1 //===- ARMInstrNEON.td - NEON support for ARM -----------------------------===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file describes the ARM NEON instruction set.
12 //===----------------------------------------------------------------------===//
14 //===----------------------------------------------------------------------===//
15 // NEON-specific DAG Nodes.
16 //===----------------------------------------------------------------------===//
18 def SDTARMVCMP : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisSameAs<1, 2>]>;
20 def NEONvceq : SDNode<"ARMISD::VCEQ", SDTARMVCMP>;
21 def NEONvcge : SDNode<"ARMISD::VCGE", SDTARMVCMP>;
22 def NEONvcgeu : SDNode<"ARMISD::VCGEU", SDTARMVCMP>;
23 def NEONvcgt : SDNode<"ARMISD::VCGT", SDTARMVCMP>;
24 def NEONvcgtu : SDNode<"ARMISD::VCGTU", SDTARMVCMP>;
25 def NEONvtst : SDNode<"ARMISD::VTST", SDTARMVCMP>;
27 // Types for vector shift by immediates. The "SHX" version is for long and
28 // narrow operations where the source and destination vectors have different
29 // types. The "SHINS" version is for shift and insert operations.
30 def SDTARMVSH : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
32 def SDTARMVSHX : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
34 def SDTARMVSHINS : SDTypeProfile<1, 3, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
35 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
37 def NEONvshl : SDNode<"ARMISD::VSHL", SDTARMVSH>;
38 def NEONvshrs : SDNode<"ARMISD::VSHRs", SDTARMVSH>;
39 def NEONvshru : SDNode<"ARMISD::VSHRu", SDTARMVSH>;
40 def NEONvshlls : SDNode<"ARMISD::VSHLLs", SDTARMVSHX>;
41 def NEONvshllu : SDNode<"ARMISD::VSHLLu", SDTARMVSHX>;
42 def NEONvshlli : SDNode<"ARMISD::VSHLLi", SDTARMVSHX>;
43 def NEONvshrn : SDNode<"ARMISD::VSHRN", SDTARMVSHX>;
45 def NEONvrshrs : SDNode<"ARMISD::VRSHRs", SDTARMVSH>;
46 def NEONvrshru : SDNode<"ARMISD::VRSHRu", SDTARMVSH>;
47 def NEONvrshrn : SDNode<"ARMISD::VRSHRN", SDTARMVSHX>;
49 def NEONvqshls : SDNode<"ARMISD::VQSHLs", SDTARMVSH>;
50 def NEONvqshlu : SDNode<"ARMISD::VQSHLu", SDTARMVSH>;
51 def NEONvqshlsu : SDNode<"ARMISD::VQSHLsu", SDTARMVSH>;
52 def NEONvqshrns : SDNode<"ARMISD::VQSHRNs", SDTARMVSHX>;
53 def NEONvqshrnu : SDNode<"ARMISD::VQSHRNu", SDTARMVSHX>;
54 def NEONvqshrnsu : SDNode<"ARMISD::VQSHRNsu", SDTARMVSHX>;
56 def NEONvqrshrns : SDNode<"ARMISD::VQRSHRNs", SDTARMVSHX>;
57 def NEONvqrshrnu : SDNode<"ARMISD::VQRSHRNu", SDTARMVSHX>;
58 def NEONvqrshrnsu : SDNode<"ARMISD::VQRSHRNsu", SDTARMVSHX>;
60 def NEONvsli : SDNode<"ARMISD::VSLI", SDTARMVSHINS>;
61 def NEONvsri : SDNode<"ARMISD::VSRI", SDTARMVSHINS>;
63 def SDTARMVGETLN : SDTypeProfile<1, 2, [SDTCisVT<0, i32>, SDTCisInt<1>,
65 def NEONvgetlaneu : SDNode<"ARMISD::VGETLANEu", SDTARMVGETLN>;
66 def NEONvgetlanes : SDNode<"ARMISD::VGETLANEs", SDTARMVGETLN>;
68 def NEONvdup : SDNode<"ARMISD::VDUP", SDTypeProfile<1, 1, [SDTCisVec<0>]>>;
70 // VDUPLANE can produce a quad-register result from a double-register source,
71 // so the result is not constrained to match the source.
72 def NEONvduplane : SDNode<"ARMISD::VDUPLANE",
73 SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisVec<1>,
76 def SDTARMVEXT : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
77 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
78 def NEONvext : SDNode<"ARMISD::VEXT", SDTARMVEXT>;
80 def SDTARMVSHUF : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisSameAs<0, 1>]>;
81 def NEONvrev64 : SDNode<"ARMISD::VREV64", SDTARMVSHUF>;
82 def NEONvrev32 : SDNode<"ARMISD::VREV32", SDTARMVSHUF>;
83 def NEONvrev16 : SDNode<"ARMISD::VREV16", SDTARMVSHUF>;
85 def SDTARMVSHUF2 : SDTypeProfile<2, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
88 def NEONzip : SDNode<"ARMISD::VZIP", SDTARMVSHUF2>;
89 def NEONuzp : SDNode<"ARMISD::VUZP", SDTARMVSHUF2>;
90 def NEONtrn : SDNode<"ARMISD::VTRN", SDTARMVSHUF2>;
92 def SDTARMFMAX : SDTypeProfile<1, 2, [SDTCisVT<0, f32>, SDTCisSameAs<0, 1>,
94 def NEONfmax : SDNode<"ARMISD::FMAX", SDTARMFMAX>;
95 def NEONfmin : SDNode<"ARMISD::FMIN", SDTARMFMAX>;
97 //===----------------------------------------------------------------------===//
98 // NEON operand definitions
99 //===----------------------------------------------------------------------===//
101 def h8imm : Operand<i8> {
102 let PrintMethod = "printHex8ImmOperand";
104 def h16imm : Operand<i16> {
105 let PrintMethod = "printHex16ImmOperand";
107 def h32imm : Operand<i32> {
108 let PrintMethod = "printHex32ImmOperand";
110 def h64imm : Operand<i64> {
111 let PrintMethod = "printHex64ImmOperand";
114 //===----------------------------------------------------------------------===//
115 // NEON load / store instructions
116 //===----------------------------------------------------------------------===//
119 // Use vldmia to load a Q register as a D register pair.
120 // This is equivalent to VLDMD except that it has a Q register operand
121 // instead of a pair of D registers.
123 : AXDI5<(outs QPR:$dst), (ins addrmode5:$addr, pred:$p),
124 IndexModeNone, IIC_fpLoadm,
125 "vldm${addr:submode}${p}\t${addr:base}, ${dst:dregpair}", "", []>;
128 let mayStore = 1 in {
129 // Use vstmia to store a Q register as a D register pair.
130 // This is equivalent to VSTMD except that it has a Q register operand
131 // instead of a pair of D registers.
133 : AXDI5<(outs), (ins QPR:$src, addrmode5:$addr, pred:$p),
134 IndexModeNone, IIC_fpStorem,
135 "vstm${addr:submode}${p}\t${addr:base}, ${src:dregpair}", "", []>;
138 let mayLoad = 1, hasExtraDefRegAllocReq = 1 in {
140 // VLD1 : Vector Load (multiple single elements)
141 class VLD1D<bits<4> op7_4, string Dt>
142 : NLdSt<0,0b10,0b0111,op7_4, (outs DPR:$dst),
143 (ins addrmode6:$addr), IIC_VLD1,
144 "vld1", Dt, "\\{$dst\\}, $addr", "", []>;
145 class VLD1Q<bits<4> op7_4, string Dt>
146 : NLdSt<0,0b10,0b1010,op7_4, (outs DPR:$dst1, DPR:$dst2),
147 (ins addrmode6:$addr), IIC_VLD1,
148 "vld1", Dt, "\\{$dst1, $dst2\\}, $addr", "", []>;
150 def VLD1d8 : VLD1D<0b0000, "8">;
151 def VLD1d16 : VLD1D<0b0100, "16">;
152 def VLD1d32 : VLD1D<0b1000, "32">;
153 def VLD1d64 : VLD1D<0b1100, "64">;
155 def VLD1q8 : VLD1Q<0b0000, "8">;
156 def VLD1q16 : VLD1Q<0b0100, "16">;
157 def VLD1q32 : VLD1Q<0b1000, "32">;
158 def VLD1q64 : VLD1Q<0b1100, "64">;
160 // ...with address register writeback:
161 class VLD1DWB<bits<4> op7_4, string Dt>
162 : NLdSt<0,0b10,0b0111,op7_4, (outs DPR:$dst, GPR:$wb),
163 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD1,
164 "vld1", Dt, "\\{$dst\\}, $addr$offset",
165 "$addr.addr = $wb", []>;
166 class VLD1QWB<bits<4> op7_4, string Dt>
167 : NLdSt<0,0b10,0b1010,op7_4, (outs QPR:$dst, GPR:$wb),
168 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD1,
169 "vld1", Dt, "${dst:dregpair}, $addr$offset",
170 "$addr.addr = $wb", []>;
172 def VLD1d8_UPD : VLD1DWB<0b0000, "8">;
173 def VLD1d16_UPD : VLD1DWB<0b0100, "16">;
174 def VLD1d32_UPD : VLD1DWB<0b1000, "32">;
175 def VLD1d64_UPD : VLD1DWB<0b1100, "64">;
177 def VLD1q8_UPD : VLD1QWB<0b0000, "8">;
178 def VLD1q16_UPD : VLD1QWB<0b0100, "16">;
179 def VLD1q32_UPD : VLD1QWB<0b1000, "32">;
180 def VLD1q64_UPD : VLD1QWB<0b1100, "64">;
182 // ...with 3 registers (some of these are only for the disassembler):
183 class VLD1D3<bits<4> op7_4, string Dt>
184 : NLdSt<0,0b10,0b0110,op7_4, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3),
185 (ins addrmode6:$addr), IIC_VLD1, "vld1", Dt,
186 "\\{$dst1, $dst2, $dst3\\}, $addr", "", []>;
187 class VLD1D3WB<bits<4> op7_4, string Dt>
188 : NLdSt<0,0b10,0b0110,op7_4, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, GPR:$wb),
189 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD1, "vld1", Dt,
190 "\\{$dst1, $dst2, $dst3\\}, $addr$offset", "$addr.addr = $wb", []>;
192 def VLD1d8T : VLD1D3<0b0000, "8">;
193 def VLD1d16T : VLD1D3<0b0100, "16">;
194 def VLD1d32T : VLD1D3<0b1000, "32">;
195 def VLD1d64T : VLD1D3<0b1100, "64">;
197 def VLD1d8T_UPD : VLD1D3WB<0b0000, "8">;
198 def VLD1d16T_UPD : VLD1D3WB<0b0100, "16">;
199 def VLD1d32T_UPD : VLD1D3WB<0b1000, "32">;
200 def VLD1d64T_UPD : VLD1D3WB<0b1100, "64">;
202 // ...with 4 registers (some of these are only for the disassembler):
203 class VLD1D4<bits<4> op7_4, string Dt>
204 : NLdSt<0,0b10,0b0010,op7_4,(outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
205 (ins addrmode6:$addr), IIC_VLD1, "vld1", Dt,
206 "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr", "", []>;
207 class VLD1D4WB<bits<4> op7_4, string Dt>
208 : NLdSt<0,0b10,0b0010,op7_4,
209 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
210 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD1, "vld1", Dt,
211 "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr$offset", "$addr.addr = $wb",
214 def VLD1d8Q : VLD1D4<0b0000, "8">;
215 def VLD1d16Q : VLD1D4<0b0100, "16">;
216 def VLD1d32Q : VLD1D4<0b1000, "32">;
217 def VLD1d64Q : VLD1D4<0b1100, "64">;
219 def VLD1d8Q_UPD : VLD1D4WB<0b0000, "8">;
220 def VLD1d16Q_UPD : VLD1D4WB<0b0100, "16">;
221 def VLD1d32Q_UPD : VLD1D4WB<0b1000, "32">;
222 def VLD1d64Q_UPD : VLD1D4WB<0b1100, "64">;
224 // VLD2 : Vector Load (multiple 2-element structures)
225 class VLD2D<bits<4> op11_8, bits<4> op7_4, string Dt>
226 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2),
227 (ins addrmode6:$addr), IIC_VLD2,
228 "vld2", Dt, "\\{$dst1, $dst2\\}, $addr", "", []>;
229 class VLD2Q<bits<4> op7_4, string Dt>
230 : NLdSt<0, 0b10, 0b0011, op7_4,
231 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
232 (ins addrmode6:$addr), IIC_VLD2,
233 "vld2", Dt, "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr", "", []>;
235 def VLD2d8 : VLD2D<0b1000, 0b0000, "8">;
236 def VLD2d16 : VLD2D<0b1000, 0b0100, "16">;
237 def VLD2d32 : VLD2D<0b1000, 0b1000, "32">;
239 def VLD2q8 : VLD2Q<0b0000, "8">;
240 def VLD2q16 : VLD2Q<0b0100, "16">;
241 def VLD2q32 : VLD2Q<0b1000, "32">;
243 // ...with address register writeback:
244 class VLD2DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
245 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2, GPR:$wb),
246 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD2,
247 "vld2", Dt, "\\{$dst1, $dst2\\}, $addr$offset",
248 "$addr.addr = $wb", []>;
249 class VLD2QWB<bits<4> op7_4, string Dt>
250 : NLdSt<0, 0b10, 0b0011, op7_4,
251 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
252 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD2,
253 "vld2", Dt, "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr$offset",
254 "$addr.addr = $wb", []>;
256 def VLD2d8_UPD : VLD2DWB<0b1000, 0b0000, "8">;
257 def VLD2d16_UPD : VLD2DWB<0b1000, 0b0100, "16">;
258 def VLD2d32_UPD : VLD2DWB<0b1000, 0b1000, "32">;
260 def VLD2q8_UPD : VLD2QWB<0b0000, "8">;
261 def VLD2q16_UPD : VLD2QWB<0b0100, "16">;
262 def VLD2q32_UPD : VLD2QWB<0b1000, "32">;
264 // ...with double-spaced registers (for disassembly only):
265 def VLD2b8 : VLD2D<0b1001, 0b0000, "8">;
266 def VLD2b16 : VLD2D<0b1001, 0b0100, "16">;
267 def VLD2b32 : VLD2D<0b1001, 0b1000, "32">;
268 def VLD2b8_UPD : VLD2DWB<0b1001, 0b0000, "8">;
269 def VLD2b16_UPD : VLD2DWB<0b1001, 0b0100, "16">;
270 def VLD2b32_UPD : VLD2DWB<0b1001, 0b1000, "32">;
272 // VLD3 : Vector Load (multiple 3-element structures)
273 class VLD3D<bits<4> op11_8, bits<4> op7_4, string Dt>
274 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3),
275 (ins addrmode6:$addr), IIC_VLD3,
276 "vld3", Dt, "\\{$dst1, $dst2, $dst3\\}, $addr", "", []>;
278 def VLD3d8 : VLD3D<0b0100, 0b0000, "8">;
279 def VLD3d16 : VLD3D<0b0100, 0b0100, "16">;
280 def VLD3d32 : VLD3D<0b0100, 0b1000, "32">;
282 // ...with address register writeback:
283 class VLD3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
284 : NLdSt<0, 0b10, op11_8, op7_4,
285 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, GPR:$wb),
286 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD3,
287 "vld3", Dt, "\\{$dst1, $dst2, $dst3\\}, $addr$offset",
288 "$addr.addr = $wb", []>;
290 def VLD3d8_UPD : VLD3DWB<0b0100, 0b0000, "8">;
291 def VLD3d16_UPD : VLD3DWB<0b0100, 0b0100, "16">;
292 def VLD3d32_UPD : VLD3DWB<0b0100, 0b1000, "32">;
294 // ...with double-spaced registers (non-updating versions for disassembly only):
295 def VLD3q8 : VLD3D<0b0101, 0b0000, "8">;
296 def VLD3q16 : VLD3D<0b0101, 0b0100, "16">;
297 def VLD3q32 : VLD3D<0b0101, 0b1000, "32">;
298 def VLD3q8_UPD : VLD3DWB<0b0101, 0b0000, "8">;
299 def VLD3q16_UPD : VLD3DWB<0b0101, 0b0100, "16">;
300 def VLD3q32_UPD : VLD3DWB<0b0101, 0b1000, "32">;
302 // ...alternate versions to be allocated odd register numbers:
303 def VLD3q8odd_UPD : VLD3DWB<0b0101, 0b0000, "8">;
304 def VLD3q16odd_UPD : VLD3DWB<0b0101, 0b0100, "16">;
305 def VLD3q32odd_UPD : VLD3DWB<0b0101, 0b1000, "32">;
307 // VLD4 : Vector Load (multiple 4-element structures)
308 class VLD4D<bits<4> op11_8, bits<4> op7_4, string Dt>
309 : NLdSt<0, 0b10, op11_8, op7_4,
310 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
311 (ins addrmode6:$addr), IIC_VLD4,
312 "vld4", Dt, "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr", "", []>;
314 def VLD4d8 : VLD4D<0b0000, 0b0000, "8">;
315 def VLD4d16 : VLD4D<0b0000, 0b0100, "16">;
316 def VLD4d32 : VLD4D<0b0000, 0b1000, "32">;
318 // ...with address register writeback:
319 class VLD4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
320 : NLdSt<0, 0b10, op11_8, op7_4,
321 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
322 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD4,
323 "vld4", Dt, "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr$offset",
324 "$addr.addr = $wb", []>;
326 def VLD4d8_UPD : VLD4DWB<0b0000, 0b0000, "8">;
327 def VLD4d16_UPD : VLD4DWB<0b0000, 0b0100, "16">;
328 def VLD4d32_UPD : VLD4DWB<0b0000, 0b1000, "32">;
330 // ...with double-spaced registers (non-updating versions for disassembly only):
331 def VLD4q8 : VLD4D<0b0001, 0b0000, "8">;
332 def VLD4q16 : VLD4D<0b0001, 0b0100, "16">;
333 def VLD4q32 : VLD4D<0b0001, 0b1000, "32">;
334 def VLD4q8_UPD : VLD4DWB<0b0001, 0b0000, "8">;
335 def VLD4q16_UPD : VLD4DWB<0b0001, 0b0100, "16">;
336 def VLD4q32_UPD : VLD4DWB<0b0001, 0b1000, "32">;
338 // ...alternate versions to be allocated odd register numbers:
339 def VLD4q8odd_UPD : VLD4DWB<0b0001, 0b0000, "8">;
340 def VLD4q16odd_UPD : VLD4DWB<0b0001, 0b0100, "16">;
341 def VLD4q32odd_UPD : VLD4DWB<0b0001, 0b1000, "32">;
343 // VLD1LN : Vector Load (single element to one lane)
344 // FIXME: Not yet implemented.
346 // VLD2LN : Vector Load (single 2-element structure to one lane)
347 class VLD2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
348 : NLdSt<1, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2),
349 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, nohash_imm:$lane),
350 IIC_VLD2, "vld2", Dt, "\\{$dst1[$lane], $dst2[$lane]\\}, $addr",
351 "$src1 = $dst1, $src2 = $dst2", []>;
353 def VLD2LNd8 : VLD2LN<0b0001, {?,?,?,?}, "8">;
354 def VLD2LNd16 : VLD2LN<0b0101, {?,?,0,?}, "16">;
355 def VLD2LNd32 : VLD2LN<0b1001, {?,0,?,?}, "32">;
357 // ...with double-spaced registers:
358 def VLD2LNq16 : VLD2LN<0b0101, {?,?,1,?}, "16">;
359 def VLD2LNq32 : VLD2LN<0b1001, {?,1,?,?}, "32">;
361 // ...alternate versions to be allocated odd register numbers:
362 def VLD2LNq16odd : VLD2LN<0b0101, {?,?,1,?}, "16">;
363 def VLD2LNq32odd : VLD2LN<0b1001, {?,1,?,?}, "32">;
365 // ...with address register writeback:
366 class VLD2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
367 : NLdSt<1, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2, GPR:$wb),
368 (ins addrmode6:$addr, am6offset:$offset,
369 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VLD2, "vld2", Dt,
370 "\\{$dst1[$lane], $dst2[$lane]\\}, $addr$offset",
371 "$src1 = $dst1, $src2 = $dst2, $addr.addr = $wb", []>;
373 def VLD2LNd8_UPD : VLD2LNWB<0b0001, {?,?,?,?}, "8">;
374 def VLD2LNd16_UPD : VLD2LNWB<0b0101, {?,?,0,?}, "16">;
375 def VLD2LNd32_UPD : VLD2LNWB<0b1001, {?,0,?,?}, "32">;
377 def VLD2LNq16_UPD : VLD2LNWB<0b0101, {?,?,1,?}, "16">;
378 def VLD2LNq32_UPD : VLD2LNWB<0b1001, {?,1,?,?}, "32">;
380 // VLD3LN : Vector Load (single 3-element structure to one lane)
381 class VLD3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
382 : NLdSt<1, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3),
383 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3,
384 nohash_imm:$lane), IIC_VLD3, "vld3", Dt,
385 "\\{$dst1[$lane], $dst2[$lane], $dst3[$lane]\\}, $addr",
386 "$src1 = $dst1, $src2 = $dst2, $src3 = $dst3", []>;
388 def VLD3LNd8 : VLD3LN<0b0010, {?,?,?,0}, "8">;
389 def VLD3LNd16 : VLD3LN<0b0110, {?,?,0,0}, "16">;
390 def VLD3LNd32 : VLD3LN<0b1010, {?,0,0,0}, "32">;
392 // ...with double-spaced registers:
393 def VLD3LNq16 : VLD3LN<0b0110, {?,?,1,0}, "16">;
394 def VLD3LNq32 : VLD3LN<0b1010, {?,1,0,0}, "32">;
396 // ...alternate versions to be allocated odd register numbers:
397 def VLD3LNq16odd : VLD3LN<0b0110, {?,?,1,0}, "16">;
398 def VLD3LNq32odd : VLD3LN<0b1010, {?,1,0,0}, "32">;
400 // ...with address register writeback:
401 class VLD3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
402 : NLdSt<1, 0b10, op11_8, op7_4,
403 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, GPR:$wb),
404 (ins addrmode6:$addr, am6offset:$offset,
405 DPR:$src1, DPR:$src2, DPR:$src3, nohash_imm:$lane),
406 IIC_VLD3, "vld3", Dt,
407 "\\{$dst1[$lane], $dst2[$lane], $dst3[$lane]\\}, $addr$offset",
408 "$src1 = $dst1, $src2 = $dst2, $src3 = $dst3, $addr.addr = $wb",
411 def VLD3LNd8_UPD : VLD3LNWB<0b0010, {?,?,?,0}, "8">;
412 def VLD3LNd16_UPD : VLD3LNWB<0b0110, {?,?,0,0}, "16">;
413 def VLD3LNd32_UPD : VLD3LNWB<0b1010, {?,0,0,0}, "32">;
415 def VLD3LNq16_UPD : VLD3LNWB<0b0110, {?,?,1,0}, "16">;
416 def VLD3LNq32_UPD : VLD3LNWB<0b1010, {?,1,0,0}, "32">;
418 // VLD4LN : Vector Load (single 4-element structure to one lane)
419 class VLD4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
420 : NLdSt<1, 0b10, op11_8, op7_4,
421 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
422 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
423 nohash_imm:$lane), IIC_VLD4, "vld4", Dt,
424 "\\{$dst1[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $addr",
425 "$src1 = $dst1, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4", []>;
427 def VLD4LNd8 : VLD4LN<0b0011, {?,?,?,?}, "8">;
428 def VLD4LNd16 : VLD4LN<0b0111, {?,?,0,?}, "16">;
429 def VLD4LNd32 : VLD4LN<0b1011, {?,0,?,?}, "32">;
431 // ...with double-spaced registers:
432 def VLD4LNq16 : VLD4LN<0b0111, {?,?,1,?}, "16">;
433 def VLD4LNq32 : VLD4LN<0b1011, {?,1,?,?}, "32">;
435 // ...alternate versions to be allocated odd register numbers:
436 def VLD4LNq16odd : VLD4LN<0b0111, {?,?,1,?}, "16">;
437 def VLD4LNq32odd : VLD4LN<0b1011, {?,1,?,?}, "32">;
439 // ...with address register writeback:
440 class VLD4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
441 : NLdSt<1, 0b10, op11_8, op7_4,
442 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
443 (ins addrmode6:$addr, am6offset:$offset,
444 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
445 IIC_VLD4, "vld4", Dt,
446 "\\{$dst1[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $addr$offset",
447 "$src1 = $dst1, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4, $addr.addr = $wb",
450 def VLD4LNd8_UPD : VLD4LNWB<0b0011, {?,?,?,?}, "8">;
451 def VLD4LNd16_UPD : VLD4LNWB<0b0111, {?,?,0,?}, "16">;
452 def VLD4LNd32_UPD : VLD4LNWB<0b1011, {?,0,?,?}, "32">;
454 def VLD4LNq16_UPD : VLD4LNWB<0b0111, {?,?,1,?}, "16">;
455 def VLD4LNq32_UPD : VLD4LNWB<0b1011, {?,1,?,?}, "32">;
457 // VLD1DUP : Vector Load (single element to all lanes)
458 // VLD2DUP : Vector Load (single 2-element structure to all lanes)
459 // VLD3DUP : Vector Load (single 3-element structure to all lanes)
460 // VLD4DUP : Vector Load (single 4-element structure to all lanes)
461 // FIXME: Not yet implemented.
