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 def SDTARMFMAX : SDTypeProfile<1, 2, [SDTCisVT<0, f32>, SDTCisSameAs<0, 1>,
94 def NEONfmax : SDNode<"ARMISD::FMAX", SDTARMFMAX>;
95 def NEONfmin : SDNode<"ARMISD::FMIN", SDTARMFMAX>;
97 //===----------------------------------------------------------------------===//
98 // NEON operand definitions
99 //===----------------------------------------------------------------------===//
101 def h8imm : Operand<i8> {
102 let PrintMethod = "printHex8ImmOperand";
104 def h16imm : Operand<i16> {
105 let PrintMethod = "printHex16ImmOperand";
107 def h32imm : Operand<i32> {
108 let PrintMethod = "printHex32ImmOperand";
110 def h64imm : Operand<i64> {
111 let PrintMethod = "printHex64ImmOperand";
114 //===----------------------------------------------------------------------===//
115 // NEON load / store instructions
116 //===----------------------------------------------------------------------===//
119 // Use vldmia to load a Q register as a D register pair.
120 // This is equivalent to VLDMD except that it has a Q register operand
121 // instead of a pair of D registers.
123 : AXDI5<(outs QPR:$dst), (ins addrmode5:$addr, pred:$p),
124 IndexModeNone, IIC_fpLoadm,
125 "vldm${addr:submode}${p}\t${addr:base}, ${dst:dregpair}", "", []>;
127 : AXDI5<(outs QPR:$dst, GPR:$wb), (ins addrmode5:$addr, pred:$p),
128 IndexModeUpd, IIC_fpLoadm,
129 "vldm${addr:submode}${p}\t${addr:base}!, ${dst:dregpair}",
130 "$addr.base = $wb", []>;
132 // Use vld1 to load a Q register as a D register pair.
133 // This alternative to VLDMQ allows an alignment to be specified.
134 // This is equivalent to VLD1q64 except that it has a Q register operand.
136 : NLdSt<0,0b10,0b1010,0b1100, (outs QPR:$dst), (ins addrmode6:$addr),
137 IIC_VLD1, "vld1", "64", "${dst:dregpair}, $addr", "", []>;
139 : NLdSt<0,0b10,0b1010,0b1100, (outs QPR:$dst, GPR:$wb),
140 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD1, "vld1", "64",
141 "${dst:dregpair}, $addr$offset", "$addr.addr = $wb", []>;
144 let mayStore = 1 in {
145 // Use vstmia to store a Q register as a D register pair.
146 // This is equivalent to VSTMD except that it has a Q register operand
147 // instead of a pair of D registers.
149 : AXDI5<(outs), (ins QPR:$src, addrmode5:$addr, pred:$p),
150 IndexModeNone, IIC_fpStorem,
151 "vstm${addr:submode}${p}\t${addr:base}, ${src:dregpair}", "", []>;
153 : AXDI5<(outs GPR:$wb), (ins QPR:$src, addrmode5:$addr, pred:$p),
154 IndexModeUpd, IIC_fpStorem,
155 "vstm${addr:submode}${p}\t${addr:base}!, ${src:dregpair}",
156 "$addr.base = $wb", []>;
158 // Use vst1 to store a Q register as a D register pair.
159 // This alternative to VSTMQ allows an alignment to be specified.
160 // This is equivalent to VST1q64 except that it has a Q register operand.
162 : NLdSt<0,0b00,0b1010,0b1100, (outs), (ins addrmode6:$addr, QPR:$src),
163 IIC_VST, "vst1", "64", "${src:dregpair}, $addr", "", []>;
165 : NLdSt<0,0b00,0b1010,0b1100, (outs GPR:$wb),
166 (ins addrmode6:$addr, am6offset:$offset, QPR:$src),
167 IIC_VST, "vst1", "64", "{$src:dregpair}, $addr$offset",
168 "$addr.addr = $wb", []>;
171 let mayLoad = 1, hasExtraDefRegAllocReq = 1 in {
173 // VLD1 : Vector Load (multiple single elements)
174 class VLD1D<bits<4> op7_4, string Dt>
175 : NLdSt<0,0b10,0b0111,op7_4, (outs DPR:$dst),
176 (ins addrmode6:$addr), IIC_VLD1,
177 "vld1", Dt, "\\{$dst\\}, $addr", "", []>;
178 class VLD1Q<bits<4> op7_4, string Dt>
179 : NLdSt<0,0b10,0b1010,op7_4, (outs DPR:$dst1, DPR:$dst2),
180 (ins addrmode6:$addr), IIC_VLD1,
181 "vld1", Dt, "\\{$dst1, $dst2\\}, $addr", "", []>;
183 def VLD1d8 : VLD1D<0b0000, "8">;
184 def VLD1d16 : VLD1D<0b0100, "16">;
185 def VLD1d32 : VLD1D<0b1000, "32">;
186 def VLD1d64 : VLD1D<0b1100, "64">;
188 def VLD1q8 : VLD1Q<0b0000, "8">;
189 def VLD1q16 : VLD1Q<0b0100, "16">;
190 def VLD1q32 : VLD1Q<0b1000, "32">;
191 def VLD1q64 : VLD1Q<0b1100, "64">;
193 // ...with address register writeback:
194 class VLD1DWB<bits<4> op7_4, string Dt>
195 : NLdSt<0,0b10,0b0111,op7_4, (outs DPR:$dst, GPR:$wb),
196 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD1,
197 "vld1", Dt, "\\{$dst\\}, $addr$offset",
198 "$addr.addr = $wb", []>;
199 class VLD1QWB<bits<4> op7_4, string Dt>
200 : NLdSt<0,0b10,0b1010,op7_4, (outs QPR:$dst, GPR:$wb),
201 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD1,
202 "vld1", Dt, "${dst:dregpair}, $addr$offset",
203 "$addr.addr = $wb", []>;
205 def VLD1d8_UPD : VLD1DWB<0b0000, "8">;
206 def VLD1d16_UPD : VLD1DWB<0b0100, "16">;
207 def VLD1d32_UPD : VLD1DWB<0b1000, "32">;
208 def VLD1d64_UPD : VLD1DWB<0b1100, "64">;
210 def VLD1q8_UPD : VLD1QWB<0b0000, "8">;
211 def VLD1q16_UPD : VLD1QWB<0b0100, "16">;
212 def VLD1q32_UPD : VLD1QWB<0b1000, "32">;
213 def VLD1q64_UPD : VLD1QWB<0b1100, "64">;
215 // ...with 3 registers (some of these are only for the disassembler):
216 class VLD1D3<bits<4> op7_4, string Dt>
217 : NLdSt<0,0b10,0b0110,op7_4, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3),
218 (ins addrmode6:$addr), IIC_VLD1, "vld1", Dt,
219 "\\{$dst1, $dst2, $dst3\\}, $addr", "", []>;
220 class VLD1D3WB<bits<4> op7_4, string Dt>
221 : NLdSt<0,0b10,0b0110,op7_4, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, GPR:$wb),
222 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD1, "vld1", Dt,
223 "\\{$dst1, $dst2, $dst3\\}, $addr$offset", "$addr.addr = $wb", []>;
225 def VLD1d8T : VLD1D3<0b0000, "8">;
226 def VLD1d16T : VLD1D3<0b0100, "16">;
227 def VLD1d32T : VLD1D3<0b1000, "32">;
228 def VLD1d64T : VLD1D3<0b1100, "64">;
230 def VLD1d8T_UPD : VLD1D3WB<0b0000, "8">;
231 def VLD1d16T_UPD : VLD1D3WB<0b0100, "16">;
232 def VLD1d32T_UPD : VLD1D3WB<0b1000, "32">;
233 def VLD1d64T_UPD : VLD1D3WB<0b1100, "64">;
235 // ...with 4 registers (some of these are only for the disassembler):
236 class VLD1D4<bits<4> op7_4, string Dt>
237 : NLdSt<0,0b10,0b0010,op7_4,(outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
238 (ins addrmode6:$addr), IIC_VLD1, "vld1", Dt,
239 "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr", "", []>;
240 class VLD1D4WB<bits<4> op7_4, string Dt>
241 : NLdSt<0,0b10,0b0010,op7_4,
242 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
243 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD1, "vld1", Dt,
244 "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr$offset", "$addr.addr = $wb",
247 def VLD1d8Q : VLD1D4<0b0000, "8">;
248 def VLD1d16Q : VLD1D4<0b0100, "16">;
249 def VLD1d32Q : VLD1D4<0b1000, "32">;
250 def VLD1d64Q : VLD1D4<0b1100, "64">;
252 def VLD1d8Q_UPD : VLD1D4WB<0b0000, "8">;
253 def VLD1d16Q_UPD : VLD1D4WB<0b0100, "16">;
254 def VLD1d32Q_UPD : VLD1D4WB<0b1000, "32">;
255 def VLD1d64Q_UPD : VLD1D4WB<0b1100, "64">;
257 // VLD2 : Vector Load (multiple 2-element structures)
258 class VLD2D<bits<4> op11_8, bits<4> op7_4, string Dt>
259 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2),
260 (ins addrmode6:$addr), IIC_VLD2,
261 "vld2", Dt, "\\{$dst1, $dst2\\}, $addr", "", []>;
262 class VLD2Q<bits<4> op7_4, string Dt>
263 : NLdSt<0, 0b10, 0b0011, op7_4,
264 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
265 (ins addrmode6:$addr), IIC_VLD2,
266 "vld2", Dt, "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr", "", []>;
268 def VLD2d8 : VLD2D<0b1000, 0b0000, "8">;
269 def VLD2d16 : VLD2D<0b1000, 0b0100, "16">;
270 def VLD2d32 : VLD2D<0b1000, 0b1000, "32">;
272 def VLD2q8 : VLD2Q<0b0000, "8">;
273 def VLD2q16 : VLD2Q<0b0100, "16">;
274 def VLD2q32 : VLD2Q<0b1000, "32">;
276 // ...with address register writeback:
277 class VLD2DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
278 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2, GPR:$wb),
279 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD2,
280 "vld2", Dt, "\\{$dst1, $dst2\\}, $addr$offset",
281 "$addr.addr = $wb", []>;
282 class VLD2QWB<bits<4> op7_4, string Dt>
283 : NLdSt<0, 0b10, 0b0011, op7_4,
284 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
285 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD2,
286 "vld2", Dt, "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr$offset",
287 "$addr.addr = $wb", []>;
289 def VLD2d8_UPD : VLD2DWB<0b1000, 0b0000, "8">;
290 def VLD2d16_UPD : VLD2DWB<0b1000, 0b0100, "16">;
291 def VLD2d32_UPD : VLD2DWB<0b1000, 0b1000, "32">;
293 def VLD2q8_UPD : VLD2QWB<0b0000, "8">;
294 def VLD2q16_UPD : VLD2QWB<0b0100, "16">;
295 def VLD2q32_UPD : VLD2QWB<0b1000, "32">;
297 // ...with double-spaced registers (for disassembly only):
298 def VLD2b8 : VLD2D<0b1001, 0b0000, "8">;
299 def VLD2b16 : VLD2D<0b1001, 0b0100, "16">;
300 def VLD2b32 : VLD2D<0b1001, 0b1000, "32">;
301 def VLD2b8_UPD : VLD2DWB<0b1001, 0b0000, "8">;
302 def VLD2b16_UPD : VLD2DWB<0b1001, 0b0100, "16">;
303 def VLD2b32_UPD : VLD2DWB<0b1001, 0b1000, "32">;
305 // VLD3 : Vector Load (multiple 3-element structures)
306 class VLD3D<bits<4> op11_8, bits<4> op7_4, string Dt>
307 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3),
308 (ins addrmode6:$addr), IIC_VLD3,
309 "vld3", Dt, "\\{$dst1, $dst2, $dst3\\}, $addr", "", []>;
311 def VLD3d8 : VLD3D<0b0100, 0b0000, "8">;
312 def VLD3d16 : VLD3D<0b0100, 0b0100, "16">;
313 def VLD3d32 : VLD3D<0b0100, 0b1000, "32">;
315 // ...with address register writeback:
316 class VLD3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
317 : NLdSt<0, 0b10, op11_8, op7_4,
318 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, GPR:$wb),
319 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD3,
320 "vld3", Dt, "\\{$dst1, $dst2, $dst3\\}, $addr$offset",
321 "$addr.addr = $wb", []>;
323 def VLD3d8_UPD : VLD3DWB<0b0100, 0b0000, "8">;
324 def VLD3d16_UPD : VLD3DWB<0b0100, 0b0100, "16">;
325 def VLD3d32_UPD : VLD3DWB<0b0100, 0b1000, "32">;
327 // ...with double-spaced registers (non-updating versions for disassembly only):
328 def VLD3q8 : VLD3D<0b0101, 0b0000, "8">;
329 def VLD3q16 : VLD3D<0b0101, 0b0100, "16">;
330 def VLD3q32 : VLD3D<0b0101, 0b1000, "32">;
331 def VLD3q8_UPD : VLD3DWB<0b0101, 0b0000, "8">;
332 def VLD3q16_UPD : VLD3DWB<0b0101, 0b0100, "16">;
333 def VLD3q32_UPD : VLD3DWB<0b0101, 0b1000, "32">;
335 // ...alternate versions to be allocated odd register numbers:
336 def VLD3q8odd_UPD : VLD3DWB<0b0101, 0b0000, "8">;
337 def VLD3q16odd_UPD : VLD3DWB<0b0101, 0b0100, "16">;
338 def VLD3q32odd_UPD : VLD3DWB<0b0101, 0b1000, "32">;
340 // VLD4 : Vector Load (multiple 4-element structures)
341 class VLD4D<bits<4> op11_8, bits<4> op7_4, string Dt>
342 : NLdSt<0, 0b10, op11_8, op7_4,
343 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
344 (ins addrmode6:$addr), IIC_VLD4,
345 "vld4", Dt, "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr", "", []>;
347 def VLD4d8 : VLD4D<0b0000, 0b0000, "8">;
348 def VLD4d16 : VLD4D<0b0000, 0b0100, "16">;
349 def VLD4d32 : VLD4D<0b0000, 0b1000, "32">;
351 // ...with address register writeback:
352 class VLD4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
353 : NLdSt<0, 0b10, op11_8, op7_4,
354 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
355 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD4,
356 "vld4", Dt, "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr$offset",
357 "$addr.addr = $wb", []>;
359 def VLD4d8_UPD : VLD4DWB<0b0000, 0b0000, "8">;
360 def VLD4d16_UPD : VLD4DWB<0b0000, 0b0100, "16">;
361 def VLD4d32_UPD : VLD4DWB<0b0000, 0b1000, "32">;
363 // ...with double-spaced registers (non-updating versions for disassembly only):
364 def VLD4q8 : VLD4D<0b0001, 0b0000, "8">;
365 def VLD4q16 : VLD4D<0b0001, 0b0100, "16">;
366 def VLD4q32 : VLD4D<0b0001, 0b1000, "32">;
367 def VLD4q8_UPD : VLD4DWB<0b0001, 0b0000, "8">;
368 def VLD4q16_UPD : VLD4DWB<0b0001, 0b0100, "16">;
369 def VLD4q32_UPD : VLD4DWB<0b0001, 0b1000, "32">;
371 // ...alternate versions to be allocated odd register numbers:
372 def VLD4q8odd_UPD : VLD4DWB<0b0001, 0b0000, "8">;
373 def VLD4q16odd_UPD : VLD4DWB<0b0001, 0b0100, "16">;
374 def VLD4q32odd_UPD : VLD4DWB<0b0001, 0b1000, "32">;
376 // VLD1LN : Vector Load (single element to one lane)
377 // FIXME: Not yet implemented.
379 // VLD2LN : Vector Load (single 2-element structure to one lane)
380 class VLD2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
381 : NLdSt<1, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2),
382 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, nohash_imm:$lane),
383 IIC_VLD2, "vld2", Dt, "\\{$dst1[$lane], $dst2[$lane]\\}, $addr",
384 "$src1 = $dst1, $src2 = $dst2", []>;
386 def VLD2LNd8 : VLD2LN<0b0001, {?,?,?,?}, "8">;
387 def VLD2LNd16 : VLD2LN<0b0101, {?,?,0,?}, "16">;
388 def VLD2LNd32 : VLD2LN<0b1001, {?,0,?,?}, "32">;
390 // ...with double-spaced registers:
391 def VLD2LNq16 : VLD2LN<0b0101, {?,?,1,?}, "16">;
392 def VLD2LNq32 : VLD2LN<0b1001, {?,1,?,?}, "32">;
394 // ...alternate versions to be allocated odd register numbers:
395 def VLD2LNq16odd : VLD2LN<0b0101, {?,?,1,?}, "16">;
396 def VLD2LNq32odd : VLD2LN<0b1001, {?,1,?,?}, "32">;
398 // ...with address register writeback:
399 class VLD2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
400 : NLdSt<1, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2, GPR:$wb),
401 (ins addrmode6:$addr, am6offset:$offset,
402 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VLD2, "vld2", Dt,
403 "\\{$dst1[$lane], $dst2[$lane]\\}, $addr$offset",
404 "$src1 = $dst1, $src2 = $dst2, $addr.addr = $wb", []>;
406 def VLD2LNd8_UPD : VLD2LNWB<0b0001, {?,?,?,?}, "8">;
407 def VLD2LNd16_UPD : VLD2LNWB<0b0101, {?,?,0,?}, "16">;
408 def VLD2LNd32_UPD : VLD2LNWB<0b1001, {?,0,?,?}, "32">;
410 def VLD2LNq16_UPD : VLD2LNWB<0b0101, {?,?,1,?}, "16">;
411 def VLD2LNq32_UPD : VLD2LNWB<0b1001, {?,1,?,?}, "32">;
413 // VLD3LN : Vector Load (single 3-element structure to one lane)
414 class VLD3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
415 : NLdSt<1, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3),
416 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3,
417 nohash_imm:$lane), IIC_VLD3, "vld3", Dt,
418 "\\{$dst1[$lane], $dst2[$lane], $dst3[$lane]\\}, $addr",
419 "$src1 = $dst1, $src2 = $dst2, $src3 = $dst3", []>;
421 def VLD3LNd8 : VLD3LN<0b0010, {?,?,?,0}, "8">;
422 def VLD3LNd16 : VLD3LN<0b0110, {?,?,0,0}, "16">;
423 def VLD3LNd32 : VLD3LN<0b1010, {?,0,0,0}, "32">;
425 // ...with double-spaced registers:
426 def VLD3LNq16 : VLD3LN<0b0110, {?,?,1,0}, "16">;
427 def VLD3LNq32 : VLD3LN<0b1010, {?,1,0,0}, "32">;
429 // ...alternate versions to be allocated odd register numbers:
430 def VLD3LNq16odd : VLD3LN<0b0110, {?,?,1,0}, "16">;
431 def VLD3LNq32odd : VLD3LN<0b1010, {?,1,0,0}, "32">;
433 // ...with address register writeback:
434 class VLD3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
435 : NLdSt<1, 0b10, op11_8, op7_4,
436 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, GPR:$wb),
437 (ins addrmode6:$addr, am6offset:$offset,
438 DPR:$src1, DPR:$src2, DPR:$src3, nohash_imm:$lane),
439 IIC_VLD3, "vld3", Dt,
440 "\\{$dst1[$lane], $dst2[$lane], $dst3[$lane]\\}, $addr$offset",
441 "$src1 = $dst1, $src2 = $dst2, $src3 = $dst3, $addr.addr = $wb",
444 def VLD3LNd8_UPD : VLD3LNWB<0b0010, {?,?,?,0}, "8">;
445 def VLD3LNd16_UPD : VLD3LNWB<0b0110, {?,?,0,0}, "16">;
446 def VLD3LNd32_UPD : VLD3LNWB<0b1010, {?,0,0,0}, "32">;
448 def VLD3LNq16_UPD : VLD3LNWB<0b0110, {?,?,1,0}, "16">;
449 def VLD3LNq32_UPD : VLD3LNWB<0b1010, {?,1,0,0}, "32">;
451 // VLD4LN : Vector Load (single 4-element structure to one lane)
452 class VLD4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
453 : NLdSt<1, 0b10, op11_8, op7_4,
454 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
455 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
456 nohash_imm:$lane), IIC_VLD4, "vld4", Dt,
457 "\\{$dst1[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $addr",
458 "$src1 = $dst1, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4", []>;
460 def VLD4LNd8 : VLD4LN<0b0011, {?,?,?,?}, "8">;
461 def VLD4LNd16 : VLD4LN<0b0111, {?,?,0,?}, "16">;
462 def VLD4LNd32 : VLD4LN<0b1011, {?,0,?,?}, "32">;
464 // ...with double-spaced registers:
465 def VLD4LNq16 : VLD4LN<0b0111, {?,?,1,?}, "16">;
466 def VLD4LNq32 : VLD4LN<0b1011, {?,1,?,?}, "32">;
468 // ...alternate versions to be allocated odd register numbers:
469 def VLD4LNq16odd : VLD4LN<0b0111, {?,?,1,?}, "16">;
470 def VLD4LNq32odd : VLD4LN<0b1011, {?,1,?,?}, "32">;
472 // ...with address register writeback:
473 class VLD4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
474 : NLdSt<1, 0b10, op11_8, op7_4,
475 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
476 (ins addrmode6:$addr, am6offset:$offset,
477 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
478 IIC_VLD4, "vld4", Dt,
479 "\\{$dst1[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $addr$offset",
480 "$src1 = $dst1, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4, $addr.addr = $wb",
483 def VLD4LNd8_UPD : VLD4LNWB<0b0011, {?,?,?,?}, "8">;
484 def VLD4LNd16_UPD : VLD4LNWB<0b0111, {?,?,0,?}, "16">;
485 def VLD4LNd32_UPD : VLD4LNWB<0b1011, {?,0,?,?}, "32">;
487 def VLD4LNq16_UPD : VLD4LNWB<0b0111, {?,?,1,?}, "16">;
488 def VLD4LNq32_UPD : VLD4LNWB<0b1011, {?,1,?,?}, "32">;
490 // VLD1DUP : Vector Load (single element to all lanes)
491 // VLD2DUP : Vector Load (single 2-element structure to all lanes)
492 // VLD3DUP : Vector Load (single 3-element structure to all lanes)
493 // VLD4DUP : Vector Load (single 4-element structure to all lanes)
494 // FIXME: Not yet implemented.
