defm VMAXs : N3VInt_QHS<0,0,0b0110,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2116 IIC_VBINi4Q, "vmax", "s", int_arm_neon_vmaxs, 1>;
2117 defm VMAXu : N3VInt_QHS<1,0,0b0110,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2118 IIC_VBINi4Q, "vmax", "u", int_arm_neon_vmaxu, 1>;
2119 def VMAXfd : N3VDInt<0, 0, 0b00, 0b1111, 0, IIC_VBIND, "vmax", "f32",
2120 v2f32, v2f32, int_arm_neon_vmaxs, 1>;
2121 def VMAXfq : N3VQInt<0, 0, 0b00, 0b1111, 0, IIC_VBINQ, "vmax", "f32",
2122 v4f32, v4f32, int_arm_neon_vmaxs, 1>;
2124 // VMIN : Vector Minimum
2125 defm VMINs : N3VInt_QHS<0,0,0b0110,1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2126 IIC_VBINi4Q, "vmin", "s", int_arm_neon_vmins, 1>;
2127 defm VMINu : N3VInt_QHS<1,0,0b0110,1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2128 IIC_VBINi4Q, "vmin", "u", int_arm_neon_vminu, 1>;
2129 def VMINfd : N3VDInt<0, 0, 0b10, 0b1111, 0, IIC_VBIND, "vmin", "f32",
2130 v2f32, v2f32, int_arm_neon_vmins, 1>;
2131 def VMINfq : N3VQInt<0, 0, 0b10, 0b1111, 0, IIC_VBINQ, "vmin", "f32",
2132 v4f32, v4f32, int_arm_neon_vmins, 1>;
2134 // Vector Pairwise Operations.
2136 // VPADD : Vector Pairwise Add
2137 def VPADDi8 : N3VDInt<0, 0, 0b00, 0b1011, 1, IIC_VBINiD, "vpadd", "i8",
2138 v8i8, v8i8, int_arm_neon_vpadd, 0>;
2139 def VPADDi16 : N3VDInt<0, 0, 0b01, 0b1011, 1, IIC_VBINiD, "vpadd", "i16",
2140 v4i16, v4i16, int_arm_neon_vpadd, 0>;
2141 def VPADDi32 : N3VDInt<0, 0, 0b10, 0b1011, 1, IIC_VBINiD, "vpadd", "i32",
2142 v2i32, v2i32, int_arm_neon_vpadd, 0>;
2143 def VPADDf : N3VDInt<1, 0, 0b00, 0b1101, 0, IIC_VBIND, "vpadd", "f32",
2144 v2f32, v2f32, int_arm_neon_vpadd, 0>;
2146 // VPADDL : Vector Pairwise Add Long
2147 defm VPADDLs : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00100, 0, "vpaddl", "s",
2148 int_arm_neon_vpaddls>;
2149 defm VPADDLu : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00101, 0, "vpaddl", "u",
2150 int_arm_neon_vpaddlu>;
2152 // VPADAL : Vector Pairwise Add and Accumulate Long
2153 defm VPADALs : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01100, 0, "vpadal", "s",
2154 int_arm_neon_vpadals>;
2155 defm VPADALu : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01101, 0, "vpadal", "u",
2156 int_arm_neon_vpadalu>;
2158 // VPMAX : Vector Pairwise Maximum
2159 def VPMAXs8 : N3VDInt<0, 0, 0b00, 0b1010, 0, IIC_VBINi4D, "vpmax", "s8",
2160 v8i8, v8i8, int_arm_neon_vpmaxs, 0>;
2161 def VPMAXs16 : N3VDInt<0, 0, 0b01, 0b1010, 0, IIC_VBINi4D, "vpmax", "s16",
2162 v4i16, v4i16, int_arm_neon_vpmaxs, 0>;
2163 def VPMAXs32 : N3VDInt<0, 0, 0b10, 0b1010, 0, IIC_VBINi4D, "vpmax", "s32",
2164 v2i32, v2i32, int_arm_neon_vpmaxs, 0>;
2165 def VPMAXu8 : N3VDInt<1, 0, 0b00, 0b1010, 0, IIC_VBINi4D, "vpmax", "u8",
2166 v8i8, v8i8, int_arm_neon_vpmaxu, 0>;
2167 def VPMAXu16 : N3VDInt<1, 0, 0b01, 0b1010, 0, IIC_VBINi4D, "vpmax", "u16",
2168 v4i16, v4i16, int_arm_neon_vpmaxu, 0>;
2169 def VPMAXu32 : N3VDInt<1, 0, 0b10, 0b1010, 0, IIC_VBINi4D, "vpmax", "u32",
2170 v2i32, v2i32, int_arm_neon_vpmaxu, 0>;
2171 def VPMAXf : N3VDInt<1, 0, 0b00, 0b1111, 0, IIC_VBINi4D, "vpmax", "f32",
2172 v2f32, v2f32, int_arm_neon_vpmaxs, 0>;
2174 // VPMIN : Vector Pairwise Minimum
2175 def VPMINs8 : N3VDInt<0, 0, 0b00, 0b1010, 1, IIC_VBINi4D, "vpmin", "s8",
2176 v8i8, v8i8, int_arm_neon_vpmins, 0>;
2177 def VPMINs16 : N3VDInt<0, 0, 0b01, 0b1010, 1, IIC_VBINi4D, "vpmin", "s16",
2178 v4i16, v4i16, int_arm_neon_vpmins, 0>;
2179 def VPMINs32 : N3VDInt<0, 0, 0b10, 0b1010, 1, IIC_VBINi4D, "vpmin", "s32",
2180 v2i32, v2i32, int_arm_neon_vpmins, 0>;
2181 def VPMINu8 : N3VDInt<1, 0, 0b00, 0b1010, 1, IIC_VBINi4D, "vpmin", "u8",
2182 v8i8, v8i8, int_arm_neon_vpminu, 0>;
2183 def VPMINu16 : N3VDInt<1, 0, 0b01, 0b1010, 1, IIC_VBINi4D, "vpmin", "u16",
2184 v4i16, v4i16, int_arm_neon_vpminu, 0>;
2185 def VPMINu32 : N3VDInt<1, 0, 0b10, 0b1010, 1, IIC_VBINi4D, "vpmin", "u32",
2186 v2i32, v2i32, int_arm_neon_vpminu, 0>;
2187 def VPMINf : N3VDInt<1, 0, 0b10, 0b1111, 0, IIC_VBINi4D, "vpmin", "f32",
2188 v2f32, v2f32, int_arm_neon_vpmins, 0>;
2190 // Vector Reciprocal and Reciprocal Square Root Estimate and Step.