462 } // mayLoad = 1, hasExtraDefRegAllocReq = 1
464 let mayStore = 1, hasExtraSrcRegAllocReq = 1 in {
466 // VST1 : Vector Store (multiple single elements)
467 class VST1D<bits<4> op7_4, string Dt>
468 : NLdSt<0,0b00,0b0111,op7_4, (outs), (ins addrmode6:$addr, DPR:$src), IIC_VST,
469 "vst1", Dt, "\\{$src\\}, $addr", "", []>;
470 class VST1Q<bits<4> op7_4, string Dt>
471 : NLdSt<0,0b00,0b1010,op7_4, (outs),
472 (ins addrmode6:$addr, DPR:$src1, DPR:$src2), IIC_VST,
473 "vst1", Dt, "\\{$src1, $src2\\}, $addr", "", []>;
475 def VST1d8 : VST1D<0b0000, "8">;
476 def VST1d16 : VST1D<0b0100, "16">;
477 def VST1d32 : VST1D<0b1000, "32">;
478 def VST1d64 : VST1D<0b1100, "64">;
480 def VST1q8 : VST1Q<0b0000, "8">;
481 def VST1q16 : VST1Q<0b0100, "16">;
482 def VST1q32 : VST1Q<0b1000, "32">;
483 def VST1q64 : VST1Q<0b1100, "64">;
485 // ...with address register writeback:
486 class VST1DWB<bits<4> op7_4, string Dt>
487 : NLdSt<0, 0b00, 0b0111, op7_4, (outs GPR:$wb),
488 (ins addrmode6:$addr, am6offset:$offset, DPR:$src), IIC_VST,
489 "vst1", Dt, "\\{$src\\}, $addr$offset", "$addr.addr = $wb", []>;
490 class VST1QWB<bits<4> op7_4, string Dt>
491 : NLdSt<0, 0b00, 0b1010, op7_4, (outs GPR:$wb),
492 (ins addrmode6:$addr, am6offset:$offset, QPR:$src), IIC_VST,
493 "vst1", Dt, "${src:dregpair}, $addr$offset", "$addr.addr = $wb", []>;
495 def VST1d8_UPD : VST1DWB<0b0000, "8">;
496 def VST1d16_UPD : VST1DWB<0b0100, "16">;
497 def VST1d32_UPD : VST1DWB<0b1000, "32">;
498 def VST1d64_UPD : VST1DWB<0b1100, "64">;
500 def VST1q8_UPD : VST1QWB<0b0000, "8">;
501 def VST1q16_UPD : VST1QWB<0b0100, "16">;
502 def VST1q32_UPD : VST1QWB<0b1000, "32">;
503 def VST1q64_UPD : VST1QWB<0b1100, "64">;
505 // ...with 3 registers (some of these are only for the disassembler):
506 class VST1D3<bits<4> op7_4, string Dt>
507 : NLdSt<0, 0b00, 0b0110, op7_4, (outs),
508 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3),
509 IIC_VST, "vst1", Dt, "\\{$src1, $src2, $src3\\}, $addr", "", []>;
510 class VST1D3WB<bits<4> op7_4, string Dt>
511 : NLdSt<0, 0b00, 0b0110, op7_4, (outs GPR:$wb),
512 (ins addrmode6:$addr, am6offset:$offset,
513 DPR:$src1, DPR:$src2, DPR:$src3),
514 IIC_VST, "vst1", Dt, "\\{$src1, $src2, $src3\\}, $addr$offset",
515 "$addr.addr = $wb", []>;
517 def VST1d8T : VST1D3<0b0000, "8">;
518 def VST1d16T : VST1D3<0b0100, "16">;
519 def VST1d32T : VST1D3<0b1000, "32">;
520 def VST1d64T : VST1D3<0b1100, "64">;
522 def VST1d8T_UPD : VST1D3WB<0b0000, "8">;
523 def VST1d16T_UPD : VST1D3WB<0b0100, "16">;
524 def VST1d32T_UPD : VST1D3WB<0b1000, "32">;
525 def VST1d64T_UPD : VST1D3WB<0b1100, "64">;
527 // ...with 4 registers (some of these are only for the disassembler):
528 class VST1D4<bits<4> op7_4, string Dt>
529 : NLdSt<0, 0b00, 0b0010, op7_4, (outs),
530 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
531 IIC_VST, "vst1", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr", "",
533 class VST1D4WB<bits<4> op7_4, string Dt>
534 : NLdSt<0, 0b00, 0b0010, op7_4, (outs GPR:$wb),
535 (ins addrmode6:$addr, am6offset:$offset,
536 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
537 IIC_VST, "vst1", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr$offset",
538 "$addr.addr = $wb", []>;
540 def VST1d8Q : VST1D4<0b0000, "8">;
541 def VST1d16Q : VST1D4<0b0100, "16">;
542 def VST1d32Q : VST1D4<0b1000, "32">;
543 def VST1d64Q : VST1D4<0b1100, "64">;
545 def VST1d8Q_UPD : VST1D4WB<0b0000, "8">;
546 def VST1d16Q_UPD : VST1D4WB<0b0100, "16">;
547 def VST1d32Q_UPD : VST1D4WB<0b1000, "32">;
548 def VST1d64Q_UPD : VST1D4WB<0b1100, "64">;
550 // VST2 : Vector Store (multiple 2-element structures)
551 class VST2D<bits<4> op11_8, bits<4> op7_4, string Dt>
552 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
553 (ins addrmode6:$addr, DPR:$src1, DPR:$src2),
554 IIC_VST, "vst2", Dt, "\\{$src1, $src2\\}, $addr", "", []>;
555 class VST2Q<bits<4> op7_4, string Dt>
556 : NLdSt<0, 0b00, 0b0011, op7_4, (outs),
557 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
558 IIC_VST, "vst2", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr",
561 def VST2d8 : VST2D<0b1000, 0b0000, "8">;
562 def VST2d16 : VST2D<0b1000, 0b0100, "16">;
563 def VST2d32 : VST2D<0b1000, 0b1000, "32">;
565 def VST2q8 : VST2Q<0b0000, "8">;
566 def VST2q16 : VST2Q<0b0100, "16">;
567 def VST2q32 : VST2Q<0b1000, "32">;
569 // ...with address register writeback:
570 class VST2DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
571 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
572 (ins addrmode6:$addr, am6offset:$offset, DPR:$src1, DPR:$src2),
573 IIC_VST, "vst2", Dt, "\\{$src1, $src2\\}, $addr$offset",
574 "$addr.addr = $wb", []>;
575 class VST2QWB<bits<4> op7_4, string Dt>
576 : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
577 (ins addrmode6:$addr, am6offset:$offset,
578 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
579 IIC_VST, "vst2", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr$offset",
580 "$addr.addr = $wb", []>;
582 def VST2d8_UPD : VST2DWB<0b1000, 0b0000, "8">;
583 def VST2d16_UPD : VST2DWB<0b1000, 0b0100, "16">;
584 def VST2d32_UPD : VST2DWB<0b1000, 0b1000, "32">;
586 def VST2q8_UPD : VST2QWB<0b0000, "8">;
587 def VST2q16_UPD : VST2QWB<0b0100, "16">;
588 def VST2q32_UPD : VST2QWB<0b1000, "32">;
590 // ...with double-spaced registers (for disassembly only):
591 def VST2b8 : VST2D<0b1001, 0b0000, "8">;
592 def VST2b16 : VST2D<0b1001, 0b0100, "16">;
593 def VST2b32 : VST2D<0b1001, 0b1000, "32">;
594 def VST2b8_UPD : VST2DWB<0b1001, 0b0000, "8">;
595 def VST2b16_UPD : VST2DWB<0b1001, 0b0100, "16">;
596 def VST2b32_UPD : VST2DWB<0b1001, 0b1000, "32">;
598 // VST3 : Vector Store (multiple 3-element structures)
599 class VST3D<bits<4> op11_8, bits<4> op7_4, string Dt>
600 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
601 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3), IIC_VST,
602 "vst3", Dt, "\\{$src1, $src2, $src3\\}, $addr", "", []>;
604 def VST3d8 : VST3D<0b0100, 0b0000, "8">;
605 def VST3d16 : VST3D<0b0100, 0b0100, "16">;
606 def VST3d32 : VST3D<0b0100, 0b1000, "32">;
608 // ...with address register writeback:
609 class VST3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
610 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
611 (ins addrmode6:$addr, am6offset:$offset,
612 DPR:$src1, DPR:$src2, DPR:$src3), IIC_VST,
613 "vst3", Dt, "\\{$src1, $src2, $src3\\}, $addr$offset",
614 "$addr.addr = $wb", []>;
616 def VST3d8_UPD : VST3DWB<0b0100, 0b0000, "8">;
617 def VST3d16_UPD : VST3DWB<0b0100, 0b0100, "16">;
618 def VST3d32_UPD : VST3DWB<0b0100, 0b1000, "32">;
620 // ...with double-spaced registers (non-updating versions for disassembly only):
621 def VST3q8 : VST3D<0b0101, 0b0000, "8">;
622 def VST3q16 : VST3D<0b0101, 0b0100, "16">;
623 def VST3q32 : VST3D<0b0101, 0b1000, "32">;
624 def VST3q8_UPD : VST3DWB<0b0101, 0b0000, "8">;
625 def VST3q16_UPD : VST3DWB<0b0101, 0b0100, "16">;
626 def VST3q32_UPD : VST3DWB<0b0101, 0b1000, "32">;
628 // ...alternate versions to be allocated odd register numbers:
629 def VST3q8odd_UPD : VST3DWB<0b0101, 0b0000, "8">;
630 def VST3q16odd_UPD : VST3DWB<0b0101, 0b0100, "16">;
631 def VST3q32odd_UPD : VST3DWB<0b0101, 0b1000, "32">;
633 // VST4 : Vector Store (multiple 4-element structures)
634 class VST4D<bits<4> op11_8, bits<4> op7_4, string Dt>
635 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
636 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
637 IIC_VST, "vst4", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr",
640 def VST4d8 : VST4D<0b0000, 0b0000, "8">;
641 def VST4d16 : VST4D<0b0000, 0b0100, "16">;
642 def VST4d32 : VST4D<0b0000, 0b1000, "32">;
644 // ...with address register writeback:
645 class VST4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
646 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
647 (ins addrmode6:$addr, am6offset:$offset,
648 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST,
649 "vst4", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr$offset",
650 "$addr.addr = $wb", []>;
652 def VST4d8_UPD : VST4DWB<0b0000, 0b0000, "8">;
653 def VST4d16_UPD : VST4DWB<0b0000, 0b0100, "16">;
654 def VST4d32_UPD : VST4DWB<0b0000, 0b1000, "32">;
656 // ...with double-spaced registers (non-updating versions for disassembly only):
657 def VST4q8 : VST4D<0b0001, 0b0000, "8">;
658 def VST4q16 : VST4D<0b0001, 0b0100, "16">;
659 def VST4q32 : VST4D<0b0001, 0b1000, "32">;
660 def VST4q8_UPD : VST4DWB<0b0001, 0b0000, "8">;
661 def VST4q16_UPD : VST4DWB<0b0001, 0b0100, "16">;
662 def VST4q32_UPD : VST4DWB<0b0001, 0b1000, "32">;
664 // ...alternate versions to be allocated odd register numbers:
665 def VST4q8odd_UPD : VST4DWB<0b0001, 0b0000, "8">;
666 def VST4q16odd_UPD : VST4DWB<0b0001, 0b0100, "16">;
667 def VST4q32odd_UPD : VST4DWB<0b0001, 0b1000, "32">;
669 // VST1LN : Vector Store (single element from one lane)
670 // FIXME: Not yet implemented.
672 // VST2LN : Vector Store (single 2-element structure from one lane)
673 class VST2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
674 : NLdSt<1, 0b00, op11_8, op7_4, (outs),
675 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, nohash_imm:$lane),
676 IIC_VST, "vst2", Dt, "\\{$src1[$lane], $src2[$lane]\\}, $addr",
679 def VST2LNd8 : VST2LN<0b0001, {?,?,?,?}, "8">;
680 def VST2LNd16 : VST2LN<0b0101, {?,?,0,?}, "16">;
681 def VST2LNd32 : VST2LN<0b1001, {?,0,?,?}, "32">;
683 // ...with double-spaced registers:
684 def VST2LNq16 : VST2LN<0b0101, {?,?,1,?}, "16">;
685 def VST2LNq32 : VST2LN<0b1001, {?,1,?,?}, "32">;
687 // ...alternate versions to be allocated odd register numbers:
688 def VST2LNq16odd : VST2LN<0b0101, {?,?,1,?}, "16">;
689 def VST2LNq32odd : VST2LN<0b1001, {?,1,?,?}, "32">;
691 // ...with address register writeback:
692 class VST2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
693 : NLdSt<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
694 (ins addrmode6:$addr, am6offset:$offset,
695 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VST, "vst2", Dt,
696 "\\{$src1[$lane], $src2[$lane]\\}, $addr$offset",
697 "$addr.addr = $wb", []>;
699 def VST2LNd8_UPD : VST2LNWB<0b0001, {?,?,?,?}, "8">;
700 def VST2LNd16_UPD : VST2LNWB<0b0101, {?,?,0,?}, "16">;
701 def VST2LNd32_UPD : VST2LNWB<0b1001, {?,0,?,?}, "32">;
703 def VST2LNq16_UPD : VST2LNWB<0b0101, {?,?,1,?}, "16">;
704 def VST2LNq32_UPD : VST2LNWB<0b1001, {?,1,?,?}, "32">;
706 // VST3LN : Vector Store (single 3-element structure from one lane)
707 class VST3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
708 : NLdSt<1, 0b00, op11_8, op7_4, (outs),
709 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3,
710 nohash_imm:$lane), IIC_VST, "vst3", Dt,
711 "\\{$src1[$lane], $src2[$lane], $src3[$lane]\\}, $addr", "", []>;
713 def VST3LNd8 : VST3LN<0b0010, {?,?,?,0}, "8">;
714 def VST3LNd16 : VST3LN<0b0110, {?,?,0,0}, "16">;
715 def VST3LNd32 : VST3LN<0b1010, {?,0,0,0}, "32">;
717 // ...with double-spaced registers:
718 def VST3LNq16 : VST3LN<0b0110, {?,?,1,0}, "16">;
719 def VST3LNq32 : VST3LN<0b1010, {?,1,0,0}, "32">;
721 // ...alternate versions to be allocated odd register numbers:
722 def VST3LNq16odd : VST3LN<0b0110, {?,?,1,0}, "16">;
723 def VST3LNq32odd : VST3LN<0b1010, {?,1,0,0}, "32">;
725 // ...with address register writeback:
726 class VST3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
727 : NLdSt<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
728 (ins addrmode6:$addr, am6offset:$offset,
729 DPR:$src1, DPR:$src2, DPR:$src3, nohash_imm:$lane),
731 "\\{$src1[$lane], $src2[$lane], $src3[$lane]\\}, $addr$offset",
732 "$addr.addr = $wb", []>;
734 def VST3LNd8_UPD : VST3LNWB<0b0010, {?,?,?,0}, "8">;
735 def VST3LNd16_UPD : VST3LNWB<0b0110, {?,?,0,0}, "16">;
736 def VST3LNd32_UPD : VST3LNWB<0b1010, {?,0,0,0}, "32">;
738 def VST3LNq16_UPD : VST3LNWB<0b0110, {?,?,1,0}, "16">;
739 def VST3LNq32_UPD : VST3LNWB<0b1010, {?,1,0,0}, "32">;
741 // VST4LN : Vector Store (single 4-element structure from one lane)
742 class VST4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
743 : NLdSt<1, 0b00, op11_8, op7_4, (outs),
744 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
745 nohash_imm:$lane), IIC_VST, "vst4", Dt,
746 "\\{$src1[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $addr",
749 def VST4LNd8 : VST4LN<0b0011, {?,?,?,?}, "8">;
750 def VST4LNd16 : VST4LN<0b0111, {?,?,0,?}, "16">;
751 def VST4LNd32 : VST4LN<0b1011, {?,0,?,?}, "32">;
753 // ...with double-spaced registers:
754 def VST4LNq16 : VST4LN<0b0111, {?,?,1,?}, "16">;
755 def VST4LNq32 : VST4LN<0b1011, {?,1,?,?}, "32">;
757 // ...alternate versions to be allocated odd register numbers:
758 def VST4LNq16odd : VST4LN<0b0111, {?,?,1,?}, "16">;
759 def VST4LNq32odd : VST4LN<0b1011, {?,1,?,?}, "32">;
761 // ...with address register writeback:
762 class VST4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
763 : NLdSt<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
764 (ins addrmode6:$addr, am6offset:$offset,
765 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
767 "\\{$src1[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $addr$offset",
768 "$addr.addr = $wb", []>;
770 def VST4LNd8_UPD : VST4LNWB<0b0011, {?,?,?,?}, "8">;
771 def VST4LNd16_UPD : VST4LNWB<0b0111, {?,?,0,?}, "16">;
772 def VST4LNd32_UPD : VST4LNWB<0b1011, {?,0,?,?}, "32">;
774 def VST4LNq16_UPD : VST4LNWB<0b0111, {?,?,1,?}, "16">;
775 def VST4LNq32_UPD : VST4LNWB<0b1011, {?,1,?,?}, "32">;
777 } // mayStore = 1, hasExtraSrcRegAllocReq = 1
780 //===----------------------------------------------------------------------===//
781 // NEON pattern fragments
782 //===----------------------------------------------------------------------===//
784 // Extract D sub-registers of Q registers.
785 // (arm_dsubreg_0 is 5; arm_dsubreg_1 is 6)
786 def DSubReg_i8_reg : SDNodeXForm<imm, [{
787 return CurDAG->getTargetConstant(5 + N->getZExtValue() / 8, MVT::i32);
789 def DSubReg_i16_reg : SDNodeXForm<imm, [{
790 return CurDAG->getTargetConstant(5 + N->getZExtValue() / 4, MVT::i32);
792 def DSubReg_i32_reg : SDNodeXForm<imm, [{
793 return CurDAG->getTargetConstant(5 + N->getZExtValue() / 2, MVT::i32);
795 def DSubReg_f64_reg : SDNodeXForm<imm, [{
796 return CurDAG->getTargetConstant(5 + N->getZExtValue(), MVT::i32);
798 def DSubReg_f64_other_reg : SDNodeXForm<imm, [{
799 return CurDAG->getTargetConstant(5 + (1 - N->getZExtValue()), MVT::i32);
802 // Extract S sub-registers of Q/D registers.
803 // (arm_ssubreg_0 is 1; arm_ssubreg_1 is 2; etc.)
804 def SSubReg_f32_reg : SDNodeXForm<imm, [{
805 return CurDAG->getTargetConstant(1 + N->getZExtValue(), MVT::i32);
808 // Translate lane numbers from Q registers to D subregs.
809 def SubReg_i8_lane : SDNodeXForm<imm, [{
810 return CurDAG->getTargetConstant(N->getZExtValue() & 7, MVT::i32);
812 def SubReg_i16_lane : SDNodeXForm<imm, [{
813 return CurDAG->getTargetConstant(N->getZExtValue() & 3, MVT::i32);
815 def SubReg_i32_lane : SDNodeXForm<imm, [{
816 return CurDAG->getTargetConstant(N->getZExtValue() & 1, MVT::i32);
819 //===----------------------------------------------------------------------===//
820 // Instruction Classes
821 //===----------------------------------------------------------------------===//
823 // Basic 2-register operations: single-, double- and quad-register.
824 class N2VS<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
825 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
826 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
827 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
828 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src),
829 IIC_VUNAD, OpcodeStr, Dt, "$dst, $src", "", []>;
830 class N2VD<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
831 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
832 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
833 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
834 (ins DPR:$src), IIC_VUNAD, OpcodeStr, Dt,"$dst, $src", "",
835 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src))))]>;
836 class N2VQ<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
837 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
838 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
839 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
840 (ins QPR:$src), IIC_VUNAQ, OpcodeStr, Dt,"$dst, $src", "",
841 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src))))]>;
843 // Basic 2-register intrinsics, both double- and quad-register.
844 class N2VDInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
845 bits<2> op17_16, bits<5> op11_7, bit op4,
846 InstrItinClass itin, string OpcodeStr, string Dt,
847 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
848 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
849 (ins DPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
850 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
851 class N2VQInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
852 bits<2> op17_16, bits<5> op11_7, bit op4,
853 InstrItinClass itin, string OpcodeStr, string Dt,
854 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
855 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
856 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
857 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
859 // Narrow 2-register intrinsics.
860 class N2VNInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
861 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
862 InstrItinClass itin, string OpcodeStr, string Dt,
863 ValueType TyD, ValueType TyQ, Intrinsic IntOp>
864 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$dst),
865 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
866 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src))))]>;
868 // Long 2-register intrinsics (currently only used for VMOVL).
869 class N2VLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
870 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
871 InstrItinClass itin, string OpcodeStr, string Dt,
872 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
873 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$dst),
874 (ins DPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
875 [(set QPR:$dst, (TyQ (IntOp (TyD DPR:$src))))]>;
877 // 2-register shuffles (VTRN/VZIP/VUZP), both double- and quad-register.
878 class N2VDShuffle<bits<2> op19_18, bits<5> op11_7, string OpcodeStr, string Dt>
879 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 0, 0, (outs DPR:$dst1, DPR:$dst2),
880 (ins DPR:$src1, DPR:$src2), IIC_VPERMD,
881 OpcodeStr, Dt, "$dst1, $dst2",
882 "$src1 = $dst1, $src2 = $dst2", []>;
883 class N2VQShuffle<bits<2> op19_18, bits<5> op11_7,
884 InstrItinClass itin, string OpcodeStr, string Dt>
885 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 1, 0, (outs QPR:$dst1, QPR:$dst2),
886 (ins QPR:$src1, QPR:$src2), itin, OpcodeStr, Dt, "$dst1, $dst2",
887 "$src1 = $dst1, $src2 = $dst2", []>;
889 // Basic 3-register operations: single-, double- and quad-register.
