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 //===----------------------------------------------------------------------===//
118 let mayLoad = 1, neverHasSideEffects = 1 in {
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}, ${dst:dregpair}", "", []>;
127 // Use vld1 to load a Q register as a D register pair.
128 // This alternative to VLDMQ allows an alignment to be specified.
129 // This is equivalent to VLD1q64 except that it has a Q register operand.
131 : NLdSt<0,0b10,0b1010,0b1100, (outs QPR:$dst), (ins addrmode6:$addr),
132 IIC_VLD1, "vld1", "64", "${dst:dregpair}, $addr", "", []>;
133 } // mayLoad = 1, neverHasSideEffects = 1
135 let mayStore = 1, neverHasSideEffects = 1 in {
136 // Use vstmia to store a Q register as a D register pair.
137 // This is equivalent to VSTMD except that it has a Q register operand
138 // instead of a pair of D registers.
140 : AXDI5<(outs), (ins QPR:$src, addrmode5:$addr, pred:$p),
141 IndexModeNone, IIC_fpStorem,
142 "vstm${addr:submode}${p}\t${addr:base}, ${src:dregpair}", "", []>;
144 // Use vst1 to store a Q register as a D register pair.
145 // This alternative to VSTMQ allows an alignment to be specified.
146 // This is equivalent to VST1q64 except that it has a Q register operand.
148 : NLdSt<0,0b00,0b1010,0b1100, (outs), (ins addrmode6:$addr, QPR:$src),
149 IIC_VST, "vst1", "64", "${src:dregpair}, $addr", "", []>;
150 } // mayStore = 1, neverHasSideEffects = 1
152 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
154 // VLD1 : Vector Load (multiple single elements)
155 class VLD1D<bits<4> op7_4, string Dt>
156 : NLdSt<0,0b10,0b0111,op7_4, (outs DPR:$dst),
157 (ins addrmode6:$addr), IIC_VLD1,
158 "vld1", Dt, "\\{$dst\\}, $addr", "", []>;
159 class VLD1Q<bits<4> op7_4, string Dt>
160 : NLdSt<0,0b10,0b1010,op7_4, (outs DPR:$dst1, DPR:$dst2),
161 (ins addrmode6:$addr), IIC_VLD1,
162 "vld1", Dt, "\\{$dst1, $dst2\\}, $addr", "", []>;
164 def VLD1d8 : VLD1D<0b0000, "8">;
165 def VLD1d16 : VLD1D<0b0100, "16">;
166 def VLD1d32 : VLD1D<0b1000, "32">;
167 def VLD1d64 : VLD1D<0b1100, "64">;
169 def VLD1q8 : VLD1Q<0b0000, "8">;
170 def VLD1q16 : VLD1Q<0b0100, "16">;
171 def VLD1q32 : VLD1Q<0b1000, "32">;
172 def VLD1q64 : VLD1Q<0b1100, "64">;
174 // ...with address register writeback:
175 class VLD1DWB<bits<4> op7_4, string Dt>
176 : NLdSt<0,0b10,0b0111,op7_4, (outs DPR:$dst, GPR:$wb),
177 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD1,
178 "vld1", Dt, "\\{$dst\\}, $addr$offset",
179 "$addr.addr = $wb", []>;
180 class VLD1QWB<bits<4> op7_4, string Dt>
181 : NLdSt<0,0b10,0b1010,op7_4, (outs QPR:$dst, GPR:$wb),
182 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD1,
183 "vld1", Dt, "${dst:dregpair}, $addr$offset",
184 "$addr.addr = $wb", []>;
186 def VLD1d8_UPD : VLD1DWB<0b0000, "8">;
187 def VLD1d16_UPD : VLD1DWB<0b0100, "16">;
188 def VLD1d32_UPD : VLD1DWB<0b1000, "32">;
189 def VLD1d64_UPD : VLD1DWB<0b1100, "64">;
191 def VLD1q8_UPD : VLD1QWB<0b0000, "8">;
192 def VLD1q16_UPD : VLD1QWB<0b0100, "16">;
193 def VLD1q32_UPD : VLD1QWB<0b1000, "32">;
194 def VLD1q64_UPD : VLD1QWB<0b1100, "64">;
196 // ...with 3 registers (some of these are only for the disassembler):
197 class VLD1D3<bits<4> op7_4, string Dt>
198 : NLdSt<0,0b10,0b0110,op7_4, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3),
199 (ins addrmode6:$addr), IIC_VLD1, "vld1", Dt,
200 "\\{$dst1, $dst2, $dst3\\}, $addr", "", []>;
201 class VLD1D3WB<bits<4> op7_4, string Dt>
202 : NLdSt<0,0b10,0b0110,op7_4, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, GPR:$wb),
203 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD1, "vld1", Dt,
204 "\\{$dst1, $dst2, $dst3\\}, $addr$offset", "$addr.addr = $wb", []>;
206 def VLD1d8T : VLD1D3<0b0000, "8">;
207 def VLD1d16T : VLD1D3<0b0100, "16">;
208 def VLD1d32T : VLD1D3<0b1000, "32">;
209 def VLD1d64T : VLD1D3<0b1100, "64">;
211 def VLD1d8T_UPD : VLD1D3WB<0b0000, "8">;
212 def VLD1d16T_UPD : VLD1D3WB<0b0100, "16">;
213 def VLD1d32T_UPD : VLD1D3WB<0b1000, "32">;
214 def VLD1d64T_UPD : VLD1D3WB<0b1100, "64">;
216 // ...with 4 registers (some of these are only for the disassembler):
217 class VLD1D4<bits<4> op7_4, string Dt>
218 : NLdSt<0,0b10,0b0010,op7_4,(outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
219 (ins addrmode6:$addr), IIC_VLD1, "vld1", Dt,
220 "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr", "", []>;
221 class VLD1D4WB<bits<4> op7_4, string Dt>
222 : NLdSt<0,0b10,0b0010,op7_4,
223 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
224 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD1, "vld1", Dt,
225 "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr$offset", "$addr.addr = $wb",
228 def VLD1d8Q : VLD1D4<0b0000, "8">;
229 def VLD1d16Q : VLD1D4<0b0100, "16">;
230 def VLD1d32Q : VLD1D4<0b1000, "32">;
231 def VLD1d64Q : VLD1D4<0b1100, "64">;
233 def VLD1d8Q_UPD : VLD1D4WB<0b0000, "8">;
234 def VLD1d16Q_UPD : VLD1D4WB<0b0100, "16">;
235 def VLD1d32Q_UPD : VLD1D4WB<0b1000, "32">;
236 def VLD1d64Q_UPD : VLD1D4WB<0b1100, "64">;
238 // VLD2 : Vector Load (multiple 2-element structures)
239 class VLD2D<bits<4> op11_8, bits<4> op7_4, string Dt>
240 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2),
241 (ins addrmode6:$addr), IIC_VLD2,
242 "vld2", Dt, "\\{$dst1, $dst2\\}, $addr", "", []>;
243 class VLD2Q<bits<4> op7_4, string Dt>
244 : NLdSt<0, 0b10, 0b0011, op7_4,
245 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
246 (ins addrmode6:$addr), IIC_VLD2,
247 "vld2", Dt, "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr", "", []>;
249 def VLD2d8 : VLD2D<0b1000, 0b0000, "8">;
250 def VLD2d16 : VLD2D<0b1000, 0b0100, "16">;
251 def VLD2d32 : VLD2D<0b1000, 0b1000, "32">;
253 def VLD2q8 : VLD2Q<0b0000, "8">;
254 def VLD2q16 : VLD2Q<0b0100, "16">;
255 def VLD2q32 : VLD2Q<0b1000, "32">;
257 // ...with address register writeback:
258 class VLD2DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
259 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2, GPR:$wb),
260 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD2,
261 "vld2", Dt, "\\{$dst1, $dst2\\}, $addr$offset",
262 "$addr.addr = $wb", []>;
263 class VLD2QWB<bits<4> op7_4, string Dt>
264 : NLdSt<0, 0b10, 0b0011, op7_4,
265 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
266 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD2,
267 "vld2", Dt, "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr$offset",
268 "$addr.addr = $wb", []>;
270 def VLD2d8_UPD : VLD2DWB<0b1000, 0b0000, "8">;
271 def VLD2d16_UPD : VLD2DWB<0b1000, 0b0100, "16">;
272 def VLD2d32_UPD : VLD2DWB<0b1000, 0b1000, "32">;
274 def VLD2q8_UPD : VLD2QWB<0b0000, "8">;
275 def VLD2q16_UPD : VLD2QWB<0b0100, "16">;
276 def VLD2q32_UPD : VLD2QWB<0b1000, "32">;
278 // ...with double-spaced registers (for disassembly only):
279 def VLD2b8 : VLD2D<0b1001, 0b0000, "8">;
280 def VLD2b16 : VLD2D<0b1001, 0b0100, "16">;
281 def VLD2b32 : VLD2D<0b1001, 0b1000, "32">;
282 def VLD2b8_UPD : VLD2DWB<0b1001, 0b0000, "8">;
283 def VLD2b16_UPD : VLD2DWB<0b1001, 0b0100, "16">;
284 def VLD2b32_UPD : VLD2DWB<0b1001, 0b1000, "32">;
286 // VLD3 : Vector Load (multiple 3-element structures)
287 class VLD3D<bits<4> op11_8, bits<4> op7_4, string Dt>
288 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3),
289 (ins addrmode6:$addr), IIC_VLD3,
290 "vld3", Dt, "\\{$dst1, $dst2, $dst3\\}, $addr", "", []>;
292 def VLD3d8 : VLD3D<0b0100, 0b0000, "8">;
293 def VLD3d16 : VLD3D<0b0100, 0b0100, "16">;
294 def VLD3d32 : VLD3D<0b0100, 0b1000, "32">;
296 // ...with address register writeback:
297 class VLD3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
298 : NLdSt<0, 0b10, op11_8, op7_4,
299 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, GPR:$wb),
300 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD3,
301 "vld3", Dt, "\\{$dst1, $dst2, $dst3\\}, $addr$offset",
302 "$addr.addr = $wb", []>;
304 def VLD3d8_UPD : VLD3DWB<0b0100, 0b0000, "8">;
305 def VLD3d16_UPD : VLD3DWB<0b0100, 0b0100, "16">;
306 def VLD3d32_UPD : VLD3DWB<0b0100, 0b1000, "32">;
308 // ...with double-spaced registers (non-updating versions for disassembly only):
309 def VLD3q8 : VLD3D<0b0101, 0b0000, "8">;
310 def VLD3q16 : VLD3D<0b0101, 0b0100, "16">;
311 def VLD3q32 : VLD3D<0b0101, 0b1000, "32">;
312 def VLD3q8_UPD : VLD3DWB<0b0101, 0b0000, "8">;
313 def VLD3q16_UPD : VLD3DWB<0b0101, 0b0100, "16">;
314 def VLD3q32_UPD : VLD3DWB<0b0101, 0b1000, "32">;
316 // ...alternate versions to be allocated odd register numbers:
317 def VLD3q8odd_UPD : VLD3DWB<0b0101, 0b0000, "8">;
318 def VLD3q16odd_UPD : VLD3DWB<0b0101, 0b0100, "16">;
319 def VLD3q32odd_UPD : VLD3DWB<0b0101, 0b1000, "32">;
321 // VLD4 : Vector Load (multiple 4-element structures)
322 class VLD4D<bits<4> op11_8, bits<4> op7_4, string Dt>
323 : NLdSt<0, 0b10, op11_8, op7_4,
324 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
325 (ins addrmode6:$addr), IIC_VLD4,
326 "vld4", Dt, "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr", "", []>;
328 def VLD4d8 : VLD4D<0b0000, 0b0000, "8">;
329 def VLD4d16 : VLD4D<0b0000, 0b0100, "16">;
330 def VLD4d32 : VLD4D<0b0000, 0b1000, "32">;
332 // ...with address register writeback:
333 class VLD4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
334 : NLdSt<0, 0b10, op11_8, op7_4,
335 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
336 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD4,
337 "vld4", Dt, "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr$offset",
338 "$addr.addr = $wb", []>;
340 def VLD4d8_UPD : VLD4DWB<0b0000, 0b0000, "8">;
341 def VLD4d16_UPD : VLD4DWB<0b0000, 0b0100, "16">;
342 def VLD4d32_UPD : VLD4DWB<0b0000, 0b1000, "32">;
344 // ...with double-spaced registers (non-updating versions for disassembly only):
345 def VLD4q8 : VLD4D<0b0001, 0b0000, "8">;
346 def VLD4q16 : VLD4D<0b0001, 0b0100, "16">;
347 def VLD4q32 : VLD4D<0b0001, 0b1000, "32">;
348 def VLD4q8_UPD : VLD4DWB<0b0001, 0b0000, "8">;
349 def VLD4q16_UPD : VLD4DWB<0b0001, 0b0100, "16">;
350 def VLD4q32_UPD : VLD4DWB<0b0001, 0b1000, "32">;
352 // ...alternate versions to be allocated odd register numbers:
353 def VLD4q8odd_UPD : VLD4DWB<0b0001, 0b0000, "8">;
354 def VLD4q16odd_UPD : VLD4DWB<0b0001, 0b0100, "16">;
355 def VLD4q32odd_UPD : VLD4DWB<0b0001, 0b1000, "32">;
357 // VLD1LN : Vector Load (single element to one lane)
358 // FIXME: Not yet implemented.
360 // VLD2LN : Vector Load (single 2-element structure to one lane)
361 class VLD2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
362 : NLdSt<1, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2),
363 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, nohash_imm:$lane),
364 IIC_VLD2, "vld2", Dt, "\\{$dst1[$lane], $dst2[$lane]\\}, $addr",
365 "$src1 = $dst1, $src2 = $dst2", []>;
367 def VLD2LNd8 : VLD2LN<0b0001, {?,?,?,?}, "8">;
368 def VLD2LNd16 : VLD2LN<0b0101, {?,?,0,?}, "16">;
369 def VLD2LNd32 : VLD2LN<0b1001, {?,0,?,?}, "32">;
371 // ...with double-spaced registers:
372 def VLD2LNq16 : VLD2LN<0b0101, {?,?,1,?}, "16">;
373 def VLD2LNq32 : VLD2LN<0b1001, {?,1,?,?}, "32">;
375 // ...alternate versions to be allocated odd register numbers:
376 def VLD2LNq16odd : VLD2LN<0b0101, {?,?,1,?}, "16">;
377 def VLD2LNq32odd : VLD2LN<0b1001, {?,1,?,?}, "32">;
379 // ...with address register writeback:
380 class VLD2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
381 : NLdSt<1, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2, GPR:$wb),
382 (ins addrmode6:$addr, am6offset:$offset,
383 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VLD2, "vld2", Dt,
384 "\\{$dst1[$lane], $dst2[$lane]\\}, $addr$offset",
385 "$src1 = $dst1, $src2 = $dst2, $addr.addr = $wb", []>;
387 def VLD2LNd8_UPD : VLD2LNWB<0b0001, {?,?,?,?}, "8">;
388 def VLD2LNd16_UPD : VLD2LNWB<0b0101, {?,?,0,?}, "16">;
389 def VLD2LNd32_UPD : VLD2LNWB<0b1001, {?,0,?,?}, "32">;
391 def VLD2LNq16_UPD : VLD2LNWB<0b0101, {?,?,1,?}, "16">;
392 def VLD2LNq32_UPD : VLD2LNWB<0b1001, {?,1,?,?}, "32">;
394 // VLD3LN : Vector Load (single 3-element structure to one lane)
395 class VLD3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
396 : NLdSt<1, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3),
397 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3,
398 nohash_imm:$lane), IIC_VLD3, "vld3", Dt,
399 "\\{$dst1[$lane], $dst2[$lane], $dst3[$lane]\\}, $addr",
400 "$src1 = $dst1, $src2 = $dst2, $src3 = $dst3", []>;
402 def VLD3LNd8 : VLD3LN<0b0010, {?,?,?,0}, "8">;
403 def VLD3LNd16 : VLD3LN<0b0110, {?,?,0,0}, "16">;
404 def VLD3LNd32 : VLD3LN<0b1010, {?,0,0,0}, "32">;
406 // ...with double-spaced registers:
407 def VLD3LNq16 : VLD3LN<0b0110, {?,?,1,0}, "16">;
408 def VLD3LNq32 : VLD3LN<0b1010, {?,1,0,0}, "32">;
410 // ...alternate versions to be allocated odd register numbers:
411 def VLD3LNq16odd : VLD3LN<0b0110, {?,?,1,0}, "16">;
412 def VLD3LNq32odd : VLD3LN<0b1010, {?,1,0,0}, "32">;
414 // ...with address register writeback:
415 class VLD3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
416 : NLdSt<1, 0b10, op11_8, op7_4,
417 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, GPR:$wb),
418 (ins addrmode6:$addr, am6offset:$offset,
419 DPR:$src1, DPR:$src2, DPR:$src3, nohash_imm:$lane),
420 IIC_VLD3, "vld3", Dt,
421 "\\{$dst1[$lane], $dst2[$lane], $dst3[$lane]\\}, $addr$offset",
422 "$src1 = $dst1, $src2 = $dst2, $src3 = $dst3, $addr.addr = $wb",
425 def VLD3LNd8_UPD : VLD3LNWB<0b0010, {?,?,?,0}, "8">;
426 def VLD3LNd16_UPD : VLD3LNWB<0b0110, {?,?,0,0}, "16">;
427 def VLD3LNd32_UPD : VLD3LNWB<0b1010, {?,0,0,0}, "32">;
429 def VLD3LNq16_UPD : VLD3LNWB<0b0110, {?,?,1,0}, "16">;
430 def VLD3LNq32_UPD : VLD3LNWB<0b1010, {?,1,0,0}, "32">;
432 // VLD4LN : Vector Load (single 4-element structure to one lane)
433 class VLD4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
434 : NLdSt<1, 0b10, op11_8, op7_4,
435 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
436 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
437 nohash_imm:$lane), IIC_VLD4, "vld4", Dt,
438 "\\{$dst1[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $addr",
439 "$src1 = $dst1, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4", []>;
441 def VLD4LNd8 : VLD4LN<0b0011, {?,?,?,?}, "8">;
442 def VLD4LNd16 : VLD4LN<0b0111, {?,?,0,?}, "16">;
443 def VLD4LNd32 : VLD4LN<0b1011, {?,0,?,?}, "32">;
445 // ...with double-spaced registers:
446 def VLD4LNq16 : VLD4LN<0b0111, {?,?,1,?}, "16">;
447 def VLD4LNq32 : VLD4LN<0b1011, {?,1,?,?}, "32">;
449 // ...alternate versions to be allocated odd register numbers:
450 def VLD4LNq16odd : VLD4LN<0b0111, {?,?,1,?}, "16">;
451 def VLD4LNq32odd : VLD4LN<0b1011, {?,1,?,?}, "32">;
453 // ...with address register writeback:
454 class VLD4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
455 : NLdSt<1, 0b10, op11_8, op7_4,
456 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
457 (ins addrmode6:$addr, am6offset:$offset,
458 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
459 IIC_VLD4, "vld4", Dt,
460 "\\{$dst1[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $addr$offset",
461 "$src1 = $dst1, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4, $addr.addr = $wb",
464 def VLD4LNd8_UPD : VLD4LNWB<0b0011, {?,?,?,?}, "8">;
465 def VLD4LNd16_UPD : VLD4LNWB<0b0111, {?,?,0,?}, "16">;
466 def VLD4LNd32_UPD : VLD4LNWB<0b1011, {?,0,?,?}, "32">;
468 def VLD4LNq16_UPD : VLD4LNWB<0b0111, {?,?,1,?}, "16">;
469 def VLD4LNq32_UPD : VLD4LNWB<0b1011, {?,1,?,?}, "32">;
471 // VLD1DUP : Vector Load (single element to all lanes)
472 // VLD2DUP : Vector Load (single 2-element structure to all lanes)
473 // VLD3DUP : Vector Load (single 3-element structure to all lanes)
474 // VLD4DUP : Vector Load (single 4-element structure to all lanes)
475 // FIXME: Not yet implemented.
