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 class N2VS<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
858 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
859 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
860 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
861 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src),
862 IIC_VUNAD, OpcodeStr, Dt, "$dst, $src", "", []>;
863 class N2VD<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
864 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
865 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
866 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
867 (ins DPR:$src), IIC_VUNAD, OpcodeStr, Dt,"$dst, $src", "",
868 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src))))]>;
869 class N2VQ<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
870 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
871 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
872 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
873 (ins QPR:$src), IIC_VUNAQ, OpcodeStr, Dt,"$dst, $src", "",
874 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src))))]>;
876 // Basic 2-register intrinsics, both double- and quad-register.
877 class N2VDInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
878 bits<2> op17_16, bits<5> op11_7, bit op4,
879 InstrItinClass itin, string OpcodeStr, string Dt,
880 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
881 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
882 (ins DPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
883 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
884 class N2VQInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
885 bits<2> op17_16, bits<5> op11_7, bit op4,
886 InstrItinClass itin, string OpcodeStr, string Dt,
887 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
888 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
889 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
890 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
892 // Narrow 2-register intrinsics.
893 class N2VNInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
894 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
895 InstrItinClass itin, string OpcodeStr, string Dt,
896 ValueType TyD, ValueType TyQ, Intrinsic IntOp>
897 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$dst),
898 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
899 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src))))]>;
901 // Long 2-register intrinsics (currently only used for VMOVL).
902 class N2VLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
903 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
904 InstrItinClass itin, string OpcodeStr, string Dt,
905 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
906 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$dst),
907 (ins DPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
908 [(set QPR:$dst, (TyQ (IntOp (TyD DPR:$src))))]>;
910 // 2-register shuffles (VTRN/VZIP/VUZP), both double- and quad-register.
911 class N2VDShuffle<bits<2> op19_18, bits<5> op11_7, string OpcodeStr, string Dt>
912 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 0, 0, (outs DPR:$dst1, DPR:$dst2),
913 (ins DPR:$src1, DPR:$src2), IIC_VPERMD,
914 OpcodeStr, Dt, "$dst1, $dst2",
915 "$src1 = $dst1, $src2 = $dst2", []>;
916 class N2VQShuffle<bits<2> op19_18, bits<5> op11_7,
917 InstrItinClass itin, string OpcodeStr, string Dt>
918 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 1, 0, (outs QPR:$dst1, QPR:$dst2),
919 (ins QPR:$src1, QPR:$src2), itin, OpcodeStr, Dt, "$dst1, $dst2",
920 "$src1 = $dst1, $src2 = $dst2", []>;
922 // Basic 3-register operations: single-, double- and quad-register.
923 class N3VS<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
924 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
925 SDNode OpNode, bit Commutable>
926 : N3V<op24, op23, op21_20, op11_8, 0, op4,
927 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src1, DPR_VFP2:$src2), IIC_VBIND,
928 OpcodeStr, Dt, "$dst, $src1, $src2", "", []> {
929 let isCommutable = Commutable;
932 class N3VD<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
933 InstrItinClass itin, string OpcodeStr, string Dt,
934 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
935 : N3V<op24, op23, op21_20, op11_8, 0, op4,
936 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), itin,
937 OpcodeStr, Dt, "$dst, $src1, $src2", "",
938 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src1), (OpTy DPR:$src2))))]> {
939 let isCommutable = Commutable;
941 // Same as N3VD but no data type.
942 class N3VDX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
943 InstrItinClass itin, string OpcodeStr,
944 ValueType ResTy, ValueType OpTy,
945 SDNode OpNode, bit Commutable>
946 : N3VX<op24, op23, op21_20, op11_8, 0, op4,
947 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), itin,
948 OpcodeStr, "$dst, $src1, $src2", "",
949 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src1), (OpTy DPR:$src2))))]>{
950 let isCommutable = Commutable;
952 class N3VDSL<bits<2> op21_20, bits<4> op11_8,
953 InstrItinClass itin, string OpcodeStr, string Dt,
954 ValueType Ty, SDNode ShOp>
955 : N3V<0, 1, op21_20, op11_8, 1, 0,
956 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
957 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
959 (Ty (ShOp (Ty DPR:$src1),
960 (Ty (NEONvduplane (Ty DPR_VFP2:$src2), imm:$lane)))))]>{
961 let isCommutable = 0;
963 class N3VDSL16<bits<2> op21_20, bits<4> op11_8,
964 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
965 : N3V<0, 1, op21_20, op11_8, 1, 0,
966 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
967 IIC_VMULi16D, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
969 (Ty (ShOp (Ty DPR:$src1),
970 (Ty (NEONvduplane (Ty DPR_8:$src2), imm:$lane)))))]> {
971 let isCommutable = 0;
974 class N3VQ<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
975 InstrItinClass itin, string OpcodeStr, string Dt,
976 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
977 : N3V<op24, op23, op21_20, op11_8, 1, op4,
978 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), itin,
979 OpcodeStr, Dt, "$dst, $src1, $src2", "",
980 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src1), (OpTy QPR:$src2))))]> {
981 let isCommutable = Commutable;
983 class N3VQX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
984 InstrItinClass itin, string OpcodeStr,
985 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
986 : N3VX<op24, op23, op21_20, op11_8, 1, op4,
987 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), itin,
988 OpcodeStr, "$dst, $src1, $src2", "",
989 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src1), (OpTy QPR:$src2))))]>{
990 let isCommutable = Commutable;
992 class N3VQSL<bits<2> op21_20, bits<4> op11_8,
993 InstrItinClass itin, string OpcodeStr, string Dt,
994 ValueType ResTy, ValueType OpTy, SDNode ShOp>
995 : N3V<1, 1, op21_20, op11_8, 1, 0,
996 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
997 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
998 [(set (ResTy QPR:$dst),
999 (ResTy (ShOp (ResTy QPR:$src1),
1000 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1002 let isCommutable = 0;
1004 class N3VQSL16<bits<2> op21_20, bits<4> op11_8, string OpcodeStr, string Dt,
1005 ValueType ResTy, ValueType OpTy, SDNode ShOp>
1006 : N3V<1, 1, op21_20, op11_8, 1, 0,
1007 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1008 IIC_VMULi16Q, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1009 [(set (ResTy QPR:$dst),
1010 (ResTy (ShOp (ResTy QPR:$src1),
1011 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
1013 let isCommutable = 0;
1016 // Basic 3-register intrinsics, both double- and quad-register.
1017 class N3VDInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1018 InstrItinClass itin, string OpcodeStr, string Dt,
1019 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
1020 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1021 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), itin,
1022 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1023 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src1), (OpTy DPR:$src2))))]> {
1024 let isCommutable = Commutable;
1026 class N3VDIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1027 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
1028 : N3V<0, 1, op21_20, op11_8, 1, 0,
1029 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1030 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1031 [(set (Ty DPR:$dst),
1032 (Ty (IntOp (Ty DPR:$src1),
1033 (Ty (NEONvduplane (Ty DPR_VFP2:$src2),
1035 let isCommutable = 0;
1037 class N3VDIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1038 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
1039 : N3V<0, 1, op21_20, op11_8, 1, 0,
1040 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1041 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1042 [(set (Ty DPR:$dst),
1043 (Ty (IntOp (Ty DPR:$src1),
1044 (Ty (NEONvduplane (Ty DPR_8:$src2),
1046 let isCommutable = 0;
1049 class N3VQInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1050 InstrItinClass itin, string OpcodeStr, string Dt,
1051 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
1052 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1053 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), itin,
1054 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1055 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src1), (OpTy QPR:$src2))))]> {
1056 let isCommutable = Commutable;
1058 class N3VQIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1059 string OpcodeStr, string Dt,
1060 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1061 : N3V<1, 1, op21_20, op11_8, 1, 0,
1062 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1063 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1064 [(set (ResTy QPR:$dst),
1065 (ResTy (IntOp (ResTy QPR:$src1),
1066 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1068 let isCommutable = 0;
1070 class N3VQIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1071 string OpcodeStr, string Dt,
1072 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1073 : N3V<1, 1, op21_20, op11_8, 1, 0,
1074 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1075 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1076 [(set (ResTy QPR:$dst),
1077 (ResTy (IntOp (ResTy QPR:$src1),
1078 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
1080 let isCommutable = 0;
1083 // Multiply-Add/Sub operations: single-, double- and quad-register.
1084 class N3VSMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1085 InstrItinClass itin, string OpcodeStr, string Dt,
1086 ValueType Ty, SDNode MulOp, SDNode OpNode>
1087 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1088 (outs DPR_VFP2:$dst),
1089 (ins DPR_VFP2:$src1, DPR_VFP2:$src2, DPR_VFP2:$src3), itin,
1090 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst", []>;
1092 class N3VDMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1093 InstrItinClass itin, string OpcodeStr, string Dt,
1094 ValueType Ty, SDNode MulOp, SDNode OpNode>
1095 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1096 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3), itin,
1097 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1098 [(set DPR:$dst, (Ty (OpNode DPR:$src1,
1099 (Ty (MulOp DPR:$src2, DPR:$src3)))))]>;
1100 class N3VDMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1101 string OpcodeStr, string Dt,
1102 ValueType Ty, SDNode MulOp, SDNode ShOp>
1103 : N3V<0, 1, op21_20, op11_8, 1, 0,
1105 (ins DPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane), itin,
1106 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1107 [(set (Ty DPR:$dst),
1108 (Ty (ShOp (Ty DPR:$src1),
1109 (Ty (MulOp DPR:$src2,
1110 (Ty (NEONvduplane (Ty DPR_VFP2:$src3),
1112 class N3VDMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1113 string OpcodeStr, string Dt,
1114 ValueType Ty, SDNode MulOp, SDNode ShOp>
1115 : N3V<0, 1, op21_20, op11_8, 1, 0,
1117 (ins DPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane), itin,
1118 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1119 [(set (Ty DPR:$dst),
1120 (Ty (ShOp (Ty DPR:$src1),
1121 (Ty (MulOp DPR:$src2,
1122 (Ty (NEONvduplane (Ty DPR_8:$src3),
1125 class N3VQMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1126 InstrItinClass itin, string OpcodeStr, string Dt, ValueType Ty,
1127 SDNode MulOp, SDNode OpNode>
1128 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1129 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3), itin,
1130 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1131 [(set QPR:$dst, (Ty (OpNode QPR:$src1,
1132 (Ty (MulOp QPR:$src2, QPR:$src3)))))]>;
1133 class N3VQMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1134 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1135 SDNode MulOp, SDNode ShOp>
1136 : N3V<1, 1, op21_20, op11_8, 1, 0,
1138 (ins QPR:$src1, QPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane), itin,
1139 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1140 [(set (ResTy QPR:$dst),
1141 (ResTy (ShOp (ResTy QPR:$src1),
1142 (ResTy (MulOp QPR:$src2,
1143 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src3),
1145 class N3VQMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1146 string OpcodeStr, string Dt,
1147 ValueType ResTy, ValueType OpTy,
1148 SDNode MulOp, SDNode ShOp>
1149 : N3V<1, 1, op21_20, op11_8, 1, 0,
1151 (ins QPR:$src1, QPR:$src2, DPR_8:$src3, nohash_imm:$lane), itin,
1152 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1153 [(set (ResTy QPR:$dst),
1154 (ResTy (ShOp (ResTy QPR:$src1),
1155 (ResTy (MulOp QPR:$src2,
1156 (ResTy (NEONvduplane (OpTy DPR_8:$src3),
1159 // Neon 3-argument intrinsics, both double- and quad-register.
1160 // The destination register is also used as the first source operand register.
1161 class N3VDInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1162 InstrItinClass itin, string OpcodeStr, string Dt,
1163 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1164 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1165 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3), itin,
1166 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1167 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src1),
1168 (OpTy DPR:$src2), (OpTy DPR:$src3))))]>;
1169 class N3VQInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1170 InstrItinClass itin, string OpcodeStr, string Dt,
1171 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1172 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1173 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3), itin,
1174 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1175 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src1),
1176 (OpTy QPR:$src2), (OpTy QPR:$src3))))]>;
1178 // Neon Long 3-argument intrinsic. The destination register is
1179 // a quad-register and is also used as the first source operand register.
1180 class N3VLInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1181 InstrItinClass itin, string OpcodeStr, string Dt,
1182 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
1183 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1184 (outs QPR:$dst), (ins QPR:$src1, DPR:$src2, DPR:$src3), itin,
1185 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1187 (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$src2), (TyD DPR:$src3))))]>;
1188 class N3VLInt3SL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1189 string OpcodeStr, string Dt,
1190 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1191 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1193 (ins QPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane), itin,
1194 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1195 [(set (ResTy QPR:$dst),
1196 (ResTy (IntOp (ResTy QPR:$src1),
1198 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src3),
1200 class N3VLInt3SL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1201 InstrItinClass itin, string OpcodeStr, string Dt,
1202 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1203 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1205 (ins QPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane), itin,
1206 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1207 [(set (ResTy QPR:$dst),
1208 (ResTy (IntOp (ResTy QPR:$src1),
1210 (OpTy (NEONvduplane (OpTy DPR_8:$src3),
1213 // Narrowing 3-register intrinsics.
1214 class N3VNInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1215 string OpcodeStr, string Dt, ValueType TyD, ValueType TyQ,
1216 Intrinsic IntOp, bit Commutable>
1217 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1218 (outs DPR:$dst), (ins QPR:$src1, QPR:$src2), IIC_VBINi4D,
1219 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1220 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src1), (TyQ QPR:$src2))))]> {
1221 let isCommutable = Commutable;
1224 // Long 3-register intrinsics.
1225 class N3VLInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1226 InstrItinClass itin, string OpcodeStr, string Dt,
1227 ValueType TyQ, ValueType TyD, Intrinsic IntOp, bit Commutable>
1228 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1229 (outs QPR:$dst), (ins DPR:$src1, DPR:$src2), itin,
1230 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1231 [(set QPR:$dst, (TyQ (IntOp (TyD DPR:$src1), (TyD DPR:$src2))))]> {
1232 let isCommutable = Commutable;
1234 class N3VLIntSL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1235 string OpcodeStr, string Dt,
1236 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1237 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1238 (outs QPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1239 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1240 [(set (ResTy QPR:$dst),
1241 (ResTy (IntOp (OpTy DPR:$src1),
1242 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1244 class N3VLIntSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1245 InstrItinClass itin, string OpcodeStr, string Dt,
1246 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1247 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1248 (outs QPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1249 itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1250 [(set (ResTy QPR:$dst),
1251 (ResTy (IntOp (OpTy DPR:$src1),
1252 (OpTy (NEONvduplane (OpTy DPR_8:$src2),
1255 // Wide 3-register intrinsics.
1256 class N3VWInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1257 string OpcodeStr, string Dt, ValueType TyQ, ValueType TyD,
1258 Intrinsic IntOp, bit Commutable>
1259 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1260 (outs QPR:$dst), (ins QPR:$src1, DPR:$src2), IIC_VSUBiD,
1261 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1262 [(set QPR:$dst, (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$src2))))]> {
1263 let isCommutable = Commutable;
1266 // Pairwise long 2-register intrinsics, both double- and quad-register.
1267 class N2VDPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1268 bits<2> op17_16, bits<5> op11_7, bit op4,
1269 string OpcodeStr, string Dt,
1270 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1271 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
1272 (ins DPR:$src), IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
1273 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
1274 class N2VQPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1275 bits<2> op17_16, bits<5> op11_7, bit op4,
1276 string OpcodeStr, string Dt,
1277 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1278 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
1279 (ins QPR:$src), IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
1280 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
1282 // Pairwise long 2-register accumulate intrinsics,
1283 // both double- and quad-register.
1284 // The destination register is also used as the first source operand register.
1285 class N2VDPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1286 bits<2> op17_16, bits<5> op11_7, bit op4,
1287 string OpcodeStr, string Dt,
1288 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1289 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
1290 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), IIC_VPALiD,
1291 OpcodeStr, Dt, "$dst, $src2", "$src1 = $dst",
1292 [(set DPR:$dst, (ResTy (IntOp (ResTy DPR:$src1), (OpTy DPR:$src2))))]>;
1293 class N2VQPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1294 bits<2> op17_16, bits<5> op11_7, bit op4,
1295 string OpcodeStr, string Dt,
1296 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1297 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4,
1298 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), IIC_VPALiQ,
1299 OpcodeStr, Dt, "$dst, $src2", "$src1 = $dst",
1300 [(set QPR:$dst, (ResTy (IntOp (ResTy QPR:$src1), (OpTy QPR:$src2))))]>;
1302 // Shift by immediate,
1303 // both double- and quad-register.
1304 class N2VDSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1305 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1306 ValueType Ty, SDNode OpNode>
1307 : N2VImm<op24, op23, op11_8, op7, 0, op4,
1308 (outs DPR:$dst), (ins DPR:$src, i32imm:$SIMM), f, itin,
1309 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1310 [(set DPR:$dst, (Ty (OpNode (Ty DPR:$src), (i32 imm:$SIMM))))]>;
1311 class N2VQSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1312 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1313 ValueType Ty, SDNode OpNode>
1314 : N2VImm<op24, op23, op11_8, op7, 1, op4,
1315 (outs QPR:$dst), (ins QPR:$src, i32imm:$SIMM), f, itin,
1316 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1317 [(set QPR:$dst, (Ty (OpNode (Ty QPR:$src), (i32 imm:$SIMM))))]>;
1319 // Long shift by immediate.
1320 class N2VLSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
1321 string OpcodeStr, string Dt,
1322 ValueType ResTy, ValueType OpTy, SDNode OpNode>
1323 : N2VImm<op24, op23, op11_8, op7, op6, op4,
1324 (outs QPR:$dst), (ins DPR:$src, i32imm:$SIMM), N2RegVShLFrm,
1325 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1326 [(set QPR:$dst, (ResTy (OpNode (OpTy DPR:$src),
1327 (i32 imm:$SIMM))))]>;
1329 // Narrow shift by immediate.
1330 class N2VNSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
1331 InstrItinClass itin, string OpcodeStr, string Dt,
1332 ValueType ResTy, ValueType OpTy, SDNode OpNode>
1333 : N2VImm<op24, op23, op11_8, op7, op6, op4,
1334 (outs DPR:$dst), (ins QPR:$src, i32imm:$SIMM), N2RegVShRFrm, itin,
1335 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1336 [(set DPR:$dst, (ResTy (OpNode (OpTy QPR:$src),
1337 (i32 imm:$SIMM))))]>;
1339 // Shift right by immediate and accumulate,
1340 // both double- and quad-register.
