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 SDTARMVMOVIMM : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisVT<1, i32>]>;
69 def NEONvmovImm : SDNode<"ARMISD::VMOVIMM", SDTARMVMOVIMM>;
70 def NEONvmvnImm : SDNode<"ARMISD::VMVNIMM", SDTARMVMOVIMM>;
72 def NEONvdup : SDNode<"ARMISD::VDUP", SDTypeProfile<1, 1, [SDTCisVec<0>]>>;
74 // VDUPLANE can produce a quad-register result from a double-register source,
75 // so the result is not constrained to match the source.
76 def NEONvduplane : SDNode<"ARMISD::VDUPLANE",
77 SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisVec<1>,
80 def SDTARMVEXT : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
81 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
82 def NEONvext : SDNode<"ARMISD::VEXT", SDTARMVEXT>;
84 def SDTARMVSHUF : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisSameAs<0, 1>]>;
85 def NEONvrev64 : SDNode<"ARMISD::VREV64", SDTARMVSHUF>;
86 def NEONvrev32 : SDNode<"ARMISD::VREV32", SDTARMVSHUF>;
87 def NEONvrev16 : SDNode<"ARMISD::VREV16", SDTARMVSHUF>;
89 def SDTARMVSHUF2 : SDTypeProfile<2, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
92 def NEONzip : SDNode<"ARMISD::VZIP", SDTARMVSHUF2>;
93 def NEONuzp : SDNode<"ARMISD::VUZP", SDTARMVSHUF2>;
94 def NEONtrn : SDNode<"ARMISD::VTRN", SDTARMVSHUF2>;
96 def SDTARMFMAX : SDTypeProfile<1, 2, [SDTCisVT<0, f32>, SDTCisSameAs<0, 1>,
98 def NEONfmax : SDNode<"ARMISD::FMAX", SDTARMFMAX>;
99 def NEONfmin : SDNode<"ARMISD::FMIN", SDTARMFMAX>;
101 def NEONimmAllZerosV: PatLeaf<(NEONvmovImm (i32 timm)), [{
102 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
103 unsigned EltBits = 0;
104 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
105 return (EltBits == 32 && EltVal == 0);
108 def NEONimmAllOnesV: PatLeaf<(NEONvmovImm (i32 timm)), [{
109 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
110 unsigned EltBits = 0;
111 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
112 return (EltBits == 8 && EltVal == 0xff);
115 //===----------------------------------------------------------------------===//
116 // NEON operand definitions
117 //===----------------------------------------------------------------------===//
119 def nModImm : Operand<i32> {
120 let PrintMethod = "printNEONModImmOperand";
123 //===----------------------------------------------------------------------===//
124 // NEON load / store instructions
125 //===----------------------------------------------------------------------===//
127 let mayLoad = 1, neverHasSideEffects = 1 in {
128 // Use vldmia to load a Q register as a D register pair.
129 // This is equivalent to VLDMD except that it has a Q register operand
130 // instead of a pair of D registers.
132 : AXDI5<(outs QPR:$dst), (ins addrmode5:$addr, pred:$p),
133 IndexModeNone, IIC_fpLoadm,
134 "vldm${addr:submode}${p}\t${addr:base}, ${dst:dregpair}", "", []>;
136 // Use vld1 to load a Q register as a D register pair.
137 // This alternative to VLDMQ allows an alignment to be specified.
138 // This is equivalent to VLD1q64 except that it has a Q register operand.
140 : NLdSt<0,0b10,0b1010,0b1100, (outs QPR:$dst), (ins addrmode6:$addr),
141 IIC_VLD1, "vld1", "64", "${dst:dregpair}, $addr", "", []>;
142 } // mayLoad = 1, neverHasSideEffects = 1
144 let mayStore = 1, neverHasSideEffects = 1 in {
145 // Use vstmia to store a Q register as a D register pair.
146 // This is equivalent to VSTMD except that it has a Q register operand
147 // instead of a pair of D registers.
149 : AXDI5<(outs), (ins QPR:$src, addrmode5:$addr, pred:$p),
150 IndexModeNone, IIC_fpStorem,
151 "vstm${addr:submode}${p}\t${addr:base}, ${src:dregpair}", "", []>;
153 // Use vst1 to store a Q register as a D register pair.
154 // This alternative to VSTMQ allows an alignment to be specified.
155 // This is equivalent to VST1q64 except that it has a Q register operand.
157 : NLdSt<0,0b00,0b1010,0b1100, (outs), (ins addrmode6:$addr, QPR:$src),
158 IIC_VST, "vst1", "64", "${src:dregpair}, $addr", "", []>;
159 } // mayStore = 1, neverHasSideEffects = 1
161 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
163 // VLD1 : Vector Load (multiple single elements)
164 class VLD1D<bits<4> op7_4, string Dt>
165 : NLdSt<0,0b10,0b0111,op7_4, (outs DPR:$dst),
166 (ins addrmode6:$addr), IIC_VLD1,
167 "vld1", Dt, "\\{$dst\\}, $addr", "", []>;
168 class VLD1Q<bits<4> op7_4, string Dt>
169 : NLdSt<0,0b10,0b1010,op7_4, (outs DPR:$dst1, DPR:$dst2),
170 (ins addrmode6:$addr), IIC_VLD1,
171 "vld1", Dt, "\\{$dst1, $dst2\\}, $addr", "", []>;
173 def VLD1d8 : VLD1D<0b0000, "8">;
174 def VLD1d16 : VLD1D<0b0100, "16">;
175 def VLD1d32 : VLD1D<0b1000, "32">;
176 def VLD1d64 : VLD1D<0b1100, "64">;
178 def VLD1q8 : VLD1Q<0b0000, "8">;
179 def VLD1q16 : VLD1Q<0b0100, "16">;
180 def VLD1q32 : VLD1Q<0b1000, "32">;
181 def VLD1q64 : VLD1Q<0b1100, "64">;
183 // ...with address register writeback:
184 class VLD1DWB<bits<4> op7_4, string Dt>
185 : NLdSt<0,0b10,0b0111,op7_4, (outs DPR:$dst, GPR:$wb),
186 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD1,
187 "vld1", Dt, "\\{$dst\\}, $addr$offset",
188 "$addr.addr = $wb", []>;
189 class VLD1QWB<bits<4> op7_4, string Dt>
190 : NLdSt<0,0b10,0b1010,op7_4, (outs QPR:$dst, GPR:$wb),
191 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD1,
192 "vld1", Dt, "${dst:dregpair}, $addr$offset",
193 "$addr.addr = $wb", []>;
195 def VLD1d8_UPD : VLD1DWB<0b0000, "8">;
196 def VLD1d16_UPD : VLD1DWB<0b0100, "16">;
197 def VLD1d32_UPD : VLD1DWB<0b1000, "32">;
198 def VLD1d64_UPD : VLD1DWB<0b1100, "64">;
200 def VLD1q8_UPD : VLD1QWB<0b0000, "8">;
201 def VLD1q16_UPD : VLD1QWB<0b0100, "16">;
202 def VLD1q32_UPD : VLD1QWB<0b1000, "32">;
203 def VLD1q64_UPD : VLD1QWB<0b1100, "64">;
205 // ...with 3 registers (some of these are only for the disassembler):
206 class VLD1D3<bits<4> op7_4, string Dt>
207 : NLdSt<0,0b10,0b0110,op7_4, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3),
208 (ins addrmode6:$addr), IIC_VLD1, "vld1", Dt,
209 "\\{$dst1, $dst2, $dst3\\}, $addr", "", []>;
210 class VLD1D3WB<bits<4> op7_4, string Dt>
211 : NLdSt<0,0b10,0b0110,op7_4, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, GPR:$wb),
212 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD1, "vld1", Dt,
213 "\\{$dst1, $dst2, $dst3\\}, $addr$offset", "$addr.addr = $wb", []>;
215 def VLD1d8T : VLD1D3<0b0000, "8">;
216 def VLD1d16T : VLD1D3<0b0100, "16">;
217 def VLD1d32T : VLD1D3<0b1000, "32">;
218 def VLD1d64T : VLD1D3<0b1100, "64">;
220 def VLD1d8T_UPD : VLD1D3WB<0b0000, "8">;
221 def VLD1d16T_UPD : VLD1D3WB<0b0100, "16">;
222 def VLD1d32T_UPD : VLD1D3WB<0b1000, "32">;
223 def VLD1d64T_UPD : VLD1D3WB<0b1100, "64">;
225 // ...with 4 registers (some of these are only for the disassembler):
226 class VLD1D4<bits<4> op7_4, string Dt>
227 : NLdSt<0,0b10,0b0010,op7_4,(outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
228 (ins addrmode6:$addr), IIC_VLD1, "vld1", Dt,
229 "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr", "", []>;
230 class VLD1D4WB<bits<4> op7_4, string Dt>
231 : NLdSt<0,0b10,0b0010,op7_4,
232 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
233 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD1, "vld1", Dt,
234 "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr$offset", "$addr.addr = $wb",
237 def VLD1d8Q : VLD1D4<0b0000, "8">;
238 def VLD1d16Q : VLD1D4<0b0100, "16">;
239 def VLD1d32Q : VLD1D4<0b1000, "32">;
240 def VLD1d64Q : VLD1D4<0b1100, "64">;
242 def VLD1d8Q_UPD : VLD1D4WB<0b0000, "8">;
243 def VLD1d16Q_UPD : VLD1D4WB<0b0100, "16">;
244 def VLD1d32Q_UPD : VLD1D4WB<0b1000, "32">;
245 def VLD1d64Q_UPD : VLD1D4WB<0b1100, "64">;
247 // VLD2 : Vector Load (multiple 2-element structures)
248 class VLD2D<bits<4> op11_8, bits<4> op7_4, string Dt>
249 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2),
250 (ins addrmode6:$addr), IIC_VLD2,
251 "vld2", Dt, "\\{$dst1, $dst2\\}, $addr", "", []>;
252 class VLD2Q<bits<4> op7_4, string Dt>
253 : NLdSt<0, 0b10, 0b0011, op7_4,
254 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
255 (ins addrmode6:$addr), IIC_VLD2,
256 "vld2", Dt, "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr", "", []>;
258 def VLD2d8 : VLD2D<0b1000, 0b0000, "8">;
259 def VLD2d16 : VLD2D<0b1000, 0b0100, "16">;
260 def VLD2d32 : VLD2D<0b1000, 0b1000, "32">;
262 def VLD2q8 : VLD2Q<0b0000, "8">;
263 def VLD2q16 : VLD2Q<0b0100, "16">;
264 def VLD2q32 : VLD2Q<0b1000, "32">;
266 // ...with address register writeback:
267 class VLD2DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
268 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2, GPR:$wb),
269 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD2,
270 "vld2", Dt, "\\{$dst1, $dst2\\}, $addr$offset",
271 "$addr.addr = $wb", []>;
272 class VLD2QWB<bits<4> op7_4, string Dt>
273 : NLdSt<0, 0b10, 0b0011, op7_4,
274 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
275 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD2,
276 "vld2", Dt, "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr$offset",
277 "$addr.addr = $wb", []>;
279 def VLD2d8_UPD : VLD2DWB<0b1000, 0b0000, "8">;
280 def VLD2d16_UPD : VLD2DWB<0b1000, 0b0100, "16">;
281 def VLD2d32_UPD : VLD2DWB<0b1000, 0b1000, "32">;
283 def VLD2q8_UPD : VLD2QWB<0b0000, "8">;
284 def VLD2q16_UPD : VLD2QWB<0b0100, "16">;
285 def VLD2q32_UPD : VLD2QWB<0b1000, "32">;
287 // ...with double-spaced registers (for disassembly only):
288 def VLD2b8 : VLD2D<0b1001, 0b0000, "8">;
289 def VLD2b16 : VLD2D<0b1001, 0b0100, "16">;
290 def VLD2b32 : VLD2D<0b1001, 0b1000, "32">;
291 def VLD2b8_UPD : VLD2DWB<0b1001, 0b0000, "8">;
292 def VLD2b16_UPD : VLD2DWB<0b1001, 0b0100, "16">;
293 def VLD2b32_UPD : VLD2DWB<0b1001, 0b1000, "32">;
295 // VLD3 : Vector Load (multiple 3-element structures)
296 class VLD3D<bits<4> op11_8, bits<4> op7_4, string Dt>
297 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3),
298 (ins addrmode6:$addr), IIC_VLD3,
299 "vld3", Dt, "\\{$dst1, $dst2, $dst3\\}, $addr", "", []>;
301 def VLD3d8 : VLD3D<0b0100, 0b0000, "8">;
302 def VLD3d16 : VLD3D<0b0100, 0b0100, "16">;
303 def VLD3d32 : VLD3D<0b0100, 0b1000, "32">;
305 // ...with address register writeback:
306 class VLD3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
307 : NLdSt<0, 0b10, op11_8, op7_4,
308 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, GPR:$wb),
309 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD3,
310 "vld3", Dt, "\\{$dst1, $dst2, $dst3\\}, $addr$offset",
311 "$addr.addr = $wb", []>;
313 def VLD3d8_UPD : VLD3DWB<0b0100, 0b0000, "8">;
314 def VLD3d16_UPD : VLD3DWB<0b0100, 0b0100, "16">;
315 def VLD3d32_UPD : VLD3DWB<0b0100, 0b1000, "32">;
317 // ...with double-spaced registers (non-updating versions for disassembly only):
318 def VLD3q8 : VLD3D<0b0101, 0b0000, "8">;
319 def VLD3q16 : VLD3D<0b0101, 0b0100, "16">;
320 def VLD3q32 : VLD3D<0b0101, 0b1000, "32">;
321 def VLD3q8_UPD : VLD3DWB<0b0101, 0b0000, "8">;
322 def VLD3q16_UPD : VLD3DWB<0b0101, 0b0100, "16">;
323 def VLD3q32_UPD : VLD3DWB<0b0101, 0b1000, "32">;
325 // ...alternate versions to be allocated odd register numbers:
326 def VLD3q8odd_UPD : VLD3DWB<0b0101, 0b0000, "8">;
327 def VLD3q16odd_UPD : VLD3DWB<0b0101, 0b0100, "16">;
328 def VLD3q32odd_UPD : VLD3DWB<0b0101, 0b1000, "32">;
330 // VLD4 : Vector Load (multiple 4-element structures)
331 class VLD4D<bits<4> op11_8, bits<4> op7_4, string Dt>
332 : NLdSt<0, 0b10, op11_8, op7_4,
333 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
334 (ins addrmode6:$addr), IIC_VLD4,
335 "vld4", Dt, "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr", "", []>;
337 def VLD4d8 : VLD4D<0b0000, 0b0000, "8">;
338 def VLD4d16 : VLD4D<0b0000, 0b0100, "16">;
339 def VLD4d32 : VLD4D<0b0000, 0b1000, "32">;
341 // ...with address register writeback:
342 class VLD4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
343 : NLdSt<0, 0b10, op11_8, op7_4,
344 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
345 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD4,
346 "vld4", Dt, "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr$offset",
347 "$addr.addr = $wb", []>;
349 def VLD4d8_UPD : VLD4DWB<0b0000, 0b0000, "8">;
350 def VLD4d16_UPD : VLD4DWB<0b0000, 0b0100, "16">;
351 def VLD4d32_UPD : VLD4DWB<0b0000, 0b1000, "32">;
353 // ...with double-spaced registers (non-updating versions for disassembly only):
354 def VLD4q8 : VLD4D<0b0001, 0b0000, "8">;
355 def VLD4q16 : VLD4D<0b0001, 0b0100, "16">;
356 def VLD4q32 : VLD4D<0b0001, 0b1000, "32">;
357 def VLD4q8_UPD : VLD4DWB<0b0001, 0b0000, "8">;
358 def VLD4q16_UPD : VLD4DWB<0b0001, 0b0100, "16">;
359 def VLD4q32_UPD : VLD4DWB<0b0001, 0b1000, "32">;
361 // ...alternate versions to be allocated odd register numbers:
362 def VLD4q8odd_UPD : VLD4DWB<0b0001, 0b0000, "8">;
363 def VLD4q16odd_UPD : VLD4DWB<0b0001, 0b0100, "16">;
364 def VLD4q32odd_UPD : VLD4DWB<0b0001, 0b1000, "32">;
366 // VLD1LN : Vector Load (single element to one lane)
367 // FIXME: Not yet implemented.
369 // VLD2LN : Vector Load (single 2-element structure to one lane)
370 class VLD2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
371 : NLdSt<1, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2),
372 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, nohash_imm:$lane),
373 IIC_VLD2, "vld2", Dt, "\\{$dst1[$lane], $dst2[$lane]\\}, $addr",
374 "$src1 = $dst1, $src2 = $dst2", []>;
376 def VLD2LNd8 : VLD2LN<0b0001, {?,?,?,?}, "8">;
377 def VLD2LNd16 : VLD2LN<0b0101, {?,?,0,?}, "16">;
378 def VLD2LNd32 : VLD2LN<0b1001, {?,0,?,?}, "32">;
380 // ...with double-spaced registers:
381 def VLD2LNq16 : VLD2LN<0b0101, {?,?,1,?}, "16">;
382 def VLD2LNq32 : VLD2LN<0b1001, {?,1,?,?}, "32">;
384 // ...alternate versions to be allocated odd register numbers:
385 def VLD2LNq16odd : VLD2LN<0b0101, {?,?,1,?}, "16">;
386 def VLD2LNq32odd : VLD2LN<0b1001, {?,1,?,?}, "32">;
388 // ...with address register writeback:
389 class VLD2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
390 : NLdSt<1, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2, GPR:$wb),
391 (ins addrmode6:$addr, am6offset:$offset,
392 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VLD2, "vld2", Dt,
393 "\\{$dst1[$lane], $dst2[$lane]\\}, $addr$offset",
394 "$src1 = $dst1, $src2 = $dst2, $addr.addr = $wb", []>;
396 def VLD2LNd8_UPD : VLD2LNWB<0b0001, {?,?,?,?}, "8">;
397 def VLD2LNd16_UPD : VLD2LNWB<0b0101, {?,?,0,?}, "16">;
398 def VLD2LNd32_UPD : VLD2LNWB<0b1001, {?,0,?,?}, "32">;
400 def VLD2LNq16_UPD : VLD2LNWB<0b0101, {?,?,1,?}, "16">;
401 def VLD2LNq32_UPD : VLD2LNWB<0b1001, {?,1,?,?}, "32">;
403 // VLD3LN : Vector Load (single 3-element structure to one lane)
404 class VLD3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
405 : NLdSt<1, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3),
406 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3,
407 nohash_imm:$lane), IIC_VLD3, "vld3", Dt,
408 "\\{$dst1[$lane], $dst2[$lane], $dst3[$lane]\\}, $addr",
409 "$src1 = $dst1, $src2 = $dst2, $src3 = $dst3", []>;
411 def VLD3LNd8 : VLD3LN<0b0010, {?,?,?,0}, "8">;
412 def VLD3LNd16 : VLD3LN<0b0110, {?,?,0,0}, "16">;
413 def VLD3LNd32 : VLD3LN<0b1010, {?,0,0,0}, "32">;
415 // ...with double-spaced registers:
416 def VLD3LNq16 : VLD3LN<0b0110, {?,?,1,0}, "16">;
417 def VLD3LNq32 : VLD3LN<0b1010, {?,1,0,0}, "32">;
419 // ...alternate versions to be allocated odd register numbers:
420 def VLD3LNq16odd : VLD3LN<0b0110, {?,?,1,0}, "16">;
421 def VLD3LNq32odd : VLD3LN<0b1010, {?,1,0,0}, "32">;
423 // ...with address register writeback:
424 class VLD3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
425 : NLdSt<1, 0b10, op11_8, op7_4,
426 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, GPR:$wb),
427 (ins addrmode6:$addr, am6offset:$offset,
428 DPR:$src1, DPR:$src2, DPR:$src3, nohash_imm:$lane),
429 IIC_VLD3, "vld3", Dt,
430 "\\{$dst1[$lane], $dst2[$lane], $dst3[$lane]\\}, $addr$offset",
431 "$src1 = $dst1, $src2 = $dst2, $src3 = $dst3, $addr.addr = $wb",
434 def VLD3LNd8_UPD : VLD3LNWB<0b0010, {?,?,?,0}, "8">;
435 def VLD3LNd16_UPD : VLD3LNWB<0b0110, {?,?,0,0}, "16">;
436 def VLD3LNd32_UPD : VLD3LNWB<0b1010, {?,0,0,0}, "32">;
438 def VLD3LNq16_UPD : VLD3LNWB<0b0110, {?,?,1,0}, "16">;
439 def VLD3LNq32_UPD : VLD3LNWB<0b1010, {?,1,0,0}, "32">;
441 // VLD4LN : Vector Load (single 4-element structure to one lane)
442 class VLD4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
443 : NLdSt<1, 0b10, op11_8, op7_4,
444 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
445 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
446 nohash_imm:$lane), IIC_VLD4, "vld4", Dt,
447 "\\{$dst1[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $addr",
448 "$src1 = $dst1, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4", []>;
450 def VLD4LNd8 : VLD4LN<0b0011, {?,?,?,?}, "8">;
451 def VLD4LNd16 : VLD4LN<0b0111, {?,?,0,?}, "16">;
452 def VLD4LNd32 : VLD4LN<0b1011, {?,0,?,?}, "32">;
454 // ...with double-spaced registers:
455 def VLD4LNq16 : VLD4LN<0b0111, {?,?,1,?}, "16">;
456 def VLD4LNq32 : VLD4LN<0b1011, {?,1,?,?}, "32">;
458 // ...alternate versions to be allocated odd register numbers:
459 def VLD4LNq16odd : VLD4LN<0b0111, {?,?,1,?}, "16">;
460 def VLD4LNq32odd : VLD4LN<0b1011, {?,1,?,?}, "32">;
462 // ...with address register writeback:
463 class VLD4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
464 : NLdSt<1, 0b10, op11_8, op7_4,
465 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
466 (ins addrmode6:$addr, am6offset:$offset,
467 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
468 IIC_VLD4, "vld4", Dt,
469 "\\{$dst1[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $addr$offset",
470 "$src1 = $dst1, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4, $addr.addr = $wb",
473 def VLD4LNd8_UPD : VLD4LNWB<0b0011, {?,?,?,?}, "8">;
474 def VLD4LNd16_UPD : VLD4LNWB<0b0111, {?,?,0,?}, "16">;
475 def VLD4LNd32_UPD : VLD4LNWB<0b1011, {?,0,?,?}, "32">;
477 def VLD4LNq16_UPD : VLD4LNWB<0b0111, {?,?,1,?}, "16">;
478 def VLD4LNq32_UPD : VLD4LNWB<0b1011, {?,1,?,?}, "32">;
480 // VLD1DUP : Vector Load (single element to all lanes)
481 // VLD2DUP : Vector Load (single 2-element structure to all lanes)
482 // VLD3DUP : Vector Load (single 3-element structure to all lanes)
483 // VLD4DUP : Vector Load (single 4-element structure to all lanes)
484 // FIXME: Not yet implemented.
485 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
487 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
489 // Classes for VST* pseudo-instructions with multi-register operands.
