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);
816 // Extract S sub-registers of Q/D registers.
817 def SSubReg_f32_reg : SDNodeXForm<imm, [{
818 assert(ARM::ssub_3 == ARM::ssub_0+3 && "Unexpected subreg numbering");
819 return CurDAG->getTargetConstant(ARM::ssub_0 + N->getZExtValue(), MVT::i32);
822 // Translate lane numbers from Q registers to D subregs.
823 def SubReg_i8_lane : SDNodeXForm<imm, [{
824 return CurDAG->getTargetConstant(N->getZExtValue() & 7, MVT::i32);
826 def SubReg_i16_lane : SDNodeXForm<imm, [{
827 return CurDAG->getTargetConstant(N->getZExtValue() & 3, MVT::i32);
829 def SubReg_i32_lane : SDNodeXForm<imm, [{
830 return CurDAG->getTargetConstant(N->getZExtValue() & 1, MVT::i32);
833 //===----------------------------------------------------------------------===//
834 // Instruction Classes
835 //===----------------------------------------------------------------------===//
837 // Basic 2-register operations: single-, double- and quad-register.
838 class N2VS<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
839 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
840 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
841 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
842 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src),
843 IIC_VUNAD, OpcodeStr, Dt, "$dst, $src", "", []>;
844 class N2VD<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
845 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
846 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
847 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
848 (ins DPR:$src), IIC_VUNAD, OpcodeStr, Dt,"$dst, $src", "",
849 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src))))]>;
850 class N2VQ<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
851 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
852 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
853 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
854 (ins QPR:$src), IIC_VUNAQ, OpcodeStr, Dt,"$dst, $src", "",
855 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src))))]>;
857 // Basic 2-register intrinsics, both double- and quad-register.
858 class N2VDInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
859 bits<2> op17_16, bits<5> op11_7, bit op4,
860 InstrItinClass itin, string OpcodeStr, string Dt,
861 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
862 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
863 (ins DPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
864 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
865 class N2VQInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
866 bits<2> op17_16, bits<5> op11_7, bit op4,
867 InstrItinClass itin, string OpcodeStr, string Dt,
868 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
869 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
870 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
871 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
873 // Narrow 2-register intrinsics.
874 class N2VNInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
875 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
876 InstrItinClass itin, string OpcodeStr, string Dt,
877 ValueType TyD, ValueType TyQ, Intrinsic IntOp>
878 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$dst),
879 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
880 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src))))]>;
882 // Long 2-register intrinsics (currently only used for VMOVL).
883 class N2VLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
884 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
885 InstrItinClass itin, string OpcodeStr, string Dt,
886 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
887 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$dst),
888 (ins DPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
889 [(set QPR:$dst, (TyQ (IntOp (TyD DPR:$src))))]>;
891 // 2-register shuffles (VTRN/VZIP/VUZP), both double- and quad-register.
892 class N2VDShuffle<bits<2> op19_18, bits<5> op11_7, string OpcodeStr, string Dt>
893 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 0, 0, (outs DPR:$dst1, DPR:$dst2),
894 (ins DPR:$src1, DPR:$src2), IIC_VPERMD,
895 OpcodeStr, Dt, "$dst1, $dst2",
896 "$src1 = $dst1, $src2 = $dst2", []>;
897 class N2VQShuffle<bits<2> op19_18, bits<5> op11_7,
898 InstrItinClass itin, string OpcodeStr, string Dt>
899 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 1, 0, (outs QPR:$dst1, QPR:$dst2),
900 (ins QPR:$src1, QPR:$src2), itin, OpcodeStr, Dt, "$dst1, $dst2",
901 "$src1 = $dst1, $src2 = $dst2", []>;
903 // Basic 3-register operations: single-, double- and quad-register.
904 class N3VS<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
905 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
906 SDNode OpNode, bit Commutable>
907 : N3V<op24, op23, op21_20, op11_8, 0, op4,
908 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src1, DPR_VFP2:$src2), N3RegFrm,
909 IIC_VBIND, OpcodeStr, Dt, "$dst, $src1, $src2", "", []> {
910 let isCommutable = Commutable;
913 class N3VD<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
914 InstrItinClass itin, string OpcodeStr, string Dt,
915 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
916 : N3V<op24, op23, op21_20, op11_8, 0, op4,
917 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
918 OpcodeStr, Dt, "$dst, $src1, $src2", "",
919 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src1), (OpTy DPR:$src2))))]> {
920 let isCommutable = Commutable;
922 // Same as N3VD but no data type.
923 class N3VDX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
924 InstrItinClass itin, string OpcodeStr,
925 ValueType ResTy, ValueType OpTy,
926 SDNode OpNode, bit Commutable>
927 : N3VX<op24, op23, op21_20, op11_8, 0, op4,
928 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
929 OpcodeStr, "$dst, $src1, $src2", "",
930 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src1), (OpTy DPR:$src2))))]>{
931 let isCommutable = Commutable;
934 class N3VDSL<bits<2> op21_20, bits<4> op11_8,
935 InstrItinClass itin, string OpcodeStr, string Dt,
936 ValueType Ty, SDNode ShOp>
937 : N3V<0, 1, op21_20, op11_8, 1, 0,
938 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
939 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
941 (Ty (ShOp (Ty DPR:$src1),
942 (Ty (NEONvduplane (Ty DPR_VFP2:$src2),imm:$lane)))))]> {
943 let isCommutable = 0;
945 class N3VDSL16<bits<2> op21_20, bits<4> op11_8,
946 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
947 : N3V<0, 1, op21_20, op11_8, 1, 0,
948 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
949 NVMulSLFrm, IIC_VMULi16D, OpcodeStr, Dt,"$dst, $src1, $src2[$lane]","",
951 (Ty (ShOp (Ty DPR:$src1),
952 (Ty (NEONvduplane (Ty DPR_8:$src2), imm:$lane)))))]> {
953 let isCommutable = 0;
956 class N3VQ<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
957 InstrItinClass itin, string OpcodeStr, string Dt,
958 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
959 : N3V<op24, op23, op21_20, op11_8, 1, op4,
960 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), N3RegFrm, itin,
961 OpcodeStr, Dt, "$dst, $src1, $src2", "",
962 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src1), (OpTy QPR:$src2))))]> {
963 let isCommutable = Commutable;
965 class N3VQX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
966 InstrItinClass itin, string OpcodeStr,
967 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
968 : N3VX<op24, op23, op21_20, op11_8, 1, op4,
969 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), N3RegFrm, itin,
970 OpcodeStr, "$dst, $src1, $src2", "",
971 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src1), (OpTy QPR:$src2))))]>{
972 let isCommutable = Commutable;
974 class N3VQSL<bits<2> op21_20, bits<4> op11_8,
975 InstrItinClass itin, string OpcodeStr, string Dt,
976 ValueType ResTy, ValueType OpTy, SDNode ShOp>
977 : N3V<1, 1, op21_20, op11_8, 1, 0,
978 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
979 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
980 [(set (ResTy QPR:$dst),
981 (ResTy (ShOp (ResTy QPR:$src1),
982 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
984 let isCommutable = 0;
986 class N3VQSL16<bits<2> op21_20, bits<4> op11_8, string OpcodeStr, string Dt,
987 ValueType ResTy, ValueType OpTy, SDNode ShOp>
988 : N3V<1, 1, op21_20, op11_8, 1, 0,
989 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
990 NVMulSLFrm, IIC_VMULi16Q, OpcodeStr, Dt,"$dst, $src1, $src2[$lane]","",
991 [(set (ResTy QPR:$dst),
992 (ResTy (ShOp (ResTy QPR:$src1),
993 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
995 let isCommutable = 0;
998 // Basic 3-register intrinsics, both double- and quad-register.
999 class N3VDInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1000 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1001 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
1002 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1003 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), f, itin,
1004 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1005 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src1), (OpTy DPR:$src2))))]> {
1006 let isCommutable = Commutable;
1008 class N3VDIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1009 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
1010 : N3V<0, 1, op21_20, op11_8, 1, 0,
1011 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1012 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1013 [(set (Ty DPR:$dst),
1014 (Ty (IntOp (Ty DPR:$src1),
1015 (Ty (NEONvduplane (Ty DPR_VFP2:$src2),
1017 let isCommutable = 0;
1019 class N3VDIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1020 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
1021 : N3V<0, 1, op21_20, op11_8, 1, 0,
1022 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1023 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1024 [(set (Ty DPR:$dst),
1025 (Ty (IntOp (Ty DPR:$src1),
1026 (Ty (NEONvduplane (Ty DPR_8:$src2), imm:$lane)))))]> {
1027 let isCommutable = 0;
1030 class N3VQInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1031 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1032 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
1033 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1034 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), f, itin,
1035 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1036 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src1), (OpTy QPR:$src2))))]> {
1037 let isCommutable = Commutable;
1039 class N3VQIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1040 string OpcodeStr, string Dt,
1041 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1042 : N3V<1, 1, op21_20, op11_8, 1, 0,
1043 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1044 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1045 [(set (ResTy QPR:$dst),
1046 (ResTy (IntOp (ResTy QPR:$src1),
1047 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1049 let isCommutable = 0;
1051 class N3VQIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1052 string OpcodeStr, string Dt,
1053 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1054 : N3V<1, 1, op21_20, op11_8, 1, 0,
1055 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1056 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1057 [(set (ResTy QPR:$dst),
1058 (ResTy (IntOp (ResTy QPR:$src1),
1059 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
1061 let isCommutable = 0;
1064 // Multiply-Add/Sub operations: single-, double- and quad-register.
1065 class N3VSMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1066 InstrItinClass itin, string OpcodeStr, string Dt,
1067 ValueType Ty, SDNode MulOp, SDNode OpNode>
1068 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1069 (outs DPR_VFP2:$dst),
1070 (ins DPR_VFP2:$src1, DPR_VFP2:$src2, DPR_VFP2:$src3), N3RegFrm, itin,
1071 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst", []>;
1073 class N3VDMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1074 InstrItinClass itin, string OpcodeStr, string Dt,
1075 ValueType Ty, SDNode MulOp, SDNode OpNode>
1076 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1077 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3), N3RegFrm, itin,
1078 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1079 [(set DPR:$dst, (Ty (OpNode DPR:$src1,
1080 (Ty (MulOp DPR:$src2, DPR:$src3)))))]>;
1081 class N3VDMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1082 string OpcodeStr, string Dt,
1083 ValueType Ty, SDNode MulOp, SDNode ShOp>
1084 : N3V<0, 1, op21_20, op11_8, 1, 0,
1086 (ins DPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1088 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1089 [(set (Ty DPR:$dst),
1090 (Ty (ShOp (Ty DPR:$src1),
1091 (Ty (MulOp DPR:$src2,
1092 (Ty (NEONvduplane (Ty DPR_VFP2:$src3),
1094 class N3VDMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1095 string OpcodeStr, string Dt,
1096 ValueType Ty, SDNode MulOp, SDNode ShOp>
1097 : N3V<0, 1, op21_20, op11_8, 1, 0,
1099 (ins DPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane),
1101 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1102 [(set (Ty DPR:$dst),
1103 (Ty (ShOp (Ty DPR:$src1),
1104 (Ty (MulOp DPR:$src2,
1105 (Ty (NEONvduplane (Ty DPR_8:$src3),
1108 class N3VQMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1109 InstrItinClass itin, string OpcodeStr, string Dt, ValueType Ty,
1110 SDNode MulOp, SDNode OpNode>
1111 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1112 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3), N3RegFrm, itin,
1113 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1114 [(set QPR:$dst, (Ty (OpNode QPR:$src1,
1115 (Ty (MulOp QPR:$src2, QPR:$src3)))))]>;
1116 class N3VQMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1117 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1118 SDNode MulOp, SDNode ShOp>
1119 : N3V<1, 1, op21_20, op11_8, 1, 0,
1121 (ins QPR:$src1, QPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1123 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1124 [(set (ResTy QPR:$dst),
1125 (ResTy (ShOp (ResTy QPR:$src1),
1126 (ResTy (MulOp QPR:$src2,
1127 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src3),
1129 class N3VQMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1130 string OpcodeStr, string Dt,
1131 ValueType ResTy, ValueType OpTy,
1132 SDNode MulOp, SDNode ShOp>
1133 : N3V<1, 1, op21_20, op11_8, 1, 0,
1135 (ins QPR:$src1, QPR:$src2, DPR_8:$src3, nohash_imm:$lane),
1137 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1138 [(set (ResTy QPR:$dst),
1139 (ResTy (ShOp (ResTy QPR:$src1),
1140 (ResTy (MulOp QPR:$src2,
1141 (ResTy (NEONvduplane (OpTy DPR_8:$src3),
1144 // Neon 3-argument intrinsics, both double- and quad-register.
1145 // The destination register is also used as the first source operand register.
1146 class N3VDInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1147 InstrItinClass itin, string OpcodeStr, string Dt,
1148 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1149 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1150 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3), N3RegFrm, itin,
1151 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1152 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src1),
1153 (OpTy DPR:$src2), (OpTy DPR:$src3))))]>;
1154 class N3VQInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1155 InstrItinClass itin, string OpcodeStr, string Dt,
1156 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1157 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1158 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3), N3RegFrm, itin,
1159 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1160 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src1),
1161 (OpTy QPR:$src2), (OpTy QPR:$src3))))]>;
1163 // Neon Long 3-argument intrinsic. The destination register is
1164 // a quad-register and is also used as the first source operand register.
1165 class N3VLInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1166 InstrItinClass itin, string OpcodeStr, string Dt,
1167 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
1168 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1169 (outs QPR:$dst), (ins QPR:$src1, DPR:$src2, DPR:$src3), N3RegFrm, itin,
1170 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1172 (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$src2), (TyD DPR:$src3))))]>;
1173 class N3VLInt3SL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1174 string OpcodeStr, string Dt,
1175 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1176 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1178 (ins QPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1180 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1181 [(set (ResTy QPR:$dst),
1182 (ResTy (IntOp (ResTy QPR:$src1),
1184 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src3),
1186 class N3VLInt3SL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1187 InstrItinClass itin, string OpcodeStr, string Dt,
1188 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1189 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1191 (ins QPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane),
1193 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1194 [(set (ResTy QPR:$dst),
1195 (ResTy (IntOp (ResTy QPR:$src1),
1197 (OpTy (NEONvduplane (OpTy DPR_8:$src3),
1200 // Narrowing 3-register intrinsics.
1201 class N3VNInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1202 string OpcodeStr, string Dt, ValueType TyD, ValueType TyQ,
1203 Intrinsic IntOp, bit Commutable>
1204 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1205 (outs DPR:$dst), (ins QPR:$src1, QPR:$src2), N3RegFrm, IIC_VBINi4D,
1206 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1207 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src1), (TyQ QPR:$src2))))]> {
1208 let isCommutable = Commutable;
1211 // Long 3-register intrinsics.
1212 class N3VLInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1213 InstrItinClass itin, string OpcodeStr, string Dt,
1214 ValueType TyQ, ValueType TyD, Intrinsic IntOp, bit Commutable>
1215 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1216 (outs QPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
1217 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1218 [(set QPR:$dst, (TyQ (IntOp (TyD DPR:$src1), (TyD DPR:$src2))))]> {
1219 let isCommutable = Commutable;
1221 class N3VLIntSL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1222 string OpcodeStr, string Dt,
1223 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1224 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1225 (outs QPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1226 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1227 [(set (ResTy QPR:$dst),
1228 (ResTy (IntOp (OpTy DPR:$src1),
1229 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1231 class N3VLIntSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1232 InstrItinClass itin, string OpcodeStr, string Dt,
1233 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1234 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1235 (outs QPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1236 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1237 [(set (ResTy QPR:$dst),
1238 (ResTy (IntOp (OpTy DPR:$src1),
1239 (OpTy (NEONvduplane (OpTy DPR_8:$src2),
1242 // Wide 3-register intrinsics.
1243 class N3VWInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1244 string OpcodeStr, string Dt, ValueType TyQ, ValueType TyD,
1245 Intrinsic IntOp, bit Commutable>
1246 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1247 (outs QPR:$dst), (ins QPR:$src1, DPR:$src2), N3RegFrm, IIC_VSUBiD,
1248 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1249 [(set QPR:$dst, (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$src2))))]> {
1250 let isCommutable = Commutable;
1253 // Pairwise long 2-register intrinsics, both double- and quad-register.
1254 class N2VDPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1255 bits<2> op17_16, bits<5> op11_7, bit op4,
1256 string OpcodeStr, string Dt,
1257 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1258 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
1259 (ins DPR:$src), IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
1260 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
1261 class N2VQPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1262 bits<2> op17_16, bits<5> op11_7, bit op4,
1263 string OpcodeStr, string Dt,
1264 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1265 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
1266 (ins QPR:$src), IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
1267 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
1269 // Pairwise long 2-register accumulate intrinsics,
1270 // both double- and quad-register.
1271 // The destination register is also used as the first source operand register.
1272 class N2VDPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1273 bits<2> op17_16, bits<5> op11_7, bit op4,
1274 string OpcodeStr, string Dt,
1275 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1276 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
1277 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), IIC_VPALiD,
1278 OpcodeStr, Dt, "$dst, $src2", "$src1 = $dst",
1279 [(set DPR:$dst, (ResTy (IntOp (ResTy DPR:$src1), (OpTy DPR:$src2))))]>;
1280 class N2VQPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1281 bits<2> op17_16, bits<5> op11_7, bit op4,
1282 string OpcodeStr, string Dt,
1283 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1284 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4,
1285 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), IIC_VPALiQ,
1286 OpcodeStr, Dt, "$dst, $src2", "$src1 = $dst",
1287 [(set QPR:$dst, (ResTy (IntOp (ResTy QPR:$src1), (OpTy QPR:$src2))))]>;
1289 // Shift by immediate,
1290 // both double- and quad-register.