890 class N3VS<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
891 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
892 SDNode OpNode, bit Commutable>
893 : N3V<op24, op23, op21_20, op11_8, 0, op4,
894 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src1, DPR_VFP2:$src2), N3RegFrm,
895 IIC_VBIND, OpcodeStr, Dt, "$dst, $src1, $src2", "", []> {
896 let isCommutable = Commutable;
899 class N3VD<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
900 InstrItinClass itin, string OpcodeStr, string Dt,
901 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
902 : N3V<op24, op23, op21_20, op11_8, 0, op4,
903 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
904 OpcodeStr, Dt, "$dst, $src1, $src2", "",
905 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src1), (OpTy DPR:$src2))))]> {
906 let isCommutable = Commutable;
908 // Same as N3VD but no data type.
909 class N3VDX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
910 InstrItinClass itin, string OpcodeStr,
911 ValueType ResTy, ValueType OpTy,
912 SDNode OpNode, bit Commutable>
913 : N3VX<op24, op23, op21_20, op11_8, 0, op4,
914 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
915 OpcodeStr, "$dst, $src1, $src2", "",
916 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src1), (OpTy DPR:$src2))))]>{
917 let isCommutable = Commutable;
920 class N3VDSL<bits<2> op21_20, bits<4> op11_8,
921 InstrItinClass itin, string OpcodeStr, string Dt,
922 ValueType Ty, SDNode ShOp>
923 : N3V<0, 1, op21_20, op11_8, 1, 0,
924 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
925 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
927 (Ty (ShOp (Ty DPR:$src1),
928 (Ty (NEONvduplane (Ty DPR_VFP2:$src2),imm:$lane)))))]> {
929 let isCommutable = 0;
931 class N3VDSL16<bits<2> op21_20, bits<4> op11_8,
932 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
933 : N3V<0, 1, op21_20, op11_8, 1, 0,
934 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
935 NVMulSLFrm, IIC_VMULi16D, OpcodeStr, Dt,"$dst, $src1, $src2[$lane]","",
937 (Ty (ShOp (Ty DPR:$src1),
938 (Ty (NEONvduplane (Ty DPR_8:$src2), imm:$lane)))))]> {
939 let isCommutable = 0;
942 class N3VQ<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
943 InstrItinClass itin, string OpcodeStr, string Dt,
944 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
945 : N3V<op24, op23, op21_20, op11_8, 1, op4,
946 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), N3RegFrm, itin,
947 OpcodeStr, Dt, "$dst, $src1, $src2", "",
948 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src1), (OpTy QPR:$src2))))]> {
949 let isCommutable = Commutable;
951 class N3VQX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
952 InstrItinClass itin, string OpcodeStr,
953 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
954 : N3VX<op24, op23, op21_20, op11_8, 1, op4,
955 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), N3RegFrm, itin,
956 OpcodeStr, "$dst, $src1, $src2", "",
957 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src1), (OpTy QPR:$src2))))]>{
958 let isCommutable = Commutable;
960 class N3VQSL<bits<2> op21_20, bits<4> op11_8,
961 InstrItinClass itin, string OpcodeStr, string Dt,
962 ValueType ResTy, ValueType OpTy, SDNode ShOp>
963 : N3V<1, 1, op21_20, op11_8, 1, 0,
964 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
965 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
966 [(set (ResTy QPR:$dst),
967 (ResTy (ShOp (ResTy QPR:$src1),
968 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
970 let isCommutable = 0;
972 class N3VQSL16<bits<2> op21_20, bits<4> op11_8, string OpcodeStr, string Dt,
973 ValueType ResTy, ValueType OpTy, SDNode ShOp>
974 : N3V<1, 1, op21_20, op11_8, 1, 0,
975 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
976 NVMulSLFrm, IIC_VMULi16Q, OpcodeStr, Dt,"$dst, $src1, $src2[$lane]","",
977 [(set (ResTy QPR:$dst),
978 (ResTy (ShOp (ResTy QPR:$src1),
979 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
981 let isCommutable = 0;
984 // Basic 3-register intrinsics, both double- and quad-register.
985 class N3VDInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
986 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
987 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
988 : N3V<op24, op23, op21_20, op11_8, 0, op4,
989 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), f, itin,
990 OpcodeStr, Dt, "$dst, $src1, $src2", "",
991 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src1), (OpTy DPR:$src2))))]> {
992 let isCommutable = Commutable;
994 class N3VDIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
995 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
996 : N3V<0, 1, op21_20, op11_8, 1, 0,
997 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
998 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1000 (Ty (IntOp (Ty DPR:$src1),
1001 (Ty (NEONvduplane (Ty DPR_VFP2:$src2),
1003 let isCommutable = 0;
1005 class N3VDIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1006 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
1007 : N3V<0, 1, op21_20, op11_8, 1, 0,
1008 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1009 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1010 [(set (Ty DPR:$dst),
1011 (Ty (IntOp (Ty DPR:$src1),
1012 (Ty (NEONvduplane (Ty DPR_8:$src2), imm:$lane)))))]> {
1013 let isCommutable = 0;
1016 class N3VQInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1017 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1018 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
1019 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1020 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), f, itin,
1021 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1022 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src1), (OpTy QPR:$src2))))]> {
1023 let isCommutable = Commutable;
1025 class N3VQIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1026 string OpcodeStr, string Dt,
1027 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1028 : N3V<1, 1, op21_20, op11_8, 1, 0,
1029 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1030 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1031 [(set (ResTy QPR:$dst),
1032 (ResTy (IntOp (ResTy QPR:$src1),
1033 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1035 let isCommutable = 0;
1037 class N3VQIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1038 string OpcodeStr, string Dt,
1039 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1040 : N3V<1, 1, op21_20, op11_8, 1, 0,
1041 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1042 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1043 [(set (ResTy QPR:$dst),
1044 (ResTy (IntOp (ResTy QPR:$src1),
1045 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
1047 let isCommutable = 0;
1050 // Multiply-Add/Sub operations: single-, double- and quad-register.
1051 class N3VSMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1052 InstrItinClass itin, string OpcodeStr, string Dt,
1053 ValueType Ty, SDNode MulOp, SDNode OpNode>
1054 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1055 (outs DPR_VFP2:$dst),
1056 (ins DPR_VFP2:$src1, DPR_VFP2:$src2, DPR_VFP2:$src3), N3RegFrm, itin,
1057 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst", []>;
1059 class N3VDMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1060 InstrItinClass itin, string OpcodeStr, string Dt,
1061 ValueType Ty, SDNode MulOp, SDNode OpNode>
1062 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1063 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3), N3RegFrm, itin,
1064 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1065 [(set DPR:$dst, (Ty (OpNode DPR:$src1,
1066 (Ty (MulOp DPR:$src2, DPR:$src3)))))]>;
1067 class N3VDMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1068 string OpcodeStr, string Dt,
1069 ValueType Ty, SDNode MulOp, SDNode ShOp>
1070 : N3V<0, 1, op21_20, op11_8, 1, 0,
1072 (ins DPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1074 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1075 [(set (Ty DPR:$dst),
1076 (Ty (ShOp (Ty DPR:$src1),
1077 (Ty (MulOp DPR:$src2,
1078 (Ty (NEONvduplane (Ty DPR_VFP2:$src3),
1080 class N3VDMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1081 string OpcodeStr, string Dt,
1082 ValueType Ty, SDNode MulOp, SDNode ShOp>
1083 : N3V<0, 1, op21_20, op11_8, 1, 0,
1085 (ins DPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane),
1087 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1088 [(set (Ty DPR:$dst),
1089 (Ty (ShOp (Ty DPR:$src1),
1090 (Ty (MulOp DPR:$src2,
1091 (Ty (NEONvduplane (Ty DPR_8:$src3),
1094 class N3VQMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1095 InstrItinClass itin, string OpcodeStr, string Dt, ValueType Ty,
1096 SDNode MulOp, SDNode OpNode>
1097 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1098 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3), N3RegFrm, itin,
1099 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1100 [(set QPR:$dst, (Ty (OpNode QPR:$src1,
1101 (Ty (MulOp QPR:$src2, QPR:$src3)))))]>;
1102 class N3VQMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1103 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1104 SDNode MulOp, SDNode ShOp>
1105 : N3V<1, 1, op21_20, op11_8, 1, 0,
1107 (ins QPR:$src1, QPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1109 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1110 [(set (ResTy QPR:$dst),
1111 (ResTy (ShOp (ResTy QPR:$src1),
1112 (ResTy (MulOp QPR:$src2,
1113 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src3),
1115 class N3VQMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1116 string OpcodeStr, string Dt,
1117 ValueType ResTy, ValueType OpTy,
1118 SDNode MulOp, SDNode ShOp>
1119 : N3V<1, 1, op21_20, op11_8, 1, 0,
1121 (ins QPR:$src1, QPR:$src2, DPR_8:$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_8:$src3),
1130 // Neon 3-argument intrinsics, both double- and quad-register.
1131 // The destination register is also used as the first source operand register.
1132 class N3VDInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1133 InstrItinClass itin, string OpcodeStr, string Dt,
1134 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1135 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1136 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3), N3RegFrm, itin,
1137 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1138 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src1),
1139 (OpTy DPR:$src2), (OpTy DPR:$src3))))]>;
1140 class N3VQInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1141 InstrItinClass itin, string OpcodeStr, string Dt,
1142 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1143 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1144 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3), N3RegFrm, itin,
1145 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1146 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src1),
1147 (OpTy QPR:$src2), (OpTy QPR:$src3))))]>;
1149 // Neon Long 3-argument intrinsic. The destination register is
1150 // a quad-register and is also used as the first source operand register.
1151 class N3VLInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1152 InstrItinClass itin, string OpcodeStr, string Dt,
1153 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
1154 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1155 (outs QPR:$dst), (ins QPR:$src1, DPR:$src2, DPR:$src3), N3RegFrm, itin,
1156 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1158 (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$src2), (TyD DPR:$src3))))]>;
1159 class N3VLInt3SL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1160 string OpcodeStr, string Dt,
1161 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1162 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1164 (ins QPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1166 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1167 [(set (ResTy QPR:$dst),
1168 (ResTy (IntOp (ResTy QPR:$src1),
1170 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src3),
1172 class N3VLInt3SL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1173 InstrItinClass itin, string OpcodeStr, string Dt,
1174 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1175 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1177 (ins QPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane),
1179 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1180 [(set (ResTy QPR:$dst),
1181 (ResTy (IntOp (ResTy QPR:$src1),
1183 (OpTy (NEONvduplane (OpTy DPR_8:$src3),
1186 // Narrowing 3-register intrinsics.
1187 class N3VNInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1188 string OpcodeStr, string Dt, ValueType TyD, ValueType TyQ,
1189 Intrinsic IntOp, bit Commutable>
1190 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1191 (outs DPR:$dst), (ins QPR:$src1, QPR:$src2), N3RegFrm, IIC_VBINi4D,
1192 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1193 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src1), (TyQ QPR:$src2))))]> {
1194 let isCommutable = Commutable;
1197 // Long 3-register intrinsics.
1198 class N3VLInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1199 InstrItinClass itin, string OpcodeStr, string Dt,
1200 ValueType TyQ, ValueType TyD, Intrinsic IntOp, bit Commutable>
1201 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1202 (outs QPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
1203 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1204 [(set QPR:$dst, (TyQ (IntOp (TyD DPR:$src1), (TyD DPR:$src2))))]> {
1205 let isCommutable = Commutable;
1207 class N3VLIntSL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1208 string OpcodeStr, string Dt,
1209 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1210 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1211 (outs QPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1212 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1213 [(set (ResTy QPR:$dst),
1214 (ResTy (IntOp (OpTy DPR:$src1),
1215 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1217 class N3VLIntSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1218 InstrItinClass itin, string OpcodeStr, string Dt,
1219 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1220 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1221 (outs QPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1222 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1223 [(set (ResTy QPR:$dst),
1224 (ResTy (IntOp (OpTy DPR:$src1),
1225 (OpTy (NEONvduplane (OpTy DPR_8:$src2),
1228 // Wide 3-register intrinsics.
1229 class N3VWInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1230 string OpcodeStr, string Dt, ValueType TyQ, ValueType TyD,
1231 Intrinsic IntOp, bit Commutable>
1232 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1233 (outs QPR:$dst), (ins QPR:$src1, DPR:$src2), N3RegFrm, IIC_VSUBiD,
1234 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1235 [(set QPR:$dst, (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$src2))))]> {
1236 let isCommutable = Commutable;
1239 // Pairwise long 2-register intrinsics, both double- and quad-register.
1240 class N2VDPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1241 bits<2> op17_16, bits<5> op11_7, bit op4,
1242 string OpcodeStr, string Dt,
1243 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1244 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
1245 (ins DPR:$src), IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
1246 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
1247 class N2VQPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1248 bits<2> op17_16, bits<5> op11_7, bit op4,
1249 string OpcodeStr, string Dt,
1250 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1251 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
1252 (ins QPR:$src), IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
1253 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
1255 // Pairwise long 2-register accumulate intrinsics,
1256 // both double- and quad-register.
1257 // The destination register is also used as the first source operand register.
1258 class N2VDPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1259 bits<2> op17_16, bits<5> op11_7, bit op4,
1260 string OpcodeStr, string Dt,
1261 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1262 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
1263 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), IIC_VPALiD,
1264 OpcodeStr, Dt, "$dst, $src2", "$src1 = $dst",
1265 [(set DPR:$dst, (ResTy (IntOp (ResTy DPR:$src1), (OpTy DPR:$src2))))]>;
1266 class N2VQPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1267 bits<2> op17_16, bits<5> op11_7, bit op4,
1268 string OpcodeStr, string Dt,
1269 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1270 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4,
1271 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), IIC_VPALiQ,
1272 OpcodeStr, Dt, "$dst, $src2", "$src1 = $dst",
1273 [(set QPR:$dst, (ResTy (IntOp (ResTy QPR:$src1), (OpTy QPR:$src2))))]>;
1275 // Shift by immediate,
1276 // both double- and quad-register.
1277 class N2VDSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1278 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1279 ValueType Ty, SDNode OpNode>
1280 : N2VImm<op24, op23, op11_8, op7, 0, op4,
1281 (outs DPR:$dst), (ins DPR:$src, i32imm:$SIMM), f, itin,
1282 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1283 [(set DPR:$dst, (Ty (OpNode (Ty DPR:$src), (i32 imm:$SIMM))))]>;
1284 class N2VQSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1285 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1286 ValueType Ty, SDNode OpNode>
1287 : N2VImm<op24, op23, op11_8, op7, 1, op4,
1288 (outs QPR:$dst), (ins QPR:$src, i32imm:$SIMM), f, itin,
1289 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1290 [(set QPR:$dst, (Ty (OpNode (Ty QPR:$src), (i32 imm:$SIMM))))]>;
1292 // Long shift by immediate.
1293 class N2VLSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
1294 string OpcodeStr, string Dt,
1295 ValueType ResTy, ValueType OpTy, SDNode OpNode>
1296 : N2VImm<op24, op23, op11_8, op7, op6, op4,
1297 (outs QPR:$dst), (ins DPR:$src, i32imm:$SIMM), N2RegVShLFrm,
1298 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1299 [(set QPR:$dst, (ResTy (OpNode (OpTy DPR:$src),
1300 (i32 imm:$SIMM))))]>;
1302 // Narrow shift by immediate.
1303 class N2VNSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
1304 InstrItinClass itin, string OpcodeStr, string Dt,
1305 ValueType ResTy, ValueType OpTy, SDNode OpNode>
1306 : N2VImm<op24, op23, op11_8, op7, op6, op4,
1307 (outs DPR:$dst), (ins QPR:$src, i32imm:$SIMM), N2RegVShRFrm, itin,
1308 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1309 [(set DPR:$dst, (ResTy (OpNode (OpTy QPR:$src),
1310 (i32 imm:$SIMM))))]>;
1312 // Shift right by immediate and accumulate,
1313 // both double- and quad-register.
1314 class N2VDShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1315 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1316 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$dst),
1317 (ins DPR:$src1, DPR:$src2, i32imm:$SIMM), N2RegVShRFrm, IIC_VPALiD,
1318 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1319 [(set DPR:$dst, (Ty (add DPR:$src1,
1320 (Ty (ShOp DPR:$src2, (i32 imm:$SIMM))))))]>;
1321 class N2VQShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1322 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1323 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$dst),
1324 (ins QPR:$src1, QPR:$src2, i32imm:$SIMM), N2RegVShRFrm, IIC_VPALiD,
1325 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1326 [(set QPR:$dst, (Ty (add QPR:$src1,
1327 (Ty (ShOp QPR:$src2, (i32 imm:$SIMM))))))]>;
1329 // Shift by immediate and insert,
1330 // both double- and quad-register.
1331 class N2VDShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1332 Format f, string OpcodeStr, string Dt, ValueType Ty,SDNode ShOp>
1333 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$dst),
1334 (ins DPR:$src1, DPR:$src2, i32imm:$SIMM), f, IIC_VSHLiD,
1335 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1336 [(set DPR:$dst, (Ty (ShOp DPR:$src1, DPR:$src2, (i32 imm:$SIMM))))]>;
1337 class N2VQShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1338 Format f, string OpcodeStr, string Dt, ValueType Ty,SDNode ShOp>
1339 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$dst),
1340 (ins QPR:$src1, QPR:$src2, i32imm:$SIMM), f, IIC_VSHLiQ,
1341 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1342 [(set QPR:$dst, (Ty (ShOp QPR:$src1, QPR:$src2, (i32 imm:$SIMM))))]>;
1344 // Convert, with fractional bits immediate,
1345 // both double- and quad-register.
1346 class N2VCvtD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1347 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1349 : N2VImm<op24, op23, op11_8, op7, 0, op4,
1350 (outs DPR:$dst), (ins DPR:$src, i32imm:$SIMM), NVCVTFrm,
1351 IIC_VUNAD, OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1352 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src), (i32 imm:$SIMM))))]>;
1353 class N2VCvtQ<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1354 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1356 : N2VImm<op24, op23, op11_8, op7, 1, op4,
1357 (outs QPR:$dst), (ins QPR:$src, i32imm:$SIMM), NVCVTFrm,
1358 IIC_VUNAQ, OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1359 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src), (i32 imm:$SIMM))))]>;
1361 //===----------------------------------------------------------------------===//
1363 //===----------------------------------------------------------------------===//
1365 // Abbreviations used in multiclass suffixes:
1366 // Q = quarter int (8 bit) elements
1367 // H = half int (16 bit) elements
1368 // S = single int (32 bit) elements
1369 // D = double int (64 bit) elements
1371 // Neon 2-register vector operations -- for disassembly only.
1373 // First with only element sizes of 8, 16 and 32 bits:
1374 multiclass N2V_QHS_cmp<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1375 bits<5> op11_7, bit op4, string opc, string Dt,
1377 // 64-bit vector types.
1378 def v8i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 0, op4,
1379 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1380 opc, !strconcat(Dt, "8"), asm, "", []>;
1381 def v4i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 0, op4,
1382 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1383 opc, !strconcat(Dt, "16"), asm, "", []>;
1384 def v2i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
1385 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1386 opc, !strconcat(Dt, "32"), asm, "", []>;
1387 def v2f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
1388 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1389 opc, "f32", asm, "", []> {
1390 let Inst{10} = 1; // overwrite F = 1
1393 // 128-bit vector types.
1394 def v16i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 1, op4,
1395 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1396 opc, !strconcat(Dt, "8"), asm, "", []>;
1397 def v8i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 1, op4,
1398 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1399 opc, !strconcat(Dt, "16"), asm, "", []>;
1400 def v4i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
1401 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1402 opc, !strconcat(Dt, "32"), asm, "", []>;
1403 def v4f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
1404 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1405 opc, "f32", asm, "", []> {
1406 let Inst{10} = 1; // overwrite F = 1
1410 // Neon 3-register vector operations.
1412 // First with only element sizes of 8, 16 and 32 bits:
1413 multiclass N3V_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1414 InstrItinClass itinD16, InstrItinClass itinD32,
1415 InstrItinClass itinQ16, InstrItinClass itinQ32,
1416 string OpcodeStr, string Dt,
1417 SDNode OpNode, bit Commutable = 0> {
1418 // 64-bit vector types.
1419 def v8i8 : N3VD<op24, op23, 0b00, op11_8, op4, itinD16,
1420 OpcodeStr, !strconcat(Dt, "8"),
1421 v8i8, v8i8, OpNode, Commutable>;
1422 def v4i16 : N3VD<op24, op23, 0b01, op11_8, op4, itinD16,
1423 OpcodeStr, !strconcat(Dt, "16"),
1424 v4i16, v4i16, OpNode, Commutable>;
1425 def v2i32 : N3VD<op24, op23, 0b10, op11_8, op4, itinD32,
1426 OpcodeStr, !strconcat(Dt, "32"),
1427 v2i32, v2i32, OpNode, Commutable>;
1429 // 128-bit vector types.
1430 def v16i8 : N3VQ<op24, op23, 0b00, op11_8, op4, itinQ16,
1431 OpcodeStr, !strconcat(Dt, "8"),
1432 v16i8, v16i8, OpNode, Commutable>;
1433 def v8i16 : N3VQ<op24, op23, 0b01, op11_8, op4, itinQ16,
1434 OpcodeStr, !strconcat(Dt, "16"),
1435 v8i16, v8i16, OpNode, Commutable>;
1436 def v4i32 : N3VQ<op24, op23, 0b10, op11_8, op4, itinQ32,
1437 OpcodeStr, !strconcat(Dt, "32"),
1438 v4i32, v4i32, OpNode, Commutable>;
1441 multiclass N3VSL_HS<bits<4> op11_8, string OpcodeStr, string Dt, SDNode ShOp> {
1442 def v4i16 : N3VDSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
1444 def v2i32 : N3VDSL<0b10, op11_8, IIC_VMULi32D, OpcodeStr, !strconcat(Dt,"32"),
1446 def v8i16 : N3VQSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
1447 v8i16, v4i16, ShOp>;
1448 def v4i32 : N3VQSL<0b10, op11_8, IIC_VMULi32Q, OpcodeStr, !strconcat(Dt,"32"),
1449 v4i32, v2i32, ShOp>;
1452 // ....then also with element size 64 bits:
1453 multiclass N3V_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1454 InstrItinClass itinD, InstrItinClass itinQ,
1455 string OpcodeStr, string Dt,
1456 SDNode OpNode, bit Commutable = 0>
1457 : N3V_QHS<op24, op23, op11_8, op4, itinD, itinD, itinQ, itinQ,
1458 OpcodeStr, Dt, OpNode, Commutable> {
1459 def v1i64 : N3VD<op24, op23, 0b11, op11_8, op4, itinD,
1460 OpcodeStr, !strconcat(Dt, "64"),
1461 v1i64, v1i64, OpNode, Commutable>;
1462 def v2i64 : N3VQ<op24, op23, 0b11, op11_8, op4, itinQ,
1463 OpcodeStr, !strconcat(Dt, "64"),
1464 v2i64, v2i64, OpNode, Commutable>;
1468 // Neon Narrowing 2-register vector intrinsics,
1469 // source operand element sizes of 16, 32 and 64 bits:
1470 multiclass N2VNInt_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1471 bits<5> op11_7, bit op6, bit op4,
1472 InstrItinClass itin, string OpcodeStr, string Dt,
1474 def v8i8 : N2VNInt<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
1475 itin, OpcodeStr, !strconcat(Dt, "16"),
1476 v8i8, v8i16, IntOp>;
1477 def v4i16 : N2VNInt<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
1478 itin, OpcodeStr, !strconcat(Dt, "32"),
1479 v4i16, v4i32, IntOp>;
1480 def v2i32 : N2VNInt<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
1481 itin, OpcodeStr, !strconcat(Dt, "64"),
1482 v2i32, v2i64, IntOp>;
1486 // Neon Lengthening 2-register vector intrinsic (currently specific to VMOVL).
