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 // Use vldmia to load a Q register as a D register pair.
128 // This is equivalent to VLDMD except that it has a Q register operand
129 // instead of a pair of D registers.
131 : AXDI5<(outs QPR:$dst), (ins addrmode4:$addr, pred:$p),
132 IndexModeNone, IIC_fpLoadm,
133 "vldm${addr:submode}${p}\t$addr, ${dst:dregpair}", "",
134 [(set QPR:$dst, (v2f64 (load addrmode4:$addr)))]>;
136 let mayLoad = 1, neverHasSideEffects = 1 in {
137 // Use vld1 to load a Q register as a D register pair.
138 // This alternative to VLDMQ allows an alignment to be specified.
139 // This is equivalent to VLD1q64 except that it has a Q register operand.
141 : NLdSt<0,0b10,0b1010,0b1100, (outs QPR:$dst), (ins addrmode6:$addr),
142 IIC_VLD1, "vld1", "64", "${dst:dregpair}, $addr", "", []>;
143 } // mayLoad = 1, neverHasSideEffects = 1
145 // Use vstmia to store a Q register as a D register pair.
146 // This is equivalent to VSTMD except that it has a Q register operand
147 // instead of a pair of D registers.
149 : AXDI5<(outs), (ins QPR:$src, addrmode4:$addr, pred:$p),
150 IndexModeNone, IIC_fpStorem,
151 "vstm${addr:submode}${p}\t$addr, ${src:dregpair}", "",
152 [(store (v2f64 QPR:$src), addrmode4:$addr)]>;
154 let mayStore = 1, neverHasSideEffects = 1 in {
155 // Use vst1 to store a Q register as a D register pair.
156 // This alternative to VSTMQ allows an alignment to be specified.
157 // This is equivalent to VST1q64 except that it has a Q register operand.
159 : NLdSt<0,0b00,0b1010,0b1100, (outs), (ins addrmode6:$addr, QPR:$src),
160 IIC_VST, "vst1", "64", "${src:dregpair}, $addr", "", []>;
161 } // mayStore = 1, neverHasSideEffects = 1
163 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
165 // VLD1 : Vector Load (multiple single elements)
166 class VLD1D<bits<4> op7_4, string Dt>
167 : NLdSt<0,0b10,0b0111,op7_4, (outs DPR:$dst),
168 (ins addrmode6:$addr), IIC_VLD1,
169 "vld1", Dt, "\\{$dst\\}, $addr", "", []>;
170 class VLD1Q<bits<4> op7_4, string Dt>
171 : NLdSt<0,0b10,0b1010,op7_4, (outs DPR:$dst1, DPR:$dst2),
172 (ins addrmode6:$addr), IIC_VLD1,
173 "vld1", Dt, "\\{$dst1, $dst2\\}, $addr", "", []>;
175 def VLD1d8 : VLD1D<0b0000, "8">;
176 def VLD1d16 : VLD1D<0b0100, "16">;
177 def VLD1d32 : VLD1D<0b1000, "32">;
178 def VLD1d64 : VLD1D<0b1100, "64">;
180 def VLD1q8 : VLD1Q<0b0000, "8">;
181 def VLD1q16 : VLD1Q<0b0100, "16">;
182 def VLD1q32 : VLD1Q<0b1000, "32">;
183 def VLD1q64 : VLD1Q<0b1100, "64">;
185 // ...with address register writeback:
186 class VLD1DWB<bits<4> op7_4, string Dt>
187 : NLdSt<0,0b10,0b0111,op7_4, (outs DPR:$dst, GPR:$wb),
188 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD1,
189 "vld1", Dt, "\\{$dst\\}, $addr$offset",
190 "$addr.addr = $wb", []>;
191 class VLD1QWB<bits<4> op7_4, string Dt>
192 : NLdSt<0,0b10,0b1010,op7_4, (outs QPR:$dst, GPR:$wb),
193 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD1,
194 "vld1", Dt, "${dst:dregpair}, $addr$offset",
195 "$addr.addr = $wb", []>;
197 def VLD1d8_UPD : VLD1DWB<0b0000, "8">;
198 def VLD1d16_UPD : VLD1DWB<0b0100, "16">;
199 def VLD1d32_UPD : VLD1DWB<0b1000, "32">;
200 def VLD1d64_UPD : VLD1DWB<0b1100, "64">;
202 def VLD1q8_UPD : VLD1QWB<0b0000, "8">;
203 def VLD1q16_UPD : VLD1QWB<0b0100, "16">;
204 def VLD1q32_UPD : VLD1QWB<0b1000, "32">;
205 def VLD1q64_UPD : VLD1QWB<0b1100, "64">;
207 // ...with 3 registers (some of these are only for the disassembler):
208 class VLD1D3<bits<4> op7_4, string Dt>
209 : NLdSt<0,0b10,0b0110,op7_4, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3),
210 (ins addrmode6:$addr), IIC_VLD1, "vld1", Dt,
211 "\\{$dst1, $dst2, $dst3\\}, $addr", "", []>;
212 class VLD1D3WB<bits<4> op7_4, string Dt>
213 : NLdSt<0,0b10,0b0110,op7_4, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, GPR:$wb),
214 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD1, "vld1", Dt,
215 "\\{$dst1, $dst2, $dst3\\}, $addr$offset", "$addr.addr = $wb", []>;
217 def VLD1d8T : VLD1D3<0b0000, "8">;
218 def VLD1d16T : VLD1D3<0b0100, "16">;
219 def VLD1d32T : VLD1D3<0b1000, "32">;
220 def VLD1d64T : VLD1D3<0b1100, "64">;
222 def VLD1d8T_UPD : VLD1D3WB<0b0000, "8">;
223 def VLD1d16T_UPD : VLD1D3WB<0b0100, "16">;
224 def VLD1d32T_UPD : VLD1D3WB<0b1000, "32">;
225 def VLD1d64T_UPD : VLD1D3WB<0b1100, "64">;
227 // ...with 4 registers (some of these are only for the disassembler):
228 class VLD1D4<bits<4> op7_4, string Dt>
229 : NLdSt<0,0b10,0b0010,op7_4,(outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
230 (ins addrmode6:$addr), IIC_VLD1, "vld1", Dt,
231 "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr", "", []>;
232 class VLD1D4WB<bits<4> op7_4, string Dt>
233 : NLdSt<0,0b10,0b0010,op7_4,
234 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
235 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD1, "vld1", Dt,
236 "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr$offset", "$addr.addr = $wb",
239 def VLD1d8Q : VLD1D4<0b0000, "8">;
240 def VLD1d16Q : VLD1D4<0b0100, "16">;
241 def VLD1d32Q : VLD1D4<0b1000, "32">;
242 def VLD1d64Q : VLD1D4<0b1100, "64">;
244 def VLD1d8Q_UPD : VLD1D4WB<0b0000, "8">;
245 def VLD1d16Q_UPD : VLD1D4WB<0b0100, "16">;
246 def VLD1d32Q_UPD : VLD1D4WB<0b1000, "32">;
247 def VLD1d64Q_UPD : VLD1D4WB<0b1100, "64">;
249 // VLD2 : Vector Load (multiple 2-element structures)
250 class VLD2D<bits<4> op11_8, bits<4> op7_4, string Dt>
251 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2),
252 (ins addrmode6:$addr), IIC_VLD2,
253 "vld2", Dt, "\\{$dst1, $dst2\\}, $addr", "", []>;
254 class VLD2Q<bits<4> op7_4, string Dt>
255 : NLdSt<0, 0b10, 0b0011, op7_4,
256 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
257 (ins addrmode6:$addr), IIC_VLD2,
258 "vld2", Dt, "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr", "", []>;
260 def VLD2d8 : VLD2D<0b1000, 0b0000, "8">;
261 def VLD2d16 : VLD2D<0b1000, 0b0100, "16">;
262 def VLD2d32 : VLD2D<0b1000, 0b1000, "32">;
264 def VLD2q8 : VLD2Q<0b0000, "8">;
265 def VLD2q16 : VLD2Q<0b0100, "16">;
266 def VLD2q32 : VLD2Q<0b1000, "32">;
268 // ...with address register writeback:
269 class VLD2DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
270 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2, GPR:$wb),
271 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD2,
272 "vld2", Dt, "\\{$dst1, $dst2\\}, $addr$offset",
273 "$addr.addr = $wb", []>;
274 class VLD2QWB<bits<4> op7_4, string Dt>
275 : NLdSt<0, 0b10, 0b0011, op7_4,
276 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
277 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD2,
278 "vld2", Dt, "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr$offset",
279 "$addr.addr = $wb", []>;
281 def VLD2d8_UPD : VLD2DWB<0b1000, 0b0000, "8">;
282 def VLD2d16_UPD : VLD2DWB<0b1000, 0b0100, "16">;
283 def VLD2d32_UPD : VLD2DWB<0b1000, 0b1000, "32">;
285 def VLD2q8_UPD : VLD2QWB<0b0000, "8">;
286 def VLD2q16_UPD : VLD2QWB<0b0100, "16">;
287 def VLD2q32_UPD : VLD2QWB<0b1000, "32">;
289 // ...with double-spaced registers (for disassembly only):
290 def VLD2b8 : VLD2D<0b1001, 0b0000, "8">;
291 def VLD2b16 : VLD2D<0b1001, 0b0100, "16">;
292 def VLD2b32 : VLD2D<0b1001, 0b1000, "32">;
293 def VLD2b8_UPD : VLD2DWB<0b1001, 0b0000, "8">;
294 def VLD2b16_UPD : VLD2DWB<0b1001, 0b0100, "16">;
295 def VLD2b32_UPD : VLD2DWB<0b1001, 0b1000, "32">;
297 // VLD3 : Vector Load (multiple 3-element structures)
298 class VLD3D<bits<4> op11_8, bits<4> op7_4, string Dt>
299 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3),
300 (ins addrmode6:$addr), IIC_VLD3,
301 "vld3", Dt, "\\{$dst1, $dst2, $dst3\\}, $addr", "", []>;
303 def VLD3d8 : VLD3D<0b0100, 0b0000, "8">;
304 def VLD3d16 : VLD3D<0b0100, 0b0100, "16">;
305 def VLD3d32 : VLD3D<0b0100, 0b1000, "32">;
307 // ...with address register writeback:
308 class VLD3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
309 : NLdSt<0, 0b10, op11_8, op7_4,
310 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, GPR:$wb),
311 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD3,
312 "vld3", Dt, "\\{$dst1, $dst2, $dst3\\}, $addr$offset",
313 "$addr.addr = $wb", []>;
315 def VLD3d8_UPD : VLD3DWB<0b0100, 0b0000, "8">;
316 def VLD3d16_UPD : VLD3DWB<0b0100, 0b0100, "16">;
317 def VLD3d32_UPD : VLD3DWB<0b0100, 0b1000, "32">;
319 // ...with double-spaced registers (non-updating versions for disassembly only):
320 def VLD3q8 : VLD3D<0b0101, 0b0000, "8">;
321 def VLD3q16 : VLD3D<0b0101, 0b0100, "16">;
322 def VLD3q32 : VLD3D<0b0101, 0b1000, "32">;
323 def VLD3q8_UPD : VLD3DWB<0b0101, 0b0000, "8">;
324 def VLD3q16_UPD : VLD3DWB<0b0101, 0b0100, "16">;
325 def VLD3q32_UPD : VLD3DWB<0b0101, 0b1000, "32">;
327 // ...alternate versions to be allocated odd register numbers:
328 def VLD3q8odd_UPD : VLD3DWB<0b0101, 0b0000, "8">;
329 def VLD3q16odd_UPD : VLD3DWB<0b0101, 0b0100, "16">;
330 def VLD3q32odd_UPD : VLD3DWB<0b0101, 0b1000, "32">;
332 // VLD4 : Vector Load (multiple 4-element structures)
333 class VLD4D<bits<4> op11_8, bits<4> op7_4, string Dt>
334 : NLdSt<0, 0b10, op11_8, op7_4,
335 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
336 (ins addrmode6:$addr), IIC_VLD4,
337 "vld4", Dt, "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr", "", []>;
339 def VLD4d8 : VLD4D<0b0000, 0b0000, "8">;
340 def VLD4d16 : VLD4D<0b0000, 0b0100, "16">;
341 def VLD4d32 : VLD4D<0b0000, 0b1000, "32">;
343 // ...with address register writeback:
344 class VLD4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
345 : NLdSt<0, 0b10, op11_8, op7_4,
346 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
347 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD4,
348 "vld4", Dt, "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr$offset",
349 "$addr.addr = $wb", []>;
351 def VLD4d8_UPD : VLD4DWB<0b0000, 0b0000, "8">;
352 def VLD4d16_UPD : VLD4DWB<0b0000, 0b0100, "16">;
353 def VLD4d32_UPD : VLD4DWB<0b0000, 0b1000, "32">;
355 // ...with double-spaced registers (non-updating versions for disassembly only):
356 def VLD4q8 : VLD4D<0b0001, 0b0000, "8">;
357 def VLD4q16 : VLD4D<0b0001, 0b0100, "16">;
358 def VLD4q32 : VLD4D<0b0001, 0b1000, "32">;
359 def VLD4q8_UPD : VLD4DWB<0b0001, 0b0000, "8">;
360 def VLD4q16_UPD : VLD4DWB<0b0001, 0b0100, "16">;
361 def VLD4q32_UPD : VLD4DWB<0b0001, 0b1000, "32">;
363 // ...alternate versions to be allocated odd register numbers:
364 def VLD4q8odd_UPD : VLD4DWB<0b0001, 0b0000, "8">;
365 def VLD4q16odd_UPD : VLD4DWB<0b0001, 0b0100, "16">;
366 def VLD4q32odd_UPD : VLD4DWB<0b0001, 0b1000, "32">;
368 // VLD1LN : Vector Load (single element to one lane)
369 // FIXME: Not yet implemented.
371 // VLD2LN : Vector Load (single 2-element structure to one lane)
372 class VLD2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
373 : NLdSt<1, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2),
374 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, nohash_imm:$lane),
375 IIC_VLD2, "vld2", Dt, "\\{$dst1[$lane], $dst2[$lane]\\}, $addr",
376 "$src1 = $dst1, $src2 = $dst2", []>;
378 def VLD2LNd8 : VLD2LN<0b0001, {?,?,?,?}, "8">;
379 def VLD2LNd16 : VLD2LN<0b0101, {?,?,0,?}, "16">;
380 def VLD2LNd32 : VLD2LN<0b1001, {?,0,?,?}, "32">;
382 // ...with double-spaced registers:
383 def VLD2LNq16 : VLD2LN<0b0101, {?,?,1,?}, "16">;
384 def VLD2LNq32 : VLD2LN<0b1001, {?,1,?,?}, "32">;
386 // ...alternate versions to be allocated odd register numbers:
387 def VLD2LNq16odd : VLD2LN<0b0101, {?,?,1,?}, "16">;
388 def VLD2LNq32odd : VLD2LN<0b1001, {?,1,?,?}, "32">;
390 // ...with address register writeback:
391 class VLD2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
392 : NLdSt<1, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2, GPR:$wb),
393 (ins addrmode6:$addr, am6offset:$offset,
394 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VLD2, "vld2", Dt,
395 "\\{$dst1[$lane], $dst2[$lane]\\}, $addr$offset",
396 "$src1 = $dst1, $src2 = $dst2, $addr.addr = $wb", []>;
398 def VLD2LNd8_UPD : VLD2LNWB<0b0001, {?,?,?,?}, "8">;
399 def VLD2LNd16_UPD : VLD2LNWB<0b0101, {?,?,0,?}, "16">;
400 def VLD2LNd32_UPD : VLD2LNWB<0b1001, {?,0,?,?}, "32">;
402 def VLD2LNq16_UPD : VLD2LNWB<0b0101, {?,?,1,?}, "16">;
403 def VLD2LNq32_UPD : VLD2LNWB<0b1001, {?,1,?,?}, "32">;
405 // VLD3LN : Vector Load (single 3-element structure to one lane)
406 class VLD3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
407 : NLdSt<1, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3),
408 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3,
409 nohash_imm:$lane), IIC_VLD3, "vld3", Dt,
410 "\\{$dst1[$lane], $dst2[$lane], $dst3[$lane]\\}, $addr",
411 "$src1 = $dst1, $src2 = $dst2, $src3 = $dst3", []>;
413 def VLD3LNd8 : VLD3LN<0b0010, {?,?,?,0}, "8">;
414 def VLD3LNd16 : VLD3LN<0b0110, {?,?,0,0}, "16">;
415 def VLD3LNd32 : VLD3LN<0b1010, {?,0,0,0}, "32">;
417 // ...with double-spaced registers:
418 def VLD3LNq16 : VLD3LN<0b0110, {?,?,1,0}, "16">;
419 def VLD3LNq32 : VLD3LN<0b1010, {?,1,0,0}, "32">;
421 // ...alternate versions to be allocated odd register numbers:
422 def VLD3LNq16odd : VLD3LN<0b0110, {?,?,1,0}, "16">;
423 def VLD3LNq32odd : VLD3LN<0b1010, {?,1,0,0}, "32">;
425 // ...with address register writeback:
426 class VLD3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
427 : NLdSt<1, 0b10, op11_8, op7_4,
428 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, GPR:$wb),
429 (ins addrmode6:$addr, am6offset:$offset,
430 DPR:$src1, DPR:$src2, DPR:$src3, nohash_imm:$lane),
431 IIC_VLD3, "vld3", Dt,
432 "\\{$dst1[$lane], $dst2[$lane], $dst3[$lane]\\}, $addr$offset",
433 "$src1 = $dst1, $src2 = $dst2, $src3 = $dst3, $addr.addr = $wb",
436 def VLD3LNd8_UPD : VLD3LNWB<0b0010, {?,?,?,0}, "8">;
437 def VLD3LNd16_UPD : VLD3LNWB<0b0110, {?,?,0,0}, "16">;
438 def VLD3LNd32_UPD : VLD3LNWB<0b1010, {?,0,0,0}, "32">;
440 def VLD3LNq16_UPD : VLD3LNWB<0b0110, {?,?,1,0}, "16">;
441 def VLD3LNq32_UPD : VLD3LNWB<0b1010, {?,1,0,0}, "32">;
443 // VLD4LN : Vector Load (single 4-element structure to one lane)
444 class VLD4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
445 : NLdSt<1, 0b10, op11_8, op7_4,
446 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
447 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
448 nohash_imm:$lane), IIC_VLD4, "vld4", Dt,
449 "\\{$dst1[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $addr",
450 "$src1 = $dst1, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4", []>;
452 def VLD4LNd8 : VLD4LN<0b0011, {?,?,?,?}, "8">;
453 def VLD4LNd16 : VLD4LN<0b0111, {?,?,0,?}, "16">;
454 def VLD4LNd32 : VLD4LN<0b1011, {?,0,?,?}, "32">;
456 // ...with double-spaced registers:
457 def VLD4LNq16 : VLD4LN<0b0111, {?,?,1,?}, "16">;
458 def VLD4LNq32 : VLD4LN<0b1011, {?,1,?,?}, "32">;
460 // ...alternate versions to be allocated odd register numbers:
461 def VLD4LNq16odd : VLD4LN<0b0111, {?,?,1,?}, "16">;
462 def VLD4LNq32odd : VLD4LN<0b1011, {?,1,?,?}, "32">;
464 // ...with address register writeback:
465 class VLD4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
466 : NLdSt<1, 0b10, op11_8, op7_4,
467 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
468 (ins addrmode6:$addr, am6offset:$offset,
469 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
470 IIC_VLD4, "vld4", Dt,
471 "\\{$dst1[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $addr$offset",
472 "$src1 = $dst1, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4, $addr.addr = $wb",
475 def VLD4LNd8_UPD : VLD4LNWB<0b0011, {?,?,?,?}, "8">;
476 def VLD4LNd16_UPD : VLD4LNWB<0b0111, {?,?,0,?}, "16">;
477 def VLD4LNd32_UPD : VLD4LNWB<0b1011, {?,0,?,?}, "32">;
479 def VLD4LNq16_UPD : VLD4LNWB<0b0111, {?,?,1,?}, "16">;
480 def VLD4LNq32_UPD : VLD4LNWB<0b1011, {?,1,?,?}, "32">;
482 // VLD1DUP : Vector Load (single element to all lanes)
483 // VLD2DUP : Vector Load (single 2-element structure to all lanes)
484 // VLD3DUP : Vector Load (single 3-element structure to all lanes)
485 // VLD4DUP : Vector Load (single 4-element structure to all lanes)
486 // FIXME: Not yet implemented.
487 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
489 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
491 // Classes for VST* pseudo-instructions with multi-register operands.
492 // These are expanded to real instructions after register allocation.