1291 class N2VDSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1292 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1293 ValueType Ty, SDNode OpNode>
1294 : N2VImm<op24, op23, op11_8, op7, 0, op4,
1295 (outs DPR:$dst), (ins DPR:$src, i32imm:$SIMM), f, itin,
1296 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1297 [(set DPR:$dst, (Ty (OpNode (Ty DPR:$src), (i32 imm:$SIMM))))]>;
1298 class N2VQSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1299 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1300 ValueType Ty, SDNode OpNode>
1301 : N2VImm<op24, op23, op11_8, op7, 1, op4,
1302 (outs QPR:$dst), (ins QPR:$src, i32imm:$SIMM), f, itin,
1303 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1304 [(set QPR:$dst, (Ty (OpNode (Ty QPR:$src), (i32 imm:$SIMM))))]>;
1306 // Long shift by immediate.
1307 class N2VLSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
1308 string OpcodeStr, string Dt,
1309 ValueType ResTy, ValueType OpTy, SDNode OpNode>
1310 : N2VImm<op24, op23, op11_8, op7, op6, op4,
1311 (outs QPR:$dst), (ins DPR:$src, i32imm:$SIMM), N2RegVShLFrm,
1312 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1313 [(set QPR:$dst, (ResTy (OpNode (OpTy DPR:$src),
1314 (i32 imm:$SIMM))))]>;
1316 // Narrow shift by immediate.
1317 class N2VNSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
1318 InstrItinClass itin, string OpcodeStr, string Dt,
1319 ValueType ResTy, ValueType OpTy, SDNode OpNode>
1320 : N2VImm<op24, op23, op11_8, op7, op6, op4,
1321 (outs DPR:$dst), (ins QPR:$src, i32imm:$SIMM), N2RegVShRFrm, itin,
1322 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1323 [(set DPR:$dst, (ResTy (OpNode (OpTy QPR:$src),
1324 (i32 imm:$SIMM))))]>;
1326 // Shift right by immediate and accumulate,
1327 // both double- and quad-register.
1328 class N2VDShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1329 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1330 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$dst),
1331 (ins DPR:$src1, DPR:$src2, i32imm:$SIMM), N2RegVShRFrm, IIC_VPALiD,
1332 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1333 [(set DPR:$dst, (Ty (add DPR:$src1,
1334 (Ty (ShOp DPR:$src2, (i32 imm:$SIMM))))))]>;
1335 class N2VQShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1336 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1337 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$dst),
1338 (ins QPR:$src1, QPR:$src2, i32imm:$SIMM), N2RegVShRFrm, IIC_VPALiD,
1339 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1340 [(set QPR:$dst, (Ty (add QPR:$src1,
1341 (Ty (ShOp QPR:$src2, (i32 imm:$SIMM))))))]>;
1343 // Shift by immediate and insert,
1344 // both double- and quad-register.
1345 class N2VDShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1346 Format f, string OpcodeStr, string Dt, ValueType Ty,SDNode ShOp>
1347 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$dst),
1348 (ins DPR:$src1, DPR:$src2, i32imm:$SIMM), f, IIC_VSHLiD,
1349 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1350 [(set DPR:$dst, (Ty (ShOp DPR:$src1, DPR:$src2, (i32 imm:$SIMM))))]>;
1351 class N2VQShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1352 Format f, string OpcodeStr, string Dt, ValueType Ty,SDNode ShOp>
1353 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$dst),
1354 (ins QPR:$src1, QPR:$src2, i32imm:$SIMM), f, IIC_VSHLiQ,
1355 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1356 [(set QPR:$dst, (Ty (ShOp QPR:$src1, QPR:$src2, (i32 imm:$SIMM))))]>;
1358 // Convert, with fractional bits immediate,
1359 // both double- and quad-register.
1360 class N2VCvtD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1361 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1363 : N2VImm<op24, op23, op11_8, op7, 0, op4,
1364 (outs DPR:$dst), (ins DPR:$src, i32imm:$SIMM), NVCVTFrm,
1365 IIC_VUNAD, OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1366 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src), (i32 imm:$SIMM))))]>;
1367 class N2VCvtQ<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1368 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1370 : N2VImm<op24, op23, op11_8, op7, 1, op4,
1371 (outs QPR:$dst), (ins QPR:$src, i32imm:$SIMM), NVCVTFrm,
1372 IIC_VUNAQ, OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1373 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src), (i32 imm:$SIMM))))]>;
1375 //===----------------------------------------------------------------------===//
1377 //===----------------------------------------------------------------------===//
1379 // Abbreviations used in multiclass suffixes:
1380 // Q = quarter int (8 bit) elements
1381 // H = half int (16 bit) elements
1382 // S = single int (32 bit) elements
1383 // D = double int (64 bit) elements
1385 // Neon 2-register vector operations -- for disassembly only.
1387 // First with only element sizes of 8, 16 and 32 bits:
1388 multiclass N2V_QHS_cmp<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1389 bits<5> op11_7, bit op4, string opc, string Dt,
1391 // 64-bit vector types.
1392 def v8i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 0, op4,
1393 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1394 opc, !strconcat(Dt, "8"), asm, "", []>;
1395 def v4i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 0, op4,
1396 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1397 opc, !strconcat(Dt, "16"), asm, "", []>;
1398 def v2i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
1399 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1400 opc, !strconcat(Dt, "32"), asm, "", []>;
1401 def v2f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
1402 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1403 opc, "f32", asm, "", []> {
1404 let Inst{10} = 1; // overwrite F = 1
1407 // 128-bit vector types.
1408 def v16i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 1, op4,
1409 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1410 opc, !strconcat(Dt, "8"), asm, "", []>;
1411 def v8i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 1, op4,
1412 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1413 opc, !strconcat(Dt, "16"), asm, "", []>;
1414 def v4i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
1415 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1416 opc, !strconcat(Dt, "32"), asm, "", []>;
1417 def v4f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
1418 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1419 opc, "f32", asm, "", []> {
1420 let Inst{10} = 1; // overwrite F = 1
1424 // Neon 3-register vector operations.
1426 // First with only element sizes of 8, 16 and 32 bits:
1427 multiclass N3V_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1428 InstrItinClass itinD16, InstrItinClass itinD32,
1429 InstrItinClass itinQ16, InstrItinClass itinQ32,
1430 string OpcodeStr, string Dt,
1431 SDNode OpNode, bit Commutable = 0> {
1432 // 64-bit vector types.
1433 def v8i8 : N3VD<op24, op23, 0b00, op11_8, op4, itinD16,
1434 OpcodeStr, !strconcat(Dt, "8"),
1435 v8i8, v8i8, OpNode, Commutable>;
1436 def v4i16 : N3VD<op24, op23, 0b01, op11_8, op4, itinD16,
1437 OpcodeStr, !strconcat(Dt, "16"),
1438 v4i16, v4i16, OpNode, Commutable>;
1439 def v2i32 : N3VD<op24, op23, 0b10, op11_8, op4, itinD32,
1440 OpcodeStr, !strconcat(Dt, "32"),
1441 v2i32, v2i32, OpNode, Commutable>;
1443 // 128-bit vector types.
1444 def v16i8 : N3VQ<op24, op23, 0b00, op11_8, op4, itinQ16,
1445 OpcodeStr, !strconcat(Dt, "8"),
1446 v16i8, v16i8, OpNode, Commutable>;
1447 def v8i16 : N3VQ<op24, op23, 0b01, op11_8, op4, itinQ16,
1448 OpcodeStr, !strconcat(Dt, "16"),
1449 v8i16, v8i16, OpNode, Commutable>;
1450 def v4i32 : N3VQ<op24, op23, 0b10, op11_8, op4, itinQ32,
1451 OpcodeStr, !strconcat(Dt, "32"),
1452 v4i32, v4i32, OpNode, Commutable>;
1455 multiclass N3VSL_HS<bits<4> op11_8, string OpcodeStr, string Dt, SDNode ShOp> {
1456 def v4i16 : N3VDSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
1458 def v2i32 : N3VDSL<0b10, op11_8, IIC_VMULi32D, OpcodeStr, !strconcat(Dt,"32"),
1460 def v8i16 : N3VQSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
1461 v8i16, v4i16, ShOp>;
1462 def v4i32 : N3VQSL<0b10, op11_8, IIC_VMULi32Q, OpcodeStr, !strconcat(Dt,"32"),
1463 v4i32, v2i32, ShOp>;
1466 // ....then also with element size 64 bits:
1467 multiclass N3V_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1468 InstrItinClass itinD, InstrItinClass itinQ,
1469 string OpcodeStr, string Dt,
1470 SDNode OpNode, bit Commutable = 0>
1471 : N3V_QHS<op24, op23, op11_8, op4, itinD, itinD, itinQ, itinQ,
1472 OpcodeStr, Dt, OpNode, Commutable> {
1473 def v1i64 : N3VD<op24, op23, 0b11, op11_8, op4, itinD,
1474 OpcodeStr, !strconcat(Dt, "64"),
1475 v1i64, v1i64, OpNode, Commutable>;
1476 def v2i64 : N3VQ<op24, op23, 0b11, op11_8, op4, itinQ,
1477 OpcodeStr, !strconcat(Dt, "64"),
1478 v2i64, v2i64, OpNode, Commutable>;
1482 // Neon Narrowing 2-register vector intrinsics,
1483 // source operand element sizes of 16, 32 and 64 bits:
1484 multiclass N2VNInt_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1485 bits<5> op11_7, bit op6, bit op4,
1486 InstrItinClass itin, string OpcodeStr, string Dt,
1488 def v8i8 : N2VNInt<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
1489 itin, OpcodeStr, !strconcat(Dt, "16"),
1490 v8i8, v8i16, IntOp>;
1491 def v4i16 : N2VNInt<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
1492 itin, OpcodeStr, !strconcat(Dt, "32"),
1493 v4i16, v4i32, IntOp>;
1494 def v2i32 : N2VNInt<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
1495 itin, OpcodeStr, !strconcat(Dt, "64"),
1496 v2i32, v2i64, IntOp>;
1500 // Neon Lengthening 2-register vector intrinsic (currently specific to VMOVL).
1501 // source operand element sizes of 16, 32 and 64 bits:
1502 multiclass N2VLInt_QHS<bits<2> op24_23, bits<5> op11_7, bit op6, bit op4,
1503 string OpcodeStr, string Dt, Intrinsic IntOp> {
1504 def v8i16 : N2VLInt<op24_23, 0b00, 0b10, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
1505 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
1506 def v4i32 : N2VLInt<op24_23, 0b01, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
1507 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
1508 def v2i64 : N2VLInt<op24_23, 0b10, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
1509 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1513 // Neon 3-register vector intrinsics.
1515 // First with only element sizes of 16 and 32 bits:
1516 multiclass N3VInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
1517 InstrItinClass itinD16, InstrItinClass itinD32,
1518 InstrItinClass itinQ16, InstrItinClass itinQ32,
1519 string OpcodeStr, string Dt,
1520 Intrinsic IntOp, bit Commutable = 0> {
1521 // 64-bit vector types.
1522 def v4i16 : N3VDInt<op24, op23, 0b01, op11_8, op4, f, itinD16,
1523 OpcodeStr, !strconcat(Dt, "16"),
1524 v4i16, v4i16, IntOp, Commutable>;
1525 def v2i32 : N3VDInt<op24, op23, 0b10, op11_8, op4, f, itinD32,
1526 OpcodeStr, !strconcat(Dt, "32"),
1527 v2i32, v2i32, IntOp, Commutable>;
1529 // 128-bit vector types.
1530 def v8i16 : N3VQInt<op24, op23, 0b01, op11_8, op4, f, itinQ16,
1531 OpcodeStr, !strconcat(Dt, "16"),
1532 v8i16, v8i16, IntOp, Commutable>;
1533 def v4i32 : N3VQInt<op24, op23, 0b10, op11_8, op4, f, itinQ32,
1534 OpcodeStr, !strconcat(Dt, "32"),
1535 v4i32, v4i32, IntOp, Commutable>;
1538 multiclass N3VIntSL_HS<bits<4> op11_8,
1539 InstrItinClass itinD16, InstrItinClass itinD32,
1540 InstrItinClass itinQ16, InstrItinClass itinQ32,
1541 string OpcodeStr, string Dt, Intrinsic IntOp> {
1542 def v4i16 : N3VDIntSL16<0b01, op11_8, itinD16,
1543 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp>;
1544 def v2i32 : N3VDIntSL<0b10, op11_8, itinD32,
1545 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp>;
1546 def v8i16 : N3VQIntSL16<0b01, op11_8, itinQ16,
1547 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16, IntOp>;
1548 def v4i32 : N3VQIntSL<0b10, op11_8, itinQ32,
1549 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32, IntOp>;
1552 // ....then also with element size of 8 bits:
1553 multiclass N3VInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
1554 InstrItinClass itinD16, InstrItinClass itinD32,
1555 InstrItinClass itinQ16, InstrItinClass itinQ32,
1556 string OpcodeStr, string Dt,
1557 Intrinsic IntOp, bit Commutable = 0>
1558 : N3VInt_HS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
1559 OpcodeStr, Dt, IntOp, Commutable> {
1560 def v8i8 : N3VDInt<op24, op23, 0b00, op11_8, op4, f, itinD16,
1561 OpcodeStr, !strconcat(Dt, "8"),
1562 v8i8, v8i8, IntOp, Commutable>;
1563 def v16i8 : N3VQInt<op24, op23, 0b00, op11_8, op4, f, itinQ16,
1564 OpcodeStr, !strconcat(Dt, "8"),
1565 v16i8, v16i8, IntOp, Commutable>;
1568 // ....then also with element size of 64 bits:
1569 multiclass N3VInt_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
1570 InstrItinClass itinD16, InstrItinClass itinD32,
1571 InstrItinClass itinQ16, InstrItinClass itinQ32,
1572 string OpcodeStr, string Dt,
1573 Intrinsic IntOp, bit Commutable = 0>
1574 : N3VInt_QHS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
1575 OpcodeStr, Dt, IntOp, Commutable> {
1576 def v1i64 : N3VDInt<op24, op23, 0b11, op11_8, op4, f, itinD32,
1577 OpcodeStr, !strconcat(Dt, "64"),
1578 v1i64, v1i64, IntOp, Commutable>;
1579 def v2i64 : N3VQInt<op24, op23, 0b11, op11_8, op4, f, itinQ32,
1580 OpcodeStr, !strconcat(Dt, "64"),
1581 v2i64, v2i64, IntOp, Commutable>;
1584 // Neon Narrowing 3-register vector intrinsics,
1585 // source operand element sizes of 16, 32 and 64 bits:
1586 multiclass N3VNInt_HSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1587 string OpcodeStr, string Dt,
1588 Intrinsic IntOp, bit Commutable = 0> {
1589 def v8i8 : N3VNInt<op24, op23, 0b00, op11_8, op4,
1590 OpcodeStr, !strconcat(Dt, "16"),
1591 v8i8, v8i16, IntOp, Commutable>;
1592 def v4i16 : N3VNInt<op24, op23, 0b01, op11_8, op4,
1593 OpcodeStr, !strconcat(Dt, "32"),
1594 v4i16, v4i32, IntOp, Commutable>;
1595 def v2i32 : N3VNInt<op24, op23, 0b10, op11_8, op4,
1596 OpcodeStr, !strconcat(Dt, "64"),
1597 v2i32, v2i64, IntOp, Commutable>;
1601 // Neon Long 3-register vector intrinsics.
1603 // First with only element sizes of 16 and 32 bits:
1604 multiclass N3VLInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
1605 InstrItinClass itin16, InstrItinClass itin32,
1606 string OpcodeStr, string Dt,
1607 Intrinsic IntOp, bit Commutable = 0> {
1608 def v4i32 : N3VLInt<op24, op23, 0b01, op11_8, op4, itin16,
1609 OpcodeStr, !strconcat(Dt, "16"),
1610 v4i32, v4i16, IntOp, Commutable>;
1611 def v2i64 : N3VLInt<op24, op23, 0b10, op11_8, op4, itin32,
1612 OpcodeStr, !strconcat(Dt, "32"),
1613 v2i64, v2i32, IntOp, Commutable>;
1616 multiclass N3VLIntSL_HS<bit op24, bits<4> op11_8,
1617 InstrItinClass itin, string OpcodeStr, string Dt,
1619 def v4i16 : N3VLIntSL16<op24, 0b01, op11_8, itin,
1620 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
1621 def v2i32 : N3VLIntSL<op24, 0b10, op11_8, itin,
1622 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1625 // ....then also with element size of 8 bits:
1626 multiclass N3VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1627 InstrItinClass itin16, InstrItinClass itin32,
1628 string OpcodeStr, string Dt,
1629 Intrinsic IntOp, bit Commutable = 0>
1630 : N3VLInt_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt,
1631 IntOp, Commutable> {
1632 def v8i16 : N3VLInt<op24, op23, 0b00, op11_8, op4, itin16,
1633 OpcodeStr, !strconcat(Dt, "8"),
1634 v8i16, v8i8, IntOp, Commutable>;
1638 // Neon Wide 3-register vector intrinsics,
1639 // source operand element sizes of 8, 16 and 32 bits:
1640 multiclass N3VWInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1641 string OpcodeStr, string Dt,
1642 Intrinsic IntOp, bit Commutable = 0> {
1643 def v8i16 : N3VWInt<op24, op23, 0b00, op11_8, op4,
1644 OpcodeStr, !strconcat(Dt, "8"),
1645 v8i16, v8i8, IntOp, Commutable>;
1646 def v4i32 : N3VWInt<op24, op23, 0b01, op11_8, op4,
1647 OpcodeStr, !strconcat(Dt, "16"),
1648 v4i32, v4i16, IntOp, Commutable>;
1649 def v2i64 : N3VWInt<op24, op23, 0b10, op11_8, op4,
1650 OpcodeStr, !strconcat(Dt, "32"),
1651 v2i64, v2i32, IntOp, Commutable>;
1655 // Neon Multiply-Op vector operations,
1656 // element sizes of 8, 16 and 32 bits:
1657 multiclass N3VMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1658 InstrItinClass itinD16, InstrItinClass itinD32,
1659 InstrItinClass itinQ16, InstrItinClass itinQ32,
1660 string OpcodeStr, string Dt, SDNode OpNode> {
1661 // 64-bit vector types.