495 } // mayLoad = 1, hasExtraDefRegAllocReq = 1
497 let mayStore = 1, hasExtraSrcRegAllocReq = 1 in {
499 // VST1 : Vector Store (multiple single elements)
500 class VST1D<bits<4> op7_4, string Dt>
501 : NLdSt<0,0b00,0b0111,op7_4, (outs), (ins addrmode6:$addr, DPR:$src), IIC_VST,
502 "vst1", Dt, "\\{$src\\}, $addr", "", []>;
503 class VST1Q<bits<4> op7_4, string Dt>
504 : NLdSt<0,0b00,0b1010,op7_4, (outs),
505 (ins addrmode6:$addr, DPR:$src1, DPR:$src2), IIC_VST,
506 "vst1", Dt, "\\{$src1, $src2\\}, $addr", "", []>;
508 def VST1d8 : VST1D<0b0000, "8">;
509 def VST1d16 : VST1D<0b0100, "16">;
510 def VST1d32 : VST1D<0b1000, "32">;
511 def VST1d64 : VST1D<0b1100, "64">;
513 def VST1q8 : VST1Q<0b0000, "8">;
514 def VST1q16 : VST1Q<0b0100, "16">;
515 def VST1q32 : VST1Q<0b1000, "32">;
516 def VST1q64 : VST1Q<0b1100, "64">;
518 // ...with address register writeback:
519 class VST1DWB<bits<4> op7_4, string Dt>
520 : NLdSt<0, 0b00, 0b0111, op7_4, (outs GPR:$wb),
521 (ins addrmode6:$addr, am6offset:$offset, DPR:$src), IIC_VST,
522 "vst1", Dt, "\\{$src\\}, $addr$offset", "$addr.addr = $wb", []>;
523 class VST1QWB<bits<4> op7_4, string Dt>
524 : NLdSt<0, 0b00, 0b1010, op7_4, (outs GPR:$wb),
525 (ins addrmode6:$addr, am6offset:$offset, QPR:$src), IIC_VST,
526 "vst1", Dt, "${src:dregpair}, $addr$offset", "$addr.addr = $wb", []>;
528 def VST1d8_UPD : VST1DWB<0b0000, "8">;
529 def VST1d16_UPD : VST1DWB<0b0100, "16">;
530 def VST1d32_UPD : VST1DWB<0b1000, "32">;
531 def VST1d64_UPD : VST1DWB<0b1100, "64">;
533 def VST1q8_UPD : VST1QWB<0b0000, "8">;
534 def VST1q16_UPD : VST1QWB<0b0100, "16">;
535 def VST1q32_UPD : VST1QWB<0b1000, "32">;
536 def VST1q64_UPD : VST1QWB<0b1100, "64">;
538 // ...with 3 registers (some of these are only for the disassembler):
539 class VST1D3<bits<4> op7_4, string Dt>
540 : NLdSt<0, 0b00, 0b0110, op7_4, (outs),
541 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3),
542 IIC_VST, "vst1", Dt, "\\{$src1, $src2, $src3\\}, $addr", "", []>;
543 class VST1D3WB<bits<4> op7_4, string Dt>
544 : NLdSt<0, 0b00, 0b0110, op7_4, (outs GPR:$wb),
545 (ins addrmode6:$addr, am6offset:$offset,
546 DPR:$src1, DPR:$src2, DPR:$src3),
547 IIC_VST, "vst1", Dt, "\\{$src1, $src2, $src3\\}, $addr$offset",
548 "$addr.addr = $wb", []>;
550 def VST1d8T : VST1D3<0b0000, "8">;
551 def VST1d16T : VST1D3<0b0100, "16">;
552 def VST1d32T : VST1D3<0b1000, "32">;
553 def VST1d64T : VST1D3<0b1100, "64">;
555 def VST1d8T_UPD : VST1D3WB<0b0000, "8">;
556 def VST1d16T_UPD : VST1D3WB<0b0100, "16">;
557 def VST1d32T_UPD : VST1D3WB<0b1000, "32">;
558 def VST1d64T_UPD : VST1D3WB<0b1100, "64">;
560 // ...with 4 registers (some of these are only for the disassembler):
561 class VST1D4<bits<4> op7_4, string Dt>
562 : NLdSt<0, 0b00, 0b0010, op7_4, (outs),
563 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
564 IIC_VST, "vst1", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr", "",
566 class VST1D4WB<bits<4> op7_4, string Dt>
567 : NLdSt<0, 0b00, 0b0010, op7_4, (outs GPR:$wb),
568 (ins addrmode6:$addr, am6offset:$offset,
569 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
570 IIC_VST, "vst1", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr$offset",
571 "$addr.addr = $wb", []>;
573 def VST1d8Q : VST1D4<0b0000, "8">;
574 def VST1d16Q : VST1D4<0b0100, "16">;
575 def VST1d32Q : VST1D4<0b1000, "32">;
576 def VST1d64Q : VST1D4<0b1100, "64">;
578 def VST1d8Q_UPD : VST1D4WB<0b0000, "8">;
579 def VST1d16Q_UPD : VST1D4WB<0b0100, "16">;
580 def VST1d32Q_UPD : VST1D4WB<0b1000, "32">;
581 def VST1d64Q_UPD : VST1D4WB<0b1100, "64">;
583 // VST2 : Vector Store (multiple 2-element structures)
584 class VST2D<bits<4> op11_8, bits<4> op7_4, string Dt>
585 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
586 (ins addrmode6:$addr, DPR:$src1, DPR:$src2),
587 IIC_VST, "vst2", Dt, "\\{$src1, $src2\\}, $addr", "", []>;
588 class VST2Q<bits<4> op7_4, string Dt>
589 : NLdSt<0, 0b00, 0b0011, op7_4, (outs),
590 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
591 IIC_VST, "vst2", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr",
594 def VST2d8 : VST2D<0b1000, 0b0000, "8">;
595 def VST2d16 : VST2D<0b1000, 0b0100, "16">;
596 def VST2d32 : VST2D<0b1000, 0b1000, "32">;
598 def VST2q8 : VST2Q<0b0000, "8">;
599 def VST2q16 : VST2Q<0b0100, "16">;
600 def VST2q32 : VST2Q<0b1000, "32">;
602 // ...with address register writeback:
603 class VST2DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
604 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
605 (ins addrmode6:$addr, am6offset:$offset, DPR:$src1, DPR:$src2),
606 IIC_VST, "vst2", Dt, "\\{$src1, $src2\\}, $addr$offset",
607 "$addr.addr = $wb", []>;
608 class VST2QWB<bits<4> op7_4, string Dt>
609 : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
610 (ins addrmode6:$addr, am6offset:$offset,
611 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
612 IIC_VST, "vst2", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr$offset",
613 "$addr.addr = $wb", []>;
615 def VST2d8_UPD : VST2DWB<0b1000, 0b0000, "8">;
616 def VST2d16_UPD : VST2DWB<0b1000, 0b0100, "16">;
617 def VST2d32_UPD : VST2DWB<0b1000, 0b1000, "32">;
619 def VST2q8_UPD : VST2QWB<0b0000, "8">;
620 def VST2q16_UPD : VST2QWB<0b0100, "16">;
621 def VST2q32_UPD : VST2QWB<0b1000, "32">;
623 // ...with double-spaced registers (for disassembly only):
624 def VST2b8 : VST2D<0b1001, 0b0000, "8">;
625 def VST2b16 : VST2D<0b1001, 0b0100, "16">;
626 def VST2b32 : VST2D<0b1001, 0b1000, "32">;
627 def VST2b8_UPD : VST2DWB<0b1001, 0b0000, "8">;
628 def VST2b16_UPD : VST2DWB<0b1001, 0b0100, "16">;
629 def VST2b32_UPD : VST2DWB<0b1001, 0b1000, "32">;
631 // VST3 : Vector Store (multiple 3-element structures)
632 class VST3D<bits<4> op11_8, bits<4> op7_4, string Dt>
633 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
634 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3), IIC_VST,
635 "vst3", Dt, "\\{$src1, $src2, $src3\\}, $addr", "", []>;
637 def VST3d8 : VST3D<0b0100, 0b0000, "8">;
638 def VST3d16 : VST3D<0b0100, 0b0100, "16">;
639 def VST3d32 : VST3D<0b0100, 0b1000, "32">;
641 // ...with address register writeback:
642 class VST3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
643 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
644 (ins addrmode6:$addr, am6offset:$offset,
645 DPR:$src1, DPR:$src2, DPR:$src3), IIC_VST,
646 "vst3", Dt, "\\{$src1, $src2, $src3\\}, $addr$offset",
647 "$addr.addr = $wb", []>;
649 def VST3d8_UPD : VST3DWB<0b0100, 0b0000, "8">;
650 def VST3d16_UPD : VST3DWB<0b0100, 0b0100, "16">;
651 def VST3d32_UPD : VST3DWB<0b0100, 0b1000, "32">;
653 // ...with double-spaced registers (non-updating versions for disassembly only):
654 def VST3q8 : VST3D<0b0101, 0b0000, "8">;
655 def VST3q16 : VST3D<0b0101, 0b0100, "16">;
656 def VST3q32 : VST3D<0b0101, 0b1000, "32">;
657 def VST3q8_UPD : VST3DWB<0b0101, 0b0000, "8">;
658 def VST3q16_UPD : VST3DWB<0b0101, 0b0100, "16">;
659 def VST3q32_UPD : VST3DWB<0b0101, 0b1000, "32">;
661 // ...alternate versions to be allocated odd register numbers:
662 def VST3q8odd_UPD : VST3DWB<0b0101, 0b0000, "8">;
663 def VST3q16odd_UPD : VST3DWB<0b0101, 0b0100, "16">;
664 def VST3q32odd_UPD : VST3DWB<0b0101, 0b1000, "32">;
666 // VST4 : Vector Store (multiple 4-element structures)
667 class VST4D<bits<4> op11_8, bits<4> op7_4, string Dt>
668 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
669 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
670 IIC_VST, "vst4", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr",
673 def VST4d8 : VST4D<0b0000, 0b0000, "8">;
674 def VST4d16 : VST4D<0b0000, 0b0100, "16">;
675 def VST4d32 : VST4D<0b0000, 0b1000, "32">;
677 // ...with address register writeback:
678 class VST4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
679 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
680 (ins addrmode6:$addr, am6offset:$offset,
681 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST,
682 "vst4", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr$offset",
683 "$addr.addr = $wb", []>;
685 def VST4d8_UPD : VST4DWB<0b0000, 0b0000, "8">;
686 def VST4d16_UPD : VST4DWB<0b0000, 0b0100, "16">;
687 def VST4d32_UPD : VST4DWB<0b0000, 0b1000, "32">;
689 // ...with double-spaced registers (non-updating versions for disassembly only):
690 def VST4q8 : VST4D<0b0001, 0b0000, "8">;
691 def VST4q16 : VST4D<0b0001, 0b0100, "16">;
692 def VST4q32 : VST4D<0b0001, 0b1000, "32">;
693 def VST4q8_UPD : VST4DWB<0b0001, 0b0000, "8">;
694 def VST4q16_UPD : VST4DWB<0b0001, 0b0100, "16">;
695 def VST4q32_UPD : VST4DWB<0b0001, 0b1000, "32">;
697 // ...alternate versions to be allocated odd register numbers:
698 def VST4q8odd_UPD : VST4DWB<0b0001, 0b0000, "8">;
699 def VST4q16odd_UPD : VST4DWB<0b0001, 0b0100, "16">;
700 def VST4q32odd_UPD : VST4DWB<0b0001, 0b1000, "32">;
702 // VST1LN : Vector Store (single element from one lane)
703 // FIXME: Not yet implemented.
705 // VST2LN : Vector Store (single 2-element structure from one lane)
706 class VST2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
707 : NLdSt<1, 0b00, op11_8, op7_4, (outs),
708 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, nohash_imm:$lane),
709 IIC_VST, "vst2", Dt, "\\{$src1[$lane], $src2[$lane]\\}, $addr",
712 def VST2LNd8 : VST2LN<0b0001, {?,?,?,?}, "8">;
713 def VST2LNd16 : VST2LN<0b0101, {?,?,0,?}, "16">;
714 def VST2LNd32 : VST2LN<0b1001, {?,0,?,?}, "32">;
716 // ...with double-spaced registers:
717 def VST2LNq16 : VST2LN<0b0101, {?,?,1,?}, "16">;
718 def VST2LNq32 : VST2LN<0b1001, {?,1,?,?}, "32">;
720 // ...alternate versions to be allocated odd register numbers:
721 def VST2LNq16odd : VST2LN<0b0101, {?,?,1,?}, "16">;
722 def VST2LNq32odd : VST2LN<0b1001, {?,1,?,?}, "32">;
724 // ...with address register writeback:
725 class VST2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
726 : NLdSt<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
727 (ins addrmode6:$addr, am6offset:$offset,
728 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VST, "vst2", Dt,
729 "\\{$src1[$lane], $src2[$lane]\\}, $addr$offset",
730 "$addr.addr = $wb", []>;
732 def VST2LNd8_UPD : VST2LNWB<0b0001, {?,?,?,?}, "8">;
733 def VST2LNd16_UPD : VST2LNWB<0b0101, {?,?,0,?}, "16">;
734 def VST2LNd32_UPD : VST2LNWB<0b1001, {?,0,?,?}, "32">;
736 def VST2LNq16_UPD : VST2LNWB<0b0101, {?,?,1,?}, "16">;
737 def VST2LNq32_UPD : VST2LNWB<0b1001, {?,1,?,?}, "32">;
739 // VST3LN : Vector Store (single 3-element structure from one lane)
740 class VST3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
741 : NLdSt<1, 0b00, op11_8, op7_4, (outs),
742 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3,
743 nohash_imm:$lane), IIC_VST, "vst3", Dt,
744 "\\{$src1[$lane], $src2[$lane], $src3[$lane]\\}, $addr", "", []>;
746 def VST3LNd8 : VST3LN<0b0010, {?,?,?,0}, "8">;
747 def VST3LNd16 : VST3LN<0b0110, {?,?,0,0}, "16">;
748 def VST3LNd32 : VST3LN<0b1010, {?,0,0,0}, "32">;
750 // ...with double-spaced registers:
751 def VST3LNq16 : VST3LN<0b0110, {?,?,1,0}, "16">;
752 def VST3LNq32 : VST3LN<0b1010, {?,1,0,0}, "32">;
754 // ...alternate versions to be allocated odd register numbers:
755 def VST3LNq16odd : VST3LN<0b0110, {?,?,1,0}, "16">;
756 def VST3LNq32odd : VST3LN<0b1010, {?,1,0,0}, "32">;
758 // ...with address register writeback:
759 class VST3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
760 : NLdSt<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
761 (ins addrmode6:$addr, am6offset:$offset,
762 DPR:$src1, DPR:$src2, DPR:$src3, nohash_imm:$lane),
764 "\\{$src1[$lane], $src2[$lane], $src3[$lane]\\}, $addr$offset",
765 "$addr.addr = $wb", []>;
767 def VST3LNd8_UPD : VST3LNWB<0b0010, {?,?,?,0}, "8">;
768 def VST3LNd16_UPD : VST3LNWB<0b0110, {?,?,0,0}, "16">;
769 def VST3LNd32_UPD : VST3LNWB<0b1010, {?,0,0,0}, "32">;
771 def VST3LNq16_UPD : VST3LNWB<0b0110, {?,?,1,0}, "16">;
772 def VST3LNq32_UPD : VST3LNWB<0b1010, {?,1,0,0}, "32">;
774 // VST4LN : Vector Store (single 4-element structure from one lane)
775 class VST4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
776 : NLdSt<1, 0b00, op11_8, op7_4, (outs),
777 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
778 nohash_imm:$lane), IIC_VST, "vst4", Dt,
779 "\\{$src1[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $addr",
782 def VST4LNd8 : VST4LN<0b0011, {?,?,?,?}, "8">;
783 def VST4LNd16 : VST4LN<0b0111, {?,?,0,?}, "16">;
784 def VST4LNd32 : VST4LN<0b1011, {?,0,?,?}, "32">;
786 // ...with double-spaced registers:
787 def VST4LNq16 : VST4LN<0b0111, {?,?,1,?}, "16">;
788 def VST4LNq32 : VST4LN<0b1011, {?,1,?,?}, "32">;
790 // ...alternate versions to be allocated odd register numbers:
791 def VST4LNq16odd : VST4LN<0b0111, {?,?,1,?}, "16">;
792 def VST4LNq32odd : VST4LN<0b1011, {?,1,?,?}, "32">;
794 // ...with address register writeback:
795 class VST4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
796 : NLdSt<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
797 (ins addrmode6:$addr, am6offset:$offset,
798 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
800 "\\{$src1[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $addr$offset",
801 "$addr.addr = $wb", []>;
803 def VST4LNd8_UPD : VST4LNWB<0b0011, {?,?,?,?}, "8">;
804 def VST4LNd16_UPD : VST4LNWB<0b0111, {?,?,0,?}, "16">;
805 def VST4LNd32_UPD : VST4LNWB<0b1011, {?,0,?,?}, "32">;
807 def VST4LNq16_UPD : VST4LNWB<0b0111, {?,?,1,?}, "16">;
808 def VST4LNq32_UPD : VST4LNWB<0b1011, {?,1,?,?}, "32">;
810 } // mayStore = 1, hasExtraSrcRegAllocReq = 1
813 //===----------------------------------------------------------------------===//
814 // NEON pattern fragments
815 //===----------------------------------------------------------------------===//
817 // Extract D sub-registers of Q registers.
818 // (arm_dsubreg_0 is 5; arm_dsubreg_1 is 6)
819 def DSubReg_i8_reg : SDNodeXForm<imm, [{
820 return CurDAG->getTargetConstant(5 + N->getZExtValue() / 8, MVT::i32);
822 def DSubReg_i16_reg : SDNodeXForm<imm, [{
823 return CurDAG->getTargetConstant(5 + N->getZExtValue() / 4, MVT::i32);
825 def DSubReg_i32_reg : SDNodeXForm<imm, [{
826 return CurDAG->getTargetConstant(5 + N->getZExtValue() / 2, MVT::i32);
828 def DSubReg_f64_reg : SDNodeXForm<imm, [{
829 return CurDAG->getTargetConstant(5 + N->getZExtValue(), MVT::i32);
831 def DSubReg_f64_other_reg : SDNodeXForm<imm, [{
832 return CurDAG->getTargetConstant(5 + (1 - N->getZExtValue()), MVT::i32);
835 // Extract S sub-registers of Q/D registers.
836 // (arm_ssubreg_0 is 1; arm_ssubreg_1 is 2; etc.)
837 def SSubReg_f32_reg : SDNodeXForm<imm, [{
838 return CurDAG->getTargetConstant(1 + N->getZExtValue(), MVT::i32);
841 // Translate lane numbers from Q registers to D subregs.
842 def SubReg_i8_lane : SDNodeXForm<imm, [{
843 return CurDAG->getTargetConstant(N->getZExtValue() & 7, MVT::i32);
845 def SubReg_i16_lane : SDNodeXForm<imm, [{
846 return CurDAG->getTargetConstant(N->getZExtValue() & 3, MVT::i32);
848 def SubReg_i32_lane : SDNodeXForm<imm, [{
849 return CurDAG->getTargetConstant(N->getZExtValue() & 1, MVT::i32);
852 //===----------------------------------------------------------------------===//
853 // Instruction Classes
854 //===----------------------------------------------------------------------===//
856 // Basic 2-register operations: single-, double- and quad-register.
857 // This is used for NVCVTFrm form.
858 class N2VS<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
859 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
860 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
861 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
862 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src),
863 IIC_VUNAD, OpcodeStr, Dt, "$dst, $src", "", []>;
864 class N2VD<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
865 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
866 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
867 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
868 (ins DPR:$src), IIC_VUNAD, OpcodeStr, Dt,"$dst, $src", "",
869 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src))))]>;
870 class N2VQ<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
871 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
872 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
873 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
874 (ins QPR:$src), IIC_VUNAQ, OpcodeStr, Dt,"$dst, $src", "",
875 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src))))]>;
877 // Basic 2-register intrinsics, both double- and quad-register.
878 class N2VDInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
879 bits<2> op17_16, bits<5> op11_7, bit op4,
880 InstrItinClass itin, string OpcodeStr, string Dt,
881 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
882 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
883 (ins DPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
884 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
885 class N2VQInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
886 bits<2> op17_16, bits<5> op11_7, bit op4,
887 InstrItinClass itin, string OpcodeStr, string Dt,
888 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
889 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
890 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
891 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
893 // Narrow 2-register intrinsics.
894 class N2VNInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
895 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
896 InstrItinClass itin, string OpcodeStr, string Dt,
897 ValueType TyD, ValueType TyQ, Intrinsic IntOp>
898 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$dst),
899 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
900 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src))))]>;
902 // Long 2-register intrinsics (currently only used for VMOVL).
903 class N2VLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
904 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
905 InstrItinClass itin, string OpcodeStr, string Dt,
906 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
907 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$dst),
908 (ins DPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
909 [(set QPR:$dst, (TyQ (IntOp (TyD DPR:$src))))]>;
911 // 2-register shuffles (VTRN/VZIP/VUZP), both double- and quad-register.
912 class N2VDShuffle<bits<2> op19_18, bits<5> op11_7, string OpcodeStr, string Dt>
913 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 0, 0, (outs DPR:$dst1, DPR:$dst2),
914 (ins DPR:$src1, DPR:$src2), IIC_VPERMD,
915 OpcodeStr, Dt, "$dst1, $dst2",
916 "$src1 = $dst1, $src2 = $dst2", []>;
917 class N2VQShuffle<bits<2> op19_18, bits<5> op11_7,
918 InstrItinClass itin, string OpcodeStr, string Dt>
919 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 1, 0, (outs QPR:$dst1, QPR:$dst2),
920 (ins QPR:$src1, QPR:$src2), itin, OpcodeStr, Dt, "$dst1, $dst2",
921 "$src1 = $dst1, $src2 = $dst2", []>;
923 // Basic 3-register operations: single-, double- and quad-register.
924 class N3VS<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
925 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
926 SDNode OpNode, bit Commutable>
927 : N3V<op24, op23, op21_20, op11_8, 0, op4,
928 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src1, DPR_VFP2:$src2), IIC_VBIND,
929 OpcodeStr, Dt, "$dst, $src1, $src2", "", []> {
930 let isCommutable = Commutable;
933 class N3VD<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
934 InstrItinClass itin, string OpcodeStr, string Dt,
935 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
936 : N3V<op24, op23, op21_20, op11_8, 0, op4,
937 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), itin,
938 OpcodeStr, Dt, "$dst, $src1, $src2", "",
939 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src1), (OpTy DPR:$src2))))]> {
940 let isCommutable = Commutable;
942 // Same as N3VD but no data type.
943 class N3VDX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
944 InstrItinClass itin, string OpcodeStr,
945 ValueType ResTy, ValueType OpTy,
946 SDNode OpNode, bit Commutable>
947 : N3VX<op24, op23, op21_20, op11_8, 0, op4,
948 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), itin,
949 OpcodeStr, "$dst, $src1, $src2", "",
950 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src1), (OpTy DPR:$src2))))]>{
951 let isCommutable = Commutable;
953 class N3VDSL<bits<2> op21_20, bits<4> op11_8,
954 InstrItinClass itin, string OpcodeStr, string Dt,
955 ValueType Ty, SDNode ShOp>
956 : N3V<0, 1, op21_20, op11_8, 1, 0,
957 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
958 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
960 (Ty (ShOp (Ty DPR:$src1),
961 (Ty (NEONvduplane (Ty DPR_VFP2:$src2), imm:$lane)))))]>{
962 let isCommutable = 0;
964 class N3VDSL16<bits<2> op21_20, bits<4> op11_8,
965 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
966 : N3V<0, 1, op21_20, op11_8, 1, 0,
967 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
968 IIC_VMULi16D, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
970 (Ty (ShOp (Ty DPR:$src1),
971 (Ty (NEONvduplane (Ty DPR_8:$src2), imm:$lane)))))]> {
972 let isCommutable = 0;
975 class N3VQ<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
976 InstrItinClass itin, string OpcodeStr, string Dt,
977 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
978 : N3V<op24, op23, op21_20, op11_8, 1, op4,
979 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), itin,
980 OpcodeStr, Dt, "$dst, $src1, $src2", "",
981 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src1), (OpTy QPR:$src2))))]> {
982 let isCommutable = Commutable;
984 class N3VQX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
985 InstrItinClass itin, string OpcodeStr,
986 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
987 : N3VX<op24, op23, op21_20, op11_8, 1, op4,
988 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), itin,
989 OpcodeStr, "$dst, $src1, $src2", "",
990 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src1), (OpTy QPR:$src2))))]>{
991 let isCommutable = Commutable;
993 class N3VQSL<bits<2> op21_20, bits<4> op11_8,
994 InstrItinClass itin, string OpcodeStr, string Dt,
995 ValueType ResTy, ValueType OpTy, SDNode ShOp>
996 : N3V<1, 1, op21_20, op11_8, 1, 0,
997 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
998 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
999 [(set (ResTy QPR:$dst),
1000 (ResTy (ShOp (ResTy QPR:$src1),
1001 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1003 let isCommutable = 0;
1005 class N3VQSL16<bits<2> op21_20, bits<4> op11_8, string OpcodeStr, string Dt,
1006 ValueType ResTy, ValueType OpTy, SDNode ShOp>
1007 : N3V<1, 1, op21_20, op11_8, 1, 0,
1008 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1009 IIC_VMULi16Q, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1010 [(set (ResTy QPR:$dst),
1011 (ResTy (ShOp (ResTy QPR:$src1),
1012 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
1014 let isCommutable = 0;
1017 // Basic 3-register intrinsics, both double- and quad-register.