490 // These are expanded to real instructions after register allocation.
492 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src), IIC_VST, "">;
494 : PseudoNLdSt<(outs GPR:$wb),
495 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src), IIC_VST,
497 class VSTQQQQWBPseudo
498 : PseudoNLdSt<(outs GPR:$wb),
499 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), IIC_VST,
502 // VST1 : Vector Store (multiple single elements)
503 class VST1D<bits<4> op7_4, string Dt>
504 : NLdSt<0,0b00,0b0111,op7_4, (outs), (ins addrmode6:$addr, DPR:$src), IIC_VST,
505 "vst1", Dt, "\\{$src\\}, $addr", "", []>;
506 class VST1Q<bits<4> op7_4, string Dt>
507 : NLdSt<0,0b00,0b1010,op7_4, (outs),
508 (ins addrmode6:$addr, DPR:$src1, DPR:$src2), IIC_VST,
509 "vst1", Dt, "\\{$src1, $src2\\}, $addr", "", []>;
511 def VST1d8 : VST1D<0b0000, "8">;
512 def VST1d16 : VST1D<0b0100, "16">;
513 def VST1d32 : VST1D<0b1000, "32">;
514 def VST1d64 : VST1D<0b1100, "64">;
516 def VST1q8 : VST1Q<0b0000, "8">;
517 def VST1q16 : VST1Q<0b0100, "16">;
518 def VST1q32 : VST1Q<0b1000, "32">;
519 def VST1q64 : VST1Q<0b1100, "64">;
521 // ...with address register writeback:
522 class VST1DWB<bits<4> op7_4, string Dt>
523 : NLdSt<0, 0b00, 0b0111, op7_4, (outs GPR:$wb),
524 (ins addrmode6:$addr, am6offset:$offset, DPR:$src), IIC_VST,
525 "vst1", Dt, "\\{$src\\}, $addr$offset", "$addr.addr = $wb", []>;
526 class VST1QWB<bits<4> op7_4, string Dt>
527 : NLdSt<0, 0b00, 0b1010, op7_4, (outs GPR:$wb),
528 (ins addrmode6:$addr, am6offset:$offset, QPR:$src), IIC_VST,
529 "vst1", Dt, "${src:dregpair}, $addr$offset", "$addr.addr = $wb", []>;
531 def VST1d8_UPD : VST1DWB<0b0000, "8">;
532 def VST1d16_UPD : VST1DWB<0b0100, "16">;
533 def VST1d32_UPD : VST1DWB<0b1000, "32">;
534 def VST1d64_UPD : VST1DWB<0b1100, "64">;
536 def VST1q8_UPD : VST1QWB<0b0000, "8">;
537 def VST1q16_UPD : VST1QWB<0b0100, "16">;
538 def VST1q32_UPD : VST1QWB<0b1000, "32">;
539 def VST1q64_UPD : VST1QWB<0b1100, "64">;
541 // ...with 3 registers (some of these are only for the disassembler):
542 class VST1D3<bits<4> op7_4, string Dt>
543 : NLdSt<0, 0b00, 0b0110, op7_4, (outs),
544 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3),
545 IIC_VST, "vst1", Dt, "\\{$src1, $src2, $src3\\}, $addr", "", []>;
546 class VST1D3WB<bits<4> op7_4, string Dt>
547 : NLdSt<0, 0b00, 0b0110, op7_4, (outs GPR:$wb),
548 (ins addrmode6:$addr, am6offset:$offset,
549 DPR:$src1, DPR:$src2, DPR:$src3),
550 IIC_VST, "vst1", Dt, "\\{$src1, $src2, $src3\\}, $addr$offset",
551 "$addr.addr = $wb", []>;
553 def VST1d8T : VST1D3<0b0000, "8">;
554 def VST1d16T : VST1D3<0b0100, "16">;
555 def VST1d32T : VST1D3<0b1000, "32">;
556 def VST1d64T : VST1D3<0b1100, "64">;
558 def VST1d8T_UPD : VST1D3WB<0b0000, "8">;
559 def VST1d16T_UPD : VST1D3WB<0b0100, "16">;
560 def VST1d32T_UPD : VST1D3WB<0b1000, "32">;
561 def VST1d64T_UPD : VST1D3WB<0b1100, "64">;
563 // ...with 4 registers (some of these are only for the disassembler):
564 class VST1D4<bits<4> op7_4, string Dt>
565 : NLdSt<0, 0b00, 0b0010, op7_4, (outs),
566 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
567 IIC_VST, "vst1", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr", "",
569 class VST1D4WB<bits<4> op7_4, string Dt>
570 : NLdSt<0, 0b00, 0b0010, op7_4, (outs GPR:$wb),
571 (ins addrmode6:$addr, am6offset:$offset,
572 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
573 IIC_VST, "vst1", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr$offset",
574 "$addr.addr = $wb", []>;
576 def VST1d8Q : VST1D4<0b0000, "8">;
577 def VST1d16Q : VST1D4<0b0100, "16">;
578 def VST1d32Q : VST1D4<0b1000, "32">;
579 def VST1d64Q : VST1D4<0b1100, "64">;
581 def VST1d8Q_UPD : VST1D4WB<0b0000, "8">;
582 def VST1d16Q_UPD : VST1D4WB<0b0100, "16">;
583 def VST1d32Q_UPD : VST1D4WB<0b1000, "32">;
584 def VST1d64Q_UPD : VST1D4WB<0b1100, "64">;
586 def VST1d64QPseudo : VSTQQPseudo;
587 def VST1d64QPseudo_UPD : VSTQQWBPseudo;
589 // VST2 : Vector Store (multiple 2-element structures)
590 class VST2D<bits<4> op11_8, bits<4> op7_4, string Dt>
591 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
592 (ins addrmode6:$addr, DPR:$src1, DPR:$src2),
593 IIC_VST, "vst2", Dt, "\\{$src1, $src2\\}, $addr", "", []>;
594 class VST2Q<bits<4> op7_4, string Dt>
595 : NLdSt<0, 0b00, 0b0011, op7_4, (outs),
596 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
597 IIC_VST, "vst2", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr",
600 def VST2d8 : VST2D<0b1000, 0b0000, "8">;
601 def VST2d16 : VST2D<0b1000, 0b0100, "16">;
602 def VST2d32 : VST2D<0b1000, 0b1000, "32">;
604 def VST2q8 : VST2Q<0b0000, "8">;
605 def VST2q16 : VST2Q<0b0100, "16">;
606 def VST2q32 : VST2Q<0b1000, "32">;
608 // ...with address register writeback:
609 class VST2DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
610 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
611 (ins addrmode6:$addr, am6offset:$offset, DPR:$src1, DPR:$src2),
612 IIC_VST, "vst2", Dt, "\\{$src1, $src2\\}, $addr$offset",
613 "$addr.addr = $wb", []>;
614 class VST2QWB<bits<4> op7_4, string Dt>
615 : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
616 (ins addrmode6:$addr, am6offset:$offset,
617 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
618 IIC_VST, "vst2", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr$offset",
619 "$addr.addr = $wb", []>;
621 def VST2d8_UPD : VST2DWB<0b1000, 0b0000, "8">;
622 def VST2d16_UPD : VST2DWB<0b1000, 0b0100, "16">;
623 def VST2d32_UPD : VST2DWB<0b1000, 0b1000, "32">;
625 def VST2q8_UPD : VST2QWB<0b0000, "8">;
626 def VST2q16_UPD : VST2QWB<0b0100, "16">;
627 def VST2q32_UPD : VST2QWB<0b1000, "32">;
629 // ...with double-spaced registers (for disassembly only):
630 def VST2b8 : VST2D<0b1001, 0b0000, "8">;
631 def VST2b16 : VST2D<0b1001, 0b0100, "16">;
632 def VST2b32 : VST2D<0b1001, 0b1000, "32">;
633 def VST2b8_UPD : VST2DWB<0b1001, 0b0000, "8">;
634 def VST2b16_UPD : VST2DWB<0b1001, 0b0100, "16">;
635 def VST2b32_UPD : VST2DWB<0b1001, 0b1000, "32">;
637 // VST3 : Vector Store (multiple 3-element structures)
638 class VST3D<bits<4> op11_8, bits<4> op7_4, string Dt>
639 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
640 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3), IIC_VST,
641 "vst3", Dt, "\\{$src1, $src2, $src3\\}, $addr", "", []>;
643 def VST3d8 : VST3D<0b0100, 0b0000, "8">;
644 def VST3d16 : VST3D<0b0100, 0b0100, "16">;
645 def VST3d32 : VST3D<0b0100, 0b1000, "32">;
647 // ...with address register writeback:
648 class VST3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
649 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
650 (ins addrmode6:$addr, am6offset:$offset,
651 DPR:$src1, DPR:$src2, DPR:$src3), IIC_VST,
652 "vst3", Dt, "\\{$src1, $src2, $src3\\}, $addr$offset",
653 "$addr.addr = $wb", []>;
655 def VST3d8_UPD : VST3DWB<0b0100, 0b0000, "8">;
656 def VST3d16_UPD : VST3DWB<0b0100, 0b0100, "16">;
657 def VST3d32_UPD : VST3DWB<0b0100, 0b1000, "32">;
659 // ...with double-spaced registers (non-updating versions for disassembly only):
660 def VST3q8 : VST3D<0b0101, 0b0000, "8">;
661 def VST3q16 : VST3D<0b0101, 0b0100, "16">;
662 def VST3q32 : VST3D<0b0101, 0b1000, "32">;
663 def VST3q8_UPD : VST3DWB<0b0101, 0b0000, "8">;
664 def VST3q16_UPD : VST3DWB<0b0101, 0b0100, "16">;
665 def VST3q32_UPD : VST3DWB<0b0101, 0b1000, "32">;
667 // ...alternate versions to be allocated odd register numbers:
668 def VST3q8odd_UPD : VST3DWB<0b0101, 0b0000, "8">;
669 def VST3q16odd_UPD : VST3DWB<0b0101, 0b0100, "16">;
670 def VST3q32odd_UPD : VST3DWB<0b0101, 0b1000, "32">;
672 // VST4 : Vector Store (multiple 4-element structures)
673 class VST4D<bits<4> op11_8, bits<4> op7_4, string Dt>
674 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
675 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
676 IIC_VST, "vst4", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr",
679 def VST4d8 : VST4D<0b0000, 0b0000, "8">;
680 def VST4d16 : VST4D<0b0000, 0b0100, "16">;
681 def VST4d32 : VST4D<0b0000, 0b1000, "32">;
683 def VST4d8Pseudo : VSTQQPseudo;
684 def VST4d16Pseudo : VSTQQPseudo;
685 def VST4d32Pseudo : VSTQQPseudo;
687 // ...with address register writeback:
688 class VST4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
689 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
690 (ins addrmode6:$addr, am6offset:$offset,
691 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST,
692 "vst4", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr$offset",
693 "$addr.addr = $wb", []>;
695 def VST4d8_UPD : VST4DWB<0b0000, 0b0000, "8">;
696 def VST4d16_UPD : VST4DWB<0b0000, 0b0100, "16">;
697 def VST4d32_UPD : VST4DWB<0b0000, 0b1000, "32">;
699 def VST4d8Pseudo_UPD : VSTQQWBPseudo;
700 def VST4d16Pseudo_UPD : VSTQQWBPseudo;
701 def VST4d32Pseudo_UPD : VSTQQWBPseudo;
703 // ...with double-spaced registers (non-updating versions for disassembly only):
704 def VST4q8 : VST4D<0b0001, 0b0000, "8">;
705 def VST4q16 : VST4D<0b0001, 0b0100, "16">;
706 def VST4q32 : VST4D<0b0001, 0b1000, "32">;
707 def VST4q8_UPD : VST4DWB<0b0001, 0b0000, "8">;
708 def VST4q16_UPD : VST4DWB<0b0001, 0b0100, "16">;
709 def VST4q32_UPD : VST4DWB<0b0001, 0b1000, "32">;
711 def VST4q8Pseudo_UPD : VSTQQQQWBPseudo;
712 def VST4q16Pseudo_UPD : VSTQQQQWBPseudo;
713 def VST4q32Pseudo_UPD : VSTQQQQWBPseudo;
715 // ...alternate versions to be allocated odd register numbers:
716 def VST4q8oddPseudo_UPD : VSTQQQQWBPseudo;
717 def VST4q16oddPseudo_UPD : VSTQQQQWBPseudo;
718 def VST4q32oddPseudo_UPD : VSTQQQQWBPseudo;
720 // VST1LN : Vector Store (single element from one lane)
721 // FIXME: Not yet implemented.
723 // VST2LN : Vector Store (single 2-element structure from one lane)
724 class VST2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
725 : NLdSt<1, 0b00, op11_8, op7_4, (outs),
726 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, nohash_imm:$lane),
727 IIC_VST, "vst2", Dt, "\\{$src1[$lane], $src2[$lane]\\}, $addr",
730 def VST2LNd8 : VST2LN<0b0001, {?,?,?,?}, "8">;
731 def VST2LNd16 : VST2LN<0b0101, {?,?,0,?}, "16">;
732 def VST2LNd32 : VST2LN<0b1001, {?,0,?,?}, "32">;
734 // ...with double-spaced registers:
735 def VST2LNq16 : VST2LN<0b0101, {?,?,1,?}, "16">;
736 def VST2LNq32 : VST2LN<0b1001, {?,1,?,?}, "32">;
738 // ...alternate versions to be allocated odd register numbers:
739 def VST2LNq16odd : VST2LN<0b0101, {?,?,1,?}, "16">;
740 def VST2LNq32odd : VST2LN<0b1001, {?,1,?,?}, "32">;
742 // ...with address register writeback:
743 class VST2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
744 : NLdSt<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
745 (ins addrmode6:$addr, am6offset:$offset,
746 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VST, "vst2", Dt,
747 "\\{$src1[$lane], $src2[$lane]\\}, $addr$offset",
748 "$addr.addr = $wb", []>;
750 def VST2LNd8_UPD : VST2LNWB<0b0001, {?,?,?,?}, "8">;
751 def VST2LNd16_UPD : VST2LNWB<0b0101, {?,?,0,?}, "16">;
752 def VST2LNd32_UPD : VST2LNWB<0b1001, {?,0,?,?}, "32">;
754 def VST2LNq16_UPD : VST2LNWB<0b0101, {?,?,1,?}, "16">;
755 def VST2LNq32_UPD : VST2LNWB<0b1001, {?,1,?,?}, "32">;
757 // VST3LN : Vector Store (single 3-element structure from one lane)
758 class VST3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
759 : NLdSt<1, 0b00, op11_8, op7_4, (outs),
760 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3,
761 nohash_imm:$lane), IIC_VST, "vst3", Dt,
762 "\\{$src1[$lane], $src2[$lane], $src3[$lane]\\}, $addr", "", []>;
764 def VST3LNd8 : VST3LN<0b0010, {?,?,?,0}, "8">;
765 def VST3LNd16 : VST3LN<0b0110, {?,?,0,0}, "16">;
766 def VST3LNd32 : VST3LN<0b1010, {?,0,0,0}, "32">;
768 // ...with double-spaced registers:
769 def VST3LNq16 : VST3LN<0b0110, {?,?,1,0}, "16">;
770 def VST3LNq32 : VST3LN<0b1010, {?,1,0,0}, "32">;
772 // ...alternate versions to be allocated odd register numbers:
773 def VST3LNq16odd : VST3LN<0b0110, {?,?,1,0}, "16">;
774 def VST3LNq32odd : VST3LN<0b1010, {?,1,0,0}, "32">;
776 // ...with address register writeback:
777 class VST3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
778 : NLdSt<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
779 (ins addrmode6:$addr, am6offset:$offset,
780 DPR:$src1, DPR:$src2, DPR:$src3, nohash_imm:$lane),
782 "\\{$src1[$lane], $src2[$lane], $src3[$lane]\\}, $addr$offset",
783 "$addr.addr = $wb", []>;
785 def VST3LNd8_UPD : VST3LNWB<0b0010, {?,?,?,0}, "8">;
786 def VST3LNd16_UPD : VST3LNWB<0b0110, {?,?,0,0}, "16">;
787 def VST3LNd32_UPD : VST3LNWB<0b1010, {?,0,0,0}, "32">;
789 def VST3LNq16_UPD : VST3LNWB<0b0110, {?,?,1,0}, "16">;
790 def VST3LNq32_UPD : VST3LNWB<0b1010, {?,1,0,0}, "32">;
792 // VST4LN : Vector Store (single 4-element structure from one lane)
793 class VST4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
794 : NLdSt<1, 0b00, op11_8, op7_4, (outs),
795 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
796 nohash_imm:$lane), IIC_VST, "vst4", Dt,
797 "\\{$src1[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $addr",
800 def VST4LNd8 : VST4LN<0b0011, {?,?,?,?}, "8">;
801 def VST4LNd16 : VST4LN<0b0111, {?,?,0,?}, "16">;
802 def VST4LNd32 : VST4LN<0b1011, {?,0,?,?}, "32">;
804 // ...with double-spaced registers:
805 def VST4LNq16 : VST4LN<0b0111, {?,?,1,?}, "16">;
806 def VST4LNq32 : VST4LN<0b1011, {?,1,?,?}, "32">;
808 // ...alternate versions to be allocated odd register numbers:
809 def VST4LNq16odd : VST4LN<0b0111, {?,?,1,?}, "16">;
810 def VST4LNq32odd : VST4LN<0b1011, {?,1,?,?}, "32">;
812 // ...with address register writeback:
813 class VST4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
814 : NLdSt<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
815 (ins addrmode6:$addr, am6offset:$offset,
816 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
818 "\\{$src1[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $addr$offset",
819 "$addr.addr = $wb", []>;
821 def VST4LNd8_UPD : VST4LNWB<0b0011, {?,?,?,?}, "8">;
822 def VST4LNd16_UPD : VST4LNWB<0b0111, {?,?,0,?}, "16">;
823 def VST4LNd32_UPD : VST4LNWB<0b1011, {?,0,?,?}, "32">;
825 def VST4LNq16_UPD : VST4LNWB<0b0111, {?,?,1,?}, "16">;
826 def VST4LNq32_UPD : VST4LNWB<0b1011, {?,1,?,?}, "32">;
828 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
831 //===----------------------------------------------------------------------===//
832 // NEON pattern fragments
833 //===----------------------------------------------------------------------===//
835 // Extract D sub-registers of Q registers.
836 def DSubReg_i8_reg : SDNodeXForm<imm, [{
837 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
838 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/8, MVT::i32);
840 def DSubReg_i16_reg : SDNodeXForm<imm, [{
841 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
842 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/4, MVT::i32);
844 def DSubReg_i32_reg : SDNodeXForm<imm, [{
845 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
846 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/2, MVT::i32);
848 def DSubReg_f64_reg : SDNodeXForm<imm, [{
849 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
850 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue(), MVT::i32);
853 // Extract S sub-registers of Q/D registers.
854 def SSubReg_f32_reg : SDNodeXForm<imm, [{
855 assert(ARM::ssub_3 == ARM::ssub_0+3 && "Unexpected subreg numbering");
856 return CurDAG->getTargetConstant(ARM::ssub_0 + N->getZExtValue(), MVT::i32);
859 // Translate lane numbers from Q registers to D subregs.
860 def SubReg_i8_lane : SDNodeXForm<imm, [{
861 return CurDAG->getTargetConstant(N->getZExtValue() & 7, MVT::i32);
863 def SubReg_i16_lane : SDNodeXForm<imm, [{
864 return CurDAG->getTargetConstant(N->getZExtValue() & 3, MVT::i32);
866 def SubReg_i32_lane : SDNodeXForm<imm, [{
867 return CurDAG->getTargetConstant(N->getZExtValue() & 1, MVT::i32);
870 //===----------------------------------------------------------------------===//
871 // Instruction Classes
872 //===----------------------------------------------------------------------===//
874 // Basic 2-register operations: single-, double- and quad-register.
875 class N2VS<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
876 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
877 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
878 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
879 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src),
880 IIC_VUNAD, OpcodeStr, Dt, "$dst, $src", "", []>;
881 class N2VD<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
882 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
883 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
884 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
885 (ins DPR:$src), IIC_VUNAD, OpcodeStr, Dt,"$dst, $src", "",
886 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src))))]>;
887 class N2VQ<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
888 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
889 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
890 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
891 (ins QPR:$src), IIC_VUNAQ, OpcodeStr, Dt,"$dst, $src", "",
892 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src))))]>;
894 // Basic 2-register intrinsics, both double- and quad-register.
895 class N2VDInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
896 bits<2> op17_16, bits<5> op11_7, bit op4,
897 InstrItinClass itin, string OpcodeStr, string Dt,
898 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
899 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
900 (ins DPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
901 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
902 class N2VQInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
903 bits<2> op17_16, bits<5> op11_7, bit op4,
904 InstrItinClass itin, string OpcodeStr, string Dt,
905 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
906 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
907 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
908 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
910 // Narrow 2-register intrinsics.
911 class N2VNInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
912 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
913 InstrItinClass itin, string OpcodeStr, string Dt,
914 ValueType TyD, ValueType TyQ, Intrinsic IntOp>
915 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$dst),
916 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
917 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src))))]>;
919 // Long 2-register operations (currently only used for VMOVL).
920 class N2VL<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
921 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
922 InstrItinClass itin, string OpcodeStr, string Dt,
923 ValueType TyQ, ValueType TyD, SDNode OpNode>
924 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$dst),
925 (ins DPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
926 [(set QPR:$dst, (TyQ (OpNode (TyD DPR:$src))))]>;
928 // 2-register shuffles (VTRN/VZIP/VUZP), both double- and quad-register.
929 class N2VDShuffle<bits<2> op19_18, bits<5> op11_7, string OpcodeStr, string Dt>
930 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 0, 0, (outs DPR:$dst1, DPR:$dst2),
931 (ins DPR:$src1, DPR:$src2), IIC_VPERMD,
932 OpcodeStr, Dt, "$dst1, $dst2",
933 "$src1 = $dst1, $src2 = $dst2", []>;
934 class N2VQShuffle<bits<2> op19_18, bits<5> op11_7,
935 InstrItinClass itin, string OpcodeStr, string Dt>
936 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 1, 0, (outs QPR:$dst1, QPR:$dst2),
937 (ins QPR:$src1, QPR:$src2), itin, OpcodeStr, Dt, "$dst1, $dst2",
938 "$src1 = $dst1, $src2 = $dst2", []>;
940 // Basic 3-register operations: single-, double- and quad-register.
941 class N3VS<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
942 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
943 SDNode OpNode, bit Commutable>
944 : N3V<op24, op23, op21_20, op11_8, 0, op4,
945 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src1, DPR_VFP2:$src2), N3RegFrm,
946 IIC_VBIND, OpcodeStr, Dt, "$dst, $src1, $src2", "", []> {
947 let isCommutable = Commutable;
950 class N3VD<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
951 InstrItinClass itin, string OpcodeStr, string Dt,
952 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
953 : N3V<op24, op23, op21_20, op11_8, 0, op4,
954 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
955 OpcodeStr, Dt, "$dst, $src1, $src2", "",
956 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src1), (OpTy DPR:$src2))))]> {
957 let isCommutable = Commutable;
959 // Same as N3VD but no data type.