476 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
478 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
480 // VST1 : Vector Store (multiple single elements)
481 class VST1D<bits<4> op7_4, string Dt>
482 : NLdSt<0,0b00,0b0111,op7_4, (outs), (ins addrmode6:$addr, DPR:$src), IIC_VST,
483 "vst1", Dt, "\\{$src\\}, $addr", "", []>;
484 class VST1Q<bits<4> op7_4, string Dt>
485 : NLdSt<0,0b00,0b1010,op7_4, (outs),
486 (ins addrmode6:$addr, DPR:$src1, DPR:$src2), IIC_VST,
487 "vst1", Dt, "\\{$src1, $src2\\}, $addr", "", []>;
489 def VST1d8 : VST1D<0b0000, "8">;
490 def VST1d16 : VST1D<0b0100, "16">;
491 def VST1d32 : VST1D<0b1000, "32">;
492 def VST1d64 : VST1D<0b1100, "64">;
494 def VST1q8 : VST1Q<0b0000, "8">;
495 def VST1q16 : VST1Q<0b0100, "16">;
496 def VST1q32 : VST1Q<0b1000, "32">;
497 def VST1q64 : VST1Q<0b1100, "64">;
499 // ...with address register writeback:
500 class VST1DWB<bits<4> op7_4, string Dt>
501 : NLdSt<0, 0b00, 0b0111, op7_4, (outs GPR:$wb),
502 (ins addrmode6:$addr, am6offset:$offset, DPR:$src), IIC_VST,
503 "vst1", Dt, "\\{$src\\}, $addr$offset", "$addr.addr = $wb", []>;
504 class VST1QWB<bits<4> op7_4, string Dt>
505 : NLdSt<0, 0b00, 0b1010, op7_4, (outs GPR:$wb),
506 (ins addrmode6:$addr, am6offset:$offset, QPR:$src), IIC_VST,
507 "vst1", Dt, "${src:dregpair}, $addr$offset", "$addr.addr = $wb", []>;
509 def VST1d8_UPD : VST1DWB<0b0000, "8">;
510 def VST1d16_UPD : VST1DWB<0b0100, "16">;
511 def VST1d32_UPD : VST1DWB<0b1000, "32">;
512 def VST1d64_UPD : VST1DWB<0b1100, "64">;
514 def VST1q8_UPD : VST1QWB<0b0000, "8">;
515 def VST1q16_UPD : VST1QWB<0b0100, "16">;
516 def VST1q32_UPD : VST1QWB<0b1000, "32">;
517 def VST1q64_UPD : VST1QWB<0b1100, "64">;
519 // ...with 3 registers (some of these are only for the disassembler):
520 class VST1D3<bits<4> op7_4, string Dt>
521 : NLdSt<0, 0b00, 0b0110, op7_4, (outs),
522 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3),
523 IIC_VST, "vst1", Dt, "\\{$src1, $src2, $src3\\}, $addr", "", []>;
524 class VST1D3WB<bits<4> op7_4, string Dt>
525 : NLdSt<0, 0b00, 0b0110, op7_4, (outs GPR:$wb),
526 (ins addrmode6:$addr, am6offset:$offset,
527 DPR:$src1, DPR:$src2, DPR:$src3),
528 IIC_VST, "vst1", Dt, "\\{$src1, $src2, $src3\\}, $addr$offset",
529 "$addr.addr = $wb", []>;
531 def VST1d8T : VST1D3<0b0000, "8">;
532 def VST1d16T : VST1D3<0b0100, "16">;
533 def VST1d32T : VST1D3<0b1000, "32">;
534 def VST1d64T : VST1D3<0b1100, "64">;
536 def VST1d8T_UPD : VST1D3WB<0b0000, "8">;
537 def VST1d16T_UPD : VST1D3WB<0b0100, "16">;
538 def VST1d32T_UPD : VST1D3WB<0b1000, "32">;
539 def VST1d64T_UPD : VST1D3WB<0b1100, "64">;
541 // ...with 4 registers (some of these are only for the disassembler):
542 class VST1D4<bits<4> op7_4, string Dt>
543 : NLdSt<0, 0b00, 0b0010, op7_4, (outs),
544 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
545 IIC_VST, "vst1", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr", "",
547 class VST1D4WB<bits<4> op7_4, string Dt>
548 : NLdSt<0, 0b00, 0b0010, op7_4, (outs GPR:$wb),
549 (ins addrmode6:$addr, am6offset:$offset,
550 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
551 IIC_VST, "vst1", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr$offset",
552 "$addr.addr = $wb", []>;
554 def VST1d8Q : VST1D4<0b0000, "8">;
555 def VST1d16Q : VST1D4<0b0100, "16">;
556 def VST1d32Q : VST1D4<0b1000, "32">;
557 def VST1d64Q : VST1D4<0b1100, "64">;
559 def VST1d8Q_UPD : VST1D4WB<0b0000, "8">;
560 def VST1d16Q_UPD : VST1D4WB<0b0100, "16">;
561 def VST1d32Q_UPD : VST1D4WB<0b1000, "32">;
562 def VST1d64Q_UPD : VST1D4WB<0b1100, "64">;
564 // VST2 : Vector Store (multiple 2-element structures)
565 class VST2D<bits<4> op11_8, bits<4> op7_4, string Dt>
566 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
567 (ins addrmode6:$addr, DPR:$src1, DPR:$src2),
568 IIC_VST, "vst2", Dt, "\\{$src1, $src2\\}, $addr", "", []>;
569 class VST2Q<bits<4> op7_4, string Dt>
570 : NLdSt<0, 0b00, 0b0011, op7_4, (outs),
571 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
572 IIC_VST, "vst2", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr",
575 def VST2d8 : VST2D<0b1000, 0b0000, "8">;
576 def VST2d16 : VST2D<0b1000, 0b0100, "16">;
577 def VST2d32 : VST2D<0b1000, 0b1000, "32">;
579 def VST2q8 : VST2Q<0b0000, "8">;
580 def VST2q16 : VST2Q<0b0100, "16">;
581 def VST2q32 : VST2Q<0b1000, "32">;
583 // ...with address register writeback:
584 class VST2DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
585 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
586 (ins addrmode6:$addr, am6offset:$offset, DPR:$src1, DPR:$src2),
587 IIC_VST, "vst2", Dt, "\\{$src1, $src2\\}, $addr$offset",
588 "$addr.addr = $wb", []>;
589 class VST2QWB<bits<4> op7_4, string Dt>
590 : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
591 (ins addrmode6:$addr, am6offset:$offset,
592 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
593 IIC_VST, "vst2", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr$offset",
594 "$addr.addr = $wb", []>;
596 def VST2d8_UPD : VST2DWB<0b1000, 0b0000, "8">;
597 def VST2d16_UPD : VST2DWB<0b1000, 0b0100, "16">;
598 def VST2d32_UPD : VST2DWB<0b1000, 0b1000, "32">;
600 def VST2q8_UPD : VST2QWB<0b0000, "8">;
601 def VST2q16_UPD : VST2QWB<0b0100, "16">;
602 def VST2q32_UPD : VST2QWB<0b1000, "32">;
604 // ...with double-spaced registers (for disassembly only):
605 def VST2b8 : VST2D<0b1001, 0b0000, "8">;
606 def VST2b16 : VST2D<0b1001, 0b0100, "16">;
607 def VST2b32 : VST2D<0b1001, 0b1000, "32">;
608 def VST2b8_UPD : VST2DWB<0b1001, 0b0000, "8">;
609 def VST2b16_UPD : VST2DWB<0b1001, 0b0100, "16">;
610 def VST2b32_UPD : VST2DWB<0b1001, 0b1000, "32">;
612 // VST3 : Vector Store (multiple 3-element structures)
613 class VST3D<bits<4> op11_8, bits<4> op7_4, string Dt>
614 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
615 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3), IIC_VST,
616 "vst3", Dt, "\\{$src1, $src2, $src3\\}, $addr", "", []>;
618 def VST3d8 : VST3D<0b0100, 0b0000, "8">;
619 def VST3d16 : VST3D<0b0100, 0b0100, "16">;
620 def VST3d32 : VST3D<0b0100, 0b1000, "32">;
622 // ...with address register writeback:
623 class VST3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
624 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
625 (ins addrmode6:$addr, am6offset:$offset,
626 DPR:$src1, DPR:$src2, DPR:$src3), IIC_VST,
627 "vst3", Dt, "\\{$src1, $src2, $src3\\}, $addr$offset",
628 "$addr.addr = $wb", []>;
630 def VST3d8_UPD : VST3DWB<0b0100, 0b0000, "8">;
631 def VST3d16_UPD : VST3DWB<0b0100, 0b0100, "16">;
632 def VST3d32_UPD : VST3DWB<0b0100, 0b1000, "32">;
634 // ...with double-spaced registers (non-updating versions for disassembly only):
635 def VST3q8 : VST3D<0b0101, 0b0000, "8">;
636 def VST3q16 : VST3D<0b0101, 0b0100, "16">;
637 def VST3q32 : VST3D<0b0101, 0b1000, "32">;
638 def VST3q8_UPD : VST3DWB<0b0101, 0b0000, "8">;
639 def VST3q16_UPD : VST3DWB<0b0101, 0b0100, "16">;
640 def VST3q32_UPD : VST3DWB<0b0101, 0b1000, "32">;
642 // ...alternate versions to be allocated odd register numbers:
643 def VST3q8odd_UPD : VST3DWB<0b0101, 0b0000, "8">;
644 def VST3q16odd_UPD : VST3DWB<0b0101, 0b0100, "16">;
645 def VST3q32odd_UPD : VST3DWB<0b0101, 0b1000, "32">;
647 // VST4 : Vector Store (multiple 4-element structures)
648 class VST4D<bits<4> op11_8, bits<4> op7_4, string Dt>
649 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
650 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
651 IIC_VST, "vst4", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr",
654 def VST4d8 : VST4D<0b0000, 0b0000, "8">;
655 def VST4d16 : VST4D<0b0000, 0b0100, "16">;
656 def VST4d32 : VST4D<0b0000, 0b1000, "32">;
658 // ...with address register writeback:
659 class VST4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
660 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
661 (ins addrmode6:$addr, am6offset:$offset,
662 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST,
663 "vst4", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr$offset",
664 "$addr.addr = $wb", []>;
666 def VST4d8_UPD : VST4DWB<0b0000, 0b0000, "8">;
667 def VST4d16_UPD : VST4DWB<0b0000, 0b0100, "16">;
668 def VST4d32_UPD : VST4DWB<0b0000, 0b1000, "32">;
670 // ...with double-spaced registers (non-updating versions for disassembly only):
671 def VST4q8 : VST4D<0b0001, 0b0000, "8">;
672 def VST4q16 : VST4D<0b0001, 0b0100, "16">;
673 def VST4q32 : VST4D<0b0001, 0b1000, "32">;
674 def VST4q8_UPD : VST4DWB<0b0001, 0b0000, "8">;
675 def VST4q16_UPD : VST4DWB<0b0001, 0b0100, "16">;
676 def VST4q32_UPD : VST4DWB<0b0001, 0b1000, "32">;
678 // ...alternate versions to be allocated odd register numbers:
679 def VST4q8odd_UPD : VST4DWB<0b0001, 0b0000, "8">;
680 def VST4q16odd_UPD : VST4DWB<0b0001, 0b0100, "16">;
681 def VST4q32odd_UPD : VST4DWB<0b0001, 0b1000, "32">;
683 // VST1LN : Vector Store (single element from one lane)
684 // FIXME: Not yet implemented.
686 // VST2LN : Vector Store (single 2-element structure from one lane)
687 class VST2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
688 : NLdSt<1, 0b00, op11_8, op7_4, (outs),
689 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, nohash_imm:$lane),
690 IIC_VST, "vst2", Dt, "\\{$src1[$lane], $src2[$lane]\\}, $addr",
693 def VST2LNd8 : VST2LN<0b0001, {?,?,?,?}, "8">;
694 def VST2LNd16 : VST2LN<0b0101, {?,?,0,?}, "16">;
695 def VST2LNd32 : VST2LN<0b1001, {?,0,?,?}, "32">;
697 // ...with double-spaced registers:
698 def VST2LNq16 : VST2LN<0b0101, {?,?,1,?}, "16">;
699 def VST2LNq32 : VST2LN<0b1001, {?,1,?,?}, "32">;
701 // ...alternate versions to be allocated odd register numbers:
702 def VST2LNq16odd : VST2LN<0b0101, {?,?,1,?}, "16">;
703 def VST2LNq32odd : VST2LN<0b1001, {?,1,?,?}, "32">;
705 // ...with address register writeback:
706 class VST2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
707 : NLdSt<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
708 (ins addrmode6:$addr, am6offset:$offset,
709 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VST, "vst2", Dt,
710 "\\{$src1[$lane], $src2[$lane]\\}, $addr$offset",
711 "$addr.addr = $wb", []>;
713 def VST2LNd8_UPD : VST2LNWB<0b0001, {?,?,?,?}, "8">;
714 def VST2LNd16_UPD : VST2LNWB<0b0101, {?,?,0,?}, "16">;
715 def VST2LNd32_UPD : VST2LNWB<0b1001, {?,0,?,?}, "32">;
717 def VST2LNq16_UPD : VST2LNWB<0b0101, {?,?,1,?}, "16">;
718 def VST2LNq32_UPD : VST2LNWB<0b1001, {?,1,?,?}, "32">;
720 // VST3LN : Vector Store (single 3-element structure from one lane)
721 class VST3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
722 : NLdSt<1, 0b00, op11_8, op7_4, (outs),
723 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3,
724 nohash_imm:$lane), IIC_VST, "vst3", Dt,
725 "\\{$src1[$lane], $src2[$lane], $src3[$lane]\\}, $addr", "", []>;
727 def VST3LNd8 : VST3LN<0b0010, {?,?,?,0}, "8">;
728 def VST3LNd16 : VST3LN<0b0110, {?,?,0,0}, "16">;
729 def VST3LNd32 : VST3LN<0b1010, {?,0,0,0}, "32">;
731 // ...with double-spaced registers:
732 def VST3LNq16 : VST3LN<0b0110, {?,?,1,0}, "16">;
733 def VST3LNq32 : VST3LN<0b1010, {?,1,0,0}, "32">;
735 // ...alternate versions to be allocated odd register numbers:
736 def VST3LNq16odd : VST3LN<0b0110, {?,?,1,0}, "16">;
737 def VST3LNq32odd : VST3LN<0b1010, {?,1,0,0}, "32">;
739 // ...with address register writeback:
740 class VST3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
741 : NLdSt<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
742 (ins addrmode6:$addr, am6offset:$offset,
743 DPR:$src1, DPR:$src2, DPR:$src3, nohash_imm:$lane),
745 "\\{$src1[$lane], $src2[$lane], $src3[$lane]\\}, $addr$offset",
746 "$addr.addr = $wb", []>;
748 def VST3LNd8_UPD : VST3LNWB<0b0010, {?,?,?,0}, "8">;
749 def VST3LNd16_UPD : VST3LNWB<0b0110, {?,?,0,0}, "16">;
750 def VST3LNd32_UPD : VST3LNWB<0b1010, {?,0,0,0}, "32">;
752 def VST3LNq16_UPD : VST3LNWB<0b0110, {?,?,1,0}, "16">;
753 def VST3LNq32_UPD : VST3LNWB<0b1010, {?,1,0,0}, "32">;
755 // VST4LN : Vector Store (single 4-element structure from one lane)
756 class VST4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
757 : NLdSt<1, 0b00, op11_8, op7_4, (outs),
758 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
759 nohash_imm:$lane), IIC_VST, "vst4", Dt,
760 "\\{$src1[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $addr",
763 def VST4LNd8 : VST4LN<0b0011, {?,?,?,?}, "8">;
764 def VST4LNd16 : VST4LN<0b0111, {?,?,0,?}, "16">;
765 def VST4LNd32 : VST4LN<0b1011, {?,0,?,?}, "32">;
767 // ...with double-spaced registers:
768 def VST4LNq16 : VST4LN<0b0111, {?,?,1,?}, "16">;
769 def VST4LNq32 : VST4LN<0b1011, {?,1,?,?}, "32">;
771 // ...alternate versions to be allocated odd register numbers:
772 def VST4LNq16odd : VST4LN<0b0111, {?,?,1,?}, "16">;
773 def VST4LNq32odd : VST4LN<0b1011, {?,1,?,?}, "32">;
775 // ...with address register writeback:
776 class VST4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
777 : NLdSt<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
778 (ins addrmode6:$addr, am6offset:$offset,
779 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
781 "\\{$src1[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $addr$offset",
782 "$addr.addr = $wb", []>;
784 def VST4LNd8_UPD : VST4LNWB<0b0011, {?,?,?,?}, "8">;
785 def VST4LNd16_UPD : VST4LNWB<0b0111, {?,?,0,?}, "16">;
786 def VST4LNd32_UPD : VST4LNWB<0b1011, {?,0,?,?}, "32">;
788 def VST4LNq16_UPD : VST4LNWB<0b0111, {?,?,1,?}, "16">;
789 def VST4LNq32_UPD : VST4LNWB<0b1011, {?,1,?,?}, "32">;
791 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
794 //===----------------------------------------------------------------------===//
795 // NEON pattern fragments
796 //===----------------------------------------------------------------------===//
798 // Extract D sub-registers of Q registers.
799 def DSubReg_i8_reg : SDNodeXForm<imm, [{
800 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
801 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/8, MVT::i32);
803 def DSubReg_i16_reg : SDNodeXForm<imm, [{
804 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
805 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/4, MVT::i32);
807 def DSubReg_i32_reg : SDNodeXForm<imm, [{
808 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
809 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/2, MVT::i32);
811 def DSubReg_f64_reg : SDNodeXForm<imm, [{
812 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
813 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue(), MVT::i32);
815 def DSubReg_f64_other_reg : SDNodeXForm<imm, [{
816 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
817 return CurDAG->getTargetConstant(ARM::dsub_0 + (1 - N->getZExtValue()),
821 // Extract S sub-registers of Q/D registers.
822 def SSubReg_f32_reg : SDNodeXForm<imm, [{
823 assert(ARM::ssub_3 == ARM::ssub_0+3 && "Unexpected subreg numbering");
824 return CurDAG->getTargetConstant(ARM::ssub_0 + N->getZExtValue(), MVT::i32);
827 // Translate lane numbers from Q registers to D subregs.
828 def SubReg_i8_lane : SDNodeXForm<imm, [{
829 return CurDAG->getTargetConstant(N->getZExtValue() & 7, MVT::i32);
831 def SubReg_i16_lane : SDNodeXForm<imm, [{
832 return CurDAG->getTargetConstant(N->getZExtValue() & 3, MVT::i32);
834 def SubReg_i32_lane : SDNodeXForm<imm, [{
835 return CurDAG->getTargetConstant(N->getZExtValue() & 1, MVT::i32);
838 //===----------------------------------------------------------------------===//
839 // Instruction Classes
840 //===----------------------------------------------------------------------===//
842 // Basic 2-register operations: single-, double- and quad-register.
843 class N2VS<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
844 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
845 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
846 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
847 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src),
848 IIC_VUNAD, OpcodeStr, Dt, "$dst, $src", "", []>;
849 class N2VD<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
850 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
851 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
852 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
853 (ins DPR:$src), IIC_VUNAD, OpcodeStr, Dt,"$dst, $src", "",
854 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src))))]>;
855 class N2VQ<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
856 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
857 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
858 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
859 (ins QPR:$src), IIC_VUNAQ, OpcodeStr, Dt,"$dst, $src", "",
860 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src))))]>;
862 // Basic 2-register intrinsics, both double- and quad-register.
863 class N2VDInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
864 bits<2> op17_16, bits<5> op11_7, bit op4,
865 InstrItinClass itin, string OpcodeStr, string Dt,
866 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
867 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
868 (ins DPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
869 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
870 class N2VQInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
871 bits<2> op17_16, bits<5> op11_7, bit op4,
872 InstrItinClass itin, string OpcodeStr, string Dt,
873 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
874 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
875 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
876 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
878 // Narrow 2-register intrinsics.
879 class N2VNInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
880 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
881 InstrItinClass itin, string OpcodeStr, string Dt,
882 ValueType TyD, ValueType TyQ, Intrinsic IntOp>
883 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$dst),
884 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
885 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src))))]>;
887 // Long 2-register intrinsics (currently only used for VMOVL).
888 class N2VLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
889 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
890 InstrItinClass itin, string OpcodeStr, string Dt,
891 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
892 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$dst),
893 (ins DPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
894 [(set QPR:$dst, (TyQ (IntOp (TyD DPR:$src))))]>;
896 // 2-register shuffles (VTRN/VZIP/VUZP), both double- and quad-register.
897 class N2VDShuffle<bits<2> op19_18, bits<5> op11_7, string OpcodeStr, string Dt>
898 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 0, 0, (outs DPR:$dst1, DPR:$dst2),
899 (ins DPR:$src1, DPR:$src2), IIC_VPERMD,
900 OpcodeStr, Dt, "$dst1, $dst2",
901 "$src1 = $dst1, $src2 = $dst2", []>;
902 class N2VQShuffle<bits<2> op19_18, bits<5> op11_7,
903 InstrItinClass itin, string OpcodeStr, string Dt>
904 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 1, 0, (outs QPR:$dst1, QPR:$dst2),
905 (ins QPR:$src1, QPR:$src2), itin, OpcodeStr, Dt, "$dst1, $dst2",
906 "$src1 = $dst1, $src2 = $dst2", []>;
908 // Basic 3-register operations: single-, double- and quad-register.
909 class N3VS<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
910 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
911 SDNode OpNode, bit Commutable>
912 : N3V<op24, op23, op21_20, op11_8, 0, op4,
913 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src1, DPR_VFP2:$src2), N3RegFrm,
914 IIC_VBIND, OpcodeStr, Dt, "$dst, $src1, $src2", "", []> {
915 let isCommutable = Commutable;
918 class N3VD<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
919 InstrItinClass itin, string OpcodeStr, string Dt,
920 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
921 : N3V<op24, op23, op21_20, op11_8, 0, op4,
922 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
923 OpcodeStr, Dt, "$dst, $src1, $src2", "",
924 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src1), (OpTy DPR:$src2))))]> {
925 let isCommutable = Commutable;
927 // Same as N3VD but no data type.
928 class N3VDX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
929 InstrItinClass itin, string OpcodeStr,
930 ValueType ResTy, ValueType OpTy,
931 SDNode OpNode, bit Commutable>
932 : N3VX<op24, op23, op21_20, op11_8, 0, op4,
933 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
934 OpcodeStr, "$dst, $src1, $src2", "",
935 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src1), (OpTy DPR:$src2))))]>{
936 let isCommutable = Commutable;
939 class N3VDSL<bits<2> op21_20, bits<4> op11_8,
940 InstrItinClass itin, string OpcodeStr, string Dt,
941 ValueType Ty, SDNode ShOp>
942 : N3V<0, 1, op21_20, op11_8, 1, 0,
943 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
944 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
946 (Ty (ShOp (Ty DPR:$src1),
947 (Ty (NEONvduplane (Ty DPR_VFP2:$src2),imm:$lane)))))]> {
948 let isCommutable = 0;
950 class N3VDSL16<bits<2> op21_20, bits<4> op11_8,
951 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
952 : N3V<0, 1, op21_20, op11_8, 1, 0,
953 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
954 NVMulSLFrm, IIC_VMULi16D, OpcodeStr, Dt,"$dst, $src1, $src2[$lane]","",
956 (Ty (ShOp (Ty DPR:$src1),
957 (Ty (NEONvduplane (Ty DPR_8:$src2), imm:$lane)))))]> {
958 let isCommutable = 0;
961 class N3VQ<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
962 InstrItinClass itin, string OpcodeStr, string Dt,
963 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
964 : N3V<op24, op23, op21_20, op11_8, 1, op4,
965 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), N3RegFrm, itin,
966 OpcodeStr, Dt, "$dst, $src1, $src2", "",
967 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src1), (OpTy QPR:$src2))))]> {
968 let isCommutable = Commutable;
970 class N3VQX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
971 InstrItinClass itin, string OpcodeStr,
972 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
973 : N3VX<op24, op23, op21_20, op11_8, 1, op4,
974 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), N3RegFrm, itin,
975 OpcodeStr, "$dst, $src1, $src2", "",
976 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src1), (OpTy QPR:$src2))))]>{
977 let isCommutable = Commutable;
979 class N3VQSL<bits<2> op21_20, bits<4> op11_8,
980 InstrItinClass itin, string OpcodeStr, string Dt,
981 ValueType ResTy, ValueType OpTy, SDNode ShOp>
982 : N3V<1, 1, op21_20, op11_8, 1, 0,
983 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
984 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
985 [(set (ResTy QPR:$dst),
986 (ResTy (ShOp (ResTy QPR:$src1),
987 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
989 let isCommutable = 0;
991 class N3VQSL16<bits<2> op21_20, bits<4> op11_8, string OpcodeStr, string Dt,
992 ValueType ResTy, ValueType OpTy, SDNode ShOp>
993 : N3V<1, 1, op21_20, op11_8, 1, 0,
994 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
995 NVMulSLFrm, IIC_VMULi16Q, OpcodeStr, Dt,"$dst, $src1, $src2[$lane]","",
996 [(set (ResTy QPR:$dst),
997 (ResTy (ShOp (ResTy QPR:$src1),
998 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
1000 let isCommutable = 0;
1003 // Basic 3-register intrinsics, both double- and quad-register.