1341 class N2VDShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1342 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1343 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$dst),
1344 (ins DPR:$src1, DPR:$src2, i32imm:$SIMM), N2RegVShRFrm, IIC_VPALiD,
1345 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1346 [(set DPR:$dst, (Ty (add DPR:$src1,
1347 (Ty (ShOp DPR:$src2, (i32 imm:$SIMM))))))]>;
1348 class N2VQShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1349 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1350 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$dst),
1351 (ins QPR:$src1, QPR:$src2, i32imm:$SIMM), N2RegVShRFrm, IIC_VPALiD,
1352 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1353 [(set QPR:$dst, (Ty (add QPR:$src1,
1354 (Ty (ShOp QPR:$src2, (i32 imm:$SIMM))))))]>;
1356 // Shift by immediate and insert,
1357 // both double- and quad-register.
1358 class N2VDShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1359 Format f, string OpcodeStr, string Dt, ValueType Ty,SDNode ShOp>
1360 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$dst),
1361 (ins DPR:$src1, DPR:$src2, i32imm:$SIMM), f, IIC_VSHLiD,
1362 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1363 [(set DPR:$dst, (Ty (ShOp DPR:$src1, DPR:$src2, (i32 imm:$SIMM))))]>;
1364 class N2VQShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1365 Format f, string OpcodeStr, string Dt, ValueType Ty,SDNode ShOp>
1366 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$dst),
1367 (ins QPR:$src1, QPR:$src2, i32imm:$SIMM), f, IIC_VSHLiQ,
1368 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1369 [(set QPR:$dst, (Ty (ShOp QPR:$src1, QPR:$src2, (i32 imm:$SIMM))))]>;
1371 // Convert, with fractional bits immediate,
1372 // both double- and quad-register.
1373 class N2VCvtD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1374 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1376 : N2VImm<op24, op23, op11_8, op7, 0, op4,
1377 (outs DPR:$dst), (ins DPR:$src, i32imm:$SIMM), NVCVTFrm,
1378 IIC_VUNAD, OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1379 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src), (i32 imm:$SIMM))))]>;
1380 class N2VCvtQ<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1381 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1383 : N2VImm<op24, op23, op11_8, op7, 1, op4,
1384 (outs QPR:$dst), (ins QPR:$src, i32imm:$SIMM), NVCVTFrm,
1385 IIC_VUNAQ, OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1386 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src), (i32 imm:$SIMM))))]>;
1388 //===----------------------------------------------------------------------===//
1390 //===----------------------------------------------------------------------===//
1392 // Abbreviations used in multiclass suffixes:
1393 // Q = quarter int (8 bit) elements
1394 // H = half int (16 bit) elements
1395 // S = single int (32 bit) elements
1396 // D = double int (64 bit) elements
1398 // Neon 2-register vector operations -- for disassembly only.
1400 // First with only element sizes of 8, 16 and 32 bits:
1401 multiclass N2V_QHS_cmp<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1402 bits<5> op11_7, bit op4, string opc, string Dt,
1404 // 64-bit vector types.
1405 def v8i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 0, op4,
1406 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1407 opc, !strconcat(Dt, "8"), asm, "", []>;
1408 def v4i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 0, op4,
1409 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1410 opc, !strconcat(Dt, "16"), asm, "", []>;
1411 def v2i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
1412 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1413 opc, !strconcat(Dt, "32"), asm, "", []>;
1414 def v2f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
1415 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1416 opc, "f32", asm, "", []> {
1417 let Inst{10} = 1; // overwrite F = 1
1420 // 128-bit vector types.
1421 def v16i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 1, op4,
1422 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1423 opc, !strconcat(Dt, "8"), asm, "", []>;
1424 def v8i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 1, op4,
1425 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1426 opc, !strconcat(Dt, "16"), asm, "", []>;
1427 def v4i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
1428 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1429 opc, !strconcat(Dt, "32"), asm, "", []>;
1430 def v4f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
1431 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1432 opc, "f32", asm, "", []> {
1433 let Inst{10} = 1; // overwrite F = 1
1437 // Neon 3-register vector operations.
1439 // First with only element sizes of 8, 16 and 32 bits:
1440 multiclass N3V_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1441 InstrItinClass itinD16, InstrItinClass itinD32,
1442 InstrItinClass itinQ16, InstrItinClass itinQ32,
1443 string OpcodeStr, string Dt,
1444 SDNode OpNode, bit Commutable = 0> {
1445 // 64-bit vector types.
1446 def v8i8 : N3VD<op24, op23, 0b00, op11_8, op4, itinD16,
1447 OpcodeStr, !strconcat(Dt, "8"),
1448 v8i8, v8i8, OpNode, Commutable>;
1449 def v4i16 : N3VD<op24, op23, 0b01, op11_8, op4, itinD16,
1450 OpcodeStr, !strconcat(Dt, "16"),
1451 v4i16, v4i16, OpNode, Commutable>;
1452 def v2i32 : N3VD<op24, op23, 0b10, op11_8, op4, itinD32,
1453 OpcodeStr, !strconcat(Dt, "32"),
1454 v2i32, v2i32, OpNode, Commutable>;
1456 // 128-bit vector types.
1457 def v16i8 : N3VQ<op24, op23, 0b00, op11_8, op4, itinQ16,
1458 OpcodeStr, !strconcat(Dt, "8"),
1459 v16i8, v16i8, OpNode, Commutable>;
1460 def v8i16 : N3VQ<op24, op23, 0b01, op11_8, op4, itinQ16,
1461 OpcodeStr, !strconcat(Dt, "16"),
1462 v8i16, v8i16, OpNode, Commutable>;
1463 def v4i32 : N3VQ<op24, op23, 0b10, op11_8, op4, itinQ32,
1464 OpcodeStr, !strconcat(Dt, "32"),
1465 v4i32, v4i32, OpNode, Commutable>;
1468 multiclass N3VSL_HS<bits<4> op11_8, string OpcodeStr, string Dt, SDNode ShOp> {
1469 def v4i16 : N3VDSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
1471 def v2i32 : N3VDSL<0b10, op11_8, IIC_VMULi32D, OpcodeStr, !strconcat(Dt,"32"),
1473 def v8i16 : N3VQSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
1474 v8i16, v4i16, ShOp>;
1475 def v4i32 : N3VQSL<0b10, op11_8, IIC_VMULi32Q, OpcodeStr, !strconcat(Dt,"32"),
1476 v4i32, v2i32, ShOp>;
1479 // ....then also with element size 64 bits:
1480 multiclass N3V_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1481 InstrItinClass itinD, InstrItinClass itinQ,
1482 string OpcodeStr, string Dt,
1483 SDNode OpNode, bit Commutable = 0>
1484 : N3V_QHS<op24, op23, op11_8, op4, itinD, itinD, itinQ, itinQ,
1485 OpcodeStr, Dt, OpNode, Commutable> {
1486 def v1i64 : N3VD<op24, op23, 0b11, op11_8, op4, itinD,
1487 OpcodeStr, !strconcat(Dt, "64"),
1488 v1i64, v1i64, OpNode, Commutable>;
1489 def v2i64 : N3VQ<op24, op23, 0b11, op11_8, op4, itinQ,
1490 OpcodeStr, !strconcat(Dt, "64"),
1491 v2i64, v2i64, OpNode, Commutable>;
1495 // Neon Narrowing 2-register vector intrinsics,
1496 // source operand element sizes of 16, 32 and 64 bits:
1497 multiclass N2VNInt_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1498 bits<5> op11_7, bit op6, bit op4,
1499 InstrItinClass itin, string OpcodeStr, string Dt,
1501 def v8i8 : N2VNInt<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
1502 itin, OpcodeStr, !strconcat(Dt, "16"),
1503 v8i8, v8i16, IntOp>;
1504 def v4i16 : N2VNInt<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
1505 itin, OpcodeStr, !strconcat(Dt, "32"),
1506 v4i16, v4i32, IntOp>;
1507 def v2i32 : N2VNInt<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
1508 itin, OpcodeStr, !strconcat(Dt, "64"),
1509 v2i32, v2i64, IntOp>;
1513 // Neon Lengthening 2-register vector intrinsic (currently specific to VMOVL).
1514 // source operand element sizes of 16, 32 and 64 bits:
1515 multiclass N2VLInt_QHS<bits<2> op24_23, bits<5> op11_7, bit op6, bit op4,
1516 string OpcodeStr, string Dt, Intrinsic IntOp> {
1517 def v8i16 : N2VLInt<op24_23, 0b00, 0b10, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
1518 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
1519 def v4i32 : N2VLInt<op24_23, 0b01, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
1520 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
1521 def v2i64 : N2VLInt<op24_23, 0b10, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
1522 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1526 // Neon 3-register vector intrinsics.
1528 // First with only element sizes of 16 and 32 bits:
1529 multiclass N3VInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
1530 InstrItinClass itinD16, InstrItinClass itinD32,
1531 InstrItinClass itinQ16, InstrItinClass itinQ32,
1532 string OpcodeStr, string Dt,
1533 Intrinsic IntOp, bit Commutable = 0> {
1534 // 64-bit vector types.
1535 def v4i16 : N3VDInt<op24, op23, 0b01, op11_8, op4, itinD16,
1536 OpcodeStr, !strconcat(Dt, "16"),
1537 v4i16, v4i16, IntOp, Commutable>;
1538 def v2i32 : N3VDInt<op24, op23, 0b10, op11_8, op4, itinD32,
1539 OpcodeStr, !strconcat(Dt, "32"),
1540 v2i32, v2i32, IntOp, Commutable>;
1542 // 128-bit vector types.
1543 def v8i16 : N3VQInt<op24, op23, 0b01, op11_8, op4, itinQ16,
1544 OpcodeStr, !strconcat(Dt, "16"),
1545 v8i16, v8i16, IntOp, Commutable>;
1546 def v4i32 : N3VQInt<op24, op23, 0b10, op11_8, op4, itinQ32,
1547 OpcodeStr, !strconcat(Dt, "32"),
1548 v4i32, v4i32, IntOp, Commutable>;
1551 multiclass N3VIntSL_HS<bits<4> op11_8,
1552 InstrItinClass itinD16, InstrItinClass itinD32,
1553 InstrItinClass itinQ16, InstrItinClass itinQ32,
1554 string OpcodeStr, string Dt, Intrinsic IntOp> {
1555 def v4i16 : N3VDIntSL16<0b01, op11_8, itinD16,
1556 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp>;
1557 def v2i32 : N3VDIntSL<0b10, op11_8, itinD32,
1558 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp>;
1559 def v8i16 : N3VQIntSL16<0b01, op11_8, itinQ16,
1560 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16, IntOp>;
1561 def v4i32 : N3VQIntSL<0b10, op11_8, itinQ32,
1562 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32, IntOp>;
1565 // ....then also with element size of 8 bits:
1566 multiclass N3VInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1567 InstrItinClass itinD16, InstrItinClass itinD32,
1568 InstrItinClass itinQ16, InstrItinClass itinQ32,
1569 string OpcodeStr, string Dt,
1570 Intrinsic IntOp, bit Commutable = 0>
1571 : N3VInt_HS<op24, op23, op11_8, op4, itinD16, itinD32, itinQ16, itinQ32,
1572 OpcodeStr, Dt, IntOp, Commutable> {
1573 def v8i8 : N3VDInt<op24, op23, 0b00, op11_8, op4, itinD16,
1574 OpcodeStr, !strconcat(Dt, "8"),
1575 v8i8, v8i8, IntOp, Commutable>;
1576 def v16i8 : N3VQInt<op24, op23, 0b00, op11_8, op4, itinQ16,
1577 OpcodeStr, !strconcat(Dt, "8"),
1578 v16i8, v16i8, IntOp, Commutable>;
1581 // ....then also with element size of 64 bits:
1582 multiclass N3VInt_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1583 InstrItinClass itinD16, InstrItinClass itinD32,
1584 InstrItinClass itinQ16, InstrItinClass itinQ32,
1585 string OpcodeStr, string Dt,
1586 Intrinsic IntOp, bit Commutable = 0>
1587 : N3VInt_QHS<op24, op23, op11_8, op4, itinD16, itinD32, itinQ16, itinQ32,
1588 OpcodeStr, Dt, IntOp, Commutable> {
1589 def v1i64 : N3VDInt<op24, op23, 0b11, op11_8, op4, itinD32,
1590 OpcodeStr, !strconcat(Dt, "64"),
1591 v1i64, v1i64, IntOp, Commutable>;
1592 def v2i64 : N3VQInt<op24, op23, 0b11, op11_8, op4, itinQ32,
1593 OpcodeStr, !strconcat(Dt, "64"),
1594 v2i64, v2i64, IntOp, Commutable>;
1597 // N3VSh_QHSD is similar to N3VInt_QHSD, except that it is for 3-Register Vector
1598 // Shift Instructions (N3RegVShFrm), which do not follow the N3RegFrm's operand
1599 // order of D:Vd N:Vn M:Vm.
1601 // The operand order of N3RegVShFrm is D:Vd M:Vm N:Vn (notice that M:Vm is the
1602 // first src operand).
1603 class N3VDSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1604 InstrItinClass itin, string OpcodeStr, string Dt,
1605 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
1606 : N3Vf<op24, op23, op21_20, op11_8, 0, op4,
1607 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegVShFrm,
1608 itin, OpcodeStr, Dt, "$dst, $src1, $src2", "",
1609 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src1), (OpTy DPR:$src2))))]> {
1610 let isCommutable = Commutable;
1612 class N3VQSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1613 InstrItinClass itin, string OpcodeStr, string Dt,
1614 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
1615 : N3Vf<op24, op23, op21_20, op11_8, 1, op4,
1616 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), N3RegVShFrm,
1617 itin, OpcodeStr, Dt, "$dst, $src1, $src2", "",
1618 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src1), (OpTy QPR:$src2))))]> {
1619 let isCommutable = Commutable;
1621 multiclass N3VSh_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1622 InstrItinClass itinD16, InstrItinClass itinD32,
1623 InstrItinClass itinQ16, InstrItinClass itinQ32,
1624 string OpcodeStr, string Dt,
1625 Intrinsic IntOp, bit Commutable> {
1626 def v4i16 : N3VDSh<op24, op23, 0b01, op11_8, op4, itinD16,
1627 OpcodeStr, !strconcat(Dt, "16"),
1628 v4i16, v4i16, IntOp, Commutable>;
1629 def v2i32 : N3VDSh<op24, op23, 0b10, op11_8, op4, itinD32,
1630 OpcodeStr, !strconcat(Dt, "32"),
1631 v2i32, v2i32, IntOp, Commutable>;
1632 def v8i16 : N3VQSh<op24, op23, 0b01, op11_8, op4, itinQ16,
1633 OpcodeStr, !strconcat(Dt, "16"),
1634 v8i16, v8i16, IntOp, Commutable>;
1635 def v4i32 : N3VQSh<op24, op23, 0b10, op11_8, op4, itinQ32,
1636 OpcodeStr, !strconcat(Dt, "32"),
1637 v4i32, v4i32, IntOp, Commutable>;
1638 def v8i8 : N3VDSh<op24, op23, 0b00, op11_8, op4, itinD16,
1639 OpcodeStr, !strconcat(Dt, "8"),
1640 v8i8, v8i8, IntOp, Commutable>;
1641 def v16i8 : N3VQSh<op24, op23, 0b00, op11_8, op4, itinQ16,
1642 OpcodeStr, !strconcat(Dt, "8"),
1643 v16i8, v16i8, IntOp, Commutable>;
1644 def v1i64 : N3VDSh<op24, op23, 0b11, op11_8, op4,
1645 itinD32, OpcodeStr, !strconcat(Dt, "64"),
1646 v1i64, v1i64, IntOp, Commutable>;
1647 def v2i64 : N3VQSh<op24, op23, 0b11, op11_8, op4,
1648 itinQ32, OpcodeStr, !strconcat(Dt, "64"),
1649 v2i64, v2i64, IntOp, Commutable>;
1652 // Neon Narrowing 3-register vector intrinsics,
1653 // source operand element sizes of 16, 32 and 64 bits:
1654 multiclass N3VNInt_HSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1655 string OpcodeStr, string Dt,
1656 Intrinsic IntOp, bit Commutable = 0> {
1657 def v8i8 : N3VNInt<op24, op23, 0b00, op11_8, op4,
1658 OpcodeStr, !strconcat(Dt, "16"),
1659 v8i8, v8i16, IntOp, Commutable>;
1660 def v4i16 : N3VNInt<op24, op23, 0b01, op11_8, op4,
1661 OpcodeStr, !strconcat(Dt, "32"),
1662 v4i16, v4i32, IntOp, Commutable>;
1663 def v2i32 : N3VNInt<op24, op23, 0b10, op11_8, op4,
1664 OpcodeStr, !strconcat(Dt, "64"),
1665 v2i32, v2i64, IntOp, Commutable>;
1669 // Neon Long 3-register vector intrinsics.
1671 // First with only element sizes of 16 and 32 bits:
1672 multiclass N3VLInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
1673 InstrItinClass itin, string OpcodeStr, string Dt,
1674 Intrinsic IntOp, bit Commutable = 0> {
1675 def v4i32 : N3VLInt<op24, op23, 0b01, op11_8, op4, itin,
1676 OpcodeStr, !strconcat(Dt, "16"),
1677 v4i32, v4i16, IntOp, Commutable>;
1678 def v2i64 : N3VLInt<op24, op23, 0b10, op11_8, op4, itin,
1679 OpcodeStr, !strconcat(Dt, "32"),
1680 v2i64, v2i32, IntOp, Commutable>;
1683 multiclass N3VLIntSL_HS<bit op24, bits<4> op11_8,
1684 InstrItinClass itin, string OpcodeStr, string Dt,
1686 def v4i16 : N3VLIntSL16<op24, 0b01, op11_8, itin,
1687 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
1688 def v2i32 : N3VLIntSL<op24, 0b10, op11_8, itin,
1689 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1692 // ....then also with element size of 8 bits:
1693 multiclass N3VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1694 InstrItinClass itin, string OpcodeStr, string Dt,
1695 Intrinsic IntOp, bit Commutable = 0>
1696 : N3VLInt_HS<op24, op23, op11_8, op4, itin, OpcodeStr, Dt,
1697 IntOp, Commutable> {
1698 def v8i16 : N3VLInt<op24, op23, 0b00, op11_8, op4, itin,
1699 OpcodeStr, !strconcat(Dt, "8"),
1700 v8i16, v8i8, IntOp, Commutable>;
1704 // Neon Wide 3-register vector intrinsics,
1705 // source operand element sizes of 8, 16 and 32 bits:
1706 multiclass N3VWInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1707 string OpcodeStr, string Dt,
1708 Intrinsic IntOp, bit Commutable = 0> {
1709 def v8i16 : N3VWInt<op24, op23, 0b00, op11_8, op4,
1710 OpcodeStr, !strconcat(Dt, "8"),
1711 v8i16, v8i8, IntOp, Commutable>;
1712 def v4i32 : N3VWInt<op24, op23, 0b01, op11_8, op4,
1713 OpcodeStr, !strconcat(Dt, "16"),
1714 v4i32, v4i16, IntOp, Commutable>;
1715 def v2i64 : N3VWInt<op24, op23, 0b10, op11_8, op4,
1716 OpcodeStr, !strconcat(Dt, "32"),
1717 v2i64, v2i32, IntOp, Commutable>;
1721 // Neon Multiply-Op vector operations,
1722 // element sizes of 8, 16 and 32 bits:
1723 multiclass N3VMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1724 InstrItinClass itinD16, InstrItinClass itinD32,
1725 InstrItinClass itinQ16, InstrItinClass itinQ32,
1726 string OpcodeStr, string Dt, SDNode OpNode> {
1727 // 64-bit vector types.