1487 // source operand element sizes of 16, 32 and 64 bits:
1488 multiclass N2VLInt_QHS<bits<2> op24_23, bits<5> op11_7, bit op6, bit op4,
1489 string OpcodeStr, string Dt, Intrinsic IntOp> {
1490 def v8i16 : N2VLInt<op24_23, 0b00, 0b10, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
1491 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
1492 def v4i32 : N2VLInt<op24_23, 0b01, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
1493 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
1494 def v2i64 : N2VLInt<op24_23, 0b10, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
1495 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1499 // Neon 3-register vector intrinsics.
1501 // First with only element sizes of 16 and 32 bits:
1502 multiclass N3VInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
1503 InstrItinClass itinD16, InstrItinClass itinD32,
1504 InstrItinClass itinQ16, InstrItinClass itinQ32,
1505 string OpcodeStr, string Dt,
1506 Intrinsic IntOp, bit Commutable = 0> {
1507 // 64-bit vector types.
1508 def v4i16 : N3VDInt<op24, op23, 0b01, op11_8, op4, f, itinD16,
1509 OpcodeStr, !strconcat(Dt, "16"),
1510 v4i16, v4i16, IntOp, Commutable>;
1511 def v2i32 : N3VDInt<op24, op23, 0b10, op11_8, op4, f, itinD32,
1512 OpcodeStr, !strconcat(Dt, "32"),
1513 v2i32, v2i32, IntOp, Commutable>;
1515 // 128-bit vector types.
1516 def v8i16 : N3VQInt<op24, op23, 0b01, op11_8, op4, f, itinQ16,
1517 OpcodeStr, !strconcat(Dt, "16"),
1518 v8i16, v8i16, IntOp, Commutable>;
1519 def v4i32 : N3VQInt<op24, op23, 0b10, op11_8, op4, f, itinQ32,
1520 OpcodeStr, !strconcat(Dt, "32"),
1521 v4i32, v4i32, IntOp, Commutable>;
1524 multiclass N3VIntSL_HS<bits<4> op11_8,
1525 InstrItinClass itinD16, InstrItinClass itinD32,
1526 InstrItinClass itinQ16, InstrItinClass itinQ32,
1527 string OpcodeStr, string Dt, Intrinsic IntOp> {
1528 def v4i16 : N3VDIntSL16<0b01, op11_8, itinD16,
1529 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp>;
1530 def v2i32 : N3VDIntSL<0b10, op11_8, itinD32,
1531 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp>;
1532 def v8i16 : N3VQIntSL16<0b01, op11_8, itinQ16,
1533 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16, IntOp>;
1534 def v4i32 : N3VQIntSL<0b10, op11_8, itinQ32,
1535 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32, IntOp>;
1538 // ....then also with element size of 8 bits:
1539 multiclass N3VInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
1540 InstrItinClass itinD16, InstrItinClass itinD32,
1541 InstrItinClass itinQ16, InstrItinClass itinQ32,
1542 string OpcodeStr, string Dt,
1543 Intrinsic IntOp, bit Commutable = 0>
1544 : N3VInt_HS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
1545 OpcodeStr, Dt, IntOp, Commutable> {
1546 def v8i8 : N3VDInt<op24, op23, 0b00, op11_8, op4, f, itinD16,
1547 OpcodeStr, !strconcat(Dt, "8"),
1548 v8i8, v8i8, IntOp, Commutable>;
1549 def v16i8 : N3VQInt<op24, op23, 0b00, op11_8, op4, f, itinQ16,
1550 OpcodeStr, !strconcat(Dt, "8"),
1551 v16i8, v16i8, IntOp, Commutable>;
1554 // ....then also with element size of 64 bits:
1555 multiclass N3VInt_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
1556 InstrItinClass itinD16, InstrItinClass itinD32,
1557 InstrItinClass itinQ16, InstrItinClass itinQ32,
1558 string OpcodeStr, string Dt,
1559 Intrinsic IntOp, bit Commutable = 0>
1560 : N3VInt_QHS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
1561 OpcodeStr, Dt, IntOp, Commutable> {
1562 def v1i64 : N3VDInt<op24, op23, 0b11, op11_8, op4, f, itinD32,
1563 OpcodeStr, !strconcat(Dt, "64"),
1564 v1i64, v1i64, IntOp, Commutable>;
1565 def v2i64 : N3VQInt<op24, op23, 0b11, op11_8, op4, f, itinQ32,
1566 OpcodeStr, !strconcat(Dt, "64"),
1567 v2i64, v2i64, IntOp, Commutable>;
1570 // Neon Narrowing 3-register vector intrinsics,
1571 // source operand element sizes of 16, 32 and 64 bits:
1572 multiclass N3VNInt_HSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1573 string OpcodeStr, string Dt,
1574 Intrinsic IntOp, bit Commutable = 0> {
1575 def v8i8 : N3VNInt<op24, op23, 0b00, op11_8, op4,
1576 OpcodeStr, !strconcat(Dt, "16"),
1577 v8i8, v8i16, IntOp, Commutable>;
1578 def v4i16 : N3VNInt<op24, op23, 0b01, op11_8, op4,
1579 OpcodeStr, !strconcat(Dt, "32"),
1580 v4i16, v4i32, IntOp, Commutable>;
1581 def v2i32 : N3VNInt<op24, op23, 0b10, op11_8, op4,
1582 OpcodeStr, !strconcat(Dt, "64"),
1583 v2i32, v2i64, IntOp, Commutable>;
1587 // Neon Long 3-register vector intrinsics.
1589 // First with only element sizes of 16 and 32 bits:
1590 multiclass N3VLInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
1591 InstrItinClass itin16, InstrItinClass itin32,
1592 string OpcodeStr, string Dt,
1593 Intrinsic IntOp, bit Commutable = 0> {
1594 def v4i32 : N3VLInt<op24, op23, 0b01, op11_8, op4, itin16,
1595 OpcodeStr, !strconcat(Dt, "16"),
1596 v4i32, v4i16, IntOp, Commutable>;
1597 def v2i64 : N3VLInt<op24, op23, 0b10, op11_8, op4, itin32,
1598 OpcodeStr, !strconcat(Dt, "32"),
1599 v2i64, v2i32, IntOp, Commutable>;
1602 multiclass N3VLIntSL_HS<bit op24, bits<4> op11_8,
1603 InstrItinClass itin, string OpcodeStr, string Dt,
1605 def v4i16 : N3VLIntSL16<op24, 0b01, op11_8, itin,
1606 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
1607 def v2i32 : N3VLIntSL<op24, 0b10, op11_8, itin,
1608 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1611 // ....then also with element size of 8 bits:
1612 multiclass N3VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1613 InstrItinClass itin16, InstrItinClass itin32,
1614 string OpcodeStr, string Dt,
1615 Intrinsic IntOp, bit Commutable = 0>
1616 : N3VLInt_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt,
1617 IntOp, Commutable> {
1618 def v8i16 : N3VLInt<op24, op23, 0b00, op11_8, op4, itin16,
1619 OpcodeStr, !strconcat(Dt, "8"),
1620 v8i16, v8i8, IntOp, Commutable>;
1624 // Neon Wide 3-register vector intrinsics,
1625 // source operand element sizes of 8, 16 and 32 bits:
1626 multiclass N3VWInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1627 string OpcodeStr, string Dt,
1628 Intrinsic IntOp, bit Commutable = 0> {
1629 def v8i16 : N3VWInt<op24, op23, 0b00, op11_8, op4,
1630 OpcodeStr, !strconcat(Dt, "8"),
1631 v8i16, v8i8, IntOp, Commutable>;
1632 def v4i32 : N3VWInt<op24, op23, 0b01, op11_8, op4,
1633 OpcodeStr, !strconcat(Dt, "16"),
1634 v4i32, v4i16, IntOp, Commutable>;
1635 def v2i64 : N3VWInt<op24, op23, 0b10, op11_8, op4,
1636 OpcodeStr, !strconcat(Dt, "32"),
1637 v2i64, v2i32, IntOp, Commutable>;
1641 // Neon Multiply-Op vector operations,
1642 // element sizes of 8, 16 and 32 bits:
1643 multiclass N3VMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1644 InstrItinClass itinD16, InstrItinClass itinD32,
1645 InstrItinClass itinQ16, InstrItinClass itinQ32,
1646 string OpcodeStr, string Dt, SDNode OpNode> {
1647 // 64-bit vector types.
1648 def v8i8 : N3VDMulOp<op24, op23, 0b00, op11_8, op4, itinD16,
1649 OpcodeStr, !strconcat(Dt, "8"), v8i8, mul, OpNode>;
1650 def v4i16 : N3VDMulOp<op24, op23, 0b01, op11_8, op4, itinD16,
1651 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, OpNode>;
1652 def v2i32 : N3VDMulOp<op24, op23, 0b10, op11_8, op4, itinD32,
1653 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, OpNode>;
1655 // 128-bit vector types.
1656 def v16i8 : N3VQMulOp<op24, op23, 0b00, op11_8, op4, itinQ16,
1657 OpcodeStr, !strconcat(Dt, "8"), v16i8, mul, OpNode>;
1658 def v8i16 : N3VQMulOp<op24, op23, 0b01, op11_8, op4, itinQ16,
1659 OpcodeStr, !strconcat(Dt, "16"), v8i16, mul, OpNode>;
1660 def v4i32 : N3VQMulOp<op24, op23, 0b10, op11_8, op4, itinQ32,
1661 OpcodeStr, !strconcat(Dt, "32"), v4i32, mul, OpNode>;
1664 multiclass N3VMulOpSL_HS<bits<4> op11_8,
1665 InstrItinClass itinD16, InstrItinClass itinD32,
1666 InstrItinClass itinQ16, InstrItinClass itinQ32,
1667 string OpcodeStr, string Dt, SDNode ShOp> {
1668 def v4i16 : N3VDMulOpSL16<0b01, op11_8, itinD16,
1669 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, ShOp>;
1670 def v2i32 : N3VDMulOpSL<0b10, op11_8, itinD32,
1671 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, ShOp>;
1672 def v8i16 : N3VQMulOpSL16<0b01, op11_8, itinQ16,
1673 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16,
1675 def v4i32 : N3VQMulOpSL<0b10, op11_8, itinQ32,
1676 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32,
1680 // Neon 3-argument intrinsics,
1681 // element sizes of 8, 16 and 32 bits:
1682 multiclass N3VInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1683 InstrItinClass itinD, InstrItinClass itinQ,
1684 string OpcodeStr, string Dt, Intrinsic IntOp> {
1685 // 64-bit vector types.
1686 def v8i8 : N3VDInt3<op24, op23, 0b00, op11_8, op4, itinD,
1687 OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
1688 def v4i16 : N3VDInt3<op24, op23, 0b01, op11_8, op4, itinD,
1689 OpcodeStr, !strconcat(Dt, "16"), v4i16, v4i16, IntOp>;
1690 def v2i32 : N3VDInt3<op24, op23, 0b10, op11_8, op4, itinD,
1691 OpcodeStr, !strconcat(Dt, "32"), v2i32, v2i32, IntOp>;
1693 // 128-bit vector types.
1694 def v16i8 : N3VQInt3<op24, op23, 0b00, op11_8, op4, itinQ,
1695 OpcodeStr, !strconcat(Dt, "8"), v16i8, v16i8, IntOp>;
1696 def v8i16 : N3VQInt3<op24, op23, 0b01, op11_8, op4, itinQ,
1697 OpcodeStr, !strconcat(Dt, "16"), v8i16, v8i16, IntOp>;
1698 def v4i32 : N3VQInt3<op24, op23, 0b10, op11_8, op4, itinQ,
1699 OpcodeStr, !strconcat(Dt, "32"), v4i32, v4i32, IntOp>;
1703 // Neon Long 3-argument intrinsics.
1705 // First with only element sizes of 16 and 32 bits:
1706 multiclass N3VLInt3_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
1707 InstrItinClass itin16, InstrItinClass itin32,
1708 string OpcodeStr, string Dt, Intrinsic IntOp> {
1709 def v4i32 : N3VLInt3<op24, op23, 0b01, op11_8, op4, itin16,
1710 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
1711 def v2i64 : N3VLInt3<op24, op23, 0b10, op11_8, op4, itin32,
1712 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1715 multiclass N3VLInt3SL_HS<bit op24, bits<4> op11_8,
1716 string OpcodeStr, string Dt, Intrinsic IntOp> {
1717 def v4i16 : N3VLInt3SL16<op24, 0b01, op11_8, IIC_VMACi16D,
1718 OpcodeStr, !strconcat(Dt,"16"), v4i32, v4i16, IntOp>;
1719 def v2i32 : N3VLInt3SL<op24, 0b10, op11_8, IIC_VMACi32D,
1720 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1723 // ....then also with element size of 8 bits:
1724 multiclass N3VLInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1725 InstrItinClass itin16, InstrItinClass itin32,
1726 string OpcodeStr, string Dt, Intrinsic IntOp>
1727 : N3VLInt3_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt, IntOp> {
1728 def v8i16 : N3VLInt3<op24, op23, 0b00, op11_8, op4, itin16,
1729 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
1733 // Neon 2-register vector intrinsics,
1734 // element sizes of 8, 16 and 32 bits:
1735 multiclass N2VInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1736 bits<5> op11_7, bit op4,
1737 InstrItinClass itinD, InstrItinClass itinQ,
1738 string OpcodeStr, string Dt, Intrinsic IntOp> {
1739 // 64-bit vector types.
1740 def v8i8 : N2VDInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1741 itinD, OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
1742 def v4i16 : N2VDInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1743 itinD, OpcodeStr, !strconcat(Dt, "16"),v4i16,v4i16,IntOp>;
1744 def v2i32 : N2VDInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1745 itinD, OpcodeStr, !strconcat(Dt, "32"),v2i32,v2i32,IntOp>;
1747 // 128-bit vector types.
1748 def v16i8 : N2VQInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1749 itinQ, OpcodeStr, !strconcat(Dt, "8"), v16i8,v16i8,IntOp>;
1750 def v8i16 : N2VQInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1751 itinQ, OpcodeStr, !strconcat(Dt, "16"),v8i16,v8i16,IntOp>;
1752 def v4i32 : N2VQInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1753 itinQ, OpcodeStr, !strconcat(Dt, "32"),v4i32,v4i32,IntOp>;
1757 // Neon Pairwise long 2-register intrinsics,
1758 // element sizes of 8, 16 and 32 bits:
1759 multiclass N2VPLInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1760 bits<5> op11_7, bit op4,
1761 string OpcodeStr, string Dt, Intrinsic IntOp> {
1762 // 64-bit vector types.
1763 def v8i8 : N2VDPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1764 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
1765 def v4i16 : N2VDPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1766 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
1767 def v2i32 : N2VDPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1768 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
1770 // 128-bit vector types.
1771 def v16i8 : N2VQPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1772 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
1773 def v8i16 : N2VQPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1774 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
1775 def v4i32 : N2VQPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1776 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
1780 // Neon Pairwise long 2-register accumulate intrinsics,
1781 // element sizes of 8, 16 and 32 bits:
1782 multiclass N2VPLInt2_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1783 bits<5> op11_7, bit op4,
1784 string OpcodeStr, string Dt, Intrinsic IntOp> {
1785 // 64-bit vector types.
1786 def v8i8 : N2VDPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1787 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
1788 def v4i16 : N2VDPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1789 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
1790 def v2i32 : N2VDPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1791 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
1793 // 128-bit vector types.
1794 def v16i8 : N2VQPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1795 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
1796 def v8i16 : N2VQPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1797 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
1798 def v4i32 : N2VQPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1799 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
1803 // Neon 2-register vector shift by immediate,
1804 // with f of either N2RegVShLFrm or N2RegVShRFrm
1805 // element sizes of 8, 16, 32 and 64 bits:
1806 multiclass N2VSh_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1807 InstrItinClass itin, string OpcodeStr, string Dt,
1808 SDNode OpNode, Format f> {
1809 // 64-bit vector types.
1810 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
1811 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
1812 let Inst{21-19} = 0b001; // imm6 = 001xxx
1814 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
1815 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
1816 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1818 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
1819 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
1820 let Inst{21} = 0b1; // imm6 = 1xxxxx
1822 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, f, itin,
1823 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
1826 // 128-bit vector types.
1827 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
1828 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
1829 let Inst{21-19} = 0b001; // imm6 = 001xxx
1831 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
1832 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
1833 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1835 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
1836 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
1837 let Inst{21} = 0b1; // imm6 = 1xxxxx
1839 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, f, itin,
1840 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
1844 // Neon Shift-Accumulate vector operations,
1845 // element sizes of 8, 16, 32 and 64 bits:
1846 multiclass N2VShAdd_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1847 string OpcodeStr, string Dt, SDNode ShOp> {
1848 // 64-bit vector types.
1849 def v8i8 : N2VDShAdd<op24, op23, op11_8, 0, op4,
1850 OpcodeStr, !strconcat(Dt, "8"), v8i8, ShOp> {
1851 let Inst{21-19} = 0b001; // imm6 = 001xxx
1853 def v4i16 : N2VDShAdd<op24, op23, op11_8, 0, op4,
1854 OpcodeStr, !strconcat(Dt, "16"), v4i16, ShOp> {
1855 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1857 def v2i32 : N2VDShAdd<op24, op23, op11_8, 0, op4,
1858 OpcodeStr, !strconcat(Dt, "32"), v2i32, ShOp> {
1859 let Inst{21} = 0b1; // imm6 = 1xxxxx
1861 def v1i64 : N2VDShAdd<op24, op23, op11_8, 1, op4,
1862 OpcodeStr, !strconcat(Dt, "64"), v1i64, ShOp>;
1865 // 128-bit vector types.
1866 def v16i8 : N2VQShAdd<op24, op23, op11_8, 0, op4,
1867 OpcodeStr, !strconcat(Dt, "8"), v16i8, ShOp> {
1868 let Inst{21-19} = 0b001; // imm6 = 001xxx
1870 def v8i16 : N2VQShAdd<op24, op23, op11_8, 0, op4,
1871 OpcodeStr, !strconcat(Dt, "16"), v8i16, ShOp> {
1872 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1874 def v4i32 : N2VQShAdd<op24, op23, op11_8, 0, op4,
1875 OpcodeStr, !strconcat(Dt, "32"), v4i32, ShOp> {
1876 let Inst{21} = 0b1; // imm6 = 1xxxxx
1878 def v2i64 : N2VQShAdd<op24, op23, op11_8, 1, op4,
1879 OpcodeStr, !strconcat(Dt, "64"), v2i64, ShOp>;
1884 // Neon Shift-Insert vector operations,
1885 // with f of either N2RegVShLFrm or N2RegVShRFrm
1886 // element sizes of 8, 16, 32 and 64 bits:
1887 multiclass N2VShIns_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1888 string OpcodeStr, SDNode ShOp,
1890 // 64-bit vector types.
1891 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4,
1892 f, OpcodeStr, "8", v8i8, ShOp> {
1893 let Inst{21-19} = 0b001; // imm6 = 001xxx
1895 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4,
1896 f, OpcodeStr, "16", v4i16, ShOp> {
1897 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1899 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4,
1900 f, OpcodeStr, "32", v2i32, ShOp> {
1901 let Inst{21} = 0b1; // imm6 = 1xxxxx
1903 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4,
1904 f, OpcodeStr, "64", v1i64, ShOp>;
1907 // 128-bit vector types.
1908 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4,
1909 f, OpcodeStr, "8", v16i8, ShOp> {
1910 let Inst{21-19} = 0b001; // imm6 = 001xxx
1912 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4,
1913 f, OpcodeStr, "16", v8i16, ShOp> {
1914 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1916 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4,
1917 f, OpcodeStr, "32", v4i32, ShOp> {
1918 let Inst{21} = 0b1; // imm6 = 1xxxxx
1920 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4,
1921 f, OpcodeStr, "64", v2i64, ShOp>;
1925 // Neon Shift Long operations,
1926 // element sizes of 8, 16, 32 bits:
1927 multiclass N2VLSh_QHS<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
1928 bit op4, string OpcodeStr, string Dt, SDNode OpNode> {
1929 def v8i16 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
1930 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode> {
1931 let Inst{21-19} = 0b001; // imm6 = 001xxx
1933 def v4i32 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
1934 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode> {
1935 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1937 def v2i64 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
1938 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode> {
1939 let Inst{21} = 0b1; // imm6 = 1xxxxx
1943 // Neon Shift Narrow operations,
1944 // element sizes of 16, 32, 64 bits:
1945 multiclass N2VNSh_HSD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
1946 bit op4, InstrItinClass itin, string OpcodeStr, string Dt,
1948 def v8i8 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
1949 OpcodeStr, !strconcat(Dt, "16"), v8i8, v8i16, OpNode> {
1950 let Inst{21-19} = 0b001; // imm6 = 001xxx
1952 def v4i16 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
1953 OpcodeStr, !strconcat(Dt, "32"), v4i16, v4i32, OpNode> {
1954 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1956 def v2i32 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
1957 OpcodeStr, !strconcat(Dt, "64"), v2i32, v2i64, OpNode> {
1958 let Inst{21} = 0b1; // imm6 = 1xxxxx
1962 //===----------------------------------------------------------------------===//
1963 // Instruction Definitions.
1964 //===----------------------------------------------------------------------===//
1966 // Vector Add Operations.
1968 // VADD : Vector Add (integer and floating-point)
1969 defm VADD : N3V_QHSD<0, 0, 0b1000, 0, IIC_VBINiD, IIC_VBINiQ, "vadd", "i",
1971 def VADDfd : N3VD<0, 0, 0b00, 0b1101, 0, IIC_VBIND, "vadd", "f32",
1972 v2f32, v2f32, fadd, 1>;
1973 def VADDfq : N3VQ<0, 0, 0b00, 0b1101, 0, IIC_VBINQ, "vadd", "f32",
1974 v4f32, v4f32, fadd, 1>;
1975 // VADDL : Vector Add Long (Q = D + D)
1976 defm VADDLs : N3VLInt_QHS<0,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
1977 "vaddl", "s", int_arm_neon_vaddls, 1>;
1978 defm VADDLu : N3VLInt_QHS<1,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
1979 "vaddl", "u", int_arm_neon_vaddlu, 1>;
1980 // VADDW : Vector Add Wide (Q = Q + D)
1981 defm VADDWs : N3VWInt_QHS<0,1,0b0001,0, "vaddw", "s", int_arm_neon_vaddws, 0>;
1982 defm VADDWu : N3VWInt_QHS<1,1,0b0001,0, "vaddw", "u", int_arm_neon_vaddwu, 0>;
1983 // VHADD : Vector Halving Add
1984 defm VHADDs : N3VInt_QHS<0, 0, 0b0000, 0, N3RegFrm,
1985 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
1986 "vhadd", "s", int_arm_neon_vhadds, 1>;
1987 defm VHADDu : N3VInt_QHS<1, 0, 0b0000, 0, N3RegFrm,
1988 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
1989 "vhadd", "u", int_arm_neon_vhaddu, 1>;
1990 // VRHADD : Vector Rounding Halving Add
1991 defm VRHADDs : N3VInt_QHS<0, 0, 0b0001, 0, N3RegFrm,
1992 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
1993 "vrhadd", "s", int_arm_neon_vrhadds, 1>;
1994 defm VRHADDu : N3VInt_QHS<1, 0, 0b0001, 0, N3RegFrm,
1995 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
1996 "vrhadd", "u", int_arm_neon_vrhaddu, 1>;
1997 // VQADD : Vector Saturating Add
1998 defm VQADDs : N3VInt_QHSD<0, 0, 0b0000, 1, N3RegFrm,
1999 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2000 "vqadd", "s", int_arm_neon_vqadds, 1>;
2001 defm VQADDu : N3VInt_QHSD<1, 0, 0b0000, 1, N3RegFrm,
2002 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2003 "vqadd", "u", int_arm_neon_vqaddu, 1>;
2004 // VADDHN : Vector Add and Narrow Returning High Half (D = Q + Q)
2005 defm VADDHN : N3VNInt_HSD<0,1,0b0100,0, "vaddhn", "i",
2006 int_arm_neon_vaddhn, 1>;
2007 // VRADDHN : Vector Rounding Add and Narrow Returning High Half (D = Q + Q)
2008 defm VRADDHN : N3VNInt_HSD<1,1,0b0100,0, "vraddhn", "i",
2009 int_arm_neon_vraddhn, 1>;
2011 // Vector Multiply Operations.