1018 class N3VDInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1019 InstrItinClass itin, string OpcodeStr, string Dt,
1020 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
1021 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1022 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), itin,
1023 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1024 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src1), (OpTy DPR:$src2))))]> {
1025 let isCommutable = Commutable;
1027 class N3VDIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1028 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
1029 : N3V<0, 1, op21_20, op11_8, 1, 0,
1030 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1031 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1032 [(set (Ty DPR:$dst),
1033 (Ty (IntOp (Ty DPR:$src1),
1034 (Ty (NEONvduplane (Ty DPR_VFP2:$src2),
1036 let isCommutable = 0;
1038 class N3VDIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1039 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
1040 : N3V<0, 1, op21_20, op11_8, 1, 0,
1041 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1042 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1043 [(set (Ty DPR:$dst),
1044 (Ty (IntOp (Ty DPR:$src1),
1045 (Ty (NEONvduplane (Ty DPR_8:$src2),
1047 let isCommutable = 0;
1050 class N3VQInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1051 InstrItinClass itin, string OpcodeStr, string Dt,
1052 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
1053 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1054 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), itin,
1055 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1056 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src1), (OpTy QPR:$src2))))]> {
1057 let isCommutable = Commutable;
1059 class N3VQIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1060 string OpcodeStr, string Dt,
1061 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1062 : N3V<1, 1, op21_20, op11_8, 1, 0,
1063 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1064 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1065 [(set (ResTy QPR:$dst),
1066 (ResTy (IntOp (ResTy QPR:$src1),
1067 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1069 let isCommutable = 0;
1071 class N3VQIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1072 string OpcodeStr, string Dt,
1073 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1074 : N3V<1, 1, op21_20, op11_8, 1, 0,
1075 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1076 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1077 [(set (ResTy QPR:$dst),
1078 (ResTy (IntOp (ResTy QPR:$src1),
1079 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
1081 let isCommutable = 0;
1084 // Multiply-Add/Sub operations: single-, double- and quad-register.
1085 class N3VSMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1086 InstrItinClass itin, string OpcodeStr, string Dt,
1087 ValueType Ty, SDNode MulOp, SDNode OpNode>
1088 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1089 (outs DPR_VFP2:$dst),
1090 (ins DPR_VFP2:$src1, DPR_VFP2:$src2, DPR_VFP2:$src3), itin,
1091 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst", []>;
1093 class N3VDMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1094 InstrItinClass itin, string OpcodeStr, string Dt,
1095 ValueType Ty, SDNode MulOp, SDNode OpNode>
1096 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1097 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3), itin,
1098 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1099 [(set DPR:$dst, (Ty (OpNode DPR:$src1,
1100 (Ty (MulOp DPR:$src2, DPR:$src3)))))]>;
1101 class N3VDMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1102 string OpcodeStr, string Dt,
1103 ValueType Ty, SDNode MulOp, SDNode ShOp>
1104 : N3V<0, 1, op21_20, op11_8, 1, 0,
1106 (ins DPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane), itin,
1107 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1108 [(set (Ty DPR:$dst),
1109 (Ty (ShOp (Ty DPR:$src1),
1110 (Ty (MulOp DPR:$src2,
1111 (Ty (NEONvduplane (Ty DPR_VFP2:$src3),
1113 class N3VDMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1114 string OpcodeStr, string Dt,
1115 ValueType Ty, SDNode MulOp, SDNode ShOp>
1116 : N3V<0, 1, op21_20, op11_8, 1, 0,
1118 (ins DPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane), itin,
1119 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1120 [(set (Ty DPR:$dst),
1121 (Ty (ShOp (Ty DPR:$src1),
1122 (Ty (MulOp DPR:$src2,
1123 (Ty (NEONvduplane (Ty DPR_8:$src3),
1126 class N3VQMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1127 InstrItinClass itin, string OpcodeStr, string Dt, ValueType Ty,
1128 SDNode MulOp, SDNode OpNode>
1129 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1130 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3), itin,
1131 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1132 [(set QPR:$dst, (Ty (OpNode QPR:$src1,
1133 (Ty (MulOp QPR:$src2, QPR:$src3)))))]>;
1134 class N3VQMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1135 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1136 SDNode MulOp, SDNode ShOp>
1137 : N3V<1, 1, op21_20, op11_8, 1, 0,
1139 (ins QPR:$src1, QPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane), itin,
1140 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1141 [(set (ResTy QPR:$dst),
1142 (ResTy (ShOp (ResTy QPR:$src1),
1143 (ResTy (MulOp QPR:$src2,
1144 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src3),
1146 class N3VQMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1147 string OpcodeStr, string Dt,
1148 ValueType ResTy, ValueType OpTy,
1149 SDNode MulOp, SDNode ShOp>
1150 : N3V<1, 1, op21_20, op11_8, 1, 0,
1152 (ins QPR:$src1, QPR:$src2, DPR_8:$src3, nohash_imm:$lane), itin,
1153 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1154 [(set (ResTy QPR:$dst),
1155 (ResTy (ShOp (ResTy QPR:$src1),
1156 (ResTy (MulOp QPR:$src2,
1157 (ResTy (NEONvduplane (OpTy DPR_8:$src3),
1160 // Neon 3-argument intrinsics, both double- and quad-register.
1161 // The destination register is also used as the first source operand register.
1162 class N3VDInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1163 InstrItinClass itin, string OpcodeStr, string Dt,
1164 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1165 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1166 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3), itin,
1167 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1168 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src1),
1169 (OpTy DPR:$src2), (OpTy DPR:$src3))))]>;
1170 class N3VQInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1171 InstrItinClass itin, string OpcodeStr, string Dt,
1172 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1173 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1174 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3), itin,
1175 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1176 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src1),
1177 (OpTy QPR:$src2), (OpTy QPR:$src3))))]>;
1179 // Neon Long 3-argument intrinsic. The destination register is
1180 // a quad-register and is also used as the first source operand register.
1181 class N3VLInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1182 InstrItinClass itin, string OpcodeStr, string Dt,
1183 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
1184 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1185 (outs QPR:$dst), (ins QPR:$src1, DPR:$src2, DPR:$src3), itin,
1186 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1188 (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$src2), (TyD DPR:$src3))))]>;
1189 class N3VLInt3SL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1190 string OpcodeStr, string Dt,
1191 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1192 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1194 (ins QPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane), itin,
1195 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1196 [(set (ResTy QPR:$dst),
1197 (ResTy (IntOp (ResTy QPR:$src1),
1199 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src3),
1201 class N3VLInt3SL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1202 InstrItinClass itin, string OpcodeStr, string Dt,
1203 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1204 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1206 (ins QPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane), itin,
1207 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1208 [(set (ResTy QPR:$dst),
1209 (ResTy (IntOp (ResTy QPR:$src1),
1211 (OpTy (NEONvduplane (OpTy DPR_8:$src3),
1214 // Narrowing 3-register intrinsics.
1215 class N3VNInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1216 string OpcodeStr, string Dt, ValueType TyD, ValueType TyQ,
1217 Intrinsic IntOp, bit Commutable>
1218 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1219 (outs DPR:$dst), (ins QPR:$src1, QPR:$src2), IIC_VBINi4D,
1220 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1221 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src1), (TyQ QPR:$src2))))]> {
1222 let isCommutable = Commutable;
1225 // Long 3-register intrinsics.
1226 class N3VLInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1227 InstrItinClass itin, string OpcodeStr, string Dt,
1228 ValueType TyQ, ValueType TyD, Intrinsic IntOp, bit Commutable>
1229 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1230 (outs QPR:$dst), (ins DPR:$src1, DPR:$src2), itin,
1231 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1232 [(set QPR:$dst, (TyQ (IntOp (TyD DPR:$src1), (TyD DPR:$src2))))]> {
1233 let isCommutable = Commutable;
1235 class N3VLIntSL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1236 string OpcodeStr, string Dt,
1237 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1238 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1239 (outs QPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1240 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1241 [(set (ResTy QPR:$dst),
1242 (ResTy (IntOp (OpTy DPR:$src1),
1243 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1245 class N3VLIntSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1246 InstrItinClass itin, string OpcodeStr, string Dt,
1247 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1248 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1249 (outs QPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1250 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1251 [(set (ResTy QPR:$dst),
1252 (ResTy (IntOp (OpTy DPR:$src1),
1253 (OpTy (NEONvduplane (OpTy DPR_8:$src2),
1256 // Wide 3-register intrinsics.
1257 class N3VWInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1258 string OpcodeStr, string Dt, ValueType TyQ, ValueType TyD,
1259 Intrinsic IntOp, bit Commutable>
1260 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1261 (outs QPR:$dst), (ins QPR:$src1, DPR:$src2), IIC_VSUBiD,
1262 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1263 [(set QPR:$dst, (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$src2))))]> {
1264 let isCommutable = Commutable;
1267 // Pairwise long 2-register intrinsics, both double- and quad-register.
1268 class N2VDPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1269 bits<2> op17_16, bits<5> op11_7, bit op4,
1270 string OpcodeStr, string Dt,
1271 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1272 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
1273 (ins DPR:$src), IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
1274 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
1275 class N2VQPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1276 bits<2> op17_16, bits<5> op11_7, bit op4,
1277 string OpcodeStr, string Dt,
1278 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1279 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
1280 (ins QPR:$src), IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
1281 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
1283 // Pairwise long 2-register accumulate intrinsics,
1284 // both double- and quad-register.
1285 // The destination register is also used as the first source operand register.
1286 class N2VDPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1287 bits<2> op17_16, bits<5> op11_7, bit op4,
1288 string OpcodeStr, string Dt,
1289 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1290 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
1291 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), IIC_VPALiD,
1292 OpcodeStr, Dt, "$dst, $src2", "$src1 = $dst",
1293 [(set DPR:$dst, (ResTy (IntOp (ResTy DPR:$src1), (OpTy DPR:$src2))))]>;
1294 class N2VQPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1295 bits<2> op17_16, bits<5> op11_7, bit op4,
1296 string OpcodeStr, string Dt,
1297 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1298 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4,
1299 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), IIC_VPALiQ,
1300 OpcodeStr, Dt, "$dst, $src2", "$src1 = $dst",
1301 [(set QPR:$dst, (ResTy (IntOp (ResTy QPR:$src1), (OpTy QPR:$src2))))]>;
1303 // Shift by immediate,
1304 // both double- and quad-register.
1305 class N2VDSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1306 InstrItinClass itin, string OpcodeStr, string Dt,
1307 ValueType Ty, SDNode OpNode>
1308 : N2VImm<op24, op23, op11_8, op7, 0, op4,
1309 (outs DPR:$dst), (ins DPR:$src, i32imm:$SIMM), itin,
1310 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1311 [(set DPR:$dst, (Ty (OpNode (Ty DPR:$src), (i32 imm:$SIMM))))]>;
1312 class N2VQSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1313 InstrItinClass itin, string OpcodeStr, string Dt,
1314 ValueType Ty, SDNode OpNode>
1315 : N2VImm<op24, op23, op11_8, op7, 1, op4,
1316 (outs QPR:$dst), (ins QPR:$src, i32imm:$SIMM), itin,
1317 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1318 [(set QPR:$dst, (Ty (OpNode (Ty QPR:$src), (i32 imm:$SIMM))))]>;
1320 // Long shift by immediate.
1321 class N2VLSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
1322 string OpcodeStr, string Dt,
1323 ValueType ResTy, ValueType OpTy, SDNode OpNode>
1324 : N2VImm<op24, op23, op11_8, op7, op6, op4,
1325 (outs QPR:$dst), (ins DPR:$src, i32imm:$SIMM), IIC_VSHLiD,
1326 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1327 [(set QPR:$dst, (ResTy (OpNode (OpTy DPR:$src),
1328 (i32 imm:$SIMM))))]>;
1330 // Narrow shift by immediate.
1331 class N2VNSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
1332 InstrItinClass itin, string OpcodeStr, string Dt,
1333 ValueType ResTy, ValueType OpTy, SDNode OpNode>
1334 : N2VImm<op24, op23, op11_8, op7, op6, op4,
1335 (outs DPR:$dst), (ins QPR:$src, i32imm:$SIMM), itin,
1336 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1337 [(set DPR:$dst, (ResTy (OpNode (OpTy QPR:$src),
1338 (i32 imm:$SIMM))))]>;
1340 // Shift right by immediate and accumulate,
1341 // both double- and quad-register.
1342 class N2VDShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1343 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1344 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$dst),
1345 (ins DPR:$src1, DPR:$src2, i32imm:$SIMM), IIC_VPALiD,
1346 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1347 [(set DPR:$dst, (Ty (add DPR:$src1,
1348 (Ty (ShOp DPR:$src2, (i32 imm:$SIMM))))))]>;
1349 class N2VQShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1350 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1351 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$dst),
1352 (ins QPR:$src1, QPR:$src2, i32imm:$SIMM), IIC_VPALiD,
1353 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1354 [(set QPR:$dst, (Ty (add QPR:$src1,
1355 (Ty (ShOp QPR:$src2, (i32 imm:$SIMM))))))]>;
1357 // Shift by immediate and insert,
1358 // both double- and quad-register.
1359 class N2VDShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1360 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1361 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$dst),
1362 (ins DPR:$src1, DPR:$src2, i32imm:$SIMM), IIC_VSHLiD,
1363 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1364 [(set DPR:$dst, (Ty (ShOp DPR:$src1, DPR:$src2, (i32 imm:$SIMM))))]>;
1365 class N2VQShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1366 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1367 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$dst),
1368 (ins QPR:$src1, QPR:$src2, i32imm:$SIMM), IIC_VSHLiQ,
1369 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1370 [(set QPR:$dst, (Ty (ShOp QPR:$src1, QPR:$src2, (i32 imm:$SIMM))))]>;
1372 // Convert, with fractional bits immediate,
1373 // both double- and quad-register.
1374 class N2VCvtD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1375 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1377 : N2VImm<op24, op23, op11_8, op7, 0, op4,
1378 (outs DPR:$dst), (ins DPR:$src, i32imm:$SIMM), IIC_VUNAD,
1379 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1380 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src), (i32 imm:$SIMM))))]>;
1381 class N2VCvtQ<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1382 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1384 : N2VImm<op24, op23, op11_8, op7, 1, op4,
1385 (outs QPR:$dst), (ins QPR:$src, i32imm:$SIMM), IIC_VUNAQ,
1386 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1387 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src), (i32 imm:$SIMM))))]>;
1389 //===----------------------------------------------------------------------===//
1391 //===----------------------------------------------------------------------===//
1393 // Abbreviations used in multiclass suffixes:
1394 // Q = quarter int (8 bit) elements
1395 // H = half int (16 bit) elements
1396 // S = single int (32 bit) elements
1397 // D = double int (64 bit) elements
1399 // Neon 2-register vector operations -- for disassembly only.
1401 // First with only element sizes of 8, 16 and 32 bits:
1402 multiclass N2V_QHS_cmp<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1403 bits<5> op11_7, bit op4, string opc, string Dt,
1405 // 64-bit vector types.
1406 def v8i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 0, op4,
1407 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1408 opc, !strconcat(Dt, "8"), asm, "", []>;
1409 def v4i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 0, op4,
1410 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1411 opc, !strconcat(Dt, "16"), asm, "", []>;
1412 def v2i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
1413 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1414 opc, !strconcat(Dt, "32"), asm, "", []>;
1415 def v2f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
1416 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1417 opc, "f32", asm, "", []> {
1418 let Inst{10} = 1; // overwrite F = 1
1421 // 128-bit vector types.
1422 def v16i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 1, op4,
1423 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1424 opc, !strconcat(Dt, "8"), asm, "", []>;
1425 def v8i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 1, op4,
1426 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1427 opc, !strconcat(Dt, "16"), asm, "", []>;
1428 def v4i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
1429 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1430 opc, !strconcat(Dt, "32"), asm, "", []>;
1431 def v4f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
1432 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1433 opc, "f32", asm, "", []> {
1434 let Inst{10} = 1; // overwrite F = 1
1438 // Neon 3-register vector operations.
1440 // First with only element sizes of 8, 16 and 32 bits:
1441 multiclass N3V_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1442 InstrItinClass itinD16, InstrItinClass itinD32,
1443 InstrItinClass itinQ16, InstrItinClass itinQ32,
1444 string OpcodeStr, string Dt,
1445 SDNode OpNode, bit Commutable = 0> {
1446 // 64-bit vector types.
1447 def v8i8 : N3VD<op24, op23, 0b00, op11_8, op4, itinD16,
1448 OpcodeStr, !strconcat(Dt, "8"),
1449 v8i8, v8i8, OpNode, Commutable>;
1450 def v4i16 : N3VD<op24, op23, 0b01, op11_8, op4, itinD16,
1451 OpcodeStr, !strconcat(Dt, "16"),
1452 v4i16, v4i16, OpNode, Commutable>;
1453 def v2i32 : N3VD<op24, op23, 0b10, op11_8, op4, itinD32,
1454 OpcodeStr, !strconcat(Dt, "32"),
1455 v2i32, v2i32, OpNode, Commutable>;
1457 // 128-bit vector types.
1458 def v16i8 : N3VQ<op24, op23, 0b00, op11_8, op4, itinQ16,
1459 OpcodeStr, !strconcat(Dt, "8"),
1460 v16i8, v16i8, OpNode, Commutable>;
1461 def v8i16 : N3VQ<op24, op23, 0b01, op11_8, op4, itinQ16,
1462 OpcodeStr, !strconcat(Dt, "16"),
1463 v8i16, v8i16, OpNode, Commutable>;
1464 def v4i32 : N3VQ<op24, op23, 0b10, op11_8, op4, itinQ32,
1465 OpcodeStr, !strconcat(Dt, "32"),
1466 v4i32, v4i32, OpNode, Commutable>;
1469 multiclass N3VSL_HS<bits<4> op11_8, string OpcodeStr, string Dt, SDNode ShOp> {
1470 def v4i16 : N3VDSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
1472 def v2i32 : N3VDSL<0b10, op11_8, IIC_VMULi32D, OpcodeStr, !strconcat(Dt,"32"),
1474 def v8i16 : N3VQSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
1475 v8i16, v4i16, ShOp>;
1476 def v4i32 : N3VQSL<0b10, op11_8, IIC_VMULi32Q, OpcodeStr, !strconcat(Dt,"32"),
1477 v4i32, v2i32, ShOp>;
1480 // ....then also with element size 64 bits:
1481 multiclass N3V_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1482 InstrItinClass itinD, InstrItinClass itinQ,
1483 string OpcodeStr, string Dt,
1484 SDNode OpNode, bit Commutable = 0>
1485 : N3V_QHS<op24, op23, op11_8, op4, itinD, itinD, itinQ, itinQ,
1486 OpcodeStr, Dt, OpNode, Commutable> {
1487 def v1i64 : N3VD<op24, op23, 0b11, op11_8, op4, itinD,
1488 OpcodeStr, !strconcat(Dt, "64"),
1489 v1i64, v1i64, OpNode, Commutable>;
1490 def v2i64 : N3VQ<op24, op23, 0b11, op11_8, op4, itinQ,
1491 OpcodeStr, !strconcat(Dt, "64"),
1492 v2i64, v2i64, OpNode, Commutable>;
1496 // Neon Narrowing 2-register vector intrinsics,
1497 // source operand element sizes of 16, 32 and 64 bits:
1498 multiclass N2VNInt_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1499 bits<5> op11_7, bit op6, bit op4,
1500 InstrItinClass itin, string OpcodeStr, string Dt,
1502 def v8i8 : N2VNInt<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
1503 itin, OpcodeStr, !strconcat(Dt, "16"),
1504 v8i8, v8i16, IntOp>;
1505 def v4i16 : N2VNInt<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
1506 itin, OpcodeStr, !strconcat(Dt, "32"),
1507 v4i16, v4i32, IntOp>;
1508 def v2i32 : N2VNInt<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
1509 itin, OpcodeStr, !strconcat(Dt, "64"),
1510 v2i32, v2i64, IntOp>;
1514 // Neon Lengthening 2-register vector intrinsic (currently specific to VMOVL).
1515 // source operand element sizes of 16, 32 and 64 bits:
1516 multiclass N2VLInt_QHS<bits<2> op24_23, bits<5> op11_7, bit op6, bit op4,
1517 string OpcodeStr, string Dt, Intrinsic IntOp> {
1518 def v8i16 : N2VLInt<op24_23, 0b00, 0b10, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
1519 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
1520 def v4i32 : N2VLInt<op24_23, 0b01, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
1521 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
1522 def v2i64 : N2VLInt<op24_23, 0b10, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
1523 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1527 // Neon 3-register vector intrinsics.
1529 // First with only element sizes of 16 and 32 bits:
1530 multiclass N3VInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
1531 InstrItinClass itinD16, InstrItinClass itinD32,
1532 InstrItinClass itinQ16, InstrItinClass itinQ32,
1533 string OpcodeStr, string Dt,
1534 Intrinsic IntOp, bit Commutable = 0> {
1535 // 64-bit vector types.
1536 def v4i16 : N3VDInt<op24, op23, 0b01, op11_8, op4, itinD16,
1537 OpcodeStr, !strconcat(Dt, "16"),
1538 v4i16, v4i16, IntOp, Commutable>;
1539 def v2i32 : N3VDInt<op24, op23, 0b10, op11_8, op4, itinD32,
1540 OpcodeStr, !strconcat(Dt, "32"),
1541 v2i32, v2i32, IntOp, Commutable>;
1543 // 128-bit vector types.