960 class N3VDX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
961 InstrItinClass itin, string OpcodeStr,
962 ValueType ResTy, ValueType OpTy,
963 SDNode OpNode, bit Commutable>
964 : N3VX<op24, op23, op21_20, op11_8, 0, op4,
965 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
966 OpcodeStr, "$dst, $src1, $src2", "",
967 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src1), (OpTy DPR:$src2))))]>{
968 let isCommutable = Commutable;
971 class N3VDSL<bits<2> op21_20, bits<4> op11_8,
972 InstrItinClass itin, string OpcodeStr, string Dt,
973 ValueType Ty, SDNode ShOp>
974 : N3V<0, 1, op21_20, op11_8, 1, 0,
975 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
976 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
978 (Ty (ShOp (Ty DPR:$src1),
979 (Ty (NEONvduplane (Ty DPR_VFP2:$src2),imm:$lane)))))]> {
980 let isCommutable = 0;
982 class N3VDSL16<bits<2> op21_20, bits<4> op11_8,
983 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
984 : N3V<0, 1, op21_20, op11_8, 1, 0,
985 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
986 NVMulSLFrm, IIC_VMULi16D, OpcodeStr, Dt,"$dst, $src1, $src2[$lane]","",
988 (Ty (ShOp (Ty DPR:$src1),
989 (Ty (NEONvduplane (Ty DPR_8:$src2), imm:$lane)))))]> {
990 let isCommutable = 0;
993 class N3VQ<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
994 InstrItinClass itin, string OpcodeStr, string Dt,
995 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
996 : N3V<op24, op23, op21_20, op11_8, 1, op4,
997 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), N3RegFrm, itin,
998 OpcodeStr, Dt, "$dst, $src1, $src2", "",
999 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src1), (OpTy QPR:$src2))))]> {
1000 let isCommutable = Commutable;
1002 class N3VQX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1003 InstrItinClass itin, string OpcodeStr,
1004 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
1005 : N3VX<op24, op23, op21_20, op11_8, 1, op4,
1006 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), N3RegFrm, itin,
1007 OpcodeStr, "$dst, $src1, $src2", "",
1008 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src1), (OpTy QPR:$src2))))]>{
1009 let isCommutable = Commutable;
1011 class N3VQSL<bits<2> op21_20, bits<4> op11_8,
1012 InstrItinClass itin, string OpcodeStr, string Dt,
1013 ValueType ResTy, ValueType OpTy, SDNode ShOp>
1014 : N3V<1, 1, op21_20, op11_8, 1, 0,
1015 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1016 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1017 [(set (ResTy QPR:$dst),
1018 (ResTy (ShOp (ResTy QPR:$src1),
1019 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1021 let isCommutable = 0;
1023 class N3VQSL16<bits<2> op21_20, bits<4> op11_8, string OpcodeStr, string Dt,
1024 ValueType ResTy, ValueType OpTy, SDNode ShOp>
1025 : N3V<1, 1, op21_20, op11_8, 1, 0,
1026 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1027 NVMulSLFrm, IIC_VMULi16Q, OpcodeStr, Dt,"$dst, $src1, $src2[$lane]","",
1028 [(set (ResTy QPR:$dst),
1029 (ResTy (ShOp (ResTy QPR:$src1),
1030 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
1032 let isCommutable = 0;
1035 // Basic 3-register intrinsics, both double- and quad-register.
1036 class N3VDInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1037 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1038 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
1039 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1040 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), f, itin,
1041 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1042 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src1), (OpTy DPR:$src2))))]> {
1043 let isCommutable = Commutable;
1045 class N3VDIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1046 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
1047 : N3V<0, 1, op21_20, op11_8, 1, 0,
1048 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1049 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1050 [(set (Ty DPR:$dst),
1051 (Ty (IntOp (Ty DPR:$src1),
1052 (Ty (NEONvduplane (Ty DPR_VFP2:$src2),
1054 let isCommutable = 0;
1056 class N3VDIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1057 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
1058 : N3V<0, 1, op21_20, op11_8, 1, 0,
1059 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1060 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1061 [(set (Ty DPR:$dst),
1062 (Ty (IntOp (Ty DPR:$src1),
1063 (Ty (NEONvduplane (Ty DPR_8:$src2), imm:$lane)))))]> {
1064 let isCommutable = 0;
1067 class N3VQInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1068 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1069 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
1070 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1071 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), f, itin,
1072 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1073 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src1), (OpTy QPR:$src2))))]> {
1074 let isCommutable = Commutable;
1076 class N3VQIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1077 string OpcodeStr, string Dt,
1078 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1079 : N3V<1, 1, op21_20, op11_8, 1, 0,
1080 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1081 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1082 [(set (ResTy QPR:$dst),
1083 (ResTy (IntOp (ResTy QPR:$src1),
1084 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1086 let isCommutable = 0;
1088 class N3VQIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1089 string OpcodeStr, string Dt,
1090 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1091 : N3V<1, 1, op21_20, op11_8, 1, 0,
1092 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1093 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1094 [(set (ResTy QPR:$dst),
1095 (ResTy (IntOp (ResTy QPR:$src1),
1096 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
1098 let isCommutable = 0;
1101 // Multiply-Add/Sub operations: single-, double- and quad-register.
1102 class N3VSMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1103 InstrItinClass itin, string OpcodeStr, string Dt,
1104 ValueType Ty, SDNode MulOp, SDNode OpNode>
1105 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1106 (outs DPR_VFP2:$dst),
1107 (ins DPR_VFP2:$src1, DPR_VFP2:$src2, DPR_VFP2:$src3), N3RegFrm, itin,
1108 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst", []>;
1110 class N3VDMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1111 InstrItinClass itin, string OpcodeStr, string Dt,
1112 ValueType Ty, SDNode MulOp, SDNode OpNode>
1113 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1114 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3), N3RegFrm, itin,
1115 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1116 [(set DPR:$dst, (Ty (OpNode DPR:$src1,
1117 (Ty (MulOp DPR:$src2, DPR:$src3)))))]>;
1118 class N3VDMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1119 string OpcodeStr, string Dt,
1120 ValueType Ty, SDNode MulOp, SDNode ShOp>
1121 : N3V<0, 1, op21_20, op11_8, 1, 0,
1123 (ins DPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1125 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1126 [(set (Ty DPR:$dst),
1127 (Ty (ShOp (Ty DPR:$src1),
1128 (Ty (MulOp DPR:$src2,
1129 (Ty (NEONvduplane (Ty DPR_VFP2:$src3),
1131 class N3VDMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1132 string OpcodeStr, string Dt,
1133 ValueType Ty, SDNode MulOp, SDNode ShOp>
1134 : N3V<0, 1, op21_20, op11_8, 1, 0,
1136 (ins DPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane),
1138 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1139 [(set (Ty DPR:$dst),
1140 (Ty (ShOp (Ty DPR:$src1),
1141 (Ty (MulOp DPR:$src2,
1142 (Ty (NEONvduplane (Ty DPR_8:$src3),
1145 class N3VQMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1146 InstrItinClass itin, string OpcodeStr, string Dt, ValueType Ty,
1147 SDNode MulOp, SDNode OpNode>
1148 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1149 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3), N3RegFrm, itin,
1150 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1151 [(set QPR:$dst, (Ty (OpNode QPR:$src1,
1152 (Ty (MulOp QPR:$src2, QPR:$src3)))))]>;
1153 class N3VQMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1154 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1155 SDNode MulOp, SDNode ShOp>
1156 : N3V<1, 1, op21_20, op11_8, 1, 0,
1158 (ins QPR:$src1, QPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1160 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1161 [(set (ResTy QPR:$dst),
1162 (ResTy (ShOp (ResTy QPR:$src1),
1163 (ResTy (MulOp QPR:$src2,
1164 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src3),
1166 class N3VQMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1167 string OpcodeStr, string Dt,
1168 ValueType ResTy, ValueType OpTy,
1169 SDNode MulOp, SDNode ShOp>
1170 : N3V<1, 1, op21_20, op11_8, 1, 0,
1172 (ins QPR:$src1, QPR:$src2, DPR_8:$src3, nohash_imm:$lane),
1174 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1175 [(set (ResTy QPR:$dst),
1176 (ResTy (ShOp (ResTy QPR:$src1),
1177 (ResTy (MulOp QPR:$src2,
1178 (ResTy (NEONvduplane (OpTy DPR_8:$src3),
1181 // Neon 3-argument intrinsics, both double- and quad-register.
1182 // The destination register is also used as the first source operand register.
1183 class N3VDInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1184 InstrItinClass itin, string OpcodeStr, string Dt,
1185 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1186 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1187 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3), N3RegFrm, itin,
1188 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1189 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src1),
1190 (OpTy DPR:$src2), (OpTy DPR:$src3))))]>;
1191 class N3VQInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1192 InstrItinClass itin, string OpcodeStr, string Dt,
1193 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1194 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1195 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3), N3RegFrm, itin,
1196 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1197 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src1),
1198 (OpTy QPR:$src2), (OpTy QPR:$src3))))]>;
1200 // Neon Long 3-argument intrinsic. The destination register is
1201 // a quad-register and is also used as the first source operand register.
1202 class N3VLInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1203 InstrItinClass itin, string OpcodeStr, string Dt,
1204 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
1205 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1206 (outs QPR:$dst), (ins QPR:$src1, DPR:$src2, DPR:$src3), N3RegFrm, itin,
1207 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1209 (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$src2), (TyD DPR:$src3))))]>;
1210 class N3VLInt3SL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1211 string OpcodeStr, string Dt,
1212 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1213 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1215 (ins QPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1217 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1218 [(set (ResTy QPR:$dst),
1219 (ResTy (IntOp (ResTy QPR:$src1),
1221 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src3),
1223 class N3VLInt3SL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1224 InstrItinClass itin, string OpcodeStr, string Dt,
1225 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1226 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1228 (ins QPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane),
1230 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1231 [(set (ResTy QPR:$dst),
1232 (ResTy (IntOp (ResTy QPR:$src1),
1234 (OpTy (NEONvduplane (OpTy DPR_8:$src3),
1237 // Narrowing 3-register intrinsics.
1238 class N3VNInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1239 string OpcodeStr, string Dt, ValueType TyD, ValueType TyQ,
1240 Intrinsic IntOp, bit Commutable>
1241 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1242 (outs DPR:$dst), (ins QPR:$src1, QPR:$src2), N3RegFrm, IIC_VBINi4D,
1243 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1244 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src1), (TyQ QPR:$src2))))]> {
1245 let isCommutable = Commutable;
1248 // Long 3-register intrinsics.
1249 class N3VLInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1250 InstrItinClass itin, string OpcodeStr, string Dt,
1251 ValueType TyQ, ValueType TyD, Intrinsic IntOp, bit Commutable>
1252 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1253 (outs QPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
1254 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1255 [(set QPR:$dst, (TyQ (IntOp (TyD DPR:$src1), (TyD DPR:$src2))))]> {
1256 let isCommutable = Commutable;
1258 class N3VLIntSL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1259 string OpcodeStr, string Dt,
1260 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1261 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1262 (outs QPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1263 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1264 [(set (ResTy QPR:$dst),
1265 (ResTy (IntOp (OpTy DPR:$src1),
1266 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1268 class N3VLIntSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1269 InstrItinClass itin, string OpcodeStr, string Dt,
1270 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1271 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1272 (outs QPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1273 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1274 [(set (ResTy QPR:$dst),
1275 (ResTy (IntOp (OpTy DPR:$src1),
1276 (OpTy (NEONvduplane (OpTy DPR_8:$src2),
1279 // Wide 3-register intrinsics.
1280 class N3VWInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1281 string OpcodeStr, string Dt, ValueType TyQ, ValueType TyD,
1282 Intrinsic IntOp, bit Commutable>
1283 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1284 (outs QPR:$dst), (ins QPR:$src1, DPR:$src2), N3RegFrm, IIC_VSUBiD,
1285 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1286 [(set QPR:$dst, (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$src2))))]> {
1287 let isCommutable = Commutable;
1290 // Pairwise long 2-register intrinsics, both double- and quad-register.
1291 class N2VDPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1292 bits<2> op17_16, bits<5> op11_7, bit op4,
1293 string OpcodeStr, string Dt,
1294 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1295 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
1296 (ins DPR:$src), IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
1297 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
1298 class N2VQPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1299 bits<2> op17_16, bits<5> op11_7, bit op4,
1300 string OpcodeStr, string Dt,
1301 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1302 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
1303 (ins QPR:$src), IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
1304 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
1306 // Pairwise long 2-register accumulate intrinsics,
1307 // both double- and quad-register.
1308 // The destination register is also used as the first source operand register.
1309 class N2VDPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1310 bits<2> op17_16, bits<5> op11_7, bit op4,
1311 string OpcodeStr, string Dt,
1312 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1313 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
1314 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), IIC_VPALiD,
1315 OpcodeStr, Dt, "$dst, $src2", "$src1 = $dst",
1316 [(set DPR:$dst, (ResTy (IntOp (ResTy DPR:$src1), (OpTy DPR:$src2))))]>;
1317 class N2VQPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1318 bits<2> op17_16, bits<5> op11_7, bit op4,
1319 string OpcodeStr, string Dt,
1320 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1321 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4,
1322 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), IIC_VPALiQ,
1323 OpcodeStr, Dt, "$dst, $src2", "$src1 = $dst",
1324 [(set QPR:$dst, (ResTy (IntOp (ResTy QPR:$src1), (OpTy QPR:$src2))))]>;
1326 // Shift by immediate,
1327 // both double- and quad-register.
1328 class N2VDSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1329 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1330 ValueType Ty, SDNode OpNode>
1331 : N2VImm<op24, op23, op11_8, op7, 0, op4,
1332 (outs DPR:$dst), (ins DPR:$src, i32imm:$SIMM), f, itin,
1333 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1334 [(set DPR:$dst, (Ty (OpNode (Ty DPR:$src), (i32 imm:$SIMM))))]>;
1335 class N2VQSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1336 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1337 ValueType Ty, SDNode OpNode>
1338 : N2VImm<op24, op23, op11_8, op7, 1, op4,
1339 (outs QPR:$dst), (ins QPR:$src, i32imm:$SIMM), f, itin,
1340 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1341 [(set QPR:$dst, (Ty (OpNode (Ty QPR:$src), (i32 imm:$SIMM))))]>;
1343 // Long shift by immediate.
1344 class N2VLSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
1345 string OpcodeStr, string Dt,
1346 ValueType ResTy, ValueType OpTy, SDNode OpNode>
1347 : N2VImm<op24, op23, op11_8, op7, op6, op4,
1348 (outs QPR:$dst), (ins DPR:$src, i32imm:$SIMM), N2RegVShLFrm,
1349 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1350 [(set QPR:$dst, (ResTy (OpNode (OpTy DPR:$src),
1351 (i32 imm:$SIMM))))]>;
1353 // Narrow shift by immediate.
1354 class N2VNSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
1355 InstrItinClass itin, string OpcodeStr, string Dt,
1356 ValueType ResTy, ValueType OpTy, SDNode OpNode>
1357 : N2VImm<op24, op23, op11_8, op7, op6, op4,
1358 (outs DPR:$dst), (ins QPR:$src, i32imm:$SIMM), N2RegVShRFrm, itin,
1359 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1360 [(set DPR:$dst, (ResTy (OpNode (OpTy QPR:$src),
1361 (i32 imm:$SIMM))))]>;
1363 // Shift right by immediate and accumulate,
1364 // both double- and quad-register.
1365 class N2VDShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1366 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1367 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$dst),
1368 (ins DPR:$src1, DPR:$src2, i32imm:$SIMM), N2RegVShRFrm, IIC_VPALiD,
1369 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1370 [(set DPR:$dst, (Ty (add DPR:$src1,
1371 (Ty (ShOp DPR:$src2, (i32 imm:$SIMM))))))]>;
1372 class N2VQShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1373 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1374 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$dst),
1375 (ins QPR:$src1, QPR:$src2, i32imm:$SIMM), N2RegVShRFrm, IIC_VPALiD,
1376 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1377 [(set QPR:$dst, (Ty (add QPR:$src1,
1378 (Ty (ShOp QPR:$src2, (i32 imm:$SIMM))))))]>;
1380 // Shift by immediate and insert,
1381 // both double- and quad-register.
1382 class N2VDShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1383 Format f, string OpcodeStr, string Dt, ValueType Ty,SDNode ShOp>
1384 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$dst),
1385 (ins DPR:$src1, DPR:$src2, i32imm:$SIMM), f, IIC_VSHLiD,
1386 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1387 [(set DPR:$dst, (Ty (ShOp DPR:$src1, DPR:$src2, (i32 imm:$SIMM))))]>;
1388 class N2VQShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1389 Format f, string OpcodeStr, string Dt, ValueType Ty,SDNode ShOp>
1390 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$dst),
1391 (ins QPR:$src1, QPR:$src2, i32imm:$SIMM), f, IIC_VSHLiQ,
1392 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1393 [(set QPR:$dst, (Ty (ShOp QPR:$src1, QPR:$src2, (i32 imm:$SIMM))))]>;
1395 // Convert, with fractional bits immediate,
1396 // both double- and quad-register.
1397 class N2VCvtD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1398 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1400 : N2VImm<op24, op23, op11_8, op7, 0, op4,
1401 (outs DPR:$dst), (ins DPR:$src, i32imm:$SIMM), NVCVTFrm,
1402 IIC_VUNAD, OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1403 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src), (i32 imm:$SIMM))))]>;
1404 class N2VCvtQ<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1405 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1407 : N2VImm<op24, op23, op11_8, op7, 1, op4,
1408 (outs QPR:$dst), (ins QPR:$src, i32imm:$SIMM), NVCVTFrm,
1409 IIC_VUNAQ, OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1410 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src), (i32 imm:$SIMM))))]>;
1412 //===----------------------------------------------------------------------===//
1414 //===----------------------------------------------------------------------===//
1416 // Abbreviations used in multiclass suffixes:
1417 // Q = quarter int (8 bit) elements
1418 // H = half int (16 bit) elements
1419 // S = single int (32 bit) elements
1420 // D = double int (64 bit) elements
1422 // Neon 2-register vector operations -- for disassembly only.
1424 // First with only element sizes of 8, 16 and 32 bits:
1425 multiclass N2V_QHS_cmp<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1426 bits<5> op11_7, bit op4, string opc, string Dt,
1428 // 64-bit vector types.
1429 def v8i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 0, op4,
1430 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1431 opc, !strconcat(Dt, "8"), asm, "", []>;
1432 def v4i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 0, op4,
1433 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1434 opc, !strconcat(Dt, "16"), asm, "", []>;
1435 def v2i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
1436 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1437 opc, !strconcat(Dt, "32"), asm, "", []>;
1438 def v2f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
1439 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1440 opc, "f32", asm, "", []> {
1441 let Inst{10} = 1; // overwrite F = 1
1444 // 128-bit vector types.
1445 def v16i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 1, op4,
1446 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1447 opc, !strconcat(Dt, "8"), asm, "", []>;
1448 def v8i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 1, op4,
1449 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1450 opc, !strconcat(Dt, "16"), asm, "", []>;
1451 def v4i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
1452 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1453 opc, !strconcat(Dt, "32"), asm, "", []>;
1454 def v4f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
1455 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1456 opc, "f32", asm, "", []> {
1457 let Inst{10} = 1; // overwrite F = 1
1461 // Neon 3-register vector operations.
1463 // First with only element sizes of 8, 16 and 32 bits:
1464 multiclass N3V_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1465 InstrItinClass itinD16, InstrItinClass itinD32,
1466 InstrItinClass itinQ16, InstrItinClass itinQ32,
1467 string OpcodeStr, string Dt,
1468 SDNode OpNode, bit Commutable = 0> {
1469 // 64-bit vector types.
1470 def v8i8 : N3VD<op24, op23, 0b00, op11_8, op4, itinD16,
1471 OpcodeStr, !strconcat(Dt, "8"),
1472 v8i8, v8i8, OpNode, Commutable>;
1473 def v4i16 : N3VD<op24, op23, 0b01, op11_8, op4, itinD16,
1474 OpcodeStr, !strconcat(Dt, "16"),
1475 v4i16, v4i16, OpNode, Commutable>;
1476 def v2i32 : N3VD<op24, op23, 0b10, op11_8, op4, itinD32,
1477 OpcodeStr, !strconcat(Dt, "32"),
1478 v2i32, v2i32, OpNode, Commutable>;
1480 // 128-bit vector types.
1481 def v16i8 : N3VQ<op24, op23, 0b00, op11_8, op4, itinQ16,
1482 OpcodeStr, !strconcat(Dt, "8"),
1483 v16i8, v16i8, OpNode, Commutable>;
1484 def v8i16 : N3VQ<op24, op23, 0b01, op11_8, op4, itinQ16,
1485 OpcodeStr, !strconcat(Dt, "16"),
1486 v8i16, v8i16, OpNode, Commutable>;
1487 def v4i32 : N3VQ<op24, op23, 0b10, op11_8, op4, itinQ32,
1488 OpcodeStr, !strconcat(Dt, "32"),
1489 v4i32, v4i32, OpNode, Commutable>;
1492 multiclass N3VSL_HS<bits<4> op11_8, string OpcodeStr, string Dt, SDNode ShOp> {
1493 def v4i16 : N3VDSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
1495 def v2i32 : N3VDSL<0b10, op11_8, IIC_VMULi32D, OpcodeStr, !strconcat(Dt,"32"),
1497 def v8i16 : N3VQSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
1498 v8i16, v4i16, ShOp>;
1499 def v4i32 : N3VQSL<0b10, op11_8, IIC_VMULi32Q, OpcodeStr, !strconcat(Dt,"32"),
1500 v4i32, v2i32, ShOp>;
1503 // ....then also with element size 64 bits:
1504 multiclass N3V_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1505 InstrItinClass itinD, InstrItinClass itinQ,
1506 string OpcodeStr, string Dt,
1507 SDNode OpNode, bit Commutable = 0>
1508 : N3V_QHS<op24, op23, op11_8, op4, itinD, itinD, itinQ, itinQ,
1509 OpcodeStr, Dt, OpNode, Commutable> {
1510 def v1i64 : N3VD<op24, op23, 0b11, op11_8, op4, itinD,
1511 OpcodeStr, !strconcat(Dt, "64"),
1512 v1i64, v1i64, OpNode, Commutable>;
1513 def v2i64 : N3VQ<op24, op23, 0b11, op11_8, op4, itinQ,
1514 OpcodeStr, !strconcat(Dt, "64"),
1515 v2i64, v2i64, OpNode, Commutable>;
1519 // Neon Narrowing 2-register vector intrinsics,
1520 // source operand element sizes of 16, 32 and 64 bits:
1521 multiclass N2VNInt_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1522 bits<5> op11_7, bit op6, bit op4,
1523 InstrItinClass itin, string OpcodeStr, string Dt,
1525 def v8i8 : N2VNInt<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
1526 itin, OpcodeStr, !strconcat(Dt, "16"),
1527 v8i8, v8i16, IntOp>;
1528 def v4i16 : N2VNInt<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
1529 itin, OpcodeStr, !strconcat(Dt, "32"),
1530 v4i16, v4i32, IntOp>;
1531 def v2i32 : N2VNInt<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
1532 itin, OpcodeStr, !strconcat(Dt, "64"),
1533 v2i32, v2i64, IntOp>;
1537 // Neon Lengthening 2-register vector intrinsic (currently specific to VMOVL).