1728 def v8i8 : N3VDMulOp<op24, op23, 0b00, op11_8, op4, itinD16,
1729 OpcodeStr, !strconcat(Dt, "8"), v8i8, mul, OpNode>;
1730 def v4i16 : N3VDMulOp<op24, op23, 0b01, op11_8, op4, itinD16,
1731 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, OpNode>;
1732 def v2i32 : N3VDMulOp<op24, op23, 0b10, op11_8, op4, itinD32,
1733 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, OpNode>;
1735 // 128-bit vector types.
1736 def v16i8 : N3VQMulOp<op24, op23, 0b00, op11_8, op4, itinQ16,
1737 OpcodeStr, !strconcat(Dt, "8"), v16i8, mul, OpNode>;
1738 def v8i16 : N3VQMulOp<op24, op23, 0b01, op11_8, op4, itinQ16,
1739 OpcodeStr, !strconcat(Dt, "16"), v8i16, mul, OpNode>;
1740 def v4i32 : N3VQMulOp<op24, op23, 0b10, op11_8, op4, itinQ32,
1741 OpcodeStr, !strconcat(Dt, "32"), v4i32, mul, OpNode>;
1744 multiclass N3VMulOpSL_HS<bits<4> op11_8,
1745 InstrItinClass itinD16, InstrItinClass itinD32,
1746 InstrItinClass itinQ16, InstrItinClass itinQ32,
1747 string OpcodeStr, string Dt, SDNode ShOp> {
1748 def v4i16 : N3VDMulOpSL16<0b01, op11_8, itinD16,
1749 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, ShOp>;
1750 def v2i32 : N3VDMulOpSL<0b10, op11_8, itinD32,
1751 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, ShOp>;
1752 def v8i16 : N3VQMulOpSL16<0b01, op11_8, itinQ16,
1753 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16,
1755 def v4i32 : N3VQMulOpSL<0b10, op11_8, itinQ32,
1756 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32,
1760 // Neon 3-argument intrinsics,
1761 // element sizes of 8, 16 and 32 bits:
1762 multiclass N3VInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1763 string OpcodeStr, string Dt, Intrinsic IntOp> {
1764 // 64-bit vector types.
1765 def v8i8 : N3VDInt3<op24, op23, 0b00, op11_8, op4, IIC_VMACi16D,
1766 OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
1767 def v4i16 : N3VDInt3<op24, op23, 0b01, op11_8, op4, IIC_VMACi16D,
1768 OpcodeStr, !strconcat(Dt, "16"), v4i16, v4i16, IntOp>;
1769 def v2i32 : N3VDInt3<op24, op23, 0b10, op11_8, op4, IIC_VMACi32D,
1770 OpcodeStr, !strconcat(Dt, "32"), v2i32, v2i32, IntOp>;
1772 // 128-bit vector types.
1773 def v16i8 : N3VQInt3<op24, op23, 0b00, op11_8, op4, IIC_VMACi16Q,
1774 OpcodeStr, !strconcat(Dt, "8"), v16i8, v16i8, IntOp>;
1775 def v8i16 : N3VQInt3<op24, op23, 0b01, op11_8, op4, IIC_VMACi16Q,
1776 OpcodeStr, !strconcat(Dt, "16"), v8i16, v8i16, IntOp>;
1777 def v4i32 : N3VQInt3<op24, op23, 0b10, op11_8, op4, IIC_VMACi32Q,
1778 OpcodeStr, !strconcat(Dt, "32"), v4i32, v4i32, IntOp>;
1782 // Neon Long 3-argument intrinsics.
1784 // First with only element sizes of 16 and 32 bits:
1785 multiclass N3VLInt3_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
1786 string OpcodeStr, string Dt, Intrinsic IntOp> {
1787 def v4i32 : N3VLInt3<op24, op23, 0b01, op11_8, op4, IIC_VMACi16D,
1788 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
1789 def v2i64 : N3VLInt3<op24, op23, 0b10, op11_8, op4, IIC_VMACi16D,
1790 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1793 multiclass N3VLInt3SL_HS<bit op24, bits<4> op11_8,
1794 string OpcodeStr, string Dt, Intrinsic IntOp> {
1795 def v4i16 : N3VLInt3SL16<op24, 0b01, op11_8, IIC_VMACi16D,
1796 OpcodeStr, !strconcat(Dt,"16"), v4i32, v4i16, IntOp>;
1797 def v2i32 : N3VLInt3SL<op24, 0b10, op11_8, IIC_VMACi32D,
1798 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1801 // ....then also with element size of 8 bits:
1802 multiclass N3VLInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1803 string OpcodeStr, string Dt, Intrinsic IntOp>
1804 : N3VLInt3_HS<op24, op23, op11_8, op4, OpcodeStr, Dt, IntOp> {
1805 def v8i16 : N3VLInt3<op24, op23, 0b00, op11_8, op4, IIC_VMACi16D,
1806 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
1810 // Neon 2-register vector intrinsics,
1811 // element sizes of 8, 16 and 32 bits:
1812 multiclass N2VInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1813 bits<5> op11_7, bit op4,
1814 InstrItinClass itinD, InstrItinClass itinQ,
1815 string OpcodeStr, string Dt, Intrinsic IntOp> {
1816 // 64-bit vector types.
1817 def v8i8 : N2VDInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1818 itinD, OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
1819 def v4i16 : N2VDInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1820 itinD, OpcodeStr, !strconcat(Dt, "16"),v4i16,v4i16,IntOp>;
1821 def v2i32 : N2VDInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1822 itinD, OpcodeStr, !strconcat(Dt, "32"),v2i32,v2i32,IntOp>;
1824 // 128-bit vector types.
1825 def v16i8 : N2VQInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1826 itinQ, OpcodeStr, !strconcat(Dt, "8"), v16i8,v16i8,IntOp>;
1827 def v8i16 : N2VQInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1828 itinQ, OpcodeStr, !strconcat(Dt, "16"),v8i16,v8i16,IntOp>;
1829 def v4i32 : N2VQInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1830 itinQ, OpcodeStr, !strconcat(Dt, "32"),v4i32,v4i32,IntOp>;
1834 // Neon Pairwise long 2-register intrinsics,
1835 // element sizes of 8, 16 and 32 bits:
1836 multiclass N2VPLInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1837 bits<5> op11_7, bit op4,
1838 string OpcodeStr, string Dt, Intrinsic IntOp> {
1839 // 64-bit vector types.
1840 def v8i8 : N2VDPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1841 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
1842 def v4i16 : N2VDPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1843 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
1844 def v2i32 : N2VDPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1845 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
1847 // 128-bit vector types.
1848 def v16i8 : N2VQPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1849 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
1850 def v8i16 : N2VQPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1851 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
1852 def v4i32 : N2VQPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1853 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
1857 // Neon Pairwise long 2-register accumulate intrinsics,
1858 // element sizes of 8, 16 and 32 bits:
1859 multiclass N2VPLInt2_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1860 bits<5> op11_7, bit op4,
1861 string OpcodeStr, string Dt, Intrinsic IntOp> {
1862 // 64-bit vector types.
1863 def v8i8 : N2VDPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1864 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
1865 def v4i16 : N2VDPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1866 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
1867 def v2i32 : N2VDPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1868 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
1870 // 128-bit vector types.
1871 def v16i8 : N2VQPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1872 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
1873 def v8i16 : N2VQPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1874 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
1875 def v4i32 : N2VQPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1876 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
1880 // Neon 2-register vector shift by immediate,
1881 // with f of either N2RegVShLFrm or N2RegVShRFrm
1882 // element sizes of 8, 16, 32 and 64 bits:
1883 multiclass N2VSh_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1884 InstrItinClass itin, string OpcodeStr, string Dt,
1885 SDNode OpNode, Format f> {
1886 // 64-bit vector types.
1887 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
1888 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
1889 let Inst{21-19} = 0b001; // imm6 = 001xxx
1891 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
1892 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
1893 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1895 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
1896 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
1897 let Inst{21} = 0b1; // imm6 = 1xxxxx
1899 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, f, itin,
1900 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
1903 // 128-bit vector types.
1904 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
1905 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
1906 let Inst{21-19} = 0b001; // imm6 = 001xxx
1908 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
1909 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
1910 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1912 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
1913 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
1914 let Inst{21} = 0b1; // imm6 = 1xxxxx
1916 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, f, itin,
1917 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
1921 // Neon Shift-Accumulate vector operations,
1922 // element sizes of 8, 16, 32 and 64 bits:
1923 multiclass N2VShAdd_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1924 string OpcodeStr, string Dt, SDNode ShOp> {
1925 // 64-bit vector types.
1926 def v8i8 : N2VDShAdd<op24, op23, op11_8, 0, op4,
1927 OpcodeStr, !strconcat(Dt, "8"), v8i8, ShOp> {
1928 let Inst{21-19} = 0b001; // imm6 = 001xxx
1930 def v4i16 : N2VDShAdd<op24, op23, op11_8, 0, op4,
1931 OpcodeStr, !strconcat(Dt, "16"), v4i16, ShOp> {
1932 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1934 def v2i32 : N2VDShAdd<op24, op23, op11_8, 0, op4,
1935 OpcodeStr, !strconcat(Dt, "32"), v2i32, ShOp> {
1936 let Inst{21} = 0b1; // imm6 = 1xxxxx
1938 def v1i64 : N2VDShAdd<op24, op23, op11_8, 1, op4,
1939 OpcodeStr, !strconcat(Dt, "64"), v1i64, ShOp>;
1942 // 128-bit vector types.
1943 def v16i8 : N2VQShAdd<op24, op23, op11_8, 0, op4,
1944 OpcodeStr, !strconcat(Dt, "8"), v16i8, ShOp> {
1945 let Inst{21-19} = 0b001; // imm6 = 001xxx
1947 def v8i16 : N2VQShAdd<op24, op23, op11_8, 0, op4,
1948 OpcodeStr, !strconcat(Dt, "16"), v8i16, ShOp> {
1949 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1951 def v4i32 : N2VQShAdd<op24, op23, op11_8, 0, op4,
1952 OpcodeStr, !strconcat(Dt, "32"), v4i32, ShOp> {
1953 let Inst{21} = 0b1; // imm6 = 1xxxxx
1955 def v2i64 : N2VQShAdd<op24, op23, op11_8, 1, op4,
1956 OpcodeStr, !strconcat(Dt, "64"), v2i64, ShOp>;
1961 // Neon Shift-Insert vector operations,
1962 // with f of either N2RegVShLFrm or N2RegVShRFrm
1963 // element sizes of 8, 16, 32 and 64 bits:
1964 multiclass N2VShIns_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1965 string OpcodeStr, SDNode ShOp,
1967 // 64-bit vector types.
1968 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4,
1969 f, OpcodeStr, "8", v8i8, ShOp> {
1970 let Inst{21-19} = 0b001; // imm6 = 001xxx
1972 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4,
1973 f, OpcodeStr, "16", v4i16, ShOp> {
1974 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1976 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4,
1977 f, OpcodeStr, "32", v2i32, ShOp> {
1978 let Inst{21} = 0b1; // imm6 = 1xxxxx
1980 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4,
1981 f, OpcodeStr, "64", v1i64, ShOp>;
1984 // 128-bit vector types.
1985 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4,
1986 f, OpcodeStr, "8", v16i8, ShOp> {
1987 let Inst{21-19} = 0b001; // imm6 = 001xxx
1989 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4,
1990 f, OpcodeStr, "16", v8i16, ShOp> {
1991 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1993 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4,
1994 f, OpcodeStr, "32", v4i32, ShOp> {
1995 let Inst{21} = 0b1; // imm6 = 1xxxxx
1997 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4,
1998 f, OpcodeStr, "64", v2i64, ShOp>;
2002 // Neon Shift Long operations,
2003 // element sizes of 8, 16, 32 bits:
2004 multiclass N2VLSh_QHS<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
2005 bit op4, string OpcodeStr, string Dt, SDNode OpNode> {
2006 def v8i16 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
2007 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode> {
2008 let Inst{21-19} = 0b001; // imm6 = 001xxx
2010 def v4i32 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
2011 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode> {
2012 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2014 def v2i64 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
2015 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode> {
2016 let Inst{21} = 0b1; // imm6 = 1xxxxx
2020 // Neon Shift Narrow operations,
2021 // element sizes of 16, 32, 64 bits:
2022 multiclass N2VNSh_HSD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
2023 bit op4, InstrItinClass itin, string OpcodeStr, string Dt,
2025 def v8i8 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
2026 OpcodeStr, !strconcat(Dt, "16"), v8i8, v8i16, OpNode> {
2027 let Inst{21-19} = 0b001; // imm6 = 001xxx
2029 def v4i16 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
2030 OpcodeStr, !strconcat(Dt, "32"), v4i16, v4i32, OpNode> {
2031 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2033 def v2i32 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
2034 OpcodeStr, !strconcat(Dt, "64"), v2i32, v2i64, OpNode> {
2035 let Inst{21} = 0b1; // imm6 = 1xxxxx
2039 //===----------------------------------------------------------------------===//
2040 // Instruction Definitions.
2041 //===----------------------------------------------------------------------===//
2043 // Vector Add Operations.
2045 // VADD : Vector Add (integer and floating-point)
2046 defm VADD : N3V_QHSD<0, 0, 0b1000, 0, IIC_VBINiD, IIC_VBINiQ, "vadd", "i",
2048 def VADDfd : N3VD<0, 0, 0b00, 0b1101, 0, IIC_VBIND, "vadd", "f32",
2049 v2f32, v2f32, fadd, 1>;
2050 def VADDfq : N3VQ<0, 0, 0b00, 0b1101, 0, IIC_VBINQ, "vadd", "f32",
2051 v4f32, v4f32, fadd, 1>;
2052 // VADDL : Vector Add Long (Q = D + D)
2053 defm VADDLs : N3VLInt_QHS<0,1,0b0000,0, IIC_VSHLiD, "vaddl", "s",
2054 int_arm_neon_vaddls, 1>;
2055 defm VADDLu : N3VLInt_QHS<1,1,0b0000,0, IIC_VSHLiD, "vaddl", "u",
2056 int_arm_neon_vaddlu, 1>;
2057 // VADDW : Vector Add Wide (Q = Q + D)
2058 defm VADDWs : N3VWInt_QHS<0,1,0b0001,0, "vaddw", "s", int_arm_neon_vaddws, 0>;
2059 defm VADDWu : N3VWInt_QHS<1,1,0b0001,0, "vaddw", "u", int_arm_neon_vaddwu, 0>;
2060 // VHADD : Vector Halving Add
2061 defm VHADDs : N3VInt_QHS<0,0,0b0000,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2062 IIC_VBINi4Q, "vhadd", "s", int_arm_neon_vhadds, 1>;
2063 defm VHADDu : N3VInt_QHS<1,0,0b0000,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2064 IIC_VBINi4Q, "vhadd", "u", int_arm_neon_vhaddu, 1>;
2065 // VRHADD : Vector Rounding Halving Add
2066 defm VRHADDs : N3VInt_QHS<0,0,0b0001,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2067 IIC_VBINi4Q, "vrhadd", "s", int_arm_neon_vrhadds, 1>;
2068 defm VRHADDu : N3VInt_QHS<1,0,0b0001,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2069 IIC_VBINi4Q, "vrhadd", "u", int_arm_neon_vrhaddu, 1>;
2070 // VQADD : Vector Saturating Add
2071 defm VQADDs : N3VInt_QHSD<0,0,0b0000,1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2072 IIC_VBINi4Q, "vqadd", "s", int_arm_neon_vqadds, 1>;
2073 defm VQADDu : N3VInt_QHSD<1,0,0b0000,1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2074 IIC_VBINi4Q, "vqadd", "u", int_arm_neon_vqaddu, 1>;
2075 // VADDHN : Vector Add and Narrow Returning High Half (D = Q + Q)
2076 defm VADDHN : N3VNInt_HSD<0,1,0b0100,0, "vaddhn", "i",
2077 int_arm_neon_vaddhn, 1>;
2078 // VRADDHN : Vector Rounding Add and Narrow Returning High Half (D = Q + Q)
2079 defm VRADDHN : N3VNInt_HSD<1,1,0b0100,0, "vraddhn", "i",
2080 int_arm_neon_vraddhn, 1>;
2082 // Vector Multiply Operations.