494 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src), IIC_VST, "">;
496 : PseudoNLdSt<(outs GPR:$wb),
497 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src), IIC_VST,
499 class VSTQQQQWBPseudo
500 : PseudoNLdSt<(outs GPR:$wb),
501 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), IIC_VST,
504 // VST1 : Vector Store (multiple single elements)
505 class VST1D<bits<4> op7_4, string Dt>
506 : NLdSt<0,0b00,0b0111,op7_4, (outs), (ins addrmode6:$addr, DPR:$src), IIC_VST,
507 "vst1", Dt, "\\{$src\\}, $addr", "", []>;
508 class VST1Q<bits<4> op7_4, string Dt>
509 : NLdSt<0,0b00,0b1010,op7_4, (outs),
510 (ins addrmode6:$addr, DPR:$src1, DPR:$src2), IIC_VST,
511 "vst1", Dt, "\\{$src1, $src2\\}, $addr", "", []>;
513 def VST1d8 : VST1D<0b0000, "8">;
514 def VST1d16 : VST1D<0b0100, "16">;
515 def VST1d32 : VST1D<0b1000, "32">;
516 def VST1d64 : VST1D<0b1100, "64">;
518 def VST1q8 : VST1Q<0b0000, "8">;
519 def VST1q16 : VST1Q<0b0100, "16">;
520 def VST1q32 : VST1Q<0b1000, "32">;
521 def VST1q64 : VST1Q<0b1100, "64">;
523 // ...with address register writeback:
524 class VST1DWB<bits<4> op7_4, string Dt>
525 : NLdSt<0, 0b00, 0b0111, op7_4, (outs GPR:$wb),
526 (ins addrmode6:$addr, am6offset:$offset, DPR:$src), IIC_VST,
527 "vst1", Dt, "\\{$src\\}, $addr$offset", "$addr.addr = $wb", []>;
528 class VST1QWB<bits<4> op7_4, string Dt>
529 : NLdSt<0, 0b00, 0b1010, op7_4, (outs GPR:$wb),
530 (ins addrmode6:$addr, am6offset:$offset, QPR:$src), IIC_VST,
531 "vst1", Dt, "${src:dregpair}, $addr$offset", "$addr.addr = $wb", []>;
533 def VST1d8_UPD : VST1DWB<0b0000, "8">;
534 def VST1d16_UPD : VST1DWB<0b0100, "16">;
535 def VST1d32_UPD : VST1DWB<0b1000, "32">;
536 def VST1d64_UPD : VST1DWB<0b1100, "64">;
538 def VST1q8_UPD : VST1QWB<0b0000, "8">;
539 def VST1q16_UPD : VST1QWB<0b0100, "16">;
540 def VST1q32_UPD : VST1QWB<0b1000, "32">;
541 def VST1q64_UPD : VST1QWB<0b1100, "64">;
543 // ...with 3 registers (some of these are only for the disassembler):
544 class VST1D3<bits<4> op7_4, string Dt>
545 : NLdSt<0, 0b00, 0b0110, op7_4, (outs),
546 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3),
547 IIC_VST, "vst1", Dt, "\\{$src1, $src2, $src3\\}, $addr", "", []>;
548 class VST1D3WB<bits<4> op7_4, string Dt>
549 : NLdSt<0, 0b00, 0b0110, op7_4, (outs GPR:$wb),
550 (ins addrmode6:$addr, am6offset:$offset,
551 DPR:$src1, DPR:$src2, DPR:$src3),
552 IIC_VST, "vst1", Dt, "\\{$src1, $src2, $src3\\}, $addr$offset",
553 "$addr.addr = $wb", []>;
555 def VST1d8T : VST1D3<0b0000, "8">;
556 def VST1d16T : VST1D3<0b0100, "16">;
557 def VST1d32T : VST1D3<0b1000, "32">;
558 def VST1d64T : VST1D3<0b1100, "64">;
560 def VST1d8T_UPD : VST1D3WB<0b0000, "8">;
561 def VST1d16T_UPD : VST1D3WB<0b0100, "16">;
562 def VST1d32T_UPD : VST1D3WB<0b1000, "32">;
563 def VST1d64T_UPD : VST1D3WB<0b1100, "64">;
565 def VST1d64TPseudo : VSTQQPseudo;
566 def VST1d64TPseudo_UPD : VSTQQWBPseudo;
568 // ...with 4 registers (some of these are only for the disassembler):
569 class VST1D4<bits<4> op7_4, string Dt>
570 : NLdSt<0, 0b00, 0b0010, op7_4, (outs),
571 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
572 IIC_VST, "vst1", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr", "",
574 class VST1D4WB<bits<4> op7_4, string Dt>
575 : NLdSt<0, 0b00, 0b0010, op7_4, (outs GPR:$wb),
576 (ins addrmode6:$addr, am6offset:$offset,
577 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
578 IIC_VST, "vst1", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr$offset",
579 "$addr.addr = $wb", []>;
581 def VST1d8Q : VST1D4<0b0000, "8">;
582 def VST1d16Q : VST1D4<0b0100, "16">;
583 def VST1d32Q : VST1D4<0b1000, "32">;
584 def VST1d64Q : VST1D4<0b1100, "64">;
586 def VST1d8Q_UPD : VST1D4WB<0b0000, "8">;
587 def VST1d16Q_UPD : VST1D4WB<0b0100, "16">;
588 def VST1d32Q_UPD : VST1D4WB<0b1000, "32">;
589 def VST1d64Q_UPD : VST1D4WB<0b1100, "64">;
591 def VST1d64QPseudo : VSTQQPseudo;
592 def VST1d64QPseudo_UPD : VSTQQWBPseudo;
594 // VST2 : Vector Store (multiple 2-element structures)
595 class VST2D<bits<4> op11_8, bits<4> op7_4, string Dt>
596 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
597 (ins addrmode6:$addr, DPR:$src1, DPR:$src2),
598 IIC_VST, "vst2", Dt, "\\{$src1, $src2\\}, $addr", "", []>;
599 class VST2Q<bits<4> op7_4, string Dt>
600 : NLdSt<0, 0b00, 0b0011, op7_4, (outs),
601 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
602 IIC_VST, "vst2", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr",
605 def VST2d8 : VST2D<0b1000, 0b0000, "8">;
606 def VST2d16 : VST2D<0b1000, 0b0100, "16">;
607 def VST2d32 : VST2D<0b1000, 0b1000, "32">;
609 def VST2q8 : VST2Q<0b0000, "8">;
610 def VST2q16 : VST2Q<0b0100, "16">;
611 def VST2q32 : VST2Q<0b1000, "32">;
613 // ...with address register writeback:
614 class VST2DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
615 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
616 (ins addrmode6:$addr, am6offset:$offset, DPR:$src1, DPR:$src2),
617 IIC_VST, "vst2", Dt, "\\{$src1, $src2\\}, $addr$offset",
618 "$addr.addr = $wb", []>;
619 class VST2QWB<bits<4> op7_4, string Dt>
620 : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
621 (ins addrmode6:$addr, am6offset:$offset,
622 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
623 IIC_VST, "vst2", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr$offset",
624 "$addr.addr = $wb", []>;
626 def VST2d8_UPD : VST2DWB<0b1000, 0b0000, "8">;
627 def VST2d16_UPD : VST2DWB<0b1000, 0b0100, "16">;
628 def VST2d32_UPD : VST2DWB<0b1000, 0b1000, "32">;
630 def VST2q8_UPD : VST2QWB<0b0000, "8">;
631 def VST2q16_UPD : VST2QWB<0b0100, "16">;
632 def VST2q32_UPD : VST2QWB<0b1000, "32">;
634 // ...with double-spaced registers (for disassembly only):
635 def VST2b8 : VST2D<0b1001, 0b0000, "8">;
636 def VST2b16 : VST2D<0b1001, 0b0100, "16">;
637 def VST2b32 : VST2D<0b1001, 0b1000, "32">;
638 def VST2b8_UPD : VST2DWB<0b1001, 0b0000, "8">;
639 def VST2b16_UPD : VST2DWB<0b1001, 0b0100, "16">;
640 def VST2b32_UPD : VST2DWB<0b1001, 0b1000, "32">;
642 // VST3 : Vector Store (multiple 3-element structures)
643 class VST3D<bits<4> op11_8, bits<4> op7_4, string Dt>
644 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
645 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3), IIC_VST,
646 "vst3", Dt, "\\{$src1, $src2, $src3\\}, $addr", "", []>;
648 def VST3d8 : VST3D<0b0100, 0b0000, "8">;
649 def VST3d16 : VST3D<0b0100, 0b0100, "16">;
650 def VST3d32 : VST3D<0b0100, 0b1000, "32">;
652 def VST3d8Pseudo : VSTQQPseudo;
653 def VST3d16Pseudo : VSTQQPseudo;
654 def VST3d32Pseudo : VSTQQPseudo;
656 // ...with address register writeback:
657 class VST3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
658 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
659 (ins addrmode6:$addr, am6offset:$offset,
660 DPR:$src1, DPR:$src2, DPR:$src3), IIC_VST,
661 "vst3", Dt, "\\{$src1, $src2, $src3\\}, $addr$offset",
662 "$addr.addr = $wb", []>;
664 def VST3d8_UPD : VST3DWB<0b0100, 0b0000, "8">;
665 def VST3d16_UPD : VST3DWB<0b0100, 0b0100, "16">;
666 def VST3d32_UPD : VST3DWB<0b0100, 0b1000, "32">;
668 def VST3d8Pseudo_UPD : VSTQQWBPseudo;
669 def VST3d16Pseudo_UPD : VSTQQWBPseudo;
670 def VST3d32Pseudo_UPD : VSTQQWBPseudo;
672 // ...with double-spaced registers (non-updating versions for disassembly only):
673 def VST3q8 : VST3D<0b0101, 0b0000, "8">;
674 def VST3q16 : VST3D<0b0101, 0b0100, "16">;
675 def VST3q32 : VST3D<0b0101, 0b1000, "32">;
676 def VST3q8_UPD : VST3DWB<0b0101, 0b0000, "8">;
677 def VST3q16_UPD : VST3DWB<0b0101, 0b0100, "16">;
678 def VST3q32_UPD : VST3DWB<0b0101, 0b1000, "32">;
680 def VST3q8Pseudo_UPD : VSTQQQQWBPseudo;
681 def VST3q16Pseudo_UPD : VSTQQQQWBPseudo;
682 def VST3q32Pseudo_UPD : VSTQQQQWBPseudo;
684 // ...alternate versions to be allocated odd register numbers:
685 def VST3q8oddPseudo_UPD : VSTQQQQWBPseudo;
686 def VST3q16oddPseudo_UPD : VSTQQQQWBPseudo;
687 def VST3q32oddPseudo_UPD : VSTQQQQWBPseudo;
689 // VST4 : Vector Store (multiple 4-element structures)
690 class VST4D<bits<4> op11_8, bits<4> op7_4, string Dt>
691 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
692 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
693 IIC_VST, "vst4", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr",
696 def VST4d8 : VST4D<0b0000, 0b0000, "8">;
697 def VST4d16 : VST4D<0b0000, 0b0100, "16">;
698 def VST4d32 : VST4D<0b0000, 0b1000, "32">;
700 def VST4d8Pseudo : VSTQQPseudo;
701 def VST4d16Pseudo : VSTQQPseudo;
702 def VST4d32Pseudo : VSTQQPseudo;
704 // ...with address register writeback:
705 class VST4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
706 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
707 (ins addrmode6:$addr, am6offset:$offset,
708 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST,
709 "vst4", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr$offset",
710 "$addr.addr = $wb", []>;
712 def VST4d8_UPD : VST4DWB<0b0000, 0b0000, "8">;
713 def VST4d16_UPD : VST4DWB<0b0000, 0b0100, "16">;
714 def VST4d32_UPD : VST4DWB<0b0000, 0b1000, "32">;
716 def VST4d8Pseudo_UPD : VSTQQWBPseudo;
717 def VST4d16Pseudo_UPD : VSTQQWBPseudo;
718 def VST4d32Pseudo_UPD : VSTQQWBPseudo;
720 // ...with double-spaced registers (non-updating versions for disassembly only):
721 def VST4q8 : VST4D<0b0001, 0b0000, "8">;
722 def VST4q16 : VST4D<0b0001, 0b0100, "16">;
723 def VST4q32 : VST4D<0b0001, 0b1000, "32">;
724 def VST4q8_UPD : VST4DWB<0b0001, 0b0000, "8">;
725 def VST4q16_UPD : VST4DWB<0b0001, 0b0100, "16">;
726 def VST4q32_UPD : VST4DWB<0b0001, 0b1000, "32">;
728 def VST4q8Pseudo_UPD : VSTQQQQWBPseudo;
729 def VST4q16Pseudo_UPD : VSTQQQQWBPseudo;
730 def VST4q32Pseudo_UPD : VSTQQQQWBPseudo;
732 // ...alternate versions to be allocated odd register numbers:
733 def VST4q8oddPseudo_UPD : VSTQQQQWBPseudo;
734 def VST4q16oddPseudo_UPD : VSTQQQQWBPseudo;
735 def VST4q32oddPseudo_UPD : VSTQQQQWBPseudo;
737 // VST1LN : Vector Store (single element from one lane)
738 // FIXME: Not yet implemented.
740 // VST2LN : Vector Store (single 2-element structure from one lane)
741 class VST2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
742 : NLdSt<1, 0b00, op11_8, op7_4, (outs),
743 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, nohash_imm:$lane),
744 IIC_VST, "vst2", Dt, "\\{$src1[$lane], $src2[$lane]\\}, $addr",
747 def VST2LNd8 : VST2LN<0b0001, {?,?,?,?}, "8">;
748 def VST2LNd16 : VST2LN<0b0101, {?,?,0,?}, "16">;
749 def VST2LNd32 : VST2LN<0b1001, {?,0,?,?}, "32">;
751 // ...with double-spaced registers:
752 def VST2LNq16 : VST2LN<0b0101, {?,?,1,?}, "16">;
753 def VST2LNq32 : VST2LN<0b1001, {?,1,?,?}, "32">;
755 // ...alternate versions to be allocated odd register numbers:
756 def VST2LNq16odd : VST2LN<0b0101, {?,?,1,?}, "16">;
757 def VST2LNq32odd : VST2LN<0b1001, {?,1,?,?}, "32">;
759 // ...with address register writeback:
760 class VST2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
761 : NLdSt<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
762 (ins addrmode6:$addr, am6offset:$offset,
763 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VST, "vst2", Dt,
764 "\\{$src1[$lane], $src2[$lane]\\}, $addr$offset",
765 "$addr.addr = $wb", []>;
767 def VST2LNd8_UPD : VST2LNWB<0b0001, {?,?,?,?}, "8">;
768 def VST2LNd16_UPD : VST2LNWB<0b0101, {?,?,0,?}, "16">;
769 def VST2LNd32_UPD : VST2LNWB<0b1001, {?,0,?,?}, "32">;
771 def VST2LNq16_UPD : VST2LNWB<0b0101, {?,?,1,?}, "16">;
772 def VST2LNq32_UPD : VST2LNWB<0b1001, {?,1,?,?}, "32">;
774 // VST3LN : Vector Store (single 3-element structure from one lane)
775 class VST3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
776 : NLdSt<1, 0b00, op11_8, op7_4, (outs),
777 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3,
778 nohash_imm:$lane), IIC_VST, "vst3", Dt,
779 "\\{$src1[$lane], $src2[$lane], $src3[$lane]\\}, $addr", "", []>;
781 def VST3LNd8 : VST3LN<0b0010, {?,?,?,0}, "8">;
782 def VST3LNd16 : VST3LN<0b0110, {?,?,0,0}, "16">;
783 def VST3LNd32 : VST3LN<0b1010, {?,0,0,0}, "32">;
785 // ...with double-spaced registers:
786 def VST3LNq16 : VST3LN<0b0110, {?,?,1,0}, "16">;
787 def VST3LNq32 : VST3LN<0b1010, {?,1,0,0}, "32">;
789 // ...alternate versions to be allocated odd register numbers:
790 def VST3LNq16odd : VST3LN<0b0110, {?,?,1,0}, "16">;
791 def VST3LNq32odd : VST3LN<0b1010, {?,1,0,0}, "32">;
793 // ...with address register writeback:
794 class VST3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
795 : NLdSt<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
796 (ins addrmode6:$addr, am6offset:$offset,
797 DPR:$src1, DPR:$src2, DPR:$src3, nohash_imm:$lane),
799 "\\{$src1[$lane], $src2[$lane], $src3[$lane]\\}, $addr$offset",
800 "$addr.addr = $wb", []>;
802 def VST3LNd8_UPD : VST3LNWB<0b0010, {?,?,?,0}, "8">;
803 def VST3LNd16_UPD : VST3LNWB<0b0110, {?,?,0,0}, "16">;
804 def VST3LNd32_UPD : VST3LNWB<0b1010, {?,0,0,0}, "32">;
806 def VST3LNq16_UPD : VST3LNWB<0b0110, {?,?,1,0}, "16">;
807 def VST3LNq32_UPD : VST3LNWB<0b1010, {?,1,0,0}, "32">;
809 // VST4LN : Vector Store (single 4-element structure from one lane)
810 class VST4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
811 : NLdSt<1, 0b00, op11_8, op7_4, (outs),
812 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
813 nohash_imm:$lane), IIC_VST, "vst4", Dt,
814 "\\{$src1[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $addr",
817 def VST4LNd8 : VST4LN<0b0011, {?,?,?,?}, "8">;
818 def VST4LNd16 : VST4LN<0b0111, {?,?,0,?}, "16">;
819 def VST4LNd32 : VST4LN<0b1011, {?,0,?,?}, "32">;
821 // ...with double-spaced registers:
822 def VST4LNq16 : VST4LN<0b0111, {?,?,1,?}, "16">;
823 def VST4LNq32 : VST4LN<0b1011, {?,1,?,?}, "32">;
825 // ...alternate versions to be allocated odd register numbers:
826 def VST4LNq16odd : VST4LN<0b0111, {?,?,1,?}, "16">;
827 def VST4LNq32odd : VST4LN<0b1011, {?,1,?,?}, "32">;
829 // ...with address register writeback:
830 class VST4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
831 : NLdSt<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
832 (ins addrmode6:$addr, am6offset:$offset,
833 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
835 "\\{$src1[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $addr$offset",
836 "$addr.addr = $wb", []>;
838 def VST4LNd8_UPD : VST4LNWB<0b0011, {?,?,?,?}, "8">;
839 def VST4LNd16_UPD : VST4LNWB<0b0111, {?,?,0,?}, "16">;
840 def VST4LNd32_UPD : VST4LNWB<0b1011, {?,0,?,?}, "32">;
842 def VST4LNq16_UPD : VST4LNWB<0b0111, {?,?,1,?}, "16">;
843 def VST4LNq32_UPD : VST4LNWB<0b1011, {?,1,?,?}, "32">;
845 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
848 //===----------------------------------------------------------------------===//
849 // NEON pattern fragments
850 //===----------------------------------------------------------------------===//
852 // Extract D sub-registers of Q registers.
853 def DSubReg_i8_reg : SDNodeXForm<imm, [{
854 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
855 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/8, MVT::i32);
857 def DSubReg_i16_reg : SDNodeXForm<imm, [{
858 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
859 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/4, MVT::i32);
861 def DSubReg_i32_reg : SDNodeXForm<imm, [{
862 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
863 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/2, MVT::i32);
865 def DSubReg_f64_reg : SDNodeXForm<imm, [{
866 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
867 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue(), MVT::i32);
870 // Extract S sub-registers of Q/D registers.
871 def SSubReg_f32_reg : SDNodeXForm<imm, [{
872 assert(ARM::ssub_3 == ARM::ssub_0+3 && "Unexpected subreg numbering");
873 return CurDAG->getTargetConstant(ARM::ssub_0 + N->getZExtValue(), MVT::i32);
876 // Translate lane numbers from Q registers to D subregs.
877 def SubReg_i8_lane : SDNodeXForm<imm, [{
878 return CurDAG->getTargetConstant(N->getZExtValue() & 7, MVT::i32);
880 def SubReg_i16_lane : SDNodeXForm<imm, [{
881 return CurDAG->getTargetConstant(N->getZExtValue() & 3, MVT::i32);
883 def SubReg_i32_lane : SDNodeXForm<imm, [{
884 return CurDAG->getTargetConstant(N->getZExtValue() & 1, MVT::i32);
887 //===----------------------------------------------------------------------===//
888 // Instruction Classes
889 //===----------------------------------------------------------------------===//
891 // Basic 2-register operations: single-, double- and quad-register.
892 class N2VS<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
893 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
894 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
895 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
896 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src),
897 IIC_VUNAD, OpcodeStr, Dt, "$dst, $src", "", []>;
898 class N2VD<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
899 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
900 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
901 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
902 (ins DPR:$src), IIC_VUNAD, OpcodeStr, Dt,"$dst, $src", "",
903 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src))))]>;
904 class N2VQ<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
905 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
906 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
907 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
908 (ins QPR:$src), IIC_VUNAQ, OpcodeStr, Dt,"$dst, $src", "",
909 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src))))]>;
911 // Basic 2-register intrinsics, both double- and quad-register.
912 class N2VDInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
913 bits<2> op17_16, bits<5> op11_7, bit op4,
914 InstrItinClass itin, string OpcodeStr, string Dt,
915 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
916 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
917 (ins DPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
918 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
919 class N2VQInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
920 bits<2> op17_16, bits<5> op11_7, bit op4,
921 InstrItinClass itin, string OpcodeStr, string Dt,
922 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
923 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
924 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
925 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
927 // Narrow 2-register intrinsics.
928 class N2VNInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
929 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
930 InstrItinClass itin, string OpcodeStr, string Dt,
931 ValueType TyD, ValueType TyQ, Intrinsic IntOp>
932 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$dst),
933 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
934 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src))))]>;
936 // Long 2-register operations (currently only used for VMOVL).
937 class N2VL<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
938 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
939 InstrItinClass itin, string OpcodeStr, string Dt,
940 ValueType TyQ, ValueType TyD, SDNode OpNode>
941 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$dst),
942 (ins DPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
943 [(set QPR:$dst, (TyQ (OpNode (TyD DPR:$src))))]>;
945 // 2-register shuffles (VTRN/VZIP/VUZP), both double- and quad-register.
946 class N2VDShuffle<bits<2> op19_18, bits<5> op11_7, string OpcodeStr, string Dt>
947 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 0, 0, (outs DPR:$dst1, DPR:$dst2),
948 (ins DPR:$src1, DPR:$src2), IIC_VPERMD,
949 OpcodeStr, Dt, "$dst1, $dst2",
950 "$src1 = $dst1, $src2 = $dst2", []>;
951 class N2VQShuffle<bits<2> op19_18, bits<5> op11_7,
952 InstrItinClass itin, string OpcodeStr, string Dt>
953 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 1, 0, (outs QPR:$dst1, QPR:$dst2),
954 (ins QPR:$src1, QPR:$src2), itin, OpcodeStr, Dt, "$dst1, $dst2",
955 "$src1 = $dst1, $src2 = $dst2", []>;
957 // Basic 3-register operations: single-, double- and quad-register.
958 class N3VS<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
959 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
960 SDNode OpNode, bit Commutable>
961 : N3V<op24, op23, op21_20, op11_8, 0, op4,
962 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src1, DPR_VFP2:$src2), N3RegFrm,
963 IIC_VBIND, OpcodeStr, Dt, "$dst, $src1, $src2", "", []> {
964 let isCommutable = Commutable;
967 class N3VD<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
968 InstrItinClass itin, string OpcodeStr, string Dt,
969 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
970 : N3V<op24, op23, op21_20, op11_8, 0, op4,
971 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
972 OpcodeStr, Dt, "$dst, $src1, $src2", "",
973 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src1), (OpTy DPR:$src2))))]> {
974 let isCommutable = Commutable;
976 // Same as N3VD but no data type.