1538 // source operand element sizes of 16, 32 and 64 bits:
1539 multiclass N2VL_QHS<bits<2> op24_23, bits<5> op11_7, bit op6, bit op4,
1540 string OpcodeStr, string Dt, SDNode OpNode> {
1541 def v8i16 : N2VL<op24_23, 0b00, 0b10, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
1542 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode>;
1543 def v4i32 : N2VL<op24_23, 0b01, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
1544 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
1545 def v2i64 : N2VL<op24_23, 0b10, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
1546 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
1550 // Neon 3-register vector intrinsics.
1552 // First with only element sizes of 16 and 32 bits:
1553 multiclass N3VInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
1554 InstrItinClass itinD16, InstrItinClass itinD32,
1555 InstrItinClass itinQ16, InstrItinClass itinQ32,
1556 string OpcodeStr, string Dt,
1557 Intrinsic IntOp, bit Commutable = 0> {
1558 // 64-bit vector types.
1559 def v4i16 : N3VDInt<op24, op23, 0b01, op11_8, op4, f, itinD16,
1560 OpcodeStr, !strconcat(Dt, "16"),
1561 v4i16, v4i16, IntOp, Commutable>;
1562 def v2i32 : N3VDInt<op24, op23, 0b10, op11_8, op4, f, itinD32,
1563 OpcodeStr, !strconcat(Dt, "32"),
1564 v2i32, v2i32, IntOp, Commutable>;
1566 // 128-bit vector types.
1567 def v8i16 : N3VQInt<op24, op23, 0b01, op11_8, op4, f, itinQ16,
1568 OpcodeStr, !strconcat(Dt, "16"),
1569 v8i16, v8i16, IntOp, Commutable>;
1570 def v4i32 : N3VQInt<op24, op23, 0b10, op11_8, op4, f, itinQ32,
1571 OpcodeStr, !strconcat(Dt, "32"),
1572 v4i32, v4i32, IntOp, Commutable>;
1575 multiclass N3VIntSL_HS<bits<4> op11_8,
1576 InstrItinClass itinD16, InstrItinClass itinD32,
1577 InstrItinClass itinQ16, InstrItinClass itinQ32,
1578 string OpcodeStr, string Dt, Intrinsic IntOp> {
1579 def v4i16 : N3VDIntSL16<0b01, op11_8, itinD16,
1580 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp>;
1581 def v2i32 : N3VDIntSL<0b10, op11_8, itinD32,
1582 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp>;
1583 def v8i16 : N3VQIntSL16<0b01, op11_8, itinQ16,
1584 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16, IntOp>;
1585 def v4i32 : N3VQIntSL<0b10, op11_8, itinQ32,
1586 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32, IntOp>;
1589 // ....then also with element size of 8 bits:
1590 multiclass N3VInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
1591 InstrItinClass itinD16, InstrItinClass itinD32,
1592 InstrItinClass itinQ16, InstrItinClass itinQ32,
1593 string OpcodeStr, string Dt,
1594 Intrinsic IntOp, bit Commutable = 0>
1595 : N3VInt_HS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
1596 OpcodeStr, Dt, IntOp, Commutable> {
1597 def v8i8 : N3VDInt<op24, op23, 0b00, op11_8, op4, f, itinD16,
1598 OpcodeStr, !strconcat(Dt, "8"),
1599 v8i8, v8i8, IntOp, Commutable>;
1600 def v16i8 : N3VQInt<op24, op23, 0b00, op11_8, op4, f, itinQ16,
1601 OpcodeStr, !strconcat(Dt, "8"),
1602 v16i8, v16i8, IntOp, Commutable>;
1605 // ....then also with element size of 64 bits:
1606 multiclass N3VInt_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
1607 InstrItinClass itinD16, InstrItinClass itinD32,
1608 InstrItinClass itinQ16, InstrItinClass itinQ32,
1609 string OpcodeStr, string Dt,
1610 Intrinsic IntOp, bit Commutable = 0>
1611 : N3VInt_QHS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
1612 OpcodeStr, Dt, IntOp, Commutable> {
1613 def v1i64 : N3VDInt<op24, op23, 0b11, op11_8, op4, f, itinD32,
1614 OpcodeStr, !strconcat(Dt, "64"),
1615 v1i64, v1i64, IntOp, Commutable>;
1616 def v2i64 : N3VQInt<op24, op23, 0b11, op11_8, op4, f, itinQ32,
1617 OpcodeStr, !strconcat(Dt, "64"),
1618 v2i64, v2i64, IntOp, Commutable>;
1621 // Neon Narrowing 3-register vector intrinsics,
1622 // source operand element sizes of 16, 32 and 64 bits:
1623 multiclass N3VNInt_HSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1624 string OpcodeStr, string Dt,
1625 Intrinsic IntOp, bit Commutable = 0> {
1626 def v8i8 : N3VNInt<op24, op23, 0b00, op11_8, op4,
1627 OpcodeStr, !strconcat(Dt, "16"),
1628 v8i8, v8i16, IntOp, Commutable>;
1629 def v4i16 : N3VNInt<op24, op23, 0b01, op11_8, op4,
1630 OpcodeStr, !strconcat(Dt, "32"),
1631 v4i16, v4i32, IntOp, Commutable>;
1632 def v2i32 : N3VNInt<op24, op23, 0b10, op11_8, op4,
1633 OpcodeStr, !strconcat(Dt, "64"),
1634 v2i32, v2i64, IntOp, Commutable>;
1638 // Neon Long 3-register vector intrinsics.
1640 // First with only element sizes of 16 and 32 bits:
1641 multiclass N3VLInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
1642 InstrItinClass itin16, InstrItinClass itin32,
1643 string OpcodeStr, string Dt,
1644 Intrinsic IntOp, bit Commutable = 0> {
1645 def v4i32 : N3VLInt<op24, op23, 0b01, op11_8, op4, itin16,
1646 OpcodeStr, !strconcat(Dt, "16"),
1647 v4i32, v4i16, IntOp, Commutable>;
1648 def v2i64 : N3VLInt<op24, op23, 0b10, op11_8, op4, itin32,
1649 OpcodeStr, !strconcat(Dt, "32"),
1650 v2i64, v2i32, IntOp, Commutable>;
1653 multiclass N3VLIntSL_HS<bit op24, bits<4> op11_8,
1654 InstrItinClass itin, string OpcodeStr, string Dt,
1656 def v4i16 : N3VLIntSL16<op24, 0b01, op11_8, itin,
1657 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
1658 def v2i32 : N3VLIntSL<op24, 0b10, op11_8, itin,
1659 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1662 // ....then also with element size of 8 bits:
1663 multiclass N3VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1664 InstrItinClass itin16, InstrItinClass itin32,
1665 string OpcodeStr, string Dt,
1666 Intrinsic IntOp, bit Commutable = 0>
1667 : N3VLInt_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt,
1668 IntOp, Commutable> {
1669 def v8i16 : N3VLInt<op24, op23, 0b00, op11_8, op4, itin16,
1670 OpcodeStr, !strconcat(Dt, "8"),
1671 v8i16, v8i8, IntOp, Commutable>;
1675 // Neon Wide 3-register vector intrinsics,
1676 // source operand element sizes of 8, 16 and 32 bits:
1677 multiclass N3VWInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1678 string OpcodeStr, string Dt,
1679 Intrinsic IntOp, bit Commutable = 0> {
1680 def v8i16 : N3VWInt<op24, op23, 0b00, op11_8, op4,
1681 OpcodeStr, !strconcat(Dt, "8"),
1682 v8i16, v8i8, IntOp, Commutable>;
1683 def v4i32 : N3VWInt<op24, op23, 0b01, op11_8, op4,
1684 OpcodeStr, !strconcat(Dt, "16"),
1685 v4i32, v4i16, IntOp, Commutable>;
1686 def v2i64 : N3VWInt<op24, op23, 0b10, op11_8, op4,
1687 OpcodeStr, !strconcat(Dt, "32"),
1688 v2i64, v2i32, IntOp, Commutable>;
1692 // Neon Multiply-Op vector operations,
1693 // element sizes of 8, 16 and 32 bits:
1694 multiclass N3VMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1695 InstrItinClass itinD16, InstrItinClass itinD32,
1696 InstrItinClass itinQ16, InstrItinClass itinQ32,
1697 string OpcodeStr, string Dt, SDNode OpNode> {
1698 // 64-bit vector types.
1699 def v8i8 : N3VDMulOp<op24, op23, 0b00, op11_8, op4, itinD16,
1700 OpcodeStr, !strconcat(Dt, "8"), v8i8, mul, OpNode>;
1701 def v4i16 : N3VDMulOp<op24, op23, 0b01, op11_8, op4, itinD16,
1702 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, OpNode>;
1703 def v2i32 : N3VDMulOp<op24, op23, 0b10, op11_8, op4, itinD32,
1704 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, OpNode>;
1706 // 128-bit vector types.
1707 def v16i8 : N3VQMulOp<op24, op23, 0b00, op11_8, op4, itinQ16,
1708 OpcodeStr, !strconcat(Dt, "8"), v16i8, mul, OpNode>;
1709 def v8i16 : N3VQMulOp<op24, op23, 0b01, op11_8, op4, itinQ16,
1710 OpcodeStr, !strconcat(Dt, "16"), v8i16, mul, OpNode>;
1711 def v4i32 : N3VQMulOp<op24, op23, 0b10, op11_8, op4, itinQ32,
1712 OpcodeStr, !strconcat(Dt, "32"), v4i32, mul, OpNode>;
1715 multiclass N3VMulOpSL_HS<bits<4> op11_8,
1716 InstrItinClass itinD16, InstrItinClass itinD32,
1717 InstrItinClass itinQ16, InstrItinClass itinQ32,
1718 string OpcodeStr, string Dt, SDNode ShOp> {
1719 def v4i16 : N3VDMulOpSL16<0b01, op11_8, itinD16,
1720 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, ShOp>;
1721 def v2i32 : N3VDMulOpSL<0b10, op11_8, itinD32,
1722 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, ShOp>;
1723 def v8i16 : N3VQMulOpSL16<0b01, op11_8, itinQ16,
1724 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16,
1726 def v4i32 : N3VQMulOpSL<0b10, op11_8, itinQ32,
1727 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32,
1731 // Neon 3-argument intrinsics,
1732 // element sizes of 8, 16 and 32 bits:
1733 multiclass N3VInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1734 InstrItinClass itinD, InstrItinClass itinQ,
1735 string OpcodeStr, string Dt, Intrinsic IntOp> {
1736 // 64-bit vector types.
1737 def v8i8 : N3VDInt3<op24, op23, 0b00, op11_8, op4, itinD,
1738 OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
1739 def v4i16 : N3VDInt3<op24, op23, 0b01, op11_8, op4, itinD,
1740 OpcodeStr, !strconcat(Dt, "16"), v4i16, v4i16, IntOp>;
1741 def v2i32 : N3VDInt3<op24, op23, 0b10, op11_8, op4, itinD,
1742 OpcodeStr, !strconcat(Dt, "32"), v2i32, v2i32, IntOp>;
1744 // 128-bit vector types.
1745 def v16i8 : N3VQInt3<op24, op23, 0b00, op11_8, op4, itinQ,
1746 OpcodeStr, !strconcat(Dt, "8"), v16i8, v16i8, IntOp>;
1747 def v8i16 : N3VQInt3<op24, op23, 0b01, op11_8, op4, itinQ,
1748 OpcodeStr, !strconcat(Dt, "16"), v8i16, v8i16, IntOp>;
1749 def v4i32 : N3VQInt3<op24, op23, 0b10, op11_8, op4, itinQ,
1750 OpcodeStr, !strconcat(Dt, "32"), v4i32, v4i32, IntOp>;
1754 // Neon Long 3-argument intrinsics.
1756 // First with only element sizes of 16 and 32 bits:
1757 multiclass N3VLInt3_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
1758 InstrItinClass itin16, InstrItinClass itin32,
1759 string OpcodeStr, string Dt, Intrinsic IntOp> {
1760 def v4i32 : N3VLInt3<op24, op23, 0b01, op11_8, op4, itin16,
1761 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
1762 def v2i64 : N3VLInt3<op24, op23, 0b10, op11_8, op4, itin32,
1763 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1766 multiclass N3VLInt3SL_HS<bit op24, bits<4> op11_8,
1767 string OpcodeStr, string Dt, Intrinsic IntOp> {
1768 def v4i16 : N3VLInt3SL16<op24, 0b01, op11_8, IIC_VMACi16D,
1769 OpcodeStr, !strconcat(Dt,"16"), v4i32, v4i16, IntOp>;
1770 def v2i32 : N3VLInt3SL<op24, 0b10, op11_8, IIC_VMACi32D,
1771 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1774 // ....then also with element size of 8 bits:
1775 multiclass N3VLInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1776 InstrItinClass itin16, InstrItinClass itin32,
1777 string OpcodeStr, string Dt, Intrinsic IntOp>
1778 : N3VLInt3_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt, IntOp> {
1779 def v8i16 : N3VLInt3<op24, op23, 0b00, op11_8, op4, itin16,
1780 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
1784 // Neon 2-register vector intrinsics,
1785 // element sizes of 8, 16 and 32 bits:
1786 multiclass N2VInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1787 bits<5> op11_7, bit op4,
1788 InstrItinClass itinD, InstrItinClass itinQ,
1789 string OpcodeStr, string Dt, Intrinsic IntOp> {
1790 // 64-bit vector types.
1791 def v8i8 : N2VDInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1792 itinD, OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
1793 def v4i16 : N2VDInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1794 itinD, OpcodeStr, !strconcat(Dt, "16"),v4i16,v4i16,IntOp>;
1795 def v2i32 : N2VDInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1796 itinD, OpcodeStr, !strconcat(Dt, "32"),v2i32,v2i32,IntOp>;
1798 // 128-bit vector types.
1799 def v16i8 : N2VQInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1800 itinQ, OpcodeStr, !strconcat(Dt, "8"), v16i8,v16i8,IntOp>;
1801 def v8i16 : N2VQInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1802 itinQ, OpcodeStr, !strconcat(Dt, "16"),v8i16,v8i16,IntOp>;
1803 def v4i32 : N2VQInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1804 itinQ, OpcodeStr, !strconcat(Dt, "32"),v4i32,v4i32,IntOp>;
1808 // Neon Pairwise long 2-register intrinsics,
1809 // element sizes of 8, 16 and 32 bits:
1810 multiclass N2VPLInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1811 bits<5> op11_7, bit op4,
1812 string OpcodeStr, string Dt, Intrinsic IntOp> {
1813 // 64-bit vector types.
1814 def v8i8 : N2VDPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1815 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
1816 def v4i16 : N2VDPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1817 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
1818 def v2i32 : N2VDPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1819 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
1821 // 128-bit vector types.
1822 def v16i8 : N2VQPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1823 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
1824 def v8i16 : N2VQPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1825 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
1826 def v4i32 : N2VQPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1827 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
1831 // Neon Pairwise long 2-register accumulate intrinsics,
1832 // element sizes of 8, 16 and 32 bits:
1833 multiclass N2VPLInt2_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1834 bits<5> op11_7, bit op4,
1835 string OpcodeStr, string Dt, Intrinsic IntOp> {
1836 // 64-bit vector types.
1837 def v8i8 : N2VDPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1838 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
1839 def v4i16 : N2VDPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1840 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
1841 def v2i32 : N2VDPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1842 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
1844 // 128-bit vector types.
1845 def v16i8 : N2VQPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1846 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
1847 def v8i16 : N2VQPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1848 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
1849 def v4i32 : N2VQPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1850 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
1854 // Neon 2-register vector shift by immediate,
1855 // with f of either N2RegVShLFrm or N2RegVShRFrm
1856 // element sizes of 8, 16, 32 and 64 bits:
1857 multiclass N2VSh_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1858 InstrItinClass itin, string OpcodeStr, string Dt,
1859 SDNode OpNode, Format f> {
1860 // 64-bit vector types.
1861 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
1862 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
1863 let Inst{21-19} = 0b001; // imm6 = 001xxx
1865 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
1866 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
1867 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1869 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
1870 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
1871 let Inst{21} = 0b1; // imm6 = 1xxxxx
1873 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, f, itin,
1874 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
1877 // 128-bit vector types.
1878 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
1879 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
1880 let Inst{21-19} = 0b001; // imm6 = 001xxx
1882 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
1883 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
1884 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1886 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
1887 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
1888 let Inst{21} = 0b1; // imm6 = 1xxxxx
1890 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, f, itin,
1891 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
1895 // Neon Shift-Accumulate vector operations,
1896 // element sizes of 8, 16, 32 and 64 bits:
1897 multiclass N2VShAdd_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1898 string OpcodeStr, string Dt, SDNode ShOp> {
1899 // 64-bit vector types.
1900 def v8i8 : N2VDShAdd<op24, op23, op11_8, 0, op4,
1901 OpcodeStr, !strconcat(Dt, "8"), v8i8, ShOp> {
1902 let Inst{21-19} = 0b001; // imm6 = 001xxx
1904 def v4i16 : N2VDShAdd<op24, op23, op11_8, 0, op4,
1905 OpcodeStr, !strconcat(Dt, "16"), v4i16, ShOp> {
1906 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1908 def v2i32 : N2VDShAdd<op24, op23, op11_8, 0, op4,
1909 OpcodeStr, !strconcat(Dt, "32"), v2i32, ShOp> {
1910 let Inst{21} = 0b1; // imm6 = 1xxxxx
1912 def v1i64 : N2VDShAdd<op24, op23, op11_8, 1, op4,
1913 OpcodeStr, !strconcat(Dt, "64"), v1i64, ShOp>;
1916 // 128-bit vector types.
1917 def v16i8 : N2VQShAdd<op24, op23, op11_8, 0, op4,
1918 OpcodeStr, !strconcat(Dt, "8"), v16i8, ShOp> {
1919 let Inst{21-19} = 0b001; // imm6 = 001xxx
1921 def v8i16 : N2VQShAdd<op24, op23, op11_8, 0, op4,
1922 OpcodeStr, !strconcat(Dt, "16"), v8i16, ShOp> {
1923 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1925 def v4i32 : N2VQShAdd<op24, op23, op11_8, 0, op4,
1926 OpcodeStr, !strconcat(Dt, "32"), v4i32, ShOp> {
1927 let Inst{21} = 0b1; // imm6 = 1xxxxx
1929 def v2i64 : N2VQShAdd<op24, op23, op11_8, 1, op4,
1930 OpcodeStr, !strconcat(Dt, "64"), v2i64, ShOp>;
1935 // Neon Shift-Insert vector operations,
1936 // with f of either N2RegVShLFrm or N2RegVShRFrm
1937 // element sizes of 8, 16, 32 and 64 bits:
1938 multiclass N2VShIns_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1939 string OpcodeStr, SDNode ShOp,
1941 // 64-bit vector types.
1942 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4,
1943 f, OpcodeStr, "8", v8i8, ShOp> {
1944 let Inst{21-19} = 0b001; // imm6 = 001xxx
1946 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4,
1947 f, OpcodeStr, "16", v4i16, ShOp> {
1948 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1950 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4,
1951 f, OpcodeStr, "32", v2i32, ShOp> {
1952 let Inst{21} = 0b1; // imm6 = 1xxxxx
1954 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4,
1955 f, OpcodeStr, "64", v1i64, ShOp>;
1958 // 128-bit vector types.
1959 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4,
1960 f, OpcodeStr, "8", v16i8, ShOp> {
1961 let Inst{21-19} = 0b001; // imm6 = 001xxx
1963 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4,
1964 f, OpcodeStr, "16", v8i16, ShOp> {
1965 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1967 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4,
1968 f, OpcodeStr, "32", v4i32, ShOp> {
1969 let Inst{21} = 0b1; // imm6 = 1xxxxx
1971 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4,
1972 f, OpcodeStr, "64", v2i64, ShOp>;
1976 // Neon Shift Long operations,
1977 // element sizes of 8, 16, 32 bits:
1978 multiclass N2VLSh_QHS<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
1979 bit op4, string OpcodeStr, string Dt, SDNode OpNode> {
1980 def v8i16 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
1981 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode> {
1982 let Inst{21-19} = 0b001; // imm6 = 001xxx
1984 def v4i32 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
1985 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode> {
1986 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1988 def v2i64 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
1989 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode> {
1990 let Inst{21} = 0b1; // imm6 = 1xxxxx
1994 // Neon Shift Narrow operations,
1995 // element sizes of 16, 32, 64 bits:
1996 multiclass N2VNSh_HSD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
1997 bit op4, InstrItinClass itin, string OpcodeStr, string Dt,
1999 def v8i8 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
2000 OpcodeStr, !strconcat(Dt, "16"), v8i8, v8i16, OpNode> {
2001 let Inst{21-19} = 0b001; // imm6 = 001xxx
2003 def v4i16 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
2004 OpcodeStr, !strconcat(Dt, "32"), v4i16, v4i32, OpNode> {
2005 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2007 def v2i32 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
2008 OpcodeStr, !strconcat(Dt, "64"), v2i32, v2i64, OpNode> {
2009 let Inst{21} = 0b1; // imm6 = 1xxxxx
2013 //===----------------------------------------------------------------------===//
2014 // Instruction Definitions.
2015 //===----------------------------------------------------------------------===//
2017 // Vector Add Operations.