1544 def v8i16 : N3VQInt<op24, op23, 0b01, op11_8, op4, itinQ16,
1545 OpcodeStr, !strconcat(Dt, "16"),
1546 v8i16, v8i16, IntOp, Commutable>;
1547 def v4i32 : N3VQInt<op24, op23, 0b10, op11_8, op4, itinQ32,
1548 OpcodeStr, !strconcat(Dt, "32"),
1549 v4i32, v4i32, IntOp, Commutable>;
1552 multiclass N3VIntSL_HS<bits<4> op11_8,
1553 InstrItinClass itinD16, InstrItinClass itinD32,
1554 InstrItinClass itinQ16, InstrItinClass itinQ32,
1555 string OpcodeStr, string Dt, Intrinsic IntOp> {
1556 def v4i16 : N3VDIntSL16<0b01, op11_8, itinD16,
1557 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp>;
1558 def v2i32 : N3VDIntSL<0b10, op11_8, itinD32,
1559 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp>;
1560 def v8i16 : N3VQIntSL16<0b01, op11_8, itinQ16,
1561 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16, IntOp>;
1562 def v4i32 : N3VQIntSL<0b10, op11_8, itinQ32,
1563 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32, IntOp>;
1566 // ....then also with element size of 8 bits:
1567 multiclass N3VInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1568 InstrItinClass itinD16, InstrItinClass itinD32,
1569 InstrItinClass itinQ16, InstrItinClass itinQ32,
1570 string OpcodeStr, string Dt,
1571 Intrinsic IntOp, bit Commutable = 0>
1572 : N3VInt_HS<op24, op23, op11_8, op4, itinD16, itinD32, itinQ16, itinQ32,
1573 OpcodeStr, Dt, IntOp, Commutable> {
1574 def v8i8 : N3VDInt<op24, op23, 0b00, op11_8, op4, itinD16,
1575 OpcodeStr, !strconcat(Dt, "8"),
1576 v8i8, v8i8, IntOp, Commutable>;
1577 def v16i8 : N3VQInt<op24, op23, 0b00, op11_8, op4, itinQ16,
1578 OpcodeStr, !strconcat(Dt, "8"),
1579 v16i8, v16i8, IntOp, Commutable>;
1582 // ....then also with element size of 64 bits:
1583 multiclass N3VInt_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1584 InstrItinClass itinD16, InstrItinClass itinD32,
1585 InstrItinClass itinQ16, InstrItinClass itinQ32,
1586 string OpcodeStr, string Dt,
1587 Intrinsic IntOp, bit Commutable = 0>
1588 : N3VInt_QHS<op24, op23, op11_8, op4, itinD16, itinD32, itinQ16, itinQ32,
1589 OpcodeStr, Dt, IntOp, Commutable> {
1590 def v1i64 : N3VDInt<op24, op23, 0b11, op11_8, op4, itinD32,
1591 OpcodeStr, !strconcat(Dt, "64"),
1592 v1i64, v1i64, IntOp, Commutable>;
1593 def v2i64 : N3VQInt<op24, op23, 0b11, op11_8, op4, itinQ32,
1594 OpcodeStr, !strconcat(Dt, "64"),
1595 v2i64, v2i64, IntOp, Commutable>;
1599 // Neon Narrowing 3-register vector intrinsics,
1600 // source operand element sizes of 16, 32 and 64 bits:
1601 multiclass N3VNInt_HSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1602 string OpcodeStr, string Dt,
1603 Intrinsic IntOp, bit Commutable = 0> {
1604 def v8i8 : N3VNInt<op24, op23, 0b00, op11_8, op4,
1605 OpcodeStr, !strconcat(Dt, "16"),
1606 v8i8, v8i16, IntOp, Commutable>;
1607 def v4i16 : N3VNInt<op24, op23, 0b01, op11_8, op4,
1608 OpcodeStr, !strconcat(Dt, "32"),
1609 v4i16, v4i32, IntOp, Commutable>;
1610 def v2i32 : N3VNInt<op24, op23, 0b10, op11_8, op4,
1611 OpcodeStr, !strconcat(Dt, "64"),
1612 v2i32, v2i64, IntOp, Commutable>;
1616 // Neon Long 3-register vector intrinsics.
1618 // First with only element sizes of 16 and 32 bits:
1619 multiclass N3VLInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
1620 InstrItinClass itin, string OpcodeStr, string Dt,
1621 Intrinsic IntOp, bit Commutable = 0> {
1622 def v4i32 : N3VLInt<op24, op23, 0b01, op11_8, op4, itin,
1623 OpcodeStr, !strconcat(Dt, "16"),
1624 v4i32, v4i16, IntOp, Commutable>;
1625 def v2i64 : N3VLInt<op24, op23, 0b10, op11_8, op4, itin,
1626 OpcodeStr, !strconcat(Dt, "32"),
1627 v2i64, v2i32, IntOp, Commutable>;
1630 multiclass N3VLIntSL_HS<bit op24, bits<4> op11_8,
1631 InstrItinClass itin, string OpcodeStr, string Dt,
1633 def v4i16 : N3VLIntSL16<op24, 0b01, op11_8, itin,
1634 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
1635 def v2i32 : N3VLIntSL<op24, 0b10, op11_8, itin,
1636 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1639 // ....then also with element size of 8 bits:
1640 multiclass N3VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1641 InstrItinClass itin, string OpcodeStr, string Dt,
1642 Intrinsic IntOp, bit Commutable = 0>
1643 : N3VLInt_HS<op24, op23, op11_8, op4, itin, OpcodeStr, Dt,
1644 IntOp, Commutable> {
1645 def v8i16 : N3VLInt<op24, op23, 0b00, op11_8, op4, itin,
1646 OpcodeStr, !strconcat(Dt, "8"),
1647 v8i16, v8i8, IntOp, Commutable>;
1651 // Neon Wide 3-register vector intrinsics,
1652 // source operand element sizes of 8, 16 and 32 bits:
1653 multiclass N3VWInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1654 string OpcodeStr, string Dt,
1655 Intrinsic IntOp, bit Commutable = 0> {
1656 def v8i16 : N3VWInt<op24, op23, 0b00, op11_8, op4,
1657 OpcodeStr, !strconcat(Dt, "8"),
1658 v8i16, v8i8, IntOp, Commutable>;
1659 def v4i32 : N3VWInt<op24, op23, 0b01, op11_8, op4,
1660 OpcodeStr, !strconcat(Dt, "16"),
1661 v4i32, v4i16, IntOp, Commutable>;
1662 def v2i64 : N3VWInt<op24, op23, 0b10, op11_8, op4,
1663 OpcodeStr, !strconcat(Dt, "32"),
1664 v2i64, v2i32, IntOp, Commutable>;
1668 // Neon Multiply-Op vector operations,
1669 // element sizes of 8, 16 and 32 bits:
1670 multiclass N3VMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1671 InstrItinClass itinD16, InstrItinClass itinD32,
1672 InstrItinClass itinQ16, InstrItinClass itinQ32,
1673 string OpcodeStr, string Dt, SDNode OpNode> {
1674 // 64-bit vector types.
1675 def v8i8 : N3VDMulOp<op24, op23, 0b00, op11_8, op4, itinD16,
1676 OpcodeStr, !strconcat(Dt, "8"), v8i8, mul, OpNode>;
1677 def v4i16 : N3VDMulOp<op24, op23, 0b01, op11_8, op4, itinD16,
1678 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, OpNode>;
1679 def v2i32 : N3VDMulOp<op24, op23, 0b10, op11_8, op4, itinD32,
1680 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, OpNode>;
1682 // 128-bit vector types.
1683 def v16i8 : N3VQMulOp<op24, op23, 0b00, op11_8, op4, itinQ16,
1684 OpcodeStr, !strconcat(Dt, "8"), v16i8, mul, OpNode>;
1685 def v8i16 : N3VQMulOp<op24, op23, 0b01, op11_8, op4, itinQ16,
1686 OpcodeStr, !strconcat(Dt, "16"), v8i16, mul, OpNode>;
1687 def v4i32 : N3VQMulOp<op24, op23, 0b10, op11_8, op4, itinQ32,
1688 OpcodeStr, !strconcat(Dt, "32"), v4i32, mul, OpNode>;
1691 multiclass N3VMulOpSL_HS<bits<4> op11_8,
1692 InstrItinClass itinD16, InstrItinClass itinD32,
1693 InstrItinClass itinQ16, InstrItinClass itinQ32,
1694 string OpcodeStr, string Dt, SDNode ShOp> {
1695 def v4i16 : N3VDMulOpSL16<0b01, op11_8, itinD16,
1696 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, ShOp>;
1697 def v2i32 : N3VDMulOpSL<0b10, op11_8, itinD32,
1698 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, ShOp>;
1699 def v8i16 : N3VQMulOpSL16<0b01, op11_8, itinQ16,
1700 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16,
1702 def v4i32 : N3VQMulOpSL<0b10, op11_8, itinQ32,
1703 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32,
1707 // Neon 3-argument intrinsics,
1708 // element sizes of 8, 16 and 32 bits:
1709 multiclass N3VInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1710 string OpcodeStr, string Dt, Intrinsic IntOp> {
1711 // 64-bit vector types.
1712 def v8i8 : N3VDInt3<op24, op23, 0b00, op11_8, op4, IIC_VMACi16D,
1713 OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
1714 def v4i16 : N3VDInt3<op24, op23, 0b01, op11_8, op4, IIC_VMACi16D,
1715 OpcodeStr, !strconcat(Dt, "16"), v4i16, v4i16, IntOp>;
1716 def v2i32 : N3VDInt3<op24, op23, 0b10, op11_8, op4, IIC_VMACi32D,
1717 OpcodeStr, !strconcat(Dt, "32"), v2i32, v2i32, IntOp>;
1719 // 128-bit vector types.
1720 def v16i8 : N3VQInt3<op24, op23, 0b00, op11_8, op4, IIC_VMACi16Q,
1721 OpcodeStr, !strconcat(Dt, "8"), v16i8, v16i8, IntOp>;
1722 def v8i16 : N3VQInt3<op24, op23, 0b01, op11_8, op4, IIC_VMACi16Q,
1723 OpcodeStr, !strconcat(Dt, "16"), v8i16, v8i16, IntOp>;
1724 def v4i32 : N3VQInt3<op24, op23, 0b10, op11_8, op4, IIC_VMACi32Q,
1725 OpcodeStr, !strconcat(Dt, "32"), v4i32, v4i32, IntOp>;
1729 // Neon Long 3-argument intrinsics.
1731 // First with only element sizes of 16 and 32 bits:
1732 multiclass N3VLInt3_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
1733 string OpcodeStr, string Dt, Intrinsic IntOp> {
1734 def v4i32 : N3VLInt3<op24, op23, 0b01, op11_8, op4, IIC_VMACi16D,
1735 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
1736 def v2i64 : N3VLInt3<op24, op23, 0b10, op11_8, op4, IIC_VMACi16D,
1737 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1740 multiclass N3VLInt3SL_HS<bit op24, bits<4> op11_8,
1741 string OpcodeStr, string Dt, Intrinsic IntOp> {
1742 def v4i16 : N3VLInt3SL16<op24, 0b01, op11_8, IIC_VMACi16D,
1743 OpcodeStr, !strconcat(Dt,"16"), v4i32, v4i16, IntOp>;
1744 def v2i32 : N3VLInt3SL<op24, 0b10, op11_8, IIC_VMACi32D,
1745 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1748 // ....then also with element size of 8 bits:
1749 multiclass N3VLInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1750 string OpcodeStr, string Dt, Intrinsic IntOp>
1751 : N3VLInt3_HS<op24, op23, op11_8, op4, OpcodeStr, Dt, IntOp> {
1752 def v8i16 : N3VLInt3<op24, op23, 0b00, op11_8, op4, IIC_VMACi16D,
1753 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
1757 // Neon 2-register vector intrinsics,
1758 // element sizes of 8, 16 and 32 bits:
1759 multiclass N2VInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1760 bits<5> op11_7, bit op4,
1761 InstrItinClass itinD, InstrItinClass itinQ,
1762 string OpcodeStr, string Dt, Intrinsic IntOp> {
1763 // 64-bit vector types.
1764 def v8i8 : N2VDInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1765 itinD, OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
1766 def v4i16 : N2VDInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1767 itinD, OpcodeStr, !strconcat(Dt, "16"),v4i16,v4i16,IntOp>;
1768 def v2i32 : N2VDInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1769 itinD, OpcodeStr, !strconcat(Dt, "32"),v2i32,v2i32,IntOp>;
1771 // 128-bit vector types.
1772 def v16i8 : N2VQInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1773 itinQ, OpcodeStr, !strconcat(Dt, "8"), v16i8,v16i8,IntOp>;
1774 def v8i16 : N2VQInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1775 itinQ, OpcodeStr, !strconcat(Dt, "16"),v8i16,v8i16,IntOp>;
1776 def v4i32 : N2VQInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1777 itinQ, OpcodeStr, !strconcat(Dt, "32"),v4i32,v4i32,IntOp>;
1781 // Neon Pairwise long 2-register intrinsics,
1782 // element sizes of 8, 16 and 32 bits:
1783 multiclass N2VPLInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1784 bits<5> op11_7, bit op4,
1785 string OpcodeStr, string Dt, Intrinsic IntOp> {
1786 // 64-bit vector types.
1787 def v8i8 : N2VDPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1788 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
1789 def v4i16 : N2VDPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1790 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
1791 def v2i32 : N2VDPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1792 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
1794 // 128-bit vector types.
1795 def v16i8 : N2VQPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1796 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
1797 def v8i16 : N2VQPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1798 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
1799 def v4i32 : N2VQPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1800 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
1804 // Neon Pairwise long 2-register accumulate intrinsics,
1805 // element sizes of 8, 16 and 32 bits:
1806 multiclass N2VPLInt2_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1807 bits<5> op11_7, bit op4,
1808 string OpcodeStr, string Dt, Intrinsic IntOp> {
1809 // 64-bit vector types.
1810 def v8i8 : N2VDPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1811 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
1812 def v4i16 : N2VDPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1813 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
1814 def v2i32 : N2VDPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1815 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
1817 // 128-bit vector types.
1818 def v16i8 : N2VQPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1819 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
1820 def v8i16 : N2VQPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1821 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
1822 def v4i32 : N2VQPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1823 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
1827 // Neon 2-register vector shift by immediate,
1828 // element sizes of 8, 16, 32 and 64 bits:
1829 multiclass N2VSh_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1830 InstrItinClass itin, string OpcodeStr, string Dt,
1832 // 64-bit vector types.
1833 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, itin,
1834 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
1835 let Inst{21-19} = 0b001; // imm6 = 001xxx
1837 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, itin,
1838 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
1839 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1841 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, itin,
1842 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
1843 let Inst{21} = 0b1; // imm6 = 1xxxxx
1845 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, itin,
1846 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
1849 // 128-bit vector types.
1850 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, itin,
1851 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
1852 let Inst{21-19} = 0b001; // imm6 = 001xxx
1854 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, itin,
1855 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
1856 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1858 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, itin,
1859 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
1860 let Inst{21} = 0b1; // imm6 = 1xxxxx
1862 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, itin,
1863 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
1868 // Neon Shift-Accumulate vector operations,
1869 // element sizes of 8, 16, 32 and 64 bits:
1870 multiclass N2VShAdd_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1871 string OpcodeStr, string Dt, SDNode ShOp> {
1872 // 64-bit vector types.
1873 def v8i8 : N2VDShAdd<op24, op23, op11_8, 0, op4,
1874 OpcodeStr, !strconcat(Dt, "8"), v8i8, ShOp> {
1875 let Inst{21-19} = 0b001; // imm6 = 001xxx
1877 def v4i16 : N2VDShAdd<op24, op23, op11_8, 0, op4,
1878 OpcodeStr, !strconcat(Dt, "16"), v4i16, ShOp> {
1879 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1881 def v2i32 : N2VDShAdd<op24, op23, op11_8, 0, op4,
1882 OpcodeStr, !strconcat(Dt, "32"), v2i32, ShOp> {
1883 let Inst{21} = 0b1; // imm6 = 1xxxxx
1885 def v1i64 : N2VDShAdd<op24, op23, op11_8, 1, op4,
1886 OpcodeStr, !strconcat(Dt, "64"), v1i64, ShOp>;
1889 // 128-bit vector types.
1890 def v16i8 : N2VQShAdd<op24, op23, op11_8, 0, op4,
1891 OpcodeStr, !strconcat(Dt, "8"), v16i8, ShOp> {
1892 let Inst{21-19} = 0b001; // imm6 = 001xxx
1894 def v8i16 : N2VQShAdd<op24, op23, op11_8, 0, op4,
1895 OpcodeStr, !strconcat(Dt, "16"), v8i16, ShOp> {
1896 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1898 def v4i32 : N2VQShAdd<op24, op23, op11_8, 0, op4,
1899 OpcodeStr, !strconcat(Dt, "32"), v4i32, ShOp> {
1900 let Inst{21} = 0b1; // imm6 = 1xxxxx
1902 def v2i64 : N2VQShAdd<op24, op23, op11_8, 1, op4,
1903 OpcodeStr, !strconcat(Dt, "64"), v2i64, ShOp>;
1908 // Neon Shift-Insert vector operations,
1909 // element sizes of 8, 16, 32 and 64 bits:
1910 multiclass N2VShIns_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1911 string OpcodeStr, SDNode ShOp> {
1912 // 64-bit vector types.
1913 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4,
1914 OpcodeStr, "8", v8i8, ShOp> {
1915 let Inst{21-19} = 0b001; // imm6 = 001xxx
1917 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4,
1918 OpcodeStr, "16", v4i16, ShOp> {
1919 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1921 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4,
1922 OpcodeStr, "32", v2i32, ShOp> {
1923 let Inst{21} = 0b1; // imm6 = 1xxxxx
1925 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4,
1926 OpcodeStr, "64", v1i64, ShOp>;
1929 // 128-bit vector types.
1930 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4,
1931 OpcodeStr, "8", v16i8, ShOp> {
1932 let Inst{21-19} = 0b001; // imm6 = 001xxx
1934 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4,
1935 OpcodeStr, "16", v8i16, ShOp> {
1936 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1938 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4,
1939 OpcodeStr, "32", v4i32, ShOp> {
1940 let Inst{21} = 0b1; // imm6 = 1xxxxx
1942 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4,
1943 OpcodeStr, "64", v2i64, ShOp>;
1947 // Neon Shift Long operations,
1948 // element sizes of 8, 16, 32 bits:
1949 multiclass N2VLSh_QHS<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
1950 bit op4, string OpcodeStr, string Dt, SDNode OpNode> {
1951 def v8i16 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
1952 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode> {
1953 let Inst{21-19} = 0b001; // imm6 = 001xxx
1955 def v4i32 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
1956 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode> {
1957 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1959 def v2i64 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
1960 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode> {
1961 let Inst{21} = 0b1; // imm6 = 1xxxxx
1965 // Neon Shift Narrow operations,
1966 // element sizes of 16, 32, 64 bits:
1967 multiclass N2VNSh_HSD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
1968 bit op4, InstrItinClass itin, string OpcodeStr, string Dt,
1970 def v8i8 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
1971 OpcodeStr, !strconcat(Dt, "16"), v8i8, v8i16, OpNode> {
1972 let Inst{21-19} = 0b001; // imm6 = 001xxx
1974 def v4i16 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
1975 OpcodeStr, !strconcat(Dt, "32"), v4i16, v4i32, OpNode> {
1976 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1978 def v2i32 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
1979 OpcodeStr, !strconcat(Dt, "64"), v2i32, v2i64, OpNode> {
1980 let Inst{21} = 0b1; // imm6 = 1xxxxx
1984 //===----------------------------------------------------------------------===//
1985 // Instruction Definitions.
1986 //===----------------------------------------------------------------------===//
1988 // Vector Add Operations.
1990 // VADD : Vector Add (integer and floating-point)
1991 defm VADD : N3V_QHSD<0, 0, 0b1000, 0, IIC_VBINiD, IIC_VBINiQ, "vadd", "i",
1993 def VADDfd : N3VD<0, 0, 0b00, 0b1101, 0, IIC_VBIND, "vadd", "f32",
1994 v2f32, v2f32, fadd, 1>;
1995 def VADDfq : N3VQ<0, 0, 0b00, 0b1101, 0, IIC_VBINQ, "vadd", "f32",
1996 v4f32, v4f32, fadd, 1>;
1997 // VADDL : Vector Add Long (Q = D + D)
1998 defm VADDLs : N3VLInt_QHS<0,1,0b0000,0, IIC_VSHLiD, "vaddl", "s",
1999 int_arm_neon_vaddls, 1>;
2000 defm VADDLu : N3VLInt_QHS<1,1,0b0000,0, IIC_VSHLiD, "vaddl", "u",
2001 int_arm_neon_vaddlu, 1>;
2002 // VADDW : Vector Add Wide (Q = Q + D)
2003 defm VADDWs : N3VWInt_QHS<0,1,0b0001,0, "vaddw", "s", int_arm_neon_vaddws, 0>;
2004 defm VADDWu : N3VWInt_QHS<1,1,0b0001,0, "vaddw", "u", int_arm_neon_vaddwu, 0>;
2005 // VHADD : Vector Halving Add
2006 defm VHADDs : N3VInt_QHS<0,0,0b0000,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2007 IIC_VBINi4Q, "vhadd", "s", int_arm_neon_vhadds, 1>;
2008 defm VHADDu : N3VInt_QHS<1,0,0b0000,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2009 IIC_VBINi4Q, "vhadd", "u", int_arm_neon_vhaddu, 1>;
2010 // VRHADD : Vector Rounding Halving Add
2011 defm VRHADDs : N3VInt_QHS<0,0,0b0001,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2012 IIC_VBINi4Q, "vrhadd", "s", int_arm_neon_vrhadds, 1>;
2013 defm VRHADDu : N3VInt_QHS<1,0,0b0001,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2014 IIC_VBINi4Q, "vrhadd", "u", int_arm_neon_vrhaddu, 1>;
2015 // VQADD : Vector Saturating Add
2016 defm VQADDs : N3VInt_QHSD<0,0,0b0000,1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2017 IIC_VBINi4Q, "vqadd", "s", int_arm_neon_vqadds, 1>;
2018 defm VQADDu : N3VInt_QHSD<1,0,0b0000,1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2019 IIC_VBINi4Q, "vqadd", "u", int_arm_neon_vqaddu, 1>;
2020 // VADDHN : Vector Add and Narrow Returning High Half (D = Q + Q)
2021 defm VADDHN : N3VNInt_HSD<0,1,0b0100,0, "vaddhn", "i",
2022 int_arm_neon_vaddhn, 1>;
2023 // VRADDHN : Vector Rounding Add and Narrow Returning High Half (D = Q + Q)
2024 defm VRADDHN : N3VNInt_HSD<1,1,0b0100,0, "vraddhn", "i",
2025 int_arm_neon_vraddhn, 1>;
2027 // Vector Multiply Operations.