1662 def v8i8 : N3VDMulOp<op24, op23, 0b00, op11_8, op4, itinD16,
1663 OpcodeStr, !strconcat(Dt, "8"), v8i8, mul, OpNode>;
1664 def v4i16 : N3VDMulOp<op24, op23, 0b01, op11_8, op4, itinD16,
1665 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, OpNode>;
1666 def v2i32 : N3VDMulOp<op24, op23, 0b10, op11_8, op4, itinD32,
1667 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, OpNode>;
1669 // 128-bit vector types.
1670 def v16i8 : N3VQMulOp<op24, op23, 0b00, op11_8, op4, itinQ16,
1671 OpcodeStr, !strconcat(Dt, "8"), v16i8, mul, OpNode>;
1672 def v8i16 : N3VQMulOp<op24, op23, 0b01, op11_8, op4, itinQ16,
1673 OpcodeStr, !strconcat(Dt, "16"), v8i16, mul, OpNode>;
1674 def v4i32 : N3VQMulOp<op24, op23, 0b10, op11_8, op4, itinQ32,
1675 OpcodeStr, !strconcat(Dt, "32"), v4i32, mul, OpNode>;
1678 multiclass N3VMulOpSL_HS<bits<4> op11_8,
1679 InstrItinClass itinD16, InstrItinClass itinD32,
1680 InstrItinClass itinQ16, InstrItinClass itinQ32,
1681 string OpcodeStr, string Dt, SDNode ShOp> {
1682 def v4i16 : N3VDMulOpSL16<0b01, op11_8, itinD16,
1683 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, ShOp>;
1684 def v2i32 : N3VDMulOpSL<0b10, op11_8, itinD32,
1685 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, ShOp>;
1686 def v8i16 : N3VQMulOpSL16<0b01, op11_8, itinQ16,
1687 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16,
1689 def v4i32 : N3VQMulOpSL<0b10, op11_8, itinQ32,
1690 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32,
1694 // Neon 3-argument intrinsics,
1695 // element sizes of 8, 16 and 32 bits:
1696 multiclass N3VInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1697 InstrItinClass itinD, InstrItinClass itinQ,
1698 string OpcodeStr, string Dt, Intrinsic IntOp> {
1699 // 64-bit vector types.
1700 def v8i8 : N3VDInt3<op24, op23, 0b00, op11_8, op4, itinD,
1701 OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
1702 def v4i16 : N3VDInt3<op24, op23, 0b01, op11_8, op4, itinD,
1703 OpcodeStr, !strconcat(Dt, "16"), v4i16, v4i16, IntOp>;
1704 def v2i32 : N3VDInt3<op24, op23, 0b10, op11_8, op4, itinD,
1705 OpcodeStr, !strconcat(Dt, "32"), v2i32, v2i32, IntOp>;
1707 // 128-bit vector types.
1708 def v16i8 : N3VQInt3<op24, op23, 0b00, op11_8, op4, itinQ,
1709 OpcodeStr, !strconcat(Dt, "8"), v16i8, v16i8, IntOp>;
1710 def v8i16 : N3VQInt3<op24, op23, 0b01, op11_8, op4, itinQ,
1711 OpcodeStr, !strconcat(Dt, "16"), v8i16, v8i16, IntOp>;
1712 def v4i32 : N3VQInt3<op24, op23, 0b10, op11_8, op4, itinQ,
1713 OpcodeStr, !strconcat(Dt, "32"), v4i32, v4i32, IntOp>;
1717 // Neon Long 3-argument intrinsics.
1719 // First with only element sizes of 16 and 32 bits:
1720 multiclass N3VLInt3_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
1721 InstrItinClass itin16, InstrItinClass itin32,
1722 string OpcodeStr, string Dt, Intrinsic IntOp> {
1723 def v4i32 : N3VLInt3<op24, op23, 0b01, op11_8, op4, itin16,
1724 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
1725 def v2i64 : N3VLInt3<op24, op23, 0b10, op11_8, op4, itin32,
1726 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1729 multiclass N3VLInt3SL_HS<bit op24, bits<4> op11_8,
1730 string OpcodeStr, string Dt, Intrinsic IntOp> {
1731 def v4i16 : N3VLInt3SL16<op24, 0b01, op11_8, IIC_VMACi16D,
1732 OpcodeStr, !strconcat(Dt,"16"), v4i32, v4i16, IntOp>;
1733 def v2i32 : N3VLInt3SL<op24, 0b10, op11_8, IIC_VMACi32D,
1734 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1737 // ....then also with element size of 8 bits:
1738 multiclass N3VLInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1739 InstrItinClass itin16, InstrItinClass itin32,
1740 string OpcodeStr, string Dt, Intrinsic IntOp>
1741 : N3VLInt3_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt, IntOp> {
1742 def v8i16 : N3VLInt3<op24, op23, 0b00, op11_8, op4, itin16,
1743 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
1747 // Neon 2-register vector intrinsics,
1748 // element sizes of 8, 16 and 32 bits:
1749 multiclass N2VInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1750 bits<5> op11_7, bit op4,
1751 InstrItinClass itinD, InstrItinClass itinQ,
1752 string OpcodeStr, string Dt, Intrinsic IntOp> {
1753 // 64-bit vector types.
1754 def v8i8 : N2VDInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1755 itinD, OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
1756 def v4i16 : N2VDInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1757 itinD, OpcodeStr, !strconcat(Dt, "16"),v4i16,v4i16,IntOp>;
1758 def v2i32 : N2VDInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1759 itinD, OpcodeStr, !strconcat(Dt, "32"),v2i32,v2i32,IntOp>;
1761 // 128-bit vector types.
1762 def v16i8 : N2VQInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1763 itinQ, OpcodeStr, !strconcat(Dt, "8"), v16i8,v16i8,IntOp>;
1764 def v8i16 : N2VQInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1765 itinQ, OpcodeStr, !strconcat(Dt, "16"),v8i16,v8i16,IntOp>;
1766 def v4i32 : N2VQInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1767 itinQ, OpcodeStr, !strconcat(Dt, "32"),v4i32,v4i32,IntOp>;
1771 // Neon Pairwise long 2-register intrinsics,
1772 // element sizes of 8, 16 and 32 bits:
1773 multiclass N2VPLInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1774 bits<5> op11_7, bit op4,
1775 string OpcodeStr, string Dt, Intrinsic IntOp> {
1776 // 64-bit vector types.
1777 def v8i8 : N2VDPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1778 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
1779 def v4i16 : N2VDPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1780 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
1781 def v2i32 : N2VDPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1782 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
1784 // 128-bit vector types.
1785 def v16i8 : N2VQPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1786 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
1787 def v8i16 : N2VQPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1788 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
1789 def v4i32 : N2VQPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1790 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
1794 // Neon Pairwise long 2-register accumulate intrinsics,
1795 // element sizes of 8, 16 and 32 bits:
1796 multiclass N2VPLInt2_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1797 bits<5> op11_7, bit op4,
1798 string OpcodeStr, string Dt, Intrinsic IntOp> {
1799 // 64-bit vector types.
1800 def v8i8 : N2VDPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1801 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
1802 def v4i16 : N2VDPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1803 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
1804 def v2i32 : N2VDPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1805 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
1807 // 128-bit vector types.
1808 def v16i8 : N2VQPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1809 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
1810 def v8i16 : N2VQPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1811 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
1812 def v4i32 : N2VQPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1813 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
1817 // Neon 2-register vector shift by immediate,
1818 // with f of either N2RegVShLFrm or N2RegVShRFrm
1819 // element sizes of 8, 16, 32 and 64 bits:
1820 multiclass N2VSh_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1821 InstrItinClass itin, string OpcodeStr, string Dt,
1822 SDNode OpNode, Format f> {
1823 // 64-bit vector types.
1824 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
1825 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
1826 let Inst{21-19} = 0b001; // imm6 = 001xxx
1828 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
1829 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
1830 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1832 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
1833 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
1834 let Inst{21} = 0b1; // imm6 = 1xxxxx
1836 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, f, itin,
1837 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
1840 // 128-bit vector types.
1841 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
1842 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
1843 let Inst{21-19} = 0b001; // imm6 = 001xxx
1845 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
1846 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
1847 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1849 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
1850 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
1851 let Inst{21} = 0b1; // imm6 = 1xxxxx
1853 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, f, itin,
1854 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
1858 // Neon Shift-Accumulate vector operations,
1859 // element sizes of 8, 16, 32 and 64 bits:
1860 multiclass N2VShAdd_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1861 string OpcodeStr, string Dt, SDNode ShOp> {
1862 // 64-bit vector types.
1863 def v8i8 : N2VDShAdd<op24, op23, op11_8, 0, op4,
1864 OpcodeStr, !strconcat(Dt, "8"), v8i8, ShOp> {
1865 let Inst{21-19} = 0b001; // imm6 = 001xxx
1867 def v4i16 : N2VDShAdd<op24, op23, op11_8, 0, op4,
1868 OpcodeStr, !strconcat(Dt, "16"), v4i16, ShOp> {
1869 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1871 def v2i32 : N2VDShAdd<op24, op23, op11_8, 0, op4,
1872 OpcodeStr, !strconcat(Dt, "32"), v2i32, ShOp> {
1873 let Inst{21} = 0b1; // imm6 = 1xxxxx
1875 def v1i64 : N2VDShAdd<op24, op23, op11_8, 1, op4,
1876 OpcodeStr, !strconcat(Dt, "64"), v1i64, ShOp>;
1879 // 128-bit vector types.
1880 def v16i8 : N2VQShAdd<op24, op23, op11_8, 0, op4,
1881 OpcodeStr, !strconcat(Dt, "8"), v16i8, ShOp> {
1882 let Inst{21-19} = 0b001; // imm6 = 001xxx
1884 def v8i16 : N2VQShAdd<op24, op23, op11_8, 0, op4,
1885 OpcodeStr, !strconcat(Dt, "16"), v8i16, ShOp> {
1886 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1888 def v4i32 : N2VQShAdd<op24, op23, op11_8, 0, op4,
1889 OpcodeStr, !strconcat(Dt, "32"), v4i32, ShOp> {
1890 let Inst{21} = 0b1; // imm6 = 1xxxxx
1892 def v2i64 : N2VQShAdd<op24, op23, op11_8, 1, op4,
1893 OpcodeStr, !strconcat(Dt, "64"), v2i64, ShOp>;
1898 // Neon Shift-Insert vector operations,
1899 // with f of either N2RegVShLFrm or N2RegVShRFrm
1900 // element sizes of 8, 16, 32 and 64 bits:
1901 multiclass N2VShIns_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1902 string OpcodeStr, SDNode ShOp,
1904 // 64-bit vector types.
1905 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4,
1906 f, OpcodeStr, "8", v8i8, ShOp> {
1907 let Inst{21-19} = 0b001; // imm6 = 001xxx
1909 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4,
1910 f, OpcodeStr, "16", v4i16, ShOp> {
1911 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1913 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4,
1914 f, OpcodeStr, "32", v2i32, ShOp> {
1915 let Inst{21} = 0b1; // imm6 = 1xxxxx
1917 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4,
1918 f, OpcodeStr, "64", v1i64, ShOp>;
1921 // 128-bit vector types.
1922 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4,
1923 f, OpcodeStr, "8", v16i8, ShOp> {
1924 let Inst{21-19} = 0b001; // imm6 = 001xxx
1926 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4,
1927 f, OpcodeStr, "16", v8i16, ShOp> {
1928 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1930 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4,
1931 f, OpcodeStr, "32", v4i32, ShOp> {
1932 let Inst{21} = 0b1; // imm6 = 1xxxxx
1934 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4,
1935 f, OpcodeStr, "64", v2i64, ShOp>;
1939 // Neon Shift Long operations,
1940 // element sizes of 8, 16, 32 bits:
1941 multiclass N2VLSh_QHS<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
1942 bit op4, string OpcodeStr, string Dt, SDNode OpNode> {
1943 def v8i16 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
1944 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode> {
1945 let Inst{21-19} = 0b001; // imm6 = 001xxx
1947 def v4i32 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
1948 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode> {
1949 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1951 def v2i64 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
1952 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode> {
1953 let Inst{21} = 0b1; // imm6 = 1xxxxx
1957 // Neon Shift Narrow operations,
1958 // element sizes of 16, 32, 64 bits:
1959 multiclass N2VNSh_HSD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
1960 bit op4, InstrItinClass itin, string OpcodeStr, string Dt,
1962 def v8i8 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
1963 OpcodeStr, !strconcat(Dt, "16"), v8i8, v8i16, OpNode> {
1964 let Inst{21-19} = 0b001; // imm6 = 001xxx
1966 def v4i16 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
1967 OpcodeStr, !strconcat(Dt, "32"), v4i16, v4i32, OpNode> {
1968 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1970 def v2i32 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
1971 OpcodeStr, !strconcat(Dt, "64"), v2i32, v2i64, OpNode> {
1972 let Inst{21} = 0b1; // imm6 = 1xxxxx
1976 //===----------------------------------------------------------------------===//
1977 // Instruction Definitions.
1978 //===----------------------------------------------------------------------===//
1980 // Vector Add Operations.
1982 // VADD : Vector Add (integer and floating-point)
1983 defm VADD : N3V_QHSD<0, 0, 0b1000, 0, IIC_VBINiD, IIC_VBINiQ, "vadd", "i",
1985 def VADDfd : N3VD<0, 0, 0b00, 0b1101, 0, IIC_VBIND, "vadd", "f32",
1986 v2f32, v2f32, fadd, 1>;
1987 def VADDfq : N3VQ<0, 0, 0b00, 0b1101, 0, IIC_VBINQ, "vadd", "f32",
1988 v4f32, v4f32, fadd, 1>;
1989 // VADDL : Vector Add Long (Q = D + D)
1990 defm VADDLs : N3VLInt_QHS<0,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
1991 "vaddl", "s", int_arm_neon_vaddls, 1>;
1992 defm VADDLu : N3VLInt_QHS<1,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
1993 "vaddl", "u", int_arm_neon_vaddlu, 1>;
1994 // VADDW : Vector Add Wide (Q = Q + D)
1995 defm VADDWs : N3VWInt_QHS<0,1,0b0001,0, "vaddw", "s", int_arm_neon_vaddws, 0>;
1996 defm VADDWu : N3VWInt_QHS<1,1,0b0001,0, "vaddw", "u", int_arm_neon_vaddwu, 0>;
1997 // VHADD : Vector Halving Add
1998 defm VHADDs : N3VInt_QHS<0, 0, 0b0000, 0, N3RegFrm,
1999 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2000 "vhadd", "s", int_arm_neon_vhadds, 1>;
2001 defm VHADDu : N3VInt_QHS<1, 0, 0b0000, 0, N3RegFrm,
2002 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2003 "vhadd", "u", int_arm_neon_vhaddu, 1>;
2004 // VRHADD : Vector Rounding Halving Add
2005 defm VRHADDs : N3VInt_QHS<0, 0, 0b0001, 0, N3RegFrm,
2006 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2007 "vrhadd", "s", int_arm_neon_vrhadds, 1>;
2008 defm VRHADDu : N3VInt_QHS<1, 0, 0b0001, 0, N3RegFrm,
2009 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2010 "vrhadd", "u", int_arm_neon_vrhaddu, 1>;
2011 // VQADD : Vector Saturating Add
2012 defm VQADDs : N3VInt_QHSD<0, 0, 0b0000, 1, N3RegFrm,
2013 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2014 "vqadd", "s", int_arm_neon_vqadds, 1>;
2015 defm VQADDu : N3VInt_QHSD<1, 0, 0b0000, 1, N3RegFrm,
2016 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2017 "vqadd", "u", int_arm_neon_vqaddu, 1>;
2018 // VADDHN : Vector Add and Narrow Returning High Half (D = Q + Q)
2019 defm VADDHN : N3VNInt_HSD<0,1,0b0100,0, "vaddhn", "i",
2020 int_arm_neon_vaddhn, 1>;
2021 // VRADDHN : Vector Rounding Add and Narrow Returning High Half (D = Q + Q)
2022 defm VRADDHN : N3VNInt_HSD<1,1,0b0100,0, "vraddhn", "i",
2023 int_arm_neon_vraddhn, 1>;
2025 // Vector Multiply Operations.