2084 // VMUL : Vector Multiply (integer, polynomial and floating-point)
2085 defm VMUL : N3V_QHS<0, 0, 0b1001, 1, IIC_VMULi16D, IIC_VMULi32D,
2086 IIC_VMULi16Q, IIC_VMULi32Q, "vmul", "i", mul, 1>;
2087 def VMULpd : N3VDInt<1, 0, 0b00, 0b1001, 1, IIC_VMULi16D, "vmul", "p8",
2088 v8i8, v8i8, int_arm_neon_vmulp, 1>;
2089 def VMULpq : N3VQInt<1, 0, 0b00, 0b1001, 1, IIC_VMULi16Q, "vmul", "p8",
2090 v16i8, v16i8, int_arm_neon_vmulp, 1>;
2091 def VMULfd : N3VD<1, 0, 0b00, 0b1101, 1, IIC_VBIND, "vmul", "f32",
2092 v2f32, v2f32, fmul, 1>;
2093 def VMULfq : N3VQ<1, 0, 0b00, 0b1101, 1, IIC_VBINQ, "vmul", "f32",
2094 v4f32, v4f32, fmul, 1>;
2095 defm VMULsl : N3VSL_HS<0b1000, "vmul", "i", mul>;
2096 def VMULslfd : N3VDSL<0b10, 0b1001, IIC_VBIND, "vmul", "f32", v2f32, fmul>;
2097 def VMULslfq : N3VQSL<0b10, 0b1001, IIC_VBINQ, "vmul", "f32", v4f32,
2100 def : Pat<(v8i16 (mul (v8i16 QPR:$src1),
2101 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
2102 (v8i16 (VMULslv8i16 (v8i16 QPR:$src1),
2103 (v4i16 (EXTRACT_SUBREG QPR:$src2,
2104 (DSubReg_i16_reg imm:$lane))),
2105 (SubReg_i16_lane imm:$lane)))>;
2106 def : Pat<(v4i32 (mul (v4i32 QPR:$src1),
2107 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
2108 (v4i32 (VMULslv4i32 (v4i32 QPR:$src1),
2109 (v2i32 (EXTRACT_SUBREG QPR:$src2,
2110 (DSubReg_i32_reg imm:$lane))),
2111 (SubReg_i32_lane imm:$lane)))>;
2112 def : Pat<(v4f32 (fmul (v4f32 QPR:$src1),
2113 (v4f32 (NEONvduplane (v4f32 QPR:$src2), imm:$lane)))),
2114 (v4f32 (VMULslfq (v4f32 QPR:$src1),
2115 (v2f32 (EXTRACT_SUBREG QPR:$src2,
2116 (DSubReg_i32_reg imm:$lane))),
2117 (SubReg_i32_lane imm:$lane)))>;
2119 // VQDMULH : Vector Saturating Doubling Multiply Returning High Half
2120 defm VQDMULH : N3VInt_HS<0, 0, 0b1011, 0, IIC_VMULi16D, IIC_VMULi32D,
2121 IIC_VMULi16Q, IIC_VMULi32Q,
2122 "vqdmulh", "s", int_arm_neon_vqdmulh, 1>;
2123 defm VQDMULHsl: N3VIntSL_HS<0b1100, IIC_VMULi16D, IIC_VMULi32D,
2124 IIC_VMULi16Q, IIC_VMULi32Q,
2125 "vqdmulh", "s", int_arm_neon_vqdmulh>;
2126 def : Pat<(v8i16 (int_arm_neon_vqdmulh (v8i16 QPR:$src1),
2127 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
2129 (v8i16 (VQDMULHslv8i16 (v8i16 QPR:$src1),
2130 (v4i16 (EXTRACT_SUBREG QPR:$src2,
2131 (DSubReg_i16_reg imm:$lane))),
2132 (SubReg_i16_lane imm:$lane)))>;
2133 def : Pat<(v4i32 (int_arm_neon_vqdmulh (v4i32 QPR:$src1),
2134 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
2136 (v4i32 (VQDMULHslv4i32 (v4i32 QPR:$src1),
2137 (v2i32 (EXTRACT_SUBREG QPR:$src2,
2138 (DSubReg_i32_reg imm:$lane))),
2139 (SubReg_i32_lane imm:$lane)))>;
2141 // VQRDMULH : Vector Rounding Saturating Doubling Multiply Returning High Half
2142 defm VQRDMULH : N3VInt_HS<1, 0, 0b1011, 0, IIC_VMULi16D, IIC_VMULi32D,
2143 IIC_VMULi16Q, IIC_VMULi32Q,
2144 "vqrdmulh", "s", int_arm_neon_vqrdmulh, 1>;
2145 defm VQRDMULHsl : N3VIntSL_HS<0b1101, IIC_VMULi16D, IIC_VMULi32D,
2146 IIC_VMULi16Q, IIC_VMULi32Q,
2147 "vqrdmulh", "s", int_arm_neon_vqrdmulh>;
2148 def : Pat<(v8i16 (int_arm_neon_vqrdmulh (v8i16 QPR:$src1),
2149 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
2151 (v8i16 (VQRDMULHslv8i16 (v8i16 QPR:$src1),
2152 (v4i16 (EXTRACT_SUBREG QPR:$src2,
2153 (DSubReg_i16_reg imm:$lane))),
2154 (SubReg_i16_lane imm:$lane)))>;
2155 def : Pat<(v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$src1),
2156 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
2158 (v4i32 (VQRDMULHslv4i32 (v4i32 QPR:$src1),
2159 (v2i32 (EXTRACT_SUBREG QPR:$src2,
2160 (DSubReg_i32_reg imm:$lane))),
2161 (SubReg_i32_lane imm:$lane)))>;
2163 // VMULL : Vector Multiply Long (integer and polynomial) (Q = D * D)
2164 defm VMULLs : N3VLInt_QHS<0,1,0b1100,0, IIC_VMULi16D, "vmull", "s",
2165 int_arm_neon_vmulls, 1>;
2166 defm VMULLu : N3VLInt_QHS<1,1,0b1100,0, IIC_VMULi16D, "vmull", "u",
2167 int_arm_neon_vmullu, 1>;
2168 def VMULLp : N3VLInt<0, 1, 0b00, 0b1110, 0, IIC_VMULi16D, "vmull", "p8",
2169 v8i16, v8i8, int_arm_neon_vmullp, 1>;
2170 defm VMULLsls : N3VLIntSL_HS<0, 0b1010, IIC_VMULi16D, "vmull", "s",
2171 int_arm_neon_vmulls>;
2172 defm VMULLslu : N3VLIntSL_HS<1, 0b1010, IIC_VMULi16D, "vmull", "u",
2173 int_arm_neon_vmullu>;
2175 // VQDMULL : Vector Saturating Doubling Multiply Long (Q = D * D)
2176 defm VQDMULL : N3VLInt_HS<0,1,0b1101,0, IIC_VMULi16D, "vqdmull", "s",
2177 int_arm_neon_vqdmull, 1>;
2178 defm VQDMULLsl: N3VLIntSL_HS<0, 0b1011, IIC_VMULi16D, "vqdmull", "s",
2179 int_arm_neon_vqdmull>;
2181 // Vector Multiply-Accumulate and Multiply-Subtract Operations.
2183 // VMLA : Vector Multiply Accumulate (integer and floating-point)
2184 defm VMLA : N3VMulOp_QHS<0, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
2185 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
2186 def VMLAfd : N3VDMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACD, "vmla", "f32",
2188 def VMLAfq : N3VQMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACQ, "vmla", "f32",
2190 defm VMLAsl : N3VMulOpSL_HS<0b0000, IIC_VMACi16D, IIC_VMACi32D,
2191 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
2192 def VMLAslfd : N3VDMulOpSL<0b10, 0b0001, IIC_VMACD, "vmla", "f32",
2194 def VMLAslfq : N3VQMulOpSL<0b10, 0b0001, IIC_VMACQ, "vmla", "f32",
2195 v4f32, v2f32, fmul, fadd>;
2197 def : Pat<(v8i16 (add (v8i16 QPR:$src1),
2198 (mul (v8i16 QPR:$src2),
2199 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
2200 (v8i16 (VMLAslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
2201 (v4i16 (EXTRACT_SUBREG QPR:$src3,
2202 (DSubReg_i16_reg imm:$lane))),
2203 (SubReg_i16_lane imm:$lane)))>;
2205 def : Pat<(v4i32 (add (v4i32 QPR:$src1),
2206 (mul (v4i32 QPR:$src2),
2207 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
2208 (v4i32 (VMLAslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
2209 (v2i32 (EXTRACT_SUBREG QPR:$src3,
2210 (DSubReg_i32_reg imm:$lane))),
2211 (SubReg_i32_lane imm:$lane)))>;
2213 def : Pat<(v4f32 (fadd (v4f32 QPR:$src1),
2214 (fmul (v4f32 QPR:$src2),
2215 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
2216 (v4f32 (VMLAslfq (v4f32 QPR:$src1),
2218 (v2f32 (EXTRACT_SUBREG QPR:$src3,
2219 (DSubReg_i32_reg imm:$lane))),
2220 (SubReg_i32_lane imm:$lane)))>;
2222 // VMLAL : Vector Multiply Accumulate Long (Q += D * D)
2223 defm VMLALs : N3VLInt3_QHS<0,1,0b1000,0, "vmlal", "s", int_arm_neon_vmlals>;
2224 defm VMLALu : N3VLInt3_QHS<1,1,0b1000,0, "vmlal", "u", int_arm_neon_vmlalu>;
2226 defm VMLALsls : N3VLInt3SL_HS<0, 0b0010, "vmlal", "s", int_arm_neon_vmlals>;
2227 defm VMLALslu : N3VLInt3SL_HS<1, 0b0010, "vmlal", "u", int_arm_neon_vmlalu>;
2229 // VQDMLAL : Vector Saturating Doubling Multiply Accumulate Long (Q += D * D)
2230 defm VQDMLAL : N3VLInt3_HS<0, 1, 0b1001, 0, "vqdmlal", "s",
2231 int_arm_neon_vqdmlal>;
2232 defm VQDMLALsl: N3VLInt3SL_HS<0, 0b0011, "vqdmlal", "s", int_arm_neon_vqdmlal>;
2234 // VMLS : Vector Multiply Subtract (integer and floating-point)
2235 defm VMLS : N3VMulOp_QHS<1, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
2236 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
2237 def VMLSfd : N3VDMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACD, "vmls", "f32",
2239 def VMLSfq : N3VQMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACQ, "vmls", "f32",
2241 defm VMLSsl : N3VMulOpSL_HS<0b0100, IIC_VMACi16D, IIC_VMACi32D,
2242 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
2243 def VMLSslfd : N3VDMulOpSL<0b10, 0b0101, IIC_VMACD, "vmls", "f32",
2245 def VMLSslfq : N3VQMulOpSL<0b10, 0b0101, IIC_VMACQ, "vmls", "f32",
2246 v4f32, v2f32, fmul, fsub>;
2248 def : Pat<(v8i16 (sub (v8i16 QPR:$src1),
2249 (mul (v8i16 QPR:$src2),
2250 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
2251 (v8i16 (VMLSslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
2252 (v4i16 (EXTRACT_SUBREG QPR:$src3,
2253 (DSubReg_i16_reg imm:$lane))),
2254 (SubReg_i16_lane imm:$lane)))>;
2256 def : Pat<(v4i32 (sub (v4i32 QPR:$src1),
2257 (mul (v4i32 QPR:$src2),
2258 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
2259 (v4i32 (VMLSslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
2260 (v2i32 (EXTRACT_SUBREG QPR:$src3,
2261 (DSubReg_i32_reg imm:$lane))),
2262 (SubReg_i32_lane imm:$lane)))>;
2264 def : Pat<(v4f32 (fsub (v4f32 QPR:$src1),
2265 (fmul (v4f32 QPR:$src2),
2266 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
2267 (v4f32 (VMLSslfq (v4f32 QPR:$src1), (v4f32 QPR:$src2),
2268 (v2f32 (EXTRACT_SUBREG QPR:$src3,
2269 (DSubReg_i32_reg imm:$lane))),
2270 (SubReg_i32_lane imm:$lane)))>;
2272 // VMLSL : Vector Multiply Subtract Long (Q -= D * D)
2273 defm VMLSLs : N3VLInt3_QHS<0,1,0b1010,0, "vmlsl", "s", int_arm_neon_vmlsls>;
2274 defm VMLSLu : N3VLInt3_QHS<1,1,0b1010,0, "vmlsl", "u", int_arm_neon_vmlslu>;
2276 defm VMLSLsls : N3VLInt3SL_HS<0, 0b0110, "vmlsl", "s", int_arm_neon_vmlsls>;
2277 defm VMLSLslu : N3VLInt3SL_HS<1, 0b0110, "vmlsl", "u", int_arm_neon_vmlslu>;
2279 // VQDMLSL : Vector Saturating Doubling Multiply Subtract Long (Q -= D * D)
2280 defm VQDMLSL : N3VLInt3_HS<0, 1, 0b1011, 0, "vqdmlsl", "s",
2281 int_arm_neon_vqdmlsl>;
2282 defm VQDMLSLsl: N3VLInt3SL_HS<0, 0b111, "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
2284 // Vector Subtract Operations.
2286 // VSUB : Vector Subtract (integer and floating-point)
2287 defm VSUB : N3V_QHSD<1, 0, 0b1000, 0, IIC_VSUBiD, IIC_VSUBiQ,
2288 "vsub", "i", sub, 0>;
2289 def VSUBfd : N3VD<0, 0, 0b10, 0b1101, 0, IIC_VBIND, "vsub", "f32",
2290 v2f32, v2f32, fsub, 0>;
2291 def VSUBfq : N3VQ<0, 0, 0b10, 0b1101, 0, IIC_VBINQ, "vsub", "f32",
2292 v4f32, v4f32, fsub, 0>;
2293 // VSUBL : Vector Subtract Long (Q = D - D)
2294 defm VSUBLs : N3VLInt_QHS<0,1,0b0010,0, IIC_VSHLiD, "vsubl", "s",
2295 int_arm_neon_vsubls, 1>;
2296 defm VSUBLu : N3VLInt_QHS<1,1,0b0010,0, IIC_VSHLiD, "vsubl", "u",
2297 int_arm_neon_vsublu, 1>;
2298 // VSUBW : Vector Subtract Wide (Q = Q - D)
2299 defm VSUBWs : N3VWInt_QHS<0,1,0b0011,0, "vsubw", "s", int_arm_neon_vsubws, 0>;
2300 defm VSUBWu : N3VWInt_QHS<1,1,0b0011,0, "vsubw", "u", int_arm_neon_vsubwu, 0>;
2301 // VHSUB : Vector Halving Subtract
2302 defm VHSUBs : N3VInt_QHS<0, 0, 0b0010, 0, IIC_VBINi4D, IIC_VBINi4D,
2303 IIC_VBINi4Q, IIC_VBINi4Q,
2304 "vhsub", "s", int_arm_neon_vhsubs, 0>;
2305 defm VHSUBu : N3VInt_QHS<1, 0, 0b0010, 0, IIC_VBINi4D, IIC_VBINi4D,
2306 IIC_VBINi4Q, IIC_VBINi4Q,
2307 "vhsub", "u", int_arm_neon_vhsubu, 0>;
2308 // VQSUB : Vector Saturing Subtract
2309 defm VQSUBs : N3VInt_QHSD<0, 0, 0b0010, 1, IIC_VBINi4D, IIC_VBINi4D,
2310 IIC_VBINi4Q, IIC_VBINi4Q,
2311 "vqsub", "s", int_arm_neon_vqsubs, 0>;
2312 defm VQSUBu : N3VInt_QHSD<1, 0, 0b0010, 1, IIC_VBINi4D, IIC_VBINi4D,
2313 IIC_VBINi4Q, IIC_VBINi4Q,
2314 "vqsub", "u", int_arm_neon_vqsubu, 0>;
2315 // VSUBHN : Vector Subtract and Narrow Returning High Half (D = Q - Q)
2316 defm VSUBHN : N3VNInt_HSD<0,1,0b0110,0, "vsubhn", "i",
2317 int_arm_neon_vsubhn, 0>;
2318 // VRSUBHN : Vector Rounding Subtract and Narrow Returning High Half (D=Q-Q)
2319 defm VRSUBHN : N3VNInt_HSD<1,1,0b0110,0, "vrsubhn", "i",
2320 int_arm_neon_vrsubhn, 0>;
2322 // Vector Comparisons.
2324 // VCEQ : Vector Compare Equal
2325 defm VCEQ : N3V_QHS<1, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2326 IIC_VBINi4Q, "vceq", "i", NEONvceq, 1>;
2327 def VCEQfd : N3VD<0,0,0b00,0b1110,0, IIC_VBIND, "vceq", "f32", v2i32, v2f32,
2329 def VCEQfq : N3VQ<0,0,0b00,0b1110,0, IIC_VBINQ, "vceq", "f32", v4i32, v4f32,
2331 // For disassembly only.
2332 defm VCEQz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00010, 0, "vceq", "i",
2335 // VCGE : Vector Compare Greater Than or Equal
2336 defm VCGEs : N3V_QHS<0, 0, 0b0011, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2337 IIC_VBINi4Q, "vcge", "s", NEONvcge, 0>;
2338 defm VCGEu : N3V_QHS<1, 0, 0b0011, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2339 IIC_VBINi4Q, "vcge", "u", NEONvcgeu, 0>;
2340 def VCGEfd : N3VD<1,0,0b00,0b1110,0, IIC_VBIND, "vcge", "f32", v2i32, v2f32,
2342 def VCGEfq : N3VQ<1,0,0b00,0b1110,0, IIC_VBINQ, "vcge", "f32", v4i32, v4f32,
2344 // For disassembly only.
2345 defm VCGEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00001, 0, "vcge", "s",
2347 // For disassembly only.
2348 defm VCLEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00011, 0, "vcle", "s",
2351 // VCGT : Vector Compare Greater Than
2352 defm VCGTs : N3V_QHS<0, 0, 0b0011, 0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2353 IIC_VBINi4Q, "vcgt", "s", NEONvcgt, 0>;
2354 defm VCGTu : N3V_QHS<1, 0, 0b0011, 0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2355 IIC_VBINi4Q, "vcgt", "u", NEONvcgtu, 0>;
2356 def VCGTfd : N3VD<1,0,0b10,0b1110,0, IIC_VBIND, "vcgt", "f32", v2i32, v2f32,
2358 def VCGTfq : N3VQ<1,0,0b10,0b1110,0, IIC_VBINQ, "vcgt", "f32", v4i32, v4f32,
2360 // For disassembly only.
2361 defm VCGTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00000, 0, "vcgt", "s",
2363 // For disassembly only.
2364 defm VCLTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00100, 0, "vclt", "s",
2367 // VACGE : Vector Absolute Compare Greater Than or Equal (aka VCAGE)
2368 def VACGEd : N3VDInt<1, 0, 0b00, 0b1110, 1, IIC_VBIND, "vacge", "f32",
2369 v2i32, v2f32, int_arm_neon_vacged, 0>;
2370 def VACGEq : N3VQInt<1, 0, 0b00, 0b1110, 1, IIC_VBINQ, "vacge", "f32",
2371 v4i32, v4f32, int_arm_neon_vacgeq, 0>;
2372 // VACGT : Vector Absolute Compare Greater Than (aka VCAGT)
2373 def VACGTd : N3VDInt<1, 0, 0b10, 0b1110, 1, IIC_VBIND, "vacgt", "f32",
2374 v2i32, v2f32, int_arm_neon_vacgtd, 0>;
2375 def VACGTq : N3VQInt<1, 0, 0b10, 0b1110, 1, IIC_VBINQ, "vacgt", "f32",
2376 v4i32, v4f32, int_arm_neon_vacgtq, 0>;
2377 // VTST : Vector Test Bits
2378 defm VTST : N3V_QHS<0, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2379 IIC_VBINi4Q, "vtst", "", NEONvtst, 1>;
2381 // Vector Bitwise Operations.