2029 // VMUL : Vector Multiply (integer, polynomial and floating-point)
2030 defm VMUL : N3V_QHS<0, 0, 0b1001, 1, IIC_VMULi16D, IIC_VMULi32D,
2031 IIC_VMULi16Q, IIC_VMULi32Q, "vmul", "i", mul, 1>;
2032 def VMULpd : N3VDInt<1, 0, 0b00, 0b1001, 1, IIC_VMULi16D, "vmul", "p8",
2033 v8i8, v8i8, int_arm_neon_vmulp, 1>;
2034 def VMULpq : N3VQInt<1, 0, 0b00, 0b1001, 1, IIC_VMULi16Q, "vmul", "p8",
2035 v16i8, v16i8, int_arm_neon_vmulp, 1>;
2036 def VMULfd : N3VD<1, 0, 0b00, 0b1101, 1, IIC_VBIND, "vmul", "f32",
2037 v2f32, v2f32, fmul, 1>;
2038 def VMULfq : N3VQ<1, 0, 0b00, 0b1101, 1, IIC_VBINQ, "vmul", "f32",
2039 v4f32, v4f32, fmul, 1>;
2040 defm VMULsl : N3VSL_HS<0b1000, "vmul", "i", mul>;
2041 def VMULslfd : N3VDSL<0b10, 0b1001, IIC_VBIND, "vmul", "f32", v2f32, fmul>;
2042 def VMULslfq : N3VQSL<0b10, 0b1001, IIC_VBINQ, "vmul", "f32", v4f32,
2045 def : Pat<(v8i16 (mul (v8i16 QPR:$src1),
2046 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
2047 (v8i16 (VMULslv8i16 (v8i16 QPR:$src1),
2048 (v4i16 (EXTRACT_SUBREG QPR:$src2,
2049 (DSubReg_i16_reg imm:$lane))),
2050 (SubReg_i16_lane imm:$lane)))>;
2051 def : Pat<(v4i32 (mul (v4i32 QPR:$src1),
2052 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
2053 (v4i32 (VMULslv4i32 (v4i32 QPR:$src1),
2054 (v2i32 (EXTRACT_SUBREG QPR:$src2,
2055 (DSubReg_i32_reg imm:$lane))),
2056 (SubReg_i32_lane imm:$lane)))>;
2057 def : Pat<(v4f32 (fmul (v4f32 QPR:$src1),
2058 (v4f32 (NEONvduplane (v4f32 QPR:$src2), imm:$lane)))),
2059 (v4f32 (VMULslfq (v4f32 QPR:$src1),
2060 (v2f32 (EXTRACT_SUBREG QPR:$src2,
2061 (DSubReg_i32_reg imm:$lane))),
2062 (SubReg_i32_lane imm:$lane)))>;
2064 // VQDMULH : Vector Saturating Doubling Multiply Returning High Half
2065 defm VQDMULH : N3VInt_HS<0, 0, 0b1011, 0, IIC_VMULi16D, IIC_VMULi32D,
2066 IIC_VMULi16Q, IIC_VMULi32Q,
2067 "vqdmulh", "s", int_arm_neon_vqdmulh, 1>;
2068 defm VQDMULHsl: N3VIntSL_HS<0b1100, IIC_VMULi16D, IIC_VMULi32D,
2069 IIC_VMULi16Q, IIC_VMULi32Q,
2070 "vqdmulh", "s", int_arm_neon_vqdmulh>;
2071 def : Pat<(v8i16 (int_arm_neon_vqdmulh (v8i16 QPR:$src1),
2072 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
2074 (v8i16 (VQDMULHslv8i16 (v8i16 QPR:$src1),
2075 (v4i16 (EXTRACT_SUBREG QPR:$src2,
2076 (DSubReg_i16_reg imm:$lane))),
2077 (SubReg_i16_lane imm:$lane)))>;
2078 def : Pat<(v4i32 (int_arm_neon_vqdmulh (v4i32 QPR:$src1),
2079 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
2081 (v4i32 (VQDMULHslv4i32 (v4i32 QPR:$src1),
2082 (v2i32 (EXTRACT_SUBREG QPR:$src2,
2083 (DSubReg_i32_reg imm:$lane))),
2084 (SubReg_i32_lane imm:$lane)))>;
2086 // VQRDMULH : Vector Rounding Saturating Doubling Multiply Returning High Half
2087 defm VQRDMULH : N3VInt_HS<1, 0, 0b1011, 0, IIC_VMULi16D, IIC_VMULi32D,
2088 IIC_VMULi16Q, IIC_VMULi32Q,
2089 "vqrdmulh", "s", int_arm_neon_vqrdmulh, 1>;
2090 defm VQRDMULHsl : N3VIntSL_HS<0b1101, IIC_VMULi16D, IIC_VMULi32D,
2091 IIC_VMULi16Q, IIC_VMULi32Q,
2092 "vqrdmulh", "s", int_arm_neon_vqrdmulh>;
2093 def : Pat<(v8i16 (int_arm_neon_vqrdmulh (v8i16 QPR:$src1),
2094 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
2096 (v8i16 (VQRDMULHslv8i16 (v8i16 QPR:$src1),
2097 (v4i16 (EXTRACT_SUBREG QPR:$src2,
2098 (DSubReg_i16_reg imm:$lane))),
2099 (SubReg_i16_lane imm:$lane)))>;
2100 def : Pat<(v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$src1),
2101 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
2103 (v4i32 (VQRDMULHslv4i32 (v4i32 QPR:$src1),
2104 (v2i32 (EXTRACT_SUBREG QPR:$src2,
2105 (DSubReg_i32_reg imm:$lane))),
2106 (SubReg_i32_lane imm:$lane)))>;
2108 // VMULL : Vector Multiply Long (integer and polynomial) (Q = D * D)
2109 defm VMULLs : N3VLInt_QHS<0,1,0b1100,0, IIC_VMULi16D, "vmull", "s",
2110 int_arm_neon_vmulls, 1>;
2111 defm VMULLu : N3VLInt_QHS<1,1,0b1100,0, IIC_VMULi16D, "vmull", "u",
2112 int_arm_neon_vmullu, 1>;
2113 def VMULLp : N3VLInt<0, 1, 0b00, 0b1110, 0, IIC_VMULi16D, "vmull", "p8",
2114 v8i16, v8i8, int_arm_neon_vmullp, 1>;
2115 defm VMULLsls : N3VLIntSL_HS<0, 0b1010, IIC_VMULi16D, "vmull", "s",
2116 int_arm_neon_vmulls>;
2117 defm VMULLslu : N3VLIntSL_HS<1, 0b1010, IIC_VMULi16D, "vmull", "u",
2118 int_arm_neon_vmullu>;
2120 // VQDMULL : Vector Saturating Doubling Multiply Long (Q = D * D)
2121 defm VQDMULL : N3VLInt_HS<0,1,0b1101,0, IIC_VMULi16D, "vqdmull", "s",
2122 int_arm_neon_vqdmull, 1>;
2123 defm VQDMULLsl: N3VLIntSL_HS<0, 0b1011, IIC_VMULi16D, "vqdmull", "s",
2124 int_arm_neon_vqdmull>;
2126 // Vector Multiply-Accumulate and Multiply-Subtract Operations.
2128 // VMLA : Vector Multiply Accumulate (integer and floating-point)
2129 defm VMLA : N3VMulOp_QHS<0, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
2130 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
2131 def VMLAfd : N3VDMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACD, "vmla", "f32",
2133 def VMLAfq : N3VQMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACQ, "vmla", "f32",
2135 defm VMLAsl : N3VMulOpSL_HS<0b0000, IIC_VMACi16D, IIC_VMACi32D,
2136 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
2137 def VMLAslfd : N3VDMulOpSL<0b10, 0b0001, IIC_VMACD, "vmla", "f32",
2139 def VMLAslfq : N3VQMulOpSL<0b10, 0b0001, IIC_VMACQ, "vmla", "f32",
2140 v4f32, v2f32, fmul, fadd>;
2142 def : Pat<(v8i16 (add (v8i16 QPR:$src1),
2143 (mul (v8i16 QPR:$src2),
2144 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
2145 (v8i16 (VMLAslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
2146 (v4i16 (EXTRACT_SUBREG QPR:$src3,
2147 (DSubReg_i16_reg imm:$lane))),
2148 (SubReg_i16_lane imm:$lane)))>;
2150 def : Pat<(v4i32 (add (v4i32 QPR:$src1),
2151 (mul (v4i32 QPR:$src2),
2152 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
2153 (v4i32 (VMLAslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
2154 (v2i32 (EXTRACT_SUBREG QPR:$src3,
2155 (DSubReg_i32_reg imm:$lane))),
2156 (SubReg_i32_lane imm:$lane)))>;
2158 def : Pat<(v4f32 (fadd (v4f32 QPR:$src1),
2159 (fmul (v4f32 QPR:$src2),
2160 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
2161 (v4f32 (VMLAslfq (v4f32 QPR:$src1),
2163 (v2f32 (EXTRACT_SUBREG QPR:$src3,
2164 (DSubReg_i32_reg imm:$lane))),
2165 (SubReg_i32_lane imm:$lane)))>;
2167 // VMLAL : Vector Multiply Accumulate Long (Q += D * D)
2168 defm VMLALs : N3VLInt3_QHS<0,1,0b1000,0, "vmlal", "s", int_arm_neon_vmlals>;
2169 defm VMLALu : N3VLInt3_QHS<1,1,0b1000,0, "vmlal", "u", int_arm_neon_vmlalu>;
2171 defm VMLALsls : N3VLInt3SL_HS<0, 0b0010, "vmlal", "s", int_arm_neon_vmlals>;
2172 defm VMLALslu : N3VLInt3SL_HS<1, 0b0010, "vmlal", "u", int_arm_neon_vmlalu>;
2174 // VQDMLAL : Vector Saturating Doubling Multiply Accumulate Long (Q += D * D)
2175 defm VQDMLAL : N3VLInt3_HS<0, 1, 0b1001, 0, "vqdmlal", "s",
2176 int_arm_neon_vqdmlal>;
2177 defm VQDMLALsl: N3VLInt3SL_HS<0, 0b0011, "vqdmlal", "s", int_arm_neon_vqdmlal>;
2179 // VMLS : Vector Multiply Subtract (integer and floating-point)
2180 defm VMLS : N3VMulOp_QHS<1, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
2181 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
2182 def VMLSfd : N3VDMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACD, "vmls", "f32",
2184 def VMLSfq : N3VQMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACQ, "vmls", "f32",
2186 defm VMLSsl : N3VMulOpSL_HS<0b0100, IIC_VMACi16D, IIC_VMACi32D,
2187 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
2188 def VMLSslfd : N3VDMulOpSL<0b10, 0b0101, IIC_VMACD, "vmls", "f32",
2190 def VMLSslfq : N3VQMulOpSL<0b10, 0b0101, IIC_VMACQ, "vmls", "f32",
2191 v4f32, v2f32, fmul, fsub>;
2193 def : Pat<(v8i16 (sub (v8i16 QPR:$src1),
2194 (mul (v8i16 QPR:$src2),
2195 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
2196 (v8i16 (VMLSslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
2197 (v4i16 (EXTRACT_SUBREG QPR:$src3,
2198 (DSubReg_i16_reg imm:$lane))),
2199 (SubReg_i16_lane imm:$lane)))>;
2201 def : Pat<(v4i32 (sub (v4i32 QPR:$src1),
2202 (mul (v4i32 QPR:$src2),
2203 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
2204 (v4i32 (VMLSslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
2205 (v2i32 (EXTRACT_SUBREG QPR:$src3,
2206 (DSubReg_i32_reg imm:$lane))),
2207 (SubReg_i32_lane imm:$lane)))>;
2209 def : Pat<(v4f32 (fsub (v4f32 QPR:$src1),
2210 (fmul (v4f32 QPR:$src2),
2211 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
2212 (v4f32 (VMLSslfq (v4f32 QPR:$src1), (v4f32 QPR:$src2),
2213 (v2f32 (EXTRACT_SUBREG QPR:$src3,
2214 (DSubReg_i32_reg imm:$lane))),
2215 (SubReg_i32_lane imm:$lane)))>;
2217 // VMLSL : Vector Multiply Subtract Long (Q -= D * D)
2218 defm VMLSLs : N3VLInt3_QHS<0,1,0b1010,0, "vmlsl", "s", int_arm_neon_vmlsls>;
2219 defm VMLSLu : N3VLInt3_QHS<1,1,0b1010,0, "vmlsl", "u", int_arm_neon_vmlslu>;
2221 defm VMLSLsls : N3VLInt3SL_HS<0, 0b0110, "vmlsl", "s", int_arm_neon_vmlsls>;
2222 defm VMLSLslu : N3VLInt3SL_HS<1, 0b0110, "vmlsl", "u", int_arm_neon_vmlslu>;
2224 // VQDMLSL : Vector Saturating Doubling Multiply Subtract Long (Q -= D * D)
2225 defm VQDMLSL : N3VLInt3_HS<0, 1, 0b1011, 0, "vqdmlsl", "s",
2226 int_arm_neon_vqdmlsl>;
2227 defm VQDMLSLsl: N3VLInt3SL_HS<0, 0b111, "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
2229 // Vector Subtract Operations.
2231 // VSUB : Vector Subtract (integer and floating-point)
2232 defm VSUB : N3V_QHSD<1, 0, 0b1000, 0, IIC_VSUBiD, IIC_VSUBiQ,
2233 "vsub", "i", sub, 0>;
2234 def VSUBfd : N3VD<0, 0, 0b10, 0b1101, 0, IIC_VBIND, "vsub", "f32",
2235 v2f32, v2f32, fsub, 0>;
2236 def VSUBfq : N3VQ<0, 0, 0b10, 0b1101, 0, IIC_VBINQ, "vsub", "f32",
2237 v4f32, v4f32, fsub, 0>;
2238 // VSUBL : Vector Subtract Long (Q = D - D)
2239 defm VSUBLs : N3VLInt_QHS<0,1,0b0010,0, IIC_VSHLiD, "vsubl", "s",
2240 int_arm_neon_vsubls, 1>;
2241 defm VSUBLu : N3VLInt_QHS<1,1,0b0010,0, IIC_VSHLiD, "vsubl", "u",
2242 int_arm_neon_vsublu, 1>;
2243 // VSUBW : Vector Subtract Wide (Q = Q - D)
2244 defm VSUBWs : N3VWInt_QHS<0,1,0b0011,0, "vsubw", "s", int_arm_neon_vsubws, 0>;
2245 defm VSUBWu : N3VWInt_QHS<1,1,0b0011,0, "vsubw", "u", int_arm_neon_vsubwu, 0>;
2246 // VHSUB : Vector Halving Subtract
2247 defm VHSUBs : N3VInt_QHS<0, 0, 0b0010, 0, IIC_VBINi4D, IIC_VBINi4D,
2248 IIC_VBINi4Q, IIC_VBINi4Q,
2249 "vhsub", "s", int_arm_neon_vhsubs, 0>;
2250 defm VHSUBu : N3VInt_QHS<1, 0, 0b0010, 0, IIC_VBINi4D, IIC_VBINi4D,
2251 IIC_VBINi4Q, IIC_VBINi4Q,
2252 "vhsub", "u", int_arm_neon_vhsubu, 0>;
2253 // VQSUB : Vector Saturing Subtract
2254 defm VQSUBs : N3VInt_QHSD<0, 0, 0b0010, 1, IIC_VBINi4D, IIC_VBINi4D,
2255 IIC_VBINi4Q, IIC_VBINi4Q,
2256 "vqsub", "s", int_arm_neon_vqsubs, 0>;
2257 defm VQSUBu : N3VInt_QHSD<1, 0, 0b0010, 1, IIC_VBINi4D, IIC_VBINi4D,
2258 IIC_VBINi4Q, IIC_VBINi4Q,
2259 "vqsub", "u", int_arm_neon_vqsubu, 0>;
2260 // VSUBHN : Vector Subtract and Narrow Returning High Half (D = Q - Q)
2261 defm VSUBHN : N3VNInt_HSD<0,1,0b0110,0, "vsubhn", "i",
2262 int_arm_neon_vsubhn, 0>;
2263 // VRSUBHN : Vector Rounding Subtract and Narrow Returning High Half (D=Q-Q)
2264 defm VRSUBHN : N3VNInt_HSD<1,1,0b0110,0, "vrsubhn", "i",
2265 int_arm_neon_vrsubhn, 0>;
2267 // Vector Comparisons.
2269 // VCEQ : Vector Compare Equal
2270 defm VCEQ : N3V_QHS<1, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2271 IIC_VBINi4Q, "vceq", "i", NEONvceq, 1>;
2272 def VCEQfd : N3VD<0,0,0b00,0b1110,0, IIC_VBIND, "vceq", "f32", v2i32, v2f32,
2274 def VCEQfq : N3VQ<0,0,0b00,0b1110,0, IIC_VBINQ, "vceq", "f32", v4i32, v4f32,
2276 // For disassembly only.
2277 defm VCEQz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00010, 0, "vceq", "i",
2280 // VCGE : Vector Compare Greater Than or Equal
2281 defm VCGEs : N3V_QHS<0, 0, 0b0011, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2282 IIC_VBINi4Q, "vcge", "s", NEONvcge, 0>;
2283 defm VCGEu : N3V_QHS<1, 0, 0b0011, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2284 IIC_VBINi4Q, "vcge", "u", NEONvcgeu, 0>;
2285 def VCGEfd : N3VD<1,0,0b00,0b1110,0, IIC_VBIND, "vcge", "f32",
2286 v2i32, v2f32, NEONvcge, 0>;
2287 def VCGEfq : N3VQ<1,0,0b00,0b1110,0, IIC_VBINQ, "vcge", "f32", v4i32, v4f32,
2289 // For disassembly only.
2290 defm VCGEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00001, 0, "vcge", "s",
2292 // For disassembly only.
2293 defm VCLEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00011, 0, "vcle", "s",
2296 // VCGT : Vector Compare Greater Than
2297 defm VCGTs : N3V_QHS<0, 0, 0b0011, 0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2298 IIC_VBINi4Q, "vcgt", "s", NEONvcgt, 0>;
2299 defm VCGTu : N3V_QHS<1, 0, 0b0011, 0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2300 IIC_VBINi4Q, "vcgt", "u", NEONvcgtu, 0>;
2301 def VCGTfd : N3VD<1,0,0b10,0b1110,0, IIC_VBIND, "vcgt", "f32", v2i32, v2f32,
2303 def VCGTfq : N3VQ<1,0,0b10,0b1110,0, IIC_VBINQ, "vcgt", "f32", v4i32, v4f32,
2305 // For disassembly only.
2306 defm VCGTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00000, 0, "vcgt", "s",
2308 // For disassembly only.
2309 defm VCLTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00100, 0, "vclt", "s",
2312 // VACGE : Vector Absolute Compare Greater Than or Equal (aka VCAGE)
2313 def VACGEd : N3VDInt<1, 0, 0b00, 0b1110, 1, IIC_VBIND, "vacge", "f32",
2314 v2i32, v2f32, int_arm_neon_vacged, 0>;
2315 def VACGEq : N3VQInt<1, 0, 0b00, 0b1110, 1, IIC_VBINQ, "vacge", "f32",
2316 v4i32, v4f32, int_arm_neon_vacgeq, 0>;
2317 // VACGT : Vector Absolute Compare Greater Than (aka VCAGT)
2318 def VACGTd : N3VDInt<1, 0, 0b10, 0b1110, 1, IIC_VBIND, "vacgt", "f32",
2319 v2i32, v2f32, int_arm_neon_vacgtd, 0>;
2320 def VACGTq : N3VQInt<1, 0, 0b10, 0b1110, 1, IIC_VBINQ, "vacgt", "f32",
2321 v4i32, v4f32, int_arm_neon_vacgtq, 0>;
2322 // VTST : Vector Test Bits
2323 defm VTST : N3V_QHS<0, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2324 IIC_VBINi4Q, "vtst", "", NEONvtst, 1>;
2326 // Vector Bitwise Operations.
2328 // VAND : Vector Bitwise AND
2329 def VANDd : N3VDX<0, 0, 0b00, 0b0001, 1, IIC_VBINiD, "vand",
2330 v2i32, v2i32, and, 1>;
2331 def VANDq : N3VQX<0, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "vand",
2332 v4i32, v4i32, and, 1>;
2334 // VEOR : Vector Bitwise Exclusive OR
2335 def VEORd : N3VDX<1, 0, 0b00, 0b0001, 1, IIC_VBINiD, "veor",
2336 v2i32, v2i32, xor, 1>;
2337 def VEORq : N3VQX<1, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "veor",
2338 v4i32, v4i32, xor, 1>;
2340 // VORR : Vector Bitwise OR
2341 def VORRd : N3VDX<0, 0, 0b10, 0b0001, 1, IIC_VBINiD, "vorr",
2342 v2i32, v2i32, or, 1>;
2343 def VORRq : N3VQX<0, 0, 0b10, 0b0001, 1, IIC_VBINiQ, "vorr",
2344 v4i32, v4i32, or, 1>;
2346 // VBIC : Vector Bitwise Bit Clear (AND NOT)
2347 def VBICd : N3VX<0, 0, 0b01, 0b0001, 0, 1, (outs DPR:$dst),
2348 (ins DPR:$src1, DPR:$src2), IIC_VBINiD,
2349 "vbic", "$dst, $src1, $src2", "",
2350 [(set DPR:$dst, (v2i32 (and DPR:$src1,
2351 (vnot_conv DPR:$src2))))]>;
2352 def VBICq : N3VX<0, 0, 0b01, 0b0001, 1, 1, (outs QPR:$dst),
2353 (ins QPR:$src1, QPR:$src2), IIC_VBINiQ,
2354 "vbic", "$dst, $src1, $src2", "",
2355 [(set QPR:$dst, (v4i32 (and QPR:$src1,
2356 (vnot_conv QPR:$src2))))]>;
2358 // VORN : Vector Bitwise OR NOT
2359 def VORNd : N3VX<0, 0, 0b11, 0b0001, 0, 1, (outs DPR:$dst),
2360 (ins DPR:$src1, DPR:$src2), IIC_VBINiD,
2361 "vorn", "$dst, $src1, $src2", "",
2362 [(set DPR:$dst, (v2i32 (or DPR:$src1,
2363 (vnot_conv DPR:$src2))))]>;
2364 def VORNq : N3VX<0, 0, 0b11, 0b0001, 1, 1, (outs QPR:$dst),
2365 (ins QPR:$src1, QPR:$src2), IIC_VBINiQ,
2366 "vorn", "$dst, $src1, $src2", "",
2367 [(set QPR:$dst, (v4i32 (or QPR:$src1,
2368 (vnot_conv QPR:$src2))))]>;
2370 // VMVN : Vector Bitwise NOT
2371 def VMVNd : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 0, 0,
2372 (outs DPR:$dst), (ins DPR:$src), IIC_VSHLiD,
2373 "vmvn", "$dst, $src", "",
2374 [(set DPR:$dst, (v2i32 (vnot DPR:$src)))]>;
2375 def VMVNq : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 1, 0,
2376 (outs QPR:$dst), (ins QPR:$src), IIC_VSHLiD,
2377 "vmvn", "$dst, $src", "",
2378 [(set QPR:$dst, (v4i32 (vnot QPR:$src)))]>;
2379 def : Pat<(v2i32 (vnot_conv DPR:$src)), (VMVNd DPR:$src)>;
2380 def : Pat<(v4i32 (vnot_conv QPR:$src)), (VMVNq QPR:$src)>;
2382 // VBSL : Vector Bitwise Select
2383 def VBSLd : N3VX<1, 0, 0b01, 0b0001, 0, 1, (outs DPR:$dst),
2384 (ins DPR:$src1, DPR:$src2, DPR:$src3), IIC_VCNTiD,
2385 "vbsl", "$dst, $src2, $src3", "$src1 = $dst",
2387 (v2i32 (or (and DPR:$src2, DPR:$src1),
2388 (and DPR:$src3, (vnot_conv DPR:$src1)))))]>;
2389 def VBSLq : N3VX<1, 0, 0b01, 0b0001, 1, 1, (outs QPR:$dst),
2390 (ins QPR:$src1, QPR:$src2, QPR:$src3), IIC_VCNTiQ,
2391 "vbsl", "$dst, $src2, $src3", "$src1 = $dst",
2393 (v4i32 (or (and QPR:$src2, QPR:$src1),
2394 (and QPR:$src3, (vnot_conv QPR:$src1)))))]>;
2396 // VBIF : Vector Bitwise Insert if False
2397 // like VBSL but with: "vbif $dst, $src3, $src1", "$src2 = $dst",
2398 def VBIFd : N3VX<1, 0, 0b11, 0b0001, 0, 1,
2399 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3),
2400 IIC_VBINiD, "vbif", "$dst, $src2, $src3", "$src1 = $dst",
2401 [/* For disassembly only; pattern left blank */]>;
2402 def VBIFq : N3VX<1, 0, 0b11, 0b0001, 1, 1,
2403 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3),
2404 IIC_VBINiQ, "vbif", "$dst, $src2, $src3", "$src1 = $dst",
2405 [/* For disassembly only; pattern left blank */]>;
2407 // VBIT : Vector Bitwise Insert if True
2408 // like VBSL but with: "vbit $dst, $src2, $src1", "$src3 = $dst",
2409 def VBITd : N3VX<1, 0, 0b10, 0b0001, 0, 1,
2410 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3),
2411 IIC_VBINiD, "vbit", "$dst, $src2, $src3", "$src1 = $dst",
2412 [/* For disassembly only; pattern left blank */]>;
2413 def VBITq : N3VX<1, 0, 0b10, 0b0001, 1, 1,
2414 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3),
2415 IIC_VBINiQ, "vbit", "$dst, $src2, $src3", "$src1 = $dst",
2416 [/* For disassembly only; pattern left blank */]>;
2418 // VBIT/VBIF are not yet implemented. The TwoAddress pass will not go looking
2419 // for equivalent operations with different register constraints; it just
2422 // Vector Absolute Differences.