2019 // VADD : Vector Add (integer and floating-point)
2020 defm VADD : N3V_QHSD<0, 0, 0b1000, 0, IIC_VBINiD, IIC_VBINiQ, "vadd", "i",
2022 def VADDfd : N3VD<0, 0, 0b00, 0b1101, 0, IIC_VBIND, "vadd", "f32",
2023 v2f32, v2f32, fadd, 1>;
2024 def VADDfq : N3VQ<0, 0, 0b00, 0b1101, 0, IIC_VBINQ, "vadd", "f32",
2025 v4f32, v4f32, fadd, 1>;
2026 // VADDL : Vector Add Long (Q = D + D)
2027 defm VADDLs : N3VLInt_QHS<0,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
2028 "vaddl", "s", int_arm_neon_vaddls, 1>;
2029 defm VADDLu : N3VLInt_QHS<1,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
2030 "vaddl", "u", int_arm_neon_vaddlu, 1>;
2031 // VADDW : Vector Add Wide (Q = Q + D)
2032 defm VADDWs : N3VWInt_QHS<0,1,0b0001,0, "vaddw", "s", int_arm_neon_vaddws, 0>;
2033 defm VADDWu : N3VWInt_QHS<1,1,0b0001,0, "vaddw", "u", int_arm_neon_vaddwu, 0>;
2034 // VHADD : Vector Halving Add
2035 defm VHADDs : N3VInt_QHS<0, 0, 0b0000, 0, N3RegFrm,
2036 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2037 "vhadd", "s", int_arm_neon_vhadds, 1>;
2038 defm VHADDu : N3VInt_QHS<1, 0, 0b0000, 0, N3RegFrm,
2039 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2040 "vhadd", "u", int_arm_neon_vhaddu, 1>;
2041 // VRHADD : Vector Rounding Halving Add
2042 defm VRHADDs : N3VInt_QHS<0, 0, 0b0001, 0, N3RegFrm,
2043 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2044 "vrhadd", "s", int_arm_neon_vrhadds, 1>;
2045 defm VRHADDu : N3VInt_QHS<1, 0, 0b0001, 0, N3RegFrm,
2046 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2047 "vrhadd", "u", int_arm_neon_vrhaddu, 1>;
2048 // VQADD : Vector Saturating Add
2049 defm VQADDs : N3VInt_QHSD<0, 0, 0b0000, 1, N3RegFrm,
2050 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2051 "vqadd", "s", int_arm_neon_vqadds, 1>;
2052 defm VQADDu : N3VInt_QHSD<1, 0, 0b0000, 1, N3RegFrm,
2053 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2054 "vqadd", "u", int_arm_neon_vqaddu, 1>;
2055 // VADDHN : Vector Add and Narrow Returning High Half (D = Q + Q)
2056 defm VADDHN : N3VNInt_HSD<0,1,0b0100,0, "vaddhn", "i",
2057 int_arm_neon_vaddhn, 1>;
2058 // VRADDHN : Vector Rounding Add and Narrow Returning High Half (D = Q + Q)
2059 defm VRADDHN : N3VNInt_HSD<1,1,0b0100,0, "vraddhn", "i",
2060 int_arm_neon_vraddhn, 1>;
2062 // Vector Multiply Operations.
2064 // VMUL : Vector Multiply (integer, polynomial and floating-point)
2065 defm VMUL : N3V_QHS<0, 0, 0b1001, 1, IIC_VMULi16D, IIC_VMULi32D,
2066 IIC_VMULi16Q, IIC_VMULi32Q, "vmul", "i", mul, 1>;
2067 def VMULpd : N3VDInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16D, "vmul",
2068 "p8", v8i8, v8i8, int_arm_neon_vmulp, 1>;
2069 def VMULpq : N3VQInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16Q, "vmul",
2070 "p8", v16i8, v16i8, int_arm_neon_vmulp, 1>;
2071 def VMULfd : N3VD<1, 0, 0b00, 0b1101, 1, IIC_VBIND, "vmul", "f32",
2072 v2f32, v2f32, fmul, 1>;
2073 def VMULfq : N3VQ<1, 0, 0b00, 0b1101, 1, IIC_VBINQ, "vmul", "f32",
2074 v4f32, v4f32, fmul, 1>;
2075 defm VMULsl : N3VSL_HS<0b1000, "vmul", "i", mul>;
2076 def VMULslfd : N3VDSL<0b10, 0b1001, IIC_VBIND, "vmul", "f32", v2f32, fmul>;
2077 def VMULslfq : N3VQSL<0b10, 0b1001, IIC_VBINQ, "vmul", "f32", v4f32,
2080 def : Pat<(v8i16 (mul (v8i16 QPR:$src1),
2081 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
2082 (v8i16 (VMULslv8i16 (v8i16 QPR:$src1),
2083 (v4i16 (EXTRACT_SUBREG QPR:$src2,
2084 (DSubReg_i16_reg imm:$lane))),
2085 (SubReg_i16_lane imm:$lane)))>;
2086 def : Pat<(v4i32 (mul (v4i32 QPR:$src1),
2087 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
2088 (v4i32 (VMULslv4i32 (v4i32 QPR:$src1),
2089 (v2i32 (EXTRACT_SUBREG QPR:$src2,
2090 (DSubReg_i32_reg imm:$lane))),
2091 (SubReg_i32_lane imm:$lane)))>;
2092 def : Pat<(v4f32 (fmul (v4f32 QPR:$src1),
2093 (v4f32 (NEONvduplane (v4f32 QPR:$src2), imm:$lane)))),
2094 (v4f32 (VMULslfq (v4f32 QPR:$src1),
2095 (v2f32 (EXTRACT_SUBREG QPR:$src2,
2096 (DSubReg_i32_reg imm:$lane))),
2097 (SubReg_i32_lane imm:$lane)))>;
2099 // VQDMULH : Vector Saturating Doubling Multiply Returning High Half
2100 defm VQDMULH : N3VInt_HS<0, 0, 0b1011, 0, N3RegFrm, IIC_VMULi16D, IIC_VMULi32D,
2101 IIC_VMULi16Q, IIC_VMULi32Q,
2102 "vqdmulh", "s", int_arm_neon_vqdmulh, 1>;
2103 defm VQDMULHsl: N3VIntSL_HS<0b1100, IIC_VMULi16D, IIC_VMULi32D,
2104 IIC_VMULi16Q, IIC_VMULi32Q,
2105 "vqdmulh", "s", int_arm_neon_vqdmulh>;
2106 def : Pat<(v8i16 (int_arm_neon_vqdmulh (v8i16 QPR:$src1),
2107 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
2109 (v8i16 (VQDMULHslv8i16 (v8i16 QPR:$src1),
2110 (v4i16 (EXTRACT_SUBREG QPR:$src2,
2111 (DSubReg_i16_reg imm:$lane))),
2112 (SubReg_i16_lane imm:$lane)))>;
2113 def : Pat<(v4i32 (int_arm_neon_vqdmulh (v4i32 QPR:$src1),
2114 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
2116 (v4i32 (VQDMULHslv4i32 (v4i32 QPR:$src1),
2117 (v2i32 (EXTRACT_SUBREG QPR:$src2,
2118 (DSubReg_i32_reg imm:$lane))),
2119 (SubReg_i32_lane imm:$lane)))>;
2121 // VQRDMULH : Vector Rounding Saturating Doubling Multiply Returning High Half
2122 defm VQRDMULH : N3VInt_HS<1, 0, 0b1011, 0, N3RegFrm,
2123 IIC_VMULi16D,IIC_VMULi32D,IIC_VMULi16Q,IIC_VMULi32Q,
2124 "vqrdmulh", "s", int_arm_neon_vqrdmulh, 1>;
2125 defm VQRDMULHsl : N3VIntSL_HS<0b1101, IIC_VMULi16D, IIC_VMULi32D,
2126 IIC_VMULi16Q, IIC_VMULi32Q,
2127 "vqrdmulh", "s", int_arm_neon_vqrdmulh>;
2128 def : Pat<(v8i16 (int_arm_neon_vqrdmulh (v8i16 QPR:$src1),
2129 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
2131 (v8i16 (VQRDMULHslv8i16 (v8i16 QPR:$src1),
2132 (v4i16 (EXTRACT_SUBREG QPR:$src2,
2133 (DSubReg_i16_reg imm:$lane))),
2134 (SubReg_i16_lane imm:$lane)))>;
2135 def : Pat<(v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$src1),
2136 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
2138 (v4i32 (VQRDMULHslv4i32 (v4i32 QPR:$src1),
2139 (v2i32 (EXTRACT_SUBREG QPR:$src2,
2140 (DSubReg_i32_reg imm:$lane))),
2141 (SubReg_i32_lane imm:$lane)))>;
2143 // VMULL : Vector Multiply Long (integer and polynomial) (Q = D * D)
2144 defm VMULLs : N3VLInt_QHS<0,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
2145 "vmull", "s", int_arm_neon_vmulls, 1>;
2146 defm VMULLu : N3VLInt_QHS<1,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
2147 "vmull", "u", int_arm_neon_vmullu, 1>;
2148 def VMULLp : N3VLInt<0, 1, 0b00, 0b1110, 0, IIC_VMULi16D, "vmull", "p8",
2149 v8i16, v8i8, int_arm_neon_vmullp, 1>;
2150 defm VMULLsls : N3VLIntSL_HS<0, 0b1010, IIC_VMULi16D, "vmull", "s",
2151 int_arm_neon_vmulls>;
2152 defm VMULLslu : N3VLIntSL_HS<1, 0b1010, IIC_VMULi16D, "vmull", "u",
2153 int_arm_neon_vmullu>;
2155 // VQDMULL : Vector Saturating Doubling Multiply Long (Q = D * D)
2156 defm VQDMULL : N3VLInt_HS<0,1,0b1101,0, IIC_VMULi16D, IIC_VMULi32D,
2157 "vqdmull", "s", int_arm_neon_vqdmull, 1>;
2158 defm VQDMULLsl: N3VLIntSL_HS<0, 0b1011, IIC_VMULi16D,
2159 "vqdmull", "s", int_arm_neon_vqdmull>;
2161 // Vector Multiply-Accumulate and Multiply-Subtract Operations.
2163 // VMLA : Vector Multiply Accumulate (integer and floating-point)
2164 defm VMLA : N3VMulOp_QHS<0, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
2165 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
2166 def VMLAfd : N3VDMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACD, "vmla", "f32",
2168 def VMLAfq : N3VQMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACQ, "vmla", "f32",
2170 defm VMLAsl : N3VMulOpSL_HS<0b0000, IIC_VMACi16D, IIC_VMACi32D,
2171 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
2172 def VMLAslfd : N3VDMulOpSL<0b10, 0b0001, IIC_VMACD, "vmla", "f32",
2174 def VMLAslfq : N3VQMulOpSL<0b10, 0b0001, IIC_VMACQ, "vmla", "f32",
2175 v4f32, v2f32, fmul, fadd>;
2177 def : Pat<(v8i16 (add (v8i16 QPR:$src1),
2178 (mul (v8i16 QPR:$src2),
2179 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
2180 (v8i16 (VMLAslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
2181 (v4i16 (EXTRACT_SUBREG QPR:$src3,
2182 (DSubReg_i16_reg imm:$lane))),
2183 (SubReg_i16_lane imm:$lane)))>;
2185 def : Pat<(v4i32 (add (v4i32 QPR:$src1),
2186 (mul (v4i32 QPR:$src2),
2187 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
2188 (v4i32 (VMLAslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
2189 (v2i32 (EXTRACT_SUBREG QPR:$src3,
2190 (DSubReg_i32_reg imm:$lane))),
2191 (SubReg_i32_lane imm:$lane)))>;
2193 def : Pat<(v4f32 (fadd (v4f32 QPR:$src1),
2194 (fmul (v4f32 QPR:$src2),
2195 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
2196 (v4f32 (VMLAslfq (v4f32 QPR:$src1),
2198 (v2f32 (EXTRACT_SUBREG QPR:$src3,
2199 (DSubReg_i32_reg imm:$lane))),
2200 (SubReg_i32_lane imm:$lane)))>;
2202 // VMLAL : Vector Multiply Accumulate Long (Q += D * D)
2203 defm VMLALs : N3VLInt3_QHS<0,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
2204 "vmlal", "s", int_arm_neon_vmlals>;
2205 defm VMLALu : N3VLInt3_QHS<1,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
2206 "vmlal", "u", int_arm_neon_vmlalu>;
2208 defm VMLALsls : N3VLInt3SL_HS<0, 0b0010, "vmlal", "s", int_arm_neon_vmlals>;
2209 defm VMLALslu : N3VLInt3SL_HS<1, 0b0010, "vmlal", "u", int_arm_neon_vmlalu>;
2211 // VQDMLAL : Vector Saturating Doubling Multiply Accumulate Long (Q += D * D)
2212 defm VQDMLAL : N3VLInt3_HS<0, 1, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
2213 "vqdmlal", "s", int_arm_neon_vqdmlal>;
2214 defm VQDMLALsl: N3VLInt3SL_HS<0, 0b0011, "vqdmlal", "s", int_arm_neon_vqdmlal>;
2216 // VMLS : Vector Multiply Subtract (integer and floating-point)
2217 defm VMLS : N3VMulOp_QHS<1, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
2218 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
2219 def VMLSfd : N3VDMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACD, "vmls", "f32",
2221 def VMLSfq : N3VQMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACQ, "vmls", "f32",
2223 defm VMLSsl : N3VMulOpSL_HS<0b0100, IIC_VMACi16D, IIC_VMACi32D,
2224 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
2225 def VMLSslfd : N3VDMulOpSL<0b10, 0b0101, IIC_VMACD, "vmls", "f32",
2227 def VMLSslfq : N3VQMulOpSL<0b10, 0b0101, IIC_VMACQ, "vmls", "f32",
2228 v4f32, v2f32, fmul, fsub>;
2230 def : Pat<(v8i16 (sub (v8i16 QPR:$src1),
2231 (mul (v8i16 QPR:$src2),
2232 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
2233 (v8i16 (VMLSslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
2234 (v4i16 (EXTRACT_SUBREG QPR:$src3,
2235 (DSubReg_i16_reg imm:$lane))),
2236 (SubReg_i16_lane imm:$lane)))>;
2238 def : Pat<(v4i32 (sub (v4i32 QPR:$src1),
2239 (mul (v4i32 QPR:$src2),
2240 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
2241 (v4i32 (VMLSslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
2242 (v2i32 (EXTRACT_SUBREG QPR:$src3,
2243 (DSubReg_i32_reg imm:$lane))),
2244 (SubReg_i32_lane imm:$lane)))>;
2246 def : Pat<(v4f32 (fsub (v4f32 QPR:$src1),
2247 (fmul (v4f32 QPR:$src2),
2248 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
2249 (v4f32 (VMLSslfq (v4f32 QPR:$src1), (v4f32 QPR:$src2),
2250 (v2f32 (EXTRACT_SUBREG QPR:$src3,
2251 (DSubReg_i32_reg imm:$lane))),
2252 (SubReg_i32_lane imm:$lane)))>;
2254 // VMLSL : Vector Multiply Subtract Long (Q -= D * D)
2255 defm VMLSLs : N3VLInt3_QHS<0,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
2256 "vmlsl", "s", int_arm_neon_vmlsls>;
2257 defm VMLSLu : N3VLInt3_QHS<1,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
2258 "vmlsl", "u", int_arm_neon_vmlslu>;
2260 defm VMLSLsls : N3VLInt3SL_HS<0, 0b0110, "vmlsl", "s", int_arm_neon_vmlsls>;
2261 defm VMLSLslu : N3VLInt3SL_HS<1, 0b0110, "vmlsl", "u", int_arm_neon_vmlslu>;
2263 // VQDMLSL : Vector Saturating Doubling Multiply Subtract Long (Q -= D * D)
2264 defm VQDMLSL : N3VLInt3_HS<0, 1, 0b1011, 0, IIC_VMACi16D, IIC_VMACi32D,
2265 "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
2266 defm VQDMLSLsl: N3VLInt3SL_HS<0, 0b111, "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
2268 // Vector Subtract Operations.
2270 // VSUB : Vector Subtract (integer and floating-point)
2271 defm VSUB : N3V_QHSD<1, 0, 0b1000, 0, IIC_VSUBiD, IIC_VSUBiQ,
2272 "vsub", "i", sub, 0>;
2273 def VSUBfd : N3VD<0, 0, 0b10, 0b1101, 0, IIC_VBIND, "vsub", "f32",
2274 v2f32, v2f32, fsub, 0>;
2275 def VSUBfq : N3VQ<0, 0, 0b10, 0b1101, 0, IIC_VBINQ, "vsub", "f32",
2276 v4f32, v4f32, fsub, 0>;
2277 // VSUBL : Vector Subtract Long (Q = D - D)
2278 defm VSUBLs : N3VLInt_QHS<0,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
2279 "vsubl", "s", int_arm_neon_vsubls, 1>;
2280 defm VSUBLu : N3VLInt_QHS<1,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
2281 "vsubl", "u", int_arm_neon_vsublu, 1>;
2282 // VSUBW : Vector Subtract Wide (Q = Q - D)
2283 defm VSUBWs : N3VWInt_QHS<0,1,0b0011,0, "vsubw", "s", int_arm_neon_vsubws, 0>;
2284 defm VSUBWu : N3VWInt_QHS<1,1,0b0011,0, "vsubw", "u", int_arm_neon_vsubwu, 0>;
2285 // VHSUB : Vector Halving Subtract
2286 defm VHSUBs : N3VInt_QHS<0, 0, 0b0010, 0, N3RegFrm,
2287 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2288 "vhsub", "s", int_arm_neon_vhsubs, 0>;
2289 defm VHSUBu : N3VInt_QHS<1, 0, 0b0010, 0, N3RegFrm,
2290 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2291 "vhsub", "u", int_arm_neon_vhsubu, 0>;
2292 // VQSUB : Vector Saturing Subtract
2293 defm VQSUBs : N3VInt_QHSD<0, 0, 0b0010, 1, N3RegFrm,
2294 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2295 "vqsub", "s", int_arm_neon_vqsubs, 0>;
2296 defm VQSUBu : N3VInt_QHSD<1, 0, 0b0010, 1, N3RegFrm,
2297 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2298 "vqsub", "u", int_arm_neon_vqsubu, 0>;
2299 // VSUBHN : Vector Subtract and Narrow Returning High Half (D = Q - Q)
2300 defm VSUBHN : N3VNInt_HSD<0,1,0b0110,0, "vsubhn", "i",
2301 int_arm_neon_vsubhn, 0>;
2302 // VRSUBHN : Vector Rounding Subtract and Narrow Returning High Half (D=Q-Q)
2303 defm VRSUBHN : N3VNInt_HSD<1,1,0b0110,0, "vrsubhn", "i",
2304 int_arm_neon_vrsubhn, 0>;
2306 // Vector Comparisons.
2308 // VCEQ : Vector Compare Equal
2309 defm VCEQ : N3V_QHS<1, 0, 0b1000, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
2310 IIC_VSUBi4Q, "vceq", "i", NEONvceq, 1>;
2311 def VCEQfd : N3VD<0,0,0b00,0b1110,0, IIC_VBIND, "vceq", "f32", v2i32, v2f32,
2313 def VCEQfq : N3VQ<0,0,0b00,0b1110,0, IIC_VBINQ, "vceq", "f32", v4i32, v4f32,
2315 // For disassembly only.
2316 defm VCEQz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00010, 0, "vceq", "i",
2319 // VCGE : Vector Compare Greater Than or Equal
2320 defm VCGEs : N3V_QHS<0, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
2321 IIC_VSUBi4Q, "vcge", "s", NEONvcge, 0>;
2322 defm VCGEu : N3V_QHS<1, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
2323 IIC_VSUBi4Q, "vcge", "u", NEONvcgeu, 0>;
2324 def VCGEfd : N3VD<1,0,0b00,0b1110,0, IIC_VBIND, "vcge", "f32", v2i32, v2f32,
2326 def VCGEfq : N3VQ<1,0,0b00,0b1110,0, IIC_VBINQ, "vcge", "f32", v4i32, v4f32,
2328 // For disassembly only.
2329 defm VCGEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00001, 0, "vcge", "s",
2331 // For disassembly only.
2332 defm VCLEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00011, 0, "vcle", "s",
2335 // VCGT : Vector Compare Greater Than
2336 defm VCGTs : N3V_QHS<0, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
2337 IIC_VSUBi4Q, "vcgt", "s", NEONvcgt, 0>;
2338 defm VCGTu : N3V_QHS<1, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
2339 IIC_VSUBi4Q, "vcgt", "u", NEONvcgtu, 0>;
2340 def VCGTfd : N3VD<1,0,0b10,0b1110,0, IIC_VBIND, "vcgt", "f32", v2i32, v2f32,
2342 def VCGTfq : N3VQ<1,0,0b10,0b1110,0, IIC_VBINQ, "vcgt", "f32", v4i32, v4f32,
2344 // For disassembly only.
2345 defm VCGTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00000, 0, "vcgt", "s",
2347 // For disassembly only.
2348 defm VCLTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00100, 0, "vclt", "s",
2351 // VACGE : Vector Absolute Compare Greater Than or Equal (aka VCAGE)
2352 def VACGEd : N3VDInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacge",
2353 "f32", v2i32, v2f32, int_arm_neon_vacged, 0>;
2354 def VACGEq : N3VQInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacge",
2355 "f32", v4i32, v4f32, int_arm_neon_vacgeq, 0>;
2356 // VACGT : Vector Absolute Compare Greater Than (aka VCAGT)
2357 def VACGTd : N3VDInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacgt",
2358 "f32", v2i32, v2f32, int_arm_neon_vacgtd, 0>;
2359 def VACGTq : N3VQInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacgt",
2360 "f32", v4i32, v4f32, int_arm_neon_vacgtq, 0>;
2361 // VTST : Vector Test Bits
2362 defm VTST : N3V_QHS<0, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2363 IIC_VBINi4Q, "vtst", "", NEONvtst, 1>;
2365 // Vector Bitwise Operations.