2013 // VMUL : Vector Multiply (integer, polynomial and floating-point)
2014 defm VMUL : N3V_QHS<0, 0, 0b1001, 1, IIC_VMULi16D, IIC_VMULi32D,
2015 IIC_VMULi16Q, IIC_VMULi32Q, "vmul", "i", mul, 1>;
2016 def VMULpd : N3VDInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16D, "vmul",
2017 "p8", v8i8, v8i8, int_arm_neon_vmulp, 1>;
2018 def VMULpq : N3VQInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16Q, "vmul",
2019 "p8", v16i8, v16i8, int_arm_neon_vmulp, 1>;
2020 def VMULfd : N3VD<1, 0, 0b00, 0b1101, 1, IIC_VBIND, "vmul", "f32",
2021 v2f32, v2f32, fmul, 1>;
2022 def VMULfq : N3VQ<1, 0, 0b00, 0b1101, 1, IIC_VBINQ, "vmul", "f32",
2023 v4f32, v4f32, fmul, 1>;
2024 defm VMULsl : N3VSL_HS<0b1000, "vmul", "i", mul>;
2025 def VMULslfd : N3VDSL<0b10, 0b1001, IIC_VBIND, "vmul", "f32", v2f32, fmul>;
2026 def VMULslfq : N3VQSL<0b10, 0b1001, IIC_VBINQ, "vmul", "f32", v4f32,
2029 def : Pat<(v8i16 (mul (v8i16 QPR:$src1),
2030 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
2031 (v8i16 (VMULslv8i16 (v8i16 QPR:$src1),
2032 (v4i16 (EXTRACT_SUBREG QPR:$src2,
2033 (DSubReg_i16_reg imm:$lane))),
2034 (SubReg_i16_lane imm:$lane)))>;
2035 def : Pat<(v4i32 (mul (v4i32 QPR:$src1),
2036 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
2037 (v4i32 (VMULslv4i32 (v4i32 QPR:$src1),
2038 (v2i32 (EXTRACT_SUBREG QPR:$src2,
2039 (DSubReg_i32_reg imm:$lane))),
2040 (SubReg_i32_lane imm:$lane)))>;
2041 def : Pat<(v4f32 (fmul (v4f32 QPR:$src1),
2042 (v4f32 (NEONvduplane (v4f32 QPR:$src2), imm:$lane)))),
2043 (v4f32 (VMULslfq (v4f32 QPR:$src1),
2044 (v2f32 (EXTRACT_SUBREG QPR:$src2,
2045 (DSubReg_i32_reg imm:$lane))),
2046 (SubReg_i32_lane imm:$lane)))>;
2048 // VQDMULH : Vector Saturating Doubling Multiply Returning High Half
2049 defm VQDMULH : N3VInt_HS<0, 0, 0b1011, 0, N3RegFrm, IIC_VMULi16D, IIC_VMULi32D,
2050 IIC_VMULi16Q, IIC_VMULi32Q,
2051 "vqdmulh", "s", int_arm_neon_vqdmulh, 1>;
2052 defm VQDMULHsl: N3VIntSL_HS<0b1100, IIC_VMULi16D, IIC_VMULi32D,
2053 IIC_VMULi16Q, IIC_VMULi32Q,
2054 "vqdmulh", "s", int_arm_neon_vqdmulh>;
2055 def : Pat<(v8i16 (int_arm_neon_vqdmulh (v8i16 QPR:$src1),
2056 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
2058 (v8i16 (VQDMULHslv8i16 (v8i16 QPR:$src1),
2059 (v4i16 (EXTRACT_SUBREG QPR:$src2,
2060 (DSubReg_i16_reg imm:$lane))),
2061 (SubReg_i16_lane imm:$lane)))>;
2062 def : Pat<(v4i32 (int_arm_neon_vqdmulh (v4i32 QPR:$src1),
2063 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
2065 (v4i32 (VQDMULHslv4i32 (v4i32 QPR:$src1),
2066 (v2i32 (EXTRACT_SUBREG QPR:$src2,
2067 (DSubReg_i32_reg imm:$lane))),
2068 (SubReg_i32_lane imm:$lane)))>;
2070 // VQRDMULH : Vector Rounding Saturating Doubling Multiply Returning High Half
2071 defm VQRDMULH : N3VInt_HS<1, 0, 0b1011, 0, N3RegFrm,
2072 IIC_VMULi16D,IIC_VMULi32D,IIC_VMULi16Q,IIC_VMULi32Q,
2073 "vqrdmulh", "s", int_arm_neon_vqrdmulh, 1>;
2074 defm VQRDMULHsl : N3VIntSL_HS<0b1101, IIC_VMULi16D, IIC_VMULi32D,
2075 IIC_VMULi16Q, IIC_VMULi32Q,
2076 "vqrdmulh", "s", int_arm_neon_vqrdmulh>;
2077 def : Pat<(v8i16 (int_arm_neon_vqrdmulh (v8i16 QPR:$src1),
2078 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
2080 (v8i16 (VQRDMULHslv8i16 (v8i16 QPR:$src1),
2081 (v4i16 (EXTRACT_SUBREG QPR:$src2,
2082 (DSubReg_i16_reg imm:$lane))),
2083 (SubReg_i16_lane imm:$lane)))>;
2084 def : Pat<(v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$src1),
2085 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
2087 (v4i32 (VQRDMULHslv4i32 (v4i32 QPR:$src1),
2088 (v2i32 (EXTRACT_SUBREG QPR:$src2,
2089 (DSubReg_i32_reg imm:$lane))),
2090 (SubReg_i32_lane imm:$lane)))>;
2092 // VMULL : Vector Multiply Long (integer and polynomial) (Q = D * D)
2093 defm VMULLs : N3VLInt_QHS<0,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
2094 "vmull", "s", int_arm_neon_vmulls, 1>;
2095 defm VMULLu : N3VLInt_QHS<1,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
2096 "vmull", "u", int_arm_neon_vmullu, 1>;
2097 def VMULLp : N3VLInt<0, 1, 0b00, 0b1110, 0, IIC_VMULi16D, "vmull", "p8",
2098 v8i16, v8i8, int_arm_neon_vmullp, 1>;
2099 defm VMULLsls : N3VLIntSL_HS<0, 0b1010, IIC_VMULi16D, "vmull", "s",
2100 int_arm_neon_vmulls>;
2101 defm VMULLslu : N3VLIntSL_HS<1, 0b1010, IIC_VMULi16D, "vmull", "u",
2102 int_arm_neon_vmullu>;
2104 // VQDMULL : Vector Saturating Doubling Multiply Long (Q = D * D)
2105 defm VQDMULL : N3VLInt_HS<0,1,0b1101,0, IIC_VMULi16D, IIC_VMULi32D,
2106 "vqdmull", "s", int_arm_neon_vqdmull, 1>;
2107 defm VQDMULLsl: N3VLIntSL_HS<0, 0b1011, IIC_VMULi16D,
2108 "vqdmull", "s", int_arm_neon_vqdmull>;
2110 // Vector Multiply-Accumulate and Multiply-Subtract Operations.
2112 // VMLA : Vector Multiply Accumulate (integer and floating-point)
2113 defm VMLA : N3VMulOp_QHS<0, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
2114 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
2115 def VMLAfd : N3VDMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACD, "vmla", "f32",
2117 def VMLAfq : N3VQMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACQ, "vmla", "f32",
2119 defm VMLAsl : N3VMulOpSL_HS<0b0000, IIC_VMACi16D, IIC_VMACi32D,
2120 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
2121 def VMLAslfd : N3VDMulOpSL<0b10, 0b0001, IIC_VMACD, "vmla", "f32",
2123 def VMLAslfq : N3VQMulOpSL<0b10, 0b0001, IIC_VMACQ, "vmla", "f32",
2124 v4f32, v2f32, fmul, fadd>;
2126 def : Pat<(v8i16 (add (v8i16 QPR:$src1),
2127 (mul (v8i16 QPR:$src2),
2128 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
2129 (v8i16 (VMLAslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
2130 (v4i16 (EXTRACT_SUBREG QPR:$src3,
2131 (DSubReg_i16_reg imm:$lane))),
2132 (SubReg_i16_lane imm:$lane)))>;
2134 def : Pat<(v4i32 (add (v4i32 QPR:$src1),
2135 (mul (v4i32 QPR:$src2),
2136 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
2137 (v4i32 (VMLAslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
2138 (v2i32 (EXTRACT_SUBREG QPR:$src3,
2139 (DSubReg_i32_reg imm:$lane))),
2140 (SubReg_i32_lane imm:$lane)))>;
2142 def : Pat<(v4f32 (fadd (v4f32 QPR:$src1),
2143 (fmul (v4f32 QPR:$src2),
2144 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
2145 (v4f32 (VMLAslfq (v4f32 QPR:$src1),
2147 (v2f32 (EXTRACT_SUBREG QPR:$src3,
2148 (DSubReg_i32_reg imm:$lane))),
2149 (SubReg_i32_lane imm:$lane)))>;
2151 // VMLAL : Vector Multiply Accumulate Long (Q += D * D)
2152 defm VMLALs : N3VLInt3_QHS<0,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
2153 "vmlal", "s", int_arm_neon_vmlals>;
2154 defm VMLALu : N3VLInt3_QHS<1,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
2155 "vmlal", "u", int_arm_neon_vmlalu>;
2157 defm VMLALsls : N3VLInt3SL_HS<0, 0b0010, "vmlal", "s", int_arm_neon_vmlals>;
2158 defm VMLALslu : N3VLInt3SL_HS<1, 0b0010, "vmlal", "u", int_arm_neon_vmlalu>;
2160 // VQDMLAL : Vector Saturating Doubling Multiply Accumulate Long (Q += D * D)
2161 defm VQDMLAL : N3VLInt3_HS<0, 1, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
2162 "vqdmlal", "s", int_arm_neon_vqdmlal>;
2163 defm VQDMLALsl: N3VLInt3SL_HS<0, 0b0011, "vqdmlal", "s", int_arm_neon_vqdmlal>;
2165 // VMLS : Vector Multiply Subtract (integer and floating-point)
2166 defm VMLS : N3VMulOp_QHS<1, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
2167 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
2168 def VMLSfd : N3VDMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACD, "vmls", "f32",
2170 def VMLSfq : N3VQMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACQ, "vmls", "f32",
2172 defm VMLSsl : N3VMulOpSL_HS<0b0100, IIC_VMACi16D, IIC_VMACi32D,
2173 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
2174 def VMLSslfd : N3VDMulOpSL<0b10, 0b0101, IIC_VMACD, "vmls", "f32",
2176 def VMLSslfq : N3VQMulOpSL<0b10, 0b0101, IIC_VMACQ, "vmls", "f32",
2177 v4f32, v2f32, fmul, fsub>;
2179 def : Pat<(v8i16 (sub (v8i16 QPR:$src1),
2180 (mul (v8i16 QPR:$src2),
2181 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
2182 (v8i16 (VMLSslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
2183 (v4i16 (EXTRACT_SUBREG QPR:$src3,
2184 (DSubReg_i16_reg imm:$lane))),
2185 (SubReg_i16_lane imm:$lane)))>;
2187 def : Pat<(v4i32 (sub (v4i32 QPR:$src1),
2188 (mul (v4i32 QPR:$src2),
2189 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
2190 (v4i32 (VMLSslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
2191 (v2i32 (EXTRACT_SUBREG QPR:$src3,
2192 (DSubReg_i32_reg imm:$lane))),
2193 (SubReg_i32_lane imm:$lane)))>;
2195 def : Pat<(v4f32 (fsub (v4f32 QPR:$src1),
2196 (fmul (v4f32 QPR:$src2),
2197 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
2198 (v4f32 (VMLSslfq (v4f32 QPR:$src1), (v4f32 QPR:$src2),
2199 (v2f32 (EXTRACT_SUBREG QPR:$src3,
2200 (DSubReg_i32_reg imm:$lane))),
2201 (SubReg_i32_lane imm:$lane)))>;
2203 // VMLSL : Vector Multiply Subtract Long (Q -= D * D)
2204 defm VMLSLs : N3VLInt3_QHS<0,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
2205 "vmlsl", "s", int_arm_neon_vmlsls>;
2206 defm VMLSLu : N3VLInt3_QHS<1,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
2207 "vmlsl", "u", int_arm_neon_vmlslu>;
2209 defm VMLSLsls : N3VLInt3SL_HS<0, 0b0110, "vmlsl", "s", int_arm_neon_vmlsls>;
2210 defm VMLSLslu : N3VLInt3SL_HS<1, 0b0110, "vmlsl", "u", int_arm_neon_vmlslu>;
2212 // VQDMLSL : Vector Saturating Doubling Multiply Subtract Long (Q -= D * D)
2213 defm VQDMLSL : N3VLInt3_HS<0, 1, 0b1011, 0, IIC_VMACi16D, IIC_VMACi32D,
2214 "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
2215 defm VQDMLSLsl: N3VLInt3SL_HS<0, 0b111, "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
2217 // Vector Subtract Operations.
2219 // VSUB : Vector Subtract (integer and floating-point)
2220 defm VSUB : N3V_QHSD<1, 0, 0b1000, 0, IIC_VSUBiD, IIC_VSUBiQ,
2221 "vsub", "i", sub, 0>;
2222 def VSUBfd : N3VD<0, 0, 0b10, 0b1101, 0, IIC_VBIND, "vsub", "f32",
2223 v2f32, v2f32, fsub, 0>;
2224 def VSUBfq : N3VQ<0, 0, 0b10, 0b1101, 0, IIC_VBINQ, "vsub", "f32",
2225 v4f32, v4f32, fsub, 0>;
2226 // VSUBL : Vector Subtract Long (Q = D - D)
2227 defm VSUBLs : N3VLInt_QHS<0,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
2228 "vsubl", "s", int_arm_neon_vsubls, 1>;
2229 defm VSUBLu : N3VLInt_QHS<1,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
2230 "vsubl", "u", int_arm_neon_vsublu, 1>;
2231 // VSUBW : Vector Subtract Wide (Q = Q - D)
2232 defm VSUBWs : N3VWInt_QHS<0,1,0b0011,0, "vsubw", "s", int_arm_neon_vsubws, 0>;
2233 defm VSUBWu : N3VWInt_QHS<1,1,0b0011,0, "vsubw", "u", int_arm_neon_vsubwu, 0>;
2234 // VHSUB : Vector Halving Subtract
2235 defm VHSUBs : N3VInt_QHS<0, 0, 0b0010, 0, N3RegFrm,
2236 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2237 "vhsub", "s", int_arm_neon_vhsubs, 0>;
2238 defm VHSUBu : N3VInt_QHS<1, 0, 0b0010, 0, N3RegFrm,
2239 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2240 "vhsub", "u", int_arm_neon_vhsubu, 0>;
2241 // VQSUB : Vector Saturing Subtract
2242 defm VQSUBs : N3VInt_QHSD<0, 0, 0b0010, 1, N3RegFrm,
2243 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2244 "vqsub", "s", int_arm_neon_vqsubs, 0>;
2245 defm VQSUBu : N3VInt_QHSD<1, 0, 0b0010, 1, N3RegFrm,
2246 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2247 "vqsub", "u", int_arm_neon_vqsubu, 0>;
2248 // VSUBHN : Vector Subtract and Narrow Returning High Half (D = Q - Q)
2249 defm VSUBHN : N3VNInt_HSD<0,1,0b0110,0, "vsubhn", "i",
2250 int_arm_neon_vsubhn, 0>;
2251 // VRSUBHN : Vector Rounding Subtract and Narrow Returning High Half (D=Q-Q)
2252 defm VRSUBHN : N3VNInt_HSD<1,1,0b0110,0, "vrsubhn", "i",
2253 int_arm_neon_vrsubhn, 0>;
2255 // Vector Comparisons.
2257 // VCEQ : Vector Compare Equal
2258 defm VCEQ : N3V_QHS<1, 0, 0b1000, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
2259 IIC_VSUBi4Q, "vceq", "i", NEONvceq, 1>;
2260 def VCEQfd : N3VD<0,0,0b00,0b1110,0, IIC_VBIND, "vceq", "f32", v2i32, v2f32,
2262 def VCEQfq : N3VQ<0,0,0b00,0b1110,0, IIC_VBINQ, "vceq", "f32", v4i32, v4f32,
2264 // For disassembly only.
2265 defm VCEQz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00010, 0, "vceq", "i",
2268 // VCGE : Vector Compare Greater Than or Equal
2269 defm VCGEs : N3V_QHS<0, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
2270 IIC_VSUBi4Q, "vcge", "s", NEONvcge, 0>;
2271 defm VCGEu : N3V_QHS<1, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
2272 IIC_VSUBi4Q, "vcge", "u", NEONvcgeu, 0>;
2273 def VCGEfd : N3VD<1,0,0b00,0b1110,0, IIC_VBIND, "vcge", "f32", v2i32, v2f32,
2275 def VCGEfq : N3VQ<1,0,0b00,0b1110,0, IIC_VBINQ, "vcge", "f32", v4i32, v4f32,
2277 // For disassembly only.
2278 defm VCGEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00001, 0, "vcge", "s",
2280 // For disassembly only.
2281 defm VCLEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00011, 0, "vcle", "s",
2284 // VCGT : Vector Compare Greater Than
2285 defm VCGTs : N3V_QHS<0, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
2286 IIC_VSUBi4Q, "vcgt", "s", NEONvcgt, 0>;
2287 defm VCGTu : N3V_QHS<1, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
2288 IIC_VSUBi4Q, "vcgt", "u", NEONvcgtu, 0>;
2289 def VCGTfd : N3VD<1,0,0b10,0b1110,0, IIC_VBIND, "vcgt", "f32", v2i32, v2f32,
2291 def VCGTfq : N3VQ<1,0,0b10,0b1110,0, IIC_VBINQ, "vcgt", "f32", v4i32, v4f32,
2293 // For disassembly only.
2294 defm VCGTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00000, 0, "vcgt", "s",
2296 // For disassembly only.
2297 defm VCLTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00100, 0, "vclt", "s",
2300 // VACGE : Vector Absolute Compare Greater Than or Equal (aka VCAGE)
2301 def VACGEd : N3VDInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacge",
2302 "f32", v2i32, v2f32, int_arm_neon_vacged, 0>;
2303 def VACGEq : N3VQInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacge",
2304 "f32", v4i32, v4f32, int_arm_neon_vacgeq, 0>;
2305 // VACGT : Vector Absolute Compare Greater Than (aka VCAGT)
2306 def VACGTd : N3VDInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacgt",
2307 "f32", v2i32, v2f32, int_arm_neon_vacgtd, 0>;
2308 def VACGTq : N3VQInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacgt",
2309 "f32", v4i32, v4f32, int_arm_neon_vacgtq, 0>;
2310 // VTST : Vector Test Bits
2311 defm VTST : N3V_QHS<0, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2312 IIC_VBINi4Q, "vtst", "", NEONvtst, 1>;
2314 // Vector Bitwise Operations.