2424 // VABD : Vector Absolute Difference
2425 defm VABDs : N3VInt_QHS<0, 0, 0b0111, 0, IIC_VBINi4D, IIC_VBINi4D,
2426 IIC_VBINi4Q, IIC_VBINi4Q,
2427 "vabd", "s", int_arm_neon_vabds, 0>;
2428 defm VABDu : N3VInt_QHS<1, 0, 0b0111, 0, IIC_VBINi4D, IIC_VBINi4D,
2429 IIC_VBINi4Q, IIC_VBINi4Q,
2430 "vabd", "u", int_arm_neon_vabdu, 0>;
2431 def VABDfd : N3VDInt<1, 0, 0b10, 0b1101, 0, IIC_VBIND,
2432 "vabd", "f32", v2f32, v2f32, int_arm_neon_vabds, 0>;
2433 def VABDfq : N3VQInt<1, 0, 0b10, 0b1101, 0, IIC_VBINQ,
2434 "vabd", "f32", v4f32, v4f32, int_arm_neon_vabds, 0>;
2436 // VABDL : Vector Absolute Difference Long (Q = | D - D |)
2437 defm VABDLs : N3VLInt_QHS<0,1,0b0111,0, IIC_VBINi4Q,
2438 "vabdl", "s", int_arm_neon_vabdls, 0>;
2439 defm VABDLu : N3VLInt_QHS<1,1,0b0111,0, IIC_VBINi4Q,
2440 "vabdl", "u", int_arm_neon_vabdlu, 0>;
2442 // VABA : Vector Absolute Difference and Accumulate
2443 defm VABAs : N3VInt3_QHS<0,0,0b0111,1, "vaba", "s", int_arm_neon_vabas>;
2444 defm VABAu : N3VInt3_QHS<1,0,0b0111,1, "vaba", "u", int_arm_neon_vabau>;
2446 // VABAL : Vector Absolute Difference and Accumulate Long (Q += | D - D |)
2447 defm VABALs : N3VLInt3_QHS<0,1,0b0101,0, "vabal", "s", int_arm_neon_vabals>;
2448 defm VABALu : N3VLInt3_QHS<1,1,0b0101,0, "vabal", "u", int_arm_neon_vabalu>;
2450 // Vector Maximum and Minimum.
2452 // VMAX : Vector Maximum
2453 defm VMAXs : N3VInt_QHS<0,0,0b0110,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2454 IIC_VBINi4Q, "vmax", "s", int_arm_neon_vmaxs, 1>;
2455 defm VMAXu : N3VInt_QHS<1,0,0b0110,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2456 IIC_VBINi4Q, "vmax", "u", int_arm_neon_vmaxu, 1>;
2457 def VMAXfd : N3VDInt<0, 0, 0b00, 0b1111, 0, IIC_VBIND, "vmax", "f32",
2458 v2f32, v2f32, int_arm_neon_vmaxs, 1>;
2459 def VMAXfq : N3VQInt<0, 0, 0b00, 0b1111, 0, IIC_VBINQ, "vmax", "f32",
2460 v4f32, v4f32, int_arm_neon_vmaxs, 1>;
2462 // VMIN : Vector Minimum
2463 defm VMINs : N3VInt_QHS<0,0,0b0110,1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2464 IIC_VBINi4Q, "vmin", "s", int_arm_neon_vmins, 1>;
2465 defm VMINu : N3VInt_QHS<1,0,0b0110,1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2466 IIC_VBINi4Q, "vmin", "u", int_arm_neon_vminu, 1>;
2467 def VMINfd : N3VDInt<0, 0, 0b10, 0b1111, 0, IIC_VBIND, "vmin", "f32",
2468 v2f32, v2f32, int_arm_neon_vmins, 1>;
2469 def VMINfq : N3VQInt<0, 0, 0b10, 0b1111, 0, IIC_VBINQ, "vmin", "f32",
2470 v4f32, v4f32, int_arm_neon_vmins, 1>;
2472 // Vector Pairwise Operations.
2474 // VPADD : Vector Pairwise Add
2475 def VPADDi8 : N3VDInt<0, 0, 0b00, 0b1011, 1, IIC_VBINiD, "vpadd", "i8",
2476 v8i8, v8i8, int_arm_neon_vpadd, 0>;
2477 def VPADDi16 : N3VDInt<0, 0, 0b01, 0b1011, 1, IIC_VBINiD, "vpadd", "i16",
2478 v4i16, v4i16, int_arm_neon_vpadd, 0>;
2479 def VPADDi32 : N3VDInt<0, 0, 0b10, 0b1011, 1, IIC_VBINiD, "vpadd", "i32",
2480 v2i32, v2i32, int_arm_neon_vpadd, 0>;
2481 def VPADDf : N3VDInt<1, 0, 0b00, 0b1101, 0, IIC_VBIND, "vpadd", "f32",
2482 v2f32, v2f32, int_arm_neon_vpadd, 0>;
2484 // VPADDL : Vector Pairwise Add Long
2485 defm VPADDLs : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00100, 0, "vpaddl", "s",
2486 int_arm_neon_vpaddls>;
2487 defm VPADDLu : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00101, 0, "vpaddl", "u",
2488 int_arm_neon_vpaddlu>;
2490 // VPADAL : Vector Pairwise Add and Accumulate Long
2491 defm VPADALs : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01100, 0, "vpadal", "s",
2492 int_arm_neon_vpadals>;
2493 defm VPADALu : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01101, 0, "vpadal", "u",
2494 int_arm_neon_vpadalu>;
2496 // VPMAX : Vector Pairwise Maximum
2497 def VPMAXs8 : N3VDInt<0, 0, 0b00, 0b1010, 0, IIC_VBINi4D, "vpmax", "s8",
2498 v8i8, v8i8, int_arm_neon_vpmaxs, 0>;
2499 def VPMAXs16 : N3VDInt<0, 0, 0b01, 0b1010, 0, IIC_VBINi4D, "vpmax", "s16",
2500 v4i16, v4i16, int_arm_neon_vpmaxs, 0>;
2501 def VPMAXs32 : N3VDInt<0, 0, 0b10, 0b1010, 0, IIC_VBINi4D, "vpmax", "s32",
2502 v2i32, v2i32, int_arm_neon_vpmaxs, 0>;
2503 def VPMAXu8 : N3VDInt<1, 0, 0b00, 0b1010, 0, IIC_VBINi4D, "vpmax", "u8",
2504 v8i8, v8i8, int_arm_neon_vpmaxu, 0>;
2505 def VPMAXu16 : N3VDInt<1, 0, 0b01, 0b1010, 0, IIC_VBINi4D, "vpmax", "u16",
2506 v4i16, v4i16, int_arm_neon_vpmaxu, 0>;
2507 def VPMAXu32 : N3VDInt<1, 0, 0b10, 0b1010, 0, IIC_VBINi4D, "vpmax", "u32",
2508 v2i32, v2i32, int_arm_neon_vpmaxu, 0>;
2509 def VPMAXf : N3VDInt<1, 0, 0b00, 0b1111, 0, IIC_VBINi4D, "vpmax", "f32",
2510 v2f32, v2f32, int_arm_neon_vpmaxs, 0>;
2512 // VPMIN : Vector Pairwise Minimum
2513 def VPMINs8 : N3VDInt<0, 0, 0b00, 0b1010, 1, IIC_VBINi4D, "vpmin", "s8",
2514 v8i8, v8i8, int_arm_neon_vpmins, 0>;
2515 def VPMINs16 : N3VDInt<0, 0, 0b01, 0b1010, 1, IIC_VBINi4D, "vpmin", "s16",
2516 v4i16, v4i16, int_arm_neon_vpmins, 0>;
2517 def VPMINs32 : N3VDInt<0, 0, 0b10, 0b1010, 1, IIC_VBINi4D, "vpmin", "s32",
2518 v2i32, v2i32, int_arm_neon_vpmins, 0>;
2519 def VPMINu8 : N3VDInt<1, 0, 0b00, 0b1010, 1, IIC_VBINi4D, "vpmin", "u8",
2520 v8i8, v8i8, int_arm_neon_vpminu, 0>;
2521 def VPMINu16 : N3VDInt<1, 0, 0b01, 0b1010, 1, IIC_VBINi4D, "vpmin", "u16",
2522 v4i16, v4i16, int_arm_neon_vpminu, 0>;
2523 def VPMINu32 : N3VDInt<1, 0, 0b10, 0b1010, 1, IIC_VBINi4D, "vpmin", "u32",
2524 v2i32, v2i32, int_arm_neon_vpminu, 0>;
2525 def VPMINf : N3VDInt<1, 0, 0b10, 0b1111, 0, IIC_VBINi4D, "vpmin", "f32",
2526 v2f32, v2f32, int_arm_neon_vpmins, 0>;
2528 // Vector Reciprocal and Reciprocal Square Root Estimate and Step.
2530 // VRECPE : Vector Reciprocal Estimate
2531 def VRECPEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
2532 IIC_VUNAD, "vrecpe", "u32",
2533 v2i32, v2i32, int_arm_neon_vrecpe>;
2534 def VRECPEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
2535 IIC_VUNAQ, "vrecpe", "u32",
2536 v4i32, v4i32, int_arm_neon_vrecpe>;
2537 def VRECPEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
2538 IIC_VUNAD, "vrecpe", "f32",
2539 v2f32, v2f32, int_arm_neon_vrecpe>;
2540 def VRECPEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
2541 IIC_VUNAQ, "vrecpe", "f32",
2542 v4f32, v4f32, int_arm_neon_vrecpe>;
2544 // VRECPS : Vector Reciprocal Step
2545 def VRECPSfd : N3VDInt<0, 0, 0b00, 0b1111, 1,
2546 IIC_VRECSD, "vrecps", "f32",
2547 v2f32, v2f32, int_arm_neon_vrecps, 1>;
2548 def VRECPSfq : N3VQInt<0, 0, 0b00, 0b1111, 1,
2549 IIC_VRECSQ, "vrecps", "f32",
2550 v4f32, v4f32, int_arm_neon_vrecps, 1>;
2552 // VRSQRTE : Vector Reciprocal Square Root Estimate
2553 def VRSQRTEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
2554 IIC_VUNAD, "vrsqrte", "u32",
2555 v2i32, v2i32, int_arm_neon_vrsqrte>;
2556 def VRSQRTEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
2557 IIC_VUNAQ, "vrsqrte", "u32",
2558 v4i32, v4i32, int_arm_neon_vrsqrte>;
2559 def VRSQRTEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
2560 IIC_VUNAD, "vrsqrte", "f32",
2561 v2f32, v2f32, int_arm_neon_vrsqrte>;
2562 def VRSQRTEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
2563 IIC_VUNAQ, "vrsqrte", "f32",
2564 v4f32, v4f32, int_arm_neon_vrsqrte>;
2566 // VRSQRTS : Vector Reciprocal Square Root Step
2567 def VRSQRTSfd : N3VDInt<0, 0, 0b10, 0b1111, 1,
2568 IIC_VRECSD, "vrsqrts", "f32",
2569 v2f32, v2f32, int_arm_neon_vrsqrts, 1>;
2570 def VRSQRTSfq : N3VQInt<0, 0, 0b10, 0b1111, 1,
2571 IIC_VRECSQ, "vrsqrts", "f32",
2572 v4f32, v4f32, int_arm_neon_vrsqrts, 1>;
2576 // VSHL : Vector Shift
2577 defm VSHLs : N3VInt_QHSD<0, 0, 0b0100, 0, IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ,
2578 IIC_VSHLiQ, "vshl", "s", int_arm_neon_vshifts, 0>;
2579 defm VSHLu : N3VInt_QHSD<1, 0, 0b0100, 0, IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ,
2580 IIC_VSHLiQ, "vshl", "u", int_arm_neon_vshiftu, 0>;
2581 // VSHL : Vector Shift Left (Immediate)
2582 defm VSHLi : N2VSh_QHSD<0, 1, 0b0101, 1, IIC_VSHLiD, "vshl", "i", NEONvshl>;
2583 // VSHR : Vector Shift Right (Immediate)
2584 defm VSHRs : N2VSh_QHSD<0, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "s", NEONvshrs>;
2585 defm VSHRu : N2VSh_QHSD<1, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "u", NEONvshru>;
2587 // VSHLL : Vector Shift Left Long
2588 defm VSHLLs : N2VLSh_QHS<0, 1, 0b1010, 0, 0, 1, "vshll", "s", NEONvshlls>;
2589 defm VSHLLu : N2VLSh_QHS<1, 1, 0b1010, 0, 0, 1, "vshll", "u", NEONvshllu>;
2591 // VSHLL : Vector Shift Left Long (with maximum shift count)
2592 class N2VLShMax<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
2593 bit op6, bit op4, string OpcodeStr, string Dt, ValueType ResTy,
2594 ValueType OpTy, SDNode OpNode>
2595 : N2VLSh<op24, op23, op11_8, op7, op6, op4, OpcodeStr, Dt,
2596 ResTy, OpTy, OpNode> {
2597 let Inst{21-16} = op21_16;
2599 def VSHLLi8 : N2VLShMax<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll", "i8",
2600 v8i16, v8i8, NEONvshlli>;
2601 def VSHLLi16 : N2VLShMax<1, 1, 0b110110, 0b0011, 0, 0, 0, "vshll", "i16",
2602 v4i32, v4i16, NEONvshlli>;
2603 def VSHLLi32 : N2VLShMax<1, 1, 0b111010, 0b0011, 0, 0, 0, "vshll", "i32",
2604 v2i64, v2i32, NEONvshlli>;
2606 // VSHRN : Vector Shift Right and Narrow
2607 defm VSHRN : N2VNSh_HSD<0,1,0b1000,0,0,1, IIC_VSHLiD, "vshrn", "i",
2610 // VRSHL : Vector Rounding Shift
2611 defm VRSHLs : N3VInt_QHSD<0,0,0b0101,0, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2612 IIC_VSHLi4Q, "vrshl", "s", int_arm_neon_vrshifts,0>;
2613 defm VRSHLu : N3VInt_QHSD<1,0,0b0101,0, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2614 IIC_VSHLi4Q, "vrshl", "u", int_arm_neon_vrshiftu,0>;
2615 // VRSHR : Vector Rounding Shift Right
2616 defm VRSHRs : N2VSh_QHSD<0,1,0b0010,1, IIC_VSHLi4D, "vrshr", "s", NEONvrshrs>;
2617 defm VRSHRu : N2VSh_QHSD<1,1,0b0010,1, IIC_VSHLi4D, "vrshr", "u", NEONvrshru>;
2619 // VRSHRN : Vector Rounding Shift Right and Narrow
2620 defm VRSHRN : N2VNSh_HSD<0, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vrshrn", "i",
2623 // VQSHL : Vector Saturating Shift
2624 defm VQSHLs : N3VInt_QHSD<0,0,0b0100,1, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2625 IIC_VSHLi4Q, "vqshl", "s", int_arm_neon_vqshifts,0>;
2626 defm VQSHLu : N3VInt_QHSD<1,0,0b0100,1, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2627 IIC_VSHLi4Q, "vqshl", "u", int_arm_neon_vqshiftu,0>;
2628 // VQSHL : Vector Saturating Shift Left (Immediate)
2629 defm VQSHLsi : N2VSh_QHSD<0,1,0b0111,1, IIC_VSHLi4D, "vqshl", "s", NEONvqshls>;
2630 defm VQSHLui : N2VSh_QHSD<1,1,0b0111,1, IIC_VSHLi4D, "vqshl", "u", NEONvqshlu>;
2631 // VQSHLU : Vector Saturating Shift Left (Immediate, Unsigned)
2632 defm VQSHLsu : N2VSh_QHSD<1,1,0b0110,1, IIC_VSHLi4D, "vqshlu","s",NEONvqshlsu>;
2634 // VQSHRN : Vector Saturating Shift Right and Narrow
2635 defm VQSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "s",
2637 defm VQSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "u",
2640 // VQSHRUN : Vector Saturating Shift Right and Narrow (Unsigned)
2641 defm VQSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 0, 1, IIC_VSHLi4D, "vqshrun", "s",
2644 // VQRSHL : Vector Saturating Rounding Shift
2645 defm VQRSHLs : N3VInt_QHSD<0,0,0b0101,1, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2646 IIC_VSHLi4Q, "vqrshl", "s",
2647 int_arm_neon_vqrshifts, 0>;
2648 defm VQRSHLu : N3VInt_QHSD<1,0,0b0101,1, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2649 IIC_VSHLi4Q, "vqrshl", "u",
2650 int_arm_neon_vqrshiftu, 0>;
2652 // VQRSHRN : Vector Saturating Rounding Shift Right and Narrow
2653 defm VQRSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "s",
2655 defm VQRSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "u",
2658 // VQRSHRUN : Vector Saturating Rounding Shift Right and Narrow (Unsigned)
2659 defm VQRSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vqrshrun", "s",
2662 // VSRA : Vector Shift Right and Accumulate
2663 defm VSRAs : N2VShAdd_QHSD<0, 1, 0b0001, 1, "vsra", "s", NEONvshrs>;
2664 defm VSRAu : N2VShAdd_QHSD<1, 1, 0b0001, 1, "vsra", "u", NEONvshru>;
2665 // VRSRA : Vector Rounding Shift Right and Accumulate
2666 defm VRSRAs : N2VShAdd_QHSD<0, 1, 0b0011, 1, "vrsra", "s", NEONvrshrs>;
2667 defm VRSRAu : N2VShAdd_QHSD<1, 1, 0b0011, 1, "vrsra", "u", NEONvrshru>;
2669 // VSLI : Vector Shift Left and Insert
2670 defm VSLI : N2VShIns_QHSD<1, 1, 0b0101, 1, "vsli", NEONvsli>;
2671 // VSRI : Vector Shift Right and Insert
2672 defm VSRI : N2VShIns_QHSD<1, 1, 0b0100, 1, "vsri", NEONvsri>;
2674 // Vector Absolute and Saturating Absolute.
2676 // VABS : Vector Absolute Value
2677 defm VABS : N2VInt_QHS<0b11, 0b11, 0b01, 0b00110, 0,
2678 IIC_VUNAiD, IIC_VUNAiQ, "vabs", "s",
2680 def VABSfd : N2VDInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
2681 IIC_VUNAD, "vabs", "f32",
2682 v2f32, v2f32, int_arm_neon_vabs>;
2683 def VABSfq : N2VQInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
2684 IIC_VUNAQ, "vabs", "f32",
2685 v4f32, v4f32, int_arm_neon_vabs>;
2687 // VQABS : Vector Saturating Absolute Value
2688 defm VQABS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01110, 0,
2689 IIC_VQUNAiD, IIC_VQUNAiQ, "vqabs", "s",
2690 int_arm_neon_vqabs>;
2694 def vneg : PatFrag<(ops node:$in), (sub immAllZerosV, node:$in)>;
2695 def vneg_conv : PatFrag<(ops node:$in), (sub immAllZerosV_bc, node:$in)>;
2697 class VNEGD<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
2698 : N2V<0b11, 0b11, size, 0b01, 0b00111, 0, 0, (outs DPR:$dst), (ins DPR:$src),
2699 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
2700 [(set DPR:$dst, (Ty (vneg DPR:$src)))]>;
2701 class VNEGQ<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
2702 : N2V<0b11, 0b11, size, 0b01, 0b00111, 1, 0, (outs QPR:$dst), (ins QPR:$src),
2703 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
2704 [(set QPR:$dst, (Ty (vneg QPR:$src)))]>;
2706 // VNEG : Vector Negate
2707 def VNEGs8d : VNEGD<0b00, "vneg", "s8", v8i8>;
2708 def VNEGs16d : VNEGD<0b01, "vneg", "s16", v4i16>;
2709 def VNEGs32d : VNEGD<0b10, "vneg", "s32", v2i32>;
2710 def VNEGs8q : VNEGQ<0b00, "vneg", "s8", v16i8>;
2711 def VNEGs16q : VNEGQ<0b01, "vneg", "s16", v8i16>;
2712 def VNEGs32q : VNEGQ<0b10, "vneg", "s32", v4i32>;
2714 // VNEG : Vector Negate (floating-point)
2715 def VNEGfd : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
2716 (outs DPR:$dst), (ins DPR:$src), IIC_VUNAD,
2717 "vneg", "f32", "$dst, $src", "",
2718 [(set DPR:$dst, (v2f32 (fneg DPR:$src)))]>;
2719 def VNEGf32q : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 1, 0,
2720 (outs QPR:$dst), (ins QPR:$src), IIC_VUNAQ,
2721 "vneg", "f32", "$dst, $src", "",
2722 [(set QPR:$dst, (v4f32 (fneg QPR:$src)))]>;
2724 def : Pat<(v8i8 (vneg_conv DPR:$src)), (VNEGs8d DPR:$src)>;
2725 def : Pat<(v4i16 (vneg_conv DPR:$src)), (VNEGs16d DPR:$src)>;
2726 def : Pat<(v2i32 (vneg_conv DPR:$src)), (VNEGs32d DPR:$src)>;
2727 def : Pat<(v16i8 (vneg_conv QPR:$src)), (VNEGs8q QPR:$src)>;
2728 def : Pat<(v8i16 (vneg_conv QPR:$src)), (VNEGs16q QPR:$src)>;
2729 def : Pat<(v4i32 (vneg_conv QPR:$src)), (VNEGs32q QPR:$src)>;
2731 // VQNEG : Vector Saturating Negate
2732 defm VQNEG : N2VInt_QHS<0b11, 0b11, 0b00, 0b01111, 0,
2733 IIC_VQUNAiD, IIC_VQUNAiQ, "vqneg", "s",
2734 int_arm_neon_vqneg>;
2736 // Vector Bit Counting Operations.