2367 def vnotd : PatFrag<(ops node:$in),
2368 (xor node:$in, (bitconvert (v8i8 NEONimmAllOnesV)))>;
2369 def vnotq : PatFrag<(ops node:$in),
2370 (xor node:$in, (bitconvert (v16i8 NEONimmAllOnesV)))>;
2373 // VAND : Vector Bitwise AND
2374 def VANDd : N3VDX<0, 0, 0b00, 0b0001, 1, IIC_VBINiD, "vand",
2375 v2i32, v2i32, and, 1>;
2376 def VANDq : N3VQX<0, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "vand",
2377 v4i32, v4i32, and, 1>;
2379 // VEOR : Vector Bitwise Exclusive OR
2380 def VEORd : N3VDX<1, 0, 0b00, 0b0001, 1, IIC_VBINiD, "veor",
2381 v2i32, v2i32, xor, 1>;
2382 def VEORq : N3VQX<1, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "veor",
2383 v4i32, v4i32, xor, 1>;
2385 // VORR : Vector Bitwise OR
2386 def VORRd : N3VDX<0, 0, 0b10, 0b0001, 1, IIC_VBINiD, "vorr",
2387 v2i32, v2i32, or, 1>;
2388 def VORRq : N3VQX<0, 0, 0b10, 0b0001, 1, IIC_VBINiQ, "vorr",
2389 v4i32, v4i32, or, 1>;
2391 // VBIC : Vector Bitwise Bit Clear (AND NOT)
2392 def VBICd : N3VX<0, 0, 0b01, 0b0001, 0, 1, (outs DPR:$dst),
2393 (ins DPR:$src1, DPR:$src2), N3RegFrm, IIC_VBINiD,
2394 "vbic", "$dst, $src1, $src2", "",
2395 [(set DPR:$dst, (v2i32 (and DPR:$src1,
2396 (vnotd DPR:$src2))))]>;
2397 def VBICq : N3VX<0, 0, 0b01, 0b0001, 1, 1, (outs QPR:$dst),
2398 (ins QPR:$src1, QPR:$src2), N3RegFrm, IIC_VBINiQ,
2399 "vbic", "$dst, $src1, $src2", "",
2400 [(set QPR:$dst, (v4i32 (and QPR:$src1,
2401 (vnotq QPR:$src2))))]>;
2403 // VORN : Vector Bitwise OR NOT
2404 def VORNd : N3VX<0, 0, 0b11, 0b0001, 0, 1, (outs DPR:$dst),
2405 (ins DPR:$src1, DPR:$src2), N3RegFrm, IIC_VBINiD,
2406 "vorn", "$dst, $src1, $src2", "",
2407 [(set DPR:$dst, (v2i32 (or DPR:$src1,
2408 (vnotd DPR:$src2))))]>;
2409 def VORNq : N3VX<0, 0, 0b11, 0b0001, 1, 1, (outs QPR:$dst),
2410 (ins QPR:$src1, QPR:$src2), N3RegFrm, IIC_VBINiQ,
2411 "vorn", "$dst, $src1, $src2", "",
2412 [(set QPR:$dst, (v4i32 (or QPR:$src1,
2413 (vnotq QPR:$src2))))]>;
2415 // VMVN : Vector Bitwise NOT (Immediate)
2417 let isReMaterializable = 1 in {
2418 def VMVNv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 1, 1, (outs DPR:$dst),
2419 (ins nModImm:$SIMM), IIC_VMOVImm,
2420 "vmvn", "i16", "$dst, $SIMM", "",
2421 [(set DPR:$dst, (v4i16 (NEONvmvnImm timm:$SIMM)))]>;
2422 def VMVNv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 1, 1, (outs QPR:$dst),
2423 (ins nModImm:$SIMM), IIC_VMOVImm,
2424 "vmvn", "i16", "$dst, $SIMM", "",
2425 [(set QPR:$dst, (v8i16 (NEONvmvnImm timm:$SIMM)))]>;
2427 def VMVNv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 1, 1, (outs DPR:$dst),
2428 (ins nModImm:$SIMM), IIC_VMOVImm,
2429 "vmvn", "i32", "$dst, $SIMM", "",
2430 [(set DPR:$dst, (v2i32 (NEONvmvnImm timm:$SIMM)))]>;
2431 def VMVNv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 1, 1, (outs QPR:$dst),
2432 (ins nModImm:$SIMM), IIC_VMOVImm,
2433 "vmvn", "i32", "$dst, $SIMM", "",
2434 [(set QPR:$dst, (v4i32 (NEONvmvnImm timm:$SIMM)))]>;
2437 // VMVN : Vector Bitwise NOT
2438 def VMVNd : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 0, 0,
2439 (outs DPR:$dst), (ins DPR:$src), IIC_VSUBiD,
2440 "vmvn", "$dst, $src", "",
2441 [(set DPR:$dst, (v2i32 (vnotd DPR:$src)))]>;
2442 def VMVNq : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 1, 0,
2443 (outs QPR:$dst), (ins QPR:$src), IIC_VSUBiD,
2444 "vmvn", "$dst, $src", "",
2445 [(set QPR:$dst, (v4i32 (vnotq QPR:$src)))]>;
2446 def : Pat<(v2i32 (vnotd DPR:$src)), (VMVNd DPR:$src)>;
2447 def : Pat<(v4i32 (vnotq QPR:$src)), (VMVNq QPR:$src)>;
2449 // VBSL : Vector Bitwise Select
2450 def VBSLd : N3VX<1, 0, 0b01, 0b0001, 0, 1, (outs DPR:$dst),
2451 (ins DPR:$src1, DPR:$src2, DPR:$src3),
2452 N3RegFrm, IIC_VCNTiD,
2453 "vbsl", "$dst, $src2, $src3", "$src1 = $dst",
2455 (v2i32 (or (and DPR:$src2, DPR:$src1),
2456 (and DPR:$src3, (vnotd DPR:$src1)))))]>;
2457 def VBSLq : N3VX<1, 0, 0b01, 0b0001, 1, 1, (outs QPR:$dst),
2458 (ins QPR:$src1, QPR:$src2, QPR:$src3),
2459 N3RegFrm, IIC_VCNTiQ,
2460 "vbsl", "$dst, $src2, $src3", "$src1 = $dst",
2462 (v4i32 (or (and QPR:$src2, QPR:$src1),
2463 (and QPR:$src3, (vnotq QPR:$src1)))))]>;
2465 // VBIF : Vector Bitwise Insert if False
2466 // like VBSL but with: "vbif $dst, $src3, $src1", "$src2 = $dst",
2467 def VBIFd : N3VX<1, 0, 0b11, 0b0001, 0, 1,
2468 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3),
2469 N3RegFrm, IIC_VBINiD,
2470 "vbif", "$dst, $src2, $src3", "$src1 = $dst",
2471 [/* For disassembly only; pattern left blank */]>;
2472 def VBIFq : N3VX<1, 0, 0b11, 0b0001, 1, 1,
2473 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3),
2474 N3RegFrm, IIC_VBINiQ,
2475 "vbif", "$dst, $src2, $src3", "$src1 = $dst",
2476 [/* For disassembly only; pattern left blank */]>;
2478 // VBIT : Vector Bitwise Insert if True
2479 // like VBSL but with: "vbit $dst, $src2, $src1", "$src3 = $dst",
2480 def VBITd : N3VX<1, 0, 0b10, 0b0001, 0, 1,
2481 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3),
2482 N3RegFrm, IIC_VBINiD,
2483 "vbit", "$dst, $src2, $src3", "$src1 = $dst",
2484 [/* For disassembly only; pattern left blank */]>;
2485 def VBITq : N3VX<1, 0, 0b10, 0b0001, 1, 1,
2486 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3),
2487 N3RegFrm, IIC_VBINiQ,
2488 "vbit", "$dst, $src2, $src3", "$src1 = $dst",
2489 [/* For disassembly only; pattern left blank */]>;
2491 // VBIT/VBIF are not yet implemented. The TwoAddress pass will not go looking
2492 // for equivalent operations with different register constraints; it just
2495 // Vector Absolute Differences.
2497 // VABD : Vector Absolute Difference
2498 defm VABDs : N3VInt_QHS<0, 0, 0b0111, 0, N3RegFrm,
2499 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2500 "vabd", "s", int_arm_neon_vabds, 0>;
2501 defm VABDu : N3VInt_QHS<1, 0, 0b0111, 0, N3RegFrm,
2502 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2503 "vabd", "u", int_arm_neon_vabdu, 0>;
2504 def VABDfd : N3VDInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBIND,
2505 "vabd", "f32", v2f32, v2f32, int_arm_neon_vabds, 0>;
2506 def VABDfq : N3VQInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBINQ,
2507 "vabd", "f32", v4f32, v4f32, int_arm_neon_vabds, 0>;
2509 // VABDL : Vector Absolute Difference Long (Q = | D - D |)
2510 defm VABDLs : N3VLInt_QHS<0,1,0b0111,0, IIC_VSUBi4Q, IIC_VSUBi4Q,
2511 "vabdl", "s", int_arm_neon_vabdls, 0>;
2512 defm VABDLu : N3VLInt_QHS<1,1,0b0111,0, IIC_VSUBi4Q, IIC_VSUBi4Q,
2513 "vabdl", "u", int_arm_neon_vabdlu, 0>;
2515 // VABA : Vector Absolute Difference and Accumulate
2516 defm VABAs : N3VInt3_QHS<0,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
2517 "vaba", "s", int_arm_neon_vabas>;
2518 defm VABAu : N3VInt3_QHS<1,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
2519 "vaba", "u", int_arm_neon_vabau>;
2521 // VABAL : Vector Absolute Difference and Accumulate Long (Q += | D - D |)
2522 defm VABALs : N3VLInt3_QHS<0,1,0b0101,0, IIC_VABAD, IIC_VABAD,
2523 "vabal", "s", int_arm_neon_vabals>;
2524 defm VABALu : N3VLInt3_QHS<1,1,0b0101,0, IIC_VABAD, IIC_VABAD,
2525 "vabal", "u", int_arm_neon_vabalu>;
2527 // Vector Maximum and Minimum.
2529 // VMAX : Vector Maximum
2530 defm VMAXs : N3VInt_QHS<0, 0, 0b0110, 0, N3RegFrm,
2531 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2532 "vmax", "s", int_arm_neon_vmaxs, 1>;
2533 defm VMAXu : N3VInt_QHS<1, 0, 0b0110, 0, N3RegFrm,
2534 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2535 "vmax", "u", int_arm_neon_vmaxu, 1>;
2536 def VMAXfd : N3VDInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBIND,
2538 v2f32, v2f32, int_arm_neon_vmaxs, 1>;
2539 def VMAXfq : N3VQInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBINQ,
2541 v4f32, v4f32, int_arm_neon_vmaxs, 1>;
2543 // VMIN : Vector Minimum
2544 defm VMINs : N3VInt_QHS<0, 0, 0b0110, 1, N3RegFrm,
2545 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2546 "vmin", "s", int_arm_neon_vmins, 1>;
2547 defm VMINu : N3VInt_QHS<1, 0, 0b0110, 1, N3RegFrm,
2548 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2549 "vmin", "u", int_arm_neon_vminu, 1>;
2550 def VMINfd : N3VDInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBIND,
2552 v2f32, v2f32, int_arm_neon_vmins, 1>;
2553 def VMINfq : N3VQInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBINQ,
2555 v4f32, v4f32, int_arm_neon_vmins, 1>;
2557 // Vector Pairwise Operations.
2559 // VPADD : Vector Pairwise Add
2560 def VPADDi8 : N3VDInt<0, 0, 0b00, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
2562 v8i8, v8i8, int_arm_neon_vpadd, 0>;
2563 def VPADDi16 : N3VDInt<0, 0, 0b01, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
2565 v4i16, v4i16, int_arm_neon_vpadd, 0>;
2566 def VPADDi32 : N3VDInt<0, 0, 0b10, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
2568 v2i32, v2i32, int_arm_neon_vpadd, 0>;
2569 def VPADDf : N3VDInt<1, 0, 0b00, 0b1101, 0, N3RegFrm,
2570 IIC_VBIND, "vpadd", "f32",
2571 v2f32, v2f32, int_arm_neon_vpadd, 0>;
2573 // VPADDL : Vector Pairwise Add Long
2574 defm VPADDLs : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00100, 0, "vpaddl", "s",
2575 int_arm_neon_vpaddls>;
2576 defm VPADDLu : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00101, 0, "vpaddl", "u",
2577 int_arm_neon_vpaddlu>;
2579 // VPADAL : Vector Pairwise Add and Accumulate Long
2580 defm VPADALs : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01100, 0, "vpadal", "s",
2581 int_arm_neon_vpadals>;
2582 defm VPADALu : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01101, 0, "vpadal", "u",
2583 int_arm_neon_vpadalu>;
2585 // VPMAX : Vector Pairwise Maximum
2586 def VPMAXs8 : N3VDInt<0, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
2587 "s8", v8i8, v8i8, int_arm_neon_vpmaxs, 0>;
2588 def VPMAXs16 : N3VDInt<0, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
2589 "s16", v4i16, v4i16, int_arm_neon_vpmaxs, 0>;
2590 def VPMAXs32 : N3VDInt<0, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
2591 "s32", v2i32, v2i32, int_arm_neon_vpmaxs, 0>;
2592 def VPMAXu8 : N3VDInt<1, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
2593 "u8", v8i8, v8i8, int_arm_neon_vpmaxu, 0>;
2594 def VPMAXu16 : N3VDInt<1, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
2595 "u16", v4i16, v4i16, int_arm_neon_vpmaxu, 0>;
2596 def VPMAXu32 : N3VDInt<1, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
2597 "u32", v2i32, v2i32, int_arm_neon_vpmaxu, 0>;
2598 def VPMAXf : N3VDInt<1, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
2599 "f32", v2f32, v2f32, int_arm_neon_vpmaxs, 0>;
2601 // VPMIN : Vector Pairwise Minimum
2602 def VPMINs8 : N3VDInt<0, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
2603 "s8", v8i8, v8i8, int_arm_neon_vpmins, 0>;
2604 def VPMINs16 : N3VDInt<0, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
2605 "s16", v4i16, v4i16, int_arm_neon_vpmins, 0>;
2606 def VPMINs32 : N3VDInt<0, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
2607 "s32", v2i32, v2i32, int_arm_neon_vpmins, 0>;
2608 def VPMINu8 : N3VDInt<1, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
2609 "u8", v8i8, v8i8, int_arm_neon_vpminu, 0>;
2610 def VPMINu16 : N3VDInt<1, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
2611 "u16", v4i16, v4i16, int_arm_neon_vpminu, 0>;
2612 def VPMINu32 : N3VDInt<1, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
2613 "u32", v2i32, v2i32, int_arm_neon_vpminu, 0>;
2614 def VPMINf : N3VDInt<1, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VSUBi4D, "vpmin",
2615 "f32", v2f32, v2f32, int_arm_neon_vpmins, 0>;
2617 // Vector Reciprocal and Reciprocal Square Root Estimate and Step.
2619 // VRECPE : Vector Reciprocal Estimate
2620 def VRECPEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
2621 IIC_VUNAD, "vrecpe", "u32",
2622 v2i32, v2i32, int_arm_neon_vrecpe>;
2623 def VRECPEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
2624 IIC_VUNAQ, "vrecpe", "u32",
2625 v4i32, v4i32, int_arm_neon_vrecpe>;
2626 def VRECPEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
2627 IIC_VUNAD, "vrecpe", "f32",
2628 v2f32, v2f32, int_arm_neon_vrecpe>;
2629 def VRECPEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
2630 IIC_VUNAQ, "vrecpe", "f32",
2631 v4f32, v4f32, int_arm_neon_vrecpe>;
2633 // VRECPS : Vector Reciprocal Step
2634 def VRECPSfd : N3VDInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
2635 IIC_VRECSD, "vrecps", "f32",
2636 v2f32, v2f32, int_arm_neon_vrecps, 1>;
2637 def VRECPSfq : N3VQInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
2638 IIC_VRECSQ, "vrecps", "f32",
2639 v4f32, v4f32, int_arm_neon_vrecps, 1>;
2641 // VRSQRTE : Vector Reciprocal Square Root Estimate
2642 def VRSQRTEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
2643 IIC_VUNAD, "vrsqrte", "u32",
2644 v2i32, v2i32, int_arm_neon_vrsqrte>;
2645 def VRSQRTEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
2646 IIC_VUNAQ, "vrsqrte", "u32",
2647 v4i32, v4i32, int_arm_neon_vrsqrte>;
2648 def VRSQRTEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
2649 IIC_VUNAD, "vrsqrte", "f32",
2650 v2f32, v2f32, int_arm_neon_vrsqrte>;
2651 def VRSQRTEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
2652 IIC_VUNAQ, "vrsqrte", "f32",
2653 v4f32, v4f32, int_arm_neon_vrsqrte>;
2655 // VRSQRTS : Vector Reciprocal Square Root Step
2656 def VRSQRTSfd : N3VDInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
2657 IIC_VRECSD, "vrsqrts", "f32",
2658 v2f32, v2f32, int_arm_neon_vrsqrts, 1>;
2659 def VRSQRTSfq : N3VQInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
2660 IIC_VRECSQ, "vrsqrts", "f32",
2661 v4f32, v4f32, int_arm_neon_vrsqrts, 1>;
2665 // VSHL : Vector Shift
2666 defm VSHLs : N3VInt_QHSD<0, 0, 0b0100, 0, N3RegVShFrm,
2667 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
2668 "vshl", "s", int_arm_neon_vshifts, 0>;
2669 defm VSHLu : N3VInt_QHSD<1, 0, 0b0100, 0, N3RegVShFrm,
2670 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
2671 "vshl", "u", int_arm_neon_vshiftu, 0>;
2672 // VSHL : Vector Shift Left (Immediate)
2673 defm VSHLi : N2VSh_QHSD<0, 1, 0b0101, 1, IIC_VSHLiD, "vshl", "i", NEONvshl,
2675 // VSHR : Vector Shift Right (Immediate)
2676 defm VSHRs : N2VSh_QHSD<0, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "s", NEONvshrs,
2678 defm VSHRu : N2VSh_QHSD<1, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "u", NEONvshru,
2681 // VSHLL : Vector Shift Left Long
2682 defm VSHLLs : N2VLSh_QHS<0, 1, 0b1010, 0, 0, 1, "vshll", "s", NEONvshlls>;
2683 defm VSHLLu : N2VLSh_QHS<1, 1, 0b1010, 0, 0, 1, "vshll", "u", NEONvshllu>;
2685 // VSHLL : Vector Shift Left Long (with maximum shift count)
2686 class N2VLShMax<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
2687 bit op6, bit op4, string OpcodeStr, string Dt, ValueType ResTy,
2688 ValueType OpTy, SDNode OpNode>
2689 : N2VLSh<op24, op23, op11_8, op7, op6, op4, OpcodeStr, Dt,
2690 ResTy, OpTy, OpNode> {
2691 let Inst{21-16} = op21_16;
2693 def VSHLLi8 : N2VLShMax<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll", "i8",
2694 v8i16, v8i8, NEONvshlli>;
2695 def VSHLLi16 : N2VLShMax<1, 1, 0b110110, 0b0011, 0, 0, 0, "vshll", "i16",
2696 v4i32, v4i16, NEONvshlli>;
2697 def VSHLLi32 : N2VLShMax<1, 1, 0b111010, 0b0011, 0, 0, 0, "vshll", "i32",
2698 v2i64, v2i32, NEONvshlli>;
2700 // VSHRN : Vector Shift Right and Narrow
2701 defm VSHRN : N2VNSh_HSD<0,1,0b1000,0,0,1, IIC_VSHLiD, "vshrn", "i",
2704 // VRSHL : Vector Rounding Shift
2705 defm VRSHLs : N3VInt_QHSD<0, 0, 0b0101, 0, N3RegVShFrm,
2706 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
2707 "vrshl", "s", int_arm_neon_vrshifts, 0>;
2708 defm VRSHLu : N3VInt_QHSD<1, 0, 0b0101, 0, N3RegVShFrm,
2709 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
2710 "vrshl", "u", int_arm_neon_vrshiftu, 0>;
2711 // VRSHR : Vector Rounding Shift Right
2712 defm VRSHRs : N2VSh_QHSD<0,1,0b0010,1, IIC_VSHLi4D, "vrshr", "s", NEONvrshrs,
2714 defm VRSHRu : N2VSh_QHSD<1,1,0b0010,1, IIC_VSHLi4D, "vrshr", "u", NEONvrshru,
2717 // VRSHRN : Vector Rounding Shift Right and Narrow
2718 defm VRSHRN : N2VNSh_HSD<0, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vrshrn", "i",
2721 // VQSHL : Vector Saturating Shift
2722 defm VQSHLs : N3VInt_QHSD<0, 0, 0b0100, 1, N3RegVShFrm,
2723 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
2724 "vqshl", "s", int_arm_neon_vqshifts, 0>;
2725 defm VQSHLu : N3VInt_QHSD<1, 0, 0b0100, 1, N3RegVShFrm,
2726 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
2727 "vqshl", "u", int_arm_neon_vqshiftu, 0>;
2728 // VQSHL : Vector Saturating Shift Left (Immediate)
2729 defm VQSHLsi : N2VSh_QHSD<0,1,0b0111,1, IIC_VSHLi4D, "vqshl", "s",NEONvqshls,
2731 defm VQSHLui : N2VSh_QHSD<1,1,0b0111,1, IIC_VSHLi4D, "vqshl", "u",NEONvqshlu,
2733 // VQSHLU : Vector Saturating Shift Left (Immediate, Unsigned)
2734 defm VQSHLsu : N2VSh_QHSD<1,1,0b0110,1, IIC_VSHLi4D,"vqshlu","s",NEONvqshlsu,
2737 // VQSHRN : Vector Saturating Shift Right and Narrow
2738 defm VQSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "s",
2740 defm VQSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "u",
2743 // VQSHRUN : Vector Saturating Shift Right and Narrow (Unsigned)
2744 defm VQSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 0, 1, IIC_VSHLi4D, "vqshrun", "s",
2747 // VQRSHL : Vector Saturating Rounding Shift
2748 defm VQRSHLs : N3VInt_QHSD<0, 0, 0b0101, 1, N3RegVShFrm,
2749 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
2750 "vqrshl", "s", int_arm_neon_vqrshifts, 0>;
2751 defm VQRSHLu : N3VInt_QHSD<1, 0, 0b0101, 1, N3RegVShFrm,
2752 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
2753 "vqrshl", "u", int_arm_neon_vqrshiftu, 0>;
2755 // VQRSHRN : Vector Saturating Rounding Shift Right and Narrow
2756 defm VQRSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "s",
2758 defm VQRSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "u",
2761 // VQRSHRUN : Vector Saturating Rounding Shift Right and Narrow (Unsigned)
2762 defm VQRSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vqrshrun", "s",
2765 // VSRA : Vector Shift Right and Accumulate
2766 defm VSRAs : N2VShAdd_QHSD<0, 1, 0b0001, 1, "vsra", "s", NEONvshrs>;
2767 defm VSRAu : N2VShAdd_QHSD<1, 1, 0b0001, 1, "vsra", "u", NEONvshru>;
2768 // VRSRA : Vector Rounding Shift Right and Accumulate
2769 defm VRSRAs : N2VShAdd_QHSD<0, 1, 0b0011, 1, "vrsra", "s", NEONvrshrs>;
2770 defm VRSRAu : N2VShAdd_QHSD<1, 1, 0b0011, 1, "vrsra", "u", NEONvrshru>;
2772 // VSLI : Vector Shift Left and Insert
2773 defm VSLI : N2VShIns_QHSD<1, 1, 0b0101, 1, "vsli", NEONvsli, N2RegVShLFrm>;
2774 // VSRI : Vector Shift Right and Insert
2775 defm VSRI : N2VShIns_QHSD<1, 1, 0b0100, 1, "vsri", NEONvsri, N2RegVShRFrm>;
2777 // Vector Absolute and Saturating Absolute.