2383 // VAND : Vector Bitwise AND
2384 def VANDd : N3VDX<0, 0, 0b00, 0b0001, 1, IIC_VBINiD, "vand",
2385 v2i32, v2i32, and, 1>;
2386 def VANDq : N3VQX<0, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "vand",
2387 v4i32, v4i32, and, 1>;
2389 // VEOR : Vector Bitwise Exclusive OR
2390 def VEORd : N3VDX<1, 0, 0b00, 0b0001, 1, IIC_VBINiD, "veor",
2391 v2i32, v2i32, xor, 1>;
2392 def VEORq : N3VQX<1, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "veor",
2393 v4i32, v4i32, xor, 1>;
2395 // VORR : Vector Bitwise OR
2396 def VORRd : N3VDX<0, 0, 0b10, 0b0001, 1, IIC_VBINiD, "vorr",
2397 v2i32, v2i32, or, 1>;
2398 def VORRq : N3VQX<0, 0, 0b10, 0b0001, 1, IIC_VBINiQ, "vorr",
2399 v4i32, v4i32, or, 1>;
2401 // VBIC : Vector Bitwise Bit Clear (AND NOT)
2402 def VBICd : N3VX<0, 0, 0b01, 0b0001, 0, 1, (outs DPR:$dst),
2403 (ins DPR:$src1, DPR:$src2), IIC_VBINiD,
2404 "vbic", "$dst, $src1, $src2", "",
2405 [(set DPR:$dst, (v2i32 (and DPR:$src1,
2406 (vnot_conv DPR:$src2))))]>;
2407 def VBICq : N3VX<0, 0, 0b01, 0b0001, 1, 1, (outs QPR:$dst),
2408 (ins QPR:$src1, QPR:$src2), IIC_VBINiQ,
2409 "vbic", "$dst, $src1, $src2", "",
2410 [(set QPR:$dst, (v4i32 (and QPR:$src1,
2411 (vnot_conv QPR:$src2))))]>;
2413 // VORN : Vector Bitwise OR NOT
2414 def VORNd : N3VX<0, 0, 0b11, 0b0001, 0, 1, (outs DPR:$dst),
2415 (ins DPR:$src1, DPR:$src2), IIC_VBINiD,
2416 "vorn", "$dst, $src1, $src2", "",
2417 [(set DPR:$dst, (v2i32 (or DPR:$src1,
2418 (vnot_conv DPR:$src2))))]>;
2419 def VORNq : N3VX<0, 0, 0b11, 0b0001, 1, 1, (outs QPR:$dst),
2420 (ins QPR:$src1, QPR:$src2), IIC_VBINiQ,
2421 "vorn", "$dst, $src1, $src2", "",
2422 [(set QPR:$dst, (v4i32 (or QPR:$src1,
2423 (vnot_conv QPR:$src2))))]>;
2425 // VMVN : Vector Bitwise NOT
2426 def VMVNd : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 0, 0,
2427 (outs DPR:$dst), (ins DPR:$src), IIC_VSHLiD,
2428 "vmvn", "$dst, $src", "",
2429 [(set DPR:$dst, (v2i32 (vnot DPR:$src)))]>;
2430 def VMVNq : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 1, 0,
2431 (outs QPR:$dst), (ins QPR:$src), IIC_VSHLiD,
2432 "vmvn", "$dst, $src", "",
2433 [(set QPR:$dst, (v4i32 (vnot QPR:$src)))]>;
2434 def : Pat<(v2i32 (vnot_conv DPR:$src)), (VMVNd DPR:$src)>;
2435 def : Pat<(v4i32 (vnot_conv QPR:$src)), (VMVNq QPR:$src)>;
2437 // VBSL : Vector Bitwise Select
2438 def VBSLd : N3VX<1, 0, 0b01, 0b0001, 0, 1, (outs DPR:$dst),
2439 (ins DPR:$src1, DPR:$src2, DPR:$src3), IIC_VCNTiD,
2440 "vbsl", "$dst, $src2, $src3", "$src1 = $dst",
2442 (v2i32 (or (and DPR:$src2, DPR:$src1),
2443 (and DPR:$src3, (vnot_conv DPR:$src1)))))]>;
2444 def VBSLq : N3VX<1, 0, 0b01, 0b0001, 1, 1, (outs QPR:$dst),
2445 (ins QPR:$src1, QPR:$src2, QPR:$src3), IIC_VCNTiQ,
2446 "vbsl", "$dst, $src2, $src3", "$src1 = $dst",
2448 (v4i32 (or (and QPR:$src2, QPR:$src1),
2449 (and QPR:$src3, (vnot_conv QPR:$src1)))))]>;
2451 // VBIF : Vector Bitwise Insert if False
2452 // like VBSL but with: "vbif $dst, $src3, $src1", "$src2 = $dst",
2453 def VBIFd : N3VX<1, 0, 0b11, 0b0001, 0, 1,
2454 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3),
2455 IIC_VBINiD, "vbif", "$dst, $src2, $src3", "$src1 = $dst",
2456 [/* For disassembly only; pattern left blank */]>;
2457 def VBIFq : N3VX<1, 0, 0b11, 0b0001, 1, 1,
2458 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3),
2459 IIC_VBINiQ, "vbif", "$dst, $src2, $src3", "$src1 = $dst",
2460 [/* For disassembly only; pattern left blank */]>;
2462 // VBIT : Vector Bitwise Insert if True
2463 // like VBSL but with: "vbit $dst, $src2, $src1", "$src3 = $dst",
2464 def VBITd : N3VX<1, 0, 0b10, 0b0001, 0, 1,
2465 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3),
2466 IIC_VBINiD, "vbit", "$dst, $src2, $src3", "$src1 = $dst",
2467 [/* For disassembly only; pattern left blank */]>;
2468 def VBITq : N3VX<1, 0, 0b10, 0b0001, 1, 1,
2469 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3),
2470 IIC_VBINiQ, "vbit", "$dst, $src2, $src3", "$src1 = $dst",
2471 [/* For disassembly only; pattern left blank */]>;
2473 // VBIT/VBIF are not yet implemented. The TwoAddress pass will not go looking
2474 // for equivalent operations with different register constraints; it just
2477 // Vector Absolute Differences.
2479 // VABD : Vector Absolute Difference
2480 defm VABDs : N3VInt_QHS<0, 0, 0b0111, 0, IIC_VBINi4D, IIC_VBINi4D,
2481 IIC_VBINi4Q, IIC_VBINi4Q,
2482 "vabd", "s", int_arm_neon_vabds, 0>;
2483 defm VABDu : N3VInt_QHS<1, 0, 0b0111, 0, IIC_VBINi4D, IIC_VBINi4D,
2484 IIC_VBINi4Q, IIC_VBINi4Q,
2485 "vabd", "u", int_arm_neon_vabdu, 0>;
2486 def VABDfd : N3VDInt<1, 0, 0b10, 0b1101, 0, IIC_VBIND,
2487 "vabd", "f32", v2f32, v2f32, int_arm_neon_vabds, 0>;
2488 def VABDfq : N3VQInt<1, 0, 0b10, 0b1101, 0, IIC_VBINQ,
2489 "vabd", "f32", v4f32, v4f32, int_arm_neon_vabds, 0>;
2491 // VABDL : Vector Absolute Difference Long (Q = | D - D |)
2492 defm VABDLs : N3VLInt_QHS<0,1,0b0111,0, IIC_VBINi4Q,
2493 "vabdl", "s", int_arm_neon_vabdls, 0>;
2494 defm VABDLu : N3VLInt_QHS<1,1,0b0111,0, IIC_VBINi4Q,
2495 "vabdl", "u", int_arm_neon_vabdlu, 0>;
2497 // VABA : Vector Absolute Difference and Accumulate
2498 defm VABAs : N3VInt3_QHS<0,0,0b0111,1, "vaba", "s", int_arm_neon_vabas>;
2499 defm VABAu : N3VInt3_QHS<1,0,0b0111,1, "vaba", "u", int_arm_neon_vabau>;
2501 // VABAL : Vector Absolute Difference and Accumulate Long (Q += | D - D |)
2502 defm VABALs : N3VLInt3_QHS<0,1,0b0101,0, "vabal", "s", int_arm_neon_vabals>;
2503 defm VABALu : N3VLInt3_QHS<1,1,0b0101,0, "vabal", "u", int_arm_neon_vabalu>;
2505 // Vector Maximum and Minimum.
2507 // VMAX : Vector Maximum
2508 defm VMAXs : N3VInt_QHS<0,0,0b0110,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2509 IIC_VBINi4Q, "vmax", "s", int_arm_neon_vmaxs, 1>;
2510 defm VMAXu : N3VInt_QHS<1,0,0b0110,0, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2511 IIC_VBINi4Q, "vmax", "u", int_arm_neon_vmaxu, 1>;
2512 def VMAXfd : N3VDInt<0, 0, 0b00, 0b1111, 0, IIC_VBIND, "vmax", "f32",
2513 v2f32, v2f32, int_arm_neon_vmaxs, 1>;
2514 def VMAXfq : N3VQInt<0, 0, 0b00, 0b1111, 0, IIC_VBINQ, "vmax", "f32",
2515 v4f32, v4f32, int_arm_neon_vmaxs, 1>;
2517 // VMIN : Vector Minimum
2518 defm VMINs : N3VInt_QHS<0,0,0b0110,1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2519 IIC_VBINi4Q, "vmin", "s", int_arm_neon_vmins, 1>;
2520 defm VMINu : N3VInt_QHS<1,0,0b0110,1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2521 IIC_VBINi4Q, "vmin", "u", int_arm_neon_vminu, 1>;
2522 def VMINfd : N3VDInt<0, 0, 0b10, 0b1111, 0, IIC_VBIND, "vmin", "f32",
2523 v2f32, v2f32, int_arm_neon_vmins, 1>;
2524 def VMINfq : N3VQInt<0, 0, 0b10, 0b1111, 0, IIC_VBINQ, "vmin", "f32",
2525 v4f32, v4f32, int_arm_neon_vmins, 1>;
2527 // Vector Pairwise Operations.
2529 // VPADD : Vector Pairwise Add
2530 def VPADDi8 : N3VDInt<0, 0, 0b00, 0b1011, 1, IIC_VBINiD, "vpadd", "i8",
2531 v8i8, v8i8, int_arm_neon_vpadd, 0>;
2532 def VPADDi16 : N3VDInt<0, 0, 0b01, 0b1011, 1, IIC_VBINiD, "vpadd", "i16",
2533 v4i16, v4i16, int_arm_neon_vpadd, 0>;
2534 def VPADDi32 : N3VDInt<0, 0, 0b10, 0b1011, 1, IIC_VBINiD, "vpadd", "i32",
2535 v2i32, v2i32, int_arm_neon_vpadd, 0>;
2536 def VPADDf : N3VDInt<1, 0, 0b00, 0b1101, 0, IIC_VBIND, "vpadd", "f32",
2537 v2f32, v2f32, int_arm_neon_vpadd, 0>;
2539 // VPADDL : Vector Pairwise Add Long
2540 defm VPADDLs : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00100, 0, "vpaddl", "s",
2541 int_arm_neon_vpaddls>;
2542 defm VPADDLu : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00101, 0, "vpaddl", "u",
2543 int_arm_neon_vpaddlu>;
2545 // VPADAL : Vector Pairwise Add and Accumulate Long
2546 defm VPADALs : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01100, 0, "vpadal", "s",
2547 int_arm_neon_vpadals>;
2548 defm VPADALu : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01101, 0, "vpadal", "u",
2549 int_arm_neon_vpadalu>;
2551 // VPMAX : Vector Pairwise Maximum
2552 def VPMAXs8 : N3VDInt<0, 0, 0b00, 0b1010, 0, IIC_VBINi4D, "vpmax", "s8",
2553 v8i8, v8i8, int_arm_neon_vpmaxs, 0>;
2554 def VPMAXs16 : N3VDInt<0, 0, 0b01, 0b1010, 0, IIC_VBINi4D, "vpmax", "s16",
2555 v4i16, v4i16, int_arm_neon_vpmaxs, 0>;
2556 def VPMAXs32 : N3VDInt<0, 0, 0b10, 0b1010, 0, IIC_VBINi4D, "vpmax", "s32",
2557 v2i32, v2i32, int_arm_neon_vpmaxs, 0>;
2558 def VPMAXu8 : N3VDInt<1, 0, 0b00, 0b1010, 0, IIC_VBINi4D, "vpmax", "u8",
2559 v8i8, v8i8, int_arm_neon_vpmaxu, 0>;
2560 def VPMAXu16 : N3VDInt<1, 0, 0b01, 0b1010, 0, IIC_VBINi4D, "vpmax", "u16",
2561 v4i16, v4i16, int_arm_neon_vpmaxu, 0>;
2562 def VPMAXu32 : N3VDInt<1, 0, 0b10, 0b1010, 0, IIC_VBINi4D, "vpmax", "u32",
2563 v2i32, v2i32, int_arm_neon_vpmaxu, 0>;
2564 def VPMAXf : N3VDInt<1, 0, 0b00, 0b1111, 0, IIC_VBINi4D, "vpmax", "f32",
2565 v2f32, v2f32, int_arm_neon_vpmaxs, 0>;
2567 // VPMIN : Vector Pairwise Minimum
2568 def VPMINs8 : N3VDInt<0, 0, 0b00, 0b1010, 1, IIC_VBINi4D, "vpmin", "s8",
2569 v8i8, v8i8, int_arm_neon_vpmins, 0>;
2570 def VPMINs16 : N3VDInt<0, 0, 0b01, 0b1010, 1, IIC_VBINi4D, "vpmin", "s16",
2571 v4i16, v4i16, int_arm_neon_vpmins, 0>;
2572 def VPMINs32 : N3VDInt<0, 0, 0b10, 0b1010, 1, IIC_VBINi4D, "vpmin", "s32",
2573 v2i32, v2i32, int_arm_neon_vpmins, 0>;
2574 def VPMINu8 : N3VDInt<1, 0, 0b00, 0b1010, 1, IIC_VBINi4D, "vpmin", "u8",
2575 v8i8, v8i8, int_arm_neon_vpminu, 0>;
2576 def VPMINu16 : N3VDInt<1, 0, 0b01, 0b1010, 1, IIC_VBINi4D, "vpmin", "u16",
2577 v4i16, v4i16, int_arm_neon_vpminu, 0>;
2578 def VPMINu32 : N3VDInt<1, 0, 0b10, 0b1010, 1, IIC_VBINi4D, "vpmin", "u32",
2579 v2i32, v2i32, int_arm_neon_vpminu, 0>;
2580 def VPMINf : N3VDInt<1, 0, 0b10, 0b1111, 0, IIC_VBINi4D, "vpmin", "f32",
2581 v2f32, v2f32, int_arm_neon_vpmins, 0>;
2583 // Vector Reciprocal and Reciprocal Square Root Estimate and Step.