2192 // VRECPE : Vector Reciprocal Estimate
2193 def VRECPEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
2194 IIC_VUNAD, "vrecpe", "u32",
2195 v2i32, v2i32, int_arm_neon_vrecpe>;
2196 def VRECPEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
2197 IIC_VUNAQ, "vrecpe", "u32",
2198 v4i32, v4i32, int_arm_neon_vrecpe>;
2199 def VRECPEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
2200 IIC_VUNAD, "vrecpe", "f32",
2201 v2f32, v2f32, int_arm_neon_vrecpe>;
2202 def VRECPEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
2203 IIC_VUNAQ, "vrecpe", "f32",
2204 v4f32, v4f32, int_arm_neon_vrecpe>;
2206 // VRECPS : Vector Reciprocal Step
2207 def VRECPSfd : N3VDInt<0, 0, 0b00, 0b1111, 1,
2208 IIC_VRECSD, "vrecps", "f32",
2209 v2f32, v2f32, int_arm_neon_vrecps, 1>;
2210 def VRECPSfq : N3VQInt<0, 0, 0b00, 0b1111, 1,
2211 IIC_VRECSQ, "vrecps", "f32",
2212 v4f32, v4f32, int_arm_neon_vrecps, 1>;
2214 // VRSQRTE : Vector Reciprocal Square Root Estimate
2215 def VRSQRTEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
2216 IIC_VUNAD, "vrsqrte", "u32",
2217 v2i32, v2i32, int_arm_neon_vrsqrte>;
2218 def VRSQRTEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
2219 IIC_VUNAQ, "vrsqrte", "u32",
2220 v4i32, v4i32, int_arm_neon_vrsqrte>;
2221 def VRSQRTEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
2222 IIC_VUNAD, "vrsqrte", "f32",
2223 v2f32, v2f32, int_arm_neon_vrsqrte>;
2224 def VRSQRTEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
2225 IIC_VUNAQ, "vrsqrte", "f32",
2226 v4f32, v4f32, int_arm_neon_vrsqrte>;
2228 // VRSQRTS : Vector Reciprocal Square Root Step
2229 def VRSQRTSfd : N3VDInt<0, 0, 0b10, 0b1111, 1,
2230 IIC_VRECSD, "vrsqrts", "f32",
2231 v2f32, v2f32, int_arm_neon_vrsqrts, 1>;
2232 def VRSQRTSfq : N3VQInt<0, 0, 0b10, 0b1111, 1,
2233 IIC_VRECSQ, "vrsqrts", "f32",
2234 v4f32, v4f32, int_arm_neon_vrsqrts, 1>;
2238 // VSHL : Vector Shift
2239 defm VSHLs : N3VInt_QHSD<0, 0, 0b0100, 0, IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ,
2240 IIC_VSHLiQ, "vshl", "s", int_arm_neon_vshifts, 0>;
2241 defm VSHLu : N3VInt_QHSD<1, 0, 0b0100, 0, IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ,
2242 IIC_VSHLiQ, "vshl", "u", int_arm_neon_vshiftu, 0>;
2243 // VSHL : Vector Shift Left (Immediate)
2244 defm VSHLi : N2VSh_QHSD<0, 1, 0b0101, 1, IIC_VSHLiD, "vshl", "i", NEONvshl>;
2245 // VSHR : Vector Shift Right (Immediate)
2246 defm VSHRs : N2VSh_QHSD<0, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "s", NEONvshrs>;
2247 defm VSHRu : N2VSh_QHSD<1, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "u", NEONvshru>;
2249 // VSHLL : Vector Shift Left Long
2250 defm VSHLLs : N2VLSh_QHS<0, 1, 0b1010, 0, 0, 1, "vshll", "s", NEONvshlls>;
2251 defm VSHLLu : N2VLSh_QHS<1, 1, 0b1010, 0, 0, 1, "vshll", "u", NEONvshllu>;
2253 // VSHLL : Vector Shift Left Long (with maximum shift count)
2254 class N2VLShMax<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
2255 bit op6, bit op4, string OpcodeStr, string Dt, ValueType ResTy,
2256 ValueType OpTy, SDNode OpNode>
2257 : N2VLSh<op24, op23, op11_8, op7, op6, op4, OpcodeStr, Dt,
2258 ResTy, OpTy, OpNode> {
2259 let Inst{21-16} = op21_16;
2261 def VSHLLi8 : N2VLShMax<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll", "i8",
2262 v8i16, v8i8, NEONvshlli>;
2263 def VSHLLi16 : N2VLShMax<1, 1, 0b110110, 0b0011, 0, 0, 0, "vshll", "i16",
2264 v4i32, v4i16, NEONvshlli>;
2265 def VSHLLi32 : N2VLShMax<1, 1, 0b111010, 0b0011, 0, 0, 0, "vshll", "i32",
2266 v2i64, v2i32, NEONvshlli>;
2268 // VSHRN : Vector Shift Right and Narrow
2269 defm VSHRN : N2VNSh_HSD<0,1,0b1000,0,0,1, IIC_VSHLiD, "vshrn", "i",
2272 // VRSHL : Vector Rounding Shift
2273 defm VRSHLs : N3VInt_QHSD<0,0,0b0101,0, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2274 IIC_VSHLi4Q, "vrshl", "s", int_arm_neon_vrshifts,0>;
2275 defm VRSHLu : N3VInt_QHSD<1,0,0b0101,0, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2276 IIC_VSHLi4Q, "vrshl", "u", int_arm_neon_vrshiftu,0>;
2277 // VRSHR : Vector Rounding Shift Right
2278 defm VRSHRs : N2VSh_QHSD<0,1,0b0010,1, IIC_VSHLi4D, "vrshr", "s", NEONvrshrs>;
2279 defm VRSHRu : N2VSh_QHSD<1,1,0b0010,1, IIC_VSHLi4D, "vrshr", "u", NEONvrshru>;
2281 // VRSHRN : Vector Rounding Shift Right and Narrow
2282 defm VRSHRN : N2VNSh_HSD<0, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vrshrn", "i",
2285 // VQSHL : Vector Saturating Shift
2286 defm VQSHLs : N3VInt_QHSD<0,0,0b0100,1, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2287 IIC_VSHLi4Q, "vqshl", "s", int_arm_neon_vqshifts,0>;
2288 defm VQSHLu : N3VInt_QHSD<1,0,0b0100,1, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2289 IIC_VSHLi4Q, "vqshl", "u", int_arm_neon_vqshiftu,0>;
2290 // VQSHL : Vector Saturating Shift Left (Immediate)
2291 defm VQSHLsi : N2VSh_QHSD<0,1,0b0111,1, IIC_VSHLi4D, "vqshl", "s", NEONvqshls>;
2292 defm VQSHLui : N2VSh_QHSD<1,1,0b0111,1, IIC_VSHLi4D, "vqshl", "u", NEONvqshlu>;
2293 // VQSHLU : Vector Saturating Shift Left (Immediate, Unsigned)
2294 defm VQSHLsu : N2VSh_QHSD<1,1,0b0110,1, IIC_VSHLi4D, "vqshlu","s",NEONvqshlsu>;
2296 // VQSHRN : Vector Saturating Shift Right and Narrow
2297 defm VQSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "s",
2299 defm VQSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "u",
2302 // VQSHRUN : Vector Saturating Shift Right and Narrow (Unsigned)
2303 defm VQSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 0, 1, IIC_VSHLi4D, "vqshrun", "s",
2306 // VQRSHL : Vector Saturating Rounding Shift
2307 defm VQRSHLs : N3VInt_QHSD<0,0,0b0101,1, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2308 IIC_VSHLi4Q, "vqrshl", "s",
2309 int_arm_neon_vqrshifts, 0>;
2310 defm VQRSHLu : N3VInt_QHSD<1,0,0b0101,1, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2311 IIC_VSHLi4Q, "vqrshl", "u",
2312 int_arm_neon_vqrshiftu, 0>;
2314 // VQRSHRN : Vector Saturating Rounding Shift Right and Narrow
2315 defm VQRSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "s",
2317 defm VQRSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "u",
2320 // VQRSHRUN : Vector Saturating Rounding Shift Right and Narrow (Unsigned)
2321 defm VQRSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vqrshrun", "s",
2324 // VSRA : Vector Shift Right and Accumulate
2325 defm VSRAs : N2VShAdd_QHSD<0, 1, 0b0001, 1, "vsra", "s", NEONvshrs>;
2326 defm VSRAu : N2VShAdd_QHSD<1, 1, 0b0001, 1, "vsra", "u", NEONvshru>;
2327 // VRSRA : Vector Rounding Shift Right and Accumulate
2328 defm VRSRAs : N2VShAdd_QHSD<0, 1, 0b0011, 1, "vrsra", "s", NEONvrshrs>;
2329 defm VRSRAu : N2VShAdd_QHSD<1, 1, 0b0011, 1, "vrsra", "u", NEONvrshru>;
2331 // VSLI : Vector Shift Left and Insert
2332 defm VSLI : N2VShIns_QHSD<1, 1, 0b0101, 1, "vsli", NEONvsli>;
2333 // VSRI : Vector Shift Right and Insert
2334 defm VSRI : N2VShIns_QHSD<1, 1, 0b0100, 1, "vsri", NEONvsri>;
2336 // Vector Absolute and Saturating Absolute.