977 class N3VDX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
978 InstrItinClass itin, string OpcodeStr,
979 ValueType ResTy, ValueType OpTy,
980 SDNode OpNode, bit Commutable>
981 : N3VX<op24, op23, op21_20, op11_8, 0, op4,
982 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
983 OpcodeStr, "$dst, $src1, $src2", "",
984 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src1), (OpTy DPR:$src2))))]>{
985 let isCommutable = Commutable;
988 class N3VDSL<bits<2> op21_20, bits<4> op11_8,
989 InstrItinClass itin, string OpcodeStr, string Dt,
990 ValueType Ty, SDNode ShOp>
991 : N3V<0, 1, op21_20, op11_8, 1, 0,
992 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
993 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
995 (Ty (ShOp (Ty DPR:$src1),
996 (Ty (NEONvduplane (Ty DPR_VFP2:$src2),imm:$lane)))))]> {
997 let isCommutable = 0;
999 class N3VDSL16<bits<2> op21_20, bits<4> op11_8,
1000 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1001 : N3V<0, 1, op21_20, op11_8, 1, 0,
1002 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1003 NVMulSLFrm, IIC_VMULi16D, OpcodeStr, Dt,"$dst, $src1, $src2[$lane]","",
1004 [(set (Ty DPR:$dst),
1005 (Ty (ShOp (Ty DPR:$src1),
1006 (Ty (NEONvduplane (Ty DPR_8:$src2), imm:$lane)))))]> {
1007 let isCommutable = 0;
1010 class N3VQ<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1011 InstrItinClass itin, string OpcodeStr, string Dt,
1012 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
1013 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1014 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), N3RegFrm, itin,
1015 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1016 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src1), (OpTy QPR:$src2))))]> {
1017 let isCommutable = Commutable;
1019 class N3VQX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1020 InstrItinClass itin, string OpcodeStr,
1021 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
1022 : N3VX<op24, op23, op21_20, op11_8, 1, op4,
1023 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), N3RegFrm, itin,
1024 OpcodeStr, "$dst, $src1, $src2", "",
1025 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src1), (OpTy QPR:$src2))))]>{
1026 let isCommutable = Commutable;
1028 class N3VQSL<bits<2> op21_20, bits<4> op11_8,
1029 InstrItinClass itin, string OpcodeStr, string Dt,
1030 ValueType ResTy, ValueType OpTy, SDNode ShOp>
1031 : N3V<1, 1, op21_20, op11_8, 1, 0,
1032 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1033 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1034 [(set (ResTy QPR:$dst),
1035 (ResTy (ShOp (ResTy QPR:$src1),
1036 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1038 let isCommutable = 0;
1040 class N3VQSL16<bits<2> op21_20, bits<4> op11_8, string OpcodeStr, string Dt,
1041 ValueType ResTy, ValueType OpTy, SDNode ShOp>
1042 : N3V<1, 1, op21_20, op11_8, 1, 0,
1043 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1044 NVMulSLFrm, IIC_VMULi16Q, OpcodeStr, Dt,"$dst, $src1, $src2[$lane]","",
1045 [(set (ResTy QPR:$dst),
1046 (ResTy (ShOp (ResTy QPR:$src1),
1047 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
1049 let isCommutable = 0;
1052 // Basic 3-register intrinsics, both double- and quad-register.
1053 class N3VDInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1054 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1055 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
1056 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1057 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), f, itin,
1058 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1059 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src1), (OpTy DPR:$src2))))]> {
1060 let isCommutable = Commutable;
1062 class N3VDIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1063 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
1064 : N3V<0, 1, op21_20, op11_8, 1, 0,
1065 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1066 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1067 [(set (Ty DPR:$dst),
1068 (Ty (IntOp (Ty DPR:$src1),
1069 (Ty (NEONvduplane (Ty DPR_VFP2:$src2),
1071 let isCommutable = 0;
1073 class N3VDIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1074 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
1075 : N3V<0, 1, op21_20, op11_8, 1, 0,
1076 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1077 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1078 [(set (Ty DPR:$dst),
1079 (Ty (IntOp (Ty DPR:$src1),
1080 (Ty (NEONvduplane (Ty DPR_8:$src2), imm:$lane)))))]> {
1081 let isCommutable = 0;
1084 class N3VQInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1085 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1086 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
1087 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1088 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), f, itin,
1089 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1090 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src1), (OpTy QPR:$src2))))]> {
1091 let isCommutable = Commutable;
1093 class N3VQIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1094 string OpcodeStr, string Dt,
1095 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1096 : N3V<1, 1, op21_20, op11_8, 1, 0,
1097 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1098 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1099 [(set (ResTy QPR:$dst),
1100 (ResTy (IntOp (ResTy QPR:$src1),
1101 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1103 let isCommutable = 0;
1105 class N3VQIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1106 string OpcodeStr, string Dt,
1107 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1108 : N3V<1, 1, op21_20, op11_8, 1, 0,
1109 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1110 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1111 [(set (ResTy QPR:$dst),
1112 (ResTy (IntOp (ResTy QPR:$src1),
1113 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
1115 let isCommutable = 0;
1118 // Multiply-Add/Sub operations: single-, double- and quad-register.
1119 class N3VSMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1120 InstrItinClass itin, string OpcodeStr, string Dt,
1121 ValueType Ty, SDNode MulOp, SDNode OpNode>
1122 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1123 (outs DPR_VFP2:$dst),
1124 (ins DPR_VFP2:$src1, DPR_VFP2:$src2, DPR_VFP2:$src3), N3RegFrm, itin,
1125 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst", []>;
1127 class N3VDMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1128 InstrItinClass itin, string OpcodeStr, string Dt,
1129 ValueType Ty, SDNode MulOp, SDNode OpNode>
1130 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1131 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3), N3RegFrm, itin,
1132 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1133 [(set DPR:$dst, (Ty (OpNode DPR:$src1,
1134 (Ty (MulOp DPR:$src2, DPR:$src3)))))]>;
1135 class N3VDMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1136 string OpcodeStr, string Dt,
1137 ValueType Ty, SDNode MulOp, SDNode ShOp>
1138 : N3V<0, 1, op21_20, op11_8, 1, 0,
1140 (ins DPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1142 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1143 [(set (Ty DPR:$dst),
1144 (Ty (ShOp (Ty DPR:$src1),
1145 (Ty (MulOp DPR:$src2,
1146 (Ty (NEONvduplane (Ty DPR_VFP2:$src3),
1148 class N3VDMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1149 string OpcodeStr, string Dt,
1150 ValueType Ty, SDNode MulOp, SDNode ShOp>
1151 : N3V<0, 1, op21_20, op11_8, 1, 0,
1153 (ins DPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane),
1155 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1156 [(set (Ty DPR:$dst),
1157 (Ty (ShOp (Ty DPR:$src1),
1158 (Ty (MulOp DPR:$src2,
1159 (Ty (NEONvduplane (Ty DPR_8:$src3),
1162 class N3VQMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1163 InstrItinClass itin, string OpcodeStr, string Dt, ValueType Ty,
1164 SDNode MulOp, SDNode OpNode>
1165 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1166 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3), N3RegFrm, itin,
1167 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1168 [(set QPR:$dst, (Ty (OpNode QPR:$src1,
1169 (Ty (MulOp QPR:$src2, QPR:$src3)))))]>;
1170 class N3VQMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1171 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1172 SDNode MulOp, SDNode ShOp>
1173 : N3V<1, 1, op21_20, op11_8, 1, 0,
1175 (ins QPR:$src1, QPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1177 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1178 [(set (ResTy QPR:$dst),
1179 (ResTy (ShOp (ResTy QPR:$src1),
1180 (ResTy (MulOp QPR:$src2,
1181 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src3),
1183 class N3VQMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1184 string OpcodeStr, string Dt,
1185 ValueType ResTy, ValueType OpTy,
1186 SDNode MulOp, SDNode ShOp>
1187 : N3V<1, 1, op21_20, op11_8, 1, 0,
1189 (ins QPR:$src1, QPR:$src2, DPR_8:$src3, nohash_imm:$lane),
1191 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1192 [(set (ResTy QPR:$dst),
1193 (ResTy (ShOp (ResTy QPR:$src1),
1194 (ResTy (MulOp QPR:$src2,
1195 (ResTy (NEONvduplane (OpTy DPR_8:$src3),
1198 // Neon 3-argument intrinsics, both double- and quad-register.
1199 // The destination register is also used as the first source operand register.
1200 class N3VDInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1201 InstrItinClass itin, string OpcodeStr, string Dt,
1202 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1203 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1204 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3), N3RegFrm, itin,
1205 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1206 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src1),
1207 (OpTy DPR:$src2), (OpTy DPR:$src3))))]>;
1208 class N3VQInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1209 InstrItinClass itin, string OpcodeStr, string Dt,
1210 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1211 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1212 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3), N3RegFrm, itin,
1213 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1214 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src1),
1215 (OpTy QPR:$src2), (OpTy QPR:$src3))))]>;
1217 // Neon Long 3-argument intrinsic. The destination register is
1218 // a quad-register and is also used as the first source operand register.
1219 class N3VLInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1220 InstrItinClass itin, string OpcodeStr, string Dt,
1221 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
1222 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1223 (outs QPR:$dst), (ins QPR:$src1, DPR:$src2, DPR:$src3), N3RegFrm, itin,
1224 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1226 (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$src2), (TyD DPR:$src3))))]>;
1227 class N3VLInt3SL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1228 string OpcodeStr, string Dt,
1229 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1230 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1232 (ins QPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1234 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1235 [(set (ResTy QPR:$dst),
1236 (ResTy (IntOp (ResTy QPR:$src1),
1238 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src3),
1240 class N3VLInt3SL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1241 InstrItinClass itin, string OpcodeStr, string Dt,
1242 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1243 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1245 (ins QPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane),
1247 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1248 [(set (ResTy QPR:$dst),
1249 (ResTy (IntOp (ResTy QPR:$src1),
1251 (OpTy (NEONvduplane (OpTy DPR_8:$src3),
1254 // Narrowing 3-register intrinsics.
1255 class N3VNInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1256 string OpcodeStr, string Dt, ValueType TyD, ValueType TyQ,
1257 Intrinsic IntOp, bit Commutable>
1258 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1259 (outs DPR:$dst), (ins QPR:$src1, QPR:$src2), N3RegFrm, IIC_VBINi4D,
1260 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1261 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src1), (TyQ QPR:$src2))))]> {
1262 let isCommutable = Commutable;
1265 // Long 3-register intrinsics.
1266 class N3VLInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1267 InstrItinClass itin, string OpcodeStr, string Dt,
1268 ValueType TyQ, ValueType TyD, Intrinsic IntOp, bit Commutable>
1269 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1270 (outs QPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
1271 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1272 [(set QPR:$dst, (TyQ (IntOp (TyD DPR:$src1), (TyD DPR:$src2))))]> {
1273 let isCommutable = Commutable;
1275 class N3VLIntSL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1276 string OpcodeStr, string Dt,
1277 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1278 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1279 (outs QPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1280 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1281 [(set (ResTy QPR:$dst),
1282 (ResTy (IntOp (OpTy DPR:$src1),
1283 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1285 class N3VLIntSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1286 InstrItinClass itin, string OpcodeStr, string Dt,
1287 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1288 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1289 (outs QPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1290 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1291 [(set (ResTy QPR:$dst),
1292 (ResTy (IntOp (OpTy DPR:$src1),
1293 (OpTy (NEONvduplane (OpTy DPR_8:$src2),
1296 // Wide 3-register intrinsics.
1297 class N3VWInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1298 string OpcodeStr, string Dt, ValueType TyQ, ValueType TyD,
1299 Intrinsic IntOp, bit Commutable>
1300 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1301 (outs QPR:$dst), (ins QPR:$src1, DPR:$src2), N3RegFrm, IIC_VSUBiD,
1302 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1303 [(set QPR:$dst, (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$src2))))]> {
1304 let isCommutable = Commutable;
1307 // Pairwise long 2-register intrinsics, both double- and quad-register.
1308 class N2VDPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1309 bits<2> op17_16, bits<5> op11_7, bit op4,
1310 string OpcodeStr, string Dt,
1311 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1312 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
1313 (ins DPR:$src), IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
1314 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
1315 class N2VQPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1316 bits<2> op17_16, bits<5> op11_7, bit op4,
1317 string OpcodeStr, string Dt,
1318 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1319 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
1320 (ins QPR:$src), IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
1321 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
1323 // Pairwise long 2-register accumulate intrinsics,
1324 // both double- and quad-register.
1325 // The destination register is also used as the first source operand register.
1326 class N2VDPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1327 bits<2> op17_16, bits<5> op11_7, bit op4,
1328 string OpcodeStr, string Dt,
1329 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1330 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
1331 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), IIC_VPALiD,
1332 OpcodeStr, Dt, "$dst, $src2", "$src1 = $dst",
1333 [(set DPR:$dst, (ResTy (IntOp (ResTy DPR:$src1), (OpTy DPR:$src2))))]>;
1334 class N2VQPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1335 bits<2> op17_16, bits<5> op11_7, bit op4,
1336 string OpcodeStr, string Dt,
1337 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1338 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4,
1339 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), IIC_VPALiQ,
1340 OpcodeStr, Dt, "$dst, $src2", "$src1 = $dst",
1341 [(set QPR:$dst, (ResTy (IntOp (ResTy QPR:$src1), (OpTy QPR:$src2))))]>;
1343 // Shift by immediate,
1344 // both double- and quad-register.
1345 class N2VDSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1346 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1347 ValueType Ty, SDNode OpNode>
1348 : N2VImm<op24, op23, op11_8, op7, 0, op4,
1349 (outs DPR:$dst), (ins DPR:$src, i32imm:$SIMM), f, itin,
1350 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1351 [(set DPR:$dst, (Ty (OpNode (Ty DPR:$src), (i32 imm:$SIMM))))]>;
1352 class N2VQSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1353 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1354 ValueType Ty, SDNode OpNode>
1355 : N2VImm<op24, op23, op11_8, op7, 1, op4,
1356 (outs QPR:$dst), (ins QPR:$src, i32imm:$SIMM), f, itin,
1357 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1358 [(set QPR:$dst, (Ty (OpNode (Ty QPR:$src), (i32 imm:$SIMM))))]>;
1360 // Long shift by immediate.
1361 class N2VLSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
1362 string OpcodeStr, string Dt,
1363 ValueType ResTy, ValueType OpTy, SDNode OpNode>
1364 : N2VImm<op24, op23, op11_8, op7, op6, op4,
1365 (outs QPR:$dst), (ins DPR:$src, i32imm:$SIMM), N2RegVShLFrm,
1366 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1367 [(set QPR:$dst, (ResTy (OpNode (OpTy DPR:$src),
1368 (i32 imm:$SIMM))))]>;
1370 // Narrow shift by immediate.
1371 class N2VNSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
1372 InstrItinClass itin, string OpcodeStr, string Dt,
1373 ValueType ResTy, ValueType OpTy, SDNode OpNode>
1374 : N2VImm<op24, op23, op11_8, op7, op6, op4,
1375 (outs DPR:$dst), (ins QPR:$src, i32imm:$SIMM), N2RegVShRFrm, itin,
1376 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1377 [(set DPR:$dst, (ResTy (OpNode (OpTy QPR:$src),
1378 (i32 imm:$SIMM))))]>;
1380 // Shift right by immediate and accumulate,
1381 // both double- and quad-register.
1382 class N2VDShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1383 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), N2RegVShRFrm, IIC_VPALiD,
1386 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1387 [(set DPR:$dst, (Ty (add DPR:$src1,
1388 (Ty (ShOp DPR:$src2, (i32 imm:$SIMM))))))]>;
1389 class N2VQShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1390 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1391 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$dst),
1392 (ins QPR:$src1, QPR:$src2, i32imm:$SIMM), N2RegVShRFrm, IIC_VPALiD,
1393 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1394 [(set QPR:$dst, (Ty (add QPR:$src1,
1395 (Ty (ShOp QPR:$src2, (i32 imm:$SIMM))))))]>;
1397 // Shift by immediate and insert,
1398 // both double- and quad-register.
1399 class N2VDShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1400 Format f, string OpcodeStr, string Dt, ValueType Ty,SDNode ShOp>
1401 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$dst),
1402 (ins DPR:$src1, DPR:$src2, i32imm:$SIMM), f, IIC_VSHLiD,
1403 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1404 [(set DPR:$dst, (Ty (ShOp DPR:$src1, DPR:$src2, (i32 imm:$SIMM))))]>;
1405 class N2VQShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1406 Format f, string OpcodeStr, string Dt, ValueType Ty,SDNode ShOp>
1407 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$dst),
1408 (ins QPR:$src1, QPR:$src2, i32imm:$SIMM), f, IIC_VSHLiQ,
1409 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1410 [(set QPR:$dst, (Ty (ShOp QPR:$src1, QPR:$src2, (i32 imm:$SIMM))))]>;
1412 // Convert, with fractional bits immediate,
1413 // both double- and quad-register.
1414 class N2VCvtD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1415 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1417 : N2VImm<op24, op23, op11_8, op7, 0, op4,
1418 (outs DPR:$dst), (ins DPR:$src, i32imm:$SIMM), NVCVTFrm,
1419 IIC_VUNAD, OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1420 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src), (i32 imm:$SIMM))))]>;
1421 class N2VCvtQ<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1422 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1424 : N2VImm<op24, op23, op11_8, op7, 1, op4,
1425 (outs QPR:$dst), (ins QPR:$src, i32imm:$SIMM), NVCVTFrm,
1426 IIC_VUNAQ, OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1427 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src), (i32 imm:$SIMM))))]>;
1429 //===----------------------------------------------------------------------===//
1431 //===----------------------------------------------------------------------===//
1433 // Abbreviations used in multiclass suffixes:
1434 // Q = quarter int (8 bit) elements
1435 // H = half int (16 bit) elements
1436 // S = single int (32 bit) elements
1437 // D = double int (64 bit) elements
1439 // Neon 2-register vector operations -- for disassembly only.
1441 // First with only element sizes of 8, 16 and 32 bits:
1442 multiclass N2V_QHS_cmp<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1443 bits<5> op11_7, bit op4, string opc, string Dt,
1445 // 64-bit vector types.
1446 def v8i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 0, op4,
1447 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1448 opc, !strconcat(Dt, "8"), asm, "", []>;
1449 def v4i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 0, op4,
1450 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1451 opc, !strconcat(Dt, "16"), asm, "", []>;
1452 def v2i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
1453 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1454 opc, !strconcat(Dt, "32"), asm, "", []>;
1455 def v2f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
1456 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1457 opc, "f32", asm, "", []> {
1458 let Inst{10} = 1; // overwrite F = 1
1461 // 128-bit vector types.
1462 def v16i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 1, op4,
1463 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1464 opc, !strconcat(Dt, "8"), asm, "", []>;
1465 def v8i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 1, op4,
1466 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1467 opc, !strconcat(Dt, "16"), asm, "", []>;
1468 def v4i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
1469 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1470 opc, !strconcat(Dt, "32"), asm, "", []>;
1471 def v4f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
1472 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1473 opc, "f32", asm, "", []> {
1474 let Inst{10} = 1; // overwrite F = 1
1478 // Neon 3-register vector operations.
1480 // First with only element sizes of 8, 16 and 32 bits:
1481 multiclass N3V_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1482 InstrItinClass itinD16, InstrItinClass itinD32,
1483 InstrItinClass itinQ16, InstrItinClass itinQ32,
1484 string OpcodeStr, string Dt,
1485 SDNode OpNode, bit Commutable = 0> {
1486 // 64-bit vector types.
1487 def v8i8 : N3VD<op24, op23, 0b00, op11_8, op4, itinD16,
1488 OpcodeStr, !strconcat(Dt, "8"),
1489 v8i8, v8i8, OpNode, Commutable>;
1490 def v4i16 : N3VD<op24, op23, 0b01, op11_8, op4, itinD16,
1491 OpcodeStr, !strconcat(Dt, "16"),
1492 v4i16, v4i16, OpNode, Commutable>;
1493 def v2i32 : N3VD<op24, op23, 0b10, op11_8, op4, itinD32,
1494 OpcodeStr, !strconcat(Dt, "32"),
1495 v2i32, v2i32, OpNode, Commutable>;
1497 // 128-bit vector types.
1498 def v16i8 : N3VQ<op24, op23, 0b00, op11_8, op4, itinQ16,
1499 OpcodeStr, !strconcat(Dt, "8"),
1500 v16i8, v16i8, OpNode, Commutable>;
1501 def v8i16 : N3VQ<op24, op23, 0b01, op11_8, op4, itinQ16,
1502 OpcodeStr, !strconcat(Dt, "16"),
1503 v8i16, v8i16, OpNode, Commutable>;
1504 def v4i32 : N3VQ<op24, op23, 0b10, op11_8, op4, itinQ32,
1505 OpcodeStr, !strconcat(Dt, "32"),
1506 v4i32, v4i32, OpNode, Commutable>;
1509 multiclass N3VSL_HS<bits<4> op11_8, string OpcodeStr, string Dt, SDNode ShOp> {
1510 def v4i16 : N3VDSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
1512 def v2i32 : N3VDSL<0b10, op11_8, IIC_VMULi32D, OpcodeStr, !strconcat(Dt,"32"),
1514 def v8i16 : N3VQSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
1515 v8i16, v4i16, ShOp>;
1516 def v4i32 : N3VQSL<0b10, op11_8, IIC_VMULi32Q, OpcodeStr, !strconcat(Dt,"32"),
1517 v4i32, v2i32, ShOp>;
1520 // ....then also with element size 64 bits:
1521 multiclass N3V_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1522 InstrItinClass itinD, InstrItinClass itinQ,
1523 string OpcodeStr, string Dt,
1524 SDNode OpNode, bit Commutable = 0>
1525 : N3V_QHS<op24, op23, op11_8, op4, itinD, itinD, itinQ, itinQ,
1526 OpcodeStr, Dt, OpNode, Commutable> {
1527 def v1i64 : N3VD<op24, op23, 0b11, op11_8, op4, itinD,
1528 OpcodeStr, !strconcat(Dt, "64"),
1529 v1i64, v1i64, OpNode, Commutable>;
1530 def v2i64 : N3VQ<op24, op23, 0b11, op11_8, op4, itinQ,
1531 OpcodeStr, !strconcat(Dt, "64"),
1532 v2i64, v2i64, OpNode, Commutable>;
1536 // Neon Narrowing 2-register vector intrinsics,
1537 // source operand element sizes of 16, 32 and 64 bits:
1538 multiclass N2VNInt_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1539 bits<5> op11_7, bit op6, bit op4,
1540 InstrItinClass itin, string OpcodeStr, string Dt,
1542 def v8i8 : N2VNInt<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
1543 itin, OpcodeStr, !strconcat(Dt, "16"),
1544 v8i8, v8i16, IntOp>;
1545 def v4i16 : N2VNInt<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
1546 itin, OpcodeStr, !strconcat(Dt, "32"),
1547 v4i16, v4i32, IntOp>;
1548 def v2i32 : N2VNInt<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
1549 itin, OpcodeStr, !strconcat(Dt, "64"),
1550 v2i32, v2i64, IntOp>;
1554 // Neon Lengthening 2-register vector intrinsic (currently specific to VMOVL).
1555 // source operand element sizes of 16, 32 and 64 bits:
1556 multiclass N2VL_QHS<bits<2> op24_23, bits<5> op11_7, bit op6, bit op4,
1557 string OpcodeStr, string Dt, SDNode OpNode> {
1558 def v8i16 : N2VL<op24_23, 0b00, 0b10, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
1559 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode>;
1560 def v4i32 : N2VL<op24_23, 0b01, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
1561 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
1562 def v2i64 : N2VL<op24_23, 0b10, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
1563 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
1567 // Neon 3-register vector intrinsics.