1004 class N3VDInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1005 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1006 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
1007 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1008 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), f, itin,
1009 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1010 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src1), (OpTy DPR:$src2))))]> {
1011 let isCommutable = Commutable;
1013 class N3VDIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1014 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
1015 : N3V<0, 1, op21_20, op11_8, 1, 0,
1016 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1017 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1018 [(set (Ty DPR:$dst),
1019 (Ty (IntOp (Ty DPR:$src1),
1020 (Ty (NEONvduplane (Ty DPR_VFP2:$src2),
1022 let isCommutable = 0;
1024 class N3VDIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1025 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
1026 : N3V<0, 1, op21_20, op11_8, 1, 0,
1027 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1028 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1029 [(set (Ty DPR:$dst),
1030 (Ty (IntOp (Ty DPR:$src1),
1031 (Ty (NEONvduplane (Ty DPR_8:$src2), imm:$lane)))))]> {
1032 let isCommutable = 0;
1035 class N3VQInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1036 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1037 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
1038 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1039 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), f, itin,
1040 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1041 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src1), (OpTy QPR:$src2))))]> {
1042 let isCommutable = Commutable;
1044 class N3VQIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1045 string OpcodeStr, string Dt,
1046 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1047 : N3V<1, 1, op21_20, op11_8, 1, 0,
1048 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1049 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1050 [(set (ResTy QPR:$dst),
1051 (ResTy (IntOp (ResTy QPR:$src1),
1052 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1054 let isCommutable = 0;
1056 class N3VQIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1057 string OpcodeStr, string Dt,
1058 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1059 : N3V<1, 1, op21_20, op11_8, 1, 0,
1060 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1061 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1062 [(set (ResTy QPR:$dst),
1063 (ResTy (IntOp (ResTy QPR:$src1),
1064 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
1066 let isCommutable = 0;
1069 // Multiply-Add/Sub operations: single-, double- and quad-register.
1070 class N3VSMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1071 InstrItinClass itin, string OpcodeStr, string Dt,
1072 ValueType Ty, SDNode MulOp, SDNode OpNode>
1073 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1074 (outs DPR_VFP2:$dst),
1075 (ins DPR_VFP2:$src1, DPR_VFP2:$src2, DPR_VFP2:$src3), N3RegFrm, itin,
1076 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst", []>;
1078 class N3VDMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1079 InstrItinClass itin, string OpcodeStr, string Dt,
1080 ValueType Ty, SDNode MulOp, SDNode OpNode>
1081 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1082 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3), N3RegFrm, itin,
1083 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1084 [(set DPR:$dst, (Ty (OpNode DPR:$src1,
1085 (Ty (MulOp DPR:$src2, DPR:$src3)))))]>;
1086 class N3VDMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1087 string OpcodeStr, string Dt,
1088 ValueType Ty, SDNode MulOp, SDNode ShOp>
1089 : N3V<0, 1, op21_20, op11_8, 1, 0,
1091 (ins DPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1093 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1094 [(set (Ty DPR:$dst),
1095 (Ty (ShOp (Ty DPR:$src1),
1096 (Ty (MulOp DPR:$src2,
1097 (Ty (NEONvduplane (Ty DPR_VFP2:$src3),
1099 class N3VDMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1100 string OpcodeStr, string Dt,
1101 ValueType Ty, SDNode MulOp, SDNode ShOp>
1102 : N3V<0, 1, op21_20, op11_8, 1, 0,
1104 (ins DPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane),
1106 OpcodeStr, Dt, "$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_8:$src3),
1113 class N3VQMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1114 InstrItinClass itin, string OpcodeStr, string Dt, ValueType Ty,
1115 SDNode MulOp, SDNode OpNode>
1116 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1117 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3), N3RegFrm, itin,
1118 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1119 [(set QPR:$dst, (Ty (OpNode QPR:$src1,
1120 (Ty (MulOp QPR:$src2, QPR:$src3)))))]>;
1121 class N3VQMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1122 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1123 SDNode MulOp, SDNode ShOp>
1124 : N3V<1, 1, op21_20, op11_8, 1, 0,
1126 (ins QPR:$src1, QPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1128 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1129 [(set (ResTy QPR:$dst),
1130 (ResTy (ShOp (ResTy QPR:$src1),
1131 (ResTy (MulOp QPR:$src2,
1132 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src3),
1134 class N3VQMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1135 string OpcodeStr, string Dt,
1136 ValueType ResTy, ValueType OpTy,
1137 SDNode MulOp, SDNode ShOp>
1138 : N3V<1, 1, op21_20, op11_8, 1, 0,
1140 (ins QPR:$src1, QPR:$src2, DPR_8:$src3, nohash_imm:$lane),
1142 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1143 [(set (ResTy QPR:$dst),
1144 (ResTy (ShOp (ResTy QPR:$src1),
1145 (ResTy (MulOp QPR:$src2,
1146 (ResTy (NEONvduplane (OpTy DPR_8:$src3),
1149 // Neon 3-argument intrinsics, both double- and quad-register.
1150 // The destination register is also used as the first source operand register.
1151 class N3VDInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1152 InstrItinClass itin, string OpcodeStr, string Dt,
1153 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1154 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1155 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3), N3RegFrm, itin,
1156 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1157 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src1),
1158 (OpTy DPR:$src2), (OpTy DPR:$src3))))]>;
1159 class N3VQInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1160 InstrItinClass itin, string OpcodeStr, string Dt,
1161 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1162 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1163 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3), N3RegFrm, itin,
1164 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1165 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src1),
1166 (OpTy QPR:$src2), (OpTy QPR:$src3))))]>;
1168 // Neon Long 3-argument intrinsic. The destination register is
1169 // a quad-register and is also used as the first source operand register.
1170 class N3VLInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1171 InstrItinClass itin, string OpcodeStr, string Dt,
1172 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
1173 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1174 (outs QPR:$dst), (ins QPR:$src1, DPR:$src2, DPR:$src3), N3RegFrm, itin,
1175 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1177 (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$src2), (TyD DPR:$src3))))]>;
1178 class N3VLInt3SL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1179 string OpcodeStr, string Dt,
1180 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1181 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1183 (ins QPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1185 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1186 [(set (ResTy QPR:$dst),
1187 (ResTy (IntOp (ResTy QPR:$src1),
1189 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src3),
1191 class N3VLInt3SL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1192 InstrItinClass itin, string OpcodeStr, string Dt,
1193 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1194 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1196 (ins QPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane),
1198 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1199 [(set (ResTy QPR:$dst),
1200 (ResTy (IntOp (ResTy QPR:$src1),
1202 (OpTy (NEONvduplane (OpTy DPR_8:$src3),
1205 // Narrowing 3-register intrinsics.
1206 class N3VNInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1207 string OpcodeStr, string Dt, ValueType TyD, ValueType TyQ,
1208 Intrinsic IntOp, bit Commutable>
1209 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1210 (outs DPR:$dst), (ins QPR:$src1, QPR:$src2), N3RegFrm, IIC_VBINi4D,
1211 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1212 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src1), (TyQ QPR:$src2))))]> {
1213 let isCommutable = Commutable;
1216 // Long 3-register intrinsics.
1217 class N3VLInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1218 InstrItinClass itin, string OpcodeStr, string Dt,
1219 ValueType TyQ, ValueType TyD, Intrinsic IntOp, bit Commutable>
1220 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1221 (outs QPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
1222 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1223 [(set QPR:$dst, (TyQ (IntOp (TyD DPR:$src1), (TyD DPR:$src2))))]> {
1224 let isCommutable = Commutable;
1226 class N3VLIntSL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1227 string OpcodeStr, string Dt,
1228 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1229 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1230 (outs QPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1231 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1232 [(set (ResTy QPR:$dst),
1233 (ResTy (IntOp (OpTy DPR:$src1),
1234 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1236 class N3VLIntSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1237 InstrItinClass itin, string OpcodeStr, string Dt,
1238 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1239 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1240 (outs QPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1241 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1242 [(set (ResTy QPR:$dst),
1243 (ResTy (IntOp (OpTy DPR:$src1),
1244 (OpTy (NEONvduplane (OpTy DPR_8:$src2),
1247 // Wide 3-register intrinsics.
1248 class N3VWInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1249 string OpcodeStr, string Dt, ValueType TyQ, ValueType TyD,
1250 Intrinsic IntOp, bit Commutable>
1251 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1252 (outs QPR:$dst), (ins QPR:$src1, DPR:$src2), N3RegFrm, IIC_VSUBiD,
1253 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1254 [(set QPR:$dst, (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$src2))))]> {
1255 let isCommutable = Commutable;
1258 // Pairwise long 2-register intrinsics, both double- and quad-register.
1259 class N2VDPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1260 bits<2> op17_16, bits<5> op11_7, bit op4,
1261 string OpcodeStr, string Dt,
1262 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1263 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
1264 (ins DPR:$src), IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
1265 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
1266 class N2VQPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1267 bits<2> op17_16, bits<5> op11_7, bit op4,
1268 string OpcodeStr, string Dt,
1269 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1270 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
1271 (ins QPR:$src), IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
1272 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
1274 // Pairwise long 2-register accumulate intrinsics,
1275 // both double- and quad-register.
1276 // The destination register is also used as the first source operand register.
1277 class N2VDPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1278 bits<2> op17_16, bits<5> op11_7, bit op4,
1279 string OpcodeStr, string Dt,
1280 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1281 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
1282 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), IIC_VPALiD,
1283 OpcodeStr, Dt, "$dst, $src2", "$src1 = $dst",
1284 [(set DPR:$dst, (ResTy (IntOp (ResTy DPR:$src1), (OpTy DPR:$src2))))]>;
1285 class N2VQPLInt2<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, 1, op4,
1290 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), IIC_VPALiQ,
1291 OpcodeStr, Dt, "$dst, $src2", "$src1 = $dst",
1292 [(set QPR:$dst, (ResTy (IntOp (ResTy QPR:$src1), (OpTy QPR:$src2))))]>;
1294 // Shift by immediate,
1295 // both double- and quad-register.
1296 class N2VDSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1297 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1298 ValueType Ty, SDNode OpNode>
1299 : N2VImm<op24, op23, op11_8, op7, 0, op4,
1300 (outs DPR:$dst), (ins DPR:$src, i32imm:$SIMM), f, itin,
1301 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1302 [(set DPR:$dst, (Ty (OpNode (Ty DPR:$src), (i32 imm:$SIMM))))]>;
1303 class N2VQSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1304 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1305 ValueType Ty, SDNode OpNode>
1306 : N2VImm<op24, op23, op11_8, op7, 1, op4,
1307 (outs QPR:$dst), (ins QPR:$src, i32imm:$SIMM), f, itin,
1308 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1309 [(set QPR:$dst, (Ty (OpNode (Ty QPR:$src), (i32 imm:$SIMM))))]>;
1311 // Long shift by immediate.
1312 class N2VLSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
1313 string OpcodeStr, string Dt,
1314 ValueType ResTy, ValueType OpTy, SDNode OpNode>
1315 : N2VImm<op24, op23, op11_8, op7, op6, op4,
1316 (outs QPR:$dst), (ins DPR:$src, i32imm:$SIMM), N2RegVShLFrm,
1317 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1318 [(set QPR:$dst, (ResTy (OpNode (OpTy DPR:$src),
1319 (i32 imm:$SIMM))))]>;
1321 // Narrow shift by immediate.
1322 class N2VNSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
1323 InstrItinClass itin, string OpcodeStr, string Dt,
1324 ValueType ResTy, ValueType OpTy, SDNode OpNode>
1325 : N2VImm<op24, op23, op11_8, op7, op6, op4,
1326 (outs DPR:$dst), (ins QPR:$src, i32imm:$SIMM), N2RegVShRFrm, itin,
1327 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1328 [(set DPR:$dst, (ResTy (OpNode (OpTy QPR:$src),
1329 (i32 imm:$SIMM))))]>;
1331 // Shift right by immediate and accumulate,
1332 // both double- and quad-register.
1333 class N2VDShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1334 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1335 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$dst),
1336 (ins DPR:$src1, DPR:$src2, i32imm:$SIMM), N2RegVShRFrm, IIC_VPALiD,
1337 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1338 [(set DPR:$dst, (Ty (add DPR:$src1,
1339 (Ty (ShOp DPR:$src2, (i32 imm:$SIMM))))))]>;
1340 class N2VQShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1341 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1342 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$dst),
1343 (ins QPR:$src1, QPR:$src2, i32imm:$SIMM), N2RegVShRFrm, IIC_VPALiD,
1344 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1345 [(set QPR:$dst, (Ty (add QPR:$src1,
1346 (Ty (ShOp QPR:$src2, (i32 imm:$SIMM))))))]>;
1348 // Shift by immediate and insert,
1349 // both double- and quad-register.
1350 class N2VDShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1351 Format f, string OpcodeStr, string Dt, ValueType Ty,SDNode ShOp>
1352 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$dst),
1353 (ins DPR:$src1, DPR:$src2, i32imm:$SIMM), f, IIC_VSHLiD,
1354 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1355 [(set DPR:$dst, (Ty (ShOp DPR:$src1, DPR:$src2, (i32 imm:$SIMM))))]>;
1356 class N2VQShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1357 Format f, string OpcodeStr, string Dt, ValueType Ty,SDNode ShOp>
1358 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$dst),
1359 (ins QPR:$src1, QPR:$src2, i32imm:$SIMM), f, IIC_VSHLiQ,
1360 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1361 [(set QPR:$dst, (Ty (ShOp QPR:$src1, QPR:$src2, (i32 imm:$SIMM))))]>;
1363 // Convert, with fractional bits immediate,
1364 // both double- and quad-register.
1365 class N2VCvtD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1366 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1368 : N2VImm<op24, op23, op11_8, op7, 0, op4,
1369 (outs DPR:$dst), (ins DPR:$src, i32imm:$SIMM), NVCVTFrm,
1370 IIC_VUNAD, OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1371 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src), (i32 imm:$SIMM))))]>;
1372 class N2VCvtQ<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1373 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1375 : N2VImm<op24, op23, op11_8, op7, 1, op4,
1376 (outs QPR:$dst), (ins QPR:$src, i32imm:$SIMM), NVCVTFrm,
1377 IIC_VUNAQ, OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1378 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src), (i32 imm:$SIMM))))]>;
1380 //===----------------------------------------------------------------------===//
1382 //===----------------------------------------------------------------------===//
1384 // Abbreviations used in multiclass suffixes:
1385 // Q = quarter int (8 bit) elements
1386 // H = half int (16 bit) elements
1387 // S = single int (32 bit) elements
1388 // D = double int (64 bit) elements
1390 // Neon 2-register vector operations -- for disassembly only.
1392 // First with only element sizes of 8, 16 and 32 bits:
1393 multiclass N2V_QHS_cmp<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1394 bits<5> op11_7, bit op4, string opc, string Dt,
1396 // 64-bit vector types.
1397 def v8i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 0, op4,
1398 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1399 opc, !strconcat(Dt, "8"), asm, "", []>;
1400 def v4i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 0, op4,
1401 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1402 opc, !strconcat(Dt, "16"), asm, "", []>;
1403 def v2i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
1404 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1405 opc, !strconcat(Dt, "32"), asm, "", []>;
1406 def v2f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
1407 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1408 opc, "f32", asm, "", []> {
1409 let Inst{10} = 1; // overwrite F = 1
1412 // 128-bit vector types.
1413 def v16i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 1, op4,
1414 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1415 opc, !strconcat(Dt, "8"), asm, "", []>;
1416 def v8i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 1, op4,
1417 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1418 opc, !strconcat(Dt, "16"), asm, "", []>;
1419 def v4i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
1420 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1421 opc, !strconcat(Dt, "32"), asm, "", []>;
1422 def v4f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
1423 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1424 opc, "f32", asm, "", []> {
1425 let Inst{10} = 1; // overwrite F = 1
1429 // Neon 3-register vector operations.
1431 // First with only element sizes of 8, 16 and 32 bits:
1432 multiclass N3V_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1433 InstrItinClass itinD16, InstrItinClass itinD32,
1434 InstrItinClass itinQ16, InstrItinClass itinQ32,
1435 string OpcodeStr, string Dt,
1436 SDNode OpNode, bit Commutable = 0> {
1437 // 64-bit vector types.
1438 def v8i8 : N3VD<op24, op23, 0b00, op11_8, op4, itinD16,
1439 OpcodeStr, !strconcat(Dt, "8"),
1440 v8i8, v8i8, OpNode, Commutable>;
1441 def v4i16 : N3VD<op24, op23, 0b01, op11_8, op4, itinD16,
1442 OpcodeStr, !strconcat(Dt, "16"),
1443 v4i16, v4i16, OpNode, Commutable>;
1444 def v2i32 : N3VD<op24, op23, 0b10, op11_8, op4, itinD32,
1445 OpcodeStr, !strconcat(Dt, "32"),
1446 v2i32, v2i32, OpNode, Commutable>;
1448 // 128-bit vector types.
1449 def v16i8 : N3VQ<op24, op23, 0b00, op11_8, op4, itinQ16,
1450 OpcodeStr, !strconcat(Dt, "8"),
1451 v16i8, v16i8, OpNode, Commutable>;
1452 def v8i16 : N3VQ<op24, op23, 0b01, op11_8, op4, itinQ16,
1453 OpcodeStr, !strconcat(Dt, "16"),
1454 v8i16, v8i16, OpNode, Commutable>;
1455 def v4i32 : N3VQ<op24, op23, 0b10, op11_8, op4, itinQ32,
1456 OpcodeStr, !strconcat(Dt, "32"),
1457 v4i32, v4i32, OpNode, Commutable>;
1460 multiclass N3VSL_HS<bits<4> op11_8, string OpcodeStr, string Dt, SDNode ShOp> {
1461 def v4i16 : N3VDSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
1463 def v2i32 : N3VDSL<0b10, op11_8, IIC_VMULi32D, OpcodeStr, !strconcat(Dt,"32"),
1465 def v8i16 : N3VQSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
1466 v8i16, v4i16, ShOp>;
1467 def v4i32 : N3VQSL<0b10, op11_8, IIC_VMULi32Q, OpcodeStr, !strconcat(Dt,"32"),
1468 v4i32, v2i32, ShOp>;
1471 // ....then also with element size 64 bits:
1472 multiclass N3V_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1473 InstrItinClass itinD, InstrItinClass itinQ,
1474 string OpcodeStr, string Dt,
1475 SDNode OpNode, bit Commutable = 0>
1476 : N3V_QHS<op24, op23, op11_8, op4, itinD, itinD, itinQ, itinQ,
1477 OpcodeStr, Dt, OpNode, Commutable> {
1478 def v1i64 : N3VD<op24, op23, 0b11, op11_8, op4, itinD,
1479 OpcodeStr, !strconcat(Dt, "64"),
1480 v1i64, v1i64, OpNode, Commutable>;
1481 def v2i64 : N3VQ<op24, op23, 0b11, op11_8, op4, itinQ,
1482 OpcodeStr, !strconcat(Dt, "64"),
1483 v2i64, v2i64, OpNode, Commutable>;
1487 // Neon Narrowing 2-register vector intrinsics,
1488 // source operand element sizes of 16, 32 and 64 bits:
1489 multiclass N2VNInt_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1490 bits<5> op11_7, bit op6, bit op4,
1491 InstrItinClass itin, string OpcodeStr, string Dt,
1493 def v8i8 : N2VNInt<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
1494 itin, OpcodeStr, !strconcat(Dt, "16"),
1495 v8i8, v8i16, IntOp>;
1496 def v4i16 : N2VNInt<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
1497 itin, OpcodeStr, !strconcat(Dt, "32"),
1498 v4i16, v4i32, IntOp>;
1499 def v2i32 : N2VNInt<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
1500 itin, OpcodeStr, !strconcat(Dt, "64"),
1501 v2i32, v2i64, IntOp>;
1505 // Neon Lengthening 2-register vector intrinsic (currently specific to VMOVL).
1506 // source operand element sizes of 16, 32 and 64 bits:
1507 multiclass N2VLInt_QHS<bits<2> op24_23, bits<5> op11_7, bit op6, bit op4,
1508 string OpcodeStr, string Dt, Intrinsic IntOp> {
1509 def v8i16 : N2VLInt<op24_23, 0b00, 0b10, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
1510 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
1511 def v4i32 : N2VLInt<op24_23, 0b01, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
1512 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
1513 def v2i64 : N2VLInt<op24_23, 0b10, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
1514 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1518 // Neon 3-register vector intrinsics.
1520 // First with only element sizes of 16 and 32 bits:
1521 multiclass N3VInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
1522 InstrItinClass itinD16, InstrItinClass itinD32,
1523 InstrItinClass itinQ16, InstrItinClass itinQ32,
1524 string OpcodeStr, string Dt,
1525 Intrinsic IntOp, bit Commutable = 0> {
1526 // 64-bit vector types.
1527 def v4i16 : N3VDInt<op24, op23, 0b01, op11_8, op4, f, itinD16,
1528 OpcodeStr, !strconcat(Dt, "16"),
1529 v4i16, v4i16, IntOp, Commutable>;
1530 def v2i32 : N3VDInt<op24, op23, 0b10, op11_8, op4, f, itinD32,
1531 OpcodeStr, !strconcat(Dt, "32"),
1532 v2i32, v2i32, IntOp, Commutable>;
1534 // 128-bit vector types.