2338 // VABS : Vector Absolute Value
2339 defm VABS : N2VInt_QHS<0b11, 0b11, 0b01, 0b00110, 0,
2340 IIC_VUNAiD, IIC_VUNAiQ, "vabs", "s",
2342 def VABSfd : N2VDInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
2343 IIC_VUNAD, "vabs", "f32",
2344 v2f32, v2f32, int_arm_neon_vabs>;
2345 def VABSfq : N2VQInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
2346 IIC_VUNAQ, "vabs", "f32",
2347 v4f32, v4f32, int_arm_neon_vabs>;
2349 // VQABS : Vector Saturating Absolute Value
2350 defm VQABS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01110, 0,
2351 IIC_VQUNAiD, IIC_VQUNAiQ, "vqabs", "s",
2352 int_arm_neon_vqabs>;
2356 def vneg : PatFrag<(ops node:$in), (sub immAllZerosV, node:$in)>;
2357 def vneg_conv : PatFrag<(ops node:$in), (sub immAllZerosV_bc, node:$in)>;
2359 class VNEGD<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
2360 : N2V<0b11, 0b11, size, 0b01, 0b00111, 0, 0, (outs DPR:$dst), (ins DPR:$src),
2361 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
2362 [(set DPR:$dst, (Ty (vneg DPR:$src)))]>;
2363 class VNEGQ<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
2364 : N2V<0b11, 0b11, size, 0b01, 0b00111, 1, 0, (outs QPR:$dst), (ins QPR:$src),
2365 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
2366 [(set QPR:$dst, (Ty (vneg QPR:$src)))]>;
2368 // VNEG : Vector Negate
2369 def VNEGs8d : VNEGD<0b00, "vneg", "s8", v8i8>;
2370 def VNEGs16d : VNEGD<0b01, "vneg", "s16", v4i16>;
2371 def VNEGs32d : VNEGD<0b10, "vneg", "s32", v2i32>;
2372 def VNEGs8q : VNEGQ<0b00, "vneg", "s8", v16i8>;
2373 def VNEGs16q : VNEGQ<0b01, "vneg", "s16", v8i16>;
2374 def VNEGs32q : VNEGQ<0b10, "vneg", "s32", v4i32>;
2376 // VNEG : Vector Negate (floating-point)
2377 def VNEGfd : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
2378 (outs DPR:$dst), (ins DPR:$src), IIC_VUNAD,
2379 "vneg", "f32", "$dst, $src", "",
2380 [(set DPR:$dst, (v2f32 (fneg DPR:$src)))]>;
2381 def VNEGf32q : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 1, 0,
2382 (outs QPR:$dst), (ins QPR:$src), IIC_VUNAQ,
2383 "vneg", "f32", "$dst, $src", "",
2384 [(set QPR:$dst, (v4f32 (fneg QPR:$src)))]>;
2386 def : Pat<(v8i8 (vneg_conv DPR:$src)), (VNEGs8d DPR:$src)>;
2387 def : Pat<(v4i16 (vneg_conv DPR:$src)), (VNEGs16d DPR:$src)>;
2388 def : Pat<(v2i32 (vneg_conv DPR:$src)), (VNEGs32d DPR:$src)>;
2389 def : Pat<(v16i8 (vneg_conv QPR:$src)), (VNEGs8q QPR:$src)>;
2390 def : Pat<(v8i16 (vneg_conv QPR:$src)), (VNEGs16q QPR:$src)>;
2391 def : Pat<(v4i32 (vneg_conv QPR:$src)), (VNEGs32q QPR:$src)>;
2393 // VQNEG : Vector Saturating Negate
2394 defm VQNEG : N2VInt_QHS<0b11, 0b11, 0b00, 0b01111, 0,
2395 IIC_VQUNAiD, IIC_VQUNAiQ, "vqneg", "s",
2396 int_arm_neon_vqneg>;
2398 // Vector Bit Counting Operations.
2400 // VCLS : Vector Count Leading Sign Bits
2401 defm VCLS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01000, 0,
2402 IIC_VCNTiD, IIC_VCNTiQ, "vcls", "s",
2404 // VCLZ : Vector Count Leading Zeros
2405 defm VCLZ : N2VInt_QHS<0b11, 0b11, 0b00, 0b01001, 0,
2406 IIC_VCNTiD, IIC_VCNTiQ, "vclz", "i",
2408 // VCNT : Vector Count One Bits
2409 def VCNTd : N2VDInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
2410 IIC_VCNTiD, "vcnt", "8",
2411 v8i8, v8i8, int_arm_neon_vcnt>;
2412 def VCNTq : N2VQInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
2413 IIC_VCNTiQ, "vcnt", "8",
2414 v16i8, v16i8, int_arm_neon_vcnt>;
2416 // Vector Move Operations.
2418 // VMOV : Vector Move (Register)
2420 def VMOVDneon: N3VX<0, 0, 0b10, 0b0001, 0, 1, (outs DPR:$dst), (ins DPR:$src),
2421 IIC_VMOVD, "vmov", "$dst, $src", "", []>;
2422 def VMOVQ : N3VX<0, 0, 0b10, 0b0001, 1, 1, (outs QPR:$dst), (ins QPR:$src),
2423 IIC_VMOVD, "vmov", "$dst, $src", "", []>;
2425 // VMOV : Vector Move (Immediate)
2427 // VMOV_get_imm8 xform function: convert build_vector to VMOV.i8 imm.
2428 def VMOV_get_imm8 : SDNodeXForm<build_vector, [{
2429 return ARM::getVMOVImm(N, 1, *CurDAG);
2431 def vmovImm8 : PatLeaf<(build_vector), [{
2432 return ARM::getVMOVImm(N, 1, *CurDAG).getNode() != 0;
2435 // VMOV_get_imm16 xform function: convert build_vector to VMOV.i16 imm.
2436 def VMOV_get_imm16 : SDNodeXForm<build_vector, [{
2437 return ARM::getVMOVImm(N, 2, *CurDAG);
2439 def vmovImm16 : PatLeaf<(build_vector), [{
2440 return ARM::getVMOVImm(N, 2, *CurDAG).getNode() != 0;
2441 }], VMOV_get_imm16>;
2443 // VMOV_get_imm32 xform function: convert build_vector to VMOV.i32 imm.
2444 def VMOV_get_imm32 : SDNodeXForm<build_vector, [{
2445 return ARM::getVMOVImm(N, 4, *CurDAG);
2447 def vmovImm32 : PatLeaf<(build_vector), [{
2448 return ARM::getVMOVImm(N, 4, *CurDAG).getNode() != 0;
2449 }], VMOV_get_imm32>;
2451 // VMOV_get_imm64 xform function: convert build_vector to VMOV.i64 imm.
2452 def VMOV_get_imm64 : SDNodeXForm<build_vector, [{
2453 return ARM::getVMOVImm(N, 8, *CurDAG);
2455 def vmovImm64 : PatLeaf<(build_vector), [{
2456 return ARM::getVMOVImm(N, 8, *CurDAG).getNode() != 0;
2457 }], VMOV_get_imm64>;
2459 // Note: Some of the cmode bits in the following VMOV instructions need to
2460 // be encoded based on the immed values.