1569 // First with only element sizes of 16 and 32 bits:
1570 multiclass N3VInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
1571 InstrItinClass itinD16, InstrItinClass itinD32,
1572 InstrItinClass itinQ16, InstrItinClass itinQ32,
1573 string OpcodeStr, string Dt,
1574 Intrinsic IntOp, bit Commutable = 0> {
1575 // 64-bit vector types.
1576 def v4i16 : N3VDInt<op24, op23, 0b01, op11_8, op4, f, itinD16,
1577 OpcodeStr, !strconcat(Dt, "16"),
1578 v4i16, v4i16, IntOp, Commutable>;
1579 def v2i32 : N3VDInt<op24, op23, 0b10, op11_8, op4, f, itinD32,
1580 OpcodeStr, !strconcat(Dt, "32"),
1581 v2i32, v2i32, IntOp, Commutable>;
1583 // 128-bit vector types.
1584 def v8i16 : N3VQInt<op24, op23, 0b01, op11_8, op4, f, itinQ16,
1585 OpcodeStr, !strconcat(Dt, "16"),
1586 v8i16, v8i16, IntOp, Commutable>;
1587 def v4i32 : N3VQInt<op24, op23, 0b10, op11_8, op4, f, itinQ32,
1588 OpcodeStr, !strconcat(Dt, "32"),
1589 v4i32, v4i32, IntOp, Commutable>;
1592 multiclass N3VIntSL_HS<bits<4> op11_8,
1593 InstrItinClass itinD16, InstrItinClass itinD32,
1594 InstrItinClass itinQ16, InstrItinClass itinQ32,
1595 string OpcodeStr, string Dt, Intrinsic IntOp> {
1596 def v4i16 : N3VDIntSL16<0b01, op11_8, itinD16,
1597 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp>;
1598 def v2i32 : N3VDIntSL<0b10, op11_8, itinD32,
1599 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp>;
1600 def v8i16 : N3VQIntSL16<0b01, op11_8, itinQ16,
1601 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16, IntOp>;
1602 def v4i32 : N3VQIntSL<0b10, op11_8, itinQ32,
1603 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32, IntOp>;
1606 // ....then also with element size of 8 bits:
1607 multiclass N3VInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
1608 InstrItinClass itinD16, InstrItinClass itinD32,
1609 InstrItinClass itinQ16, InstrItinClass itinQ32,
1610 string OpcodeStr, string Dt,
1611 Intrinsic IntOp, bit Commutable = 0>
1612 : N3VInt_HS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
1613 OpcodeStr, Dt, IntOp, Commutable> {
1614 def v8i8 : N3VDInt<op24, op23, 0b00, op11_8, op4, f, itinD16,
1615 OpcodeStr, !strconcat(Dt, "8"),
1616 v8i8, v8i8, IntOp, Commutable>;
1617 def v16i8 : N3VQInt<op24, op23, 0b00, op11_8, op4, f, itinQ16,
1618 OpcodeStr, !strconcat(Dt, "8"),
1619 v16i8, v16i8, IntOp, Commutable>;
1622 // ....then also with element size of 64 bits:
1623 multiclass N3VInt_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
1624 InstrItinClass itinD16, InstrItinClass itinD32,
1625 InstrItinClass itinQ16, InstrItinClass itinQ32,
1626 string OpcodeStr, string Dt,
1627 Intrinsic IntOp, bit Commutable = 0>
1628 : N3VInt_QHS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
1629 OpcodeStr, Dt, IntOp, Commutable> {
1630 def v1i64 : N3VDInt<op24, op23, 0b11, op11_8, op4, f, itinD32,
1631 OpcodeStr, !strconcat(Dt, "64"),
1632 v1i64, v1i64, IntOp, Commutable>;
1633 def v2i64 : N3VQInt<op24, op23, 0b11, op11_8, op4, f, itinQ32,
1634 OpcodeStr, !strconcat(Dt, "64"),
1635 v2i64, v2i64, IntOp, Commutable>;
1638 // Neon Narrowing 3-register vector intrinsics,
1639 // source operand element sizes of 16, 32 and 64 bits:
1640 multiclass N3VNInt_HSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1641 string OpcodeStr, string Dt,
1642 Intrinsic IntOp, bit Commutable = 0> {
1643 def v8i8 : N3VNInt<op24, op23, 0b00, op11_8, op4,
1644 OpcodeStr, !strconcat(Dt, "16"),
1645 v8i8, v8i16, IntOp, Commutable>;
1646 def v4i16 : N3VNInt<op24, op23, 0b01, op11_8, op4,
1647 OpcodeStr, !strconcat(Dt, "32"),
1648 v4i16, v4i32, IntOp, Commutable>;
1649 def v2i32 : N3VNInt<op24, op23, 0b10, op11_8, op4,
1650 OpcodeStr, !strconcat(Dt, "64"),
1651 v2i32, v2i64, IntOp, Commutable>;
1655 // Neon Long 3-register vector intrinsics.
1657 // First with only element sizes of 16 and 32 bits:
1658 multiclass N3VLInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
1659 InstrItinClass itin16, InstrItinClass itin32,
1660 string OpcodeStr, string Dt,
1661 Intrinsic IntOp, bit Commutable = 0> {
1662 def v4i32 : N3VLInt<op24, op23, 0b01, op11_8, op4, itin16,
1663 OpcodeStr, !strconcat(Dt, "16"),
1664 v4i32, v4i16, IntOp, Commutable>;
1665 def v2i64 : N3VLInt<op24, op23, 0b10, op11_8, op4, itin32,
1666 OpcodeStr, !strconcat(Dt, "32"),
1667 v2i64, v2i32, IntOp, Commutable>;
1670 multiclass N3VLIntSL_HS<bit op24, bits<4> op11_8,
1671 InstrItinClass itin, string OpcodeStr, string Dt,
1673 def v4i16 : N3VLIntSL16<op24, 0b01, op11_8, itin,
1674 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
1675 def v2i32 : N3VLIntSL<op24, 0b10, op11_8, itin,
1676 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1679 // ....then also with element size of 8 bits:
1680 multiclass N3VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1681 InstrItinClass itin16, InstrItinClass itin32,
1682 string OpcodeStr, string Dt,
1683 Intrinsic IntOp, bit Commutable = 0>
1684 : N3VLInt_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt,
1685 IntOp, Commutable> {
1686 def v8i16 : N3VLInt<op24, op23, 0b00, op11_8, op4, itin16,
1687 OpcodeStr, !strconcat(Dt, "8"),
1688 v8i16, v8i8, IntOp, Commutable>;
1692 // Neon Wide 3-register vector intrinsics,
1693 // source operand element sizes of 8, 16 and 32 bits:
1694 multiclass N3VWInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1695 string OpcodeStr, string Dt,
1696 Intrinsic IntOp, bit Commutable = 0> {
1697 def v8i16 : N3VWInt<op24, op23, 0b00, op11_8, op4,
1698 OpcodeStr, !strconcat(Dt, "8"),
1699 v8i16, v8i8, IntOp, Commutable>;
1700 def v4i32 : N3VWInt<op24, op23, 0b01, op11_8, op4,
1701 OpcodeStr, !strconcat(Dt, "16"),
1702 v4i32, v4i16, IntOp, Commutable>;
1703 def v2i64 : N3VWInt<op24, op23, 0b10, op11_8, op4,
1704 OpcodeStr, !strconcat(Dt, "32"),
1705 v2i64, v2i32, IntOp, Commutable>;
1709 // Neon Multiply-Op vector operations,
1710 // element sizes of 8, 16 and 32 bits:
1711 multiclass N3VMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1712 InstrItinClass itinD16, InstrItinClass itinD32,
1713 InstrItinClass itinQ16, InstrItinClass itinQ32,
1714 string OpcodeStr, string Dt, SDNode OpNode> {
1715 // 64-bit vector types.
1716 def v8i8 : N3VDMulOp<op24, op23, 0b00, op11_8, op4, itinD16,
1717 OpcodeStr, !strconcat(Dt, "8"), v8i8, mul, OpNode>;
1718 def v4i16 : N3VDMulOp<op24, op23, 0b01, op11_8, op4, itinD16,
1719 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, OpNode>;
1720 def v2i32 : N3VDMulOp<op24, op23, 0b10, op11_8, op4, itinD32,
1721 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, OpNode>;
1723 // 128-bit vector types.
1724 def v16i8 : N3VQMulOp<op24, op23, 0b00, op11_8, op4, itinQ16,
1725 OpcodeStr, !strconcat(Dt, "8"), v16i8, mul, OpNode>;
1726 def v8i16 : N3VQMulOp<op24, op23, 0b01, op11_8, op4, itinQ16,
1727 OpcodeStr, !strconcat(Dt, "16"), v8i16, mul, OpNode>;
1728 def v4i32 : N3VQMulOp<op24, op23, 0b10, op11_8, op4, itinQ32,
1729 OpcodeStr, !strconcat(Dt, "32"), v4i32, mul, OpNode>;
1732 multiclass N3VMulOpSL_HS<bits<4> op11_8,
1733 InstrItinClass itinD16, InstrItinClass itinD32,
1734 InstrItinClass itinQ16, InstrItinClass itinQ32,
1735 string OpcodeStr, string Dt, SDNode ShOp> {
1736 def v4i16 : N3VDMulOpSL16<0b01, op11_8, itinD16,
1737 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, ShOp>;
1738 def v2i32 : N3VDMulOpSL<0b10, op11_8, itinD32,
1739 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, ShOp>;
1740 def v8i16 : N3VQMulOpSL16<0b01, op11_8, itinQ16,
1741 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16,
1743 def v4i32 : N3VQMulOpSL<0b10, op11_8, itinQ32,
1744 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32,
1748 // Neon 3-argument intrinsics,
1749 // element sizes of 8, 16 and 32 bits:
1750 multiclass N3VInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1751 InstrItinClass itinD, InstrItinClass itinQ,
1752 string OpcodeStr, string Dt, Intrinsic IntOp> {
1753 // 64-bit vector types.
1754 def v8i8 : N3VDInt3<op24, op23, 0b00, op11_8, op4, itinD,
1755 OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
1756 def v4i16 : N3VDInt3<op24, op23, 0b01, op11_8, op4, itinD,
1757 OpcodeStr, !strconcat(Dt, "16"), v4i16, v4i16, IntOp>;
1758 def v2i32 : N3VDInt3<op24, op23, 0b10, op11_8, op4, itinD,
1759 OpcodeStr, !strconcat(Dt, "32"), v2i32, v2i32, IntOp>;
1761 // 128-bit vector types.
1762 def v16i8 : N3VQInt3<op24, op23, 0b00, op11_8, op4, itinQ,
1763 OpcodeStr, !strconcat(Dt, "8"), v16i8, v16i8, IntOp>;
1764 def v8i16 : N3VQInt3<op24, op23, 0b01, op11_8, op4, itinQ,
1765 OpcodeStr, !strconcat(Dt, "16"), v8i16, v8i16, IntOp>;
1766 def v4i32 : N3VQInt3<op24, op23, 0b10, op11_8, op4, itinQ,
1767 OpcodeStr, !strconcat(Dt, "32"), v4i32, v4i32, IntOp>;
1771 // Neon Long 3-argument intrinsics.
1773 // First with only element sizes of 16 and 32 bits:
1774 multiclass N3VLInt3_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
1775 InstrItinClass itin16, InstrItinClass itin32,
1776 string OpcodeStr, string Dt, Intrinsic IntOp> {
1777 def v4i32 : N3VLInt3<op24, op23, 0b01, op11_8, op4, itin16,
1778 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
1779 def v2i64 : N3VLInt3<op24, op23, 0b10, op11_8, op4, itin32,
1780 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1783 multiclass N3VLInt3SL_HS<bit op24, bits<4> op11_8,
1784 string OpcodeStr, string Dt, Intrinsic IntOp> {
1785 def v4i16 : N3VLInt3SL16<op24, 0b01, op11_8, IIC_VMACi16D,
1786 OpcodeStr, !strconcat(Dt,"16"), v4i32, v4i16, IntOp>;
1787 def v2i32 : N3VLInt3SL<op24, 0b10, op11_8, IIC_VMACi32D,
1788 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1791 // ....then also with element size of 8 bits:
1792 multiclass N3VLInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1793 InstrItinClass itin16, InstrItinClass itin32,
1794 string OpcodeStr, string Dt, Intrinsic IntOp>
1795 : N3VLInt3_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt, IntOp> {
1796 def v8i16 : N3VLInt3<op24, op23, 0b00, op11_8, op4, itin16,
1797 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
1801 // Neon 2-register vector intrinsics,
1802 // element sizes of 8, 16 and 32 bits:
1803 multiclass N2VInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1804 bits<5> op11_7, bit op4,
1805 InstrItinClass itinD, InstrItinClass itinQ,
1806 string OpcodeStr, string Dt, Intrinsic IntOp> {
1807 // 64-bit vector types.
1808 def v8i8 : N2VDInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1809 itinD, OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
1810 def v4i16 : N2VDInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1811 itinD, OpcodeStr, !strconcat(Dt, "16"),v4i16,v4i16,IntOp>;
1812 def v2i32 : N2VDInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1813 itinD, OpcodeStr, !strconcat(Dt, "32"),v2i32,v2i32,IntOp>;
1815 // 128-bit vector types.
1816 def v16i8 : N2VQInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1817 itinQ, OpcodeStr, !strconcat(Dt, "8"), v16i8,v16i8,IntOp>;
1818 def v8i16 : N2VQInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1819 itinQ, OpcodeStr, !strconcat(Dt, "16"),v8i16,v8i16,IntOp>;
1820 def v4i32 : N2VQInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1821 itinQ, OpcodeStr, !strconcat(Dt, "32"),v4i32,v4i32,IntOp>;
1825 // Neon Pairwise long 2-register intrinsics,
1826 // element sizes of 8, 16 and 32 bits:
1827 multiclass N2VPLInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1828 bits<5> op11_7, bit op4,
1829 string OpcodeStr, string Dt, Intrinsic IntOp> {
1830 // 64-bit vector types.
1831 def v8i8 : N2VDPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1832 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
1833 def v4i16 : N2VDPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1834 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
1835 def v2i32 : N2VDPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1836 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
1838 // 128-bit vector types.
1839 def v16i8 : N2VQPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1840 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
1841 def v8i16 : N2VQPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1842 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
1843 def v4i32 : N2VQPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1844 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
1848 // Neon Pairwise long 2-register accumulate intrinsics,
1849 // element sizes of 8, 16 and 32 bits:
1850 multiclass N2VPLInt2_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1851 bits<5> op11_7, bit op4,
1852 string OpcodeStr, string Dt, Intrinsic IntOp> {
1853 // 64-bit vector types.
1854 def v8i8 : N2VDPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1855 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
1856 def v4i16 : N2VDPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1857 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
1858 def v2i32 : N2VDPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1859 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
1861 // 128-bit vector types.
1862 def v16i8 : N2VQPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1863 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
1864 def v8i16 : N2VQPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1865 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
1866 def v4i32 : N2VQPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1867 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
1871 // Neon 2-register vector shift by immediate,
1872 // with f of either N2RegVShLFrm or N2RegVShRFrm
1873 // element sizes of 8, 16, 32 and 64 bits:
1874 multiclass N2VSh_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1875 InstrItinClass itin, string OpcodeStr, string Dt,
1876 SDNode OpNode, Format f> {
1877 // 64-bit vector types.
1878 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
1879 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
1880 let Inst{21-19} = 0b001; // imm6 = 001xxx
1882 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
1883 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
1884 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1886 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
1887 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
1888 let Inst{21} = 0b1; // imm6 = 1xxxxx
1890 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, f, itin,
1891 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
1894 // 128-bit vector types.
1895 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
1896 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
1897 let Inst{21-19} = 0b001; // imm6 = 001xxx
1899 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
1900 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
1901 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1903 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
1904 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
1905 let Inst{21} = 0b1; // imm6 = 1xxxxx
1907 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, f, itin,
1908 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
1912 // Neon Shift-Accumulate vector operations,
1913 // element sizes of 8, 16, 32 and 64 bits:
1914 multiclass N2VShAdd_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1915 string OpcodeStr, string Dt, SDNode ShOp> {
1916 // 64-bit vector types.
1917 def v8i8 : N2VDShAdd<op24, op23, op11_8, 0, op4,
1918 OpcodeStr, !strconcat(Dt, "8"), v8i8, ShOp> {
1919 let Inst{21-19} = 0b001; // imm6 = 001xxx
1921 def v4i16 : N2VDShAdd<op24, op23, op11_8, 0, op4,
1922 OpcodeStr, !strconcat(Dt, "16"), v4i16, ShOp> {
1923 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1925 def v2i32 : N2VDShAdd<op24, op23, op11_8, 0, op4,
1926 OpcodeStr, !strconcat(Dt, "32"), v2i32, ShOp> {
1927 let Inst{21} = 0b1; // imm6 = 1xxxxx
1929 def v1i64 : N2VDShAdd<op24, op23, op11_8, 1, op4,
1930 OpcodeStr, !strconcat(Dt, "64"), v1i64, ShOp>;
1933 // 128-bit vector types.
1934 def v16i8 : N2VQShAdd<op24, op23, op11_8, 0, op4,
1935 OpcodeStr, !strconcat(Dt, "8"), v16i8, ShOp> {
1936 let Inst{21-19} = 0b001; // imm6 = 001xxx
1938 def v8i16 : N2VQShAdd<op24, op23, op11_8, 0, op4,
1939 OpcodeStr, !strconcat(Dt, "16"), v8i16, ShOp> {
1940 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1942 def v4i32 : N2VQShAdd<op24, op23, op11_8, 0, op4,
1943 OpcodeStr, !strconcat(Dt, "32"), v4i32, ShOp> {
1944 let Inst{21} = 0b1; // imm6 = 1xxxxx
1946 def v2i64 : N2VQShAdd<op24, op23, op11_8, 1, op4,
1947 OpcodeStr, !strconcat(Dt, "64"), v2i64, ShOp>;
1952 // Neon Shift-Insert vector operations,
1953 // with f of either N2RegVShLFrm or N2RegVShRFrm
1954 // element sizes of 8, 16, 32 and 64 bits:
1955 multiclass N2VShIns_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1956 string OpcodeStr, SDNode ShOp,
1958 // 64-bit vector types.
1959 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4,
1960 f, OpcodeStr, "8", v8i8, ShOp> {
1961 let Inst{21-19} = 0b001; // imm6 = 001xxx
1963 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4,
1964 f, OpcodeStr, "16", v4i16, ShOp> {
1965 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1967 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4,
1968 f, OpcodeStr, "32", v2i32, ShOp> {
1969 let Inst{21} = 0b1; // imm6 = 1xxxxx
1971 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4,
1972 f, OpcodeStr, "64", v1i64, ShOp>;
1975 // 128-bit vector types.
1976 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4,
1977 f, OpcodeStr, "8", v16i8, ShOp> {
1978 let Inst{21-19} = 0b001; // imm6 = 001xxx
1980 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4,
1981 f, OpcodeStr, "16", v8i16, ShOp> {
1982 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1984 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4,
1985 f, OpcodeStr, "32", v4i32, ShOp> {
1986 let Inst{21} = 0b1; // imm6 = 1xxxxx
1988 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4,
1989 f, OpcodeStr, "64", v2i64, ShOp>;
1993 // Neon Shift Long operations,
1994 // element sizes of 8, 16, 32 bits:
1995 multiclass N2VLSh_QHS<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
1996 bit op4, string OpcodeStr, string Dt, SDNode OpNode> {
1997 def v8i16 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
1998 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode> {
1999 let Inst{21-19} = 0b001; // imm6 = 001xxx
2001 def v4i32 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
2002 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode> {
2003 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2005 def v2i64 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
2006 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode> {
2007 let Inst{21} = 0b1; // imm6 = 1xxxxx
2011 // Neon Shift Narrow operations,
2012 // element sizes of 16, 32, 64 bits:
2013 multiclass N2VNSh_HSD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
2014 bit op4, InstrItinClass itin, string OpcodeStr, string Dt,
2016 def v8i8 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
2017 OpcodeStr, !strconcat(Dt, "16"), v8i8, v8i16, OpNode> {
2018 let Inst{21-19} = 0b001; // imm6 = 001xxx
2020 def v4i16 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
2021 OpcodeStr, !strconcat(Dt, "32"), v4i16, v4i32, OpNode> {
2022 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2024 def v2i32 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
2025 OpcodeStr, !strconcat(Dt, "64"), v2i32, v2i64, OpNode> {
2026 let Inst{21} = 0b1; // imm6 = 1xxxxx
2030 //===----------------------------------------------------------------------===//
2031 // Instruction Definitions.
2032 //===----------------------------------------------------------------------===//
2034 // Vector Add Operations.
2036 // VADD : Vector Add (integer and floating-point)
2037 defm VADD : N3V_QHSD<0, 0, 0b1000, 0, IIC_VBINiD, IIC_VBINiQ, "vadd", "i",
2039 def VADDfd : N3VD<0, 0, 0b00, 0b1101, 0, IIC_VBIND, "vadd", "f32",
2040 v2f32, v2f32, fadd, 1>;
2041 def VADDfq : N3VQ<0, 0, 0b00, 0b1101, 0, IIC_VBINQ, "vadd", "f32",
2042 v4f32, v4f32, fadd, 1>;
2043 // VADDL : Vector Add Long (Q = D + D)
2044 defm VADDLs : N3VLInt_QHS<0,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
2045 "vaddl", "s", int_arm_neon_vaddls, 1>;
2046 defm VADDLu : N3VLInt_QHS<1,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
2047 "vaddl", "u", int_arm_neon_vaddlu, 1>;
2048 // VADDW : Vector Add Wide (Q = Q + D)
2049 defm VADDWs : N3VWInt_QHS<0,1,0b0001,0, "vaddw", "s", int_arm_neon_vaddws, 0>;
2050 defm VADDWu : N3VWInt_QHS<1,1,0b0001,0, "vaddw", "u", int_arm_neon_vaddwu, 0>;
2051 // VHADD : Vector Halving Add
2052 defm VHADDs : N3VInt_QHS<0, 0, 0b0000, 0, N3RegFrm,
2053 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2054 "vhadd", "s", int_arm_neon_vhadds, 1>;
2055 defm VHADDu : N3VInt_QHS<1, 0, 0b0000, 0, N3RegFrm,
2056 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2057 "vhadd", "u", int_arm_neon_vhaddu, 1>;
2058 // VRHADD : Vector Rounding Halving Add
2059 defm VRHADDs : N3VInt_QHS<0, 0, 0b0001, 0, N3RegFrm,
2060 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2061 "vrhadd", "s", int_arm_neon_vrhadds, 1>;
2062 defm VRHADDu : N3VInt_QHS<1, 0, 0b0001, 0, N3RegFrm,
2063 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2064 "vrhadd", "u", int_arm_neon_vrhaddu, 1>;
2065 // VQADD : Vector Saturating Add
2066 defm VQADDs : N3VInt_QHSD<0, 0, 0b0000, 1, N3RegFrm,
2067 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2068 "vqadd", "s", int_arm_neon_vqadds, 1>;
2069 defm VQADDu : N3VInt_QHSD<1, 0, 0b0000, 1, N3RegFrm,
2070 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2071 "vqadd", "u", int_arm_neon_vqaddu, 1>;
2072 // VADDHN : Vector Add and Narrow Returning High Half (D = Q + Q)
2073 defm VADDHN : N3VNInt_HSD<0,1,0b0100,0, "vaddhn", "i",
2074 int_arm_neon_vaddhn, 1>;
2075 // VRADDHN : Vector Rounding Add and Narrow Returning High Half (D = Q + Q)
2076 defm VRADDHN : N3VNInt_HSD<1,1,0b0100,0, "vraddhn", "i",
2077 int_arm_neon_vraddhn, 1>;
2079 // Vector Multiply Operations.