2779 // VABS : Vector Absolute Value
2780 defm VABS : N2VInt_QHS<0b11, 0b11, 0b01, 0b00110, 0,
2781 IIC_VUNAiD, IIC_VUNAiQ, "vabs", "s",
2783 def VABSfd : N2VDInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
2784 IIC_VUNAD, "vabs", "f32",
2785 v2f32, v2f32, int_arm_neon_vabs>;
2786 def VABSfq : N2VQInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
2787 IIC_VUNAQ, "vabs", "f32",
2788 v4f32, v4f32, int_arm_neon_vabs>;
2790 // VQABS : Vector Saturating Absolute Value
2791 defm VQABS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01110, 0,
2792 IIC_VQUNAiD, IIC_VQUNAiQ, "vqabs", "s",
2793 int_arm_neon_vqabs>;
2797 def vnegd : PatFrag<(ops node:$in),
2798 (sub (bitconvert (v2i32 NEONimmAllZerosV)), node:$in)>;
2799 def vnegq : PatFrag<(ops node:$in),
2800 (sub (bitconvert (v4i32 NEONimmAllZerosV)), node:$in)>;
2802 class VNEGD<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
2803 : N2V<0b11, 0b11, size, 0b01, 0b00111, 0, 0, (outs DPR:$dst), (ins DPR:$src),
2804 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
2805 [(set DPR:$dst, (Ty (vnegd DPR:$src)))]>;
2806 class VNEGQ<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
2807 : N2V<0b11, 0b11, size, 0b01, 0b00111, 1, 0, (outs QPR:$dst), (ins QPR:$src),
2808 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
2809 [(set QPR:$dst, (Ty (vnegq QPR:$src)))]>;
2811 // VNEG : Vector Negate (integer)
2812 def VNEGs8d : VNEGD<0b00, "vneg", "s8", v8i8>;
2813 def VNEGs16d : VNEGD<0b01, "vneg", "s16", v4i16>;
2814 def VNEGs32d : VNEGD<0b10, "vneg", "s32", v2i32>;
2815 def VNEGs8q : VNEGQ<0b00, "vneg", "s8", v16i8>;
2816 def VNEGs16q : VNEGQ<0b01, "vneg", "s16", v8i16>;
2817 def VNEGs32q : VNEGQ<0b10, "vneg", "s32", v4i32>;
2819 // VNEG : Vector Negate (floating-point)
2820 def VNEGfd : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
2821 (outs DPR:$dst), (ins DPR:$src), IIC_VUNAD,
2822 "vneg", "f32", "$dst, $src", "",
2823 [(set DPR:$dst, (v2f32 (fneg DPR:$src)))]>;
2824 def VNEGf32q : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 1, 0,
2825 (outs QPR:$dst), (ins QPR:$src), IIC_VUNAQ,
2826 "vneg", "f32", "$dst, $src", "",
2827 [(set QPR:$dst, (v4f32 (fneg QPR:$src)))]>;
2829 def : Pat<(v8i8 (vnegd DPR:$src)), (VNEGs8d DPR:$src)>;
2830 def : Pat<(v4i16 (vnegd DPR:$src)), (VNEGs16d DPR:$src)>;
2831 def : Pat<(v2i32 (vnegd DPR:$src)), (VNEGs32d DPR:$src)>;
2832 def : Pat<(v16i8 (vnegq QPR:$src)), (VNEGs8q QPR:$src)>;
2833 def : Pat<(v8i16 (vnegq QPR:$src)), (VNEGs16q QPR:$src)>;
2834 def : Pat<(v4i32 (vnegq QPR:$src)), (VNEGs32q QPR:$src)>;
2836 // VQNEG : Vector Saturating Negate
2837 defm VQNEG : N2VInt_QHS<0b11, 0b11, 0b00, 0b01111, 0,
2838 IIC_VQUNAiD, IIC_VQUNAiQ, "vqneg", "s",
2839 int_arm_neon_vqneg>;
2841 // Vector Bit Counting Operations.
2843 // VCLS : Vector Count Leading Sign Bits
2844 defm VCLS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01000, 0,
2845 IIC_VCNTiD, IIC_VCNTiQ, "vcls", "s",
2847 // VCLZ : Vector Count Leading Zeros
2848 defm VCLZ : N2VInt_QHS<0b11, 0b11, 0b00, 0b01001, 0,
2849 IIC_VCNTiD, IIC_VCNTiQ, "vclz", "i",
2851 // VCNT : Vector Count One Bits
2852 def VCNTd : N2VDInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
2853 IIC_VCNTiD, "vcnt", "8",
2854 v8i8, v8i8, int_arm_neon_vcnt>;
2855 def VCNTq : N2VQInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
2856 IIC_VCNTiQ, "vcnt", "8",
2857 v16i8, v16i8, int_arm_neon_vcnt>;
2859 // Vector Swap -- for disassembly only.
2860 def VSWPd : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 0, 0,
2861 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
2862 "vswp", "$dst, $src", "", []>;
2863 def VSWPq : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 1, 0,
2864 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
2865 "vswp", "$dst, $src", "", []>;
2867 // Vector Move Operations.
2869 // VMOV : Vector Move (Register)
2871 let neverHasSideEffects = 1 in {
2872 def VMOVDneon: N3VX<0, 0, 0b10, 0b0001, 0, 1, (outs DPR:$dst), (ins DPR:$src),
2873 N3RegFrm, IIC_VMOVD, "vmov", "$dst, $src", "", []>;
2874 def VMOVQ : N3VX<0, 0, 0b10, 0b0001, 1, 1, (outs QPR:$dst), (ins QPR:$src),
2875 N3RegFrm, IIC_VMOVD, "vmov", "$dst, $src", "", []>;
2877 // Pseudo vector move instructions for QQ and QQQQ registers. This should
2878 // be expanded after register allocation is completed.
2879 def VMOVQQ : PseudoInst<(outs QQPR:$dst), (ins QQPR:$src),
2880 NoItinerary, "${:comment} vmov\t$dst, $src", []>;
2882 def VMOVQQQQ : PseudoInst<(outs QQQQPR:$dst), (ins QQQQPR:$src),
2883 NoItinerary, "${:comment} vmov\t$dst, $src", []>;
2884 } // neverHasSideEffects
2886 // VMOV : Vector Move (Immediate)
2888 let isReMaterializable = 1 in {
2889 def VMOVv8i8 : N1ModImm<1, 0b000, 0b1110, 0, 0, 0, 1, (outs DPR:$dst),
2890 (ins nModImm:$SIMM), IIC_VMOVImm,
2891 "vmov", "i8", "$dst, $SIMM", "",
2892 [(set DPR:$dst, (v8i8 (NEONvmovImm timm:$SIMM)))]>;
2893 def VMOVv16i8 : N1ModImm<1, 0b000, 0b1110, 0, 1, 0, 1, (outs QPR:$dst),
2894 (ins nModImm:$SIMM), IIC_VMOVImm,
2895 "vmov", "i8", "$dst, $SIMM", "",
2896 [(set QPR:$dst, (v16i8 (NEONvmovImm timm:$SIMM)))]>;
2898 def VMOVv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 0, 1, (outs DPR:$dst),
2899 (ins nModImm:$SIMM), IIC_VMOVImm,
2900 "vmov", "i16", "$dst, $SIMM", "",
2901 [(set DPR:$dst, (v4i16 (NEONvmovImm timm:$SIMM)))]>;
2902 def VMOVv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 0, 1, (outs QPR:$dst),
2903 (ins nModImm:$SIMM), IIC_VMOVImm,
2904 "vmov", "i16", "$dst, $SIMM", "",
2905 [(set QPR:$dst, (v8i16 (NEONvmovImm timm:$SIMM)))]>;
2907 def VMOVv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 0, 1, (outs DPR:$dst),
2908 (ins nModImm:$SIMM), IIC_VMOVImm,
2909 "vmov", "i32", "$dst, $SIMM", "",
2910 [(set DPR:$dst, (v2i32 (NEONvmovImm timm:$SIMM)))]>;
2911 def VMOVv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 0, 1, (outs QPR:$dst),
2912 (ins nModImm:$SIMM), IIC_VMOVImm,
2913 "vmov", "i32", "$dst, $SIMM", "",
2914 [(set QPR:$dst, (v4i32 (NEONvmovImm timm:$SIMM)))]>;
2916 def VMOVv1i64 : N1ModImm<1, 0b000, 0b1110, 0, 0, 1, 1, (outs DPR:$dst),
2917 (ins nModImm:$SIMM), IIC_VMOVImm,
2918 "vmov", "i64", "$dst, $SIMM", "",
2919 [(set DPR:$dst, (v1i64 (NEONvmovImm timm:$SIMM)))]>;
2920 def VMOVv2i64 : N1ModImm<1, 0b000, 0b1110, 0, 1, 1, 1, (outs QPR:$dst),
2921 (ins nModImm:$SIMM), IIC_VMOVImm,
2922 "vmov", "i64", "$dst, $SIMM", "",
2923 [(set QPR:$dst, (v2i64 (NEONvmovImm timm:$SIMM)))]>;
2924 } // isReMaterializable
2926 // VMOV : Vector Get Lane (move scalar to ARM core register)
2928 def VGETLNs8 : NVGetLane<{1,1,1,0,0,1,?,1}, 0b1011, {?,?},
2929 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2930 IIC_VMOVSI, "vmov", "s8", "$dst, $src[$lane]",
2931 [(set GPR:$dst, (NEONvgetlanes (v8i8 DPR:$src),
2933 def VGETLNs16 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, {?,1},
2934 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2935 IIC_VMOVSI, "vmov", "s16", "$dst, $src[$lane]",
2936 [(set GPR:$dst, (NEONvgetlanes (v4i16 DPR:$src),
2938 def VGETLNu8 : NVGetLane<{1,1,1,0,1,1,?,1}, 0b1011, {?,?},
2939 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2940 IIC_VMOVSI, "vmov", "u8", "$dst, $src[$lane]",
2941 [(set GPR:$dst, (NEONvgetlaneu (v8i8 DPR:$src),
2943 def VGETLNu16 : NVGetLane<{1,1,1,0,1,0,?,1}, 0b1011, {?,1},
2944 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2945 IIC_VMOVSI, "vmov", "u16", "$dst, $src[$lane]",
2946 [(set GPR:$dst, (NEONvgetlaneu (v4i16 DPR:$src),
2948 def VGETLNi32 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, 0b00,
2949 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2950 IIC_VMOVSI, "vmov", "32", "$dst, $src[$lane]",
2951 [(set GPR:$dst, (extractelt (v2i32 DPR:$src),
2953 // def VGETLNf32: see FMRDH and FMRDL in ARMInstrVFP.td
2954 def : Pat<(NEONvgetlanes (v16i8 QPR:$src), imm:$lane),
2955 (VGETLNs8 (v8i8 (EXTRACT_SUBREG QPR:$src,
2956 (DSubReg_i8_reg imm:$lane))),
2957 (SubReg_i8_lane imm:$lane))>;
2958 def : Pat<(NEONvgetlanes (v8i16 QPR:$src), imm:$lane),
2959 (VGETLNs16 (v4i16 (EXTRACT_SUBREG QPR:$src,
2960 (DSubReg_i16_reg imm:$lane))),
2961 (SubReg_i16_lane imm:$lane))>;
2962 def : Pat<(NEONvgetlaneu (v16i8 QPR:$src), imm:$lane),
2963 (VGETLNu8 (v8i8 (EXTRACT_SUBREG QPR:$src,
2964 (DSubReg_i8_reg imm:$lane))),
2965 (SubReg_i8_lane imm:$lane))>;
2966 def : Pat<(NEONvgetlaneu (v8i16 QPR:$src), imm:$lane),
2967 (VGETLNu16 (v4i16 (EXTRACT_SUBREG QPR:$src,
2968 (DSubReg_i16_reg imm:$lane))),
2969 (SubReg_i16_lane imm:$lane))>;
2970 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
2971 (VGETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src,
2972 (DSubReg_i32_reg imm:$lane))),
2973 (SubReg_i32_lane imm:$lane))>;
2974 def : Pat<(extractelt (v2f32 DPR:$src1), imm:$src2),
2975 (EXTRACT_SUBREG (v2f32 (COPY_TO_REGCLASS (v2f32 DPR:$src1),DPR_VFP2)),
2976 (SSubReg_f32_reg imm:$src2))>;
2977 def : Pat<(extractelt (v4f32 QPR:$src1), imm:$src2),
2978 (EXTRACT_SUBREG (v4f32 (COPY_TO_REGCLASS (v4f32 QPR:$src1),QPR_VFP2)),
2979 (SSubReg_f32_reg imm:$src2))>;
2980 //def : Pat<(extractelt (v2i64 QPR:$src1), imm:$src2),
2981 // (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
2982 def : Pat<(extractelt (v2f64 QPR:$src1), imm:$src2),
2983 (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
2986 // VMOV : Vector Set Lane (move ARM core register to scalar)
2988 let Constraints = "$src1 = $dst" in {
2989 def VSETLNi8 : NVSetLane<{1,1,1,0,0,1,?,0}, 0b1011, {?,?}, (outs DPR:$dst),
2990 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
2991 IIC_VMOVISL, "vmov", "8", "$dst[$lane], $src2",
2992 [(set DPR:$dst, (vector_insert (v8i8 DPR:$src1),
2993 GPR:$src2, imm:$lane))]>;
2994 def VSETLNi16 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, {?,1}, (outs DPR:$dst),
2995 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
2996 IIC_VMOVISL, "vmov", "16", "$dst[$lane], $src2",
2997 [(set DPR:$dst, (vector_insert (v4i16 DPR:$src1),
2998 GPR:$src2, imm:$lane))]>;
2999 def VSETLNi32 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, 0b00, (outs DPR:$dst),
3000 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
3001 IIC_VMOVISL, "vmov", "32", "$dst[$lane], $src2",
3002 [(set DPR:$dst, (insertelt (v2i32 DPR:$src1),
3003 GPR:$src2, imm:$lane))]>;
3005 def : Pat<(vector_insert (v16i8 QPR:$src1), GPR:$src2, imm:$lane),
3006 (v16i8 (INSERT_SUBREG QPR:$src1,
3007 (v8i8 (VSETLNi8 (v8i8 (EXTRACT_SUBREG QPR:$src1,
3008 (DSubReg_i8_reg imm:$lane))),
3009 GPR:$src2, (SubReg_i8_lane imm:$lane))),
3010 (DSubReg_i8_reg imm:$lane)))>;
3011 def : Pat<(vector_insert (v8i16 QPR:$src1), GPR:$src2, imm:$lane),
3012 (v8i16 (INSERT_SUBREG QPR:$src1,
3013 (v4i16 (VSETLNi16 (v4i16 (EXTRACT_SUBREG QPR:$src1,
3014 (DSubReg_i16_reg imm:$lane))),
3015 GPR:$src2, (SubReg_i16_lane imm:$lane))),
3016 (DSubReg_i16_reg imm:$lane)))>;
3017 def : Pat<(insertelt (v4i32 QPR:$src1), GPR:$src2, imm:$lane),
3018 (v4i32 (INSERT_SUBREG QPR:$src1,
3019 (v2i32 (VSETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src1,
3020 (DSubReg_i32_reg imm:$lane))),
3021 GPR:$src2, (SubReg_i32_lane imm:$lane))),
3022 (DSubReg_i32_reg imm:$lane)))>;
3024 def : Pat<(v2f32 (insertelt DPR:$src1, SPR:$src2, imm:$src3)),
3025 (INSERT_SUBREG (v2f32 (COPY_TO_REGCLASS DPR:$src1, DPR_VFP2)),
3026 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
3027 def : Pat<(v4f32 (insertelt QPR:$src1, SPR:$src2, imm:$src3)),
3028 (INSERT_SUBREG (v4f32 (COPY_TO_REGCLASS QPR:$src1, QPR_VFP2)),
3029 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
3031 //def : Pat<(v2i64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
3032 // (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
3033 def : Pat<(v2f64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
3034 (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
3036 def : Pat<(v2f32 (scalar_to_vector SPR:$src)),
3037 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
3038 def : Pat<(v2f64 (scalar_to_vector (f64 DPR:$src))),
3039 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
3040 def : Pat<(v4f32 (scalar_to_vector SPR:$src)),
3041 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
3043 def : Pat<(v8i8 (scalar_to_vector GPR:$src)),
3044 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
3045 def : Pat<(v4i16 (scalar_to_vector GPR:$src)),
3046 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
3047 def : Pat<(v2i32 (scalar_to_vector GPR:$src)),
3048 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
3050 def : Pat<(v16i8 (scalar_to_vector GPR:$src)),
3051 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3052 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
3054 def : Pat<(v8i16 (scalar_to_vector GPR:$src)),
3055 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
3056 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
3058 def : Pat<(v4i32 (scalar_to_vector GPR:$src)),
3059 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
3060 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
3063 // VDUP : Vector Duplicate (from ARM core register to all elements)
3065 class VDUPD<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
3066 : NVDup<opcod1, 0b1011, opcod3, (outs DPR:$dst), (ins GPR:$src),
3067 IIC_VMOVIS, "vdup", Dt, "$dst, $src",
3068 [(set DPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
3069 class VDUPQ<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
3070 : NVDup<opcod1, 0b1011, opcod3, (outs QPR:$dst), (ins GPR:$src),
3071 IIC_VMOVIS, "vdup", Dt, "$dst, $src",
3072 [(set QPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
3074 def VDUP8d : VDUPD<0b11101100, 0b00, "8", v8i8>;
3075 def VDUP16d : VDUPD<0b11101000, 0b01, "16", v4i16>;
3076 def VDUP32d : VDUPD<0b11101000, 0b00, "32", v2i32>;
3077 def VDUP8q : VDUPQ<0b11101110, 0b00, "8", v16i8>;
3078 def VDUP16q : VDUPQ<0b11101010, 0b01, "16", v8i16>;
3079 def VDUP32q : VDUPQ<0b11101010, 0b00, "32", v4i32>;
3081 def VDUPfd : NVDup<0b11101000, 0b1011, 0b00, (outs DPR:$dst), (ins GPR:$src),
3082 IIC_VMOVIS, "vdup", "32", "$dst, $src",
3083 [(set DPR:$dst, (v2f32 (NEONvdup
3084 (f32 (bitconvert GPR:$src)))))]>;
3085 def VDUPfq : NVDup<0b11101010, 0b1011, 0b00, (outs QPR:$dst), (ins GPR:$src),
3086 IIC_VMOVIS, "vdup", "32", "$dst, $src",
3087 [(set QPR:$dst, (v4f32 (NEONvdup
3088 (f32 (bitconvert GPR:$src)))))]>;
3090 // VDUP : Vector Duplicate Lane (from scalar to all elements)
3092 class VDUPLND<bits<4> op19_16, string OpcodeStr, string Dt,
3094 : NVDupLane<op19_16, 0, (outs DPR:$dst), (ins DPR:$src, nohash_imm:$lane),
3095 IIC_VMOVD, OpcodeStr, Dt, "$dst, $src[$lane]",
3096 [(set DPR:$dst, (Ty (NEONvduplane (Ty DPR:$src), imm:$lane)))]>;
3098 class VDUPLNQ<bits<4> op19_16, string OpcodeStr, string Dt,
3099 ValueType ResTy, ValueType OpTy>
3100 : NVDupLane<op19_16, 1, (outs QPR:$dst), (ins DPR:$src, nohash_imm:$lane),
3101 IIC_VMOVD, OpcodeStr, Dt, "$dst, $src[$lane]",
3102 [(set QPR:$dst, (ResTy (NEONvduplane (OpTy DPR:$src),
3105 // Inst{19-16} is partially specified depending on the element size.
3107 def VDUPLN8d : VDUPLND<{?,?,?,1}, "vdup", "8", v8i8>;
3108 def VDUPLN16d : VDUPLND<{?,?,1,0}, "vdup", "16", v4i16>;
3109 def VDUPLN32d : VDUPLND<{?,1,0,0}, "vdup", "32", v2i32>;
3110 def VDUPLNfd : VDUPLND<{?,1,0,0}, "vdup", "32", v2f32>;
3111 def VDUPLN8q : VDUPLNQ<{?,?,?,1}, "vdup", "8", v16i8, v8i8>;
3112 def VDUPLN16q : VDUPLNQ<{?,?,1,0}, "vdup", "16", v8i16, v4i16>;
3113 def VDUPLN32q : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4i32, v2i32>;
3114 def VDUPLNfq : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4f32, v2f32>;
3116 def : Pat<(v16i8 (NEONvduplane (v16i8 QPR:$src), imm:$lane)),
3117 (v16i8 (VDUPLN8q (v8i8 (EXTRACT_SUBREG QPR:$src,
3118 (DSubReg_i8_reg imm:$lane))),
3119 (SubReg_i8_lane imm:$lane)))>;
3120 def : Pat<(v8i16 (NEONvduplane (v8i16 QPR:$src), imm:$lane)),
3121 (v8i16 (VDUPLN16q (v4i16 (EXTRACT_SUBREG QPR:$src,
3122 (DSubReg_i16_reg imm:$lane))),
3123 (SubReg_i16_lane imm:$lane)))>;
3124 def : Pat<(v4i32 (NEONvduplane (v4i32 QPR:$src), imm:$lane)),
3125 (v4i32 (VDUPLN32q (v2i32 (EXTRACT_SUBREG QPR:$src,
3126 (DSubReg_i32_reg imm:$lane))),
3127 (SubReg_i32_lane imm:$lane)))>;
3128 def : Pat<(v4f32 (NEONvduplane (v4f32 QPR:$src), imm:$lane)),
3129 (v4f32 (VDUPLNfq (v2f32 (EXTRACT_SUBREG QPR:$src,
3130 (DSubReg_i32_reg imm:$lane))),
3131 (SubReg_i32_lane imm:$lane)))>;
3133 def VDUPfdf : N2V<0b11, 0b11, {?,1}, {0,0}, 0b11000, 0, 0,
3134 (outs DPR:$dst), (ins SPR:$src),
3135 IIC_VMOVD, "vdup", "32", "$dst, ${src:lane}", "",
3136 [(set DPR:$dst, (v2f32 (NEONvdup (f32 SPR:$src))))]>;
3138 def VDUPfqf : N2V<0b11, 0b11, {?,1}, {0,0}, 0b11000, 1, 0,
3139 (outs QPR:$dst), (ins SPR:$src),
3140 IIC_VMOVD, "vdup", "32", "$dst, ${src:lane}", "",
3141 [(set QPR:$dst, (v4f32 (NEONvdup (f32 SPR:$src))))]>;
3143 // VMOVN : Vector Narrowing Move
3144 defm VMOVN : N2VNInt_HSD<0b11,0b11,0b10,0b00100,0,0, IIC_VMOVD,
3145 "vmovn", "i", int_arm_neon_vmovn>;
3146 // VQMOVN : Vector Saturating Narrowing Move
3147 defm VQMOVNs : N2VNInt_HSD<0b11,0b11,0b10,0b00101,0,0, IIC_VQUNAiD,
3148 "vqmovn", "s", int_arm_neon_vqmovns>;
3149 defm VQMOVNu : N2VNInt_HSD<0b11,0b11,0b10,0b00101,1,0, IIC_VQUNAiD,
3150 "vqmovn", "u", int_arm_neon_vqmovnu>;
3151 defm VQMOVNsu : N2VNInt_HSD<0b11,0b11,0b10,0b00100,1,0, IIC_VQUNAiD,
3152 "vqmovun", "s", int_arm_neon_vqmovnsu>;
3153 // VMOVL : Vector Lengthening Move
3154 defm VMOVLs : N2VL_QHS<0b01,0b10100,0,1, "vmovl", "s", sext>;
3155 defm VMOVLu : N2VL_QHS<0b11,0b10100,0,1, "vmovl", "u", zext>;
3157 // Vector Conversions.
3159 // VCVT : Vector Convert Between Floating-Point and Integers
3160 def VCVTf2sd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
3161 v2i32, v2f32, fp_to_sint>;
3162 def VCVTf2ud : N2VD<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
3163 v2i32, v2f32, fp_to_uint>;
3164 def VCVTs2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
3165 v2f32, v2i32, sint_to_fp>;
3166 def VCVTu2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
3167 v2f32, v2i32, uint_to_fp>;
3169 def VCVTf2sq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
3170 v4i32, v4f32, fp_to_sint>;
3171 def VCVTf2uq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
3172 v4i32, v4f32, fp_to_uint>;
3173 def VCVTs2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
3174 v4f32, v4i32, sint_to_fp>;
3175 def VCVTu2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
3176 v4f32, v4i32, uint_to_fp>;
3178 // VCVT : Vector Convert Between Floating-Point and Fixed-Point.