1535 def v8i16 : N3VQInt<op24, op23, 0b01, op11_8, op4, f, itinQ16,
1536 OpcodeStr, !strconcat(Dt, "16"),
1537 v8i16, v8i16, IntOp, Commutable>;
1538 def v4i32 : N3VQInt<op24, op23, 0b10, op11_8, op4, f, itinQ32,
1539 OpcodeStr, !strconcat(Dt, "32"),
1540 v4i32, v4i32, IntOp, Commutable>;
1543 multiclass N3VIntSL_HS<bits<4> op11_8,
1544 InstrItinClass itinD16, InstrItinClass itinD32,
1545 InstrItinClass itinQ16, InstrItinClass itinQ32,
1546 string OpcodeStr, string Dt, Intrinsic IntOp> {
1547 def v4i16 : N3VDIntSL16<0b01, op11_8, itinD16,
1548 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp>;
1549 def v2i32 : N3VDIntSL<0b10, op11_8, itinD32,
1550 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp>;
1551 def v8i16 : N3VQIntSL16<0b01, op11_8, itinQ16,
1552 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16, IntOp>;
1553 def v4i32 : N3VQIntSL<0b10, op11_8, itinQ32,
1554 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32, IntOp>;
1557 // ....then also with element size of 8 bits:
1558 multiclass N3VInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
1559 InstrItinClass itinD16, InstrItinClass itinD32,
1560 InstrItinClass itinQ16, InstrItinClass itinQ32,
1561 string OpcodeStr, string Dt,
1562 Intrinsic IntOp, bit Commutable = 0>
1563 : N3VInt_HS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
1564 OpcodeStr, Dt, IntOp, Commutable> {
1565 def v8i8 : N3VDInt<op24, op23, 0b00, op11_8, op4, f, itinD16,
1566 OpcodeStr, !strconcat(Dt, "8"),
1567 v8i8, v8i8, IntOp, Commutable>;
1568 def v16i8 : N3VQInt<op24, op23, 0b00, op11_8, op4, f, itinQ16,
1569 OpcodeStr, !strconcat(Dt, "8"),
1570 v16i8, v16i8, IntOp, Commutable>;
1573 // ....then also with element size of 64 bits:
1574 multiclass N3VInt_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
1575 InstrItinClass itinD16, InstrItinClass itinD32,
1576 InstrItinClass itinQ16, InstrItinClass itinQ32,
1577 string OpcodeStr, string Dt,
1578 Intrinsic IntOp, bit Commutable = 0>
1579 : N3VInt_QHS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
1580 OpcodeStr, Dt, IntOp, Commutable> {
1581 def v1i64 : N3VDInt<op24, op23, 0b11, op11_8, op4, f, itinD32,
1582 OpcodeStr, !strconcat(Dt, "64"),
1583 v1i64, v1i64, IntOp, Commutable>;
1584 def v2i64 : N3VQInt<op24, op23, 0b11, op11_8, op4, f, itinQ32,
1585 OpcodeStr, !strconcat(Dt, "64"),
1586 v2i64, v2i64, IntOp, Commutable>;
1589 // Neon Narrowing 3-register vector intrinsics,
1590 // source operand element sizes of 16, 32 and 64 bits:
1591 multiclass N3VNInt_HSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1592 string OpcodeStr, string Dt,
1593 Intrinsic IntOp, bit Commutable = 0> {
1594 def v8i8 : N3VNInt<op24, op23, 0b00, op11_8, op4,
1595 OpcodeStr, !strconcat(Dt, "16"),
1596 v8i8, v8i16, IntOp, Commutable>;
1597 def v4i16 : N3VNInt<op24, op23, 0b01, op11_8, op4,
1598 OpcodeStr, !strconcat(Dt, "32"),
1599 v4i16, v4i32, IntOp, Commutable>;
1600 def v2i32 : N3VNInt<op24, op23, 0b10, op11_8, op4,
1601 OpcodeStr, !strconcat(Dt, "64"),
1602 v2i32, v2i64, IntOp, Commutable>;
1606 // Neon Long 3-register vector intrinsics.
1608 // First with only element sizes of 16 and 32 bits:
1609 multiclass N3VLInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
1610 InstrItinClass itin16, InstrItinClass itin32,
1611 string OpcodeStr, string Dt,
1612 Intrinsic IntOp, bit Commutable = 0> {
1613 def v4i32 : N3VLInt<op24, op23, 0b01, op11_8, op4, itin16,
1614 OpcodeStr, !strconcat(Dt, "16"),
1615 v4i32, v4i16, IntOp, Commutable>;
1616 def v2i64 : N3VLInt<op24, op23, 0b10, op11_8, op4, itin32,
1617 OpcodeStr, !strconcat(Dt, "32"),
1618 v2i64, v2i32, IntOp, Commutable>;
1621 multiclass N3VLIntSL_HS<bit op24, bits<4> op11_8,
1622 InstrItinClass itin, string OpcodeStr, string Dt,
1624 def v4i16 : N3VLIntSL16<op24, 0b01, op11_8, itin,
1625 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
1626 def v2i32 : N3VLIntSL<op24, 0b10, op11_8, itin,
1627 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1630 // ....then also with element size of 8 bits:
1631 multiclass N3VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1632 InstrItinClass itin16, InstrItinClass itin32,
1633 string OpcodeStr, string Dt,
1634 Intrinsic IntOp, bit Commutable = 0>
1635 : N3VLInt_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt,
1636 IntOp, Commutable> {
1637 def v8i16 : N3VLInt<op24, op23, 0b00, op11_8, op4, itin16,
1638 OpcodeStr, !strconcat(Dt, "8"),
1639 v8i16, v8i8, IntOp, Commutable>;
1643 // Neon Wide 3-register vector intrinsics,
1644 // source operand element sizes of 8, 16 and 32 bits:
1645 multiclass N3VWInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1646 string OpcodeStr, string Dt,
1647 Intrinsic IntOp, bit Commutable = 0> {
1648 def v8i16 : N3VWInt<op24, op23, 0b00, op11_8, op4,
1649 OpcodeStr, !strconcat(Dt, "8"),
1650 v8i16, v8i8, IntOp, Commutable>;
1651 def v4i32 : N3VWInt<op24, op23, 0b01, op11_8, op4,
1652 OpcodeStr, !strconcat(Dt, "16"),
1653 v4i32, v4i16, IntOp, Commutable>;
1654 def v2i64 : N3VWInt<op24, op23, 0b10, op11_8, op4,
1655 OpcodeStr, !strconcat(Dt, "32"),
1656 v2i64, v2i32, IntOp, Commutable>;
1660 // Neon Multiply-Op vector operations,
1661 // element sizes of 8, 16 and 32 bits:
1662 multiclass N3VMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1663 InstrItinClass itinD16, InstrItinClass itinD32,
1664 InstrItinClass itinQ16, InstrItinClass itinQ32,
1665 string OpcodeStr, string Dt, SDNode OpNode> {
1666 // 64-bit vector types.
1667 def v8i8 : N3VDMulOp<op24, op23, 0b00, op11_8, op4, itinD16,
1668 OpcodeStr, !strconcat(Dt, "8"), v8i8, mul, OpNode>;
1669 def v4i16 : N3VDMulOp<op24, op23, 0b01, op11_8, op4, itinD16,
1670 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, OpNode>;
1671 def v2i32 : N3VDMulOp<op24, op23, 0b10, op11_8, op4, itinD32,
1672 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, OpNode>;
1674 // 128-bit vector types.
1675 def v16i8 : N3VQMulOp<op24, op23, 0b00, op11_8, op4, itinQ16,
1676 OpcodeStr, !strconcat(Dt, "8"), v16i8, mul, OpNode>;
1677 def v8i16 : N3VQMulOp<op24, op23, 0b01, op11_8, op4, itinQ16,
1678 OpcodeStr, !strconcat(Dt, "16"), v8i16, mul, OpNode>;
1679 def v4i32 : N3VQMulOp<op24, op23, 0b10, op11_8, op4, itinQ32,
1680 OpcodeStr, !strconcat(Dt, "32"), v4i32, mul, OpNode>;
1683 multiclass N3VMulOpSL_HS<bits<4> op11_8,
1684 InstrItinClass itinD16, InstrItinClass itinD32,
1685 InstrItinClass itinQ16, InstrItinClass itinQ32,
1686 string OpcodeStr, string Dt, SDNode ShOp> {
1687 def v4i16 : N3VDMulOpSL16<0b01, op11_8, itinD16,
1688 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, ShOp>;
1689 def v2i32 : N3VDMulOpSL<0b10, op11_8, itinD32,
1690 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, ShOp>;
1691 def v8i16 : N3VQMulOpSL16<0b01, op11_8, itinQ16,
1692 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16,
1694 def v4i32 : N3VQMulOpSL<0b10, op11_8, itinQ32,
1695 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32,
1699 // Neon 3-argument intrinsics,
1700 // element sizes of 8, 16 and 32 bits:
1701 multiclass N3VInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1702 InstrItinClass itinD, InstrItinClass itinQ,
1703 string OpcodeStr, string Dt, Intrinsic IntOp> {
1704 // 64-bit vector types.
1705 def v8i8 : N3VDInt3<op24, op23, 0b00, op11_8, op4, itinD,
1706 OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
1707 def v4i16 : N3VDInt3<op24, op23, 0b01, op11_8, op4, itinD,
1708 OpcodeStr, !strconcat(Dt, "16"), v4i16, v4i16, IntOp>;
1709 def v2i32 : N3VDInt3<op24, op23, 0b10, op11_8, op4, itinD,
1710 OpcodeStr, !strconcat(Dt, "32"), v2i32, v2i32, IntOp>;
1712 // 128-bit vector types.
1713 def v16i8 : N3VQInt3<op24, op23, 0b00, op11_8, op4, itinQ,
1714 OpcodeStr, !strconcat(Dt, "8"), v16i8, v16i8, IntOp>;
1715 def v8i16 : N3VQInt3<op24, op23, 0b01, op11_8, op4, itinQ,
1716 OpcodeStr, !strconcat(Dt, "16"), v8i16, v8i16, IntOp>;
1717 def v4i32 : N3VQInt3<op24, op23, 0b10, op11_8, op4, itinQ,
1718 OpcodeStr, !strconcat(Dt, "32"), v4i32, v4i32, IntOp>;
1722 // Neon Long 3-argument intrinsics.
1724 // First with only element sizes of 16 and 32 bits:
1725 multiclass N3VLInt3_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
1726 InstrItinClass itin16, InstrItinClass itin32,
1727 string OpcodeStr, string Dt, Intrinsic IntOp> {
1728 def v4i32 : N3VLInt3<op24, op23, 0b01, op11_8, op4, itin16,
1729 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
1730 def v2i64 : N3VLInt3<op24, op23, 0b10, op11_8, op4, itin32,
1731 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1734 multiclass N3VLInt3SL_HS<bit op24, bits<4> op11_8,
1735 string OpcodeStr, string Dt, Intrinsic IntOp> {
1736 def v4i16 : N3VLInt3SL16<op24, 0b01, op11_8, IIC_VMACi16D,
1737 OpcodeStr, !strconcat(Dt,"16"), v4i32, v4i16, IntOp>;
1738 def v2i32 : N3VLInt3SL<op24, 0b10, op11_8, IIC_VMACi32D,
1739 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1742 // ....then also with element size of 8 bits:
1743 multiclass N3VLInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1744 InstrItinClass itin16, InstrItinClass itin32,
1745 string OpcodeStr, string Dt, Intrinsic IntOp>
1746 : N3VLInt3_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt, IntOp> {
1747 def v8i16 : N3VLInt3<op24, op23, 0b00, op11_8, op4, itin16,
1748 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
1752 // Neon 2-register vector intrinsics,
1753 // element sizes of 8, 16 and 32 bits:
1754 multiclass N2VInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1755 bits<5> op11_7, bit op4,
1756 InstrItinClass itinD, InstrItinClass itinQ,
1757 string OpcodeStr, string Dt, Intrinsic IntOp> {
1758 // 64-bit vector types.
1759 def v8i8 : N2VDInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1760 itinD, OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
1761 def v4i16 : N2VDInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1762 itinD, OpcodeStr, !strconcat(Dt, "16"),v4i16,v4i16,IntOp>;
1763 def v2i32 : N2VDInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1764 itinD, OpcodeStr, !strconcat(Dt, "32"),v2i32,v2i32,IntOp>;
1766 // 128-bit vector types.
1767 def v16i8 : N2VQInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1768 itinQ, OpcodeStr, !strconcat(Dt, "8"), v16i8,v16i8,IntOp>;
1769 def v8i16 : N2VQInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1770 itinQ, OpcodeStr, !strconcat(Dt, "16"),v8i16,v8i16,IntOp>;
1771 def v4i32 : N2VQInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1772 itinQ, OpcodeStr, !strconcat(Dt, "32"),v4i32,v4i32,IntOp>;
1776 // Neon Pairwise long 2-register intrinsics,
1777 // element sizes of 8, 16 and 32 bits:
1778 multiclass N2VPLInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1779 bits<5> op11_7, bit op4,
1780 string OpcodeStr, string Dt, Intrinsic IntOp> {
1781 // 64-bit vector types.
1782 def v8i8 : N2VDPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1783 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
1784 def v4i16 : N2VDPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1785 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
1786 def v2i32 : N2VDPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1787 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
1789 // 128-bit vector types.
1790 def v16i8 : N2VQPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1791 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
1792 def v8i16 : N2VQPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1793 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
1794 def v4i32 : N2VQPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1795 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
1799 // Neon Pairwise long 2-register accumulate intrinsics,
1800 // element sizes of 8, 16 and 32 bits:
1801 multiclass N2VPLInt2_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1802 bits<5> op11_7, bit op4,
1803 string OpcodeStr, string Dt, Intrinsic IntOp> {
1804 // 64-bit vector types.
1805 def v8i8 : N2VDPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1806 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
1807 def v4i16 : N2VDPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1808 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
1809 def v2i32 : N2VDPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1810 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
1812 // 128-bit vector types.
1813 def v16i8 : N2VQPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1814 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
1815 def v8i16 : N2VQPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1816 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
1817 def v4i32 : N2VQPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1818 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
1822 // Neon 2-register vector shift by immediate,
1823 // with f of either N2RegVShLFrm or N2RegVShRFrm
1824 // element sizes of 8, 16, 32 and 64 bits:
1825 multiclass N2VSh_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1826 InstrItinClass itin, string OpcodeStr, string Dt,
1827 SDNode OpNode, Format f> {
1828 // 64-bit vector types.
1829 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
1830 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
1831 let Inst{21-19} = 0b001; // imm6 = 001xxx
1833 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
1834 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
1835 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1837 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
1838 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
1839 let Inst{21} = 0b1; // imm6 = 1xxxxx
1841 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, f, itin,
1842 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
1845 // 128-bit vector types.
1846 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
1847 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
1848 let Inst{21-19} = 0b001; // imm6 = 001xxx
1850 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
1851 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
1852 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1854 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
1855 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
1856 let Inst{21} = 0b1; // imm6 = 1xxxxx
1858 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, f, itin,
1859 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
1863 // Neon Shift-Accumulate vector operations,
1864 // element sizes of 8, 16, 32 and 64 bits:
1865 multiclass N2VShAdd_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1866 string OpcodeStr, string Dt, SDNode ShOp> {
1867 // 64-bit vector types.
1868 def v8i8 : N2VDShAdd<op24, op23, op11_8, 0, op4,
1869 OpcodeStr, !strconcat(Dt, "8"), v8i8, ShOp> {
1870 let Inst{21-19} = 0b001; // imm6 = 001xxx
1872 def v4i16 : N2VDShAdd<op24, op23, op11_8, 0, op4,
1873 OpcodeStr, !strconcat(Dt, "16"), v4i16, ShOp> {
1874 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1876 def v2i32 : N2VDShAdd<op24, op23, op11_8, 0, op4,
1877 OpcodeStr, !strconcat(Dt, "32"), v2i32, ShOp> {
1878 let Inst{21} = 0b1; // imm6 = 1xxxxx
1880 def v1i64 : N2VDShAdd<op24, op23, op11_8, 1, op4,
1881 OpcodeStr, !strconcat(Dt, "64"), v1i64, ShOp>;
1884 // 128-bit vector types.
1885 def v16i8 : N2VQShAdd<op24, op23, op11_8, 0, op4,
1886 OpcodeStr, !strconcat(Dt, "8"), v16i8, ShOp> {
1887 let Inst{21-19} = 0b001; // imm6 = 001xxx
1889 def v8i16 : N2VQShAdd<op24, op23, op11_8, 0, op4,
1890 OpcodeStr, !strconcat(Dt, "16"), v8i16, ShOp> {
1891 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1893 def v4i32 : N2VQShAdd<op24, op23, op11_8, 0, op4,
1894 OpcodeStr, !strconcat(Dt, "32"), v4i32, ShOp> {
1895 let Inst{21} = 0b1; // imm6 = 1xxxxx
1897 def v2i64 : N2VQShAdd<op24, op23, op11_8, 1, op4,
1898 OpcodeStr, !strconcat(Dt, "64"), v2i64, ShOp>;
1903 // Neon Shift-Insert vector operations,
1904 // with f of either N2RegVShLFrm or N2RegVShRFrm
1905 // element sizes of 8, 16, 32 and 64 bits:
1906 multiclass N2VShIns_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1907 string OpcodeStr, SDNode ShOp,
1909 // 64-bit vector types.
1910 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4,
1911 f, OpcodeStr, "8", v8i8, ShOp> {
1912 let Inst{21-19} = 0b001; // imm6 = 001xxx
1914 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4,
1915 f, OpcodeStr, "16", v4i16, ShOp> {
1916 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1918 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4,
1919 f, OpcodeStr, "32", v2i32, ShOp> {
1920 let Inst{21} = 0b1; // imm6 = 1xxxxx
1922 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4,
1923 f, OpcodeStr, "64", v1i64, ShOp>;
1926 // 128-bit vector types.
1927 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4,
1928 f, OpcodeStr, "8", v16i8, ShOp> {
1929 let Inst{21-19} = 0b001; // imm6 = 001xxx
1931 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4,
1932 f, OpcodeStr, "16", v8i16, ShOp> {
1933 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1935 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4,
1936 f, OpcodeStr, "32", v4i32, ShOp> {
1937 let Inst{21} = 0b1; // imm6 = 1xxxxx
1939 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4,
1940 f, OpcodeStr, "64", v2i64, ShOp>;
1944 // Neon Shift Long operations,
1945 // element sizes of 8, 16, 32 bits:
1946 multiclass N2VLSh_QHS<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
1947 bit op4, string OpcodeStr, string Dt, SDNode OpNode> {
1948 def v8i16 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
1949 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode> {
1950 let Inst{21-19} = 0b001; // imm6 = 001xxx
1952 def v4i32 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
1953 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode> {
1954 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1956 def v2i64 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
1957 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode> {
1958 let Inst{21} = 0b1; // imm6 = 1xxxxx
1962 // Neon Shift Narrow operations,
1963 // element sizes of 16, 32, 64 bits:
1964 multiclass N2VNSh_HSD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
1965 bit op4, InstrItinClass itin, string OpcodeStr, string Dt,
1967 def v8i8 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
1968 OpcodeStr, !strconcat(Dt, "16"), v8i8, v8i16, OpNode> {
1969 let Inst{21-19} = 0b001; // imm6 = 001xxx
1971 def v4i16 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
1972 OpcodeStr, !strconcat(Dt, "32"), v4i16, v4i32, OpNode> {
1973 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1975 def v2i32 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
1976 OpcodeStr, !strconcat(Dt, "64"), v2i32, v2i64, OpNode> {
1977 let Inst{21} = 0b1; // imm6 = 1xxxxx
1981 //===----------------------------------------------------------------------===//
1982 // Instruction Definitions.
1983 //===----------------------------------------------------------------------===//
1985 // Vector Add Operations.
1987 // VADD : Vector Add (integer and floating-point)
1988 defm VADD : N3V_QHSD<0, 0, 0b1000, 0, IIC_VBINiD, IIC_VBINiQ, "vadd", "i",
1990 def VADDfd : N3VD<0, 0, 0b00, 0b1101, 0, IIC_VBIND, "vadd", "f32",
1991 v2f32, v2f32, fadd, 1>;
1992 def VADDfq : N3VQ<0, 0, 0b00, 0b1101, 0, IIC_VBINQ, "vadd", "f32",
1993 v4f32, v4f32, fadd, 1>;
1994 // VADDL : Vector Add Long (Q = D + D)
1995 defm VADDLs : N3VLInt_QHS<0,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
1996 "vaddl", "s", int_arm_neon_vaddls, 1>;
1997 defm VADDLu : N3VLInt_QHS<1,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
1998 "vaddl", "u", int_arm_neon_vaddlu, 1>;
1999 // VADDW : Vector Add Wide (Q = Q + D)
2000 defm VADDWs : N3VWInt_QHS<0,1,0b0001,0, "vaddw", "s", int_arm_neon_vaddws, 0>;
2001 defm VADDWu : N3VWInt_QHS<1,1,0b0001,0, "vaddw", "u", int_arm_neon_vaddwu, 0>;
2002 // VHADD : Vector Halving Add
2003 defm VHADDs : N3VInt_QHS<0, 0, 0b0000, 0, N3RegFrm,
2004 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2005 "vhadd", "s", int_arm_neon_vhadds, 1>;
2006 defm VHADDu : N3VInt_QHS<1, 0, 0b0000, 0, N3RegFrm,
2007 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2008 "vhadd", "u", int_arm_neon_vhaddu, 1>;
2009 // VRHADD : Vector Rounding Halving Add
2010 defm VRHADDs : N3VInt_QHS<0, 0, 0b0001, 0, N3RegFrm,
2011 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2012 "vrhadd", "s", int_arm_neon_vrhadds, 1>;
2013 defm VRHADDu : N3VInt_QHS<1, 0, 0b0001, 0, N3RegFrm,
2014 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2015 "vrhadd", "u", int_arm_neon_vrhaddu, 1>;
2016 // VQADD : Vector Saturating Add
2017 defm VQADDs : N3VInt_QHSD<0, 0, 0b0000, 1, N3RegFrm,
2018 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2019 "vqadd", "s", int_arm_neon_vqadds, 1>;
2020 defm VQADDu : N3VInt_QHSD<1, 0, 0b0000, 1, N3RegFrm,
2021 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2022 "vqadd", "u", int_arm_neon_vqaddu, 1>;
2023 // VADDHN : Vector Add and Narrow Returning High Half (D = Q + Q)
2024 defm VADDHN : N3VNInt_HSD<0,1,0b0100,0, "vaddhn", "i",
2025 int_arm_neon_vaddhn, 1>;
2026 // VRADDHN : Vector Rounding Add and Narrow Returning High Half (D = Q + Q)
2027 defm VRADDHN : N3VNInt_HSD<1,1,0b0100,0, "vraddhn", "i",
2028 int_arm_neon_vraddhn, 1>;
2030 // Vector Multiply Operations.