2081 // VMUL : Vector Multiply (integer, polynomial and floating-point)
2082 defm VMUL : N3V_QHS<0, 0, 0b1001, 1, IIC_VMULi16D, IIC_VMULi32D,
2083 IIC_VMULi16Q, IIC_VMULi32Q, "vmul", "i", mul, 1>;
2084 def VMULpd : N3VDInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16D, "vmul",
2085 "p8", v8i8, v8i8, int_arm_neon_vmulp, 1>;
2086 def VMULpq : N3VQInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16Q, "vmul",
2087 "p8", v16i8, v16i8, int_arm_neon_vmulp, 1>;
2088 def VMULfd : N3VD<1, 0, 0b00, 0b1101, 1, IIC_VBIND, "vmul", "f32",
2089 v2f32, v2f32, fmul, 1>;
2090 def VMULfq : N3VQ<1, 0, 0b00, 0b1101, 1, IIC_VBINQ, "vmul", "f32",
2091 v4f32, v4f32, fmul, 1>;
2092 defm VMULsl : N3VSL_HS<0b1000, "vmul", "i", mul>;
2093 def VMULslfd : N3VDSL<0b10, 0b1001, IIC_VBIND, "vmul", "f32", v2f32, fmul>;
2094 def VMULslfq : N3VQSL<0b10, 0b1001, IIC_VBINQ, "vmul", "f32", v4f32,
2097 def : Pat<(v8i16 (mul (v8i16 QPR:$src1),
2098 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
2099 (v8i16 (VMULslv8i16 (v8i16 QPR:$src1),
2100 (v4i16 (EXTRACT_SUBREG QPR:$src2,
2101 (DSubReg_i16_reg imm:$lane))),
2102 (SubReg_i16_lane imm:$lane)))>;
2103 def : Pat<(v4i32 (mul (v4i32 QPR:$src1),
2104 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
2105 (v4i32 (VMULslv4i32 (v4i32 QPR:$src1),
2106 (v2i32 (EXTRACT_SUBREG QPR:$src2,
2107 (DSubReg_i32_reg imm:$lane))),
2108 (SubReg_i32_lane imm:$lane)))>;
2109 def : Pat<(v4f32 (fmul (v4f32 QPR:$src1),
2110 (v4f32 (NEONvduplane (v4f32 QPR:$src2), imm:$lane)))),
2111 (v4f32 (VMULslfq (v4f32 QPR:$src1),
2112 (v2f32 (EXTRACT_SUBREG QPR:$src2,
2113 (DSubReg_i32_reg imm:$lane))),
2114 (SubReg_i32_lane imm:$lane)))>;
2116 // VQDMULH : Vector Saturating Doubling Multiply Returning High Half
2117 defm VQDMULH : N3VInt_HS<0, 0, 0b1011, 0, N3RegFrm, IIC_VMULi16D, IIC_VMULi32D,
2118 IIC_VMULi16Q, IIC_VMULi32Q,
2119 "vqdmulh", "s", int_arm_neon_vqdmulh, 1>;
2120 defm VQDMULHsl: N3VIntSL_HS<0b1100, IIC_VMULi16D, IIC_VMULi32D,
2121 IIC_VMULi16Q, IIC_VMULi32Q,
2122 "vqdmulh", "s", int_arm_neon_vqdmulh>;
2123 def : Pat<(v8i16 (int_arm_neon_vqdmulh (v8i16 QPR:$src1),
2124 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
2126 (v8i16 (VQDMULHslv8i16 (v8i16 QPR:$src1),
2127 (v4i16 (EXTRACT_SUBREG QPR:$src2,
2128 (DSubReg_i16_reg imm:$lane))),
2129 (SubReg_i16_lane imm:$lane)))>;
2130 def : Pat<(v4i32 (int_arm_neon_vqdmulh (v4i32 QPR:$src1),
2131 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
2133 (v4i32 (VQDMULHslv4i32 (v4i32 QPR:$src1),
2134 (v2i32 (EXTRACT_SUBREG QPR:$src2,
2135 (DSubReg_i32_reg imm:$lane))),
2136 (SubReg_i32_lane imm:$lane)))>;
2138 // VQRDMULH : Vector Rounding Saturating Doubling Multiply Returning High Half
2139 defm VQRDMULH : N3VInt_HS<1, 0, 0b1011, 0, N3RegFrm,
2140 IIC_VMULi16D,IIC_VMULi32D,IIC_VMULi16Q,IIC_VMULi32Q,
2141 "vqrdmulh", "s", int_arm_neon_vqrdmulh, 1>;
2142 defm VQRDMULHsl : N3VIntSL_HS<0b1101, IIC_VMULi16D, IIC_VMULi32D,
2143 IIC_VMULi16Q, IIC_VMULi32Q,
2144 "vqrdmulh", "s", int_arm_neon_vqrdmulh>;
2145 def : Pat<(v8i16 (int_arm_neon_vqrdmulh (v8i16 QPR:$src1),
2146 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
2148 (v8i16 (VQRDMULHslv8i16 (v8i16 QPR:$src1),
2149 (v4i16 (EXTRACT_SUBREG QPR:$src2,
2150 (DSubReg_i16_reg imm:$lane))),
2151 (SubReg_i16_lane imm:$lane)))>;
2152 def : Pat<(v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$src1),
2153 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
2155 (v4i32 (VQRDMULHslv4i32 (v4i32 QPR:$src1),
2156 (v2i32 (EXTRACT_SUBREG QPR:$src2,
2157 (DSubReg_i32_reg imm:$lane))),
2158 (SubReg_i32_lane imm:$lane)))>;
2160 // VMULL : Vector Multiply Long (integer and polynomial) (Q = D * D)
2161 defm VMULLs : N3VLInt_QHS<0,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
2162 "vmull", "s", int_arm_neon_vmulls, 1>;
2163 defm VMULLu : N3VLInt_QHS<1,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
2164 "vmull", "u", int_arm_neon_vmullu, 1>;
2165 def VMULLp : N3VLInt<0, 1, 0b00, 0b1110, 0, IIC_VMULi16D, "vmull", "p8",
2166 v8i16, v8i8, int_arm_neon_vmullp, 1>;
2167 defm VMULLsls : N3VLIntSL_HS<0, 0b1010, IIC_VMULi16D, "vmull", "s",
2168 int_arm_neon_vmulls>;
2169 defm VMULLslu : N3VLIntSL_HS<1, 0b1010, IIC_VMULi16D, "vmull", "u",
2170 int_arm_neon_vmullu>;
2172 // VQDMULL : Vector Saturating Doubling Multiply Long (Q = D * D)
2173 defm VQDMULL : N3VLInt_HS<0,1,0b1101,0, IIC_VMULi16D, IIC_VMULi32D,
2174 "vqdmull", "s", int_arm_neon_vqdmull, 1>;
2175 defm VQDMULLsl: N3VLIntSL_HS<0, 0b1011, IIC_VMULi16D,
2176 "vqdmull", "s", int_arm_neon_vqdmull>;
2178 // Vector Multiply-Accumulate and Multiply-Subtract Operations.
2180 // VMLA : Vector Multiply Accumulate (integer and floating-point)
2181 defm VMLA : N3VMulOp_QHS<0, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
2182 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
2183 def VMLAfd : N3VDMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACD, "vmla", "f32",
2185 def VMLAfq : N3VQMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACQ, "vmla", "f32",
2187 defm VMLAsl : N3VMulOpSL_HS<0b0000, IIC_VMACi16D, IIC_VMACi32D,
2188 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
2189 def VMLAslfd : N3VDMulOpSL<0b10, 0b0001, IIC_VMACD, "vmla", "f32",
2191 def VMLAslfq : N3VQMulOpSL<0b10, 0b0001, IIC_VMACQ, "vmla", "f32",
2192 v4f32, v2f32, fmul, fadd>;
2194 def : Pat<(v8i16 (add (v8i16 QPR:$src1),
2195 (mul (v8i16 QPR:$src2),
2196 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
2197 (v8i16 (VMLAslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
2198 (v4i16 (EXTRACT_SUBREG QPR:$src3,
2199 (DSubReg_i16_reg imm:$lane))),
2200 (SubReg_i16_lane imm:$lane)))>;
2202 def : Pat<(v4i32 (add (v4i32 QPR:$src1),
2203 (mul (v4i32 QPR:$src2),
2204 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
2205 (v4i32 (VMLAslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
2206 (v2i32 (EXTRACT_SUBREG QPR:$src3,
2207 (DSubReg_i32_reg imm:$lane))),
2208 (SubReg_i32_lane imm:$lane)))>;
2210 def : Pat<(v4f32 (fadd (v4f32 QPR:$src1),
2211 (fmul (v4f32 QPR:$src2),
2212 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
2213 (v4f32 (VMLAslfq (v4f32 QPR:$src1),
2215 (v2f32 (EXTRACT_SUBREG QPR:$src3,
2216 (DSubReg_i32_reg imm:$lane))),
2217 (SubReg_i32_lane imm:$lane)))>;
2219 // VMLAL : Vector Multiply Accumulate Long (Q += D * D)
2220 defm VMLALs : N3VLInt3_QHS<0,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
2221 "vmlal", "s", int_arm_neon_vmlals>;
2222 defm VMLALu : N3VLInt3_QHS<1,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
2223 "vmlal", "u", int_arm_neon_vmlalu>;
2225 defm VMLALsls : N3VLInt3SL_HS<0, 0b0010, "vmlal", "s", int_arm_neon_vmlals>;
2226 defm VMLALslu : N3VLInt3SL_HS<1, 0b0010, "vmlal", "u", int_arm_neon_vmlalu>;
2228 // VQDMLAL : Vector Saturating Doubling Multiply Accumulate Long (Q += D * D)
2229 defm VQDMLAL : N3VLInt3_HS<0, 1, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
2230 "vqdmlal", "s", int_arm_neon_vqdmlal>;
2231 defm VQDMLALsl: N3VLInt3SL_HS<0, 0b0011, "vqdmlal", "s", int_arm_neon_vqdmlal>;
2233 // VMLS : Vector Multiply Subtract (integer and floating-point)
2234 defm VMLS : N3VMulOp_QHS<1, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
2235 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
2236 def VMLSfd : N3VDMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACD, "vmls", "f32",
2238 def VMLSfq : N3VQMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACQ, "vmls", "f32",
2240 defm VMLSsl : N3VMulOpSL_HS<0b0100, IIC_VMACi16D, IIC_VMACi32D,
2241 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
2242 def VMLSslfd : N3VDMulOpSL<0b10, 0b0101, IIC_VMACD, "vmls", "f32",
2244 def VMLSslfq : N3VQMulOpSL<0b10, 0b0101, IIC_VMACQ, "vmls", "f32",
2245 v4f32, v2f32, fmul, fsub>;
2247 def : Pat<(v8i16 (sub (v8i16 QPR:$src1),
2248 (mul (v8i16 QPR:$src2),
2249 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
2250 (v8i16 (VMLSslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
2251 (v4i16 (EXTRACT_SUBREG QPR:$src3,
2252 (DSubReg_i16_reg imm:$lane))),
2253 (SubReg_i16_lane imm:$lane)))>;
2255 def : Pat<(v4i32 (sub (v4i32 QPR:$src1),
2256 (mul (v4i32 QPR:$src2),
2257 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
2258 (v4i32 (VMLSslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
2259 (v2i32 (EXTRACT_SUBREG QPR:$src3,
2260 (DSubReg_i32_reg imm:$lane))),
2261 (SubReg_i32_lane imm:$lane)))>;
2263 def : Pat<(v4f32 (fsub (v4f32 QPR:$src1),
2264 (fmul (v4f32 QPR:$src2),
2265 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
2266 (v4f32 (VMLSslfq (v4f32 QPR:$src1), (v4f32 QPR:$src2),
2267 (v2f32 (EXTRACT_SUBREG QPR:$src3,
2268 (DSubReg_i32_reg imm:$lane))),
2269 (SubReg_i32_lane imm:$lane)))>;
2271 // VMLSL : Vector Multiply Subtract Long (Q -= D * D)
2272 defm VMLSLs : N3VLInt3_QHS<0,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
2273 "vmlsl", "s", int_arm_neon_vmlsls>;
2274 defm VMLSLu : N3VLInt3_QHS<1,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
2275 "vmlsl", "u", int_arm_neon_vmlslu>;
2277 defm VMLSLsls : N3VLInt3SL_HS<0, 0b0110, "vmlsl", "s", int_arm_neon_vmlsls>;
2278 defm VMLSLslu : N3VLInt3SL_HS<1, 0b0110, "vmlsl", "u", int_arm_neon_vmlslu>;
2280 // VQDMLSL : Vector Saturating Doubling Multiply Subtract Long (Q -= D * D)
2281 defm VQDMLSL : N3VLInt3_HS<0, 1, 0b1011, 0, IIC_VMACi16D, IIC_VMACi32D,
2282 "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
2283 defm VQDMLSLsl: N3VLInt3SL_HS<0, 0b111, "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
2285 // Vector Subtract Operations.
2287 // VSUB : Vector Subtract (integer and floating-point)
2288 defm VSUB : N3V_QHSD<1, 0, 0b1000, 0, IIC_VSUBiD, IIC_VSUBiQ,
2289 "vsub", "i", sub, 0>;
2290 def VSUBfd : N3VD<0, 0, 0b10, 0b1101, 0, IIC_VBIND, "vsub", "f32",
2291 v2f32, v2f32, fsub, 0>;
2292 def VSUBfq : N3VQ<0, 0, 0b10, 0b1101, 0, IIC_VBINQ, "vsub", "f32",
2293 v4f32, v4f32, fsub, 0>;
2294 // VSUBL : Vector Subtract Long (Q = D - D)
2295 defm VSUBLs : N3VLInt_QHS<0,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
2296 "vsubl", "s", int_arm_neon_vsubls, 1>;
2297 defm VSUBLu : N3VLInt_QHS<1,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
2298 "vsubl", "u", int_arm_neon_vsublu, 1>;
2299 // VSUBW : Vector Subtract Wide (Q = Q - D)
2300 defm VSUBWs : N3VWInt_QHS<0,1,0b0011,0, "vsubw", "s", int_arm_neon_vsubws, 0>;
2301 defm VSUBWu : N3VWInt_QHS<1,1,0b0011,0, "vsubw", "u", int_arm_neon_vsubwu, 0>;
2302 // VHSUB : Vector Halving Subtract
2303 defm VHSUBs : N3VInt_QHS<0, 0, 0b0010, 0, N3RegFrm,
2304 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2305 "vhsub", "s", int_arm_neon_vhsubs, 0>;
2306 defm VHSUBu : N3VInt_QHS<1, 0, 0b0010, 0, N3RegFrm,
2307 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2308 "vhsub", "u", int_arm_neon_vhsubu, 0>;
2309 // VQSUB : Vector Saturing Subtract
2310 defm VQSUBs : N3VInt_QHSD<0, 0, 0b0010, 1, N3RegFrm,
2311 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2312 "vqsub", "s", int_arm_neon_vqsubs, 0>;
2313 defm VQSUBu : N3VInt_QHSD<1, 0, 0b0010, 1, N3RegFrm,
2314 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2315 "vqsub", "u", int_arm_neon_vqsubu, 0>;
2316 // VSUBHN : Vector Subtract and Narrow Returning High Half (D = Q - Q)
2317 defm VSUBHN : N3VNInt_HSD<0,1,0b0110,0, "vsubhn", "i",
2318 int_arm_neon_vsubhn, 0>;
2319 // VRSUBHN : Vector Rounding Subtract and Narrow Returning High Half (D=Q-Q)
2320 defm VRSUBHN : N3VNInt_HSD<1,1,0b0110,0, "vrsubhn", "i",
2321 int_arm_neon_vrsubhn, 0>;
2323 // Vector Comparisons.
2325 // VCEQ : Vector Compare Equal
2326 defm VCEQ : N3V_QHS<1, 0, 0b1000, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
2327 IIC_VSUBi4Q, "vceq", "i", NEONvceq, 1>;
2328 def VCEQfd : N3VD<0,0,0b00,0b1110,0, IIC_VBIND, "vceq", "f32", v2i32, v2f32,
2330 def VCEQfq : N3VQ<0,0,0b00,0b1110,0, IIC_VBINQ, "vceq", "f32", v4i32, v4f32,
2332 // For disassembly only.
2333 defm VCEQz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00010, 0, "vceq", "i",
2336 // VCGE : Vector Compare Greater Than or Equal
2337 defm VCGEs : N3V_QHS<0, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
2338 IIC_VSUBi4Q, "vcge", "s", NEONvcge, 0>;
2339 defm VCGEu : N3V_QHS<1, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
2340 IIC_VSUBi4Q, "vcge", "u", NEONvcgeu, 0>;
2341 def VCGEfd : N3VD<1,0,0b00,0b1110,0, IIC_VBIND, "vcge", "f32", v2i32, v2f32,
2343 def VCGEfq : N3VQ<1,0,0b00,0b1110,0, IIC_VBINQ, "vcge", "f32", v4i32, v4f32,
2345 // For disassembly only.
2346 defm VCGEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00001, 0, "vcge", "s",
2348 // For disassembly only.
2349 defm VCLEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00011, 0, "vcle", "s",
2352 // VCGT : Vector Compare Greater Than
2353 defm VCGTs : N3V_QHS<0, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
2354 IIC_VSUBi4Q, "vcgt", "s", NEONvcgt, 0>;
2355 defm VCGTu : N3V_QHS<1, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
2356 IIC_VSUBi4Q, "vcgt", "u", NEONvcgtu, 0>;
2357 def VCGTfd : N3VD<1,0,0b10,0b1110,0, IIC_VBIND, "vcgt", "f32", v2i32, v2f32,
2359 def VCGTfq : N3VQ<1,0,0b10,0b1110,0, IIC_VBINQ, "vcgt", "f32", v4i32, v4f32,
2361 // For disassembly only.
2362 defm VCGTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00000, 0, "vcgt", "s",
2364 // For disassembly only.
2365 defm VCLTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00100, 0, "vclt", "s",
2368 // VACGE : Vector Absolute Compare Greater Than or Equal (aka VCAGE)
2369 def VACGEd : N3VDInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacge",
2370 "f32", v2i32, v2f32, int_arm_neon_vacged, 0>;
2371 def VACGEq : N3VQInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacge",
2372 "f32", v4i32, v4f32, int_arm_neon_vacgeq, 0>;
2373 // VACGT : Vector Absolute Compare Greater Than (aka VCAGT)
2374 def VACGTd : N3VDInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacgt",
2375 "f32", v2i32, v2f32, int_arm_neon_vacgtd, 0>;
2376 def VACGTq : N3VQInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacgt",
2377 "f32", v4i32, v4f32, int_arm_neon_vacgtq, 0>;
2378 // VTST : Vector Test Bits
2379 defm VTST : N3V_QHS<0, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2380 IIC_VBINi4Q, "vtst", "", NEONvtst, 1>;
2382 // Vector Bitwise Operations.