2462 def VMOVv8i8 : N1ModImm<1, 0b000, 0b1110, 0, 0, 0, 1, (outs DPR:$dst),
2463 (ins h8imm:$SIMM), IIC_VMOVImm,
2464 "vmov", "i8", "$dst, $SIMM", "",
2465 [(set DPR:$dst, (v8i8 vmovImm8:$SIMM))]>;
2466 def VMOVv16i8 : N1ModImm<1, 0b000, 0b1110, 0, 1, 0, 1, (outs QPR:$dst),
2467 (ins h8imm:$SIMM), IIC_VMOVImm,
2468 "vmov", "i8", "$dst, $SIMM", "",
2469 [(set QPR:$dst, (v16i8 vmovImm8:$SIMM))]>;
2471 def VMOVv4i16 : N1ModImm<1, 0b000, {1,0,?,?}, 0, 0, {?}, 1, (outs DPR:$dst),
2472 (ins h16imm:$SIMM), IIC_VMOVImm,
2473 "vmov", "i16", "$dst, $SIMM", "",
2474 [(set DPR:$dst, (v4i16 vmovImm16:$SIMM))]>;
2475 def VMOVv8i16 : N1ModImm<1, 0b000, {1,0,?,?}, 0, 1, {?}, 1, (outs QPR:$dst),
2476 (ins h16imm:$SIMM), IIC_VMOVImm,
2477 "vmov", "i16", "$dst, $SIMM", "",
2478 [(set QPR:$dst, (v8i16 vmovImm16:$SIMM))]>;
2480 def VMOVv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, {?}, 1, (outs DPR:$dst),
2481 (ins h32imm:$SIMM), IIC_VMOVImm,
2482 "vmov", "i32", "$dst, $SIMM", "",
2483 [(set DPR:$dst, (v2i32 vmovImm32:$SIMM))]>;
2484 def VMOVv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, {?}, 1, (outs QPR:$dst),
2485 (ins h32imm:$SIMM), IIC_VMOVImm,
2486 "vmov", "i32", "$dst, $SIMM", "",
2487 [(set QPR:$dst, (v4i32 vmovImm32:$SIMM))]>;
2489 def VMOVv1i64 : N1ModImm<1, 0b000, 0b1110, 0, 0, 1, 1, (outs DPR:$dst),
2490 (ins h64imm:$SIMM), IIC_VMOVImm,
2491 "vmov", "i64", "$dst, $SIMM", "",
2492 [(set DPR:$dst, (v1i64 vmovImm64:$SIMM))]>;
2493 def VMOVv2i64 : N1ModImm<1, 0b000, 0b1110, 0, 1, 1, 1, (outs QPR:$dst),
2494 (ins h64imm:$SIMM), IIC_VMOVImm,
2495 "vmov", "i64", "$dst, $SIMM", "",
2496 [(set QPR:$dst, (v2i64 vmovImm64:$SIMM))]>;
2498 // VMOV : Vector Get Lane (move scalar to ARM core register)
2500 def VGETLNs8 : NVGetLane<{1,1,1,0,0,1,?,1}, 0b1011, {?,?},
2501 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2502 IIC_VMOVSI, "vmov", "s8", "$dst, $src[$lane]",
2503 [(set GPR:$dst, (NEONvgetlanes (v8i8 DPR:$src),
2505 def VGETLNs16 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, {?,1},
2506 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2507 IIC_VMOVSI, "vmov", "s16", "$dst, $src[$lane]",
2508 [(set GPR:$dst, (NEONvgetlanes (v4i16 DPR:$src),
2510 def VGETLNu8 : NVGetLane<{1,1,1,0,1,1,?,1}, 0b1011, {?,?},
2511 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2512 IIC_VMOVSI, "vmov", "u8", "$dst, $src[$lane]",
2513 [(set GPR:$dst, (NEONvgetlaneu (v8i8 DPR:$src),
2515 def VGETLNu16 : NVGetLane<{1,1,1,0,1,0,?,1}, 0b1011, {?,1},
2516 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2517 IIC_VMOVSI, "vmov", "u16", "$dst, $src[$lane]",
2518 [(set GPR:$dst, (NEONvgetlaneu (v4i16 DPR:$src),
2520 def VGETLNi32 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, 0b00,
2521 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2522 IIC_VMOVSI, "vmov", "32", "$dst, $src[$lane]",
2523 [(set GPR:$dst, (extractelt (v2i32 DPR:$src),
2525 // def VGETLNf32: see FMRDH and FMRDL in ARMInstrVFP.td
2526 def : Pat<(NEONvgetlanes (v16i8 QPR:$src), imm:$lane),
2527 (VGETLNs8 (v8i8 (EXTRACT_SUBREG QPR:$src,
2528 (DSubReg_i8_reg imm:$lane))),
2529 (SubReg_i8_lane imm:$lane))>;
2530 def : Pat<(NEONvgetlanes (v8i16 QPR:$src), imm:$lane),
2531 (VGETLNs16 (v4i16 (EXTRACT_SUBREG QPR:$src,
2532 (DSubReg_i16_reg imm:$lane))),
2533 (SubReg_i16_lane imm:$lane))>;
2534 def : Pat<(NEONvgetlaneu (v16i8 QPR:$src), imm:$lane),
2535 (VGETLNu8 (v8i8 (EXTRACT_SUBREG QPR:$src,
2536 (DSubReg_i8_reg imm:$lane))),
2537 (SubReg_i8_lane imm:$lane))>;
2538 def : Pat<(NEONvgetlaneu (v8i16 QPR:$src), imm:$lane),
2539 (VGETLNu16 (v4i16 (EXTRACT_SUBREG QPR:$src,
2540 (DSubReg_i16_reg imm:$lane))),
2541 (SubReg_i16_lane imm:$lane))>;
2542 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
2543 (VGETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src,
2544 (DSubReg_i32_reg imm:$lane))),
2545 (SubReg_i32_lane imm:$lane))>;
2546 def : Pat<(extractelt (v2f32 DPR:$src1), imm:$src2),
2547 (EXTRACT_SUBREG (v2f32 (COPY_TO_REGCLASS (v2f32 DPR:$src1),DPR_VFP2)),
2548 (SSubReg_f32_reg imm:$src2))>;
2549 def : Pat<(extractelt (v4f32 QPR:$src1), imm:$src2),
2550 (EXTRACT_SUBREG (v4f32 (COPY_TO_REGCLASS (v4f32 QPR:$src1),QPR_VFP2)),
2551 (SSubReg_f32_reg imm:$src2))>;
2552 //def : Pat<(extractelt (v2i64 QPR:$src1), imm:$src2),
2553 // (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
2554 def : Pat<(extractelt (v2f64 QPR:$src1), imm:$src2),
2555 (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
2558 // VMOV : Vector Set Lane (move ARM core register to scalar)
2560 let Constraints = "$src1 = $dst" in {
2561 def VSETLNi8 : NVSetLane<{1,1,1,0,0,1,?,0}, 0b1011, {?,?}, (outs DPR:$dst),
2562 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
2563 IIC_VMOVISL, "vmov", "8", "$dst[$lane], $src2",
2564 [(set DPR:$dst, (vector_insert (v8i8 DPR:$src1),
2565 GPR:$src2, imm:$lane))]>;
2566 def VSETLNi16 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, {?,1}, (outs DPR:$dst),
2567 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
2568 IIC_VMOVISL, "vmov", "16", "$dst[$lane], $src2",
2569 [(set DPR:$dst, (vector_insert (v4i16 DPR:$src1),
2570 GPR:$src2, imm:$lane))]>;
2571 def VSETLNi32 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, 0b00, (outs DPR:$dst),
2572 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
2573 IIC_VMOVISL, "vmov", "32", "$dst[$lane], $src2",
2574 [(set DPR:$dst, (insertelt (v2i32 DPR:$src1),
2575 GPR:$src2, imm:$lane))]>;
2577 def : Pat<(vector_insert (v16i8 QPR:$src1), GPR:$src2, imm:$lane),
2578 (v16i8 (INSERT_SUBREG QPR:$src1,
2579 (VSETLNi8 (v8i8 (EXTRACT_SUBREG QPR:$src1,
2580 (DSubReg_i8_reg imm:$lane))),
2581 GPR:$src2, (SubReg_i8_lane imm:$lane)),
2582 (DSubReg_i8_reg imm:$lane)))>;
2583 def : Pat<(vector_insert (v8i16 QPR:$src1), GPR:$src2, imm:$lane),
2584 (v8i16 (INSERT_SUBREG QPR:$src1,
2585 (VSETLNi16 (v4i16 (EXTRACT_SUBREG QPR:$src1,
2586 (DSubReg_i16_reg imm:$lane))),
2587 GPR:$src2, (SubReg_i16_lane imm:$lane)),
2588 (DSubReg_i16_reg imm:$lane)))>;
2589 def : Pat<(insertelt (v4i32 QPR:$src1), GPR:$src2, imm:$lane),
2590 (v4i32 (INSERT_SUBREG QPR:$src1,
2591 (VSETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src1,
2592 (DSubReg_i32_reg imm:$lane))),
2593 GPR:$src2, (SubReg_i32_lane imm:$lane)),
2594 (DSubReg_i32_reg imm:$lane)))>;
2596 def : Pat<(v2f32 (insertelt DPR:$src1, SPR:$src2, imm:$src3)),
2597 (INSERT_SUBREG (v2f32 (COPY_TO_REGCLASS DPR:$src1, DPR_VFP2)),
2598 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
2599 def : Pat<(v4f32 (insertelt QPR:$src1, SPR:$src2, imm:$src3)),
2600 (INSERT_SUBREG (v4f32 (COPY_TO_REGCLASS QPR:$src1, QPR_VFP2)),
2601 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