2027 // VMUL : Vector Multiply (integer, polynomial and floating-point)
2028 defm VMUL : N3V_QHS<0, 0, 0b1001, 1, IIC_VMULi16D, IIC_VMULi32D,
2029 IIC_VMULi16Q, IIC_VMULi32Q, "vmul", "i", mul, 1>;
2030 def VMULpd : N3VDInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16D, "vmul",
2031 "p8", v8i8, v8i8, int_arm_neon_vmulp, 1>;
2032 def VMULpq : N3VQInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16Q, "vmul",
2033 "p8", v16i8, v16i8, int_arm_neon_vmulp, 1>;
2034 def VMULfd : N3VD<1, 0, 0b00, 0b1101, 1, IIC_VBIND, "vmul", "f32",
2035 v2f32, v2f32, fmul, 1>;
2036 def VMULfq : N3VQ<1, 0, 0b00, 0b1101, 1, IIC_VBINQ, "vmul", "f32",
2037 v4f32, v4f32, fmul, 1>;
2038 defm VMULsl : N3VSL_HS<0b1000, "vmul", "i", mul>;
2039 def VMULslfd : N3VDSL<0b10, 0b1001, IIC_VBIND, "vmul", "f32", v2f32, fmul>;
2040 def VMULslfq : N3VQSL<0b10, 0b1001, IIC_VBINQ, "vmul", "f32", v4f32,
2043 def : Pat<(v8i16 (mul (v8i16 QPR:$src1),
2044 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
2045 (v8i16 (VMULslv8i16 (v8i16 QPR:$src1),
2046 (v4i16 (EXTRACT_SUBREG QPR:$src2,
2047 (DSubReg_i16_reg imm:$lane))),
2048 (SubReg_i16_lane imm:$lane)))>;
2049 def : Pat<(v4i32 (mul (v4i32 QPR:$src1),
2050 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
2051 (v4i32 (VMULslv4i32 (v4i32 QPR:$src1),
2052 (v2i32 (EXTRACT_SUBREG QPR:$src2,
2053 (DSubReg_i32_reg imm:$lane))),
2054 (SubReg_i32_lane imm:$lane)))>;
2055 def : Pat<(v4f32 (fmul (v4f32 QPR:$src1),
2056 (v4f32 (NEONvduplane (v4f32 QPR:$src2), imm:$lane)))),
2057 (v4f32 (VMULslfq (v4f32 QPR:$src1),
2058 (v2f32 (EXTRACT_SUBREG QPR:$src2,
2059 (DSubReg_i32_reg imm:$lane))),
2060 (SubReg_i32_lane imm:$lane)))>;
2062 // VQDMULH : Vector Saturating Doubling Multiply Returning High Half
2063 defm VQDMULH : N3VInt_HS<0, 0, 0b1011, 0, N3RegFrm, IIC_VMULi16D, IIC_VMULi32D,
2064 IIC_VMULi16Q, IIC_VMULi32Q,
2065 "vqdmulh", "s", int_arm_neon_vqdmulh, 1>;
2066 defm VQDMULHsl: N3VIntSL_HS<0b1100, IIC_VMULi16D, IIC_VMULi32D,
2067 IIC_VMULi16Q, IIC_VMULi32Q,
2068 "vqdmulh", "s", int_arm_neon_vqdmulh>;
2069 def : Pat<(v8i16 (int_arm_neon_vqdmulh (v8i16 QPR:$src1),
2070 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
2072 (v8i16 (VQDMULHslv8i16 (v8i16 QPR:$src1),
2073 (v4i16 (EXTRACT_SUBREG QPR:$src2,
2074 (DSubReg_i16_reg imm:$lane))),
2075 (SubReg_i16_lane imm:$lane)))>;
2076 def : Pat<(v4i32 (int_arm_neon_vqdmulh (v4i32 QPR:$src1),
2077 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
2079 (v4i32 (VQDMULHslv4i32 (v4i32 QPR:$src1),
2080 (v2i32 (EXTRACT_SUBREG QPR:$src2,
2081 (DSubReg_i32_reg imm:$lane))),
2082 (SubReg_i32_lane imm:$lane)))>;
2084 // VQRDMULH : Vector Rounding Saturating Doubling Multiply Returning High Half
2085 defm VQRDMULH : N3VInt_HS<1, 0, 0b1011, 0, N3RegFrm,
2086 IIC_VMULi16D,IIC_VMULi32D,IIC_VMULi16Q,IIC_VMULi32Q,
2087 "vqrdmulh", "s", int_arm_neon_vqrdmulh, 1>;
2088 defm VQRDMULHsl : N3VIntSL_HS<0b1101, IIC_VMULi16D, IIC_VMULi32D,
2089 IIC_VMULi16Q, IIC_VMULi32Q,
2090 "vqrdmulh", "s", int_arm_neon_vqrdmulh>;
2091 def : Pat<(v8i16 (int_arm_neon_vqrdmulh (v8i16 QPR:$src1),
2092 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
2094 (v8i16 (VQRDMULHslv8i16 (v8i16 QPR:$src1),
2095 (v4i16 (EXTRACT_SUBREG QPR:$src2,
2096 (DSubReg_i16_reg imm:$lane))),
2097 (SubReg_i16_lane imm:$lane)))>;
2098 def : Pat<(v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$src1),
2099 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
2101 (v4i32 (VQRDMULHslv4i32 (v4i32 QPR:$src1),
2102 (v2i32 (EXTRACT_SUBREG QPR:$src2,
2103 (DSubReg_i32_reg imm:$lane))),
2104 (SubReg_i32_lane imm:$lane)))>;
2106 // VMULL : Vector Multiply Long (integer and polynomial) (Q = D * D)
2107 defm VMULLs : N3VLInt_QHS<0,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
2108 "vmull", "s", int_arm_neon_vmulls, 1>;
2109 defm VMULLu : N3VLInt_QHS<1,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
2110 "vmull", "u", int_arm_neon_vmullu, 1>;
2111 def VMULLp : N3VLInt<0, 1, 0b00, 0b1110, 0, IIC_VMULi16D, "vmull", "p8",
2112 v8i16, v8i8, int_arm_neon_vmullp, 1>;
2113 defm VMULLsls : N3VLIntSL_HS<0, 0b1010, IIC_VMULi16D, "vmull", "s",
2114 int_arm_neon_vmulls>;
2115 defm VMULLslu : N3VLIntSL_HS<1, 0b1010, IIC_VMULi16D, "vmull", "u",
2116 int_arm_neon_vmullu>;
2118 // VQDMULL : Vector Saturating Doubling Multiply Long (Q = D * D)
2119 defm VQDMULL : N3VLInt_HS<0,1,0b1101,0, IIC_VMULi16D, IIC_VMULi32D,
2120 "vqdmull", "s", int_arm_neon_vqdmull, 1>;
2121 defm VQDMULLsl: N3VLIntSL_HS<0, 0b1011, IIC_VMULi16D,
2122 "vqdmull", "s", int_arm_neon_vqdmull>;
2124 // Vector Multiply-Accumulate and Multiply-Subtract Operations.
2126 // VMLA : Vector Multiply Accumulate (integer and floating-point)
2127 defm VMLA : N3VMulOp_QHS<0, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
2128 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
2129 def VMLAfd : N3VDMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACD, "vmla", "f32",
2131 def VMLAfq : N3VQMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACQ, "vmla", "f32",
2133 defm VMLAsl : N3VMulOpSL_HS<0b0000, IIC_VMACi16D, IIC_VMACi32D,
2134 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
2135 def VMLAslfd : N3VDMulOpSL<0b10, 0b0001, IIC_VMACD, "vmla", "f32",
2137 def VMLAslfq : N3VQMulOpSL<0b10, 0b0001, IIC_VMACQ, "vmla", "f32",
2138 v4f32, v2f32, fmul, fadd>;
2140 def : Pat<(v8i16 (add (v8i16 QPR:$src1),
2141 (mul (v8i16 QPR:$src2),
2142 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
2143 (v8i16 (VMLAslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
2144 (v4i16 (EXTRACT_SUBREG QPR:$src3,
2145 (DSubReg_i16_reg imm:$lane))),
2146 (SubReg_i16_lane imm:$lane)))>;
2148 def : Pat<(v4i32 (add (v4i32 QPR:$src1),
2149 (mul (v4i32 QPR:$src2),
2150 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
2151 (v4i32 (VMLAslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
2152 (v2i32 (EXTRACT_SUBREG QPR:$src3,
2153 (DSubReg_i32_reg imm:$lane))),
2154 (SubReg_i32_lane imm:$lane)))>;
2156 def : Pat<(v4f32 (fadd (v4f32 QPR:$src1),
2157 (fmul (v4f32 QPR:$src2),
2158 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
2159 (v4f32 (VMLAslfq (v4f32 QPR:$src1),
2161 (v2f32 (EXTRACT_SUBREG QPR:$src3,
2162 (DSubReg_i32_reg imm:$lane))),
2163 (SubReg_i32_lane imm:$lane)))>;
2165 // VMLAL : Vector Multiply Accumulate Long (Q += D * D)
2166 defm VMLALs : N3VLInt3_QHS<0,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
2167 "vmlal", "s", int_arm_neon_vmlals>;
2168 defm VMLALu : N3VLInt3_QHS<1,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
2169 "vmlal", "u", int_arm_neon_vmlalu>;
2171 defm VMLALsls : N3VLInt3SL_HS<0, 0b0010, "vmlal", "s", int_arm_neon_vmlals>;
2172 defm VMLALslu : N3VLInt3SL_HS<1, 0b0010, "vmlal", "u", int_arm_neon_vmlalu>;
2174 // VQDMLAL : Vector Saturating Doubling Multiply Accumulate Long (Q += D * D)
2175 defm VQDMLAL : N3VLInt3_HS<0, 1, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
2176 "vqdmlal", "s", int_arm_neon_vqdmlal>;
2177 defm VQDMLALsl: N3VLInt3SL_HS<0, 0b0011, "vqdmlal", "s", int_arm_neon_vqdmlal>;
2179 // VMLS : Vector Multiply Subtract (integer and floating-point)
2180 defm VMLS : N3VMulOp_QHS<1, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
2181 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
2182 def VMLSfd : N3VDMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACD, "vmls", "f32",
2184 def VMLSfq : N3VQMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACQ, "vmls", "f32",
2186 defm VMLSsl : N3VMulOpSL_HS<0b0100, IIC_VMACi16D, IIC_VMACi32D,
2187 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
2188 def VMLSslfd : N3VDMulOpSL<0b10, 0b0101, IIC_VMACD, "vmls", "f32",
2190 def VMLSslfq : N3VQMulOpSL<0b10, 0b0101, IIC_VMACQ, "vmls", "f32",
2191 v4f32, v2f32, fmul, fsub>;
2193 def : Pat<(v8i16 (sub (v8i16 QPR:$src1),
2194 (mul (v8i16 QPR:$src2),
2195 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
2196 (v8i16 (VMLSslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
2197 (v4i16 (EXTRACT_SUBREG QPR:$src3,
2198 (DSubReg_i16_reg imm:$lane))),
2199 (SubReg_i16_lane imm:$lane)))>;
2201 def : Pat<(v4i32 (sub (v4i32 QPR:$src1),
2202 (mul (v4i32 QPR:$src2),
2203 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
2204 (v4i32 (VMLSslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
2205 (v2i32 (EXTRACT_SUBREG QPR:$src3,
2206 (DSubReg_i32_reg imm:$lane))),
2207 (SubReg_i32_lane imm:$lane)))>;
2209 def : Pat<(v4f32 (fsub (v4f32 QPR:$src1),
2210 (fmul (v4f32 QPR:$src2),
2211 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
2212 (v4f32 (VMLSslfq (v4f32 QPR:$src1), (v4f32 QPR:$src2),
2213 (v2f32 (EXTRACT_SUBREG QPR:$src3,
2214 (DSubReg_i32_reg imm:$lane))),
2215 (SubReg_i32_lane imm:$lane)))>;
2217 // VMLSL : Vector Multiply Subtract Long (Q -= D * D)
2218 defm VMLSLs : N3VLInt3_QHS<0,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
2219 "vmlsl", "s", int_arm_neon_vmlsls>;
2220 defm VMLSLu : N3VLInt3_QHS<1,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
2221 "vmlsl", "u", int_arm_neon_vmlslu>;
2223 defm VMLSLsls : N3VLInt3SL_HS<0, 0b0110, "vmlsl", "s", int_arm_neon_vmlsls>;
2224 defm VMLSLslu : N3VLInt3SL_HS<1, 0b0110, "vmlsl", "u", int_arm_neon_vmlslu>;
2226 // VQDMLSL : Vector Saturating Doubling Multiply Subtract Long (Q -= D * D)
2227 defm VQDMLSL : N3VLInt3_HS<0, 1, 0b1011, 0, IIC_VMACi16D, IIC_VMACi32D,
2228 "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
2229 defm VQDMLSLsl: N3VLInt3SL_HS<0, 0b111, "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
2231 // Vector Subtract Operations.
2233 // VSUB : Vector Subtract (integer and floating-point)
2234 defm VSUB : N3V_QHSD<1, 0, 0b1000, 0, IIC_VSUBiD, IIC_VSUBiQ,
2235 "vsub", "i", sub, 0>;
2236 def VSUBfd : N3VD<0, 0, 0b10, 0b1101, 0, IIC_VBIND, "vsub", "f32",
2237 v2f32, v2f32, fsub, 0>;
2238 def VSUBfq : N3VQ<0, 0, 0b10, 0b1101, 0, IIC_VBINQ, "vsub", "f32",
2239 v4f32, v4f32, fsub, 0>;
2240 // VSUBL : Vector Subtract Long (Q = D - D)
2241 defm VSUBLs : N3VLInt_QHS<0,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
2242 "vsubl", "s", int_arm_neon_vsubls, 1>;
2243 defm VSUBLu : N3VLInt_QHS<1,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
2244 "vsubl", "u", int_arm_neon_vsublu, 1>;
2245 // VSUBW : Vector Subtract Wide (Q = Q - D)
2246 defm VSUBWs : N3VWInt_QHS<0,1,0b0011,0, "vsubw", "s", int_arm_neon_vsubws, 0>;
2247 defm VSUBWu : N3VWInt_QHS<1,1,0b0011,0, "vsubw", "u", int_arm_neon_vsubwu, 0>;
2248 // VHSUB : Vector Halving Subtract
2249 defm VHSUBs : N3VInt_QHS<0, 0, 0b0010, 0, N3RegFrm,
2250 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2251 "vhsub", "s", int_arm_neon_vhsubs, 0>;
2252 defm VHSUBu : N3VInt_QHS<1, 0, 0b0010, 0, N3RegFrm,
2253 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2254 "vhsub", "u", int_arm_neon_vhsubu, 0>;
2255 // VQSUB : Vector Saturing Subtract
2256 defm VQSUBs : N3VInt_QHSD<0, 0, 0b0010, 1, N3RegFrm,
2257 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2258 "vqsub", "s", int_arm_neon_vqsubs, 0>;
2259 defm VQSUBu : N3VInt_QHSD<1, 0, 0b0010, 1, N3RegFrm,
2260 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2261 "vqsub", "u", int_arm_neon_vqsubu, 0>;
2262 // VSUBHN : Vector Subtract and Narrow Returning High Half (D = Q - Q)
2263 defm VSUBHN : N3VNInt_HSD<0,1,0b0110,0, "vsubhn", "i",
2264 int_arm_neon_vsubhn, 0>;
2265 // VRSUBHN : Vector Rounding Subtract and Narrow Returning High Half (D=Q-Q)
2266 defm VRSUBHN : N3VNInt_HSD<1,1,0b0110,0, "vrsubhn", "i",
2267 int_arm_neon_vrsubhn, 0>;
2269 // Vector Comparisons.
2271 // VCEQ : Vector Compare Equal
2272 defm VCEQ : N3V_QHS<1, 0, 0b1000, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
2273 IIC_VSUBi4Q, "vceq", "i", NEONvceq, 1>;
2274 def VCEQfd : N3VD<0,0,0b00,0b1110,0, IIC_VBIND, "vceq", "f32", v2i32, v2f32,
2276 def VCEQfq : N3VQ<0,0,0b00,0b1110,0, IIC_VBINQ, "vceq", "f32", v4i32, v4f32,
2278 // For disassembly only.
2279 defm VCEQz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00010, 0, "vceq", "i",
2282 // VCGE : Vector Compare Greater Than or Equal
2283 defm VCGEs : N3V_QHS<0, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
2284 IIC_VSUBi4Q, "vcge", "s", NEONvcge, 0>;
2285 defm VCGEu : N3V_QHS<1, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
2286 IIC_VSUBi4Q, "vcge", "u", NEONvcgeu, 0>;
2287 def VCGEfd : N3VD<1,0,0b00,0b1110,0, IIC_VBIND, "vcge", "f32", v2i32, v2f32,
2289 def VCGEfq : N3VQ<1,0,0b00,0b1110,0, IIC_VBINQ, "vcge", "f32", v4i32, v4f32,
2291 // For disassembly only.
2292 defm VCGEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00001, 0, "vcge", "s",
2294 // For disassembly only.
2295 defm VCLEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00011, 0, "vcle", "s",
2298 // VCGT : Vector Compare Greater Than
2299 defm VCGTs : N3V_QHS<0, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
2300 IIC_VSUBi4Q, "vcgt", "s", NEONvcgt, 0>;
2301 defm VCGTu : N3V_QHS<1, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
2302 IIC_VSUBi4Q, "vcgt", "u", NEONvcgtu, 0>;
2303 def VCGTfd : N3VD<1,0,0b10,0b1110,0, IIC_VBIND, "vcgt", "f32", v2i32, v2f32,
2305 def VCGTfq : N3VQ<1,0,0b10,0b1110,0, IIC_VBINQ, "vcgt", "f32", v4i32, v4f32,
2307 // For disassembly only.
2308 defm VCGTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00000, 0, "vcgt", "s",
2310 // For disassembly only.
2311 defm VCLTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00100, 0, "vclt", "s",
2314 // VACGE : Vector Absolute Compare Greater Than or Equal (aka VCAGE)
2315 def VACGEd : N3VDInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacge",
2316 "f32", v2i32, v2f32, int_arm_neon_vacged, 0>;
2317 def VACGEq : N3VQInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacge",
2318 "f32", v4i32, v4f32, int_arm_neon_vacgeq, 0>;
2319 // VACGT : Vector Absolute Compare Greater Than (aka VCAGT)
2320 def VACGTd : N3VDInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacgt",
2321 "f32", v2i32, v2f32, int_arm_neon_vacgtd, 0>;
2322 def VACGTq : N3VQInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacgt",
2323 "f32", v4i32, v4f32, int_arm_neon_vacgtq, 0>;
2324 // VTST : Vector Test Bits
2325 defm VTST : N3V_QHS<0, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2326 IIC_VBINi4Q, "vtst", "", NEONvtst, 1>;
2328 // Vector Bitwise Operations.