2384 def vnotd : PatFrag<(ops node:$in),
2385 (xor node:$in, (bitconvert (v8i8 NEONimmAllOnesV)))>;
2386 def vnotq : PatFrag<(ops node:$in),
2387 (xor node:$in, (bitconvert (v16i8 NEONimmAllOnesV)))>;
2390 // VAND : Vector Bitwise AND
2391 def VANDd : N3VDX<0, 0, 0b00, 0b0001, 1, IIC_VBINiD, "vand",
2392 v2i32, v2i32, and, 1>;
2393 def VANDq : N3VQX<0, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "vand",
2394 v4i32, v4i32, and, 1>;
2396 // VEOR : Vector Bitwise Exclusive OR
2397 def VEORd : N3VDX<1, 0, 0b00, 0b0001, 1, IIC_VBINiD, "veor",
2398 v2i32, v2i32, xor, 1>;
2399 def VEORq : N3VQX<1, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "veor",
2400 v4i32, v4i32, xor, 1>;
2402 // VORR : Vector Bitwise OR
2403 def VORRd : N3VDX<0, 0, 0b10, 0b0001, 1, IIC_VBINiD, "vorr",
2404 v2i32, v2i32, or, 1>;
2405 def VORRq : N3VQX<0, 0, 0b10, 0b0001, 1, IIC_VBINiQ, "vorr",
2406 v4i32, v4i32, or, 1>;
2408 // VBIC : Vector Bitwise Bit Clear (AND NOT)
2409 def VBICd : N3VX<0, 0, 0b01, 0b0001, 0, 1, (outs DPR:$dst),
2410 (ins DPR:$src1, DPR:$src2), N3RegFrm, IIC_VBINiD,
2411 "vbic", "$dst, $src1, $src2", "",
2412 [(set DPR:$dst, (v2i32 (and DPR:$src1,
2413 (vnotd DPR:$src2))))]>;
2414 def VBICq : N3VX<0, 0, 0b01, 0b0001, 1, 1, (outs QPR:$dst),
2415 (ins QPR:$src1, QPR:$src2), N3RegFrm, IIC_VBINiQ,
2416 "vbic", "$dst, $src1, $src2", "",
2417 [(set QPR:$dst, (v4i32 (and QPR:$src1,
2418 (vnotq QPR:$src2))))]>;
2420 // VORN : Vector Bitwise OR NOT
2421 def VORNd : N3VX<0, 0, 0b11, 0b0001, 0, 1, (outs DPR:$dst),
2422 (ins DPR:$src1, DPR:$src2), N3RegFrm, IIC_VBINiD,
2423 "vorn", "$dst, $src1, $src2", "",
2424 [(set DPR:$dst, (v2i32 (or DPR:$src1,
2425 (vnotd DPR:$src2))))]>;
2426 def VORNq : N3VX<0, 0, 0b11, 0b0001, 1, 1, (outs QPR:$dst),
2427 (ins QPR:$src1, QPR:$src2), N3RegFrm, IIC_VBINiQ,
2428 "vorn", "$dst, $src1, $src2", "",
2429 [(set QPR:$dst, (v4i32 (or QPR:$src1,
2430 (vnotq QPR:$src2))))]>;
2432 // VMVN : Vector Bitwise NOT (Immediate)
2434 let isReMaterializable = 1 in {
2435 def VMVNv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 1, 1, (outs DPR:$dst),
2436 (ins nModImm:$SIMM), IIC_VMOVImm,
2437 "vmvn", "i16", "$dst, $SIMM", "",
2438 [(set DPR:$dst, (v4i16 (NEONvmvnImm timm:$SIMM)))]>;
2439 def VMVNv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 1, 1, (outs QPR:$dst),
2440 (ins nModImm:$SIMM), IIC_VMOVImm,
2441 "vmvn", "i16", "$dst, $SIMM", "",
2442 [(set QPR:$dst, (v8i16 (NEONvmvnImm timm:$SIMM)))]>;
2444 def VMVNv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 1, 1, (outs DPR:$dst),
2445 (ins nModImm:$SIMM), IIC_VMOVImm,
2446 "vmvn", "i32", "$dst, $SIMM", "",
2447 [(set DPR:$dst, (v2i32 (NEONvmvnImm timm:$SIMM)))]>;
2448 def VMVNv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 1, 1, (outs QPR:$dst),
2449 (ins nModImm:$SIMM), IIC_VMOVImm,
2450 "vmvn", "i32", "$dst, $SIMM", "",
2451 [(set QPR:$dst, (v4i32 (NEONvmvnImm timm:$SIMM)))]>;
2454 // VMVN : Vector Bitwise NOT
2455 def VMVNd : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 0, 0,
2456 (outs DPR:$dst), (ins DPR:$src), IIC_VSUBiD,
2457 "vmvn", "$dst, $src", "",
2458 [(set DPR:$dst, (v2i32 (vnotd DPR:$src)))]>;
2459 def VMVNq : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 1, 0,
2460 (outs QPR:$dst), (ins QPR:$src), IIC_VSUBiD,
2461 "vmvn", "$dst, $src", "",
2462 [(set QPR:$dst, (v4i32 (vnotq QPR:$src)))]>;
2463 def : Pat<(v2i32 (vnotd DPR:$src)), (VMVNd DPR:$src)>;
2464 def : Pat<(v4i32 (vnotq QPR:$src)), (VMVNq QPR:$src)>;
2466 // VBSL : Vector Bitwise Select
2467 def VBSLd : N3VX<1, 0, 0b01, 0b0001, 0, 1, (outs DPR:$dst),
2468 (ins DPR:$src1, DPR:$src2, DPR:$src3),
2469 N3RegFrm, IIC_VCNTiD,
2470 "vbsl", "$dst, $src2, $src3", "$src1 = $dst",
2472 (v2i32 (or (and DPR:$src2, DPR:$src1),
2473 (and DPR:$src3, (vnotd DPR:$src1)))))]>;
2474 def VBSLq : N3VX<1, 0, 0b01, 0b0001, 1, 1, (outs QPR:$dst),
2475 (ins QPR:$src1, QPR:$src2, QPR:$src3),
2476 N3RegFrm, IIC_VCNTiQ,
2477 "vbsl", "$dst, $src2, $src3", "$src1 = $dst",
2479 (v4i32 (or (and QPR:$src2, QPR:$src1),
2480 (and QPR:$src3, (vnotq QPR:$src1)))))]>;
2482 // VBIF : Vector Bitwise Insert if False
2483 // like VBSL but with: "vbif $dst, $src3, $src1", "$src2 = $dst",
2484 def VBIFd : N3VX<1, 0, 0b11, 0b0001, 0, 1,
2485 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3),
2486 N3RegFrm, IIC_VBINiD,
2487 "vbif", "$dst, $src2, $src3", "$src1 = $dst",
2488 [/* For disassembly only; pattern left blank */]>;
2489 def VBIFq : N3VX<1, 0, 0b11, 0b0001, 1, 1,
2490 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3),
2491 N3RegFrm, IIC_VBINiQ,
2492 "vbif", "$dst, $src2, $src3", "$src1 = $dst",
2493 [/* For disassembly only; pattern left blank */]>;
2495 // VBIT : Vector Bitwise Insert if True
2496 // like VBSL but with: "vbit $dst, $src2, $src1", "$src3 = $dst",
2497 def VBITd : N3VX<1, 0, 0b10, 0b0001, 0, 1,
2498 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3),
2499 N3RegFrm, IIC_VBINiD,
2500 "vbit", "$dst, $src2, $src3", "$src1 = $dst",
2501 [/* For disassembly only; pattern left blank */]>;
2502 def VBITq : N3VX<1, 0, 0b10, 0b0001, 1, 1,
2503 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3),
2504 N3RegFrm, IIC_VBINiQ,
2505 "vbit", "$dst, $src2, $src3", "$src1 = $dst",
2506 [/* For disassembly only; pattern left blank */]>;
2508 // VBIT/VBIF are not yet implemented. The TwoAddress pass will not go looking
2509 // for equivalent operations with different register constraints; it just
2512 // Vector Absolute Differences.
2514 // VABD : Vector Absolute Difference
2515 defm VABDs : N3VInt_QHS<0, 0, 0b0111, 0, N3RegFrm,
2516 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2517 "vabd", "s", int_arm_neon_vabds, 0>;
2518 defm VABDu : N3VInt_QHS<1, 0, 0b0111, 0, N3RegFrm,
2519 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2520 "vabd", "u", int_arm_neon_vabdu, 0>;
2521 def VABDfd : N3VDInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBIND,
2522 "vabd", "f32", v2f32, v2f32, int_arm_neon_vabds, 0>;
2523 def VABDfq : N3VQInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBINQ,
2524 "vabd", "f32", v4f32, v4f32, int_arm_neon_vabds, 0>;
2526 // VABDL : Vector Absolute Difference Long (Q = | D - D |)
2527 defm VABDLs : N3VLInt_QHS<0,1,0b0111,0, IIC_VSUBi4Q, IIC_VSUBi4Q,
2528 "vabdl", "s", int_arm_neon_vabdls, 0>;
2529 defm VABDLu : N3VLInt_QHS<1,1,0b0111,0, IIC_VSUBi4Q, IIC_VSUBi4Q,
2530 "vabdl", "u", int_arm_neon_vabdlu, 0>;
2532 // VABA : Vector Absolute Difference and Accumulate
2533 defm VABAs : N3VInt3_QHS<0,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
2534 "vaba", "s", int_arm_neon_vabas>;
2535 defm VABAu : N3VInt3_QHS<1,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
2536 "vaba", "u", int_arm_neon_vabau>;
2538 // VABAL : Vector Absolute Difference and Accumulate Long (Q += | D - D |)
2539 defm VABALs : N3VLInt3_QHS<0,1,0b0101,0, IIC_VABAD, IIC_VABAD,
2540 "vabal", "s", int_arm_neon_vabals>;
2541 defm VABALu : N3VLInt3_QHS<1,1,0b0101,0, IIC_VABAD, IIC_VABAD,
2542 "vabal", "u", int_arm_neon_vabalu>;
2544 // Vector Maximum and Minimum.
2546 // VMAX : Vector Maximum
2547 defm VMAXs : N3VInt_QHS<0, 0, 0b0110, 0, N3RegFrm,
2548 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2549 "vmax", "s", int_arm_neon_vmaxs, 1>;
2550 defm VMAXu : N3VInt_QHS<1, 0, 0b0110, 0, N3RegFrm,
2551 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2552 "vmax", "u", int_arm_neon_vmaxu, 1>;
2553 def VMAXfd : N3VDInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBIND,
2555 v2f32, v2f32, int_arm_neon_vmaxs, 1>;
2556 def VMAXfq : N3VQInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBINQ,
2558 v4f32, v4f32, int_arm_neon_vmaxs, 1>;
2560 // VMIN : Vector Minimum
2561 defm VMINs : N3VInt_QHS<0, 0, 0b0110, 1, N3RegFrm,
2562 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2563 "vmin", "s", int_arm_neon_vmins, 1>;
2564 defm VMINu : N3VInt_QHS<1, 0, 0b0110, 1, N3RegFrm,
2565 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2566 "vmin", "u", int_arm_neon_vminu, 1>;
2567 def VMINfd : N3VDInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBIND,
2569 v2f32, v2f32, int_arm_neon_vmins, 1>;
2570 def VMINfq : N3VQInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBINQ,
2572 v4f32, v4f32, int_arm_neon_vmins, 1>;
2574 // Vector Pairwise Operations.
2576 // VPADD : Vector Pairwise Add
2577 def VPADDi8 : N3VDInt<0, 0, 0b00, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
2579 v8i8, v8i8, int_arm_neon_vpadd, 0>;
2580 def VPADDi16 : N3VDInt<0, 0, 0b01, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
2582 v4i16, v4i16, int_arm_neon_vpadd, 0>;
2583 def VPADDi32 : N3VDInt<0, 0, 0b10, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
2585 v2i32, v2i32, int_arm_neon_vpadd, 0>;
2586 def VPADDf : N3VDInt<1, 0, 0b00, 0b1101, 0, N3RegFrm,
2587 IIC_VBIND, "vpadd", "f32",
2588 v2f32, v2f32, int_arm_neon_vpadd, 0>;
2590 // VPADDL : Vector Pairwise Add Long
2591 defm VPADDLs : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00100, 0, "vpaddl", "s",
2592 int_arm_neon_vpaddls>;
2593 defm VPADDLu : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00101, 0, "vpaddl", "u",
2594 int_arm_neon_vpaddlu>;
2596 // VPADAL : Vector Pairwise Add and Accumulate Long
2597 defm VPADALs : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01100, 0, "vpadal", "s",
2598 int_arm_neon_vpadals>;
2599 defm VPADALu : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01101, 0, "vpadal", "u",
2600 int_arm_neon_vpadalu>;
2602 // VPMAX : Vector Pairwise Maximum
2603 def VPMAXs8 : N3VDInt<0, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
2604 "s8", v8i8, v8i8, int_arm_neon_vpmaxs, 0>;
2605 def VPMAXs16 : N3VDInt<0, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
2606 "s16", v4i16, v4i16, int_arm_neon_vpmaxs, 0>;
2607 def VPMAXs32 : N3VDInt<0, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
2608 "s32", v2i32, v2i32, int_arm_neon_vpmaxs, 0>;
2609 def VPMAXu8 : N3VDInt<1, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
2610 "u8", v8i8, v8i8, int_arm_neon_vpmaxu, 0>;
2611 def VPMAXu16 : N3VDInt<1, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
2612 "u16", v4i16, v4i16, int_arm_neon_vpmaxu, 0>;
2613 def VPMAXu32 : N3VDInt<1, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
2614 "u32", v2i32, v2i32, int_arm_neon_vpmaxu, 0>;
2615 def VPMAXf : N3VDInt<1, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
2616 "f32", v2f32, v2f32, int_arm_neon_vpmaxs, 0>;
2618 // VPMIN : Vector Pairwise Minimum
2619 def VPMINs8 : N3VDInt<0, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
2620 "s8", v8i8, v8i8, int_arm_neon_vpmins, 0>;
2621 def VPMINs16 : N3VDInt<0, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
2622 "s16", v4i16, v4i16, int_arm_neon_vpmins, 0>;
2623 def VPMINs32 : N3VDInt<0, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
2624 "s32", v2i32, v2i32, int_arm_neon_vpmins, 0>;
2625 def VPMINu8 : N3VDInt<1, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
2626 "u8", v8i8, v8i8, int_arm_neon_vpminu, 0>;
2627 def VPMINu16 : N3VDInt<1, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
2628 "u16", v4i16, v4i16, int_arm_neon_vpminu, 0>;
2629 def VPMINu32 : N3VDInt<1, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
2630 "u32", v2i32, v2i32, int_arm_neon_vpminu, 0>;
2631 def VPMINf : N3VDInt<1, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VSUBi4D, "vpmin",
2632 "f32", v2f32, v2f32, int_arm_neon_vpmins, 0>;
2634 // Vector Reciprocal and Reciprocal Square Root Estimate and Step.
2636 // VRECPE : Vector Reciprocal Estimate
2637 def VRECPEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
2638 IIC_VUNAD, "vrecpe", "u32",
2639 v2i32, v2i32, int_arm_neon_vrecpe>;
2640 def VRECPEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
2641 IIC_VUNAQ, "vrecpe", "u32",
2642 v4i32, v4i32, int_arm_neon_vrecpe>;
2643 def VRECPEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
2644 IIC_VUNAD, "vrecpe", "f32",
2645 v2f32, v2f32, int_arm_neon_vrecpe>;
2646 def VRECPEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
2647 IIC_VUNAQ, "vrecpe", "f32",
2648 v4f32, v4f32, int_arm_neon_vrecpe>;
2650 // VRECPS : Vector Reciprocal Step
2651 def VRECPSfd : N3VDInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
2652 IIC_VRECSD, "vrecps", "f32",
2653 v2f32, v2f32, int_arm_neon_vrecps, 1>;
2654 def VRECPSfq : N3VQInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
2655 IIC_VRECSQ, "vrecps", "f32",
2656 v4f32, v4f32, int_arm_neon_vrecps, 1>;
2658 // VRSQRTE : Vector Reciprocal Square Root Estimate
2659 def VRSQRTEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
2660 IIC_VUNAD, "vrsqrte", "u32",
2661 v2i32, v2i32, int_arm_neon_vrsqrte>;
2662 def VRSQRTEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
2663 IIC_VUNAQ, "vrsqrte", "u32",
2664 v4i32, v4i32, int_arm_neon_vrsqrte>;
2665 def VRSQRTEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
2666 IIC_VUNAD, "vrsqrte", "f32",
2667 v2f32, v2f32, int_arm_neon_vrsqrte>;
2668 def VRSQRTEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
2669 IIC_VUNAQ, "vrsqrte", "f32",
2670 v4f32, v4f32, int_arm_neon_vrsqrte>;
2672 // VRSQRTS : Vector Reciprocal Square Root Step
2673 def VRSQRTSfd : N3VDInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
2674 IIC_VRECSD, "vrsqrts", "f32",
2675 v2f32, v2f32, int_arm_neon_vrsqrts, 1>;
2676 def VRSQRTSfq : N3VQInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
2677 IIC_VRECSQ, "vrsqrts", "f32",
2678 v4f32, v4f32, int_arm_neon_vrsqrts, 1>;
2682 // VSHL : Vector Shift
2683 defm VSHLs : N3VInt_QHSD<0, 0, 0b0100, 0, N3RegVShFrm,
2684 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
2685 "vshl", "s", int_arm_neon_vshifts, 0>;
2686 defm VSHLu : N3VInt_QHSD<1, 0, 0b0100, 0, N3RegVShFrm,
2687 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
2688 "vshl", "u", int_arm_neon_vshiftu, 0>;
2689 // VSHL : Vector Shift Left (Immediate)
2690 defm VSHLi : N2VSh_QHSD<0, 1, 0b0101, 1, IIC_VSHLiD, "vshl", "i", NEONvshl,
2692 // VSHR : Vector Shift Right (Immediate)
2693 defm VSHRs : N2VSh_QHSD<0, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "s", NEONvshrs,
2695 defm VSHRu : N2VSh_QHSD<1, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "u", NEONvshru,
2698 // VSHLL : Vector Shift Left Long
2699 defm VSHLLs : N2VLSh_QHS<0, 1, 0b1010, 0, 0, 1, "vshll", "s", NEONvshlls>;
2700 defm VSHLLu : N2VLSh_QHS<1, 1, 0b1010, 0, 0, 1, "vshll", "u", NEONvshllu>;
2702 // VSHLL : Vector Shift Left Long (with maximum shift count)
2703 class N2VLShMax<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
2704 bit op6, bit op4, string OpcodeStr, string Dt, ValueType ResTy,
2705 ValueType OpTy, SDNode OpNode>
2706 : N2VLSh<op24, op23, op11_8, op7, op6, op4, OpcodeStr, Dt,
2707 ResTy, OpTy, OpNode> {
2708 let Inst{21-16} = op21_16;
2710 def VSHLLi8 : N2VLShMax<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll", "i8",
2711 v8i16, v8i8, NEONvshlli>;
2712 def VSHLLi16 : N2VLShMax<1, 1, 0b110110, 0b0011, 0, 0, 0, "vshll", "i16",
2713 v4i32, v4i16, NEONvshlli>;
2714 def VSHLLi32 : N2VLShMax<1, 1, 0b111010, 0b0011, 0, 0, 0, "vshll", "i32",
2715 v2i64, v2i32, NEONvshlli>;
2717 // VSHRN : Vector Shift Right and Narrow
2718 defm VSHRN : N2VNSh_HSD<0,1,0b1000,0,0,1, IIC_VSHLiD, "vshrn", "i",
2721 // VRSHL : Vector Rounding Shift
2722 defm VRSHLs : N3VInt_QHSD<0, 0, 0b0101, 0, N3RegVShFrm,
2723 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
2724 "vrshl", "s", int_arm_neon_vrshifts, 0>;
2725 defm VRSHLu : N3VInt_QHSD<1, 0, 0b0101, 0, N3RegVShFrm,
2726 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
2727 "vrshl", "u", int_arm_neon_vrshiftu, 0>;
2728 // VRSHR : Vector Rounding Shift Right
2729 defm VRSHRs : N2VSh_QHSD<0,1,0b0010,1, IIC_VSHLi4D, "vrshr", "s", NEONvrshrs,
2731 defm VRSHRu : N2VSh_QHSD<1,1,0b0010,1, IIC_VSHLi4D, "vrshr", "u", NEONvrshru,
2734 // VRSHRN : Vector Rounding Shift Right and Narrow
2735 defm VRSHRN : N2VNSh_HSD<0, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vrshrn", "i",
2738 // VQSHL : Vector Saturating Shift
2739 defm VQSHLs : N3VInt_QHSD<0, 0, 0b0100, 1, N3RegVShFrm,
2740 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
2741 "vqshl", "s", int_arm_neon_vqshifts, 0>;
2742 defm VQSHLu : N3VInt_QHSD<1, 0, 0b0100, 1, N3RegVShFrm,
2743 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
2744 "vqshl", "u", int_arm_neon_vqshiftu, 0>;
2745 // VQSHL : Vector Saturating Shift Left (Immediate)
2746 defm VQSHLsi : N2VSh_QHSD<0,1,0b0111,1, IIC_VSHLi4D, "vqshl", "s",NEONvqshls,
2748 defm VQSHLui : N2VSh_QHSD<1,1,0b0111,1, IIC_VSHLi4D, "vqshl", "u",NEONvqshlu,
2750 // VQSHLU : Vector Saturating Shift Left (Immediate, Unsigned)
2751 defm VQSHLsu : N2VSh_QHSD<1,1,0b0110,1, IIC_VSHLi4D,"vqshlu","s",NEONvqshlsu,
2754 // VQSHRN : Vector Saturating Shift Right and Narrow
2755 defm VQSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "s",
2757 defm VQSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "u",
2760 // VQSHRUN : Vector Saturating Shift Right and Narrow (Unsigned)
2761 defm VQSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 0, 1, IIC_VSHLi4D, "vqshrun", "s",
2764 // VQRSHL : Vector Saturating Rounding Shift
2765 defm VQRSHLs : N3VInt_QHSD<0, 0, 0b0101, 1, N3RegVShFrm,
2766 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
2767 "vqrshl", "s", int_arm_neon_vqrshifts, 0>;
2768 defm VQRSHLu : N3VInt_QHSD<1, 0, 0b0101, 1, N3RegVShFrm,
2769 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
2770 "vqrshl", "u", int_arm_neon_vqrshiftu, 0>;
2772 // VQRSHRN : Vector Saturating Rounding Shift Right and Narrow
2773 defm VQRSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "s",
2775 defm VQRSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "u",
2778 // VQRSHRUN : Vector Saturating Rounding Shift Right and Narrow (Unsigned)
2779 defm VQRSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vqrshrun", "s",
2782 // VSRA : Vector Shift Right and Accumulate
2783 defm VSRAs : N2VShAdd_QHSD<0, 1, 0b0001, 1, "vsra", "s", NEONvshrs>;
2784 defm VSRAu : N2VShAdd_QHSD<1, 1, 0b0001, 1, "vsra", "u", NEONvshru>;
2785 // VRSRA : Vector Rounding Shift Right and Accumulate
2786 defm VRSRAs : N2VShAdd_QHSD<0, 1, 0b0011, 1, "vrsra", "s", NEONvrshrs>;
2787 defm VRSRAu : N2VShAdd_QHSD<1, 1, 0b0011, 1, "vrsra", "u", NEONvrshru>;
2789 // VSLI : Vector Shift Left and Insert
2790 defm VSLI : N2VShIns_QHSD<1, 1, 0b0101, 1, "vsli", NEONvsli, N2RegVShLFrm>;
2791 // VSRI : Vector Shift Right and Insert
2792 defm VSRI : N2VShIns_QHSD<1, 1, 0b0100, 1, "vsri", NEONvsri, N2RegVShRFrm>;
2794 // Vector Absolute and Saturating Absolute.