2032 // VMUL : Vector Multiply (integer, polynomial and floating-point)
2033 defm VMUL : N3V_QHS<0, 0, 0b1001, 1, IIC_VMULi16D, IIC_VMULi32D,
2034 IIC_VMULi16Q, IIC_VMULi32Q, "vmul", "i", mul, 1>;
2035 def VMULpd : N3VDInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16D, "vmul",
2036 "p8", v8i8, v8i8, int_arm_neon_vmulp, 1>;
2037 def VMULpq : N3VQInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16Q, "vmul",
2038 "p8", v16i8, v16i8, int_arm_neon_vmulp, 1>;
2039 def VMULfd : N3VD<1, 0, 0b00, 0b1101, 1, IIC_VBIND, "vmul", "f32",
2040 v2f32, v2f32, fmul, 1>;
2041 def VMULfq : N3VQ<1, 0, 0b00, 0b1101, 1, IIC_VBINQ, "vmul", "f32",
2042 v4f32, v4f32, fmul, 1>;
2043 defm VMULsl : N3VSL_HS<0b1000, "vmul", "i", mul>;
2044 def VMULslfd : N3VDSL<0b10, 0b1001, IIC_VBIND, "vmul", "f32", v2f32, fmul>;
2045 def VMULslfq : N3VQSL<0b10, 0b1001, IIC_VBINQ, "vmul", "f32", v4f32,
2048 def : Pat<(v8i16 (mul (v8i16 QPR:$src1),
2049 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
2050 (v8i16 (VMULslv8i16 (v8i16 QPR:$src1),
2051 (v4i16 (EXTRACT_SUBREG QPR:$src2,
2052 (DSubReg_i16_reg imm:$lane))),
2053 (SubReg_i16_lane imm:$lane)))>;
2054 def : Pat<(v4i32 (mul (v4i32 QPR:$src1),
2055 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
2056 (v4i32 (VMULslv4i32 (v4i32 QPR:$src1),
2057 (v2i32 (EXTRACT_SUBREG QPR:$src2,
2058 (DSubReg_i32_reg imm:$lane))),
2059 (SubReg_i32_lane imm:$lane)))>;
2060 def : Pat<(v4f32 (fmul (v4f32 QPR:$src1),
2061 (v4f32 (NEONvduplane (v4f32 QPR:$src2), imm:$lane)))),
2062 (v4f32 (VMULslfq (v4f32 QPR:$src1),
2063 (v2f32 (EXTRACT_SUBREG QPR:$src2,
2064 (DSubReg_i32_reg imm:$lane))),
2065 (SubReg_i32_lane imm:$lane)))>;
2067 // VQDMULH : Vector Saturating Doubling Multiply Returning High Half
2068 defm VQDMULH : N3VInt_HS<0, 0, 0b1011, 0, N3RegFrm, IIC_VMULi16D, IIC_VMULi32D,
2069 IIC_VMULi16Q, IIC_VMULi32Q,
2070 "vqdmulh", "s", int_arm_neon_vqdmulh, 1>;
2071 defm VQDMULHsl: N3VIntSL_HS<0b1100, IIC_VMULi16D, IIC_VMULi32D,
2072 IIC_VMULi16Q, IIC_VMULi32Q,
2073 "vqdmulh", "s", int_arm_neon_vqdmulh>;
2074 def : Pat<(v8i16 (int_arm_neon_vqdmulh (v8i16 QPR:$src1),
2075 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
2077 (v8i16 (VQDMULHslv8i16 (v8i16 QPR:$src1),
2078 (v4i16 (EXTRACT_SUBREG QPR:$src2,
2079 (DSubReg_i16_reg imm:$lane))),
2080 (SubReg_i16_lane imm:$lane)))>;
2081 def : Pat<(v4i32 (int_arm_neon_vqdmulh (v4i32 QPR:$src1),
2082 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
2084 (v4i32 (VQDMULHslv4i32 (v4i32 QPR:$src1),
2085 (v2i32 (EXTRACT_SUBREG QPR:$src2,
2086 (DSubReg_i32_reg imm:$lane))),
2087 (SubReg_i32_lane imm:$lane)))>;
2089 // VQRDMULH : Vector Rounding Saturating Doubling Multiply Returning High Half
2090 defm VQRDMULH : N3VInt_HS<1, 0, 0b1011, 0, N3RegFrm,
2091 IIC_VMULi16D,IIC_VMULi32D,IIC_VMULi16Q,IIC_VMULi32Q,
2092 "vqrdmulh", "s", int_arm_neon_vqrdmulh, 1>;
2093 defm VQRDMULHsl : N3VIntSL_HS<0b1101, IIC_VMULi16D, IIC_VMULi32D,
2094 IIC_VMULi16Q, IIC_VMULi32Q,
2095 "vqrdmulh", "s", int_arm_neon_vqrdmulh>;
2096 def : Pat<(v8i16 (int_arm_neon_vqrdmulh (v8i16 QPR:$src1),
2097 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
2099 (v8i16 (VQRDMULHslv8i16 (v8i16 QPR:$src1),
2100 (v4i16 (EXTRACT_SUBREG QPR:$src2,
2101 (DSubReg_i16_reg imm:$lane))),
2102 (SubReg_i16_lane imm:$lane)))>;
2103 def : Pat<(v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$src1),
2104 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
2106 (v4i32 (VQRDMULHslv4i32 (v4i32 QPR:$src1),
2107 (v2i32 (EXTRACT_SUBREG QPR:$src2,
2108 (DSubReg_i32_reg imm:$lane))),
2109 (SubReg_i32_lane imm:$lane)))>;
2111 // VMULL : Vector Multiply Long (integer and polynomial) (Q = D * D)
2112 defm VMULLs : N3VLInt_QHS<0,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
2113 "vmull", "s", int_arm_neon_vmulls, 1>;
2114 defm VMULLu : N3VLInt_QHS<1,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
2115 "vmull", "u", int_arm_neon_vmullu, 1>;
2116 def VMULLp : N3VLInt<0, 1, 0b00, 0b1110, 0, IIC_VMULi16D, "vmull", "p8",
2117 v8i16, v8i8, int_arm_neon_vmullp, 1>;
2118 defm VMULLsls : N3VLIntSL_HS<0, 0b1010, IIC_VMULi16D, "vmull", "s",
2119 int_arm_neon_vmulls>;
2120 defm VMULLslu : N3VLIntSL_HS<1, 0b1010, IIC_VMULi16D, "vmull", "u",
2121 int_arm_neon_vmullu>;
2123 // VQDMULL : Vector Saturating Doubling Multiply Long (Q = D * D)
2124 defm VQDMULL : N3VLInt_HS<0,1,0b1101,0, IIC_VMULi16D, IIC_VMULi32D,
2125 "vqdmull", "s", int_arm_neon_vqdmull, 1>;
2126 defm VQDMULLsl: N3VLIntSL_HS<0, 0b1011, IIC_VMULi16D,
2127 "vqdmull", "s", int_arm_neon_vqdmull>;
2129 // Vector Multiply-Accumulate and Multiply-Subtract Operations.
2131 // VMLA : Vector Multiply Accumulate (integer and floating-point)
2132 defm VMLA : N3VMulOp_QHS<0, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
2133 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
2134 def VMLAfd : N3VDMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACD, "vmla", "f32",
2136 def VMLAfq : N3VQMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACQ, "vmla", "f32",
2138 defm VMLAsl : N3VMulOpSL_HS<0b0000, IIC_VMACi16D, IIC_VMACi32D,
2139 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
2140 def VMLAslfd : N3VDMulOpSL<0b10, 0b0001, IIC_VMACD, "vmla", "f32",
2142 def VMLAslfq : N3VQMulOpSL<0b10, 0b0001, IIC_VMACQ, "vmla", "f32",
2143 v4f32, v2f32, fmul, fadd>;
2145 def : Pat<(v8i16 (add (v8i16 QPR:$src1),
2146 (mul (v8i16 QPR:$src2),
2147 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
2148 (v8i16 (VMLAslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
2149 (v4i16 (EXTRACT_SUBREG QPR:$src3,
2150 (DSubReg_i16_reg imm:$lane))),
2151 (SubReg_i16_lane imm:$lane)))>;
2153 def : Pat<(v4i32 (add (v4i32 QPR:$src1),
2154 (mul (v4i32 QPR:$src2),
2155 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
2156 (v4i32 (VMLAslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
2157 (v2i32 (EXTRACT_SUBREG QPR:$src3,
2158 (DSubReg_i32_reg imm:$lane))),
2159 (SubReg_i32_lane imm:$lane)))>;
2161 def : Pat<(v4f32 (fadd (v4f32 QPR:$src1),
2162 (fmul (v4f32 QPR:$src2),
2163 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
2164 (v4f32 (VMLAslfq (v4f32 QPR:$src1),
2166 (v2f32 (EXTRACT_SUBREG QPR:$src3,
2167 (DSubReg_i32_reg imm:$lane))),
2168 (SubReg_i32_lane imm:$lane)))>;
2170 // VMLAL : Vector Multiply Accumulate Long (Q += D * D)
2171 defm VMLALs : N3VLInt3_QHS<0,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
2172 "vmlal", "s", int_arm_neon_vmlals>;
2173 defm VMLALu : N3VLInt3_QHS<1,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
2174 "vmlal", "u", int_arm_neon_vmlalu>;
2176 defm VMLALsls : N3VLInt3SL_HS<0, 0b0010, "vmlal", "s", int_arm_neon_vmlals>;
2177 defm VMLALslu : N3VLInt3SL_HS<1, 0b0010, "vmlal", "u", int_arm_neon_vmlalu>;
2179 // VQDMLAL : Vector Saturating Doubling Multiply Accumulate Long (Q += D * D)
2180 defm VQDMLAL : N3VLInt3_HS<0, 1, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
2181 "vqdmlal", "s", int_arm_neon_vqdmlal>;
2182 defm VQDMLALsl: N3VLInt3SL_HS<0, 0b0011, "vqdmlal", "s", int_arm_neon_vqdmlal>;
2184 // VMLS : Vector Multiply Subtract (integer and floating-point)
2185 defm VMLS : N3VMulOp_QHS<1, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
2186 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
2187 def VMLSfd : N3VDMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACD, "vmls", "f32",
2189 def VMLSfq : N3VQMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACQ, "vmls", "f32",
2191 defm VMLSsl : N3VMulOpSL_HS<0b0100, IIC_VMACi16D, IIC_VMACi32D,
2192 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
2193 def VMLSslfd : N3VDMulOpSL<0b10, 0b0101, IIC_VMACD, "vmls", "f32",
2195 def VMLSslfq : N3VQMulOpSL<0b10, 0b0101, IIC_VMACQ, "vmls", "f32",
2196 v4f32, v2f32, fmul, fsub>;
2198 def : Pat<(v8i16 (sub (v8i16 QPR:$src1),
2199 (mul (v8i16 QPR:$src2),
2200 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
2201 (v8i16 (VMLSslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
2202 (v4i16 (EXTRACT_SUBREG QPR:$src3,
2203 (DSubReg_i16_reg imm:$lane))),
2204 (SubReg_i16_lane imm:$lane)))>;
2206 def : Pat<(v4i32 (sub (v4i32 QPR:$src1),
2207 (mul (v4i32 QPR:$src2),
2208 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
2209 (v4i32 (VMLSslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
2210 (v2i32 (EXTRACT_SUBREG QPR:$src3,
2211 (DSubReg_i32_reg imm:$lane))),
2212 (SubReg_i32_lane imm:$lane)))>;
2214 def : Pat<(v4f32 (fsub (v4f32 QPR:$src1),
2215 (fmul (v4f32 QPR:$src2),
2216 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
2217 (v4f32 (VMLSslfq (v4f32 QPR:$src1), (v4f32 QPR:$src2),
2218 (v2f32 (EXTRACT_SUBREG QPR:$src3,
2219 (DSubReg_i32_reg imm:$lane))),
2220 (SubReg_i32_lane imm:$lane)))>;
2222 // VMLSL : Vector Multiply Subtract Long (Q -= D * D)
2223 defm VMLSLs : N3VLInt3_QHS<0,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
2224 "vmlsl", "s", int_arm_neon_vmlsls>;
2225 defm VMLSLu : N3VLInt3_QHS<1,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
2226 "vmlsl", "u", int_arm_neon_vmlslu>;
2228 defm VMLSLsls : N3VLInt3SL_HS<0, 0b0110, "vmlsl", "s", int_arm_neon_vmlsls>;
2229 defm VMLSLslu : N3VLInt3SL_HS<1, 0b0110, "vmlsl", "u", int_arm_neon_vmlslu>;
2231 // VQDMLSL : Vector Saturating Doubling Multiply Subtract Long (Q -= D * D)
2232 defm VQDMLSL : N3VLInt3_HS<0, 1, 0b1011, 0, IIC_VMACi16D, IIC_VMACi32D,
2233 "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
2234 defm VQDMLSLsl: N3VLInt3SL_HS<0, 0b111, "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
2236 // Vector Subtract Operations.
2238 // VSUB : Vector Subtract (integer and floating-point)
2239 defm VSUB : N3V_QHSD<1, 0, 0b1000, 0, IIC_VSUBiD, IIC_VSUBiQ,
2240 "vsub", "i", sub, 0>;
2241 def VSUBfd : N3VD<0, 0, 0b10, 0b1101, 0, IIC_VBIND, "vsub", "f32",
2242 v2f32, v2f32, fsub, 0>;
2243 def VSUBfq : N3VQ<0, 0, 0b10, 0b1101, 0, IIC_VBINQ, "vsub", "f32",
2244 v4f32, v4f32, fsub, 0>;
2245 // VSUBL : Vector Subtract Long (Q = D - D)
2246 defm VSUBLs : N3VLInt_QHS<0,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
2247 "vsubl", "s", int_arm_neon_vsubls, 1>;
2248 defm VSUBLu : N3VLInt_QHS<1,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
2249 "vsubl", "u", int_arm_neon_vsublu, 1>;
2250 // VSUBW : Vector Subtract Wide (Q = Q - D)
2251 defm VSUBWs : N3VWInt_QHS<0,1,0b0011,0, "vsubw", "s", int_arm_neon_vsubws, 0>;
2252 defm VSUBWu : N3VWInt_QHS<1,1,0b0011,0, "vsubw", "u", int_arm_neon_vsubwu, 0>;
2253 // VHSUB : Vector Halving Subtract
2254 defm VHSUBs : N3VInt_QHS<0, 0, 0b0010, 0, N3RegFrm,
2255 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2256 "vhsub", "s", int_arm_neon_vhsubs, 0>;
2257 defm VHSUBu : N3VInt_QHS<1, 0, 0b0010, 0, N3RegFrm,
2258 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2259 "vhsub", "u", int_arm_neon_vhsubu, 0>;
2260 // VQSUB : Vector Saturing Subtract
2261 defm VQSUBs : N3VInt_QHSD<0, 0, 0b0010, 1, N3RegFrm,
2262 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2263 "vqsub", "s", int_arm_neon_vqsubs, 0>;
2264 defm VQSUBu : N3VInt_QHSD<1, 0, 0b0010, 1, N3RegFrm,
2265 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2266 "vqsub", "u", int_arm_neon_vqsubu, 0>;
2267 // VSUBHN : Vector Subtract and Narrow Returning High Half (D = Q - Q)
2268 defm VSUBHN : N3VNInt_HSD<0,1,0b0110,0, "vsubhn", "i",
2269 int_arm_neon_vsubhn, 0>;
2270 // VRSUBHN : Vector Rounding Subtract and Narrow Returning High Half (D=Q-Q)
2271 defm VRSUBHN : N3VNInt_HSD<1,1,0b0110,0, "vrsubhn", "i",
2272 int_arm_neon_vrsubhn, 0>;
2274 // Vector Comparisons.
2276 // VCEQ : Vector Compare Equal
2277 defm VCEQ : N3V_QHS<1, 0, 0b1000, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
2278 IIC_VSUBi4Q, "vceq", "i", NEONvceq, 1>;
2279 def VCEQfd : N3VD<0,0,0b00,0b1110,0, IIC_VBIND, "vceq", "f32", v2i32, v2f32,
2281 def VCEQfq : N3VQ<0,0,0b00,0b1110,0, IIC_VBINQ, "vceq", "f32", v4i32, v4f32,
2283 // For disassembly only.
2284 defm VCEQz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00010, 0, "vceq", "i",
2287 // VCGE : Vector Compare Greater Than or Equal
2288 defm VCGEs : N3V_QHS<0, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
2289 IIC_VSUBi4Q, "vcge", "s", NEONvcge, 0>;
2290 defm VCGEu : N3V_QHS<1, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
2291 IIC_VSUBi4Q, "vcge", "u", NEONvcgeu, 0>;
2292 def VCGEfd : N3VD<1,0,0b00,0b1110,0, IIC_VBIND, "vcge", "f32", v2i32, v2f32,
2294 def VCGEfq : N3VQ<1,0,0b00,0b1110,0, IIC_VBINQ, "vcge", "f32", v4i32, v4f32,
2296 // For disassembly only.
2297 defm VCGEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00001, 0, "vcge", "s",
2299 // For disassembly only.
2300 defm VCLEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00011, 0, "vcle", "s",
2303 // VCGT : Vector Compare Greater Than
2304 defm VCGTs : N3V_QHS<0, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
2305 IIC_VSUBi4Q, "vcgt", "s", NEONvcgt, 0>;
2306 defm VCGTu : N3V_QHS<1, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
2307 IIC_VSUBi4Q, "vcgt", "u", NEONvcgtu, 0>;
2308 def VCGTfd : N3VD<1,0,0b10,0b1110,0, IIC_VBIND, "vcgt", "f32", v2i32, v2f32,
2310 def VCGTfq : N3VQ<1,0,0b10,0b1110,0, IIC_VBINQ, "vcgt", "f32", v4i32, v4f32,
2312 // For disassembly only.
2313 defm VCGTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00000, 0, "vcgt", "s",
2315 // For disassembly only.
2316 defm VCLTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00100, 0, "vclt", "s",
2319 // VACGE : Vector Absolute Compare Greater Than or Equal (aka VCAGE)
2320 def VACGEd : N3VDInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacge",
2321 "f32", v2i32, v2f32, int_arm_neon_vacged, 0>;
2322 def VACGEq : N3VQInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacge",
2323 "f32", v4i32, v4f32, int_arm_neon_vacgeq, 0>;
2324 // VACGT : Vector Absolute Compare Greater Than (aka VCAGT)
2325 def VACGTd : N3VDInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacgt",
2326 "f32", v2i32, v2f32, int_arm_neon_vacgtd, 0>;
2327 def VACGTq : N3VQInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacgt",
2328 "f32", v4i32, v4f32, int_arm_neon_vacgtq, 0>;
2329 // VTST : Vector Test Bits
2330 defm VTST : N3V_QHS<0, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2331 IIC_VBINi4Q, "vtst", "", NEONvtst, 1>;
2333 // Vector Bitwise Operations.
2335 def vnot8 : PatFrag<(ops node:$in),
2336 (xor node:$in, (bitconvert (v8i8 immAllOnesV)))>;
2337 def vnot16 : PatFrag<(ops node:$in),
2338 (xor node:$in, (bitconvert (v16i8 immAllOnesV)))>;
2341 // VAND : Vector Bitwise AND
2342 def VANDd : N3VDX<0, 0, 0b00, 0b0001, 1, IIC_VBINiD, "vand",
2343 v2i32, v2i32, and, 1>;
2344 def VANDq : N3VQX<0, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "vand",
2345 v4i32, v4i32, and, 1>;
2347 // VEOR : Vector Bitwise Exclusive OR
2348 def VEORd : N3VDX<1, 0, 0b00, 0b0001, 1, IIC_VBINiD, "veor",
2349 v2i32, v2i32, xor, 1>;
2350 def VEORq : N3VQX<1, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "veor",
2351 v4i32, v4i32, xor, 1>;
2353 // VORR : Vector Bitwise OR
2354 def VORRd : N3VDX<0, 0, 0b10, 0b0001, 1, IIC_VBINiD, "vorr",
2355 v2i32, v2i32, or, 1>;
2356 def VORRq : N3VQX<0, 0, 0b10, 0b0001, 1, IIC_VBINiQ, "vorr",
2357 v4i32, v4i32, or, 1>;
2359 // VBIC : Vector Bitwise Bit Clear (AND NOT)
2360 def VBICd : N3VX<0, 0, 0b01, 0b0001, 0, 1, (outs DPR:$dst),
2361 (ins DPR:$src1, DPR:$src2), N3RegFrm, IIC_VBINiD,
2362 "vbic", "$dst, $src1, $src2", "",
2363 [(set DPR:$dst, (v2i32 (and DPR:$src1,
2364 (vnot8 DPR:$src2))))]>;
2365 def VBICq : N3VX<0, 0, 0b01, 0b0001, 1, 1, (outs QPR:$dst),
2366 (ins QPR:$src1, QPR:$src2), N3RegFrm, IIC_VBINiQ,
2367 "vbic", "$dst, $src1, $src2", "",
2368 [(set QPR:$dst, (v4i32 (and QPR:$src1,
2369 (vnot16 QPR:$src2))))]>;
2371 // VORN : Vector Bitwise OR NOT
2372 def VORNd : N3VX<0, 0, 0b11, 0b0001, 0, 1, (outs DPR:$dst),
2373 (ins DPR:$src1, DPR:$src2), N3RegFrm, IIC_VBINiD,
2374 "vorn", "$dst, $src1, $src2", "",
2375 [(set DPR:$dst, (v2i32 (or DPR:$src1,
2376 (vnot8 DPR:$src2))))]>;
2377 def VORNq : N3VX<0, 0, 0b11, 0b0001, 1, 1, (outs QPR:$dst),
2378 (ins QPR:$src1, QPR:$src2), N3RegFrm, IIC_VBINiQ,
2379 "vorn", "$dst, $src1, $src2", "",
2380 [(set QPR:$dst, (v4i32 (or QPR:$src1,
2381 (vnot16 QPR:$src2))))]>;
2383 // VMVN : Vector Bitwise NOT
2384 def VMVNd : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 0, 0,
2385 (outs DPR:$dst), (ins DPR:$src), IIC_VSUBiD,
2386 "vmvn", "$dst, $src", "",
2387 [(set DPR:$dst, (v2i32 (vnot8 DPR:$src)))]>;
2388 def VMVNq : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 1, 0,
2389 (outs QPR:$dst), (ins QPR:$src), IIC_VSUBiD,
2390 "vmvn", "$dst, $src", "",
2391 [(set QPR:$dst, (v4i32 (vnot16 QPR:$src)))]>;
2392 def : Pat<(v2i32 (vnot8 DPR:$src)), (VMVNd DPR:$src)>;
2393 def : Pat<(v4i32 (vnot16 QPR:$src)), (VMVNq QPR:$src)>;
2395 // VBSL : Vector Bitwise Select
2396 def VBSLd : N3VX<1, 0, 0b01, 0b0001, 0, 1, (outs DPR:$dst),
2397 (ins DPR:$src1, DPR:$src2, DPR:$src3),
2398 N3RegFrm, IIC_VCNTiD,
2399 "vbsl", "$dst, $src2, $src3", "$src1 = $dst",
2401 (v2i32 (or (and DPR:$src2, DPR:$src1),
2402 (and DPR:$src3, (vnot8 DPR:$src1)))))]>;
2403 def VBSLq : N3VX<1, 0, 0b01, 0b0001, 1, 1, (outs QPR:$dst),
2404 (ins QPR:$src1, QPR:$src2, QPR:$src3),
2405 N3RegFrm, IIC_VCNTiQ,
2406 "vbsl", "$dst, $src2, $src3", "$src1 = $dst",
2408 (v4i32 (or (and QPR:$src2, QPR:$src1),
2409 (and QPR:$src3, (vnot16 QPR:$src1)))))]>;
2411 // VBIF : Vector Bitwise Insert if False
2412 // like VBSL but with: "vbif $dst, $src3, $src1", "$src2 = $dst",
2413 def VBIFd : N3VX<1, 0, 0b11, 0b0001, 0, 1,
2414 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3),
2415 N3RegFrm, IIC_VBINiD,
2416 "vbif", "$dst, $src2, $src3", "$src1 = $dst",
2417 [/* For disassembly only; pattern left blank */]>;
2418 def VBIFq : N3VX<1, 0, 0b11, 0b0001, 1, 1,
2419 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3),
2420 N3RegFrm, IIC_VBINiQ,
2421 "vbif", "$dst, $src2, $src3", "$src1 = $dst",
2422 [/* For disassembly only; pattern left blank */]>;
2424 // VBIT : Vector Bitwise Insert if True
2425 // like VBSL but with: "vbit $dst, $src2, $src1", "$src3 = $dst",
2426 def VBITd : N3VX<1, 0, 0b10, 0b0001, 0, 1,
2427 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3),
2428 N3RegFrm, IIC_VBINiD,
2429 "vbit", "$dst, $src2, $src3", "$src1 = $dst",
2430 [/* For disassembly only; pattern left blank */]>;
2431 def VBITq : N3VX<1, 0, 0b10, 0b0001, 1, 1,
2432 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3),
2433 N3RegFrm, IIC_VBINiQ,
2434 "vbit", "$dst, $src2, $src3", "$src1 = $dst",
2435 [/* For disassembly only; pattern left blank */]>;
2437 // VBIT/VBIF are not yet implemented. The TwoAddress pass will not go looking
2438 // for equivalent operations with different register constraints; it just
2441 // Vector Absolute Differences.