2330 def vnot8 : PatFrag<(ops node:$in),
2331 (xor node:$in, (bitconvert (v8i8 immAllOnesV)))>;
2332 def vnot16 : PatFrag<(ops node:$in),
2333 (xor node:$in, (bitconvert (v16i8 immAllOnesV)))>;
2336 // VAND : Vector Bitwise AND
2337 def VANDd : N3VDX<0, 0, 0b00, 0b0001, 1, IIC_VBINiD, "vand",
2338 v2i32, v2i32, and, 1>;
2339 def VANDq : N3VQX<0, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "vand",
2340 v4i32, v4i32, and, 1>;
2342 // VEOR : Vector Bitwise Exclusive OR
2343 def VEORd : N3VDX<1, 0, 0b00, 0b0001, 1, IIC_VBINiD, "veor",
2344 v2i32, v2i32, xor, 1>;
2345 def VEORq : N3VQX<1, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "veor",
2346 v4i32, v4i32, xor, 1>;
2348 // VORR : Vector Bitwise OR
2349 def VORRd : N3VDX<0, 0, 0b10, 0b0001, 1, IIC_VBINiD, "vorr",
2350 v2i32, v2i32, or, 1>;
2351 def VORRq : N3VQX<0, 0, 0b10, 0b0001, 1, IIC_VBINiQ, "vorr",
2352 v4i32, v4i32, or, 1>;
2354 // VBIC : Vector Bitwise Bit Clear (AND NOT)
2355 def VBICd : N3VX<0, 0, 0b01, 0b0001, 0, 1, (outs DPR:$dst),
2356 (ins DPR:$src1, DPR:$src2), N3RegFrm, IIC_VBINiD,
2357 "vbic", "$dst, $src1, $src2", "",
2358 [(set DPR:$dst, (v2i32 (and DPR:$src1,
2359 (vnot8 DPR:$src2))))]>;
2360 def VBICq : N3VX<0, 0, 0b01, 0b0001, 1, 1, (outs QPR:$dst),
2361 (ins QPR:$src1, QPR:$src2), N3RegFrm, IIC_VBINiQ,
2362 "vbic", "$dst, $src1, $src2", "",
2363 [(set QPR:$dst, (v4i32 (and QPR:$src1,
2364 (vnot16 QPR:$src2))))]>;
2366 // VORN : Vector Bitwise OR NOT
2367 def VORNd : N3VX<0, 0, 0b11, 0b0001, 0, 1, (outs DPR:$dst),
2368 (ins DPR:$src1, DPR:$src2), N3RegFrm, IIC_VBINiD,
2369 "vorn", "$dst, $src1, $src2", "",
2370 [(set DPR:$dst, (v2i32 (or DPR:$src1,
2371 (vnot8 DPR:$src2))))]>;
2372 def VORNq : N3VX<0, 0, 0b11, 0b0001, 1, 1, (outs QPR:$dst),
2373 (ins QPR:$src1, QPR:$src2), N3RegFrm, IIC_VBINiQ,
2374 "vorn", "$dst, $src1, $src2", "",
2375 [(set QPR:$dst, (v4i32 (or QPR:$src1,
2376 (vnot16 QPR:$src2))))]>;
2378 // VMVN : Vector Bitwise NOT
2379 def VMVNd : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 0, 0,
2380 (outs DPR:$dst), (ins DPR:$src), IIC_VSUBiD,
2381 "vmvn", "$dst, $src", "",
2382 [(set DPR:$dst, (v2i32 (vnot8 DPR:$src)))]>;
2383 def VMVNq : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 1, 0,
2384 (outs QPR:$dst), (ins QPR:$src), IIC_VSUBiD,
2385 "vmvn", "$dst, $src", "",
2386 [(set QPR:$dst, (v4i32 (vnot16 QPR:$src)))]>;
2387 def : Pat<(v2i32 (vnot8 DPR:$src)), (VMVNd DPR:$src)>;
2388 def : Pat<(v4i32 (vnot16 QPR:$src)), (VMVNq QPR:$src)>;
2390 // VBSL : Vector Bitwise Select
2391 def VBSLd : N3VX<1, 0, 0b01, 0b0001, 0, 1, (outs DPR:$dst),
2392 (ins DPR:$src1, DPR:$src2, DPR:$src3),
2393 N3RegFrm, IIC_VCNTiD,
2394 "vbsl", "$dst, $src2, $src3", "$src1 = $dst",
2396 (v2i32 (or (and DPR:$src2, DPR:$src1),
2397 (and DPR:$src3, (vnot8 DPR:$src1)))))]>;
2398 def VBSLq : N3VX<1, 0, 0b01, 0b0001, 1, 1, (outs QPR:$dst),
2399 (ins QPR:$src1, QPR:$src2, QPR:$src3),
2400 N3RegFrm, IIC_VCNTiQ,
2401 "vbsl", "$dst, $src2, $src3", "$src1 = $dst",
2403 (v4i32 (or (and QPR:$src2, QPR:$src1),
2404 (and QPR:$src3, (vnot16 QPR:$src1)))))]>;
2406 // VBIF : Vector Bitwise Insert if False
2407 // like VBSL but with: "vbif $dst, $src3, $src1", "$src2 = $dst",
2408 def VBIFd : N3VX<1, 0, 0b11, 0b0001, 0, 1,
2409 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3),
2410 N3RegFrm, IIC_VBINiD,
2411 "vbif", "$dst, $src2, $src3", "$src1 = $dst",
2412 [/* For disassembly only; pattern left blank */]>;
2413 def VBIFq : N3VX<1, 0, 0b11, 0b0001, 1, 1,
2414 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3),
2415 N3RegFrm, IIC_VBINiQ,
2416 "vbif", "$dst, $src2, $src3", "$src1 = $dst",
2417 [/* For disassembly only; pattern left blank */]>;
2419 // VBIT : Vector Bitwise Insert if True
2420 // like VBSL but with: "vbit $dst, $src2, $src1", "$src3 = $dst",
2421 def VBITd : N3VX<1, 0, 0b10, 0b0001, 0, 1,
2422 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3),
2423 N3RegFrm, IIC_VBINiD,
2424 "vbit", "$dst, $src2, $src3", "$src1 = $dst",
2425 [/* For disassembly only; pattern left blank */]>;
2426 def VBITq : N3VX<1, 0, 0b10, 0b0001, 1, 1,
2427 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3),
2428 N3RegFrm, IIC_VBINiQ,
2429 "vbit", "$dst, $src2, $src3", "$src1 = $dst",
2430 [/* For disassembly only; pattern left blank */]>;
2432 // VBIT/VBIF are not yet implemented. The TwoAddress pass will not go looking
2433 // for equivalent operations with different register constraints; it just
2436 // Vector Absolute Differences.
2438 // VABD : Vector Absolute Difference
2439 defm VABDs : N3VInt_QHS<0, 0, 0b0111, 0, N3RegFrm,
2440 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2441 "vabd", "s", int_arm_neon_vabds, 0>;
2442 defm VABDu : N3VInt_QHS<1, 0, 0b0111, 0, N3RegFrm,
2443 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2444 "vabd", "u", int_arm_neon_vabdu, 0>;
2445 def VABDfd : N3VDInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBIND,
2446 "vabd", "f32", v2f32, v2f32, int_arm_neon_vabds, 0>;
2447 def VABDfq : N3VQInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBINQ,
2448 "vabd", "f32", v4f32, v4f32, int_arm_neon_vabds, 0>;
2450 // VABDL : Vector Absolute Difference Long (Q = | D - D |)
2451 defm VABDLs : N3VLInt_QHS<0,1,0b0111,0, IIC_VSUBi4Q, IIC_VSUBi4Q,
2452 "vabdl", "s", int_arm_neon_vabdls, 0>;
2453 defm VABDLu : N3VLInt_QHS<1,1,0b0111,0, IIC_VSUBi4Q, IIC_VSUBi4Q,
2454 "vabdl", "u", int_arm_neon_vabdlu, 0>;
2456 // VABA : Vector Absolute Difference and Accumulate
2457 defm VABAs : N3VInt3_QHS<0,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
2458 "vaba", "s", int_arm_neon_vabas>;
2459 defm VABAu : N3VInt3_QHS<1,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
2460 "vaba", "u", int_arm_neon_vabau>;
2462 // VABAL : Vector Absolute Difference and Accumulate Long (Q += | D - D |)
2463 defm VABALs : N3VLInt3_QHS<0,1,0b0101,0, IIC_VABAD, IIC_VABAD,
2464 "vabal", "s", int_arm_neon_vabals>;
2465 defm VABALu : N3VLInt3_QHS<1,1,0b0101,0, IIC_VABAD, IIC_VABAD,
2466 "vabal", "u", int_arm_neon_vabalu>;
2468 // Vector Maximum and Minimum.
2470 // VMAX : Vector Maximum
2471 defm VMAXs : N3VInt_QHS<0, 0, 0b0110, 0, N3RegFrm,
2472 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2473 "vmax", "s", int_arm_neon_vmaxs, 1>;
2474 defm VMAXu : N3VInt_QHS<1, 0, 0b0110, 0, N3RegFrm,
2475 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2476 "vmax", "u", int_arm_neon_vmaxu, 1>;
2477 def VMAXfd : N3VDInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBIND,
2479 v2f32, v2f32, int_arm_neon_vmaxs, 1>;
2480 def VMAXfq : N3VQInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBINQ,
2482 v4f32, v4f32, int_arm_neon_vmaxs, 1>;
2484 // VMIN : Vector Minimum
2485 defm VMINs : N3VInt_QHS<0, 0, 0b0110, 1, N3RegFrm,
2486 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2487 "vmin", "s", int_arm_neon_vmins, 1>;
2488 defm VMINu : N3VInt_QHS<1, 0, 0b0110, 1, N3RegFrm,
2489 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2490 "vmin", "u", int_arm_neon_vminu, 1>;
2491 def VMINfd : N3VDInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBIND,
2493 v2f32, v2f32, int_arm_neon_vmins, 1>;
2494 def VMINfq : N3VQInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBINQ,
2496 v4f32, v4f32, int_arm_neon_vmins, 1>;
2498 // Vector Pairwise Operations.
2500 // VPADD : Vector Pairwise Add
2501 def VPADDi8 : N3VDInt<0, 0, 0b00, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
2503 v8i8, v8i8, int_arm_neon_vpadd, 0>;
2504 def VPADDi16 : N3VDInt<0, 0, 0b01, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
2506 v4i16, v4i16, int_arm_neon_vpadd, 0>;
2507 def VPADDi32 : N3VDInt<0, 0, 0b10, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
2509 v2i32, v2i32, int_arm_neon_vpadd, 0>;
2510 def VPADDf : N3VDInt<1, 0, 0b00, 0b1101, 0, N3RegFrm,
2511 IIC_VBIND, "vpadd", "f32",
2512 v2f32, v2f32, int_arm_neon_vpadd, 0>;
2514 // VPADDL : Vector Pairwise Add Long
2515 defm VPADDLs : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00100, 0, "vpaddl", "s",
2516 int_arm_neon_vpaddls>;
2517 defm VPADDLu : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00101, 0, "vpaddl", "u",
2518 int_arm_neon_vpaddlu>;
2520 // VPADAL : Vector Pairwise Add and Accumulate Long
2521 defm VPADALs : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01100, 0, "vpadal", "s",
2522 int_arm_neon_vpadals>;
2523 defm VPADALu : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01101, 0, "vpadal", "u",
2524 int_arm_neon_vpadalu>;
2526 // VPMAX : Vector Pairwise Maximum
2527 def VPMAXs8 : N3VDInt<0, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
2528 "s8", v8i8, v8i8, int_arm_neon_vpmaxs, 0>;
2529 def VPMAXs16 : N3VDInt<0, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
2530 "s16", v4i16, v4i16, int_arm_neon_vpmaxs, 0>;
2531 def VPMAXs32 : N3VDInt<0, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
2532 "s32", v2i32, v2i32, int_arm_neon_vpmaxs, 0>;
2533 def VPMAXu8 : N3VDInt<1, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
2534 "u8", v8i8, v8i8, int_arm_neon_vpmaxu, 0>;
2535 def VPMAXu16 : N3VDInt<1, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
2536 "u16", v4i16, v4i16, int_arm_neon_vpmaxu, 0>;
2537 def VPMAXu32 : N3VDInt<1, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
2538 "u32", v2i32, v2i32, int_arm_neon_vpmaxu, 0>;
2539 def VPMAXf : N3VDInt<1, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
2540 "f32", v2f32, v2f32, int_arm_neon_vpmaxs, 0>;
2542 // VPMIN : Vector Pairwise Minimum
2543 def VPMINs8 : N3VDInt<0, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
2544 "s8", v8i8, v8i8, int_arm_neon_vpmins, 0>;
2545 def VPMINs16 : N3VDInt<0, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
2546 "s16", v4i16, v4i16, int_arm_neon_vpmins, 0>;
2547 def VPMINs32 : N3VDInt<0, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
2548 "s32", v2i32, v2i32, int_arm_neon_vpmins, 0>;
2549 def VPMINu8 : N3VDInt<1, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
2550 "u8", v8i8, v8i8, int_arm_neon_vpminu, 0>;
2551 def VPMINu16 : N3VDInt<1, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
2552 "u16", v4i16, v4i16, int_arm_neon_vpminu, 0>;
2553 def VPMINu32 : N3VDInt<1, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
2554 "u32", v2i32, v2i32, int_arm_neon_vpminu, 0>;
2555 def VPMINf : N3VDInt<1, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VSUBi4D, "vpmin",
2556 "f32", v2f32, v2f32, int_arm_neon_vpmins, 0>;
2558 // Vector Reciprocal and Reciprocal Square Root Estimate and Step.
2560 // VRECPE : Vector Reciprocal Estimate
2561 def VRECPEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
2562 IIC_VUNAD, "vrecpe", "u32",
2563 v2i32, v2i32, int_arm_neon_vrecpe>;
2564 def VRECPEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
2565 IIC_VUNAQ, "vrecpe", "u32",
2566 v4i32, v4i32, int_arm_neon_vrecpe>;
2567 def VRECPEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
2568 IIC_VUNAD, "vrecpe", "f32",
2569 v2f32, v2f32, int_arm_neon_vrecpe>;
2570 def VRECPEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
2571 IIC_VUNAQ, "vrecpe", "f32",
2572 v4f32, v4f32, int_arm_neon_vrecpe>;
2574 // VRECPS : Vector Reciprocal Step
2575 def VRECPSfd : N3VDInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
2576 IIC_VRECSD, "vrecps", "f32",
2577 v2f32, v2f32, int_arm_neon_vrecps, 1>;
2578 def VRECPSfq : N3VQInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
2579 IIC_VRECSQ, "vrecps", "f32",
2580 v4f32, v4f32, int_arm_neon_vrecps, 1>;
2582 // VRSQRTE : Vector Reciprocal Square Root Estimate
2583 def VRSQRTEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
2584 IIC_VUNAD, "vrsqrte", "u32",
2585 v2i32, v2i32, int_arm_neon_vrsqrte>;
2586 def VRSQRTEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
2587 IIC_VUNAQ, "vrsqrte", "u32",
2588 v4i32, v4i32, int_arm_neon_vrsqrte>;
2589 def VRSQRTEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
2590 IIC_VUNAD, "vrsqrte", "f32",
2591 v2f32, v2f32, int_arm_neon_vrsqrte>;
2592 def VRSQRTEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
2593 IIC_VUNAQ, "vrsqrte", "f32",
2594 v4f32, v4f32, int_arm_neon_vrsqrte>;
2596 // VRSQRTS : Vector Reciprocal Square Root Step
2597 def VRSQRTSfd : N3VDInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
2598 IIC_VRECSD, "vrsqrts", "f32",
2599 v2f32, v2f32, int_arm_neon_vrsqrts, 1>;
2600 def VRSQRTSfq : N3VQInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
2601 IIC_VRECSQ, "vrsqrts", "f32",
2602 v4f32, v4f32, int_arm_neon_vrsqrts, 1>;
2606 // VSHL : Vector Shift
2607 defm VSHLs : N3VInt_QHSD<0, 0, 0b0100, 0, N3RegVShFrm,
2608 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
2609 "vshl", "s", int_arm_neon_vshifts, 0>;
2610 defm VSHLu : N3VInt_QHSD<1, 0, 0b0100, 0, N3RegVShFrm,
2611 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
2612 "vshl", "u", int_arm_neon_vshiftu, 0>;
2613 // VSHL : Vector Shift Left (Immediate)
2614 defm VSHLi : N2VSh_QHSD<0, 1, 0b0101, 1, IIC_VSHLiD, "vshl", "i", NEONvshl,
2616 // VSHR : Vector Shift Right (Immediate)
2617 defm VSHRs : N2VSh_QHSD<0, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "s", NEONvshrs,
2619 defm VSHRu : N2VSh_QHSD<1, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "u", NEONvshru,
2622 // VSHLL : Vector Shift Left Long
2623 defm VSHLLs : N2VLSh_QHS<0, 1, 0b1010, 0, 0, 1, "vshll", "s", NEONvshlls>;
2624 defm VSHLLu : N2VLSh_QHS<1, 1, 0b1010, 0, 0, 1, "vshll", "u", NEONvshllu>;
2626 // VSHLL : Vector Shift Left Long (with maximum shift count)
2627 class N2VLShMax<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
2628 bit op6, bit op4, string OpcodeStr, string Dt, ValueType ResTy,
2629 ValueType OpTy, SDNode OpNode>
2630 : N2VLSh<op24, op23, op11_8, op7, op6, op4, OpcodeStr, Dt,
2631 ResTy, OpTy, OpNode> {
2632 let Inst{21-16} = op21_16;
2634 def VSHLLi8 : N2VLShMax<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll", "i8",
2635 v8i16, v8i8, NEONvshlli>;
2636 def VSHLLi16 : N2VLShMax<1, 1, 0b110110, 0b0011, 0, 0, 0, "vshll", "i16",
2637 v4i32, v4i16, NEONvshlli>;
2638 def VSHLLi32 : N2VLShMax<1, 1, 0b111010, 0b0011, 0, 0, 0, "vshll", "i32",
2639 v2i64, v2i32, NEONvshlli>;
2641 // VSHRN : Vector Shift Right and Narrow
2642 defm VSHRN : N2VNSh_HSD<0,1,0b1000,0,0,1, IIC_VSHLiD, "vshrn", "i",
2645 // VRSHL : Vector Rounding Shift
2646 defm VRSHLs : N3VInt_QHSD<0, 0, 0b0101, 0, N3RegVShFrm,
2647 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
2648 "vrshl", "s", int_arm_neon_vrshifts, 0>;
2649 defm VRSHLu : N3VInt_QHSD<1, 0, 0b0101, 0, N3RegVShFrm,
2650 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
2651 "vrshl", "u", int_arm_neon_vrshiftu, 0>;
2652 // VRSHR : Vector Rounding Shift Right
2653 defm VRSHRs : N2VSh_QHSD<0,1,0b0010,1, IIC_VSHLi4D, "vrshr", "s", NEONvrshrs,
2655 defm VRSHRu : N2VSh_QHSD<1,1,0b0010,1, IIC_VSHLi4D, "vrshr", "u", NEONvrshru,
2658 // VRSHRN : Vector Rounding Shift Right and Narrow
2659 defm VRSHRN : N2VNSh_HSD<0, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vrshrn", "i",
2662 // VQSHL : Vector Saturating Shift
2663 defm VQSHLs : N3VInt_QHSD<0, 0, 0b0100, 1, N3RegVShFrm,
2664 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
2665 "vqshl", "s", int_arm_neon_vqshifts, 0>;
2666 defm VQSHLu : N3VInt_QHSD<1, 0, 0b0100, 1, N3RegVShFrm,
2667 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
2668 "vqshl", "u", int_arm_neon_vqshiftu, 0>;
2669 // VQSHL : Vector Saturating Shift Left (Immediate)
2670 defm VQSHLsi : N2VSh_QHSD<0,1,0b0111,1, IIC_VSHLi4D, "vqshl", "s",NEONvqshls,
2672 defm VQSHLui : N2VSh_QHSD<1,1,0b0111,1, IIC_VSHLi4D, "vqshl", "u",NEONvqshlu,
2674 // VQSHLU : Vector Saturating Shift Left (Immediate, Unsigned)
2675 defm VQSHLsu : N2VSh_QHSD<1,1,0b0110,1, IIC_VSHLi4D,"vqshlu","s",NEONvqshlsu,
2678 // VQSHRN : Vector Saturating Shift Right and Narrow
2679 defm VQSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "s",
2681 defm VQSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "u",
2684 // VQSHRUN : Vector Saturating Shift Right and Narrow (Unsigned)
2685 defm VQSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 0, 1, IIC_VSHLi4D, "vqshrun", "s",
2688 // VQRSHL : Vector Saturating Rounding Shift
2689 defm VQRSHLs : N3VInt_QHSD<0, 0, 0b0101, 1, N3RegVShFrm,
2690 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
2691 "vqrshl", "s", int_arm_neon_vqrshifts, 0>;
2692 defm VQRSHLu : N3VInt_QHSD<1, 0, 0b0101, 1, N3RegVShFrm,
2693 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
2694 "vqrshl", "u", int_arm_neon_vqrshiftu, 0>;
2696 // VQRSHRN : Vector Saturating Rounding Shift Right and Narrow
2697 defm VQRSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "s",
2699 defm VQRSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "u",
2702 // VQRSHRUN : Vector Saturating Rounding Shift Right and Narrow (Unsigned)
2703 defm VQRSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vqrshrun", "s",
2706 // VSRA : Vector Shift Right and Accumulate
2707 defm VSRAs : N2VShAdd_QHSD<0, 1, 0b0001, 1, "vsra", "s", NEONvshrs>;
2708 defm VSRAu : N2VShAdd_QHSD<1, 1, 0b0001, 1, "vsra", "u", NEONvshru>;
2709 // VRSRA : Vector Rounding Shift Right and Accumulate
2710 defm VRSRAs : N2VShAdd_QHSD<0, 1, 0b0011, 1, "vrsra", "s", NEONvrshrs>;
2711 defm VRSRAu : N2VShAdd_QHSD<1, 1, 0b0011, 1, "vrsra", "u", NEONvrshru>;
2713 // VSLI : Vector Shift Left and Insert
2714 defm VSLI : N2VShIns_QHSD<1, 1, 0b0101, 1, "vsli", NEONvsli, N2RegVShLFrm>;
2715 // VSRI : Vector Shift Right and Insert
2716 defm VSRI : N2VShIns_QHSD<1, 1, 0b0100, 1, "vsri", NEONvsri, N2RegVShRFrm>;
2718 // Vector Absolute and Saturating Absolute.