2796 // VABS : Vector Absolute Value
2797 defm VABS : N2VInt_QHS<0b11, 0b11, 0b01, 0b00110, 0,
2798 IIC_VUNAiD, IIC_VUNAiQ, "vabs", "s",
2800 def VABSfd : N2VDInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
2801 IIC_VUNAD, "vabs", "f32",
2802 v2f32, v2f32, int_arm_neon_vabs>;
2803 def VABSfq : N2VQInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
2804 IIC_VUNAQ, "vabs", "f32",
2805 v4f32, v4f32, int_arm_neon_vabs>;
2807 // VQABS : Vector Saturating Absolute Value
2808 defm VQABS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01110, 0,
2809 IIC_VQUNAiD, IIC_VQUNAiQ, "vqabs", "s",
2810 int_arm_neon_vqabs>;
2814 def vnegd : PatFrag<(ops node:$in),
2815 (sub (bitconvert (v2i32 NEONimmAllZerosV)), node:$in)>;
2816 def vnegq : PatFrag<(ops node:$in),
2817 (sub (bitconvert (v4i32 NEONimmAllZerosV)), node:$in)>;
2819 class VNEGD<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
2820 : N2V<0b11, 0b11, size, 0b01, 0b00111, 0, 0, (outs DPR:$dst), (ins DPR:$src),
2821 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
2822 [(set DPR:$dst, (Ty (vnegd DPR:$src)))]>;
2823 class VNEGQ<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
2824 : N2V<0b11, 0b11, size, 0b01, 0b00111, 1, 0, (outs QPR:$dst), (ins QPR:$src),
2825 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
2826 [(set QPR:$dst, (Ty (vnegq QPR:$src)))]>;
2828 // VNEG : Vector Negate (integer)
2829 def VNEGs8d : VNEGD<0b00, "vneg", "s8", v8i8>;
2830 def VNEGs16d : VNEGD<0b01, "vneg", "s16", v4i16>;
2831 def VNEGs32d : VNEGD<0b10, "vneg", "s32", v2i32>;
2832 def VNEGs8q : VNEGQ<0b00, "vneg", "s8", v16i8>;
2833 def VNEGs16q : VNEGQ<0b01, "vneg", "s16", v8i16>;
2834 def VNEGs32q : VNEGQ<0b10, "vneg", "s32", v4i32>;
2836 // VNEG : Vector Negate (floating-point)
2837 def VNEGfd : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
2838 (outs DPR:$dst), (ins DPR:$src), IIC_VUNAD,
2839 "vneg", "f32", "$dst, $src", "",
2840 [(set DPR:$dst, (v2f32 (fneg DPR:$src)))]>;
2841 def VNEGf32q : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 1, 0,
2842 (outs QPR:$dst), (ins QPR:$src), IIC_VUNAQ,
2843 "vneg", "f32", "$dst, $src", "",
2844 [(set QPR:$dst, (v4f32 (fneg QPR:$src)))]>;
2846 def : Pat<(v8i8 (vnegd DPR:$src)), (VNEGs8d DPR:$src)>;
2847 def : Pat<(v4i16 (vnegd DPR:$src)), (VNEGs16d DPR:$src)>;
2848 def : Pat<(v2i32 (vnegd DPR:$src)), (VNEGs32d DPR:$src)>;
2849 def : Pat<(v16i8 (vnegq QPR:$src)), (VNEGs8q QPR:$src)>;
2850 def : Pat<(v8i16 (vnegq QPR:$src)), (VNEGs16q QPR:$src)>;
2851 def : Pat<(v4i32 (vnegq QPR:$src)), (VNEGs32q QPR:$src)>;
2853 // VQNEG : Vector Saturating Negate
2854 defm VQNEG : N2VInt_QHS<0b11, 0b11, 0b00, 0b01111, 0,
2855 IIC_VQUNAiD, IIC_VQUNAiQ, "vqneg", "s",
2856 int_arm_neon_vqneg>;
2858 // Vector Bit Counting Operations.
2860 // VCLS : Vector Count Leading Sign Bits
2861 defm VCLS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01000, 0,
2862 IIC_VCNTiD, IIC_VCNTiQ, "vcls", "s",
2864 // VCLZ : Vector Count Leading Zeros
2865 defm VCLZ : N2VInt_QHS<0b11, 0b11, 0b00, 0b01001, 0,
2866 IIC_VCNTiD, IIC_VCNTiQ, "vclz", "i",
2868 // VCNT : Vector Count One Bits
2869 def VCNTd : N2VDInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
2870 IIC_VCNTiD, "vcnt", "8",
2871 v8i8, v8i8, int_arm_neon_vcnt>;
2872 def VCNTq : N2VQInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
2873 IIC_VCNTiQ, "vcnt", "8",
2874 v16i8, v16i8, int_arm_neon_vcnt>;
2876 // Vector Swap -- for disassembly only.
2877 def VSWPd : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 0, 0,
2878 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
2879 "vswp", "$dst, $src", "", []>;
2880 def VSWPq : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 1, 0,
2881 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
2882 "vswp", "$dst, $src", "", []>;
2884 // Vector Move Operations.
2886 // VMOV : Vector Move (Register)
2888 let neverHasSideEffects = 1 in {
2889 def VMOVDneon: N3VX<0, 0, 0b10, 0b0001, 0, 1, (outs DPR:$dst), (ins DPR:$src),
2890 N3RegFrm, IIC_VMOVD, "vmov", "$dst, $src", "", []>;
2891 def VMOVQ : N3VX<0, 0, 0b10, 0b0001, 1, 1, (outs QPR:$dst), (ins QPR:$src),
2892 N3RegFrm, IIC_VMOVD, "vmov", "$dst, $src", "", []>;
2894 // Pseudo vector move instructions for QQ and QQQQ registers. This should
2895 // be expanded after register allocation is completed.
2896 def VMOVQQ : PseudoInst<(outs QQPR:$dst), (ins QQPR:$src),
2897 NoItinerary, "${:comment} vmov\t$dst, $src", []>;
2899 def VMOVQQQQ : PseudoInst<(outs QQQQPR:$dst), (ins QQQQPR:$src),
2900 NoItinerary, "${:comment} vmov\t$dst, $src", []>;
2901 } // neverHasSideEffects
2903 // VMOV : Vector Move (Immediate)
2905 let isReMaterializable = 1 in {
2906 def VMOVv8i8 : N1ModImm<1, 0b000, 0b1110, 0, 0, 0, 1, (outs DPR:$dst),
2907 (ins nModImm:$SIMM), IIC_VMOVImm,
2908 "vmov", "i8", "$dst, $SIMM", "",
2909 [(set DPR:$dst, (v8i8 (NEONvmovImm timm:$SIMM)))]>;
2910 def VMOVv16i8 : N1ModImm<1, 0b000, 0b1110, 0, 1, 0, 1, (outs QPR:$dst),
2911 (ins nModImm:$SIMM), IIC_VMOVImm,
2912 "vmov", "i8", "$dst, $SIMM", "",
2913 [(set QPR:$dst, (v16i8 (NEONvmovImm timm:$SIMM)))]>;
2915 def VMOVv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 0, 1, (outs DPR:$dst),
2916 (ins nModImm:$SIMM), IIC_VMOVImm,
2917 "vmov", "i16", "$dst, $SIMM", "",
2918 [(set DPR:$dst, (v4i16 (NEONvmovImm timm:$SIMM)))]>;
2919 def VMOVv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 0, 1, (outs QPR:$dst),
2920 (ins nModImm:$SIMM), IIC_VMOVImm,
2921 "vmov", "i16", "$dst, $SIMM", "",
2922 [(set QPR:$dst, (v8i16 (NEONvmovImm timm:$SIMM)))]>;
2924 def VMOVv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 0, 1, (outs DPR:$dst),
2925 (ins nModImm:$SIMM), IIC_VMOVImm,
2926 "vmov", "i32", "$dst, $SIMM", "",
2927 [(set DPR:$dst, (v2i32 (NEONvmovImm timm:$SIMM)))]>;
2928 def VMOVv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 0, 1, (outs QPR:$dst),
2929 (ins nModImm:$SIMM), IIC_VMOVImm,
2930 "vmov", "i32", "$dst, $SIMM", "",
2931 [(set QPR:$dst, (v4i32 (NEONvmovImm timm:$SIMM)))]>;
2933 def VMOVv1i64 : N1ModImm<1, 0b000, 0b1110, 0, 0, 1, 1, (outs DPR:$dst),
2934 (ins nModImm:$SIMM), IIC_VMOVImm,
2935 "vmov", "i64", "$dst, $SIMM", "",
2936 [(set DPR:$dst, (v1i64 (NEONvmovImm timm:$SIMM)))]>;
2937 def VMOVv2i64 : N1ModImm<1, 0b000, 0b1110, 0, 1, 1, 1, (outs QPR:$dst),
2938 (ins nModImm:$SIMM), IIC_VMOVImm,
2939 "vmov", "i64", "$dst, $SIMM", "",
2940 [(set QPR:$dst, (v2i64 (NEONvmovImm timm:$SIMM)))]>;
2941 } // isReMaterializable
2943 // VMOV : Vector Get Lane (move scalar to ARM core register)
2945 def VGETLNs8 : NVGetLane<{1,1,1,0,0,1,?,1}, 0b1011, {?,?},
2946 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2947 IIC_VMOVSI, "vmov", "s8", "$dst, $src[$lane]",
2948 [(set GPR:$dst, (NEONvgetlanes (v8i8 DPR:$src),
2950 def VGETLNs16 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, {?,1},
2951 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2952 IIC_VMOVSI, "vmov", "s16", "$dst, $src[$lane]",
2953 [(set GPR:$dst, (NEONvgetlanes (v4i16 DPR:$src),
2955 def VGETLNu8 : NVGetLane<{1,1,1,0,1,1,?,1}, 0b1011, {?,?},
2956 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2957 IIC_VMOVSI, "vmov", "u8", "$dst, $src[$lane]",
2958 [(set GPR:$dst, (NEONvgetlaneu (v8i8 DPR:$src),
2960 def VGETLNu16 : NVGetLane<{1,1,1,0,1,0,?,1}, 0b1011, {?,1},
2961 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2962 IIC_VMOVSI, "vmov", "u16", "$dst, $src[$lane]",
2963 [(set GPR:$dst, (NEONvgetlaneu (v4i16 DPR:$src),
2965 def VGETLNi32 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, 0b00,
2966 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2967 IIC_VMOVSI, "vmov", "32", "$dst, $src[$lane]",
2968 [(set GPR:$dst, (extractelt (v2i32 DPR:$src),
2970 // def VGETLNf32: see FMRDH and FMRDL in ARMInstrVFP.td
2971 def : Pat<(NEONvgetlanes (v16i8 QPR:$src), imm:$lane),
2972 (VGETLNs8 (v8i8 (EXTRACT_SUBREG QPR:$src,
2973 (DSubReg_i8_reg imm:$lane))),
2974 (SubReg_i8_lane imm:$lane))>;
2975 def : Pat<(NEONvgetlanes (v8i16 QPR:$src), imm:$lane),
2976 (VGETLNs16 (v4i16 (EXTRACT_SUBREG QPR:$src,
2977 (DSubReg_i16_reg imm:$lane))),
2978 (SubReg_i16_lane imm:$lane))>;
2979 def : Pat<(NEONvgetlaneu (v16i8 QPR:$src), imm:$lane),
2980 (VGETLNu8 (v8i8 (EXTRACT_SUBREG QPR:$src,
2981 (DSubReg_i8_reg imm:$lane))),
2982 (SubReg_i8_lane imm:$lane))>;
2983 def : Pat<(NEONvgetlaneu (v8i16 QPR:$src), imm:$lane),
2984 (VGETLNu16 (v4i16 (EXTRACT_SUBREG QPR:$src,
2985 (DSubReg_i16_reg imm:$lane))),
2986 (SubReg_i16_lane imm:$lane))>;
2987 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
2988 (VGETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src,
2989 (DSubReg_i32_reg imm:$lane))),
2990 (SubReg_i32_lane imm:$lane))>;
2991 def : Pat<(extractelt (v2f32 DPR:$src1), imm:$src2),
2992 (EXTRACT_SUBREG (v2f32 (COPY_TO_REGCLASS (v2f32 DPR:$src1),DPR_VFP2)),
2993 (SSubReg_f32_reg imm:$src2))>;
2994 def : Pat<(extractelt (v4f32 QPR:$src1), imm:$src2),
2995 (EXTRACT_SUBREG (v4f32 (COPY_TO_REGCLASS (v4f32 QPR:$src1),QPR_VFP2)),
2996 (SSubReg_f32_reg imm:$src2))>;
2997 //def : Pat<(extractelt (v2i64 QPR:$src1), imm:$src2),
2998 // (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
2999 def : Pat<(extractelt (v2f64 QPR:$src1), imm:$src2),
3000 (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
3003 // VMOV : Vector Set Lane (move ARM core register to scalar)
3005 let Constraints = "$src1 = $dst" in {
3006 def VSETLNi8 : NVSetLane<{1,1,1,0,0,1,?,0}, 0b1011, {?,?}, (outs DPR:$dst),
3007 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
3008 IIC_VMOVISL, "vmov", "8", "$dst[$lane], $src2",
3009 [(set DPR:$dst, (vector_insert (v8i8 DPR:$src1),
3010 GPR:$src2, imm:$lane))]>;
3011 def VSETLNi16 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, {?,1}, (outs DPR:$dst),
3012 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
3013 IIC_VMOVISL, "vmov", "16", "$dst[$lane], $src2",
3014 [(set DPR:$dst, (vector_insert (v4i16 DPR:$src1),
3015 GPR:$src2, imm:$lane))]>;
3016 def VSETLNi32 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, 0b00, (outs DPR:$dst),
3017 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
3018 IIC_VMOVISL, "vmov", "32", "$dst[$lane], $src2",
3019 [(set DPR:$dst, (insertelt (v2i32 DPR:$src1),
3020 GPR:$src2, imm:$lane))]>;
3022 def : Pat<(vector_insert (v16i8 QPR:$src1), GPR:$src2, imm:$lane),
3023 (v16i8 (INSERT_SUBREG QPR:$src1,
3024 (v8i8 (VSETLNi8 (v8i8 (EXTRACT_SUBREG QPR:$src1,
3025 (DSubReg_i8_reg imm:$lane))),
3026 GPR:$src2, (SubReg_i8_lane imm:$lane))),
3027 (DSubReg_i8_reg imm:$lane)))>;
3028 def : Pat<(vector_insert (v8i16 QPR:$src1), GPR:$src2, imm:$lane),
3029 (v8i16 (INSERT_SUBREG QPR:$src1,
3030 (v4i16 (VSETLNi16 (v4i16 (EXTRACT_SUBREG QPR:$src1,
3031 (DSubReg_i16_reg imm:$lane))),
3032 GPR:$src2, (SubReg_i16_lane imm:$lane))),
3033 (DSubReg_i16_reg imm:$lane)))>;
3034 def : Pat<(insertelt (v4i32 QPR:$src1), GPR:$src2, imm:$lane),
3035 (v4i32 (INSERT_SUBREG QPR:$src1,
3036 (v2i32 (VSETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src1,
3037 (DSubReg_i32_reg imm:$lane))),
3038 GPR:$src2, (SubReg_i32_lane imm:$lane))),
3039 (DSubReg_i32_reg imm:$lane)))>;
3041 def : Pat<(v2f32 (insertelt DPR:$src1, SPR:$src2, imm:$src3)),
3042 (INSERT_SUBREG (v2f32 (COPY_TO_REGCLASS DPR:$src1, DPR_VFP2)),
3043 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
3044 def : Pat<(v4f32 (insertelt QPR:$src1, SPR:$src2, imm:$src3)),
3045 (INSERT_SUBREG (v4f32 (COPY_TO_REGCLASS QPR:$src1, QPR_VFP2)),
3046 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
3048 //def : Pat<(v2i64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
3049 // (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
3050 def : Pat<(v2f64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
3051 (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
3053 def : Pat<(v2f32 (scalar_to_vector SPR:$src)),
3054 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
3055 def : Pat<(v2f64 (scalar_to_vector (f64 DPR:$src))),
3056 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
3057 def : Pat<(v4f32 (scalar_to_vector SPR:$src)),
3058 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
3060 def : Pat<(v8i8 (scalar_to_vector GPR:$src)),
3061 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
3062 def : Pat<(v4i16 (scalar_to_vector GPR:$src)),
3063 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
3064 def : Pat<(v2i32 (scalar_to_vector GPR:$src)),
3065 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
3067 def : Pat<(v16i8 (scalar_to_vector GPR:$src)),
3068 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3069 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
3071 def : Pat<(v8i16 (scalar_to_vector GPR:$src)),
3072 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
3073 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
3075 def : Pat<(v4i32 (scalar_to_vector GPR:$src)),
3076 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
3077 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
3080 // VDUP : Vector Duplicate (from ARM core register to all elements)
3082 class VDUPD<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
3083 : NVDup<opcod1, 0b1011, opcod3, (outs DPR:$dst), (ins GPR:$src),
3084 IIC_VMOVIS, "vdup", Dt, "$dst, $src",
3085 [(set DPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
3086 class VDUPQ<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
3087 : NVDup<opcod1, 0b1011, opcod3, (outs QPR:$dst), (ins GPR:$src),
3088 IIC_VMOVIS, "vdup", Dt, "$dst, $src",
3089 [(set QPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
3091 def VDUP8d : VDUPD<0b11101100, 0b00, "8", v8i8>;
3092 def VDUP16d : VDUPD<0b11101000, 0b01, "16", v4i16>;
3093 def VDUP32d : VDUPD<0b11101000, 0b00, "32", v2i32>;
3094 def VDUP8q : VDUPQ<0b11101110, 0b00, "8", v16i8>;
3095 def VDUP16q : VDUPQ<0b11101010, 0b01, "16", v8i16>;
3096 def VDUP32q : VDUPQ<0b11101010, 0b00, "32", v4i32>;
3098 def VDUPfd : NVDup<0b11101000, 0b1011, 0b00, (outs DPR:$dst), (ins GPR:$src),
3099 IIC_VMOVIS, "vdup", "32", "$dst, $src",
3100 [(set DPR:$dst, (v2f32 (NEONvdup
3101 (f32 (bitconvert GPR:$src)))))]>;
3102 def VDUPfq : NVDup<0b11101010, 0b1011, 0b00, (outs QPR:$dst), (ins GPR:$src),
3103 IIC_VMOVIS, "vdup", "32", "$dst, $src",
3104 [(set QPR:$dst, (v4f32 (NEONvdup
3105 (f32 (bitconvert GPR:$src)))))]>;
3107 // VDUP : Vector Duplicate Lane (from scalar to all elements)
3109 class VDUPLND<bits<4> op19_16, string OpcodeStr, string Dt,
3111 : NVDupLane<op19_16, 0, (outs DPR:$dst), (ins DPR:$src, nohash_imm:$lane),
3112 IIC_VMOVD, OpcodeStr, Dt, "$dst, $src[$lane]",
3113 [(set DPR:$dst, (Ty (NEONvduplane (Ty DPR:$src), imm:$lane)))]>;
3115 class VDUPLNQ<bits<4> op19_16, string OpcodeStr, string Dt,
3116 ValueType ResTy, ValueType OpTy>
3117 : NVDupLane<op19_16, 1, (outs QPR:$dst), (ins DPR:$src, nohash_imm:$lane),
3118 IIC_VMOVD, OpcodeStr, Dt, "$dst, $src[$lane]",
3119 [(set QPR:$dst, (ResTy (NEONvduplane (OpTy DPR:$src),
3122 // Inst{19-16} is partially specified depending on the element size.
3124 def VDUPLN8d : VDUPLND<{?,?,?,1}, "vdup", "8", v8i8>;
3125 def VDUPLN16d : VDUPLND<{?,?,1,0}, "vdup", "16", v4i16>;
3126 def VDUPLN32d : VDUPLND<{?,1,0,0}, "vdup", "32", v2i32>;
3127 def VDUPLNfd : VDUPLND<{?,1,0,0}, "vdup", "32", v2f32>;
3128 def VDUPLN8q : VDUPLNQ<{?,?,?,1}, "vdup", "8", v16i8, v8i8>;
3129 def VDUPLN16q : VDUPLNQ<{?,?,1,0}, "vdup", "16", v8i16, v4i16>;
3130 def VDUPLN32q : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4i32, v2i32>;
3131 def VDUPLNfq : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4f32, v2f32>;
3133 def : Pat<(v16i8 (NEONvduplane (v16i8 QPR:$src), imm:$lane)),
3134 (v16i8 (VDUPLN8q (v8i8 (EXTRACT_SUBREG QPR:$src,
3135 (DSubReg_i8_reg imm:$lane))),
3136 (SubReg_i8_lane imm:$lane)))>;
3137 def : Pat<(v8i16 (NEONvduplane (v8i16 QPR:$src), imm:$lane)),
3138 (v8i16 (VDUPLN16q (v4i16 (EXTRACT_SUBREG QPR:$src,
3139 (DSubReg_i16_reg imm:$lane))),
3140 (SubReg_i16_lane imm:$lane)))>;
3141 def : Pat<(v4i32 (NEONvduplane (v4i32 QPR:$src), imm:$lane)),
3142 (v4i32 (VDUPLN32q (v2i32 (EXTRACT_SUBREG QPR:$src,
3143 (DSubReg_i32_reg imm:$lane))),
3144 (SubReg_i32_lane imm:$lane)))>;
3145 def : Pat<(v4f32 (NEONvduplane (v4f32 QPR:$src), imm:$lane)),
3146 (v4f32 (VDUPLNfq (v2f32 (EXTRACT_SUBREG QPR:$src,
3147 (DSubReg_i32_reg imm:$lane))),
3148 (SubReg_i32_lane imm:$lane)))>;
3150 def VDUPfdf : N2V<0b11, 0b11, {?,1}, {0,0}, 0b11000, 0, 0,
3151 (outs DPR:$dst), (ins SPR:$src),
3152 IIC_VMOVD, "vdup", "32", "$dst, ${src:lane}", "",
3153 [(set DPR:$dst, (v2f32 (NEONvdup (f32 SPR:$src))))]>;
3155 def VDUPfqf : N2V<0b11, 0b11, {?,1}, {0,0}, 0b11000, 1, 0,
3156 (outs QPR:$dst), (ins SPR:$src),
3157 IIC_VMOVD, "vdup", "32", "$dst, ${src:lane}", "",
3158 [(set QPR:$dst, (v4f32 (NEONvdup (f32 SPR:$src))))]>;
3160 // VMOVN : Vector Narrowing Move
3161 defm VMOVN : N2VNInt_HSD<0b11,0b11,0b10,0b00100,0,0, IIC_VMOVD,
3162 "vmovn", "i", int_arm_neon_vmovn>;
3163 // VQMOVN : Vector Saturating Narrowing Move
3164 defm VQMOVNs : N2VNInt_HSD<0b11,0b11,0b10,0b00101,0,0, IIC_VQUNAiD,
3165 "vqmovn", "s", int_arm_neon_vqmovns>;
3166 defm VQMOVNu : N2VNInt_HSD<0b11,0b11,0b10,0b00101,1,0, IIC_VQUNAiD,
3167 "vqmovn", "u", int_arm_neon_vqmovnu>;
3168 defm VQMOVNsu : N2VNInt_HSD<0b11,0b11,0b10,0b00100,1,0, IIC_VQUNAiD,
3169 "vqmovun", "s", int_arm_neon_vqmovnsu>;
3170 // VMOVL : Vector Lengthening Move
3171 defm VMOVLs : N2VL_QHS<0b01,0b10100,0,1, "vmovl", "s", sext>;
3172 defm VMOVLu : N2VL_QHS<0b11,0b10100,0,1, "vmovl", "u", zext>;
3174 // Vector Conversions.
3176 // VCVT : Vector Convert Between Floating-Point and Integers
3177 def VCVTf2sd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
3178 v2i32, v2f32, fp_to_sint>;
3179 def VCVTf2ud : N2VD<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
3180 v2i32, v2f32, fp_to_uint>;
3181 def VCVTs2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
3182 v2f32, v2i32, sint_to_fp>;
3183 def VCVTu2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
3184 v2f32, v2i32, uint_to_fp>;
3186 def VCVTf2sq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
3187 v4i32, v4f32, fp_to_sint>;
3188 def VCVTf2uq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
3189 v4i32, v4f32, fp_to_uint>;
3190 def VCVTs2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
3191 v4f32, v4i32, sint_to_fp>;
3192 def VCVTu2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
3193 v4f32, v4i32, uint_to_fp>;
3195 // VCVT : Vector Convert Between Floating-Point and Fixed-Point.