2603 //def : Pat<(v2i64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
2604 // (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
2605 def : Pat<(v2f64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
2606 (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
2608 def : Pat<(v2f32 (scalar_to_vector SPR:$src)),
2609 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$src, arm_ssubreg_0)>;
2610 def : Pat<(v2f64 (scalar_to_vector DPR:$src)),
2611 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), DPR:$src, arm_dsubreg_0)>;
2612 def : Pat<(v4f32 (scalar_to_vector SPR:$src)),
2613 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), SPR:$src, arm_ssubreg_0)>;
2615 def : Pat<(v8i8 (scalar_to_vector GPR:$src)),
2616 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
2617 def : Pat<(v4i16 (scalar_to_vector GPR:$src)),
2618 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
2619 def : Pat<(v2i32 (scalar_to_vector GPR:$src)),
2620 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
2622 def : Pat<(v16i8 (scalar_to_vector GPR:$src)),
2623 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
2624 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
2626 def : Pat<(v8i16 (scalar_to_vector GPR:$src)),
2627 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
2628 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
2630 def : Pat<(v4i32 (scalar_to_vector GPR:$src)),
2631 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
2632 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
2635 // VDUP : Vector Duplicate (from ARM core register to all elements)
2637 class VDUPD<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
2638 : NVDup<opcod1, 0b1011, opcod3, (outs DPR:$dst), (ins GPR:$src),
2639 IIC_VMOVIS, "vdup", Dt, "$dst, $src",
2640 [(set DPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
2641 class VDUPQ<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
2642 : NVDup<opcod1, 0b1011, opcod3, (outs QPR:$dst), (ins GPR:$src),
2643 IIC_VMOVIS, "vdup", Dt, "$dst, $src",
2644 [(set QPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
2646 def VDUP8d : VDUPD<0b11101100, 0b00, "8", v8i8>;
2647 def VDUP16d : VDUPD<0b11101000, 0b01, "16", v4i16>;
2648 def VDUP32d : VDUPD<0b11101000, 0b00, "32", v2i32>;
2649 def VDUP8q : VDUPQ<0b11101110, 0b00, "8", v16i8>;
2650 def VDUP16q : VDUPQ<0b11101010, 0b01, "16", v8i16>;
2651 def VDUP32q : VDUPQ<0b11101010, 0b00, "32", v4i32>;
2653 def VDUPfd : NVDup<0b11101000, 0b1011, 0b00, (outs DPR:$dst), (ins GPR:$src),
2654 IIC_VMOVIS, "vdup", "32", "$dst, $src",
2655 [(set DPR:$dst, (v2f32 (NEONvdup
2656 (f32 (bitconvert GPR:$src)))))]>;
2657 def VDUPfq : NVDup<0b11101010, 0b1011, 0b00, (outs QPR:$dst), (ins GPR:$src),
2658 IIC_VMOVIS, "vdup", "32", "$dst, $src",
2659 [(set QPR:$dst, (v4f32 (NEONvdup
2660 (f32 (bitconvert GPR:$src)))))]>;
2662 // VDUP : Vector Duplicate Lane (from scalar to all elements)
2664 class VDUPLND<bits<2> op19_18, bits<2> op17_16,
2665 string OpcodeStr, string Dt, ValueType Ty>
2666 : N2V<0b11, 0b11, op19_18, op17_16, 0b11000, 0, 0,
2667 (outs DPR:$dst), (ins DPR:$src, nohash_imm:$lane), IIC_VMOVD,
2668 OpcodeStr, Dt, "$dst, $src[$lane]", "",
2669 [(set DPR:$dst, (Ty (NEONvduplane (Ty DPR:$src), imm:$lane)))]>;
2671 class VDUPLNQ<bits<2> op19_18, bits<2> op17_16, string OpcodeStr, string Dt,
2672 ValueType ResTy, ValueType OpTy>
2673 : N2V<0b11, 0b11, op19_18, op17_16, 0b11000, 1, 0,
2674 (outs QPR:$dst), (ins DPR:$src, nohash_imm:$lane), IIC_VMOVD,
2675 OpcodeStr, Dt, "$dst, $src[$lane]", "",
2676 [(set QPR:$dst, (ResTy (NEONvduplane (OpTy DPR:$src), imm:$lane)))]>;
2678 // Inst{19-16} is partially specified depending on the element size.
2680 def VDUPLN8d : VDUPLND<{?,?}, {?,1}, "vdup", "8", v8i8>;
2681 def VDUPLN16d : VDUPLND<{?,?}, {1,0}, "vdup", "16", v4i16>;
2682 def VDUPLN32d : VDUPLND<{?,1}, {0,0}, "vdup", "32", v2i32>;
2683 def VDUPLNfd : VDUPLND<{?,1}, {0,0}, "vdup", "32", v2f32>;
2684 def VDUPLN8q : VDUPLNQ<{?,?}, {?,1}, "vdup", "8", v16i8, v8i8>;
2685 def VDUPLN16q : VDUPLNQ<{?,?}, {1,0}, "vdup", "16", v8i16, v4i16>;
2686 def VDUPLN32q : VDUPLNQ<{?,1}, {0,0}, "vdup", "32", v4i32, v2i32>;
2687 def VDUPLNfq : VDUPLNQ<{?,1}, {0,0}, "vdup", "32", v4f32, v2f32>;
2689 def : Pat<(v16i8 (NEONvduplane (v16i8 QPR:$src), imm:$lane)),
2690 (v16i8 (VDUPLN8q (v8i8 (EXTRACT_SUBREG QPR:$src,
2691 (DSubReg_i8_reg imm:$lane))),
2692 (SubReg_i8_lane imm:$lane)))>;
2693 def : Pat<(v8i16 (NEONvduplane (v8i16 QPR:$src), imm:$lane)),
2694 (v8i16 (VDUPLN16q (v4i16 (EXTRACT_SUBREG QPR:$src,
2695 (DSubReg_i16_reg imm:$lane))),
2696 (SubReg_i16_lane imm:$lane)))>;
2697 def : Pat<(v4i32 (NEONvduplane (v4i32 QPR:$src), imm:$lane)),
2698 (v4i32 (VDUPLN32q (v2i32 (EXTRACT_SUBREG QPR:$src,
2699 (DSubReg_i32_reg imm:$lane))),
2700 (SubReg_i32_lane imm:$lane)))>;
2701 def : Pat<(v4f32 (NEONvduplane (v4f32 QPR:$src), imm:$lane)),
2702 (v4f32 (VDUPLNfq (v2f32 (EXTRACT_SUBREG QPR:$src,
2703 (DSubReg_i32_reg imm:$lane))),
2704 (SubReg_i32_lane imm:$lane)))>;
2706 def VDUPfdf : N2V<0b11, 0b11, {?,1}, {0,0}, 0b11000, 0, 0,
2707 (outs DPR:$dst), (ins SPR:$src),
2708 IIC_VMOVD, "vdup", "32", "$dst, ${src:lane}", "",
2709 [(set DPR:$dst, (v2f32 (NEONvdup (f32 SPR:$src))))]>;
2711 def VDUPfqf : N2V<0b11, 0b11, {?,1}, {0,0}, 0b11000, 1, 0,
2712 (outs QPR:$dst), (ins SPR:$src),
2713 IIC_VMOVD, "vdup", "32", "$dst, ${src:lane}", "",
2714 [(set QPR:$dst, (v4f32 (NEONvdup (f32 SPR:$src))))]>;
2716 def : Pat<(v2i64 (NEONvduplane (v2i64 QPR:$src), imm:$lane)),
2717 (INSERT_SUBREG QPR:$src,
2718 (i64 (EXTRACT_SUBREG QPR:$src,
2719 (DSubReg_f64_reg imm:$lane))),
2720 (DSubReg_f64_other_reg imm:$lane))>;
2721 def : Pat<(v2f64 (NEONvduplane (v2f64 QPR:$src), imm:$lane)),
2722 (INSERT_SUBREG QPR:$src,
2723 (f64 (EXTRACT_SUBREG QPR:$src,
2724 (DSubReg_f64_reg imm:$lane))),
2725 (DSubReg_f64_other_reg imm:$lane))>;
2727 // VMOVN : Vector Narrowing Move
2728 defm VMOVN : N2VNInt_HSD<0b11,0b11,0b10,0b00100,0,0, IIC_VMOVD,
2729 "vmovn", "i", int_arm_neon_vmovn>;
2730 // VQMOVN : Vector Saturating Narrowing Move
2731 defm VQMOVNs : N2VNInt_HSD<0b11,0b11,0b10,0b00101,0,0, IIC_VQUNAiD,
2732 "vqmovn", "s", int_arm_neon_vqmovns>;
2733 defm VQMOVNu : N2VNInt_HSD<0b11,0b11,0b10,0b00101,1,0, IIC_VQUNAiD,
2734 "vqmovn", "u", int_arm_neon_vqmovnu>;
2735 defm VQMOVNsu : N2VNInt_HSD<0b11,0b11,0b10,0b00100,1,0, IIC_VQUNAiD,
2736 "vqmovun", "s", int_arm_neon_vqmovnsu>;
2737 // VMOVL : Vector Lengthening Move
2738 defm VMOVLs : N2VLInt_QHS<0b01,0b10100,0,1, "vmovl", "s",
2739 int_arm_neon_vmovls>;
2740 defm VMOVLu : N2VLInt_QHS<0b11,0b10100,0,1, "vmovl", "u",
2741 int_arm_neon_vmovlu>;
2743 // Vector Conversions.