2720 // VABS : Vector Absolute Value
2721 defm VABS : N2VInt_QHS<0b11, 0b11, 0b01, 0b00110, 0,
2722 IIC_VUNAiD, IIC_VUNAiQ, "vabs", "s",
2724 def VABSfd : N2VDInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
2725 IIC_VUNAD, "vabs", "f32",
2726 v2f32, v2f32, int_arm_neon_vabs>;
2727 def VABSfq : N2VQInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
2728 IIC_VUNAQ, "vabs", "f32",
2729 v4f32, v4f32, int_arm_neon_vabs>;
2731 // VQABS : Vector Saturating Absolute Value
2732 defm VQABS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01110, 0,
2733 IIC_VQUNAiD, IIC_VQUNAiQ, "vqabs", "s",
2734 int_arm_neon_vqabs>;
2738 def vneg : PatFrag<(ops node:$in), (sub immAllZerosV, node:$in)>;
2739 def vneg8 : PatFrag<(ops node:$in),
2740 (sub (bitconvert (v8i8 immAllZerosV)), node:$in)>;
2741 def vneg16 : PatFrag<(ops node:$in),
2742 (sub (bitconvert (v16i8 immAllZerosV)), node:$in)>;
2744 class VNEGD<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
2745 : N2V<0b11, 0b11, size, 0b01, 0b00111, 0, 0, (outs DPR:$dst), (ins DPR:$src),
2746 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
2747 [(set DPR:$dst, (Ty (vneg8 DPR:$src)))]>;
2748 class VNEGQ<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
2749 : N2V<0b11, 0b11, size, 0b01, 0b00111, 1, 0, (outs QPR:$dst), (ins QPR:$src),
2750 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
2751 [(set QPR:$dst, (Ty (vneg16 QPR:$src)))]>;
2753 // VNEG : Vector Negate (integer)
2754 def VNEGs8d : VNEGD<0b00, "vneg", "s8", v8i8>;
2755 def VNEGs16d : VNEGD<0b01, "vneg", "s16", v4i16>;
2756 def VNEGs32d : VNEGD<0b10, "vneg", "s32", v2i32>;
2757 def VNEGs8q : VNEGQ<0b00, "vneg", "s8", v16i8>;
2758 def VNEGs16q : VNEGQ<0b01, "vneg", "s16", v8i16>;
2759 def VNEGs32q : VNEGQ<0b10, "vneg", "s32", v4i32>;
2761 // VNEG : Vector Negate (floating-point)
2762 def VNEGfd : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
2763 (outs DPR:$dst), (ins DPR:$src), IIC_VUNAD,
2764 "vneg", "f32", "$dst, $src", "",
2765 [(set DPR:$dst, (v2f32 (fneg DPR:$src)))]>;
2766 def VNEGf32q : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 1, 0,
2767 (outs QPR:$dst), (ins QPR:$src), IIC_VUNAQ,
2768 "vneg", "f32", "$dst, $src", "",
2769 [(set QPR:$dst, (v4f32 (fneg QPR:$src)))]>;
2771 def : Pat<(v8i8 (vneg8 DPR:$src)), (VNEGs8d DPR:$src)>;
2772 def : Pat<(v4i16 (vneg8 DPR:$src)), (VNEGs16d DPR:$src)>;
2773 def : Pat<(v2i32 (vneg8 DPR:$src)), (VNEGs32d DPR:$src)>;
2774 def : Pat<(v16i8 (vneg16 QPR:$src)), (VNEGs8q QPR:$src)>;
2775 def : Pat<(v8i16 (vneg16 QPR:$src)), (VNEGs16q QPR:$src)>;
2776 def : Pat<(v4i32 (vneg16 QPR:$src)), (VNEGs32q QPR:$src)>;
2778 // VQNEG : Vector Saturating Negate
2779 defm VQNEG : N2VInt_QHS<0b11, 0b11, 0b00, 0b01111, 0,
2780 IIC_VQUNAiD, IIC_VQUNAiQ, "vqneg", "s",
2781 int_arm_neon_vqneg>;
2783 // Vector Bit Counting Operations.
2785 // VCLS : Vector Count Leading Sign Bits
2786 defm VCLS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01000, 0,
2787 IIC_VCNTiD, IIC_VCNTiQ, "vcls", "s",
2789 // VCLZ : Vector Count Leading Zeros
2790 defm VCLZ : N2VInt_QHS<0b11, 0b11, 0b00, 0b01001, 0,
2791 IIC_VCNTiD, IIC_VCNTiQ, "vclz", "i",
2793 // VCNT : Vector Count One Bits
2794 def VCNTd : N2VDInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
2795 IIC_VCNTiD, "vcnt", "8",
2796 v8i8, v8i8, int_arm_neon_vcnt>;
2797 def VCNTq : N2VQInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
2798 IIC_VCNTiQ, "vcnt", "8",
2799 v16i8, v16i8, int_arm_neon_vcnt>;
2801 // Vector Swap -- for disassembly only.
2802 def VSWPd : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 0, 0,
2803 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
2804 "vswp", "$dst, $src", "", []>;
2805 def VSWPq : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 1, 0,
2806 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
2807 "vswp", "$dst, $src", "", []>;
2809 // Vector Move Operations.
2811 // VMOV : Vector Move (Register)
2813 let neverHasSideEffects = 1 in {
2814 def VMOVDneon: N3VX<0, 0, 0b10, 0b0001, 0, 1, (outs DPR:$dst), (ins DPR:$src),
2815 N3RegFrm, IIC_VMOVD, "vmov", "$dst, $src", "", []>;
2816 def VMOVQ : N3VX<0, 0, 0b10, 0b0001, 1, 1, (outs QPR:$dst), (ins QPR:$src),
2817 N3RegFrm, IIC_VMOVD, "vmov", "$dst, $src", "", []>;
2819 // Pseudo vector move instructions for QQ and QQQQ registers. This should
2820 // be expanded after register allocation is completed.
2821 def VMOVQQ : PseudoInst<(outs QQPR:$dst), (ins QQPR:$src),
2822 NoItinerary, "${:comment} vmov\t$dst, $src", []>;
2824 def VMOVQQQQ : PseudoInst<(outs QQQQPR:$dst), (ins QQQQPR:$src),
2825 NoItinerary, "${:comment} vmov\t$dst, $src", []>;
2826 } // neverHasSideEffects
2828 // VMOV : Vector Move (Immediate)
2830 // VMOV_get_imm8 xform function: convert build_vector to VMOV.i8 imm.
2831 def VMOV_get_imm8 : SDNodeXForm<build_vector, [{
2832 return ARM::getVMOVImm(N, 1, *CurDAG);
2834 def vmovImm8 : PatLeaf<(build_vector), [{
2835 return ARM::getVMOVImm(N, 1, *CurDAG).getNode() != 0;
2838 // VMOV_get_imm16 xform function: convert build_vector to VMOV.i16 imm.
2839 def VMOV_get_imm16 : SDNodeXForm<build_vector, [{
2840 return ARM::getVMOVImm(N, 2, *CurDAG);
2842 def vmovImm16 : PatLeaf<(build_vector), [{
2843 return ARM::getVMOVImm(N, 2, *CurDAG).getNode() != 0;
2844 }], VMOV_get_imm16>;
2846 // VMOV_get_imm32 xform function: convert build_vector to VMOV.i32 imm.
2847 def VMOV_get_imm32 : SDNodeXForm<build_vector, [{
2848 return ARM::getVMOVImm(N, 4, *CurDAG);
2850 def vmovImm32 : PatLeaf<(build_vector), [{
2851 return ARM::getVMOVImm(N, 4, *CurDAG).getNode() != 0;
2852 }], VMOV_get_imm32>;
2854 // VMOV_get_imm64 xform function: convert build_vector to VMOV.i64 imm.
2855 def VMOV_get_imm64 : SDNodeXForm<build_vector, [{
2856 return ARM::getVMOVImm(N, 8, *CurDAG);
2858 def vmovImm64 : PatLeaf<(build_vector), [{
2859 return ARM::getVMOVImm(N, 8, *CurDAG).getNode() != 0;
2860 }], VMOV_get_imm64>;
2862 // Note: Some of the cmode bits in the following VMOV instructions need to
2863 // be encoded based on the immed values.
2865 let isReMaterializable = 1 in {
2866 def VMOVv8i8 : N1ModImm<1, 0b000, 0b1110, 0, 0, 0, 1, (outs DPR:$dst),
2867 (ins h8imm:$SIMM), IIC_VMOVImm,
2868 "vmov", "i8", "$dst, $SIMM", "",
2869 [(set DPR:$dst, (v8i8 vmovImm8:$SIMM))]>;
2870 def VMOVv16i8 : N1ModImm<1, 0b000, 0b1110, 0, 1, 0, 1, (outs QPR:$dst),
2871 (ins h8imm:$SIMM), IIC_VMOVImm,
2872 "vmov", "i8", "$dst, $SIMM", "",
2873 [(set QPR:$dst, (v16i8 vmovImm8:$SIMM))]>;
2875 def VMOVv4i16 : N1ModImm<1, 0b000, {1,0,?,?}, 0, 0, {?}, 1, (outs DPR:$dst),
2876 (ins h16imm:$SIMM), IIC_VMOVImm,
2877 "vmov", "i16", "$dst, $SIMM", "",
2878 [(set DPR:$dst, (v4i16 vmovImm16:$SIMM))]>;
2879 def VMOVv8i16 : N1ModImm<1, 0b000, {1,0,?,?}, 0, 1, {?}, 1, (outs QPR:$dst),
2880 (ins h16imm:$SIMM), IIC_VMOVImm,
2881 "vmov", "i16", "$dst, $SIMM", "",
2882 [(set QPR:$dst, (v8i16 vmovImm16:$SIMM))]>;
2884 def VMOVv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, {?}, 1, (outs DPR:$dst),
2885 (ins h32imm:$SIMM), IIC_VMOVImm,
2886 "vmov", "i32", "$dst, $SIMM", "",
2887 [(set DPR:$dst, (v2i32 vmovImm32:$SIMM))]>;
2888 def VMOVv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, {?}, 1, (outs QPR:$dst),
2889 (ins h32imm:$SIMM), IIC_VMOVImm,
2890 "vmov", "i32", "$dst, $SIMM", "",
2891 [(set QPR:$dst, (v4i32 vmovImm32:$SIMM))]>;
2893 def VMOVv1i64 : N1ModImm<1, 0b000, 0b1110, 0, 0, 1, 1, (outs DPR:$dst),
2894 (ins h64imm:$SIMM), IIC_VMOVImm,
2895 "vmov", "i64", "$dst, $SIMM", "",
2896 [(set DPR:$dst, (v1i64 vmovImm64:$SIMM))]>;
2897 def VMOVv2i64 : N1ModImm<1, 0b000, 0b1110, 0, 1, 1, 1, (outs QPR:$dst),
2898 (ins h64imm:$SIMM), IIC_VMOVImm,
2899 "vmov", "i64", "$dst, $SIMM", "",
2900 [(set QPR:$dst, (v2i64 vmovImm64:$SIMM))]>;
2901 } // isReMaterializable
2903 // VMOV : Vector Get Lane (move scalar to ARM core register)
2905 def VGETLNs8 : NVGetLane<{1,1,1,0,0,1,?,1}, 0b1011, {?,?},
2906 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2907 IIC_VMOVSI, "vmov", "s8", "$dst, $src[$lane]",
2908 [(set GPR:$dst, (NEONvgetlanes (v8i8 DPR:$src),
2910 def VGETLNs16 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, {?,1},
2911 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2912 IIC_VMOVSI, "vmov", "s16", "$dst, $src[$lane]",
2913 [(set GPR:$dst, (NEONvgetlanes (v4i16 DPR:$src),
2915 def VGETLNu8 : NVGetLane<{1,1,1,0,1,1,?,1}, 0b1011, {?,?},
2916 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2917 IIC_VMOVSI, "vmov", "u8", "$dst, $src[$lane]",
2918 [(set GPR:$dst, (NEONvgetlaneu (v8i8 DPR:$src),
2920 def VGETLNu16 : NVGetLane<{1,1,1,0,1,0,?,1}, 0b1011, {?,1},
2921 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2922 IIC_VMOVSI, "vmov", "u16", "$dst, $src[$lane]",
2923 [(set GPR:$dst, (NEONvgetlaneu (v4i16 DPR:$src),
2925 def VGETLNi32 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, 0b00,
2926 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2927 IIC_VMOVSI, "vmov", "32", "$dst, $src[$lane]",
2928 [(set GPR:$dst, (extractelt (v2i32 DPR:$src),
2930 // def VGETLNf32: see FMRDH and FMRDL in ARMInstrVFP.td
2931 def : Pat<(NEONvgetlanes (v16i8 QPR:$src), imm:$lane),
2932 (VGETLNs8 (v8i8 (EXTRACT_SUBREG QPR:$src,
2933 (DSubReg_i8_reg imm:$lane))),
2934 (SubReg_i8_lane imm:$lane))>;
2935 def : Pat<(NEONvgetlanes (v8i16 QPR:$src), imm:$lane),
2936 (VGETLNs16 (v4i16 (EXTRACT_SUBREG QPR:$src,
2937 (DSubReg_i16_reg imm:$lane))),
2938 (SubReg_i16_lane imm:$lane))>;
2939 def : Pat<(NEONvgetlaneu (v16i8 QPR:$src), imm:$lane),
2940 (VGETLNu8 (v8i8 (EXTRACT_SUBREG QPR:$src,
2941 (DSubReg_i8_reg imm:$lane))),
2942 (SubReg_i8_lane imm:$lane))>;
2943 def : Pat<(NEONvgetlaneu (v8i16 QPR:$src), imm:$lane),
2944 (VGETLNu16 (v4i16 (EXTRACT_SUBREG QPR:$src,
2945 (DSubReg_i16_reg imm:$lane))),
2946 (SubReg_i16_lane imm:$lane))>;
2947 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
2948 (VGETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src,
2949 (DSubReg_i32_reg imm:$lane))),
2950 (SubReg_i32_lane imm:$lane))>;
2951 def : Pat<(extractelt (v2f32 DPR:$src1), imm:$src2),
2952 (EXTRACT_SUBREG (v2f32 (COPY_TO_REGCLASS (v2f32 DPR:$src1),DPR_VFP2)),
2953 (SSubReg_f32_reg imm:$src2))>;
2954 def : Pat<(extractelt (v4f32 QPR:$src1), imm:$src2),
2955 (EXTRACT_SUBREG (v4f32 (COPY_TO_REGCLASS (v4f32 QPR:$src1),QPR_VFP2)),
2956 (SSubReg_f32_reg imm:$src2))>;
2957 //def : Pat<(extractelt (v2i64 QPR:$src1), imm:$src2),
2958 // (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
2959 def : Pat<(extractelt (v2f64 QPR:$src1), imm:$src2),
2960 (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
2963 // VMOV : Vector Set Lane (move ARM core register to scalar)
2965 let Constraints = "$src1 = $dst" in {
2966 def VSETLNi8 : NVSetLane<{1,1,1,0,0,1,?,0}, 0b1011, {?,?}, (outs DPR:$dst),
2967 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
2968 IIC_VMOVISL, "vmov", "8", "$dst[$lane], $src2",
2969 [(set DPR:$dst, (vector_insert (v8i8 DPR:$src1),
2970 GPR:$src2, imm:$lane))]>;
2971 def VSETLNi16 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, {?,1}, (outs DPR:$dst),
2972 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
2973 IIC_VMOVISL, "vmov", "16", "$dst[$lane], $src2",
2974 [(set DPR:$dst, (vector_insert (v4i16 DPR:$src1),
2975 GPR:$src2, imm:$lane))]>;
2976 def VSETLNi32 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, 0b00, (outs DPR:$dst),
2977 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
2978 IIC_VMOVISL, "vmov", "32", "$dst[$lane], $src2",
2979 [(set DPR:$dst, (insertelt (v2i32 DPR:$src1),
2980 GPR:$src2, imm:$lane))]>;
2982 def : Pat<(vector_insert (v16i8 QPR:$src1), GPR:$src2, imm:$lane),
2983 (v16i8 (INSERT_SUBREG QPR:$src1,
2984 (v8i8 (VSETLNi8 (v8i8 (EXTRACT_SUBREG QPR:$src1,
2985 (DSubReg_i8_reg imm:$lane))),
2986 GPR:$src2, (SubReg_i8_lane imm:$lane))),