2585 // VRECPE : Vector Reciprocal Estimate
2586 def VRECPEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
2587 IIC_VUNAD, "vrecpe", "u32",
2588 v2i32, v2i32, int_arm_neon_vrecpe>;
2589 def VRECPEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
2590 IIC_VUNAQ, "vrecpe", "u32",
2591 v4i32, v4i32, int_arm_neon_vrecpe>;
2592 def VRECPEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
2593 IIC_VUNAD, "vrecpe", "f32",
2594 v2f32, v2f32, int_arm_neon_vrecpe>;
2595 def VRECPEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
2596 IIC_VUNAQ, "vrecpe", "f32",
2597 v4f32, v4f32, int_arm_neon_vrecpe>;
2599 // VRECPS : Vector Reciprocal Step
2600 def VRECPSfd : N3VDInt<0, 0, 0b00, 0b1111, 1,
2601 IIC_VRECSD, "vrecps", "f32",
2602 v2f32, v2f32, int_arm_neon_vrecps, 1>;
2603 def VRECPSfq : N3VQInt<0, 0, 0b00, 0b1111, 1,
2604 IIC_VRECSQ, "vrecps", "f32",
2605 v4f32, v4f32, int_arm_neon_vrecps, 1>;
2607 // VRSQRTE : Vector Reciprocal Square Root Estimate
2608 def VRSQRTEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
2609 IIC_VUNAD, "vrsqrte", "u32",
2610 v2i32, v2i32, int_arm_neon_vrsqrte>;
2611 def VRSQRTEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
2612 IIC_VUNAQ, "vrsqrte", "u32",
2613 v4i32, v4i32, int_arm_neon_vrsqrte>;
2614 def VRSQRTEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
2615 IIC_VUNAD, "vrsqrte", "f32",
2616 v2f32, v2f32, int_arm_neon_vrsqrte>;
2617 def VRSQRTEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
2618 IIC_VUNAQ, "vrsqrte", "f32",
2619 v4f32, v4f32, int_arm_neon_vrsqrte>;
2621 // VRSQRTS : Vector Reciprocal Square Root Step
2622 def VRSQRTSfd : N3VDInt<0, 0, 0b10, 0b1111, 1,
2623 IIC_VRECSD, "vrsqrts", "f32",
2624 v2f32, v2f32, int_arm_neon_vrsqrts, 1>;
2625 def VRSQRTSfq : N3VQInt<0, 0, 0b10, 0b1111, 1,
2626 IIC_VRECSQ, "vrsqrts", "f32",
2627 v4f32, v4f32, int_arm_neon_vrsqrts, 1>;
2631 // VSHL : Vector Shift
2632 defm VSHLs : N3VSh_QHSD<0, 0, 0b0100, 0, IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ,
2633 IIC_VSHLiQ, "vshl", "s", int_arm_neon_vshifts, 0>;
2634 defm VSHLu : N3VSh_QHSD<1, 0, 0b0100, 0, IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ,
2635 IIC_VSHLiQ, "vshl", "u", int_arm_neon_vshiftu, 0>;
2636 // VSHL : Vector Shift Left (Immediate)
2637 defm VSHLi : N2VSh_QHSD<0, 1, 0b0101, 1, IIC_VSHLiD, "vshl", "i", NEONvshl,
2639 // VSHR : Vector Shift Right (Immediate)
2640 defm VSHRs : N2VSh_QHSD<0, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "s", NEONvshrs,
2642 defm VSHRu : N2VSh_QHSD<1, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "u", NEONvshru,
2645 // VSHLL : Vector Shift Left Long
2646 defm VSHLLs : N2VLSh_QHS<0, 1, 0b1010, 0, 0, 1, "vshll", "s", NEONvshlls>;
2647 defm VSHLLu : N2VLSh_QHS<1, 1, 0b1010, 0, 0, 1, "vshll", "u", NEONvshllu>;
2649 // VSHLL : Vector Shift Left Long (with maximum shift count)
2650 class N2VLShMax<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
2651 bit op6, bit op4, string OpcodeStr, string Dt, ValueType ResTy,
2652 ValueType OpTy, SDNode OpNode>
2653 : N2VLSh<op24, op23, op11_8, op7, op6, op4, OpcodeStr, Dt,
2654 ResTy, OpTy, OpNode> {
2655 let Inst{21-16} = op21_16;
2657 def VSHLLi8 : N2VLShMax<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll", "i8",
2658 v8i16, v8i8, NEONvshlli>;
2659 def VSHLLi16 : N2VLShMax<1, 1, 0b110110, 0b0011, 0, 0, 0, "vshll", "i16",
2660 v4i32, v4i16, NEONvshlli>;
2661 def VSHLLi32 : N2VLShMax<1, 1, 0b111010, 0b0011, 0, 0, 0, "vshll", "i32",
2662 v2i64, v2i32, NEONvshlli>;
2664 // VSHRN : Vector Shift Right and Narrow
2665 defm VSHRN : N2VNSh_HSD<0,1,0b1000,0,0,1, IIC_VSHLiD, "vshrn", "i",
2668 // VRSHL : Vector Rounding Shift
2669 defm VRSHLs : N3VSh_QHSD<0,0,0b0101,0,IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2670 IIC_VSHLi4Q,"vrshl", "s", int_arm_neon_vrshifts,0>;
2671 defm VRSHLu : N3VSh_QHSD<1,0,0b0101,0,IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2672 IIC_VSHLi4Q,"vrshl", "u", int_arm_neon_vrshiftu,0>;
2673 // VRSHR : Vector Rounding Shift Right
2674 defm VRSHRs : N2VSh_QHSD<0,1,0b0010,1, IIC_VSHLi4D, "vrshr", "s", NEONvrshrs,
2676 defm VRSHRu : N2VSh_QHSD<1,1,0b0010,1, IIC_VSHLi4D, "vrshr", "u", NEONvrshru,
2679 // VRSHRN : Vector Rounding Shift Right and Narrow
2680 defm VRSHRN : N2VNSh_HSD<0, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vrshrn", "i",
2683 // VQSHL : Vector Saturating Shift
2684 defm VQSHLs : N3VSh_QHSD<0,0,0b0100,1,IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2685 IIC_VSHLi4Q, "vqshl", "s", int_arm_neon_vqshifts,0>;
2686 defm VQSHLu : N3VSh_QHSD<1,0,0b0100,1,IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2687 IIC_VSHLi4Q, "vqshl", "u", int_arm_neon_vqshiftu,0>;
2688 // VQSHL : Vector Saturating Shift Left (Immediate)
2689 defm VQSHLsi : N2VSh_QHSD<0,1,0b0111,1, IIC_VSHLi4D, "vqshl", "s",NEONvqshls,
2691 defm VQSHLui : N2VSh_QHSD<1,1,0b0111,1, IIC_VSHLi4D, "vqshl", "u",NEONvqshlu,
2693 // VQSHLU : Vector Saturating Shift Left (Immediate, Unsigned)
2694 defm VQSHLsu : N2VSh_QHSD<1,1,0b0110,1, IIC_VSHLi4D,"vqshlu","s",NEONvqshlsu,
2697 // VQSHRN : Vector Saturating Shift Right and Narrow
2698 defm VQSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "s",
2700 defm VQSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "u",
2703 // VQSHRUN : Vector Saturating Shift Right and Narrow (Unsigned)
2704 defm VQSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 0, 1, IIC_VSHLi4D, "vqshrun", "s",
2707 // VQRSHL : Vector Saturating Rounding Shift
2708 defm VQRSHLs : N3VSh_QHSD<0,0,0b0101,1, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2709 IIC_VSHLi4Q, "vqrshl", "s",
2710 int_arm_neon_vqrshifts, 0>;
2711 defm VQRSHLu : N3VSh_QHSD<1,0,0b0101,1, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
2712 IIC_VSHLi4Q, "vqrshl", "u",
2713 int_arm_neon_vqrshiftu, 0>;
2715 // VQRSHRN : Vector Saturating Rounding Shift Right and Narrow
2716 defm VQRSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "s",
2718 defm VQRSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "u",
2721 // VQRSHRUN : Vector Saturating Rounding Shift Right and Narrow (Unsigned)
2722 defm VQRSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vqrshrun", "s",
2725 // VSRA : Vector Shift Right and Accumulate
2726 defm VSRAs : N2VShAdd_QHSD<0, 1, 0b0001, 1, "vsra", "s", NEONvshrs>;
2727 defm VSRAu : N2VShAdd_QHSD<1, 1, 0b0001, 1, "vsra", "u", NEONvshru>;
2728 // VRSRA : Vector Rounding Shift Right and Accumulate
2729 defm VRSRAs : N2VShAdd_QHSD<0, 1, 0b0011, 1, "vrsra", "s", NEONvrshrs>;
2730 defm VRSRAu : N2VShAdd_QHSD<1, 1, 0b0011, 1, "vrsra", "u", NEONvrshru>;
2732 // VSLI : Vector Shift Left and Insert
2733 defm VSLI : N2VShIns_QHSD<1, 1, 0b0101, 1, "vsli", NEONvsli, N2RegVShLFrm>;
2734 // VSRI : Vector Shift Right and Insert
2735 defm VSRI : N2VShIns_QHSD<1, 1, 0b0100, 1, "vsri", NEONvsri, N2RegVShRFrm>;
2737 // Vector Absolute and Saturating Absolute.
2739 // VABS : Vector Absolute Value
2740 defm VABS : N2VInt_QHS<0b11, 0b11, 0b01, 0b00110, 0,
2741 IIC_VUNAiD, IIC_VUNAiQ, "vabs", "s",
2743 def VABSfd : N2VDInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
2744 IIC_VUNAD, "vabs", "f32",
2745 v2f32, v2f32, int_arm_neon_vabs>;
2746 def VABSfq : N2VQInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
2747 IIC_VUNAQ, "vabs", "f32",
2748 v4f32, v4f32, int_arm_neon_vabs>;
2750 // VQABS : Vector Saturating Absolute Value
2751 defm VQABS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01110, 0,
2752 IIC_VQUNAiD, IIC_VQUNAiQ, "vqabs", "s",
2753 int_arm_neon_vqabs>;
2757 def vneg : PatFrag<(ops node:$in), (sub immAllZerosV, node:$in)>;
2758 def vneg_conv : PatFrag<(ops node:$in), (sub immAllZerosV_bc, node:$in)>;
2760 class VNEGD<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
2761 : N2V<0b11, 0b11, size, 0b01, 0b00111, 0, 0, (outs DPR:$dst), (ins DPR:$src),
2762 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
2763 [(set DPR:$dst, (Ty (vneg DPR:$src)))]>;
2764 class VNEGQ<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
2765 : N2V<0b11, 0b11, size, 0b01, 0b00111, 1, 0, (outs QPR:$dst), (ins QPR:$src),
2766 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
2767 [(set QPR:$dst, (Ty (vneg QPR:$src)))]>;
2769 // VNEG : Vector Negate
2770 def VNEGs8d : VNEGD<0b00, "vneg", "s8", v8i8>;
2771 def VNEGs16d : VNEGD<0b01, "vneg", "s16", v4i16>;
2772 def VNEGs32d : VNEGD<0b10, "vneg", "s32", v2i32>;
2773 def VNEGs8q : VNEGQ<0b00, "vneg", "s8", v16i8>;
2774 def VNEGs16q : VNEGQ<0b01, "vneg", "s16", v8i16>;
2775 def VNEGs32q : VNEGQ<0b10, "vneg", "s32", v4i32>;
2777 // VNEG : Vector Negate (floating-point)
2778 def VNEGfd : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
2779 (outs DPR:$dst), (ins DPR:$src), IIC_VUNAD,
2780 "vneg", "f32", "$dst, $src", "",
2781 [(set DPR:$dst, (v2f32 (fneg DPR:$src)))]>;
2782 def VNEGf32q : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 1, 0,
2783 (outs QPR:$dst), (ins QPR:$src), IIC_VUNAQ,
2784 "vneg", "f32", "$dst, $src", "",
2785 [(set QPR:$dst, (v4f32 (fneg QPR:$src)))]>;
2787 def : Pat<(v8i8 (vneg_conv DPR:$src)), (VNEGs8d DPR:$src)>;
2788 def : Pat<(v4i16 (vneg_conv DPR:$src)), (VNEGs16d DPR:$src)>;
2789 def : Pat<(v2i32 (vneg_conv DPR:$src)), (VNEGs32d DPR:$src)>;
2790 def : Pat<(v16i8 (vneg_conv QPR:$src)), (VNEGs8q QPR:$src)>;
2791 def : Pat<(v8i16 (vneg_conv QPR:$src)), (VNEGs16q QPR:$src)>;
2792 def : Pat<(v4i32 (vneg_conv QPR:$src)), (VNEGs32q QPR:$src)>;
2794 // VQNEG : Vector Saturating Negate
2795 defm VQNEG : N2VInt_QHS<0b11, 0b11, 0b00, 0b01111, 0,
2796 IIC_VQUNAiD, IIC_VQUNAiQ, "vqneg", "s",
2797 int_arm_neon_vqneg>;
2799 // Vector Bit Counting Operations.
2801 // VCLS : Vector Count Leading Sign Bits
2802 defm VCLS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01000, 0,
2803 IIC_VCNTiD, IIC_VCNTiQ, "vcls", "s",
2805 // VCLZ : Vector Count Leading Zeros
2806 defm VCLZ : N2VInt_QHS<0b11, 0b11, 0b00, 0b01001, 0,
2807 IIC_VCNTiD, IIC_VCNTiQ, "vclz", "i",
2809 // VCNT : Vector Count One Bits
2810 def VCNTd : N2VDInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
2811 IIC_VCNTiD, "vcnt", "8",
2812 v8i8, v8i8, int_arm_neon_vcnt>;
2813 def VCNTq : N2VQInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
2814 IIC_VCNTiQ, "vcnt", "8",
2815 v16i8, v16i8, int_arm_neon_vcnt>;
2817 // Vector Swap -- for disassembly only.
2818 def VSWPd : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 0, 0,
2819 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
2820 "vswp", "$dst, $src", "", []>;
2821 def VSWPq : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 1, 0,
2822 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
2823 "vswp", "$dst, $src", "", []>;
2825 // Vector Move Operations.
2827 // VMOV : Vector Move (Register)
2829 def VMOVDneon: N3VX<0, 0, 0b10, 0b0001, 0, 1, (outs DPR:$dst), (ins DPR:$src),
2830 IIC_VMOVD, "vmov", "$dst, $src", "", []>;
2831 def VMOVQ : N3VX<0, 0, 0b10, 0b0001, 1, 1, (outs QPR:$dst), (ins QPR:$src),
2832 IIC_VMOVD, "vmov", "$dst, $src", "", []>;
2834 // VMOV : Vector Move (Immediate)
2836 // VMOV_get_imm8 xform function: convert build_vector to VMOV.i8 imm.
2837 def VMOV_get_imm8 : SDNodeXForm<build_vector, [{
2838 return ARM::getVMOVImm(N, 1, *CurDAG);
2840 def vmovImm8 : PatLeaf<(build_vector), [{
2841 return ARM::getVMOVImm(N, 1, *CurDAG).getNode() != 0;
2844 // VMOV_get_imm16 xform function: convert build_vector to VMOV.i16 imm.
2845 def VMOV_get_imm16 : SDNodeXForm<build_vector, [{
2846 return ARM::getVMOVImm(N, 2, *CurDAG);
2848 def vmovImm16 : PatLeaf<(build_vector), [{
2849 return ARM::getVMOVImm(N, 2, *CurDAG).getNode() != 0;
2850 }], VMOV_get_imm16>;
2852 // VMOV_get_imm32 xform function: convert build_vector to VMOV.i32 imm.
2853 def VMOV_get_imm32 : SDNodeXForm<build_vector, [{
2854 return ARM::getVMOVImm(N, 4, *CurDAG);
2856 def vmovImm32 : PatLeaf<(build_vector), [{
2857 return ARM::getVMOVImm(N, 4, *CurDAG).getNode() != 0;
2858 }], VMOV_get_imm32>;
2860 // VMOV_get_imm64 xform function: convert build_vector to VMOV.i64 imm.
2861 def VMOV_get_imm64 : SDNodeXForm<build_vector, [{
2862 return ARM::getVMOVImm(N, 8, *CurDAG);
2864 def vmovImm64 : PatLeaf<(build_vector), [{
2865 return ARM::getVMOVImm(N, 8, *CurDAG).getNode() != 0;
2866 }], VMOV_get_imm64>;
2868 // Note: Some of the cmode bits in the following VMOV instructions need to
2869 // be encoded based on the immed values.
2871 def VMOVv8i8 : N1ModImm<1, 0b000, 0b1110, 0, 0, 0, 1, (outs DPR:$dst),
2872 (ins h8imm:$SIMM), IIC_VMOVImm,
2873 "vmov", "i8", "$dst, $SIMM", "",
2874 [(set DPR:$dst, (v8i8 vmovImm8:$SIMM))]>;
2875 def VMOVv16i8 : N1ModImm<1, 0b000, 0b1110, 0, 1, 0, 1, (outs QPR:$dst),
2876 (ins h8imm:$SIMM), IIC_VMOVImm,
2877 "vmov", "i8", "$dst, $SIMM", "",
2878 [(set QPR:$dst, (v16i8 vmovImm8:$SIMM))]>;
2880 def VMOVv4i16 : N1ModImm<1, 0b000, {1,0,?,?}, 0, 0, {?}, 1, (outs DPR:$dst),
2881 (ins h16imm:$SIMM), IIC_VMOVImm,
2882 "vmov", "i16", "$dst, $SIMM", "",
2883 [(set DPR:$dst, (v4i16 vmovImm16:$SIMM))]>;
2884 def VMOVv8i16 : N1ModImm<1, 0b000, {1,0,?,?}, 0, 1, {?}, 1, (outs QPR:$dst),
2885 (ins h16imm:$SIMM), IIC_VMOVImm,
2886 "vmov", "i16", "$dst, $SIMM", "",
2887 [(set QPR:$dst, (v8i16 vmovImm16:$SIMM))]>;
2889 def VMOVv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, {?}, 1, (outs DPR:$dst),
2890 (ins h32imm:$SIMM), IIC_VMOVImm,
2891 "vmov", "i32", "$dst, $SIMM", "",
2892 [(set DPR:$dst, (v2i32 vmovImm32:$SIMM))]>;
2893 def VMOVv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, {?}, 1, (outs QPR:$dst),
2894 (ins h32imm:$SIMM), IIC_VMOVImm,
2895 "vmov", "i32", "$dst, $SIMM", "",
2896 [(set QPR:$dst, (v4i32 vmovImm32:$SIMM))]>;
2898 def VMOVv1i64 : N1ModImm<1, 0b000, 0b1110, 0, 0, 1, 1, (outs DPR:$dst),
2899 (ins h64imm:$SIMM), IIC_VMOVImm,
2900 "vmov", "i64", "$dst, $SIMM", "",
2901 [(set DPR:$dst, (v1i64 vmovImm64:$SIMM))]>;
2902 def VMOVv2i64 : N1ModImm<1, 0b000, 0b1110, 0, 1, 1, 1, (outs QPR:$dst),
2903 (ins h64imm:$SIMM), IIC_VMOVImm,
2904 "vmov", "i64", "$dst, $SIMM", "",
2905 [(set QPR:$dst, (v2i64 vmovImm64:$SIMM))]>;
2907 // VMOV : Vector Get Lane (move scalar to ARM core register)
2909 def VGETLNs8 : NVGetLane<{1,1,1,0,0,1,?,1}, 0b1011, {?,?},
2910 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2911 IIC_VMOVSI, "vmov", "s8", "$dst, $src[$lane]",
2912 [(set GPR:$dst, (NEONvgetlanes (v8i8 DPR:$src),
2914 def VGETLNs16 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, {?,1},
2915 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2916 IIC_VMOVSI, "vmov", "s16", "$dst, $src[$lane]",
2917 [(set GPR:$dst, (NEONvgetlanes (v4i16 DPR:$src),
2919 def VGETLNu8 : NVGetLane<{1,1,1,0,1,1,?,1}, 0b1011, {?,?},
2920 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2921 IIC_VMOVSI, "vmov", "u8", "$dst, $src[$lane]",
2922 [(set GPR:$dst, (NEONvgetlaneu (v8i8 DPR:$src),
2924 def VGETLNu16 : NVGetLane<{1,1,1,0,1,0,?,1}, 0b1011, {?,1},
2925 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2926 IIC_VMOVSI, "vmov", "u16", "$dst, $src[$lane]",
2927 [(set GPR:$dst, (NEONvgetlaneu (v4i16 DPR:$src),
2929 def VGETLNi32 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, 0b00,