2443 // VABD : Vector Absolute Difference
2444 defm VABDs : N3VInt_QHS<0, 0, 0b0111, 0, N3RegFrm,
2445 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2446 "vabd", "s", int_arm_neon_vabds, 0>;
2447 defm VABDu : N3VInt_QHS<1, 0, 0b0111, 0, N3RegFrm,
2448 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2449 "vabd", "u", int_arm_neon_vabdu, 0>;
2450 def VABDfd : N3VDInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBIND,
2451 "vabd", "f32", v2f32, v2f32, int_arm_neon_vabds, 0>;
2452 def VABDfq : N3VQInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBINQ,
2453 "vabd", "f32", v4f32, v4f32, int_arm_neon_vabds, 0>;
2455 // VABDL : Vector Absolute Difference Long (Q = | D - D |)
2456 defm VABDLs : N3VLInt_QHS<0,1,0b0111,0, IIC_VSUBi4Q, IIC_VSUBi4Q,
2457 "vabdl", "s", int_arm_neon_vabdls, 0>;
2458 defm VABDLu : N3VLInt_QHS<1,1,0b0111,0, IIC_VSUBi4Q, IIC_VSUBi4Q,
2459 "vabdl", "u", int_arm_neon_vabdlu, 0>;
2461 // VABA : Vector Absolute Difference and Accumulate
2462 defm VABAs : N3VInt3_QHS<0,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
2463 "vaba", "s", int_arm_neon_vabas>;
2464 defm VABAu : N3VInt3_QHS<1,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
2465 "vaba", "u", int_arm_neon_vabau>;
2467 // VABAL : Vector Absolute Difference and Accumulate Long (Q += | D - D |)
2468 defm VABALs : N3VLInt3_QHS<0,1,0b0101,0, IIC_VABAD, IIC_VABAD,
2469 "vabal", "s", int_arm_neon_vabals>;
2470 defm VABALu : N3VLInt3_QHS<1,1,0b0101,0, IIC_VABAD, IIC_VABAD,
2471 "vabal", "u", int_arm_neon_vabalu>;
2473 // Vector Maximum and Minimum.
2475 // VMAX : Vector Maximum
2476 defm VMAXs : N3VInt_QHS<0, 0, 0b0110, 0, N3RegFrm,
2477 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2478 "vmax", "s", int_arm_neon_vmaxs, 1>;
2479 defm VMAXu : N3VInt_QHS<1, 0, 0b0110, 0, N3RegFrm,
2480 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2481 "vmax", "u", int_arm_neon_vmaxu, 1>;
2482 def VMAXfd : N3VDInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBIND,
2484 v2f32, v2f32, int_arm_neon_vmaxs, 1>;
2485 def VMAXfq : N3VQInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBINQ,
2487 v4f32, v4f32, int_arm_neon_vmaxs, 1>;
2489 // VMIN : Vector Minimum
2490 defm VMINs : N3VInt_QHS<0, 0, 0b0110, 1, N3RegFrm,
2491 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2492 "vmin", "s", int_arm_neon_vmins, 1>;
2493 defm VMINu : N3VInt_QHS<1, 0, 0b0110, 1, N3RegFrm,
2494 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2495 "vmin", "u", int_arm_neon_vminu, 1>;
2496 def VMINfd : N3VDInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBIND,
2498 v2f32, v2f32, int_arm_neon_vmins, 1>;
2499 def VMINfq : N3VQInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBINQ,
2501 v4f32, v4f32, int_arm_neon_vmins, 1>;
2503 // Vector Pairwise Operations.
2505 // VPADD : Vector Pairwise Add
2506 def VPADDi8 : N3VDInt<0, 0, 0b00, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
2508 v8i8, v8i8, int_arm_neon_vpadd, 0>;
2509 def VPADDi16 : N3VDInt<0, 0, 0b01, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
2511 v4i16, v4i16, int_arm_neon_vpadd, 0>;
2512 def VPADDi32 : N3VDInt<0, 0, 0b10, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
2514 v2i32, v2i32, int_arm_neon_vpadd, 0>;
2515 def VPADDf : N3VDInt<1, 0, 0b00, 0b1101, 0, N3RegFrm,
2516 IIC_VBIND, "vpadd", "f32",
2517 v2f32, v2f32, int_arm_neon_vpadd, 0>;
2519 // VPADDL : Vector Pairwise Add Long
2520 defm VPADDLs : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00100, 0, "vpaddl", "s",
2521 int_arm_neon_vpaddls>;
2522 defm VPADDLu : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00101, 0, "vpaddl", "u",
2523 int_arm_neon_vpaddlu>;
2525 // VPADAL : Vector Pairwise Add and Accumulate Long
2526 defm VPADALs : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01100, 0, "vpadal", "s",
2527 int_arm_neon_vpadals>;
2528 defm VPADALu : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01101, 0, "vpadal", "u",
2529 int_arm_neon_vpadalu>;
2531 // VPMAX : Vector Pairwise Maximum
2532 def VPMAXs8 : N3VDInt<0, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
2533 "s8", v8i8, v8i8, int_arm_neon_vpmaxs, 0>;
2534 def VPMAXs16 : N3VDInt<0, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
2535 "s16", v4i16, v4i16, int_arm_neon_vpmaxs, 0>;
2536 def VPMAXs32 : N3VDInt<0, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
2537 "s32", v2i32, v2i32, int_arm_neon_vpmaxs, 0>;
2538 def VPMAXu8 : N3VDInt<1, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
2539 "u8", v8i8, v8i8, int_arm_neon_vpmaxu, 0>;
2540 def VPMAXu16 : N3VDInt<1, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
2541 "u16", v4i16, v4i16, int_arm_neon_vpmaxu, 0>;
2542 def VPMAXu32 : N3VDInt<1, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
2543 "u32", v2i32, v2i32, int_arm_neon_vpmaxu, 0>;
2544 def VPMAXf : N3VDInt<1, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
2545 "f32", v2f32, v2f32, int_arm_neon_vpmaxs, 0>;
2547 // VPMIN : Vector Pairwise Minimum
2548 def VPMINs8 : N3VDInt<0, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
2549 "s8", v8i8, v8i8, int_arm_neon_vpmins, 0>;
2550 def VPMINs16 : N3VDInt<0, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
2551 "s16", v4i16, v4i16, int_arm_neon_vpmins, 0>;
2552 def VPMINs32 : N3VDInt<0, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
2553 "s32", v2i32, v2i32, int_arm_neon_vpmins, 0>;
2554 def VPMINu8 : N3VDInt<1, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
2555 "u8", v8i8, v8i8, int_arm_neon_vpminu, 0>;
2556 def VPMINu16 : N3VDInt<1, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
2557 "u16", v4i16, v4i16, int_arm_neon_vpminu, 0>;
2558 def VPMINu32 : N3VDInt<1, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
2559 "u32", v2i32, v2i32, int_arm_neon_vpminu, 0>;
2560 def VPMINf : N3VDInt<1, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VSUBi4D, "vpmin",
2561 "f32", v2f32, v2f32, int_arm_neon_vpmins, 0>;
2563 // Vector Reciprocal and Reciprocal Square Root Estimate and Step.
2565 // VRECPE : Vector Reciprocal Estimate
2566 def VRECPEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
2567 IIC_VUNAD, "vrecpe", "u32",
2568 v2i32, v2i32, int_arm_neon_vrecpe>;
2569 def VRECPEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
2570 IIC_VUNAQ, "vrecpe", "u32",
2571 v4i32, v4i32, int_arm_neon_vrecpe>;
2572 def VRECPEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
2573 IIC_VUNAD, "vrecpe", "f32",
2574 v2f32, v2f32, int_arm_neon_vrecpe>;
2575 def VRECPEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
2576 IIC_VUNAQ, "vrecpe", "f32",
2577 v4f32, v4f32, int_arm_neon_vrecpe>;
2579 // VRECPS : Vector Reciprocal Step
2580 def VRECPSfd : N3VDInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
2581 IIC_VRECSD, "vrecps", "f32",
2582 v2f32, v2f32, int_arm_neon_vrecps, 1>;
2583 def VRECPSfq : N3VQInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
2584 IIC_VRECSQ, "vrecps", "f32",
2585 v4f32, v4f32, int_arm_neon_vrecps, 1>;
2587 // VRSQRTE : Vector Reciprocal Square Root Estimate
2588 def VRSQRTEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
2589 IIC_VUNAD, "vrsqrte", "u32",
2590 v2i32, v2i32, int_arm_neon_vrsqrte>;
2591 def VRSQRTEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
2592 IIC_VUNAQ, "vrsqrte", "u32",
2593 v4i32, v4i32, int_arm_neon_vrsqrte>;
2594 def VRSQRTEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
2595 IIC_VUNAD, "vrsqrte", "f32",
2596 v2f32, v2f32, int_arm_neon_vrsqrte>;
2597 def VRSQRTEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
2598 IIC_VUNAQ, "vrsqrte", "f32",
2599 v4f32, v4f32, int_arm_neon_vrsqrte>;
2601 // VRSQRTS : Vector Reciprocal Square Root Step
2602 def VRSQRTSfd : N3VDInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
2603 IIC_VRECSD, "vrsqrts", "f32",
2604 v2f32, v2f32, int_arm_neon_vrsqrts, 1>;
2605 def VRSQRTSfq : N3VQInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
2606 IIC_VRECSQ, "vrsqrts", "f32",
2607 v4f32, v4f32, int_arm_neon_vrsqrts, 1>;
2611 // VSHL : Vector Shift
2612 defm VSHLs : N3VInt_QHSD<0, 0, 0b0100, 0, N3RegVShFrm,
2613 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
2614 "vshl", "s", int_arm_neon_vshifts, 0>;
2615 defm VSHLu : N3VInt_QHSD<1, 0, 0b0100, 0, N3RegVShFrm,
2616 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
2617 "vshl", "u", int_arm_neon_vshiftu, 0>;
2618 // VSHL : Vector Shift Left (Immediate)
2619 defm VSHLi : N2VSh_QHSD<0, 1, 0b0101, 1, IIC_VSHLiD, "vshl", "i", NEONvshl,
2621 // VSHR : Vector Shift Right (Immediate)
2622 defm VSHRs : N2VSh_QHSD<0, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "s", NEONvshrs,
2624 defm VSHRu : N2VSh_QHSD<1, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "u", NEONvshru,
2627 // VSHLL : Vector Shift Left Long
2628 defm VSHLLs : N2VLSh_QHS<0, 1, 0b1010, 0, 0, 1, "vshll", "s", NEONvshlls>;
2629 defm VSHLLu : N2VLSh_QHS<1, 1, 0b1010, 0, 0, 1, "vshll", "u", NEONvshllu>;
2631 // VSHLL : Vector Shift Left Long (with maximum shift count)
2632 class N2VLShMax<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
2633 bit op6, bit op4, string OpcodeStr, string Dt, ValueType ResTy,
2634 ValueType OpTy, SDNode OpNode>
2635 : N2VLSh<op24, op23, op11_8, op7, op6, op4, OpcodeStr, Dt,
2636 ResTy, OpTy, OpNode> {
2637 let Inst{21-16} = op21_16;
2639 def VSHLLi8 : N2VLShMax<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll", "i8",
2640 v8i16, v8i8, NEONvshlli>;
2641 def VSHLLi16 : N2VLShMax<1, 1, 0b110110, 0b0011, 0, 0, 0, "vshll", "i16",
2642 v4i32, v4i16, NEONvshlli>;
2643 def VSHLLi32 : N2VLShMax<1, 1, 0b111010, 0b0011, 0, 0, 0, "vshll", "i32",
2644 v2i64, v2i32, NEONvshlli>;
2646 // VSHRN : Vector Shift Right and Narrow
2647 defm VSHRN : N2VNSh_HSD<0,1,0b1000,0,0,1, IIC_VSHLiD, "vshrn", "i",
2650 // VRSHL : Vector Rounding Shift
2651 defm VRSHLs : N3VInt_QHSD<0, 0, 0b0101, 0, N3RegVShFrm,
2652 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
2653 "vrshl", "s", int_arm_neon_vrshifts, 0>;
2654 defm VRSHLu : N3VInt_QHSD<1, 0, 0b0101, 0, N3RegVShFrm,
2655 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
2656 "vrshl", "u", int_arm_neon_vrshiftu, 0>;
2657 // VRSHR : Vector Rounding Shift Right
2658 defm VRSHRs : N2VSh_QHSD<0,1,0b0010,1, IIC_VSHLi4D, "vrshr", "s", NEONvrshrs,
2660 defm VRSHRu : N2VSh_QHSD<1,1,0b0010,1, IIC_VSHLi4D, "vrshr", "u", NEONvrshru,
2663 // VRSHRN : Vector Rounding Shift Right and Narrow
2664 defm VRSHRN : N2VNSh_HSD<0, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vrshrn", "i",
2667 // VQSHL : Vector Saturating Shift
2668 defm VQSHLs : N3VInt_QHSD<0, 0, 0b0100, 1, N3RegVShFrm,
2669 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
2670 "vqshl", "s", int_arm_neon_vqshifts, 0>;
2671 defm VQSHLu : N3VInt_QHSD<1, 0, 0b0100, 1, N3RegVShFrm,
2672 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
2673 "vqshl", "u", int_arm_neon_vqshiftu, 0>;
2674 // VQSHL : Vector Saturating Shift Left (Immediate)
2675 defm VQSHLsi : N2VSh_QHSD<0,1,0b0111,1, IIC_VSHLi4D, "vqshl", "s",NEONvqshls,
2677 defm VQSHLui : N2VSh_QHSD<1,1,0b0111,1, IIC_VSHLi4D, "vqshl", "u",NEONvqshlu,
2679 // VQSHLU : Vector Saturating Shift Left (Immediate, Unsigned)
2680 defm VQSHLsu : N2VSh_QHSD<1,1,0b0110,1, IIC_VSHLi4D,"vqshlu","s",NEONvqshlsu,
2683 // VQSHRN : Vector Saturating Shift Right and Narrow
2684 defm VQSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "s",
2686 defm VQSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "u",
2689 // VQSHRUN : Vector Saturating Shift Right and Narrow (Unsigned)
2690 defm VQSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 0, 1, IIC_VSHLi4D, "vqshrun", "s",
2693 // VQRSHL : Vector Saturating Rounding Shift
2694 defm VQRSHLs : N3VInt_QHSD<0, 0, 0b0101, 1, N3RegVShFrm,
2695 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
2696 "vqrshl", "s", int_arm_neon_vqrshifts, 0>;
2697 defm VQRSHLu : N3VInt_QHSD<1, 0, 0b0101, 1, N3RegVShFrm,
2698 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
2699 "vqrshl", "u", int_arm_neon_vqrshiftu, 0>;
2701 // VQRSHRN : Vector Saturating Rounding Shift Right and Narrow
2702 defm VQRSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "s",
2704 defm VQRSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "u",
2707 // VQRSHRUN : Vector Saturating Rounding Shift Right and Narrow (Unsigned)
2708 defm VQRSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vqrshrun", "s",
2711 // VSRA : Vector Shift Right and Accumulate
2712 defm VSRAs : N2VShAdd_QHSD<0, 1, 0b0001, 1, "vsra", "s", NEONvshrs>;
2713 defm VSRAu : N2VShAdd_QHSD<1, 1, 0b0001, 1, "vsra", "u", NEONvshru>;
2714 // VRSRA : Vector Rounding Shift Right and Accumulate
2715 defm VRSRAs : N2VShAdd_QHSD<0, 1, 0b0011, 1, "vrsra", "s", NEONvrshrs>;
2716 defm VRSRAu : N2VShAdd_QHSD<1, 1, 0b0011, 1, "vrsra", "u", NEONvrshru>;
2718 // VSLI : Vector Shift Left and Insert
2719 defm VSLI : N2VShIns_QHSD<1, 1, 0b0101, 1, "vsli", NEONvsli, N2RegVShLFrm>;
2720 // VSRI : Vector Shift Right and Insert
2721 defm VSRI : N2VShIns_QHSD<1, 1, 0b0100, 1, "vsri", NEONvsri, N2RegVShRFrm>;
2723 // Vector Absolute and Saturating Absolute.
2725 // VABS : Vector Absolute Value
2726 defm VABS : N2VInt_QHS<0b11, 0b11, 0b01, 0b00110, 0,
2727 IIC_VUNAiD, IIC_VUNAiQ, "vabs", "s",
2729 def VABSfd : N2VDInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
2730 IIC_VUNAD, "vabs", "f32",
2731 v2f32, v2f32, int_arm_neon_vabs>;
2732 def VABSfq : N2VQInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
2733 IIC_VUNAQ, "vabs", "f32",
2734 v4f32, v4f32, int_arm_neon_vabs>;
2736 // VQABS : Vector Saturating Absolute Value
2737 defm VQABS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01110, 0,
2738 IIC_VQUNAiD, IIC_VQUNAiQ, "vqabs", "s",
2739 int_arm_neon_vqabs>;
2743 def vneg : PatFrag<(ops node:$in), (sub immAllZerosV, node:$in)>;
2744 def vneg8 : PatFrag<(ops node:$in),
2745 (sub (bitconvert (v8i8 immAllZerosV)), node:$in)>;
2746 def vneg16 : PatFrag<(ops node:$in),
2747 (sub (bitconvert (v16i8 immAllZerosV)), node:$in)>;
2749 class VNEGD<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
2750 : N2V<0b11, 0b11, size, 0b01, 0b00111, 0, 0, (outs DPR:$dst), (ins DPR:$src),
2751 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
2752 [(set DPR:$dst, (Ty (vneg8 DPR:$src)))]>;
2753 class VNEGQ<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
2754 : N2V<0b11, 0b11, size, 0b01, 0b00111, 1, 0, (outs QPR:$dst), (ins QPR:$src),
2755 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
2756 [(set QPR:$dst, (Ty (vneg16 QPR:$src)))]>;
2758 // VNEG : Vector Negate (integer)
2759 def VNEGs8d : VNEGD<0b00, "vneg", "s8", v8i8>;
2760 def VNEGs16d : VNEGD<0b01, "vneg", "s16", v4i16>;
2761 def VNEGs32d : VNEGD<0b10, "vneg", "s32", v2i32>;
2762 def VNEGs8q : VNEGQ<0b00, "vneg", "s8", v16i8>;
2763 def VNEGs16q : VNEGQ<0b01, "vneg", "s16", v8i16>;
2764 def VNEGs32q : VNEGQ<0b10, "vneg", "s32", v4i32>;
2766 // VNEG : Vector Negate (floating-point)
2767 def VNEGfd : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
2768 (outs DPR:$dst), (ins DPR:$src), IIC_VUNAD,
2769 "vneg", "f32", "$dst, $src", "",
2770 [(set DPR:$dst, (v2f32 (fneg DPR:$src)))]>;
2771 def VNEGf32q : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 1, 0,
2772 (outs QPR:$dst), (ins QPR:$src), IIC_VUNAQ,
2773 "vneg", "f32", "$dst, $src", "",
2774 [(set QPR:$dst, (v4f32 (fneg QPR:$src)))]>;
2776 def : Pat<(v8i8 (vneg8 DPR:$src)), (VNEGs8d DPR:$src)>;
2777 def : Pat<(v4i16 (vneg8 DPR:$src)), (VNEGs16d DPR:$src)>;
2778 def : Pat<(v2i32 (vneg8 DPR:$src)), (VNEGs32d DPR:$src)>;
2779 def : Pat<(v16i8 (vneg16 QPR:$src)), (VNEGs8q QPR:$src)>;
2780 def : Pat<(v8i16 (vneg16 QPR:$src)), (VNEGs16q QPR:$src)>;
2781 def : Pat<(v4i32 (vneg16 QPR:$src)), (VNEGs32q QPR:$src)>;
2783 // VQNEG : Vector Saturating Negate
2784 defm VQNEG : N2VInt_QHS<0b11, 0b11, 0b00, 0b01111, 0,
2785 IIC_VQUNAiD, IIC_VQUNAiQ, "vqneg", "s",
2786 int_arm_neon_vqneg>;
2788 // Vector Bit Counting Operations.