2987 (DSubReg_i8_reg imm:$lane)))>;
2988 def : Pat<(vector_insert (v8i16 QPR:$src1), GPR:$src2, imm:$lane),
2989 (v8i16 (INSERT_SUBREG QPR:$src1,
2990 (v4i16 (VSETLNi16 (v4i16 (EXTRACT_SUBREG QPR:$src1,
2991 (DSubReg_i16_reg imm:$lane))),
2992 GPR:$src2, (SubReg_i16_lane imm:$lane))),
2993 (DSubReg_i16_reg imm:$lane)))>;
2994 def : Pat<(insertelt (v4i32 QPR:$src1), GPR:$src2, imm:$lane),
2995 (v4i32 (INSERT_SUBREG QPR:$src1,
2996 (v2i32 (VSETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src1,
2997 (DSubReg_i32_reg imm:$lane))),
2998 GPR:$src2, (SubReg_i32_lane imm:$lane))),
2999 (DSubReg_i32_reg imm:$lane)))>;
3001 def : Pat<(v2f32 (insertelt DPR:$src1, SPR:$src2, imm:$src3)),
3002 (INSERT_SUBREG (v2f32 (COPY_TO_REGCLASS DPR:$src1, DPR_VFP2)),
3003 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
3004 def : Pat<(v4f32 (insertelt QPR:$src1, SPR:$src2, imm:$src3)),
3005 (INSERT_SUBREG (v4f32 (COPY_TO_REGCLASS QPR:$src1, QPR_VFP2)),
3006 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
3008 //def : Pat<(v2i64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
3009 // (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
3010 def : Pat<(v2f64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
3011 (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
3013 def : Pat<(v2f32 (scalar_to_vector SPR:$src)),
3014 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
3015 def : Pat<(v2f64 (scalar_to_vector (f64 DPR:$src))),
3016 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
3017 def : Pat<(v4f32 (scalar_to_vector SPR:$src)),
3018 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
3020 def : Pat<(v8i8 (scalar_to_vector GPR:$src)),
3021 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
3022 def : Pat<(v4i16 (scalar_to_vector GPR:$src)),
3023 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
3024 def : Pat<(v2i32 (scalar_to_vector GPR:$src)),
3025 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
3027 def : Pat<(v16i8 (scalar_to_vector GPR:$src)),
3028 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3029 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
3031 def : Pat<(v8i16 (scalar_to_vector GPR:$src)),
3032 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
3033 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
3035 def : Pat<(v4i32 (scalar_to_vector GPR:$src)),
3036 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
3037 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
3040 // VDUP : Vector Duplicate (from ARM core register to all elements)
3042 class VDUPD<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
3043 : NVDup<opcod1, 0b1011, opcod3, (outs DPR:$dst), (ins GPR:$src),
3044 IIC_VMOVIS, "vdup", Dt, "$dst, $src",
3045 [(set DPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
3046 class VDUPQ<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
3047 : NVDup<opcod1, 0b1011, opcod3, (outs QPR:$dst), (ins GPR:$src),
3048 IIC_VMOVIS, "vdup", Dt, "$dst, $src",
3049 [(set QPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
3051 def VDUP8d : VDUPD<0b11101100, 0b00, "8", v8i8>;
3052 def VDUP16d : VDUPD<0b11101000, 0b01, "16", v4i16>;
3053 def VDUP32d : VDUPD<0b11101000, 0b00, "32", v2i32>;
3054 def VDUP8q : VDUPQ<0b11101110, 0b00, "8", v16i8>;
3055 def VDUP16q : VDUPQ<0b11101010, 0b01, "16", v8i16>;
3056 def VDUP32q : VDUPQ<0b11101010, 0b00, "32", v4i32>;
3058 def VDUPfd : NVDup<0b11101000, 0b1011, 0b00, (outs DPR:$dst), (ins GPR:$src),
3059 IIC_VMOVIS, "vdup", "32", "$dst, $src",
3060 [(set DPR:$dst, (v2f32 (NEONvdup
3061 (f32 (bitconvert GPR:$src)))))]>;
3062 def VDUPfq : NVDup<0b11101010, 0b1011, 0b00, (outs QPR:$dst), (ins GPR:$src),
3063 IIC_VMOVIS, "vdup", "32", "$dst, $src",
3064 [(set QPR:$dst, (v4f32 (NEONvdup
3065 (f32 (bitconvert GPR:$src)))))]>;
3067 // VDUP : Vector Duplicate Lane (from scalar to all elements)
3069 class VDUPLND<bits<4> op19_16, string OpcodeStr, string Dt,
3071 : NVDupLane<op19_16, 0, (outs DPR:$dst), (ins DPR:$src, nohash_imm:$lane),
3072 IIC_VMOVD, OpcodeStr, Dt, "$dst, $src[$lane]",
3073 [(set DPR:$dst, (Ty (NEONvduplane (Ty DPR:$src), imm:$lane)))]>;
3075 class VDUPLNQ<bits<4> op19_16, string OpcodeStr, string Dt,
3076 ValueType ResTy, ValueType OpTy>
3077 : NVDupLane<op19_16, 1, (outs QPR:$dst), (ins DPR:$src, nohash_imm:$lane),
3078 IIC_VMOVD, OpcodeStr, Dt, "$dst, $src[$lane]",
3079 [(set QPR:$dst, (ResTy (NEONvduplane (OpTy DPR:$src),
3082 // Inst{19-16} is partially specified depending on the element size.
3084 def VDUPLN8d : VDUPLND<{?,?,?,1}, "vdup", "8", v8i8>;
3085 def VDUPLN16d : VDUPLND<{?,?,1,0}, "vdup", "16", v4i16>;
3086 def VDUPLN32d : VDUPLND<{?,1,0,0}, "vdup", "32", v2i32>;
3087 def VDUPLNfd : VDUPLND<{?,1,0,0}, "vdup", "32", v2f32>;
3088 def VDUPLN8q : VDUPLNQ<{?,?,?,1}, "vdup", "8", v16i8, v8i8>;
3089 def VDUPLN16q : VDUPLNQ<{?,?,1,0}, "vdup", "16", v8i16, v4i16>;
3090 def VDUPLN32q : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4i32, v2i32>;
3091 def VDUPLNfq : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4f32, v2f32>;
3093 def : Pat<(v16i8 (NEONvduplane (v16i8 QPR:$src), imm:$lane)),
3094 (v16i8 (VDUPLN8q (v8i8 (EXTRACT_SUBREG QPR:$src,
3095 (DSubReg_i8_reg imm:$lane))),
3096 (SubReg_i8_lane imm:$lane)))>;
3097 def : Pat<(v8i16 (NEONvduplane (v8i16 QPR:$src), imm:$lane)),
3098 (v8i16 (VDUPLN16q (v4i16 (EXTRACT_SUBREG QPR:$src,
3099 (DSubReg_i16_reg imm:$lane))),
3100 (SubReg_i16_lane imm:$lane)))>;
3101 def : Pat<(v4i32 (NEONvduplane (v4i32 QPR:$src), imm:$lane)),
3102 (v4i32 (VDUPLN32q (v2i32 (EXTRACT_SUBREG QPR:$src,
3103 (DSubReg_i32_reg imm:$lane))),
3104 (SubReg_i32_lane imm:$lane)))>;
3105 def : Pat<(v4f32 (NEONvduplane (v4f32 QPR:$src), imm:$lane)),
3106 (v4f32 (VDUPLNfq (v2f32 (EXTRACT_SUBREG QPR:$src,
3107 (DSubReg_i32_reg imm:$lane))),
3108 (SubReg_i32_lane imm:$lane)))>;
3110 def VDUPfdf : N2V<0b11, 0b11, {?,1}, {0,0}, 0b11000, 0, 0,
3111 (outs DPR:$dst), (ins SPR:$src),
3112 IIC_VMOVD, "vdup", "32", "$dst, ${src:lane}", "",
3113 [(set DPR:$dst, (v2f32 (NEONvdup (f32 SPR:$src))))]>;
3115 def VDUPfqf : N2V<0b11, 0b11, {?,1}, {0,0}, 0b11000, 1, 0,
3116 (outs QPR:$dst), (ins SPR:$src),
3117 IIC_VMOVD, "vdup", "32", "$dst, ${src:lane}", "",
3118 [(set QPR:$dst, (v4f32 (NEONvdup (f32 SPR:$src))))]>;
3120 // VMOVN : Vector Narrowing Move
3121 defm VMOVN : N2VNInt_HSD<0b11,0b11,0b10,0b00100,0,0, IIC_VMOVD,
3122 "vmovn", "i", int_arm_neon_vmovn>;
3123 // VQMOVN : Vector Saturating Narrowing Move
3124 defm VQMOVNs : N2VNInt_HSD<0b11,0b11,0b10,0b00101,0,0, IIC_VQUNAiD,
3125 "vqmovn", "s", int_arm_neon_vqmovns>;
3126 defm VQMOVNu : N2VNInt_HSD<0b11,0b11,0b10,0b00101,1,0, IIC_VQUNAiD,
3127 "vqmovn", "u", int_arm_neon_vqmovnu>;
3128 defm VQMOVNsu : N2VNInt_HSD<0b11,0b11,0b10,0b00100,1,0, IIC_VQUNAiD,
3129 "vqmovun", "s", int_arm_neon_vqmovnsu>;
3130 // VMOVL : Vector Lengthening Move
3131 defm VMOVLs : N2VLInt_QHS<0b01,0b10100,0,1, "vmovl", "s",
3132 int_arm_neon_vmovls>;
3133 defm VMOVLu : N2VLInt_QHS<0b11,0b10100,0,1, "vmovl", "u",
3134 int_arm_neon_vmovlu>;
3136 // Vector Conversions.
3138 // VCVT : Vector Convert Between Floating-Point and Integers
3139 def VCVTf2sd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
3140 v2i32, v2f32, fp_to_sint>;
3141 def VCVTf2ud : N2VD<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
3142 v2i32, v2f32, fp_to_uint>;
3143 def VCVTs2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
3144 v2f32, v2i32, sint_to_fp>;
3145 def VCVTu2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
3146 v2f32, v2i32, uint_to_fp>;
3148 def VCVTf2sq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
3149 v4i32, v4f32, fp_to_sint>;
3150 def VCVTf2uq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
3151 v4i32, v4f32, fp_to_uint>;
3152 def VCVTs2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
3153 v4f32, v4i32, sint_to_fp>;
3154 def VCVTu2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
3155 v4f32, v4i32, uint_to_fp>;
3157 // VCVT : Vector Convert Between Floating-Point and Fixed-Point.
3158 def VCVTf2xsd : N2VCvtD<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
3159 v2i32, v2f32, int_arm_neon_vcvtfp2fxs>;
3160 def VCVTf2xud : N2VCvtD<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
3161 v2i32, v2f32, int_arm_neon_vcvtfp2fxu>;
3162 def VCVTxs2fd : N2VCvtD<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
3163 v2f32, v2i32, int_arm_neon_vcvtfxs2fp>;
3164 def VCVTxu2fd : N2VCvtD<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
3165 v2f32, v2i32, int_arm_neon_vcvtfxu2fp>;
3167 def VCVTf2xsq : N2VCvtQ<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
3168 v4i32, v4f32, int_arm_neon_vcvtfp2fxs>;
3169 def VCVTf2xuq : N2VCvtQ<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
3170 v4i32, v4f32, int_arm_neon_vcvtfp2fxu>;
3171 def VCVTxs2fq : N2VCvtQ<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
3172 v4f32, v4i32, int_arm_neon_vcvtfxs2fp>;
3173 def VCVTxu2fq : N2VCvtQ<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
3174 v4f32, v4i32, int_arm_neon_vcvtfxu2fp>;
3178 // VREV64 : Vector Reverse elements within 64-bit doublewords
3180 class VREV64D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3181 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 0, 0, (outs DPR:$dst),
3182 (ins DPR:$src), IIC_VMOVD,
3183 OpcodeStr, Dt, "$dst, $src", "",
3184 [(set DPR:$dst, (Ty (NEONvrev64 (Ty DPR:$src))))]>;
3185 class VREV64Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3186 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 1, 0, (outs QPR:$dst),
3187 (ins QPR:$src), IIC_VMOVD,
3188 OpcodeStr, Dt, "$dst, $src", "",
3189 [(set QPR:$dst, (Ty (NEONvrev64 (Ty QPR:$src))))]>;
3191 def VREV64d8 : VREV64D<0b00, "vrev64", "8", v8i8>;
3192 def VREV64d16 : VREV64D<0b01, "vrev64", "16", v4i16>;
3193 def VREV64d32 : VREV64D<0b10, "vrev64", "32", v2i32>;
3194 def VREV64df : VREV64D<0b10, "vrev64", "32", v2f32>;
3196 def VREV64q8 : VREV64Q<0b00, "vrev64", "8", v16i8>;
3197 def VREV64q16 : VREV64Q<0b01, "vrev64", "16", v8i16>;
3198 def VREV64q32 : VREV64Q<0b10, "vrev64", "32", v4i32>;
3199 def VREV64qf : VREV64Q<0b10, "vrev64", "32", v4f32>;
3201 // VREV32 : Vector Reverse elements within 32-bit words
3203 class VREV32D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3204 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 0, 0, (outs DPR:$dst),
3205 (ins DPR:$src), IIC_VMOVD,
3206 OpcodeStr, Dt, "$dst, $src", "",
3207 [(set DPR:$dst, (Ty (NEONvrev32 (Ty DPR:$src))))]>;
3208 class VREV32Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3209 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 1, 0, (outs QPR:$dst),
3210 (ins QPR:$src), IIC_VMOVD,
3211 OpcodeStr, Dt, "$dst, $src", "",
3212 [(set QPR:$dst, (Ty (NEONvrev32 (Ty QPR:$src))))]>;
3214 def VREV32d8 : VREV32D<0b00, "vrev32", "8", v8i8>;
3215 def VREV32d16 : VREV32D<0b01, "vrev32", "16", v4i16>;
3217 def VREV32q8 : VREV32Q<0b00, "vrev32", "8", v16i8>;
3218 def VREV32q16 : VREV32Q<0b01, "vrev32", "16", v8i16>;
3220 // VREV16 : Vector Reverse elements within 16-bit halfwords
3222 class VREV16D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3223 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 0, 0, (outs DPR:$dst),
3224 (ins DPR:$src), IIC_VMOVD,
3225 OpcodeStr, Dt, "$dst, $src", "",
3226 [(set DPR:$dst, (Ty (NEONvrev16 (Ty DPR:$src))))]>;
3227 class VREV16Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3228 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 1, 0, (outs QPR:$dst),
3229 (ins QPR:$src), IIC_VMOVD,
3230 OpcodeStr, Dt, "$dst, $src", "",
3231 [(set QPR:$dst, (Ty (NEONvrev16 (Ty QPR:$src))))]>;
3233 def VREV16d8 : VREV16D<0b00, "vrev16", "8", v8i8>;
3234 def VREV16q8 : VREV16Q<0b00, "vrev16", "8", v16i8>;
3236 // Other Vector Shuffles.