2930 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2931 IIC_VMOVSI, "vmov", "32", "$dst, $src[$lane]",
2932 [(set GPR:$dst, (extractelt (v2i32 DPR:$src),
2934 // def VGETLNf32: see FMRDH and FMRDL in ARMInstrVFP.td
2935 def : Pat<(NEONvgetlanes (v16i8 QPR:$src), imm:$lane),
2936 (VGETLNs8 (v8i8 (EXTRACT_SUBREG QPR:$src,
2937 (DSubReg_i8_reg imm:$lane))),
2938 (SubReg_i8_lane imm:$lane))>;
2939 def : Pat<(NEONvgetlanes (v8i16 QPR:$src), imm:$lane),
2940 (VGETLNs16 (v4i16 (EXTRACT_SUBREG QPR:$src,
2941 (DSubReg_i16_reg imm:$lane))),
2942 (SubReg_i16_lane imm:$lane))>;
2943 def : Pat<(NEONvgetlaneu (v16i8 QPR:$src), imm:$lane),
2944 (VGETLNu8 (v8i8 (EXTRACT_SUBREG QPR:$src,
2945 (DSubReg_i8_reg imm:$lane))),
2946 (SubReg_i8_lane imm:$lane))>;
2947 def : Pat<(NEONvgetlaneu (v8i16 QPR:$src), imm:$lane),
2948 (VGETLNu16 (v4i16 (EXTRACT_SUBREG QPR:$src,
2949 (DSubReg_i16_reg imm:$lane))),
2950 (SubReg_i16_lane imm:$lane))>;
2951 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
2952 (VGETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src,
2953 (DSubReg_i32_reg imm:$lane))),
2954 (SubReg_i32_lane imm:$lane))>;
2955 def : Pat<(extractelt (v2f32 DPR:$src1), imm:$src2),
2956 (EXTRACT_SUBREG (v2f32 (COPY_TO_REGCLASS (v2f32 DPR:$src1),DPR_VFP2)),
2957 (SSubReg_f32_reg imm:$src2))>;
2958 def : Pat<(extractelt (v4f32 QPR:$src1), imm:$src2),
2959 (EXTRACT_SUBREG (v4f32 (COPY_TO_REGCLASS (v4f32 QPR:$src1),QPR_VFP2)),
2960 (SSubReg_f32_reg imm:$src2))>;
2961 //def : Pat<(extractelt (v2i64 QPR:$src1), imm:$src2),
2962 // (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
2963 def : Pat<(extractelt (v2f64 QPR:$src1), imm:$src2),
2964 (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
2967 // VMOV : Vector Set Lane (move ARM core register to scalar)
2969 let Constraints = "$src1 = $dst" in {
2970 def VSETLNi8 : NVSetLane<{1,1,1,0,0,1,?,0}, 0b1011, {?,?}, (outs DPR:$dst),
2971 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
2972 IIC_VMOVISL, "vmov", "8", "$dst[$lane], $src2",
2973 [(set DPR:$dst, (vector_insert (v8i8 DPR:$src1),
2974 GPR:$src2, imm:$lane))]>;
2975 def VSETLNi16 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, {?,1}, (outs DPR:$dst),
2976 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
2977 IIC_VMOVISL, "vmov", "16", "$dst[$lane], $src2",
2978 [(set DPR:$dst, (vector_insert (v4i16 DPR:$src1),
2979 GPR:$src2, imm:$lane))]>;
2980 def VSETLNi32 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, 0b00, (outs DPR:$dst),
2981 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
2982 IIC_VMOVISL, "vmov", "32", "$dst[$lane], $src2",
2983 [(set DPR:$dst, (insertelt (v2i32 DPR:$src1),
2984 GPR:$src2, imm:$lane))]>;
2986 def : Pat<(vector_insert (v16i8 QPR:$src1), GPR:$src2, imm:$lane),
2987 (v16i8 (INSERT_SUBREG QPR:$src1,
2988 (v8i8 (VSETLNi8 (v8i8 (EXTRACT_SUBREG QPR:$src1,
2989 (DSubReg_i8_reg imm:$lane))),
2990 GPR:$src2, (SubReg_i8_lane imm:$lane))),
2991 (DSubReg_i8_reg imm:$lane)))>;
2992 def : Pat<(vector_insert (v8i16 QPR:$src1), GPR:$src2, imm:$lane),
2993 (v8i16 (INSERT_SUBREG QPR:$src1,
2994 (v4i16 (VSETLNi16 (v4i16 (EXTRACT_SUBREG QPR:$src1,
2995 (DSubReg_i16_reg imm:$lane))),
2996 GPR:$src2, (SubReg_i16_lane imm:$lane))),
2997 (DSubReg_i16_reg imm:$lane)))>;
2998 def : Pat<(insertelt (v4i32 QPR:$src1), GPR:$src2, imm:$lane),
2999 (v4i32 (INSERT_SUBREG QPR:$src1,
3000 (v2i32 (VSETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src1,
3001 (DSubReg_i32_reg imm:$lane))),
3002 GPR:$src2, (SubReg_i32_lane imm:$lane))),
3003 (DSubReg_i32_reg imm:$lane)))>;
3005 def : Pat<(v2f32 (insertelt DPR:$src1, SPR:$src2, imm:$src3)),
3006 (INSERT_SUBREG (v2f32 (COPY_TO_REGCLASS DPR:$src1, DPR_VFP2)),
3007 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
3008 def : Pat<(v4f32 (insertelt QPR:$src1, SPR:$src2, imm:$src3)),
3009 (INSERT_SUBREG (v4f32 (COPY_TO_REGCLASS QPR:$src1, QPR_VFP2)),
3010 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
3012 //def : Pat<(v2i64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
3013 // (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
3014 def : Pat<(v2f64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
3015 (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
3017 def : Pat<(v2f32 (scalar_to_vector SPR:$src)),
3018 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$src, arm_ssubreg_0)>;
3019 def : Pat<(v2f64 (scalar_to_vector (f64 DPR:$src))),
3020 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), DPR:$src, arm_dsubreg_0)>;
3021 def : Pat<(v4f32 (scalar_to_vector SPR:$src)),
3022 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), SPR:$src, arm_ssubreg_0)>;
3024 def : Pat<(v8i8 (scalar_to_vector GPR:$src)),
3025 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
3026 def : Pat<(v4i16 (scalar_to_vector GPR:$src)),
3027 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
3028 def : Pat<(v2i32 (scalar_to_vector GPR:$src)),
3029 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
3031 def : Pat<(v16i8 (scalar_to_vector GPR:$src)),
3032 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3033 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
3035 def : Pat<(v8i16 (scalar_to_vector GPR:$src)),
3036 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
3037 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
3039 def : Pat<(v4i32 (scalar_to_vector GPR:$src)),
3040 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
3041 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
3044 // VDUP : Vector Duplicate (from ARM core register to all elements)
3046 class VDUPD<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
3047 : NVDup<opcod1, 0b1011, opcod3, (outs DPR:$dst), (ins GPR:$src),
3048 IIC_VMOVIS, "vdup", Dt, "$dst, $src",
3049 [(set DPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
3050 class VDUPQ<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
3051 : NVDup<opcod1, 0b1011, opcod3, (outs QPR:$dst), (ins GPR:$src),
3052 IIC_VMOVIS, "vdup", Dt, "$dst, $src",
3053 [(set QPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
3055 def VDUP8d : VDUPD<0b11101100, 0b00, "8", v8i8>;
3056 def VDUP16d : VDUPD<0b11101000, 0b01, "16", v4i16>;
3057 def VDUP32d : VDUPD<0b11101000, 0b00, "32", v2i32>;
3058 def VDUP8q : VDUPQ<0b11101110, 0b00, "8", v16i8>;
3059 def VDUP16q : VDUPQ<0b11101010, 0b01, "16", v8i16>;
3060 def VDUP32q : VDUPQ<0b11101010, 0b00, "32", v4i32>;
3062 def VDUPfd : NVDup<0b11101000, 0b1011, 0b00, (outs DPR:$dst), (ins GPR:$src),
3063 IIC_VMOVIS, "vdup", "32", "$dst, $src",
3064 [(set DPR:$dst, (v2f32 (NEONvdup
3065 (f32 (bitconvert GPR:$src)))))]>;
3066 def VDUPfq : NVDup<0b11101010, 0b1011, 0b00, (outs QPR:$dst), (ins GPR:$src),
3067 IIC_VMOVIS, "vdup", "32", "$dst, $src",
3068 [(set QPR:$dst, (v4f32 (NEONvdup
3069 (f32 (bitconvert GPR:$src)))))]>;
3071 // VDUP : Vector Duplicate Lane (from scalar to all elements)
3073 class VDUPLND<bits<4> op19_16, string OpcodeStr, string Dt,
3075 : NVDupLane<op19_16, 0, (outs DPR:$dst), (ins DPR:$src, nohash_imm:$lane),
3076 IIC_VMOVD, OpcodeStr, Dt, "$dst, $src[$lane]",
3077 [(set DPR:$dst, (Ty (NEONvduplane (Ty DPR:$src), imm:$lane)))]>;
3079 class VDUPLNQ<bits<4> op19_16, string OpcodeStr, string Dt,
3080 ValueType ResTy, ValueType OpTy>
3081 : NVDupLane<op19_16, 1, (outs QPR:$dst), (ins DPR:$src, nohash_imm:$lane),
3082 IIC_VMOVD, OpcodeStr, Dt, "$dst, $src[$lane]",
3083 [(set QPR:$dst, (ResTy (NEONvduplane (OpTy DPR:$src),
3086 // Inst{19-16} is partially specified depending on the element size.
3088 def VDUPLN8d : VDUPLND<{?,?,?,1}, "vdup", "8", v8i8>;
3089 def VDUPLN16d : VDUPLND<{?,?,1,0}, "vdup", "16", v4i16>;
3090 def VDUPLN32d : VDUPLND<{?,1,0,0}, "vdup", "32", v2i32>;
3091 def VDUPLNfd : VDUPLND<{?,1,0,0}, "vdup", "32", v2f32>;
3092 def VDUPLN8q : VDUPLNQ<{?,?,?,1}, "vdup", "8", v16i8, v8i8>;
3093 def VDUPLN16q : VDUPLNQ<{?,?,1,0}, "vdup", "16", v8i16, v4i16>;
3094 def VDUPLN32q : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4i32, v2i32>;
3095 def VDUPLNfq : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4f32, v2f32>;
3097 def : Pat<(v16i8 (NEONvduplane (v16i8 QPR:$src), imm:$lane)),
3098 (v16i8 (VDUPLN8q (v8i8 (EXTRACT_SUBREG QPR:$src,
3099 (DSubReg_i8_reg imm:$lane))),
3100 (SubReg_i8_lane imm:$lane)))>;
3101 def : Pat<(v8i16 (NEONvduplane (v8i16 QPR:$src), imm:$lane)),
3102 (v8i16 (VDUPLN16q (v4i16 (EXTRACT_SUBREG QPR:$src,
3103 (DSubReg_i16_reg imm:$lane))),
3104 (SubReg_i16_lane imm:$lane)))>;
3105 def : Pat<(v4i32 (NEONvduplane (v4i32 QPR:$src), imm:$lane)),
3106 (v4i32 (VDUPLN32q (v2i32 (EXTRACT_SUBREG QPR:$src,
3107 (DSubReg_i32_reg imm:$lane))),
3108 (SubReg_i32_lane imm:$lane)))>;
3109 def : Pat<(v4f32 (NEONvduplane (v4f32 QPR:$src), imm:$lane)),
3110 (v4f32 (VDUPLNfq (v2f32 (EXTRACT_SUBREG QPR:$src,
3111 (DSubReg_i32_reg imm:$lane))),
3112 (SubReg_i32_lane imm:$lane)))>;
3114 def VDUPfdf : N2V<0b11, 0b11, {?,1}, {0,0}, 0b11000, 0, 0,
3115 (outs DPR:$dst), (ins SPR:$src),
3116 IIC_VMOVD, "vdup", "32", "$dst, ${src:lane}", "",
3117 [(set DPR:$dst, (v2f32 (NEONvdup (f32 SPR:$src))))]>;
3119 def VDUPfqf : N2V<0b11, 0b11, {?,1}, {0,0}, 0b11000, 1, 0,
3120 (outs QPR:$dst), (ins SPR:$src),
3121 IIC_VMOVD, "vdup", "32", "$dst, ${src:lane}", "",
3122 [(set QPR:$dst, (v4f32 (NEONvdup (f32 SPR:$src))))]>;
3124 def : Pat<(v2i64 (NEONvduplane (v2i64 QPR:$src), imm:$lane)),
3125 (INSERT_SUBREG QPR:$src,
3126 (i64 (EXTRACT_SUBREG QPR:$src,
3127 (DSubReg_f64_reg imm:$lane))),
3128 (DSubReg_f64_other_reg imm:$lane))>;
3129 def : Pat<(v2f64 (NEONvduplane (v2f64 QPR:$src), imm:$lane)),
3130 (INSERT_SUBREG QPR:$src,
3131 (f64 (EXTRACT_SUBREG QPR:$src,
3132 (DSubReg_f64_reg imm:$lane))),
3133 (DSubReg_f64_other_reg imm:$lane))>;
3135 // VMOVN : Vector Narrowing Move
3136 defm VMOVN : N2VNInt_HSD<0b11,0b11,0b10,0b00100,0,0, IIC_VMOVD,
3137 "vmovn", "i", int_arm_neon_vmovn>;
3138 // VQMOVN : Vector Saturating Narrowing Move
3139 defm VQMOVNs : N2VNInt_HSD<0b11,0b11,0b10,0b00101,0,0, IIC_VQUNAiD,
3140 "vqmovn", "s", int_arm_neon_vqmovns>;
3141 defm VQMOVNu : N2VNInt_HSD<0b11,0b11,0b10,0b00101,1,0, IIC_VQUNAiD,
3142 "vqmovn", "u", int_arm_neon_vqmovnu>;
3143 defm VQMOVNsu : N2VNInt_HSD<0b11,0b11,0b10,0b00100,1,0, IIC_VQUNAiD,
3144 "vqmovun", "s", int_arm_neon_vqmovnsu>;
3145 // VMOVL : Vector Lengthening Move
3146 defm VMOVLs : N2VLInt_QHS<0b01,0b10100,0,1, "vmovl", "s",
3147 int_arm_neon_vmovls>;
3148 defm VMOVLu : N2VLInt_QHS<0b11,0b10100,0,1, "vmovl", "u",
3149 int_arm_neon_vmovlu>;
3151 // Vector Conversions.
3153 // VCVT : Vector Convert Between Floating-Point and Integers
3154 def VCVTf2sd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
3155 v2i32, v2f32, fp_to_sint>;
3156 def VCVTf2ud : N2VD<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
3157 v2i32, v2f32, fp_to_uint>;
3158 def VCVTs2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
3159 v2f32, v2i32, sint_to_fp>;
3160 def VCVTu2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
3161 v2f32, v2i32, uint_to_fp>;
3163 def VCVTf2sq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
3164 v4i32, v4f32, fp_to_sint>;
3165 def VCVTf2uq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
3166 v4i32, v4f32, fp_to_uint>;
3167 def VCVTs2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
3168 v4f32, v4i32, sint_to_fp>;
3169 def VCVTu2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
3170 v4f32, v4i32, uint_to_fp>;
3172 // VCVT : Vector Convert Between Floating-Point and Fixed-Point.
3173 def VCVTf2xsd : N2VCvtD<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
3174 v2i32, v2f32, int_arm_neon_vcvtfp2fxs>;
3175 def VCVTf2xud : N2VCvtD<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
3176 v2i32, v2f32, int_arm_neon_vcvtfp2fxu>;
3177 def VCVTxs2fd : N2VCvtD<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
3178 v2f32, v2i32, int_arm_neon_vcvtfxs2fp>;
3179 def VCVTxu2fd : N2VCvtD<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
3180 v2f32, v2i32, int_arm_neon_vcvtfxu2fp>;
3182 def VCVTf2xsq : N2VCvtQ<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
3183 v4i32, v4f32, int_arm_neon_vcvtfp2fxs>;
3184 def VCVTf2xuq : N2VCvtQ<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
3185 v4i32, v4f32, int_arm_neon_vcvtfp2fxu>;
3186 def VCVTxs2fq : N2VCvtQ<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
3187 v4f32, v4i32, int_arm_neon_vcvtfxs2fp>;
3188 def VCVTxu2fq : N2VCvtQ<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
3189 v4f32, v4i32, int_arm_neon_vcvtfxu2fp>;
3193 // VREV64 : Vector Reverse elements within 64-bit doublewords
3195 class VREV64D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3196 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 0, 0, (outs DPR:$dst),
3197 (ins DPR:$src), IIC_VMOVD,
3198 OpcodeStr, Dt, "$dst, $src", "",
3199 [(set DPR:$dst, (Ty (NEONvrev64 (Ty DPR:$src))))]>;
3200 class VREV64Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3201 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 1, 0, (outs QPR:$dst),
3202 (ins QPR:$src), IIC_VMOVD,
3203 OpcodeStr, Dt, "$dst, $src", "",
3204 [(set QPR:$dst, (Ty (NEONvrev64 (Ty QPR:$src))))]>;
3206 def VREV64d8 : VREV64D<0b00, "vrev64", "8", v8i8>;
3207 def VREV64d16 : VREV64D<0b01, "vrev64", "16", v4i16>;
3208 def VREV64d32 : VREV64D<0b10, "vrev64", "32", v2i32>;
3209 def VREV64df : VREV64D<0b10, "vrev64", "32", v2f32>;
3211 def VREV64q8 : VREV64Q<0b00, "vrev64", "8", v16i8>;
3212 def VREV64q16 : VREV64Q<0b01, "vrev64", "16", v8i16>;
3213 def VREV64q32 : VREV64Q<0b10, "vrev64", "32", v4i32>;
3214 def VREV64qf : VREV64Q<0b10, "vrev64", "32", v4f32>;
3216 // VREV32 : Vector Reverse elements within 32-bit words
3218 class VREV32D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3219 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 0, 0, (outs DPR:$dst),
3220 (ins DPR:$src), IIC_VMOVD,
3221 OpcodeStr, Dt, "$dst, $src", "",
3222 [(set DPR:$dst, (Ty (NEONvrev32 (Ty DPR:$src))))]>;
3223 class VREV32Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3224 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 1, 0, (outs QPR:$dst),
3225 (ins QPR:$src), IIC_VMOVD,
3226 OpcodeStr, Dt, "$dst, $src", "",
3227 [(set QPR:$dst, (Ty (NEONvrev32 (Ty QPR:$src))))]>;
3229 def VREV32d8 : VREV32D<0b00, "vrev32", "8", v8i8>;
3230 def VREV32d16 : VREV32D<0b01, "vrev32", "16", v4i16>;
3232 def VREV32q8 : VREV32Q<0b00, "vrev32", "8", v16i8>;
3233 def VREV32q16 : VREV32Q<0b01, "vrev32", "16", v8i16>;
3235 // VREV16 : Vector Reverse elements within 16-bit halfwords
3237 class VREV16D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3238 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 0, 0, (outs DPR:$dst),
3239 (ins DPR:$src), IIC_VMOVD,
3240 OpcodeStr, Dt, "$dst, $src", "",
3241 [(set DPR:$dst, (Ty (NEONvrev16 (Ty DPR:$src))))]>;
3242 class VREV16Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3243 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 1, 0, (outs QPR:$dst),
3244 (ins QPR:$src), IIC_VMOVD,
3245 OpcodeStr, Dt, "$dst, $src", "",
3246 [(set QPR:$dst, (Ty (NEONvrev16 (Ty QPR:$src))))]>;
3248 def VREV16d8 : VREV16D<0b00, "vrev16", "8", v8i8>;
3249 def VREV16q8 : VREV16Q<0b00, "vrev16", "8", v16i8>;
3251 // Other Vector Shuffles.