2790 // VCLS : Vector Count Leading Sign Bits
2791 defm VCLS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01000, 0,
2792 IIC_VCNTiD, IIC_VCNTiQ, "vcls", "s",
2794 // VCLZ : Vector Count Leading Zeros
2795 defm VCLZ : N2VInt_QHS<0b11, 0b11, 0b00, 0b01001, 0,
2796 IIC_VCNTiD, IIC_VCNTiQ, "vclz", "i",
2798 // VCNT : Vector Count One Bits
2799 def VCNTd : N2VDInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
2800 IIC_VCNTiD, "vcnt", "8",
2801 v8i8, v8i8, int_arm_neon_vcnt>;
2802 def VCNTq : N2VQInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
2803 IIC_VCNTiQ, "vcnt", "8",
2804 v16i8, v16i8, int_arm_neon_vcnt>;
2806 // Vector Swap -- for disassembly only.
2807 def VSWPd : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 0, 0,
2808 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
2809 "vswp", "$dst, $src", "", []>;
2810 def VSWPq : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 1, 0,
2811 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
2812 "vswp", "$dst, $src", "", []>;
2814 // Vector Move Operations.
2816 // VMOV : Vector Move (Register)
2818 let neverHasSideEffects = 1 in {
2819 def VMOVDneon: N3VX<0, 0, 0b10, 0b0001, 0, 1, (outs DPR:$dst), (ins DPR:$src),
2820 N3RegFrm, IIC_VMOVD, "vmov", "$dst, $src", "", []>;
2821 def VMOVQ : N3VX<0, 0, 0b10, 0b0001, 1, 1, (outs QPR:$dst), (ins QPR:$src),
2822 N3RegFrm, IIC_VMOVD, "vmov", "$dst, $src", "", []>;
2824 // Pseudo vector move instructions for QQ and QQQQ registers. This should
2825 // be expanded after register allocation is completed.
2826 def VMOVQQ : PseudoInst<(outs QQPR:$dst), (ins QQPR:$src),
2827 NoItinerary, "${:comment} vmov\t$dst, $src", []>;
2829 def VMOVQQQQ : PseudoInst<(outs QQQQPR:$dst), (ins QQQQPR:$src),
2830 NoItinerary, "${:comment} vmov\t$dst, $src", []>;
2831 } // neverHasSideEffects
2833 // VMOV : Vector Move (Immediate)
2835 // VMOV_get_imm8 xform function: convert build_vector to VMOV.i8 imm.
2836 def VMOV_get_imm8 : SDNodeXForm<build_vector, [{
2837 return ARM::getVMOVImm(N, 1, *CurDAG);
2839 def vmovImm8 : PatLeaf<(build_vector), [{
2840 return ARM::getVMOVImm(N, 1, *CurDAG).getNode() != 0;
2843 // VMOV_get_imm16 xform function: convert build_vector to VMOV.i16 imm.
2844 def VMOV_get_imm16 : SDNodeXForm<build_vector, [{
2845 return ARM::getVMOVImm(N, 2, *CurDAG);
2847 def vmovImm16 : PatLeaf<(build_vector), [{
2848 return ARM::getVMOVImm(N, 2, *CurDAG).getNode() != 0;
2849 }], VMOV_get_imm16>;
2851 // VMOV_get_imm32 xform function: convert build_vector to VMOV.i32 imm.
2852 def VMOV_get_imm32 : SDNodeXForm<build_vector, [{
2853 return ARM::getVMOVImm(N, 4, *CurDAG);
2855 def vmovImm32 : PatLeaf<(build_vector), [{
2856 return ARM::getVMOVImm(N, 4, *CurDAG).getNode() != 0;
2857 }], VMOV_get_imm32>;
2859 // VMOV_get_imm64 xform function: convert build_vector to VMOV.i64 imm.
2860 def VMOV_get_imm64 : SDNodeXForm<build_vector, [{
2861 return ARM::getVMOVImm(N, 8, *CurDAG);
2863 def vmovImm64 : PatLeaf<(build_vector), [{
2864 return ARM::getVMOVImm(N, 8, *CurDAG).getNode() != 0;
2865 }], VMOV_get_imm64>;
2867 // Note: Some of the cmode bits in the following VMOV instructions need to
2868 // be encoded based on the immed values.
2870 let isReMaterializable = 1 in {
2871 def VMOVv8i8 : N1ModImm<1, 0b000, 0b1110, 0, 0, 0, 1, (outs DPR:$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:$SIMM))]>;
2906 } // isReMaterializable
2908 // VMOV : Vector Get Lane (move scalar to ARM core register)
2910 def VGETLNs8 : NVGetLane<{1,1,1,0,0,1,?,1}, 0b1011, {?,?},
2911 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2912 IIC_VMOVSI, "vmov", "s8", "$dst, $src[$lane]",
2913 [(set GPR:$dst, (NEONvgetlanes (v8i8 DPR:$src),
2915 def VGETLNs16 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, {?,1},
2916 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2917 IIC_VMOVSI, "vmov", "s16", "$dst, $src[$lane]",
2918 [(set GPR:$dst, (NEONvgetlanes (v4i16 DPR:$src),
2920 def VGETLNu8 : NVGetLane<{1,1,1,0,1,1,?,1}, 0b1011, {?,?},
2921 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2922 IIC_VMOVSI, "vmov", "u8", "$dst, $src[$lane]",
2923 [(set GPR:$dst, (NEONvgetlaneu (v8i8 DPR:$src),
2925 def VGETLNu16 : NVGetLane<{1,1,1,0,1,0,?,1}, 0b1011, {?,1},
2926 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2927 IIC_VMOVSI, "vmov", "u16", "$dst, $src[$lane]",
2928 [(set GPR:$dst, (NEONvgetlaneu (v4i16 DPR:$src),
2930 def VGETLNi32 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, 0b00,
2931 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2932 IIC_VMOVSI, "vmov", "32", "$dst, $src[$lane]",
2933 [(set GPR:$dst, (extractelt (v2i32 DPR:$src),
2935 // def VGETLNf32: see FMRDH and FMRDL in ARMInstrVFP.td
2936 def : Pat<(NEONvgetlanes (v16i8 QPR:$src), imm:$lane),
2937 (VGETLNs8 (v8i8 (EXTRACT_SUBREG QPR:$src,
2938 (DSubReg_i8_reg imm:$lane))),
2939 (SubReg_i8_lane imm:$lane))>;
2940 def : Pat<(NEONvgetlanes (v8i16 QPR:$src), imm:$lane),
2941 (VGETLNs16 (v4i16 (EXTRACT_SUBREG QPR:$src,
2942 (DSubReg_i16_reg imm:$lane))),
2943 (SubReg_i16_lane imm:$lane))>;
2944 def : Pat<(NEONvgetlaneu (v16i8 QPR:$src), imm:$lane),
2945 (VGETLNu8 (v8i8 (EXTRACT_SUBREG QPR:$src,
2946 (DSubReg_i8_reg imm:$lane))),
2947 (SubReg_i8_lane imm:$lane))>;
2948 def : Pat<(NEONvgetlaneu (v8i16 QPR:$src), imm:$lane),
2949 (VGETLNu16 (v4i16 (EXTRACT_SUBREG QPR:$src,
2950 (DSubReg_i16_reg imm:$lane))),
2951 (SubReg_i16_lane imm:$lane))>;
2952 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
2953 (VGETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src,
2954 (DSubReg_i32_reg imm:$lane))),
2955 (SubReg_i32_lane imm:$lane))>;
2956 def : Pat<(extractelt (v2f32 DPR:$src1), imm:$src2),
2957 (EXTRACT_SUBREG (v2f32 (COPY_TO_REGCLASS (v2f32 DPR:$src1),DPR_VFP2)),
2958 (SSubReg_f32_reg imm:$src2))>;
2959 def : Pat<(extractelt (v4f32 QPR:$src1), imm:$src2),
2960 (EXTRACT_SUBREG (v4f32 (COPY_TO_REGCLASS (v4f32 QPR:$src1),QPR_VFP2)),
2961 (SSubReg_f32_reg imm:$src2))>;
2962 //def : Pat<(extractelt (v2i64 QPR:$src1), imm:$src2),
2963 // (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
2964 def : Pat<(extractelt (v2f64 QPR:$src1), imm:$src2),
2965 (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
2968 // VMOV : Vector Set Lane (move ARM core register to scalar)
2970 let Constraints = "$src1 = $dst" in {
2971 def VSETLNi8 : NVSetLane<{1,1,1,0,0,1,?,0}, 0b1011, {?,?}, (outs DPR:$dst),
2972 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
2973 IIC_VMOVISL, "vmov", "8", "$dst[$lane], $src2",
2974 [(set DPR:$dst, (vector_insert (v8i8 DPR:$src1),
2975 GPR:$src2, imm:$lane))]>;
2976 def VSETLNi16 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, {?,1}, (outs DPR:$dst),
2977 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
2978 IIC_VMOVISL, "vmov", "16", "$dst[$lane], $src2",
2979 [(set DPR:$dst, (vector_insert (v4i16 DPR:$src1),
2980 GPR:$src2, imm:$lane))]>;
2981 def VSETLNi32 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, 0b00, (outs DPR:$dst),
2982 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
2983 IIC_VMOVISL, "vmov", "32", "$dst[$lane], $src2",
2984 [(set DPR:$dst, (insertelt (v2i32 DPR:$src1),
2985 GPR:$src2, imm:$lane))]>;
2987 def : Pat<(vector_insert (v16i8 QPR:$src1), GPR:$src2, imm:$lane),
2988 (v16i8 (INSERT_SUBREG QPR:$src1,
2989 (v8i8 (VSETLNi8 (v8i8 (EXTRACT_SUBREG QPR:$src1,
2990 (DSubReg_i8_reg imm:$lane))),
2991 GPR:$src2, (SubReg_i8_lane imm:$lane))),
2992 (DSubReg_i8_reg imm:$lane)))>;
2993 def : Pat<(vector_insert (v8i16 QPR:$src1), GPR:$src2, imm:$lane),
2994 (v8i16 (INSERT_SUBREG QPR:$src1,
2995 (v4i16 (VSETLNi16 (v4i16 (EXTRACT_SUBREG QPR:$src1,
2996 (DSubReg_i16_reg imm:$lane))),
2997 GPR:$src2, (SubReg_i16_lane imm:$lane))),
2998 (DSubReg_i16_reg imm:$lane)))>;
2999 def : Pat<(insertelt (v4i32 QPR:$src1), GPR:$src2, imm:$lane),
3000 (v4i32 (INSERT_SUBREG QPR:$src1,
3001 (v2i32 (VSETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src1,
3002 (DSubReg_i32_reg imm:$lane))),
3003 GPR:$src2, (SubReg_i32_lane imm:$lane))),
3004 (DSubReg_i32_reg imm:$lane)))>;
3006 def : Pat<(v2f32 (insertelt DPR:$src1, SPR:$src2, imm:$src3)),
3007 (INSERT_SUBREG (v2f32 (COPY_TO_REGCLASS DPR:$src1, DPR_VFP2)),
3008 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
3009 def : Pat<(v4f32 (insertelt QPR:$src1, SPR:$src2, imm:$src3)),
3010 (INSERT_SUBREG (v4f32 (COPY_TO_REGCLASS QPR:$src1, QPR_VFP2)),
3011 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
3013 //def : Pat<(v2i64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
3014 // (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
3015 def : Pat<(v2f64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
3016 (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
3018 def : Pat<(v2f32 (scalar_to_vector SPR:$src)),
3019 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
3020 def : Pat<(v2f64 (scalar_to_vector (f64 DPR:$src))),
3021 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
3022 def : Pat<(v4f32 (scalar_to_vector SPR:$src)),
3023 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
3025 def : Pat<(v8i8 (scalar_to_vector GPR:$src)),
3026 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
3027 def : Pat<(v4i16 (scalar_to_vector GPR:$src)),
3028 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
3029 def : Pat<(v2i32 (scalar_to_vector GPR:$src)),
3030 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
3032 def : Pat<(v16i8 (scalar_to_vector GPR:$src)),
3033 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3034 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
3036 def : Pat<(v8i16 (scalar_to_vector GPR:$src)),
3037 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
3038 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
3040 def : Pat<(v4i32 (scalar_to_vector GPR:$src)),
3041 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
3042 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
3045 // VDUP : Vector Duplicate (from ARM core register to all elements)
3047 class VDUPD<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
3048 : NVDup<opcod1, 0b1011, opcod3, (outs DPR:$dst), (ins GPR:$src),
3049 IIC_VMOVIS, "vdup", Dt, "$dst, $src",
3050 [(set DPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
3051 class VDUPQ<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
3052 : NVDup<opcod1, 0b1011, opcod3, (outs QPR:$dst), (ins GPR:$src),
3053 IIC_VMOVIS, "vdup", Dt, "$dst, $src",
3054 [(set QPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
3056 def VDUP8d : VDUPD<0b11101100, 0b00, "8", v8i8>;
3057 def VDUP16d : VDUPD<0b11101000, 0b01, "16", v4i16>;
3058 def VDUP32d : VDUPD<0b11101000, 0b00, "32", v2i32>;
3059 def VDUP8q : VDUPQ<0b11101110, 0b00, "8", v16i8>;
3060 def VDUP16q : VDUPQ<0b11101010, 0b01, "16", v8i16>;
3061 def VDUP32q : VDUPQ<0b11101010, 0b00, "32", v4i32>;
3063 def VDUPfd : NVDup<0b11101000, 0b1011, 0b00, (outs DPR:$dst), (ins GPR:$src),
3064 IIC_VMOVIS, "vdup", "32", "$dst, $src",
3065 [(set DPR:$dst, (v2f32 (NEONvdup
3066 (f32 (bitconvert GPR:$src)))))]>;
3067 def VDUPfq : NVDup<0b11101010, 0b1011, 0b00, (outs QPR:$dst), (ins GPR:$src),
3068 IIC_VMOVIS, "vdup", "32", "$dst, $src",
3069 [(set QPR:$dst, (v4f32 (NEONvdup
3070 (f32 (bitconvert GPR:$src)))))]>;
3072 // VDUP : Vector Duplicate Lane (from scalar to all elements)
3074 class VDUPLND<bits<4> op19_16, string OpcodeStr, string Dt,
3076 : NVDupLane<op19_16, 0, (outs DPR:$dst), (ins DPR:$src, nohash_imm:$lane),
3077 IIC_VMOVD, OpcodeStr, Dt, "$dst, $src[$lane]",
3078 [(set DPR:$dst, (Ty (NEONvduplane (Ty DPR:$src), imm:$lane)))]>;
3080 class VDUPLNQ<bits<4> op19_16, string OpcodeStr, string Dt,
3081 ValueType ResTy, ValueType OpTy>
3082 : NVDupLane<op19_16, 1, (outs QPR:$dst), (ins DPR:$src, nohash_imm:$lane),
3083 IIC_VMOVD, OpcodeStr, Dt, "$dst, $src[$lane]",
3084 [(set QPR:$dst, (ResTy (NEONvduplane (OpTy DPR:$src),
3087 // Inst{19-16} is partially specified depending on the element size.
3089 def VDUPLN8d : VDUPLND<{?,?,?,1}, "vdup", "8", v8i8>;
3090 def VDUPLN16d : VDUPLND<{?,?,1,0}, "vdup", "16", v4i16>;
3091 def VDUPLN32d : VDUPLND<{?,1,0,0}, "vdup", "32", v2i32>;
3092 def VDUPLNfd : VDUPLND<{?,1,0,0}, "vdup", "32", v2f32>;
3093 def VDUPLN8q : VDUPLNQ<{?,?,?,1}, "vdup", "8", v16i8, v8i8>;
3094 def VDUPLN16q : VDUPLNQ<{?,?,1,0}, "vdup", "16", v8i16, v4i16>;
3095 def VDUPLN32q : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4i32, v2i32>;
3096 def VDUPLNfq : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4f32, v2f32>;
3098 def : Pat<(v16i8 (NEONvduplane (v16i8 QPR:$src), imm:$lane)),
3099 (v16i8 (VDUPLN8q (v8i8 (EXTRACT_SUBREG QPR:$src,
3100 (DSubReg_i8_reg imm:$lane))),
3101 (SubReg_i8_lane imm:$lane)))>;
3102 def : Pat<(v8i16 (NEONvduplane (v8i16 QPR:$src), imm:$lane)),
3103 (v8i16 (VDUPLN16q (v4i16 (EXTRACT_SUBREG QPR:$src,
3104 (DSubReg_i16_reg imm:$lane))),
3105 (SubReg_i16_lane imm:$lane)))>;
3106 def : Pat<(v4i32 (NEONvduplane (v4i32 QPR:$src), imm:$lane)),
3107 (v4i32 (VDUPLN32q (v2i32 (EXTRACT_SUBREG QPR:$src,
3108 (DSubReg_i32_reg imm:$lane))),
3109 (SubReg_i32_lane imm:$lane)))>;
3110 def : Pat<(v4f32 (NEONvduplane (v4f32 QPR:$src), imm:$lane)),
3111 (v4f32 (VDUPLNfq (v2f32 (EXTRACT_SUBREG QPR:$src,
3112 (DSubReg_i32_reg imm:$lane))),
3113 (SubReg_i32_lane imm:$lane)))>;
3115 def VDUPfdf : N2V<0b11, 0b11, {?,1}, {0,0}, 0b11000, 0, 0,
3116 (outs DPR:$dst), (ins SPR:$src),
3117 IIC_VMOVD, "vdup", "32", "$dst, ${src:lane}", "",
3118 [(set DPR:$dst, (v2f32 (NEONvdup (f32 SPR:$src))))]>;
3120 def VDUPfqf : N2V<0b11, 0b11, {?,1}, {0,0}, 0b11000, 1, 0,
3121 (outs QPR:$dst), (ins SPR:$src),
3122 IIC_VMOVD, "vdup", "32", "$dst, ${src:lane}", "",
3123 [(set QPR:$dst, (v4f32 (NEONvdup (f32 SPR:$src))))]>;
3125 def : Pat<(v2i64 (NEONvduplane (v2i64 QPR:$src), imm:$lane)),
3126 (INSERT_SUBREG QPR:$src,
3127 (i64 (EXTRACT_SUBREG QPR:$src,
3128 (DSubReg_f64_reg imm:$lane))),
3129 (DSubReg_f64_other_reg imm:$lane))>;
3130 def : Pat<(v2f64 (NEONvduplane (v2f64 QPR:$src), imm:$lane)),
3131 (INSERT_SUBREG QPR:$src,
3132 (f64 (EXTRACT_SUBREG QPR:$src,
3133 (DSubReg_f64_reg imm:$lane))),
3134 (DSubReg_f64_other_reg imm:$lane))>;
3136 // VMOVN : Vector Narrowing Move
3137 defm VMOVN : N2VNInt_HSD<0b11,0b11,0b10,0b00100,0,0, IIC_VMOVD,
3138 "vmovn", "i", int_arm_neon_vmovn>;
3139 // VQMOVN : Vector Saturating Narrowing Move
3140 defm VQMOVNs : N2VNInt_HSD<0b11,0b11,0b10,0b00101,0,0, IIC_VQUNAiD,
3141 "vqmovn", "s", int_arm_neon_vqmovns>;
3142 defm VQMOVNu : N2VNInt_HSD<0b11,0b11,0b10,0b00101,1,0, IIC_VQUNAiD,
3143 "vqmovn", "u", int_arm_neon_vqmovnu>;
3144 defm VQMOVNsu : N2VNInt_HSD<0b11,0b11,0b10,0b00100,1,0, IIC_VQUNAiD,
3145 "vqmovun", "s", int_arm_neon_vqmovnsu>;
3146 // VMOVL : Vector Lengthening Move
3147 defm VMOVLs : N2VLInt_QHS<0b01,0b10100,0,1, "vmovl", "s",
3148 int_arm_neon_vmovls>;
3149 defm VMOVLu : N2VLInt_QHS<0b11,0b10100,0,1, "vmovl", "u",
3150 int_arm_neon_vmovlu>;
3152 // Vector Conversions.
3154 // VCVT : Vector Convert Between Floating-Point and Integers
3155 def VCVTf2sd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
3156 v2i32, v2f32, fp_to_sint>;
3157 def VCVTf2ud : N2VD<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
3158 v2i32, v2f32, fp_to_uint>;
3159 def VCVTs2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
3160 v2f32, v2i32, sint_to_fp>;
3161 def VCVTu2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
3162 v2f32, v2i32, uint_to_fp>;
3164 def VCVTf2sq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
3165 v4i32, v4f32, fp_to_sint>;
3166 def VCVTf2uq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
3167 v4i32, v4f32, fp_to_uint>;
3168 def VCVTs2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
3169 v4f32, v4i32, sint_to_fp>;
3170 def VCVTu2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
3171 v4f32, v4i32, uint_to_fp>;
3173 // VCVT : Vector Convert Between Floating-Point and Fixed-Point.