2316 def vnot8 : PatFrag<(ops node:$in),
2317 (xor node:$in, (bitconvert (v8i8 immAllOnesV)))>;
2318 def vnot16 : PatFrag<(ops node:$in),
2319 (xor node:$in, (bitconvert (v16i8 immAllOnesV)))>;
2322 // VAND : Vector Bitwise AND
2323 def VANDd : N3VDX<0, 0, 0b00, 0b0001, 1, IIC_VBINiD, "vand",
2324 v2i32, v2i32, and, 1>;
2325 def VANDq : N3VQX<0, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "vand",
2326 v4i32, v4i32, and, 1>;
2328 // VEOR : Vector Bitwise Exclusive OR
2329 def VEORd : N3VDX<1, 0, 0b00, 0b0001, 1, IIC_VBINiD, "veor",
2330 v2i32, v2i32, xor, 1>;
2331 def VEORq : N3VQX<1, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "veor",
2332 v4i32, v4i32, xor, 1>;
2334 // VORR : Vector Bitwise OR
2335 def VORRd : N3VDX<0, 0, 0b10, 0b0001, 1, IIC_VBINiD, "vorr",
2336 v2i32, v2i32, or, 1>;
2337 def VORRq : N3VQX<0, 0, 0b10, 0b0001, 1, IIC_VBINiQ, "vorr",
2338 v4i32, v4i32, or, 1>;
2340 // VBIC : Vector Bitwise Bit Clear (AND NOT)
2341 def VBICd : N3VX<0, 0, 0b01, 0b0001, 0, 1, (outs DPR:$dst),
2342 (ins DPR:$src1, DPR:$src2), N3RegFrm, IIC_VBINiD,
2343 "vbic", "$dst, $src1, $src2", "",
2344 [(set DPR:$dst, (v2i32 (and DPR:$src1,
2345 (vnot8 DPR:$src2))))]>;
2346 def VBICq : N3VX<0, 0, 0b01, 0b0001, 1, 1, (outs QPR:$dst),
2347 (ins QPR:$src1, QPR:$src2), N3RegFrm, IIC_VBINiQ,
2348 "vbic", "$dst, $src1, $src2", "",
2349 [(set QPR:$dst, (v4i32 (and QPR:$src1,
2350 (vnot16 QPR:$src2))))]>;
2352 // VORN : Vector Bitwise OR NOT
2353 def VORNd : N3VX<0, 0, 0b11, 0b0001, 0, 1, (outs DPR:$dst),
2354 (ins DPR:$src1, DPR:$src2), N3RegFrm, IIC_VBINiD,
2355 "vorn", "$dst, $src1, $src2", "",
2356 [(set DPR:$dst, (v2i32 (or DPR:$src1,
2357 (vnot8 DPR:$src2))))]>;
2358 def VORNq : N3VX<0, 0, 0b11, 0b0001, 1, 1, (outs QPR:$dst),
2359 (ins QPR:$src1, QPR:$src2), N3RegFrm, IIC_VBINiQ,
2360 "vorn", "$dst, $src1, $src2", "",
2361 [(set QPR:$dst, (v4i32 (or QPR:$src1,
2362 (vnot16 QPR:$src2))))]>;
2364 // VMVN : Vector Bitwise NOT
2365 def VMVNd : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 0, 0,
2366 (outs DPR:$dst), (ins DPR:$src), IIC_VSUBiD,
2367 "vmvn", "$dst, $src", "",
2368 [(set DPR:$dst, (v2i32 (vnot8 DPR:$src)))]>;
2369 def VMVNq : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 1, 0,
2370 (outs QPR:$dst), (ins QPR:$src), IIC_VSUBiD,
2371 "vmvn", "$dst, $src", "",
2372 [(set QPR:$dst, (v4i32 (vnot16 QPR:$src)))]>;
2373 def : Pat<(v2i32 (vnot8 DPR:$src)), (VMVNd DPR:$src)>;
2374 def : Pat<(v4i32 (vnot16 QPR:$src)), (VMVNq QPR:$src)>;
2376 // VBSL : Vector Bitwise Select
2377 def VBSLd : N3VX<1, 0, 0b01, 0b0001, 0, 1, (outs DPR:$dst),
2378 (ins DPR:$src1, DPR:$src2, DPR:$src3),
2379 N3RegFrm, IIC_VCNTiD,
2380 "vbsl", "$dst, $src2, $src3", "$src1 = $dst",
2382 (v2i32 (or (and DPR:$src2, DPR:$src1),
2383 (and DPR:$src3, (vnot8 DPR:$src1)))))]>;
2384 def VBSLq : N3VX<1, 0, 0b01, 0b0001, 1, 1, (outs QPR:$dst),
2385 (ins QPR:$src1, QPR:$src2, QPR:$src3),
2386 N3RegFrm, IIC_VCNTiQ,
2387 "vbsl", "$dst, $src2, $src3", "$src1 = $dst",
2389 (v4i32 (or (and QPR:$src2, QPR:$src1),
2390 (and QPR:$src3, (vnot16 QPR:$src1)))))]>;
2392 // VBIF : Vector Bitwise Insert if False
2393 // like VBSL but with: "vbif $dst, $src3, $src1", "$src2 = $dst",
2394 def VBIFd : N3VX<1, 0, 0b11, 0b0001, 0, 1,
2395 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3),
2396 N3RegFrm, IIC_VBINiD,
2397 "vbif", "$dst, $src2, $src3", "$src1 = $dst",
2398 [/* For disassembly only; pattern left blank */]>;
2399 def VBIFq : N3VX<1, 0, 0b11, 0b0001, 1, 1,
2400 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3),
2401 N3RegFrm, IIC_VBINiQ,
2402 "vbif", "$dst, $src2, $src3", "$src1 = $dst",
2403 [/* For disassembly only; pattern left blank */]>;
2405 // VBIT : Vector Bitwise Insert if True
2406 // like VBSL but with: "vbit $dst, $src2, $src1", "$src3 = $dst",
2407 def VBITd : N3VX<1, 0, 0b10, 0b0001, 0, 1,
2408 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3),
2409 N3RegFrm, IIC_VBINiD,
2410 "vbit", "$dst, $src2, $src3", "$src1 = $dst",
2411 [/* For disassembly only; pattern left blank */]>;
2412 def VBITq : N3VX<1, 0, 0b10, 0b0001, 1, 1,
2413 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3),
2414 N3RegFrm, IIC_VBINiQ,
2415 "vbit", "$dst, $src2, $src3", "$src1 = $dst",
2416 [/* For disassembly only; pattern left blank */]>;
2418 // VBIT/VBIF are not yet implemented. The TwoAddress pass will not go looking
2419 // for equivalent operations with different register constraints; it just
2422 // Vector Absolute Differences.
2424 // VABD : Vector Absolute Difference
2425 defm VABDs : N3VInt_QHS<0, 0, 0b0111, 0, N3RegFrm,
2426 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2427 "vabd", "s", int_arm_neon_vabds, 0>;
2428 defm VABDu : N3VInt_QHS<1, 0, 0b0111, 0, N3RegFrm,
2429 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2430 "vabd", "u", int_arm_neon_vabdu, 0>;
2431 def VABDfd : N3VDInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBIND,
2432 "vabd", "f32", v2f32, v2f32, int_arm_neon_vabds, 0>;
2433 def VABDfq : N3VQInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBINQ,
2434 "vabd", "f32", v4f32, v4f32, int_arm_neon_vabds, 0>;
2436 // VABDL : Vector Absolute Difference Long (Q = | D - D |)
2437 defm VABDLs : N3VLInt_QHS<0,1,0b0111,0, IIC_VSUBi4Q, IIC_VSUBi4Q,
2438 "vabdl", "s", int_arm_neon_vabdls, 0>;
2439 defm VABDLu : N3VLInt_QHS<1,1,0b0111,0, IIC_VSUBi4Q, IIC_VSUBi4Q,
2440 "vabdl", "u", int_arm_neon_vabdlu, 0>;
2442 // VABA : Vector Absolute Difference and Accumulate
2443 defm VABAs : N3VInt3_QHS<0,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
2444 "vaba", "s", int_arm_neon_vabas>;
2445 defm VABAu : N3VInt3_QHS<1,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
2446 "vaba", "u", int_arm_neon_vabau>;
2448 // VABAL : Vector Absolute Difference and Accumulate Long (Q += | D - D |)
2449 defm VABALs : N3VLInt3_QHS<0,1,0b0101,0, IIC_VABAD, IIC_VABAD,
2450 "vabal", "s", int_arm_neon_vabals>;
2451 defm VABALu : N3VLInt3_QHS<1,1,0b0101,0, IIC_VABAD, IIC_VABAD,
2452 "vabal", "u", int_arm_neon_vabalu>;
2454 // Vector Maximum and Minimum.
2456 // VMAX : Vector Maximum
2457 defm VMAXs : N3VInt_QHS<0, 0, 0b0110, 0, N3RegFrm,
2458 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2459 "vmax", "s", int_arm_neon_vmaxs, 1>;
2460 defm VMAXu : N3VInt_QHS<1, 0, 0b0110, 0, N3RegFrm,
2461 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2462 "vmax", "u", int_arm_neon_vmaxu, 1>;
2463 def VMAXfd : N3VDInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBIND,
2465 v2f32, v2f32, int_arm_neon_vmaxs, 1>;
2466 def VMAXfq : N3VQInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBINQ,
2468 v4f32, v4f32, int_arm_neon_vmaxs, 1>;
2470 // VMIN : Vector Minimum
2471 defm VMINs : N3VInt_QHS<0, 0, 0b0110, 1, N3RegFrm,
2472 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2473 "vmin", "s", int_arm_neon_vmins, 1>;
2474 defm VMINu : N3VInt_QHS<1, 0, 0b0110, 1, N3RegFrm,
2475 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2476 "vmin", "u", int_arm_neon_vminu, 1>;
2477 def VMINfd : N3VDInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBIND,
2479 v2f32, v2f32, int_arm_neon_vmins, 1>;
2480 def VMINfq : N3VQInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBINQ,
2482 v4f32, v4f32, int_arm_neon_vmins, 1>;
2484 // Vector Pairwise Operations.
2486 // VPADD : Vector Pairwise Add
2487 def VPADDi8 : N3VDInt<0, 0, 0b00, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
2489 v8i8, v8i8, int_arm_neon_vpadd, 0>;
2490 def VPADDi16 : N3VDInt<0, 0, 0b01, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
2492 v4i16, v4i16, int_arm_neon_vpadd, 0>;
2493 def VPADDi32 : N3VDInt<0, 0, 0b10, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
2495 v2i32, v2i32, int_arm_neon_vpadd, 0>;
2496 def VPADDf : N3VDInt<1, 0, 0b00, 0b1101, 0, N3RegFrm,
2497 IIC_VBIND, "vpadd", "f32",
2498 v2f32, v2f32, int_arm_neon_vpadd, 0>;
2500 // VPADDL : Vector Pairwise Add Long
2501 defm VPADDLs : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00100, 0, "vpaddl", "s",
2502 int_arm_neon_vpaddls>;
2503 defm VPADDLu : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00101, 0, "vpaddl", "u",
2504 int_arm_neon_vpaddlu>;
2506 // VPADAL : Vector Pairwise Add and Accumulate Long
2507 defm VPADALs : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01100, 0, "vpadal", "s",
2508 int_arm_neon_vpadals>;
2509 defm VPADALu : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01101, 0, "vpadal", "u",
2510 int_arm_neon_vpadalu>;
2512 // VPMAX : Vector Pairwise Maximum
2513 def VPMAXs8 : N3VDInt<0, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
2514 "s8", v8i8, v8i8, int_arm_neon_vpmaxs, 0>;
2515 def VPMAXs16 : N3VDInt<0, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
2516 "s16", v4i16, v4i16, int_arm_neon_vpmaxs, 0>;
2517 def VPMAXs32 : N3VDInt<0, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
2518 "s32", v2i32, v2i32, int_arm_neon_vpmaxs, 0>;
2519 def VPMAXu8 : N3VDInt<1, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
2520 "u8", v8i8, v8i8, int_arm_neon_vpmaxu, 0>;
2521 def VPMAXu16 : N3VDInt<1, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
2522 "u16", v4i16, v4i16, int_arm_neon_vpmaxu, 0>;
2523 def VPMAXu32 : N3VDInt<1, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
2524 "u32", v2i32, v2i32, int_arm_neon_vpmaxu, 0>;
2525 def VPMAXf : N3VDInt<1, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
2526 "f32", v2f32, v2f32, int_arm_neon_vpmaxs, 0>;
2528 // VPMIN : Vector Pairwise Minimum
2529 def VPMINs8 : N3VDInt<0, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
2530 "s8", v8i8, v8i8, int_arm_neon_vpmins, 0>;
2531 def VPMINs16 : N3VDInt<0, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
2532 "s16", v4i16, v4i16, int_arm_neon_vpmins, 0>;
2533 def VPMINs32 : N3VDInt<0, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
2534 "s32", v2i32, v2i32, int_arm_neon_vpmins, 0>;
2535 def VPMINu8 : N3VDInt<1, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
2536 "u8", v8i8, v8i8, int_arm_neon_vpminu, 0>;
2537 def VPMINu16 : N3VDInt<1, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
2538 "u16", v4i16, v4i16, int_arm_neon_vpminu, 0>;
2539 def VPMINu32 : N3VDInt<1, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
2540 "u32", v2i32, v2i32, int_arm_neon_vpminu, 0>;
2541 def VPMINf : N3VDInt<1, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VSUBi4D, "vpmin",
2542 "f32", v2f32, v2f32, int_arm_neon_vpmins, 0>;
2544 // Vector Reciprocal and Reciprocal Square Root Estimate and Step.
2546 // VRECPE : Vector Reciprocal Estimate
2547 def VRECPEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
2548 IIC_VUNAD, "vrecpe", "u32",
2549 v2i32, v2i32, int_arm_neon_vrecpe>;
2550 def VRECPEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
2551 IIC_VUNAQ, "vrecpe", "u32",
2552 v4i32, v4i32, int_arm_neon_vrecpe>;
2553 def VRECPEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
2554 IIC_VUNAD, "vrecpe", "f32",
2555 v2f32, v2f32, int_arm_neon_vrecpe>;
2556 def VRECPEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
2557 IIC_VUNAQ, "vrecpe", "f32",
2558 v4f32, v4f32, int_arm_neon_vrecpe>;
2560 // VRECPS : Vector Reciprocal Step
2561 def VRECPSfd : N3VDInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
2562 IIC_VRECSD, "vrecps", "f32",
2563 v2f32, v2f32, int_arm_neon_vrecps, 1>;
2564 def VRECPSfq : N3VQInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
2565 IIC_VRECSQ, "vrecps", "f32",
2566 v4f32, v4f32, int_arm_neon_vrecps, 1>;
2568 // VRSQRTE : Vector Reciprocal Square Root Estimate
2569 def VRSQRTEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
2570 IIC_VUNAD, "vrsqrte", "u32",
2571 v2i32, v2i32, int_arm_neon_vrsqrte>;
2572 def VRSQRTEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
2573 IIC_VUNAQ, "vrsqrte", "u32",
2574 v4i32, v4i32, int_arm_neon_vrsqrte>;
2575 def VRSQRTEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
2576 IIC_VUNAD, "vrsqrte", "f32",
2577 v2f32, v2f32, int_arm_neon_vrsqrte>;
2578 def VRSQRTEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
2579 IIC_VUNAQ, "vrsqrte", "f32",
2580 v4f32, v4f32, int_arm_neon_vrsqrte>;
2582 // VRSQRTS : Vector Reciprocal Square Root Step
2583 def VRSQRTSfd : N3VDInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
2584 IIC_VRECSD, "vrsqrts", "f32",
2585 v2f32, v2f32, int_arm_neon_vrsqrts, 1>;
2586 def VRSQRTSfq : N3VQInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
2587 IIC_VRECSQ, "vrsqrts", "f32",
2588 v4f32, v4f32, int_arm_neon_vrsqrts, 1>;
2592 // VSHL : Vector Shift
2593 defm VSHLs : N3VInt_QHSD<0, 0, 0b0100, 0, N3RegVShFrm,
2594 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
2595 "vshl", "s", int_arm_neon_vshifts, 0>;
2596 defm VSHLu : N3VInt_QHSD<1, 0, 0b0100, 0, N3RegVShFrm,
2597 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
2598 "vshl", "u", int_arm_neon_vshiftu, 0>;
2599 // VSHL : Vector Shift Left (Immediate)
2600 defm VSHLi : N2VSh_QHSD<0, 1, 0b0101, 1, IIC_VSHLiD, "vshl", "i", NEONvshl,
2602 // VSHR : Vector Shift Right (Immediate)
2603 defm VSHRs : N2VSh_QHSD<0, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "s", NEONvshrs,
2605 defm VSHRu : N2VSh_QHSD<1, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "u", NEONvshru,
2608 // VSHLL : Vector Shift Left Long
2609 defm VSHLLs : N2VLSh_QHS<0, 1, 0b1010, 0, 0, 1, "vshll", "s", NEONvshlls>;
2610 defm VSHLLu : N2VLSh_QHS<1, 1, 0b1010, 0, 0, 1, "vshll", "u", NEONvshllu>;
2612 // VSHLL : Vector Shift Left Long (with maximum shift count)
2613 class N2VLShMax<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
2614 bit op6, bit op4, string OpcodeStr, string Dt, ValueType ResTy,
2615 ValueType OpTy, SDNode OpNode>
2616 : N2VLSh<op24, op23, op11_8, op7, op6, op4, OpcodeStr, Dt,
2617 ResTy, OpTy, OpNode> {
2618 let Inst{21-16} = op21_16;
2620 def VSHLLi8 : N2VLShMax<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll", "i8",
2621 v8i16, v8i8, NEONvshlli>;
2622 def VSHLLi16 : N2VLShMax<1, 1, 0b110110, 0b0011, 0, 0, 0, "vshll", "i16",
2623 v4i32, v4i16, NEONvshlli>;
2624 def VSHLLi32 : N2VLShMax<1, 1, 0b111010, 0b0011, 0, 0, 0, "vshll", "i32",
2625 v2i64, v2i32, NEONvshlli>;
2627 // VSHRN : Vector Shift Right and Narrow
2628 defm VSHRN : N2VNSh_HSD<0,1,0b1000,0,0,1, IIC_VSHLiD, "vshrn", "i",
2631 // VRSHL : Vector Rounding Shift
2632 defm VRSHLs : N3VInt_QHSD<0, 0, 0b0101, 0, N3RegVShFrm,
2633 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
2634 "vrshl", "s", int_arm_neon_vrshifts, 0>;
2635 defm VRSHLu : N3VInt_QHSD<1, 0, 0b0101, 0, N3RegVShFrm,
2636 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
2637 "vrshl", "u", int_arm_neon_vrshiftu, 0>;
2638 // VRSHR : Vector Rounding Shift Right
2639 defm VRSHRs : N2VSh_QHSD<0,1,0b0010,1, IIC_VSHLi4D, "vrshr", "s", NEONvrshrs,
2641 defm VRSHRu : N2VSh_QHSD<1,1,0b0010,1, IIC_VSHLi4D, "vrshr", "u", NEONvrshru,
2644 // VRSHRN : Vector Rounding Shift Right and Narrow
2645 defm VRSHRN : N2VNSh_HSD<0, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vrshrn", "i",
2648 // VQSHL : Vector Saturating Shift
2649 defm VQSHLs : N3VInt_QHSD<0, 0, 0b0100, 1, N3RegVShFrm,
2650 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
2651 "vqshl", "s", int_arm_neon_vqshifts, 0>;
2652 defm VQSHLu : N3VInt_QHSD<1, 0, 0b0100, 1, N3RegVShFrm,
2653 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
2654 "vqshl", "u", int_arm_neon_vqshiftu, 0>;
2655 // VQSHL : Vector Saturating Shift Left (Immediate)
2656 defm VQSHLsi : N2VSh_QHSD<0,1,0b0111,1, IIC_VSHLi4D, "vqshl", "s",NEONvqshls,
2658 defm VQSHLui : N2VSh_QHSD<1,1,0b0111,1, IIC_VSHLi4D, "vqshl", "u",NEONvqshlu,
2660 // VQSHLU : Vector Saturating Shift Left (Immediate, Unsigned)
2661 defm VQSHLsu : N2VSh_QHSD<1,1,0b0110,1, IIC_VSHLi4D,"vqshlu","s",NEONvqshlsu,
2664 // VQSHRN : Vector Saturating Shift Right and Narrow
2665 defm VQSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "s",
2667 defm VQSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "u",
2670 // VQSHRUN : Vector Saturating Shift Right and Narrow (Unsigned)
2671 defm VQSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 0, 1, IIC_VSHLi4D, "vqshrun", "s",
2674 // VQRSHL : Vector Saturating Rounding Shift
2675 defm VQRSHLs : N3VInt_QHSD<0, 0, 0b0101, 1, N3RegVShFrm,
2676 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
2677 "vqrshl", "s", int_arm_neon_vqrshifts, 0>;
2678 defm VQRSHLu : N3VInt_QHSD<1, 0, 0b0101, 1, N3RegVShFrm,
2679 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
2680 "vqrshl", "u", int_arm_neon_vqrshiftu, 0>;
2682 // VQRSHRN : Vector Saturating Rounding Shift Right and Narrow
2683 defm VQRSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "s",
2685 defm VQRSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "u",
2688 // VQRSHRUN : Vector Saturating Rounding Shift Right and Narrow (Unsigned)
2689 defm VQRSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vqrshrun", "s",
2692 // VSRA : Vector Shift Right and Accumulate
2693 defm VSRAs : N2VShAdd_QHSD<0, 1, 0b0001, 1, "vsra", "s", NEONvshrs>;
2694 defm VSRAu : N2VShAdd_QHSD<1, 1, 0b0001, 1, "vsra", "u", NEONvshru>;
2695 // VRSRA : Vector Rounding Shift Right and Accumulate
2696 defm VRSRAs : N2VShAdd_QHSD<0, 1, 0b0011, 1, "vrsra", "s", NEONvrshrs>;
2697 defm VRSRAu : N2VShAdd_QHSD<1, 1, 0b0011, 1, "vrsra", "u", NEONvrshru>;
2699 // VSLI : Vector Shift Left and Insert
2700 defm VSLI : N2VShIns_QHSD<1, 1, 0b0101, 1, "vsli", NEONvsli, N2RegVShLFrm>;
2701 // VSRI : Vector Shift Right and Insert
2702 defm VSRI : N2VShIns_QHSD<1, 1, 0b0100, 1, "vsri", NEONvsri, N2RegVShRFrm>;
2704 // Vector Absolute and Saturating Absolute.
2706 // VABS : Vector Absolute Value
2707 defm VABS : N2VInt_QHS<0b11, 0b11, 0b01, 0b00110, 0,
2708 IIC_VUNAiD, IIC_VUNAiQ, "vabs", "s",
2710 def VABSfd : N2VDInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
2711 IIC_VUNAD, "vabs", "f32",
2712 v2f32, v2f32, int_arm_neon_vabs>;
2713 def VABSfq : N2VQInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
2714 IIC_VUNAQ, "vabs", "f32",
2715 v4f32, v4f32, int_arm_neon_vabs>;
2717 // VQABS : Vector Saturating Absolute Value
2718 defm VQABS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01110, 0,
2719 IIC_VQUNAiD, IIC_VQUNAiQ, "vqabs", "s",
2720 int_arm_neon_vqabs>;
2724 def vneg : PatFrag<(ops node:$in), (sub immAllZerosV, node:$in)>;
2725 def vneg8 : PatFrag<(ops node:$in),
2726 (sub (bitconvert (v8i8 immAllZerosV)), node:$in)>;
2727 def vneg16 : PatFrag<(ops node:$in),
2728 (sub (bitconvert (v16i8 immAllZerosV)), node:$in)>;
2730 class VNEGD<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
2731 : N2V<0b11, 0b11, size, 0b01, 0b00111, 0, 0, (outs DPR:$dst), (ins DPR:$src),
2732 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
2733 [(set DPR:$dst, (Ty (vneg8 DPR:$src)))]>;
2734 class VNEGQ<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
2735 : N2V<0b11, 0b11, size, 0b01, 0b00111, 1, 0, (outs QPR:$dst), (ins QPR:$src),
2736 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
2737 [(set QPR:$dst, (Ty (vneg16 QPR:$src)))]>;
2739 // VNEG : Vector Negate (integer)
2740 def VNEGs8d : VNEGD<0b00, "vneg", "s8", v8i8>;
2741 def VNEGs16d : VNEGD<0b01, "vneg", "s16", v4i16>;
2742 def VNEGs32d : VNEGD<0b10, "vneg", "s32", v2i32>;
2743 def VNEGs8q : VNEGQ<0b00, "vneg", "s8", v16i8>;
2744 def VNEGs16q : VNEGQ<0b01, "vneg", "s16", v8i16>;
2745 def VNEGs32q : VNEGQ<0b10, "vneg", "s32", v4i32>;
2747 // VNEG : Vector Negate (floating-point)
2748 def VNEGfd : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
2749 (outs DPR:$dst), (ins DPR:$src), IIC_VUNAD,
2750 "vneg", "f32", "$dst, $src", "",
2751 [(set DPR:$dst, (v2f32 (fneg DPR:$src)))]>;
2752 def VNEGf32q : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 1, 0,
2753 (outs QPR:$dst), (ins QPR:$src), IIC_VUNAQ,
2754 "vneg", "f32", "$dst, $src", "",
2755 [(set QPR:$dst, (v4f32 (fneg QPR:$src)))]>;
2757 def : Pat<(v8i8 (vneg8 DPR:$src)), (VNEGs8d DPR:$src)>;
2758 def : Pat<(v4i16 (vneg8 DPR:$src)), (VNEGs16d DPR:$src)>;
2759 def : Pat<(v2i32 (vneg8 DPR:$src)), (VNEGs32d DPR:$src)>;
2760 def : Pat<(v16i8 (vneg16 QPR:$src)), (VNEGs8q QPR:$src)>;
2761 def : Pat<(v8i16 (vneg16 QPR:$src)), (VNEGs16q QPR:$src)>;
2762 def : Pat<(v4i32 (vneg16 QPR:$src)), (VNEGs32q QPR:$src)>;
2764 // VQNEG : Vector Saturating Negate
2765 defm VQNEG : N2VInt_QHS<0b11, 0b11, 0b00, 0b01111, 0,
2766 IIC_VQUNAiD, IIC_VQUNAiQ, "vqneg", "s",
2767 int_arm_neon_vqneg>;
2769 // Vector Bit Counting Operations.