2738 // VCLS : Vector Count Leading Sign Bits
2739 defm VCLS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01000, 0,
2740 IIC_VCNTiD, IIC_VCNTiQ, "vcls", "s",
2742 // VCLZ : Vector Count Leading Zeros
2743 defm VCLZ : N2VInt_QHS<0b11, 0b11, 0b00, 0b01001, 0,
2744 IIC_VCNTiD, IIC_VCNTiQ, "vclz", "i",
2746 // VCNT : Vector Count One Bits
2747 def VCNTd : N2VDInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
2748 IIC_VCNTiD, "vcnt", "8",
2749 v8i8, v8i8, int_arm_neon_vcnt>;
2750 def VCNTq : N2VQInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
2751 IIC_VCNTiQ, "vcnt", "8",
2752 v16i8, v16i8, int_arm_neon_vcnt>;
2754 // Vector Swap -- for disassembly only.
2755 def VSWPd : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 0, 0,
2756 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
2757 "vswp", "$dst, $src", "", []>;
2758 def VSWPq : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 1, 0,
2759 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
2760 "vswp", "$dst, $src", "", []>;
2762 // Vector Move Operations.
2764 // VMOV : Vector Move (Register)
2766 def VMOVDneon: N3VX<0, 0, 0b10, 0b0001, 0, 1, (outs DPR:$dst), (ins DPR:$src),
2767 IIC_VMOVD, "vmov", "$dst, $src", "", []>;
2768 def VMOVQ : N3VX<0, 0, 0b10, 0b0001, 1, 1, (outs QPR:$dst), (ins QPR:$src),
2769 IIC_VMOVD, "vmov", "$dst, $src", "", []>;
2771 // VMOV : Vector Move (Immediate)
2773 // VMOV_get_imm8 xform function: convert build_vector to VMOV.i8 imm.
2774 def VMOV_get_imm8 : SDNodeXForm<build_vector, [{
2775 return ARM::getVMOVImm(N, 1, *CurDAG);
2777 def vmovImm8 : PatLeaf<(build_vector), [{
2778 return ARM::getVMOVImm(N, 1, *CurDAG).getNode() != 0;
2781 // VMOV_get_imm16 xform function: convert build_vector to VMOV.i16 imm.
2782 def VMOV_get_imm16 : SDNodeXForm<build_vector, [{
2783 return ARM::getVMOVImm(N, 2, *CurDAG);
2785 def vmovImm16 : PatLeaf<(build_vector), [{
2786 return ARM::getVMOVImm(N, 2, *CurDAG).getNode() != 0;
2787 }], VMOV_get_imm16>;
2789 // VMOV_get_imm32 xform function: convert build_vector to VMOV.i32 imm.
2790 def VMOV_get_imm32 : SDNodeXForm<build_vector, [{
2791 return ARM::getVMOVImm(N, 4, *CurDAG);
2793 def vmovImm32 : PatLeaf<(build_vector), [{
2794 return ARM::getVMOVImm(N, 4, *CurDAG).getNode() != 0;
2795 }], VMOV_get_imm32>;
2797 // VMOV_get_imm64 xform function: convert build_vector to VMOV.i64 imm.
2798 def VMOV_get_imm64 : SDNodeXForm<build_vector, [{
2799 return ARM::getVMOVImm(N, 8, *CurDAG);
2801 def vmovImm64 : PatLeaf<(build_vector), [{
2802 return ARM::getVMOVImm(N, 8, *CurDAG).getNode() != 0;
2803 }], VMOV_get_imm64>;
2805 // Note: Some of the cmode bits in the following VMOV instructions need to
2806 // be encoded based on the immed values.
2808 def VMOVv8i8 : N1ModImm<1, 0b000, 0b1110, 0, 0, 0, 1, (outs DPR:$dst),
2809 (ins h8imm:$SIMM), IIC_VMOVImm,
2810 "vmov", "i8", "$dst, $SIMM", "",
2811 [(set DPR:$dst, (v8i8 vmovImm8:$SIMM))]>;
2812 def VMOVv16i8 : N1ModImm<1, 0b000, 0b1110, 0, 1, 0, 1, (outs QPR:$dst),
2813 (ins h8imm:$SIMM), IIC_VMOVImm,
2814 "vmov", "i8", "$dst, $SIMM", "",
2815 [(set QPR:$dst, (v16i8 vmovImm8:$SIMM))]>;
2817 def VMOVv4i16 : N1ModImm<1, 0b000, {1,0,?,?}, 0, 0, {?}, 1, (outs DPR:$dst),
2818 (ins h16imm:$SIMM), IIC_VMOVImm,
2819 "vmov", "i16", "$dst, $SIMM", "",
2820 [(set DPR:$dst, (v4i16 vmovImm16:$SIMM))]>;
2821 def VMOVv8i16 : N1ModImm<1, 0b000, {1,0,?,?}, 0, 1, {?}, 1, (outs QPR:$dst),
2822 (ins h16imm:$SIMM), IIC_VMOVImm,
2823 "vmov", "i16", "$dst, $SIMM", "",
2824 [(set QPR:$dst, (v8i16 vmovImm16:$SIMM))]>;
2826 def VMOVv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, {?}, 1, (outs DPR:$dst),
2827 (ins h32imm:$SIMM), IIC_VMOVImm,
2828 "vmov", "i32", "$dst, $SIMM", "",
2829 [(set DPR:$dst, (v2i32 vmovImm32:$SIMM))]>;
2830 def VMOVv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, {?}, 1, (outs QPR:$dst),
2831 (ins h32imm:$SIMM), IIC_VMOVImm,
2832 "vmov", "i32", "$dst, $SIMM", "",
2833 [(set QPR:$dst, (v4i32 vmovImm32:$SIMM))]>;
2835 def VMOVv1i64 : N1ModImm<1, 0b000, 0b1110, 0, 0, 1, 1, (outs DPR:$dst),
2836 (ins h64imm:$SIMM), IIC_VMOVImm,
2837 "vmov", "i64", "$dst, $SIMM", "",
2838 [(set DPR:$dst, (v1i64 vmovImm64:$SIMM))]>;
2839 def VMOVv2i64 : N1ModImm<1, 0b000, 0b1110, 0, 1, 1, 1, (outs QPR:$dst),
2840 (ins h64imm:$SIMM), IIC_VMOVImm,
2841 "vmov", "i64", "$dst, $SIMM", "",
2842 [(set QPR:$dst, (v2i64 vmovImm64:$SIMM))]>;
2844 // VMOV : Vector Get Lane (move scalar to ARM core register)
2846 def VGETLNs8 : NVGetLane<{1,1,1,0,0,1,?,1}, 0b1011, {?,?},
2847 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2848 IIC_VMOVSI, "vmov", "s8", "$dst, $src[$lane]",
2849 [(set GPR:$dst, (NEONvgetlanes (v8i8 DPR:$src),
2851 def VGETLNs16 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, {?,1},
2852 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2853 IIC_VMOVSI, "vmov", "s16", "$dst, $src[$lane]",
2854 [(set GPR:$dst, (NEONvgetlanes (v4i16 DPR:$src),
2856 def VGETLNu8 : NVGetLane<{1,1,1,0,1,1,?,1}, 0b1011, {?,?},
2857 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2858 IIC_VMOVSI, "vmov", "u8", "$dst, $src[$lane]",
2859 [(set GPR:$dst, (NEONvgetlaneu (v8i8 DPR:$src),
2861 def VGETLNu16 : NVGetLane<{1,1,1,0,1,0,?,1}, 0b1011, {?,1},
2862 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2863 IIC_VMOVSI, "vmov", "u16", "$dst, $src[$lane]",
2864 [(set GPR:$dst, (NEONvgetlaneu (v4i16 DPR:$src),
2866 def VGETLNi32 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, 0b00,
2867 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2868 IIC_VMOVSI, "vmov", "32", "$dst, $src[$lane]",
2869 [(set GPR:$dst, (extractelt (v2i32 DPR:$src),
2871 // def VGETLNf32: see FMRDH and FMRDL in ARMInstrVFP.td
2872 def : Pat<(NEONvgetlanes (v16i8 QPR:$src), imm:$lane),
2873 (VGETLNs8 (v8i8 (EXTRACT_SUBREG QPR:$src,
2874 (DSubReg_i8_reg imm:$lane))),
2875 (SubReg_i8_lane imm:$lane))>;
2876 def : Pat<(NEONvgetlanes (v8i16 QPR:$src), imm:$lane),
2877 (VGETLNs16 (v4i16 (EXTRACT_SUBREG QPR:$src,
2878 (DSubReg_i16_reg imm:$lane))),
2879 (SubReg_i16_lane imm:$lane))>;
2880 def : Pat<(NEONvgetlaneu (v16i8 QPR:$src), imm:$lane),
2881 (VGETLNu8 (v8i8 (EXTRACT_SUBREG QPR:$src,
2882 (DSubReg_i8_reg imm:$lane))),
2883 (SubReg_i8_lane imm:$lane))>;
2884 def : Pat<(NEONvgetlaneu (v8i16 QPR:$src), imm:$lane),
2885 (VGETLNu16 (v4i16 (EXTRACT_SUBREG QPR:$src,
2886 (DSubReg_i16_reg imm:$lane))),
2887 (SubReg_i16_lane imm:$lane))>;
2888 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
2889 (VGETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src,
2890 (DSubReg_i32_reg imm:$lane))),
2891 (SubReg_i32_lane imm:$lane))>;
2892 def : Pat<(extractelt (v2f32 DPR:$src1), imm:$src2),
2893 (EXTRACT_SUBREG (v2f32 (COPY_TO_REGCLASS (v2f32 DPR:$src1),DPR_VFP2)),
2894 (SSubReg_f32_reg imm:$src2))>;
2895 def : Pat<(extractelt (v4f32 QPR:$src1), imm:$src2),
2896 (EXTRACT_SUBREG (v4f32 (COPY_TO_REGCLASS (v4f32 QPR:$src1),QPR_VFP2)),
2897 (SSubReg_f32_reg imm:$src2))>;
2898 //def : Pat<(extractelt (v2i64 QPR:$src1), imm:$src2),
2899 // (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
2900 def : Pat<(extractelt (v2f64 QPR:$src1), imm:$src2),
2901 (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
2904 // VMOV : Vector Set Lane (move ARM core register to scalar)
2906 let Constraints = "$src1 = $dst" in {
2907 def VSETLNi8 : NVSetLane<{1,1,1,0,0,1,?,0}, 0b1011, {?,?}, (outs DPR:$dst),
2908 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
2909 IIC_VMOVISL, "vmov", "8", "$dst[$lane], $src2",
2910 [(set DPR:$dst, (vector_insert (v8i8 DPR:$src1),
2911 GPR:$src2, imm:$lane))]>;
2912 def VSETLNi16 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, {?,1}, (outs DPR:$dst),
2913 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
2914 IIC_VMOVISL, "vmov", "16", "$dst[$lane], $src2",
2915 [(set DPR:$dst, (vector_insert (v4i16 DPR:$src1),
2916 GPR:$src2, imm:$lane))]>;
2917 def VSETLNi32 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, 0b00, (outs DPR:$dst),
2918 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
2919 IIC_VMOVISL, "vmov", "32", "$dst[$lane], $src2",
2920 [(set DPR:$dst, (insertelt (v2i32 DPR:$src1),
2921 GPR:$src2, imm:$lane))]>;
2923 def : Pat<(vector_insert (v16i8 QPR:$src1), GPR:$src2, imm:$lane),
2924 (v16i8 (INSERT_SUBREG QPR:$src1,
2925 (v8i8 (VSETLNi8 (v8i8 (EXTRACT_SUBREG QPR:$src1,
2926 (DSubReg_i8_reg imm:$lane))),
2927 GPR:$src2, (SubReg_i8_lane imm:$lane))),
2928 (DSubReg_i8_reg imm:$lane)))>;
2929 def : Pat<(vector_insert (v8i16 QPR:$src1), GPR:$src2, imm:$lane),
2930 (v8i16 (INSERT_SUBREG QPR:$src1,
2931 (v4i16 (VSETLNi16 (v4i16 (EXTRACT_SUBREG QPR:$src1,
2932 (DSubReg_i16_reg imm:$lane))),
2933 GPR:$src2, (SubReg_i16_lane imm:$lane))),
2934 (DSubReg_i16_reg imm:$lane)))>;
2935 def : Pat<(insertelt (v4i32 QPR:$src1), GPR:$src2, imm:$lane),
2936 (v4i32 (INSERT_SUBREG QPR:$src1,
2937 (v2i32 (VSETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src1,
2938 (DSubReg_i32_reg imm:$lane))),
2939 GPR:$src2, (SubReg_i32_lane imm:$lane))),
2940 (DSubReg_i32_reg imm:$lane)))>;
2942 def : Pat<(v2f32 (insertelt DPR:$src1, SPR:$src2, imm:$src3)),
2943 (INSERT_SUBREG (v2f32 (COPY_TO_REGCLASS DPR:$src1, DPR_VFP2)),
2944 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
2945 def : Pat<(v4f32 (insertelt QPR:$src1, SPR:$src2, imm:$src3)),
2946 (INSERT_SUBREG (v4f32 (COPY_TO_REGCLASS QPR:$src1, QPR_VFP2)),
2947 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
2949 //def : Pat<(v2i64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
2950 // (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
2951 def : Pat<(v2f64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
2952 (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
2954 def : Pat<(v2f32 (scalar_to_vector SPR:$src)),
2955 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$src, arm_ssubreg_0)>;
2956 def : Pat<(v2f64 (scalar_to_vector (f64 DPR:$src))),
2957 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), DPR:$src, arm_dsubreg_0)>;
2958 def : Pat<(v4f32 (scalar_to_vector SPR:$src)),
2959 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), SPR:$src, arm_ssubreg_0)>;
2961 def : Pat<(v8i8 (scalar_to_vector GPR:$src)),
2962 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
2963 def : Pat<(v4i16 (scalar_to_vector GPR:$src)),
2964 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
2965 def : Pat<(v2i32 (scalar_to_vector GPR:$src)),
2966 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
2968 def : Pat<(v16i8 (scalar_to_vector GPR:$src)),
2969 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
2970 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
2972 def : Pat<(v8i16 (scalar_to_vector GPR:$src)),
2973 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
2974 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
2976 def : Pat<(v4i32 (scalar_to_vector GPR:$src)),
2977 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
2978 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
2981 // VDUP : Vector Duplicate (from ARM core register to all elements)
2983 class VDUPD<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
2984 : NVDup<opcod1, 0b1011, opcod3, (outs DPR:$dst), (ins GPR:$src),
2985 IIC_VMOVIS, "vdup", Dt, "$dst, $src",
2986 [(set DPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
2987 class VDUPQ<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
2988 : NVDup<opcod1, 0b1011, opcod3, (outs QPR:$dst), (ins GPR:$src),
2989 IIC_VMOVIS, "vdup", Dt, "$dst, $src",
2990 [(set QPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
2992 def VDUP8d : VDUPD<0b11101100, 0b00, "8", v8i8>;
2993 def VDUP16d : VDUPD<0b11101000, 0b01, "16", v4i16>;
2994 def VDUP32d : VDUPD<0b11101000, 0b00, "32", v2i32>;
2995 def VDUP8q : VDUPQ<0b11101110, 0b00, "8", v16i8>;
2996 def VDUP16q : VDUPQ<0b11101010, 0b01, "16", v8i16>;
2997 def VDUP32q : VDUPQ<0b11101010, 0b00, "32", v4i32>;
2999 def VDUPfd : NVDup<0b11101000, 0b1011, 0b00, (outs DPR:$dst), (ins GPR:$src),
3000 IIC_VMOVIS, "vdup", "32", "$dst, $src",
3001 [(set DPR:$dst, (v2f32 (NEONvdup
3002 (f32 (bitconvert GPR:$src)))))]>;
3003 def VDUPfq : NVDup<0b11101010, 0b1011, 0b00, (outs QPR:$dst), (ins GPR:$src),
3004 IIC_VMOVIS, "vdup", "32", "$dst, $src",
3005 [(set QPR:$dst, (v4f32 (NEONvdup
3006 (f32 (bitconvert GPR:$src)))))]>;
3008 // VDUP : Vector Duplicate Lane (from scalar to all elements)
3010 class VDUPLND<bits<2> op19_18, bits<2> op17_16,
3011 string OpcodeStr, string Dt, ValueType Ty>
3012 : N2V<0b11, 0b11, op19_18, op17_16, 0b11000, 0, 0,
3013 (outs DPR:$dst), (ins DPR:$src, nohash_imm:$lane), IIC_VMOVD,
3014 OpcodeStr, Dt, "$dst, $src[$lane]", "",
3015 [(set DPR:$dst, (Ty (NEONvduplane (Ty DPR:$src), imm:$lane)))]>;
3017 class VDUPLNQ<bits<2> op19_18, bits<2> op17_16, string OpcodeStr, string Dt,
3018 ValueType ResTy, ValueType OpTy>
3019 : N2V<0b11, 0b11, op19_18, op17_16, 0b11000, 1, 0,
3020 (outs QPR:$dst), (ins DPR:$src, nohash_imm:$lane), IIC_VMOVD,
3021 OpcodeStr, Dt, "$dst, $src[$lane]", "",
3022 [(set QPR:$dst, (ResTy (NEONvduplane (OpTy DPR:$src), imm:$lane)))]>;
3024 // Inst{19-16} is partially specified depending on the element size.
3026 def VDUPLN8d : VDUPLND<{?,?}, {?,1}, "vdup", "8", v8i8>;
3027 def VDUPLN16d : VDUPLND<{?,?}, {1,0}, "vdup", "16", v4i16>;
3028 def VDUPLN32d : VDUPLND<{?,1}, {0,0}, "vdup", "32", v2i32>;
3029 def VDUPLNfd : VDUPLND<{?,1}, {0,0}, "vdup", "32", v2f32>;
3030 def VDUPLN8q : VDUPLNQ<{?,?}, {?,1}, "vdup", "8", v16i8, v8i8>;
3031 def VDUPLN16q : VDUPLNQ<{?,?}, {1,0}, "vdup", "16", v8i16, v4i16>;
3032 def VDUPLN32q : VDUPLNQ<{?,1}, {0,0}, "vdup", "32", v4i32, v2i32>;
3033 def VDUPLNfq : VDUPLNQ<{?,1}, {0,0}, "vdup", "32", v4f32, v2f32>;
3035 def : Pat<(v16i8 (NEONvduplane (v16i8 QPR:$src), imm:$lane)),
3036 (v16i8 (VDUPLN8q (v8i8 (EXTRACT_SUBREG QPR:$src,
3037 (DSubReg_i8_reg imm:$lane))),
3038 (SubReg_i8_lane imm:$lane)))>;
3039 def : Pat<(v8i16 (NEONvduplane (v8i16 QPR:$src), imm:$lane)),
3040 (v8i16 (VDUPLN16q (v4i16 (EXTRACT_SUBREG QPR:$src,
3041 (DSubReg_i16_reg imm:$lane))),
3042 (SubReg_i16_lane imm:$lane)))>;
3043 def : Pat<(v4i32 (NEONvduplane (v4i32 QPR:$src), imm:$lane)),
3044 (v4i32 (VDUPLN32q (v2i32 (EXTRACT_SUBREG QPR:$src,
3045 (DSubReg_i32_reg imm:$lane))),
3046 (SubReg_i32_lane imm:$lane)))>;
3047 def : Pat<(v4f32 (NEONvduplane (v4f32 QPR:$src), imm:$lane)),
3048 (v4f32 (VDUPLNfq (v2f32 (EXTRACT_SUBREG QPR:$src,
3049 (DSubReg_i32_reg imm:$lane))),
3050 (SubReg_i32_lane imm:$lane)))>;
3052 def VDUPfdf : N2V<0b11, 0b11, {?,1}, {0,0}, 0b11000, 0, 0,
3053 (outs DPR:$dst), (ins SPR:$src),
3054 IIC_VMOVD, "vdup", "32", "$dst, ${src:lane}", "",
3055 [(set DPR:$dst, (v2f32 (NEONvdup (f32 SPR:$src))))]>;
3057 def VDUPfqf : N2V<0b11, 0b11, {?,1}, {0,0}, 0b11000, 1, 0,
3058 (outs QPR:$dst), (ins SPR:$src),
3059 IIC_VMOVD, "vdup", "32", "$dst, ${src:lane}", "",
3060 [(set QPR:$dst, (v4f32 (NEONvdup (f32 SPR:$src))))]>;
3062 def : Pat<(v2i64 (NEONvduplane (v2i64 QPR:$src), imm:$lane)),
3063 (INSERT_SUBREG QPR:$src,
3064 (i64 (EXTRACT_SUBREG QPR:$src,
3065 (DSubReg_f64_reg imm:$lane))),
3066 (DSubReg_f64_other_reg imm:$lane))>;
3067 def : Pat<(v2f64 (NEONvduplane (v2f64 QPR:$src), imm:$lane)),
3068 (INSERT_SUBREG QPR:$src,
3069 (f64 (EXTRACT_SUBREG QPR:$src,
3070 (DSubReg_f64_reg imm:$lane))),
3071 (DSubReg_f64_other_reg imm:$lane))>;
3073 // VMOVN : Vector Narrowing Move
3074 defm VMOVN : N2VNInt_HSD<0b11,0b11,0b10,0b00100,0,0, IIC_VMOVD,
3075 "vmovn", "i", int_arm_neon_vmovn>;
3076 // VQMOVN : Vector Saturating Narrowing Move
3077 defm VQMOVNs : N2VNInt_HSD<0b11,0b11,0b10,0b00101,0,0, IIC_VQUNAiD,
3078 "vqmovn", "s", int_arm_neon_vqmovns>;
3079 defm VQMOVNu : N2VNInt_HSD<0b11,0b11,0b10,0b00101,1,0, IIC_VQUNAiD,
3080 "vqmovn", "u", int_arm_neon_vqmovnu>;
3081 defm VQMOVNsu : N2VNInt_HSD<0b11,0b11,0b10,0b00100,1,0, IIC_VQUNAiD,
3082 "vqmovun", "s", int_arm_neon_vqmovnsu>;
3083 // VMOVL : Vector Lengthening Move
3084 defm VMOVLs : N2VLInt_QHS<0b01,0b10100,0,1, "vmovl", "s",
3085 int_arm_neon_vmovls>;
3086 defm VMOVLu : N2VLInt_QHS<0b11,0b10100,0,1, "vmovl", "u",
3087 int_arm_neon_vmovlu>;
3089 // Vector Conversions.
3091 // VCVT : Vector Convert Between Floating-Point and Integers
3092 def VCVTf2sd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
3093 v2i32, v2f32, fp_to_sint>;
3094 def VCVTf2ud : N2VD<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
3095 v2i32, v2f32, fp_to_uint>;
3096 def VCVTs2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
3097 v2f32, v2i32, sint_to_fp>;
3098 def VCVTu2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
3099 v2f32, v2i32, uint_to_fp>;
3101 def VCVTf2sq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
3102 v4i32, v4f32, fp_to_sint>;
3103 def VCVTf2uq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
3104 v4i32, v4f32, fp_to_uint>;
3105 def VCVTs2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
3106 v4f32, v4i32, sint_to_fp>;
3107 def VCVTu2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
3108 v4f32, v4i32, uint_to_fp>;
3110 // VCVT : Vector Convert Between Floating-Point and Fixed-Point.