3174 def VCVTf2xsd : N2VCvtD<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
3175 v2i32, v2f32, int_arm_neon_vcvtfp2fxs>;
3176 def VCVTf2xud : N2VCvtD<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
3177 v2i32, v2f32, int_arm_neon_vcvtfp2fxu>;
3178 def VCVTxs2fd : N2VCvtD<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
3179 v2f32, v2i32, int_arm_neon_vcvtfxs2fp>;
3180 def VCVTxu2fd : N2VCvtD<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
3181 v2f32, v2i32, int_arm_neon_vcvtfxu2fp>;
3183 def VCVTf2xsq : N2VCvtQ<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
3184 v4i32, v4f32, int_arm_neon_vcvtfp2fxs>;
3185 def VCVTf2xuq : N2VCvtQ<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
3186 v4i32, v4f32, int_arm_neon_vcvtfp2fxu>;
3187 def VCVTxs2fq : N2VCvtQ<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
3188 v4f32, v4i32, int_arm_neon_vcvtfxs2fp>;
3189 def VCVTxu2fq : N2VCvtQ<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
3190 v4f32, v4i32, int_arm_neon_vcvtfxu2fp>;
3194 // VREV64 : Vector Reverse elements within 64-bit doublewords
3196 class VREV64D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3197 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 0, 0, (outs DPR:$dst),
3198 (ins DPR:$src), IIC_VMOVD,
3199 OpcodeStr, Dt, "$dst, $src", "",
3200 [(set DPR:$dst, (Ty (NEONvrev64 (Ty DPR:$src))))]>;
3201 class VREV64Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3202 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 1, 0, (outs QPR:$dst),
3203 (ins QPR:$src), IIC_VMOVD,
3204 OpcodeStr, Dt, "$dst, $src", "",
3205 [(set QPR:$dst, (Ty (NEONvrev64 (Ty QPR:$src))))]>;
3207 def VREV64d8 : VREV64D<0b00, "vrev64", "8", v8i8>;
3208 def VREV64d16 : VREV64D<0b01, "vrev64", "16", v4i16>;
3209 def VREV64d32 : VREV64D<0b10, "vrev64", "32", v2i32>;
3210 def VREV64df : VREV64D<0b10, "vrev64", "32", v2f32>;
3212 def VREV64q8 : VREV64Q<0b00, "vrev64", "8", v16i8>;
3213 def VREV64q16 : VREV64Q<0b01, "vrev64", "16", v8i16>;
3214 def VREV64q32 : VREV64Q<0b10, "vrev64", "32", v4i32>;
3215 def VREV64qf : VREV64Q<0b10, "vrev64", "32", v4f32>;
3217 // VREV32 : Vector Reverse elements within 32-bit words
3219 class VREV32D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3220 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 0, 0, (outs DPR:$dst),
3221 (ins DPR:$src), IIC_VMOVD,
3222 OpcodeStr, Dt, "$dst, $src", "",
3223 [(set DPR:$dst, (Ty (NEONvrev32 (Ty DPR:$src))))]>;
3224 class VREV32Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3225 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 1, 0, (outs QPR:$dst),
3226 (ins QPR:$src), IIC_VMOVD,
3227 OpcodeStr, Dt, "$dst, $src", "",
3228 [(set QPR:$dst, (Ty (NEONvrev32 (Ty QPR:$src))))]>;
3230 def VREV32d8 : VREV32D<0b00, "vrev32", "8", v8i8>;
3231 def VREV32d16 : VREV32D<0b01, "vrev32", "16", v4i16>;
3233 def VREV32q8 : VREV32Q<0b00, "vrev32", "8", v16i8>;
3234 def VREV32q16 : VREV32Q<0b01, "vrev32", "16", v8i16>;
3236 // VREV16 : Vector Reverse elements within 16-bit halfwords
3238 class VREV16D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3239 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 0, 0, (outs DPR:$dst),
3240 (ins DPR:$src), IIC_VMOVD,
3241 OpcodeStr, Dt, "$dst, $src", "",
3242 [(set DPR:$dst, (Ty (NEONvrev16 (Ty DPR:$src))))]>;
3243 class VREV16Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3244 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 1, 0, (outs QPR:$dst),
3245 (ins QPR:$src), IIC_VMOVD,
3246 OpcodeStr, Dt, "$dst, $src", "",
3247 [(set QPR:$dst, (Ty (NEONvrev16 (Ty QPR:$src))))]>;
3249 def VREV16d8 : VREV16D<0b00, "vrev16", "8", v8i8>;
3250 def VREV16q8 : VREV16Q<0b00, "vrev16", "8", v16i8>;
3252 // Other Vector Shuffles.
3254 // VEXT : Vector Extract
3256 class VEXTd<string OpcodeStr, string Dt, ValueType Ty>
3257 : N3V<0,1,0b11,{?,?,?,?},0,0, (outs DPR:$dst),
3258 (ins DPR:$lhs, DPR:$rhs, i32imm:$index), NVExtFrm,
3259 IIC_VEXTD, OpcodeStr, Dt, "$dst, $lhs, $rhs, $index", "",
3260 [(set DPR:$dst, (Ty (NEONvext (Ty DPR:$lhs),
3261 (Ty DPR:$rhs), imm:$index)))]>;
3263 class VEXTq<string OpcodeStr, string Dt, ValueType Ty>
3264 : N3V<0,1,0b11,{?,?,?,?},1,0, (outs QPR:$dst),
3265 (ins QPR:$lhs, QPR:$rhs, i32imm:$index), NVExtFrm,
3266 IIC_VEXTQ, OpcodeStr, Dt, "$dst, $lhs, $rhs, $index", "",
3267 [(set QPR:$dst, (Ty (NEONvext (Ty QPR:$lhs),
3268 (Ty QPR:$rhs), imm:$index)))]>;
3270 def VEXTd8 : VEXTd<"vext", "8", v8i8>;
3271 def VEXTd16 : VEXTd<"vext", "16", v4i16>;
3272 def VEXTd32 : VEXTd<"vext", "32", v2i32>;
3273 def VEXTdf : VEXTd<"vext", "32", v2f32>;
3275 def VEXTq8 : VEXTq<"vext", "8", v16i8>;
3276 def VEXTq16 : VEXTq<"vext", "16", v8i16>;
3277 def VEXTq32 : VEXTq<"vext", "32", v4i32>;
3278 def VEXTqf : VEXTq<"vext", "32", v4f32>;
3280 // VTRN : Vector Transpose
3282 def VTRNd8 : N2VDShuffle<0b00, 0b00001, "vtrn", "8">;
3283 def VTRNd16 : N2VDShuffle<0b01, 0b00001, "vtrn", "16">;
3284 def VTRNd32 : N2VDShuffle<0b10, 0b00001, "vtrn", "32">;
3286 def VTRNq8 : N2VQShuffle<0b00, 0b00001, IIC_VPERMQ, "vtrn", "8">;
3287 def VTRNq16 : N2VQShuffle<0b01, 0b00001, IIC_VPERMQ, "vtrn", "16">;
3288 def VTRNq32 : N2VQShuffle<0b10, 0b00001, IIC_VPERMQ, "vtrn", "32">;
3290 // VUZP : Vector Unzip (Deinterleave)
3292 def VUZPd8 : N2VDShuffle<0b00, 0b00010, "vuzp", "8">;
3293 def VUZPd16 : N2VDShuffle<0b01, 0b00010, "vuzp", "16">;
3294 def VUZPd32 : N2VDShuffle<0b10, 0b00010, "vuzp", "32">;
3296 def VUZPq8 : N2VQShuffle<0b00, 0b00010, IIC_VPERMQ3, "vuzp", "8">;
3297 def VUZPq16 : N2VQShuffle<0b01, 0b00010, IIC_VPERMQ3, "vuzp", "16">;
3298 def VUZPq32 : N2VQShuffle<0b10, 0b00010, IIC_VPERMQ3, "vuzp", "32">;
3300 // VZIP : Vector Zip (Interleave)
3302 def VZIPd8 : N2VDShuffle<0b00, 0b00011, "vzip", "8">;
3303 def VZIPd16 : N2VDShuffle<0b01, 0b00011, "vzip", "16">;
3304 def VZIPd32 : N2VDShuffle<0b10, 0b00011, "vzip", "32">;
3306 def VZIPq8 : N2VQShuffle<0b00, 0b00011, IIC_VPERMQ3, "vzip", "8">;
3307 def VZIPq16 : N2VQShuffle<0b01, 0b00011, IIC_VPERMQ3, "vzip", "16">;
3308 def VZIPq32 : N2VQShuffle<0b10, 0b00011, IIC_VPERMQ3, "vzip", "32">;
3310 // Vector Table Lookup and Table Extension.
3312 // VTBL : Vector Table Lookup
3314 : N3V<1,1,0b11,0b1000,0,0, (outs DPR:$dst),
3315 (ins DPR:$tbl1, DPR:$src), NVTBLFrm, IIC_VTB1,
3316 "vtbl", "8", "$dst, \\{$tbl1\\}, $src", "",
3317 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl1 DPR:$tbl1, DPR:$src)))]>;
3318 let hasExtraSrcRegAllocReq = 1 in {
3320 : N3V<1,1,0b11,0b1001,0,0, (outs DPR:$dst),
3321 (ins DPR:$tbl1, DPR:$tbl2, DPR:$src), NVTBLFrm, IIC_VTB2,
3322 "vtbl", "8", "$dst, \\{$tbl1, $tbl2\\}, $src", "",
3323 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl2
3324 DPR:$tbl1, DPR:$tbl2, DPR:$src)))]>;
3326 : N3V<1,1,0b11,0b1010,0,0, (outs DPR:$dst),
3327 (ins DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$src), NVTBLFrm, IIC_VTB3,
3328 "vtbl", "8", "$dst, \\{$tbl1, $tbl2, $tbl3\\}, $src", "",
3329 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl3
3330 DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$src)))]>;
3332 : N3V<1,1,0b11,0b1011,0,0, (outs DPR:$dst),
3333 (ins DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$src),
3335 "vtbl", "8", "$dst, \\{$tbl1, $tbl2, $tbl3, $tbl4\\}, $src", "",
3336 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl4 DPR:$tbl1, DPR:$tbl2,
3337 DPR:$tbl3, DPR:$tbl4, DPR:$src)))]>;
3338 } // hasExtraSrcRegAllocReq = 1
3340 // VTBX : Vector Table Extension
3342 : N3V<1,1,0b11,0b1000,1,0, (outs DPR:$dst),
3343 (ins DPR:$orig, DPR:$tbl1, DPR:$src), NVTBLFrm, IIC_VTBX1,
3344 "vtbx", "8", "$dst, \\{$tbl1\\}, $src", "$orig = $dst",
3345 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx1
3346 DPR:$orig, DPR:$tbl1, DPR:$src)))]>;
3347 let hasExtraSrcRegAllocReq = 1 in {
3349 : N3V<1,1,0b11,0b1001,1,0, (outs DPR:$dst),
3350 (ins DPR:$orig, DPR:$tbl1, DPR:$tbl2, DPR:$src), NVTBLFrm, IIC_VTBX2,
3351 "vtbx", "8", "$dst, \\{$tbl1, $tbl2\\}, $src", "$orig = $dst",
3352 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx2
3353 DPR:$orig, DPR:$tbl1, DPR:$tbl2, DPR:$src)))]>;
3355 : N3V<1,1,0b11,0b1010,1,0, (outs DPR:$dst),
3356 (ins DPR:$orig, DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$src),
3357 NVTBLFrm, IIC_VTBX3,
3358 "vtbx", "8", "$dst, \\{$tbl1, $tbl2, $tbl3\\}, $src", "$orig = $dst",
3359 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx3 DPR:$orig, DPR:$tbl1,
3360 DPR:$tbl2, DPR:$tbl3, DPR:$src)))]>;
3362 : N3V<1,1,0b11,0b1011,1,0, (outs DPR:$dst), (ins DPR:$orig, DPR:$tbl1,
3363 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$src), NVTBLFrm, IIC_VTBX4,
3364 "vtbx", "8", "$dst, \\{$tbl1, $tbl2, $tbl3, $tbl4\\}, $src",
3366 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx4 DPR:$orig, DPR:$tbl1,
3367 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$src)))]>;
3368 } // hasExtraSrcRegAllocReq = 1
3370 //===----------------------------------------------------------------------===//
3371 // NEON instructions for single-precision FP math
3372 //===----------------------------------------------------------------------===//
3374 class N2VSPat<SDNode OpNode, ValueType ResTy, ValueType OpTy, NeonI Inst>
3375 : NEONFPPat<(ResTy (OpNode SPR:$a)),
3376 (EXTRACT_SUBREG (OpTy (Inst (INSERT_SUBREG (OpTy (IMPLICIT_DEF)),
3380 class N3VSPat<SDNode OpNode, NeonI Inst>
3381 : NEONFPPat<(f32 (OpNode SPR:$a, SPR:$b)),
3382 (EXTRACT_SUBREG (v2f32
3383 (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3385 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3389 class N3VSMulOpPat<SDNode MulNode, SDNode OpNode, NeonI Inst>
3390 : NEONFPPat<(f32 (OpNode SPR:$acc, (f32 (MulNode SPR:$a, SPR:$b)))),
3391 (EXTRACT_SUBREG (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3393 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3395 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3399 // These need separate instructions because they must use DPR_VFP2 register
3400 // class which have SPR sub-registers.
3402 // Vector Add Operations used for single-precision FP
3403 let neverHasSideEffects = 1 in
3404 def VADDfd_sfp : N3VS<0,0,0b00,0b1101,0, "vadd", "f32", v2f32, v2f32, fadd, 1>;
3405 def : N3VSPat<fadd, VADDfd_sfp>;
3407 // Vector Sub Operations used for single-precision FP
3408 let neverHasSideEffects = 1 in
3409 def VSUBfd_sfp : N3VS<0,0,0b10,0b1101,0, "vsub", "f32", v2f32, v2f32, fsub, 0>;
3410 def : N3VSPat<fsub, VSUBfd_sfp>;
3412 // Vector Multiply Operations used for single-precision FP
3413 let neverHasSideEffects = 1 in
3414 def VMULfd_sfp : N3VS<1,0,0b00,0b1101,1, "vmul", "f32", v2f32, v2f32, fmul, 1>;
3415 def : N3VSPat<fmul, VMULfd_sfp>;
3417 // Vector Multiply-Accumulate/Subtract used for single-precision FP
3418 // vml[as].f32 can cause 4-8 cycle stalls in following ASIMD instructions, so
3419 // we want to avoid them for now. e.g., alternating vmla/vadd instructions.
3421 //let neverHasSideEffects = 1 in
3422 //def VMLAfd_sfp : N3VSMulOp<0,0,0b00,0b1101,1, IIC_VMACD, "vmla", "f32",
3423 // v2f32, fmul, fadd>;
3424 //def : N3VSMulOpPat<fmul, fadd, VMLAfd_sfp>;
3426 //let neverHasSideEffects = 1 in
3427 //def VMLSfd_sfp : N3VSMulOp<0,0,0b10,0b1101,1, IIC_VMACD, "vmls", "f32",
3428 // v2f32, fmul, fsub>;
3429 //def : N3VSMulOpPat<fmul, fsub, VMLSfd_sfp>;
3431 // Vector Absolute used for single-precision FP
3432 let neverHasSideEffects = 1 in
3433 def VABSfd_sfp : N2V<0b11, 0b11, 0b10, 0b01, 0b01110, 0, 0,
3434 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), IIC_VUNAD,
3435 "vabs", "f32", "$dst, $src", "", []>;
3436 def : N2VSPat<fabs, f32, v2f32, VABSfd_sfp>;
3438 // Vector Negate used for single-precision FP
3439 let neverHasSideEffects = 1 in
3440 def VNEGfd_sfp : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
3441 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), IIC_VUNAD,
3442 "vneg", "f32", "$dst, $src", "", []>;
3443 def : N2VSPat<fneg, f32, v2f32, VNEGfd_sfp>;
3445 // Vector Maximum used for single-precision FP
3446 let neverHasSideEffects = 1 in
3447 def VMAXfd_sfp : N3V<0, 0, 0b00, 0b1111, 0, 0, (outs DPR_VFP2:$dst),
3448 (ins DPR_VFP2:$src1, DPR_VFP2:$src2), N3RegFrm, IIC_VBIND,
3449 "vmax", "f32", "$dst, $src1, $src2", "", []>;
3450 def : N3VSPat<NEONfmax, VMAXfd_sfp>;
3452 // Vector Minimum used for single-precision FP
3453 let neverHasSideEffects = 1 in
3454 def VMINfd_sfp : N3V<0, 0, 0b00, 0b1111, 0, 0, (outs DPR_VFP2:$dst),
3455 (ins DPR_VFP2:$src1, DPR_VFP2:$src2), N3RegFrm, IIC_VBIND,
3456 "vmin", "f32", "$dst, $src1, $src2", "", []>;
3457 def : N3VSPat<NEONfmin, VMINfd_sfp>;
3459 // Vector Convert between single-precision FP and integer
3460 let neverHasSideEffects = 1 in
3461 def VCVTf2sd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
3462 v2i32, v2f32, fp_to_sint>;
3463 def : N2VSPat<arm_ftosi, f32, v2f32, VCVTf2sd_sfp>;
3465 let neverHasSideEffects = 1 in
3466 def VCVTf2ud_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
3467 v2i32, v2f32, fp_to_uint>;
3468 def : N2VSPat<arm_ftoui, f32, v2f32, VCVTf2ud_sfp>;
3470 let neverHasSideEffects = 1 in
3471 def VCVTs2fd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
3472 v2f32, v2i32, sint_to_fp>;
3473 def : N2VSPat<arm_sitof, f32, v2i32, VCVTs2fd_sfp>;
3475 let neverHasSideEffects = 1 in
3476 def VCVTu2fd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
3477 v2f32, v2i32, uint_to_fp>;
3478 def : N2VSPat<arm_uitof, f32, v2i32, VCVTu2fd_sfp>;
3480 //===----------------------------------------------------------------------===//
3481 // Non-Instruction Patterns
3482 //===----------------------------------------------------------------------===//
3485 def : Pat<(v1i64 (bitconvert (v2i32 DPR:$src))), (v1i64 DPR:$src)>;
3486 def : Pat<(v1i64 (bitconvert (v4i16 DPR:$src))), (v1i64 DPR:$src)>;
3487 def : Pat<(v1i64 (bitconvert (v8i8 DPR:$src))), (v1i64 DPR:$src)>;
3488 def : Pat<(v1i64 (bitconvert (f64 DPR:$src))), (v1i64 DPR:$src)>;
3489 def : Pat<(v1i64 (bitconvert (v2f32 DPR:$src))), (v1i64 DPR:$src)>;
3490 def : Pat<(v2i32 (bitconvert (v1i64 DPR:$src))), (v2i32 DPR:$src)>;
3491 def : Pat<(v2i32 (bitconvert (v4i16 DPR:$src))), (v2i32 DPR:$src)>;
3492 def : Pat<(v2i32 (bitconvert (v8i8 DPR:$src))), (v2i32 DPR:$src)>;
3493 def : Pat<(v2i32 (bitconvert (f64 DPR:$src))), (v2i32 DPR:$src)>;
3494 def : Pat<(v2i32 (bitconvert (v2f32 DPR:$src))), (v2i32 DPR:$src)>;
3495 def : Pat<(v4i16 (bitconvert (v1i64 DPR:$src))), (v4i16 DPR:$src)>;
3496 def : Pat<(v4i16 (bitconvert (v2i32 DPR:$src))), (v4i16 DPR:$src)>;
3497 def : Pat<(v4i16 (bitconvert (v8i8 DPR:$src))), (v4i16 DPR:$src)>;
3498 def : Pat<(v4i16 (bitconvert (f64 DPR:$src))), (v4i16 DPR:$src)>;
3499 def : Pat<(v4i16 (bitconvert (v2f32 DPR:$src))), (v4i16 DPR:$src)>;
3500 def : Pat<(v8i8 (bitconvert (v1i64 DPR:$src))), (v8i8 DPR:$src)>;
3501 def : Pat<(v8i8 (bitconvert (v2i32 DPR:$src))), (v8i8 DPR:$src)>;
3502 def : Pat<(v8i8 (bitconvert (v4i16 DPR:$src))), (v8i8 DPR:$src)>;
3503 def : Pat<(v8i8 (bitconvert (f64 DPR:$src))), (v8i8 DPR:$src)>;
3504 def : Pat<(v8i8 (bitconvert (v2f32 DPR:$src))), (v8i8 DPR:$src)>;
3505 def : Pat<(f64 (bitconvert (v1i64 DPR:$src))), (f64 DPR:$src)>;
3506 def : Pat<(f64 (bitconvert (v2i32 DPR:$src))), (f64 DPR:$src)>;
3507 def : Pat<(f64 (bitconvert (v4i16 DPR:$src))), (f64 DPR:$src)>;
3508 def : Pat<(f64 (bitconvert (v8i8 DPR:$src))), (f64 DPR:$src)>;
3509 def : Pat<(f64 (bitconvert (v2f32 DPR:$src))), (f64 DPR:$src)>;
3510 def : Pat<(v2f32 (bitconvert (f64 DPR:$src))), (v2f32 DPR:$src)>;
3511 def : Pat<(v2f32 (bitconvert (v1i64 DPR:$src))), (v2f32 DPR:$src)>;
3512 def : Pat<(v2f32 (bitconvert (v2i32 DPR:$src))), (v2f32 DPR:$src)>;
3513 def : Pat<(v2f32 (bitconvert (v4i16 DPR:$src))), (v2f32 DPR:$src)>;
3514 def : Pat<(v2f32 (bitconvert (v8i8 DPR:$src))), (v2f32 DPR:$src)>;
3516 def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (v2i64 QPR:$src)>;
3517 def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (v2i64 QPR:$src)>;
3518 def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (v2i64 QPR:$src)>;
3519 def : Pat<(v2i64 (bitconvert (v2f64 QPR:$src))), (v2i64 QPR:$src)>;
3520 def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (v2i64 QPR:$src)>;
3521 def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (v4i32 QPR:$src)>;
3522 def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (v4i32 QPR:$src)>;
3523 def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (v4i32 QPR:$src)>;
3524 def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (v4i32 QPR:$src)>;
3525 def : Pat<(v4i32 (bitconvert (v4f32 QPR:$src))), (v4i32 QPR:$src)>;
3526 def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (v8i16 QPR:$src)>;
3527 def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (v8i16 QPR:$src)>;
3528 def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (v8i16 QPR:$src)>;
3529 def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (v8i16 QPR:$src)>;
3530 def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (v8i16 QPR:$src)>;
3531 def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (v16i8 QPR:$src)>;
3532 def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (v16i8 QPR:$src)>;
3533 def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (v16i8 QPR:$src)>;
3534 def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (v16i8 QPR:$src)>;
3535 def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (v16i8 QPR:$src)>;
3536 def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (v4f32 QPR:$src)>;
3537 def : Pat<(v4f32 (bitconvert (v4i32 QPR:$src))), (v4f32 QPR:$src)>;
3538 def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (v4f32 QPR:$src)>;
3539 def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (v4f32 QPR:$src)>;
3540 def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (v4f32 QPR:$src)>;
3541 def : Pat<(v2f64 (bitconvert (v2i64 QPR:$src))), (v2f64 QPR:$src)>;
3542 def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (v2f64 QPR:$src)>;
3543 def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 QPR:$src)>;
3544 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 QPR:$src)>;
3545 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 QPR:$src)>;