2771 // VCLS : Vector Count Leading Sign Bits
2772 defm VCLS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01000, 0,
2773 IIC_VCNTiD, IIC_VCNTiQ, "vcls", "s",
2775 // VCLZ : Vector Count Leading Zeros
2776 defm VCLZ : N2VInt_QHS<0b11, 0b11, 0b00, 0b01001, 0,
2777 IIC_VCNTiD, IIC_VCNTiQ, "vclz", "i",
2779 // VCNT : Vector Count One Bits
2780 def VCNTd : N2VDInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
2781 IIC_VCNTiD, "vcnt", "8",
2782 v8i8, v8i8, int_arm_neon_vcnt>;
2783 def VCNTq : N2VQInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
2784 IIC_VCNTiQ, "vcnt", "8",
2785 v16i8, v16i8, int_arm_neon_vcnt>;
2787 // Vector Swap -- for disassembly only.
2788 def VSWPd : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 0, 0,
2789 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
2790 "vswp", "$dst, $src", "", []>;
2791 def VSWPq : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 1, 0,
2792 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
2793 "vswp", "$dst, $src", "", []>;
2795 // Vector Move Operations.
2797 // VMOV : Vector Move (Register)
2799 let neverHasSideEffects = 1 in {
2800 def VMOVDneon: N3VX<0, 0, 0b10, 0b0001, 0, 1, (outs DPR:$dst), (ins DPR:$src),
2801 N3RegFrm, IIC_VMOVD, "vmov", "$dst, $src", "", []>;
2802 def VMOVQ : N3VX<0, 0, 0b10, 0b0001, 1, 1, (outs QPR:$dst), (ins QPR:$src),
2803 N3RegFrm, IIC_VMOVD, "vmov", "$dst, $src", "", []>;
2805 // Pseudo vector move instruction for QQ (a pair of Q) registers. This should
2806 // be expanded after register allocation is completed.
2807 def VMOVQQ : PseudoInst<(outs QQPR:$dst), (ins QQPR:$src),
2808 NoItinerary, "@ vmov\t$dst, $src", []>;
2809 } // neverHasSideEffects
2811 // VMOV : Vector Move (Immediate)
2813 // VMOV_get_imm8 xform function: convert build_vector to VMOV.i8 imm.
2814 def VMOV_get_imm8 : SDNodeXForm<build_vector, [{
2815 return ARM::getVMOVImm(N, 1, *CurDAG);
2817 def vmovImm8 : PatLeaf<(build_vector), [{
2818 return ARM::getVMOVImm(N, 1, *CurDAG).getNode() != 0;
2821 // VMOV_get_imm16 xform function: convert build_vector to VMOV.i16 imm.
2822 def VMOV_get_imm16 : SDNodeXForm<build_vector, [{
2823 return ARM::getVMOVImm(N, 2, *CurDAG);
2825 def vmovImm16 : PatLeaf<(build_vector), [{
2826 return ARM::getVMOVImm(N, 2, *CurDAG).getNode() != 0;
2827 }], VMOV_get_imm16>;
2829 // VMOV_get_imm32 xform function: convert build_vector to VMOV.i32 imm.
2830 def VMOV_get_imm32 : SDNodeXForm<build_vector, [{
2831 return ARM::getVMOVImm(N, 4, *CurDAG);
2833 def vmovImm32 : PatLeaf<(build_vector), [{
2834 return ARM::getVMOVImm(N, 4, *CurDAG).getNode() != 0;
2835 }], VMOV_get_imm32>;
2837 // VMOV_get_imm64 xform function: convert build_vector to VMOV.i64 imm.
2838 def VMOV_get_imm64 : SDNodeXForm<build_vector, [{
2839 return ARM::getVMOVImm(N, 8, *CurDAG);
2841 def vmovImm64 : PatLeaf<(build_vector), [{
2842 return ARM::getVMOVImm(N, 8, *CurDAG).getNode() != 0;
2843 }], VMOV_get_imm64>;
2845 // Note: Some of the cmode bits in the following VMOV instructions need to
2846 // be encoded based on the immed values.
2848 def VMOVv8i8 : N1ModImm<1, 0b000, 0b1110, 0, 0, 0, 1, (outs DPR:$dst),
2849 (ins h8imm:$SIMM), IIC_VMOVImm,
2850 "vmov", "i8", "$dst, $SIMM", "",
2851 [(set DPR:$dst, (v8i8 vmovImm8:$SIMM))]>;
2852 def VMOVv16i8 : N1ModImm<1, 0b000, 0b1110, 0, 1, 0, 1, (outs QPR:$dst),
2853 (ins h8imm:$SIMM), IIC_VMOVImm,
2854 "vmov", "i8", "$dst, $SIMM", "",
2855 [(set QPR:$dst, (v16i8 vmovImm8:$SIMM))]>;
2857 def VMOVv4i16 : N1ModImm<1, 0b000, {1,0,?,?}, 0, 0, {?}, 1, (outs DPR:$dst),
2858 (ins h16imm:$SIMM), IIC_VMOVImm,
2859 "vmov", "i16", "$dst, $SIMM", "",
2860 [(set DPR:$dst, (v4i16 vmovImm16:$SIMM))]>;
2861 def VMOVv8i16 : N1ModImm<1, 0b000, {1,0,?,?}, 0, 1, {?}, 1, (outs QPR:$dst),
2862 (ins h16imm:$SIMM), IIC_VMOVImm,
2863 "vmov", "i16", "$dst, $SIMM", "",
2864 [(set QPR:$dst, (v8i16 vmovImm16:$SIMM))]>;
2866 def VMOVv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, {?}, 1, (outs DPR:$dst),
2867 (ins h32imm:$SIMM), IIC_VMOVImm,
2868 "vmov", "i32", "$dst, $SIMM", "",
2869 [(set DPR:$dst, (v2i32 vmovImm32:$SIMM))]>;
2870 def VMOVv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, {?}, 1, (outs QPR:$dst),
2871 (ins h32imm:$SIMM), IIC_VMOVImm,
2872 "vmov", "i32", "$dst, $SIMM", "",
2873 [(set QPR:$dst, (v4i32 vmovImm32:$SIMM))]>;
2875 def VMOVv1i64 : N1ModImm<1, 0b000, 0b1110, 0, 0, 1, 1, (outs DPR:$dst),
2876 (ins h64imm:$SIMM), IIC_VMOVImm,
2877 "vmov", "i64", "$dst, $SIMM", "",
2878 [(set DPR:$dst, (v1i64 vmovImm64:$SIMM))]>;
2879 def VMOVv2i64 : N1ModImm<1, 0b000, 0b1110, 0, 1, 1, 1, (outs QPR:$dst),
2880 (ins h64imm:$SIMM), IIC_VMOVImm,
2881 "vmov", "i64", "$dst, $SIMM", "",
2882 [(set QPR:$dst, (v2i64 vmovImm64:$SIMM))]>;
2884 // VMOV : Vector Get Lane (move scalar to ARM core register)
2886 def VGETLNs8 : NVGetLane<{1,1,1,0,0,1,?,1}, 0b1011, {?,?},
2887 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2888 IIC_VMOVSI, "vmov", "s8", "$dst, $src[$lane]",
2889 [(set GPR:$dst, (NEONvgetlanes (v8i8 DPR:$src),
2891 def VGETLNs16 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, {?,1},
2892 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2893 IIC_VMOVSI, "vmov", "s16", "$dst, $src[$lane]",
2894 [(set GPR:$dst, (NEONvgetlanes (v4i16 DPR:$src),
2896 def VGETLNu8 : NVGetLane<{1,1,1,0,1,1,?,1}, 0b1011, {?,?},
2897 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2898 IIC_VMOVSI, "vmov", "u8", "$dst, $src[$lane]",
2899 [(set GPR:$dst, (NEONvgetlaneu (v8i8 DPR:$src),
2901 def VGETLNu16 : NVGetLane<{1,1,1,0,1,0,?,1}, 0b1011, {?,1},
2902 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2903 IIC_VMOVSI, "vmov", "u16", "$dst, $src[$lane]",
2904 [(set GPR:$dst, (NEONvgetlaneu (v4i16 DPR:$src),
2906 def VGETLNi32 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, 0b00,
2907 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2908 IIC_VMOVSI, "vmov", "32", "$dst, $src[$lane]",
2909 [(set GPR:$dst, (extractelt (v2i32 DPR:$src),
2911 // def VGETLNf32: see FMRDH and FMRDL in ARMInstrVFP.td
2912 def : Pat<(NEONvgetlanes (v16i8 QPR:$src), imm:$lane),
2913 (VGETLNs8 (v8i8 (EXTRACT_SUBREG QPR:$src,
2914 (DSubReg_i8_reg imm:$lane))),
2915 (SubReg_i8_lane imm:$lane))>;
2916 def : Pat<(NEONvgetlanes (v8i16 QPR:$src), imm:$lane),
2917 (VGETLNs16 (v4i16 (EXTRACT_SUBREG QPR:$src,
2918 (DSubReg_i16_reg imm:$lane))),
2919 (SubReg_i16_lane imm:$lane))>;
2920 def : Pat<(NEONvgetlaneu (v16i8 QPR:$src), imm:$lane),
2921 (VGETLNu8 (v8i8 (EXTRACT_SUBREG QPR:$src,
2922 (DSubReg_i8_reg imm:$lane))),
2923 (SubReg_i8_lane imm:$lane))>;
2924 def : Pat<(NEONvgetlaneu (v8i16 QPR:$src), imm:$lane),
2925 (VGETLNu16 (v4i16 (EXTRACT_SUBREG QPR:$src,
2926 (DSubReg_i16_reg imm:$lane))),
2927 (SubReg_i16_lane imm:$lane))>;
2928 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
2929 (VGETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src,
2930 (DSubReg_i32_reg imm:$lane))),
2931 (SubReg_i32_lane imm:$lane))>;
2932 def : Pat<(extractelt (v2f32 DPR:$src1), imm:$src2),
2933 (EXTRACT_SUBREG (v2f32 (COPY_TO_REGCLASS (v2f32 DPR:$src1),DPR_VFP2)),
2934 (SSubReg_f32_reg imm:$src2))>;
2935 def : Pat<(extractelt (v4f32 QPR:$src1), imm:$src2),
2936 (EXTRACT_SUBREG (v4f32 (COPY_TO_REGCLASS (v4f32 QPR:$src1),QPR_VFP2)),
2937 (SSubReg_f32_reg imm:$src2))>;
2938 //def : Pat<(extractelt (v2i64 QPR:$src1), imm:$src2),
2939 // (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
2940 def : Pat<(extractelt (v2f64 QPR:$src1), imm:$src2),
2941 (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
2944 // VMOV : Vector Set Lane (move ARM core register to scalar)
2946 let Constraints = "$src1 = $dst" in {
2947 def VSETLNi8 : NVSetLane<{1,1,1,0,0,1,?,0}, 0b1011, {?,?}, (outs DPR:$dst),
2948 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
2949 IIC_VMOVISL, "vmov", "8", "$dst[$lane], $src2",
2950 [(set DPR:$dst, (vector_insert (v8i8 DPR:$src1),
2951 GPR:$src2, imm:$lane))]>;
2952 def VSETLNi16 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, {?,1}, (outs DPR:$dst),
2953 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
2954 IIC_VMOVISL, "vmov", "16", "$dst[$lane], $src2",
2955 [(set DPR:$dst, (vector_insert (v4i16 DPR:$src1),
2956 GPR:$src2, imm:$lane))]>;
2957 def VSETLNi32 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, 0b00, (outs DPR:$dst),
2958 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
2959 IIC_VMOVISL, "vmov", "32", "$dst[$lane], $src2",
2960 [(set DPR:$dst, (insertelt (v2i32 DPR:$src1),
2961 GPR:$src2, imm:$lane))]>;
2963 def : Pat<(vector_insert (v16i8 QPR:$src1), GPR:$src2, imm:$lane),
2964 (v16i8 (INSERT_SUBREG QPR:$src1,
2965 (v8i8 (VSETLNi8 (v8i8 (EXTRACT_SUBREG QPR:$src1,
2966 (DSubReg_i8_reg imm:$lane))),
2967 GPR:$src2, (SubReg_i8_lane imm:$lane))),
2968 (DSubReg_i8_reg imm:$lane)))>;
2969 def : Pat<(vector_insert (v8i16 QPR:$src1), GPR:$src2, imm:$lane),
2970 (v8i16 (INSERT_SUBREG QPR:$src1,
2971 (v4i16 (VSETLNi16 (v4i16 (EXTRACT_SUBREG QPR:$src1,
2972 (DSubReg_i16_reg imm:$lane))),
2973 GPR:$src2, (SubReg_i16_lane imm:$lane))),
2974 (DSubReg_i16_reg imm:$lane)))>;
2975 def : Pat<(insertelt (v4i32 QPR:$src1), GPR:$src2, imm:$lane),
2976 (v4i32 (INSERT_SUBREG QPR:$src1,
2977 (v2i32 (VSETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src1,
2978 (DSubReg_i32_reg imm:$lane))),
2979 GPR:$src2, (SubReg_i32_lane imm:$lane))),
2980 (DSubReg_i32_reg imm:$lane)))>;
2982 def : Pat<(v2f32 (insertelt DPR:$src1, SPR:$src2, imm:$src3)),
2983 (INSERT_SUBREG (v2f32 (COPY_TO_REGCLASS DPR:$src1, DPR_VFP2)),
2984 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
2985 def : Pat<(v4f32 (insertelt QPR:$src1, SPR:$src2, imm:$src3)),
2986 (INSERT_SUBREG (v4f32 (COPY_TO_REGCLASS QPR:$src1, QPR_VFP2)),
2987 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
2989 //def : Pat<(v2i64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
2990 // (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
2991 def : Pat<(v2f64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
2992 (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
2994 def : Pat<(v2f32 (scalar_to_vector SPR:$src)),
2995 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$src, arm_ssubreg_0)>;
2996 def : Pat<(v2f64 (scalar_to_vector (f64 DPR:$src))),
2997 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), DPR:$src, arm_dsubreg_0)>;
2998 def : Pat<(v4f32 (scalar_to_vector SPR:$src)),
2999 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), SPR:$src, arm_ssubreg_0)>;
3001 def : Pat<(v8i8 (scalar_to_vector GPR:$src)),
3002 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
3003 def : Pat<(v4i16 (scalar_to_vector GPR:$src)),
3004 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
3005 def : Pat<(v2i32 (scalar_to_vector GPR:$src)),
3006 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
3008 def : Pat<(v16i8 (scalar_to_vector GPR:$src)),
3009 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3010 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
3012 def : Pat<(v8i16 (scalar_to_vector GPR:$src)),
3013 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
3014 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
3016 def : Pat<(v4i32 (scalar_to_vector GPR:$src)),
3017 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
3018 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
3021 // VDUP : Vector Duplicate (from ARM core register to all elements)
3023 class VDUPD<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
3024 : NVDup<opcod1, 0b1011, opcod3, (outs DPR:$dst), (ins GPR:$src),
3025 IIC_VMOVIS, "vdup", Dt, "$dst, $src",
3026 [(set DPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
3027 class VDUPQ<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
3028 : NVDup<opcod1, 0b1011, opcod3, (outs QPR:$dst), (ins GPR:$src),
3029 IIC_VMOVIS, "vdup", Dt, "$dst, $src",
3030 [(set QPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
3032 def VDUP8d : VDUPD<0b11101100, 0b00, "8", v8i8>;
3033 def VDUP16d : VDUPD<0b11101000, 0b01, "16", v4i16>;
3034 def VDUP32d : VDUPD<0b11101000, 0b00, "32", v2i32>;
3035 def VDUP8q : VDUPQ<0b11101110, 0b00, "8", v16i8>;
3036 def VDUP16q : VDUPQ<0b11101010, 0b01, "16", v8i16>;
3037 def VDUP32q : VDUPQ<0b11101010, 0b00, "32", v4i32>;
3039 def VDUPfd : NVDup<0b11101000, 0b1011, 0b00, (outs DPR:$dst), (ins GPR:$src),
3040 IIC_VMOVIS, "vdup", "32", "$dst, $src",
3041 [(set DPR:$dst, (v2f32 (NEONvdup
3042 (f32 (bitconvert GPR:$src)))))]>;
3043 def VDUPfq : NVDup<0b11101010, 0b1011, 0b00, (outs QPR:$dst), (ins GPR:$src),
3044 IIC_VMOVIS, "vdup", "32", "$dst, $src",
3045 [(set QPR:$dst, (v4f32 (NEONvdup
3046 (f32 (bitconvert GPR:$src)))))]>;
3048 // VDUP : Vector Duplicate Lane (from scalar to all elements)
3050 class VDUPLND<bits<4> op19_16, string OpcodeStr, string Dt,
3052 : NVDupLane<op19_16, 0, (outs DPR:$dst), (ins DPR:$src, nohash_imm:$lane),
3053 IIC_VMOVD, OpcodeStr, Dt, "$dst, $src[$lane]",
3054 [(set DPR:$dst, (Ty (NEONvduplane (Ty DPR:$src), imm:$lane)))]>;
3056 class VDUPLNQ<bits<4> op19_16, string OpcodeStr, string Dt,
3057 ValueType ResTy, ValueType OpTy>
3058 : NVDupLane<op19_16, 1, (outs QPR:$dst), (ins DPR:$src, nohash_imm:$lane),
3059 IIC_VMOVD, OpcodeStr, Dt, "$dst, $src[$lane]",
3060 [(set QPR:$dst, (ResTy (NEONvduplane (OpTy DPR:$src),
3063 // Inst{19-16} is partially specified depending on the element size.
3065 def VDUPLN8d : VDUPLND<{?,?,?,1}, "vdup", "8", v8i8>;
3066 def VDUPLN16d : VDUPLND<{?,?,1,0}, "vdup", "16", v4i16>;
3067 def VDUPLN32d : VDUPLND<{?,1,0,0}, "vdup", "32", v2i32>;
3068 def VDUPLNfd : VDUPLND<{?,1,0,0}, "vdup", "32", v2f32>;
3069 def VDUPLN8q : VDUPLNQ<{?,?,?,1}, "vdup", "8", v16i8, v8i8>;
3070 def VDUPLN16q : VDUPLNQ<{?,?,1,0}, "vdup", "16", v8i16, v4i16>;
3071 def VDUPLN32q : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4i32, v2i32>;
3072 def VDUPLNfq : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4f32, v2f32>;
3074 def : Pat<(v16i8 (NEONvduplane (v16i8 QPR:$src), imm:$lane)),
3075 (v16i8 (VDUPLN8q (v8i8 (EXTRACT_SUBREG QPR:$src,
3076 (DSubReg_i8_reg imm:$lane))),
3077 (SubReg_i8_lane imm:$lane)))>;
3078 def : Pat<(v8i16 (NEONvduplane (v8i16 QPR:$src), imm:$lane)),
3079 (v8i16 (VDUPLN16q (v4i16 (EXTRACT_SUBREG QPR:$src,
3080 (DSubReg_i16_reg imm:$lane))),
3081 (SubReg_i16_lane imm:$lane)))>;
3082 def : Pat<(v4i32 (NEONvduplane (v4i32 QPR:$src), imm:$lane)),
3083 (v4i32 (VDUPLN32q (v2i32 (EXTRACT_SUBREG QPR:$src,
3084 (DSubReg_i32_reg imm:$lane))),
3085 (SubReg_i32_lane imm:$lane)))>;
3086 def : Pat<(v4f32 (NEONvduplane (v4f32 QPR:$src), imm:$lane)),
3087 (v4f32 (VDUPLNfq (v2f32 (EXTRACT_SUBREG QPR:$src,
3088 (DSubReg_i32_reg imm:$lane))),
3089 (SubReg_i32_lane imm:$lane)))>;
3091 def VDUPfdf : N2V<0b11, 0b11, {?,1}, {0,0}, 0b11000, 0, 0,
3092 (outs DPR:$dst), (ins SPR:$src),
3093 IIC_VMOVD, "vdup", "32", "$dst, ${src:lane}", "",
3094 [(set DPR:$dst, (v2f32 (NEONvdup (f32 SPR:$src))))]>;
3096 def VDUPfqf : N2V<0b11, 0b11, {?,1}, {0,0}, 0b11000, 1, 0,
3097 (outs QPR:$dst), (ins SPR:$src),
3098 IIC_VMOVD, "vdup", "32", "$dst, ${src:lane}", "",
3099 [(set QPR:$dst, (v4f32 (NEONvdup (f32 SPR:$src))))]>;
3101 def : Pat<(v2i64 (NEONvduplane (v2i64 QPR:$src), imm:$lane)),
3102 (INSERT_SUBREG QPR:$src,
3103 (i64 (EXTRACT_SUBREG QPR:$src,
3104 (DSubReg_f64_reg imm:$lane))),
3105 (DSubReg_f64_other_reg imm:$lane))>;
3106 def : Pat<(v2f64 (NEONvduplane (v2f64 QPR:$src), imm:$lane)),
3107 (INSERT_SUBREG QPR:$src,
3108 (f64 (EXTRACT_SUBREG QPR:$src,
3109 (DSubReg_f64_reg imm:$lane))),
3110 (DSubReg_f64_other_reg imm:$lane))>;
3112 // VMOVN : Vector Narrowing Move
3113 defm VMOVN : N2VNInt_HSD<0b11,0b11,0b10,0b00100,0,0, IIC_VMOVD,
3114 "vmovn", "i", int_arm_neon_vmovn>;
3115 // VQMOVN : Vector Saturating Narrowing Move
3116 defm VQMOVNs : N2VNInt_HSD<0b11,0b11,0b10,0b00101,0,0, IIC_VQUNAiD,
3117 "vqmovn", "s", int_arm_neon_vqmovns>;
3118 defm VQMOVNu : N2VNInt_HSD<0b11,0b11,0b10,0b00101,1,0, IIC_VQUNAiD,
3119 "vqmovn", "u", int_arm_neon_vqmovnu>;
3120 defm VQMOVNsu : N2VNInt_HSD<0b11,0b11,0b10,0b00100,1,0, IIC_VQUNAiD,
3121 "vqmovun", "s", int_arm_neon_vqmovnsu>;
3122 // VMOVL : Vector Lengthening Move
3123 defm VMOVLs : N2VLInt_QHS<0b01,0b10100,0,1, "vmovl", "s",
3124 int_arm_neon_vmovls>;
3125 defm VMOVLu : N2VLInt_QHS<0b11,0b10100,0,1, "vmovl", "u",
3126 int_arm_neon_vmovlu>;
3128 // Vector Conversions.
3130 // VCVT : Vector Convert Between Floating-Point and Integers
3131 def VCVTf2sd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
3132 v2i32, v2f32, fp_to_sint>;
3133 def VCVTf2ud : N2VD<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
3134 v2i32, v2f32, fp_to_uint>;
3135 def VCVTs2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
3136 v2f32, v2i32, sint_to_fp>;
3137 def VCVTu2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
3138 v2f32, v2i32, uint_to_fp>;
3140 def VCVTf2sq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
3141 v4i32, v4f32, fp_to_sint>;
3142 def VCVTf2uq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
3143 v4i32, v4f32, fp_to_uint>;
3144 def VCVTs2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
3145 v4f32, v4i32, sint_to_fp>;
3146 def VCVTu2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
3147 v4f32, v4i32, uint_to_fp>;
3149 // VCVT : Vector Convert Between Floating-Point and Fixed-Point.