3196 def VCVTf2xsd : N2VCvtD<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
3197 v2i32, v2f32, int_arm_neon_vcvtfp2fxs>;
3198 def VCVTf2xud : N2VCvtD<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
3199 v2i32, v2f32, int_arm_neon_vcvtfp2fxu>;
3200 def VCVTxs2fd : N2VCvtD<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
3201 v2f32, v2i32, int_arm_neon_vcvtfxs2fp>;
3202 def VCVTxu2fd : N2VCvtD<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
3203 v2f32, v2i32, int_arm_neon_vcvtfxu2fp>;
3205 def VCVTf2xsq : N2VCvtQ<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
3206 v4i32, v4f32, int_arm_neon_vcvtfp2fxs>;
3207 def VCVTf2xuq : N2VCvtQ<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
3208 v4i32, v4f32, int_arm_neon_vcvtfp2fxu>;
3209 def VCVTxs2fq : N2VCvtQ<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
3210 v4f32, v4i32, int_arm_neon_vcvtfxs2fp>;
3211 def VCVTxu2fq : N2VCvtQ<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
3212 v4f32, v4i32, int_arm_neon_vcvtfxu2fp>;
3216 // VREV64 : Vector Reverse elements within 64-bit doublewords
3218 class VREV64D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3219 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 0, 0, (outs DPR:$dst),
3220 (ins DPR:$src), IIC_VMOVD,
3221 OpcodeStr, Dt, "$dst, $src", "",
3222 [(set DPR:$dst, (Ty (NEONvrev64 (Ty DPR:$src))))]>;
3223 class VREV64Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3224 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 1, 0, (outs QPR:$dst),
3225 (ins QPR:$src), IIC_VMOVD,
3226 OpcodeStr, Dt, "$dst, $src", "",
3227 [(set QPR:$dst, (Ty (NEONvrev64 (Ty QPR:$src))))]>;
3229 def VREV64d8 : VREV64D<0b00, "vrev64", "8", v8i8>;
3230 def VREV64d16 : VREV64D<0b01, "vrev64", "16", v4i16>;
3231 def VREV64d32 : VREV64D<0b10, "vrev64", "32", v2i32>;
3232 def VREV64df : VREV64D<0b10, "vrev64", "32", v2f32>;
3234 def VREV64q8 : VREV64Q<0b00, "vrev64", "8", v16i8>;
3235 def VREV64q16 : VREV64Q<0b01, "vrev64", "16", v8i16>;
3236 def VREV64q32 : VREV64Q<0b10, "vrev64", "32", v4i32>;
3237 def VREV64qf : VREV64Q<0b10, "vrev64", "32", v4f32>;
3239 // VREV32 : Vector Reverse elements within 32-bit words
3241 class VREV32D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3242 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 0, 0, (outs DPR:$dst),
3243 (ins DPR:$src), IIC_VMOVD,
3244 OpcodeStr, Dt, "$dst, $src", "",
3245 [(set DPR:$dst, (Ty (NEONvrev32 (Ty DPR:$src))))]>;
3246 class VREV32Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3247 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 1, 0, (outs QPR:$dst),
3248 (ins QPR:$src), IIC_VMOVD,
3249 OpcodeStr, Dt, "$dst, $src", "",
3250 [(set QPR:$dst, (Ty (NEONvrev32 (Ty QPR:$src))))]>;
3252 def VREV32d8 : VREV32D<0b00, "vrev32", "8", v8i8>;
3253 def VREV32d16 : VREV32D<0b01, "vrev32", "16", v4i16>;
3255 def VREV32q8 : VREV32Q<0b00, "vrev32", "8", v16i8>;
3256 def VREV32q16 : VREV32Q<0b01, "vrev32", "16", v8i16>;
3258 // VREV16 : Vector Reverse elements within 16-bit halfwords
3260 class VREV16D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3261 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 0, 0, (outs DPR:$dst),
3262 (ins DPR:$src), IIC_VMOVD,
3263 OpcodeStr, Dt, "$dst, $src", "",
3264 [(set DPR:$dst, (Ty (NEONvrev16 (Ty DPR:$src))))]>;
3265 class VREV16Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3266 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 1, 0, (outs QPR:$dst),
3267 (ins QPR:$src), IIC_VMOVD,
3268 OpcodeStr, Dt, "$dst, $src", "",
3269 [(set QPR:$dst, (Ty (NEONvrev16 (Ty QPR:$src))))]>;
3271 def VREV16d8 : VREV16D<0b00, "vrev16", "8", v8i8>;
3272 def VREV16q8 : VREV16Q<0b00, "vrev16", "8", v16i8>;
3274 // Other Vector Shuffles.
3276 // VEXT : Vector Extract
3278 class VEXTd<string OpcodeStr, string Dt, ValueType Ty>
3279 : N3V<0,1,0b11,{?,?,?,?},0,0, (outs DPR:$dst),
3280 (ins DPR:$lhs, DPR:$rhs, i32imm:$index), NVExtFrm,
3281 IIC_VEXTD, OpcodeStr, Dt, "$dst, $lhs, $rhs, $index", "",
3282 [(set DPR:$dst, (Ty (NEONvext (Ty DPR:$lhs),
3283 (Ty DPR:$rhs), imm:$index)))]>;
3285 class VEXTq<string OpcodeStr, string Dt, ValueType Ty>
3286 : N3V<0,1,0b11,{?,?,?,?},1,0, (outs QPR:$dst),
3287 (ins QPR:$lhs, QPR:$rhs, i32imm:$index), NVExtFrm,
3288 IIC_VEXTQ, OpcodeStr, Dt, "$dst, $lhs, $rhs, $index", "",
3289 [(set QPR:$dst, (Ty (NEONvext (Ty QPR:$lhs),
3290 (Ty QPR:$rhs), imm:$index)))]>;
3292 def VEXTd8 : VEXTd<"vext", "8", v8i8>;
3293 def VEXTd16 : VEXTd<"vext", "16", v4i16>;
3294 def VEXTd32 : VEXTd<"vext", "32", v2i32>;
3295 def VEXTdf : VEXTd<"vext", "32", v2f32>;
3297 def VEXTq8 : VEXTq<"vext", "8", v16i8>;
3298 def VEXTq16 : VEXTq<"vext", "16", v8i16>;
3299 def VEXTq32 : VEXTq<"vext", "32", v4i32>;
3300 def VEXTqf : VEXTq<"vext", "32", v4f32>;
3302 // VTRN : Vector Transpose
3304 def VTRNd8 : N2VDShuffle<0b00, 0b00001, "vtrn", "8">;
3305 def VTRNd16 : N2VDShuffle<0b01, 0b00001, "vtrn", "16">;
3306 def VTRNd32 : N2VDShuffle<0b10, 0b00001, "vtrn", "32">;
3308 def VTRNq8 : N2VQShuffle<0b00, 0b00001, IIC_VPERMQ, "vtrn", "8">;
3309 def VTRNq16 : N2VQShuffle<0b01, 0b00001, IIC_VPERMQ, "vtrn", "16">;
3310 def VTRNq32 : N2VQShuffle<0b10, 0b00001, IIC_VPERMQ, "vtrn", "32">;
3312 // VUZP : Vector Unzip (Deinterleave)
3314 def VUZPd8 : N2VDShuffle<0b00, 0b00010, "vuzp", "8">;
3315 def VUZPd16 : N2VDShuffle<0b01, 0b00010, "vuzp", "16">;
3316 def VUZPd32 : N2VDShuffle<0b10, 0b00010, "vuzp", "32">;
3318 def VUZPq8 : N2VQShuffle<0b00, 0b00010, IIC_VPERMQ3, "vuzp", "8">;
3319 def VUZPq16 : N2VQShuffle<0b01, 0b00010, IIC_VPERMQ3, "vuzp", "16">;
3320 def VUZPq32 : N2VQShuffle<0b10, 0b00010, IIC_VPERMQ3, "vuzp", "32">;
3322 // VZIP : Vector Zip (Interleave)
3324 def VZIPd8 : N2VDShuffle<0b00, 0b00011, "vzip", "8">;
3325 def VZIPd16 : N2VDShuffle<0b01, 0b00011, "vzip", "16">;
3326 def VZIPd32 : N2VDShuffle<0b10, 0b00011, "vzip", "32">;
3328 def VZIPq8 : N2VQShuffle<0b00, 0b00011, IIC_VPERMQ3, "vzip", "8">;
3329 def VZIPq16 : N2VQShuffle<0b01, 0b00011, IIC_VPERMQ3, "vzip", "16">;
3330 def VZIPq32 : N2VQShuffle<0b10, 0b00011, IIC_VPERMQ3, "vzip", "32">;
3332 // Vector Table Lookup and Table Extension.
3334 // VTBL : Vector Table Lookup
3336 : N3V<1,1,0b11,0b1000,0,0, (outs DPR:$dst),
3337 (ins DPR:$tbl1, DPR:$src), NVTBLFrm, IIC_VTB1,
3338 "vtbl", "8", "$dst, \\{$tbl1\\}, $src", "",
3339 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl1 DPR:$tbl1, DPR:$src)))]>;
3340 let hasExtraSrcRegAllocReq = 1 in {
3342 : N3V<1,1,0b11,0b1001,0,0, (outs DPR:$dst),
3343 (ins DPR:$tbl1, DPR:$tbl2, DPR:$src), NVTBLFrm, IIC_VTB2,
3344 "vtbl", "8", "$dst, \\{$tbl1, $tbl2\\}, $src", "", []>;
3346 : N3V<1,1,0b11,0b1010,0,0, (outs DPR:$dst),
3347 (ins DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$src), NVTBLFrm, IIC_VTB3,
3348 "vtbl", "8", "$dst, \\{$tbl1, $tbl2, $tbl3\\}, $src", "", []>;
3350 : N3V<1,1,0b11,0b1011,0,0, (outs DPR:$dst),
3351 (ins DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$src),
3353 "vtbl", "8", "$dst, \\{$tbl1, $tbl2, $tbl3, $tbl4\\}, $src", "", []>;
3354 } // hasExtraSrcRegAllocReq = 1
3356 // VTBX : Vector Table Extension
3358 : N3V<1,1,0b11,0b1000,1,0, (outs DPR:$dst),
3359 (ins DPR:$orig, DPR:$tbl1, DPR:$src), NVTBLFrm, IIC_VTBX1,
3360 "vtbx", "8", "$dst, \\{$tbl1\\}, $src", "$orig = $dst",
3361 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx1
3362 DPR:$orig, DPR:$tbl1, DPR:$src)))]>;
3363 let hasExtraSrcRegAllocReq = 1 in {
3365 : N3V<1,1,0b11,0b1001,1,0, (outs DPR:$dst),
3366 (ins DPR:$orig, DPR:$tbl1, DPR:$tbl2, DPR:$src), NVTBLFrm, IIC_VTBX2,
3367 "vtbx", "8", "$dst, \\{$tbl1, $tbl2\\}, $src", "$orig = $dst", []>;
3369 : N3V<1,1,0b11,0b1010,1,0, (outs DPR:$dst),
3370 (ins DPR:$orig, DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$src),
3371 NVTBLFrm, IIC_VTBX3,
3372 "vtbx", "8", "$dst, \\{$tbl1, $tbl2, $tbl3\\}, $src",
3373 "$orig = $dst", []>;
3375 : N3V<1,1,0b11,0b1011,1,0, (outs DPR:$dst), (ins DPR:$orig, DPR:$tbl1,
3376 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$src), NVTBLFrm, IIC_VTBX4,
3377 "vtbx", "8", "$dst, \\{$tbl1, $tbl2, $tbl3, $tbl4\\}, $src",
3378 "$orig = $dst", []>;
3379 } // hasExtraSrcRegAllocReq = 1
3381 //===----------------------------------------------------------------------===//
3382 // NEON instructions for single-precision FP math
3383 //===----------------------------------------------------------------------===//
3385 class N2VSPat<SDNode OpNode, ValueType ResTy, ValueType OpTy, NeonI Inst>
3386 : NEONFPPat<(ResTy (OpNode SPR:$a)),
3387 (EXTRACT_SUBREG (OpTy (Inst (INSERT_SUBREG (OpTy (IMPLICIT_DEF)),
3391 class N3VSPat<SDNode OpNode, NeonI Inst>
3392 : NEONFPPat<(f32 (OpNode SPR:$a, SPR:$b)),
3393 (EXTRACT_SUBREG (v2f32
3394 (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3396 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3400 class N3VSMulOpPat<SDNode MulNode, SDNode OpNode, NeonI Inst>
3401 : NEONFPPat<(f32 (OpNode SPR:$acc, (f32 (MulNode SPR:$a, SPR:$b)))),
3402 (EXTRACT_SUBREG (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3404 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3406 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3410 // These need separate instructions because they must use DPR_VFP2 register
3411 // class which have SPR sub-registers.
3413 // Vector Add Operations used for single-precision FP
3414 let neverHasSideEffects = 1 in
3415 def VADDfd_sfp : N3VS<0,0,0b00,0b1101,0, "vadd", "f32", v2f32, v2f32, fadd, 1>;
3416 def : N3VSPat<fadd, VADDfd_sfp>;
3418 // Vector Sub Operations used for single-precision FP
3419 let neverHasSideEffects = 1 in
3420 def VSUBfd_sfp : N3VS<0,0,0b10,0b1101,0, "vsub", "f32", v2f32, v2f32, fsub, 0>;
3421 def : N3VSPat<fsub, VSUBfd_sfp>;
3423 // Vector Multiply Operations used for single-precision FP
3424 let neverHasSideEffects = 1 in
3425 def VMULfd_sfp : N3VS<1,0,0b00,0b1101,1, "vmul", "f32", v2f32, v2f32, fmul, 1>;
3426 def : N3VSPat<fmul, VMULfd_sfp>;
3428 // Vector Multiply-Accumulate/Subtract used for single-precision FP
3429 // vml[as].f32 can cause 4-8 cycle stalls in following ASIMD instructions, so
3430 // we want to avoid them for now. e.g., alternating vmla/vadd instructions.
3432 //let neverHasSideEffects = 1 in
3433 //def VMLAfd_sfp : N3VSMulOp<0,0,0b00,0b1101,1, IIC_VMACD, "vmla", "f32",
3434 // v2f32, fmul, fadd>;
3435 //def : N3VSMulOpPat<fmul, fadd, VMLAfd_sfp>;
3437 //let neverHasSideEffects = 1 in
3438 //def VMLSfd_sfp : N3VSMulOp<0,0,0b10,0b1101,1, IIC_VMACD, "vmls", "f32",
3439 // v2f32, fmul, fsub>;
3440 //def : N3VSMulOpPat<fmul, fsub, VMLSfd_sfp>;
3442 // Vector Absolute used for single-precision FP
3443 let neverHasSideEffects = 1 in
3444 def VABSfd_sfp : N2V<0b11, 0b11, 0b10, 0b01, 0b01110, 0, 0,
3445 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), IIC_VUNAD,
3446 "vabs", "f32", "$dst, $src", "", []>;
3447 def : N2VSPat<fabs, f32, v2f32, VABSfd_sfp>;
3449 // Vector Negate used for single-precision FP
3450 let neverHasSideEffects = 1 in
3451 def VNEGfd_sfp : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
3452 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), IIC_VUNAD,
3453 "vneg", "f32", "$dst, $src", "", []>;
3454 def : N2VSPat<fneg, f32, v2f32, VNEGfd_sfp>;
3456 // Vector Maximum used for single-precision FP
3457 let neverHasSideEffects = 1 in
3458 def VMAXfd_sfp : N3V<0, 0, 0b00, 0b1111, 0, 0, (outs DPR_VFP2:$dst),
3459 (ins DPR_VFP2:$src1, DPR_VFP2:$src2), N3RegFrm, IIC_VBIND,
3460 "vmax", "f32", "$dst, $src1, $src2", "", []>;
3461 def : N3VSPat<NEONfmax, VMAXfd_sfp>;
3463 // Vector Minimum used for single-precision FP
3464 let neverHasSideEffects = 1 in
3465 def VMINfd_sfp : N3V<0, 0, 0b00, 0b1111, 0, 0, (outs DPR_VFP2:$dst),
3466 (ins DPR_VFP2:$src1, DPR_VFP2:$src2), N3RegFrm, IIC_VBIND,
3467 "vmin", "f32", "$dst, $src1, $src2", "", []>;
3468 def : N3VSPat<NEONfmin, VMINfd_sfp>;
3470 // Vector Convert between single-precision FP and integer
3471 let neverHasSideEffects = 1 in
3472 def VCVTf2sd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
3473 v2i32, v2f32, fp_to_sint>;
3474 def : N2VSPat<arm_ftosi, f32, v2f32, VCVTf2sd_sfp>;
3476 let neverHasSideEffects = 1 in
3477 def VCVTf2ud_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
3478 v2i32, v2f32, fp_to_uint>;
3479 def : N2VSPat<arm_ftoui, f32, v2f32, VCVTf2ud_sfp>;
3481 let neverHasSideEffects = 1 in
3482 def VCVTs2fd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
3483 v2f32, v2i32, sint_to_fp>;
3484 def : N2VSPat<arm_sitof, f32, v2i32, VCVTs2fd_sfp>;
3486 let neverHasSideEffects = 1 in
3487 def VCVTu2fd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
3488 v2f32, v2i32, uint_to_fp>;
3489 def : N2VSPat<arm_uitof, f32, v2i32, VCVTu2fd_sfp>;
3491 //===----------------------------------------------------------------------===//
3492 // Non-Instruction Patterns
3493 //===----------------------------------------------------------------------===//
3496 def : Pat<(v1i64 (bitconvert (v2i32 DPR:$src))), (v1i64 DPR:$src)>;
3497 def : Pat<(v1i64 (bitconvert (v4i16 DPR:$src))), (v1i64 DPR:$src)>;
3498 def : Pat<(v1i64 (bitconvert (v8i8 DPR:$src))), (v1i64 DPR:$src)>;
3499 def : Pat<(v1i64 (bitconvert (f64 DPR:$src))), (v1i64 DPR:$src)>;
3500 def : Pat<(v1i64 (bitconvert (v2f32 DPR:$src))), (v1i64 DPR:$src)>;
3501 def : Pat<(v2i32 (bitconvert (v1i64 DPR:$src))), (v2i32 DPR:$src)>;
3502 def : Pat<(v2i32 (bitconvert (v4i16 DPR:$src))), (v2i32 DPR:$src)>;
3503 def : Pat<(v2i32 (bitconvert (v8i8 DPR:$src))), (v2i32 DPR:$src)>;
3504 def : Pat<(v2i32 (bitconvert (f64 DPR:$src))), (v2i32 DPR:$src)>;
3505 def : Pat<(v2i32 (bitconvert (v2f32 DPR:$src))), (v2i32 DPR:$src)>;
3506 def : Pat<(v4i16 (bitconvert (v1i64 DPR:$src))), (v4i16 DPR:$src)>;
3507 def : Pat<(v4i16 (bitconvert (v2i32 DPR:$src))), (v4i16 DPR:$src)>;
3508 def : Pat<(v4i16 (bitconvert (v8i8 DPR:$src))), (v4i16 DPR:$src)>;
3509 def : Pat<(v4i16 (bitconvert (f64 DPR:$src))), (v4i16 DPR:$src)>;
3510 def : Pat<(v4i16 (bitconvert (v2f32 DPR:$src))), (v4i16 DPR:$src)>;
3511 def : Pat<(v8i8 (bitconvert (v1i64 DPR:$src))), (v8i8 DPR:$src)>;
3512 def : Pat<(v8i8 (bitconvert (v2i32 DPR:$src))), (v8i8 DPR:$src)>;
3513 def : Pat<(v8i8 (bitconvert (v4i16 DPR:$src))), (v8i8 DPR:$src)>;
3514 def : Pat<(v8i8 (bitconvert (f64 DPR:$src))), (v8i8 DPR:$src)>;
3515 def : Pat<(v8i8 (bitconvert (v2f32 DPR:$src))), (v8i8 DPR:$src)>;
3516 def : Pat<(f64 (bitconvert (v1i64 DPR:$src))), (f64 DPR:$src)>;
3517 def : Pat<(f64 (bitconvert (v2i32 DPR:$src))), (f64 DPR:$src)>;
3518 def : Pat<(f64 (bitconvert (v4i16 DPR:$src))), (f64 DPR:$src)>;
3519 def : Pat<(f64 (bitconvert (v8i8 DPR:$src))), (f64 DPR:$src)>;
3520 def : Pat<(f64 (bitconvert (v2f32 DPR:$src))), (f64 DPR:$src)>;
3521 def : Pat<(v2f32 (bitconvert (f64 DPR:$src))), (v2f32 DPR:$src)>;
3522 def : Pat<(v2f32 (bitconvert (v1i64 DPR:$src))), (v2f32 DPR:$src)>;
3523 def : Pat<(v2f32 (bitconvert (v2i32 DPR:$src))), (v2f32 DPR:$src)>;
3524 def : Pat<(v2f32 (bitconvert (v4i16 DPR:$src))), (v2f32 DPR:$src)>;
3525 def : Pat<(v2f32 (bitconvert (v8i8 DPR:$src))), (v2f32 DPR:$src)>;
3527 def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (v2i64 QPR:$src)>;
3528 def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (v2i64 QPR:$src)>;
3529 def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (v2i64 QPR:$src)>;
3530 def : Pat<(v2i64 (bitconvert (v2f64 QPR:$src))), (v2i64 QPR:$src)>;
3531 def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (v2i64 QPR:$src)>;
3532 def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (v4i32 QPR:$src)>;
3533 def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (v4i32 QPR:$src)>;
3534 def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (v4i32 QPR:$src)>;
3535 def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (v4i32 QPR:$src)>;
3536 def : Pat<(v4i32 (bitconvert (v4f32 QPR:$src))), (v4i32 QPR:$src)>;
3537 def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (v8i16 QPR:$src)>;
3538 def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (v8i16 QPR:$src)>;
3539 def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (v8i16 QPR:$src)>;
3540 def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (v8i16 QPR:$src)>;
3541 def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (v8i16 QPR:$src)>;
3542 def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (v16i8 QPR:$src)>;
3543 def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (v16i8 QPR:$src)>;
3544 def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (v16i8 QPR:$src)>;
3545 def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (v16i8 QPR:$src)>;
3546 def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (v16i8 QPR:$src)>;
3547 def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (v4f32 QPR:$src)>;
3548 def : Pat<(v4f32 (bitconvert (v4i32 QPR:$src))), (v4f32 QPR:$src)>;
3549 def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (v4f32 QPR:$src)>;
3550 def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (v4f32 QPR:$src)>;
3551 def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (v4f32 QPR:$src)>;
3552 def : Pat<(v2f64 (bitconvert (v2i64 QPR:$src))), (v2f64 QPR:$src)>;
3553 def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (v2f64 QPR:$src)>;
3554 def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 QPR:$src)>;
3555 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 QPR:$src)>;
3556 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 QPR:$src)>;