3111 def VCVTf2xsd : N2VCvtD<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
3112 v2i32, v2f32, int_arm_neon_vcvtfp2fxs>;
3113 def VCVTf2xud : N2VCvtD<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
3114 v2i32, v2f32, int_arm_neon_vcvtfp2fxu>;
3115 def VCVTxs2fd : N2VCvtD<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
3116 v2f32, v2i32, int_arm_neon_vcvtfxs2fp>;
3117 def VCVTxu2fd : N2VCvtD<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
3118 v2f32, v2i32, int_arm_neon_vcvtfxu2fp>;
3120 def VCVTf2xsq : N2VCvtQ<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
3121 v4i32, v4f32, int_arm_neon_vcvtfp2fxs>;
3122 def VCVTf2xuq : N2VCvtQ<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
3123 v4i32, v4f32, int_arm_neon_vcvtfp2fxu>;
3124 def VCVTxs2fq : N2VCvtQ<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
3125 v4f32, v4i32, int_arm_neon_vcvtfxs2fp>;
3126 def VCVTxu2fq : N2VCvtQ<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
3127 v4f32, v4i32, int_arm_neon_vcvtfxu2fp>;
3131 // VREV64 : Vector Reverse elements within 64-bit doublewords
3133 class VREV64D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3134 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 0, 0, (outs DPR:$dst),
3135 (ins DPR:$src), IIC_VMOVD,
3136 OpcodeStr, Dt, "$dst, $src", "",
3137 [(set DPR:$dst, (Ty (NEONvrev64 (Ty DPR:$src))))]>;
3138 class VREV64Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3139 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 1, 0, (outs QPR:$dst),
3140 (ins QPR:$src), IIC_VMOVD,
3141 OpcodeStr, Dt, "$dst, $src", "",
3142 [(set QPR:$dst, (Ty (NEONvrev64 (Ty QPR:$src))))]>;
3144 def VREV64d8 : VREV64D<0b00, "vrev64", "8", v8i8>;
3145 def VREV64d16 : VREV64D<0b01, "vrev64", "16", v4i16>;
3146 def VREV64d32 : VREV64D<0b10, "vrev64", "32", v2i32>;
3147 def VREV64df : VREV64D<0b10, "vrev64", "32", v2f32>;
3149 def VREV64q8 : VREV64Q<0b00, "vrev64", "8", v16i8>;
3150 def VREV64q16 : VREV64Q<0b01, "vrev64", "16", v8i16>;
3151 def VREV64q32 : VREV64Q<0b10, "vrev64", "32", v4i32>;
3152 def VREV64qf : VREV64Q<0b10, "vrev64", "32", v4f32>;
3154 // VREV32 : Vector Reverse elements within 32-bit words
3156 class VREV32D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3157 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 0, 0, (outs DPR:$dst),
3158 (ins DPR:$src), IIC_VMOVD,
3159 OpcodeStr, Dt, "$dst, $src", "",
3160 [(set DPR:$dst, (Ty (NEONvrev32 (Ty DPR:$src))))]>;
3161 class VREV32Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3162 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 1, 0, (outs QPR:$dst),
3163 (ins QPR:$src), IIC_VMOVD,
3164 OpcodeStr, Dt, "$dst, $src", "",
3165 [(set QPR:$dst, (Ty (NEONvrev32 (Ty QPR:$src))))]>;
3167 def VREV32d8 : VREV32D<0b00, "vrev32", "8", v8i8>;
3168 def VREV32d16 : VREV32D<0b01, "vrev32", "16", v4i16>;
3170 def VREV32q8 : VREV32Q<0b00, "vrev32", "8", v16i8>;
3171 def VREV32q16 : VREV32Q<0b01, "vrev32", "16", v8i16>;
3173 // VREV16 : Vector Reverse elements within 16-bit halfwords
3175 class VREV16D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3176 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 0, 0, (outs DPR:$dst),
3177 (ins DPR:$src), IIC_VMOVD,
3178 OpcodeStr, Dt, "$dst, $src", "",
3179 [(set DPR:$dst, (Ty (NEONvrev16 (Ty DPR:$src))))]>;
3180 class VREV16Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3181 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 1, 0, (outs QPR:$dst),
3182 (ins QPR:$src), IIC_VMOVD,
3183 OpcodeStr, Dt, "$dst, $src", "",
3184 [(set QPR:$dst, (Ty (NEONvrev16 (Ty QPR:$src))))]>;
3186 def VREV16d8 : VREV16D<0b00, "vrev16", "8", v8i8>;
3187 def VREV16q8 : VREV16Q<0b00, "vrev16", "8", v16i8>;
3189 // Other Vector Shuffles.
3191 // VEXT : Vector Extract
3193 class VEXTd<string OpcodeStr, string Dt, ValueType Ty>
3194 : N3V<0,1,0b11,{?,?,?,?},0,0, (outs DPR:$dst),
3195 (ins DPR:$lhs, DPR:$rhs, i32imm:$index), IIC_VEXTD,
3196 OpcodeStr, Dt, "$dst, $lhs, $rhs, $index", "",
3197 [(set DPR:$dst, (Ty (NEONvext (Ty DPR:$lhs),
3198 (Ty DPR:$rhs), imm:$index)))]>;
3200 class VEXTq<string OpcodeStr, string Dt, ValueType Ty>
3201 : N3V<0,1,0b11,{?,?,?,?},1,0, (outs QPR:$dst),
3202 (ins QPR:$lhs, QPR:$rhs, i32imm:$index), IIC_VEXTQ,
3203 OpcodeStr, Dt, "$dst, $lhs, $rhs, $index", "",
3204 [(set QPR:$dst, (Ty (NEONvext (Ty QPR:$lhs),
3205 (Ty QPR:$rhs), imm:$index)))]>;
3207 def VEXTd8 : VEXTd<"vext", "8", v8i8>;
3208 def VEXTd16 : VEXTd<"vext", "16", v4i16>;
3209 def VEXTd32 : VEXTd<"vext", "32", v2i32>;
3210 def VEXTdf : VEXTd<"vext", "32", v2f32>;
3212 def VEXTq8 : VEXTq<"vext", "8", v16i8>;
3213 def VEXTq16 : VEXTq<"vext", "16", v8i16>;
3214 def VEXTq32 : VEXTq<"vext", "32", v4i32>;
3215 def VEXTqf : VEXTq<"vext", "32", v4f32>;
3217 // VTRN : Vector Transpose
3219 def VTRNd8 : N2VDShuffle<0b00, 0b00001, "vtrn", "8">;
3220 def VTRNd16 : N2VDShuffle<0b01, 0b00001, "vtrn", "16">;
3221 def VTRNd32 : N2VDShuffle<0b10, 0b00001, "vtrn", "32">;
3223 def VTRNq8 : N2VQShuffle<0b00, 0b00001, IIC_VPERMQ, "vtrn", "8">;
3224 def VTRNq16 : N2VQShuffle<0b01, 0b00001, IIC_VPERMQ, "vtrn", "16">;
3225 def VTRNq32 : N2VQShuffle<0b10, 0b00001, IIC_VPERMQ, "vtrn", "32">;
3227 // VUZP : Vector Unzip (Deinterleave)
3229 def VUZPd8 : N2VDShuffle<0b00, 0b00010, "vuzp", "8">;
3230 def VUZPd16 : N2VDShuffle<0b01, 0b00010, "vuzp", "16">;
3231 def VUZPd32 : N2VDShuffle<0b10, 0b00010, "vuzp", "32">;
3233 def VUZPq8 : N2VQShuffle<0b00, 0b00010, IIC_VPERMQ3, "vuzp", "8">;
3234 def VUZPq16 : N2VQShuffle<0b01, 0b00010, IIC_VPERMQ3, "vuzp", "16">;
3235 def VUZPq32 : N2VQShuffle<0b10, 0b00010, IIC_VPERMQ3, "vuzp", "32">;
3237 // VZIP : Vector Zip (Interleave)
3239 def VZIPd8 : N2VDShuffle<0b00, 0b00011, "vzip", "8">;
3240 def VZIPd16 : N2VDShuffle<0b01, 0b00011, "vzip", "16">;
3241 def VZIPd32 : N2VDShuffle<0b10, 0b00011, "vzip", "32">;
3243 def VZIPq8 : N2VQShuffle<0b00, 0b00011, IIC_VPERMQ3, "vzip", "8">;
3244 def VZIPq16 : N2VQShuffle<0b01, 0b00011, IIC_VPERMQ3, "vzip", "16">;
3245 def VZIPq32 : N2VQShuffle<0b10, 0b00011, IIC_VPERMQ3, "vzip", "32">;
3247 // Vector Table Lookup and Table Extension.
3249 // VTBL : Vector Table Lookup
3251 : N3V<1,1,0b11,0b1000,0,0, (outs DPR:$dst),
3252 (ins DPR:$tbl1, DPR:$src), IIC_VTB1,
3253 "vtbl", "8", "$dst, \\{$tbl1\\}, $src", "",
3254 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl1 DPR:$tbl1, DPR:$src)))]>;
3255 let hasExtraSrcRegAllocReq = 1 in {
3257 : N3V<1,1,0b11,0b1001,0,0, (outs DPR:$dst),
3258 (ins DPR:$tbl1, DPR:$tbl2, DPR:$src), IIC_VTB2,
3259 "vtbl", "8", "$dst, \\{$tbl1, $tbl2\\}, $src", "",
3260 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl2
3261 DPR:$tbl1, DPR:$tbl2, DPR:$src)))]>;
3263 : N3V<1,1,0b11,0b1010,0,0, (outs DPR:$dst),
3264 (ins DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$src), IIC_VTB3,
3265 "vtbl", "8", "$dst, \\{$tbl1, $tbl2, $tbl3\\}, $src", "",
3266 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl3
3267 DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$src)))]>;
3269 : N3V<1,1,0b11,0b1011,0,0, (outs DPR:$dst),
3270 (ins DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$src), IIC_VTB4,
3271 "vtbl", "8", "$dst, \\{$tbl1, $tbl2, $tbl3, $tbl4\\}, $src", "",
3272 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl4 DPR:$tbl1, DPR:$tbl2,
3273 DPR:$tbl3, DPR:$tbl4, DPR:$src)))]>;
3274 } // hasExtraSrcRegAllocReq = 1
3276 // VTBX : Vector Table Extension
3278 : N3V<1,1,0b11,0b1000,1,0, (outs DPR:$dst),
3279 (ins DPR:$orig, DPR:$tbl1, DPR:$src), IIC_VTBX1,
3280 "vtbx", "8", "$dst, \\{$tbl1\\}, $src", "$orig = $dst",
3281 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx1
3282 DPR:$orig, DPR:$tbl1, DPR:$src)))]>;
3283 let hasExtraSrcRegAllocReq = 1 in {
3285 : N3V<1,1,0b11,0b1001,1,0, (outs DPR:$dst),
3286 (ins DPR:$orig, DPR:$tbl1, DPR:$tbl2, DPR:$src), IIC_VTBX2,
3287 "vtbx", "8", "$dst, \\{$tbl1, $tbl2\\}, $src", "$orig = $dst",
3288 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx2
3289 DPR:$orig, DPR:$tbl1, DPR:$tbl2, DPR:$src)))]>;
3291 : N3V<1,1,0b11,0b1010,1,0, (outs DPR:$dst),
3292 (ins DPR:$orig, DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$src), IIC_VTBX3,
3293 "vtbx", "8", "$dst, \\{$tbl1, $tbl2, $tbl3\\}, $src", "$orig = $dst",
3294 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx3 DPR:$orig, DPR:$tbl1,
3295 DPR:$tbl2, DPR:$tbl3, DPR:$src)))]>;
3297 : N3V<1,1,0b11,0b1011,1,0, (outs DPR:$dst), (ins DPR:$orig, DPR:$tbl1,
3298 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$src), IIC_VTBX4,
3299 "vtbx", "8", "$dst, \\{$tbl1, $tbl2, $tbl3, $tbl4\\}, $src",
3301 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx4 DPR:$orig, DPR:$tbl1,
3302 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$src)))]>;
3303 } // hasExtraSrcRegAllocReq = 1
3305 //===----------------------------------------------------------------------===//
3306 // NEON instructions for single-precision FP math
3307 //===----------------------------------------------------------------------===//
3309 class N2VSPat<SDNode OpNode, ValueType ResTy, ValueType OpTy, NeonI Inst>
3310 : NEONFPPat<(ResTy (OpNode SPR:$a)),
3311 (EXTRACT_SUBREG (OpTy (Inst (INSERT_SUBREG (OpTy (IMPLICIT_DEF)),
3312 SPR:$a, arm_ssubreg_0))),
3315 class N3VSPat<SDNode OpNode, NeonI Inst>
3316 : NEONFPPat<(f32 (OpNode SPR:$a, SPR:$b)),
3317 (EXTRACT_SUBREG (v2f32
3318 (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3319 SPR:$a, arm_ssubreg_0),
3320 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3321 SPR:$b, arm_ssubreg_0))),
3324 class N3VSMulOpPat<SDNode MulNode, SDNode OpNode, NeonI Inst>
3325 : NEONFPPat<(f32 (OpNode SPR:$acc, (f32 (MulNode SPR:$a, SPR:$b)))),
3326 (EXTRACT_SUBREG (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3327 SPR:$acc, arm_ssubreg_0),
3328 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3329 SPR:$a, arm_ssubreg_0),
3330 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3331 SPR:$b, arm_ssubreg_0)),
3334 // These need separate instructions because they must use DPR_VFP2 register
3335 // class which have SPR sub-registers.
3337 // Vector Add Operations used for single-precision FP
3338 let neverHasSideEffects = 1 in
3339 def VADDfd_sfp : N3VS<0,0,0b00,0b1101,0, "vadd", "f32", v2f32, v2f32, fadd, 1>;
3340 def : N3VSPat<fadd, VADDfd_sfp>;
3342 // Vector Sub Operations used for single-precision FP
3343 let neverHasSideEffects = 1 in
3344 def VSUBfd_sfp : N3VS<0,0,0b10,0b1101,0, "vsub", "f32", v2f32, v2f32, fsub, 0>;
3345 def : N3VSPat<fsub, VSUBfd_sfp>;
3347 // Vector Multiply Operations used for single-precision FP
3348 let neverHasSideEffects = 1 in
3349 def VMULfd_sfp : N3VS<1,0,0b00,0b1101,1, "vmul", "f32", v2f32, v2f32, fmul, 1>;
3350 def : N3VSPat<fmul, VMULfd_sfp>;
3352 // Vector Multiply-Accumulate/Subtract used for single-precision FP
3353 // vml[as].f32 can cause 4-8 cycle stalls in following ASIMD instructions, so
3354 // we want to avoid them for now. e.g., alternating vmla/vadd instructions.
3356 //let neverHasSideEffects = 1 in
3357 //def VMLAfd_sfp : N3VSMulOp<0,0,0b00,0b1101,1, IIC_VMACD, "vmla", "f32",
3358 // v2f32, fmul, fadd>;
3359 //def : N3VSMulOpPat<fmul, fadd, VMLAfd_sfp>;
3361 //let neverHasSideEffects = 1 in
3362 //def VMLSfd_sfp : N3VSMulOp<0,0,0b10,0b1101,1, IIC_VMACD, "vmls", "f32",
3363 // v2f32, fmul, fsub>;
3364 //def : N3VSMulOpPat<fmul, fsub, VMLSfd_sfp>;
3366 // Vector Absolute used for single-precision FP
3367 let neverHasSideEffects = 1 in
3368 def VABSfd_sfp : N2V<0b11, 0b11, 0b10, 0b01, 0b01110, 0, 0,
3369 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), IIC_VUNAD,
3370 "vabs", "f32", "$dst, $src", "", []>;
3371 def : N2VSPat<fabs, f32, v2f32, VABSfd_sfp>;
3373 // Vector Negate used for single-precision FP
3374 let neverHasSideEffects = 1 in
3375 def VNEGfd_sfp : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
3376 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), IIC_VUNAD,
3377 "vneg", "f32", "$dst, $src", "", []>;
3378 def : N2VSPat<fneg, f32, v2f32, VNEGfd_sfp>;
3380 // Vector Maximum used for single-precision FP
3381 let neverHasSideEffects = 1 in
3382 def VMAXfd_sfp : N3V<0, 0, 0b00, 0b1111, 0, 0, (outs DPR_VFP2:$dst),
3383 (ins DPR_VFP2:$src1, DPR_VFP2:$src2), IIC_VBIND,
3384 "vmax", "f32", "$dst, $src1, $src2", "", []>;
3385 def : N3VSPat<NEONfmax, VMAXfd_sfp>;
3387 // Vector Minimum used for single-precision FP
3388 let neverHasSideEffects = 1 in
3389 def VMINfd_sfp : N3V<0, 0, 0b00, 0b1111, 0, 0, (outs DPR_VFP2:$dst),
3390 (ins DPR_VFP2:$src1, DPR_VFP2:$src2), IIC_VBIND,
3391 "vmin", "f32", "$dst, $src1, $src2", "", []>;
3392 def : N3VSPat<NEONfmin, VMINfd_sfp>;
3394 // Vector Convert between single-precision FP and integer
3395 let neverHasSideEffects = 1 in
3396 def VCVTf2sd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
3397 v2i32, v2f32, fp_to_sint>;
3398 def : N2VSPat<arm_ftosi, f32, v2f32, VCVTf2sd_sfp>;
3400 let neverHasSideEffects = 1 in
3401 def VCVTf2ud_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
3402 v2i32, v2f32, fp_to_uint>;
3403 def : N2VSPat<arm_ftoui, f32, v2f32, VCVTf2ud_sfp>;
3405 let neverHasSideEffects = 1 in
3406 def VCVTs2fd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
3407 v2f32, v2i32, sint_to_fp>;
3408 def : N2VSPat<arm_sitof, f32, v2i32, VCVTs2fd_sfp>;
3410 let neverHasSideEffects = 1 in
3411 def VCVTu2fd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
3412 v2f32, v2i32, uint_to_fp>;
3413 def : N2VSPat<arm_uitof, f32, v2i32, VCVTu2fd_sfp>;
3415 //===----------------------------------------------------------------------===//
3416 // Non-Instruction Patterns
3417 //===----------------------------------------------------------------------===//
3420 def : Pat<(v1i64 (bitconvert (v2i32 DPR:$src))), (v1i64 DPR:$src)>;
3421 def : Pat<(v1i64 (bitconvert (v4i16 DPR:$src))), (v1i64 DPR:$src)>;
3422 def : Pat<(v1i64 (bitconvert (v8i8 DPR:$src))), (v1i64 DPR:$src)>;
3423 def : Pat<(v1i64 (bitconvert (f64 DPR:$src))), (v1i64 DPR:$src)>;
3424 def : Pat<(v1i64 (bitconvert (v2f32 DPR:$src))), (v1i64 DPR:$src)>;
3425 def : Pat<(v2i32 (bitconvert (v1i64 DPR:$src))), (v2i32 DPR:$src)>;
3426 def : Pat<(v2i32 (bitconvert (v4i16 DPR:$src))), (v2i32 DPR:$src)>;
3427 def : Pat<(v2i32 (bitconvert (v8i8 DPR:$src))), (v2i32 DPR:$src)>;
3428 def : Pat<(v2i32 (bitconvert (f64 DPR:$src))), (v2i32 DPR:$src)>;
3429 def : Pat<(v2i32 (bitconvert (v2f32 DPR:$src))), (v2i32 DPR:$src)>;
3430 def : Pat<(v4i16 (bitconvert (v1i64 DPR:$src))), (v4i16 DPR:$src)>;
3431 def : Pat<(v4i16 (bitconvert (v2i32 DPR:$src))), (v4i16 DPR:$src)>;
3432 def : Pat<(v4i16 (bitconvert (v8i8 DPR:$src))), (v4i16 DPR:$src)>;
3433 def : Pat<(v4i16 (bitconvert (f64 DPR:$src))), (v4i16 DPR:$src)>;
3434 def : Pat<(v4i16 (bitconvert (v2f32 DPR:$src))), (v4i16 DPR:$src)>;
3435 def : Pat<(v8i8 (bitconvert (v1i64 DPR:$src))), (v8i8 DPR:$src)>;
3436 def : Pat<(v8i8 (bitconvert (v2i32 DPR:$src))), (v8i8 DPR:$src)>;
3437 def : Pat<(v8i8 (bitconvert (v4i16 DPR:$src))), (v8i8 DPR:$src)>;
3438 def : Pat<(v8i8 (bitconvert (f64 DPR:$src))), (v8i8 DPR:$src)>;
3439 def : Pat<(v8i8 (bitconvert (v2f32 DPR:$src))), (v8i8 DPR:$src)>;
3440 def : Pat<(f64 (bitconvert (v1i64 DPR:$src))), (f64 DPR:$src)>;
3441 def : Pat<(f64 (bitconvert (v2i32 DPR:$src))), (f64 DPR:$src)>;
3442 def : Pat<(f64 (bitconvert (v4i16 DPR:$src))), (f64 DPR:$src)>;
3443 def : Pat<(f64 (bitconvert (v8i8 DPR:$src))), (f64 DPR:$src)>;
3444 def : Pat<(f64 (bitconvert (v2f32 DPR:$src))), (f64 DPR:$src)>;
3445 def : Pat<(v2f32 (bitconvert (f64 DPR:$src))), (v2f32 DPR:$src)>;
3446 def : Pat<(v2f32 (bitconvert (v1i64 DPR:$src))), (v2f32 DPR:$src)>;
3447 def : Pat<(v2f32 (bitconvert (v2i32 DPR:$src))), (v2f32 DPR:$src)>;
3448 def : Pat<(v2f32 (bitconvert (v4i16 DPR:$src))), (v2f32 DPR:$src)>;
3449 def : Pat<(v2f32 (bitconvert (v8i8 DPR:$src))), (v2f32 DPR:$src)>;
3451 def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (v2i64 QPR:$src)>;
3452 def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (v2i64 QPR:$src)>;
3453 def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (v2i64 QPR:$src)>;
3454 def : Pat<(v2i64 (bitconvert (v2f64 QPR:$src))), (v2i64 QPR:$src)>;
3455 def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (v2i64 QPR:$src)>;
3456 def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (v4i32 QPR:$src)>;
3457 def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (v4i32 QPR:$src)>;
3458 def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (v4i32 QPR:$src)>;
3459 def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (v4i32 QPR:$src)>;
3460 def : Pat<(v4i32 (bitconvert (v4f32 QPR:$src))), (v4i32 QPR:$src)>;
3461 def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (v8i16 QPR:$src)>;
3462 def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (v8i16 QPR:$src)>;
3463 def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (v8i16 QPR:$src)>;
3464 def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (v8i16 QPR:$src)>;
3465 def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (v8i16 QPR:$src)>;
3466 def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (v16i8 QPR:$src)>;
3467 def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (v16i8 QPR:$src)>;
3468 def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (v16i8 QPR:$src)>;
3469 def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (v16i8 QPR:$src)>;
3470 def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (v16i8 QPR:$src)>;
3471 def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (v4f32 QPR:$src)>;
3472 def : Pat<(v4f32 (bitconvert (v4i32 QPR:$src))), (v4f32 QPR:$src)>;
3473 def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (v4f32 QPR:$src)>;
3474 def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (v4f32 QPR:$src)>;
3475 def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (v4f32 QPR:$src)>;
3476 def : Pat<(v2f64 (bitconvert (v2i64 QPR:$src))), (v2f64 QPR:$src)>;
3477 def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (v2f64 QPR:$src)>;
3478 def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 QPR:$src)>;
3479 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 QPR:$src)>;
3480 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 QPR:$src)>;