3253 // VEXT : Vector Extract
3255 class VEXTd<string OpcodeStr, string Dt, ValueType Ty>
3256 : N3V<0,1,0b11,{?,?,?,?},0,0, (outs DPR:$dst),
3257 (ins DPR:$lhs, DPR:$rhs, i32imm:$index), IIC_VEXTD,
3258 OpcodeStr, Dt, "$dst, $lhs, $rhs, $index", "",
3259 [(set DPR:$dst, (Ty (NEONvext (Ty DPR:$lhs),
3260 (Ty DPR:$rhs), imm:$index)))]>;
3262 class VEXTq<string OpcodeStr, string Dt, ValueType Ty>
3263 : N3V<0,1,0b11,{?,?,?,?},1,0, (outs QPR:$dst),
3264 (ins QPR:$lhs, QPR:$rhs, i32imm:$index), IIC_VEXTQ,
3265 OpcodeStr, Dt, "$dst, $lhs, $rhs, $index", "",
3266 [(set QPR:$dst, (Ty (NEONvext (Ty QPR:$lhs),
3267 (Ty QPR:$rhs), imm:$index)))]>;
3269 def VEXTd8 : VEXTd<"vext", "8", v8i8>;
3270 def VEXTd16 : VEXTd<"vext", "16", v4i16>;
3271 def VEXTd32 : VEXTd<"vext", "32", v2i32>;
3272 def VEXTdf : VEXTd<"vext", "32", v2f32>;
3274 def VEXTq8 : VEXTq<"vext", "8", v16i8>;
3275 def VEXTq16 : VEXTq<"vext", "16", v8i16>;
3276 def VEXTq32 : VEXTq<"vext", "32", v4i32>;
3277 def VEXTqf : VEXTq<"vext", "32", v4f32>;
3279 // VTRN : Vector Transpose
3281 def VTRNd8 : N2VDShuffle<0b00, 0b00001, "vtrn", "8">;
3282 def VTRNd16 : N2VDShuffle<0b01, 0b00001, "vtrn", "16">;
3283 def VTRNd32 : N2VDShuffle<0b10, 0b00001, "vtrn", "32">;
3285 def VTRNq8 : N2VQShuffle<0b00, 0b00001, IIC_VPERMQ, "vtrn", "8">;
3286 def VTRNq16 : N2VQShuffle<0b01, 0b00001, IIC_VPERMQ, "vtrn", "16">;
3287 def VTRNq32 : N2VQShuffle<0b10, 0b00001, IIC_VPERMQ, "vtrn", "32">;
3289 // VUZP : Vector Unzip (Deinterleave)
3291 def VUZPd8 : N2VDShuffle<0b00, 0b00010, "vuzp", "8">;
3292 def VUZPd16 : N2VDShuffle<0b01, 0b00010, "vuzp", "16">;
3293 def VUZPd32 : N2VDShuffle<0b10, 0b00010, "vuzp", "32">;
3295 def VUZPq8 : N2VQShuffle<0b00, 0b00010, IIC_VPERMQ3, "vuzp", "8">;
3296 def VUZPq16 : N2VQShuffle<0b01, 0b00010, IIC_VPERMQ3, "vuzp", "16">;
3297 def VUZPq32 : N2VQShuffle<0b10, 0b00010, IIC_VPERMQ3, "vuzp", "32">;
3299 // VZIP : Vector Zip (Interleave)
3301 def VZIPd8 : N2VDShuffle<0b00, 0b00011, "vzip", "8">;
3302 def VZIPd16 : N2VDShuffle<0b01, 0b00011, "vzip", "16">;
3303 def VZIPd32 : N2VDShuffle<0b10, 0b00011, "vzip", "32">;
3305 def VZIPq8 : N2VQShuffle<0b00, 0b00011, IIC_VPERMQ3, "vzip", "8">;
3306 def VZIPq16 : N2VQShuffle<0b01, 0b00011, IIC_VPERMQ3, "vzip", "16">;
3307 def VZIPq32 : N2VQShuffle<0b10, 0b00011, IIC_VPERMQ3, "vzip", "32">;
3309 // Vector Table Lookup and Table Extension.
3311 // VTBL : Vector Table Lookup
3313 : N3V<1,1,0b11,0b1000,0,0, (outs DPR:$dst),
3314 (ins DPR:$tbl1, DPR:$src), IIC_VTB1,
3315 "vtbl", "8", "$dst, \\{$tbl1\\}, $src", "",
3316 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl1 DPR:$tbl1, DPR:$src)))]>;
3317 let hasExtraSrcRegAllocReq = 1 in {
3319 : N3V<1,1,0b11,0b1001,0,0, (outs DPR:$dst),
3320 (ins DPR:$tbl1, DPR:$tbl2, DPR:$src), IIC_VTB2,
3321 "vtbl", "8", "$dst, \\{$tbl1, $tbl2\\}, $src", "",
3322 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl2
3323 DPR:$tbl1, DPR:$tbl2, DPR:$src)))]>;
3325 : N3V<1,1,0b11,0b1010,0,0, (outs DPR:$dst),
3326 (ins DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$src), IIC_VTB3,
3327 "vtbl", "8", "$dst, \\{$tbl1, $tbl2, $tbl3\\}, $src", "",
3328 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl3
3329 DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$src)))]>;
3331 : N3V<1,1,0b11,0b1011,0,0, (outs DPR:$dst),
3332 (ins DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$src), IIC_VTB4,
3333 "vtbl", "8", "$dst, \\{$tbl1, $tbl2, $tbl3, $tbl4\\}, $src", "",
3334 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl4 DPR:$tbl1, DPR:$tbl2,
3335 DPR:$tbl3, DPR:$tbl4, DPR:$src)))]>;
3336 } // hasExtraSrcRegAllocReq = 1
3338 // VTBX : Vector Table Extension
3340 : N3V<1,1,0b11,0b1000,1,0, (outs DPR:$dst),
3341 (ins DPR:$orig, DPR:$tbl1, DPR:$src), IIC_VTBX1,
3342 "vtbx", "8", "$dst, \\{$tbl1\\}, $src", "$orig = $dst",
3343 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx1
3344 DPR:$orig, DPR:$tbl1, DPR:$src)))]>;
3345 let hasExtraSrcRegAllocReq = 1 in {
3347 : N3V<1,1,0b11,0b1001,1,0, (outs DPR:$dst),
3348 (ins DPR:$orig, DPR:$tbl1, DPR:$tbl2, DPR:$src), IIC_VTBX2,
3349 "vtbx", "8", "$dst, \\{$tbl1, $tbl2\\}, $src", "$orig = $dst",
3350 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx2
3351 DPR:$orig, DPR:$tbl1, DPR:$tbl2, DPR:$src)))]>;
3353 : N3V<1,1,0b11,0b1010,1,0, (outs DPR:$dst),
3354 (ins DPR:$orig, DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$src), IIC_VTBX3,
3355 "vtbx", "8", "$dst, \\{$tbl1, $tbl2, $tbl3\\}, $src", "$orig = $dst",
3356 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx3 DPR:$orig, DPR:$tbl1,
3357 DPR:$tbl2, DPR:$tbl3, DPR:$src)))]>;
3359 : N3V<1,1,0b11,0b1011,1,0, (outs DPR:$dst), (ins DPR:$orig, DPR:$tbl1,
3360 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$src), IIC_VTBX4,
3361 "vtbx", "8", "$dst, \\{$tbl1, $tbl2, $tbl3, $tbl4\\}, $src",
3363 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx4 DPR:$orig, DPR:$tbl1,
3364 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$src)))]>;
3365 } // hasExtraSrcRegAllocReq = 1
3367 //===----------------------------------------------------------------------===//
3368 // NEON instructions for single-precision FP math
3369 //===----------------------------------------------------------------------===//
3371 class N2VSPat<SDNode OpNode, ValueType ResTy, ValueType OpTy, NeonI Inst>
3372 : NEONFPPat<(ResTy (OpNode SPR:$a)),
3373 (EXTRACT_SUBREG (OpTy (Inst (INSERT_SUBREG (OpTy (IMPLICIT_DEF)),
3374 SPR:$a, arm_ssubreg_0))),
3377 class N3VSPat<SDNode OpNode, NeonI Inst>
3378 : NEONFPPat<(f32 (OpNode SPR:$a, SPR:$b)),
3379 (EXTRACT_SUBREG (v2f32
3380 (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3381 SPR:$a, arm_ssubreg_0),
3382 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3383 SPR:$b, arm_ssubreg_0))),
3386 class N3VSMulOpPat<SDNode MulNode, SDNode OpNode, NeonI Inst>
3387 : NEONFPPat<(f32 (OpNode SPR:$acc, (f32 (MulNode SPR:$a, SPR:$b)))),
3388 (EXTRACT_SUBREG (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3389 SPR:$acc, arm_ssubreg_0),
3390 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3391 SPR:$a, arm_ssubreg_0),
3392 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3393 SPR:$b, arm_ssubreg_0)),
3396 // These need separate instructions because they must use DPR_VFP2 register
3397 // class which have SPR sub-registers.
3399 // Vector Add Operations used for single-precision FP
3400 let neverHasSideEffects = 1 in
3401 def VADDfd_sfp : N3VS<0,0,0b00,0b1101,0, "vadd", "f32", v2f32, v2f32, fadd, 1>;
3402 def : N3VSPat<fadd, VADDfd_sfp>;
3404 // Vector Sub Operations used for single-precision FP
3405 let neverHasSideEffects = 1 in
3406 def VSUBfd_sfp : N3VS<0,0,0b10,0b1101,0, "vsub", "f32", v2f32, v2f32, fsub, 0>;
3407 def : N3VSPat<fsub, VSUBfd_sfp>;
3409 // Vector Multiply Operations used for single-precision FP
3410 let neverHasSideEffects = 1 in
3411 def VMULfd_sfp : N3VS<1,0,0b00,0b1101,1, "vmul", "f32", v2f32, v2f32, fmul, 1>;
3412 def : N3VSPat<fmul, VMULfd_sfp>;
3414 // Vector Multiply-Accumulate/Subtract used for single-precision FP
3415 // vml[as].f32 can cause 4-8 cycle stalls in following ASIMD instructions, so
3416 // we want to avoid them for now. e.g., alternating vmla/vadd instructions.
3418 //let neverHasSideEffects = 1 in
3419 //def VMLAfd_sfp : N3VSMulOp<0,0,0b00,0b1101,1, IIC_VMACD, "vmla", "f32",
3420 // v2f32, fmul, fadd>;
3421 //def : N3VSMulOpPat<fmul, fadd, VMLAfd_sfp>;
3423 //let neverHasSideEffects = 1 in
3424 //def VMLSfd_sfp : N3VSMulOp<0,0,0b10,0b1101,1, IIC_VMACD, "vmls", "f32",
3425 // v2f32, fmul, fsub>;
3426 //def : N3VSMulOpPat<fmul, fsub, VMLSfd_sfp>;
3428 // Vector Absolute used for single-precision FP
3429 let neverHasSideEffects = 1 in
3430 def VABSfd_sfp : N2V<0b11, 0b11, 0b10, 0b01, 0b01110, 0, 0,
3431 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), IIC_VUNAD,
3432 "vabs", "f32", "$dst, $src", "", []>;
3433 def : N2VSPat<fabs, f32, v2f32, VABSfd_sfp>;
3435 // Vector Negate used for single-precision FP
3436 let neverHasSideEffects = 1 in
3437 def VNEGfd_sfp : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
3438 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), IIC_VUNAD,
3439 "vneg", "f32", "$dst, $src", "", []>;
3440 def : N2VSPat<fneg, f32, v2f32, VNEGfd_sfp>;
3442 // Vector Maximum used for single-precision FP
3443 let neverHasSideEffects = 1 in
3444 def VMAXfd_sfp : N3V<0, 0, 0b00, 0b1111, 0, 0, (outs DPR_VFP2:$dst),
3445 (ins DPR_VFP2:$src1, DPR_VFP2:$src2), IIC_VBIND,
3446 "vmax", "f32", "$dst, $src1, $src2", "", []>;
3447 def : N3VSPat<NEONfmax, VMAXfd_sfp>;
3449 // Vector Minimum used for single-precision FP
3450 let neverHasSideEffects = 1 in
3451 def VMINfd_sfp : N3V<0, 0, 0b00, 0b1111, 0, 0, (outs DPR_VFP2:$dst),
3452 (ins DPR_VFP2:$src1, DPR_VFP2:$src2), IIC_VBIND,
3453 "vmin", "f32", "$dst, $src1, $src2", "", []>;
3454 def : N3VSPat<NEONfmin, VMINfd_sfp>;
3456 // Vector Convert between single-precision FP and integer
3457 let neverHasSideEffects = 1 in
3458 def VCVTf2sd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
3459 v2i32, v2f32, fp_to_sint>;
3460 def : N2VSPat<arm_ftosi, f32, v2f32, VCVTf2sd_sfp>;
3462 let neverHasSideEffects = 1 in
3463 def VCVTf2ud_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
3464 v2i32, v2f32, fp_to_uint>;
3465 def : N2VSPat<arm_ftoui, f32, v2f32, VCVTf2ud_sfp>;
3467 let neverHasSideEffects = 1 in
3468 def VCVTs2fd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
3469 v2f32, v2i32, sint_to_fp>;
3470 def : N2VSPat<arm_sitof, f32, v2i32, VCVTs2fd_sfp>;
3472 let neverHasSideEffects = 1 in
3473 def VCVTu2fd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
3474 v2f32, v2i32, uint_to_fp>;
3475 def : N2VSPat<arm_uitof, f32, v2i32, VCVTu2fd_sfp>;
3477 //===----------------------------------------------------------------------===//
3478 // Non-Instruction Patterns
3479 //===----------------------------------------------------------------------===//
3482 def : Pat<(v1i64 (bitconvert (v2i32 DPR:$src))), (v1i64 DPR:$src)>;
3483 def : Pat<(v1i64 (bitconvert (v4i16 DPR:$src))), (v1i64 DPR:$src)>;
3484 def : Pat<(v1i64 (bitconvert (v8i8 DPR:$src))), (v1i64 DPR:$src)>;
3485 def : Pat<(v1i64 (bitconvert (f64 DPR:$src))), (v1i64 DPR:$src)>;
3486 def : Pat<(v1i64 (bitconvert (v2f32 DPR:$src))), (v1i64 DPR:$src)>;
3487 def : Pat<(v2i32 (bitconvert (v1i64 DPR:$src))), (v2i32 DPR:$src)>;
3488 def : Pat<(v2i32 (bitconvert (v4i16 DPR:$src))), (v2i32 DPR:$src)>;
3489 def : Pat<(v2i32 (bitconvert (v8i8 DPR:$src))), (v2i32 DPR:$src)>;
3490 def : Pat<(v2i32 (bitconvert (f64 DPR:$src))), (v2i32 DPR:$src)>;
3491 def : Pat<(v2i32 (bitconvert (v2f32 DPR:$src))), (v2i32 DPR:$src)>;
3492 def : Pat<(v4i16 (bitconvert (v1i64 DPR:$src))), (v4i16 DPR:$src)>;
3493 def : Pat<(v4i16 (bitconvert (v2i32 DPR:$src))), (v4i16 DPR:$src)>;
3494 def : Pat<(v4i16 (bitconvert (v8i8 DPR:$src))), (v4i16 DPR:$src)>;
3495 def : Pat<(v4i16 (bitconvert (f64 DPR:$src))), (v4i16 DPR:$src)>;
3496 def : Pat<(v4i16 (bitconvert (v2f32 DPR:$src))), (v4i16 DPR:$src)>;
3497 def : Pat<(v8i8 (bitconvert (v1i64 DPR:$src))), (v8i8 DPR:$src)>;
3498 def : Pat<(v8i8 (bitconvert (v2i32 DPR:$src))), (v8i8 DPR:$src)>;
3499 def : Pat<(v8i8 (bitconvert (v4i16 DPR:$src))), (v8i8 DPR:$src)>;
3500 def : Pat<(v8i8 (bitconvert (f64 DPR:$src))), (v8i8 DPR:$src)>;
3501 def : Pat<(v8i8 (bitconvert (v2f32 DPR:$src))), (v8i8 DPR:$src)>;
3502 def : Pat<(f64 (bitconvert (v1i64 DPR:$src))), (f64 DPR:$src)>;
3503 def : Pat<(f64 (bitconvert (v2i32 DPR:$src))), (f64 DPR:$src)>;
3504 def : Pat<(f64 (bitconvert (v4i16 DPR:$src))), (f64 DPR:$src)>;
3505 def : Pat<(f64 (bitconvert (v8i8 DPR:$src))), (f64 DPR:$src)>;
3506 def : Pat<(f64 (bitconvert (v2f32 DPR:$src))), (f64 DPR:$src)>;
3507 def : Pat<(v2f32 (bitconvert (f64 DPR:$src))), (v2f32 DPR:$src)>;
3508 def : Pat<(v2f32 (bitconvert (v1i64 DPR:$src))), (v2f32 DPR:$src)>;
3509 def : Pat<(v2f32 (bitconvert (v2i32 DPR:$src))), (v2f32 DPR:$src)>;
3510 def : Pat<(v2f32 (bitconvert (v4i16 DPR:$src))), (v2f32 DPR:$src)>;
3511 def : Pat<(v2f32 (bitconvert (v8i8 DPR:$src))), (v2f32 DPR:$src)>;
3513 def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (v2i64 QPR:$src)>;
3514 def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (v2i64 QPR:$src)>;
3515 def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (v2i64 QPR:$src)>;
3516 def : Pat<(v2i64 (bitconvert (v2f64 QPR:$src))), (v2i64 QPR:$src)>;
3517 def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (v2i64 QPR:$src)>;
3518 def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (v4i32 QPR:$src)>;
3519 def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (v4i32 QPR:$src)>;
3520 def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (v4i32 QPR:$src)>;
3521 def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (v4i32 QPR:$src)>;
3522 def : Pat<(v4i32 (bitconvert (v4f32 QPR:$src))), (v4i32 QPR:$src)>;
3523 def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (v8i16 QPR:$src)>;
3524 def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (v8i16 QPR:$src)>;
3525 def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (v8i16 QPR:$src)>;
3526 def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (v8i16 QPR:$src)>;
3527 def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (v8i16 QPR:$src)>;
3528 def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (v16i8 QPR:$src)>;
3529 def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (v16i8 QPR:$src)>;
3530 def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (v16i8 QPR:$src)>;
3531 def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (v16i8 QPR:$src)>;
3532 def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (v16i8 QPR:$src)>;
3533 def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (v4f32 QPR:$src)>;
3534 def : Pat<(v4f32 (bitconvert (v4i32 QPR:$src))), (v4f32 QPR:$src)>;
3535 def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (v4f32 QPR:$src)>;
3536 def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (v4f32 QPR:$src)>;
3537 def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (v4f32 QPR:$src)>;
3538 def : Pat<(v2f64 (bitconvert (v2i64 QPR:$src))), (v2f64 QPR:$src)>;
3539 def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (v2f64 QPR:$src)>;
3540 def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 QPR:$src)>;
3541 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 QPR:$src)>;
3542 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 QPR:$src)>;