2745 // VCVT : Vector Convert Between Floating-Point and Integers
2746 def VCVTf2sd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
2747 v2i32, v2f32, fp_to_sint>;
2748 def VCVTf2ud : N2VD<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
2749 v2i32, v2f32, fp_to_uint>;
2750 def VCVTs2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
2751 v2f32, v2i32, sint_to_fp>;
2752 def VCVTu2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
2753 v2f32, v2i32, uint_to_fp>;
2755 def VCVTf2sq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
2756 v4i32, v4f32, fp_to_sint>;
2757 def VCVTf2uq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
2758 v4i32, v4f32, fp_to_uint>;
2759 def VCVTs2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
2760 v4f32, v4i32, sint_to_fp>;
2761 def VCVTu2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
2762 v4f32, v4i32, uint_to_fp>;
2764 // VCVT : Vector Convert Between Floating-Point and Fixed-Point.
2765 def VCVTf2xsd : N2VCvtD<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
2766 v2i32, v2f32, int_arm_neon_vcvtfp2fxs>;
2767 def VCVTf2xud : N2VCvtD<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
2768 v2i32, v2f32, int_arm_neon_vcvtfp2fxu>;
2769 def VCVTxs2fd : N2VCvtD<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
2770 v2f32, v2i32, int_arm_neon_vcvtfxs2fp>;
2771 def VCVTxu2fd : N2VCvtD<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
2772 v2f32, v2i32, int_arm_neon_vcvtfxu2fp>;
2774 def VCVTf2xsq : N2VCvtQ<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
2775 v4i32, v4f32, int_arm_neon_vcvtfp2fxs>;
2776 def VCVTf2xuq : N2VCvtQ<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
2777 v4i32, v4f32, int_arm_neon_vcvtfp2fxu>;
2778 def VCVTxs2fq : N2VCvtQ<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
2779 v4f32, v4i32, int_arm_neon_vcvtfxs2fp>;
2780 def VCVTxu2fq : N2VCvtQ<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
2781 v4f32, v4i32, int_arm_neon_vcvtfxu2fp>;
2785 // VREV64 : Vector Reverse elements within 64-bit doublewords
2787 class VREV64D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
2788 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 0, 0, (outs DPR:$dst),
2789 (ins DPR:$src), IIC_VMOVD,
2790 OpcodeStr, Dt, "$dst, $src", "",
2791 [(set DPR:$dst, (Ty (NEONvrev64 (Ty DPR:$src))))]>;
2792 class VREV64Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
2793 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 1, 0, (outs QPR:$dst),
2794 (ins QPR:$src), IIC_VMOVD,
2795 OpcodeStr, Dt, "$dst, $src", "",
2796 [(set QPR:$dst, (Ty (NEONvrev64 (Ty QPR:$src))))]>;
2798 def VREV64d8 : VREV64D<0b00, "vrev64", "8", v8i8>;
2799 def VREV64d16 : VREV64D<0b01, "vrev64", "16", v4i16>;
2800 def VREV64d32 : VREV64D<0b10, "vrev64", "32", v2i32>;
2801 def VREV64df : VREV64D<0b10, "vrev64", "32", v2f32>;
2803 def VREV64q8 : VREV64Q<0b00, "vrev64", "8", v16i8>;
2804 def VREV64q16 : VREV64Q<0b01, "vrev64", "16", v8i16>;
2805 def VREV64q32 : VREV64Q<0b10, "vrev64", "32", v4i32>;
2806 def VREV64qf : VREV64Q<0b10, "vrev64", "32", v4f32>;
2808 // VREV32 : Vector Reverse elements within 32-bit words
2810 class VREV32D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
2811 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 0, 0, (outs DPR:$dst),
2812 (ins DPR:$src), IIC_VMOVD,
2813 OpcodeStr, Dt, "$dst, $src", "",
2814 [(set DPR:$dst, (Ty (NEONvrev32 (Ty DPR:$src))))]>;
2815 class VREV32Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
2816 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 1, 0, (outs QPR:$dst),
2817 (ins QPR:$src), IIC_VMOVD,
2818 OpcodeStr, Dt, "$dst, $src", "",
2819 [(set QPR:$dst, (Ty (NEONvrev32 (Ty QPR:$src))))]>;
2821 def VREV32d8 : VREV32D<0b00, "vrev32", "8", v8i8>;
2822 def VREV32d16 : VREV32D<0b01, "vrev32", "16", v4i16>;
2824 def VREV32q8 : VREV32Q<0b00, "vrev32", "8", v16i8>;
2825 def VREV32q16 : VREV32Q<0b01, "vrev32", "16", v8i16>;
2827 // VREV16 : Vector Reverse elements within 16-bit halfwords
2829 class VREV16D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
2830 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 0, 0, (outs DPR:$dst),
2831 (ins DPR:$src), IIC_VMOVD,
2832 OpcodeStr, Dt, "$dst, $src", "",
2833 [(set DPR:$dst, (Ty (NEONvrev16 (Ty DPR:$src))))]>;
2834 class VREV16Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
2835 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 1, 0, (outs QPR:$dst),
2836 (ins QPR:$src), IIC_VMOVD,
2837 OpcodeStr, Dt, "$dst, $src", "",
2838 [(set QPR:$dst, (Ty (NEONvrev16 (Ty QPR:$src))))]>;
2840 def VREV16d8 : VREV16D<0b00, "vrev16", "8", v8i8>;
2841 def VREV16q8 : VREV16Q<0b00, "vrev16", "8", v16i8>;
2843 // Other Vector Shuffles.