3150 def VCVTf2xsd : N2VCvtD<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
3151 v2i32, v2f32, int_arm_neon_vcvtfp2fxs>;
3152 def VCVTf2xud : N2VCvtD<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
3153 v2i32, v2f32, int_arm_neon_vcvtfp2fxu>;
3154 def VCVTxs2fd : N2VCvtD<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
3155 v2f32, v2i32, int_arm_neon_vcvtfxs2fp>;
3156 def VCVTxu2fd : N2VCvtD<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
3157 v2f32, v2i32, int_arm_neon_vcvtfxu2fp>;
3159 def VCVTf2xsq : N2VCvtQ<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
3160 v4i32, v4f32, int_arm_neon_vcvtfp2fxs>;
3161 def VCVTf2xuq : N2VCvtQ<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
3162 v4i32, v4f32, int_arm_neon_vcvtfp2fxu>;
3163 def VCVTxs2fq : N2VCvtQ<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
3164 v4f32, v4i32, int_arm_neon_vcvtfxs2fp>;
3165 def VCVTxu2fq : N2VCvtQ<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
3166 v4f32, v4i32, int_arm_neon_vcvtfxu2fp>;
3170 // VREV64 : Vector Reverse elements within 64-bit doublewords
3172 class VREV64D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3173 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 0, 0, (outs DPR:$dst),
3174 (ins DPR:$src), IIC_VMOVD,
3175 OpcodeStr, Dt, "$dst, $src", "",
3176 [(set DPR:$dst, (Ty (NEONvrev64 (Ty DPR:$src))))]>;
3177 class VREV64Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3178 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 1, 0, (outs QPR:$dst),
3179 (ins QPR:$src), IIC_VMOVD,
3180 OpcodeStr, Dt, "$dst, $src", "",
3181 [(set QPR:$dst, (Ty (NEONvrev64 (Ty QPR:$src))))]>;
3183 def VREV64d8 : VREV64D<0b00, "vrev64", "8", v8i8>;
3184 def VREV64d16 : VREV64D<0b01, "vrev64", "16", v4i16>;
3185 def VREV64d32 : VREV64D<0b10, "vrev64", "32", v2i32>;
3186 def VREV64df : VREV64D<0b10, "vrev64", "32", v2f32>;
3188 def VREV64q8 : VREV64Q<0b00, "vrev64", "8", v16i8>;
3189 def VREV64q16 : VREV64Q<0b01, "vrev64", "16", v8i16>;
3190 def VREV64q32 : VREV64Q<0b10, "vrev64", "32", v4i32>;
3191 def VREV64qf : VREV64Q<0b10, "vrev64", "32", v4f32>;
3193 // VREV32 : Vector Reverse elements within 32-bit words
3195 class VREV32D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3196 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 0, 0, (outs DPR:$dst),
3197 (ins DPR:$src), IIC_VMOVD,
3198 OpcodeStr, Dt, "$dst, $src", "",
3199 [(set DPR:$dst, (Ty (NEONvrev32 (Ty DPR:$src))))]>;
3200 class VREV32Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3201 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 1, 0, (outs QPR:$dst),
3202 (ins QPR:$src), IIC_VMOVD,
3203 OpcodeStr, Dt, "$dst, $src", "",
3204 [(set QPR:$dst, (Ty (NEONvrev32 (Ty QPR:$src))))]>;
3206 def VREV32d8 : VREV32D<0b00, "vrev32", "8", v8i8>;
3207 def VREV32d16 : VREV32D<0b01, "vrev32", "16", v4i16>;
3209 def VREV32q8 : VREV32Q<0b00, "vrev32", "8", v16i8>;
3210 def VREV32q16 : VREV32Q<0b01, "vrev32", "16", v8i16>;
3212 // VREV16 : Vector Reverse elements within 16-bit halfwords
3214 class VREV16D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3215 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 0, 0, (outs DPR:$dst),
3216 (ins DPR:$src), IIC_VMOVD,
3217 OpcodeStr, Dt, "$dst, $src", "",
3218 [(set DPR:$dst, (Ty (NEONvrev16 (Ty DPR:$src))))]>;
3219 class VREV16Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3220 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 1, 0, (outs QPR:$dst),
3221 (ins QPR:$src), IIC_VMOVD,
3222 OpcodeStr, Dt, "$dst, $src", "",
3223 [(set QPR:$dst, (Ty (NEONvrev16 (Ty QPR:$src))))]>;
3225 def VREV16d8 : VREV16D<0b00, "vrev16", "8", v8i8>;
3226 def VREV16q8 : VREV16Q<0b00, "vrev16", "8", v16i8>;
3228 // Other Vector Shuffles.
3230 // VEXT : Vector Extract
3232 class VEXTd<string OpcodeStr, string Dt, ValueType Ty>
3233 : N3V<0,1,0b11,{?,?,?,?},0,0, (outs DPR:$dst),
3234 (ins DPR:$lhs, DPR:$rhs, i32imm:$index), NVExtFrm,
3235 IIC_VEXTD, OpcodeStr, Dt, "$dst, $lhs, $rhs, $index", "",
3236 [(set DPR:$dst, (Ty (NEONvext (Ty DPR:$lhs),
3237 (Ty DPR:$rhs), imm:$index)))]>;
3239 class VEXTq<string OpcodeStr, string Dt, ValueType Ty>
3240 : N3V<0,1,0b11,{?,?,?,?},1,0, (outs QPR:$dst),
3241 (ins QPR:$lhs, QPR:$rhs, i32imm:$index), NVExtFrm,
3242 IIC_VEXTQ, OpcodeStr, Dt, "$dst, $lhs, $rhs, $index", "",
3243 [(set QPR:$dst, (Ty (NEONvext (Ty QPR:$lhs),
3244 (Ty QPR:$rhs), imm:$index)))]>;
3246 def VEXTd8 : VEXTd<"vext", "8", v8i8>;
3247 def VEXTd16 : VEXTd<"vext", "16", v4i16>;
3248 def VEXTd32 : VEXTd<"vext", "32", v2i32>;
3249 def VEXTdf : VEXTd<"vext", "32", v2f32>;
3251 def VEXTq8 : VEXTq<"vext", "8", v16i8>;
3252 def VEXTq16 : VEXTq<"vext", "16", v8i16>;
3253 def VEXTq32 : VEXTq<"vext", "32", v4i32>;
3254 def VEXTqf : VEXTq<"vext", "32", v4f32>;
3256 // VTRN : Vector Transpose
3258 def VTRNd8 : N2VDShuffle<0b00, 0b00001, "vtrn", "8">;
3259 def VTRNd16 : N2VDShuffle<0b01, 0b00001, "vtrn", "16">;
3260 def VTRNd32 : N2VDShuffle<0b10, 0b00001, "vtrn", "32">;
3262 def VTRNq8 : N2VQShuffle<0b00, 0b00001, IIC_VPERMQ, "vtrn", "8">;
3263 def VTRNq16 : N2VQShuffle<0b01, 0b00001, IIC_VPERMQ, "vtrn", "16">;
3264 def VTRNq32 : N2VQShuffle<0b10, 0b00001, IIC_VPERMQ, "vtrn", "32">;
3266 // VUZP : Vector Unzip (Deinterleave)
3268 def VUZPd8 : N2VDShuffle<0b00, 0b00010, "vuzp", "8">;
3269 def VUZPd16 : N2VDShuffle<0b01, 0b00010, "vuzp", "16">;
3270 def VUZPd32 : N2VDShuffle<0b10, 0b00010, "vuzp", "32">;
3272 def VUZPq8 : N2VQShuffle<0b00, 0b00010, IIC_VPERMQ3, "vuzp", "8">;
3273 def VUZPq16 : N2VQShuffle<0b01, 0b00010, IIC_VPERMQ3, "vuzp", "16">;
3274 def VUZPq32 : N2VQShuffle<0b10, 0b00010, IIC_VPERMQ3, "vuzp", "32">;
3276 // VZIP : Vector Zip (Interleave)
3278 def VZIPd8 : N2VDShuffle<0b00, 0b00011, "vzip", "8">;
3279 def VZIPd16 : N2VDShuffle<0b01, 0b00011, "vzip", "16">;
3280 def VZIPd32 : N2VDShuffle<0b10, 0b00011, "vzip", "32">;
3282 def VZIPq8 : N2VQShuffle<0b00, 0b00011, IIC_VPERMQ3, "vzip", "8">;
3283 def VZIPq16 : N2VQShuffle<0b01, 0b00011, IIC_VPERMQ3, "vzip", "16">;
3284 def VZIPq32 : N2VQShuffle<0b10, 0b00011, IIC_VPERMQ3, "vzip", "32">;
3286 // Vector Table Lookup and Table Extension.
3288 // VTBL : Vector Table Lookup
3290 : N3V<1,1,0b11,0b1000,0,0, (outs DPR:$dst),
3291 (ins DPR:$tbl1, DPR:$src), NVTBLFrm, IIC_VTB1,
3292 "vtbl", "8", "$dst, \\{$tbl1\\}, $src", "",
3293 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl1 DPR:$tbl1, DPR:$src)))]>;
3294 let hasExtraSrcRegAllocReq = 1 in {
3296 : N3V<1,1,0b11,0b1001,0,0, (outs DPR:$dst),
3297 (ins DPR:$tbl1, DPR:$tbl2, DPR:$src), NVTBLFrm, IIC_VTB2,
3298 "vtbl", "8", "$dst, \\{$tbl1, $tbl2\\}, $src", "",
3299 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl2
3300 DPR:$tbl1, DPR:$tbl2, DPR:$src)))]>;
3302 : N3V<1,1,0b11,0b1010,0,0, (outs DPR:$dst),
3303 (ins DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$src), NVTBLFrm, IIC_VTB3,
3304 "vtbl", "8", "$dst, \\{$tbl1, $tbl2, $tbl3\\}, $src", "",
3305 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl3
3306 DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$src)))]>;
3308 : N3V<1,1,0b11,0b1011,0,0, (outs DPR:$dst),
3309 (ins DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$src),
3311 "vtbl", "8", "$dst, \\{$tbl1, $tbl2, $tbl3, $tbl4\\}, $src", "",
3312 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl4 DPR:$tbl1, DPR:$tbl2,
3313 DPR:$tbl3, DPR:$tbl4, DPR:$src)))]>;
3314 } // hasExtraSrcRegAllocReq = 1
3316 // VTBX : Vector Table Extension
3318 : N3V<1,1,0b11,0b1000,1,0, (outs DPR:$dst),
3319 (ins DPR:$orig, DPR:$tbl1, DPR:$src), NVTBLFrm, IIC_VTBX1,
3320 "vtbx", "8", "$dst, \\{$tbl1\\}, $src", "$orig = $dst",
3321 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx1
3322 DPR:$orig, DPR:$tbl1, DPR:$src)))]>;
3323 let hasExtraSrcRegAllocReq = 1 in {
3325 : N3V<1,1,0b11,0b1001,1,0, (outs DPR:$dst),
3326 (ins DPR:$orig, DPR:$tbl1, DPR:$tbl2, DPR:$src), NVTBLFrm, IIC_VTBX2,
3327 "vtbx", "8", "$dst, \\{$tbl1, $tbl2\\}, $src", "$orig = $dst",
3328 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx2
3329 DPR:$orig, DPR:$tbl1, DPR:$tbl2, DPR:$src)))]>;
3331 : N3V<1,1,0b11,0b1010,1,0, (outs DPR:$dst),
3332 (ins DPR:$orig, DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$src),
3333 NVTBLFrm, IIC_VTBX3,
3334 "vtbx", "8", "$dst, \\{$tbl1, $tbl2, $tbl3\\}, $src", "$orig = $dst",
3335 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx3 DPR:$orig, DPR:$tbl1,
3336 DPR:$tbl2, DPR:$tbl3, DPR:$src)))]>;
3338 : N3V<1,1,0b11,0b1011,1,0, (outs DPR:$dst), (ins DPR:$orig, DPR:$tbl1,
3339 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$src), NVTBLFrm, IIC_VTBX4,
3340 "vtbx", "8", "$dst, \\{$tbl1, $tbl2, $tbl3, $tbl4\\}, $src",
3342 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx4 DPR:$orig, DPR:$tbl1,
3343 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$src)))]>;
3344 } // hasExtraSrcRegAllocReq = 1
3346 //===----------------------------------------------------------------------===//
3347 // NEON instructions for single-precision FP math
3348 //===----------------------------------------------------------------------===//
3350 class N2VSPat<SDNode OpNode, ValueType ResTy, ValueType OpTy, NeonI Inst>
3351 : NEONFPPat<(ResTy (OpNode SPR:$a)),
3352 (EXTRACT_SUBREG (OpTy (Inst (INSERT_SUBREG (OpTy (IMPLICIT_DEF)),
3353 SPR:$a, arm_ssubreg_0))),
3356 class N3VSPat<SDNode OpNode, NeonI Inst>
3357 : NEONFPPat<(f32 (OpNode SPR:$a, SPR:$b)),
3358 (EXTRACT_SUBREG (v2f32
3359 (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3360 SPR:$a, arm_ssubreg_0),
3361 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3362 SPR:$b, arm_ssubreg_0))),
3365 class N3VSMulOpPat<SDNode MulNode, SDNode OpNode, NeonI Inst>
3366 : NEONFPPat<(f32 (OpNode SPR:$acc, (f32 (MulNode SPR:$a, SPR:$b)))),
3367 (EXTRACT_SUBREG (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3368 SPR:$acc, arm_ssubreg_0),
3369 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3370 SPR:$a, arm_ssubreg_0),
3371 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3372 SPR:$b, arm_ssubreg_0)),
3375 // These need separate instructions because they must use DPR_VFP2 register
3376 // class which have SPR sub-registers.
3378 // Vector Add Operations used for single-precision FP
3379 let neverHasSideEffects = 1 in
3380 def VADDfd_sfp : N3VS<0,0,0b00,0b1101,0, "vadd", "f32", v2f32, v2f32, fadd, 1>;
3381 def : N3VSPat<fadd, VADDfd_sfp>;
3383 // Vector Sub Operations used for single-precision FP
3384 let neverHasSideEffects = 1 in
3385 def VSUBfd_sfp : N3VS<0,0,0b10,0b1101,0, "vsub", "f32", v2f32, v2f32, fsub, 0>;
3386 def : N3VSPat<fsub, VSUBfd_sfp>;
3388 // Vector Multiply Operations used for single-precision FP
3389 let neverHasSideEffects = 1 in
3390 def VMULfd_sfp : N3VS<1,0,0b00,0b1101,1, "vmul", "f32", v2f32, v2f32, fmul, 1>;
3391 def : N3VSPat<fmul, VMULfd_sfp>;
3393 // Vector Multiply-Accumulate/Subtract used for single-precision FP
3394 // vml[as].f32 can cause 4-8 cycle stalls in following ASIMD instructions, so
3395 // we want to avoid them for now. e.g., alternating vmla/vadd instructions.
3397 //let neverHasSideEffects = 1 in
3398 //def VMLAfd_sfp : N3VSMulOp<0,0,0b00,0b1101,1, IIC_VMACD, "vmla", "f32",
3399 // v2f32, fmul, fadd>;
3400 //def : N3VSMulOpPat<fmul, fadd, VMLAfd_sfp>;
3402 //let neverHasSideEffects = 1 in
3403 //def VMLSfd_sfp : N3VSMulOp<0,0,0b10,0b1101,1, IIC_VMACD, "vmls", "f32",
3404 // v2f32, fmul, fsub>;
3405 //def : N3VSMulOpPat<fmul, fsub, VMLSfd_sfp>;
3407 // Vector Absolute used for single-precision FP
3408 let neverHasSideEffects = 1 in
3409 def VABSfd_sfp : N2V<0b11, 0b11, 0b10, 0b01, 0b01110, 0, 0,
3410 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), IIC_VUNAD,
3411 "vabs", "f32", "$dst, $src", "", []>;
3412 def : N2VSPat<fabs, f32, v2f32, VABSfd_sfp>;
3414 // Vector Negate used for single-precision FP
3415 let neverHasSideEffects = 1 in
3416 def VNEGfd_sfp : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
3417 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), IIC_VUNAD,
3418 "vneg", "f32", "$dst, $src", "", []>;
3419 def : N2VSPat<fneg, f32, v2f32, VNEGfd_sfp>;
3421 // Vector Maximum used for single-precision FP
3422 let neverHasSideEffects = 1 in
3423 def VMAXfd_sfp : N3V<0, 0, 0b00, 0b1111, 0, 0, (outs DPR_VFP2:$dst),
3424 (ins DPR_VFP2:$src1, DPR_VFP2:$src2), N3RegFrm, IIC_VBIND,
3425 "vmax", "f32", "$dst, $src1, $src2", "", []>;
3426 def : N3VSPat<NEONfmax, VMAXfd_sfp>;
3428 // Vector Minimum used for single-precision FP
3429 let neverHasSideEffects = 1 in
3430 def VMINfd_sfp : N3V<0, 0, 0b00, 0b1111, 0, 0, (outs DPR_VFP2:$dst),
3431 (ins DPR_VFP2:$src1, DPR_VFP2:$src2), N3RegFrm, IIC_VBIND,
3432 "vmin", "f32", "$dst, $src1, $src2", "", []>;
3433 def : N3VSPat<NEONfmin, VMINfd_sfp>;
3435 // Vector Convert between single-precision FP and integer
3436 let neverHasSideEffects = 1 in
3437 def VCVTf2sd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
3438 v2i32, v2f32, fp_to_sint>;
3439 def : N2VSPat<arm_ftosi, f32, v2f32, VCVTf2sd_sfp>;
3441 let neverHasSideEffects = 1 in
3442 def VCVTf2ud_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
3443 v2i32, v2f32, fp_to_uint>;
3444 def : N2VSPat<arm_ftoui, f32, v2f32, VCVTf2ud_sfp>;
3446 let neverHasSideEffects = 1 in
3447 def VCVTs2fd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
3448 v2f32, v2i32, sint_to_fp>;
3449 def : N2VSPat<arm_sitof, f32, v2i32, VCVTs2fd_sfp>;
3451 let neverHasSideEffects = 1 in
3452 def VCVTu2fd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
3453 v2f32, v2i32, uint_to_fp>;
3454 def : N2VSPat<arm_uitof, f32, v2i32, VCVTu2fd_sfp>;
3456 //===----------------------------------------------------------------------===//
3457 // Non-Instruction Patterns
3458 //===----------------------------------------------------------------------===//
3461 def : Pat<(v1i64 (bitconvert (v2i32 DPR:$src))), (v1i64 DPR:$src)>;
3462 def : Pat<(v1i64 (bitconvert (v4i16 DPR:$src))), (v1i64 DPR:$src)>;
3463 def : Pat<(v1i64 (bitconvert (v8i8 DPR:$src))), (v1i64 DPR:$src)>;
3464 def : Pat<(v1i64 (bitconvert (f64 DPR:$src))), (v1i64 DPR:$src)>;
3465 def : Pat<(v1i64 (bitconvert (v2f32 DPR:$src))), (v1i64 DPR:$src)>;
3466 def : Pat<(v2i32 (bitconvert (v1i64 DPR:$src))), (v2i32 DPR:$src)>;
3467 def : Pat<(v2i32 (bitconvert (v4i16 DPR:$src))), (v2i32 DPR:$src)>;
3468 def : Pat<(v2i32 (bitconvert (v8i8 DPR:$src))), (v2i32 DPR:$src)>;
3469 def : Pat<(v2i32 (bitconvert (f64 DPR:$src))), (v2i32 DPR:$src)>;
3470 def : Pat<(v2i32 (bitconvert (v2f32 DPR:$src))), (v2i32 DPR:$src)>;
3471 def : Pat<(v4i16 (bitconvert (v1i64 DPR:$src))), (v4i16 DPR:$src)>;
3472 def : Pat<(v4i16 (bitconvert (v2i32 DPR:$src))), (v4i16 DPR:$src)>;
3473 def : Pat<(v4i16 (bitconvert (v8i8 DPR:$src))), (v4i16 DPR:$src)>;
3474 def : Pat<(v4i16 (bitconvert (f64 DPR:$src))), (v4i16 DPR:$src)>;
3475 def : Pat<(v4i16 (bitconvert (v2f32 DPR:$src))), (v4i16 DPR:$src)>;
3476 def : Pat<(v8i8 (bitconvert (v1i64 DPR:$src))), (v8i8 DPR:$src)>;
3477 def : Pat<(v8i8 (bitconvert (v2i32 DPR:$src))), (v8i8 DPR:$src)>;
3478 def : Pat<(v8i8 (bitconvert (v4i16 DPR:$src))), (v8i8 DPR:$src)>;
3479 def : Pat<(v8i8 (bitconvert (f64 DPR:$src))), (v8i8 DPR:$src)>;
3480 def : Pat<(v8i8 (bitconvert (v2f32 DPR:$src))), (v8i8 DPR:$src)>;
3481 def : Pat<(f64 (bitconvert (v1i64 DPR:$src))), (f64 DPR:$src)>;
3482 def : Pat<(f64 (bitconvert (v2i32 DPR:$src))), (f64 DPR:$src)>;
3483 def : Pat<(f64 (bitconvert (v4i16 DPR:$src))), (f64 DPR:$src)>;
3484 def : Pat<(f64 (bitconvert (v8i8 DPR:$src))), (f64 DPR:$src)>;
3485 def : Pat<(f64 (bitconvert (v2f32 DPR:$src))), (f64 DPR:$src)>;
3486 def : Pat<(v2f32 (bitconvert (f64 DPR:$src))), (v2f32 DPR:$src)>;
3487 def : Pat<(v2f32 (bitconvert (v1i64 DPR:$src))), (v2f32 DPR:$src)>;
3488 def : Pat<(v2f32 (bitconvert (v2i32 DPR:$src))), (v2f32 DPR:$src)>;
3489 def : Pat<(v2f32 (bitconvert (v4i16 DPR:$src))), (v2f32 DPR:$src)>;
3490 def : Pat<(v2f32 (bitconvert (v8i8 DPR:$src))), (v2f32 DPR:$src)>;
3492 def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (v2i64 QPR:$src)>;
3493 def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (v2i64 QPR:$src)>;
3494 def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (v2i64 QPR:$src)>;
3495 def : Pat<(v2i64 (bitconvert (v2f64 QPR:$src))), (v2i64 QPR:$src)>;
3496 def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (v2i64 QPR:$src)>;
3497 def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (v4i32 QPR:$src)>;
3498 def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (v4i32 QPR:$src)>;
3499 def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (v4i32 QPR:$src)>;
3500 def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (v4i32 QPR:$src)>;
3501 def : Pat<(v4i32 (bitconvert (v4f32 QPR:$src))), (v4i32 QPR:$src)>;
3502 def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (v8i16 QPR:$src)>;
3503 def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (v8i16 QPR:$src)>;
3504 def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (v8i16 QPR:$src)>;
3505 def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (v8i16 QPR:$src)>;
3506 def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (v8i16 QPR:$src)>;
3507 def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (v16i8 QPR:$src)>;
3508 def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (v16i8 QPR:$src)>;
3509 def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (v16i8 QPR:$src)>;
3510 def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (v16i8 QPR:$src)>;
3511 def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (v16i8 QPR:$src)>;
3512 def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (v4f32 QPR:$src)>;
3513 def : Pat<(v4f32 (bitconvert (v4i32 QPR:$src))), (v4f32 QPR:$src)>;
3514 def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (v4f32 QPR:$src)>;
3515 def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (v4f32 QPR:$src)>;
3516 def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (v4f32 QPR:$src)>;
3517 def : Pat<(v2f64 (bitconvert (v2i64 QPR:$src))), (v2f64 QPR:$src)>;
3518 def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (v2f64 QPR:$src)>;
3519 def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 QPR:$src)>;
3520 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 QPR:$src)>;
3521 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 QPR:$src)>;