3179 def VCVTf2xsd : N2VCvtD<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
3180 v2i32, v2f32, int_arm_neon_vcvtfp2fxs>;
3181 def VCVTf2xud : N2VCvtD<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
3182 v2i32, v2f32, int_arm_neon_vcvtfp2fxu>;
3183 def VCVTxs2fd : N2VCvtD<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
3184 v2f32, v2i32, int_arm_neon_vcvtfxs2fp>;
3185 def VCVTxu2fd : N2VCvtD<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
3186 v2f32, v2i32, int_arm_neon_vcvtfxu2fp>;
3188 def VCVTf2xsq : N2VCvtQ<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
3189 v4i32, v4f32, int_arm_neon_vcvtfp2fxs>;
3190 def VCVTf2xuq : N2VCvtQ<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
3191 v4i32, v4f32, int_arm_neon_vcvtfp2fxu>;
3192 def VCVTxs2fq : N2VCvtQ<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
3193 v4f32, v4i32, int_arm_neon_vcvtfxs2fp>;
3194 def VCVTxu2fq : N2VCvtQ<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
3195 v4f32, v4i32, int_arm_neon_vcvtfxu2fp>;
3199 // VREV64 : Vector Reverse elements within 64-bit doublewords
3201 class VREV64D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3202 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 0, 0, (outs DPR:$dst),
3203 (ins DPR:$src), IIC_VMOVD,
3204 OpcodeStr, Dt, "$dst, $src", "",
3205 [(set DPR:$dst, (Ty (NEONvrev64 (Ty DPR:$src))))]>;
3206 class VREV64Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3207 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 1, 0, (outs QPR:$dst),
3208 (ins QPR:$src), IIC_VMOVD,
3209 OpcodeStr, Dt, "$dst, $src", "",
3210 [(set QPR:$dst, (Ty (NEONvrev64 (Ty QPR:$src))))]>;
3212 def VREV64d8 : VREV64D<0b00, "vrev64", "8", v8i8>;
3213 def VREV64d16 : VREV64D<0b01, "vrev64", "16", v4i16>;
3214 def VREV64d32 : VREV64D<0b10, "vrev64", "32", v2i32>;
3215 def VREV64df : VREV64D<0b10, "vrev64", "32", v2f32>;
3217 def VREV64q8 : VREV64Q<0b00, "vrev64", "8", v16i8>;
3218 def VREV64q16 : VREV64Q<0b01, "vrev64", "16", v8i16>;
3219 def VREV64q32 : VREV64Q<0b10, "vrev64", "32", v4i32>;
3220 def VREV64qf : VREV64Q<0b10, "vrev64", "32", v4f32>;
3222 // VREV32 : Vector Reverse elements within 32-bit words
3224 class VREV32D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3225 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 0, 0, (outs DPR:$dst),
3226 (ins DPR:$src), IIC_VMOVD,
3227 OpcodeStr, Dt, "$dst, $src", "",
3228 [(set DPR:$dst, (Ty (NEONvrev32 (Ty DPR:$src))))]>;
3229 class VREV32Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3230 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 1, 0, (outs QPR:$dst),
3231 (ins QPR:$src), IIC_VMOVD,
3232 OpcodeStr, Dt, "$dst, $src", "",
3233 [(set QPR:$dst, (Ty (NEONvrev32 (Ty QPR:$src))))]>;
3235 def VREV32d8 : VREV32D<0b00, "vrev32", "8", v8i8>;
3236 def VREV32d16 : VREV32D<0b01, "vrev32", "16", v4i16>;
3238 def VREV32q8 : VREV32Q<0b00, "vrev32", "8", v16i8>;
3239 def VREV32q16 : VREV32Q<0b01, "vrev32", "16", v8i16>;
3241 // VREV16 : Vector Reverse elements within 16-bit halfwords
3243 class VREV16D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3244 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 0, 0, (outs DPR:$dst),
3245 (ins DPR:$src), IIC_VMOVD,
3246 OpcodeStr, Dt, "$dst, $src", "",
3247 [(set DPR:$dst, (Ty (NEONvrev16 (Ty DPR:$src))))]>;
3248 class VREV16Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3249 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 1, 0, (outs QPR:$dst),
3250 (ins QPR:$src), IIC_VMOVD,
3251 OpcodeStr, Dt, "$dst, $src", "",
3252 [(set QPR:$dst, (Ty (NEONvrev16 (Ty QPR:$src))))]>;
3254 def VREV16d8 : VREV16D<0b00, "vrev16", "8", v8i8>;
3255 def VREV16q8 : VREV16Q<0b00, "vrev16", "8", v16i8>;
3257 // Other Vector Shuffles.
3259 // VEXT : Vector Extract
3261 class VEXTd<string OpcodeStr, string Dt, ValueType Ty>
3262 : N3V<0,1,0b11,{?,?,?,?},0,0, (outs DPR:$dst),
3263 (ins DPR:$lhs, DPR:$rhs, i32imm:$index), NVExtFrm,
3264 IIC_VEXTD, OpcodeStr, Dt, "$dst, $lhs, $rhs, $index", "",
3265 [(set DPR:$dst, (Ty (NEONvext (Ty DPR:$lhs),
3266 (Ty DPR:$rhs), imm:$index)))]>;
3268 class VEXTq<string OpcodeStr, string Dt, ValueType Ty>
3269 : N3V<0,1,0b11,{?,?,?,?},1,0, (outs QPR:$dst),
3270 (ins QPR:$lhs, QPR:$rhs, i32imm:$index), NVExtFrm,
3271 IIC_VEXTQ, OpcodeStr, Dt, "$dst, $lhs, $rhs, $index", "",
3272 [(set QPR:$dst, (Ty (NEONvext (Ty QPR:$lhs),
3273 (Ty QPR:$rhs), imm:$index)))]>;
3275 def VEXTd8 : VEXTd<"vext", "8", v8i8>;
3276 def VEXTd16 : VEXTd<"vext", "16", v4i16>;
3277 def VEXTd32 : VEXTd<"vext", "32", v2i32>;
3278 def VEXTdf : VEXTd<"vext", "32", v2f32>;
3280 def VEXTq8 : VEXTq<"vext", "8", v16i8>;
3281 def VEXTq16 : VEXTq<"vext", "16", v8i16>;
3282 def VEXTq32 : VEXTq<"vext", "32", v4i32>;
3283 def VEXTqf : VEXTq<"vext", "32", v4f32>;
3285 // VTRN : Vector Transpose
3287 def VTRNd8 : N2VDShuffle<0b00, 0b00001, "vtrn", "8">;
3288 def VTRNd16 : N2VDShuffle<0b01, 0b00001, "vtrn", "16">;
3289 def VTRNd32 : N2VDShuffle<0b10, 0b00001, "vtrn", "32">;
3291 def VTRNq8 : N2VQShuffle<0b00, 0b00001, IIC_VPERMQ, "vtrn", "8">;
3292 def VTRNq16 : N2VQShuffle<0b01, 0b00001, IIC_VPERMQ, "vtrn", "16">;
3293 def VTRNq32 : N2VQShuffle<0b10, 0b00001, IIC_VPERMQ, "vtrn", "32">;
3295 // VUZP : Vector Unzip (Deinterleave)
3297 def VUZPd8 : N2VDShuffle<0b00, 0b00010, "vuzp", "8">;
3298 def VUZPd16 : N2VDShuffle<0b01, 0b00010, "vuzp", "16">;
3299 def VUZPd32 : N2VDShuffle<0b10, 0b00010, "vuzp", "32">;
3301 def VUZPq8 : N2VQShuffle<0b00, 0b00010, IIC_VPERMQ3, "vuzp", "8">;
3302 def VUZPq16 : N2VQShuffle<0b01, 0b00010, IIC_VPERMQ3, "vuzp", "16">;
3303 def VUZPq32 : N2VQShuffle<0b10, 0b00010, IIC_VPERMQ3, "vuzp", "32">;
3305 // VZIP : Vector Zip (Interleave)
3307 def VZIPd8 : N2VDShuffle<0b00, 0b00011, "vzip", "8">;
3308 def VZIPd16 : N2VDShuffle<0b01, 0b00011, "vzip", "16">;
3309 def VZIPd32 : N2VDShuffle<0b10, 0b00011, "vzip", "32">;
3311 def VZIPq8 : N2VQShuffle<0b00, 0b00011, IIC_VPERMQ3, "vzip", "8">;
3312 def VZIPq16 : N2VQShuffle<0b01, 0b00011, IIC_VPERMQ3, "vzip", "16">;
3313 def VZIPq32 : N2VQShuffle<0b10, 0b00011, IIC_VPERMQ3, "vzip", "32">;
3315 // Vector Table Lookup and Table Extension.
3317 // VTBL : Vector Table Lookup
3319 : N3V<1,1,0b11,0b1000,0,0, (outs DPR:$dst),
3320 (ins DPR:$tbl1, DPR:$src), NVTBLFrm, IIC_VTB1,
3321 "vtbl", "8", "$dst, \\{$tbl1\\}, $src", "",
3322 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl1 DPR:$tbl1, DPR:$src)))]>;
3323 let hasExtraSrcRegAllocReq = 1 in {
3325 : N3V<1,1,0b11,0b1001,0,0, (outs DPR:$dst),
3326 (ins DPR:$tbl1, DPR:$tbl2, DPR:$src), NVTBLFrm, IIC_VTB2,
3327 "vtbl", "8", "$dst, \\{$tbl1, $tbl2\\}, $src", "", []>;
3329 : N3V<1,1,0b11,0b1010,0,0, (outs DPR:$dst),
3330 (ins DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$src), NVTBLFrm, IIC_VTB3,
3331 "vtbl", "8", "$dst, \\{$tbl1, $tbl2, $tbl3\\}, $src", "", []>;
3333 : N3V<1,1,0b11,0b1011,0,0, (outs DPR:$dst),
3334 (ins DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$src),
3336 "vtbl", "8", "$dst, \\{$tbl1, $tbl2, $tbl3, $tbl4\\}, $src", "", []>;
3337 } // hasExtraSrcRegAllocReq = 1
3339 // VTBX : Vector Table Extension
3341 : N3V<1,1,0b11,0b1000,1,0, (outs DPR:$dst),
3342 (ins DPR:$orig, DPR:$tbl1, DPR:$src), NVTBLFrm, IIC_VTBX1,
3343 "vtbx", "8", "$dst, \\{$tbl1\\}, $src", "$orig = $dst",
3344 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx1
3345 DPR:$orig, DPR:$tbl1, DPR:$src)))]>;
3346 let hasExtraSrcRegAllocReq = 1 in {
3348 : N3V<1,1,0b11,0b1001,1,0, (outs DPR:$dst),
3349 (ins DPR:$orig, DPR:$tbl1, DPR:$tbl2, DPR:$src), NVTBLFrm, IIC_VTBX2,
3350 "vtbx", "8", "$dst, \\{$tbl1, $tbl2\\}, $src", "$orig = $dst", []>;
3352 : N3V<1,1,0b11,0b1010,1,0, (outs DPR:$dst),
3353 (ins DPR:$orig, DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$src),
3354 NVTBLFrm, IIC_VTBX3,
3355 "vtbx", "8", "$dst, \\{$tbl1, $tbl2, $tbl3\\}, $src",
3356 "$orig = $dst", []>;
3358 : N3V<1,1,0b11,0b1011,1,0, (outs DPR:$dst), (ins DPR:$orig, DPR:$tbl1,
3359 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$src), NVTBLFrm, IIC_VTBX4,
3360 "vtbx", "8", "$dst, \\{$tbl1, $tbl2, $tbl3, $tbl4\\}, $src",
3361 "$orig = $dst", []>;
3362 } // hasExtraSrcRegAllocReq = 1
3364 //===----------------------------------------------------------------------===//
3365 // NEON instructions for single-precision FP math
3366 //===----------------------------------------------------------------------===//
3368 class N2VSPat<SDNode OpNode, ValueType ResTy, ValueType OpTy, NeonI Inst>
3369 : NEONFPPat<(ResTy (OpNode SPR:$a)),
3370 (EXTRACT_SUBREG (OpTy (Inst (INSERT_SUBREG (OpTy (IMPLICIT_DEF)),
3374 class N3VSPat<SDNode OpNode, NeonI Inst>
3375 : NEONFPPat<(f32 (OpNode SPR:$a, SPR:$b)),
3376 (EXTRACT_SUBREG (v2f32
3377 (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3379 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3383 class N3VSMulOpPat<SDNode MulNode, SDNode OpNode, NeonI Inst>
3384 : NEONFPPat<(f32 (OpNode SPR:$acc, (f32 (MulNode SPR:$a, SPR:$b)))),
3385 (EXTRACT_SUBREG (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3387 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3389 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3393 // These need separate instructions because they must use DPR_VFP2 register
3394 // class which have SPR sub-registers.
3396 // Vector Add Operations used for single-precision FP
3397 let neverHasSideEffects = 1 in
3398 def VADDfd_sfp : N3VS<0,0,0b00,0b1101,0, "vadd", "f32", v2f32, v2f32, fadd, 1>;
3399 def : N3VSPat<fadd, VADDfd_sfp>;
3401 // Vector Sub Operations used for single-precision FP
3402 let neverHasSideEffects = 1 in
3403 def VSUBfd_sfp : N3VS<0,0,0b10,0b1101,0, "vsub", "f32", v2f32, v2f32, fsub, 0>;
3404 def : N3VSPat<fsub, VSUBfd_sfp>;
3406 // Vector Multiply Operations used for single-precision FP
3407 let neverHasSideEffects = 1 in
3408 def VMULfd_sfp : N3VS<1,0,0b00,0b1101,1, "vmul", "f32", v2f32, v2f32, fmul, 1>;
3409 def : N3VSPat<fmul, VMULfd_sfp>;
3411 // Vector Multiply-Accumulate/Subtract used for single-precision FP
3412 // vml[as].f32 can cause 4-8 cycle stalls in following ASIMD instructions, so
3413 // we want to avoid them for now. e.g., alternating vmla/vadd instructions.
3415 //let neverHasSideEffects = 1 in
3416 //def VMLAfd_sfp : N3VSMulOp<0,0,0b00,0b1101,1, IIC_VMACD, "vmla", "f32",
3417 // v2f32, fmul, fadd>;
3418 //def : N3VSMulOpPat<fmul, fadd, VMLAfd_sfp>;
3420 //let neverHasSideEffects = 1 in
3421 //def VMLSfd_sfp : N3VSMulOp<0,0,0b10,0b1101,1, IIC_VMACD, "vmls", "f32",
3422 // v2f32, fmul, fsub>;
3423 //def : N3VSMulOpPat<fmul, fsub, VMLSfd_sfp>;
3425 // Vector Absolute used for single-precision FP
3426 let neverHasSideEffects = 1 in
3427 def VABSfd_sfp : N2V<0b11, 0b11, 0b10, 0b01, 0b01110, 0, 0,
3428 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), IIC_VUNAD,
3429 "vabs", "f32", "$dst, $src", "", []>;
3430 def : N2VSPat<fabs, f32, v2f32, VABSfd_sfp>;
3432 // Vector Negate used for single-precision FP
3433 let neverHasSideEffects = 1 in
3434 def VNEGfd_sfp : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
3435 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), IIC_VUNAD,
3436 "vneg", "f32", "$dst, $src", "", []>;
3437 def : N2VSPat<fneg, f32, v2f32, VNEGfd_sfp>;
3439 // Vector Maximum used for single-precision FP
3440 let neverHasSideEffects = 1 in
3441 def VMAXfd_sfp : N3V<0, 0, 0b00, 0b1111, 0, 0, (outs DPR_VFP2:$dst),
3442 (ins DPR_VFP2:$src1, DPR_VFP2:$src2), N3RegFrm, IIC_VBIND,
3443 "vmax", "f32", "$dst, $src1, $src2", "", []>;
3444 def : N3VSPat<NEONfmax, VMAXfd_sfp>;
3446 // Vector Minimum used for single-precision FP
3447 let neverHasSideEffects = 1 in
3448 def VMINfd_sfp : N3V<0, 0, 0b00, 0b1111, 0, 0, (outs DPR_VFP2:$dst),
3449 (ins DPR_VFP2:$src1, DPR_VFP2:$src2), N3RegFrm, IIC_VBIND,
3450 "vmin", "f32", "$dst, $src1, $src2", "", []>;
3451 def : N3VSPat<NEONfmin, VMINfd_sfp>;
3453 // Vector Convert between single-precision FP and integer
3454 let neverHasSideEffects = 1 in
3455 def VCVTf2sd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
3456 v2i32, v2f32, fp_to_sint>;
3457 def : N2VSPat<arm_ftosi, f32, v2f32, VCVTf2sd_sfp>;
3459 let neverHasSideEffects = 1 in
3460 def VCVTf2ud_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
3461 v2i32, v2f32, fp_to_uint>;
3462 def : N2VSPat<arm_ftoui, f32, v2f32, VCVTf2ud_sfp>;
3464 let neverHasSideEffects = 1 in
3465 def VCVTs2fd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
3466 v2f32, v2i32, sint_to_fp>;
3467 def : N2VSPat<arm_sitof, f32, v2i32, VCVTs2fd_sfp>;
3469 let neverHasSideEffects = 1 in
3470 def VCVTu2fd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
3471 v2f32, v2i32, uint_to_fp>;
3472 def : N2VSPat<arm_uitof, f32, v2i32, VCVTu2fd_sfp>;
3474 //===----------------------------------------------------------------------===//
3475 // Non-Instruction Patterns
3476 //===----------------------------------------------------------------------===//
3479 def : Pat<(v1i64 (bitconvert (v2i32 DPR:$src))), (v1i64 DPR:$src)>;
3480 def : Pat<(v1i64 (bitconvert (v4i16 DPR:$src))), (v1i64 DPR:$src)>;
3481 def : Pat<(v1i64 (bitconvert (v8i8 DPR:$src))), (v1i64 DPR:$src)>;
3482 def : Pat<(v1i64 (bitconvert (f64 DPR:$src))), (v1i64 DPR:$src)>;
3483 def : Pat<(v1i64 (bitconvert (v2f32 DPR:$src))), (v1i64 DPR:$src)>;
3484 def : Pat<(v2i32 (bitconvert (v1i64 DPR:$src))), (v2i32 DPR:$src)>;
3485 def : Pat<(v2i32 (bitconvert (v4i16 DPR:$src))), (v2i32 DPR:$src)>;
3486 def : Pat<(v2i32 (bitconvert (v8i8 DPR:$src))), (v2i32 DPR:$src)>;
3487 def : Pat<(v2i32 (bitconvert (f64 DPR:$src))), (v2i32 DPR:$src)>;
3488 def : Pat<(v2i32 (bitconvert (v2f32 DPR:$src))), (v2i32 DPR:$src)>;
3489 def : Pat<(v4i16 (bitconvert (v1i64 DPR:$src))), (v4i16 DPR:$src)>;
3490 def : Pat<(v4i16 (bitconvert (v2i32 DPR:$src))), (v4i16 DPR:$src)>;
3491 def : Pat<(v4i16 (bitconvert (v8i8 DPR:$src))), (v4i16 DPR:$src)>;
3492 def : Pat<(v4i16 (bitconvert (f64 DPR:$src))), (v4i16 DPR:$src)>;
3493 def : Pat<(v4i16 (bitconvert (v2f32 DPR:$src))), (v4i16 DPR:$src)>;
3494 def : Pat<(v8i8 (bitconvert (v1i64 DPR:$src))), (v8i8 DPR:$src)>;
3495 def : Pat<(v8i8 (bitconvert (v2i32 DPR:$src))), (v8i8 DPR:$src)>;
3496 def : Pat<(v8i8 (bitconvert (v4i16 DPR:$src))), (v8i8 DPR:$src)>;
3497 def : Pat<(v8i8 (bitconvert (f64 DPR:$src))), (v8i8 DPR:$src)>;
3498 def : Pat<(v8i8 (bitconvert (v2f32 DPR:$src))), (v8i8 DPR:$src)>;
3499 def : Pat<(f64 (bitconvert (v1i64 DPR:$src))), (f64 DPR:$src)>;
3500 def : Pat<(f64 (bitconvert (v2i32 DPR:$src))), (f64 DPR:$src)>;
3501 def : Pat<(f64 (bitconvert (v4i16 DPR:$src))), (f64 DPR:$src)>;
3502 def : Pat<(f64 (bitconvert (v8i8 DPR:$src))), (f64 DPR:$src)>;
3503 def : Pat<(f64 (bitconvert (v2f32 DPR:$src))), (f64 DPR:$src)>;
3504 def : Pat<(v2f32 (bitconvert (f64 DPR:$src))), (v2f32 DPR:$src)>;
3505 def : Pat<(v2f32 (bitconvert (v1i64 DPR:$src))), (v2f32 DPR:$src)>;
3506 def : Pat<(v2f32 (bitconvert (v2i32 DPR:$src))), (v2f32 DPR:$src)>;
3507 def : Pat<(v2f32 (bitconvert (v4i16 DPR:$src))), (v2f32 DPR:$src)>;
3508 def : Pat<(v2f32 (bitconvert (v8i8 DPR:$src))), (v2f32 DPR:$src)>;
3510 def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (v2i64 QPR:$src)>;
3511 def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (v2i64 QPR:$src)>;
3512 def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (v2i64 QPR:$src)>;
3513 def : Pat<(v2i64 (bitconvert (v2f64 QPR:$src))), (v2i64 QPR:$src)>;
3514 def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (v2i64 QPR:$src)>;
3515 def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (v4i32 QPR:$src)>;
3516 def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (v4i32 QPR:$src)>;
3517 def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (v4i32 QPR:$src)>;
3518 def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (v4i32 QPR:$src)>;
3519 def : Pat<(v4i32 (bitconvert (v4f32 QPR:$src))), (v4i32 QPR:$src)>;
3520 def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (v8i16 QPR:$src)>;
3521 def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (v8i16 QPR:$src)>;
3522 def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (v8i16 QPR:$src)>;
3523 def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (v8i16 QPR:$src)>;
3524 def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (v8i16 QPR:$src)>;
3525 def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (v16i8 QPR:$src)>;
3526 def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (v16i8 QPR:$src)>;
3527 def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (v16i8 QPR:$src)>;
3528 def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (v16i8 QPR:$src)>;
3529 def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (v16i8 QPR:$src)>;
3530 def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (v4f32 QPR:$src)>;
3531 def : Pat<(v4f32 (bitconvert (v4i32 QPR:$src))), (v4f32 QPR:$src)>;
3532 def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (v4f32 QPR:$src)>;
3533 def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (v4f32 QPR:$src)>;
3534 def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (v4f32 QPR:$src)>;
3535 def : Pat<(v2f64 (bitconvert (v2i64 QPR:$src))), (v2f64 QPR:$src)>;
3536 def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (v2f64 QPR:$src)>;
3537 def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 QPR:$src)>;
3538 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 QPR:$src)>;
3539 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 QPR:$src)>;