2845 // VEXT : Vector Extract
2847 class VEXTd<string OpcodeStr, string Dt, ValueType Ty>
2848 : N3V<0,1,0b11,{?,?,?,?},0,0, (outs DPR:$dst),
2849 (ins DPR:$lhs, DPR:$rhs, i32imm:$index), IIC_VEXTD,
2850 OpcodeStr, Dt, "$dst, $lhs, $rhs, $index", "",
2851 [(set DPR:$dst, (Ty (NEONvext (Ty DPR:$lhs),
2852 (Ty DPR:$rhs), imm:$index)))]>;
2854 class VEXTq<string OpcodeStr, string Dt, ValueType Ty>
2855 : N3V<0,1,0b11,{?,?,?,?},1,0, (outs QPR:$dst),
2856 (ins QPR:$lhs, QPR:$rhs, i32imm:$index), IIC_VEXTQ,
2857 OpcodeStr, Dt, "$dst, $lhs, $rhs, $index", "",
2858 [(set QPR:$dst, (Ty (NEONvext (Ty QPR:$lhs),
2859 (Ty QPR:$rhs), imm:$index)))]>;
2861 def VEXTd8 : VEXTd<"vext", "8", v8i8>;
2862 def VEXTd16 : VEXTd<"vext", "16", v4i16>;
2863 def VEXTd32 : VEXTd<"vext", "32", v2i32>;
2864 def VEXTdf : VEXTd<"vext", "32", v2f32>;
2866 def VEXTq8 : VEXTq<"vext", "8", v16i8>;
2867 def VEXTq16 : VEXTq<"vext", "16", v8i16>;
2868 def VEXTq32 : VEXTq<"vext", "32", v4i32>;
2869 def VEXTqf : VEXTq<"vext", "32", v4f32>;
2871 // VTRN : Vector Transpose
2873 def VTRNd8 : N2VDShuffle<0b00, 0b00001, "vtrn", "8">;
2874 def VTRNd16 : N2VDShuffle<0b01, 0b00001, "vtrn", "16">;
2875 def VTRNd32 : N2VDShuffle<0b10, 0b00001, "vtrn", "32">;
2877 def VTRNq8 : N2VQShuffle<0b00, 0b00001, IIC_VPERMQ, "vtrn", "8">;
2878 def VTRNq16 : N2VQShuffle<0b01, 0b00001, IIC_VPERMQ, "vtrn", "16">;
2879 def VTRNq32 : N2VQShuffle<0b10, 0b00001, IIC_VPERMQ, "vtrn", "32">;
2881 // VUZP : Vector Unzip (Deinterleave)
2883 def VUZPd8 : N2VDShuffle<0b00, 0b00010, "vuzp", "8">;
2884 def VUZPd16 : N2VDShuffle<0b01, 0b00010, "vuzp", "16">;
2885 def VUZPd32 : N2VDShuffle<0b10, 0b00010, "vuzp", "32">;
2887 def VUZPq8 : N2VQShuffle<0b00, 0b00010, IIC_VPERMQ3, "vuzp", "8">;
2888 def VUZPq16 : N2VQShuffle<0b01, 0b00010, IIC_VPERMQ3, "vuzp", "16">;
2889 def VUZPq32 : N2VQShuffle<0b10, 0b00010, IIC_VPERMQ3, "vuzp", "32">;
2891 // VZIP : Vector Zip (Interleave)
2893 def VZIPd8 : N2VDShuffle<0b00, 0b00011, "vzip", "8">;
2894 def VZIPd16 : N2VDShuffle<0b01, 0b00011, "vzip", "16">;
2895 def VZIPd32 : N2VDShuffle<0b10, 0b00011, "vzip", "32">;
2897 def VZIPq8 : N2VQShuffle<0b00, 0b00011, IIC_VPERMQ3, "vzip", "8">;
2898 def VZIPq16 : N2VQShuffle<0b01, 0b00011, IIC_VPERMQ3, "vzip", "16">;
2899 def VZIPq32 : N2VQShuffle<0b10, 0b00011, IIC_VPERMQ3, "vzip", "32">;
2901 // Vector Table Lookup and Table Extension.
2903 // VTBL : Vector Table Lookup
2905 : N3V<1,1,0b11,0b1000,0,0, (outs DPR:$dst),
2906 (ins DPR:$tbl1, DPR:$src), IIC_VTB1,
2907 "vtbl", "8", "$dst, \\{$tbl1\\}, $src", "",
2908 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl1 DPR:$tbl1, DPR:$src)))]>;
2909 let hasExtraSrcRegAllocReq = 1 in {
2911 : N3V<1,1,0b11,0b1001,0,0, (outs DPR:$dst),
2912 (ins DPR:$tbl1, DPR:$tbl2, DPR:$src), IIC_VTB2,
2913 "vtbl", "8", "$dst, \\{$tbl1, $tbl2\\}, $src", "",
2914 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl2
2915 DPR:$tbl1, DPR:$tbl2, DPR:$src)))]>;
2917 : N3V<1,1,0b11,0b1010,0,0, (outs DPR:$dst),
2918 (ins DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$src), IIC_VTB3,
2919 "vtbl", "8", "$dst, \\{$tbl1, $tbl2, $tbl3\\}, $src", "",
2920 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl3
2921 DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$src)))]>;
2923 : N3V<1,1,0b11,0b1011,0,0, (outs DPR:$dst),
2924 (ins DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$src), IIC_VTB4,
2925 "vtbl", "8", "$dst, \\{$tbl1, $tbl2, $tbl3, $tbl4\\}, $src", "",
2926 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl4 DPR:$tbl1, DPR:$tbl2,
2927 DPR:$tbl3, DPR:$tbl4, DPR:$src)))]>;
2928 } // hasExtraSrcRegAllocReq = 1
2930 // VTBX : Vector Table Extension
2932 : N3V<1,1,0b11,0b1000,1,0, (outs DPR:$dst),
2933 (ins DPR:$orig, DPR:$tbl1, DPR:$src), IIC_VTBX1,
2934 "vtbx", "8", "$dst, \\{$tbl1\\}, $src", "$orig = $dst",
2935 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx1
2936 DPR:$orig, DPR:$tbl1, DPR:$src)))]>;
2937 let hasExtraSrcRegAllocReq = 1 in {
2939 : N3V<1,1,0b11,0b1001,1,0, (outs DPR:$dst),
2940 (ins DPR:$orig, DPR:$tbl1, DPR:$tbl2, DPR:$src), IIC_VTBX2,
2941 "vtbx", "8", "$dst, \\{$tbl1, $tbl2\\}, $src", "$orig = $dst",
2942 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx2
2943 DPR:$orig, DPR:$tbl1, DPR:$tbl2, DPR:$src)))]>;
2945 : N3V<1,1,0b11,0b1010,1,0, (outs DPR:$dst),
2946 (ins DPR:$orig, DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$src), IIC_VTBX3,
2947 "vtbx", "8", "$dst, \\{$tbl1, $tbl2, $tbl3\\}, $src", "$orig = $dst",
2948 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx3 DPR:$orig, DPR:$tbl1,
2949 DPR:$tbl2, DPR:$tbl3, DPR:$src)))]>;
2951 : N3V<1,1,0b11,0b1011,1,0, (outs DPR:$dst), (ins DPR:$orig, DPR:$tbl1,
2952 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$src), IIC_VTBX4,
2953 "vtbx", "8", "$dst, \\{$tbl1, $tbl2, $tbl3, $tbl4\\}, $src",
2955 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx4 DPR:$orig, DPR:$tbl1,
2956 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$src)))]>;
2957 } // hasExtraSrcRegAllocReq = 1
2959 //===----------------------------------------------------------------------===//
2960 // NEON instructions for single-precision FP math
2961 //===----------------------------------------------------------------------===//
2963 class N2VSPat<SDNode OpNode, ValueType ResTy, ValueType OpTy, NeonI Inst>
2964 : NEONFPPat<(ResTy (OpNode SPR:$a)),
2965 (EXTRACT_SUBREG (Inst (INSERT_SUBREG (OpTy (IMPLICIT_DEF)),
2966 SPR:$a, arm_ssubreg_0)),
2969 class N3VSPat<SDNode OpNode, NeonI Inst>
2970 : NEONFPPat<(f32 (OpNode SPR:$a, SPR:$b)),
2971 (EXTRACT_SUBREG (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
2972 SPR:$a, arm_ssubreg_0),
2973 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
2974 SPR:$b, arm_ssubreg_0)),
2977 class N3VSMulOpPat<SDNode MulNode, SDNode OpNode, NeonI Inst>
2978 : NEONFPPat<(f32 (OpNode SPR:$acc, (f32 (MulNode SPR:$a, SPR:$b)))),
2979 (EXTRACT_SUBREG (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
2980 SPR:$acc, arm_ssubreg_0),
2981 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
2982 SPR:$a, arm_ssubreg_0),
2983 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
2984 SPR:$b, arm_ssubreg_0)),
2987 // These need separate instructions because they must use DPR_VFP2 register
2988 // class which have SPR sub-registers.
2990 // Vector Add Operations used for single-precision FP
2991 let neverHasSideEffects = 1 in
2992 def VADDfd_sfp : N3VS<0,0,0b00,0b1101,0, "vadd", "f32", v2f32, v2f32, fadd, 1>;
2993 def : N3VSPat<fadd, VADDfd_sfp>;
2995 // Vector Sub Operations used for single-precision FP
2996 let neverHasSideEffects = 1 in
2997 def VSUBfd_sfp : N3VS<0,0,0b10,0b1101,0, "vsub", "f32", v2f32, v2f32, fsub, 0>;
2998 def : N3VSPat<fsub, VSUBfd_sfp>;
3000 // Vector Multiply Operations used for single-precision FP
3001 let neverHasSideEffects = 1 in
3002 def VMULfd_sfp : N3VS<1,0,0b00,0b1101,1, "vmul", "f32", v2f32, v2f32, fmul, 1>;
3003 def : N3VSPat<fmul, VMULfd_sfp>;
3005 // Vector Multiply-Accumulate/Subtract used for single-precision FP
3006 // vml[as].f32 can cause 4-8 cycle stalls in following ASIMD instructions, so
3007 // we want to avoid them for now. e.g., alternating vmla/vadd instructions.