3238 // VEXT : Vector Extract
3240 class VEXTd<string OpcodeStr, string Dt, ValueType Ty>
3241 : N3V<0,1,0b11,{?,?,?,?},0,0, (outs DPR:$dst),
3242 (ins DPR:$lhs, DPR:$rhs, i32imm:$index), NVExtFrm,
3243 IIC_VEXTD, OpcodeStr, Dt, "$dst, $lhs, $rhs, $index", "",
3244 [(set DPR:$dst, (Ty (NEONvext (Ty DPR:$lhs),
3245 (Ty DPR:$rhs), imm:$index)))]>;
3247 class VEXTq<string OpcodeStr, string Dt, ValueType Ty>
3248 : N3V<0,1,0b11,{?,?,?,?},1,0, (outs QPR:$dst),
3249 (ins QPR:$lhs, QPR:$rhs, i32imm:$index), NVExtFrm,
3250 IIC_VEXTQ, OpcodeStr, Dt, "$dst, $lhs, $rhs, $index", "",
3251 [(set QPR:$dst, (Ty (NEONvext (Ty QPR:$lhs),
3252 (Ty QPR:$rhs), imm:$index)))]>;
3254 def VEXTd8 : VEXTd<"vext", "8", v8i8>;
3255 def VEXTd16 : VEXTd<"vext", "16", v4i16>;
3256 def VEXTd32 : VEXTd<"vext", "32", v2i32>;
3257 def VEXTdf : VEXTd<"vext", "32", v2f32>;
3259 def VEXTq8 : VEXTq<"vext", "8", v16i8>;
3260 def VEXTq16 : VEXTq<"vext", "16", v8i16>;
3261 def VEXTq32 : VEXTq<"vext", "32", v4i32>;
3262 def VEXTqf : VEXTq<"vext", "32", v4f32>;
3264 // VTRN : Vector Transpose
3266 def VTRNd8 : N2VDShuffle<0b00, 0b00001, "vtrn", "8">;
3267 def VTRNd16 : N2VDShuffle<0b01, 0b00001, "vtrn", "16">;
3268 def VTRNd32 : N2VDShuffle<0b10, 0b00001, "vtrn", "32">;
3270 def VTRNq8 : N2VQShuffle<0b00, 0b00001, IIC_VPERMQ, "vtrn", "8">;
3271 def VTRNq16 : N2VQShuffle<0b01, 0b00001, IIC_VPERMQ, "vtrn", "16">;
3272 def VTRNq32 : N2VQShuffle<0b10, 0b00001, IIC_VPERMQ, "vtrn", "32">;
3274 // VUZP : Vector Unzip (Deinterleave)
3276 def VUZPd8 : N2VDShuffle<0b00, 0b00010, "vuzp", "8">;
3277 def VUZPd16 : N2VDShuffle<0b01, 0b00010, "vuzp", "16">;
3278 def VUZPd32 : N2VDShuffle<0b10, 0b00010, "vuzp", "32">;
3280 def VUZPq8 : N2VQShuffle<0b00, 0b00010, IIC_VPERMQ3, "vuzp", "8">;
3281 def VUZPq16 : N2VQShuffle<0b01, 0b00010, IIC_VPERMQ3, "vuzp", "16">;
3282 def VUZPq32 : N2VQShuffle<0b10, 0b00010, IIC_VPERMQ3, "vuzp", "32">;
3284 // VZIP : Vector Zip (Interleave)
3286 def VZIPd8 : N2VDShuffle<0b00, 0b00011, "vzip", "8">;
3287 def VZIPd16 : N2VDShuffle<0b01, 0b00011, "vzip", "16">;
3288 def VZIPd32 : N2VDShuffle<0b10, 0b00011, "vzip", "32">;
3290 def VZIPq8 : N2VQShuffle<0b00, 0b00011, IIC_VPERMQ3, "vzip", "8">;
3291 def VZIPq16 : N2VQShuffle<0b01, 0b00011, IIC_VPERMQ3, "vzip", "16">;
3292 def VZIPq32 : N2VQShuffle<0b10, 0b00011, IIC_VPERMQ3, "vzip", "32">;
3294 // Vector Table Lookup and Table Extension.
3296 // VTBL : Vector Table Lookup
3298 : N3V<1,1,0b11,0b1000,0,0, (outs DPR:$dst),
3299 (ins DPR:$tbl1, DPR:$src), NVTBLFrm, IIC_VTB1,
3300 "vtbl", "8", "$dst, \\{$tbl1\\}, $src", "",
3301 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl1 DPR:$tbl1, DPR:$src)))]>;
3302 let hasExtraSrcRegAllocReq = 1 in {
3304 : N3V<1,1,0b11,0b1001,0,0, (outs DPR:$dst),
3305 (ins DPR:$tbl1, DPR:$tbl2, DPR:$src), NVTBLFrm, IIC_VTB2,
3306 "vtbl", "8", "$dst, \\{$tbl1, $tbl2\\}, $src", "",
3307 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl2
3308 DPR:$tbl1, DPR:$tbl2, DPR:$src)))]>;
3310 : N3V<1,1,0b11,0b1010,0,0, (outs DPR:$dst),
3311 (ins DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$src), NVTBLFrm, IIC_VTB3,
3312 "vtbl", "8", "$dst, \\{$tbl1, $tbl2, $tbl3\\}, $src", "",
3313 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl3
3314 DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$src)))]>;
3316 : N3V<1,1,0b11,0b1011,0,0, (outs DPR:$dst),
3317 (ins DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$src),
3319 "vtbl", "8", "$dst, \\{$tbl1, $tbl2, $tbl3, $tbl4\\}, $src", "",
3320 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl4 DPR:$tbl1, DPR:$tbl2,
3321 DPR:$tbl3, DPR:$tbl4, DPR:$src)))]>;
3322 } // hasExtraSrcRegAllocReq = 1
3324 // VTBX : Vector Table Extension
3326 : N3V<1,1,0b11,0b1000,1,0, (outs DPR:$dst),
3327 (ins DPR:$orig, DPR:$tbl1, DPR:$src), NVTBLFrm, IIC_VTBX1,
3328 "vtbx", "8", "$dst, \\{$tbl1\\}, $src", "$orig = $dst",
3329 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx1
3330 DPR:$orig, DPR:$tbl1, DPR:$src)))]>;
3331 let hasExtraSrcRegAllocReq = 1 in {
3333 : N3V<1,1,0b11,0b1001,1,0, (outs DPR:$dst),
3334 (ins DPR:$orig, DPR:$tbl1, DPR:$tbl2, DPR:$src), NVTBLFrm, IIC_VTBX2,
3335 "vtbx", "8", "$dst, \\{$tbl1, $tbl2\\}, $src", "$orig = $dst",
3336 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx2
3337 DPR:$orig, DPR:$tbl1, DPR:$tbl2, DPR:$src)))]>;
3339 : N3V<1,1,0b11,0b1010,1,0, (outs DPR:$dst),
3340 (ins DPR:$orig, DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$src),
3341 NVTBLFrm, IIC_VTBX3,
3342 "vtbx", "8", "$dst, \\{$tbl1, $tbl2, $tbl3\\}, $src", "$orig = $dst",
3343 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx3 DPR:$orig, DPR:$tbl1,
3344 DPR:$tbl2, DPR:$tbl3, DPR:$src)))]>;
3346 : N3V<1,1,0b11,0b1011,1,0, (outs DPR:$dst), (ins DPR:$orig, DPR:$tbl1,
3347 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$src), NVTBLFrm, IIC_VTBX4,
3348 "vtbx", "8", "$dst, \\{$tbl1, $tbl2, $tbl3, $tbl4\\}, $src",
3350 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx4 DPR:$orig, DPR:$tbl1,
3351 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$src)))]>;
3352 } // hasExtraSrcRegAllocReq = 1
3354 //===----------------------------------------------------------------------===//
3355 // NEON instructions for single-precision FP math
3356 //===----------------------------------------------------------------------===//
3358 class N2VSPat<SDNode OpNode, ValueType ResTy, ValueType OpTy, NeonI Inst>
3359 : NEONFPPat<(ResTy (OpNode SPR:$a)),
3360 (EXTRACT_SUBREG (OpTy (Inst (INSERT_SUBREG (OpTy (IMPLICIT_DEF)),
3364 class N3VSPat<SDNode OpNode, NeonI Inst>
3365 : NEONFPPat<(f32 (OpNode SPR:$a, SPR:$b)),
3366 (EXTRACT_SUBREG (v2f32
3367 (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3369 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3373 class N3VSMulOpPat<SDNode MulNode, SDNode OpNode, NeonI Inst>
3374 : NEONFPPat<(f32 (OpNode SPR:$acc, (f32 (MulNode SPR:$a, SPR:$b)))),
3375 (EXTRACT_SUBREG (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3377 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3379 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3383 // These need separate instructions because they must use DPR_VFP2 register
3384 // class which have SPR sub-registers.
3386 // Vector Add Operations used for single-precision FP
3387 let neverHasSideEffects = 1 in
3388 def VADDfd_sfp : N3VS<0,0,0b00,0b1101,0, "vadd", "f32", v2f32, v2f32, fadd, 1>;
3389 def : N3VSPat<fadd, VADDfd_sfp>;
3391 // Vector Sub Operations used for single-precision FP
3392 let neverHasSideEffects = 1 in
3393 def VSUBfd_sfp : N3VS<0,0,0b10,0b1101,0, "vsub", "f32", v2f32, v2f32, fsub, 0>;
3394 def : N3VSPat<fsub, VSUBfd_sfp>;
3396 // Vector Multiply Operations used for single-precision FP
3397 let neverHasSideEffects = 1 in
3398 def VMULfd_sfp : N3VS<1,0,0b00,0b1101,1, "vmul", "f32", v2f32, v2f32, fmul, 1>;
3399 def : N3VSPat<fmul, VMULfd_sfp>;
3401 // Vector Multiply-Accumulate/Subtract used for single-precision FP
3402 // vml[as].f32 can cause 4-8 cycle stalls in following ASIMD instructions, so
3403 // we want to avoid them for now. e.g., alternating vmla/vadd instructions.
3405 //let neverHasSideEffects = 1 in
3406 //def VMLAfd_sfp : N3VSMulOp<0,0,0b00,0b1101,1, IIC_VMACD, "vmla", "f32",
3407 // v2f32, fmul, fadd>;
3408 //def : N3VSMulOpPat<fmul, fadd, VMLAfd_sfp>;
3410 //let neverHasSideEffects = 1 in
3411 //def VMLSfd_sfp : N3VSMulOp<0,0,0b10,0b1101,1, IIC_VMACD, "vmls", "f32",
3412 // v2f32, fmul, fsub>;
3413 //def : N3VSMulOpPat<fmul, fsub, VMLSfd_sfp>;
3415 // Vector Absolute used for single-precision FP
3416 let neverHasSideEffects = 1 in
3417 def VABSfd_sfp : N2V<0b11, 0b11, 0b10, 0b01, 0b01110, 0, 0,
3418 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), IIC_VUNAD,
3419 "vabs", "f32", "$dst, $src", "", []>;
3420 def : N2VSPat<fabs, f32, v2f32, VABSfd_sfp>;
3422 // Vector Negate used for single-precision FP
3423 let neverHasSideEffects = 1 in
3424 def VNEGfd_sfp : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
3425 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), IIC_VUNAD,
3426 "vneg", "f32", "$dst, $src", "", []>;
3427 def : N2VSPat<fneg, f32, v2f32, VNEGfd_sfp>;
3429 // Vector Maximum used for single-precision FP
3430 let neverHasSideEffects = 1 in
3431 def VMAXfd_sfp : N3V<0, 0, 0b00, 0b1111, 0, 0, (outs DPR_VFP2:$dst),
3432 (ins DPR_VFP2:$src1, DPR_VFP2:$src2), N3RegFrm, IIC_VBIND,
3433 "vmax", "f32", "$dst, $src1, $src2", "", []>;
3434 def : N3VSPat<NEONfmax, VMAXfd_sfp>;
3436 // Vector Minimum used for single-precision FP
3437 let neverHasSideEffects = 1 in
3438 def VMINfd_sfp : N3V<0, 0, 0b00, 0b1111, 0, 0, (outs DPR_VFP2:$dst),
3439 (ins DPR_VFP2:$src1, DPR_VFP2:$src2), N3RegFrm, IIC_VBIND,
3440 "vmin", "f32", "$dst, $src1, $src2", "", []>;
3441 def : N3VSPat<NEONfmin, VMINfd_sfp>;
3443 // Vector Convert between single-precision FP and integer
3444 let neverHasSideEffects = 1 in
3445 def VCVTf2sd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
3446 v2i32, v2f32, fp_to_sint>;
3447 def : N2VSPat<arm_ftosi, f32, v2f32, VCVTf2sd_sfp>;
3449 let neverHasSideEffects = 1 in
3450 def VCVTf2ud_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
3451 v2i32, v2f32, fp_to_uint>;
3452 def : N2VSPat<arm_ftoui, f32, v2f32, VCVTf2ud_sfp>;
3454 let neverHasSideEffects = 1 in
3455 def VCVTs2fd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
3456 v2f32, v2i32, sint_to_fp>;
3457 def : N2VSPat<arm_sitof, f32, v2i32, VCVTs2fd_sfp>;
3459 let neverHasSideEffects = 1 in
3460 def VCVTu2fd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
3461 v2f32, v2i32, uint_to_fp>;
3462 def : N2VSPat<arm_uitof, f32, v2i32, VCVTu2fd_sfp>;
3464 //===----------------------------------------------------------------------===//
3465 // Non-Instruction Patterns
3466 //===----------------------------------------------------------------------===//
3469 def : Pat<(v1i64 (bitconvert (v2i32 DPR:$src))), (v1i64 DPR:$src)>;
3470 def : Pat<(v1i64 (bitconvert (v4i16 DPR:$src))), (v1i64 DPR:$src)>;
3471 def : Pat<(v1i64 (bitconvert (v8i8 DPR:$src))), (v1i64 DPR:$src)>;
3472 def : Pat<(v1i64 (bitconvert (f64 DPR:$src))), (v1i64 DPR:$src)>;
3473 def : Pat<(v1i64 (bitconvert (v2f32 DPR:$src))), (v1i64 DPR:$src)>;
3474 def : Pat<(v2i32 (bitconvert (v1i64 DPR:$src))), (v2i32 DPR:$src)>;
3475 def : Pat<(v2i32 (bitconvert (v4i16 DPR:$src))), (v2i32 DPR:$src)>;
3476 def : Pat<(v2i32 (bitconvert (v8i8 DPR:$src))), (v2i32 DPR:$src)>;
3477 def : Pat<(v2i32 (bitconvert (f64 DPR:$src))), (v2i32 DPR:$src)>;
3478 def : Pat<(v2i32 (bitconvert (v2f32 DPR:$src))), (v2i32 DPR:$src)>;
3479 def : Pat<(v4i16 (bitconvert (v1i64 DPR:$src))), (v4i16 DPR:$src)>;
3480 def : Pat<(v4i16 (bitconvert (v2i32 DPR:$src))), (v4i16 DPR:$src)>;
3481 def : Pat<(v4i16 (bitconvert (v8i8 DPR:$src))), (v4i16 DPR:$src)>;
3482 def : Pat<(v4i16 (bitconvert (f64 DPR:$src))), (v4i16 DPR:$src)>;
3483 def : Pat<(v4i16 (bitconvert (v2f32 DPR:$src))), (v4i16 DPR:$src)>;
3484 def : Pat<(v8i8 (bitconvert (v1i64 DPR:$src))), (v8i8 DPR:$src)>;
3485 def : Pat<(v8i8 (bitconvert (v2i32 DPR:$src))), (v8i8 DPR:$src)>;
3486 def : Pat<(v8i8 (bitconvert (v4i16 DPR:$src))), (v8i8 DPR:$src)>;
3487 def : Pat<(v8i8 (bitconvert (f64 DPR:$src))), (v8i8 DPR:$src)>;
3488 def : Pat<(v8i8 (bitconvert (v2f32 DPR:$src))), (v8i8 DPR:$src)>;
3489 def : Pat<(f64 (bitconvert (v1i64 DPR:$src))), (f64 DPR:$src)>;
3490 def : Pat<(f64 (bitconvert (v2i32 DPR:$src))), (f64 DPR:$src)>;
3491 def : Pat<(f64 (bitconvert (v4i16 DPR:$src))), (f64 DPR:$src)>;
3492 def : Pat<(f64 (bitconvert (v8i8 DPR:$src))), (f64 DPR:$src)>;
3493 def : Pat<(f64 (bitconvert (v2f32 DPR:$src))), (f64 DPR:$src)>;
3494 def : Pat<(v2f32 (bitconvert (f64 DPR:$src))), (v2f32 DPR:$src)>;
3495 def : Pat<(v2f32 (bitconvert (v1i64 DPR:$src))), (v2f32 DPR:$src)>;
3496 def : Pat<(v2f32 (bitconvert (v2i32 DPR:$src))), (v2f32 DPR:$src)>;
3497 def : Pat<(v2f32 (bitconvert (v4i16 DPR:$src))), (v2f32 DPR:$src)>;
3498 def : Pat<(v2f32 (bitconvert (v8i8 DPR:$src))), (v2f32 DPR:$src)>;
3500 def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (v2i64 QPR:$src)>;
3501 def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (v2i64 QPR:$src)>;
3502 def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (v2i64 QPR:$src)>;
3503 def : Pat<(v2i64 (bitconvert (v2f64 QPR:$src))), (v2i64 QPR:$src)>;
3504 def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (v2i64 QPR:$src)>;
3505 def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (v4i32 QPR:$src)>;
3506 def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (v4i32 QPR:$src)>;
3507 def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (v4i32 QPR:$src)>;
3508 def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (v4i32 QPR:$src)>;
3509 def : Pat<(v4i32 (bitconvert (v4f32 QPR:$src))), (v4i32 QPR:$src)>;
3510 def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (v8i16 QPR:$src)>;
3511 def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (v8i16 QPR:$src)>;
3512 def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (v8i16 QPR:$src)>;
3513 def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (v8i16 QPR:$src)>;
3514 def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (v8i16 QPR:$src)>;
3515 def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (v16i8 QPR:$src)>;
3516 def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (v16i8 QPR:$src)>;
3517 def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (v16i8 QPR:$src)>;
3518 def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (v16i8 QPR:$src)>;
3519 def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (v16i8 QPR:$src)>;
3520 def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (v4f32 QPR:$src)>;
3521 def : Pat<(v4f32 (bitconvert (v4i32 QPR:$src))), (v4f32 QPR:$src)>;
3522 def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (v4f32 QPR:$src)>;
3523 def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (v4f32 QPR:$src)>;
3524 def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (v4f32 QPR:$src)>;
3525 def : Pat<(v2f64 (bitconvert (v2i64 QPR:$src))), (v2f64 QPR:$src)>;
3526 def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (v2f64 QPR:$src)>;
3527 def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 QPR:$src)>;
3528 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 QPR:$src)>;
3529 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 QPR:$src)>;