3009 //let neverHasSideEffects = 1 in
3010 //def VMLAfd_sfp : N3VSMulOp<0,0,0b00,0b1101,1, IIC_VMACD, "vmla", "f32",
3011 // v2f32, fmul, fadd>;
3012 //def : N3VSMulOpPat<fmul, fadd, VMLAfd_sfp>;
3014 //let neverHasSideEffects = 1 in
3015 //def VMLSfd_sfp : N3VSMulOp<0,0,0b10,0b1101,1, IIC_VMACD, "vmls", "f32",
3016 // v2f32, fmul, fsub>;
3017 //def : N3VSMulOpPat<fmul, fsub, VMLSfd_sfp>;
3019 // Vector Absolute used for single-precision FP
3020 let neverHasSideEffects = 1 in
3021 def VABSfd_sfp : N2VSInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0, IIC_VUNAD,
3022 "vabs", "f32", v2f32, v2f32, int_arm_neon_vabs>;
3023 def : N2VSPat<fabs, f32, v2f32, VABSfd_sfp>;
3025 // Vector Negate used for single-precision FP
3026 let neverHasSideEffects = 1 in
3027 def VNEGfd_sfp : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
3028 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), IIC_VUNAD,
3029 "vneg", "f32", "$dst, $src", "", []>;
3030 def : N2VSPat<fneg, f32, v2f32, VNEGfd_sfp>;
3032 // Vector Convert between single-precision FP and integer
3033 let neverHasSideEffects = 1 in
3034 def VCVTf2sd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
3035 v2i32, v2f32, fp_to_sint>;
3036 def : N2VSPat<arm_ftosi, f32, v2f32, VCVTf2sd_sfp>;
3038 let neverHasSideEffects = 1 in
3039 def VCVTf2ud_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
3040 v2i32, v2f32, fp_to_uint>;
3041 def : N2VSPat<arm_ftoui, f32, v2f32, VCVTf2ud_sfp>;
3043 let neverHasSideEffects = 1 in
3044 def VCVTs2fd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
3045 v2f32, v2i32, sint_to_fp>;
3046 def : N2VSPat<arm_sitof, f32, v2i32, VCVTs2fd_sfp>;
3048 let neverHasSideEffects = 1 in
3049 def VCVTu2fd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
3050 v2f32, v2i32, uint_to_fp>;
3051 def : N2VSPat<arm_uitof, f32, v2i32, VCVTu2fd_sfp>;
3053 //===----------------------------------------------------------------------===//
3054 // Non-Instruction Patterns
3055 //===----------------------------------------------------------------------===//
3058 def : Pat<(v1i64 (bitconvert (v2i32 DPR:$src))), (v1i64 DPR:$src)>;
3059 def : Pat<(v1i64 (bitconvert (v4i16 DPR:$src))), (v1i64 DPR:$src)>;
3060 def : Pat<(v1i64 (bitconvert (v8i8 DPR:$src))), (v1i64 DPR:$src)>;
3061 def : Pat<(v1i64 (bitconvert (f64 DPR:$src))), (v1i64 DPR:$src)>;
3062 def : Pat<(v1i64 (bitconvert (v2f32 DPR:$src))), (v1i64 DPR:$src)>;
3063 def : Pat<(v2i32 (bitconvert (v1i64 DPR:$src))), (v2i32 DPR:$src)>;
3064 def : Pat<(v2i32 (bitconvert (v4i16 DPR:$src))), (v2i32 DPR:$src)>;
3065 def : Pat<(v2i32 (bitconvert (v8i8 DPR:$src))), (v2i32 DPR:$src)>;
3066 def : Pat<(v2i32 (bitconvert (f64 DPR:$src))), (v2i32 DPR:$src)>;
3067 def : Pat<(v2i32 (bitconvert (v2f32 DPR:$src))), (v2i32 DPR:$src)>;
3068 def : Pat<(v4i16 (bitconvert (v1i64 DPR:$src))), (v4i16 DPR:$src)>;
3069 def : Pat<(v4i16 (bitconvert (v2i32 DPR:$src))), (v4i16 DPR:$src)>;
3070 def : Pat<(v4i16 (bitconvert (v8i8 DPR:$src))), (v4i16 DPR:$src)>;
3071 def : Pat<(v4i16 (bitconvert (f64 DPR:$src))), (v4i16 DPR:$src)>;
3072 def : Pat<(v4i16 (bitconvert (v2f32 DPR:$src))), (v4i16 DPR:$src)>;
3073 def : Pat<(v8i8 (bitconvert (v1i64 DPR:$src))), (v8i8 DPR:$src)>;
3074 def : Pat<(v8i8 (bitconvert (v2i32 DPR:$src))), (v8i8 DPR:$src)>;
3075 def : Pat<(v8i8 (bitconvert (v4i16 DPR:$src))), (v8i8 DPR:$src)>;
3076 def : Pat<(v8i8 (bitconvert (f64 DPR:$src))), (v8i8 DPR:$src)>;
3077 def : Pat<(v8i8 (bitconvert (v2f32 DPR:$src))), (v8i8 DPR:$src)>;
3078 def : Pat<(f64 (bitconvert (v1i64 DPR:$src))), (f64 DPR:$src)>;
3079 def : Pat<(f64 (bitconvert (v2i32 DPR:$src))), (f64 DPR:$src)>;
3080 def : Pat<(f64 (bitconvert (v4i16 DPR:$src))), (f64 DPR:$src)>;
3081 def : Pat<(f64 (bitconvert (v8i8 DPR:$src))), (f64 DPR:$src)>;
3082 def : Pat<(f64 (bitconvert (v2f32 DPR:$src))), (f64 DPR:$src)>;
3083 def : Pat<(v2f32 (bitconvert (f64 DPR:$src))), (v2f32 DPR:$src)>;
3084 def : Pat<(v2f32 (bitconvert (v1i64 DPR:$src))), (v2f32 DPR:$src)>;
3085 def : Pat<(v2f32 (bitconvert (v2i32 DPR:$src))), (v2f32 DPR:$src)>;
3086 def : Pat<(v2f32 (bitconvert (v4i16 DPR:$src))), (v2f32 DPR:$src)>;
3087 def : Pat<(v2f32 (bitconvert (v8i8 DPR:$src))), (v2f32 DPR:$src)>;
3089 def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (v2i64 QPR:$src)>;
3090 def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (v2i64 QPR:$src)>;
3091 def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (v2i64 QPR:$src)>;
3092 def : Pat<(v2i64 (bitconvert (v2f64 QPR:$src))), (v2i64 QPR:$src)>;
3093 def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (v2i64 QPR:$src)>;
3094 def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (v4i32 QPR:$src)>;
3095 def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (v4i32 QPR:$src)>;
3096 def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (v4i32 QPR:$src)>;
3097 def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (v4i32 QPR:$src)>;
3098 def : Pat<(v4i32 (bitconvert (v4f32 QPR:$src))), (v4i32 QPR:$src)>;
3099 def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (v8i16 QPR:$src)>;
3100 def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (v8i16 QPR:$src)>;
3101 def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (v8i16 QPR:$src)>;
3102 def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (v8i16 QPR:$src)>;
3103 def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (v8i16 QPR:$src)>;
3104 def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (v16i8 QPR:$src)>;
3105 def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (v16i8 QPR:$src)>;
3106 def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (v16i8 QPR:$src)>;
3107 def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (v16i8 QPR:$src)>;
3108 def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (v16i8 QPR:$src)>;
3109 def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (v4f32 QPR:$src)>;
3110 def : Pat<(v4f32 (bitconvert (v4i32 QPR:$src))), (v4f32 QPR:$src)>;
3111 def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (v4f32 QPR:$src)>;
3112 def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (v4f32 QPR:$src)>;
3113 def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (v4f32 QPR:$src)>;
3114 def : Pat<(v2f64 (bitconvert (v2i64 QPR:$src))), (v2f64 QPR:$src)>;
3115 def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (v2f64 QPR:$src)>;
3116 def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 QPR:$src)>;
3117 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 QPR:$src)>;
3118 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 QPR:$src)>;